2 * Kernel-based Virtual Machine driver for Linux
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
7 * Copyright (C) 2006 Qumranet, Inc.
8 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
11 * Avi Kivity <avi@qumranet.com>
12 * Yaniv Kamay <yaniv@qumranet.com>
14 * This work is licensed under the terms of the GNU GPL, version 2. See
15 * the COPYING file in the top-level directory.
23 #include <linux/kvm_host.h>
24 #include <linux/module.h>
25 #include <linux/kernel.h>
27 #include <linux/highmem.h>
28 #include <linux/sched.h>
29 #include <linux/moduleparam.h>
30 #include <linux/mod_devicetable.h>
31 #include <linux/trace_events.h>
32 #include <linux/slab.h>
33 #include <linux/tboot.h>
34 #include <linux/hrtimer.h>
35 #include "kvm_cache_regs.h"
42 #include <asm/virtext.h>
44 #include <asm/fpu/internal.h>
45 #include <asm/perf_event.h>
46 #include <asm/debugreg.h>
47 #include <asm/kexec.h>
49 #include <asm/irq_remapping.h>
54 #define __ex(x) __kvm_handle_fault_on_reboot(x)
55 #define __ex_clear(x, reg) \
56 ____kvm_handle_fault_on_reboot(x, "xor " reg " , " reg)
58 MODULE_AUTHOR("Qumranet");
59 MODULE_LICENSE("GPL");
61 static const struct x86_cpu_id vmx_cpu_id[] = {
62 X86_FEATURE_MATCH(X86_FEATURE_VMX),
65 MODULE_DEVICE_TABLE(x86cpu, vmx_cpu_id);
67 static bool __read_mostly enable_vpid = 1;
68 module_param_named(vpid, enable_vpid, bool, 0444);
70 static bool __read_mostly flexpriority_enabled = 1;
71 module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
73 static bool __read_mostly enable_ept = 1;
74 module_param_named(ept, enable_ept, bool, S_IRUGO);
76 static bool __read_mostly enable_unrestricted_guest = 1;
77 module_param_named(unrestricted_guest,
78 enable_unrestricted_guest, bool, S_IRUGO);
80 static bool __read_mostly enable_ept_ad_bits = 1;
81 module_param_named(eptad, enable_ept_ad_bits, bool, S_IRUGO);
83 static bool __read_mostly emulate_invalid_guest_state = true;
84 module_param(emulate_invalid_guest_state, bool, S_IRUGO);
86 static bool __read_mostly vmm_exclusive = 1;
87 module_param(vmm_exclusive, bool, S_IRUGO);
89 static bool __read_mostly fasteoi = 1;
90 module_param(fasteoi, bool, S_IRUGO);
92 static bool __read_mostly enable_apicv = 1;
93 module_param(enable_apicv, bool, S_IRUGO);
95 static bool __read_mostly enable_shadow_vmcs = 1;
96 module_param_named(enable_shadow_vmcs, enable_shadow_vmcs, bool, S_IRUGO);
98 * If nested=1, nested virtualization is supported, i.e., guests may use
99 * VMX and be a hypervisor for its own guests. If nested=0, guests may not
100 * use VMX instructions.
102 static bool __read_mostly nested = 0;
103 module_param(nested, bool, S_IRUGO);
105 static u64 __read_mostly host_xss;
107 static bool __read_mostly enable_pml = 1;
108 module_param_named(pml, enable_pml, bool, S_IRUGO);
110 #define KVM_VMX_TSC_MULTIPLIER_MAX 0xffffffffffffffffULL
112 #define KVM_GUEST_CR0_MASK (X86_CR0_NW | X86_CR0_CD)
113 #define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST (X86_CR0_WP | X86_CR0_NE)
114 #define KVM_VM_CR0_ALWAYS_ON \
115 (KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
116 #define KVM_CR4_GUEST_OWNED_BITS \
117 (X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR \
118 | X86_CR4_OSXMMEXCPT | X86_CR4_TSD)
120 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
121 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
123 #define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
125 #define VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE 5
128 * These 2 parameters are used to config the controls for Pause-Loop Exiting:
129 * ple_gap: upper bound on the amount of time between two successive
130 * executions of PAUSE in a loop. Also indicate if ple enabled.
131 * According to test, this time is usually smaller than 128 cycles.
132 * ple_window: upper bound on the amount of time a guest is allowed to execute
133 * in a PAUSE loop. Tests indicate that most spinlocks are held for
134 * less than 2^12 cycles
135 * Time is measured based on a counter that runs at the same rate as the TSC,
136 * refer SDM volume 3b section 21.6.13 & 22.1.3.
138 #define KVM_VMX_DEFAULT_PLE_GAP 128
139 #define KVM_VMX_DEFAULT_PLE_WINDOW 4096
140 #define KVM_VMX_DEFAULT_PLE_WINDOW_GROW 2
141 #define KVM_VMX_DEFAULT_PLE_WINDOW_SHRINK 0
142 #define KVM_VMX_DEFAULT_PLE_WINDOW_MAX \
143 INT_MAX / KVM_VMX_DEFAULT_PLE_WINDOW_GROW
145 static int ple_gap = KVM_VMX_DEFAULT_PLE_GAP;
146 module_param(ple_gap, int, S_IRUGO);
148 static int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
149 module_param(ple_window, int, S_IRUGO);
151 /* Default doubles per-vcpu window every exit. */
152 static int ple_window_grow = KVM_VMX_DEFAULT_PLE_WINDOW_GROW;
153 module_param(ple_window_grow, int, S_IRUGO);
155 /* Default resets per-vcpu window every exit to ple_window. */
156 static int ple_window_shrink = KVM_VMX_DEFAULT_PLE_WINDOW_SHRINK;
157 module_param(ple_window_shrink, int, S_IRUGO);
159 /* Default is to compute the maximum so we can never overflow. */
160 static int ple_window_actual_max = KVM_VMX_DEFAULT_PLE_WINDOW_MAX;
161 static int ple_window_max = KVM_VMX_DEFAULT_PLE_WINDOW_MAX;
162 module_param(ple_window_max, int, S_IRUGO);
164 extern const ulong vmx_return;
166 #define NR_AUTOLOAD_MSRS 8
167 #define VMCS02_POOL_SIZE 1
176 * Track a VMCS that may be loaded on a certain CPU. If it is (cpu!=-1), also
177 * remember whether it was VMLAUNCHed, and maintain a linked list of all VMCSs
178 * loaded on this CPU (so we can clear them if the CPU goes down).
184 struct list_head loaded_vmcss_on_cpu_link;
187 struct shared_msr_entry {
194 * struct vmcs12 describes the state that our guest hypervisor (L1) keeps for a
195 * single nested guest (L2), hence the name vmcs12. Any VMX implementation has
196 * a VMCS structure, and vmcs12 is our emulated VMX's VMCS. This structure is
197 * stored in guest memory specified by VMPTRLD, but is opaque to the guest,
198 * which must access it using VMREAD/VMWRITE/VMCLEAR instructions.
199 * More than one of these structures may exist, if L1 runs multiple L2 guests.
200 * nested_vmx_run() will use the data here to build a vmcs02: a VMCS for the
201 * underlying hardware which will be used to run L2.
202 * This structure is packed to ensure that its layout is identical across
203 * machines (necessary for live migration).
204 * If there are changes in this struct, VMCS12_REVISION must be changed.
206 typedef u64 natural_width;
207 struct __packed vmcs12 {
208 /* According to the Intel spec, a VMCS region must start with the
209 * following two fields. Then follow implementation-specific data.
214 u32 launch_state; /* set to 0 by VMCLEAR, to 1 by VMLAUNCH */
215 u32 padding[7]; /* room for future expansion */
220 u64 vm_exit_msr_store_addr;
221 u64 vm_exit_msr_load_addr;
222 u64 vm_entry_msr_load_addr;
224 u64 virtual_apic_page_addr;
225 u64 apic_access_addr;
226 u64 posted_intr_desc_addr;
228 u64 eoi_exit_bitmap0;
229 u64 eoi_exit_bitmap1;
230 u64 eoi_exit_bitmap2;
231 u64 eoi_exit_bitmap3;
233 u64 guest_physical_address;
234 u64 vmcs_link_pointer;
235 u64 guest_ia32_debugctl;
238 u64 guest_ia32_perf_global_ctrl;
246 u64 host_ia32_perf_global_ctrl;
247 u64 padding64[8]; /* room for future expansion */
249 * To allow migration of L1 (complete with its L2 guests) between
250 * machines of different natural widths (32 or 64 bit), we cannot have
251 * unsigned long fields with no explict size. We use u64 (aliased
252 * natural_width) instead. Luckily, x86 is little-endian.
254 natural_width cr0_guest_host_mask;
255 natural_width cr4_guest_host_mask;
256 natural_width cr0_read_shadow;
257 natural_width cr4_read_shadow;
258 natural_width cr3_target_value0;
259 natural_width cr3_target_value1;
260 natural_width cr3_target_value2;
261 natural_width cr3_target_value3;
262 natural_width exit_qualification;
263 natural_width guest_linear_address;
264 natural_width guest_cr0;
265 natural_width guest_cr3;
266 natural_width guest_cr4;
267 natural_width guest_es_base;
268 natural_width guest_cs_base;
269 natural_width guest_ss_base;
270 natural_width guest_ds_base;
271 natural_width guest_fs_base;
272 natural_width guest_gs_base;
273 natural_width guest_ldtr_base;
274 natural_width guest_tr_base;
275 natural_width guest_gdtr_base;
276 natural_width guest_idtr_base;
277 natural_width guest_dr7;
278 natural_width guest_rsp;
279 natural_width guest_rip;
280 natural_width guest_rflags;
281 natural_width guest_pending_dbg_exceptions;
282 natural_width guest_sysenter_esp;
283 natural_width guest_sysenter_eip;
284 natural_width host_cr0;
285 natural_width host_cr3;
286 natural_width host_cr4;
287 natural_width host_fs_base;
288 natural_width host_gs_base;
289 natural_width host_tr_base;
290 natural_width host_gdtr_base;
291 natural_width host_idtr_base;
292 natural_width host_ia32_sysenter_esp;
293 natural_width host_ia32_sysenter_eip;
294 natural_width host_rsp;
295 natural_width host_rip;
296 natural_width paddingl[8]; /* room for future expansion */
297 u32 pin_based_vm_exec_control;
298 u32 cpu_based_vm_exec_control;
299 u32 exception_bitmap;
300 u32 page_fault_error_code_mask;
301 u32 page_fault_error_code_match;
302 u32 cr3_target_count;
303 u32 vm_exit_controls;
304 u32 vm_exit_msr_store_count;
305 u32 vm_exit_msr_load_count;
306 u32 vm_entry_controls;
307 u32 vm_entry_msr_load_count;
308 u32 vm_entry_intr_info_field;
309 u32 vm_entry_exception_error_code;
310 u32 vm_entry_instruction_len;
312 u32 secondary_vm_exec_control;
313 u32 vm_instruction_error;
315 u32 vm_exit_intr_info;
316 u32 vm_exit_intr_error_code;
317 u32 idt_vectoring_info_field;
318 u32 idt_vectoring_error_code;
319 u32 vm_exit_instruction_len;
320 u32 vmx_instruction_info;
327 u32 guest_ldtr_limit;
329 u32 guest_gdtr_limit;
330 u32 guest_idtr_limit;
331 u32 guest_es_ar_bytes;
332 u32 guest_cs_ar_bytes;
333 u32 guest_ss_ar_bytes;
334 u32 guest_ds_ar_bytes;
335 u32 guest_fs_ar_bytes;
336 u32 guest_gs_ar_bytes;
337 u32 guest_ldtr_ar_bytes;
338 u32 guest_tr_ar_bytes;
339 u32 guest_interruptibility_info;
340 u32 guest_activity_state;
341 u32 guest_sysenter_cs;
342 u32 host_ia32_sysenter_cs;
343 u32 vmx_preemption_timer_value;
344 u32 padding32[7]; /* room for future expansion */
345 u16 virtual_processor_id;
347 u16 guest_es_selector;
348 u16 guest_cs_selector;
349 u16 guest_ss_selector;
350 u16 guest_ds_selector;
351 u16 guest_fs_selector;
352 u16 guest_gs_selector;
353 u16 guest_ldtr_selector;
354 u16 guest_tr_selector;
355 u16 guest_intr_status;
356 u16 host_es_selector;
357 u16 host_cs_selector;
358 u16 host_ss_selector;
359 u16 host_ds_selector;
360 u16 host_fs_selector;
361 u16 host_gs_selector;
362 u16 host_tr_selector;
366 * VMCS12_REVISION is an arbitrary id that should be changed if the content or
367 * layout of struct vmcs12 is changed. MSR_IA32_VMX_BASIC returns this id, and
368 * VMPTRLD verifies that the VMCS region that L1 is loading contains this id.
370 #define VMCS12_REVISION 0x11e57ed0
373 * VMCS12_SIZE is the number of bytes L1 should allocate for the VMXON region
374 * and any VMCS region. Although only sizeof(struct vmcs12) are used by the
375 * current implementation, 4K are reserved to avoid future complications.
377 #define VMCS12_SIZE 0x1000
379 /* Used to remember the last vmcs02 used for some recently used vmcs12s */
381 struct list_head list;
383 struct loaded_vmcs vmcs02;
387 * The nested_vmx structure is part of vcpu_vmx, and holds information we need
388 * for correct emulation of VMX (i.e., nested VMX) on this vcpu.
391 /* Has the level1 guest done vmxon? */
395 /* The guest-physical address of the current VMCS L1 keeps for L2 */
397 /* The host-usable pointer to the above */
398 struct page *current_vmcs12_page;
399 struct vmcs12 *current_vmcs12;
400 struct vmcs *current_shadow_vmcs;
402 * Indicates if the shadow vmcs must be updated with the
403 * data hold by vmcs12
405 bool sync_shadow_vmcs;
407 /* vmcs02_list cache of VMCSs recently used to run L2 guests */
408 struct list_head vmcs02_pool;
410 u64 vmcs01_tsc_offset;
411 /* L2 must run next, and mustn't decide to exit to L1. */
412 bool nested_run_pending;
414 * Guest pages referred to in vmcs02 with host-physical pointers, so
415 * we must keep them pinned while L2 runs.
417 struct page *apic_access_page;
418 struct page *virtual_apic_page;
419 struct page *pi_desc_page;
420 struct pi_desc *pi_desc;
423 u64 msr_ia32_feature_control;
425 struct hrtimer preemption_timer;
426 bool preemption_timer_expired;
428 /* to migrate it to L2 if VM_ENTRY_LOAD_DEBUG_CONTROLS is off */
434 u32 nested_vmx_procbased_ctls_low;
435 u32 nested_vmx_procbased_ctls_high;
436 u32 nested_vmx_true_procbased_ctls_low;
437 u32 nested_vmx_secondary_ctls_low;
438 u32 nested_vmx_secondary_ctls_high;
439 u32 nested_vmx_pinbased_ctls_low;
440 u32 nested_vmx_pinbased_ctls_high;
441 u32 nested_vmx_exit_ctls_low;
442 u32 nested_vmx_exit_ctls_high;
443 u32 nested_vmx_true_exit_ctls_low;
444 u32 nested_vmx_entry_ctls_low;
445 u32 nested_vmx_entry_ctls_high;
446 u32 nested_vmx_true_entry_ctls_low;
447 u32 nested_vmx_misc_low;
448 u32 nested_vmx_misc_high;
449 u32 nested_vmx_ept_caps;
450 u32 nested_vmx_vpid_caps;
453 #define POSTED_INTR_ON 0
454 #define POSTED_INTR_SN 1
456 /* Posted-Interrupt Descriptor */
458 u32 pir[8]; /* Posted interrupt requested */
461 /* bit 256 - Outstanding Notification */
463 /* bit 257 - Suppress Notification */
465 /* bit 271:258 - Reserved */
467 /* bit 279:272 - Notification Vector */
469 /* bit 287:280 - Reserved */
471 /* bit 319:288 - Notification Destination */
479 static bool pi_test_and_set_on(struct pi_desc *pi_desc)
481 return test_and_set_bit(POSTED_INTR_ON,
482 (unsigned long *)&pi_desc->control);
485 static bool pi_test_and_clear_on(struct pi_desc *pi_desc)
487 return test_and_clear_bit(POSTED_INTR_ON,
488 (unsigned long *)&pi_desc->control);
491 static int pi_test_and_set_pir(int vector, struct pi_desc *pi_desc)
493 return test_and_set_bit(vector, (unsigned long *)pi_desc->pir);
496 static inline void pi_clear_sn(struct pi_desc *pi_desc)
498 return clear_bit(POSTED_INTR_SN,
499 (unsigned long *)&pi_desc->control);
502 static inline void pi_set_sn(struct pi_desc *pi_desc)
504 return set_bit(POSTED_INTR_SN,
505 (unsigned long *)&pi_desc->control);
508 static inline int pi_test_on(struct pi_desc *pi_desc)
510 return test_bit(POSTED_INTR_ON,
511 (unsigned long *)&pi_desc->control);
514 static inline int pi_test_sn(struct pi_desc *pi_desc)
516 return test_bit(POSTED_INTR_SN,
517 (unsigned long *)&pi_desc->control);
521 struct kvm_vcpu vcpu;
522 unsigned long host_rsp;
524 bool nmi_known_unmasked;
526 u32 idt_vectoring_info;
528 struct shared_msr_entry *guest_msrs;
531 unsigned long host_idt_base;
533 u64 msr_host_kernel_gs_base;
534 u64 msr_guest_kernel_gs_base;
536 u32 vm_entry_controls_shadow;
537 u32 vm_exit_controls_shadow;
539 * loaded_vmcs points to the VMCS currently used in this vcpu. For a
540 * non-nested (L1) guest, it always points to vmcs01. For a nested
541 * guest (L2), it points to a different VMCS.
543 struct loaded_vmcs vmcs01;
544 struct loaded_vmcs *loaded_vmcs;
545 bool __launched; /* temporary, used in vmx_vcpu_run */
546 struct msr_autoload {
548 struct vmx_msr_entry guest[NR_AUTOLOAD_MSRS];
549 struct vmx_msr_entry host[NR_AUTOLOAD_MSRS];
553 u16 fs_sel, gs_sel, ldt_sel;
557 int gs_ldt_reload_needed;
558 int fs_reload_needed;
559 u64 msr_host_bndcfgs;
560 unsigned long vmcs_host_cr4; /* May not match real cr4 */
565 struct kvm_segment segs[8];
568 u32 bitmask; /* 4 bits per segment (1 bit per field) */
569 struct kvm_save_segment {
577 bool emulation_required;
579 /* Support for vnmi-less CPUs */
580 int soft_vnmi_blocked;
582 s64 vnmi_blocked_time;
585 /* Posted interrupt descriptor */
586 struct pi_desc pi_desc;
588 /* Support for a guest hypervisor (nested VMX) */
589 struct nested_vmx nested;
591 /* Dynamic PLE window. */
593 bool ple_window_dirty;
595 /* Support for PML */
596 #define PML_ENTITY_NUM 512
599 u64 current_tsc_ratio;
602 enum segment_cache_field {
611 static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
613 return container_of(vcpu, struct vcpu_vmx, vcpu);
616 static struct pi_desc *vcpu_to_pi_desc(struct kvm_vcpu *vcpu)
618 return &(to_vmx(vcpu)->pi_desc);
621 #define VMCS12_OFFSET(x) offsetof(struct vmcs12, x)
622 #define FIELD(number, name) [number] = VMCS12_OFFSET(name)
623 #define FIELD64(number, name) [number] = VMCS12_OFFSET(name), \
624 [number##_HIGH] = VMCS12_OFFSET(name)+4
627 static unsigned long shadow_read_only_fields[] = {
629 * We do NOT shadow fields that are modified when L0
630 * traps and emulates any vmx instruction (e.g. VMPTRLD,
631 * VMXON...) executed by L1.
632 * For example, VM_INSTRUCTION_ERROR is read
633 * by L1 if a vmx instruction fails (part of the error path).
634 * Note the code assumes this logic. If for some reason
635 * we start shadowing these fields then we need to
636 * force a shadow sync when L0 emulates vmx instructions
637 * (e.g. force a sync if VM_INSTRUCTION_ERROR is modified
638 * by nested_vmx_failValid)
642 VM_EXIT_INSTRUCTION_LEN,
643 IDT_VECTORING_INFO_FIELD,
644 IDT_VECTORING_ERROR_CODE,
645 VM_EXIT_INTR_ERROR_CODE,
647 GUEST_LINEAR_ADDRESS,
648 GUEST_PHYSICAL_ADDRESS
650 static int max_shadow_read_only_fields =
651 ARRAY_SIZE(shadow_read_only_fields);
653 static unsigned long shadow_read_write_fields[] = {
660 GUEST_INTERRUPTIBILITY_INFO,
673 CPU_BASED_VM_EXEC_CONTROL,
674 VM_ENTRY_EXCEPTION_ERROR_CODE,
675 VM_ENTRY_INTR_INFO_FIELD,
676 VM_ENTRY_INSTRUCTION_LEN,
677 VM_ENTRY_EXCEPTION_ERROR_CODE,
683 static int max_shadow_read_write_fields =
684 ARRAY_SIZE(shadow_read_write_fields);
686 static const unsigned short vmcs_field_to_offset_table[] = {
687 FIELD(VIRTUAL_PROCESSOR_ID, virtual_processor_id),
688 FIELD(POSTED_INTR_NV, posted_intr_nv),
689 FIELD(GUEST_ES_SELECTOR, guest_es_selector),
690 FIELD(GUEST_CS_SELECTOR, guest_cs_selector),
691 FIELD(GUEST_SS_SELECTOR, guest_ss_selector),
692 FIELD(GUEST_DS_SELECTOR, guest_ds_selector),
693 FIELD(GUEST_FS_SELECTOR, guest_fs_selector),
694 FIELD(GUEST_GS_SELECTOR, guest_gs_selector),
695 FIELD(GUEST_LDTR_SELECTOR, guest_ldtr_selector),
696 FIELD(GUEST_TR_SELECTOR, guest_tr_selector),
697 FIELD(GUEST_INTR_STATUS, guest_intr_status),
698 FIELD(HOST_ES_SELECTOR, host_es_selector),
699 FIELD(HOST_CS_SELECTOR, host_cs_selector),
700 FIELD(HOST_SS_SELECTOR, host_ss_selector),
701 FIELD(HOST_DS_SELECTOR, host_ds_selector),
702 FIELD(HOST_FS_SELECTOR, host_fs_selector),
703 FIELD(HOST_GS_SELECTOR, host_gs_selector),
704 FIELD(HOST_TR_SELECTOR, host_tr_selector),
705 FIELD64(IO_BITMAP_A, io_bitmap_a),
706 FIELD64(IO_BITMAP_B, io_bitmap_b),
707 FIELD64(MSR_BITMAP, msr_bitmap),
708 FIELD64(VM_EXIT_MSR_STORE_ADDR, vm_exit_msr_store_addr),
709 FIELD64(VM_EXIT_MSR_LOAD_ADDR, vm_exit_msr_load_addr),
710 FIELD64(VM_ENTRY_MSR_LOAD_ADDR, vm_entry_msr_load_addr),
711 FIELD64(TSC_OFFSET, tsc_offset),
712 FIELD64(VIRTUAL_APIC_PAGE_ADDR, virtual_apic_page_addr),
713 FIELD64(APIC_ACCESS_ADDR, apic_access_addr),
714 FIELD64(POSTED_INTR_DESC_ADDR, posted_intr_desc_addr),
715 FIELD64(EPT_POINTER, ept_pointer),
716 FIELD64(EOI_EXIT_BITMAP0, eoi_exit_bitmap0),
717 FIELD64(EOI_EXIT_BITMAP1, eoi_exit_bitmap1),
718 FIELD64(EOI_EXIT_BITMAP2, eoi_exit_bitmap2),
719 FIELD64(EOI_EXIT_BITMAP3, eoi_exit_bitmap3),
720 FIELD64(XSS_EXIT_BITMAP, xss_exit_bitmap),
721 FIELD64(GUEST_PHYSICAL_ADDRESS, guest_physical_address),
722 FIELD64(VMCS_LINK_POINTER, vmcs_link_pointer),
723 FIELD64(GUEST_IA32_DEBUGCTL, guest_ia32_debugctl),
724 FIELD64(GUEST_IA32_PAT, guest_ia32_pat),
725 FIELD64(GUEST_IA32_EFER, guest_ia32_efer),
726 FIELD64(GUEST_IA32_PERF_GLOBAL_CTRL, guest_ia32_perf_global_ctrl),
727 FIELD64(GUEST_PDPTR0, guest_pdptr0),
728 FIELD64(GUEST_PDPTR1, guest_pdptr1),
729 FIELD64(GUEST_PDPTR2, guest_pdptr2),
730 FIELD64(GUEST_PDPTR3, guest_pdptr3),
731 FIELD64(GUEST_BNDCFGS, guest_bndcfgs),
732 FIELD64(HOST_IA32_PAT, host_ia32_pat),
733 FIELD64(HOST_IA32_EFER, host_ia32_efer),
734 FIELD64(HOST_IA32_PERF_GLOBAL_CTRL, host_ia32_perf_global_ctrl),
735 FIELD(PIN_BASED_VM_EXEC_CONTROL, pin_based_vm_exec_control),
736 FIELD(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control),
737 FIELD(EXCEPTION_BITMAP, exception_bitmap),
738 FIELD(PAGE_FAULT_ERROR_CODE_MASK, page_fault_error_code_mask),
739 FIELD(PAGE_FAULT_ERROR_CODE_MATCH, page_fault_error_code_match),
740 FIELD(CR3_TARGET_COUNT, cr3_target_count),
741 FIELD(VM_EXIT_CONTROLS, vm_exit_controls),
742 FIELD(VM_EXIT_MSR_STORE_COUNT, vm_exit_msr_store_count),
743 FIELD(VM_EXIT_MSR_LOAD_COUNT, vm_exit_msr_load_count),
744 FIELD(VM_ENTRY_CONTROLS, vm_entry_controls),
745 FIELD(VM_ENTRY_MSR_LOAD_COUNT, vm_entry_msr_load_count),
746 FIELD(VM_ENTRY_INTR_INFO_FIELD, vm_entry_intr_info_field),
747 FIELD(VM_ENTRY_EXCEPTION_ERROR_CODE, vm_entry_exception_error_code),
748 FIELD(VM_ENTRY_INSTRUCTION_LEN, vm_entry_instruction_len),
749 FIELD(TPR_THRESHOLD, tpr_threshold),
750 FIELD(SECONDARY_VM_EXEC_CONTROL, secondary_vm_exec_control),
751 FIELD(VM_INSTRUCTION_ERROR, vm_instruction_error),
752 FIELD(VM_EXIT_REASON, vm_exit_reason),
753 FIELD(VM_EXIT_INTR_INFO, vm_exit_intr_info),
754 FIELD(VM_EXIT_INTR_ERROR_CODE, vm_exit_intr_error_code),
755 FIELD(IDT_VECTORING_INFO_FIELD, idt_vectoring_info_field),
756 FIELD(IDT_VECTORING_ERROR_CODE, idt_vectoring_error_code),
757 FIELD(VM_EXIT_INSTRUCTION_LEN, vm_exit_instruction_len),
758 FIELD(VMX_INSTRUCTION_INFO, vmx_instruction_info),
759 FIELD(GUEST_ES_LIMIT, guest_es_limit),
760 FIELD(GUEST_CS_LIMIT, guest_cs_limit),
761 FIELD(GUEST_SS_LIMIT, guest_ss_limit),
762 FIELD(GUEST_DS_LIMIT, guest_ds_limit),
763 FIELD(GUEST_FS_LIMIT, guest_fs_limit),
764 FIELD(GUEST_GS_LIMIT, guest_gs_limit),
765 FIELD(GUEST_LDTR_LIMIT, guest_ldtr_limit),
766 FIELD(GUEST_TR_LIMIT, guest_tr_limit),
767 FIELD(GUEST_GDTR_LIMIT, guest_gdtr_limit),
768 FIELD(GUEST_IDTR_LIMIT, guest_idtr_limit),
769 FIELD(GUEST_ES_AR_BYTES, guest_es_ar_bytes),
770 FIELD(GUEST_CS_AR_BYTES, guest_cs_ar_bytes),
771 FIELD(GUEST_SS_AR_BYTES, guest_ss_ar_bytes),
772 FIELD(GUEST_DS_AR_BYTES, guest_ds_ar_bytes),
773 FIELD(GUEST_FS_AR_BYTES, guest_fs_ar_bytes),
774 FIELD(GUEST_GS_AR_BYTES, guest_gs_ar_bytes),
775 FIELD(GUEST_LDTR_AR_BYTES, guest_ldtr_ar_bytes),
776 FIELD(GUEST_TR_AR_BYTES, guest_tr_ar_bytes),
777 FIELD(GUEST_INTERRUPTIBILITY_INFO, guest_interruptibility_info),
778 FIELD(GUEST_ACTIVITY_STATE, guest_activity_state),
779 FIELD(GUEST_SYSENTER_CS, guest_sysenter_cs),
780 FIELD(HOST_IA32_SYSENTER_CS, host_ia32_sysenter_cs),
781 FIELD(VMX_PREEMPTION_TIMER_VALUE, vmx_preemption_timer_value),
782 FIELD(CR0_GUEST_HOST_MASK, cr0_guest_host_mask),
783 FIELD(CR4_GUEST_HOST_MASK, cr4_guest_host_mask),
784 FIELD(CR0_READ_SHADOW, cr0_read_shadow),
785 FIELD(CR4_READ_SHADOW, cr4_read_shadow),
786 FIELD(CR3_TARGET_VALUE0, cr3_target_value0),
787 FIELD(CR3_TARGET_VALUE1, cr3_target_value1),
788 FIELD(CR3_TARGET_VALUE2, cr3_target_value2),
789 FIELD(CR3_TARGET_VALUE3, cr3_target_value3),
790 FIELD(EXIT_QUALIFICATION, exit_qualification),
791 FIELD(GUEST_LINEAR_ADDRESS, guest_linear_address),
792 FIELD(GUEST_CR0, guest_cr0),
793 FIELD(GUEST_CR3, guest_cr3),
794 FIELD(GUEST_CR4, guest_cr4),
795 FIELD(GUEST_ES_BASE, guest_es_base),
796 FIELD(GUEST_CS_BASE, guest_cs_base),
797 FIELD(GUEST_SS_BASE, guest_ss_base),
798 FIELD(GUEST_DS_BASE, guest_ds_base),
799 FIELD(GUEST_FS_BASE, guest_fs_base),
800 FIELD(GUEST_GS_BASE, guest_gs_base),
801 FIELD(GUEST_LDTR_BASE, guest_ldtr_base),
802 FIELD(GUEST_TR_BASE, guest_tr_base),
803 FIELD(GUEST_GDTR_BASE, guest_gdtr_base),
804 FIELD(GUEST_IDTR_BASE, guest_idtr_base),
805 FIELD(GUEST_DR7, guest_dr7),
806 FIELD(GUEST_RSP, guest_rsp),
807 FIELD(GUEST_RIP, guest_rip),
808 FIELD(GUEST_RFLAGS, guest_rflags),
809 FIELD(GUEST_PENDING_DBG_EXCEPTIONS, guest_pending_dbg_exceptions),
810 FIELD(GUEST_SYSENTER_ESP, guest_sysenter_esp),
811 FIELD(GUEST_SYSENTER_EIP, guest_sysenter_eip),
812 FIELD(HOST_CR0, host_cr0),
813 FIELD(HOST_CR3, host_cr3),
814 FIELD(HOST_CR4, host_cr4),
815 FIELD(HOST_FS_BASE, host_fs_base),
816 FIELD(HOST_GS_BASE, host_gs_base),
817 FIELD(HOST_TR_BASE, host_tr_base),
818 FIELD(HOST_GDTR_BASE, host_gdtr_base),
819 FIELD(HOST_IDTR_BASE, host_idtr_base),
820 FIELD(HOST_IA32_SYSENTER_ESP, host_ia32_sysenter_esp),
821 FIELD(HOST_IA32_SYSENTER_EIP, host_ia32_sysenter_eip),
822 FIELD(HOST_RSP, host_rsp),
823 FIELD(HOST_RIP, host_rip),
826 static inline short vmcs_field_to_offset(unsigned long field)
828 BUILD_BUG_ON(ARRAY_SIZE(vmcs_field_to_offset_table) > SHRT_MAX);
830 if (field >= ARRAY_SIZE(vmcs_field_to_offset_table) ||
831 vmcs_field_to_offset_table[field] == 0)
834 return vmcs_field_to_offset_table[field];
837 static inline struct vmcs12 *get_vmcs12(struct kvm_vcpu *vcpu)
839 return to_vmx(vcpu)->nested.current_vmcs12;
842 static struct page *nested_get_page(struct kvm_vcpu *vcpu, gpa_t addr)
844 struct page *page = kvm_vcpu_gfn_to_page(vcpu, addr >> PAGE_SHIFT);
845 if (is_error_page(page))
851 static void nested_release_page(struct page *page)
853 kvm_release_page_dirty(page);
856 static void nested_release_page_clean(struct page *page)
858 kvm_release_page_clean(page);
861 static unsigned long nested_ept_get_cr3(struct kvm_vcpu *vcpu);
862 static u64 construct_eptp(unsigned long root_hpa);
863 static void kvm_cpu_vmxon(u64 addr);
864 static void kvm_cpu_vmxoff(void);
865 static bool vmx_mpx_supported(void);
866 static bool vmx_xsaves_supported(void);
867 static int vmx_cpu_uses_apicv(struct kvm_vcpu *vcpu);
868 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr);
869 static void vmx_set_segment(struct kvm_vcpu *vcpu,
870 struct kvm_segment *var, int seg);
871 static void vmx_get_segment(struct kvm_vcpu *vcpu,
872 struct kvm_segment *var, int seg);
873 static bool guest_state_valid(struct kvm_vcpu *vcpu);
874 static u32 vmx_segment_access_rights(struct kvm_segment *var);
875 static void vmx_sync_pir_to_irr_dummy(struct kvm_vcpu *vcpu);
876 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx);
877 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx);
878 static int alloc_identity_pagetable(struct kvm *kvm);
880 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
881 static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
883 * We maintain a per-CPU linked-list of VMCS loaded on that CPU. This is needed
884 * when a CPU is brought down, and we need to VMCLEAR all VMCSs loaded on it.
886 static DEFINE_PER_CPU(struct list_head, loaded_vmcss_on_cpu);
887 static DEFINE_PER_CPU(struct desc_ptr, host_gdt);
890 * We maintian a per-CPU linked-list of vCPU, so in wakeup_handler() we
891 * can find which vCPU should be waken up.
893 static DEFINE_PER_CPU(struct list_head, blocked_vcpu_on_cpu);
894 static DEFINE_PER_CPU(spinlock_t, blocked_vcpu_on_cpu_lock);
896 static unsigned long *vmx_io_bitmap_a;
897 static unsigned long *vmx_io_bitmap_b;
898 static unsigned long *vmx_msr_bitmap_legacy;
899 static unsigned long *vmx_msr_bitmap_longmode;
900 static unsigned long *vmx_msr_bitmap_legacy_x2apic;
901 static unsigned long *vmx_msr_bitmap_longmode_x2apic;
902 static unsigned long *vmx_msr_bitmap_nested;
903 static unsigned long *vmx_vmread_bitmap;
904 static unsigned long *vmx_vmwrite_bitmap;
906 static bool cpu_has_load_ia32_efer;
907 static bool cpu_has_load_perf_global_ctrl;
909 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
910 static DEFINE_SPINLOCK(vmx_vpid_lock);
912 static struct vmcs_config {
916 u32 pin_based_exec_ctrl;
917 u32 cpu_based_exec_ctrl;
918 u32 cpu_based_2nd_exec_ctrl;
923 static struct vmx_capability {
928 #define VMX_SEGMENT_FIELD(seg) \
929 [VCPU_SREG_##seg] = { \
930 .selector = GUEST_##seg##_SELECTOR, \
931 .base = GUEST_##seg##_BASE, \
932 .limit = GUEST_##seg##_LIMIT, \
933 .ar_bytes = GUEST_##seg##_AR_BYTES, \
936 static const struct kvm_vmx_segment_field {
941 } kvm_vmx_segment_fields[] = {
942 VMX_SEGMENT_FIELD(CS),
943 VMX_SEGMENT_FIELD(DS),
944 VMX_SEGMENT_FIELD(ES),
945 VMX_SEGMENT_FIELD(FS),
946 VMX_SEGMENT_FIELD(GS),
947 VMX_SEGMENT_FIELD(SS),
948 VMX_SEGMENT_FIELD(TR),
949 VMX_SEGMENT_FIELD(LDTR),
952 static u64 host_efer;
954 static void ept_save_pdptrs(struct kvm_vcpu *vcpu);
957 * Keep MSR_STAR at the end, as setup_msrs() will try to optimize it
958 * away by decrementing the array size.
960 static const u32 vmx_msr_index[] = {
962 MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
964 MSR_EFER, MSR_TSC_AUX, MSR_STAR,
967 static inline bool is_page_fault(u32 intr_info)
969 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
970 INTR_INFO_VALID_MASK)) ==
971 (INTR_TYPE_HARD_EXCEPTION | PF_VECTOR | INTR_INFO_VALID_MASK);
974 static inline bool is_no_device(u32 intr_info)
976 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
977 INTR_INFO_VALID_MASK)) ==
978 (INTR_TYPE_HARD_EXCEPTION | NM_VECTOR | INTR_INFO_VALID_MASK);
981 static inline bool is_invalid_opcode(u32 intr_info)
983 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
984 INTR_INFO_VALID_MASK)) ==
985 (INTR_TYPE_HARD_EXCEPTION | UD_VECTOR | INTR_INFO_VALID_MASK);
988 static inline bool is_external_interrupt(u32 intr_info)
990 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
991 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
994 static inline bool is_machine_check(u32 intr_info)
996 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
997 INTR_INFO_VALID_MASK)) ==
998 (INTR_TYPE_HARD_EXCEPTION | MC_VECTOR | INTR_INFO_VALID_MASK);
1001 static inline bool cpu_has_vmx_msr_bitmap(void)
1003 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS;
1006 static inline bool cpu_has_vmx_tpr_shadow(void)
1008 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW;
1011 static inline bool cpu_need_tpr_shadow(struct kvm_vcpu *vcpu)
1013 return cpu_has_vmx_tpr_shadow() && lapic_in_kernel(vcpu);
1016 static inline bool cpu_has_secondary_exec_ctrls(void)
1018 return vmcs_config.cpu_based_exec_ctrl &
1019 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
1022 static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
1024 return vmcs_config.cpu_based_2nd_exec_ctrl &
1025 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
1028 static inline bool cpu_has_vmx_virtualize_x2apic_mode(void)
1030 return vmcs_config.cpu_based_2nd_exec_ctrl &
1031 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
1034 static inline bool cpu_has_vmx_apic_register_virt(void)
1036 return vmcs_config.cpu_based_2nd_exec_ctrl &
1037 SECONDARY_EXEC_APIC_REGISTER_VIRT;
1040 static inline bool cpu_has_vmx_virtual_intr_delivery(void)
1042 return vmcs_config.cpu_based_2nd_exec_ctrl &
1043 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY;
1046 static inline bool cpu_has_vmx_posted_intr(void)
1048 return IS_ENABLED(CONFIG_X86_LOCAL_APIC) &&
1049 vmcs_config.pin_based_exec_ctrl & PIN_BASED_POSTED_INTR;
1052 static inline bool cpu_has_vmx_apicv(void)
1054 return cpu_has_vmx_apic_register_virt() &&
1055 cpu_has_vmx_virtual_intr_delivery() &&
1056 cpu_has_vmx_posted_intr();
1059 static inline bool cpu_has_vmx_flexpriority(void)
1061 return cpu_has_vmx_tpr_shadow() &&
1062 cpu_has_vmx_virtualize_apic_accesses();
1065 static inline bool cpu_has_vmx_ept_execute_only(void)
1067 return vmx_capability.ept & VMX_EPT_EXECUTE_ONLY_BIT;
1070 static inline bool cpu_has_vmx_ept_2m_page(void)
1072 return vmx_capability.ept & VMX_EPT_2MB_PAGE_BIT;
1075 static inline bool cpu_has_vmx_ept_1g_page(void)
1077 return vmx_capability.ept & VMX_EPT_1GB_PAGE_BIT;
1080 static inline bool cpu_has_vmx_ept_4levels(void)
1082 return vmx_capability.ept & VMX_EPT_PAGE_WALK_4_BIT;
1085 static inline bool cpu_has_vmx_ept_ad_bits(void)
1087 return vmx_capability.ept & VMX_EPT_AD_BIT;
1090 static inline bool cpu_has_vmx_invept_context(void)
1092 return vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT;
1095 static inline bool cpu_has_vmx_invept_global(void)
1097 return vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT;
1100 static inline bool cpu_has_vmx_invvpid_single(void)
1102 return vmx_capability.vpid & VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT;
1105 static inline bool cpu_has_vmx_invvpid_global(void)
1107 return vmx_capability.vpid & VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT;
1110 static inline bool cpu_has_vmx_ept(void)
1112 return vmcs_config.cpu_based_2nd_exec_ctrl &
1113 SECONDARY_EXEC_ENABLE_EPT;
1116 static inline bool cpu_has_vmx_unrestricted_guest(void)
1118 return vmcs_config.cpu_based_2nd_exec_ctrl &
1119 SECONDARY_EXEC_UNRESTRICTED_GUEST;
1122 static inline bool cpu_has_vmx_ple(void)
1124 return vmcs_config.cpu_based_2nd_exec_ctrl &
1125 SECONDARY_EXEC_PAUSE_LOOP_EXITING;
1128 static inline bool cpu_need_virtualize_apic_accesses(struct kvm_vcpu *vcpu)
1130 return flexpriority_enabled && lapic_in_kernel(vcpu);
1133 static inline bool cpu_has_vmx_vpid(void)
1135 return vmcs_config.cpu_based_2nd_exec_ctrl &
1136 SECONDARY_EXEC_ENABLE_VPID;
1139 static inline bool cpu_has_vmx_rdtscp(void)
1141 return vmcs_config.cpu_based_2nd_exec_ctrl &
1142 SECONDARY_EXEC_RDTSCP;
1145 static inline bool cpu_has_vmx_invpcid(void)
1147 return vmcs_config.cpu_based_2nd_exec_ctrl &
1148 SECONDARY_EXEC_ENABLE_INVPCID;
1151 static inline bool cpu_has_virtual_nmis(void)
1153 return vmcs_config.pin_based_exec_ctrl & PIN_BASED_VIRTUAL_NMIS;
1156 static inline bool cpu_has_vmx_wbinvd_exit(void)
1158 return vmcs_config.cpu_based_2nd_exec_ctrl &
1159 SECONDARY_EXEC_WBINVD_EXITING;
1162 static inline bool cpu_has_vmx_shadow_vmcs(void)
1165 rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
1166 /* check if the cpu supports writing r/o exit information fields */
1167 if (!(vmx_msr & MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS))
1170 return vmcs_config.cpu_based_2nd_exec_ctrl &
1171 SECONDARY_EXEC_SHADOW_VMCS;
1174 static inline bool cpu_has_vmx_pml(void)
1176 return vmcs_config.cpu_based_2nd_exec_ctrl & SECONDARY_EXEC_ENABLE_PML;
1179 static inline bool cpu_has_vmx_tsc_scaling(void)
1181 return vmcs_config.cpu_based_2nd_exec_ctrl &
1182 SECONDARY_EXEC_TSC_SCALING;
1185 static inline bool report_flexpriority(void)
1187 return flexpriority_enabled;
1190 static inline bool nested_cpu_has(struct vmcs12 *vmcs12, u32 bit)
1192 return vmcs12->cpu_based_vm_exec_control & bit;
1195 static inline bool nested_cpu_has2(struct vmcs12 *vmcs12, u32 bit)
1197 return (vmcs12->cpu_based_vm_exec_control &
1198 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
1199 (vmcs12->secondary_vm_exec_control & bit);
1202 static inline bool nested_cpu_has_virtual_nmis(struct vmcs12 *vmcs12)
1204 return vmcs12->pin_based_vm_exec_control & PIN_BASED_VIRTUAL_NMIS;
1207 static inline bool nested_cpu_has_preemption_timer(struct vmcs12 *vmcs12)
1209 return vmcs12->pin_based_vm_exec_control &
1210 PIN_BASED_VMX_PREEMPTION_TIMER;
1213 static inline int nested_cpu_has_ept(struct vmcs12 *vmcs12)
1215 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_EPT);
1218 static inline bool nested_cpu_has_xsaves(struct vmcs12 *vmcs12)
1220 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES) &&
1221 vmx_xsaves_supported();
1224 static inline bool nested_cpu_has_virt_x2apic_mode(struct vmcs12 *vmcs12)
1226 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE);
1229 static inline bool nested_cpu_has_vpid(struct vmcs12 *vmcs12)
1231 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_VPID);
1234 static inline bool nested_cpu_has_apic_reg_virt(struct vmcs12 *vmcs12)
1236 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_APIC_REGISTER_VIRT);
1239 static inline bool nested_cpu_has_vid(struct vmcs12 *vmcs12)
1241 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
1244 static inline bool nested_cpu_has_posted_intr(struct vmcs12 *vmcs12)
1246 return vmcs12->pin_based_vm_exec_control & PIN_BASED_POSTED_INTR;
1249 static inline bool is_exception(u32 intr_info)
1251 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
1252 == (INTR_TYPE_HARD_EXCEPTION | INTR_INFO_VALID_MASK);
1255 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
1257 unsigned long exit_qualification);
1258 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
1259 struct vmcs12 *vmcs12,
1260 u32 reason, unsigned long qualification);
1262 static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
1266 for (i = 0; i < vmx->nmsrs; ++i)
1267 if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
1272 static inline void __invvpid(int ext, u16 vpid, gva_t gva)
1278 } operand = { vpid, 0, gva };
1280 asm volatile (__ex(ASM_VMX_INVVPID)
1281 /* CF==1 or ZF==1 --> rc = -1 */
1282 "; ja 1f ; ud2 ; 1:"
1283 : : "a"(&operand), "c"(ext) : "cc", "memory");
1286 static inline void __invept(int ext, u64 eptp, gpa_t gpa)
1290 } operand = {eptp, gpa};
1292 asm volatile (__ex(ASM_VMX_INVEPT)
1293 /* CF==1 or ZF==1 --> rc = -1 */
1294 "; ja 1f ; ud2 ; 1:\n"
1295 : : "a" (&operand), "c" (ext) : "cc", "memory");
1298 static struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
1302 i = __find_msr_index(vmx, msr);
1304 return &vmx->guest_msrs[i];
1308 static void vmcs_clear(struct vmcs *vmcs)
1310 u64 phys_addr = __pa(vmcs);
1313 asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0"
1314 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
1317 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
1321 static inline void loaded_vmcs_init(struct loaded_vmcs *loaded_vmcs)
1323 vmcs_clear(loaded_vmcs->vmcs);
1324 loaded_vmcs->cpu = -1;
1325 loaded_vmcs->launched = 0;
1328 static void vmcs_load(struct vmcs *vmcs)
1330 u64 phys_addr = __pa(vmcs);
1333 asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0"
1334 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
1337 printk(KERN_ERR "kvm: vmptrld %p/%llx failed\n",
1341 #ifdef CONFIG_KEXEC_CORE
1343 * This bitmap is used to indicate whether the vmclear
1344 * operation is enabled on all cpus. All disabled by
1347 static cpumask_t crash_vmclear_enabled_bitmap = CPU_MASK_NONE;
1349 static inline void crash_enable_local_vmclear(int cpu)
1351 cpumask_set_cpu(cpu, &crash_vmclear_enabled_bitmap);
1354 static inline void crash_disable_local_vmclear(int cpu)
1356 cpumask_clear_cpu(cpu, &crash_vmclear_enabled_bitmap);
1359 static inline int crash_local_vmclear_enabled(int cpu)
1361 return cpumask_test_cpu(cpu, &crash_vmclear_enabled_bitmap);
1364 static void crash_vmclear_local_loaded_vmcss(void)
1366 int cpu = raw_smp_processor_id();
1367 struct loaded_vmcs *v;
1369 if (!crash_local_vmclear_enabled(cpu))
1372 list_for_each_entry(v, &per_cpu(loaded_vmcss_on_cpu, cpu),
1373 loaded_vmcss_on_cpu_link)
1374 vmcs_clear(v->vmcs);
1377 static inline void crash_enable_local_vmclear(int cpu) { }
1378 static inline void crash_disable_local_vmclear(int cpu) { }
1379 #endif /* CONFIG_KEXEC_CORE */
1381 static void __loaded_vmcs_clear(void *arg)
1383 struct loaded_vmcs *loaded_vmcs = arg;
1384 int cpu = raw_smp_processor_id();
1386 if (loaded_vmcs->cpu != cpu)
1387 return; /* vcpu migration can race with cpu offline */
1388 if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
1389 per_cpu(current_vmcs, cpu) = NULL;
1390 crash_disable_local_vmclear(cpu);
1391 list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
1394 * we should ensure updating loaded_vmcs->loaded_vmcss_on_cpu_link
1395 * is before setting loaded_vmcs->vcpu to -1 which is done in
1396 * loaded_vmcs_init. Otherwise, other cpu can see vcpu = -1 fist
1397 * then adds the vmcs into percpu list before it is deleted.
