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_GUEST_CR0_MASK (X86_CR0_NW | X86_CR0_CD)
111 #define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST (X86_CR0_WP | X86_CR0_NE)
112 #define KVM_VM_CR0_ALWAYS_ON \
113 (KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
114 #define KVM_CR4_GUEST_OWNED_BITS \
115 (X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR \
116 | X86_CR4_OSXMMEXCPT | X86_CR4_TSD)
118 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
119 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
121 #define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
123 #define VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE 5
126 * These 2 parameters are used to config the controls for Pause-Loop Exiting:
127 * ple_gap: upper bound on the amount of time between two successive
128 * executions of PAUSE in a loop. Also indicate if ple enabled.
129 * According to test, this time is usually smaller than 128 cycles.
130 * ple_window: upper bound on the amount of time a guest is allowed to execute
131 * in a PAUSE loop. Tests indicate that most spinlocks are held for
132 * less than 2^12 cycles
133 * Time is measured based on a counter that runs at the same rate as the TSC,
134 * refer SDM volume 3b section 21.6.13 & 22.1.3.
136 #define KVM_VMX_DEFAULT_PLE_GAP 128
137 #define KVM_VMX_DEFAULT_PLE_WINDOW 4096
138 #define KVM_VMX_DEFAULT_PLE_WINDOW_GROW 2
139 #define KVM_VMX_DEFAULT_PLE_WINDOW_SHRINK 0
140 #define KVM_VMX_DEFAULT_PLE_WINDOW_MAX \
141 INT_MAX / KVM_VMX_DEFAULT_PLE_WINDOW_GROW
143 static int ple_gap = KVM_VMX_DEFAULT_PLE_GAP;
144 module_param(ple_gap, int, S_IRUGO);
146 static int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
147 module_param(ple_window, int, S_IRUGO);
149 /* Default doubles per-vcpu window every exit. */
150 static int ple_window_grow = KVM_VMX_DEFAULT_PLE_WINDOW_GROW;
151 module_param(ple_window_grow, int, S_IRUGO);
153 /* Default resets per-vcpu window every exit to ple_window. */
154 static int ple_window_shrink = KVM_VMX_DEFAULT_PLE_WINDOW_SHRINK;
155 module_param(ple_window_shrink, int, S_IRUGO);
157 /* Default is to compute the maximum so we can never overflow. */
158 static int ple_window_actual_max = KVM_VMX_DEFAULT_PLE_WINDOW_MAX;
159 static int ple_window_max = KVM_VMX_DEFAULT_PLE_WINDOW_MAX;
160 module_param(ple_window_max, int, S_IRUGO);
162 extern const ulong vmx_return;
164 #define NR_AUTOLOAD_MSRS 8
165 #define VMCS02_POOL_SIZE 1
174 * Track a VMCS that may be loaded on a certain CPU. If it is (cpu!=-1), also
175 * remember whether it was VMLAUNCHed, and maintain a linked list of all VMCSs
176 * loaded on this CPU (so we can clear them if the CPU goes down).
182 struct list_head loaded_vmcss_on_cpu_link;
185 struct shared_msr_entry {
192 * struct vmcs12 describes the state that our guest hypervisor (L1) keeps for a
193 * single nested guest (L2), hence the name vmcs12. Any VMX implementation has
194 * a VMCS structure, and vmcs12 is our emulated VMX's VMCS. This structure is
195 * stored in guest memory specified by VMPTRLD, but is opaque to the guest,
196 * which must access it using VMREAD/VMWRITE/VMCLEAR instructions.
197 * More than one of these structures may exist, if L1 runs multiple L2 guests.
198 * nested_vmx_run() will use the data here to build a vmcs02: a VMCS for the
199 * underlying hardware which will be used to run L2.
200 * This structure is packed to ensure that its layout is identical across
201 * machines (necessary for live migration).
202 * If there are changes in this struct, VMCS12_REVISION must be changed.
204 typedef u64 natural_width;
205 struct __packed vmcs12 {
206 /* According to the Intel spec, a VMCS region must start with the
207 * following two fields. Then follow implementation-specific data.
212 u32 launch_state; /* set to 0 by VMCLEAR, to 1 by VMLAUNCH */
213 u32 padding[7]; /* room for future expansion */
218 u64 vm_exit_msr_store_addr;
219 u64 vm_exit_msr_load_addr;
220 u64 vm_entry_msr_load_addr;
222 u64 virtual_apic_page_addr;
223 u64 apic_access_addr;
224 u64 posted_intr_desc_addr;
226 u64 eoi_exit_bitmap0;
227 u64 eoi_exit_bitmap1;
228 u64 eoi_exit_bitmap2;
229 u64 eoi_exit_bitmap3;
231 u64 guest_physical_address;
232 u64 vmcs_link_pointer;
233 u64 guest_ia32_debugctl;
236 u64 guest_ia32_perf_global_ctrl;
244 u64 host_ia32_perf_global_ctrl;
245 u64 padding64[8]; /* room for future expansion */
247 * To allow migration of L1 (complete with its L2 guests) between
248 * machines of different natural widths (32 or 64 bit), we cannot have
249 * unsigned long fields with no explict size. We use u64 (aliased
250 * natural_width) instead. Luckily, x86 is little-endian.
252 natural_width cr0_guest_host_mask;
253 natural_width cr4_guest_host_mask;
254 natural_width cr0_read_shadow;
255 natural_width cr4_read_shadow;
256 natural_width cr3_target_value0;
257 natural_width cr3_target_value1;
258 natural_width cr3_target_value2;
259 natural_width cr3_target_value3;
260 natural_width exit_qualification;
261 natural_width guest_linear_address;
262 natural_width guest_cr0;
263 natural_width guest_cr3;
264 natural_width guest_cr4;
265 natural_width guest_es_base;
266 natural_width guest_cs_base;
267 natural_width guest_ss_base;
268 natural_width guest_ds_base;
269 natural_width guest_fs_base;
270 natural_width guest_gs_base;
271 natural_width guest_ldtr_base;
272 natural_width guest_tr_base;
273 natural_width guest_gdtr_base;
274 natural_width guest_idtr_base;
275 natural_width guest_dr7;
276 natural_width guest_rsp;
277 natural_width guest_rip;
278 natural_width guest_rflags;
279 natural_width guest_pending_dbg_exceptions;
280 natural_width guest_sysenter_esp;
281 natural_width guest_sysenter_eip;
282 natural_width host_cr0;
283 natural_width host_cr3;
284 natural_width host_cr4;
285 natural_width host_fs_base;
286 natural_width host_gs_base;
287 natural_width host_tr_base;
288 natural_width host_gdtr_base;
289 natural_width host_idtr_base;
290 natural_width host_ia32_sysenter_esp;
291 natural_width host_ia32_sysenter_eip;
292 natural_width host_rsp;
293 natural_width host_rip;
294 natural_width paddingl[8]; /* room for future expansion */
295 u32 pin_based_vm_exec_control;
296 u32 cpu_based_vm_exec_control;
297 u32 exception_bitmap;
298 u32 page_fault_error_code_mask;
299 u32 page_fault_error_code_match;
300 u32 cr3_target_count;
301 u32 vm_exit_controls;
302 u32 vm_exit_msr_store_count;
303 u32 vm_exit_msr_load_count;
304 u32 vm_entry_controls;
305 u32 vm_entry_msr_load_count;
306 u32 vm_entry_intr_info_field;
307 u32 vm_entry_exception_error_code;
308 u32 vm_entry_instruction_len;
310 u32 secondary_vm_exec_control;
311 u32 vm_instruction_error;
313 u32 vm_exit_intr_info;
314 u32 vm_exit_intr_error_code;
315 u32 idt_vectoring_info_field;
316 u32 idt_vectoring_error_code;
317 u32 vm_exit_instruction_len;
318 u32 vmx_instruction_info;
325 u32 guest_ldtr_limit;
327 u32 guest_gdtr_limit;
328 u32 guest_idtr_limit;
329 u32 guest_es_ar_bytes;
330 u32 guest_cs_ar_bytes;
331 u32 guest_ss_ar_bytes;
332 u32 guest_ds_ar_bytes;
333 u32 guest_fs_ar_bytes;
334 u32 guest_gs_ar_bytes;
335 u32 guest_ldtr_ar_bytes;
336 u32 guest_tr_ar_bytes;
337 u32 guest_interruptibility_info;
338 u32 guest_activity_state;
339 u32 guest_sysenter_cs;
340 u32 host_ia32_sysenter_cs;
341 u32 vmx_preemption_timer_value;
342 u32 padding32[7]; /* room for future expansion */
343 u16 virtual_processor_id;
345 u16 guest_es_selector;
346 u16 guest_cs_selector;
347 u16 guest_ss_selector;
348 u16 guest_ds_selector;
349 u16 guest_fs_selector;
350 u16 guest_gs_selector;
351 u16 guest_ldtr_selector;
352 u16 guest_tr_selector;
353 u16 guest_intr_status;
354 u16 host_es_selector;
355 u16 host_cs_selector;
356 u16 host_ss_selector;
357 u16 host_ds_selector;
358 u16 host_fs_selector;
359 u16 host_gs_selector;
360 u16 host_tr_selector;
364 * VMCS12_REVISION is an arbitrary id that should be changed if the content or
365 * layout of struct vmcs12 is changed. MSR_IA32_VMX_BASIC returns this id, and
366 * VMPTRLD verifies that the VMCS region that L1 is loading contains this id.
368 #define VMCS12_REVISION 0x11e57ed0
371 * VMCS12_SIZE is the number of bytes L1 should allocate for the VMXON region
372 * and any VMCS region. Although only sizeof(struct vmcs12) are used by the
373 * current implementation, 4K are reserved to avoid future complications.
375 #define VMCS12_SIZE 0x1000
377 /* Used to remember the last vmcs02 used for some recently used vmcs12s */
379 struct list_head list;
381 struct loaded_vmcs vmcs02;
385 * The nested_vmx structure is part of vcpu_vmx, and holds information we need
386 * for correct emulation of VMX (i.e., nested VMX) on this vcpu.
389 /* Has the level1 guest done vmxon? */
393 /* The guest-physical address of the current VMCS L1 keeps for L2 */
395 /* The host-usable pointer to the above */
396 struct page *current_vmcs12_page;
397 struct vmcs12 *current_vmcs12;
398 struct vmcs *current_shadow_vmcs;
400 * Indicates if the shadow vmcs must be updated with the
401 * data hold by vmcs12
403 bool sync_shadow_vmcs;
405 /* vmcs02_list cache of VMCSs recently used to run L2 guests */
406 struct list_head vmcs02_pool;
408 u64 vmcs01_tsc_offset;
409 /* L2 must run next, and mustn't decide to exit to L1. */
410 bool nested_run_pending;
412 * Guest pages referred to in vmcs02 with host-physical pointers, so
413 * we must keep them pinned while L2 runs.
415 struct page *apic_access_page;
416 struct page *virtual_apic_page;
417 struct page *pi_desc_page;
418 struct pi_desc *pi_desc;
421 u64 msr_ia32_feature_control;
423 struct hrtimer preemption_timer;
424 bool preemption_timer_expired;
426 /* to migrate it to L2 if VM_ENTRY_LOAD_DEBUG_CONTROLS is off */
429 u32 nested_vmx_procbased_ctls_low;
430 u32 nested_vmx_procbased_ctls_high;
431 u32 nested_vmx_true_procbased_ctls_low;
432 u32 nested_vmx_secondary_ctls_low;
433 u32 nested_vmx_secondary_ctls_high;
434 u32 nested_vmx_pinbased_ctls_low;
435 u32 nested_vmx_pinbased_ctls_high;
436 u32 nested_vmx_exit_ctls_low;
437 u32 nested_vmx_exit_ctls_high;
438 u32 nested_vmx_true_exit_ctls_low;
439 u32 nested_vmx_entry_ctls_low;
440 u32 nested_vmx_entry_ctls_high;
441 u32 nested_vmx_true_entry_ctls_low;
442 u32 nested_vmx_misc_low;
443 u32 nested_vmx_misc_high;
444 u32 nested_vmx_ept_caps;
445 u32 nested_vmx_vpid_caps;
448 #define POSTED_INTR_ON 0
449 #define POSTED_INTR_SN 1
451 /* Posted-Interrupt Descriptor */
453 u32 pir[8]; /* Posted interrupt requested */
456 /* bit 256 - Outstanding Notification */
458 /* bit 257 - Suppress Notification */
460 /* bit 271:258 - Reserved */
462 /* bit 279:272 - Notification Vector */
464 /* bit 287:280 - Reserved */
466 /* bit 319:288 - Notification Destination */
474 static bool pi_test_and_set_on(struct pi_desc *pi_desc)
476 return test_and_set_bit(POSTED_INTR_ON,
477 (unsigned long *)&pi_desc->control);
480 static bool pi_test_and_clear_on(struct pi_desc *pi_desc)
482 return test_and_clear_bit(POSTED_INTR_ON,
483 (unsigned long *)&pi_desc->control);
486 static int pi_test_and_set_pir(int vector, struct pi_desc *pi_desc)
488 return test_and_set_bit(vector, (unsigned long *)pi_desc->pir);
491 static inline void pi_clear_sn(struct pi_desc *pi_desc)
493 return clear_bit(POSTED_INTR_SN,
494 (unsigned long *)&pi_desc->control);
497 static inline void pi_set_sn(struct pi_desc *pi_desc)
499 return set_bit(POSTED_INTR_SN,
500 (unsigned long *)&pi_desc->control);
503 static inline int pi_test_on(struct pi_desc *pi_desc)
505 return test_bit(POSTED_INTR_ON,
506 (unsigned long *)&pi_desc->control);
509 static inline int pi_test_sn(struct pi_desc *pi_desc)
511 return test_bit(POSTED_INTR_SN,
512 (unsigned long *)&pi_desc->control);
516 struct kvm_vcpu vcpu;
517 unsigned long host_rsp;
519 bool nmi_known_unmasked;
521 u32 idt_vectoring_info;
523 struct shared_msr_entry *guest_msrs;
526 unsigned long host_idt_base;
528 u64 msr_host_kernel_gs_base;
529 u64 msr_guest_kernel_gs_base;
531 u32 vm_entry_controls_shadow;
532 u32 vm_exit_controls_shadow;
534 * loaded_vmcs points to the VMCS currently used in this vcpu. For a
535 * non-nested (L1) guest, it always points to vmcs01. For a nested
536 * guest (L2), it points to a different VMCS.
538 struct loaded_vmcs vmcs01;
539 struct loaded_vmcs *loaded_vmcs;
540 bool __launched; /* temporary, used in vmx_vcpu_run */
541 struct msr_autoload {
543 struct vmx_msr_entry guest[NR_AUTOLOAD_MSRS];
544 struct vmx_msr_entry host[NR_AUTOLOAD_MSRS];
548 u16 fs_sel, gs_sel, ldt_sel;
552 int gs_ldt_reload_needed;
553 int fs_reload_needed;
554 u64 msr_host_bndcfgs;
555 unsigned long vmcs_host_cr4; /* May not match real cr4 */
560 struct kvm_segment segs[8];
563 u32 bitmask; /* 4 bits per segment (1 bit per field) */
564 struct kvm_save_segment {
572 bool emulation_required;
574 /* Support for vnmi-less CPUs */
575 int soft_vnmi_blocked;
577 s64 vnmi_blocked_time;
580 /* Posted interrupt descriptor */
581 struct pi_desc pi_desc;
583 /* Support for a guest hypervisor (nested VMX) */
584 struct nested_vmx nested;
586 /* Dynamic PLE window. */
588 bool ple_window_dirty;
590 /* Support for PML */
591 #define PML_ENTITY_NUM 512
595 enum segment_cache_field {
604 static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
606 return container_of(vcpu, struct vcpu_vmx, vcpu);
609 static struct pi_desc *vcpu_to_pi_desc(struct kvm_vcpu *vcpu)
611 return &(to_vmx(vcpu)->pi_desc);
614 #define VMCS12_OFFSET(x) offsetof(struct vmcs12, x)
615 #define FIELD(number, name) [number] = VMCS12_OFFSET(name)
616 #define FIELD64(number, name) [number] = VMCS12_OFFSET(name), \
617 [number##_HIGH] = VMCS12_OFFSET(name)+4
620 static unsigned long shadow_read_only_fields[] = {
622 * We do NOT shadow fields that are modified when L0
623 * traps and emulates any vmx instruction (e.g. VMPTRLD,
624 * VMXON...) executed by L1.
625 * For example, VM_INSTRUCTION_ERROR is read
626 * by L1 if a vmx instruction fails (part of the error path).