1401 loaded_vmcs_init(loaded_vmcs);
1402 crash_enable_local_vmclear(cpu);
1405 static void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
1407 int cpu = loaded_vmcs->cpu;
1410 smp_call_function_single(cpu,
1411 __loaded_vmcs_clear, loaded_vmcs, 1);
1414 static inline void vpid_sync_vcpu_single(int vpid)
1419 if (cpu_has_vmx_invvpid_single())
1420 __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vpid, 0);
1423 static inline void vpid_sync_vcpu_global(void)
1425 if (cpu_has_vmx_invvpid_global())
1426 __invvpid(VMX_VPID_EXTENT_ALL_CONTEXT, 0, 0);
1429 static inline void vpid_sync_context(int vpid)
1431 if (cpu_has_vmx_invvpid_single())
1432 vpid_sync_vcpu_single(vpid);
1434 vpid_sync_vcpu_global();
1437 static inline void ept_sync_global(void)
1439 if (cpu_has_vmx_invept_global())
1440 __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0);
1443 static inline void ept_sync_context(u64 eptp)
1446 if (cpu_has_vmx_invept_context())
1447 __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0);
1453 static __always_inline unsigned long vmcs_readl(unsigned long field)
1455 unsigned long value;
1457 asm volatile (__ex_clear(ASM_VMX_VMREAD_RDX_RAX, "%0")
1458 : "=a"(value) : "d"(field) : "cc");
1462 static __always_inline u16 vmcs_read16(unsigned long field)
1464 return vmcs_readl(field);
1467 static __always_inline u32 vmcs_read32(unsigned long field)
1469 return vmcs_readl(field);
1472 static __always_inline u64 vmcs_read64(unsigned long field)
1474 #ifdef CONFIG_X86_64
1475 return vmcs_readl(field);
1477 return vmcs_readl(field) | ((u64)vmcs_readl(field+1) << 32);
1481 static noinline void vmwrite_error(unsigned long field, unsigned long value)
1483 printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
1484 field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
1488 static void vmcs_writel(unsigned long field, unsigned long value)
1492 asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0"
1493 : "=q"(error) : "a"(value), "d"(field) : "cc");
1494 if (unlikely(error))
1495 vmwrite_error(field, value);
1498 static void vmcs_write16(unsigned long field, u16 value)
1500 vmcs_writel(field, value);
1503 static void vmcs_write32(unsigned long field, u32 value)
1505 vmcs_writel(field, value);
1508 static void vmcs_write64(unsigned long field, u64 value)
1510 vmcs_writel(field, value);
1511 #ifndef CONFIG_X86_64
1513 vmcs_writel(field+1, value >> 32);
1517 static void vmcs_clear_bits(unsigned long field, u32 mask)
1519 vmcs_writel(field, vmcs_readl(field) & ~mask);
1522 static void vmcs_set_bits(unsigned long field, u32 mask)
1524 vmcs_writel(field, vmcs_readl(field) | mask);
1527 static inline void vm_entry_controls_init(struct vcpu_vmx *vmx, u32 val)
1529 vmcs_write32(VM_ENTRY_CONTROLS, val);
1530 vmx->vm_entry_controls_shadow = val;
1533 static inline void vm_entry_controls_set(struct vcpu_vmx *vmx, u32 val)
1535 if (vmx->vm_entry_controls_shadow != val)
1536 vm_entry_controls_init(vmx, val);
1539 static inline u32 vm_entry_controls_get(struct vcpu_vmx *vmx)
1541 return vmx->vm_entry_controls_shadow;
1545 static inline void vm_entry_controls_setbit(struct vcpu_vmx *vmx, u32 val)
1547 vm_entry_controls_set(vmx, vm_entry_controls_get(vmx) | val);
1550 static inline void vm_entry_controls_clearbit(struct vcpu_vmx *vmx, u32 val)
1552 vm_entry_controls_set(vmx, vm_entry_controls_get(vmx) & ~val);
1555 static inline void vm_exit_controls_init(struct vcpu_vmx *vmx, u32 val)
1557 vmcs_write32(VM_EXIT_CONTROLS, val);
1558 vmx->vm_exit_controls_shadow = val;
1561 static inline void vm_exit_controls_set(struct vcpu_vmx *vmx, u32 val)
1563 if (vmx->vm_exit_controls_shadow != val)
1564 vm_exit_controls_init(vmx, val);
1567 static inline u32 vm_exit_controls_get(struct vcpu_vmx *vmx)
1569 return vmx->vm_exit_controls_shadow;
1573 static inline void vm_exit_controls_setbit(struct vcpu_vmx *vmx, u32 val)
1575 vm_exit_controls_set(vmx, vm_exit_controls_get(vmx) | val);
1578 static inline void vm_exit_controls_clearbit(struct vcpu_vmx *vmx, u32 val)
1580 vm_exit_controls_set(vmx, vm_exit_controls_get(vmx) & ~val);
1583 static void vmx_segment_cache_clear(struct vcpu_vmx *vmx)
1585 vmx->segment_cache.bitmask = 0;
1588 static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
1592 u32 mask = 1 << (seg * SEG_FIELD_NR + field);
1594 if (!(vmx->vcpu.arch.regs_avail & (1 << VCPU_EXREG_SEGMENTS))) {
1595 vmx->vcpu.arch.regs_avail |= (1 << VCPU_EXREG_SEGMENTS);
1596 vmx->segment_cache.bitmask = 0;
1598 ret = vmx->segment_cache.bitmask & mask;
1599 vmx->segment_cache.bitmask |= mask;
1603 static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
1605 u16 *p = &vmx->segment_cache.seg[seg].selector;
1607 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
1608 *p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
1612 static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
1614 ulong *p = &vmx->segment_cache.seg[seg].base;
1616 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
1617 *p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
1621 static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
1623 u32 *p = &vmx->segment_cache.seg[seg].limit;
1625 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
1626 *p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
1630 static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
1632 u32 *p = &vmx->segment_cache.seg[seg].ar;
1634 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
1635 *p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
1639 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
1643 eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
1644 (1u << NM_VECTOR) | (1u << DB_VECTOR) | (1u << AC_VECTOR);
1645 if ((vcpu->guest_debug &
1646 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
1647 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
1648 eb |= 1u << BP_VECTOR;
1649 if (to_vmx(vcpu)->rmode.vm86_active)
1652 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
1653 if (vcpu->fpu_active)
1654 eb &= ~(1u << NM_VECTOR);
1656 /* When we are running a nested L2 guest and L1 specified for it a
1657 * certain exception bitmap, we must trap the same exceptions and pass
1658 * them to L1. When running L2, we will only handle the exceptions
1659 * specified above if L1 did not want them.
1661 if (is_guest_mode(vcpu))
1662 eb |= get_vmcs12(vcpu)->exception_bitmap;
1664 vmcs_write32(EXCEPTION_BITMAP, eb);
1667 static void clear_atomic_switch_msr_special(struct vcpu_vmx *vmx,
1668 unsigned long entry, unsigned long exit)
1670 vm_entry_controls_clearbit(vmx, entry);
1671 vm_exit_controls_clearbit(vmx, exit);
1674 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
1677 struct msr_autoload *m = &vmx->msr_autoload;
1681 if (cpu_has_load_ia32_efer) {
1682 clear_atomic_switch_msr_special(vmx,
1683 VM_ENTRY_LOAD_IA32_EFER,
1684 VM_EXIT_LOAD_IA32_EFER);
1688 case MSR_CORE_PERF_GLOBAL_CTRL:
1689 if (cpu_has_load_perf_global_ctrl) {
1690 clear_atomic_switch_msr_special(vmx,
1691 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1692 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
1698 for (i = 0; i < m->nr; ++i)
1699 if (m->guest[i].index == msr)
1705 m->guest[i] = m->guest[m->nr];
1706 m->host[i] = m->host[m->nr];
1707 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1708 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1711 static void add_atomic_switch_msr_special(struct vcpu_vmx *vmx,
1712 unsigned long entry, unsigned long exit,
1713 unsigned long guest_val_vmcs, unsigned long host_val_vmcs,
1714 u64 guest_val, u64 host_val)
1716 vmcs_write64(guest_val_vmcs, guest_val);
1717 vmcs_write64(host_val_vmcs, host_val);
1718 vm_entry_controls_setbit(vmx, entry);
1719 vm_exit_controls_setbit(vmx, exit);
1722 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
1723 u64 guest_val, u64 host_val)
1726 struct msr_autoload *m = &vmx->msr_autoload;
1730 if (cpu_has_load_ia32_efer) {
1731 add_atomic_switch_msr_special(vmx,
1732 VM_ENTRY_LOAD_IA32_EFER,
1733 VM_EXIT_LOAD_IA32_EFER,
1736 guest_val, host_val);
1740 case MSR_CORE_PERF_GLOBAL_CTRL:
1741 if (cpu_has_load_perf_global_ctrl) {
1742 add_atomic_switch_msr_special(vmx,
1743 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1744 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL,
1745 GUEST_IA32_PERF_GLOBAL_CTRL,
1746 HOST_IA32_PERF_GLOBAL_CTRL,
1747 guest_val, host_val);
1753 for (i = 0; i < m->nr; ++i)
1754 if (m->guest[i].index == msr)
1757 if (i == NR_AUTOLOAD_MSRS) {
1758 printk_once(KERN_WARNING "Not enough msr switch entries. "
1759 "Can't add msr %x\n", msr);
1761 } else if (i == m->nr) {
1763 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1764 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1767 m->guest[i].index = msr;
1768 m->guest[i].value = guest_val;
1769 m->host[i].index = msr;
1770 m->host[i].value = host_val;
1773 static void reload_tss(void)
1776 * VT restores TR but not its size. Useless.
1778 struct desc_ptr *gdt = this_cpu_ptr(&host_gdt);
1779 struct desc_struct *descs;
1781 descs = (void *)gdt->address;
1782 descs[GDT_ENTRY_TSS].type = 9; /* available TSS */
1786 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
1791 guest_efer = vmx->vcpu.arch.efer;
1794 * NX is emulated; LMA and LME handled by hardware; SCE meaningless
1797 ignore_bits = EFER_NX | EFER_SCE;
1798 #ifdef CONFIG_X86_64
1799 ignore_bits |= EFER_LMA | EFER_LME;
1800 /* SCE is meaningful only in long mode on Intel */
1801 if (guest_efer & EFER_LMA)
1802 ignore_bits &= ~(u64)EFER_SCE;
1804 guest_efer &= ~ignore_bits;
1805 guest_efer |= host_efer & ignore_bits;
1806 vmx->guest_msrs[efer_offset].data = guest_efer;
1807 vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
1809 clear_atomic_switch_msr(vmx, MSR_EFER);
1812 * On EPT, we can't emulate NX, so we must switch EFER atomically.
1813 * On CPUs that support "load IA32_EFER", always switch EFER
1814 * atomically, since it's faster than switching it manually.
1816 if (cpu_has_load_ia32_efer ||
1817 (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX))) {
1818 guest_efer = vmx->vcpu.arch.efer;
1819 if (!(guest_efer & EFER_LMA))
1820 guest_efer &= ~EFER_LME;
1821 if (guest_efer != host_efer)
1822 add_atomic_switch_msr(vmx, MSR_EFER,
1823 guest_efer, host_efer);
1830 static unsigned long segment_base(u16 selector)
1832 struct desc_ptr *gdt = this_cpu_ptr(&host_gdt);
1833 struct desc_struct *d;
1834 unsigned long table_base;
1837 if (!(selector & ~3))
1840 table_base = gdt->address;
1842 if (selector & 4) { /* from ldt */
1843 u16 ldt_selector = kvm_read_ldt();
1845 if (!(ldt_selector & ~3))
1848 table_base = segment_base(ldt_selector);
1850 d = (struct desc_struct *)(table_base + (selector & ~7));
1851 v = get_desc_base(d);
1852 #ifdef CONFIG_X86_64
1853 if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
1854 v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
1859 static inline unsigned long kvm_read_tr_base(void)
1862 asm("str %0" : "=g"(tr));
1863 return segment_base(tr);
1866 static void vmx_save_host_state(struct kvm_vcpu *vcpu)
1868 struct vcpu_vmx *vmx = to_vmx(vcpu);
1871 if (vmx->host_state.loaded)
1874 vmx->host_state.loaded = 1;
1876 * Set host fs and gs selectors. Unfortunately, 22.2.3 does not
1877 * allow segment selectors with cpl > 0 or ti == 1.
1879 vmx->host_state.ldt_sel = kvm_read_ldt();
1880 vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
1881 savesegment(fs, vmx->host_state.fs_sel);
1882 if (!(vmx->host_state.fs_sel & 7)) {
1883 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
1884 vmx->host_state.fs_reload_needed = 0;
1886 vmcs_write16(HOST_FS_SELECTOR, 0);
1887 vmx->host_state.fs_reload_needed = 1;
1889 savesegment(gs, vmx->host_state.gs_sel);
1890 if (!(vmx->host_state.gs_sel & 7))
1891 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
1893 vmcs_write16(HOST_GS_SELECTOR, 0);
1894 vmx->host_state.gs_ldt_reload_needed = 1;
1897 #ifdef CONFIG_X86_64
1898 savesegment(ds, vmx->host_state.ds_sel);
1899 savesegment(es, vmx->host_state.es_sel);
1902 #ifdef CONFIG_X86_64
1903 vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
1904 vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
1906 vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
1907 vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
1910 #ifdef CONFIG_X86_64
1911 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1912 if (is_long_mode(&vmx->vcpu))
1913 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1915 if (boot_cpu_has(X86_FEATURE_MPX))
1916 rdmsrl(MSR_IA32_BNDCFGS, vmx->host_state.msr_host_bndcfgs);
1917 for (i = 0; i < vmx->save_nmsrs; ++i)
1918 kvm_set_shared_msr(vmx->guest_msrs[i].index,
1919 vmx->guest_msrs[i].data,
1920 vmx->guest_msrs[i].mask);
1923 static void __vmx_load_host_state(struct vcpu_vmx *vmx)
1925 if (!vmx->host_state.loaded)
1928 ++vmx->vcpu.stat.host_state_reload;
1929 vmx->host_state.loaded = 0;
1930 #ifdef CONFIG_X86_64
1931 if (is_long_mode(&vmx->vcpu))
1932 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1934 if (vmx->host_state.gs_ldt_reload_needed) {
1935 kvm_load_ldt(vmx->host_state.ldt_sel);
1936 #ifdef CONFIG_X86_64
1937 load_gs_index(vmx->host_state.gs_sel);
1939 loadsegment(gs, vmx->host_state.gs_sel);
1942 if (vmx->host_state.fs_reload_needed)
1943 loadsegment(fs, vmx->host_state.fs_sel);
1944 #ifdef CONFIG_X86_64
1945 if (unlikely(vmx->host_state.ds_sel | vmx->host_state.es_sel)) {
1946 loadsegment(ds, vmx->host_state.ds_sel);
1947 loadsegment(es, vmx->host_state.es_sel);
1951 #ifdef CONFIG_X86_64
1952 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1954 if (vmx->host_state.msr_host_bndcfgs)
1955 wrmsrl(MSR_IA32_BNDCFGS, vmx->host_state.msr_host_bndcfgs);
1957 * If the FPU is not active (through the host task or
1958 * the guest vcpu), then restore the cr0.TS bit.
1960 if (!fpregs_active() && !vmx->vcpu.guest_fpu_loaded)
1962 load_gdt(this_cpu_ptr(&host_gdt));
1965 static void vmx_load_host_state(struct vcpu_vmx *vmx)
1968 __vmx_load_host_state(vmx);
1972 static void vmx_vcpu_pi_load(struct kvm_vcpu *vcpu, int cpu)
1974 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
1975 struct pi_desc old, new;
1978 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
1979 !irq_remapping_cap(IRQ_POSTING_CAP))
1983 old.control = new.control = pi_desc->control;
1986 * If 'nv' field is POSTED_INTR_WAKEUP_VECTOR, there
1987 * are two possible cases:
1988 * 1. After running 'pre_block', context switch
1989 * happened. For this case, 'sn' was set in
1990 * vmx_vcpu_put(), so we need to clear it here.
1991 * 2. After running 'pre_block', we were blocked,
1992 * and woken up by some other guy. For this case,
1993 * we don't need to do anything, 'pi_post_block'
1994 * will do everything for us. However, we cannot
1995 * check whether it is case #1 or case #2 here
1996 * (maybe, not needed), so we also clear sn here,
1997 * I think it is not a big deal.
1999 if (pi_desc->nv != POSTED_INTR_WAKEUP_VECTOR) {
2000 if (vcpu->cpu != cpu) {
2001 dest = cpu_physical_id(cpu);
2003 if (x2apic_enabled())
2006 new.ndst = (dest << 8) & 0xFF00;
2009 /* set 'NV' to 'notification vector' */
2010 new.nv = POSTED_INTR_VECTOR;
2013 /* Allow posting non-urgent interrupts */
2015 } while (cmpxchg(&pi_desc->control, old.control,
2016 new.control) != old.control);
2019 * Switches to specified vcpu, until a matching vcpu_put(), but assumes
2020 * vcpu mutex is already taken.
2022 static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2024 struct vcpu_vmx *vmx = to_vmx(vcpu);
2025 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
2028 kvm_cpu_vmxon(phys_addr);
2029 else if (vmx->loaded_vmcs->cpu != cpu)
2030 loaded_vmcs_clear(vmx->loaded_vmcs);
2032 if (per_cpu(current_vmcs, cpu) != vmx->loaded_vmcs->vmcs) {
2033 per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
2034 vmcs_load(vmx->loaded_vmcs->vmcs);
2037 if (vmx->loaded_vmcs->cpu != cpu) {
2038 struct desc_ptr *gdt = this_cpu_ptr(&host_gdt);
2039 unsigned long sysenter_esp;
2041 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
2042 local_irq_disable();
2043 crash_disable_local_vmclear(cpu);
2046 * Read loaded_vmcs->cpu should be before fetching
2047 * loaded_vmcs->loaded_vmcss_on_cpu_link.
2048 * See the comments in __loaded_vmcs_clear().
2052 list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
2053 &per_cpu(loaded_vmcss_on_cpu, cpu));
2054 crash_enable_local_vmclear(cpu);
2058 * Linux uses per-cpu TSS and GDT, so set these when switching
2061 vmcs_writel(HOST_TR_BASE, kvm_read_tr_base()); /* 22.2.4 */
2062 vmcs_writel(HOST_GDTR_BASE, gdt->address); /* 22.2.4 */
2064 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
2065 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
2067 vmx->loaded_vmcs->cpu = cpu;
2070 /* Setup TSC multiplier */
2071 if (kvm_has_tsc_control &&
2072 vmx->current_tsc_ratio != vcpu->arch.tsc_scaling_ratio) {
2073 vmx->current_tsc_ratio = vcpu->arch.tsc_scaling_ratio;
2074 vmcs_write64(TSC_MULTIPLIER, vmx->current_tsc_ratio);
2077 vmx_vcpu_pi_load(vcpu, cpu);
2080 static void vmx_vcpu_pi_put(struct kvm_vcpu *vcpu)
2082 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
2084 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
2085 !irq_remapping_cap(IRQ_POSTING_CAP))
2088 /* Set SN when the vCPU is preempted */
2089 if (vcpu->preempted)
2093 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
2095 vmx_vcpu_pi_put(vcpu);
2097 __vmx_load_host_state(to_vmx(vcpu));
2098 if (!vmm_exclusive) {
2099 __loaded_vmcs_clear(to_vmx(vcpu)->loaded_vmcs);
2105 static void vmx_fpu_activate(struct kvm_vcpu *vcpu)
2109 if (vcpu->fpu_active)
2111 vcpu->fpu_active = 1;
2112 cr0 = vmcs_readl(GUEST_CR0);
2113 cr0 &= ~(X86_CR0_TS | X86_CR0_MP);
2114 cr0 |= kvm_read_cr0_bits(vcpu, X86_CR0_TS | X86_CR0_MP);
2115 vmcs_writel(GUEST_CR0, cr0);
2116 update_exception_bitmap(vcpu);
2117 vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
2118 if (is_guest_mode(vcpu))
2119 vcpu->arch.cr0_guest_owned_bits &=
2120 ~get_vmcs12(vcpu)->cr0_guest_host_mask;
2121 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
2124 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu);
2127 * Return the cr0 value that a nested guest would read. This is a combination
2128 * of the real cr0 used to run the guest (guest_cr0), and the bits shadowed by
2129 * its hypervisor (cr0_read_shadow).
2131 static inline unsigned long nested_read_cr0(struct vmcs12 *fields)
2133 return (fields->guest_cr0 & ~fields->cr0_guest_host_mask) |
2134 (fields->cr0_read_shadow & fields->cr0_guest_host_mask);
2136 static inline unsigned long nested_read_cr4(struct vmcs12 *fields)
2138 return (fields->guest_cr4 & ~fields->cr4_guest_host_mask) |
2139 (fields->cr4_read_shadow & fields->cr4_guest_host_mask);
2142 static void vmx_fpu_deactivate(struct kvm_vcpu *vcpu)
2144 /* Note that there is no vcpu->fpu_active = 0 here. The caller must
2145 * set this *before* calling this function.
2147 vmx_decache_cr0_guest_bits(vcpu);
2148 vmcs_set_bits(GUEST_CR0, X86_CR0_TS | X86_CR0_MP);
2149 update_exception_bitmap(vcpu);
2150 vcpu->arch.cr0_guest_owned_bits = 0;
2151 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
2152 if (is_guest_mode(vcpu)) {
2154 * L1's specified read shadow might not contain the TS bit,
2155 * so now that we turned on shadowing of this bit, we need to
2156 * set this bit of the shadow. Like in nested_vmx_run we need
2157 * nested_read_cr0(vmcs12), but vmcs12->guest_cr0 is not yet
2158 * up-to-date here because we just decached cr0.TS (and we'll
2159 * only update vmcs12->guest_cr0 on nested exit).
2161 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2162 vmcs12->guest_cr0 = (vmcs12->guest_cr0 & ~X86_CR0_TS) |
2163 (vcpu->arch.cr0 & X86_CR0_TS);
2164 vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
2166 vmcs_writel(CR0_READ_SHADOW, vcpu->arch.cr0);
2169 static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
2171 unsigned long rflags, save_rflags;
2173 if (!test_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail)) {
2174 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
2175 rflags = vmcs_readl(GUEST_RFLAGS);
2176 if (to_vmx(vcpu)->rmode.vm86_active) {
2177 rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
2178 save_rflags = to_vmx(vcpu)->rmode.save_rflags;
2179 rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
2181 to_vmx(vcpu)->rflags = rflags;
2183 return to_vmx(vcpu)->rflags;
2186 static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
2188 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
2189 to_vmx(vcpu)->rflags = rflags;
2190 if (to_vmx(vcpu)->rmode.vm86_active) {
2191 to_vmx(vcpu)->rmode.save_rflags = rflags;
2192 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
2194 vmcs_writel(GUEST_RFLAGS, rflags);
2197 static u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu)
2199 u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
2202 if (interruptibility & GUEST_INTR_STATE_STI)
2203 ret |= KVM_X86_SHADOW_INT_STI;
2204 if (interruptibility & GUEST_INTR_STATE_MOV_SS)
2205 ret |= KVM_X86_SHADOW_INT_MOV_SS;
2210 static void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
2212 u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
2213 u32 interruptibility = interruptibility_old;
2215 interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
2217 if (mask & KVM_X86_SHADOW_INT_MOV_SS)
2218 interruptibility |= GUEST_INTR_STATE_MOV_SS;
2219 else if (mask & KVM_X86_SHADOW_INT_STI)
2220 interruptibility |= GUEST_INTR_STATE_STI;
2222 if ((interruptibility != interruptibility_old))
2223 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
2226 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
2230 rip = kvm_rip_read(vcpu);
2231 rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
2232 kvm_rip_write(vcpu, rip);
2234 /* skipping an emulated instruction also counts */
2235 vmx_set_interrupt_shadow(vcpu, 0);
2239 * KVM wants to inject page-faults which it got to the guest. This function
2240 * checks whether in a nested guest, we need to inject them to L1 or L2.
2242 static int nested_vmx_check_exception(struct kvm_vcpu *vcpu, unsigned nr)
2244 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2246 if (!(vmcs12->exception_bitmap & (1u << nr)))
2249 nested_vmx_vmexit(vcpu, to_vmx(vcpu)->exit_reason,
2250 vmcs_read32(VM_EXIT_INTR_INFO),
2251 vmcs_readl(EXIT_QUALIFICATION));
2255 static void vmx_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
2256 bool has_error_code, u32 error_code,
2259 struct vcpu_vmx *vmx = to_vmx(vcpu);
2260 u32 intr_info = nr | INTR_INFO_VALID_MASK;
2262 if (!reinject && is_guest_mode(vcpu) &&
2263 nested_vmx_check_exception(vcpu, nr))
2266 if (has_error_code) {
2267 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
2268 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
2271 if (vmx->rmode.vm86_active) {
2273 if (kvm_exception_is_soft(nr))
2274 inc_eip = vcpu->arch.event_exit_inst_len;
2275 if (kvm_inject_realmode_interrupt(vcpu, nr, inc_eip) != EMULATE_DONE)
2276 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2280 if (kvm_exception_is_soft(nr)) {
2281 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2282 vmx->vcpu.arch.event_exit_inst_len);
2283 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
2285 intr_info |= INTR_TYPE_HARD_EXCEPTION;
2287 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
2290 static bool vmx_rdtscp_supported(void)
2292 return cpu_has_vmx_rdtscp();
2295 static bool vmx_invpcid_supported(void)
2297 return cpu_has_vmx_invpcid() && enable_ept;
2301 * Swap MSR entry in host/guest MSR entry array.
2303 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
2305 struct shared_msr_entry tmp;
2307 tmp = vmx->guest_msrs[to];
2308 vmx->guest_msrs[to] = vmx->guest_msrs[from];
2309 vmx->guest_msrs[from] = tmp;
2312 static void vmx_set_msr_bitmap(struct kvm_vcpu *vcpu)
2314 unsigned long *msr_bitmap;
2316 if (is_guest_mode(vcpu))
2317 msr_bitmap = vmx_msr_bitmap_nested;
2318 else if (vcpu->arch.apic_base & X2APIC_ENABLE) {
2319 if (is_long_mode(vcpu))
2320 msr_bitmap = vmx_msr_bitmap_longmode_x2apic;
2322 msr_bitmap = vmx_msr_bitmap_legacy_x2apic;
2324 if (is_long_mode(vcpu))
2325 msr_bitmap = vmx_msr_bitmap_longmode;
2327 msr_bitmap = vmx_msr_bitmap_legacy;
2330 vmcs_write64(MSR_BITMAP, __pa(msr_bitmap));
2334 * Set up the vmcs to automatically save and restore system
2335 * msrs. Don't touch the 64-bit msrs if the guest is in legacy
2336 * mode, as fiddling with msrs is very expensive.
2338 static void setup_msrs(struct vcpu_vmx *vmx)
2340 int save_nmsrs, index;
2343 #ifdef CONFIG_X86_64
2344 if (is_long_mode(&vmx->vcpu)) {
2345 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
2347 move_msr_up(vmx, index, save_nmsrs++);
2348 index = __find_msr_index(vmx, MSR_LSTAR);
2350 move_msr_up(vmx, index, save_nmsrs++);
2351 index = __find_msr_index(vmx, MSR_CSTAR);
2353 move_msr_up(vmx, index, save_nmsrs++);
2354 index = __find_msr_index(vmx, MSR_TSC_AUX);
2355 if (index >= 0 && guest_cpuid_has_rdtscp(&vmx->vcpu))
2356 move_msr_up(vmx, index, save_nmsrs++);
2358 * MSR_STAR is only needed on long mode guests, and only
2359 * if efer.sce is enabled.
2361 index = __find_msr_index(vmx, MSR_STAR);
2362 if ((index >= 0) && (vmx->vcpu.arch.efer & EFER_SCE))
2363 move_msr_up(vmx, index, save_nmsrs++);
2366 index = __find_msr_index(vmx, MSR_EFER);
2367 if (index >= 0 && update_transition_efer(vmx, index))
2368 move_msr_up(vmx, index, save_nmsrs++);
2370 vmx->save_nmsrs = save_nmsrs;
2372 if (cpu_has_vmx_msr_bitmap())
2373 vmx_set_msr_bitmap(&vmx->vcpu);
2377 * reads and returns guest's timestamp counter "register"
2378 * guest_tsc = (host_tsc * tsc multiplier) >> 48 + tsc_offset
2379 * -- Intel TSC Scaling for Virtualization White Paper, sec 1.3
2381 static u64 guest_read_tsc(struct kvm_vcpu *vcpu)
2383 u64 host_tsc, tsc_offset;
2386 tsc_offset = vmcs_read64(TSC_OFFSET);
2387 return kvm_scale_tsc(vcpu, host_tsc) + tsc_offset;
2391 * Like guest_read_tsc, but always returns L1's notion of the timestamp
2392 * counter, even if a nested guest (L2) is currently running.
2394 static u64 vmx_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
2398 tsc_offset = is_guest_mode(vcpu) ?
2399 to_vmx(vcpu)->nested.vmcs01_tsc_offset :
2400 vmcs_read64(TSC_OFFSET);
2401 return host_tsc + tsc_offset;
2404 static u64 vmx_read_tsc_offset(struct kvm_vcpu *vcpu)
2406 return vmcs_read64(TSC_OFFSET);
2410 * writes 'offset' into guest's timestamp counter offset register
2412 static void vmx_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
2414 if (is_guest_mode(vcpu)) {
2416 * We're here if L1 chose not to trap WRMSR to TSC. According
2417 * to the spec, this should set L1's TSC; The offset that L1
2418 * set for L2 remains unchanged, and still needs to be added
2419 * to the newly set TSC to get L2's TSC.
2421 struct vmcs12 *vmcs12;
2422 to_vmx(vcpu)->nested.vmcs01_tsc_offset = offset;
2423 /* recalculate vmcs02.TSC_OFFSET: */
2424 vmcs12 = get_vmcs12(vcpu);
2425 vmcs_write64(TSC_OFFSET, offset +
2426 (nested_cpu_has(vmcs12, CPU_BASED_USE_TSC_OFFSETING) ?
2427 vmcs12->tsc_offset : 0));
2429 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
2430 vmcs_read64(TSC_OFFSET), offset);
2431 vmcs_write64(TSC_OFFSET, offset);
2435 static void vmx_adjust_tsc_offset_guest(struct kvm_vcpu *vcpu, s64 adjustment)
2437 u64 offset = vmcs_read64(TSC_OFFSET);
2439 vmcs_write64(TSC_OFFSET, offset + adjustment);
2440 if (is_guest_mode(vcpu)) {
2441 /* Even when running L2, the adjustment needs to apply to L1 */
2442 to_vmx(vcpu)->nested.vmcs01_tsc_offset += adjustment;
2444 trace_kvm_write_tsc_offset(vcpu->vcpu_id, offset,
2445 offset + adjustment);
2448 static bool guest_cpuid_has_vmx(struct kvm_vcpu *vcpu)
2450 struct kvm_cpuid_entry2 *best = kvm_find_cpuid_entry(vcpu, 1, 0);
2451 return best && (best->ecx & (1 << (X86_FEATURE_VMX & 31)));
2455 * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
2456 * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
2457 * all guests if the "nested" module option is off, and can also be disabled
2458 * for a single guest by disabling its VMX cpuid bit.
2460 static inline bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
2462 return nested && guest_cpuid_has_vmx(vcpu);
2466 * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be
2467 * returned for the various VMX controls MSRs when nested VMX is enabled.
2468 * The same values should also be used to verify that vmcs12 control fields are
2469 * valid during nested entry from L1 to L2.
2470 * Each of these control msrs has a low and high 32-bit half: A low bit is on
2471 * if the corresponding bit in the (32-bit) control field *must* be on, and a
2472 * bit in the high half is on if the corresponding bit in the control field
2473 * may be on. See also vmx_control_verify().
2475 static void nested_vmx_setup_ctls_msrs(struct vcpu_vmx *vmx)
2478 * Note that as a general rule, the high half of the MSRs (bits in
2479 * the control fields which may be 1) should be initialized by the
2480 * intersection of the underlying hardware's MSR (i.e., features which
2481 * can be supported) and the list of features we want to expose -
2482 * because they are known to be properly supported in our code.
2483 * Also, usually, the low half of the MSRs (bits which must be 1) can
2484 * be set to 0, meaning that L1 may turn off any of these bits. The
2485 * reason is that if one of these bits is necessary, it will appear
2486 * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control
2487 * fields of vmcs01 and vmcs02, will turn these bits off - and
2488 * nested_vmx_exit_handled() will not pass related exits to L1.
2489 * These rules have exceptions below.
2492 /* pin-based controls */
2493 rdmsr(MSR_IA32_VMX_PINBASED_CTLS,
2494 vmx->nested.nested_vmx_pinbased_ctls_low,
2495 vmx->nested.nested_vmx_pinbased_ctls_high);
2496 vmx->nested.nested_vmx_pinbased_ctls_low |=
2497 PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
2498 vmx->nested.nested_vmx_pinbased_ctls_high &=
2499 PIN_BASED_EXT_INTR_MASK |
2500 PIN_BASED_NMI_EXITING |
2501 PIN_BASED_VIRTUAL_NMIS;
2502 vmx->nested.nested_vmx_pinbased_ctls_high |=
2503 PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
2504 PIN_BASED_VMX_PREEMPTION_TIMER;
2505 if (vmx_cpu_uses_apicv(&vmx->vcpu))
2506 vmx->nested.nested_vmx_pinbased_ctls_high |=
2507 PIN_BASED_POSTED_INTR;
2510 rdmsr(MSR_IA32_VMX_EXIT_CTLS,
2511 vmx->nested.nested_vmx_exit_ctls_low,
2512 vmx->nested.nested_vmx_exit_ctls_high);
2513 vmx->nested.nested_vmx_exit_ctls_low =
2514 VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
2516 vmx->nested.nested_vmx_exit_ctls_high &=
2517 #ifdef CONFIG_X86_64
2518 VM_EXIT_HOST_ADDR_SPACE_SIZE |
2520 VM_EXIT_LOAD_IA32_PAT | VM_EXIT_SAVE_IA32_PAT;
2521 vmx->nested.nested_vmx_exit_ctls_high |=
2522 VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR |
2523 VM_EXIT_LOAD_IA32_EFER | VM_EXIT_SAVE_IA32_EFER |
2524 VM_EXIT_SAVE_VMX_PREEMPTION_TIMER | VM_EXIT_ACK_INTR_ON_EXIT;
2526 if (vmx_mpx_supported())
2527 vmx->nested.nested_vmx_exit_ctls_high |= VM_EXIT_CLEAR_BNDCFGS;
2529 /* We support free control of debug control saving. */
2530 vmx->nested.nested_vmx_true_exit_ctls_low =
2531 vmx->nested.nested_vmx_exit_ctls_low &
2532 ~VM_EXIT_SAVE_DEBUG_CONTROLS;
2534 /* entry controls */
2535 rdmsr(MSR_IA32_VMX_ENTRY_CTLS,
2536 vmx->nested.nested_vmx_entry_ctls_low,
2537 vmx->nested.nested_vmx_entry_ctls_high);
2538 vmx->nested.nested_vmx_entry_ctls_low =
2539 VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
2540 vmx->nested.nested_vmx_entry_ctls_high &=
2541 #ifdef CONFIG_X86_64
2542 VM_ENTRY_IA32E_MODE |
2544 VM_ENTRY_LOAD_IA32_PAT;
2545 vmx->nested.nested_vmx_entry_ctls_high |=
2546 (VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR | VM_ENTRY_LOAD_IA32_EFER);
2547 if (vmx_mpx_supported())
2548 vmx->nested.nested_vmx_entry_ctls_high |= VM_ENTRY_LOAD_BNDCFGS;
2550 /* We support free control of debug control loading. */
2551 vmx->nested.nested_vmx_true_entry_ctls_low =
2552 vmx->nested.nested_vmx_entry_ctls_low &
2553 ~VM_ENTRY_LOAD_DEBUG_CONTROLS;
2555 /* cpu-based controls */
2556 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS,
2557 vmx->nested.nested_vmx_procbased_ctls_low,
2558 vmx->nested.nested_vmx_procbased_ctls_high);
2559 vmx->nested.nested_vmx_procbased_ctls_low =
2560 CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
2561 vmx->nested.nested_vmx_procbased_ctls_high &=
2562 CPU_BASED_VIRTUAL_INTR_PENDING |
2563 CPU_BASED_VIRTUAL_NMI_PENDING | CPU_BASED_USE_TSC_OFFSETING |
2564 CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING |
2565 CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING |
2566 CPU_BASED_CR3_STORE_EXITING |
2567 #ifdef CONFIG_X86_64
2568 CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING |
2570 CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING |
2571 CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_TRAP_FLAG |
2572 CPU_BASED_MONITOR_EXITING | CPU_BASED_RDPMC_EXITING |
2573 CPU_BASED_RDTSC_EXITING | CPU_BASED_PAUSE_EXITING |
2574 CPU_BASED_TPR_SHADOW | CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2576 * We can allow some features even when not supported by the
2577 * hardware. For example, L1 can specify an MSR bitmap - and we
2578 * can use it to avoid exits to L1 - even when L0 runs L2
2579 * without MSR bitmaps.
2581 vmx->nested.nested_vmx_procbased_ctls_high |=
2582 CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
2583 CPU_BASED_USE_MSR_BITMAPS;
2585 /* We support free control of CR3 access interception. */
2586 vmx->nested.nested_vmx_true_procbased_ctls_low =
2587 vmx->nested.nested_vmx_procbased_ctls_low &
2588 ~(CPU_BASED_CR3_LOAD_EXITING | CPU_BASED_CR3_STORE_EXITING);
2590 /* secondary cpu-based controls */
2591 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
2592 vmx->nested.nested_vmx_secondary_ctls_low,
2593 vmx->nested.nested_vmx_secondary_ctls_high);
2594 vmx->nested.nested_vmx_secondary_ctls_low = 0;
2595 vmx->nested.nested_vmx_secondary_ctls_high &=
2596 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2597 SECONDARY_EXEC_RDTSCP |
2598 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2599 SECONDARY_EXEC_ENABLE_VPID |
2600 SECONDARY_EXEC_APIC_REGISTER_VIRT |
2601 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2602 SECONDARY_EXEC_WBINVD_EXITING |
2603 SECONDARY_EXEC_XSAVES |
2604 SECONDARY_EXEC_PCOMMIT;
2607 /* nested EPT: emulate EPT also to L1 */
2608 vmx->nested.nested_vmx_secondary_ctls_high |=
2609 SECONDARY_EXEC_ENABLE_EPT;
2610 vmx->nested.nested_vmx_ept_caps = VMX_EPT_PAGE_WALK_4_BIT |
2611 VMX_EPTP_WB_BIT | VMX_EPT_2MB_PAGE_BIT |
2613 vmx->nested.nested_vmx_ept_caps &= vmx_capability.ept;
2615 * For nested guests, we don't do anything specific
2616 * for single context invalidation. Hence, only advertise
2617 * support for global context invalidation.