627 * Note the code assumes this logic. If for some reason
628 * we start shadowing these fields then we need to
629 * force a shadow sync when L0 emulates vmx instructions
630 * (e.g. force a sync if VM_INSTRUCTION_ERROR is modified
631 * by nested_vmx_failValid)
635 VM_EXIT_INSTRUCTION_LEN,
636 IDT_VECTORING_INFO_FIELD,
637 IDT_VECTORING_ERROR_CODE,
638 VM_EXIT_INTR_ERROR_CODE,
640 GUEST_LINEAR_ADDRESS,
641 GUEST_PHYSICAL_ADDRESS
643 static int max_shadow_read_only_fields =
644 ARRAY_SIZE(shadow_read_only_fields);
646 static unsigned long shadow_read_write_fields[] = {
653 GUEST_INTERRUPTIBILITY_INFO,
666 CPU_BASED_VM_EXEC_CONTROL,
667 VM_ENTRY_EXCEPTION_ERROR_CODE,
668 VM_ENTRY_INTR_INFO_FIELD,
669 VM_ENTRY_INSTRUCTION_LEN,
670 VM_ENTRY_EXCEPTION_ERROR_CODE,
676 static int max_shadow_read_write_fields =
677 ARRAY_SIZE(shadow_read_write_fields);
679 static const unsigned short vmcs_field_to_offset_table[] = {
680 FIELD(VIRTUAL_PROCESSOR_ID, virtual_processor_id),
681 FIELD(POSTED_INTR_NV, posted_intr_nv),
682 FIELD(GUEST_ES_SELECTOR, guest_es_selector),
683 FIELD(GUEST_CS_SELECTOR, guest_cs_selector),
684 FIELD(GUEST_SS_SELECTOR, guest_ss_selector),
685 FIELD(GUEST_DS_SELECTOR, guest_ds_selector),
686 FIELD(GUEST_FS_SELECTOR, guest_fs_selector),
687 FIELD(GUEST_GS_SELECTOR, guest_gs_selector),
688 FIELD(GUEST_LDTR_SELECTOR, guest_ldtr_selector),
689 FIELD(GUEST_TR_SELECTOR, guest_tr_selector),
690 FIELD(GUEST_INTR_STATUS, guest_intr_status),
691 FIELD(HOST_ES_SELECTOR, host_es_selector),
692 FIELD(HOST_CS_SELECTOR, host_cs_selector),
693 FIELD(HOST_SS_SELECTOR, host_ss_selector),
694 FIELD(HOST_DS_SELECTOR, host_ds_selector),
695 FIELD(HOST_FS_SELECTOR, host_fs_selector),
696 FIELD(HOST_GS_SELECTOR, host_gs_selector),
697 FIELD(HOST_TR_SELECTOR, host_tr_selector),
698 FIELD64(IO_BITMAP_A, io_bitmap_a),
699 FIELD64(IO_BITMAP_B, io_bitmap_b),
700 FIELD64(MSR_BITMAP, msr_bitmap),
701 FIELD64(VM_EXIT_MSR_STORE_ADDR, vm_exit_msr_store_addr),
702 FIELD64(VM_EXIT_MSR_LOAD_ADDR, vm_exit_msr_load_addr),
703 FIELD64(VM_ENTRY_MSR_LOAD_ADDR, vm_entry_msr_load_addr),
704 FIELD64(TSC_OFFSET, tsc_offset),
705 FIELD64(VIRTUAL_APIC_PAGE_ADDR, virtual_apic_page_addr),
706 FIELD64(APIC_ACCESS_ADDR, apic_access_addr),
707 FIELD64(POSTED_INTR_DESC_ADDR, posted_intr_desc_addr),
708 FIELD64(EPT_POINTER, ept_pointer),
709 FIELD64(EOI_EXIT_BITMAP0, eoi_exit_bitmap0),
710 FIELD64(EOI_EXIT_BITMAP1, eoi_exit_bitmap1),
711 FIELD64(EOI_EXIT_BITMAP2, eoi_exit_bitmap2),
712 FIELD64(EOI_EXIT_BITMAP3, eoi_exit_bitmap3),
713 FIELD64(XSS_EXIT_BITMAP, xss_exit_bitmap),
714 FIELD64(GUEST_PHYSICAL_ADDRESS, guest_physical_address),
715 FIELD64(VMCS_LINK_POINTER, vmcs_link_pointer),
716 FIELD64(GUEST_IA32_DEBUGCTL, guest_ia32_debugctl),
717 FIELD64(GUEST_IA32_PAT, guest_ia32_pat),
718 FIELD64(GUEST_IA32_EFER, guest_ia32_efer),
719 FIELD64(GUEST_IA32_PERF_GLOBAL_CTRL, guest_ia32_perf_global_ctrl),
720 FIELD64(GUEST_PDPTR0, guest_pdptr0),
721 FIELD64(GUEST_PDPTR1, guest_pdptr1),
722 FIELD64(GUEST_PDPTR2, guest_pdptr2),
723 FIELD64(GUEST_PDPTR3, guest_pdptr3),
724 FIELD64(GUEST_BNDCFGS, guest_bndcfgs),
725 FIELD64(HOST_IA32_PAT, host_ia32_pat),
726 FIELD64(HOST_IA32_EFER, host_ia32_efer),
727 FIELD64(HOST_IA32_PERF_GLOBAL_CTRL, host_ia32_perf_global_ctrl),
728 FIELD(PIN_BASED_VM_EXEC_CONTROL, pin_based_vm_exec_control),
729 FIELD(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control),
730 FIELD(EXCEPTION_BITMAP, exception_bitmap),
731 FIELD(PAGE_FAULT_ERROR_CODE_MASK, page_fault_error_code_mask),
732 FIELD(PAGE_FAULT_ERROR_CODE_MATCH, page_fault_error_code_match),
733 FIELD(CR3_TARGET_COUNT, cr3_target_count),
734 FIELD(VM_EXIT_CONTROLS, vm_exit_controls),
735 FIELD(VM_EXIT_MSR_STORE_COUNT, vm_exit_msr_store_count),
736 FIELD(VM_EXIT_MSR_LOAD_COUNT, vm_exit_msr_load_count),
737 FIELD(VM_ENTRY_CONTROLS, vm_entry_controls),
738 FIELD(VM_ENTRY_MSR_LOAD_COUNT, vm_entry_msr_load_count),
739 FIELD(VM_ENTRY_INTR_INFO_FIELD, vm_entry_intr_info_field),
740 FIELD(VM_ENTRY_EXCEPTION_ERROR_CODE, vm_entry_exception_error_code),
741 FIELD(VM_ENTRY_INSTRUCTION_LEN, vm_entry_instruction_len),
742 FIELD(TPR_THRESHOLD, tpr_threshold),
743 FIELD(SECONDARY_VM_EXEC_CONTROL, secondary_vm_exec_control),
744 FIELD(VM_INSTRUCTION_ERROR, vm_instruction_error),
745 FIELD(VM_EXIT_REASON, vm_exit_reason),
746 FIELD(VM_EXIT_INTR_INFO, vm_exit_intr_info),
747 FIELD(VM_EXIT_INTR_ERROR_CODE, vm_exit_intr_error_code),
748 FIELD(IDT_VECTORING_INFO_FIELD, idt_vectoring_info_field),
749 FIELD(IDT_VECTORING_ERROR_CODE, idt_vectoring_error_code),
750 FIELD(VM_EXIT_INSTRUCTION_LEN, vm_exit_instruction_len),
751 FIELD(VMX_INSTRUCTION_INFO, vmx_instruction_info),
752 FIELD(GUEST_ES_LIMIT, guest_es_limit),
753 FIELD(GUEST_CS_LIMIT, guest_cs_limit),
754 FIELD(GUEST_SS_LIMIT, guest_ss_limit),
755 FIELD(GUEST_DS_LIMIT, guest_ds_limit),
756 FIELD(GUEST_FS_LIMIT, guest_fs_limit),
757 FIELD(GUEST_GS_LIMIT, guest_gs_limit),
758 FIELD(GUEST_LDTR_LIMIT, guest_ldtr_limit),
759 FIELD(GUEST_TR_LIMIT, guest_tr_limit),
760 FIELD(GUEST_GDTR_LIMIT, guest_gdtr_limit),
761 FIELD(GUEST_IDTR_LIMIT, guest_idtr_limit),
762 FIELD(GUEST_ES_AR_BYTES, guest_es_ar_bytes),
763 FIELD(GUEST_CS_AR_BYTES, guest_cs_ar_bytes),
764 FIELD(GUEST_SS_AR_BYTES, guest_ss_ar_bytes),
765 FIELD(GUEST_DS_AR_BYTES, guest_ds_ar_bytes),
766 FIELD(GUEST_FS_AR_BYTES, guest_fs_ar_bytes),
767 FIELD(GUEST_GS_AR_BYTES, guest_gs_ar_bytes),
768 FIELD(GUEST_LDTR_AR_BYTES, guest_ldtr_ar_bytes),
769 FIELD(GUEST_TR_AR_BYTES, guest_tr_ar_bytes),
770 FIELD(GUEST_INTERRUPTIBILITY_INFO, guest_interruptibility_info),
771 FIELD(GUEST_ACTIVITY_STATE, guest_activity_state),
772 FIELD(GUEST_SYSENTER_CS, guest_sysenter_cs),
773 FIELD(HOST_IA32_SYSENTER_CS, host_ia32_sysenter_cs),
774 FIELD(VMX_PREEMPTION_TIMER_VALUE, vmx_preemption_timer_value),
775 FIELD(CR0_GUEST_HOST_MASK, cr0_guest_host_mask),
776 FIELD(CR4_GUEST_HOST_MASK, cr4_guest_host_mask),
777 FIELD(CR0_READ_SHADOW, cr0_read_shadow),
778 FIELD(CR4_READ_SHADOW, cr4_read_shadow),
779 FIELD(CR3_TARGET_VALUE0, cr3_target_value0),
780 FIELD(CR3_TARGET_VALUE1, cr3_target_value1),
781 FIELD(CR3_TARGET_VALUE2, cr3_target_value2),
782 FIELD(CR3_TARGET_VALUE3, cr3_target_value3),
783 FIELD(EXIT_QUALIFICATION, exit_qualification),
784 FIELD(GUEST_LINEAR_ADDRESS, guest_linear_address),
785 FIELD(GUEST_CR0, guest_cr0),
786 FIELD(GUEST_CR3, guest_cr3),
787 FIELD(GUEST_CR4, guest_cr4),
788 FIELD(GUEST_ES_BASE, guest_es_base),
789 FIELD(GUEST_CS_BASE, guest_cs_base),
790 FIELD(GUEST_SS_BASE, guest_ss_base),
791 FIELD(GUEST_DS_BASE, guest_ds_base),
792 FIELD(GUEST_FS_BASE, guest_fs_base),
793 FIELD(GUEST_GS_BASE, guest_gs_base),
794 FIELD(GUEST_LDTR_BASE, guest_ldtr_base),
795 FIELD(GUEST_TR_BASE, guest_tr_base),
796 FIELD(GUEST_GDTR_BASE, guest_gdtr_base),
797 FIELD(GUEST_IDTR_BASE, guest_idtr_base),
798 FIELD(GUEST_DR7, guest_dr7),
799 FIELD(GUEST_RSP, guest_rsp),
800 FIELD(GUEST_RIP, guest_rip),
801 FIELD(GUEST_RFLAGS, guest_rflags),
802 FIELD(GUEST_PENDING_DBG_EXCEPTIONS, guest_pending_dbg_exceptions),
803 FIELD(GUEST_SYSENTER_ESP, guest_sysenter_esp),
804 FIELD(GUEST_SYSENTER_EIP, guest_sysenter_eip),
805 FIELD(HOST_CR0, host_cr0),
806 FIELD(HOST_CR3, host_cr3),
807 FIELD(HOST_CR4, host_cr4),
808 FIELD(HOST_FS_BASE, host_fs_base),
809 FIELD(HOST_GS_BASE, host_gs_base),
810 FIELD(HOST_TR_BASE, host_tr_base),
811 FIELD(HOST_GDTR_BASE, host_gdtr_base),
812 FIELD(HOST_IDTR_BASE, host_idtr_base),
813 FIELD(HOST_IA32_SYSENTER_ESP, host_ia32_sysenter_esp),
814 FIELD(HOST_IA32_SYSENTER_EIP, host_ia32_sysenter_eip),
815 FIELD(HOST_RSP, host_rsp),
816 FIELD(HOST_RIP, host_rip),
819 static inline short vmcs_field_to_offset(unsigned long field)
821 BUILD_BUG_ON(ARRAY_SIZE(vmcs_field_to_offset_table) > SHRT_MAX);
823 if (field >= ARRAY_SIZE(vmcs_field_to_offset_table) ||
824 vmcs_field_to_offset_table[field] == 0)
827 return vmcs_field_to_offset_table[field];
830 static inline struct vmcs12 *get_vmcs12(struct kvm_vcpu *vcpu)
832 return to_vmx(vcpu)->nested.current_vmcs12;
835 static struct page *nested_get_page(struct kvm_vcpu *vcpu, gpa_t addr)
837 struct page *page = kvm_vcpu_gfn_to_page(vcpu, addr >> PAGE_SHIFT);
838 if (is_error_page(page))
844 static void nested_release_page(struct page *page)
846 kvm_release_page_dirty(page);
849 static void nested_release_page_clean(struct page *page)
851 kvm_release_page_clean(page);
854 static unsigned long nested_ept_get_cr3(struct kvm_vcpu *vcpu);
855 static u64 construct_eptp(unsigned long root_hpa);
856 static void kvm_cpu_vmxon(u64 addr);
857 static void kvm_cpu_vmxoff(void);
858 static bool vmx_mpx_supported(void);
859 static bool vmx_xsaves_supported(void);
860 static int vmx_cpu_uses_apicv(struct kvm_vcpu *vcpu);
861 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr);
862 static void vmx_set_segment(struct kvm_vcpu *vcpu,
863 struct kvm_segment *var, int seg);
864 static void vmx_get_segment(struct kvm_vcpu *vcpu,
865 struct kvm_segment *var, int seg);
866 static bool guest_state_valid(struct kvm_vcpu *vcpu);
867 static u32 vmx_segment_access_rights(struct kvm_segment *var);
868 static void vmx_sync_pir_to_irr_dummy(struct kvm_vcpu *vcpu);
869 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx);
870 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx);
871 static int alloc_identity_pagetable(struct kvm *kvm);
873 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
874 static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
876 * We maintain a per-CPU linked-list of VMCS loaded on that CPU. This is needed
877 * when a CPU is brought down, and we need to VMCLEAR all VMCSs loaded on it.
879 static DEFINE_PER_CPU(struct list_head, loaded_vmcss_on_cpu);
880 static DEFINE_PER_CPU(struct desc_ptr, host_gdt);
883 * We maintian a per-CPU linked-list of vCPU, so in wakeup_handler() we
884 * can find which vCPU should be waken up.
886 static DEFINE_PER_CPU(struct list_head, blocked_vcpu_on_cpu);
887 static DEFINE_PER_CPU(spinlock_t, blocked_vcpu_on_cpu_lock);
889 static unsigned long *vmx_io_bitmap_a;
890 static unsigned long *vmx_io_bitmap_b;
891 static unsigned long *vmx_msr_bitmap_legacy;
892 static unsigned long *vmx_msr_bitmap_longmode;
893 static unsigned long *vmx_msr_bitmap_legacy_x2apic;
894 static unsigned long *vmx_msr_bitmap_longmode_x2apic;
895 static unsigned long *vmx_msr_bitmap_nested;
896 static unsigned long *vmx_vmread_bitmap;
897 static unsigned long *vmx_vmwrite_bitmap;
899 static bool cpu_has_load_ia32_efer;
900 static bool cpu_has_load_perf_global_ctrl;
902 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
903 static DEFINE_SPINLOCK(vmx_vpid_lock);
905 static struct vmcs_config {
909 u32 pin_based_exec_ctrl;
910 u32 cpu_based_exec_ctrl;
911 u32 cpu_based_2nd_exec_ctrl;
916 static struct vmx_capability {
921 #define VMX_SEGMENT_FIELD(seg) \
922 [VCPU_SREG_##seg] = { \
923 .selector = GUEST_##seg##_SELECTOR, \
924 .base = GUEST_##seg##_BASE, \
925 .limit = GUEST_##seg##_LIMIT, \
926 .ar_bytes = GUEST_##seg##_AR_BYTES, \
929 static const struct kvm_vmx_segment_field {
934 } kvm_vmx_segment_fields[] = {
935 VMX_SEGMENT_FIELD(CS),
936 VMX_SEGMENT_FIELD(DS),
937 VMX_SEGMENT_FIELD(ES),
938 VMX_SEGMENT_FIELD(FS),
939 VMX_SEGMENT_FIELD(GS),
940 VMX_SEGMENT_FIELD(SS),
941 VMX_SEGMENT_FIELD(TR),
942 VMX_SEGMENT_FIELD(LDTR),
945 static u64 host_efer;
947 static void ept_save_pdptrs(struct kvm_vcpu *vcpu);
950 * Keep MSR_STAR at the end, as setup_msrs() will try to optimize it
951 * away by decrementing the array size.
953 static const u32 vmx_msr_index[] = {
955 MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
957 MSR_EFER, MSR_TSC_AUX, MSR_STAR,
960 static inline bool is_page_fault(u32 intr_info)
962 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
963 INTR_INFO_VALID_MASK)) ==
964 (INTR_TYPE_HARD_EXCEPTION | PF_VECTOR | INTR_INFO_VALID_MASK);
967 static inline bool is_no_device(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 | NM_VECTOR | INTR_INFO_VALID_MASK);
974 static inline bool is_invalid_opcode(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 | UD_VECTOR | INTR_INFO_VALID_MASK);
981 static inline bool is_external_interrupt(u32 intr_info)
983 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
984 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
987 static inline bool is_machine_check(u32 intr_info)
989 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
990 INTR_INFO_VALID_MASK)) ==
991 (INTR_TYPE_HARD_EXCEPTION | MC_VECTOR | INTR_INFO_VALID_MASK);
994 static inline bool cpu_has_vmx_msr_bitmap(void)
996 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS;
999 static inline bool cpu_has_vmx_tpr_shadow(void)
1001 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW;
1004 static inline bool cpu_need_tpr_shadow(struct kvm_vcpu *vcpu)
1006 return cpu_has_vmx_tpr_shadow() && lapic_in_kernel(vcpu);
1009 static inline bool cpu_has_secondary_exec_ctrls(void)
1011 return vmcs_config.cpu_based_exec_ctrl &
1012 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
1015 static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
1017 return vmcs_config.cpu_based_2nd_exec_ctrl &
1018 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
1021 static inline bool cpu_has_vmx_virtualize_x2apic_mode(void)
1023 return vmcs_config.cpu_based_2nd_exec_ctrl &
1024 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
1027 static inline bool cpu_has_vmx_apic_register_virt(void)
1029 return vmcs_config.cpu_based_2nd_exec_ctrl &
1030 SECONDARY_EXEC_APIC_REGISTER_VIRT;
1033 static inline bool cpu_has_vmx_virtual_intr_delivery(void)
1035 return vmcs_config.cpu_based_2nd_exec_ctrl &
1036 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY;
1039 static inline bool cpu_has_vmx_posted_intr(void)
1041 return IS_ENABLED(CONFIG_X86_LOCAL_APIC) &&
1042 vmcs_config.pin_based_exec_ctrl & PIN_BASED_POSTED_INTR;
1045 static inline bool cpu_has_vmx_apicv(void)
1047 return cpu_has_vmx_apic_register_virt() &&
1048 cpu_has_vmx_virtual_intr_delivery() &&
1049 cpu_has_vmx_posted_intr();
1052 static inline bool cpu_has_vmx_flexpriority(void)
1054 return cpu_has_vmx_tpr_shadow() &&
1055 cpu_has_vmx_virtualize_apic_accesses();
1058 static inline bool cpu_has_vmx_ept_execute_only(void)
1060 return vmx_capability.ept & VMX_EPT_EXECUTE_ONLY_BIT;
1063 static inline bool cpu_has_vmx_ept_2m_page(void)
1065 return vmx_capability.ept & VMX_EPT_2MB_PAGE_BIT;
1068 static inline bool cpu_has_vmx_ept_1g_page(void)
1070 return vmx_capability.ept & VMX_EPT_1GB_PAGE_BIT;
1073 static inline bool cpu_has_vmx_ept_4levels(void)
1075 return vmx_capability.ept & VMX_EPT_PAGE_WALK_4_BIT;
1078 static inline bool cpu_has_vmx_ept_ad_bits(void)
1080 return vmx_capability.ept & VMX_EPT_AD_BIT;
1083 static inline bool cpu_has_vmx_invept_context(void)
1085 return vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT;
1088 static inline bool cpu_has_vmx_invept_global(void)
1090 return vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT;
1093 static inline bool cpu_has_vmx_invvpid_single(void)
1095 return vmx_capability.vpid & VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT;
1098 static inline bool cpu_has_vmx_invvpid_global(void)
1100 return vmx_capability.vpid & VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT;
1103 static inline bool cpu_has_vmx_ept(void)
1105 return vmcs_config.cpu_based_2nd_exec_ctrl &
1106 SECONDARY_EXEC_ENABLE_EPT;
1109 static inline bool cpu_has_vmx_unrestricted_guest(void)
1111 return vmcs_config.cpu_based_2nd_exec_ctrl &
1112 SECONDARY_EXEC_UNRESTRICTED_GUEST;
1115 static inline bool cpu_has_vmx_ple(void)
1117 return vmcs_config.cpu_based_2nd_exec_ctrl &
1118 SECONDARY_EXEC_PAUSE_LOOP_EXITING;
1121 static inline bool cpu_need_virtualize_apic_accesses(struct kvm_vcpu *vcpu)
1123 return flexpriority_enabled && lapic_in_kernel(vcpu);
1126 static inline bool cpu_has_vmx_vpid(void)
1128 return vmcs_config.cpu_based_2nd_exec_ctrl &
1129 SECONDARY_EXEC_ENABLE_VPID;
1132 static inline bool cpu_has_vmx_rdtscp(void)
1134 return vmcs_config.cpu_based_2nd_exec_ctrl &
1135 SECONDARY_EXEC_RDTSCP;
1138 static inline bool cpu_has_vmx_invpcid(void)
1140 return vmcs_config.cpu_based_2nd_exec_ctrl &
1141 SECONDARY_EXEC_ENABLE_INVPCID;
1144 static inline bool cpu_has_virtual_nmis(void)
1146 return vmcs_config.pin_based_exec_ctrl & PIN_BASED_VIRTUAL_NMIS;
1149 static inline bool cpu_has_vmx_wbinvd_exit(void)
1151 return vmcs_config.cpu_based_2nd_exec_ctrl &
1152 SECONDARY_EXEC_WBINVD_EXITING;
1155 static inline bool cpu_has_vmx_shadow_vmcs(void)
1158 rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
1159 /* check if the cpu supports writing r/o exit information fields */
1160 if (!(vmx_msr & MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS))
1163 return vmcs_config.cpu_based_2nd_exec_ctrl &
1164 SECONDARY_EXEC_SHADOW_VMCS;
1167 static inline bool cpu_has_vmx_pml(void)
1169 return vmcs_config.cpu_based_2nd_exec_ctrl & SECONDARY_EXEC_ENABLE_PML;
1172 static inline bool report_flexpriority(void)
1174 return flexpriority_enabled;
1177 static inline bool nested_cpu_has(struct vmcs12 *vmcs12, u32 bit)
1179 return vmcs12->cpu_based_vm_exec_control & bit;
1182 static inline bool nested_cpu_has2(struct vmcs12 *vmcs12, u32 bit)
1184 return (vmcs12->cpu_based_vm_exec_control &
1185 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
1186 (vmcs12->secondary_vm_exec_control & bit);
1189 static inline bool nested_cpu_has_virtual_nmis(struct vmcs12 *vmcs12)
1191 return vmcs12->pin_based_vm_exec_control & PIN_BASED_VIRTUAL_NMIS;
1194 static inline bool nested_cpu_has_preemption_timer(struct vmcs12 *vmcs12)
1196 return vmcs12->pin_based_vm_exec_control &
1197 PIN_BASED_VMX_PREEMPTION_TIMER;
1200 static inline int nested_cpu_has_ept(struct vmcs12 *vmcs12)
1202 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_EPT);
1205 static inline bool nested_cpu_has_xsaves(struct vmcs12 *vmcs12)
1207 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES) &&
1208 vmx_xsaves_supported();
1211 static inline bool nested_cpu_has_virt_x2apic_mode(struct vmcs12 *vmcs12)
1213 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE);
1216 static inline bool nested_cpu_has_apic_reg_virt(struct vmcs12 *vmcs12)
1218 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_APIC_REGISTER_VIRT);
1221 static inline bool nested_cpu_has_vid(struct vmcs12 *vmcs12)
1223 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
1226 static inline bool nested_cpu_has_posted_intr(struct vmcs12 *vmcs12)
1228 return vmcs12->pin_based_vm_exec_control & PIN_BASED_POSTED_INTR;
1231 static inline bool is_exception(u32 intr_info)
1233 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
1234 == (INTR_TYPE_HARD_EXCEPTION | INTR_INFO_VALID_MASK);
1237 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
1239 unsigned long exit_qualification);
1240 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
1241 struct vmcs12 *vmcs12,
1242 u32 reason, unsigned long qualification);
1244 static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
1248 for (i = 0; i < vmx->nmsrs; ++i)
1249 if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
1254 static inline void __invvpid(int ext, u16 vpid, gva_t gva)
1260 } operand = { vpid, 0, gva };
1262 asm volatile (__ex(ASM_VMX_INVVPID)
1263 /* CF==1 or ZF==1 --> rc = -1 */
1264 "; ja 1f ; ud2 ; 1:"
1265 : : "a"(&operand), "c"(ext) : "cc", "memory");
1268 static inline void __invept(int ext, u64 eptp, gpa_t gpa)
1272 } operand = {eptp, gpa};
1274 asm volatile (__ex(ASM_VMX_INVEPT)
1275 /* CF==1 or ZF==1 --> rc = -1 */
1276 "; ja 1f ; ud2 ; 1:\n"
1277 : : "a" (&operand), "c" (ext) : "cc", "memory");
1280 static struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
1284 i = __find_msr_index(vmx, msr);
1286 return &vmx->guest_msrs[i];
1290 static void vmcs_clear(struct vmcs *vmcs)
1292 u64 phys_addr = __pa(vmcs);
1295 asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0"
1296 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
1299 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
1303 static inline void loaded_vmcs_init(struct loaded_vmcs *loaded_vmcs)
1305 vmcs_clear(loaded_vmcs->vmcs);
1306 loaded_vmcs->cpu = -1;
1307 loaded_vmcs->launched = 0;
1310 static void vmcs_load(struct vmcs *vmcs)
1312 u64 phys_addr = __pa(vmcs);
1315 asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0"
1316 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
1319 printk(KERN_ERR "kvm: vmptrld %p/%llx failed\n",
1323 #ifdef CONFIG_KEXEC_CORE
1325 * This bitmap is used to indicate whether the vmclear
1326 * operation is enabled on all cpus. All disabled by
1329 static cpumask_t crash_vmclear_enabled_bitmap = CPU_MASK_NONE;
1331 static inline void crash_enable_local_vmclear(int cpu)
1333 cpumask_set_cpu(cpu, &crash_vmclear_enabled_bitmap);
1336 static inline void crash_disable_local_vmclear(int cpu)
1338 cpumask_clear_cpu(cpu, &crash_vmclear_enabled_bitmap);
1341 static inline int crash_local_vmclear_enabled(int cpu)
1343 return cpumask_test_cpu(cpu, &crash_vmclear_enabled_bitmap);
1346 static void crash_vmclear_local_loaded_vmcss(void)
1348 int cpu = raw_smp_processor_id();
1349 struct loaded_vmcs *v;
1351 if (!crash_local_vmclear_enabled(cpu))
1354 list_for_each_entry(v, &per_cpu(loaded_vmcss_on_cpu, cpu),
1355 loaded_vmcss_on_cpu_link)
1356 vmcs_clear(v->vmcs);
1359 static inline void crash_enable_local_vmclear(int cpu) { }
1360 static inline void crash_disable_local_vmclear(int cpu) { }
1361 #endif /* CONFIG_KEXEC_CORE */
1363 static void __loaded_vmcs_clear(void *arg)
1365 struct loaded_vmcs *loaded_vmcs = arg;
1366 int cpu = raw_smp_processor_id();
1368 if (loaded_vmcs->cpu != cpu)
1369 return; /* vcpu migration can race with cpu offline */
1370 if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
1371 per_cpu(current_vmcs, cpu) = NULL;
1372 crash_disable_local_vmclear(cpu);
1373 list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
1376 * we should ensure updating loaded_vmcs->loaded_vmcss_on_cpu_link
1377 * is before setting loaded_vmcs->vcpu to -1 which is done in
1378 * loaded_vmcs_init. Otherwise, other cpu can see vcpu = -1 fist
1379 * then adds the vmcs into percpu list before it is deleted.