2619 vmx->nested.nested_vmx_ept_caps |= VMX_EPT_EXTENT_GLOBAL_BIT;
2621 vmx->nested.nested_vmx_ept_caps = 0;
2624 vmx->nested.nested_vmx_vpid_caps = VMX_VPID_INVVPID_BIT |
2625 VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT;
2627 vmx->nested.nested_vmx_vpid_caps = 0;
2629 if (enable_unrestricted_guest)
2630 vmx->nested.nested_vmx_secondary_ctls_high |=
2631 SECONDARY_EXEC_UNRESTRICTED_GUEST;
2633 /* miscellaneous data */
2634 rdmsr(MSR_IA32_VMX_MISC,
2635 vmx->nested.nested_vmx_misc_low,
2636 vmx->nested.nested_vmx_misc_high);
2637 vmx->nested.nested_vmx_misc_low &= VMX_MISC_SAVE_EFER_LMA;
2638 vmx->nested.nested_vmx_misc_low |=
2639 VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE |
2640 VMX_MISC_ACTIVITY_HLT;
2641 vmx->nested.nested_vmx_misc_high = 0;
2644 static inline bool vmx_control_verify(u32 control, u32 low, u32 high)
2647 * Bits 0 in high must be 0, and bits 1 in low must be 1.
2649 return ((control & high) | low) == control;
2652 static inline u64 vmx_control_msr(u32 low, u32 high)
2654 return low | ((u64)high << 32);
2657 /* Returns 0 on success, non-0 otherwise. */
2658 static int vmx_get_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
2660 struct vcpu_vmx *vmx = to_vmx(vcpu);
2662 switch (msr_index) {
2663 case MSR_IA32_VMX_BASIC:
2665 * This MSR reports some information about VMX support. We
2666 * should return information about the VMX we emulate for the
2667 * guest, and the VMCS structure we give it - not about the
2668 * VMX support of the underlying hardware.
2670 *pdata = VMCS12_REVISION | VMX_BASIC_TRUE_CTLS |
2671 ((u64)VMCS12_SIZE << VMX_BASIC_VMCS_SIZE_SHIFT) |
2672 (VMX_BASIC_MEM_TYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT);
2674 case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
2675 case MSR_IA32_VMX_PINBASED_CTLS:
2676 *pdata = vmx_control_msr(
2677 vmx->nested.nested_vmx_pinbased_ctls_low,
2678 vmx->nested.nested_vmx_pinbased_ctls_high);
2680 case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
2681 *pdata = vmx_control_msr(
2682 vmx->nested.nested_vmx_true_procbased_ctls_low,
2683 vmx->nested.nested_vmx_procbased_ctls_high);
2685 case MSR_IA32_VMX_PROCBASED_CTLS:
2686 *pdata = vmx_control_msr(
2687 vmx->nested.nested_vmx_procbased_ctls_low,
2688 vmx->nested.nested_vmx_procbased_ctls_high);
2690 case MSR_IA32_VMX_TRUE_EXIT_CTLS:
2691 *pdata = vmx_control_msr(
2692 vmx->nested.nested_vmx_true_exit_ctls_low,
2693 vmx->nested.nested_vmx_exit_ctls_high);
2695 case MSR_IA32_VMX_EXIT_CTLS:
2696 *pdata = vmx_control_msr(
2697 vmx->nested.nested_vmx_exit_ctls_low,
2698 vmx->nested.nested_vmx_exit_ctls_high);
2700 case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
2701 *pdata = vmx_control_msr(
2702 vmx->nested.nested_vmx_true_entry_ctls_low,
2703 vmx->nested.nested_vmx_entry_ctls_high);
2705 case MSR_IA32_VMX_ENTRY_CTLS:
2706 *pdata = vmx_control_msr(
2707 vmx->nested.nested_vmx_entry_ctls_low,
2708 vmx->nested.nested_vmx_entry_ctls_high);
2710 case MSR_IA32_VMX_MISC:
2711 *pdata = vmx_control_msr(
2712 vmx->nested.nested_vmx_misc_low,
2713 vmx->nested.nested_vmx_misc_high);
2716 * These MSRs specify bits which the guest must keep fixed (on or off)
2717 * while L1 is in VMXON mode (in L1's root mode, or running an L2).
2718 * We picked the standard core2 setting.
2720 #define VMXON_CR0_ALWAYSON (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
2721 #define VMXON_CR4_ALWAYSON X86_CR4_VMXE
2722 case MSR_IA32_VMX_CR0_FIXED0:
2723 *pdata = VMXON_CR0_ALWAYSON;
2725 case MSR_IA32_VMX_CR0_FIXED1:
2728 case MSR_IA32_VMX_CR4_FIXED0:
2729 *pdata = VMXON_CR4_ALWAYSON;
2731 case MSR_IA32_VMX_CR4_FIXED1:
2734 case MSR_IA32_VMX_VMCS_ENUM:
2735 *pdata = 0x2e; /* highest index: VMX_PREEMPTION_TIMER_VALUE */
2737 case MSR_IA32_VMX_PROCBASED_CTLS2:
2738 *pdata = vmx_control_msr(
2739 vmx->nested.nested_vmx_secondary_ctls_low,
2740 vmx->nested.nested_vmx_secondary_ctls_high);
2742 case MSR_IA32_VMX_EPT_VPID_CAP:
2743 /* Currently, no nested vpid support */
2744 *pdata = vmx->nested.nested_vmx_ept_caps |
2745 ((u64)vmx->nested.nested_vmx_vpid_caps << 32);
2755 * Reads an msr value (of 'msr_index') into 'pdata'.
2756 * Returns 0 on success, non-0 otherwise.
2757 * Assumes vcpu_load() was already called.
2759 static int vmx_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2761 struct shared_msr_entry *msr;
2763 switch (msr_info->index) {
2764 #ifdef CONFIG_X86_64
2766 msr_info->data = vmcs_readl(GUEST_FS_BASE);
2769 msr_info->data = vmcs_readl(GUEST_GS_BASE);
2771 case MSR_KERNEL_GS_BASE:
2772 vmx_load_host_state(to_vmx(vcpu));
2773 msr_info->data = to_vmx(vcpu)->msr_guest_kernel_gs_base;
2777 return kvm_get_msr_common(vcpu, msr_info);
2779 msr_info->data = guest_read_tsc(vcpu);
2781 case MSR_IA32_SYSENTER_CS:
2782 msr_info->data = vmcs_read32(GUEST_SYSENTER_CS);
2784 case MSR_IA32_SYSENTER_EIP:
2785 msr_info->data = vmcs_readl(GUEST_SYSENTER_EIP);
2787 case MSR_IA32_SYSENTER_ESP:
2788 msr_info->data = vmcs_readl(GUEST_SYSENTER_ESP);
2790 case MSR_IA32_BNDCFGS:
2791 if (!vmx_mpx_supported())
2793 msr_info->data = vmcs_read64(GUEST_BNDCFGS);
2795 case MSR_IA32_FEATURE_CONTROL:
2796 if (!nested_vmx_allowed(vcpu))
2798 msr_info->data = to_vmx(vcpu)->nested.msr_ia32_feature_control;
2800 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
2801 if (!nested_vmx_allowed(vcpu))
2803 return vmx_get_vmx_msr(vcpu, msr_info->index, &msr_info->data);
2805 if (!vmx_xsaves_supported())
2807 msr_info->data = vcpu->arch.ia32_xss;
2810 if (!guest_cpuid_has_rdtscp(vcpu) && !msr_info->host_initiated)
2812 /* Otherwise falls through */
2814 msr = find_msr_entry(to_vmx(vcpu), msr_info->index);
2816 msr_info->data = msr->data;
2819 return kvm_get_msr_common(vcpu, msr_info);
2825 static void vmx_leave_nested(struct kvm_vcpu *vcpu);
2828 * Writes msr value into into the appropriate "register".
2829 * Returns 0 on success, non-0 otherwise.
2830 * Assumes vcpu_load() was already called.
2832 static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2834 struct vcpu_vmx *vmx = to_vmx(vcpu);
2835 struct shared_msr_entry *msr;
2837 u32 msr_index = msr_info->index;
2838 u64 data = msr_info->data;
2840 switch (msr_index) {
2842 ret = kvm_set_msr_common(vcpu, msr_info);
2844 #ifdef CONFIG_X86_64
2846 vmx_segment_cache_clear(vmx);
2847 vmcs_writel(GUEST_FS_BASE, data);
2850 vmx_segment_cache_clear(vmx);
2851 vmcs_writel(GUEST_GS_BASE, data);
2853 case MSR_KERNEL_GS_BASE:
2854 vmx_load_host_state(vmx);
2855 vmx->msr_guest_kernel_gs_base = data;
2858 case MSR_IA32_SYSENTER_CS:
2859 vmcs_write32(GUEST_SYSENTER_CS, data);
2861 case MSR_IA32_SYSENTER_EIP:
2862 vmcs_writel(GUEST_SYSENTER_EIP, data);
2864 case MSR_IA32_SYSENTER_ESP:
2865 vmcs_writel(GUEST_SYSENTER_ESP, data);
2867 case MSR_IA32_BNDCFGS:
2868 if (!vmx_mpx_supported())
2870 vmcs_write64(GUEST_BNDCFGS, data);
2873 kvm_write_tsc(vcpu, msr_info);
2875 case MSR_IA32_CR_PAT:
2876 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2877 if (!kvm_mtrr_valid(vcpu, MSR_IA32_CR_PAT, data))
2879 vmcs_write64(GUEST_IA32_PAT, data);
2880 vcpu->arch.pat = data;
2883 ret = kvm_set_msr_common(vcpu, msr_info);
2885 case MSR_IA32_TSC_ADJUST:
2886 ret = kvm_set_msr_common(vcpu, msr_info);
2888 case MSR_IA32_FEATURE_CONTROL:
2889 if (!nested_vmx_allowed(vcpu) ||
2890 (to_vmx(vcpu)->nested.msr_ia32_feature_control &
2891 FEATURE_CONTROL_LOCKED && !msr_info->host_initiated))
2893 vmx->nested.msr_ia32_feature_control = data;
2894 if (msr_info->host_initiated && data == 0)
2895 vmx_leave_nested(vcpu);
2897 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
2898 return 1; /* they are read-only */
2900 if (!vmx_xsaves_supported())
2903 * The only supported bit as of Skylake is bit 8, but
2904 * it is not supported on KVM.
2908 vcpu->arch.ia32_xss = data;
2909 if (vcpu->arch.ia32_xss != host_xss)
2910 add_atomic_switch_msr(vmx, MSR_IA32_XSS,
2911 vcpu->arch.ia32_xss, host_xss);
2913 clear_atomic_switch_msr(vmx, MSR_IA32_XSS);
2916 if (!guest_cpuid_has_rdtscp(vcpu) && !msr_info->host_initiated)
2918 /* Check reserved bit, higher 32 bits should be zero */
2919 if ((data >> 32) != 0)
2921 /* Otherwise falls through */
2923 msr = find_msr_entry(vmx, msr_index);
2925 u64 old_msr_data = msr->data;
2927 if (msr - vmx->guest_msrs < vmx->save_nmsrs) {
2929 ret = kvm_set_shared_msr(msr->index, msr->data,
2933 msr->data = old_msr_data;
2937 ret = kvm_set_msr_common(vcpu, msr_info);
2943 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
2945 __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
2948 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
2951 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
2953 case VCPU_EXREG_PDPTR:
2955 ept_save_pdptrs(vcpu);
2962 static __init int cpu_has_kvm_support(void)
2964 return cpu_has_vmx();
2967 static __init int vmx_disabled_by_bios(void)
2971 rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
2972 if (msr & FEATURE_CONTROL_LOCKED) {
2973 /* launched w/ TXT and VMX disabled */
2974 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
2977 /* launched w/o TXT and VMX only enabled w/ TXT */
2978 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
2979 && (msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
2980 && !tboot_enabled()) {
2981 printk(KERN_WARNING "kvm: disable TXT in the BIOS or "
2982 "activate TXT before enabling KVM\n");
2985 /* launched w/o TXT and VMX disabled */
2986 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
2987 && !tboot_enabled())
2994 static void kvm_cpu_vmxon(u64 addr)
2996 asm volatile (ASM_VMX_VMXON_RAX
2997 : : "a"(&addr), "m"(addr)
3001 static int hardware_enable(void)
3003 int cpu = raw_smp_processor_id();
3004 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
3007 if (cr4_read_shadow() & X86_CR4_VMXE)
3010 INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
3011 INIT_LIST_HEAD(&per_cpu(blocked_vcpu_on_cpu, cpu));
3012 spin_lock_init(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
3015 * Now we can enable the vmclear operation in kdump
3016 * since the loaded_vmcss_on_cpu list on this cpu
3017 * has been initialized.
3019 * Though the cpu is not in VMX operation now, there
3020 * is no problem to enable the vmclear operation
3021 * for the loaded_vmcss_on_cpu list is empty!
3023 crash_enable_local_vmclear(cpu);
3025 rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
3027 test_bits = FEATURE_CONTROL_LOCKED;
3028 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
3029 if (tboot_enabled())
3030 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX;
3032 if ((old & test_bits) != test_bits) {
3033 /* enable and lock */
3034 wrmsrl(MSR_IA32_FEATURE_CONTROL, old | test_bits);
3036 cr4_set_bits(X86_CR4_VMXE);
3038 if (vmm_exclusive) {
3039 kvm_cpu_vmxon(phys_addr);
3043 native_store_gdt(this_cpu_ptr(&host_gdt));
3048 static void vmclear_local_loaded_vmcss(void)
3050 int cpu = raw_smp_processor_id();
3051 struct loaded_vmcs *v, *n;
3053 list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
3054 loaded_vmcss_on_cpu_link)
3055 __loaded_vmcs_clear(v);
3059 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
3062 static void kvm_cpu_vmxoff(void)
3064 asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
3067 static void hardware_disable(void)
3069 if (vmm_exclusive) {
3070 vmclear_local_loaded_vmcss();
3073 cr4_clear_bits(X86_CR4_VMXE);
3076 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
3077 u32 msr, u32 *result)
3079 u32 vmx_msr_low, vmx_msr_high;
3080 u32 ctl = ctl_min | ctl_opt;
3082 rdmsr(msr, vmx_msr_low, vmx_msr_high);
3084 ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
3085 ctl |= vmx_msr_low; /* bit == 1 in low word ==> must be one */
3087 /* Ensure minimum (required) set of control bits are supported. */
3095 static __init bool allow_1_setting(u32 msr, u32 ctl)
3097 u32 vmx_msr_low, vmx_msr_high;
3099 rdmsr(msr, vmx_msr_low, vmx_msr_high);
3100 return vmx_msr_high & ctl;
3103 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
3105 u32 vmx_msr_low, vmx_msr_high;
3106 u32 min, opt, min2, opt2;
3107 u32 _pin_based_exec_control = 0;
3108 u32 _cpu_based_exec_control = 0;
3109 u32 _cpu_based_2nd_exec_control = 0;
3110 u32 _vmexit_control = 0;
3111 u32 _vmentry_control = 0;
3113 min = CPU_BASED_HLT_EXITING |
3114 #ifdef CONFIG_X86_64
3115 CPU_BASED_CR8_LOAD_EXITING |
3116 CPU_BASED_CR8_STORE_EXITING |
3118 CPU_BASED_CR3_LOAD_EXITING |
3119 CPU_BASED_CR3_STORE_EXITING |
3120 CPU_BASED_USE_IO_BITMAPS |
3121 CPU_BASED_MOV_DR_EXITING |
3122 CPU_BASED_USE_TSC_OFFSETING |
3123 CPU_BASED_MWAIT_EXITING |
3124 CPU_BASED_MONITOR_EXITING |
3125 CPU_BASED_INVLPG_EXITING |
3126 CPU_BASED_RDPMC_EXITING;
3128 opt = CPU_BASED_TPR_SHADOW |
3129 CPU_BASED_USE_MSR_BITMAPS |
3130 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
3131 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
3132 &_cpu_based_exec_control) < 0)
3134 #ifdef CONFIG_X86_64
3135 if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
3136 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
3137 ~CPU_BASED_CR8_STORE_EXITING;
3139 if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
3141 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
3142 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
3143 SECONDARY_EXEC_WBINVD_EXITING |
3144 SECONDARY_EXEC_ENABLE_VPID |
3145 SECONDARY_EXEC_ENABLE_EPT |
3146 SECONDARY_EXEC_UNRESTRICTED_GUEST |
3147 SECONDARY_EXEC_PAUSE_LOOP_EXITING |
3148 SECONDARY_EXEC_RDTSCP |
3149 SECONDARY_EXEC_ENABLE_INVPCID |
3150 SECONDARY_EXEC_APIC_REGISTER_VIRT |
3151 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
3152 SECONDARY_EXEC_SHADOW_VMCS |
3153 SECONDARY_EXEC_XSAVES |
3154 SECONDARY_EXEC_ENABLE_PML |
3155 SECONDARY_EXEC_PCOMMIT |
3156 SECONDARY_EXEC_TSC_SCALING;
3157 if (adjust_vmx_controls(min2, opt2,
3158 MSR_IA32_VMX_PROCBASED_CTLS2,
3159 &_cpu_based_2nd_exec_control) < 0)
3162 #ifndef CONFIG_X86_64
3163 if (!(_cpu_based_2nd_exec_control &
3164 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
3165 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
3168 if (!(_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
3169 _cpu_based_2nd_exec_control &= ~(
3170 SECONDARY_EXEC_APIC_REGISTER_VIRT |
3171 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
3172 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
3174 if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
3175 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
3177 _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
3178 CPU_BASED_CR3_STORE_EXITING |
3179 CPU_BASED_INVLPG_EXITING);
3180 rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
3181 vmx_capability.ept, vmx_capability.vpid);
3184 min = VM_EXIT_SAVE_DEBUG_CONTROLS;
3185 #ifdef CONFIG_X86_64
3186 min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
3188 opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT |
3189 VM_EXIT_ACK_INTR_ON_EXIT | VM_EXIT_CLEAR_BNDCFGS;
3190 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
3191 &_vmexit_control) < 0)
3194 min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
3195 opt = PIN_BASED_VIRTUAL_NMIS | PIN_BASED_POSTED_INTR;
3196 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
3197 &_pin_based_exec_control) < 0)
3200 if (!(_cpu_based_2nd_exec_control &
3201 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) ||
3202 !(_vmexit_control & VM_EXIT_ACK_INTR_ON_EXIT))
3203 _pin_based_exec_control &= ~PIN_BASED_POSTED_INTR;
3205 min = VM_ENTRY_LOAD_DEBUG_CONTROLS;
3206 opt = VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_BNDCFGS;
3207 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
3208 &_vmentry_control) < 0)
3211 rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
3213 /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
3214 if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
3217 #ifdef CONFIG_X86_64
3218 /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
3219 if (vmx_msr_high & (1u<<16))
3223 /* Require Write-Back (WB) memory type for VMCS accesses. */
3224 if (((vmx_msr_high >> 18) & 15) != 6)
3227 vmcs_conf->size = vmx_msr_high & 0x1fff;
3228 vmcs_conf->order = get_order(vmcs_config.size);
3229 vmcs_conf->revision_id = vmx_msr_low;
3231 vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
3232 vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
3233 vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
3234 vmcs_conf->vmexit_ctrl = _vmexit_control;
3235 vmcs_conf->vmentry_ctrl = _vmentry_control;
3237 cpu_has_load_ia32_efer =
3238 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
3239 VM_ENTRY_LOAD_IA32_EFER)
3240 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
3241 VM_EXIT_LOAD_IA32_EFER);
3243 cpu_has_load_perf_global_ctrl =
3244 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
3245 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
3246 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
3247 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
3250 * Some cpus support VM_ENTRY_(LOAD|SAVE)_IA32_PERF_GLOBAL_CTRL
3251 * but due to arrata below it can't be used. Workaround is to use
3252 * msr load mechanism to switch IA32_PERF_GLOBAL_CTRL.
3254 * VM Exit May Incorrectly Clear IA32_PERF_GLOBAL_CTRL [34:32]
3259 * BC86,AAY89,BD102 (model 44)
3263 if (cpu_has_load_perf_global_ctrl && boot_cpu_data.x86 == 0x6) {
3264 switch (boot_cpu_data.x86_model) {
3270 cpu_has_load_perf_global_ctrl = false;
3271 printk_once(KERN_WARNING"kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
3272 "does not work properly. Using workaround\n");
3280 rdmsrl(MSR_IA32_XSS, host_xss);
3285 static struct vmcs *alloc_vmcs_cpu(int cpu)
3287 int node = cpu_to_node(cpu);
3291 pages = __alloc_pages_node(node, GFP_KERNEL, vmcs_config.order);
3294 vmcs = page_address(pages);
3295 memset(vmcs, 0, vmcs_config.size);
3296 vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
3300 static struct vmcs *alloc_vmcs(void)
3302 return alloc_vmcs_cpu(raw_smp_processor_id());
3305 static void free_vmcs(struct vmcs *vmcs)
3307 free_pages((unsigned long)vmcs, vmcs_config.order);
3311 * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
3313 static void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
3315 if (!loaded_vmcs->vmcs)
3317 loaded_vmcs_clear(loaded_vmcs);
3318 free_vmcs(loaded_vmcs->vmcs);
3319 loaded_vmcs->vmcs = NULL;
3322 static void free_kvm_area(void)
3326 for_each_possible_cpu(cpu) {
3327 free_vmcs(per_cpu(vmxarea, cpu));
3328 per_cpu(vmxarea, cpu) = NULL;
3332 static void init_vmcs_shadow_fields(void)
3336 /* No checks for read only fields yet */
3338 for (i = j = 0; i < max_shadow_read_write_fields; i++) {
3339 switch (shadow_read_write_fields[i]) {
3341 if (!vmx_mpx_supported())
3349 shadow_read_write_fields[j] =
3350 shadow_read_write_fields[i];
3353 max_shadow_read_write_fields = j;
3355 /* shadowed fields guest access without vmexit */
3356 for (i = 0; i < max_shadow_read_write_fields; i++) {
3357 clear_bit(shadow_read_write_fields[i],
3358 vmx_vmwrite_bitmap);
3359 clear_bit(shadow_read_write_fields[i],
3362 for (i = 0; i < max_shadow_read_only_fields; i++)
3363 clear_bit(shadow_read_only_fields[i],
3367 static __init int alloc_kvm_area(void)
3371 for_each_possible_cpu(cpu) {
3374 vmcs = alloc_vmcs_cpu(cpu);
3380 per_cpu(vmxarea, cpu) = vmcs;
3385 static bool emulation_required(struct kvm_vcpu *vcpu)
3387 return emulate_invalid_guest_state && !guest_state_valid(vcpu);
3390 static void fix_pmode_seg(struct kvm_vcpu *vcpu, int seg,
3391 struct kvm_segment *save)
3393 if (!emulate_invalid_guest_state) {
3395 * CS and SS RPL should be equal during guest entry according
3396 * to VMX spec, but in reality it is not always so. Since vcpu
3397 * is in the middle of the transition from real mode to
3398 * protected mode it is safe to assume that RPL 0 is a good
3401 if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
3402 save->selector &= ~SEGMENT_RPL_MASK;
3403 save->dpl = save->selector & SEGMENT_RPL_MASK;
3406 vmx_set_segment(vcpu, save, seg);
3409 static void enter_pmode(struct kvm_vcpu *vcpu)
3411 unsigned long flags;
3412 struct vcpu_vmx *vmx = to_vmx(vcpu);
3415 * Update real mode segment cache. It may be not up-to-date if sement
3416 * register was written while vcpu was in a guest mode.
3418 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
3419 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
3420 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
3421 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
3422 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
3423 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
3425 vmx->rmode.vm86_active = 0;
3427 vmx_segment_cache_clear(vmx);
3429 vmx_set_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
3431 flags = vmcs_readl(GUEST_RFLAGS);
3432 flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
3433 flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
3434 vmcs_writel(GUEST_RFLAGS, flags);
3436 vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
3437 (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
3439 update_exception_bitmap(vcpu);
3441 fix_pmode_seg(vcpu, VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
3442 fix_pmode_seg(vcpu, VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
3443 fix_pmode_seg(vcpu, VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
3444 fix_pmode_seg(vcpu, VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
3445 fix_pmode_seg(vcpu, VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
3446 fix_pmode_seg(vcpu, VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
3449 static void fix_rmode_seg(int seg, struct kvm_segment *save)
3451 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3452 struct kvm_segment var = *save;
3455 if (seg == VCPU_SREG_CS)
3458 if (!emulate_invalid_guest_state) {
3459 var.selector = var.base >> 4;
3460 var.base = var.base & 0xffff0;
3470 if (save->base & 0xf)
3471 printk_once(KERN_WARNING "kvm: segment base is not "
3472 "paragraph aligned when entering "
3473 "protected mode (seg=%d)", seg);
3476 vmcs_write16(sf->selector, var.selector);
3477 vmcs_write32(sf->base, var.base);
3478 vmcs_write32(sf->limit, var.limit);
3479 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(&var));
3482 static void enter_rmode(struct kvm_vcpu *vcpu)
3484 unsigned long flags;
3485 struct vcpu_vmx *vmx = to_vmx(vcpu);
3487 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
3488 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
3489 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
3490 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
3491 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
3492 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
3493 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
3495 vmx->rmode.vm86_active = 1;
3498 * Very old userspace does not call KVM_SET_TSS_ADDR before entering
3499 * vcpu. Warn the user that an update is overdue.
3501 if (!vcpu->kvm->arch.tss_addr)
3502 printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
3503 "called before entering vcpu\n");
3505 vmx_segment_cache_clear(vmx);
3507 vmcs_writel(GUEST_TR_BASE, vcpu->kvm->arch.tss_addr);
3508 vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
3509 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
3511 flags = vmcs_readl(GUEST_RFLAGS);
3512 vmx->rmode.save_rflags = flags;
3514 flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
3516 vmcs_writel(GUEST_RFLAGS, flags);
3517 vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
3518 update_exception_bitmap(vcpu);
3520 fix_rmode_seg(VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
3521 fix_rmode_seg(VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
3522 fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
3523 fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
3524 fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
3525 fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
3527 kvm_mmu_reset_context(vcpu);
3530 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
3532 struct vcpu_vmx *vmx = to_vmx(vcpu);
3533 struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
3539 * Force kernel_gs_base reloading before EFER changes, as control
3540 * of this msr depends on is_long_mode().
3542 vmx_load_host_state(to_vmx(vcpu));
3543 vcpu->arch.efer = efer;
3544 if (efer & EFER_LMA) {
3545 vm_entry_controls_setbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
3548 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
3550 msr->data = efer & ~EFER_LME;
3555 #ifdef CONFIG_X86_64
3557 static void enter_lmode(struct kvm_vcpu *vcpu)
3561 vmx_segment_cache_clear(to_vmx(vcpu));
3563 guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
3564 if ((guest_tr_ar & VMX_AR_TYPE_MASK) != VMX_AR_TYPE_BUSY_64_TSS) {
3565 pr_debug_ratelimited("%s: tss fixup for long mode. \n",
3567 vmcs_write32(GUEST_TR_AR_BYTES,
3568 (guest_tr_ar & ~VMX_AR_TYPE_MASK)
3569 | VMX_AR_TYPE_BUSY_64_TSS);
3571 vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
3574 static void exit_lmode(struct kvm_vcpu *vcpu)
3576 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
3577 vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
3582 static inline void __vmx_flush_tlb(struct kvm_vcpu *vcpu, int vpid)
3584 vpid_sync_context(vpid);
3586 if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
3588 ept_sync_context(construct_eptp(vcpu->arch.mmu.root_hpa));
3592 static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
3594 __vmx_flush_tlb(vcpu, to_vmx(vcpu)->vpid);
3597 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
3599 ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
3601 vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
3602 vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
3605 static void vmx_decache_cr3(struct kvm_vcpu *vcpu)
3607 if (enable_ept && is_paging(vcpu))
3608 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
3609 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
3612 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
3614 ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
3616 vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
3617 vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
3620 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
3622 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
3624 if (!test_bit(VCPU_EXREG_PDPTR,
3625 (unsigned long *)&vcpu->arch.regs_dirty))
3628 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
3629 vmcs_write64(GUEST_PDPTR0, mmu->pdptrs[0]);
3630 vmcs_write64(GUEST_PDPTR1, mmu->pdptrs[1]);
3631 vmcs_write64(GUEST_PDPTR2, mmu->pdptrs[2]);
3632 vmcs_write64(GUEST_PDPTR3, mmu->pdptrs[3]);
3636 static void ept_save_pdptrs(struct kvm_vcpu *vcpu)
3638 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
3640 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
3641 mmu->pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
3642 mmu->pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
3643 mmu->pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
3644 mmu->pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
3647 __set_bit(VCPU_EXREG_PDPTR,
3648 (unsigned long *)&vcpu->arch.regs_avail);
3649 __set_bit(VCPU_EXREG_PDPTR,
3650 (unsigned long *)&vcpu->arch.regs_dirty);
3653 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
3655 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
3657 struct kvm_vcpu *vcpu)
3659 if (!test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
3660 vmx_decache_cr3(vcpu);
3661 if (!(cr0 & X86_CR0_PG)) {
3662 /* From paging/starting to nonpaging */
3663 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
3664 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
3665 (CPU_BASED_CR3_LOAD_EXITING |
3666 CPU_BASED_CR3_STORE_EXITING));
3667 vcpu->arch.cr0 = cr0;
3668 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
3669 } else if (!is_paging(vcpu)) {
3670 /* From nonpaging to paging */
3671 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
3672 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
3673 ~(CPU_BASED_CR3_LOAD_EXITING |
3674 CPU_BASED_CR3_STORE_EXITING));
3675 vcpu->arch.cr0 = cr0;
3676 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
3679 if (!(cr0 & X86_CR0_WP))
3680 *hw_cr0 &= ~X86_CR0_WP;
3683 static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
3685 struct vcpu_vmx *vmx = to_vmx(vcpu);
3686 unsigned long hw_cr0;
3688 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK);
3689 if (enable_unrestricted_guest)
3690 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
3692 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON;
3694 if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
3697 if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
3701 #ifdef CONFIG_X86_64
3702 if (vcpu->arch.efer & EFER_LME) {
3703 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
3705 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
3711 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
3713 if (!vcpu->fpu_active)
3714 hw_cr0 |= X86_CR0_TS | X86_CR0_MP;
3716 vmcs_writel(CR0_READ_SHADOW, cr0);
3717 vmcs_writel(GUEST_CR0, hw_cr0);
3718 vcpu->arch.cr0 = cr0;
3720 /* depends on vcpu->arch.cr0 to be set to a new value */
3721 vmx->emulation_required = emulation_required(vcpu);
3724 static u64 construct_eptp(unsigned long root_hpa)
3728 /* TODO write the value reading from MSR */
3729 eptp = VMX_EPT_DEFAULT_MT |
3730 VMX_EPT_DEFAULT_GAW << VMX_EPT_GAW_EPTP_SHIFT;
3731 if (enable_ept_ad_bits)
3732 eptp |= VMX_EPT_AD_ENABLE_BIT;
3733 eptp |= (root_hpa & PAGE_MASK);
3738 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
3740 unsigned long guest_cr3;
3745 eptp = construct_eptp(cr3);
3746 vmcs_write64(EPT_POINTER, eptp);
3747 if (is_paging(vcpu) || is_guest_mode(vcpu))
3748 guest_cr3 = kvm_read_cr3(vcpu);
3750 guest_cr3 = vcpu->kvm->arch.ept_identity_map_addr;
3751 ept_load_pdptrs(vcpu);
3754 vmx_flush_tlb(vcpu);
3755 vmcs_writel(GUEST_CR3, guest_cr3);
3758 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
3761 * Pass through host's Machine Check Enable value to hw_cr4, which
3762 * is in force while we are in guest mode. Do not let guests control
3763 * this bit, even if host CR4.MCE == 0.
3765 unsigned long hw_cr4 =
3766 (cr4_read_shadow() & X86_CR4_MCE) |
3767 (cr4 & ~X86_CR4_MCE) |
3768 (to_vmx(vcpu)->rmode.vm86_active ?
3769 KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON);
3771 if (cr4 & X86_CR4_VMXE) {
3773 * To use VMXON (and later other VMX instructions), a guest
3774 * must first be able to turn on cr4.VMXE (see handle_vmon()).
3775 * So basically the check on whether to allow nested VMX
3778 if (!nested_vmx_allowed(vcpu))
3781 if (to_vmx(vcpu)->nested.vmxon &&
3782 ((cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON))
3785 vcpu->arch.cr4 = cr4;
3787 if (!is_paging(vcpu)) {
3788 hw_cr4 &= ~X86_CR4_PAE;
3789 hw_cr4 |= X86_CR4_PSE;
3790 } else if (!(cr4 & X86_CR4_PAE)) {
3791 hw_cr4 &= ~X86_CR4_PAE;
3795 if (!enable_unrestricted_guest && !is_paging(vcpu))
3797 * SMEP/SMAP is disabled if CPU is in non-paging mode in
3798 * hardware. However KVM always uses paging mode without
3799 * unrestricted guest.
3800 * To emulate this behavior, SMEP/SMAP needs to be manually
3801 * disabled when guest switches to non-paging mode.
3803 hw_cr4 &= ~(X86_CR4_SMEP | X86_CR4_SMAP);
3805 vmcs_writel(CR4_READ_SHADOW, cr4);
3806 vmcs_writel(GUEST_CR4, hw_cr4);
3810 static void vmx_get_segment(struct kvm_vcpu *vcpu,
3811 struct kvm_segment *var, int seg)
3813 struct vcpu_vmx *vmx = to_vmx(vcpu);
3816 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3817 *var = vmx->rmode.segs[seg];
3818 if (seg == VCPU_SREG_TR
3819 || var->selector == vmx_read_guest_seg_selector(vmx, seg))
3821 var->base = vmx_read_guest_seg_base(vmx, seg);
3822 var->selector = vmx_read_guest_seg_selector(vmx, seg);
3825 var->base = vmx_read_guest_seg_base(vmx, seg);
3826 var->limit = vmx_read_guest_seg_limit(vmx, seg);
3827 var->selector = vmx_read_guest_seg_selector(vmx, seg);
3828 ar = vmx_read_guest_seg_ar(vmx, seg);
3829 var->unusable = (ar >> 16) & 1;
3830 var->type = ar & 15;
3831 var->s = (ar >> 4) & 1;
3832 var->dpl = (ar >> 5) & 3;
3834 * Some userspaces do not preserve unusable property. Since usable
3835 * segment has to be present according to VMX spec we can use present
3836 * property to amend userspace bug by making unusable segment always
3837 * nonpresent. vmx_segment_access_rights() already marks nonpresent
3838 * segment as unusable.
3840 var->present = !var->unusable;
3841 var->avl = (ar >> 12) & 1;
3842 var->l = (ar >> 13) & 1;
3843 var->db = (ar >> 14) & 1;
3844 var->g = (ar >> 15) & 1;
3847 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
3849 struct kvm_segment s;
3851 if (to_vmx(vcpu)->rmode.vm86_active) {
3852 vmx_get_segment(vcpu, &s, seg);
3855 return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
3858 static int vmx_get_cpl(struct kvm_vcpu *vcpu)
3860 struct vcpu_vmx *vmx = to_vmx(vcpu);
3862 if (unlikely(vmx->rmode.vm86_active))
3865 int ar = vmx_read_guest_seg_ar(vmx, VCPU_SREG_SS);
3866 return VMX_AR_DPL(ar);
3870 static u32 vmx_segment_access_rights(struct kvm_segment *var)
3874 if (var->unusable || !var->present)
3877 ar = var->type & 15;
3878 ar |= (var->s & 1) << 4;
3879 ar |= (var->dpl & 3) << 5;
3880 ar |= (var->present & 1) << 7;
3881 ar |= (var->avl & 1) << 12;
3882 ar |= (var->l & 1) << 13;
3883 ar |= (var->db & 1) << 14;
3884 ar |= (var->g & 1) << 15;
3890 static void vmx_set_segment(struct kvm_vcpu *vcpu,
3891 struct kvm_segment *var, int seg)
3893 struct vcpu_vmx *vmx = to_vmx(vcpu);
3894 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3896 vmx_segment_cache_clear(vmx);
3898 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3899 vmx->rmode.segs[seg] = *var;
3900 if (seg == VCPU_SREG_TR)
3901 vmcs_write16(sf->selector, var->selector);
3903 fix_rmode_seg(seg, &vmx->rmode.segs[seg]);
3907 vmcs_writel(sf->base, var->base);
3908 vmcs_write32(sf->limit, var->limit);
3909 vmcs_write16(sf->selector, var->selector);
3912 * Fix the "Accessed" bit in AR field of segment registers for older
3914 * IA32 arch specifies that at the time of processor reset the
3915 * "Accessed" bit in the AR field of segment registers is 1. And qemu
3916 * is setting it to 0 in the userland code. This causes invalid guest
3917 * state vmexit when "unrestricted guest" mode is turned on.
3918 * Fix for this setup issue in cpu_reset is being pushed in the qemu
3919 * tree. Newer qemu binaries with that qemu fix would not need this
3922 if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
3923 var->type |= 0x1; /* Accessed */
3925 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(var));
3928 vmx->emulation_required = emulation_required(vcpu);
3931 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3933 u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
3935 *db = (ar >> 14) & 1;
3936 *l = (ar >> 13) & 1;
3939 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3941 dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
3942 dt->address = vmcs_readl(GUEST_IDTR_BASE);
3945 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3947 vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
3948 vmcs_writel(GUEST_IDTR_BASE, dt->address);
3951 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3953 dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
3954 dt->address = vmcs_readl(GUEST_GDTR_BASE);
3957 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3959 vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
3960 vmcs_writel(GUEST_GDTR_BASE, dt->address);
3963 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
3965 struct kvm_segment var;
3968 vmx_get_segment(vcpu, &var, seg);
3970 if (seg == VCPU_SREG_CS)
3972 ar = vmx_segment_access_rights(&var);
3974 if (var.base != (var.selector << 4))
3976 if (var.limit != 0xffff)
3984 static bool code_segment_valid(struct kvm_vcpu *vcpu)
3986 struct kvm_segment cs;
3987 unsigned int cs_rpl;
3989 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3990 cs_rpl = cs.selector & SEGMENT_RPL_MASK;
3994 if (~cs.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_ACCESSES_MASK))
3998 if (cs.type & VMX_AR_TYPE_WRITEABLE_MASK) {
3999 if (cs.dpl > cs_rpl)
4002 if (cs.dpl != cs_rpl)
4008 /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
4012 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
4014 struct kvm_segment ss;
4015 unsigned int ss_rpl;
4017 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
4018 ss_rpl = ss.selector & SEGMENT_RPL_MASK;
4022 if (ss.type != 3 && ss.type != 7)
4026 if (ss.dpl != ss_rpl) /* DPL != RPL */
4034 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
4036 struct kvm_segment var;
4039 vmx_get_segment(vcpu, &var, seg);
4040 rpl = var.selector & SEGMENT_RPL_MASK;
4048 if (~var.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_WRITEABLE_MASK)) {
4049 if (var.dpl < rpl) /* DPL < RPL */
4053 /* TODO: Add other members to kvm_segment_field to allow checking for other access
4059 static bool tr_valid(struct kvm_vcpu *vcpu)
4061 struct kvm_segment tr;
4063 vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
4067 if (tr.selector & SEGMENT_TI_MASK) /* TI = 1 */
4069 if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
4077 static bool ldtr_valid(struct kvm_vcpu *vcpu)
4079 struct kvm_segment ldtr;
4081 vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
4085 if (ldtr.selector & SEGMENT_TI_MASK) /* TI = 1 */
4095 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
4097 struct kvm_segment cs, ss;
4099 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
4100 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
4102 return ((cs.selector & SEGMENT_RPL_MASK) ==
4103 (ss.selector & SEGMENT_RPL_MASK));
4107 * Check if guest state is valid. Returns true if valid, false if
4109 * We assume that registers are always usable
4111 static bool guest_state_valid(struct kvm_vcpu *vcpu)
4113 if (enable_unrestricted_guest)
4116 /* real mode guest state checks */
4117 if (!is_protmode(vcpu) || (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
4118 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
4120 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
4122 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
4124 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
4126 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
4128 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
4131 /* protected mode guest state checks */
4132 if (!cs_ss_rpl_check(vcpu))
4134 if (!code_segment_valid(vcpu))
4136 if (!stack_segment_valid(vcpu))
4138 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
4140 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
4142 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
4144 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
4146 if (!tr_valid(vcpu))
4148 if (!ldtr_valid(vcpu))
4152 * - Add checks on RIP
4153 * - Add checks on RFLAGS
4159 static int init_rmode_tss(struct kvm *kvm)
4165 idx = srcu_read_lock(&kvm->srcu);
4166 fn = kvm->arch.tss_addr >> PAGE_SHIFT;
4167 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
4170 data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
4171 r = kvm_write_guest_page(kvm, fn++, &data,
4172 TSS_IOPB_BASE_OFFSET, sizeof(u16));
4175 r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
4178 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
4182 r = kvm_write_guest_page(kvm, fn, &data,
4183 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
4186 srcu_read_unlock(&kvm->srcu, idx);
4190 static int init_rmode_identity_map(struct kvm *kvm)
4193 pfn_t identity_map_pfn;
4199 /* Protect kvm->arch.ept_identity_pagetable_done. */
4200 mutex_lock(&kvm->slots_lock);
4202 if (likely(kvm->arch.ept_identity_pagetable_done))
4205 identity_map_pfn = kvm->arch.ept_identity_map_addr >> PAGE_SHIFT;
4207 r = alloc_identity_pagetable(kvm);
4211 idx = srcu_read_lock(&kvm->srcu);
4212 r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
4215 /* Set up identity-mapping pagetable for EPT in real mode */
4216 for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
4217 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
4218 _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
4219 r = kvm_write_guest_page(kvm, identity_map_pfn,
4220 &tmp, i * sizeof(tmp), sizeof(tmp));
4224 kvm->arch.ept_identity_pagetable_done = true;
4227 srcu_read_unlock(&kvm->srcu, idx);
4230 mutex_unlock(&kvm->slots_lock);
4234 static void seg_setup(int seg)
4236 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
4239 vmcs_write16(sf->selector, 0);
4240 vmcs_writel(sf->base, 0);
4241 vmcs_write32(sf->limit, 0xffff);
4243 if (seg == VCPU_SREG_CS)
4244 ar |= 0x08; /* code segment */
4246 vmcs_write32(sf->ar_bytes, ar);
4249 static int alloc_apic_access_page(struct kvm *kvm)
4254 mutex_lock(&kvm->slots_lock);
4255 if (kvm->arch.apic_access_page_done)
4257 r = __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
4258 APIC_DEFAULT_PHYS_BASE, PAGE_SIZE);
4262 page = gfn_to_page(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
4263 if (is_error_page(page)) {
4269 * Do not pin the page in memory, so that memory hot-unplug
4270 * is able to migrate it.
4273 kvm->arch.apic_access_page_done = true;
4275 mutex_unlock(&kvm->slots_lock);
4279 static int alloc_identity_pagetable(struct kvm *kvm)
4281 /* Called with kvm->slots_lock held. */
4285 BUG_ON(kvm->arch.ept_identity_pagetable_done);
4287 r = __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
4288 kvm->arch.ept_identity_map_addr, PAGE_SIZE);
4293 static int allocate_vpid(void)
4299 spin_lock(&vmx_vpid_lock);
4300 vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
4301 if (vpid < VMX_NR_VPIDS)
4302 __set_bit(vpid, vmx_vpid_bitmap);
4305 spin_unlock(&vmx_vpid_lock);
4309 static void free_vpid(int vpid)
4311 if (!enable_vpid || vpid == 0)
4313 spin_lock(&vmx_vpid_lock);
4314 __clear_bit(vpid, vmx_vpid_bitmap);
4315 spin_unlock(&vmx_vpid_lock);
4318 #define MSR_TYPE_R 1
4319 #define MSR_TYPE_W 2
4320 static void __vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
4323 int f = sizeof(unsigned long);
4325 if (!cpu_has_vmx_msr_bitmap())
4329 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
4330 * have the write-low and read-high bitmap offsets the wrong way round.