1383 loaded_vmcs_init(loaded_vmcs);
1384 crash_enable_local_vmclear(cpu);
1387 static void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
1389 int cpu = loaded_vmcs->cpu;
1392 smp_call_function_single(cpu,
1393 __loaded_vmcs_clear, loaded_vmcs, 1);
1396 static inline void vpid_sync_vcpu_single(int vpid)
1401 if (cpu_has_vmx_invvpid_single())
1402 __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vpid, 0);
1405 static inline void vpid_sync_vcpu_global(void)
1407 if (cpu_has_vmx_invvpid_global())
1408 __invvpid(VMX_VPID_EXTENT_ALL_CONTEXT, 0, 0);
1411 static inline void vpid_sync_context(int vpid)
1413 if (cpu_has_vmx_invvpid_single())
1414 vpid_sync_vcpu_single(vpid);
1416 vpid_sync_vcpu_global();
1419 static inline void ept_sync_global(void)
1421 if (cpu_has_vmx_invept_global())
1422 __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0);
1425 static inline void ept_sync_context(u64 eptp)
1428 if (cpu_has_vmx_invept_context())
1429 __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0);
1435 static __always_inline unsigned long vmcs_readl(unsigned long field)
1437 unsigned long value;
1439 asm volatile (__ex_clear(ASM_VMX_VMREAD_RDX_RAX, "%0")
1440 : "=a"(value) : "d"(field) : "cc");
1444 static __always_inline u16 vmcs_read16(unsigned long field)
1446 return vmcs_readl(field);
1449 static __always_inline u32 vmcs_read32(unsigned long field)
1451 return vmcs_readl(field);
1454 static __always_inline u64 vmcs_read64(unsigned long field)
1456 #ifdef CONFIG_X86_64
1457 return vmcs_readl(field);
1459 return vmcs_readl(field) | ((u64)vmcs_readl(field+1) << 32);
1463 static noinline void vmwrite_error(unsigned long field, unsigned long value)
1465 printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
1466 field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
1470 static void vmcs_writel(unsigned long field, unsigned long value)
1474 asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0"
1475 : "=q"(error) : "a"(value), "d"(field) : "cc");
1476 if (unlikely(error))
1477 vmwrite_error(field, value);
1480 static void vmcs_write16(unsigned long field, u16 value)
1482 vmcs_writel(field, value);
1485 static void vmcs_write32(unsigned long field, u32 value)
1487 vmcs_writel(field, value);
1490 static void vmcs_write64(unsigned long field, u64 value)
1492 vmcs_writel(field, value);
1493 #ifndef CONFIG_X86_64
1495 vmcs_writel(field+1, value >> 32);
1499 static void vmcs_clear_bits(unsigned long field, u32 mask)
1501 vmcs_writel(field, vmcs_readl(field) & ~mask);
1504 static void vmcs_set_bits(unsigned long field, u32 mask)
1506 vmcs_writel(field, vmcs_readl(field) | mask);
1509 static inline void vm_entry_controls_init(struct vcpu_vmx *vmx, u32 val)
1511 vmcs_write32(VM_ENTRY_CONTROLS, val);
1512 vmx->vm_entry_controls_shadow = val;
1515 static inline void vm_entry_controls_set(struct vcpu_vmx *vmx, u32 val)
1517 if (vmx->vm_entry_controls_shadow != val)
1518 vm_entry_controls_init(vmx, val);
1521 static inline u32 vm_entry_controls_get(struct vcpu_vmx *vmx)
1523 return vmx->vm_entry_controls_shadow;
1527 static inline void vm_entry_controls_setbit(struct vcpu_vmx *vmx, u32 val)
1529 vm_entry_controls_set(vmx, vm_entry_controls_get(vmx) | val);
1532 static inline void vm_entry_controls_clearbit(struct vcpu_vmx *vmx, u32 val)
1534 vm_entry_controls_set(vmx, vm_entry_controls_get(vmx) & ~val);
1537 static inline void vm_exit_controls_init(struct vcpu_vmx *vmx, u32 val)
1539 vmcs_write32(VM_EXIT_CONTROLS, val);
1540 vmx->vm_exit_controls_shadow = val;
1543 static inline void vm_exit_controls_set(struct vcpu_vmx *vmx, u32 val)
1545 if (vmx->vm_exit_controls_shadow != val)
1546 vm_exit_controls_init(vmx, val);
1549 static inline u32 vm_exit_controls_get(struct vcpu_vmx *vmx)
1551 return vmx->vm_exit_controls_shadow;
1555 static inline void vm_exit_controls_setbit(struct vcpu_vmx *vmx, u32 val)
1557 vm_exit_controls_set(vmx, vm_exit_controls_get(vmx) | val);
1560 static inline void vm_exit_controls_clearbit(struct vcpu_vmx *vmx, u32 val)
1562 vm_exit_controls_set(vmx, vm_exit_controls_get(vmx) & ~val);
1565 static void vmx_segment_cache_clear(struct vcpu_vmx *vmx)
1567 vmx->segment_cache.bitmask = 0;
1570 static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
1574 u32 mask = 1 << (seg * SEG_FIELD_NR + field);
1576 if (!(vmx->vcpu.arch.regs_avail & (1 << VCPU_EXREG_SEGMENTS))) {
1577 vmx->vcpu.arch.regs_avail |= (1 << VCPU_EXREG_SEGMENTS);
1578 vmx->segment_cache.bitmask = 0;
1580 ret = vmx->segment_cache.bitmask & mask;
1581 vmx->segment_cache.bitmask |= mask;
1585 static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
1587 u16 *p = &vmx->segment_cache.seg[seg].selector;
1589 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
1590 *p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
1594 static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
1596 ulong *p = &vmx->segment_cache.seg[seg].base;
1598 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
1599 *p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
1603 static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
1605 u32 *p = &vmx->segment_cache.seg[seg].limit;
1607 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
1608 *p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
1612 static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
1614 u32 *p = &vmx->segment_cache.seg[seg].ar;
1616 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
1617 *p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
1621 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
1625 eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
1626 (1u << NM_VECTOR) | (1u << DB_VECTOR);
1627 if ((vcpu->guest_debug &
1628 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
1629 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
1630 eb |= 1u << BP_VECTOR;
1631 if (to_vmx(vcpu)->rmode.vm86_active)
1634 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
1635 if (vcpu->fpu_active)
1636 eb &= ~(1u << NM_VECTOR);
1638 /* When we are running a nested L2 guest and L1 specified for it a
1639 * certain exception bitmap, we must trap the same exceptions and pass
1640 * them to L1. When running L2, we will only handle the exceptions
1641 * specified above if L1 did not want them.
1643 if (is_guest_mode(vcpu))
1644 eb |= get_vmcs12(vcpu)->exception_bitmap;
1646 vmcs_write32(EXCEPTION_BITMAP, eb);
1649 static void clear_atomic_switch_msr_special(struct vcpu_vmx *vmx,
1650 unsigned long entry, unsigned long exit)
1652 vm_entry_controls_clearbit(vmx, entry);
1653 vm_exit_controls_clearbit(vmx, exit);
1656 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
1659 struct msr_autoload *m = &vmx->msr_autoload;
1663 if (cpu_has_load_ia32_efer) {
1664 clear_atomic_switch_msr_special(vmx,
1665 VM_ENTRY_LOAD_IA32_EFER,
1666 VM_EXIT_LOAD_IA32_EFER);
1670 case MSR_CORE_PERF_GLOBAL_CTRL:
1671 if (cpu_has_load_perf_global_ctrl) {
1672 clear_atomic_switch_msr_special(vmx,
1673 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1674 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
1680 for (i = 0; i < m->nr; ++i)
1681 if (m->guest[i].index == msr)
1687 m->guest[i] = m->guest[m->nr];
1688 m->host[i] = m->host[m->nr];
1689 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1690 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1693 static void add_atomic_switch_msr_special(struct vcpu_vmx *vmx,
1694 unsigned long entry, unsigned long exit,
1695 unsigned long guest_val_vmcs, unsigned long host_val_vmcs,
1696 u64 guest_val, u64 host_val)
1698 vmcs_write64(guest_val_vmcs, guest_val);
1699 vmcs_write64(host_val_vmcs, host_val);
1700 vm_entry_controls_setbit(vmx, entry);
1701 vm_exit_controls_setbit(vmx, exit);
1704 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
1705 u64 guest_val, u64 host_val)
1708 struct msr_autoload *m = &vmx->msr_autoload;
1712 if (cpu_has_load_ia32_efer) {
1713 add_atomic_switch_msr_special(vmx,
1714 VM_ENTRY_LOAD_IA32_EFER,
1715 VM_EXIT_LOAD_IA32_EFER,
1718 guest_val, host_val);
1722 case MSR_CORE_PERF_GLOBAL_CTRL:
1723 if (cpu_has_load_perf_global_ctrl) {
1724 add_atomic_switch_msr_special(vmx,
1725 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1726 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL,
1727 GUEST_IA32_PERF_GLOBAL_CTRL,
1728 HOST_IA32_PERF_GLOBAL_CTRL,
1729 guest_val, host_val);
1735 for (i = 0; i < m->nr; ++i)
1736 if (m->guest[i].index == msr)
1739 if (i == NR_AUTOLOAD_MSRS) {
1740 printk_once(KERN_WARNING "Not enough msr switch entries. "
1741 "Can't add msr %x\n", msr);
1743 } else if (i == m->nr) {
1745 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1746 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1749 m->guest[i].index = msr;
1750 m->guest[i].value = guest_val;
1751 m->host[i].index = msr;
1752 m->host[i].value = host_val;
1755 static void reload_tss(void)
1758 * VT restores TR but not its size. Useless.
1760 struct desc_ptr *gdt = this_cpu_ptr(&host_gdt);
1761 struct desc_struct *descs;
1763 descs = (void *)gdt->address;
1764 descs[GDT_ENTRY_TSS].type = 9; /* available TSS */
1768 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
1773 guest_efer = vmx->vcpu.arch.efer;
1776 * NX is emulated; LMA and LME handled by hardware; SCE meaningless
1779 ignore_bits = EFER_NX | EFER_SCE;
1780 #ifdef CONFIG_X86_64
1781 ignore_bits |= EFER_LMA | EFER_LME;
1782 /* SCE is meaningful only in long mode on Intel */
1783 if (guest_efer & EFER_LMA)
1784 ignore_bits &= ~(u64)EFER_SCE;
1786 guest_efer &= ~ignore_bits;
1787 guest_efer |= host_efer & ignore_bits;
1788 vmx->guest_msrs[efer_offset].data = guest_efer;
1789 vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
1791 clear_atomic_switch_msr(vmx, MSR_EFER);
1794 * On EPT, we can't emulate NX, so we must switch EFER atomically.
1795 * On CPUs that support "load IA32_EFER", always switch EFER
1796 * atomically, since it's faster than switching it manually.
1798 if (cpu_has_load_ia32_efer ||
1799 (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX))) {
1800 guest_efer = vmx->vcpu.arch.efer;
1801 if (!(guest_efer & EFER_LMA))
1802 guest_efer &= ~EFER_LME;
1803 if (guest_efer != host_efer)
1804 add_atomic_switch_msr(vmx, MSR_EFER,
1805 guest_efer, host_efer);
1812 static unsigned long segment_base(u16 selector)
1814 struct desc_ptr *gdt = this_cpu_ptr(&host_gdt);
1815 struct desc_struct *d;
1816 unsigned long table_base;
1819 if (!(selector & ~3))
1822 table_base = gdt->address;
1824 if (selector & 4) { /* from ldt */
1825 u16 ldt_selector = kvm_read_ldt();
1827 if (!(ldt_selector & ~3))
1830 table_base = segment_base(ldt_selector);
1832 d = (struct desc_struct *)(table_base + (selector & ~7));
1833 v = get_desc_base(d);
1834 #ifdef CONFIG_X86_64
1835 if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
1836 v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
1841 static inline unsigned long kvm_read_tr_base(void)
1844 asm("str %0" : "=g"(tr));
1845 return segment_base(tr);
1848 static void vmx_save_host_state(struct kvm_vcpu *vcpu)
1850 struct vcpu_vmx *vmx = to_vmx(vcpu);
1853 if (vmx->host_state.loaded)
1856 vmx->host_state.loaded = 1;
1858 * Set host fs and gs selectors. Unfortunately, 22.2.3 does not
1859 * allow segment selectors with cpl > 0 or ti == 1.
1861 vmx->host_state.ldt_sel = kvm_read_ldt();
1862 vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
1863 savesegment(fs, vmx->host_state.fs_sel);
1864 if (!(vmx->host_state.fs_sel & 7)) {
1865 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
1866 vmx->host_state.fs_reload_needed = 0;
1868 vmcs_write16(HOST_FS_SELECTOR, 0);
1869 vmx->host_state.fs_reload_needed = 1;
1871 savesegment(gs, vmx->host_state.gs_sel);
1872 if (!(vmx->host_state.gs_sel & 7))
1873 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
1875 vmcs_write16(HOST_GS_SELECTOR, 0);
1876 vmx->host_state.gs_ldt_reload_needed = 1;
1879 #ifdef CONFIG_X86_64
1880 savesegment(ds, vmx->host_state.ds_sel);
1881 savesegment(es, vmx->host_state.es_sel);
1884 #ifdef CONFIG_X86_64
1885 vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
1886 vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
1888 vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
1889 vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
1892 #ifdef CONFIG_X86_64
1893 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1894 if (is_long_mode(&vmx->vcpu))
1895 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1897 if (boot_cpu_has(X86_FEATURE_MPX))
1898 rdmsrl(MSR_IA32_BNDCFGS, vmx->host_state.msr_host_bndcfgs);
1899 for (i = 0; i < vmx->save_nmsrs; ++i)
1900 kvm_set_shared_msr(vmx->guest_msrs[i].index,
1901 vmx->guest_msrs[i].data,
1902 vmx->guest_msrs[i].mask);
1905 static void __vmx_load_host_state(struct vcpu_vmx *vmx)
1907 if (!vmx->host_state.loaded)
1910 ++vmx->vcpu.stat.host_state_reload;
1911 vmx->host_state.loaded = 0;
1912 #ifdef CONFIG_X86_64
1913 if (is_long_mode(&vmx->vcpu))
1914 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1916 if (vmx->host_state.gs_ldt_reload_needed) {
1917 kvm_load_ldt(vmx->host_state.ldt_sel);
1918 #ifdef CONFIG_X86_64
1919 load_gs_index(vmx->host_state.gs_sel);
1921 loadsegment(gs, vmx->host_state.gs_sel);
1924 if (vmx->host_state.fs_reload_needed)
1925 loadsegment(fs, vmx->host_state.fs_sel);
1926 #ifdef CONFIG_X86_64
1927 if (unlikely(vmx->host_state.ds_sel | vmx->host_state.es_sel)) {
1928 loadsegment(ds, vmx->host_state.ds_sel);
1929 loadsegment(es, vmx->host_state.es_sel);
1933 #ifdef CONFIG_X86_64
1934 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1936 if (vmx->host_state.msr_host_bndcfgs)
1937 wrmsrl(MSR_IA32_BNDCFGS, vmx->host_state.msr_host_bndcfgs);
1939 * If the FPU is not active (through the host task or
1940 * the guest vcpu), then restore the cr0.TS bit.
1942 if (!fpregs_active() && !vmx->vcpu.guest_fpu_loaded)
1944 load_gdt(this_cpu_ptr(&host_gdt));
1947 static void vmx_load_host_state(struct vcpu_vmx *vmx)
1950 __vmx_load_host_state(vmx);
1954 static void vmx_vcpu_pi_load(struct kvm_vcpu *vcpu, int cpu)
1956 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
1957 struct pi_desc old, new;
1960 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
1961 !irq_remapping_cap(IRQ_POSTING_CAP))
1965 old.control = new.control = pi_desc->control;
1968 * If 'nv' field is POSTED_INTR_WAKEUP_VECTOR, there
1969 * are two possible cases:
1970 * 1. After running 'pre_block', context switch
1971 * happened. For this case, 'sn' was set in
1972 * vmx_vcpu_put(), so we need to clear it here.
1973 * 2. After running 'pre_block', we were blocked,
1974 * and woken up by some other guy. For this case,
1975 * we don't need to do anything, 'pi_post_block'
1976 * will do everything for us. However, we cannot
1977 * check whether it is case #1 or case #2 here
1978 * (maybe, not needed), so we also clear sn here,
1979 * I think it is not a big deal.
1981 if (pi_desc->nv != POSTED_INTR_WAKEUP_VECTOR) {
1982 if (vcpu->cpu != cpu) {
1983 dest = cpu_physical_id(cpu);
1985 if (x2apic_enabled())
1988 new.ndst = (dest << 8) & 0xFF00;
1991 /* set 'NV' to 'notification vector' */
1992 new.nv = POSTED_INTR_VECTOR;
1995 /* Allow posting non-urgent interrupts */
1997 } while (cmpxchg(&pi_desc->control, old.control,
1998 new.control) != old.control);
2001 * Switches to specified vcpu, until a matching vcpu_put(), but assumes
2002 * vcpu mutex is already taken.
2004 static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2006 struct vcpu_vmx *vmx = to_vmx(vcpu);
2007 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
2010 kvm_cpu_vmxon(phys_addr);
2011 else if (vmx->loaded_vmcs->cpu != cpu)
2012 loaded_vmcs_clear(vmx->loaded_vmcs);
2014 if (per_cpu(current_vmcs, cpu) != vmx->loaded_vmcs->vmcs) {
2015 per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
2016 vmcs_load(vmx->loaded_vmcs->vmcs);
2019 if (vmx->loaded_vmcs->cpu != cpu) {
2020 struct desc_ptr *gdt = this_cpu_ptr(&host_gdt);
2021 unsigned long sysenter_esp;
2023 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
2024 local_irq_disable();
2025 crash_disable_local_vmclear(cpu);
2028 * Read loaded_vmcs->cpu should be before fetching
2029 * loaded_vmcs->loaded_vmcss_on_cpu_link.
2030 * See the comments in __loaded_vmcs_clear().
2034 list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
2035 &per_cpu(loaded_vmcss_on_cpu, cpu));
2036 crash_enable_local_vmclear(cpu);
2040 * Linux uses per-cpu TSS and GDT, so set these when switching
2043 vmcs_writel(HOST_TR_BASE, kvm_read_tr_base()); /* 22.2.4 */
2044 vmcs_writel(HOST_GDTR_BASE, gdt->address); /* 22.2.4 */
2046 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
2047 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
2048 vmx->loaded_vmcs->cpu = cpu;
2051 vmx_vcpu_pi_load(vcpu, cpu);
2054 static void vmx_vcpu_pi_put(struct kvm_vcpu *vcpu)
2056 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
2058 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
2059 !irq_remapping_cap(IRQ_POSTING_CAP))
2062 /* Set SN when the vCPU is preempted */
2063 if (vcpu->preempted)
2067 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
2069 vmx_vcpu_pi_put(vcpu);
2071 __vmx_load_host_state(to_vmx(vcpu));
2072 if (!vmm_exclusive) {
2073 __loaded_vmcs_clear(to_vmx(vcpu)->loaded_vmcs);
2079 static void vmx_fpu_activate(struct kvm_vcpu *vcpu)
2083 if (vcpu->fpu_active)
2085 vcpu->fpu_active = 1;
2086 cr0 = vmcs_readl(GUEST_CR0);
2087 cr0 &= ~(X86_CR0_TS | X86_CR0_MP);
2088 cr0 |= kvm_read_cr0_bits(vcpu, X86_CR0_TS | X86_CR0_MP);
2089 vmcs_writel(GUEST_CR0, cr0);
2090 update_exception_bitmap(vcpu);
2091 vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
2092 if (is_guest_mode(vcpu))
2093 vcpu->arch.cr0_guest_owned_bits &=
2094 ~get_vmcs12(vcpu)->cr0_guest_host_mask;
2095 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
2098 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu);
2101 * Return the cr0 value that a nested guest would read. This is a combination
2102 * of the real cr0 used to run the guest (guest_cr0), and the bits shadowed by
2103 * its hypervisor (cr0_read_shadow).
2105 static inline unsigned long nested_read_cr0(struct vmcs12 *fields)
2107 return (fields->guest_cr0 & ~fields->cr0_guest_host_mask) |
2108 (fields->cr0_read_shadow & fields->cr0_guest_host_mask);
2110 static inline unsigned long nested_read_cr4(struct vmcs12 *fields)
2112 return (fields->guest_cr4 & ~fields->cr4_guest_host_mask) |
2113 (fields->cr4_read_shadow & fields->cr4_guest_host_mask);
2116 static void vmx_fpu_deactivate(struct kvm_vcpu *vcpu)
2118 /* Note that there is no vcpu->fpu_active = 0 here. The caller must
2119 * set this *before* calling this function.
2121 vmx_decache_cr0_guest_bits(vcpu);
2122 vmcs_set_bits(GUEST_CR0, X86_CR0_TS | X86_CR0_MP);
2123 update_exception_bitmap(vcpu);
2124 vcpu->arch.cr0_guest_owned_bits = 0;
2125 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
2126 if (is_guest_mode(vcpu)) {
2128 * L1's specified read shadow might not contain the TS bit,
2129 * so now that we turned on shadowing of this bit, we need to
2130 * set this bit of the shadow. Like in nested_vmx_run we need
2131 * nested_read_cr0(vmcs12), but vmcs12->guest_cr0 is not yet
2132 * up-to-date here because we just decached cr0.TS (and we'll
2133 * only update vmcs12->guest_cr0 on nested exit).
2135 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2136 vmcs12->guest_cr0 = (vmcs12->guest_cr0 & ~X86_CR0_TS) |
2137 (vcpu->arch.cr0 & X86_CR0_TS);
2138 vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
2140 vmcs_writel(CR0_READ_SHADOW, vcpu->arch.cr0);
2143 static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
2145 unsigned long rflags, save_rflags;
2147 if (!test_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail)) {
2148 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
2149 rflags = vmcs_readl(GUEST_RFLAGS);
2150 if (to_vmx(vcpu)->rmode.vm86_active) {
2151 rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
2152 save_rflags = to_vmx(vcpu)->rmode.save_rflags;
2153 rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
2155 to_vmx(vcpu)->rflags = rflags;
2157 return to_vmx(vcpu)->rflags;
2160 static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
2162 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
2163 to_vmx(vcpu)->rflags = rflags;
2164 if (to_vmx(vcpu)->rmode.vm86_active) {
2165 to_vmx(vcpu)->rmode.save_rflags = rflags;
2166 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
2168 vmcs_writel(GUEST_RFLAGS, rflags);
2171 static u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu)
2173 u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
2176 if (interruptibility & GUEST_INTR_STATE_STI)
2177 ret |= KVM_X86_SHADOW_INT_STI;
2178 if (interruptibility & GUEST_INTR_STATE_MOV_SS)
2179 ret |= KVM_X86_SHADOW_INT_MOV_SS;
2184 static void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
2186 u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
2187 u32 interruptibility = interruptibility_old;
2189 interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
2191 if (mask & KVM_X86_SHADOW_INT_MOV_SS)
2192 interruptibility |= GUEST_INTR_STATE_MOV_SS;
2193 else if (mask & KVM_X86_SHADOW_INT_STI)
2194 interruptibility |= GUEST_INTR_STATE_STI;
2196 if ((interruptibility != interruptibility_old))
2197 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
2200 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
2204 rip = kvm_rip_read(vcpu);
2205 rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
2206 kvm_rip_write(vcpu, rip);
2208 /* skipping an emulated instruction also counts */
2209 vmx_set_interrupt_shadow(vcpu, 0);
2213 * KVM wants to inject page-faults which it got to the guest. This function
2214 * checks whether in a nested guest, we need to inject them to L1 or L2.
2216 static int nested_vmx_check_exception(struct kvm_vcpu *vcpu, unsigned nr)
2218 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2220 if (!(vmcs12->exception_bitmap & (1u << nr)))
2223 nested_vmx_vmexit(vcpu, to_vmx(vcpu)->exit_reason,
2224 vmcs_read32(VM_EXIT_INTR_INFO),
2225 vmcs_readl(EXIT_QUALIFICATION));
2229 static void vmx_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
2230 bool has_error_code, u32 error_code,
2233 struct vcpu_vmx *vmx = to_vmx(vcpu);
2234 u32 intr_info = nr | INTR_INFO_VALID_MASK;
2236 if (!reinject && is_guest_mode(vcpu) &&
2237 nested_vmx_check_exception(vcpu, nr))
2240 if (has_error_code) {
2241 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
2242 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
2245 if (vmx->rmode.vm86_active) {
2247 if (kvm_exception_is_soft(nr))
2248 inc_eip = vcpu->arch.event_exit_inst_len;
2249 if (kvm_inject_realmode_interrupt(vcpu, nr, inc_eip) != EMULATE_DONE)
2250 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2254 if (kvm_exception_is_soft(nr)) {
2255 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2256 vmx->vcpu.arch.event_exit_inst_len);
2257 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
2259 intr_info |= INTR_TYPE_HARD_EXCEPTION;
2261 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
2264 static bool vmx_rdtscp_supported(void)
2266 return cpu_has_vmx_rdtscp();
2269 static bool vmx_invpcid_supported(void)
2271 return cpu_has_vmx_invpcid() && enable_ept;
2275 * Swap MSR entry in host/guest MSR entry array.
2277 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
2279 struct shared_msr_entry tmp;
2281 tmp = vmx->guest_msrs[to];
2282 vmx->guest_msrs[to] = vmx->guest_msrs[from];
2283 vmx->guest_msrs[from] = tmp;
2286 static void vmx_set_msr_bitmap(struct kvm_vcpu *vcpu)
2288 unsigned long *msr_bitmap;
2290 if (is_guest_mode(vcpu))
2291 msr_bitmap = vmx_msr_bitmap_nested;
2292 else if (vcpu->arch.apic_base & X2APIC_ENABLE) {
2293 if (is_long_mode(vcpu))
2294 msr_bitmap = vmx_msr_bitmap_longmode_x2apic;
2296 msr_bitmap = vmx_msr_bitmap_legacy_x2apic;
2298 if (is_long_mode(vcpu))
2299 msr_bitmap = vmx_msr_bitmap_longmode;
2301 msr_bitmap = vmx_msr_bitmap_legacy;
2304 vmcs_write64(MSR_BITMAP, __pa(msr_bitmap));
2308 * Set up the vmcs to automatically save and restore system
2309 * msrs. Don't touch the 64-bit msrs if the guest is in legacy
2310 * mode, as fiddling with msrs is very expensive.
2312 static void setup_msrs(struct vcpu_vmx *vmx)
2314 int save_nmsrs, index;
2317 #ifdef CONFIG_X86_64
2318 if (is_long_mode(&vmx->vcpu)) {
2319 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
2321 move_msr_up(vmx, index, save_nmsrs++);
2322 index = __find_msr_index(vmx, MSR_LSTAR);
2324 move_msr_up(vmx, index, save_nmsrs++);
2325 index = __find_msr_index(vmx, MSR_CSTAR);
2327 move_msr_up(vmx, index, save_nmsrs++);
2328 index = __find_msr_index(vmx, MSR_TSC_AUX);
2329 if (index >= 0 && guest_cpuid_has_rdtscp(&vmx->vcpu))
2330 move_msr_up(vmx, index, save_nmsrs++);
2332 * MSR_STAR is only needed on long mode guests, and only
2333 * if efer.sce is enabled.
2335 index = __find_msr_index(vmx, MSR_STAR);
2336 if ((index >= 0) && (vmx->vcpu.arch.efer & EFER_SCE))
2337 move_msr_up(vmx, index, save_nmsrs++);
2340 index = __find_msr_index(vmx, MSR_EFER);
2341 if (index >= 0 && update_transition_efer(vmx, index))
2342 move_msr_up(vmx, index, save_nmsrs++);
2344 vmx->save_nmsrs = save_nmsrs;
2346 if (cpu_has_vmx_msr_bitmap())
2347 vmx_set_msr_bitmap(&vmx->vcpu);
2351 * reads and returns guest's timestamp counter "register"
2352 * guest_tsc = host_tsc + tsc_offset -- 21.3
2354 static u64 guest_read_tsc(void)
2356 u64 host_tsc, tsc_offset;
2359 tsc_offset = vmcs_read64(TSC_OFFSET);
2360 return host_tsc + tsc_offset;
2364 * Like guest_read_tsc, but always returns L1's notion of the timestamp
2365 * counter, even if a nested guest (L2) is currently running.
2367 static u64 vmx_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
2371 tsc_offset = is_guest_mode(vcpu) ?
2372 to_vmx(vcpu)->nested.vmcs01_tsc_offset :
2373 vmcs_read64(TSC_OFFSET);
2374 return host_tsc + tsc_offset;
2378 * Engage any workarounds for mis-matched TSC rates. Currently limited to
2379 * software catchup for faster rates on slower CPUs.
2381 static void vmx_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
2386 if (user_tsc_khz > tsc_khz) {
2387 vcpu->arch.tsc_catchup = 1;
2388 vcpu->arch.tsc_always_catchup = 1;
2390 WARN(1, "user requested TSC rate below hardware speed\n");
2393 static u64 vmx_read_tsc_offset(struct kvm_vcpu *vcpu)
2395 return vmcs_read64(TSC_OFFSET);
2399 * writes 'offset' into guest's timestamp counter offset register
2401 static void vmx_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
2403 if (is_guest_mode(vcpu)) {
2405 * We're here if L1 chose not to trap WRMSR to TSC. According
2406 * to the spec, this should set L1's TSC; The offset that L1
2407 * set for L2 remains unchanged, and still needs to be added
2408 * to the newly set TSC to get L2's TSC.
2410 struct vmcs12 *vmcs12;
2411 to_vmx(vcpu)->nested.vmcs01_tsc_offset = offset;
2412 /* recalculate vmcs02.TSC_OFFSET: */
2413 vmcs12 = get_vmcs12(vcpu);
2414 vmcs_write64(TSC_OFFSET, offset +
2415 (nested_cpu_has(vmcs12, CPU_BASED_USE_TSC_OFFSETING) ?
2416 vmcs12->tsc_offset : 0));
2418 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
2419 vmcs_read64(TSC_OFFSET), offset);
2420 vmcs_write64(TSC_OFFSET, offset);
2424 static void vmx_adjust_tsc_offset(struct kvm_vcpu *vcpu, s64 adjustment, bool host)
2426 u64 offset = vmcs_read64(TSC_OFFSET);
2428 vmcs_write64(TSC_OFFSET, offset + adjustment);
2429 if (is_guest_mode(vcpu)) {
2430 /* Even when running L2, the adjustment needs to apply to L1 */
2431 to_vmx(vcpu)->nested.vmcs01_tsc_offset += adjustment;
2433 trace_kvm_write_tsc_offset(vcpu->vcpu_id, offset,
2434 offset + adjustment);
2437 static u64 vmx_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
2439 return target_tsc - rdtsc();
2442 static bool guest_cpuid_has_vmx(struct kvm_vcpu *vcpu)
2444 struct kvm_cpuid_entry2 *best = kvm_find_cpuid_entry(vcpu, 1, 0);
2445 return best && (best->ecx & (1 << (X86_FEATURE_VMX & 31)));
2449 * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
2450 * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
2451 * all guests if the "nested" module option is off, and can also be disabled
2452 * for a single guest by disabling its VMX cpuid bit.
2454 static inline bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
2456 return nested && guest_cpuid_has_vmx(vcpu);
2460 * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be
2461 * returned for the various VMX controls MSRs when nested VMX is enabled.
2462 * The same values should also be used to verify that vmcs12 control fields are
2463 * valid during nested entry from L1 to L2.
2464 * Each of these control msrs has a low and high 32-bit half: A low bit is on
2465 * if the corresponding bit in the (32-bit) control field *must* be on, and a
2466 * bit in the high half is on if the corresponding bit in the control field
2467 * may be on. See also vmx_control_verify().
2469 static void nested_vmx_setup_ctls_msrs(struct vcpu_vmx *vmx)
2472 * Note that as a general rule, the high half of the MSRs (bits in
2473 * the control fields which may be 1) should be initialized by the
2474 * intersection of the underlying hardware's MSR (i.e., features which
2475 * can be supported) and the list of features we want to expose -
2476 * because they are known to be properly supported in our code.
2477 * Also, usually, the low half of the MSRs (bits which must be 1) can
2478 * be set to 0, meaning that L1 may turn off any of these bits. The
2479 * reason is that if one of these bits is necessary, it will appear
2480 * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control
2481 * fields of vmcs01 and vmcs02, will turn these bits off - and
2482 * nested_vmx_exit_handled() will not pass related exits to L1.
2483 * These rules have exceptions below.