4331 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
4333 if (msr <= 0x1fff) {
4334 if (type & MSR_TYPE_R)
4336 __clear_bit(msr, msr_bitmap + 0x000 / f);
4338 if (type & MSR_TYPE_W)
4340 __clear_bit(msr, msr_bitmap + 0x800 / f);
4342 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
4344 if (type & MSR_TYPE_R)
4346 __clear_bit(msr, msr_bitmap + 0x400 / f);
4348 if (type & MSR_TYPE_W)
4350 __clear_bit(msr, msr_bitmap + 0xc00 / f);
4355 static void __vmx_enable_intercept_for_msr(unsigned long *msr_bitmap,
4358 int f = sizeof(unsigned long);
4360 if (!cpu_has_vmx_msr_bitmap())
4364 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
4365 * have the write-low and read-high bitmap offsets the wrong way round.
4366 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
4368 if (msr <= 0x1fff) {
4369 if (type & MSR_TYPE_R)
4371 __set_bit(msr, msr_bitmap + 0x000 / f);
4373 if (type & MSR_TYPE_W)
4375 __set_bit(msr, msr_bitmap + 0x800 / f);
4377 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
4379 if (type & MSR_TYPE_R)
4381 __set_bit(msr, msr_bitmap + 0x400 / f);
4383 if (type & MSR_TYPE_W)
4385 __set_bit(msr, msr_bitmap + 0xc00 / f);
4391 * If a msr is allowed by L0, we should check whether it is allowed by L1.
4392 * The corresponding bit will be cleared unless both of L0 and L1 allow it.
4394 static void nested_vmx_disable_intercept_for_msr(unsigned long *msr_bitmap_l1,
4395 unsigned long *msr_bitmap_nested,
4398 int f = sizeof(unsigned long);
4400 if (!cpu_has_vmx_msr_bitmap()) {
4406 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
4407 * have the write-low and read-high bitmap offsets the wrong way round.
4408 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
4410 if (msr <= 0x1fff) {
4411 if (type & MSR_TYPE_R &&
4412 !test_bit(msr, msr_bitmap_l1 + 0x000 / f))
4414 __clear_bit(msr, msr_bitmap_nested + 0x000 / f);
4416 if (type & MSR_TYPE_W &&
4417 !test_bit(msr, msr_bitmap_l1 + 0x800 / f))
4419 __clear_bit(msr, msr_bitmap_nested + 0x800 / f);
4421 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
4423 if (type & MSR_TYPE_R &&
4424 !test_bit(msr, msr_bitmap_l1 + 0x400 / f))
4426 __clear_bit(msr, msr_bitmap_nested + 0x400 / f);
4428 if (type & MSR_TYPE_W &&
4429 !test_bit(msr, msr_bitmap_l1 + 0xc00 / f))
4431 __clear_bit(msr, msr_bitmap_nested + 0xc00 / f);
4436 static void vmx_disable_intercept_for_msr(u32 msr, bool longmode_only)
4439 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy,
4440 msr, MSR_TYPE_R | MSR_TYPE_W);
4441 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode,
4442 msr, MSR_TYPE_R | MSR_TYPE_W);
4445 static void vmx_enable_intercept_msr_read_x2apic(u32 msr)
4447 __vmx_enable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic,
4449 __vmx_enable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic,
4453 static void vmx_disable_intercept_msr_read_x2apic(u32 msr)
4455 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic,
4457 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic,
4461 static void vmx_disable_intercept_msr_write_x2apic(u32 msr)
4463 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic,
4465 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic,
4469 static int vmx_cpu_uses_apicv(struct kvm_vcpu *vcpu)
4471 return enable_apicv && lapic_in_kernel(vcpu);
4474 static int vmx_complete_nested_posted_interrupt(struct kvm_vcpu *vcpu)
4476 struct vcpu_vmx *vmx = to_vmx(vcpu);
4481 if (vmx->nested.pi_desc &&
4482 vmx->nested.pi_pending) {
4483 vmx->nested.pi_pending = false;
4484 if (!pi_test_and_clear_on(vmx->nested.pi_desc))
4487 max_irr = find_last_bit(
4488 (unsigned long *)vmx->nested.pi_desc->pir, 256);
4493 vapic_page = kmap(vmx->nested.virtual_apic_page);
4498 __kvm_apic_update_irr(vmx->nested.pi_desc->pir, vapic_page);
4499 kunmap(vmx->nested.virtual_apic_page);
4501 status = vmcs_read16(GUEST_INTR_STATUS);
4502 if ((u8)max_irr > ((u8)status & 0xff)) {
4504 status |= (u8)max_irr;
4505 vmcs_write16(GUEST_INTR_STATUS, status);
4511 static inline bool kvm_vcpu_trigger_posted_interrupt(struct kvm_vcpu *vcpu)
4514 if (vcpu->mode == IN_GUEST_MODE) {
4515 struct vcpu_vmx *vmx = to_vmx(vcpu);
4518 * Currently, we don't support urgent interrupt,
4519 * all interrupts are recognized as non-urgent
4520 * interrupt, so we cannot post interrupts when
4523 * If the vcpu is in guest mode, it means it is
4524 * running instead of being scheduled out and
4525 * waiting in the run queue, and that's the only
4526 * case when 'SN' is set currently, warning if
4529 WARN_ON_ONCE(pi_test_sn(&vmx->pi_desc));
4531 apic->send_IPI_mask(get_cpu_mask(vcpu->cpu),
4532 POSTED_INTR_VECTOR);
4539 static int vmx_deliver_nested_posted_interrupt(struct kvm_vcpu *vcpu,
4542 struct vcpu_vmx *vmx = to_vmx(vcpu);
4544 if (is_guest_mode(vcpu) &&
4545 vector == vmx->nested.posted_intr_nv) {
4546 /* the PIR and ON have been set by L1. */
4547 kvm_vcpu_trigger_posted_interrupt(vcpu);
4549 * If a posted intr is not recognized by hardware,
4550 * we will accomplish it in the next vmentry.
4552 vmx->nested.pi_pending = true;
4553 kvm_make_request(KVM_REQ_EVENT, vcpu);
4559 * Send interrupt to vcpu via posted interrupt way.
4560 * 1. If target vcpu is running(non-root mode), send posted interrupt
4561 * notification to vcpu and hardware will sync PIR to vIRR atomically.
4562 * 2. If target vcpu isn't running(root mode), kick it to pick up the
4563 * interrupt from PIR in next vmentry.
4565 static void vmx_deliver_posted_interrupt(struct kvm_vcpu *vcpu, int vector)
4567 struct vcpu_vmx *vmx = to_vmx(vcpu);
4570 r = vmx_deliver_nested_posted_interrupt(vcpu, vector);
4574 if (pi_test_and_set_pir(vector, &vmx->pi_desc))
4577 r = pi_test_and_set_on(&vmx->pi_desc);
4578 kvm_make_request(KVM_REQ_EVENT, vcpu);
4579 if (r || !kvm_vcpu_trigger_posted_interrupt(vcpu))
4580 kvm_vcpu_kick(vcpu);
4583 static void vmx_sync_pir_to_irr(struct kvm_vcpu *vcpu)
4585 struct vcpu_vmx *vmx = to_vmx(vcpu);
4587 if (!pi_test_and_clear_on(&vmx->pi_desc))
4590 kvm_apic_update_irr(vcpu, vmx->pi_desc.pir);
4593 static void vmx_sync_pir_to_irr_dummy(struct kvm_vcpu *vcpu)
4599 * Set up the vmcs's constant host-state fields, i.e., host-state fields that
4600 * will not change in the lifetime of the guest.
4601 * Note that host-state that does change is set elsewhere. E.g., host-state
4602 * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
4604 static void vmx_set_constant_host_state(struct vcpu_vmx *vmx)
4611 vmcs_writel(HOST_CR0, read_cr0() & ~X86_CR0_TS); /* 22.2.3 */
4612 vmcs_writel(HOST_CR3, read_cr3()); /* 22.2.3 FIXME: shadow tables */
4614 /* Save the most likely value for this task's CR4 in the VMCS. */
4615 cr4 = cr4_read_shadow();
4616 vmcs_writel(HOST_CR4, cr4); /* 22.2.3, 22.2.5 */
4617 vmx->host_state.vmcs_host_cr4 = cr4;
4619 vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */
4620 #ifdef CONFIG_X86_64
4622 * Load null selectors, so we can avoid reloading them in
4623 * __vmx_load_host_state(), in case userspace uses the null selectors
4624 * too (the expected case).
4626 vmcs_write16(HOST_DS_SELECTOR, 0);
4627 vmcs_write16(HOST_ES_SELECTOR, 0);
4629 vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
4630 vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS); /* 22.2.4 */
4632 vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
4633 vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8); /* 22.2.4 */
4635 native_store_idt(&dt);
4636 vmcs_writel(HOST_IDTR_BASE, dt.address); /* 22.2.4 */
4637 vmx->host_idt_base = dt.address;
4639 vmcs_writel(HOST_RIP, vmx_return); /* 22.2.5 */
4641 rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
4642 vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
4643 rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
4644 vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl); /* 22.2.3 */
4646 if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
4647 rdmsr(MSR_IA32_CR_PAT, low32, high32);
4648 vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
4652 static void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
4654 vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
4656 vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
4657 if (is_guest_mode(&vmx->vcpu))
4658 vmx->vcpu.arch.cr4_guest_owned_bits &=
4659 ~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask;
4660 vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
4663 static u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx)
4665 u32 pin_based_exec_ctrl = vmcs_config.pin_based_exec_ctrl;
4667 if (!vmx_cpu_uses_apicv(&vmx->vcpu))
4668 pin_based_exec_ctrl &= ~PIN_BASED_POSTED_INTR;
4669 return pin_based_exec_ctrl;
4672 static u32 vmx_exec_control(struct vcpu_vmx *vmx)
4674 u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
4676 if (vmx->vcpu.arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)
4677 exec_control &= ~CPU_BASED_MOV_DR_EXITING;
4679 if (!cpu_need_tpr_shadow(&vmx->vcpu)) {
4680 exec_control &= ~CPU_BASED_TPR_SHADOW;
4681 #ifdef CONFIG_X86_64
4682 exec_control |= CPU_BASED_CR8_STORE_EXITING |
4683 CPU_BASED_CR8_LOAD_EXITING;
4687 exec_control |= CPU_BASED_CR3_STORE_EXITING |
4688 CPU_BASED_CR3_LOAD_EXITING |
4689 CPU_BASED_INVLPG_EXITING;
4690 return exec_control;
4693 static u32 vmx_secondary_exec_control(struct vcpu_vmx *vmx)
4695 u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
4696 if (!cpu_need_virtualize_apic_accesses(&vmx->vcpu))
4697 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
4699 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
4701 exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
4702 enable_unrestricted_guest = 0;
4703 /* Enable INVPCID for non-ept guests may cause performance regression. */
4704 exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
4706 if (!enable_unrestricted_guest)
4707 exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
4709 exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
4710 if (!vmx_cpu_uses_apicv(&vmx->vcpu))
4711 exec_control &= ~(SECONDARY_EXEC_APIC_REGISTER_VIRT |
4712 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4713 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
4714 /* SECONDARY_EXEC_SHADOW_VMCS is enabled when L1 executes VMPTRLD
4716 We can NOT enable shadow_vmcs here because we don't have yet
4719 exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
4722 exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
4724 /* Currently, we allow L1 guest to directly run pcommit instruction. */
4725 exec_control &= ~SECONDARY_EXEC_PCOMMIT;
4727 return exec_control;
4730 static void ept_set_mmio_spte_mask(void)
4733 * EPT Misconfigurations can be generated if the value of bits 2:0
4734 * of an EPT paging-structure entry is 110b (write/execute).
4735 * Also, magic bits (0x3ull << 62) is set to quickly identify mmio
4738 kvm_mmu_set_mmio_spte_mask((0x3ull << 62) | 0x6ull);
4741 #define VMX_XSS_EXIT_BITMAP 0
4743 * Sets up the vmcs for emulated real mode.
4745 static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
4747 #ifdef CONFIG_X86_64
4753 vmcs_write64(IO_BITMAP_A, __pa(vmx_io_bitmap_a));
4754 vmcs_write64(IO_BITMAP_B, __pa(vmx_io_bitmap_b));
4756 if (enable_shadow_vmcs) {
4757 vmcs_write64(VMREAD_BITMAP, __pa(vmx_vmread_bitmap));
4758 vmcs_write64(VMWRITE_BITMAP, __pa(vmx_vmwrite_bitmap));
4760 if (cpu_has_vmx_msr_bitmap())
4761 vmcs_write64(MSR_BITMAP, __pa(vmx_msr_bitmap_legacy));
4763 vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
4766 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, vmx_pin_based_exec_ctrl(vmx));
4768 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, vmx_exec_control(vmx));
4770 if (cpu_has_secondary_exec_ctrls())
4771 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
4772 vmx_secondary_exec_control(vmx));
4774 if (vmx_cpu_uses_apicv(&vmx->vcpu)) {
4775 vmcs_write64(EOI_EXIT_BITMAP0, 0);
4776 vmcs_write64(EOI_EXIT_BITMAP1, 0);
4777 vmcs_write64(EOI_EXIT_BITMAP2, 0);
4778 vmcs_write64(EOI_EXIT_BITMAP3, 0);
4780 vmcs_write16(GUEST_INTR_STATUS, 0);
4782 vmcs_write64(POSTED_INTR_NV, POSTED_INTR_VECTOR);
4783 vmcs_write64(POSTED_INTR_DESC_ADDR, __pa((&vmx->pi_desc)));
4787 vmcs_write32(PLE_GAP, ple_gap);
4788 vmx->ple_window = ple_window;
4789 vmx->ple_window_dirty = true;
4792 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
4793 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
4794 vmcs_write32(CR3_TARGET_COUNT, 0); /* 22.2.1 */
4796 vmcs_write16(HOST_FS_SELECTOR, 0); /* 22.2.4 */
4797 vmcs_write16(HOST_GS_SELECTOR, 0); /* 22.2.4 */
4798 vmx_set_constant_host_state(vmx);
4799 #ifdef CONFIG_X86_64
4800 rdmsrl(MSR_FS_BASE, a);
4801 vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
4802 rdmsrl(MSR_GS_BASE, a);
4803 vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
4805 vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
4806 vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
4809 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
4810 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
4811 vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host));
4812 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
4813 vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest));
4815 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
4816 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
4818 for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i) {
4819 u32 index = vmx_msr_index[i];
4820 u32 data_low, data_high;
4823 if (rdmsr_safe(index, &data_low, &data_high) < 0)
4825 if (wrmsr_safe(index, data_low, data_high) < 0)
4827 vmx->guest_msrs[j].index = i;
4828 vmx->guest_msrs[j].data = 0;
4829 vmx->guest_msrs[j].mask = -1ull;
4834 vm_exit_controls_init(vmx, vmcs_config.vmexit_ctrl);
4836 /* 22.2.1, 20.8.1 */
4837 vm_entry_controls_init(vmx, vmcs_config.vmentry_ctrl);
4839 vmcs_writel(CR0_GUEST_HOST_MASK, ~0UL);
4840 set_cr4_guest_host_mask(vmx);
4842 if (vmx_xsaves_supported())
4843 vmcs_write64(XSS_EXIT_BITMAP, VMX_XSS_EXIT_BITMAP);
4848 static void vmx_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
4850 struct vcpu_vmx *vmx = to_vmx(vcpu);
4851 struct msr_data apic_base_msr;
4854 vmx->rmode.vm86_active = 0;
4856 vmx->soft_vnmi_blocked = 0;
4858 vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
4859 kvm_set_cr8(vcpu, 0);
4862 apic_base_msr.data = APIC_DEFAULT_PHYS_BASE |
4863 MSR_IA32_APICBASE_ENABLE;
4864 if (kvm_vcpu_is_reset_bsp(vcpu))
4865 apic_base_msr.data |= MSR_IA32_APICBASE_BSP;
4866 apic_base_msr.host_initiated = true;
4867 kvm_set_apic_base(vcpu, &apic_base_msr);
4870 vmx_segment_cache_clear(vmx);
4872 seg_setup(VCPU_SREG_CS);
4873 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
4874 vmcs_write32(GUEST_CS_BASE, 0xffff0000);
4876 seg_setup(VCPU_SREG_DS);
4877 seg_setup(VCPU_SREG_ES);
4878 seg_setup(VCPU_SREG_FS);
4879 seg_setup(VCPU_SREG_GS);
4880 seg_setup(VCPU_SREG_SS);
4882 vmcs_write16(GUEST_TR_SELECTOR, 0);
4883 vmcs_writel(GUEST_TR_BASE, 0);
4884 vmcs_write32(GUEST_TR_LIMIT, 0xffff);
4885 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
4887 vmcs_write16(GUEST_LDTR_SELECTOR, 0);
4888 vmcs_writel(GUEST_LDTR_BASE, 0);
4889 vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
4890 vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
4893 vmcs_write32(GUEST_SYSENTER_CS, 0);
4894 vmcs_writel(GUEST_SYSENTER_ESP, 0);
4895 vmcs_writel(GUEST_SYSENTER_EIP, 0);
4896 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
4899 vmcs_writel(GUEST_RFLAGS, 0x02);
4900 kvm_rip_write(vcpu, 0xfff0);
4902 vmcs_writel(GUEST_GDTR_BASE, 0);
4903 vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
4905 vmcs_writel(GUEST_IDTR_BASE, 0);
4906 vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
4908 vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
4909 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
4910 vmcs_write32(GUEST_PENDING_DBG_EXCEPTIONS, 0);
4914 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */
4916 if (cpu_has_vmx_tpr_shadow() && !init_event) {
4917 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
4918 if (cpu_need_tpr_shadow(vcpu))
4919 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
4920 __pa(vcpu->arch.apic->regs));
4921 vmcs_write32(TPR_THRESHOLD, 0);
4924 kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
4926 if (vmx_cpu_uses_apicv(vcpu))
4927 memset(&vmx->pi_desc, 0, sizeof(struct pi_desc));
4930 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
4932 cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
4933 vmx_set_cr0(vcpu, cr0); /* enter rmode */
4934 vmx->vcpu.arch.cr0 = cr0;
4935 vmx_set_cr4(vcpu, 0);
4936 vmx_set_efer(vcpu, 0);
4937 vmx_fpu_activate(vcpu);
4938 update_exception_bitmap(vcpu);
4940 vpid_sync_context(vmx->vpid);
4944 * In nested virtualization, check if L1 asked to exit on external interrupts.
4945 * For most existing hypervisors, this will always return true.
4947 static bool nested_exit_on_intr(struct kvm_vcpu *vcpu)
4949 return get_vmcs12(vcpu)->pin_based_vm_exec_control &
4950 PIN_BASED_EXT_INTR_MASK;
4954 * In nested virtualization, check if L1 has set
4955 * VM_EXIT_ACK_INTR_ON_EXIT
4957 static bool nested_exit_intr_ack_set(struct kvm_vcpu *vcpu)
4959 return get_vmcs12(vcpu)->vm_exit_controls &
4960 VM_EXIT_ACK_INTR_ON_EXIT;
4963 static bool nested_exit_on_nmi(struct kvm_vcpu *vcpu)
4965 return get_vmcs12(vcpu)->pin_based_vm_exec_control &
4966 PIN_BASED_NMI_EXITING;
4969 static void enable_irq_window(struct kvm_vcpu *vcpu)
4971 u32 cpu_based_vm_exec_control;
4973 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
4974 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
4975 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
4978 static void enable_nmi_window(struct kvm_vcpu *vcpu)
4980 u32 cpu_based_vm_exec_control;
4982 if (!cpu_has_virtual_nmis() ||
4983 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
4984 enable_irq_window(vcpu);
4988 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
4989 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_NMI_PENDING;
4990 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
4993 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
4995 struct vcpu_vmx *vmx = to_vmx(vcpu);
4997 int irq = vcpu->arch.interrupt.nr;
4999 trace_kvm_inj_virq(irq);
5001 ++vcpu->stat.irq_injections;
5002 if (vmx->rmode.vm86_active) {
5004 if (vcpu->arch.interrupt.soft)
5005 inc_eip = vcpu->arch.event_exit_inst_len;
5006 if (kvm_inject_realmode_interrupt(vcpu, irq, inc_eip) != EMULATE_DONE)
5007 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5010 intr = irq | INTR_INFO_VALID_MASK;
5011 if (vcpu->arch.interrupt.soft) {
5012 intr |= INTR_TYPE_SOFT_INTR;
5013 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
5014 vmx->vcpu.arch.event_exit_inst_len);
5016 intr |= INTR_TYPE_EXT_INTR;
5017 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
5020 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
5022 struct vcpu_vmx *vmx = to_vmx(vcpu);
5024 if (is_guest_mode(vcpu))
5027 if (!cpu_has_virtual_nmis()) {
5029 * Tracking the NMI-blocked state in software is built upon
5030 * finding the next open IRQ window. This, in turn, depends on
5031 * well-behaving guests: They have to keep IRQs disabled at
5032 * least as long as the NMI handler runs. Otherwise we may
5033 * cause NMI nesting, maybe breaking the guest. But as this is
5034 * highly unlikely, we can live with the residual risk.
5036 vmx->soft_vnmi_blocked = 1;
5037 vmx->vnmi_blocked_time = 0;
5040 ++vcpu->stat.nmi_injections;
5041 vmx->nmi_known_unmasked = false;
5042 if (vmx->rmode.vm86_active) {
5043 if (kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0) != EMULATE_DONE)
5044 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5047 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
5048 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
5051 static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
5053 if (!cpu_has_virtual_nmis())
5054 return to_vmx(vcpu)->soft_vnmi_blocked;
5055 if (to_vmx(vcpu)->nmi_known_unmasked)
5057 return vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
5060 static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
5062 struct vcpu_vmx *vmx = to_vmx(vcpu);
5064 if (!cpu_has_virtual_nmis()) {
5065 if (vmx->soft_vnmi_blocked != masked) {
5066 vmx->soft_vnmi_blocked = masked;
5067 vmx->vnmi_blocked_time = 0;
5070 vmx->nmi_known_unmasked = !masked;
5072 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
5073 GUEST_INTR_STATE_NMI);
5075 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
5076 GUEST_INTR_STATE_NMI);
5080 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
5082 if (to_vmx(vcpu)->nested.nested_run_pending)
5085 if (!cpu_has_virtual_nmis() && to_vmx(vcpu)->soft_vnmi_blocked)
5088 return !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
5089 (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
5090 | GUEST_INTR_STATE_NMI));
5093 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
5095 return (!to_vmx(vcpu)->nested.nested_run_pending &&
5096 vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
5097 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
5098 (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
5101 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
5105 ret = x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, addr,
5109 kvm->arch.tss_addr = addr;
5110 return init_rmode_tss(kvm);
5113 static bool rmode_exception(struct kvm_vcpu *vcpu, int vec)
5118 * Update instruction length as we may reinject the exception
5119 * from user space while in guest debugging mode.
5121 to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
5122 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
5123 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
5127 if (vcpu->guest_debug &
5128 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
5145 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
5146 int vec, u32 err_code)
5149 * Instruction with address size override prefix opcode 0x67
5150 * Cause the #SS fault with 0 error code in VM86 mode.
5152 if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0) {
5153 if (emulate_instruction(vcpu, 0) == EMULATE_DONE) {
5154 if (vcpu->arch.halt_request) {
5155 vcpu->arch.halt_request = 0;
5156 return kvm_vcpu_halt(vcpu);
5164 * Forward all other exceptions that are valid in real mode.
5165 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
5166 * the required debugging infrastructure rework.
5168 kvm_queue_exception(vcpu, vec);
5173 * Trigger machine check on the host. We assume all the MSRs are already set up
5174 * by the CPU and that we still run on the same CPU as the MCE occurred on.
5175 * We pass a fake environment to the machine check handler because we want
5176 * the guest to be always treated like user space, no matter what context
5177 * it used internally.
5179 static void kvm_machine_check(void)
5181 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
5182 struct pt_regs regs = {
5183 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
5184 .flags = X86_EFLAGS_IF,
5187 do_machine_check(®s, 0);
5191 static int handle_machine_check(struct kvm_vcpu *vcpu)
5193 /* already handled by vcpu_run */
5197 static int handle_exception(struct kvm_vcpu *vcpu)
5199 struct vcpu_vmx *vmx = to_vmx(vcpu);
5200 struct kvm_run *kvm_run = vcpu->run;
5201 u32 intr_info, ex_no, error_code;
5202 unsigned long cr2, rip, dr6;
5204 enum emulation_result er;
5206 vect_info = vmx->idt_vectoring_info;
5207 intr_info = vmx->exit_intr_info;
5209 if (is_machine_check(intr_info))
5210 return handle_machine_check(vcpu);
5212 if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR)
5213 return 1; /* already handled by vmx_vcpu_run() */
5215 if (is_no_device(intr_info)) {
5216 vmx_fpu_activate(vcpu);
5220 if (is_invalid_opcode(intr_info)) {
5221 if (is_guest_mode(vcpu)) {
5222 kvm_queue_exception(vcpu, UD_VECTOR);
5225 er = emulate_instruction(vcpu, EMULTYPE_TRAP_UD);
5226 if (er != EMULATE_DONE)
5227 kvm_queue_exception(vcpu, UD_VECTOR);
5232 if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
5233 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
5236 * The #PF with PFEC.RSVD = 1 indicates the guest is accessing
5237 * MMIO, it is better to report an internal error.
5238 * See the comments in vmx_handle_exit.
5240 if ((vect_info & VECTORING_INFO_VALID_MASK) &&
5241 !(is_page_fault(intr_info) && !(error_code & PFERR_RSVD_MASK))) {
5242 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5243 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
5244 vcpu->run->internal.ndata = 3;
5245 vcpu->run->internal.data[0] = vect_info;
5246 vcpu->run->internal.data[1] = intr_info;
5247 vcpu->run->internal.data[2] = error_code;
5251 if (is_page_fault(intr_info)) {
5252 /* EPT won't cause page fault directly */
5254 cr2 = vmcs_readl(EXIT_QUALIFICATION);
5255 trace_kvm_page_fault(cr2, error_code);
5257 if (kvm_event_needs_reinjection(vcpu))
5258 kvm_mmu_unprotect_page_virt(vcpu, cr2);
5259 return kvm_mmu_page_fault(vcpu, cr2, error_code, NULL, 0);
5262 ex_no = intr_info & INTR_INFO_VECTOR_MASK;
5264 if (vmx->rmode.vm86_active && rmode_exception(vcpu, ex_no))
5265 return handle_rmode_exception(vcpu, ex_no, error_code);
5269 kvm_queue_exception_e(vcpu, AC_VECTOR, error_code);
5272 dr6 = vmcs_readl(EXIT_QUALIFICATION);
5273 if (!(vcpu->guest_debug &
5274 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
5275 vcpu->arch.dr6 &= ~15;
5276 vcpu->arch.dr6 |= dr6 | DR6_RTM;
5277 if (!(dr6 & ~DR6_RESERVED)) /* icebp */
5278 skip_emulated_instruction(vcpu);
5280 kvm_queue_exception(vcpu, DB_VECTOR);
5283 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
5284 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
5288 * Update instruction length as we may reinject #BP from
5289 * user space while in guest debugging mode. Reading it for
5290 * #DB as well causes no harm, it is not used in that case.
5292 vmx->vcpu.arch.event_exit_inst_len =
5293 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
5294 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5295 rip = kvm_rip_read(vcpu);
5296 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
5297 kvm_run->debug.arch.exception = ex_no;
5300 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
5301 kvm_run->ex.exception = ex_no;
5302 kvm_run->ex.error_code = error_code;
5308 static int handle_external_interrupt(struct kvm_vcpu *vcpu)
5310 ++vcpu->stat.irq_exits;
5314 static int handle_triple_fault(struct kvm_vcpu *vcpu)
5316 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
5320 static int handle_io(struct kvm_vcpu *vcpu)
5322 unsigned long exit_qualification;
5323 int size, in, string;
5326 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5327 string = (exit_qualification & 16) != 0;
5328 in = (exit_qualification & 8) != 0;
5330 ++vcpu->stat.io_exits;
5333 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
5335 port = exit_qualification >> 16;
5336 size = (exit_qualification & 7) + 1;
5337 skip_emulated_instruction(vcpu);
5339 return kvm_fast_pio_out(vcpu, size, port);
5343 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
5346 * Patch in the VMCALL instruction:
5348 hypercall[0] = 0x0f;
5349 hypercall[1] = 0x01;
5350 hypercall[2] = 0xc1;
5353 static bool nested_cr0_valid(struct kvm_vcpu *vcpu, unsigned long val)
5355 unsigned long always_on = VMXON_CR0_ALWAYSON;
5356 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5358 if (to_vmx(vcpu)->nested.nested_vmx_secondary_ctls_high &
5359 SECONDARY_EXEC_UNRESTRICTED_GUEST &&
5360 nested_cpu_has2(vmcs12, SECONDARY_EXEC_UNRESTRICTED_GUEST))
5361 always_on &= ~(X86_CR0_PE | X86_CR0_PG);
5362 return (val & always_on) == always_on;
5365 /* called to set cr0 as appropriate for a mov-to-cr0 exit. */
5366 static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
5368 if (is_guest_mode(vcpu)) {
5369 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5370 unsigned long orig_val = val;
5373 * We get here when L2 changed cr0 in a way that did not change
5374 * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
5375 * but did change L0 shadowed bits. So we first calculate the
5376 * effective cr0 value that L1 would like to write into the
5377 * hardware. It consists of the L2-owned bits from the new
5378 * value combined with the L1-owned bits from L1's guest_cr0.
5380 val = (val & ~vmcs12->cr0_guest_host_mask) |
5381 (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask);
5383 if (!nested_cr0_valid(vcpu, val))
5386 if (kvm_set_cr0(vcpu, val))
5388 vmcs_writel(CR0_READ_SHADOW, orig_val);
5391 if (to_vmx(vcpu)->nested.vmxon &&
5392 ((val & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON))
5394 return kvm_set_cr0(vcpu, val);
5398 static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
5400 if (is_guest_mode(vcpu)) {
5401 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5402 unsigned long orig_val = val;
5404 /* analogously to handle_set_cr0 */
5405 val = (val & ~vmcs12->cr4_guest_host_mask) |
5406 (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask);
5407 if (kvm_set_cr4(vcpu, val))
5409 vmcs_writel(CR4_READ_SHADOW, orig_val);
5412 return kvm_set_cr4(vcpu, val);
5415 /* called to set cr0 as approriate for clts instruction exit. */
5416 static void handle_clts(struct kvm_vcpu *vcpu)
5418 if (is_guest_mode(vcpu)) {
5420 * We get here when L2 did CLTS, and L1 didn't shadow CR0.TS
5421 * but we did (!fpu_active). We need to keep GUEST_CR0.TS on,
5422 * just pretend it's off (also in arch.cr0 for fpu_activate).
5424 vmcs_writel(CR0_READ_SHADOW,
5425 vmcs_readl(CR0_READ_SHADOW) & ~X86_CR0_TS);
5426 vcpu->arch.cr0 &= ~X86_CR0_TS;
5428 vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
5431 static int handle_cr(struct kvm_vcpu *vcpu)
5433 unsigned long exit_qualification, val;
5438 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5439 cr = exit_qualification & 15;
5440 reg = (exit_qualification >> 8) & 15;
5441 switch ((exit_qualification >> 4) & 3) {
5442 case 0: /* mov to cr */
5443 val = kvm_register_readl(vcpu, reg);
5444 trace_kvm_cr_write(cr, val);
5447 err = handle_set_cr0(vcpu, val);
5448 kvm_complete_insn_gp(vcpu, err);
5451 err = kvm_set_cr3(vcpu, val);
5452 kvm_complete_insn_gp(vcpu, err);
5455 err = handle_set_cr4(vcpu, val);
5456 kvm_complete_insn_gp(vcpu, err);
5459 u8 cr8_prev = kvm_get_cr8(vcpu);
5461 err = kvm_set_cr8(vcpu, cr8);
5462 kvm_complete_insn_gp(vcpu, err);
5463 if (lapic_in_kernel(vcpu))
5465 if (cr8_prev <= cr8)
5467 vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
5474 trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
5475 skip_emulated_instruction(vcpu);
5476 vmx_fpu_activate(vcpu);
5478 case 1: /*mov from cr*/
5481 val = kvm_read_cr3(vcpu);
5482 kvm_register_write(vcpu, reg, val);
5483 trace_kvm_cr_read(cr, val);
5484 skip_emulated_instruction(vcpu);
5487 val = kvm_get_cr8(vcpu);
5488 kvm_register_write(vcpu, reg, val);
5489 trace_kvm_cr_read(cr, val);
5490 skip_emulated_instruction(vcpu);
5495 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
5496 trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
5497 kvm_lmsw(vcpu, val);
5499 skip_emulated_instruction(vcpu);
5504 vcpu->run->exit_reason = 0;
5505 vcpu_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
5506 (int)(exit_qualification >> 4) & 3, cr);
5510 static int handle_dr(struct kvm_vcpu *vcpu)
5512 unsigned long exit_qualification;
5515 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5516 dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
5518 /* First, if DR does not exist, trigger UD */
5519 if (!kvm_require_dr(vcpu, dr))
5522 /* Do not handle if the CPL > 0, will trigger GP on re-entry */
5523 if (!kvm_require_cpl(vcpu, 0))
5525 dr7 = vmcs_readl(GUEST_DR7);
5528 * As the vm-exit takes precedence over the debug trap, we
5529 * need to emulate the latter, either for the host or the
5530 * guest debugging itself.
5532 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
5533 vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
5534 vcpu->run->debug.arch.dr7 = dr7;
5535 vcpu->run->debug.arch.pc = kvm_get_linear_rip(vcpu);
5536 vcpu->run->debug.arch.exception = DB_VECTOR;
5537 vcpu->run->exit_reason = KVM_EXIT_DEBUG;
5540 vcpu->arch.dr6 &= ~15;
5541 vcpu->arch.dr6 |= DR6_BD | DR6_RTM;
5542 kvm_queue_exception(vcpu, DB_VECTOR);
5547 if (vcpu->guest_debug == 0) {
5548 u32 cpu_based_vm_exec_control;
5550 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5551 cpu_based_vm_exec_control &= ~CPU_BASED_MOV_DR_EXITING;
5552 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
5555 * No more DR vmexits; force a reload of the debug registers
5556 * and reenter on this instruction. The next vmexit will
5557 * retrieve the full state of the debug registers.
5559 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
5563 reg = DEBUG_REG_ACCESS_REG(exit_qualification);
5564 if (exit_qualification & TYPE_MOV_FROM_DR) {
5567 if (kvm_get_dr(vcpu, dr, &val))
5569 kvm_register_write(vcpu, reg, val);
5571 if (kvm_set_dr(vcpu, dr, kvm_register_readl(vcpu, reg)))
5574 skip_emulated_instruction(vcpu);
5578 static u64 vmx_get_dr6(struct kvm_vcpu *vcpu)
5580 return vcpu->arch.dr6;
5583 static void vmx_set_dr6(struct kvm_vcpu *vcpu, unsigned long val)
5587 static void vmx_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
5589 u32 cpu_based_vm_exec_control;
5591 get_debugreg(vcpu->arch.db[0], 0);
5592 get_debugreg(vcpu->arch.db[1], 1);
5593 get_debugreg(vcpu->arch.db[2], 2);
5594 get_debugreg(vcpu->arch.db[3], 3);
5595 get_debugreg(vcpu->arch.dr6, 6);
5596 vcpu->arch.dr7 = vmcs_readl(GUEST_DR7);
5598 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
5600 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5601 cpu_based_vm_exec_control |= CPU_BASED_MOV_DR_EXITING;
5602 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
5605 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
5607 vmcs_writel(GUEST_DR7, val);
5610 static int handle_cpuid(struct kvm_vcpu *vcpu)
5612 kvm_emulate_cpuid(vcpu);
5616 static int handle_rdmsr(struct kvm_vcpu *vcpu)
5618 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
5619 struct msr_data msr_info;
5621 msr_info.index = ecx;
5622 msr_info.host_initiated = false;
5623 if (vmx_get_msr(vcpu, &msr_info)) {
5624 trace_kvm_msr_read_ex(ecx);
5625 kvm_inject_gp(vcpu, 0);
5629 trace_kvm_msr_read(ecx, msr_info.data);
5631 /* FIXME: handling of bits 32:63 of rax, rdx */
5632 vcpu->arch.regs[VCPU_REGS_RAX] = msr_info.data & -1u;
5633 vcpu->arch.regs[VCPU_REGS_RDX] = (msr_info.data >> 32) & -1u;
5634 skip_emulated_instruction(vcpu);
5638 static int handle_wrmsr(struct kvm_vcpu *vcpu)
5640 struct msr_data msr;
5641 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
5642 u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
5643 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
5647 msr.host_initiated = false;
5648 if (kvm_set_msr(vcpu, &msr) != 0) {
5649 trace_kvm_msr_write_ex(ecx, data);
5650 kvm_inject_gp(vcpu, 0);
5654 trace_kvm_msr_write(ecx, data);
5655 skip_emulated_instruction(vcpu);
5659 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
5661 kvm_make_request(KVM_REQ_EVENT, vcpu);
5665 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
5667 u32 cpu_based_vm_exec_control;
5669 /* clear pending irq */
5670 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5671 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
5672 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
5674 kvm_make_request(KVM_REQ_EVENT, vcpu);
5676 ++vcpu->stat.irq_window_exits;
5680 static int handle_halt(struct kvm_vcpu *vcpu)
5682 return kvm_emulate_halt(vcpu);
5685 static int handle_vmcall(struct kvm_vcpu *vcpu)
5687 kvm_emulate_hypercall(vcpu);
5691 static int handle_invd(struct kvm_vcpu *vcpu)
5693 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
5696 static int handle_invlpg(struct kvm_vcpu *vcpu)
5698 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5700 kvm_mmu_invlpg(vcpu, exit_qualification);
5701 skip_emulated_instruction(vcpu);
5705 static int handle_rdpmc(struct kvm_vcpu *vcpu)
5709 err = kvm_rdpmc(vcpu);
5710 kvm_complete_insn_gp(vcpu, err);
5715 static int handle_wbinvd(struct kvm_vcpu *vcpu)
5717 kvm_emulate_wbinvd(vcpu);
5721 static int handle_xsetbv(struct kvm_vcpu *vcpu)
5723 u64 new_bv = kvm_read_edx_eax(vcpu);
5724 u32 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
5726 if (kvm_set_xcr(vcpu, index, new_bv) == 0)
5727 skip_emulated_instruction(vcpu);
5731 static int handle_xsaves(struct kvm_vcpu *vcpu)
5733 skip_emulated_instruction(vcpu);
5734 WARN(1, "this should never happen\n");
5738 static int handle_xrstors(struct kvm_vcpu *vcpu)
5740 skip_emulated_instruction(vcpu);
5741 WARN(1, "this should never happen\n");
5745 static int handle_apic_access(struct kvm_vcpu *vcpu)
5747 if (likely(fasteoi)) {
5748 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5749 int access_type, offset;
5751 access_type = exit_qualification & APIC_ACCESS_TYPE;
5752 offset = exit_qualification & APIC_ACCESS_OFFSET;
5754 * Sane guest uses MOV to write EOI, with written value
5755 * not cared. So make a short-circuit here by avoiding
5756 * heavy instruction emulation.
5758 if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
5759 (offset == APIC_EOI)) {
5760 kvm_lapic_set_eoi(vcpu);
5761 skip_emulated_instruction(vcpu);
5765 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
5768 static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu)
5770 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5771 int vector = exit_qualification & 0xff;
5773 /* EOI-induced VM exit is trap-like and thus no need to adjust IP */
5774 kvm_apic_set_eoi_accelerated(vcpu, vector);
5778 static int handle_apic_write(struct kvm_vcpu *vcpu)
5780 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5781 u32 offset = exit_qualification & 0xfff;
5783 /* APIC-write VM exit is trap-like and thus no need to adjust IP */
5784 kvm_apic_write_nodecode(vcpu, offset);
5788 static int handle_task_switch(struct kvm_vcpu *vcpu)
5790 struct vcpu_vmx *vmx = to_vmx(vcpu);
5791 unsigned long exit_qualification;
5792 bool has_error_code = false;
5795 int reason, type, idt_v, idt_index;
5797 idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
5798 idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK);
5799 type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
5801 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5803 reason = (u32)exit_qualification >> 30;
5804 if (reason == TASK_SWITCH_GATE && idt_v) {
5806 case INTR_TYPE_NMI_INTR:
5807 vcpu->arch.nmi_injected = false;
5808 vmx_set_nmi_mask(vcpu, true);
5810 case INTR_TYPE_EXT_INTR:
5811 case INTR_TYPE_SOFT_INTR:
5812 kvm_clear_interrupt_queue(vcpu);
5814 case INTR_TYPE_HARD_EXCEPTION:
5815 if (vmx->idt_vectoring_info &
5816 VECTORING_INFO_DELIVER_CODE_MASK) {
5817 has_error_code = true;
5819 vmcs_read32(IDT_VECTORING_ERROR_CODE);
5822 case INTR_TYPE_SOFT_EXCEPTION:
5823 kvm_clear_exception_queue(vcpu);
5829 tss_selector = exit_qualification;
5831 if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
5832 type != INTR_TYPE_EXT_INTR &&
5833 type != INTR_TYPE_NMI_INTR))
5834 skip_emulated_instruction(vcpu);
5836 if (kvm_task_switch(vcpu, tss_selector,
5837 type == INTR_TYPE_SOFT_INTR ? idt_index : -1, reason,
5838 has_error_code, error_code) == EMULATE_FAIL) {
5839 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5840 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5841 vcpu->run->internal.ndata = 0;
5846 * TODO: What about debug traps on tss switch?
5847 * Are we supposed to inject them and update dr6?
5853 static int handle_ept_violation(struct kvm_vcpu *vcpu)
5855 unsigned long exit_qualification;
5860 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5862 gla_validity = (exit_qualification >> 7) & 0x3;
5863 if (gla_validity != 0x3 && gla_validity != 0x1 && gla_validity != 0) {
5864 printk(KERN_ERR "EPT: Handling EPT violation failed!\n");
5865 printk(KERN_ERR "EPT: GPA: 0x%lx, GVA: 0x%lx\n",
5866 (long unsigned int)vmcs_read64(GUEST_PHYSICAL_ADDRESS),
5867 vmcs_readl(GUEST_LINEAR_ADDRESS));
5868 printk(KERN_ERR "EPT: Exit qualification is 0x%lx\n",
5869 (long unsigned int)exit_qualification);
5870 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
5871 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_VIOLATION;
5876 * EPT violation happened while executing iret from NMI,
5877 * "blocked by NMI" bit has to be set before next VM entry.