2486 /* pin-based controls */
2487 rdmsr(MSR_IA32_VMX_PINBASED_CTLS,
2488 vmx->nested.nested_vmx_pinbased_ctls_low,
2489 vmx->nested.nested_vmx_pinbased_ctls_high);
2490 vmx->nested.nested_vmx_pinbased_ctls_low |=
2491 PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
2492 vmx->nested.nested_vmx_pinbased_ctls_high &=
2493 PIN_BASED_EXT_INTR_MASK |
2494 PIN_BASED_NMI_EXITING |
2495 PIN_BASED_VIRTUAL_NMIS;
2496 vmx->nested.nested_vmx_pinbased_ctls_high |=
2497 PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
2498 PIN_BASED_VMX_PREEMPTION_TIMER;
2499 if (vmx_cpu_uses_apicv(&vmx->vcpu))
2500 vmx->nested.nested_vmx_pinbased_ctls_high |=
2501 PIN_BASED_POSTED_INTR;
2504 rdmsr(MSR_IA32_VMX_EXIT_CTLS,
2505 vmx->nested.nested_vmx_exit_ctls_low,
2506 vmx->nested.nested_vmx_exit_ctls_high);
2507 vmx->nested.nested_vmx_exit_ctls_low =
2508 VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
2510 vmx->nested.nested_vmx_exit_ctls_high &=
2511 #ifdef CONFIG_X86_64
2512 VM_EXIT_HOST_ADDR_SPACE_SIZE |
2514 VM_EXIT_LOAD_IA32_PAT | VM_EXIT_SAVE_IA32_PAT;
2515 vmx->nested.nested_vmx_exit_ctls_high |=
2516 VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR |
2517 VM_EXIT_LOAD_IA32_EFER | VM_EXIT_SAVE_IA32_EFER |
2518 VM_EXIT_SAVE_VMX_PREEMPTION_TIMER | VM_EXIT_ACK_INTR_ON_EXIT;
2520 if (vmx_mpx_supported())
2521 vmx->nested.nested_vmx_exit_ctls_high |= VM_EXIT_CLEAR_BNDCFGS;
2523 /* We support free control of debug control saving. */
2524 vmx->nested.nested_vmx_true_exit_ctls_low =
2525 vmx->nested.nested_vmx_exit_ctls_low &
2526 ~VM_EXIT_SAVE_DEBUG_CONTROLS;
2528 /* entry controls */
2529 rdmsr(MSR_IA32_VMX_ENTRY_CTLS,
2530 vmx->nested.nested_vmx_entry_ctls_low,
2531 vmx->nested.nested_vmx_entry_ctls_high);
2532 vmx->nested.nested_vmx_entry_ctls_low =
2533 VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
2534 vmx->nested.nested_vmx_entry_ctls_high &=
2535 #ifdef CONFIG_X86_64
2536 VM_ENTRY_IA32E_MODE |
2538 VM_ENTRY_LOAD_IA32_PAT;
2539 vmx->nested.nested_vmx_entry_ctls_high |=
2540 (VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR | VM_ENTRY_LOAD_IA32_EFER);
2541 if (vmx_mpx_supported())
2542 vmx->nested.nested_vmx_entry_ctls_high |= VM_ENTRY_LOAD_BNDCFGS;
2544 /* We support free control of debug control loading. */
2545 vmx->nested.nested_vmx_true_entry_ctls_low =
2546 vmx->nested.nested_vmx_entry_ctls_low &
2547 ~VM_ENTRY_LOAD_DEBUG_CONTROLS;
2549 /* cpu-based controls */
2550 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS,
2551 vmx->nested.nested_vmx_procbased_ctls_low,
2552 vmx->nested.nested_vmx_procbased_ctls_high);
2553 vmx->nested.nested_vmx_procbased_ctls_low =
2554 CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
2555 vmx->nested.nested_vmx_procbased_ctls_high &=
2556 CPU_BASED_VIRTUAL_INTR_PENDING |
2557 CPU_BASED_VIRTUAL_NMI_PENDING | CPU_BASED_USE_TSC_OFFSETING |
2558 CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING |
2559 CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING |
2560 CPU_BASED_CR3_STORE_EXITING |
2561 #ifdef CONFIG_X86_64
2562 CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING |
2564 CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING |
2565 CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_TRAP_FLAG |
2566 CPU_BASED_MONITOR_EXITING | CPU_BASED_RDPMC_EXITING |
2567 CPU_BASED_RDTSC_EXITING | CPU_BASED_PAUSE_EXITING |
2568 CPU_BASED_TPR_SHADOW | CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2570 * We can allow some features even when not supported by the
2571 * hardware. For example, L1 can specify an MSR bitmap - and we
2572 * can use it to avoid exits to L1 - even when L0 runs L2
2573 * without MSR bitmaps.
2575 vmx->nested.nested_vmx_procbased_ctls_high |=
2576 CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
2577 CPU_BASED_USE_MSR_BITMAPS;
2579 /* We support free control of CR3 access interception. */
2580 vmx->nested.nested_vmx_true_procbased_ctls_low =
2581 vmx->nested.nested_vmx_procbased_ctls_low &
2582 ~(CPU_BASED_CR3_LOAD_EXITING | CPU_BASED_CR3_STORE_EXITING);
2584 /* secondary cpu-based controls */
2585 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
2586 vmx->nested.nested_vmx_secondary_ctls_low,
2587 vmx->nested.nested_vmx_secondary_ctls_high);
2588 vmx->nested.nested_vmx_secondary_ctls_low = 0;
2589 vmx->nested.nested_vmx_secondary_ctls_high &=
2590 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2591 SECONDARY_EXEC_RDTSCP |
2592 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2593 SECONDARY_EXEC_APIC_REGISTER_VIRT |
2594 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2595 SECONDARY_EXEC_WBINVD_EXITING |
2596 SECONDARY_EXEC_XSAVES |
2597 SECONDARY_EXEC_PCOMMIT;
2600 /* nested EPT: emulate EPT also to L1 */
2601 vmx->nested.nested_vmx_secondary_ctls_high |=
2602 SECONDARY_EXEC_ENABLE_EPT;
2603 vmx->nested.nested_vmx_ept_caps = VMX_EPT_PAGE_WALK_4_BIT |
2604 VMX_EPTP_WB_BIT | VMX_EPT_2MB_PAGE_BIT |
2606 vmx->nested.nested_vmx_ept_caps &= vmx_capability.ept;
2608 * For nested guests, we don't do anything specific
2609 * for single context invalidation. Hence, only advertise
2610 * support for global context invalidation.
2612 vmx->nested.nested_vmx_ept_caps |= VMX_EPT_EXTENT_GLOBAL_BIT;
2614 vmx->nested.nested_vmx_ept_caps = 0;
2616 vmx->nested.nested_vmx_vpid_caps = 0;
2618 if (enable_unrestricted_guest)
2619 vmx->nested.nested_vmx_secondary_ctls_high |=
2620 SECONDARY_EXEC_UNRESTRICTED_GUEST;
2622 /* miscellaneous data */
2623 rdmsr(MSR_IA32_VMX_MISC,
2624 vmx->nested.nested_vmx_misc_low,
2625 vmx->nested.nested_vmx_misc_high);
2626 vmx->nested.nested_vmx_misc_low &= VMX_MISC_SAVE_EFER_LMA;
2627 vmx->nested.nested_vmx_misc_low |=
2628 VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE |
2629 VMX_MISC_ACTIVITY_HLT;
2630 vmx->nested.nested_vmx_misc_high = 0;
2633 static inline bool vmx_control_verify(u32 control, u32 low, u32 high)
2636 * Bits 0 in high must be 0, and bits 1 in low must be 1.
2638 return ((control & high) | low) == control;
2641 static inline u64 vmx_control_msr(u32 low, u32 high)
2643 return low | ((u64)high << 32);
2646 /* Returns 0 on success, non-0 otherwise. */
2647 static int vmx_get_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
2649 struct vcpu_vmx *vmx = to_vmx(vcpu);
2651 switch (msr_index) {
2652 case MSR_IA32_VMX_BASIC:
2654 * This MSR reports some information about VMX support. We
2655 * should return information about the VMX we emulate for the
2656 * guest, and the VMCS structure we give it - not about the
2657 * VMX support of the underlying hardware.
2659 *pdata = VMCS12_REVISION | VMX_BASIC_TRUE_CTLS |
2660 ((u64)VMCS12_SIZE << VMX_BASIC_VMCS_SIZE_SHIFT) |
2661 (VMX_BASIC_MEM_TYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT);
2663 case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
2664 case MSR_IA32_VMX_PINBASED_CTLS:
2665 *pdata = vmx_control_msr(
2666 vmx->nested.nested_vmx_pinbased_ctls_low,
2667 vmx->nested.nested_vmx_pinbased_ctls_high);
2669 case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
2670 *pdata = vmx_control_msr(
2671 vmx->nested.nested_vmx_true_procbased_ctls_low,
2672 vmx->nested.nested_vmx_procbased_ctls_high);
2674 case MSR_IA32_VMX_PROCBASED_CTLS:
2675 *pdata = vmx_control_msr(
2676 vmx->nested.nested_vmx_procbased_ctls_low,
2677 vmx->nested.nested_vmx_procbased_ctls_high);
2679 case MSR_IA32_VMX_TRUE_EXIT_CTLS:
2680 *pdata = vmx_control_msr(
2681 vmx->nested.nested_vmx_true_exit_ctls_low,
2682 vmx->nested.nested_vmx_exit_ctls_high);
2684 case MSR_IA32_VMX_EXIT_CTLS:
2685 *pdata = vmx_control_msr(
2686 vmx->nested.nested_vmx_exit_ctls_low,
2687 vmx->nested.nested_vmx_exit_ctls_high);
2689 case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
2690 *pdata = vmx_control_msr(
2691 vmx->nested.nested_vmx_true_entry_ctls_low,
2692 vmx->nested.nested_vmx_entry_ctls_high);
2694 case MSR_IA32_VMX_ENTRY_CTLS:
2695 *pdata = vmx_control_msr(
2696 vmx->nested.nested_vmx_entry_ctls_low,
2697 vmx->nested.nested_vmx_entry_ctls_high);
2699 case MSR_IA32_VMX_MISC:
2700 *pdata = vmx_control_msr(
2701 vmx->nested.nested_vmx_misc_low,
2702 vmx->nested.nested_vmx_misc_high);
2705 * These MSRs specify bits which the guest must keep fixed (on or off)
2706 * while L1 is in VMXON mode (in L1's root mode, or running an L2).
2707 * We picked the standard core2 setting.
2709 #define VMXON_CR0_ALWAYSON (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
2710 #define VMXON_CR4_ALWAYSON X86_CR4_VMXE
2711 case MSR_IA32_VMX_CR0_FIXED0:
2712 *pdata = VMXON_CR0_ALWAYSON;
2714 case MSR_IA32_VMX_CR0_FIXED1:
2717 case MSR_IA32_VMX_CR4_FIXED0:
2718 *pdata = VMXON_CR4_ALWAYSON;
2720 case MSR_IA32_VMX_CR4_FIXED1:
2723 case MSR_IA32_VMX_VMCS_ENUM:
2724 *pdata = 0x2e; /* highest index: VMX_PREEMPTION_TIMER_VALUE */
2726 case MSR_IA32_VMX_PROCBASED_CTLS2:
2727 *pdata = vmx_control_msr(
2728 vmx->nested.nested_vmx_secondary_ctls_low,
2729 vmx->nested.nested_vmx_secondary_ctls_high);
2731 case MSR_IA32_VMX_EPT_VPID_CAP:
2732 /* Currently, no nested vpid support */
2733 *pdata = vmx->nested.nested_vmx_ept_caps;
2743 * Reads an msr value (of 'msr_index') into 'pdata'.
2744 * Returns 0 on success, non-0 otherwise.
2745 * Assumes vcpu_load() was already called.
2747 static int vmx_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2749 struct shared_msr_entry *msr;
2751 switch (msr_info->index) {
2752 #ifdef CONFIG_X86_64
2754 msr_info->data = vmcs_readl(GUEST_FS_BASE);
2757 msr_info->data = vmcs_readl(GUEST_GS_BASE);
2759 case MSR_KERNEL_GS_BASE:
2760 vmx_load_host_state(to_vmx(vcpu));
2761 msr_info->data = to_vmx(vcpu)->msr_guest_kernel_gs_base;
2765 return kvm_get_msr_common(vcpu, msr_info);
2767 msr_info->data = guest_read_tsc();
2769 case MSR_IA32_SYSENTER_CS:
2770 msr_info->data = vmcs_read32(GUEST_SYSENTER_CS);
2772 case MSR_IA32_SYSENTER_EIP:
2773 msr_info->data = vmcs_readl(GUEST_SYSENTER_EIP);
2775 case MSR_IA32_SYSENTER_ESP:
2776 msr_info->data = vmcs_readl(GUEST_SYSENTER_ESP);
2778 case MSR_IA32_BNDCFGS:
2779 if (!vmx_mpx_supported())
2781 msr_info->data = vmcs_read64(GUEST_BNDCFGS);
2783 case MSR_IA32_FEATURE_CONTROL:
2784 if (!nested_vmx_allowed(vcpu))
2786 msr_info->data = to_vmx(vcpu)->nested.msr_ia32_feature_control;
2788 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
2789 if (!nested_vmx_allowed(vcpu))
2791 return vmx_get_vmx_msr(vcpu, msr_info->index, &msr_info->data);
2793 if (!vmx_xsaves_supported())
2795 msr_info->data = vcpu->arch.ia32_xss;
2798 if (!guest_cpuid_has_rdtscp(vcpu))
2800 /* Otherwise falls through */
2802 msr = find_msr_entry(to_vmx(vcpu), msr_info->index);
2804 msr_info->data = msr->data;
2807 return kvm_get_msr_common(vcpu, msr_info);
2813 static void vmx_leave_nested(struct kvm_vcpu *vcpu);
2816 * Writes msr value into into the appropriate "register".
2817 * Returns 0 on success, non-0 otherwise.
2818 * Assumes vcpu_load() was already called.
2820 static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2822 struct vcpu_vmx *vmx = to_vmx(vcpu);
2823 struct shared_msr_entry *msr;
2825 u32 msr_index = msr_info->index;
2826 u64 data = msr_info->data;
2828 switch (msr_index) {
2830 ret = kvm_set_msr_common(vcpu, msr_info);
2832 #ifdef CONFIG_X86_64
2834 vmx_segment_cache_clear(vmx);
2835 vmcs_writel(GUEST_FS_BASE, data);
2838 vmx_segment_cache_clear(vmx);
2839 vmcs_writel(GUEST_GS_BASE, data);
2841 case MSR_KERNEL_GS_BASE:
2842 vmx_load_host_state(vmx);
2843 vmx->msr_guest_kernel_gs_base = data;
2846 case MSR_IA32_SYSENTER_CS:
2847 vmcs_write32(GUEST_SYSENTER_CS, data);
2849 case MSR_IA32_SYSENTER_EIP:
2850 vmcs_writel(GUEST_SYSENTER_EIP, data);
2852 case MSR_IA32_SYSENTER_ESP:
2853 vmcs_writel(GUEST_SYSENTER_ESP, data);
2855 case MSR_IA32_BNDCFGS:
2856 if (!vmx_mpx_supported())
2858 vmcs_write64(GUEST_BNDCFGS, data);
2861 kvm_write_tsc(vcpu, msr_info);
2863 case MSR_IA32_CR_PAT:
2864 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2865 if (!kvm_mtrr_valid(vcpu, MSR_IA32_CR_PAT, data))
2867 vmcs_write64(GUEST_IA32_PAT, data);
2868 vcpu->arch.pat = data;
2871 ret = kvm_set_msr_common(vcpu, msr_info);
2873 case MSR_IA32_TSC_ADJUST:
2874 ret = kvm_set_msr_common(vcpu, msr_info);
2876 case MSR_IA32_FEATURE_CONTROL:
2877 if (!nested_vmx_allowed(vcpu) ||
2878 (to_vmx(vcpu)->nested.msr_ia32_feature_control &
2879 FEATURE_CONTROL_LOCKED && !msr_info->host_initiated))
2881 vmx->nested.msr_ia32_feature_control = data;
2882 if (msr_info->host_initiated && data == 0)
2883 vmx_leave_nested(vcpu);
2885 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
2886 return 1; /* they are read-only */
2888 if (!vmx_xsaves_supported())
2891 * The only supported bit as of Skylake is bit 8, but
2892 * it is not supported on KVM.
2896 vcpu->arch.ia32_xss = data;
2897 if (vcpu->arch.ia32_xss != host_xss)
2898 add_atomic_switch_msr(vmx, MSR_IA32_XSS,
2899 vcpu->arch.ia32_xss, host_xss);
2901 clear_atomic_switch_msr(vmx, MSR_IA32_XSS);
2904 if (!guest_cpuid_has_rdtscp(vcpu))
2906 /* Check reserved bit, higher 32 bits should be zero */
2907 if ((data >> 32) != 0)
2909 /* Otherwise falls through */
2911 msr = find_msr_entry(vmx, msr_index);
2913 u64 old_msr_data = msr->data;
2915 if (msr - vmx->guest_msrs < vmx->save_nmsrs) {
2917 ret = kvm_set_shared_msr(msr->index, msr->data,
2921 msr->data = old_msr_data;
2925 ret = kvm_set_msr_common(vcpu, msr_info);
2931 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
2933 __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
2936 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
2939 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
2941 case VCPU_EXREG_PDPTR:
2943 ept_save_pdptrs(vcpu);
2950 static __init int cpu_has_kvm_support(void)
2952 return cpu_has_vmx();
2955 static __init int vmx_disabled_by_bios(void)
2959 rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
2960 if (msr & FEATURE_CONTROL_LOCKED) {
2961 /* launched w/ TXT and VMX disabled */
2962 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
2965 /* launched w/o TXT and VMX only enabled w/ TXT */
2966 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
2967 && (msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
2968 && !tboot_enabled()) {
2969 printk(KERN_WARNING "kvm: disable TXT in the BIOS or "
2970 "activate TXT before enabling KVM\n");
2973 /* launched w/o TXT and VMX disabled */
2974 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
2975 && !tboot_enabled())
2982 static void kvm_cpu_vmxon(u64 addr)
2984 asm volatile (ASM_VMX_VMXON_RAX
2985 : : "a"(&addr), "m"(addr)
2989 static int hardware_enable(void)
2991 int cpu = raw_smp_processor_id();
2992 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
2995 if (cr4_read_shadow() & X86_CR4_VMXE)
2998 INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
2999 INIT_LIST_HEAD(&per_cpu(blocked_vcpu_on_cpu, cpu));
3000 spin_lock_init(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
3003 * Now we can enable the vmclear operation in kdump
3004 * since the loaded_vmcss_on_cpu list on this cpu
3005 * has been initialized.
3007 * Though the cpu is not in VMX operation now, there
3008 * is no problem to enable the vmclear operation
3009 * for the loaded_vmcss_on_cpu list is empty!
3011 crash_enable_local_vmclear(cpu);
3013 rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
3015 test_bits = FEATURE_CONTROL_LOCKED;
3016 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
3017 if (tboot_enabled())
3018 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX;
3020 if ((old & test_bits) != test_bits) {
3021 /* enable and lock */
3022 wrmsrl(MSR_IA32_FEATURE_CONTROL, old | test_bits);
3024 cr4_set_bits(X86_CR4_VMXE);
3026 if (vmm_exclusive) {
3027 kvm_cpu_vmxon(phys_addr);
3031 native_store_gdt(this_cpu_ptr(&host_gdt));
3036 static void vmclear_local_loaded_vmcss(void)
3038 int cpu = raw_smp_processor_id();
3039 struct loaded_vmcs *v, *n;
3041 list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
3042 loaded_vmcss_on_cpu_link)
3043 __loaded_vmcs_clear(v);
3047 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
3050 static void kvm_cpu_vmxoff(void)
3052 asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
3055 static void hardware_disable(void)
3057 if (vmm_exclusive) {
3058 vmclear_local_loaded_vmcss();
3061 cr4_clear_bits(X86_CR4_VMXE);
3064 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
3065 u32 msr, u32 *result)
3067 u32 vmx_msr_low, vmx_msr_high;
3068 u32 ctl = ctl_min | ctl_opt;
3070 rdmsr(msr, vmx_msr_low, vmx_msr_high);
3072 ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
3073 ctl |= vmx_msr_low; /* bit == 1 in low word ==> must be one */
3075 /* Ensure minimum (required) set of control bits are supported. */
3083 static __init bool allow_1_setting(u32 msr, u32 ctl)
3085 u32 vmx_msr_low, vmx_msr_high;
3087 rdmsr(msr, vmx_msr_low, vmx_msr_high);
3088 return vmx_msr_high & ctl;
3091 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
3093 u32 vmx_msr_low, vmx_msr_high;
3094 u32 min, opt, min2, opt2;
3095 u32 _pin_based_exec_control = 0;
3096 u32 _cpu_based_exec_control = 0;
3097 u32 _cpu_based_2nd_exec_control = 0;
3098 u32 _vmexit_control = 0;
3099 u32 _vmentry_control = 0;
3101 min = CPU_BASED_HLT_EXITING |
3102 #ifdef CONFIG_X86_64
3103 CPU_BASED_CR8_LOAD_EXITING |
3104 CPU_BASED_CR8_STORE_EXITING |
3106 CPU_BASED_CR3_LOAD_EXITING |
3107 CPU_BASED_CR3_STORE_EXITING |
3108 CPU_BASED_USE_IO_BITMAPS |
3109 CPU_BASED_MOV_DR_EXITING |
3110 CPU_BASED_USE_TSC_OFFSETING |
3111 CPU_BASED_MWAIT_EXITING |
3112 CPU_BASED_MONITOR_EXITING |
3113 CPU_BASED_INVLPG_EXITING |
3114 CPU_BASED_RDPMC_EXITING;
3116 opt = CPU_BASED_TPR_SHADOW |
3117 CPU_BASED_USE_MSR_BITMAPS |
3118 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
3119 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
3120 &_cpu_based_exec_control) < 0)
3122 #ifdef CONFIG_X86_64
3123 if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
3124 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
3125 ~CPU_BASED_CR8_STORE_EXITING;
3127 if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
3129 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
3130 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
3131 SECONDARY_EXEC_WBINVD_EXITING |
3132 SECONDARY_EXEC_ENABLE_VPID |
3133 SECONDARY_EXEC_ENABLE_EPT |
3134 SECONDARY_EXEC_UNRESTRICTED_GUEST |
3135 SECONDARY_EXEC_PAUSE_LOOP_EXITING |
3136 SECONDARY_EXEC_RDTSCP |
3137 SECONDARY_EXEC_ENABLE_INVPCID |
3138 SECONDARY_EXEC_APIC_REGISTER_VIRT |
3139 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
3140 SECONDARY_EXEC_SHADOW_VMCS |
3141 SECONDARY_EXEC_XSAVES |
3142 SECONDARY_EXEC_ENABLE_PML |
3143 SECONDARY_EXEC_PCOMMIT;
3144 if (adjust_vmx_controls(min2, opt2,
3145 MSR_IA32_VMX_PROCBASED_CTLS2,
3146 &_cpu_based_2nd_exec_control) < 0)
3149 #ifndef CONFIG_X86_64
3150 if (!(_cpu_based_2nd_exec_control &
3151 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
3152 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
3155 if (!(_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
3156 _cpu_based_2nd_exec_control &= ~(
3157 SECONDARY_EXEC_APIC_REGISTER_VIRT |
3158 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
3159 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
3161 if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
3162 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
3164 _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
3165 CPU_BASED_CR3_STORE_EXITING |
3166 CPU_BASED_INVLPG_EXITING);
3167 rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
3168 vmx_capability.ept, vmx_capability.vpid);
3171 min = VM_EXIT_SAVE_DEBUG_CONTROLS;
3172 #ifdef CONFIG_X86_64
3173 min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
3175 opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT |
3176 VM_EXIT_ACK_INTR_ON_EXIT | VM_EXIT_CLEAR_BNDCFGS;
3177 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
3178 &_vmexit_control) < 0)
3181 min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
3182 opt = PIN_BASED_VIRTUAL_NMIS | PIN_BASED_POSTED_INTR;
3183 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
3184 &_pin_based_exec_control) < 0)
3187 if (!(_cpu_based_2nd_exec_control &
3188 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) ||
3189 !(_vmexit_control & VM_EXIT_ACK_INTR_ON_EXIT))
3190 _pin_based_exec_control &= ~PIN_BASED_POSTED_INTR;
3192 min = VM_ENTRY_LOAD_DEBUG_CONTROLS;
3193 opt = VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_BNDCFGS;
3194 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
3195 &_vmentry_control) < 0)
3198 rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
3200 /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
3201 if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
3204 #ifdef CONFIG_X86_64
3205 /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
3206 if (vmx_msr_high & (1u<<16))
3210 /* Require Write-Back (WB) memory type for VMCS accesses. */
3211 if (((vmx_msr_high >> 18) & 15) != 6)
3214 vmcs_conf->size = vmx_msr_high & 0x1fff;
3215 vmcs_conf->order = get_order(vmcs_config.size);
3216 vmcs_conf->revision_id = vmx_msr_low;
3218 vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
3219 vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
3220 vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
3221 vmcs_conf->vmexit_ctrl = _vmexit_control;
3222 vmcs_conf->vmentry_ctrl = _vmentry_control;
3224 cpu_has_load_ia32_efer =
3225 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
3226 VM_ENTRY_LOAD_IA32_EFER)
3227 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
3228 VM_EXIT_LOAD_IA32_EFER);
3230 cpu_has_load_perf_global_ctrl =
3231 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
3232 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
3233 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
3234 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
3237 * Some cpus support VM_ENTRY_(LOAD|SAVE)_IA32_PERF_GLOBAL_CTRL
3238 * but due to arrata below it can't be used. Workaround is to use
3239 * msr load mechanism to switch IA32_PERF_GLOBAL_CTRL.
3241 * VM Exit May Incorrectly Clear IA32_PERF_GLOBAL_CTRL [34:32]
3246 * BC86,AAY89,BD102 (model 44)
3250 if (cpu_has_load_perf_global_ctrl && boot_cpu_data.x86 == 0x6) {
3251 switch (boot_cpu_data.x86_model) {
3257 cpu_has_load_perf_global_ctrl = false;
3258 printk_once(KERN_WARNING"kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
3259 "does not work properly. Using workaround\n");
3267 rdmsrl(MSR_IA32_XSS, host_xss);
3272 static struct vmcs *alloc_vmcs_cpu(int cpu)
3274 int node = cpu_to_node(cpu);
3278 pages = __alloc_pages_node(node, GFP_KERNEL, vmcs_config.order);
3281 vmcs = page_address(pages);
3282 memset(vmcs, 0, vmcs_config.size);
3283 vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
3287 static struct vmcs *alloc_vmcs(void)
3289 return alloc_vmcs_cpu(raw_smp_processor_id());
3292 static void free_vmcs(struct vmcs *vmcs)
3294 free_pages((unsigned long)vmcs, vmcs_config.order);
3298 * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
3300 static void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
3302 if (!loaded_vmcs->vmcs)
3304 loaded_vmcs_clear(loaded_vmcs);
3305 free_vmcs(loaded_vmcs->vmcs);
3306 loaded_vmcs->vmcs = NULL;
3309 static void free_kvm_area(void)
3313 for_each_possible_cpu(cpu) {
3314 free_vmcs(per_cpu(vmxarea, cpu));
3315 per_cpu(vmxarea, cpu) = NULL;
3319 static void init_vmcs_shadow_fields(void)
3323 /* No checks for read only fields yet */
3325 for (i = j = 0; i < max_shadow_read_write_fields; i++) {
3326 switch (shadow_read_write_fields[i]) {
3328 if (!vmx_mpx_supported())
3336 shadow_read_write_fields[j] =
3337 shadow_read_write_fields[i];
3340 max_shadow_read_write_fields = j;
3342 /* shadowed fields guest access without vmexit */
3343 for (i = 0; i < max_shadow_read_write_fields; i++) {
3344 clear_bit(shadow_read_write_fields[i],
3345 vmx_vmwrite_bitmap);
3346 clear_bit(shadow_read_write_fields[i],
3349 for (i = 0; i < max_shadow_read_only_fields; i++)
3350 clear_bit(shadow_read_only_fields[i],
3354 static __init int alloc_kvm_area(void)
3358 for_each_possible_cpu(cpu) {
3361 vmcs = alloc_vmcs_cpu(cpu);
3367 per_cpu(vmxarea, cpu) = vmcs;
3372 static bool emulation_required(struct kvm_vcpu *vcpu)
3374 return emulate_invalid_guest_state && !guest_state_valid(vcpu);
3377 static void fix_pmode_seg(struct kvm_vcpu *vcpu, int seg,
3378 struct kvm_segment *save)
3380 if (!emulate_invalid_guest_state) {
3382 * CS and SS RPL should be equal during guest entry according
3383 * to VMX spec, but in reality it is not always so. Since vcpu
3384 * is in the middle of the transition from real mode to
3385 * protected mode it is safe to assume that RPL 0 is a good
3388 if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
3389 save->selector &= ~SEGMENT_RPL_MASK;
3390 save->dpl = save->selector & SEGMENT_RPL_MASK;
3393 vmx_set_segment(vcpu, save, seg);
3396 static void enter_pmode(struct kvm_vcpu *vcpu)
3398 unsigned long flags;
3399 struct vcpu_vmx *vmx = to_vmx(vcpu);
3402 * Update real mode segment cache. It may be not up-to-date if sement
3403 * register was written while vcpu was in a guest mode.
3405 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
3406 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
3407 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
3408 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
3409 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
3410 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
3412 vmx->rmode.vm86_active = 0;
3414 vmx_segment_cache_clear(vmx);
3416 vmx_set_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
3418 flags = vmcs_readl(GUEST_RFLAGS);
3419 flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
3420 flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
3421 vmcs_writel(GUEST_RFLAGS, flags);
3423 vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
3424 (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
3426 update_exception_bitmap(vcpu);
3428 fix_pmode_seg(vcpu, VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
3429 fix_pmode_seg(vcpu, VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
3430 fix_pmode_seg(vcpu, VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
3431 fix_pmode_seg(vcpu, VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
3432 fix_pmode_seg(vcpu, VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
3433 fix_pmode_seg(vcpu, VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
3436 static void fix_rmode_seg(int seg, struct kvm_segment *save)
3438 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3439 struct kvm_segment var = *save;
3442 if (seg == VCPU_SREG_CS)
3445 if (!emulate_invalid_guest_state) {
3446 var.selector = var.base >> 4;
3447 var.base = var.base & 0xffff0;
3457 if (save->base & 0xf)
3458 printk_once(KERN_WARNING "kvm: segment base is not "
3459 "paragraph aligned when entering "
3460 "protected mode (seg=%d)", seg);
3463 vmcs_write16(sf->selector, var.selector);
3464 vmcs_write32(sf->base, var.base);
3465 vmcs_write32(sf->limit, var.limit);
3466 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(&var));
3469 static void enter_rmode(struct kvm_vcpu *vcpu)
3471 unsigned long flags;
3472 struct vcpu_vmx *vmx = to_vmx(vcpu);
3474 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
3475 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
3476 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
3477 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
3478 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
3479 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
3480 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
3482 vmx->rmode.vm86_active = 1;
3485 * Very old userspace does not call KVM_SET_TSS_ADDR before entering
3486 * vcpu. Warn the user that an update is overdue.
3488 if (!vcpu->kvm->arch.tss_addr)
3489 printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
3490 "called before entering vcpu\n");
3492 vmx_segment_cache_clear(vmx);
3494 vmcs_writel(GUEST_TR_BASE, vcpu->kvm->arch.tss_addr);
3495 vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
3496 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
3498 flags = vmcs_readl(GUEST_RFLAGS);
3499 vmx->rmode.save_rflags = flags;
3501 flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
3503 vmcs_writel(GUEST_RFLAGS, flags);
3504 vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
3505 update_exception_bitmap(vcpu);
3507 fix_rmode_seg(VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
3508 fix_rmode_seg(VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
3509 fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
3510 fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
3511 fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
3512 fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
3514 kvm_mmu_reset_context(vcpu);
3517 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
3519 struct vcpu_vmx *vmx = to_vmx(vcpu);
3520 struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
3526 * Force kernel_gs_base reloading before EFER changes, as control
3527 * of this msr depends on is_long_mode().
3529 vmx_load_host_state(to_vmx(vcpu));
3530 vcpu->arch.efer = efer;
3531 if (efer & EFER_LMA) {
3532 vm_entry_controls_setbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
3535 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
3537 msr->data = efer & ~EFER_LME;
3542 #ifdef CONFIG_X86_64
3544 static void enter_lmode(struct kvm_vcpu *vcpu)
3548 vmx_segment_cache_clear(to_vmx(vcpu));
3550 guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
3551 if ((guest_tr_ar & VMX_AR_TYPE_MASK) != VMX_AR_TYPE_BUSY_64_TSS) {
3552 pr_debug_ratelimited("%s: tss fixup for long mode. \n",
3554 vmcs_write32(GUEST_TR_AR_BYTES,
3555 (guest_tr_ar & ~VMX_AR_TYPE_MASK)
3556 | VMX_AR_TYPE_BUSY_64_TSS);
3558 vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
3561 static void exit_lmode(struct kvm_vcpu *vcpu)
3563 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
3564 vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
3569 static inline void __vmx_flush_tlb(struct kvm_vcpu *vcpu, int vpid)
3571 vpid_sync_context(vpid);
3573 if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
3575 ept_sync_context(construct_eptp(vcpu->arch.mmu.root_hpa));
3579 static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
3581 __vmx_flush_tlb(vcpu, to_vmx(vcpu)->vpid);
3584 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
3586 ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
3588 vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
3589 vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
3592 static void vmx_decache_cr3(struct kvm_vcpu *vcpu)
3594 if (enable_ept && is_paging(vcpu))
3595 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
3596 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
3599 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
3601 ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
3603 vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
3604 vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
3607 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
3609 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
3611 if (!test_bit(VCPU_EXREG_PDPTR,
3612 (unsigned long *)&vcpu->arch.regs_dirty))
3615 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
3616 vmcs_write64(GUEST_PDPTR0, mmu->pdptrs[0]);
3617 vmcs_write64(GUEST_PDPTR1, mmu->pdptrs[1]);
3618 vmcs_write64(GUEST_PDPTR2, mmu->pdptrs[2]);
3619 vmcs_write64(GUEST_PDPTR3, mmu->pdptrs[3]);
3623 static void ept_save_pdptrs(struct kvm_vcpu *vcpu)
3625 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
3627 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
3628 mmu->pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
3629 mmu->pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
3630 mmu->pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
3631 mmu->pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
3634 __set_bit(VCPU_EXREG_PDPTR,
3635 (unsigned long *)&vcpu->arch.regs_avail);
3636 __set_bit(VCPU_EXREG_PDPTR,
3637 (unsigned long *)&vcpu->arch.regs_dirty);
3640 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
3642 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
3644 struct kvm_vcpu *vcpu)
3646 if (!test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
3647 vmx_decache_cr3(vcpu);
3648 if (!(cr0 & X86_CR0_PG)) {
3649 /* From paging/starting to nonpaging */
3650 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
3651 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
3652 (CPU_BASED_CR3_LOAD_EXITING |
3653 CPU_BASED_CR3_STORE_EXITING));
3654 vcpu->arch.cr0 = cr0;
3655 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
3656 } else if (!is_paging(vcpu)) {
3657 /* From nonpaging to paging */
3658 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
3659 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
3660 ~(CPU_BASED_CR3_LOAD_EXITING |
3661 CPU_BASED_CR3_STORE_EXITING));
3662 vcpu->arch.cr0 = cr0;
3663 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
3666 if (!(cr0 & X86_CR0_WP))
3667 *hw_cr0 &= ~X86_CR0_WP;
3670 static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
3672 struct vcpu_vmx *vmx = to_vmx(vcpu);
3673 unsigned long hw_cr0;
3675 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK);
3676 if (enable_unrestricted_guest)
3677 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
3679 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON;
3681 if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
3684 if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
3688 #ifdef CONFIG_X86_64
3689 if (vcpu->arch.efer & EFER_LME) {
3690 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
3692 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
3698 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
3700 if (!vcpu->fpu_active)
3701 hw_cr0 |= X86_CR0_TS | X86_CR0_MP;
3703 vmcs_writel(CR0_READ_SHADOW, cr0);
3704 vmcs_writel(GUEST_CR0, hw_cr0);
3705 vcpu->arch.cr0 = cr0;
3707 /* depends on vcpu->arch.cr0 to be set to a new value */
3708 vmx->emulation_required = emulation_required(vcpu);
3711 static u64 construct_eptp(unsigned long root_hpa)
3715 /* TODO write the value reading from MSR */
3716 eptp = VMX_EPT_DEFAULT_MT |
3717 VMX_EPT_DEFAULT_GAW << VMX_EPT_GAW_EPTP_SHIFT;
3718 if (enable_ept_ad_bits)
3719 eptp |= VMX_EPT_AD_ENABLE_BIT;
3720 eptp |= (root_hpa & PAGE_MASK);
3725 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
3727 unsigned long guest_cr3;
3732 eptp = construct_eptp(cr3);
3733 vmcs_write64(EPT_POINTER, eptp);
3734 if (is_paging(vcpu) || is_guest_mode(vcpu))
3735 guest_cr3 = kvm_read_cr3(vcpu);
3737 guest_cr3 = vcpu->kvm->arch.ept_identity_map_addr;
3738 ept_load_pdptrs(vcpu);
3741 vmx_flush_tlb(vcpu);
3742 vmcs_writel(GUEST_CR3, guest_cr3);
3745 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
3748 * Pass through host's Machine Check Enable value to hw_cr4, which
3749 * is in force while we are in guest mode. Do not let guests control
3750 * this bit, even if host CR4.MCE == 0.
3752 unsigned long hw_cr4 =
3753 (cr4_read_shadow() & X86_CR4_MCE) |
3754 (cr4 & ~X86_CR4_MCE) |
3755 (to_vmx(vcpu)->rmode.vm86_active ?
3756 KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON);
3758 if (cr4 & X86_CR4_VMXE) {
3760 * To use VMXON (and later other VMX instructions), a guest
3761 * must first be able to turn on cr4.VMXE (see handle_vmon()).
3762 * So basically the check on whether to allow nested VMX
3765 if (!nested_vmx_allowed(vcpu))
3768 if (to_vmx(vcpu)->nested.vmxon &&
3769 ((cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON))
3772 vcpu->arch.cr4 = cr4;
3774 if (!is_paging(vcpu)) {
3775 hw_cr4 &= ~X86_CR4_PAE;
3776 hw_cr4 |= X86_CR4_PSE;
3778 * SMEP/SMAP is disabled if CPU is in non-paging mode
3779 * in hardware. However KVM always uses paging mode to
3780 * emulate guest non-paging mode with TDP.
3781 * To emulate this behavior, SMEP/SMAP needs to be
3782 * manually disabled when guest switches to non-paging
3785 hw_cr4 &= ~(X86_CR4_SMEP | X86_CR4_SMAP);
3786 } else if (!(cr4 & X86_CR4_PAE)) {
3787 hw_cr4 &= ~X86_CR4_PAE;
3791 vmcs_writel(CR4_READ_SHADOW, cr4);
3792 vmcs_writel(GUEST_CR4, hw_cr4);
3796 static void vmx_get_segment(struct kvm_vcpu *vcpu,
3797 struct kvm_segment *var, int seg)
3799 struct vcpu_vmx *vmx = to_vmx(vcpu);
3802 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3803 *var = vmx->rmode.segs[seg];
3804 if (seg == VCPU_SREG_TR
3805 || var->selector == vmx_read_guest_seg_selector(vmx, seg))
3807 var->base = vmx_read_guest_seg_base(vmx, seg);
3808 var->selector = vmx_read_guest_seg_selector(vmx, seg);
3811 var->base = vmx_read_guest_seg_base(vmx, seg);
3812 var->limit = vmx_read_guest_seg_limit(vmx, seg);
3813 var->selector = vmx_read_guest_seg_selector(vmx, seg);
3814 ar = vmx_read_guest_seg_ar(vmx, seg);
3815 var->unusable = (ar >> 16) & 1;
3816 var->type = ar & 15;
3817 var->s = (ar >> 4) & 1;
3818 var->dpl = (ar >> 5) & 3;
3820 * Some userspaces do not preserve unusable property. Since usable
3821 * segment has to be present according to VMX spec we can use present
3822 * property to amend userspace bug by making unusable segment always
3823 * nonpresent. vmx_segment_access_rights() already marks nonpresent
3824 * segment as unusable.