5878 * There are errata that may cause this bit to not be set:
5881 if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5882 cpu_has_virtual_nmis() &&
5883 (exit_qualification & INTR_INFO_UNBLOCK_NMI))
5884 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, GUEST_INTR_STATE_NMI);
5886 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5887 trace_kvm_page_fault(gpa, exit_qualification);
5889 /* It is a write fault? */
5890 error_code = exit_qualification & PFERR_WRITE_MASK;
5891 /* It is a fetch fault? */
5892 error_code |= (exit_qualification << 2) & PFERR_FETCH_MASK;
5893 /* ept page table is present? */
5894 error_code |= (exit_qualification >> 3) & PFERR_PRESENT_MASK;
5896 vcpu->arch.exit_qualification = exit_qualification;
5898 return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0);
5901 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
5906 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5907 if (!kvm_io_bus_write(vcpu, KVM_FAST_MMIO_BUS, gpa, 0, NULL)) {
5908 skip_emulated_instruction(vcpu);
5909 trace_kvm_fast_mmio(gpa);
5913 ret = handle_mmio_page_fault(vcpu, gpa, true);
5914 if (likely(ret == RET_MMIO_PF_EMULATE))
5915 return x86_emulate_instruction(vcpu, gpa, 0, NULL, 0) ==
5918 if (unlikely(ret == RET_MMIO_PF_INVALID))
5919 return kvm_mmu_page_fault(vcpu, gpa, 0, NULL, 0);
5921 if (unlikely(ret == RET_MMIO_PF_RETRY))
5924 /* It is the real ept misconfig */
5927 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
5928 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_MISCONFIG;
5933 static int handle_nmi_window(struct kvm_vcpu *vcpu)
5935 u32 cpu_based_vm_exec_control;
5937 /* clear pending NMI */
5938 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5939 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
5940 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
5941 ++vcpu->stat.nmi_window_exits;
5942 kvm_make_request(KVM_REQ_EVENT, vcpu);
5947 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
5949 struct vcpu_vmx *vmx = to_vmx(vcpu);
5950 enum emulation_result err = EMULATE_DONE;
5953 bool intr_window_requested;
5954 unsigned count = 130;
5956 cpu_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5957 intr_window_requested = cpu_exec_ctrl & CPU_BASED_VIRTUAL_INTR_PENDING;
5959 while (vmx->emulation_required && count-- != 0) {
5960 if (intr_window_requested && vmx_interrupt_allowed(vcpu))
5961 return handle_interrupt_window(&vmx->vcpu);
5963 if (test_bit(KVM_REQ_EVENT, &vcpu->requests))
5966 err = emulate_instruction(vcpu, EMULTYPE_NO_REEXECUTE);
5968 if (err == EMULATE_USER_EXIT) {
5969 ++vcpu->stat.mmio_exits;
5974 if (err != EMULATE_DONE) {
5975 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5976 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5977 vcpu->run->internal.ndata = 0;
5981 if (vcpu->arch.halt_request) {
5982 vcpu->arch.halt_request = 0;
5983 ret = kvm_vcpu_halt(vcpu);
5987 if (signal_pending(current))
5997 static int __grow_ple_window(int val)
5999 if (ple_window_grow < 1)
6002 val = min(val, ple_window_actual_max);
6004 if (ple_window_grow < ple_window)
6005 val *= ple_window_grow;
6007 val += ple_window_grow;
6012 static int __shrink_ple_window(int val, int modifier, int minimum)
6017 if (modifier < ple_window)
6022 return max(val, minimum);
6025 static void grow_ple_window(struct kvm_vcpu *vcpu)
6027 struct vcpu_vmx *vmx = to_vmx(vcpu);
6028 int old = vmx->ple_window;
6030 vmx->ple_window = __grow_ple_window(old);
6032 if (vmx->ple_window != old)
6033 vmx->ple_window_dirty = true;
6035 trace_kvm_ple_window_grow(vcpu->vcpu_id, vmx->ple_window, old);
6038 static void shrink_ple_window(struct kvm_vcpu *vcpu)
6040 struct vcpu_vmx *vmx = to_vmx(vcpu);
6041 int old = vmx->ple_window;
6043 vmx->ple_window = __shrink_ple_window(old,
6044 ple_window_shrink, ple_window);
6046 if (vmx->ple_window != old)
6047 vmx->ple_window_dirty = true;
6049 trace_kvm_ple_window_shrink(vcpu->vcpu_id, vmx->ple_window, old);
6053 * ple_window_actual_max is computed to be one grow_ple_window() below
6054 * ple_window_max. (See __grow_ple_window for the reason.)
6055 * This prevents overflows, because ple_window_max is int.
6056 * ple_window_max effectively rounded down to a multiple of ple_window_grow in
6058 * ple_window_max is also prevented from setting vmx->ple_window < ple_window.
6060 static void update_ple_window_actual_max(void)
6062 ple_window_actual_max =
6063 __shrink_ple_window(max(ple_window_max, ple_window),
6064 ple_window_grow, INT_MIN);
6068 * Handler for POSTED_INTERRUPT_WAKEUP_VECTOR.
6070 static void wakeup_handler(void)
6072 struct kvm_vcpu *vcpu;
6073 int cpu = smp_processor_id();
6075 spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
6076 list_for_each_entry(vcpu, &per_cpu(blocked_vcpu_on_cpu, cpu),
6077 blocked_vcpu_list) {
6078 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
6080 if (pi_test_on(pi_desc) == 1)
6081 kvm_vcpu_kick(vcpu);
6083 spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
6086 static __init int hardware_setup(void)
6088 int r = -ENOMEM, i, msr;
6090 rdmsrl_safe(MSR_EFER, &host_efer);
6092 for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i)
6093 kvm_define_shared_msr(i, vmx_msr_index[i]);
6095 vmx_io_bitmap_a = (unsigned long *)__get_free_page(GFP_KERNEL);
6096 if (!vmx_io_bitmap_a)
6099 vmx_io_bitmap_b = (unsigned long *)__get_free_page(GFP_KERNEL);
6100 if (!vmx_io_bitmap_b)
6103 vmx_msr_bitmap_legacy = (unsigned long *)__get_free_page(GFP_KERNEL);
6104 if (!vmx_msr_bitmap_legacy)
6107 vmx_msr_bitmap_legacy_x2apic =
6108 (unsigned long *)__get_free_page(GFP_KERNEL);
6109 if (!vmx_msr_bitmap_legacy_x2apic)
6112 vmx_msr_bitmap_longmode = (unsigned long *)__get_free_page(GFP_KERNEL);
6113 if (!vmx_msr_bitmap_longmode)
6116 vmx_msr_bitmap_longmode_x2apic =
6117 (unsigned long *)__get_free_page(GFP_KERNEL);
6118 if (!vmx_msr_bitmap_longmode_x2apic)
6122 vmx_msr_bitmap_nested =
6123 (unsigned long *)__get_free_page(GFP_KERNEL);
6124 if (!vmx_msr_bitmap_nested)
6128 vmx_vmread_bitmap = (unsigned long *)__get_free_page(GFP_KERNEL);
6129 if (!vmx_vmread_bitmap)
6132 vmx_vmwrite_bitmap = (unsigned long *)__get_free_page(GFP_KERNEL);
6133 if (!vmx_vmwrite_bitmap)
6136 memset(vmx_vmread_bitmap, 0xff, PAGE_SIZE);
6137 memset(vmx_vmwrite_bitmap, 0xff, PAGE_SIZE);
6140 * Allow direct access to the PC debug port (it is often used for I/O
6141 * delays, but the vmexits simply slow things down).
6143 memset(vmx_io_bitmap_a, 0xff, PAGE_SIZE);
6144 clear_bit(0x80, vmx_io_bitmap_a);
6146 memset(vmx_io_bitmap_b, 0xff, PAGE_SIZE);
6148 memset(vmx_msr_bitmap_legacy, 0xff, PAGE_SIZE);
6149 memset(vmx_msr_bitmap_longmode, 0xff, PAGE_SIZE);
6151 memset(vmx_msr_bitmap_nested, 0xff, PAGE_SIZE);
6153 if (setup_vmcs_config(&vmcs_config) < 0) {
6158 if (boot_cpu_has(X86_FEATURE_NX))
6159 kvm_enable_efer_bits(EFER_NX);
6161 if (!cpu_has_vmx_vpid())
6163 if (!cpu_has_vmx_shadow_vmcs())
6164 enable_shadow_vmcs = 0;
6165 if (enable_shadow_vmcs)
6166 init_vmcs_shadow_fields();
6168 if (!cpu_has_vmx_ept() ||
6169 !cpu_has_vmx_ept_4levels()) {
6171 enable_unrestricted_guest = 0;
6172 enable_ept_ad_bits = 0;
6175 if (!cpu_has_vmx_ept_ad_bits())
6176 enable_ept_ad_bits = 0;
6178 if (!cpu_has_vmx_unrestricted_guest())
6179 enable_unrestricted_guest = 0;
6181 if (!cpu_has_vmx_flexpriority())
6182 flexpriority_enabled = 0;
6185 * set_apic_access_page_addr() is used to reload apic access
6186 * page upon invalidation. No need to do anything if not
6187 * using the APIC_ACCESS_ADDR VMCS field.
6189 if (!flexpriority_enabled)
6190 kvm_x86_ops->set_apic_access_page_addr = NULL;
6192 if (!cpu_has_vmx_tpr_shadow())
6193 kvm_x86_ops->update_cr8_intercept = NULL;
6195 if (enable_ept && !cpu_has_vmx_ept_2m_page())
6196 kvm_disable_largepages();
6198 if (!cpu_has_vmx_ple())
6201 if (!cpu_has_vmx_apicv())
6204 if (cpu_has_vmx_tsc_scaling()) {
6205 kvm_has_tsc_control = true;
6206 kvm_max_tsc_scaling_ratio = KVM_VMX_TSC_MULTIPLIER_MAX;
6207 kvm_tsc_scaling_ratio_frac_bits = 48;
6211 kvm_x86_ops->update_cr8_intercept = NULL;
6213 kvm_x86_ops->hwapic_irr_update = NULL;
6214 kvm_x86_ops->hwapic_isr_update = NULL;
6215 kvm_x86_ops->deliver_posted_interrupt = NULL;
6216 kvm_x86_ops->sync_pir_to_irr = vmx_sync_pir_to_irr_dummy;
6219 vmx_disable_intercept_for_msr(MSR_FS_BASE, false);
6220 vmx_disable_intercept_for_msr(MSR_GS_BASE, false);
6221 vmx_disable_intercept_for_msr(MSR_KERNEL_GS_BASE, true);
6222 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_CS, false);
6223 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_ESP, false);
6224 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_EIP, false);
6225 vmx_disable_intercept_for_msr(MSR_IA32_BNDCFGS, true);
6227 memcpy(vmx_msr_bitmap_legacy_x2apic,
6228 vmx_msr_bitmap_legacy, PAGE_SIZE);
6229 memcpy(vmx_msr_bitmap_longmode_x2apic,
6230 vmx_msr_bitmap_longmode, PAGE_SIZE);
6232 set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
6235 for (msr = 0x800; msr <= 0x8ff; msr++)
6236 vmx_disable_intercept_msr_read_x2apic(msr);
6238 /* According SDM, in x2apic mode, the whole id reg is used.
6239 * But in KVM, it only use the highest eight bits. Need to
6241 vmx_enable_intercept_msr_read_x2apic(0x802);
6243 vmx_enable_intercept_msr_read_x2apic(0x839);
6245 vmx_disable_intercept_msr_write_x2apic(0x808);
6247 vmx_disable_intercept_msr_write_x2apic(0x80b);
6249 vmx_disable_intercept_msr_write_x2apic(0x83f);
6253 kvm_mmu_set_mask_ptes(0ull,
6254 (enable_ept_ad_bits) ? VMX_EPT_ACCESS_BIT : 0ull,
6255 (enable_ept_ad_bits) ? VMX_EPT_DIRTY_BIT : 0ull,
6256 0ull, VMX_EPT_EXECUTABLE_MASK);
6257 ept_set_mmio_spte_mask();
6262 update_ple_window_actual_max();
6265 * Only enable PML when hardware supports PML feature, and both EPT
6266 * and EPT A/D bit features are enabled -- PML depends on them to work.
6268 if (!enable_ept || !enable_ept_ad_bits || !cpu_has_vmx_pml())
6272 kvm_x86_ops->slot_enable_log_dirty = NULL;
6273 kvm_x86_ops->slot_disable_log_dirty = NULL;
6274 kvm_x86_ops->flush_log_dirty = NULL;
6275 kvm_x86_ops->enable_log_dirty_pt_masked = NULL;
6278 kvm_set_posted_intr_wakeup_handler(wakeup_handler);
6280 return alloc_kvm_area();
6283 free_page((unsigned long)vmx_vmwrite_bitmap);
6285 free_page((unsigned long)vmx_vmread_bitmap);
6288 free_page((unsigned long)vmx_msr_bitmap_nested);
6290 free_page((unsigned long)vmx_msr_bitmap_longmode_x2apic);
6292 free_page((unsigned long)vmx_msr_bitmap_longmode);
6294 free_page((unsigned long)vmx_msr_bitmap_legacy_x2apic);
6296 free_page((unsigned long)vmx_msr_bitmap_legacy);
6298 free_page((unsigned long)vmx_io_bitmap_b);
6300 free_page((unsigned long)vmx_io_bitmap_a);
6305 static __exit void hardware_unsetup(void)
6307 free_page((unsigned long)vmx_msr_bitmap_legacy_x2apic);
6308 free_page((unsigned long)vmx_msr_bitmap_longmode_x2apic);
6309 free_page((unsigned long)vmx_msr_bitmap_legacy);
6310 free_page((unsigned long)vmx_msr_bitmap_longmode);
6311 free_page((unsigned long)vmx_io_bitmap_b);
6312 free_page((unsigned long)vmx_io_bitmap_a);
6313 free_page((unsigned long)vmx_vmwrite_bitmap);
6314 free_page((unsigned long)vmx_vmread_bitmap);
6316 free_page((unsigned long)vmx_msr_bitmap_nested);
6322 * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
6323 * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
6325 static int handle_pause(struct kvm_vcpu *vcpu)
6328 grow_ple_window(vcpu);
6330 skip_emulated_instruction(vcpu);
6331 kvm_vcpu_on_spin(vcpu);
6336 static int handle_nop(struct kvm_vcpu *vcpu)
6338 skip_emulated_instruction(vcpu);
6342 static int handle_mwait(struct kvm_vcpu *vcpu)
6344 printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
6345 return handle_nop(vcpu);
6348 static int handle_monitor_trap(struct kvm_vcpu *vcpu)
6353 static int handle_monitor(struct kvm_vcpu *vcpu)
6355 printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
6356 return handle_nop(vcpu);
6360 * To run an L2 guest, we need a vmcs02 based on the L1-specified vmcs12.
6361 * We could reuse a single VMCS for all the L2 guests, but we also want the
6362 * option to allocate a separate vmcs02 for each separate loaded vmcs12 - this
6363 * allows keeping them loaded on the processor, and in the future will allow
6364 * optimizations where prepare_vmcs02 doesn't need to set all the fields on
6365 * every entry if they never change.
6366 * So we keep, in vmx->nested.vmcs02_pool, a cache of size VMCS02_POOL_SIZE
6367 * (>=0) with a vmcs02 for each recently loaded vmcs12s, most recent first.
6369 * The following functions allocate and free a vmcs02 in this pool.
6372 /* Get a VMCS from the pool to use as vmcs02 for the current vmcs12. */
6373 static struct loaded_vmcs *nested_get_current_vmcs02(struct vcpu_vmx *vmx)
6375 struct vmcs02_list *item;
6376 list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
6377 if (item->vmptr == vmx->nested.current_vmptr) {
6378 list_move(&item->list, &vmx->nested.vmcs02_pool);
6379 return &item->vmcs02;
6382 if (vmx->nested.vmcs02_num >= max(VMCS02_POOL_SIZE, 1)) {
6383 /* Recycle the least recently used VMCS. */
6384 item = list_entry(vmx->nested.vmcs02_pool.prev,
6385 struct vmcs02_list, list);
6386 item->vmptr = vmx->nested.current_vmptr;
6387 list_move(&item->list, &vmx->nested.vmcs02_pool);
6388 return &item->vmcs02;
6391 /* Create a new VMCS */
6392 item = kmalloc(sizeof(struct vmcs02_list), GFP_KERNEL);
6395 item->vmcs02.vmcs = alloc_vmcs();
6396 if (!item->vmcs02.vmcs) {
6400 loaded_vmcs_init(&item->vmcs02);
6401 item->vmptr = vmx->nested.current_vmptr;
6402 list_add(&(item->list), &(vmx->nested.vmcs02_pool));
6403 vmx->nested.vmcs02_num++;
6404 return &item->vmcs02;
6407 /* Free and remove from pool a vmcs02 saved for a vmcs12 (if there is one) */
6408 static void nested_free_vmcs02(struct vcpu_vmx *vmx, gpa_t vmptr)
6410 struct vmcs02_list *item;
6411 list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
6412 if (item->vmptr == vmptr) {
6413 free_loaded_vmcs(&item->vmcs02);
6414 list_del(&item->list);
6416 vmx->nested.vmcs02_num--;
6422 * Free all VMCSs saved for this vcpu, except the one pointed by
6423 * vmx->loaded_vmcs. We must be running L1, so vmx->loaded_vmcs
6424 * must be &vmx->vmcs01.
6426 static void nested_free_all_saved_vmcss(struct vcpu_vmx *vmx)
6428 struct vmcs02_list *item, *n;
6430 WARN_ON(vmx->loaded_vmcs != &vmx->vmcs01);
6431 list_for_each_entry_safe(item, n, &vmx->nested.vmcs02_pool, list) {
6433 * Something will leak if the above WARN triggers. Better than
6436 if (vmx->loaded_vmcs == &item->vmcs02)
6439 free_loaded_vmcs(&item->vmcs02);
6440 list_del(&item->list);
6442 vmx->nested.vmcs02_num--;
6447 * The following 3 functions, nested_vmx_succeed()/failValid()/failInvalid(),
6448 * set the success or error code of an emulated VMX instruction, as specified
6449 * by Vol 2B, VMX Instruction Reference, "Conventions".
6451 static void nested_vmx_succeed(struct kvm_vcpu *vcpu)
6453 vmx_set_rflags(vcpu, vmx_get_rflags(vcpu)
6454 & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
6455 X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF));
6458 static void nested_vmx_failInvalid(struct kvm_vcpu *vcpu)
6460 vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
6461 & ~(X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
6462 X86_EFLAGS_SF | X86_EFLAGS_OF))
6466 static void nested_vmx_failValid(struct kvm_vcpu *vcpu,
6467 u32 vm_instruction_error)
6469 if (to_vmx(vcpu)->nested.current_vmptr == -1ull) {
6471 * failValid writes the error number to the current VMCS, which
6472 * can't be done there isn't a current VMCS.
6474 nested_vmx_failInvalid(vcpu);
6477 vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
6478 & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
6479 X86_EFLAGS_SF | X86_EFLAGS_OF))
6481 get_vmcs12(vcpu)->vm_instruction_error = vm_instruction_error;
6483 * We don't need to force a shadow sync because
6484 * VM_INSTRUCTION_ERROR is not shadowed
6488 static void nested_vmx_abort(struct kvm_vcpu *vcpu, u32 indicator)
6490 /* TODO: not to reset guest simply here. */
6491 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
6492 pr_warn("kvm: nested vmx abort, indicator %d\n", indicator);
6495 static enum hrtimer_restart vmx_preemption_timer_fn(struct hrtimer *timer)
6497 struct vcpu_vmx *vmx =
6498 container_of(timer, struct vcpu_vmx, nested.preemption_timer);
6500 vmx->nested.preemption_timer_expired = true;
6501 kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
6502 kvm_vcpu_kick(&vmx->vcpu);
6504 return HRTIMER_NORESTART;
6508 * Decode the memory-address operand of a vmx instruction, as recorded on an
6509 * exit caused by such an instruction (run by a guest hypervisor).
6510 * On success, returns 0. When the operand is invalid, returns 1 and throws
6513 static int get_vmx_mem_address(struct kvm_vcpu *vcpu,
6514 unsigned long exit_qualification,
6515 u32 vmx_instruction_info, bool wr, gva_t *ret)
6519 struct kvm_segment s;
6522 * According to Vol. 3B, "Information for VM Exits Due to Instruction
6523 * Execution", on an exit, vmx_instruction_info holds most of the
6524 * addressing components of the operand. Only the displacement part
6525 * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
6526 * For how an actual address is calculated from all these components,
6527 * refer to Vol. 1, "Operand Addressing".
6529 int scaling = vmx_instruction_info & 3;
6530 int addr_size = (vmx_instruction_info >> 7) & 7;
6531 bool is_reg = vmx_instruction_info & (1u << 10);
6532 int seg_reg = (vmx_instruction_info >> 15) & 7;
6533 int index_reg = (vmx_instruction_info >> 18) & 0xf;
6534 bool index_is_valid = !(vmx_instruction_info & (1u << 22));
6535 int base_reg = (vmx_instruction_info >> 23) & 0xf;
6536 bool base_is_valid = !(vmx_instruction_info & (1u << 27));
6539 kvm_queue_exception(vcpu, UD_VECTOR);
6543 /* Addr = segment_base + offset */
6544 /* offset = base + [index * scale] + displacement */
6545 off = exit_qualification; /* holds the displacement */
6547 off += kvm_register_read(vcpu, base_reg);
6549 off += kvm_register_read(vcpu, index_reg)<<scaling;
6550 vmx_get_segment(vcpu, &s, seg_reg);
6551 *ret = s.base + off;
6553 if (addr_size == 1) /* 32 bit */
6556 /* Checks for #GP/#SS exceptions. */
6558 if (is_protmode(vcpu)) {
6559 /* Protected mode: apply checks for segment validity in the
6561 * - segment type check (#GP(0) may be thrown)
6562 * - usability check (#GP(0)/#SS(0))
6563 * - limit check (#GP(0)/#SS(0))
6566 /* #GP(0) if the destination operand is located in a
6567 * read-only data segment or any code segment.
6569 exn = ((s.type & 0xa) == 0 || (s.type & 8));
6571 /* #GP(0) if the source operand is located in an
6572 * execute-only code segment
6574 exn = ((s.type & 0xa) == 8);
6577 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
6580 if (is_long_mode(vcpu)) {
6581 /* Long mode: #GP(0)/#SS(0) if the memory address is in a
6582 * non-canonical form. This is an only check for long mode.
6584 exn = is_noncanonical_address(*ret);
6585 } else if (is_protmode(vcpu)) {
6586 /* Protected mode: #GP(0)/#SS(0) if the segment is unusable.
6588 exn = (s.unusable != 0);
6589 /* Protected mode: #GP(0)/#SS(0) if the memory
6590 * operand is outside the segment limit.
6592 exn = exn || (off + sizeof(u64) > s.limit);
6595 kvm_queue_exception_e(vcpu,
6596 seg_reg == VCPU_SREG_SS ?
6597 SS_VECTOR : GP_VECTOR,
6606 * This function performs the various checks including
6607 * - if it's 4KB aligned
6608 * - No bits beyond the physical address width are set
6609 * - Returns 0 on success or else 1
6610 * (Intel SDM Section 30.3)
6612 static int nested_vmx_check_vmptr(struct kvm_vcpu *vcpu, int exit_reason,
6617 struct x86_exception e;
6619 struct vcpu_vmx *vmx = to_vmx(vcpu);
6620 int maxphyaddr = cpuid_maxphyaddr(vcpu);
6622 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
6623 vmcs_read32(VMX_INSTRUCTION_INFO), false, &gva))
6626 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &vmptr,
6627 sizeof(vmptr), &e)) {
6628 kvm_inject_page_fault(vcpu, &e);
6632 switch (exit_reason) {
6633 case EXIT_REASON_VMON:
6636 * The first 4 bytes of VMXON region contain the supported
6637 * VMCS revision identifier
6639 * Note - IA32_VMX_BASIC[48] will never be 1
6640 * for the nested case;
6641 * which replaces physical address width with 32
6644 if (!PAGE_ALIGNED(vmptr) || (vmptr >> maxphyaddr)) {
6645 nested_vmx_failInvalid(vcpu);
6646 skip_emulated_instruction(vcpu);
6650 page = nested_get_page(vcpu, vmptr);
6652 *(u32 *)kmap(page) != VMCS12_REVISION) {
6653 nested_vmx_failInvalid(vcpu);
6655 skip_emulated_instruction(vcpu);
6659 vmx->nested.vmxon_ptr = vmptr;
6661 case EXIT_REASON_VMCLEAR:
6662 if (!PAGE_ALIGNED(vmptr) || (vmptr >> maxphyaddr)) {
6663 nested_vmx_failValid(vcpu,
6664 VMXERR_VMCLEAR_INVALID_ADDRESS);
6665 skip_emulated_instruction(vcpu);
6669 if (vmptr == vmx->nested.vmxon_ptr) {
6670 nested_vmx_failValid(vcpu,
6671 VMXERR_VMCLEAR_VMXON_POINTER);
6672 skip_emulated_instruction(vcpu);
6676 case EXIT_REASON_VMPTRLD:
6677 if (!PAGE_ALIGNED(vmptr) || (vmptr >> maxphyaddr)) {
6678 nested_vmx_failValid(vcpu,
6679 VMXERR_VMPTRLD_INVALID_ADDRESS);
6680 skip_emulated_instruction(vcpu);
6684 if (vmptr == vmx->nested.vmxon_ptr) {
6685 nested_vmx_failValid(vcpu,
6686 VMXERR_VMCLEAR_VMXON_POINTER);
6687 skip_emulated_instruction(vcpu);
6692 return 1; /* shouldn't happen */
6701 * Emulate the VMXON instruction.
6702 * Currently, we just remember that VMX is active, and do not save or even
6703 * inspect the argument to VMXON (the so-called "VMXON pointer") because we
6704 * do not currently need to store anything in that guest-allocated memory
6705 * region. Consequently, VMCLEAR and VMPTRLD also do not verify that the their
6706 * argument is different from the VMXON pointer (which the spec says they do).
6708 static int handle_vmon(struct kvm_vcpu *vcpu)
6710 struct kvm_segment cs;
6711 struct vcpu_vmx *vmx = to_vmx(vcpu);
6712 struct vmcs *shadow_vmcs;
6713 const u64 VMXON_NEEDED_FEATURES = FEATURE_CONTROL_LOCKED
6714 | FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
6716 /* The Intel VMX Instruction Reference lists a bunch of bits that
6717 * are prerequisite to running VMXON, most notably cr4.VMXE must be
6718 * set to 1 (see vmx_set_cr4() for when we allow the guest to set this).
6719 * Otherwise, we should fail with #UD. We test these now:
6721 if (!kvm_read_cr4_bits(vcpu, X86_CR4_VMXE) ||
6722 !kvm_read_cr0_bits(vcpu, X86_CR0_PE) ||
6723 (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
6724 kvm_queue_exception(vcpu, UD_VECTOR);
6728 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
6729 if (is_long_mode(vcpu) && !cs.l) {
6730 kvm_queue_exception(vcpu, UD_VECTOR);
6734 if (vmx_get_cpl(vcpu)) {
6735 kvm_inject_gp(vcpu, 0);
6739 if (nested_vmx_check_vmptr(vcpu, EXIT_REASON_VMON, NULL))
6742 if (vmx->nested.vmxon) {
6743 nested_vmx_failValid(vcpu, VMXERR_VMXON_IN_VMX_ROOT_OPERATION);
6744 skip_emulated_instruction(vcpu);
6748 if ((vmx->nested.msr_ia32_feature_control & VMXON_NEEDED_FEATURES)
6749 != VMXON_NEEDED_FEATURES) {
6750 kvm_inject_gp(vcpu, 0);
6754 if (enable_shadow_vmcs) {
6755 shadow_vmcs = alloc_vmcs();
6758 /* mark vmcs as shadow */
6759 shadow_vmcs->revision_id |= (1u << 31);
6760 /* init shadow vmcs */
6761 vmcs_clear(shadow_vmcs);
6762 vmx->nested.current_shadow_vmcs = shadow_vmcs;
6765 INIT_LIST_HEAD(&(vmx->nested.vmcs02_pool));
6766 vmx->nested.vmcs02_num = 0;
6768 hrtimer_init(&vmx->nested.preemption_timer, CLOCK_MONOTONIC,
6770 vmx->nested.preemption_timer.function = vmx_preemption_timer_fn;
6772 vmx->nested.vmxon = true;
6774 skip_emulated_instruction(vcpu);
6775 nested_vmx_succeed(vcpu);
6780 * Intel's VMX Instruction Reference specifies a common set of prerequisites
6781 * for running VMX instructions (except VMXON, whose prerequisites are
6782 * slightly different). It also specifies what exception to inject otherwise.
6784 static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
6786 struct kvm_segment cs;
6787 struct vcpu_vmx *vmx = to_vmx(vcpu);
6789 if (!vmx->nested.vmxon) {
6790 kvm_queue_exception(vcpu, UD_VECTOR);
6794 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
6795 if ((vmx_get_rflags(vcpu) & X86_EFLAGS_VM) ||
6796 (is_long_mode(vcpu) && !cs.l)) {
6797 kvm_queue_exception(vcpu, UD_VECTOR);
6801 if (vmx_get_cpl(vcpu)) {
6802 kvm_inject_gp(vcpu, 0);
6809 static inline void nested_release_vmcs12(struct vcpu_vmx *vmx)
6811 if (vmx->nested.current_vmptr == -1ull)
6814 /* current_vmptr and current_vmcs12 are always set/reset together */
6815 if (WARN_ON(vmx->nested.current_vmcs12 == NULL))
6818 if (enable_shadow_vmcs) {
6819 /* copy to memory all shadowed fields in case
6820 they were modified */
6821 copy_shadow_to_vmcs12(vmx);
6822 vmx->nested.sync_shadow_vmcs = false;
6823 vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL,
6824 SECONDARY_EXEC_SHADOW_VMCS);
6825 vmcs_write64(VMCS_LINK_POINTER, -1ull);
6827 vmx->nested.posted_intr_nv = -1;
6828 kunmap(vmx->nested.current_vmcs12_page);
6829 nested_release_page(vmx->nested.current_vmcs12_page);
6830 vmx->nested.current_vmptr = -1ull;
6831 vmx->nested.current_vmcs12 = NULL;
6835 * Free whatever needs to be freed from vmx->nested when L1 goes down, or
6836 * just stops using VMX.
6838 static void free_nested(struct vcpu_vmx *vmx)
6840 if (!vmx->nested.vmxon)
6843 vmx->nested.vmxon = false;
6844 free_vpid(vmx->nested.vpid02);
6845 nested_release_vmcs12(vmx);
6846 if (enable_shadow_vmcs)
6847 free_vmcs(vmx->nested.current_shadow_vmcs);
6848 /* Unpin physical memory we referred to in current vmcs02 */
6849 if (vmx->nested.apic_access_page) {
6850 nested_release_page(vmx->nested.apic_access_page);
6851 vmx->nested.apic_access_page = NULL;
6853 if (vmx->nested.virtual_apic_page) {
6854 nested_release_page(vmx->nested.virtual_apic_page);
6855 vmx->nested.virtual_apic_page = NULL;
6857 if (vmx->nested.pi_desc_page) {
6858 kunmap(vmx->nested.pi_desc_page);
6859 nested_release_page(vmx->nested.pi_desc_page);
6860 vmx->nested.pi_desc_page = NULL;
6861 vmx->nested.pi_desc = NULL;
6864 nested_free_all_saved_vmcss(vmx);
6867 /* Emulate the VMXOFF instruction */
6868 static int handle_vmoff(struct kvm_vcpu *vcpu)
6870 if (!nested_vmx_check_permission(vcpu))
6872 free_nested(to_vmx(vcpu));
6873 skip_emulated_instruction(vcpu);
6874 nested_vmx_succeed(vcpu);
6878 /* Emulate the VMCLEAR instruction */
6879 static int handle_vmclear(struct kvm_vcpu *vcpu)
6881 struct vcpu_vmx *vmx = to_vmx(vcpu);
6883 struct vmcs12 *vmcs12;
6886 if (!nested_vmx_check_permission(vcpu))
6889 if (nested_vmx_check_vmptr(vcpu, EXIT_REASON_VMCLEAR, &vmptr))
6892 if (vmptr == vmx->nested.current_vmptr)
6893 nested_release_vmcs12(vmx);
6895 page = nested_get_page(vcpu, vmptr);
6898 * For accurate processor emulation, VMCLEAR beyond available
6899 * physical memory should do nothing at all. However, it is
6900 * possible that a nested vmx bug, not a guest hypervisor bug,
6901 * resulted in this case, so let's shut down before doing any
6904 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
6907 vmcs12 = kmap(page);
6908 vmcs12->launch_state = 0;
6910 nested_release_page(page);
6912 nested_free_vmcs02(vmx, vmptr);
6914 skip_emulated_instruction(vcpu);
6915 nested_vmx_succeed(vcpu);
6919 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch);
6921 /* Emulate the VMLAUNCH instruction */
6922 static int handle_vmlaunch(struct kvm_vcpu *vcpu)
6924 return nested_vmx_run(vcpu, true);
6927 /* Emulate the VMRESUME instruction */
6928 static int handle_vmresume(struct kvm_vcpu *vcpu)
6931 return nested_vmx_run(vcpu, false);
6934 enum vmcs_field_type {
6935 VMCS_FIELD_TYPE_U16 = 0,
6936 VMCS_FIELD_TYPE_U64 = 1,
6937 VMCS_FIELD_TYPE_U32 = 2,
6938 VMCS_FIELD_TYPE_NATURAL_WIDTH = 3
6941 static inline int vmcs_field_type(unsigned long field)
6943 if (0x1 & field) /* the *_HIGH fields are all 32 bit */
6944 return VMCS_FIELD_TYPE_U32;
6945 return (field >> 13) & 0x3 ;
6948 static inline int vmcs_field_readonly(unsigned long field)
6950 return (((field >> 10) & 0x3) == 1);
6954 * Read a vmcs12 field. Since these can have varying lengths and we return
6955 * one type, we chose the biggest type (u64) and zero-extend the return value
6956 * to that size. Note that the caller, handle_vmread, might need to use only
6957 * some of the bits we return here (e.g., on 32-bit guests, only 32 bits of
6958 * 64-bit fields are to be returned).
6960 static inline int vmcs12_read_any(struct kvm_vcpu *vcpu,
6961 unsigned long field, u64 *ret)
6963 short offset = vmcs_field_to_offset(field);
6969 p = ((char *)(get_vmcs12(vcpu))) + offset;
6971 switch (vmcs_field_type(field)) {
6972 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
6973 *ret = *((natural_width *)p);
6975 case VMCS_FIELD_TYPE_U16:
6978 case VMCS_FIELD_TYPE_U32:
6981 case VMCS_FIELD_TYPE_U64:
6991 static inline int vmcs12_write_any(struct kvm_vcpu *vcpu,
6992 unsigned long field, u64 field_value){
6993 short offset = vmcs_field_to_offset(field);
6994 char *p = ((char *) get_vmcs12(vcpu)) + offset;
6998 switch (vmcs_field_type(field)) {
6999 case VMCS_FIELD_TYPE_U16:
7000 *(u16 *)p = field_value;
7002 case VMCS_FIELD_TYPE_U32:
7003 *(u32 *)p = field_value;
7005 case VMCS_FIELD_TYPE_U64:
7006 *(u64 *)p = field_value;
7008 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7009 *(natural_width *)p = field_value;
7018 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx)
7021 unsigned long field;
7023 struct vmcs *shadow_vmcs = vmx->nested.current_shadow_vmcs;
7024 const unsigned long *fields = shadow_read_write_fields;
7025 const int num_fields = max_shadow_read_write_fields;
7029 vmcs_load(shadow_vmcs);
7031 for (i = 0; i < num_fields; i++) {
7033 switch (vmcs_field_type(field)) {
7034 case VMCS_FIELD_TYPE_U16:
7035 field_value = vmcs_read16(field);
7037 case VMCS_FIELD_TYPE_U32:
7038 field_value = vmcs_read32(field);
7040 case VMCS_FIELD_TYPE_U64:
7041 field_value = vmcs_read64(field);
7043 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7044 field_value = vmcs_readl(field);
7050 vmcs12_write_any(&vmx->vcpu, field, field_value);
7053 vmcs_clear(shadow_vmcs);
7054 vmcs_load(vmx->loaded_vmcs->vmcs);
7059 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx)
7061 const unsigned long *fields[] = {
7062 shadow_read_write_fields,
7063 shadow_read_only_fields
7065 const int max_fields[] = {
7066 max_shadow_read_write_fields,
7067 max_shadow_read_only_fields
7070 unsigned long field;
7071 u64 field_value = 0;
7072 struct vmcs *shadow_vmcs = vmx->nested.current_shadow_vmcs;
7074 vmcs_load(shadow_vmcs);
7076 for (q = 0; q < ARRAY_SIZE(fields); q++) {
7077 for (i = 0; i < max_fields[q]; i++) {
7078 field = fields[q][i];
7079 vmcs12_read_any(&vmx->vcpu, field, &field_value);
7081 switch (vmcs_field_type(field)) {
7082 case VMCS_FIELD_TYPE_U16:
7083 vmcs_write16(field, (u16)field_value);
7085 case VMCS_FIELD_TYPE_U32:
7086 vmcs_write32(field, (u32)field_value);
7088 case VMCS_FIELD_TYPE_U64:
7089 vmcs_write64(field, (u64)field_value);
7091 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7092 vmcs_writel(field, (long)field_value);
7101 vmcs_clear(shadow_vmcs);
7102 vmcs_load(vmx->loaded_vmcs->vmcs);
7106 * VMX instructions which assume a current vmcs12 (i.e., that VMPTRLD was
7107 * used before) all generate the same failure when it is missing.
7109 static int nested_vmx_check_vmcs12(struct kvm_vcpu *vcpu)
7111 struct vcpu_vmx *vmx = to_vmx(vcpu);
7112 if (vmx->nested.current_vmptr == -1ull) {
7113 nested_vmx_failInvalid(vcpu);
7114 skip_emulated_instruction(vcpu);
7120 static int handle_vmread(struct kvm_vcpu *vcpu)
7122 unsigned long field;
7124 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7125 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7128 if (!nested_vmx_check_permission(vcpu) ||
7129 !nested_vmx_check_vmcs12(vcpu))
7132 /* Decode instruction info and find the field to read */
7133 field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
7134 /* Read the field, zero-extended to a u64 field_value */
7135 if (vmcs12_read_any(vcpu, field, &field_value) < 0) {
7136 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
7137 skip_emulated_instruction(vcpu);
7141 * Now copy part of this value to register or memory, as requested.
7142 * Note that the number of bits actually copied is 32 or 64 depending
7143 * on the guest's mode (32 or 64 bit), not on the given field's length.
7145 if (vmx_instruction_info & (1u << 10)) {
7146 kvm_register_writel(vcpu, (((vmx_instruction_info) >> 3) & 0xf),
7149 if (get_vmx_mem_address(vcpu, exit_qualification,
7150 vmx_instruction_info, true, &gva))
7152 /* _system ok, as nested_vmx_check_permission verified cpl=0 */
7153 kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, gva,
7154 &field_value, (is_long_mode(vcpu) ? 8 : 4), NULL);
7157 nested_vmx_succeed(vcpu);
7158 skip_emulated_instruction(vcpu);
7163 static int handle_vmwrite(struct kvm_vcpu *vcpu)
7165 unsigned long field;
7167 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7168 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7169 /* The value to write might be 32 or 64 bits, depending on L1's long
7170 * mode, and eventually we need to write that into a field of several
7171 * possible lengths. The code below first zero-extends the value to 64
7172 * bit (field_value), and then copies only the approriate number of
7173 * bits into the vmcs12 field.
7175 u64 field_value = 0;
7176 struct x86_exception e;
7178 if (!nested_vmx_check_permission(vcpu) ||
7179 !nested_vmx_check_vmcs12(vcpu))
7182 if (vmx_instruction_info & (1u << 10))
7183 field_value = kvm_register_readl(vcpu,
7184 (((vmx_instruction_info) >> 3) & 0xf));
7186 if (get_vmx_mem_address(vcpu, exit_qualification,
7187 vmx_instruction_info, false, &gva))
7189 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva,
7190 &field_value, (is_64_bit_mode(vcpu) ? 8 : 4), &e)) {
7191 kvm_inject_page_fault(vcpu, &e);
7197 field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
7198 if (vmcs_field_readonly(field)) {
7199 nested_vmx_failValid(vcpu,
7200 VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
7201 skip_emulated_instruction(vcpu);
7205 if (vmcs12_write_any(vcpu, field, field_value) < 0) {
7206 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
7207 skip_emulated_instruction(vcpu);
7211 nested_vmx_succeed(vcpu);
7212 skip_emulated_instruction(vcpu);
7216 /* Emulate the VMPTRLD instruction */
7217 static int handle_vmptrld(struct kvm_vcpu *vcpu)
7219 struct vcpu_vmx *vmx = to_vmx(vcpu);
7222 if (!nested_vmx_check_permission(vcpu))
7225 if (nested_vmx_check_vmptr(vcpu, EXIT_REASON_VMPTRLD, &vmptr))
7228 if (vmx->nested.current_vmptr != vmptr) {
7229 struct vmcs12 *new_vmcs12;
7231 page = nested_get_page(vcpu, vmptr);
7233 nested_vmx_failInvalid(vcpu);
7234 skip_emulated_instruction(vcpu);
7237 new_vmcs12 = kmap(page);
7238 if (new_vmcs12->revision_id != VMCS12_REVISION) {
7240 nested_release_page_clean(page);
7241 nested_vmx_failValid(vcpu,
7242 VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
7243 skip_emulated_instruction(vcpu);
7247 nested_release_vmcs12(vmx);
7248 vmx->nested.current_vmptr = vmptr;
7249 vmx->nested.current_vmcs12 = new_vmcs12;
7250 vmx->nested.current_vmcs12_page = page;
7251 if (enable_shadow_vmcs) {
7252 vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
7253 SECONDARY_EXEC_SHADOW_VMCS);
7254 vmcs_write64(VMCS_LINK_POINTER,
7255 __pa(vmx->nested.current_shadow_vmcs));
7256 vmx->nested.sync_shadow_vmcs = true;
7260 nested_vmx_succeed(vcpu);
7261 skip_emulated_instruction(vcpu);
7265 /* Emulate the VMPTRST instruction */
7266 static int handle_vmptrst(struct kvm_vcpu *vcpu)
7268 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7269 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7271 struct x86_exception e;
7273 if (!nested_vmx_check_permission(vcpu))
7276 if (get_vmx_mem_address(vcpu, exit_qualification,
7277 vmx_instruction_info, true, &vmcs_gva))
7279 /* ok to use *_system, as nested_vmx_check_permission verified cpl=0 */
7280 if (kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, vmcs_gva,
7281 (void *)&to_vmx(vcpu)->nested.current_vmptr,
7283 kvm_inject_page_fault(vcpu, &e);
7286 nested_vmx_succeed(vcpu);
7287 skip_emulated_instruction(vcpu);
7291 /* Emulate the INVEPT instruction */
7292 static int handle_invept(struct kvm_vcpu *vcpu)
7294 struct vcpu_vmx *vmx = to_vmx(vcpu);
7295 u32 vmx_instruction_info, types;
7298 struct x86_exception e;
7303 if (!(vmx->nested.nested_vmx_secondary_ctls_high &
7304 SECONDARY_EXEC_ENABLE_EPT) ||
7305 !(vmx->nested.nested_vmx_ept_caps & VMX_EPT_INVEPT_BIT)) {
7306 kvm_queue_exception(vcpu, UD_VECTOR);
7310 if (!nested_vmx_check_permission(vcpu))
7313 if (!kvm_read_cr0_bits(vcpu, X86_CR0_PE)) {
7314 kvm_queue_exception(vcpu, UD_VECTOR);
7318 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7319 type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
7321 types = (vmx->nested.nested_vmx_ept_caps >> VMX_EPT_EXTENT_SHIFT) & 6;
7323 if (!(types & (1UL << type))) {
7324 nested_vmx_failValid(vcpu,
7325 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
7329 /* According to the Intel VMX instruction reference, the memory
7330 * operand is read even if it isn't needed (e.g., for type==global)
7332 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
7333 vmx_instruction_info, false, &gva))
7335 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &operand,
7336 sizeof(operand), &e)) {
7337 kvm_inject_page_fault(vcpu, &e);
7342 case VMX_EPT_EXTENT_GLOBAL:
7343 kvm_mmu_sync_roots(vcpu);
7344 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
7345 nested_vmx_succeed(vcpu);
7348 /* Trap single context invalidation invept calls */
7353 skip_emulated_instruction(vcpu);
7357 static int handle_invvpid(struct kvm_vcpu *vcpu)
7359 struct vcpu_vmx *vmx = to_vmx(vcpu);
7360 u32 vmx_instruction_info;
7361 unsigned long type, types;
7363 struct x86_exception e;
7366 if (!(vmx->nested.nested_vmx_secondary_ctls_high &
7367 SECONDARY_EXEC_ENABLE_VPID) ||
7368 !(vmx->nested.nested_vmx_vpid_caps & VMX_VPID_INVVPID_BIT)) {
7369 kvm_queue_exception(vcpu, UD_VECTOR);
7373 if (!nested_vmx_check_permission(vcpu))
7376 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7377 type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
7379 types = (vmx->nested.nested_vmx_vpid_caps >> 8) & 0x7;
7381 if (!(types & (1UL << type))) {
7382 nested_vmx_failValid(vcpu,
7383 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
7387 /* according to the intel vmx instruction reference, the memory
7388 * operand is read even if it isn't needed (e.g., for type==global)
7390 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
7391 vmx_instruction_info, false, &gva))
7393 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &vpid,
7395 kvm_inject_page_fault(vcpu, &e);
7400 case VMX_VPID_EXTENT_ALL_CONTEXT:
7401 __vmx_flush_tlb(vcpu, to_vmx(vcpu)->nested.vpid02);
7402 nested_vmx_succeed(vcpu);
7405 /* Trap single context invalidation invvpid calls */
7410 skip_emulated_instruction(vcpu);
7414 static int handle_pml_full(struct kvm_vcpu *vcpu)
7416 unsigned long exit_qualification;
7418 trace_kvm_pml_full(vcpu->vcpu_id);
7420 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7423 * PML buffer FULL happened while executing iret from NMI,
7424 * "blocked by NMI" bit has to be set before next VM entry.