3826 var->present = !var->unusable;
3827 var->avl = (ar >> 12) & 1;
3828 var->l = (ar >> 13) & 1;
3829 var->db = (ar >> 14) & 1;
3830 var->g = (ar >> 15) & 1;
3833 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
3835 struct kvm_segment s;
3837 if (to_vmx(vcpu)->rmode.vm86_active) {
3838 vmx_get_segment(vcpu, &s, seg);
3841 return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
3844 static int vmx_get_cpl(struct kvm_vcpu *vcpu)
3846 struct vcpu_vmx *vmx = to_vmx(vcpu);
3848 if (unlikely(vmx->rmode.vm86_active))
3851 int ar = vmx_read_guest_seg_ar(vmx, VCPU_SREG_SS);
3852 return VMX_AR_DPL(ar);
3856 static u32 vmx_segment_access_rights(struct kvm_segment *var)
3860 if (var->unusable || !var->present)
3863 ar = var->type & 15;
3864 ar |= (var->s & 1) << 4;
3865 ar |= (var->dpl & 3) << 5;
3866 ar |= (var->present & 1) << 7;
3867 ar |= (var->avl & 1) << 12;
3868 ar |= (var->l & 1) << 13;
3869 ar |= (var->db & 1) << 14;
3870 ar |= (var->g & 1) << 15;
3876 static void vmx_set_segment(struct kvm_vcpu *vcpu,
3877 struct kvm_segment *var, int seg)
3879 struct vcpu_vmx *vmx = to_vmx(vcpu);
3880 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3882 vmx_segment_cache_clear(vmx);
3884 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3885 vmx->rmode.segs[seg] = *var;
3886 if (seg == VCPU_SREG_TR)
3887 vmcs_write16(sf->selector, var->selector);
3889 fix_rmode_seg(seg, &vmx->rmode.segs[seg]);
3893 vmcs_writel(sf->base, var->base);
3894 vmcs_write32(sf->limit, var->limit);
3895 vmcs_write16(sf->selector, var->selector);
3898 * Fix the "Accessed" bit in AR field of segment registers for older
3900 * IA32 arch specifies that at the time of processor reset the
3901 * "Accessed" bit in the AR field of segment registers is 1. And qemu
3902 * is setting it to 0 in the userland code. This causes invalid guest
3903 * state vmexit when "unrestricted guest" mode is turned on.
3904 * Fix for this setup issue in cpu_reset is being pushed in the qemu
3905 * tree. Newer qemu binaries with that qemu fix would not need this
3908 if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
3909 var->type |= 0x1; /* Accessed */
3911 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(var));
3914 vmx->emulation_required = emulation_required(vcpu);
3917 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3919 u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
3921 *db = (ar >> 14) & 1;
3922 *l = (ar >> 13) & 1;
3925 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3927 dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
3928 dt->address = vmcs_readl(GUEST_IDTR_BASE);
3931 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3933 vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
3934 vmcs_writel(GUEST_IDTR_BASE, dt->address);
3937 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3939 dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
3940 dt->address = vmcs_readl(GUEST_GDTR_BASE);
3943 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3945 vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
3946 vmcs_writel(GUEST_GDTR_BASE, dt->address);
3949 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
3951 struct kvm_segment var;
3954 vmx_get_segment(vcpu, &var, seg);
3956 if (seg == VCPU_SREG_CS)
3958 ar = vmx_segment_access_rights(&var);
3960 if (var.base != (var.selector << 4))
3962 if (var.limit != 0xffff)
3970 static bool code_segment_valid(struct kvm_vcpu *vcpu)
3972 struct kvm_segment cs;
3973 unsigned int cs_rpl;
3975 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3976 cs_rpl = cs.selector & SEGMENT_RPL_MASK;
3980 if (~cs.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_ACCESSES_MASK))
3984 if (cs.type & VMX_AR_TYPE_WRITEABLE_MASK) {
3985 if (cs.dpl > cs_rpl)
3988 if (cs.dpl != cs_rpl)
3994 /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
3998 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
4000 struct kvm_segment ss;
4001 unsigned int ss_rpl;
4003 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
4004 ss_rpl = ss.selector & SEGMENT_RPL_MASK;
4008 if (ss.type != 3 && ss.type != 7)
4012 if (ss.dpl != ss_rpl) /* DPL != RPL */
4020 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
4022 struct kvm_segment var;
4025 vmx_get_segment(vcpu, &var, seg);
4026 rpl = var.selector & SEGMENT_RPL_MASK;
4034 if (~var.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_WRITEABLE_MASK)) {
4035 if (var.dpl < rpl) /* DPL < RPL */
4039 /* TODO: Add other members to kvm_segment_field to allow checking for other access
4045 static bool tr_valid(struct kvm_vcpu *vcpu)
4047 struct kvm_segment tr;
4049 vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
4053 if (tr.selector & SEGMENT_TI_MASK) /* TI = 1 */
4055 if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
4063 static bool ldtr_valid(struct kvm_vcpu *vcpu)
4065 struct kvm_segment ldtr;
4067 vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
4071 if (ldtr.selector & SEGMENT_TI_MASK) /* TI = 1 */
4081 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
4083 struct kvm_segment cs, ss;
4085 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
4086 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
4088 return ((cs.selector & SEGMENT_RPL_MASK) ==
4089 (ss.selector & SEGMENT_RPL_MASK));
4093 * Check if guest state is valid. Returns true if valid, false if
4095 * We assume that registers are always usable
4097 static bool guest_state_valid(struct kvm_vcpu *vcpu)
4099 if (enable_unrestricted_guest)
4102 /* real mode guest state checks */
4103 if (!is_protmode(vcpu) || (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
4104 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
4106 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
4108 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
4110 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
4112 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
4114 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
4117 /* protected mode guest state checks */
4118 if (!cs_ss_rpl_check(vcpu))
4120 if (!code_segment_valid(vcpu))
4122 if (!stack_segment_valid(vcpu))
4124 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
4126 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
4128 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
4130 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
4132 if (!tr_valid(vcpu))
4134 if (!ldtr_valid(vcpu))
4138 * - Add checks on RIP
4139 * - Add checks on RFLAGS
4145 static int init_rmode_tss(struct kvm *kvm)
4151 idx = srcu_read_lock(&kvm->srcu);
4152 fn = kvm->arch.tss_addr >> PAGE_SHIFT;
4153 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
4156 data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
4157 r = kvm_write_guest_page(kvm, fn++, &data,
4158 TSS_IOPB_BASE_OFFSET, sizeof(u16));
4161 r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
4164 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
4168 r = kvm_write_guest_page(kvm, fn, &data,
4169 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
4172 srcu_read_unlock(&kvm->srcu, idx);
4176 static int init_rmode_identity_map(struct kvm *kvm)
4179 pfn_t identity_map_pfn;
4185 /* Protect kvm->arch.ept_identity_pagetable_done. */
4186 mutex_lock(&kvm->slots_lock);
4188 if (likely(kvm->arch.ept_identity_pagetable_done))
4191 identity_map_pfn = kvm->arch.ept_identity_map_addr >> PAGE_SHIFT;
4193 r = alloc_identity_pagetable(kvm);
4197 idx = srcu_read_lock(&kvm->srcu);
4198 r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
4201 /* Set up identity-mapping pagetable for EPT in real mode */
4202 for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
4203 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
4204 _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
4205 r = kvm_write_guest_page(kvm, identity_map_pfn,
4206 &tmp, i * sizeof(tmp), sizeof(tmp));
4210 kvm->arch.ept_identity_pagetable_done = true;
4213 srcu_read_unlock(&kvm->srcu, idx);
4216 mutex_unlock(&kvm->slots_lock);
4220 static void seg_setup(int seg)
4222 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
4225 vmcs_write16(sf->selector, 0);
4226 vmcs_writel(sf->base, 0);
4227 vmcs_write32(sf->limit, 0xffff);
4229 if (seg == VCPU_SREG_CS)
4230 ar |= 0x08; /* code segment */
4232 vmcs_write32(sf->ar_bytes, ar);
4235 static int alloc_apic_access_page(struct kvm *kvm)
4240 mutex_lock(&kvm->slots_lock);
4241 if (kvm->arch.apic_access_page_done)
4243 r = __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
4244 APIC_DEFAULT_PHYS_BASE, PAGE_SIZE);
4248 page = gfn_to_page(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
4249 if (is_error_page(page)) {
4255 * Do not pin the page in memory, so that memory hot-unplug
4256 * is able to migrate it.
4259 kvm->arch.apic_access_page_done = true;
4261 mutex_unlock(&kvm->slots_lock);
4265 static int alloc_identity_pagetable(struct kvm *kvm)
4267 /* Called with kvm->slots_lock held. */
4271 BUG_ON(kvm->arch.ept_identity_pagetable_done);
4273 r = __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
4274 kvm->arch.ept_identity_map_addr, PAGE_SIZE);
4279 static int allocate_vpid(void)
4285 spin_lock(&vmx_vpid_lock);
4286 vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
4287 if (vpid < VMX_NR_VPIDS)
4288 __set_bit(vpid, vmx_vpid_bitmap);
4291 spin_unlock(&vmx_vpid_lock);
4295 static void free_vpid(int vpid)
4297 if (!enable_vpid || vpid == 0)
4299 spin_lock(&vmx_vpid_lock);
4300 __clear_bit(vpid, vmx_vpid_bitmap);
4301 spin_unlock(&vmx_vpid_lock);
4304 #define MSR_TYPE_R 1
4305 #define MSR_TYPE_W 2
4306 static void __vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
4309 int f = sizeof(unsigned long);
4311 if (!cpu_has_vmx_msr_bitmap())
4315 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
4316 * have the write-low and read-high bitmap offsets the wrong way round.
4317 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
4319 if (msr <= 0x1fff) {
4320 if (type & MSR_TYPE_R)
4322 __clear_bit(msr, msr_bitmap + 0x000 / f);
4324 if (type & MSR_TYPE_W)
4326 __clear_bit(msr, msr_bitmap + 0x800 / f);
4328 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
4330 if (type & MSR_TYPE_R)
4332 __clear_bit(msr, msr_bitmap + 0x400 / f);
4334 if (type & MSR_TYPE_W)
4336 __clear_bit(msr, msr_bitmap + 0xc00 / f);
4341 static void __vmx_enable_intercept_for_msr(unsigned long *msr_bitmap,
4344 int f = sizeof(unsigned long);
4346 if (!cpu_has_vmx_msr_bitmap())
4350 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
4351 * have the write-low and read-high bitmap offsets the wrong way round.
4352 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
4354 if (msr <= 0x1fff) {
4355 if (type & MSR_TYPE_R)
4357 __set_bit(msr, msr_bitmap + 0x000 / f);
4359 if (type & MSR_TYPE_W)
4361 __set_bit(msr, msr_bitmap + 0x800 / f);
4363 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
4365 if (type & MSR_TYPE_R)
4367 __set_bit(msr, msr_bitmap + 0x400 / f);
4369 if (type & MSR_TYPE_W)
4371 __set_bit(msr, msr_bitmap + 0xc00 / f);
4377 * If a msr is allowed by L0, we should check whether it is allowed by L1.
4378 * The corresponding bit will be cleared unless both of L0 and L1 allow it.
4380 static void nested_vmx_disable_intercept_for_msr(unsigned long *msr_bitmap_l1,
4381 unsigned long *msr_bitmap_nested,
4384 int f = sizeof(unsigned long);
4386 if (!cpu_has_vmx_msr_bitmap()) {
4392 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
4393 * have the write-low and read-high bitmap offsets the wrong way round.
4394 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
4396 if (msr <= 0x1fff) {
4397 if (type & MSR_TYPE_R &&
4398 !test_bit(msr, msr_bitmap_l1 + 0x000 / f))
4400 __clear_bit(msr, msr_bitmap_nested + 0x000 / f);
4402 if (type & MSR_TYPE_W &&
4403 !test_bit(msr, msr_bitmap_l1 + 0x800 / f))
4405 __clear_bit(msr, msr_bitmap_nested + 0x800 / f);
4407 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
4409 if (type & MSR_TYPE_R &&
4410 !test_bit(msr, msr_bitmap_l1 + 0x400 / f))
4412 __clear_bit(msr, msr_bitmap_nested + 0x400 / f);
4414 if (type & MSR_TYPE_W &&
4415 !test_bit(msr, msr_bitmap_l1 + 0xc00 / f))
4417 __clear_bit(msr, msr_bitmap_nested + 0xc00 / f);
4422 static void vmx_disable_intercept_for_msr(u32 msr, bool longmode_only)
4425 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy,
4426 msr, MSR_TYPE_R | MSR_TYPE_W);
4427 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode,
4428 msr, MSR_TYPE_R | MSR_TYPE_W);
4431 static void vmx_enable_intercept_msr_read_x2apic(u32 msr)
4433 __vmx_enable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic,
4435 __vmx_enable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic,
4439 static void vmx_disable_intercept_msr_read_x2apic(u32 msr)
4441 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic,
4443 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic,
4447 static void vmx_disable_intercept_msr_write_x2apic(u32 msr)
4449 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic,
4451 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic,
4455 static int vmx_cpu_uses_apicv(struct kvm_vcpu *vcpu)
4457 return enable_apicv && lapic_in_kernel(vcpu);
4460 static int vmx_complete_nested_posted_interrupt(struct kvm_vcpu *vcpu)
4462 struct vcpu_vmx *vmx = to_vmx(vcpu);
4467 if (vmx->nested.pi_desc &&
4468 vmx->nested.pi_pending) {
4469 vmx->nested.pi_pending = false;
4470 if (!pi_test_and_clear_on(vmx->nested.pi_desc))
4473 max_irr = find_last_bit(
4474 (unsigned long *)vmx->nested.pi_desc->pir, 256);
4479 vapic_page = kmap(vmx->nested.virtual_apic_page);
4484 __kvm_apic_update_irr(vmx->nested.pi_desc->pir, vapic_page);
4485 kunmap(vmx->nested.virtual_apic_page);
4487 status = vmcs_read16(GUEST_INTR_STATUS);
4488 if ((u8)max_irr > ((u8)status & 0xff)) {
4490 status |= (u8)max_irr;
4491 vmcs_write16(GUEST_INTR_STATUS, status);
4497 static inline bool kvm_vcpu_trigger_posted_interrupt(struct kvm_vcpu *vcpu)
4500 if (vcpu->mode == IN_GUEST_MODE) {
4501 struct vcpu_vmx *vmx = to_vmx(vcpu);
4504 * Currently, we don't support urgent interrupt,
4505 * all interrupts are recognized as non-urgent
4506 * interrupt, so we cannot post interrupts when
4509 * If the vcpu is in guest mode, it means it is
4510 * running instead of being scheduled out and
4511 * waiting in the run queue, and that's the only
4512 * case when 'SN' is set currently, warning if
4515 WARN_ON_ONCE(pi_test_sn(&vmx->pi_desc));
4517 apic->send_IPI_mask(get_cpu_mask(vcpu->cpu),
4518 POSTED_INTR_VECTOR);
4525 static int vmx_deliver_nested_posted_interrupt(struct kvm_vcpu *vcpu,
4528 struct vcpu_vmx *vmx = to_vmx(vcpu);
4530 if (is_guest_mode(vcpu) &&
4531 vector == vmx->nested.posted_intr_nv) {
4532 /* the PIR and ON have been set by L1. */
4533 kvm_vcpu_trigger_posted_interrupt(vcpu);
4535 * If a posted intr is not recognized by hardware,
4536 * we will accomplish it in the next vmentry.
4538 vmx->nested.pi_pending = true;
4539 kvm_make_request(KVM_REQ_EVENT, vcpu);
4545 * Send interrupt to vcpu via posted interrupt way.
4546 * 1. If target vcpu is running(non-root mode), send posted interrupt
4547 * notification to vcpu and hardware will sync PIR to vIRR atomically.
4548 * 2. If target vcpu isn't running(root mode), kick it to pick up the
4549 * interrupt from PIR in next vmentry.
4551 static void vmx_deliver_posted_interrupt(struct kvm_vcpu *vcpu, int vector)
4553 struct vcpu_vmx *vmx = to_vmx(vcpu);
4556 r = vmx_deliver_nested_posted_interrupt(vcpu, vector);
4560 if (pi_test_and_set_pir(vector, &vmx->pi_desc))
4563 r = pi_test_and_set_on(&vmx->pi_desc);
4564 kvm_make_request(KVM_REQ_EVENT, vcpu);
4565 if (r || !kvm_vcpu_trigger_posted_interrupt(vcpu))
4566 kvm_vcpu_kick(vcpu);
4569 static void vmx_sync_pir_to_irr(struct kvm_vcpu *vcpu)
4571 struct vcpu_vmx *vmx = to_vmx(vcpu);
4573 if (!pi_test_and_clear_on(&vmx->pi_desc))
4576 kvm_apic_update_irr(vcpu, vmx->pi_desc.pir);
4579 static void vmx_sync_pir_to_irr_dummy(struct kvm_vcpu *vcpu)
4585 * Set up the vmcs's constant host-state fields, i.e., host-state fields that
4586 * will not change in the lifetime of the guest.
4587 * Note that host-state that does change is set elsewhere. E.g., host-state
4588 * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
4590 static void vmx_set_constant_host_state(struct vcpu_vmx *vmx)
4597 vmcs_writel(HOST_CR0, read_cr0() & ~X86_CR0_TS); /* 22.2.3 */
4598 vmcs_writel(HOST_CR3, read_cr3()); /* 22.2.3 FIXME: shadow tables */
4600 /* Save the most likely value for this task's CR4 in the VMCS. */
4601 cr4 = cr4_read_shadow();
4602 vmcs_writel(HOST_CR4, cr4); /* 22.2.3, 22.2.5 */
4603 vmx->host_state.vmcs_host_cr4 = cr4;
4605 vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */
4606 #ifdef CONFIG_X86_64
4608 * Load null selectors, so we can avoid reloading them in
4609 * __vmx_load_host_state(), in case userspace uses the null selectors
4610 * too (the expected case).
4612 vmcs_write16(HOST_DS_SELECTOR, 0);
4613 vmcs_write16(HOST_ES_SELECTOR, 0);
4615 vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
4616 vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS); /* 22.2.4 */
4618 vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
4619 vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8); /* 22.2.4 */
4621 native_store_idt(&dt);
4622 vmcs_writel(HOST_IDTR_BASE, dt.address); /* 22.2.4 */
4623 vmx->host_idt_base = dt.address;
4625 vmcs_writel(HOST_RIP, vmx_return); /* 22.2.5 */
4627 rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
4628 vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
4629 rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
4630 vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl); /* 22.2.3 */
4632 if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
4633 rdmsr(MSR_IA32_CR_PAT, low32, high32);
4634 vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
4638 static void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
4640 vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
4642 vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
4643 if (is_guest_mode(&vmx->vcpu))
4644 vmx->vcpu.arch.cr4_guest_owned_bits &=
4645 ~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask;
4646 vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
4649 static u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx)
4651 u32 pin_based_exec_ctrl = vmcs_config.pin_based_exec_ctrl;
4653 if (!vmx_cpu_uses_apicv(&vmx->vcpu))
4654 pin_based_exec_ctrl &= ~PIN_BASED_POSTED_INTR;
4655 return pin_based_exec_ctrl;
4658 static u32 vmx_exec_control(struct vcpu_vmx *vmx)
4660 u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
4662 if (vmx->vcpu.arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)
4663 exec_control &= ~CPU_BASED_MOV_DR_EXITING;
4665 if (!cpu_need_tpr_shadow(&vmx->vcpu)) {
4666 exec_control &= ~CPU_BASED_TPR_SHADOW;
4667 #ifdef CONFIG_X86_64
4668 exec_control |= CPU_BASED_CR8_STORE_EXITING |
4669 CPU_BASED_CR8_LOAD_EXITING;
4673 exec_control |= CPU_BASED_CR3_STORE_EXITING |
4674 CPU_BASED_CR3_LOAD_EXITING |
4675 CPU_BASED_INVLPG_EXITING;
4676 return exec_control;
4679 static u32 vmx_secondary_exec_control(struct vcpu_vmx *vmx)
4681 u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
4682 if (!cpu_need_virtualize_apic_accesses(&vmx->vcpu))
4683 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
4685 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
4687 exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
4688 enable_unrestricted_guest = 0;
4689 /* Enable INVPCID for non-ept guests may cause performance regression. */
4690 exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
4692 if (!enable_unrestricted_guest)
4693 exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
4695 exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
4696 if (!vmx_cpu_uses_apicv(&vmx->vcpu))
4697 exec_control &= ~(SECONDARY_EXEC_APIC_REGISTER_VIRT |
4698 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4699 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
4700 /* SECONDARY_EXEC_SHADOW_VMCS is enabled when L1 executes VMPTRLD
4702 We can NOT enable shadow_vmcs here because we don't have yet
4705 exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
4706 /* PML is enabled/disabled in creating/destorying vcpu */
4707 exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
4709 /* Currently, we allow L1 guest to directly run pcommit instruction. */
4710 exec_control &= ~SECONDARY_EXEC_PCOMMIT;
4712 return exec_control;
4715 static void ept_set_mmio_spte_mask(void)
4718 * EPT Misconfigurations can be generated if the value of bits 2:0
4719 * of an EPT paging-structure entry is 110b (write/execute).