7426 if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
7427 cpu_has_virtual_nmis() &&
7428 (exit_qualification & INTR_INFO_UNBLOCK_NMI))
7429 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
7430 GUEST_INTR_STATE_NMI);
7433 * PML buffer already flushed at beginning of VMEXIT. Nothing to do
7434 * here.., and there's no userspace involvement needed for PML.
7439 static int handle_pcommit(struct kvm_vcpu *vcpu)
7441 /* we never catch pcommit instruct for L1 guest. */
7447 * The exit handlers return 1 if the exit was handled fully and guest execution
7448 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
7449 * to be done to userspace and return 0.
7451 static int (*const kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
7452 [EXIT_REASON_EXCEPTION_NMI] = handle_exception,
7453 [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt,
7454 [EXIT_REASON_TRIPLE_FAULT] = handle_triple_fault,
7455 [EXIT_REASON_NMI_WINDOW] = handle_nmi_window,
7456 [EXIT_REASON_IO_INSTRUCTION] = handle_io,
7457 [EXIT_REASON_CR_ACCESS] = handle_cr,
7458 [EXIT_REASON_DR_ACCESS] = handle_dr,
7459 [EXIT_REASON_CPUID] = handle_cpuid,
7460 [EXIT_REASON_MSR_READ] = handle_rdmsr,
7461 [EXIT_REASON_MSR_WRITE] = handle_wrmsr,
7462 [EXIT_REASON_PENDING_INTERRUPT] = handle_interrupt_window,
7463 [EXIT_REASON_HLT] = handle_halt,
7464 [EXIT_REASON_INVD] = handle_invd,
7465 [EXIT_REASON_INVLPG] = handle_invlpg,
7466 [EXIT_REASON_RDPMC] = handle_rdpmc,
7467 [EXIT_REASON_VMCALL] = handle_vmcall,
7468 [EXIT_REASON_VMCLEAR] = handle_vmclear,
7469 [EXIT_REASON_VMLAUNCH] = handle_vmlaunch,
7470 [EXIT_REASON_VMPTRLD] = handle_vmptrld,
7471 [EXIT_REASON_VMPTRST] = handle_vmptrst,
7472 [EXIT_REASON_VMREAD] = handle_vmread,
7473 [EXIT_REASON_VMRESUME] = handle_vmresume,
7474 [EXIT_REASON_VMWRITE] = handle_vmwrite,
7475 [EXIT_REASON_VMOFF] = handle_vmoff,
7476 [EXIT_REASON_VMON] = handle_vmon,
7477 [EXIT_REASON_TPR_BELOW_THRESHOLD] = handle_tpr_below_threshold,
7478 [EXIT_REASON_APIC_ACCESS] = handle_apic_access,
7479 [EXIT_REASON_APIC_WRITE] = handle_apic_write,
7480 [EXIT_REASON_EOI_INDUCED] = handle_apic_eoi_induced,
7481 [EXIT_REASON_WBINVD] = handle_wbinvd,
7482 [EXIT_REASON_XSETBV] = handle_xsetbv,
7483 [EXIT_REASON_TASK_SWITCH] = handle_task_switch,
7484 [EXIT_REASON_MCE_DURING_VMENTRY] = handle_machine_check,
7485 [EXIT_REASON_EPT_VIOLATION] = handle_ept_violation,
7486 [EXIT_REASON_EPT_MISCONFIG] = handle_ept_misconfig,
7487 [EXIT_REASON_PAUSE_INSTRUCTION] = handle_pause,
7488 [EXIT_REASON_MWAIT_INSTRUCTION] = handle_mwait,
7489 [EXIT_REASON_MONITOR_TRAP_FLAG] = handle_monitor_trap,
7490 [EXIT_REASON_MONITOR_INSTRUCTION] = handle_monitor,
7491 [EXIT_REASON_INVEPT] = handle_invept,
7492 [EXIT_REASON_INVVPID] = handle_invvpid,
7493 [EXIT_REASON_XSAVES] = handle_xsaves,
7494 [EXIT_REASON_XRSTORS] = handle_xrstors,
7495 [EXIT_REASON_PML_FULL] = handle_pml_full,
7496 [EXIT_REASON_PCOMMIT] = handle_pcommit,
7499 static const int kvm_vmx_max_exit_handlers =
7500 ARRAY_SIZE(kvm_vmx_exit_handlers);
7502 static bool nested_vmx_exit_handled_io(struct kvm_vcpu *vcpu,
7503 struct vmcs12 *vmcs12)
7505 unsigned long exit_qualification;
7506 gpa_t bitmap, last_bitmap;
7511 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
7512 return nested_cpu_has(vmcs12, CPU_BASED_UNCOND_IO_EXITING);
7514 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7516 port = exit_qualification >> 16;
7517 size = (exit_qualification & 7) + 1;
7519 last_bitmap = (gpa_t)-1;
7524 bitmap = vmcs12->io_bitmap_a;
7525 else if (port < 0x10000)
7526 bitmap = vmcs12->io_bitmap_b;
7529 bitmap += (port & 0x7fff) / 8;
7531 if (last_bitmap != bitmap)
7532 if (kvm_vcpu_read_guest(vcpu, bitmap, &b, 1))
7534 if (b & (1 << (port & 7)))
7539 last_bitmap = bitmap;
7546 * Return 1 if we should exit from L2 to L1 to handle an MSR access access,
7547 * rather than handle it ourselves in L0. I.e., check whether L1 expressed
7548 * disinterest in the current event (read or write a specific MSR) by using an
7549 * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps.
7551 static bool nested_vmx_exit_handled_msr(struct kvm_vcpu *vcpu,
7552 struct vmcs12 *vmcs12, u32 exit_reason)
7554 u32 msr_index = vcpu->arch.regs[VCPU_REGS_RCX];
7557 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
7561 * The MSR_BITMAP page is divided into four 1024-byte bitmaps,
7562 * for the four combinations of read/write and low/high MSR numbers.
7563 * First we need to figure out which of the four to use:
7565 bitmap = vmcs12->msr_bitmap;
7566 if (exit_reason == EXIT_REASON_MSR_WRITE)
7568 if (msr_index >= 0xc0000000) {
7569 msr_index -= 0xc0000000;
7573 /* Then read the msr_index'th bit from this bitmap: */
7574 if (msr_index < 1024*8) {
7576 if (kvm_vcpu_read_guest(vcpu, bitmap + msr_index/8, &b, 1))
7578 return 1 & (b >> (msr_index & 7));
7580 return true; /* let L1 handle the wrong parameter */
7584 * Return 1 if we should exit from L2 to L1 to handle a CR access exit,
7585 * rather than handle it ourselves in L0. I.e., check if L1 wanted to
7586 * intercept (via guest_host_mask etc.) the current event.
7588 static bool nested_vmx_exit_handled_cr(struct kvm_vcpu *vcpu,
7589 struct vmcs12 *vmcs12)
7591 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7592 int cr = exit_qualification & 15;
7593 int reg = (exit_qualification >> 8) & 15;
7594 unsigned long val = kvm_register_readl(vcpu, reg);
7596 switch ((exit_qualification >> 4) & 3) {
7597 case 0: /* mov to cr */
7600 if (vmcs12->cr0_guest_host_mask &
7601 (val ^ vmcs12->cr0_read_shadow))
7605 if ((vmcs12->cr3_target_count >= 1 &&
7606 vmcs12->cr3_target_value0 == val) ||
7607 (vmcs12->cr3_target_count >= 2 &&
7608 vmcs12->cr3_target_value1 == val) ||
7609 (vmcs12->cr3_target_count >= 3 &&
7610 vmcs12->cr3_target_value2 == val) ||
7611 (vmcs12->cr3_target_count >= 4 &&
7612 vmcs12->cr3_target_value3 == val))
7614 if (nested_cpu_has(vmcs12, CPU_BASED_CR3_LOAD_EXITING))
7618 if (vmcs12->cr4_guest_host_mask &
7619 (vmcs12->cr4_read_shadow ^ val))
7623 if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING))
7629 if ((vmcs12->cr0_guest_host_mask & X86_CR0_TS) &&
7630 (vmcs12->cr0_read_shadow & X86_CR0_TS))
7633 case 1: /* mov from cr */
7636 if (vmcs12->cpu_based_vm_exec_control &
7637 CPU_BASED_CR3_STORE_EXITING)
7641 if (vmcs12->cpu_based_vm_exec_control &
7642 CPU_BASED_CR8_STORE_EXITING)
7649 * lmsw can change bits 1..3 of cr0, and only set bit 0 of
7650 * cr0. Other attempted changes are ignored, with no exit.
7652 if (vmcs12->cr0_guest_host_mask & 0xe &
7653 (val ^ vmcs12->cr0_read_shadow))
7655 if ((vmcs12->cr0_guest_host_mask & 0x1) &&
7656 !(vmcs12->cr0_read_shadow & 0x1) &&
7665 * Return 1 if we should exit from L2 to L1 to handle an exit, or 0 if we
7666 * should handle it ourselves in L0 (and then continue L2). Only call this
7667 * when in is_guest_mode (L2).
7669 static bool nested_vmx_exit_handled(struct kvm_vcpu *vcpu)
7671 u32 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
7672 struct vcpu_vmx *vmx = to_vmx(vcpu);
7673 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7674 u32 exit_reason = vmx->exit_reason;
7676 trace_kvm_nested_vmexit(kvm_rip_read(vcpu), exit_reason,
7677 vmcs_readl(EXIT_QUALIFICATION),
7678 vmx->idt_vectoring_info,
7680 vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
7683 if (vmx->nested.nested_run_pending)
7686 if (unlikely(vmx->fail)) {
7687 pr_info_ratelimited("%s failed vm entry %x\n", __func__,
7688 vmcs_read32(VM_INSTRUCTION_ERROR));
7692 switch (exit_reason) {
7693 case EXIT_REASON_EXCEPTION_NMI:
7694 if (!is_exception(intr_info))
7696 else if (is_page_fault(intr_info))
7698 else if (is_no_device(intr_info) &&
7699 !(vmcs12->guest_cr0 & X86_CR0_TS))
7701 return vmcs12->exception_bitmap &
7702 (1u << (intr_info & INTR_INFO_VECTOR_MASK));
7703 case EXIT_REASON_EXTERNAL_INTERRUPT:
7705 case EXIT_REASON_TRIPLE_FAULT:
7707 case EXIT_REASON_PENDING_INTERRUPT:
7708 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_INTR_PENDING);
7709 case EXIT_REASON_NMI_WINDOW:
7710 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_NMI_PENDING);
7711 case EXIT_REASON_TASK_SWITCH:
7713 case EXIT_REASON_CPUID:
7714 if (kvm_register_read(vcpu, VCPU_REGS_RAX) == 0xa)
7717 case EXIT_REASON_HLT:
7718 return nested_cpu_has(vmcs12, CPU_BASED_HLT_EXITING);
7719 case EXIT_REASON_INVD:
7721 case EXIT_REASON_INVLPG:
7722 return nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
7723 case EXIT_REASON_RDPMC:
7724 return nested_cpu_has(vmcs12, CPU_BASED_RDPMC_EXITING);
7725 case EXIT_REASON_RDTSC: case EXIT_REASON_RDTSCP:
7726 return nested_cpu_has(vmcs12, CPU_BASED_RDTSC_EXITING);
7727 case EXIT_REASON_VMCALL: case EXIT_REASON_VMCLEAR:
7728 case EXIT_REASON_VMLAUNCH: case EXIT_REASON_VMPTRLD:
7729 case EXIT_REASON_VMPTRST: case EXIT_REASON_VMREAD:
7730 case EXIT_REASON_VMRESUME: case EXIT_REASON_VMWRITE:
7731 case EXIT_REASON_VMOFF: case EXIT_REASON_VMON:
7732 case EXIT_REASON_INVEPT: case EXIT_REASON_INVVPID:
7734 * VMX instructions trap unconditionally. This allows L1 to
7735 * emulate them for its L2 guest, i.e., allows 3-level nesting!
7738 case EXIT_REASON_CR_ACCESS:
7739 return nested_vmx_exit_handled_cr(vcpu, vmcs12);
7740 case EXIT_REASON_DR_ACCESS:
7741 return nested_cpu_has(vmcs12, CPU_BASED_MOV_DR_EXITING);
7742 case EXIT_REASON_IO_INSTRUCTION:
7743 return nested_vmx_exit_handled_io(vcpu, vmcs12);
7744 case EXIT_REASON_MSR_READ:
7745 case EXIT_REASON_MSR_WRITE:
7746 return nested_vmx_exit_handled_msr(vcpu, vmcs12, exit_reason);
7747 case EXIT_REASON_INVALID_STATE:
7749 case EXIT_REASON_MWAIT_INSTRUCTION:
7750 return nested_cpu_has(vmcs12, CPU_BASED_MWAIT_EXITING);
7751 case EXIT_REASON_MONITOR_TRAP_FLAG:
7752 return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_TRAP_FLAG);
7753 case EXIT_REASON_MONITOR_INSTRUCTION:
7754 return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_EXITING);
7755 case EXIT_REASON_PAUSE_INSTRUCTION:
7756 return nested_cpu_has(vmcs12, CPU_BASED_PAUSE_EXITING) ||
7757 nested_cpu_has2(vmcs12,
7758 SECONDARY_EXEC_PAUSE_LOOP_EXITING);
7759 case EXIT_REASON_MCE_DURING_VMENTRY:
7761 case EXIT_REASON_TPR_BELOW_THRESHOLD:
7762 return nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW);
7763 case EXIT_REASON_APIC_ACCESS:
7764 return nested_cpu_has2(vmcs12,
7765 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
7766 case EXIT_REASON_APIC_WRITE:
7767 case EXIT_REASON_EOI_INDUCED:
7768 /* apic_write and eoi_induced should exit unconditionally. */
7770 case EXIT_REASON_EPT_VIOLATION:
7772 * L0 always deals with the EPT violation. If nested EPT is
7773 * used, and the nested mmu code discovers that the address is
7774 * missing in the guest EPT table (EPT12), the EPT violation
7775 * will be injected with nested_ept_inject_page_fault()
7778 case EXIT_REASON_EPT_MISCONFIG:
7780 * L2 never uses directly L1's EPT, but rather L0's own EPT
7781 * table (shadow on EPT) or a merged EPT table that L0 built
7782 * (EPT on EPT). So any problems with the structure of the
7783 * table is L0's fault.
7786 case EXIT_REASON_WBINVD:
7787 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_WBINVD_EXITING);
7788 case EXIT_REASON_XSETBV:
7790 case EXIT_REASON_XSAVES: case EXIT_REASON_XRSTORS:
7792 * This should never happen, since it is not possible to
7793 * set XSS to a non-zero value---neither in L1 nor in L2.
7794 * If if it were, XSS would have to be checked against
7795 * the XSS exit bitmap in vmcs12.
7797 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES);
7798 case EXIT_REASON_PCOMMIT:
7799 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_PCOMMIT);
7805 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
7807 *info1 = vmcs_readl(EXIT_QUALIFICATION);
7808 *info2 = vmcs_read32(VM_EXIT_INTR_INFO);
7811 static int vmx_create_pml_buffer(struct vcpu_vmx *vmx)
7813 struct page *pml_pg;
7815 pml_pg = alloc_page(GFP_KERNEL | __GFP_ZERO);
7819 vmx->pml_pg = pml_pg;
7821 vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
7822 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
7827 static void vmx_destroy_pml_buffer(struct vcpu_vmx *vmx)
7830 __free_page(vmx->pml_pg);
7835 static void vmx_flush_pml_buffer(struct kvm_vcpu *vcpu)
7837 struct vcpu_vmx *vmx = to_vmx(vcpu);
7841 pml_idx = vmcs_read16(GUEST_PML_INDEX);
7843 /* Do nothing if PML buffer is empty */
7844 if (pml_idx == (PML_ENTITY_NUM - 1))
7847 /* PML index always points to next available PML buffer entity */
7848 if (pml_idx >= PML_ENTITY_NUM)
7853 pml_buf = page_address(vmx->pml_pg);
7854 for (; pml_idx < PML_ENTITY_NUM; pml_idx++) {
7857 gpa = pml_buf[pml_idx];
7858 WARN_ON(gpa & (PAGE_SIZE - 1));
7859 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
7862 /* reset PML index */
7863 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
7867 * Flush all vcpus' PML buffer and update logged GPAs to dirty_bitmap.
7868 * Called before reporting dirty_bitmap to userspace.
7870 static void kvm_flush_pml_buffers(struct kvm *kvm)
7873 struct kvm_vcpu *vcpu;
7875 * We only need to kick vcpu out of guest mode here, as PML buffer
7876 * is flushed at beginning of all VMEXITs, and it's obvious that only
7877 * vcpus running in guest are possible to have unflushed GPAs in PML
7880 kvm_for_each_vcpu(i, vcpu, kvm)
7881 kvm_vcpu_kick(vcpu);
7884 static void vmx_dump_sel(char *name, uint32_t sel)
7886 pr_err("%s sel=0x%04x, attr=0x%05x, limit=0x%08x, base=0x%016lx\n",
7887 name, vmcs_read32(sel),
7888 vmcs_read32(sel + GUEST_ES_AR_BYTES - GUEST_ES_SELECTOR),
7889 vmcs_read32(sel + GUEST_ES_LIMIT - GUEST_ES_SELECTOR),
7890 vmcs_readl(sel + GUEST_ES_BASE - GUEST_ES_SELECTOR));
7893 static void vmx_dump_dtsel(char *name, uint32_t limit)
7895 pr_err("%s limit=0x%08x, base=0x%016lx\n",
7896 name, vmcs_read32(limit),
7897 vmcs_readl(limit + GUEST_GDTR_BASE - GUEST_GDTR_LIMIT));
7900 static void dump_vmcs(void)
7902 u32 vmentry_ctl = vmcs_read32(VM_ENTRY_CONTROLS);
7903 u32 vmexit_ctl = vmcs_read32(VM_EXIT_CONTROLS);
7904 u32 cpu_based_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
7905 u32 pin_based_exec_ctrl = vmcs_read32(PIN_BASED_VM_EXEC_CONTROL);
7906 u32 secondary_exec_control = 0;
7907 unsigned long cr4 = vmcs_readl(GUEST_CR4);
7908 u64 efer = vmcs_readl(GUEST_IA32_EFER);
7911 if (cpu_has_secondary_exec_ctrls())
7912 secondary_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
7914 pr_err("*** Guest State ***\n");
7915 pr_err("CR0: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
7916 vmcs_readl(GUEST_CR0), vmcs_readl(CR0_READ_SHADOW),
7917 vmcs_readl(CR0_GUEST_HOST_MASK));
7918 pr_err("CR4: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
7919 cr4, vmcs_readl(CR4_READ_SHADOW), vmcs_readl(CR4_GUEST_HOST_MASK));
7920 pr_err("CR3 = 0x%016lx\n", vmcs_readl(GUEST_CR3));
7921 if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT) &&
7922 (cr4 & X86_CR4_PAE) && !(efer & EFER_LMA))
7924 pr_err("PDPTR0 = 0x%016lx PDPTR1 = 0x%016lx\n",
7925 vmcs_readl(GUEST_PDPTR0), vmcs_readl(GUEST_PDPTR1));
7926 pr_err("PDPTR2 = 0x%016lx PDPTR3 = 0x%016lx\n",
7927 vmcs_readl(GUEST_PDPTR2), vmcs_readl(GUEST_PDPTR3));
7929 pr_err("RSP = 0x%016lx RIP = 0x%016lx\n",
7930 vmcs_readl(GUEST_RSP), vmcs_readl(GUEST_RIP));
7931 pr_err("RFLAGS=0x%08lx DR7 = 0x%016lx\n",
7932 vmcs_readl(GUEST_RFLAGS), vmcs_readl(GUEST_DR7));
7933 pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
7934 vmcs_readl(GUEST_SYSENTER_ESP),
7935 vmcs_read32(GUEST_SYSENTER_CS), vmcs_readl(GUEST_SYSENTER_EIP));
7936 vmx_dump_sel("CS: ", GUEST_CS_SELECTOR);
7937 vmx_dump_sel("DS: ", GUEST_DS_SELECTOR);
7938 vmx_dump_sel("SS: ", GUEST_SS_SELECTOR);
7939 vmx_dump_sel("ES: ", GUEST_ES_SELECTOR);
7940 vmx_dump_sel("FS: ", GUEST_FS_SELECTOR);
7941 vmx_dump_sel("GS: ", GUEST_GS_SELECTOR);
7942 vmx_dump_dtsel("GDTR:", GUEST_GDTR_LIMIT);
7943 vmx_dump_sel("LDTR:", GUEST_LDTR_SELECTOR);
7944 vmx_dump_dtsel("IDTR:", GUEST_IDTR_LIMIT);
7945 vmx_dump_sel("TR: ", GUEST_TR_SELECTOR);
7946 if ((vmexit_ctl & (VM_EXIT_SAVE_IA32_PAT | VM_EXIT_SAVE_IA32_EFER)) ||
7947 (vmentry_ctl & (VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_IA32_EFER)))
7948 pr_err("EFER = 0x%016llx PAT = 0x%016lx\n",
7949 efer, vmcs_readl(GUEST_IA32_PAT));
7950 pr_err("DebugCtl = 0x%016lx DebugExceptions = 0x%016lx\n",
7951 vmcs_readl(GUEST_IA32_DEBUGCTL),
7952 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS));
7953 if (vmentry_ctl & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
7954 pr_err("PerfGlobCtl = 0x%016lx\n",
7955 vmcs_readl(GUEST_IA32_PERF_GLOBAL_CTRL));
7956 if (vmentry_ctl & VM_ENTRY_LOAD_BNDCFGS)
7957 pr_err("BndCfgS = 0x%016lx\n", vmcs_readl(GUEST_BNDCFGS));
7958 pr_err("Interruptibility = %08x ActivityState = %08x\n",
7959 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO),
7960 vmcs_read32(GUEST_ACTIVITY_STATE));
7961 if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
7962 pr_err("InterruptStatus = %04x\n",
7963 vmcs_read16(GUEST_INTR_STATUS));
7965 pr_err("*** Host State ***\n");
7966 pr_err("RIP = 0x%016lx RSP = 0x%016lx\n",
7967 vmcs_readl(HOST_RIP), vmcs_readl(HOST_RSP));
7968 pr_err("CS=%04x SS=%04x DS=%04x ES=%04x FS=%04x GS=%04x TR=%04x\n",
7969 vmcs_read16(HOST_CS_SELECTOR), vmcs_read16(HOST_SS_SELECTOR),
7970 vmcs_read16(HOST_DS_SELECTOR), vmcs_read16(HOST_ES_SELECTOR),
7971 vmcs_read16(HOST_FS_SELECTOR), vmcs_read16(HOST_GS_SELECTOR),
7972 vmcs_read16(HOST_TR_SELECTOR));
7973 pr_err("FSBase=%016lx GSBase=%016lx TRBase=%016lx\n",
7974 vmcs_readl(HOST_FS_BASE), vmcs_readl(HOST_GS_BASE),
7975 vmcs_readl(HOST_TR_BASE));
7976 pr_err("GDTBase=%016lx IDTBase=%016lx\n",
7977 vmcs_readl(HOST_GDTR_BASE), vmcs_readl(HOST_IDTR_BASE));
7978 pr_err("CR0=%016lx CR3=%016lx CR4=%016lx\n",
7979 vmcs_readl(HOST_CR0), vmcs_readl(HOST_CR3),
7980 vmcs_readl(HOST_CR4));
7981 pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
7982 vmcs_readl(HOST_IA32_SYSENTER_ESP),
7983 vmcs_read32(HOST_IA32_SYSENTER_CS),
7984 vmcs_readl(HOST_IA32_SYSENTER_EIP));
7985 if (vmexit_ctl & (VM_EXIT_LOAD_IA32_PAT | VM_EXIT_LOAD_IA32_EFER))
7986 pr_err("EFER = 0x%016lx PAT = 0x%016lx\n",
7987 vmcs_readl(HOST_IA32_EFER), vmcs_readl(HOST_IA32_PAT));
7988 if (vmexit_ctl & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
7989 pr_err("PerfGlobCtl = 0x%016lx\n",
7990 vmcs_readl(HOST_IA32_PERF_GLOBAL_CTRL));
7992 pr_err("*** Control State ***\n");
7993 pr_err("PinBased=%08x CPUBased=%08x SecondaryExec=%08x\n",
7994 pin_based_exec_ctrl, cpu_based_exec_ctrl, secondary_exec_control);
7995 pr_err("EntryControls=%08x ExitControls=%08x\n", vmentry_ctl, vmexit_ctl);
7996 pr_err("ExceptionBitmap=%08x PFECmask=%08x PFECmatch=%08x\n",
7997 vmcs_read32(EXCEPTION_BITMAP),
7998 vmcs_read32(PAGE_FAULT_ERROR_CODE_MASK),
7999 vmcs_read32(PAGE_FAULT_ERROR_CODE_MATCH));
8000 pr_err("VMEntry: intr_info=%08x errcode=%08x ilen=%08x\n",
8001 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
8002 vmcs_read32(VM_ENTRY_EXCEPTION_ERROR_CODE),
8003 vmcs_read32(VM_ENTRY_INSTRUCTION_LEN));
8004 pr_err("VMExit: intr_info=%08x errcode=%08x ilen=%08x\n",
8005 vmcs_read32(VM_EXIT_INTR_INFO),
8006 vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
8007 vmcs_read32(VM_EXIT_INSTRUCTION_LEN));
8008 pr_err(" reason=%08x qualification=%016lx\n",
8009 vmcs_read32(VM_EXIT_REASON), vmcs_readl(EXIT_QUALIFICATION));
8010 pr_err("IDTVectoring: info=%08x errcode=%08x\n",
8011 vmcs_read32(IDT_VECTORING_INFO_FIELD),
8012 vmcs_read32(IDT_VECTORING_ERROR_CODE));
8013 pr_err("TSC Offset = 0x%016lx\n", vmcs_readl(TSC_OFFSET));
8014 if (secondary_exec_control & SECONDARY_EXEC_TSC_SCALING)
8015 pr_err("TSC Multiplier = 0x%016lx\n",
8016 vmcs_readl(TSC_MULTIPLIER));
8017 if (cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW)
8018 pr_err("TPR Threshold = 0x%02x\n", vmcs_read32(TPR_THRESHOLD));
8019 if (pin_based_exec_ctrl & PIN_BASED_POSTED_INTR)
8020 pr_err("PostedIntrVec = 0x%02x\n", vmcs_read16(POSTED_INTR_NV));
8021 if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT))
8022 pr_err("EPT pointer = 0x%016lx\n", vmcs_readl(EPT_POINTER));
8023 n = vmcs_read32(CR3_TARGET_COUNT);
8024 for (i = 0; i + 1 < n; i += 4)
8025 pr_err("CR3 target%u=%016lx target%u=%016lx\n",
8026 i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2),
8027 i + 1, vmcs_readl(CR3_TARGET_VALUE0 + i * 2 + 2));
8029 pr_err("CR3 target%u=%016lx\n",
8030 i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2));
8031 if (secondary_exec_control & SECONDARY_EXEC_PAUSE_LOOP_EXITING)
8032 pr_err("PLE Gap=%08x Window=%08x\n",
8033 vmcs_read32(PLE_GAP), vmcs_read32(PLE_WINDOW));
8034 if (secondary_exec_control & SECONDARY_EXEC_ENABLE_VPID)
8035 pr_err("Virtual processor ID = 0x%04x\n",
8036 vmcs_read16(VIRTUAL_PROCESSOR_ID));
8040 * The guest has exited. See if we can fix it or if we need userspace
8043 static int vmx_handle_exit(struct kvm_vcpu *vcpu)
8045 struct vcpu_vmx *vmx = to_vmx(vcpu);
8046 u32 exit_reason = vmx->exit_reason;
8047 u32 vectoring_info = vmx->idt_vectoring_info;
8049 trace_kvm_exit(exit_reason, vcpu, KVM_ISA_VMX);
8052 * Flush logged GPAs PML buffer, this will make dirty_bitmap more
8053 * updated. Another good is, in kvm_vm_ioctl_get_dirty_log, before
8054 * querying dirty_bitmap, we only need to kick all vcpus out of guest
8055 * mode as if vcpus is in root mode, the PML buffer must has been
8059 vmx_flush_pml_buffer(vcpu);
8061 /* If guest state is invalid, start emulating */
8062 if (vmx->emulation_required)
8063 return handle_invalid_guest_state(vcpu);
8065 if (is_guest_mode(vcpu) && nested_vmx_exit_handled(vcpu)) {
8066 nested_vmx_vmexit(vcpu, exit_reason,
8067 vmcs_read32(VM_EXIT_INTR_INFO),
8068 vmcs_readl(EXIT_QUALIFICATION));
8072 if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
8074 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
8075 vcpu->run->fail_entry.hardware_entry_failure_reason
8080 if (unlikely(vmx->fail)) {
8081 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
8082 vcpu->run->fail_entry.hardware_entry_failure_reason
8083 = vmcs_read32(VM_INSTRUCTION_ERROR);
8089 * Do not try to fix EXIT_REASON_EPT_MISCONFIG if it caused by
8090 * delivery event since it indicates guest is accessing MMIO.
8091 * The vm-exit can be triggered again after return to guest that
8092 * will cause infinite loop.
8094 if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
8095 (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
8096 exit_reason != EXIT_REASON_EPT_VIOLATION &&
8097 exit_reason != EXIT_REASON_TASK_SWITCH)) {
8098 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
8099 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_DELIVERY_EV;
8100 vcpu->run->internal.ndata = 2;
8101 vcpu->run->internal.data[0] = vectoring_info;
8102 vcpu->run->internal.data[1] = exit_reason;
8106 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked &&
8107 !(is_guest_mode(vcpu) && nested_cpu_has_virtual_nmis(
8108 get_vmcs12(vcpu))))) {
8109 if (vmx_interrupt_allowed(vcpu)) {
8110 vmx->soft_vnmi_blocked = 0;
8111 } else if (vmx->vnmi_blocked_time > 1000000000LL &&
8112 vcpu->arch.nmi_pending) {
8114 * This CPU don't support us in finding the end of an
8115 * NMI-blocked window if the guest runs with IRQs
8116 * disabled. So we pull the trigger after 1 s of
8117 * futile waiting, but inform the user about this.
8119 printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
8120 "state on VCPU %d after 1 s timeout\n",
8121 __func__, vcpu->vcpu_id);
8122 vmx->soft_vnmi_blocked = 0;
8126 if (exit_reason < kvm_vmx_max_exit_handlers
8127 && kvm_vmx_exit_handlers[exit_reason])
8128 return kvm_vmx_exit_handlers[exit_reason](vcpu);
8130 WARN_ONCE(1, "vmx: unexpected exit reason 0x%x\n", exit_reason);
8131 kvm_queue_exception(vcpu, UD_VECTOR);
8136 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
8138 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
8140 if (is_guest_mode(vcpu) &&
8141 nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
8144 if (irr == -1 || tpr < irr) {
8145 vmcs_write32(TPR_THRESHOLD, 0);
8149 vmcs_write32(TPR_THRESHOLD, irr);
8152 static void vmx_set_virtual_x2apic_mode(struct kvm_vcpu *vcpu, bool set)
8154 u32 sec_exec_control;
8157 * There is not point to enable virtualize x2apic without enable
8160 if (!cpu_has_vmx_virtualize_x2apic_mode() ||
8161 !vmx_cpu_uses_apicv(vcpu))
8164 if (!cpu_need_tpr_shadow(vcpu))
8167 sec_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
8170 sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
8171 sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
8173 sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
8174 sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
8176 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, sec_exec_control);
8178 vmx_set_msr_bitmap(vcpu);
8181 static void vmx_set_apic_access_page_addr(struct kvm_vcpu *vcpu, hpa_t hpa)
8183 struct vcpu_vmx *vmx = to_vmx(vcpu);
8186 * Currently we do not handle the nested case where L2 has an
8187 * APIC access page of its own; that page is still pinned.
8188 * Hence, we skip the case where the VCPU is in guest mode _and_
8189 * L1 prepared an APIC access page for L2.
8191 * For the case where L1 and L2 share the same APIC access page
8192 * (flexpriority=Y but SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES clear
8193 * in the vmcs12), this function will only update either the vmcs01
8194 * or the vmcs02. If the former, the vmcs02 will be updated by
8195 * prepare_vmcs02. If the latter, the vmcs01 will be updated in
8196 * the next L2->L1 exit.
8198 if (!is_guest_mode(vcpu) ||
8199 !nested_cpu_has2(vmx->nested.current_vmcs12,
8200 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
8201 vmcs_write64(APIC_ACCESS_ADDR, hpa);
8204 static void vmx_hwapic_isr_update(struct kvm *kvm, int isr)
8212 status = vmcs_read16(GUEST_INTR_STATUS);
8217 vmcs_write16(GUEST_INTR_STATUS, status);
8221 static void vmx_set_rvi(int vector)
8229 status = vmcs_read16(GUEST_INTR_STATUS);
8230 old = (u8)status & 0xff;
8231 if ((u8)vector != old) {
8233 status |= (u8)vector;
8234 vmcs_write16(GUEST_INTR_STATUS, status);
8238 static void vmx_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
8240 if (!is_guest_mode(vcpu)) {
8241 vmx_set_rvi(max_irr);
8249 * In guest mode. If a vmexit is needed, vmx_check_nested_events
8252 if (nested_exit_on_intr(vcpu))
8256 * Else, fall back to pre-APICv interrupt injection since L2
8257 * is run without virtual interrupt delivery.
8259 if (!kvm_event_needs_reinjection(vcpu) &&
8260 vmx_interrupt_allowed(vcpu)) {
8261 kvm_queue_interrupt(vcpu, max_irr, false);
8262 vmx_inject_irq(vcpu);
8266 static void vmx_load_eoi_exitmap(struct kvm_vcpu *vcpu)
8268 u64 *eoi_exit_bitmap = vcpu->arch.eoi_exit_bitmap;
8269 if (!vmx_cpu_uses_apicv(vcpu))
8272 vmcs_write64(EOI_EXIT_BITMAP0, eoi_exit_bitmap[0]);
8273 vmcs_write64(EOI_EXIT_BITMAP1, eoi_exit_bitmap[1]);
8274 vmcs_write64(EOI_EXIT_BITMAP2, eoi_exit_bitmap[2]);
8275 vmcs_write64(EOI_EXIT_BITMAP3, eoi_exit_bitmap[3]);
8278 static void vmx_complete_atomic_exit(struct vcpu_vmx *vmx)
8282 if (!(vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY
8283 || vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI))
8286 vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8287 exit_intr_info = vmx->exit_intr_info;
8289 /* Handle machine checks before interrupts are enabled */
8290 if (is_machine_check(exit_intr_info))
8291 kvm_machine_check();
8293 /* We need to handle NMIs before interrupts are enabled */
8294 if ((exit_intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR &&
8295 (exit_intr_info & INTR_INFO_VALID_MASK)) {
8296 kvm_before_handle_nmi(&vmx->vcpu);
8298 kvm_after_handle_nmi(&vmx->vcpu);
8302 static void vmx_handle_external_intr(struct kvm_vcpu *vcpu)
8304 u32 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8307 * If external interrupt exists, IF bit is set in rflags/eflags on the
8308 * interrupt stack frame, and interrupt will be enabled on a return
8309 * from interrupt handler.
8311 if ((exit_intr_info & (INTR_INFO_VALID_MASK | INTR_INFO_INTR_TYPE_MASK))
8312 == (INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR)) {
8313 unsigned int vector;
8314 unsigned long entry;
8316 struct vcpu_vmx *vmx = to_vmx(vcpu);
8317 #ifdef CONFIG_X86_64
8321 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
8322 desc = (gate_desc *)vmx->host_idt_base + vector;
8323 entry = gate_offset(*desc);
8325 #ifdef CONFIG_X86_64
8326 "mov %%" _ASM_SP ", %[sp]\n\t"
8327 "and $0xfffffffffffffff0, %%" _ASM_SP "\n\t"
8332 "orl $0x200, (%%" _ASM_SP ")\n\t"
8333 __ASM_SIZE(push) " $%c[cs]\n\t"
8334 "call *%[entry]\n\t"
8336 #ifdef CONFIG_X86_64
8341 [ss]"i"(__KERNEL_DS),
8342 [cs]"i"(__KERNEL_CS)
8348 static bool vmx_has_high_real_mode_segbase(void)
8350 return enable_unrestricted_guest || emulate_invalid_guest_state;
8353 static bool vmx_mpx_supported(void)
8355 return (vmcs_config.vmexit_ctrl & VM_EXIT_CLEAR_BNDCFGS) &&
8356 (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_BNDCFGS);
8359 static bool vmx_xsaves_supported(void)
8361 return vmcs_config.cpu_based_2nd_exec_ctrl &
8362 SECONDARY_EXEC_XSAVES;
8365 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
8370 bool idtv_info_valid;
8372 idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
8374 if (cpu_has_virtual_nmis()) {
8375 if (vmx->nmi_known_unmasked)
8378 * Can't use vmx->exit_intr_info since we're not sure what
8379 * the exit reason is.
8381 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8382 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
8383 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
8385 * SDM 3: 27.7.1.2 (September 2008)
8386 * Re-set bit "block by NMI" before VM entry if vmexit caused by
8387 * a guest IRET fault.
8388 * SDM 3: 23.2.2 (September 2008)
8389 * Bit 12 is undefined in any of the following cases:
8390 * If the VM exit sets the valid bit in the IDT-vectoring
8391 * information field.
8392 * If the VM exit is due to a double fault.
8394 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
8395 vector != DF_VECTOR && !idtv_info_valid)
8396 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
8397 GUEST_INTR_STATE_NMI);
8399 vmx->nmi_known_unmasked =
8400 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
8401 & GUEST_INTR_STATE_NMI);
8402 } else if (unlikely(vmx->soft_vnmi_blocked))
8403 vmx->vnmi_blocked_time +=
8404 ktime_to_ns(ktime_sub(ktime_get(), vmx->entry_time));
8407 static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu,
8408 u32 idt_vectoring_info,
8409 int instr_len_field,
8410 int error_code_field)
8414 bool idtv_info_valid;
8416 idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
8418 vcpu->arch.nmi_injected = false;
8419 kvm_clear_exception_queue(vcpu);
8420 kvm_clear_interrupt_queue(vcpu);
8422 if (!idtv_info_valid)
8425 kvm_make_request(KVM_REQ_EVENT, vcpu);
8427 vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
8428 type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
8431 case INTR_TYPE_NMI_INTR:
8432 vcpu->arch.nmi_injected = true;
8434 * SDM 3: 27.7.1.2 (September 2008)
8435 * Clear bit "block by NMI" before VM entry if a NMI
8438 vmx_set_nmi_mask(vcpu, false);
8440 case INTR_TYPE_SOFT_EXCEPTION:
8441 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
8443 case INTR_TYPE_HARD_EXCEPTION:
8444 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
8445 u32 err = vmcs_read32(error_code_field);
8446 kvm_requeue_exception_e(vcpu, vector, err);
8448 kvm_requeue_exception(vcpu, vector);
8450 case INTR_TYPE_SOFT_INTR:
8451 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
8453 case INTR_TYPE_EXT_INTR:
8454 kvm_queue_interrupt(vcpu, vector, type == INTR_TYPE_SOFT_INTR);
8461 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
8463 __vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info,
8464 VM_EXIT_INSTRUCTION_LEN,
8465 IDT_VECTORING_ERROR_CODE);
8468 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
8470 __vmx_complete_interrupts(vcpu,
8471 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
8472 VM_ENTRY_INSTRUCTION_LEN,
8473 VM_ENTRY_EXCEPTION_ERROR_CODE);
8475 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
8478 static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
8481 struct perf_guest_switch_msr *msrs;
8483 msrs = perf_guest_get_msrs(&nr_msrs);
8488 for (i = 0; i < nr_msrs; i++)
8489 if (msrs[i].host == msrs[i].guest)
8490 clear_atomic_switch_msr(vmx, msrs[i].msr);
8492 add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
8496 static void __noclone vmx_vcpu_run(struct kvm_vcpu *vcpu)
8498 struct vcpu_vmx *vmx = to_vmx(vcpu);
8499 unsigned long debugctlmsr, cr4;
8501 /* Record the guest's net vcpu time for enforced NMI injections. */
8502 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked))
8503 vmx->entry_time = ktime_get();
8505 /* Don't enter VMX if guest state is invalid, let the exit handler
8506 start emulation until we arrive back to a valid state */
8507 if (vmx->emulation_required)
8510 if (vmx->ple_window_dirty) {
8511 vmx->ple_window_dirty = false;
8512 vmcs_write32(PLE_WINDOW, vmx->ple_window);
8515 if (vmx->nested.sync_shadow_vmcs) {
8516 copy_vmcs12_to_shadow(vmx);
8517 vmx->nested.sync_shadow_vmcs = false;
8520 if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
8521 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
8522 if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
8523 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
8525 cr4 = cr4_read_shadow();
8526 if (unlikely(cr4 != vmx->host_state.vmcs_host_cr4)) {
8527 vmcs_writel(HOST_CR4, cr4);
8528 vmx->host_state.vmcs_host_cr4 = cr4;
8531 /* When single-stepping over STI and MOV SS, we must clear the
8532 * corresponding interruptibility bits in the guest state. Otherwise
8533 * vmentry fails as it then expects bit 14 (BS) in pending debug
8534 * exceptions being set, but that's not correct for the guest debugging
8536 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8537 vmx_set_interrupt_shadow(vcpu, 0);
8539 atomic_switch_perf_msrs(vmx);
8540 debugctlmsr = get_debugctlmsr();
8542 vmx->__launched = vmx->loaded_vmcs->launched;
8544 /* Store host registers */
8545 "push %%" _ASM_DX "; push %%" _ASM_BP ";"
8546 "push %%" _ASM_CX " \n\t" /* placeholder for guest rcx */
8547 "push %%" _ASM_CX " \n\t"
8548 "cmp %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
8550 "mov %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
8551 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
8553 /* Reload cr2 if changed */
8554 "mov %c[cr2](%0), %%" _ASM_AX " \n\t"
8555 "mov %%cr2, %%" _ASM_DX " \n\t"
8556 "cmp %%" _ASM_AX ", %%" _ASM_DX " \n\t"
8558 "mov %%" _ASM_AX", %%cr2 \n\t"
8560 /* Check if vmlaunch of vmresume is needed */
8561 "cmpl $0, %c[launched](%0) \n\t"
8562 /* Load guest registers. Don't clobber flags. */
8563 "mov %c[rax](%0), %%" _ASM_AX " \n\t"
8564 "mov %c[rbx](%0), %%" _ASM_BX " \n\t"
8565 "mov %c[rdx](%0), %%" _ASM_DX " \n\t"
8566 "mov %c[rsi](%0), %%" _ASM_SI " \n\t"
8567 "mov %c[rdi](%0), %%" _ASM_DI " \n\t"
8568 "mov %c[rbp](%0), %%" _ASM_BP " \n\t"
8569 #ifdef CONFIG_X86_64
8570 "mov %c[r8](%0), %%r8 \n\t"
8571 "mov %c[r9](%0), %%r9 \n\t"
8572 "mov %c[r10](%0), %%r10 \n\t"
8573 "mov %c[r11](%0), %%r11 \n\t"
8574 "mov %c[r12](%0), %%r12 \n\t"
8575 "mov %c[r13](%0), %%r13 \n\t"
8576 "mov %c[r14](%0), %%r14 \n\t"
8577 "mov %c[r15](%0), %%r15 \n\t"
8579 "mov %c[rcx](%0), %%" _ASM_CX " \n\t" /* kills %0 (ecx) */
8581 /* Enter guest mode */
8583 __ex(ASM_VMX_VMLAUNCH) "\n\t"
8585 "1: " __ex(ASM_VMX_VMRESUME) "\n\t"
8587 /* Save guest registers, load host registers, keep flags */
8588 "mov %0, %c[wordsize](%%" _ASM_SP ") \n\t"
8590 "mov %%" _ASM_AX ", %c[rax](%0) \n\t"
8591 "mov %%" _ASM_BX ", %c[rbx](%0) \n\t"
8592 __ASM_SIZE(pop) " %c[rcx](%0) \n\t"
8593 "mov %%" _ASM_DX ", %c[rdx](%0) \n\t"
8594 "mov %%" _ASM_SI ", %c[rsi](%0) \n\t"
8595 "mov %%" _ASM_DI ", %c[rdi](%0) \n\t"
8596 "mov %%" _ASM_BP ", %c[rbp](%0) \n\t"
8597 #ifdef CONFIG_X86_64
8598 "mov %%r8, %c[r8](%0) \n\t"
8599 "mov %%r9, %c[r9](%0) \n\t"
8600 "mov %%r10, %c[r10](%0) \n\t"
8601 "mov %%r11, %c[r11](%0) \n\t"
8602 "mov %%r12, %c[r12](%0) \n\t"
8603 "mov %%r13, %c[r13](%0) \n\t"
8604 "mov %%r14, %c[r14](%0) \n\t"
8605 "mov %%r15, %c[r15](%0) \n\t"
8607 "mov %%cr2, %%" _ASM_AX " \n\t"
8608 "mov %%" _ASM_AX ", %c[cr2](%0) \n\t"
8610 "pop %%" _ASM_BP "; pop %%" _ASM_DX " \n\t"
8611 "setbe %c[fail](%0) \n\t"
8612 ".pushsection .rodata \n\t"
8613 ".global vmx_return \n\t"
8614 "vmx_return: " _ASM_PTR " 2b \n\t"
8616 : : "c"(vmx), "d"((unsigned long)HOST_RSP),
8617 [launched]"i"(offsetof(struct vcpu_vmx, __launched)),
8618 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
8619 [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
8620 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
8621 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
8622 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
8623 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
8624 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
8625 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
8626 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
8627 #ifdef CONFIG_X86_64
8628 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
8629 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
8630 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
8631 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
8632 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
8633 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
8634 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
8635 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
8637 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2)),
8638 [wordsize]"i"(sizeof(ulong))
8640 #ifdef CONFIG_X86_64
8641 , "rax", "rbx", "rdi", "rsi"
8642 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
8644 , "eax", "ebx", "edi", "esi"
8648 /* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
8650 update_debugctlmsr(debugctlmsr);
8652 #ifndef CONFIG_X86_64
8654 * The sysexit path does not restore ds/es, so we must set them to
8655 * a reasonable value ourselves.
8657 * We can't defer this to vmx_load_host_state() since that function
8658 * may be executed in interrupt context, which saves and restore segments
8659 * around it, nullifying its effect.
8661 loadsegment(ds, __USER_DS);
8662 loadsegment(es, __USER_DS);
8665 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
8666 | (1 << VCPU_EXREG_RFLAGS)
8667 | (1 << VCPU_EXREG_PDPTR)
8668 | (1 << VCPU_EXREG_SEGMENTS)
8669 | (1 << VCPU_EXREG_CR3));
8670 vcpu->arch.regs_dirty = 0;
8672 vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
8674 vmx->loaded_vmcs->launched = 1;
8676 vmx->exit_reason = vmcs_read32(VM_EXIT_REASON);
8679 * the KVM_REQ_EVENT optimization bit is only on for one entry, and if
8680 * we did not inject a still-pending event to L1 now because of
8681 * nested_run_pending, we need to re-enable this bit.
8683 if (vmx->nested.nested_run_pending)
8684 kvm_make_request(KVM_REQ_EVENT, vcpu);
8686 vmx->nested.nested_run_pending = 0;
8688 vmx_complete_atomic_exit(vmx);
8689 vmx_recover_nmi_blocking(vmx);
8690 vmx_complete_interrupts(vmx);
8693 static void vmx_load_vmcs01(struct kvm_vcpu *vcpu)
8695 struct vcpu_vmx *vmx = to_vmx(vcpu);
8698 if (vmx->loaded_vmcs == &vmx->vmcs01)
8702 vmx->loaded_vmcs = &vmx->vmcs01;
8704 vmx_vcpu_load(vcpu, cpu);
8709 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
8711 struct vcpu_vmx *vmx = to_vmx(vcpu);
8714 vmx_destroy_pml_buffer(vmx);
8715 free_vpid(vmx->vpid);
8716 leave_guest_mode(vcpu);
8717 vmx_load_vmcs01(vcpu);
8719 free_loaded_vmcs(vmx->loaded_vmcs);
8720 kfree(vmx->guest_msrs);
8721 kvm_vcpu_uninit(vcpu);
8722 kmem_cache_free(kvm_vcpu_cache, vmx);
8725 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
8728 struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
8732 return ERR_PTR(-ENOMEM);
8734 vmx->vpid = allocate_vpid();
8736 err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
8740 vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
8741 BUILD_BUG_ON(ARRAY_SIZE(vmx_msr_index) * sizeof(vmx->guest_msrs[0])
8745 if (!vmx->guest_msrs) {
8749 vmx->loaded_vmcs = &vmx->vmcs01;
8750 vmx->loaded_vmcs->vmcs = alloc_vmcs();
8751 if (!vmx->loaded_vmcs->vmcs)
8754 kvm_cpu_vmxon(__pa(per_cpu(vmxarea, raw_smp_processor_id())));
8755 loaded_vmcs_init(vmx->loaded_vmcs);
8760 vmx_vcpu_load(&vmx->vcpu, cpu);
8761 vmx->vcpu.cpu = cpu;
8762 err = vmx_vcpu_setup(vmx);
8763 vmx_vcpu_put(&vmx->vcpu);
8767 if (cpu_need_virtualize_apic_accesses(&vmx->vcpu)) {
8768 err = alloc_apic_access_page(kvm);
8774 if (!kvm->arch.ept_identity_map_addr)
8775 kvm->arch.ept_identity_map_addr =
8776 VMX_EPT_IDENTITY_PAGETABLE_ADDR;
8777 err = init_rmode_identity_map(kvm);
8783 nested_vmx_setup_ctls_msrs(vmx);
8784 vmx->nested.vpid02 = allocate_vpid();
8787 vmx->nested.posted_intr_nv = -1;
8788 vmx->nested.current_vmptr = -1ull;
8789 vmx->nested.current_vmcs12 = NULL;
8792 * If PML is turned on, failure on enabling PML just results in failure
8793 * of creating the vcpu, therefore we can simplify PML logic (by
8794 * avoiding dealing with cases, such as enabling PML partially on vcpus
8795 * for the guest, etc.
8798 err = vmx_create_pml_buffer(vmx);
8806 free_vpid(vmx->nested.vpid02);
8807 free_loaded_vmcs(vmx->loaded_vmcs);
8809 kfree(vmx->guest_msrs);
8811 kvm_vcpu_uninit(&vmx->vcpu);
8813 free_vpid(vmx->vpid);
8814 kmem_cache_free(kvm_vcpu_cache, vmx);
8815 return ERR_PTR(err);
8818 static void __init vmx_check_processor_compat(void *rtn)
8820 struct vmcs_config vmcs_conf;
8823 if (setup_vmcs_config(&vmcs_conf) < 0)
8825 if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
8826 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
8827 smp_processor_id());
8832 static int get_ept_level(void)
8834 return VMX_EPT_DEFAULT_GAW + 1;
8837 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
8842 /* For VT-d and EPT combination
8843 * 1. MMIO: always map as UC
8845 * a. VT-d without snooping control feature: can't guarantee the
8846 * result, try to trust guest.
8847 * b. VT-d with snooping control feature: snooping control feature of
8848 * VT-d engine can guarantee the cache correctness. Just set it
8849 * to WB to keep consistent with host. So the same as item 3.
8850 * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
8851 * consistent with host MTRR
8854 cache = MTRR_TYPE_UNCACHABLE;
8858 if (!kvm_arch_has_noncoherent_dma(vcpu->kvm)) {
8859 ipat = VMX_EPT_IPAT_BIT;
8860 cache = MTRR_TYPE_WRBACK;
8864 if (kvm_read_cr0(vcpu) & X86_CR0_CD) {
8865 ipat = VMX_EPT_IPAT_BIT;
8866 if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
8867 cache = MTRR_TYPE_WRBACK;
8869 cache = MTRR_TYPE_UNCACHABLE;
8873 cache = kvm_mtrr_get_guest_memory_type(vcpu, gfn);
8876 return (cache << VMX_EPT_MT_EPTE_SHIFT) | ipat;
8879 static int vmx_get_lpage_level(void)
8881 if (enable_ept && !cpu_has_vmx_ept_1g_page())
8882 return PT_DIRECTORY_LEVEL;
8884 /* For shadow and EPT supported 1GB page */
8885 return PT_PDPE_LEVEL;
8888 static void vmcs_set_secondary_exec_control(u32 new_ctl)
8891 * These bits in the secondary execution controls field
8892 * are dynamic, the others are mostly based on the hypervisor
8893 * architecture and the guest's CPUID. Do not touch the
8897 SECONDARY_EXEC_SHADOW_VMCS |
8898 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
8899 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
8901 u32 cur_ctl = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
8903 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
8904 (new_ctl & ~mask) | (cur_ctl & mask));
8907 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
8909 struct kvm_cpuid_entry2 *best;
8910 struct vcpu_vmx *vmx = to_vmx(vcpu);
8911 u32 secondary_exec_ctl = vmx_secondary_exec_control(vmx);
8913 if (vmx_rdtscp_supported()) {
8914 bool rdtscp_enabled = guest_cpuid_has_rdtscp(vcpu);
8915 if (!rdtscp_enabled)
8916 secondary_exec_ctl &= ~SECONDARY_EXEC_RDTSCP;
8920 vmx->nested.nested_vmx_secondary_ctls_high |=
8921 SECONDARY_EXEC_RDTSCP;
8923 vmx->nested.nested_vmx_secondary_ctls_high &=
8924 ~SECONDARY_EXEC_RDTSCP;
8928 /* Exposing INVPCID only when PCID is exposed */
8929 best = kvm_find_cpuid_entry(vcpu, 0x7, 0);
8930 if (vmx_invpcid_supported() &&
8931 (!best || !(best->ebx & bit(X86_FEATURE_INVPCID)) ||
8932 !guest_cpuid_has_pcid(vcpu))) {
8933 secondary_exec_ctl &= ~SECONDARY_EXEC_ENABLE_INVPCID;
8936 best->ebx &= ~bit(X86_FEATURE_INVPCID);
8939 if (cpu_has_secondary_exec_ctrls())
8940 vmcs_set_secondary_exec_control(secondary_exec_ctl);
8942 if (static_cpu_has(X86_FEATURE_PCOMMIT) && nested) {
8943 if (guest_cpuid_has_pcommit(vcpu))
8944 vmx->nested.nested_vmx_secondary_ctls_high |=
8945 SECONDARY_EXEC_PCOMMIT;
8947 vmx->nested.nested_vmx_secondary_ctls_high &=
8948 ~SECONDARY_EXEC_PCOMMIT;
8952 static void vmx_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
8954 if (func == 1 && nested)
8955 entry->ecx |= bit(X86_FEATURE_VMX);
8958 static void nested_ept_inject_page_fault(struct kvm_vcpu *vcpu,
8959 struct x86_exception *fault)
8961 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
8964 if (fault->error_code & PFERR_RSVD_MASK)
8965 exit_reason = EXIT_REASON_EPT_MISCONFIG;
8967 exit_reason = EXIT_REASON_EPT_VIOLATION;
8968 nested_vmx_vmexit(vcpu, exit_reason, 0, vcpu->arch.exit_qualification);
8969 vmcs12->guest_physical_address = fault->address;
8972 /* Callbacks for nested_ept_init_mmu_context: */
8974 static unsigned long nested_ept_get_cr3(struct kvm_vcpu *vcpu)
8976 /* return the page table to be shadowed - in our case, EPT12 */
8977 return get_vmcs12(vcpu)->ept_pointer;
8980 static void nested_ept_init_mmu_context(struct kvm_vcpu *vcpu)
8982 WARN_ON(mmu_is_nested(vcpu));
8983 kvm_init_shadow_ept_mmu(vcpu,
8984 to_vmx(vcpu)->nested.nested_vmx_ept_caps &
8985 VMX_EPT_EXECUTE_ONLY_BIT);
8986 vcpu->arch.mmu.set_cr3 = vmx_set_cr3;
8987 vcpu->arch.mmu.get_cr3 = nested_ept_get_cr3;
8988 vcpu->arch.mmu.inject_page_fault = nested_ept_inject_page_fault;
8990 vcpu->arch.walk_mmu = &vcpu->arch.nested_mmu;
8993 static void nested_ept_uninit_mmu_context(struct kvm_vcpu *vcpu)
8995 vcpu->arch.walk_mmu = &vcpu->arch.mmu;
8998 static bool nested_vmx_is_page_fault_vmexit(struct vmcs12 *vmcs12,
9001 bool inequality, bit;
9003 bit = (vmcs12->exception_bitmap & (1u << PF_VECTOR)) != 0;
9005 (error_code & vmcs12->page_fault_error_code_mask) !=
9006 vmcs12->page_fault_error_code_match;
9007 return inequality ^ bit;
9010 static void vmx_inject_page_fault_nested(struct kvm_vcpu *vcpu,
9011 struct x86_exception *fault)
9013 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
9015 WARN_ON(!is_guest_mode(vcpu));
9017 if (nested_vmx_is_page_fault_vmexit(vmcs12, fault->error_code))
9018 nested_vmx_vmexit(vcpu, to_vmx(vcpu)->exit_reason,
9019 vmcs_read32(VM_EXIT_INTR_INFO),
9020 vmcs_readl(EXIT_QUALIFICATION));
9022 kvm_inject_page_fault(vcpu, fault);
9025 static bool nested_get_vmcs12_pages(struct kvm_vcpu *vcpu,
9026 struct vmcs12 *vmcs12)
9028 struct vcpu_vmx *vmx = to_vmx(vcpu);
9029 int maxphyaddr = cpuid_maxphyaddr(vcpu);
9031 if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
9032 if (!PAGE_ALIGNED(vmcs12->apic_access_addr) ||
9033 vmcs12->apic_access_addr >> maxphyaddr)
9037 * Translate L1 physical address to host physical
9038 * address for vmcs02. Keep the page pinned, so this
9039 * physical address remains valid. We keep a reference
9040 * to it so we can release it later.
9042 if (vmx->nested.apic_access_page) /* shouldn't happen */
9043 nested_release_page(vmx->nested.apic_access_page);
9044 vmx->nested.apic_access_page =
9045 nested_get_page(vcpu, vmcs12->apic_access_addr);
9048 if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
9049 if (!PAGE_ALIGNED(vmcs12->virtual_apic_page_addr) ||
9050 vmcs12->virtual_apic_page_addr >> maxphyaddr)
9053 if (vmx->nested.virtual_apic_page) /* shouldn't happen */
9054 nested_release_page(vmx->nested.virtual_apic_page);
9055 vmx->nested.virtual_apic_page =
9056 nested_get_page(vcpu, vmcs12->virtual_apic_page_addr);
9059 * Failing the vm entry is _not_ what the processor does
9060 * but it's basically the only possibility we have.
9061 * We could still enter the guest if CR8 load exits are
9062 * enabled, CR8 store exits are enabled, and virtualize APIC
9063 * access is disabled; in this case the processor would never
9064 * use the TPR shadow and we could simply clear the bit from
9065 * the execution control. But such a configuration is useless,
9066 * so let's keep the code simple.
9068 if (!vmx->nested.virtual_apic_page)
9072 if (nested_cpu_has_posted_intr(vmcs12)) {
9073 if (!IS_ALIGNED(vmcs12->posted_intr_desc_addr, 64) ||
9074 vmcs12->posted_intr_desc_addr >> maxphyaddr)
9077 if (vmx->nested.pi_desc_page) { /* shouldn't happen */
9078 kunmap(vmx->nested.pi_desc_page);
9079 nested_release_page(vmx->nested.pi_desc_page);
9081 vmx->nested.pi_desc_page =
9082 nested_get_page(vcpu, vmcs12->posted_intr_desc_addr);
9083 if (!vmx->nested.pi_desc_page)
9086 vmx->nested.pi_desc =
9087 (struct pi_desc *)kmap(vmx->nested.pi_desc_page);
9088 if (!vmx->nested.pi_desc) {
9089 nested_release_page_clean(vmx->nested.pi_desc_page);
9092 vmx->nested.pi_desc =
9093 (struct pi_desc *)((void *)vmx->nested.pi_desc +
9094 (unsigned long)(vmcs12->posted_intr_desc_addr &
9101 static void vmx_start_preemption_timer(struct kvm_vcpu *vcpu)
9103 u64 preemption_timeout = get_vmcs12(vcpu)->vmx_preemption_timer_value;
9104 struct vcpu_vmx *vmx = to_vmx(vcpu);
9106 if (vcpu->arch.virtual_tsc_khz == 0)
9109 /* Make sure short timeouts reliably trigger an immediate vmexit.
9110 * hrtimer_start does not guarantee this. */
9111 if (preemption_timeout <= 1) {
9112 vmx_preemption_timer_fn(&vmx->nested.preemption_timer);
9116 preemption_timeout <<= VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
9117 preemption_timeout *= 1000000;
9118 do_div(preemption_timeout, vcpu->arch.virtual_tsc_khz);
9119 hrtimer_start(&vmx->nested.preemption_timer,
9120 ns_to_ktime(preemption_timeout), HRTIMER_MODE_REL);
9123 static int nested_vmx_check_msr_bitmap_controls(struct kvm_vcpu *vcpu,
9124 struct vmcs12 *vmcs12)
9129 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
9132 if (vmcs12_read_any(vcpu, MSR_BITMAP, &addr)) {
9136 maxphyaddr = cpuid_maxphyaddr(vcpu);
9138 if (!PAGE_ALIGNED(vmcs12->msr_bitmap) ||
9139 ((addr + PAGE_SIZE) >> maxphyaddr))
9146 * Merge L0's and L1's MSR bitmap, return false to indicate that
9147 * we do not use the hardware.
9149 static inline bool nested_vmx_merge_msr_bitmap(struct kvm_vcpu *vcpu,
9150 struct vmcs12 *vmcs12)
9154 unsigned long *msr_bitmap;
9156 if (!nested_cpu_has_virt_x2apic_mode(vmcs12))
9159 page = nested_get_page(vcpu, vmcs12->msr_bitmap);
9164 msr_bitmap = (unsigned long *)kmap(page);
9166 nested_release_page_clean(page);
9171 if (nested_cpu_has_virt_x2apic_mode(vmcs12)) {
9172 if (nested_cpu_has_apic_reg_virt(vmcs12))
9173 for (msr = 0x800; msr <= 0x8ff; msr++)
9174 nested_vmx_disable_intercept_for_msr(
9176 vmx_msr_bitmap_nested,
9178 /* TPR is allowed */
9179 nested_vmx_disable_intercept_for_msr(msr_bitmap,
9180 vmx_msr_bitmap_nested,
9181 APIC_BASE_MSR + (APIC_TASKPRI >> 4),
9182 MSR_TYPE_R | MSR_TYPE_W);
9183 if (nested_cpu_has_vid(vmcs12)) {
9184 /* EOI and self-IPI are allowed */
9185 nested_vmx_disable_intercept_for_msr(
9187 vmx_msr_bitmap_nested,
9188 APIC_BASE_MSR + (APIC_EOI >> 4),
9190 nested_vmx_disable_intercept_for_msr(
9192 vmx_msr_bitmap_nested,
9193 APIC_BASE_MSR + (APIC_SELF_IPI >> 4),
9198 * Enable reading intercept of all the x2apic
9199 * MSRs. We should not rely on vmcs12 to do any
9200 * optimizations here, it may have been modified
9203 for (msr = 0x800; msr <= 0x8ff; msr++)
9204 __vmx_enable_intercept_for_msr(
9205 vmx_msr_bitmap_nested,
9209 __vmx_enable_intercept_for_msr(
9210 vmx_msr_bitmap_nested,
9211 APIC_BASE_MSR + (APIC_TASKPRI >> 4),
9213 __vmx_enable_intercept_for_msr(
9214 vmx_msr_bitmap_nested,
9215 APIC_BASE_MSR + (APIC_EOI >> 4),
9217 __vmx_enable_intercept_for_msr(
9218 vmx_msr_bitmap_nested,
9219 APIC_BASE_MSR + (APIC_SELF_IPI >> 4),
9223 nested_release_page_clean(page);
9228 static int nested_vmx_check_apicv_controls(struct kvm_vcpu *vcpu,
9229 struct vmcs12 *vmcs12)
9231 if (!nested_cpu_has_virt_x2apic_mode(vmcs12) &&
9232 !nested_cpu_has_apic_reg_virt(vmcs12) &&
9233 !nested_cpu_has_vid(vmcs12) &&
9234 !nested_cpu_has_posted_intr(vmcs12))
9238 * If virtualize x2apic mode is enabled,
9239 * virtualize apic access must be disabled.
9241 if (nested_cpu_has_virt_x2apic_mode(vmcs12) &&
9242 nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
9246 * If virtual interrupt delivery is enabled,
9247 * we must exit on external interrupts.
9249 if (nested_cpu_has_vid(vmcs12) &&
9250 !nested_exit_on_intr(vcpu))
9254 * bits 15:8 should be zero in posted_intr_nv,
9255 * the descriptor address has been already checked
9256 * in nested_get_vmcs12_pages.
9258 if (nested_cpu_has_posted_intr(vmcs12) &&
9259 (!nested_cpu_has_vid(vmcs12) ||
9260 !nested_exit_intr_ack_set(vcpu) ||
9261 vmcs12->posted_intr_nv & 0xff00))
9264 /* tpr shadow is needed by all apicv features. */
9265 if (!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
9271 static int nested_vmx_check_msr_switch(struct kvm_vcpu *vcpu,
9272 unsigned long count_field,
9273 unsigned long addr_field)
9278 if (vmcs12_read_any(vcpu, count_field, &count) ||
9279 vmcs12_read_any(vcpu, addr_field, &addr)) {
9285 maxphyaddr = cpuid_maxphyaddr(vcpu);
9286 if (!IS_ALIGNED(addr, 16) || addr >> maxphyaddr ||
9287 (addr + count * sizeof(struct vmx_msr_entry) - 1) >> maxphyaddr) {
9288 pr_warn_ratelimited(
9289 "nVMX: invalid MSR switch (0x%lx, %d, %llu, 0x%08llx)",
9290 addr_field, maxphyaddr, count, addr);
9296 static int nested_vmx_check_msr_switch_controls(struct kvm_vcpu *vcpu,
9297 struct vmcs12 *vmcs12)
9299 if (vmcs12->vm_exit_msr_load_count == 0 &&
9300 vmcs12->vm_exit_msr_store_count == 0 &&
9301 vmcs12->vm_entry_msr_load_count == 0)
9302 return 0; /* Fast path */
9303 if (nested_vmx_check_msr_switch(vcpu, VM_EXIT_MSR_LOAD_COUNT,
9304 VM_EXIT_MSR_LOAD_ADDR) ||
9305 nested_vmx_check_msr_switch(vcpu, VM_EXIT_MSR_STORE_COUNT,
9306 VM_EXIT_MSR_STORE_ADDR) ||
9307 nested_vmx_check_msr_switch(vcpu, VM_ENTRY_MSR_LOAD_COUNT,
9308 VM_ENTRY_MSR_LOAD_ADDR))
9313 static int nested_vmx_msr_check_common(struct kvm_vcpu *vcpu,
9314 struct vmx_msr_entry *e)
9316 /* x2APIC MSR accesses are not allowed */
9317 if (vcpu->arch.apic_base & X2APIC_ENABLE && e->index >> 8 == 0x8)
9319 if (e->index == MSR_IA32_UCODE_WRITE || /* SDM Table 35-2 */
9320 e->index == MSR_IA32_UCODE_REV)
9322 if (e->reserved != 0)
9327 static int nested_vmx_load_msr_check(struct kvm_vcpu *vcpu,
9328 struct vmx_msr_entry *e)
9330 if (e->index == MSR_FS_BASE ||
9331 e->index == MSR_GS_BASE ||
9332 e->index == MSR_IA32_SMM_MONITOR_CTL || /* SMM is not supported */
9333 nested_vmx_msr_check_common(vcpu, e))
9338 static int nested_vmx_store_msr_check(struct kvm_vcpu *vcpu,
9339 struct vmx_msr_entry *e)
9341 if (e->index == MSR_IA32_SMBASE || /* SMM is not supported */
9342 nested_vmx_msr_check_common(vcpu, e))
9348 * Load guest's/host's msr at nested entry/exit.
9349 * return 0 for success, entry index for failure.
9351 static u32 nested_vmx_load_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
9354 struct vmx_msr_entry e;
9355 struct msr_data msr;
9357 msr.host_initiated = false;
9358 for (i = 0; i < count; i++) {
9359 if (kvm_vcpu_read_guest(vcpu, gpa + i * sizeof(e),
9361 pr_warn_ratelimited(
9362 "%s cannot read MSR entry (%u, 0x%08llx)\n",
9363 __func__, i, gpa + i * sizeof(e));
9366 if (nested_vmx_load_msr_check(vcpu, &e)) {
9367 pr_warn_ratelimited(
9368 "%s check failed (%u, 0x%x, 0x%x)\n",
9369 __func__, i, e.index, e.reserved);
9372 msr.index = e.index;
9374 if (kvm_set_msr(vcpu, &msr)) {
9375 pr_warn_ratelimited(
9376 "%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
9377 __func__, i, e.index, e.value);
9386 static int nested_vmx_store_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
9389 struct vmx_msr_entry e;
9391 for (i = 0; i < count; i++) {
9392 struct msr_data msr_info;
9393 if (kvm_vcpu_read_guest(vcpu,
9394 gpa + i * sizeof(e),
9395 &e, 2 * sizeof(u32))) {
9396 pr_warn_ratelimited(
9397 "%s cannot read MSR entry (%u, 0x%08llx)\n",
9398 __func__, i, gpa + i * sizeof(e));
9401 if (nested_vmx_store_msr_check(vcpu, &e)) {
9402 pr_warn_ratelimited(
9403 "%s check failed (%u, 0x%x, 0x%x)\n",
9404 __func__, i, e.index, e.reserved);
9407 msr_info.host_initiated = false;
9408 msr_info.index = e.index;
9409 if (kvm_get_msr(vcpu, &msr_info)) {
9410 pr_warn_ratelimited(
9411 "%s cannot read MSR (%u, 0x%x)\n",
9412 __func__, i, e.index);
9415 if (kvm_vcpu_write_guest(vcpu,
9416 gpa + i * sizeof(e) +
9417 offsetof(struct vmx_msr_entry, value),
9418 &msr_info.data, sizeof(msr_info.data))) {
9419 pr_warn_ratelimited(
9420 "%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
9421 __func__, i, e.index, msr_info.data);
9429 * prepare_vmcs02 is called when the L1 guest hypervisor runs its nested
9430 * L2 guest. L1 has a vmcs for L2 (vmcs12), and this function "merges" it
9431 * with L0's requirements for its guest (a.k.a. vmcs01), so we can run the L2
9432 * guest in a way that will both be appropriate to L1's requests, and our
9433 * needs. In addition to modifying the active vmcs (which is vmcs02), this
9434 * function also has additional necessary side-effects, like setting various
9435 * vcpu->arch fields.
9437 static void prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
9439 struct vcpu_vmx *vmx = to_vmx(vcpu);
9442 vmcs_write16(GUEST_ES_SELECTOR, vmcs12->guest_es_selector);
9443 vmcs_write16(GUEST_CS_SELECTOR, vmcs12->guest_cs_selector);
9444 vmcs_write16(GUEST_SS_SELECTOR, vmcs12->guest_ss_selector);
9445 vmcs_write16(GUEST_DS_SELECTOR, vmcs12->guest_ds_selector);
9446 vmcs_write16(GUEST_FS_SELECTOR, vmcs12->guest_fs_selector);
9447 vmcs_write16(GUEST_GS_SELECTOR, vmcs12->guest_gs_selector);
9448 vmcs_write16(GUEST_LDTR_SELECTOR, vmcs12->guest_ldtr_selector);
9449 vmcs_write16(GUEST_TR_SELECTOR, vmcs12->guest_tr_selector);
9450 vmcs_write32(GUEST_ES_LIMIT, vmcs12->guest_es_limit);
9451 vmcs_write32(GUEST_CS_LIMIT, vmcs12->guest_cs_limit);
9452 vmcs_write32(GUEST_SS_LIMIT, vmcs12->guest_ss_limit);
9453 vmcs_write32(GUEST_DS_LIMIT, vmcs12->guest_ds_limit);
9454 vmcs_write32(GUEST_FS_LIMIT, vmcs12->guest_fs_limit);
9455 vmcs_write32(GUEST_GS_LIMIT, vmcs12->guest_gs_limit);
9456 vmcs_write32(GUEST_LDTR_LIMIT, vmcs12->guest_ldtr_limit);
9457 vmcs_write32(GUEST_TR_LIMIT, vmcs12->guest_tr_limit);
9458 vmcs_write32(GUEST_GDTR_LIMIT, vmcs12->guest_gdtr_limit);
9459 vmcs_write32(GUEST_IDTR_LIMIT, vmcs12->guest_idtr_limit);
9460 vmcs_write32(GUEST_ES_AR_BYTES, vmcs12->guest_es_ar_bytes);
9461 vmcs_write32(GUEST_CS_AR_BYTES, vmcs12->guest_cs_ar_bytes);
9462 vmcs_write32(GUEST_SS_AR_BYTES, vmcs12->guest_ss_ar_bytes);
9463 vmcs_write32(GUEST_DS_AR_BYTES, vmcs12->guest_ds_ar_bytes);
9464 vmcs_write32(GUEST_FS_AR_BYTES, vmcs12->guest_fs_ar_bytes);
9465 vmcs_write32(GUEST_GS_AR_BYTES, vmcs12->guest_gs_ar_bytes);
9466 vmcs_write32(GUEST_LDTR_AR_BYTES, vmcs12->guest_ldtr_ar_bytes);
9467 vmcs_write32(GUEST_TR_AR_BYTES, vmcs12->guest_tr_ar_bytes);
9468 vmcs_writel(GUEST_ES_BASE, vmcs12->guest_es_base);
9469 vmcs_writel(GUEST_CS_BASE, vmcs12->guest_cs_base);
9470 vmcs_writel(GUEST_SS_BASE, vmcs12->guest_ss_base);
9471 vmcs_writel(GUEST_DS_BASE, vmcs12->guest_ds_base);
9472 vmcs_writel(GUEST_FS_BASE, vmcs12->guest_fs_base);
9473 vmcs_writel(GUEST_GS_BASE, vmcs12->guest_gs_base);
9474 vmcs_writel(GUEST_LDTR_BASE, vmcs12->guest_ldtr_base);
9475 vmcs_writel(GUEST_TR_BASE, vmcs12->guest_tr_base);
9476 vmcs_writel(GUEST_GDTR_BASE, vmcs12->guest_gdtr_base);
9477 vmcs_writel(GUEST_IDTR_BASE, vmcs12->guest_idtr_base);
9479 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS) {
9480 kvm_set_dr(vcpu, 7, vmcs12->guest_dr7);
9481 vmcs_write64(GUEST_IA32_DEBUGCTL, vmcs12->guest_ia32_debugctl);
9483 kvm_set_dr(vcpu, 7, vcpu->arch.dr7);
9484 vmcs_write64(GUEST_IA32_DEBUGCTL, vmx->nested.vmcs01_debugctl);
9486 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
9487 vmcs12->vm_entry_intr_info_field);
9488 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
9489 vmcs12->vm_entry_exception_error_code);
9490 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
9491 vmcs12->vm_entry_instruction_len);
9492 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
9493 vmcs12->guest_interruptibility_info);
9494 vmcs_write32(GUEST_SYSENTER_CS, vmcs12->guest_sysenter_cs);
9495 vmx_set_rflags(vcpu, vmcs12->guest_rflags);
9496 vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
9497 vmcs12->guest_pending_dbg_exceptions);
9498 vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->guest_sysenter_esp);
9499 vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->guest_sysenter_eip);
9501 if (nested_cpu_has_xsaves(vmcs12))
9502 vmcs_write64(XSS_EXIT_BITMAP, vmcs12->xss_exit_bitmap);
9503 vmcs_write64(VMCS_LINK_POINTER, -1ull);
9505 exec_control = vmcs12->pin_based_vm_exec_control;
9506 exec_control |= vmcs_config.pin_based_exec_ctrl;
9507 exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
9509 if (nested_cpu_has_posted_intr(vmcs12)) {
9511 * Note that we use L0's vector here and in
9512 * vmx_deliver_nested_posted_interrupt.
9514 vmx->nested.posted_intr_nv = vmcs12->posted_intr_nv;
9515 vmx->nested.pi_pending = false;
9516 vmcs_write64(POSTED_INTR_NV, POSTED_INTR_VECTOR);
9517 vmcs_write64(POSTED_INTR_DESC_ADDR,
9518 page_to_phys(vmx->nested.pi_desc_page) +
9519 (unsigned long)(vmcs12->posted_intr_desc_addr &
9522 exec_control &= ~PIN_BASED_POSTED_INTR;
9524 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, exec_control);
9526 vmx->nested.preemption_timer_expired = false;
9527 if (nested_cpu_has_preemption_timer(vmcs12))
9528 vmx_start_preemption_timer(vcpu);
9531 * Whether page-faults are trapped is determined by a combination of
9532 * 3 settings: PFEC_MASK, PFEC_MATCH and EXCEPTION_BITMAP.PF.
9533 * If enable_ept, L0 doesn't care about page faults and we should
9534 * set all of these to L1's desires. However, if !enable_ept, L0 does
9535 * care about (at least some) page faults, and because it is not easy
9536 * (if at all possible?) to merge L0 and L1's desires, we simply ask
9537 * to exit on each and every L2 page fault. This is done by setting
9538 * MASK=MATCH=0 and (see below) EB.PF=1.
9539 * Note that below we don't need special code to set EB.PF beyond the
9540 * "or"ing of the EB of vmcs01 and vmcs12, because when enable_ept,
9541 * vmcs01's EB.PF is 0 so the "or" will take vmcs12's value, and when
9542 * !enable_ept, EB.PF is 1, so the "or" will always be 1.
9544 * A problem with this approach (when !enable_ept) is that L1 may be
9545 * injected with more page faults than it asked for. This could have
9546 * caused problems, but in practice existing hypervisors don't care.
9547 * To fix this, we will need to emulate the PFEC checking (on the L1
9548 * page tables), using walk_addr(), when injecting PFs to L1.
9550 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK,
9551 enable_ept ? vmcs12->page_fault_error_code_mask : 0);
9552 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH,
9553 enable_ept ? vmcs12->page_fault_error_code_match : 0);
9555 if (cpu_has_secondary_exec_ctrls()) {
9556 exec_control = vmx_secondary_exec_control(vmx);
9558 /* Take the following fields only from vmcs12 */
9559 exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
9560 SECONDARY_EXEC_RDTSCP |
9561 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
9562 SECONDARY_EXEC_APIC_REGISTER_VIRT |
9563 SECONDARY_EXEC_PCOMMIT);
9564 if (nested_cpu_has(vmcs12,
9565 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS))
9566 exec_control |= vmcs12->secondary_vm_exec_control;
9568 if (exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES) {
9570 * If translation failed, no matter: This feature asks
9571 * to exit when accessing the given address, and if it
9572 * can never be accessed, this feature won't do
9575 if (!vmx->nested.apic_access_page)
9577 ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
9579 vmcs_write64(APIC_ACCESS_ADDR,
9580 page_to_phys(vmx->nested.apic_access_page));
9581 } else if (!(nested_cpu_has_virt_x2apic_mode(vmcs12)) &&
9582 cpu_need_virtualize_apic_accesses(&vmx->vcpu)) {
9584 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
9585 kvm_vcpu_reload_apic_access_page(vcpu);
9588 if (exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) {
9589 vmcs_write64(EOI_EXIT_BITMAP0,
9590 vmcs12->eoi_exit_bitmap0);
9591 vmcs_write64(EOI_EXIT_BITMAP1,
9592 vmcs12->eoi_exit_bitmap1);
9593 vmcs_write64(EOI_EXIT_BITMAP2,
9594 vmcs12->eoi_exit_bitmap2);
9595 vmcs_write64(EOI_EXIT_BITMAP3,
9596 vmcs12->eoi_exit_bitmap3);
9597 vmcs_write16(GUEST_INTR_STATUS,
9598 vmcs12->guest_intr_status);
9601 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
9606 * Set host-state according to L0's settings (vmcs12 is irrelevant here)
9607 * Some constant fields are set here by vmx_set_constant_host_state().
9608 * Other fields are different per CPU, and will be set later when
9609 * vmx_vcpu_load() is called, and when vmx_save_host_state() is called.
9611 vmx_set_constant_host_state(vmx);
9614 * HOST_RSP is normally set correctly in vmx_vcpu_run() just before
9615 * entry, but only if the current (host) sp changed from the value
9616 * we wrote last (vmx->host_rsp). This cache is no longer relevant
9617 * if we switch vmcs, and rather than hold a separate cache per vmcs,
9618 * here we just force the write to happen on entry.
9622 exec_control = vmx_exec_control(vmx); /* L0's desires */
9623 exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
9624 exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
9625 exec_control &= ~CPU_BASED_TPR_SHADOW;
9626 exec_control |= vmcs12->cpu_based_vm_exec_control;
9628 if (exec_control & CPU_BASED_TPR_SHADOW) {
9629 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
9630 page_to_phys(vmx->nested.virtual_apic_page));
9631 vmcs_write32(TPR_THRESHOLD, vmcs12->tpr_threshold);
9634 if (cpu_has_vmx_msr_bitmap() &&
9635 exec_control & CPU_BASED_USE_MSR_BITMAPS) {
9636 nested_vmx_merge_msr_bitmap(vcpu, vmcs12);
9637 /* MSR_BITMAP will be set by following vmx_set_efer. */
9639 exec_control &= ~CPU_BASED_USE_MSR_BITMAPS;
9642 * Merging of IO bitmap not currently supported.
9643 * Rather, exit every time.
9645 exec_control &= ~CPU_BASED_USE_IO_BITMAPS;
9646 exec_control |= CPU_BASED_UNCOND_IO_EXITING;
9648 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
9650 /* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the
9651 * bitwise-or of what L1 wants to trap for L2, and what we want to
9652 * trap. Note that CR0.TS also needs updating - we do this later.
9654 update_exception_bitmap(vcpu);
9655 vcpu->arch.cr0_guest_owned_bits &= ~vmcs12->cr0_guest_host_mask;
9656 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
9658 /* L2->L1 exit controls are emulated - the hardware exit is to L0 so
9659 * we should use its exit controls. Note that VM_EXIT_LOAD_IA32_EFER
9660 * bits are further modified by vmx_set_efer() below.
9662 vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
9664 /* vmcs12's VM_ENTRY_LOAD_IA32_EFER and VM_ENTRY_IA32E_MODE are
9665 * emulated by vmx_set_efer(), below.
9667 vm_entry_controls_init(vmx,
9668 (vmcs12->vm_entry_controls & ~VM_ENTRY_LOAD_IA32_EFER &
9669 ~VM_ENTRY_IA32E_MODE) |
9670 (vmcs_config.vmentry_ctrl & ~VM_ENTRY_IA32E_MODE));
9672 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT) {
9673 vmcs_write64(GUEST_IA32_PAT, vmcs12->guest_ia32_pat);
9674 vcpu->arch.pat = vmcs12->guest_ia32_pat;
9675 } else if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
9676 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
9679 set_cr4_guest_host_mask(vmx);
9681 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS)
9682 vmcs_write64(GUEST_BNDCFGS, vmcs12->guest_bndcfgs);
9684 if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING)
9685 vmcs_write64(TSC_OFFSET,
9686 vmx->nested.vmcs01_tsc_offset + vmcs12->tsc_offset);
9688 vmcs_write64(TSC_OFFSET, vmx->nested.vmcs01_tsc_offset);
9692 * There is no direct mapping between vpid02 and vpid12, the
9693 * vpid02 is per-vCPU for L0 and reused while the value of
9694 * vpid12 is changed w/ one invvpid during nested vmentry.
9695 * The vpid12 is allocated by L1 for L2, so it will not
9696 * influence global bitmap(for vpid01 and vpid02 allocation)
9697 * even if spawn a lot of nested vCPUs.
9699 if (nested_cpu_has_vpid(vmcs12) && vmx->nested.vpid02) {
9700 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->nested.vpid02);
9701 if (vmcs12->virtual_processor_id != vmx->nested.last_vpid) {
9702 vmx->nested.last_vpid = vmcs12->virtual_processor_id;
9703 __vmx_flush_tlb(vcpu, to_vmx(vcpu)->nested.vpid02);
9706 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
9707 vmx_flush_tlb(vcpu);
9712 if (nested_cpu_has_ept(vmcs12)) {
9713 kvm_mmu_unload(vcpu);
9714 nested_ept_init_mmu_context(vcpu);
9717 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER)
9718 vcpu->arch.efer = vmcs12->guest_ia32_efer;
9719 else if (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE)
9720 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
9722 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
9723 /* Note: modifies VM_ENTRY/EXIT_CONTROLS and GUEST/HOST_IA32_EFER */
9724 vmx_set_efer(vcpu, vcpu->arch.efer);
9727 * This sets GUEST_CR0 to vmcs12->guest_cr0, with possibly a modified
9728 * TS bit (for lazy fpu) and bits which we consider mandatory enabled.
9729 * The CR0_READ_SHADOW is what L2 should have expected to read given
9730 * the specifications by L1; It's not enough to take
9731 * vmcs12->cr0_read_shadow because on our cr0_guest_host_mask we we
9732 * have more bits than L1 expected.
9734 vmx_set_cr0(vcpu, vmcs12->guest_cr0);
9735 vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
9737 vmx_set_cr4(vcpu, vmcs12->guest_cr4);
9738 vmcs_writel(CR4_READ_SHADOW, nested_read_cr4(vmcs12));
9740 /* shadow page tables on either EPT or shadow page tables */
9741 kvm_set_cr3(vcpu, vmcs12->guest_cr3);
9742 kvm_mmu_reset_context(vcpu);
9745 vcpu->arch.walk_mmu->inject_page_fault = vmx_inject_page_fault_nested;
9748 * L1 may access the L2's PDPTR, so save them to construct vmcs12
9751 vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
9752 vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
9753 vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
9754 vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
9757 kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->guest_rsp);
9758 kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->guest_rip);
9762 * nested_vmx_run() handles a nested entry, i.e., a VMLAUNCH or VMRESUME on L1
9763 * for running an L2 nested guest.
9765 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
9767 struct vmcs12 *vmcs12;
9768 struct vcpu_vmx *vmx = to_vmx(vcpu);
9770 struct loaded_vmcs *vmcs02;
9774 if (!nested_vmx_check_permission(vcpu) ||
9775 !nested_vmx_check_vmcs12(vcpu))
9778 skip_emulated_instruction(vcpu);
9779 vmcs12 = get_vmcs12(vcpu);
9781 if (enable_shadow_vmcs)
9782 copy_shadow_to_vmcs12(vmx);
9785 * The nested entry process starts with enforcing various prerequisites
9786 * on vmcs12 as required by the Intel SDM, and act appropriately when
9787 * they fail: As the SDM explains, some conditions should cause the
9788 * instruction to fail, while others will cause the instruction to seem
9789 * to succeed, but return an EXIT_REASON_INVALID_STATE.
9790 * To speed up the normal (success) code path, we should avoid checking
9791 * for misconfigurations which will anyway be caught by the processor
9792 * when using the merged vmcs02.
9794 if (vmcs12->launch_state == launch) {
9795 nested_vmx_failValid(vcpu,
9796 launch ? VMXERR_VMLAUNCH_NONCLEAR_VMCS
9797 : VMXERR_VMRESUME_NONLAUNCHED_VMCS);
9801 if (vmcs12->guest_activity_state != GUEST_ACTIVITY_ACTIVE &&
9802 vmcs12->guest_activity_state != GUEST_ACTIVITY_HLT) {
9803 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
9807 if (!nested_get_vmcs12_pages(vcpu, vmcs12)) {
9808 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
9812 if (nested_vmx_check_msr_bitmap_controls(vcpu, vmcs12)) {
9813 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
9817 if (nested_vmx_check_apicv_controls(vcpu, vmcs12)) {
9818 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
9822 if (nested_vmx_check_msr_switch_controls(vcpu, vmcs12)) {
9823 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
9827 if (!vmx_control_verify(vmcs12->cpu_based_vm_exec_control,
9828 vmx->nested.nested_vmx_true_procbased_ctls_low,
9829 vmx->nested.nested_vmx_procbased_ctls_high) ||
9830 !vmx_control_verify(vmcs12->secondary_vm_exec_control,
9831 vmx->nested.nested_vmx_secondary_ctls_low,
9832 vmx->nested.nested_vmx_secondary_ctls_high) ||
9833 !vmx_control_verify(vmcs12->pin_based_vm_exec_control,
9834 vmx->nested.nested_vmx_pinbased_ctls_low,
9835 vmx->nested.nested_vmx_pinbased_ctls_high) ||
9836 !vmx_control_verify(vmcs12->vm_exit_controls,
9837 vmx->nested.nested_vmx_true_exit_ctls_low,
9838 vmx->nested.nested_vmx_exit_ctls_high) ||
9839 !vmx_control_verify(vmcs12->vm_entry_controls,
9840 vmx->nested.nested_vmx_true_entry_ctls_low,
9841 vmx->nested.nested_vmx_entry_ctls_high))
9843 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
9847 if (((vmcs12->host_cr0 & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON) ||
9848 ((vmcs12->host_cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON)) {
9849 nested_vmx_failValid(vcpu,
9850 VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
9854 if (!nested_cr0_valid(vcpu, vmcs12->guest_cr0) ||
9855 ((vmcs12->guest_cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON)) {
9856 nested_vmx_entry_failure(vcpu, vmcs12,
9857 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
9860 if (vmcs12->vmcs_link_pointer != -1ull) {
9861 nested_vmx_entry_failure(vcpu, vmcs12,
9862 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_VMCS_LINK_PTR);
9867 * If the load IA32_EFER VM-entry control is 1, the following checks
9868 * are performed on the field for the IA32_EFER MSR:
9869 * - Bits reserved in the IA32_EFER MSR must be 0.
9870 * - Bit 10 (corresponding to IA32_EFER.LMA) must equal the value of
9871 * the IA-32e mode guest VM-exit control. It must also be identical
9872 * to bit 8 (LME) if bit 31 in the CR0 field (corresponding to
9875 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER) {
9876 ia32e = (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) != 0;
9877 if (!kvm_valid_efer(vcpu, vmcs12->guest_ia32_efer) ||
9878 ia32e != !!(vmcs12->guest_ia32_efer & EFER_LMA) ||
9879 ((vmcs12->guest_cr0 & X86_CR0_PG) &&
9880 ia32e != !!(vmcs12->guest_ia32_efer & EFER_LME))) {
9881 nested_vmx_entry_failure(vcpu, vmcs12,
9882 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
9888 * If the load IA32_EFER VM-exit control is 1, bits reserved in the
9889 * IA32_EFER MSR must be 0 in the field for that register. In addition,
9890 * the values of the LMA and LME bits in the field must each be that of
9891 * the host address-space size VM-exit control.
9893 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER) {
9894 ia32e = (vmcs12->vm_exit_controls &
9895 VM_EXIT_HOST_ADDR_SPACE_SIZE) != 0;
9896 if (!kvm_valid_efer(vcpu, vmcs12->host_ia32_efer) ||
9897 ia32e != !!(vmcs12->host_ia32_efer & EFER_LMA) ||
9898 ia32e != !!(vmcs12->host_ia32_efer & EFER_LME)) {
9899 nested_vmx_entry_failure(vcpu, vmcs12,
9900 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
9906 * We're finally done with prerequisite checking, and can start with
9910 vmcs02 = nested_get_current_vmcs02(vmx);
9914 enter_guest_mode(vcpu);
9916 vmx->nested.vmcs01_tsc_offset = vmcs_read64(TSC_OFFSET);
9918 if (!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS))
9919 vmx->nested.vmcs01_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
9922 vmx->loaded_vmcs = vmcs02;
9924 vmx_vcpu_load(vcpu, cpu);
9928 vmx_segment_cache_clear(vmx);
9930 prepare_vmcs02(vcpu, vmcs12);
9932 msr_entry_idx = nested_vmx_load_msr(vcpu,
9933 vmcs12->vm_entry_msr_load_addr,
9934 vmcs12->vm_entry_msr_load_count);
9935 if (msr_entry_idx) {
9936 leave_guest_mode(vcpu);
9937 vmx_load_vmcs01(vcpu);
9938 nested_vmx_entry_failure(vcpu, vmcs12,
9939 EXIT_REASON_MSR_LOAD_FAIL, msr_entry_idx);
9943 vmcs12->launch_state = 1;
9945 if (vmcs12->guest_activity_state == GUEST_ACTIVITY_HLT)
9946 return kvm_vcpu_halt(vcpu);
9948 vmx->nested.nested_run_pending = 1;
9951 * Note no nested_vmx_succeed or nested_vmx_fail here. At this point
9952 * we are no longer running L1, and VMLAUNCH/VMRESUME has not yet
9953 * returned as far as L1 is concerned. It will only return (and set
9954 * the success flag) when L2 exits (see nested_vmx_vmexit()).
9960 * On a nested exit from L2 to L1, vmcs12.guest_cr0 might not be up-to-date
9961 * because L2 may have changed some cr0 bits directly (CRO_GUEST_HOST_MASK).
9962 * This function returns the new value we should put in vmcs12.guest_cr0.
9963 * It's not enough to just return the vmcs02 GUEST_CR0. Rather,
9964 * 1. Bits that neither L0 nor L1 trapped, were set directly by L2 and are now
9965 * available in vmcs02 GUEST_CR0. (Note: It's enough to check that L0
9966 * didn't trap the bit, because if L1 did, so would L0).
9967 * 2. Bits that L1 asked to trap (and therefore L0 also did) could not have
9968 * been modified by L2, and L1 knows it. So just leave the old value of
9969 * the bit from vmcs12.guest_cr0. Note that the bit from vmcs02 GUEST_CR0
9970 * isn't relevant, because if L0 traps this bit it can set it to anything.
9971 * 3. Bits that L1 didn't trap, but L0 did. L1 believes the guest could have
9972 * changed these bits, and therefore they need to be updated, but L0
9973 * didn't necessarily allow them to be changed in GUEST_CR0 - and rather
9974 * put them in vmcs02 CR0_READ_SHADOW. So take these bits from there.
9976 static inline unsigned long
9977 vmcs12_guest_cr0(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
9980 /*1*/ (vmcs_readl(GUEST_CR0) & vcpu->arch.cr0_guest_owned_bits) |
9981 /*2*/ (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask) |
9982 /*3*/ (vmcs_readl(CR0_READ_SHADOW) & ~(vmcs12->cr0_guest_host_mask |
9983 vcpu->arch.cr0_guest_owned_bits));
9986 static inline unsigned long
9987 vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
9990 /*1*/ (vmcs_readl(GUEST_CR4) & vcpu->arch.cr4_guest_owned_bits) |
9991 /*2*/ (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask) |
9992 /*3*/ (vmcs_readl(CR4_READ_SHADOW) & ~(vmcs12->cr4_guest_host_mask |
9993 vcpu->arch.cr4_guest_owned_bits));
9996 static void vmcs12_save_pending_event(struct kvm_vcpu *vcpu,
9997 struct vmcs12 *vmcs12)
10002 if (vcpu->arch.exception.pending && vcpu->arch.exception.reinject) {
10003 nr = vcpu->arch.exception.nr;
10004 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
10006 if (kvm_exception_is_soft(nr)) {
10007 vmcs12->vm_exit_instruction_len =
10008 vcpu->arch.event_exit_inst_len;
10009 idt_vectoring |= INTR_TYPE_SOFT_EXCEPTION;
10011 idt_vectoring |= INTR_TYPE_HARD_EXCEPTION;
10013 if (vcpu->arch.exception.has_error_code) {
10014 idt_vectoring |= VECTORING_INFO_DELIVER_CODE_MASK;
10015 vmcs12->idt_vectoring_error_code =
10016 vcpu->arch.exception.error_code;
10019 vmcs12->idt_vectoring_info_field = idt_vectoring;
10020 } else if (vcpu->arch.nmi_injected) {
10021 vmcs12->idt_vectoring_info_field =
10022 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR;
10023 } else if (vcpu->arch.interrupt.pending) {
10024 nr = vcpu->arch.interrupt.nr;
10025 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
10027 if (vcpu->arch.interrupt.soft) {
10028 idt_vectoring |= INTR_TYPE_SOFT_INTR;
10029 vmcs12->vm_entry_instruction_len =
10030 vcpu->arch.event_exit_inst_len;
10032 idt_vectoring |= INTR_TYPE_EXT_INTR;
10034 vmcs12->idt_vectoring_info_field = idt_vectoring;
10038 static int vmx_check_nested_events(struct kvm_vcpu *vcpu, bool external_intr)
10040 struct vcpu_vmx *vmx = to_vmx(vcpu);
10042 if (nested_cpu_has_preemption_timer(get_vmcs12(vcpu)) &&
10043 vmx->nested.preemption_timer_expired) {
10044 if (vmx->nested.nested_run_pending)
10046 nested_vmx_vmexit(vcpu, EXIT_REASON_PREEMPTION_TIMER, 0, 0);
10050 if (vcpu->arch.nmi_pending && nested_exit_on_nmi(vcpu)) {
10051 if (vmx->nested.nested_run_pending ||
10052 vcpu->arch.interrupt.pending)
10054 nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
10055 NMI_VECTOR | INTR_TYPE_NMI_INTR |
10056 INTR_INFO_VALID_MASK, 0);
10058 * The NMI-triggered VM exit counts as injection:
10059 * clear this one and block further NMIs.
10061 vcpu->arch.nmi_pending = 0;
10062 vmx_set_nmi_mask(vcpu, true);
10066 if ((kvm_cpu_has_interrupt(vcpu) || external_intr) &&
10067 nested_exit_on_intr(vcpu)) {
10068 if (vmx->nested.nested_run_pending)
10070 nested_vmx_vmexit(vcpu, EXIT_REASON_EXTERNAL_INTERRUPT, 0, 0);
10074 return vmx_complete_nested_posted_interrupt(vcpu);
10077 static u32 vmx_get_preemption_timer_value(struct kvm_vcpu *vcpu)
10079 ktime_t remaining =
10080 hrtimer_get_remaining(&to_vmx(vcpu)->nested.preemption_timer);
10083 if (ktime_to_ns(remaining) <= 0)
10086 value = ktime_to_ns(remaining) * vcpu->arch.virtual_tsc_khz;
10087 do_div(value, 1000000);
10088 return value >> VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
10092 * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
10093 * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),
10094 * and this function updates it to reflect the changes to the guest state while
10095 * L2 was running (and perhaps made some exits which were handled directly by L0
10096 * without going back to L1), and to reflect the exit reason.
10097 * Note that we do not have to copy here all VMCS fields, just those that
10098 * could have changed by the L2 guest or the exit - i.e., the guest-state and
10099 * exit-information fields only. Other fields are modified by L1 with VMWRITE,
10100 * which already writes to vmcs12 directly.
10102 static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
10103 u32 exit_reason, u32 exit_intr_info,
10104 unsigned long exit_qualification)
10106 /* update guest state fields: */
10107 vmcs12->guest_cr0 = vmcs12_guest_cr0(vcpu, vmcs12);
10108 vmcs12->guest_cr4 = vmcs12_guest_cr4(vcpu, vmcs12);
10110 vmcs12->guest_rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
10111 vmcs12->guest_rip = kvm_register_read(vcpu, VCPU_REGS_RIP);
10112 vmcs12->guest_rflags = vmcs_readl(GUEST_RFLAGS);
10114 vmcs12->guest_es_selector = vmcs_read16(GUEST_ES_SELECTOR);
10115 vmcs12->guest_cs_selector = vmcs_read16(GUEST_CS_SELECTOR);
10116 vmcs12->guest_ss_selector = vmcs_read16(GUEST_SS_SELECTOR);
10117 vmcs12->guest_ds_selector = vmcs_read16(GUEST_DS_SELECTOR);
10118 vmcs12->guest_fs_selector = vmcs_read16(GUEST_FS_SELECTOR);
10119 vmcs12->guest_gs_selector = vmcs_read16(GUEST_GS_SELECTOR);
10120 vmcs12->guest_ldtr_selector = vmcs_read16(GUEST_LDTR_SELECTOR);
10121 vmcs12->guest_tr_selector = vmcs_read16(GUEST_TR_SELECTOR);
10122 vmcs12->guest_es_limit = vmcs_read32(GUEST_ES_LIMIT);
10123 vmcs12->guest_cs_limit = vmcs_read32(GUEST_CS_LIMIT);
10124 vmcs12->guest_ss_limit = vmcs_read32(GUEST_SS_LIMIT);
10125 vmcs12->guest_ds_limit = vmcs_read32(GUEST_DS_LIMIT);
10126 vmcs12->guest_fs_limit = vmcs_read32(GUEST_FS_LIMIT);
10127 vmcs12->guest_gs_limit = vmcs_read32(GUEST_GS_LIMIT);
10128 vmcs12->guest_ldtr_limit = vmcs_read32(GUEST_LDTR_LIMIT);
10129 vmcs12->guest_tr_limit = vmcs_read32(GUEST_TR_LIMIT);
10130 vmcs12->guest_gdtr_limit = vmcs_read32(GUEST_GDTR_LIMIT);
10131 vmcs12->guest_idtr_limit = vmcs_read32(GUEST_IDTR_LIMIT);
10132 vmcs12->guest_es_ar_bytes = vmcs_read32(GUEST_ES_AR_BYTES);
10133 vmcs12->guest_cs_ar_bytes = vmcs_read32(GUEST_CS_AR_BYTES);
10134 vmcs12->guest_ss_ar_bytes = vmcs_read32(GUEST_SS_AR_BYTES);
10135 vmcs12->guest_ds_ar_bytes = vmcs_read32(GUEST_DS_AR_BYTES);
10136 vmcs12->guest_fs_ar_bytes = vmcs_read32(GUEST_FS_AR_BYTES);
10137 vmcs12->guest_gs_ar_bytes = vmcs_read32(GUEST_GS_AR_BYTES);
10138 vmcs12->guest_ldtr_ar_bytes = vmcs_read32(GUEST_LDTR_AR_BYTES);
10139 vmcs12->guest_tr_ar_bytes = vmcs_read32(GUEST_TR_AR_BYTES);
10140 vmcs12->guest_es_base = vmcs_readl(GUEST_ES_BASE);
10141 vmcs12->guest_cs_base = vmcs_readl(GUEST_CS_BASE);
10142 vmcs12->guest_ss_base = vmcs_readl(GUEST_SS_BASE);
10143 vmcs12->guest_ds_base = vmcs_readl(GUEST_DS_BASE);
10144 vmcs12->guest_fs_base = vmcs_readl(GUEST_FS_BASE);
10145 vmcs12->guest_gs_base = vmcs_readl(GUEST_GS_BASE);
10146 vmcs12->guest_ldtr_base = vmcs_readl(GUEST_LDTR_BASE);
10147 vmcs12->guest_tr_base = vmcs_readl(GUEST_TR_BASE);
10148 vmcs12->guest_gdtr_base = vmcs_readl(GUEST_GDTR_BASE);
10149 vmcs12->guest_idtr_base = vmcs_readl(GUEST_IDTR_BASE);
10151 vmcs12->guest_interruptibility_info =
10152 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
10153 vmcs12->guest_pending_dbg_exceptions =
10154 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS);
10155 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
10156 vmcs12->guest_activity_state = GUEST_ACTIVITY_HLT;
10158 vmcs12->guest_activity_state = GUEST_ACTIVITY_ACTIVE;
10160 if (nested_cpu_has_preemption_timer(vmcs12)) {
10161 if (vmcs12->vm_exit_controls &
10162 VM_EXIT_SAVE_VMX_PREEMPTION_TIMER)
10163 vmcs12->vmx_preemption_timer_value =
10164 vmx_get_preemption_timer_value(vcpu);
10165 hrtimer_cancel(&to_vmx(vcpu)->nested.preemption_timer);
10169 * In some cases (usually, nested EPT), L2 is allowed to change its
10170 * own CR3 without exiting. If it has changed it, we must keep it.
10171 * Of course, if L0 is using shadow page tables, GUEST_CR3 was defined
10172 * by L0, not L1 or L2, so we mustn't unconditionally copy it to vmcs12.
10174 * Additionally, restore L2's PDPTR to vmcs12.
10177 vmcs12->guest_cr3 = vmcs_read64(GUEST_CR3);
10178 vmcs12->guest_pdptr0 = vmcs_read64(GUEST_PDPTR0);
10179 vmcs12->guest_pdptr1 = vmcs_read64(GUEST_PDPTR1);
10180 vmcs12->guest_pdptr2 = vmcs_read64(GUEST_PDPTR2);
10181 vmcs12->guest_pdptr3 = vmcs_read64(GUEST_PDPTR3);
10184 if (nested_cpu_has_vid(vmcs12))
10185 vmcs12->guest_intr_status = vmcs_read16(GUEST_INTR_STATUS);
10187 vmcs12->vm_entry_controls =
10188 (vmcs12->vm_entry_controls & ~VM_ENTRY_IA32E_MODE) |
10189 (vm_entry_controls_get(to_vmx(vcpu)) & VM_ENTRY_IA32E_MODE);
10191 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_DEBUG_CONTROLS) {
10192 kvm_get_dr(vcpu, 7, (unsigned long *)&vmcs12->guest_dr7);
10193 vmcs12->guest_ia32_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
10196 /* TODO: These cannot have changed unless we have MSR bitmaps and
10197 * the relevant bit asks not to trap the change */
10198 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_PAT)
10199 vmcs12->guest_ia32_pat = vmcs_read64(GUEST_IA32_PAT);
10200 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_EFER)
10201 vmcs12->guest_ia32_efer = vcpu->arch.efer;
10202 vmcs12->guest_sysenter_cs = vmcs_read32(GUEST_SYSENTER_CS);
10203 vmcs12->guest_sysenter_esp = vmcs_readl(GUEST_SYSENTER_ESP);
10204 vmcs12->guest_sysenter_eip = vmcs_readl(GUEST_SYSENTER_EIP);
10205 if (vmx_mpx_supported())
10206 vmcs12->guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
10207 if (nested_cpu_has_xsaves(vmcs12))
10208 vmcs12->xss_exit_bitmap = vmcs_read64(XSS_EXIT_BITMAP);
10210 /* update exit information fields: */
10212 vmcs12->vm_exit_reason = exit_reason;
10213 vmcs12->exit_qualification = exit_qualification;
10215 vmcs12->vm_exit_intr_info = exit_intr_info;
10216 if ((vmcs12->vm_exit_intr_info &
10217 (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK)) ==
10218 (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK))
10219 vmcs12->vm_exit_intr_error_code =
10220 vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
10221 vmcs12->idt_vectoring_info_field = 0;
10222 vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
10223 vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
10225 if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY)) {
10226 /* vm_entry_intr_info_field is cleared on exit. Emulate this
10227 * instead of reading the real value. */
10228 vmcs12->vm_entry_intr_info_field &= ~INTR_INFO_VALID_MASK;
10231 * Transfer the event that L0 or L1 may wanted to inject into
10232 * L2 to IDT_VECTORING_INFO_FIELD.
10234 vmcs12_save_pending_event(vcpu, vmcs12);
10238 * Drop what we picked up for L2 via vmx_complete_interrupts. It is
10239 * preserved above and would only end up incorrectly in L1.
10241 vcpu->arch.nmi_injected = false;
10242 kvm_clear_exception_queue(vcpu);
10243 kvm_clear_interrupt_queue(vcpu);
10247 * A part of what we need to when the nested L2 guest exits and we want to
10248 * run its L1 parent, is to reset L1's guest state to the host state specified
10250 * This function is to be called not only on normal nested exit, but also on
10251 * a nested entry failure, as explained in Intel's spec, 3B.23.7 ("VM-Entry
10252 * Failures During or After Loading Guest State").
10253 * This function should be called when the active VMCS is L1's (vmcs01).
10255 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
10256 struct vmcs12 *vmcs12)
10258 struct kvm_segment seg;
10260 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER)
10261 vcpu->arch.efer = vmcs12->host_ia32_efer;
10262 else if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
10263 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
10265 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
10266 vmx_set_efer(vcpu, vcpu->arch.efer);
10268 kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->host_rsp);
10269 kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->host_rip);
10270 vmx_set_rflags(vcpu, X86_EFLAGS_FIXED);
10272 * Note that calling vmx_set_cr0 is important, even if cr0 hasn't
10273 * actually changed, because it depends on the current state of
10274 * fpu_active (which may have changed).
10275 * Note that vmx_set_cr0 refers to efer set above.
10277 vmx_set_cr0(vcpu, vmcs12->host_cr0);
10279 * If we did fpu_activate()/fpu_deactivate() during L2's run, we need
10280 * to apply the same changes to L1's vmcs. We just set cr0 correctly,
10281 * but we also need to update cr0_guest_host_mask and exception_bitmap.
10283 update_exception_bitmap(vcpu);
10284 vcpu->arch.cr0_guest_owned_bits = (vcpu->fpu_active ? X86_CR0_TS : 0);
10285 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
10288 * Note that CR4_GUEST_HOST_MASK is already set in the original vmcs01
10289 * (KVM doesn't change it)- no reason to call set_cr4_guest_host_mask();
10291 vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
10292 kvm_set_cr4(vcpu, vmcs12->host_cr4);
10294 nested_ept_uninit_mmu_context(vcpu);
10296 kvm_set_cr3(vcpu, vmcs12->host_cr3);
10297 kvm_mmu_reset_context(vcpu);
10300 vcpu->arch.walk_mmu->inject_page_fault = kvm_inject_page_fault;
10304 * Trivially support vpid by letting L2s share their parent
10305 * L1's vpid. TODO: move to a more elaborate solution, giving
10306 * each L2 its own vpid and exposing the vpid feature to L1.
10308 vmx_flush_tlb(vcpu);
10312 vmcs_write32(GUEST_SYSENTER_CS, vmcs12->host_ia32_sysenter_cs);
10313 vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->host_ia32_sysenter_esp);
10314 vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->host_ia32_sysenter_eip);
10315 vmcs_writel(GUEST_IDTR_BASE, vmcs12->host_idtr_base);
10316 vmcs_writel(GUEST_GDTR_BASE, vmcs12->host_gdtr_base);
10318 /* If not VM_EXIT_CLEAR_BNDCFGS, the L2 value propagates to L1. */
10319 if (vmcs12->vm_exit_controls & VM_EXIT_CLEAR_BNDCFGS)
10320 vmcs_write64(GUEST_BNDCFGS, 0);
10322 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) {
10323 vmcs_write64(GUEST_IA32_PAT, vmcs12->host_ia32_pat);
10324 vcpu->arch.pat = vmcs12->host_ia32_pat;
10326 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
10327 vmcs_write64(GUEST_IA32_PERF_GLOBAL_CTRL,
10328 vmcs12->host_ia32_perf_global_ctrl);
10330 /* Set L1 segment info according to Intel SDM
10331 27.5.2 Loading Host Segment and Descriptor-Table Registers */
10332 seg = (struct kvm_segment) {
10334 .limit = 0xFFFFFFFF,
10335 .selector = vmcs12->host_cs_selector,
10341 if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
10345 vmx_set_segment(vcpu, &seg, VCPU_SREG_CS);
10346 seg = (struct kvm_segment) {
10348 .limit = 0xFFFFFFFF,
10355 seg.selector = vmcs12->host_ds_selector;
10356 vmx_set_segment(vcpu, &seg, VCPU_SREG_DS);
10357 seg.selector = vmcs12->host_es_selector;
10358 vmx_set_segment(vcpu, &seg, VCPU_SREG_ES);
10359 seg.selector = vmcs12->host_ss_selector;
10360 vmx_set_segment(vcpu, &seg, VCPU_SREG_SS);
10361 seg.selector = vmcs12->host_fs_selector;
10362 seg.base = vmcs12->host_fs_base;
10363 vmx_set_segment(vcpu, &seg, VCPU_SREG_FS);
10364 seg.selector = vmcs12->host_gs_selector;
10365 seg.base = vmcs12->host_gs_base;
10366 vmx_set_segment(vcpu, &seg, VCPU_SREG_GS);
10367 seg = (struct kvm_segment) {
10368 .base = vmcs12->host_tr_base,
10370 .selector = vmcs12->host_tr_selector,
10374 vmx_set_segment(vcpu, &seg, VCPU_SREG_TR);
10376 kvm_set_dr(vcpu, 7, 0x400);
10377 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
10379 if (cpu_has_vmx_msr_bitmap())
10380 vmx_set_msr_bitmap(vcpu);
10382 if (nested_vmx_load_msr(vcpu, vmcs12->vm_exit_msr_load_addr,
10383 vmcs12->vm_exit_msr_load_count))
10384 nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
10388 * Emulate an exit from nested guest (L2) to L1, i.e., prepare to run L1
10389 * and modify vmcs12 to make it see what it would expect to see there if
10390 * L2 was its real guest. Must only be called when in L2 (is_guest_mode())
10392 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
10393 u32 exit_intr_info,
10394 unsigned long exit_qualification)
10396 struct vcpu_vmx *vmx = to_vmx(vcpu);
10397 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
10399 /* trying to cancel vmlaunch/vmresume is a bug */
10400 WARN_ON_ONCE(vmx->nested.nested_run_pending);
10402 leave_guest_mode(vcpu);
10403 prepare_vmcs12(vcpu, vmcs12, exit_reason, exit_intr_info,
10404 exit_qualification);
10406 if (nested_vmx_store_msr(vcpu, vmcs12->vm_exit_msr_store_addr,
10407 vmcs12->vm_exit_msr_store_count))
10408 nested_vmx_abort(vcpu, VMX_ABORT_SAVE_GUEST_MSR_FAIL);
10410 vmx_load_vmcs01(vcpu);
10412 if ((exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT)
10413 && nested_exit_intr_ack_set(vcpu)) {
10414 int irq = kvm_cpu_get_interrupt(vcpu);
10416 vmcs12->vm_exit_intr_info = irq |
10417 INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR;
10420 trace_kvm_nested_vmexit_inject(vmcs12->vm_exit_reason,
10421 vmcs12->exit_qualification,
10422 vmcs12->idt_vectoring_info_field,
10423 vmcs12->vm_exit_intr_info,
10424 vmcs12->vm_exit_intr_error_code,
10427 vm_entry_controls_init(vmx, vmcs_read32(VM_ENTRY_CONTROLS));
10428 vm_exit_controls_init(vmx, vmcs_read32(VM_EXIT_CONTROLS));
10429 vmx_segment_cache_clear(vmx);
10431 /* if no vmcs02 cache requested, remove the one we used */
10432 if (VMCS02_POOL_SIZE == 0)
10433 nested_free_vmcs02(vmx, vmx->nested.current_vmptr);
10435 load_vmcs12_host_state(vcpu, vmcs12);
10437 /* Update TSC_OFFSET if TSC was changed while L2 ran */
10438 vmcs_write64(TSC_OFFSET, vmx->nested.vmcs01_tsc_offset);
10440 /* This is needed for same reason as it was needed in prepare_vmcs02 */
10443 /* Unpin physical memory we referred to in vmcs02 */
10444 if (vmx->nested.apic_access_page) {
10445 nested_release_page(vmx->nested.apic_access_page);
10446 vmx->nested.apic_access_page = NULL;
10448 if (vmx->nested.virtual_apic_page) {
10449 nested_release_page(vmx->nested.virtual_apic_page);
10450 vmx->nested.virtual_apic_page = NULL;
10452 if (vmx->nested.pi_desc_page) {
10453 kunmap(vmx->nested.pi_desc_page);
10454 nested_release_page(vmx->nested.pi_desc_page);
10455 vmx->nested.pi_desc_page = NULL;
10456 vmx->nested.pi_desc = NULL;
10460 * We are now running in L2, mmu_notifier will force to reload the
10461 * page's hpa for L2 vmcs. Need to reload it for L1 before entering L1.
10463 kvm_vcpu_reload_apic_access_page(vcpu);
10466 * Exiting from L2 to L1, we're now back to L1 which thinks it just
10467 * finished a VMLAUNCH or VMRESUME instruction, so we need to set the
10468 * success or failure flag accordingly.
10470 if (unlikely(vmx->fail)) {
10472 nested_vmx_failValid(vcpu, vmcs_read32(VM_INSTRUCTION_ERROR));
10474 nested_vmx_succeed(vcpu);
10475 if (enable_shadow_vmcs)
10476 vmx->nested.sync_shadow_vmcs = true;
10478 /* in case we halted in L2 */
10479 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10483 * Forcibly leave nested mode in order to be able to reset the VCPU later on.
10485 static void vmx_leave_nested(struct kvm_vcpu *vcpu)
10487 if (is_guest_mode(vcpu))
10488 nested_vmx_vmexit(vcpu, -1, 0, 0);
10489 free_nested(to_vmx(vcpu));
10493 * L1's failure to enter L2 is a subset of a normal exit, as explained in
10494 * 23.7 "VM-entry failures during or after loading guest state" (this also
10495 * lists the acceptable exit-reason and exit-qualification parameters).
10496 * It should only be called before L2 actually succeeded to run, and when
10497 * vmcs01 is current (it doesn't leave_guest_mode() or switch vmcss).
10499 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
10500 struct vmcs12 *vmcs12,
10501 u32 reason, unsigned long qualification)
10503 load_vmcs12_host_state(vcpu, vmcs12);
10504 vmcs12->vm_exit_reason = reason | VMX_EXIT_REASONS_FAILED_VMENTRY;
10505 vmcs12->exit_qualification = qualification;
10506 nested_vmx_succeed(vcpu);
10507 if (enable_shadow_vmcs)
10508 to_vmx(vcpu)->nested.sync_shadow_vmcs = true;
10511 static int vmx_check_intercept(struct kvm_vcpu *vcpu,
10512 struct x86_instruction_info *info,
10513 enum x86_intercept_stage stage)
10515 return X86EMUL_CONTINUE;
10518 static void vmx_sched_in(struct kvm_vcpu *vcpu, int cpu)
10521 shrink_ple_window(vcpu);
10524 static void vmx_slot_enable_log_dirty(struct kvm *kvm,
10525 struct kvm_memory_slot *slot)
10527 kvm_mmu_slot_leaf_clear_dirty(kvm, slot);
10528 kvm_mmu_slot_largepage_remove_write_access(kvm, slot);
10531 static void vmx_slot_disable_log_dirty(struct kvm *kvm,
10532 struct kvm_memory_slot *slot)
10534 kvm_mmu_slot_set_dirty(kvm, slot);
10537 static void vmx_flush_log_dirty(struct kvm *kvm)
10539 kvm_flush_pml_buffers(kvm);
10542 static void vmx_enable_log_dirty_pt_masked(struct kvm *kvm,
10543 struct kvm_memory_slot *memslot,
10544 gfn_t offset, unsigned long mask)
10546 kvm_mmu_clear_dirty_pt_masked(kvm, memslot, offset, mask);
10550 * This routine does the following things for vCPU which is going
10551 * to be blocked if VT-d PI is enabled.
10552 * - Store the vCPU to the wakeup list, so when interrupts happen
10553 * we can find the right vCPU to wake up.
10554 * - Change the Posted-interrupt descriptor as below:
10555 * 'NDST' <-- vcpu->pre_pcpu
10556 * 'NV' <-- POSTED_INTR_WAKEUP_VECTOR
10557 * - If 'ON' is set during this process, which means at least one
10558 * interrupt is posted for this vCPU, we cannot block it, in
10559 * this case, return 1, otherwise, return 0.
10562 static int vmx_pre_block(struct kvm_vcpu *vcpu)
10564 unsigned long flags;
10566 struct pi_desc old, new;
10567 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
10569 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
10570 !irq_remapping_cap(IRQ_POSTING_CAP))
10573 vcpu->pre_pcpu = vcpu->cpu;
10574 spin_lock_irqsave(&per_cpu(blocked_vcpu_on_cpu_lock,
10575 vcpu->pre_pcpu), flags);
10576 list_add_tail(&vcpu->blocked_vcpu_list,
10577 &per_cpu(blocked_vcpu_on_cpu,
10579 spin_unlock_irqrestore(&per_cpu(blocked_vcpu_on_cpu_lock,
10580 vcpu->pre_pcpu), flags);
10583 old.control = new.control = pi_desc->control;
10586 * We should not block the vCPU if
10587 * an interrupt is posted for it.
10589 if (pi_test_on(pi_desc) == 1) {
10590 spin_lock_irqsave(&per_cpu(blocked_vcpu_on_cpu_lock,
10591 vcpu->pre_pcpu), flags);
10592 list_del(&vcpu->blocked_vcpu_list);
10593 spin_unlock_irqrestore(
10594 &per_cpu(blocked_vcpu_on_cpu_lock,
10595 vcpu->pre_pcpu), flags);
10596 vcpu->pre_pcpu = -1;
10601 WARN((pi_desc->sn == 1),
10602 "Warning: SN field of posted-interrupts "
10603 "is set before blocking\n");
10606 * Since vCPU can be preempted during this process,
10607 * vcpu->cpu could be different with pre_pcpu, we
10608 * need to set pre_pcpu as the destination of wakeup
10609 * notification event, then we can find the right vCPU
10610 * to wakeup in wakeup handler if interrupts happen
10611 * when the vCPU is in blocked state.
10613 dest = cpu_physical_id(vcpu->pre_pcpu);
10615 if (x2apic_enabled())
10618 new.ndst = (dest << 8) & 0xFF00;
10620 /* set 'NV' to 'wakeup vector' */
10621 new.nv = POSTED_INTR_WAKEUP_VECTOR;
10622 } while (cmpxchg(&pi_desc->control, old.control,
10623 new.control) != old.control);
10628 static void vmx_post_block(struct kvm_vcpu *vcpu)
10630 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
10631 struct pi_desc old, new;
10633 unsigned long flags;
10635 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
10636 !irq_remapping_cap(IRQ_POSTING_CAP))
10640 old.control = new.control = pi_desc->control;
10642 dest = cpu_physical_id(vcpu->cpu);
10644 if (x2apic_enabled())
10647 new.ndst = (dest << 8) & 0xFF00;
10649 /* Allow posting non-urgent interrupts */
10652 /* set 'NV' to 'notification vector' */
10653 new.nv = POSTED_INTR_VECTOR;
10654 } while (cmpxchg(&pi_desc->control, old.control,
10655 new.control) != old.control);
10657 if(vcpu->pre_pcpu != -1) {
10659 &per_cpu(blocked_vcpu_on_cpu_lock,
10660 vcpu->pre_pcpu), flags);
10661 list_del(&vcpu->blocked_vcpu_list);
10662 spin_unlock_irqrestore(
10663 &per_cpu(blocked_vcpu_on_cpu_lock,
10664 vcpu->pre_pcpu), flags);
10665 vcpu->pre_pcpu = -1;
10670 * vmx_update_pi_irte - set IRTE for Posted-Interrupts
10673 * @host_irq: host irq of the interrupt
10674 * @guest_irq: gsi of the interrupt
10675 * @set: set or unset PI
10676 * returns 0 on success, < 0 on failure
10678 static int vmx_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
10679 uint32_t guest_irq, bool set)
10681 struct kvm_kernel_irq_routing_entry *e;
10682 struct kvm_irq_routing_table *irq_rt;
10683 struct kvm_lapic_irq irq;
10684 struct kvm_vcpu *vcpu;
10685 struct vcpu_data vcpu_info;
10686 int idx, ret = -EINVAL;
10688 if (!kvm_arch_has_assigned_device(kvm) ||
10689 !irq_remapping_cap(IRQ_POSTING_CAP))
10692 idx = srcu_read_lock(&kvm->irq_srcu);
10693 irq_rt = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu);
10694 BUG_ON(guest_irq >= irq_rt->nr_rt_entries);
10696 hlist_for_each_entry(e, &irq_rt->map[guest_irq], link) {
10697 if (e->type != KVM_IRQ_ROUTING_MSI)
10700 * VT-d PI cannot support posting multicast/broadcast
10701 * interrupts to a vCPU, we still use interrupt remapping
10702 * for these kind of interrupts.
10704 * For lowest-priority interrupts, we only support
10705 * those with single CPU as the destination, e.g. user
10706 * configures the interrupts via /proc/irq or uses
10707 * irqbalance to make the interrupts single-CPU.
10709 * We will support full lowest-priority interrupt later.
10712 kvm_set_msi_irq(e, &irq);
10713 if (!kvm_intr_is_single_vcpu(kvm, &irq, &vcpu))
10716 vcpu_info.pi_desc_addr = __pa(vcpu_to_pi_desc(vcpu));
10717 vcpu_info.vector = irq.vector;
10719 trace_kvm_pi_irte_update(vcpu->vcpu_id, e->gsi,
10720 vcpu_info.vector, vcpu_info.pi_desc_addr, set);
10723 ret = irq_set_vcpu_affinity(host_irq, &vcpu_info);
10725 /* suppress notification event before unposting */
10726 pi_set_sn(vcpu_to_pi_desc(vcpu));
10727 ret = irq_set_vcpu_affinity(host_irq, NULL);
10728 pi_clear_sn(vcpu_to_pi_desc(vcpu));
10732 printk(KERN_INFO "%s: failed to update PI IRTE\n",
10740 srcu_read_unlock(&kvm->irq_srcu, idx);
10744 static struct kvm_x86_ops vmx_x86_ops = {
10745 .cpu_has_kvm_support = cpu_has_kvm_support,
10746 .disabled_by_bios = vmx_disabled_by_bios,
10747 .hardware_setup = hardware_setup,
10748 .hardware_unsetup = hardware_unsetup,
10749 .check_processor_compatibility = vmx_check_processor_compat,
10750 .hardware_enable = hardware_enable,
10751 .hardware_disable = hardware_disable,
10752 .cpu_has_accelerated_tpr = report_flexpriority,
10753 .cpu_has_high_real_mode_segbase = vmx_has_high_real_mode_segbase,
10755 .vcpu_create = vmx_create_vcpu,
10756 .vcpu_free = vmx_free_vcpu,
10757 .vcpu_reset = vmx_vcpu_reset,
10759 .prepare_guest_switch = vmx_save_host_state,
10760 .vcpu_load = vmx_vcpu_load,
10761 .vcpu_put = vmx_vcpu_put,
10763 .update_bp_intercept = update_exception_bitmap,
10764 .get_msr = vmx_get_msr,
10765 .set_msr = vmx_set_msr,
10766 .get_segment_base = vmx_get_segment_base,
10767 .get_segment = vmx_get_segment,
10768 .set_segment = vmx_set_segment,
10769 .get_cpl = vmx_get_cpl,
10770 .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
10771 .decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
10772 .decache_cr3 = vmx_decache_cr3,
10773 .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
10774 .set_cr0 = vmx_set_cr0,
10775 .set_cr3 = vmx_set_cr3,
10776 .set_cr4 = vmx_set_cr4,
10777 .set_efer = vmx_set_efer,
10778 .get_idt = vmx_get_idt,
10779 .set_idt = vmx_set_idt,
10780 .get_gdt = vmx_get_gdt,
10781 .set_gdt = vmx_set_gdt,
10782 .get_dr6 = vmx_get_dr6,
10783 .set_dr6 = vmx_set_dr6,
10784 .set_dr7 = vmx_set_dr7,
10785 .sync_dirty_debug_regs = vmx_sync_dirty_debug_regs,
10786 .cache_reg = vmx_cache_reg,
10787 .get_rflags = vmx_get_rflags,
10788 .set_rflags = vmx_set_rflags,
10789 .fpu_activate = vmx_fpu_activate,
10790 .fpu_deactivate = vmx_fpu_deactivate,
10792 .tlb_flush = vmx_flush_tlb,
10794 .run = vmx_vcpu_run,
10795 .handle_exit = vmx_handle_exit,
10796 .skip_emulated_instruction = skip_emulated_instruction,
10797 .set_interrupt_shadow = vmx_set_interrupt_shadow,
10798 .get_interrupt_shadow = vmx_get_interrupt_shadow,
10799 .patch_hypercall = vmx_patch_hypercall,
10800 .set_irq = vmx_inject_irq,
10801 .set_nmi = vmx_inject_nmi,
10802 .queue_exception = vmx_queue_exception,
10803 .cancel_injection = vmx_cancel_injection,
10804 .interrupt_allowed = vmx_interrupt_allowed,
10805 .nmi_allowed = vmx_nmi_allowed,
10806 .get_nmi_mask = vmx_get_nmi_mask,
10807 .set_nmi_mask = vmx_set_nmi_mask,
10808 .enable_nmi_window = enable_nmi_window,
10809 .enable_irq_window = enable_irq_window,
10810 .update_cr8_intercept = update_cr8_intercept,
10811 .set_virtual_x2apic_mode = vmx_set_virtual_x2apic_mode,
10812 .set_apic_access_page_addr = vmx_set_apic_access_page_addr,
10813 .cpu_uses_apicv = vmx_cpu_uses_apicv,
10814 .load_eoi_exitmap = vmx_load_eoi_exitmap,
10815 .hwapic_irr_update = vmx_hwapic_irr_update,
10816 .hwapic_isr_update = vmx_hwapic_isr_update,
10817 .sync_pir_to_irr = vmx_sync_pir_to_irr,
10818 .deliver_posted_interrupt = vmx_deliver_posted_interrupt,
10820 .set_tss_addr = vmx_set_tss_addr,
10821 .get_tdp_level = get_ept_level,
10822 .get_mt_mask = vmx_get_mt_mask,
10824 .get_exit_info = vmx_get_exit_info,
10826 .get_lpage_level = vmx_get_lpage_level,
10828 .cpuid_update = vmx_cpuid_update,
10830 .rdtscp_supported = vmx_rdtscp_supported,
10831 .invpcid_supported = vmx_invpcid_supported,
10833 .set_supported_cpuid = vmx_set_supported_cpuid,
10835 .has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
10837 .read_tsc_offset = vmx_read_tsc_offset,
10838 .write_tsc_offset = vmx_write_tsc_offset,
10839 .adjust_tsc_offset_guest = vmx_adjust_tsc_offset_guest,
10840 .read_l1_tsc = vmx_read_l1_tsc,
10842 .set_tdp_cr3 = vmx_set_cr3,
10844 .check_intercept = vmx_check_intercept,
10845 .handle_external_intr = vmx_handle_external_intr,
10846 .mpx_supported = vmx_mpx_supported,
10847 .xsaves_supported = vmx_xsaves_supported,
10849 .check_nested_events = vmx_check_nested_events,
10851 .sched_in = vmx_sched_in,
10853 .slot_enable_log_dirty = vmx_slot_enable_log_dirty,
10854 .slot_disable_log_dirty = vmx_slot_disable_log_dirty,
10855 .flush_log_dirty = vmx_flush_log_dirty,
10856 .enable_log_dirty_pt_masked = vmx_enable_log_dirty_pt_masked,
10858 .pre_block = vmx_pre_block,
10859 .post_block = vmx_post_block,
10861 .pmu_ops = &intel_pmu_ops,
10863 .update_pi_irte = vmx_update_pi_irte,
10866 static int __init vmx_init(void)
10868 int r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx),
10869 __alignof__(struct vcpu_vmx), THIS_MODULE);
10873 #ifdef CONFIG_KEXEC_CORE
10874 rcu_assign_pointer(crash_vmclear_loaded_vmcss,
10875 crash_vmclear_local_loaded_vmcss);
10881 static void __exit vmx_exit(void)
10883 #ifdef CONFIG_KEXEC_CORE
10884 RCU_INIT_POINTER(crash_vmclear_loaded_vmcss, NULL);
10891 module_init(vmx_init)
10892 module_exit(vmx_exit)