4720 * Also, magic bits (0x3ull << 62) is set to quickly identify mmio
4723 kvm_mmu_set_mmio_spte_mask((0x3ull << 62) | 0x6ull);
4726 #define VMX_XSS_EXIT_BITMAP 0
4728 * Sets up the vmcs for emulated real mode.
4730 static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
4732 #ifdef CONFIG_X86_64
4738 vmcs_write64(IO_BITMAP_A, __pa(vmx_io_bitmap_a));
4739 vmcs_write64(IO_BITMAP_B, __pa(vmx_io_bitmap_b));
4741 if (enable_shadow_vmcs) {
4742 vmcs_write64(VMREAD_BITMAP, __pa(vmx_vmread_bitmap));
4743 vmcs_write64(VMWRITE_BITMAP, __pa(vmx_vmwrite_bitmap));
4745 if (cpu_has_vmx_msr_bitmap())
4746 vmcs_write64(MSR_BITMAP, __pa(vmx_msr_bitmap_legacy));
4748 vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
4751 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, vmx_pin_based_exec_ctrl(vmx));
4753 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, vmx_exec_control(vmx));
4755 if (cpu_has_secondary_exec_ctrls())
4756 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
4757 vmx_secondary_exec_control(vmx));
4759 if (vmx_cpu_uses_apicv(&vmx->vcpu)) {
4760 vmcs_write64(EOI_EXIT_BITMAP0, 0);
4761 vmcs_write64(EOI_EXIT_BITMAP1, 0);
4762 vmcs_write64(EOI_EXIT_BITMAP2, 0);
4763 vmcs_write64(EOI_EXIT_BITMAP3, 0);
4765 vmcs_write16(GUEST_INTR_STATUS, 0);
4767 vmcs_write64(POSTED_INTR_NV, POSTED_INTR_VECTOR);
4768 vmcs_write64(POSTED_INTR_DESC_ADDR, __pa((&vmx->pi_desc)));
4772 vmcs_write32(PLE_GAP, ple_gap);
4773 vmx->ple_window = ple_window;
4774 vmx->ple_window_dirty = true;
4777 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
4778 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
4779 vmcs_write32(CR3_TARGET_COUNT, 0); /* 22.2.1 */
4781 vmcs_write16(HOST_FS_SELECTOR, 0); /* 22.2.4 */
4782 vmcs_write16(HOST_GS_SELECTOR, 0); /* 22.2.4 */
4783 vmx_set_constant_host_state(vmx);
4784 #ifdef CONFIG_X86_64
4785 rdmsrl(MSR_FS_BASE, a);
4786 vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
4787 rdmsrl(MSR_GS_BASE, a);
4788 vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
4790 vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
4791 vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
4794 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
4795 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
4796 vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host));
4797 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
4798 vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest));
4800 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
4801 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
4803 for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i) {
4804 u32 index = vmx_msr_index[i];
4805 u32 data_low, data_high;
4808 if (rdmsr_safe(index, &data_low, &data_high) < 0)
4810 if (wrmsr_safe(index, data_low, data_high) < 0)
4812 vmx->guest_msrs[j].index = i;
4813 vmx->guest_msrs[j].data = 0;
4814 vmx->guest_msrs[j].mask = -1ull;
4819 vm_exit_controls_init(vmx, vmcs_config.vmexit_ctrl);
4821 /* 22.2.1, 20.8.1 */
4822 vm_entry_controls_init(vmx, vmcs_config.vmentry_ctrl);
4824 vmcs_writel(CR0_GUEST_HOST_MASK, ~0UL);
4825 set_cr4_guest_host_mask(vmx);
4827 if (vmx_xsaves_supported())
4828 vmcs_write64(XSS_EXIT_BITMAP, VMX_XSS_EXIT_BITMAP);
4833 static void vmx_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
4835 struct vcpu_vmx *vmx = to_vmx(vcpu);
4836 struct msr_data apic_base_msr;
4839 vmx->rmode.vm86_active = 0;
4841 vmx->soft_vnmi_blocked = 0;
4843 vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
4844 kvm_set_cr8(vcpu, 0);
4847 apic_base_msr.data = APIC_DEFAULT_PHYS_BASE |
4848 MSR_IA32_APICBASE_ENABLE;
4849 if (kvm_vcpu_is_reset_bsp(vcpu))
4850 apic_base_msr.data |= MSR_IA32_APICBASE_BSP;
4851 apic_base_msr.host_initiated = true;
4852 kvm_set_apic_base(vcpu, &apic_base_msr);
4855 vmx_segment_cache_clear(vmx);
4857 seg_setup(VCPU_SREG_CS);
4858 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
4859 vmcs_write32(GUEST_CS_BASE, 0xffff0000);
4861 seg_setup(VCPU_SREG_DS);
4862 seg_setup(VCPU_SREG_ES);
4863 seg_setup(VCPU_SREG_FS);
4864 seg_setup(VCPU_SREG_GS);
4865 seg_setup(VCPU_SREG_SS);
4867 vmcs_write16(GUEST_TR_SELECTOR, 0);
4868 vmcs_writel(GUEST_TR_BASE, 0);
4869 vmcs_write32(GUEST_TR_LIMIT, 0xffff);
4870 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
4872 vmcs_write16(GUEST_LDTR_SELECTOR, 0);
4873 vmcs_writel(GUEST_LDTR_BASE, 0);
4874 vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
4875 vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
4878 vmcs_write32(GUEST_SYSENTER_CS, 0);
4879 vmcs_writel(GUEST_SYSENTER_ESP, 0);
4880 vmcs_writel(GUEST_SYSENTER_EIP, 0);
4881 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
4884 vmcs_writel(GUEST_RFLAGS, 0x02);
4885 kvm_rip_write(vcpu, 0xfff0);
4887 vmcs_writel(GUEST_GDTR_BASE, 0);
4888 vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
4890 vmcs_writel(GUEST_IDTR_BASE, 0);
4891 vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
4893 vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
4894 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
4895 vmcs_write32(GUEST_PENDING_DBG_EXCEPTIONS, 0);
4899 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */
4901 if (cpu_has_vmx_tpr_shadow() && !init_event) {
4902 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
4903 if (cpu_need_tpr_shadow(vcpu))
4904 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
4905 __pa(vcpu->arch.apic->regs));
4906 vmcs_write32(TPR_THRESHOLD, 0);
4909 kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
4911 if (vmx_cpu_uses_apicv(vcpu))
4912 memset(&vmx->pi_desc, 0, sizeof(struct pi_desc));
4915 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
4917 cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
4918 vmx_set_cr0(vcpu, cr0); /* enter rmode */
4919 vmx->vcpu.arch.cr0 = cr0;
4920 vmx_set_cr4(vcpu, 0);
4922 vmx_set_efer(vcpu, 0);
4923 vmx_fpu_activate(vcpu);
4924 update_exception_bitmap(vcpu);
4926 vpid_sync_context(vmx->vpid);
4930 * In nested virtualization, check if L1 asked to exit on external interrupts.
4931 * For most existing hypervisors, this will always return true.
4933 static bool nested_exit_on_intr(struct kvm_vcpu *vcpu)
4935 return get_vmcs12(vcpu)->pin_based_vm_exec_control &
4936 PIN_BASED_EXT_INTR_MASK;
4940 * In nested virtualization, check if L1 has set
4941 * VM_EXIT_ACK_INTR_ON_EXIT
4943 static bool nested_exit_intr_ack_set(struct kvm_vcpu *vcpu)
4945 return get_vmcs12(vcpu)->vm_exit_controls &
4946 VM_EXIT_ACK_INTR_ON_EXIT;
4949 static bool nested_exit_on_nmi(struct kvm_vcpu *vcpu)
4951 return get_vmcs12(vcpu)->pin_based_vm_exec_control &
4952 PIN_BASED_NMI_EXITING;
4955 static void enable_irq_window(struct kvm_vcpu *vcpu)
4957 u32 cpu_based_vm_exec_control;
4959 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
4960 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
4961 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
4964 static void enable_nmi_window(struct kvm_vcpu *vcpu)
4966 u32 cpu_based_vm_exec_control;
4968 if (!cpu_has_virtual_nmis() ||
4969 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
4970 enable_irq_window(vcpu);
4974 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
4975 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_NMI_PENDING;
4976 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
4979 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
4981 struct vcpu_vmx *vmx = to_vmx(vcpu);
4983 int irq = vcpu->arch.interrupt.nr;
4985 trace_kvm_inj_virq(irq);
4987 ++vcpu->stat.irq_injections;
4988 if (vmx->rmode.vm86_active) {
4990 if (vcpu->arch.interrupt.soft)
4991 inc_eip = vcpu->arch.event_exit_inst_len;
4992 if (kvm_inject_realmode_interrupt(vcpu, irq, inc_eip) != EMULATE_DONE)
4993 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4996 intr = irq | INTR_INFO_VALID_MASK;
4997 if (vcpu->arch.interrupt.soft) {
4998 intr |= INTR_TYPE_SOFT_INTR;
4999 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
5000 vmx->vcpu.arch.event_exit_inst_len);
5002 intr |= INTR_TYPE_EXT_INTR;
5003 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
5006 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
5008 struct vcpu_vmx *vmx = to_vmx(vcpu);
5010 if (is_guest_mode(vcpu))
5013 if (!cpu_has_virtual_nmis()) {
5015 * Tracking the NMI-blocked state in software is built upon
5016 * finding the next open IRQ window. This, in turn, depends on
5017 * well-behaving guests: They have to keep IRQs disabled at
5018 * least as long as the NMI handler runs. Otherwise we may
5019 * cause NMI nesting, maybe breaking the guest. But as this is
5020 * highly unlikely, we can live with the residual risk.
5022 vmx->soft_vnmi_blocked = 1;
5023 vmx->vnmi_blocked_time = 0;
5026 ++vcpu->stat.nmi_injections;
5027 vmx->nmi_known_unmasked = false;
5028 if (vmx->rmode.vm86_active) {
5029 if (kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0) != EMULATE_DONE)
5030 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5033 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
5034 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
5037 static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
5039 if (!cpu_has_virtual_nmis())
5040 return to_vmx(vcpu)->soft_vnmi_blocked;
5041 if (to_vmx(vcpu)->nmi_known_unmasked)
5043 return vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
5046 static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
5048 struct vcpu_vmx *vmx = to_vmx(vcpu);
5050 if (!cpu_has_virtual_nmis()) {
5051 if (vmx->soft_vnmi_blocked != masked) {
5052 vmx->soft_vnmi_blocked = masked;
5053 vmx->vnmi_blocked_time = 0;
5056 vmx->nmi_known_unmasked = !masked;
5058 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
5059 GUEST_INTR_STATE_NMI);
5061 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
5062 GUEST_INTR_STATE_NMI);
5066 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
5068 if (to_vmx(vcpu)->nested.nested_run_pending)
5071 if (!cpu_has_virtual_nmis() && to_vmx(vcpu)->soft_vnmi_blocked)
5074 return !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
5075 (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
5076 | GUEST_INTR_STATE_NMI));
5079 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
5081 return (!to_vmx(vcpu)->nested.nested_run_pending &&
5082 vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
5083 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
5084 (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
5087 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
5091 ret = x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, addr,
5095 kvm->arch.tss_addr = addr;
5096 return init_rmode_tss(kvm);
5099 static bool rmode_exception(struct kvm_vcpu *vcpu, int vec)
5104 * Update instruction length as we may reinject the exception
5105 * from user space while in guest debugging mode.
5107 to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
5108 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
5109 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
5113 if (vcpu->guest_debug &
5114 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
5131 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
5132 int vec, u32 err_code)
5135 * Instruction with address size override prefix opcode 0x67
5136 * Cause the #SS fault with 0 error code in VM86 mode.
5138 if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0) {
5139 if (emulate_instruction(vcpu, 0) == EMULATE_DONE) {
5140 if (vcpu->arch.halt_request) {
5141 vcpu->arch.halt_request = 0;
5142 return kvm_vcpu_halt(vcpu);
5150 * Forward all other exceptions that are valid in real mode.
5151 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
5152 * the required debugging infrastructure rework.
5154 kvm_queue_exception(vcpu, vec);
5159 * Trigger machine check on the host. We assume all the MSRs are already set up
5160 * by the CPU and that we still run on the same CPU as the MCE occurred on.
5161 * We pass a fake environment to the machine check handler because we want
5162 * the guest to be always treated like user space, no matter what context
5163 * it used internally.
5165 static void kvm_machine_check(void)
5167 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
5168 struct pt_regs regs = {
5169 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
5170 .flags = X86_EFLAGS_IF,
5173 do_machine_check(®s, 0);
5177 static int handle_machine_check(struct kvm_vcpu *vcpu)
5179 /* already handled by vcpu_run */
5183 static int handle_exception(struct kvm_vcpu *vcpu)
5185 struct vcpu_vmx *vmx = to_vmx(vcpu);
5186 struct kvm_run *kvm_run = vcpu->run;
5187 u32 intr_info, ex_no, error_code;
5188 unsigned long cr2, rip, dr6;
5190 enum emulation_result er;
5192 vect_info = vmx->idt_vectoring_info;
5193 intr_info = vmx->exit_intr_info;
5195 if (is_machine_check(intr_info))
5196 return handle_machine_check(vcpu);
5198 if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR)
5199 return 1; /* already handled by vmx_vcpu_run() */
5201 if (is_no_device(intr_info)) {
5202 vmx_fpu_activate(vcpu);
5206 if (is_invalid_opcode(intr_info)) {
5207 if (is_guest_mode(vcpu)) {
5208 kvm_queue_exception(vcpu, UD_VECTOR);
5211 er = emulate_instruction(vcpu, EMULTYPE_TRAP_UD);
5212 if (er != EMULATE_DONE)
5213 kvm_queue_exception(vcpu, UD_VECTOR);
5218 if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
5219 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
5222 * The #PF with PFEC.RSVD = 1 indicates the guest is accessing
5223 * MMIO, it is better to report an internal error.
5224 * See the comments in vmx_handle_exit.
5226 if ((vect_info & VECTORING_INFO_VALID_MASK) &&
5227 !(is_page_fault(intr_info) && !(error_code & PFERR_RSVD_MASK))) {
5228 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5229 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
5230 vcpu->run->internal.ndata = 3;
5231 vcpu->run->internal.data[0] = vect_info;
5232 vcpu->run->internal.data[1] = intr_info;
5233 vcpu->run->internal.data[2] = error_code;
5237 if (is_page_fault(intr_info)) {
5238 /* EPT won't cause page fault directly */
5240 cr2 = vmcs_readl(EXIT_QUALIFICATION);
5241 trace_kvm_page_fault(cr2, error_code);
5243 if (kvm_event_needs_reinjection(vcpu))
5244 kvm_mmu_unprotect_page_virt(vcpu, cr2);
5245 return kvm_mmu_page_fault(vcpu, cr2, error_code, NULL, 0);
5248 ex_no = intr_info & INTR_INFO_VECTOR_MASK;
5250 if (vmx->rmode.vm86_active && rmode_exception(vcpu, ex_no))
5251 return handle_rmode_exception(vcpu, ex_no, error_code);
5255 dr6 = vmcs_readl(EXIT_QUALIFICATION);
5256 if (!(vcpu->guest_debug &
5257 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
5258 vcpu->arch.dr6 &= ~15;
5259 vcpu->arch.dr6 |= dr6 | DR6_RTM;
5260 if (!(dr6 & ~DR6_RESERVED)) /* icebp */
5261 skip_emulated_instruction(vcpu);
5263 kvm_queue_exception(vcpu, DB_VECTOR);
5266 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
5267 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
5271 * Update instruction length as we may reinject #BP from
5272 * user space while in guest debugging mode. Reading it for
5273 * #DB as well causes no harm, it is not used in that case.
5275 vmx->vcpu.arch.event_exit_inst_len =
5276 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
5277 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5278 rip = kvm_rip_read(vcpu);
5279 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
5280 kvm_run->debug.arch.exception = ex_no;
5283 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
5284 kvm_run->ex.exception = ex_no;
5285 kvm_run->ex.error_code = error_code;
5291 static int handle_external_interrupt(struct kvm_vcpu *vcpu)
5293 ++vcpu->stat.irq_exits;
5297 static int handle_triple_fault(struct kvm_vcpu *vcpu)
5299 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
5303 static int handle_io(struct kvm_vcpu *vcpu)
5305 unsigned long exit_qualification;
5306 int size, in, string;
5309 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5310 string = (exit_qualification & 16) != 0;
5311 in = (exit_qualification & 8) != 0;
5313 ++vcpu->stat.io_exits;
5316 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
5318 port = exit_qualification >> 16;
5319 size = (exit_qualification & 7) + 1;
5320 skip_emulated_instruction(vcpu);
5322 return kvm_fast_pio_out(vcpu, size, port);
5326 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
5329 * Patch in the VMCALL instruction:
5331 hypercall[0] = 0x0f;
5332 hypercall[1] = 0x01;
5333 hypercall[2] = 0xc1;
5336 static bool nested_cr0_valid(struct kvm_vcpu *vcpu, unsigned long val)
5338 unsigned long always_on = VMXON_CR0_ALWAYSON;
5339 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5341 if (to_vmx(vcpu)->nested.nested_vmx_secondary_ctls_high &
5342 SECONDARY_EXEC_UNRESTRICTED_GUEST &&
5343 nested_cpu_has2(vmcs12, SECONDARY_EXEC_UNRESTRICTED_GUEST))
5344 always_on &= ~(X86_CR0_PE | X86_CR0_PG);
5345 return (val & always_on) == always_on;
5348 /* called to set cr0 as appropriate for a mov-to-cr0 exit. */
5349 static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
5351 if (is_guest_mode(vcpu)) {
5352 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5353 unsigned long orig_val = val;
5356 * We get here when L2 changed cr0 in a way that did not change
5357 * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
5358 * but did change L0 shadowed bits. So we first calculate the
5359 * effective cr0 value that L1 would like to write into the
5360 * hardware. It consists of the L2-owned bits from the new
5361 * value combined with the L1-owned bits from L1's guest_cr0.
5363 val = (val & ~vmcs12->cr0_guest_host_mask) |
5364 (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask);
5366 if (!nested_cr0_valid(vcpu, val))
5369 if (kvm_set_cr0(vcpu, val))
5371 vmcs_writel(CR0_READ_SHADOW, orig_val);
5374 if (to_vmx(vcpu)->nested.vmxon &&
5375 ((val & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON))
5377 return kvm_set_cr0(vcpu, val);
5381 static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
5383 if (is_guest_mode(vcpu)) {
5384 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5385 unsigned long orig_val = val;
5387 /* analogously to handle_set_cr0 */
5388 val = (val & ~vmcs12->cr4_guest_host_mask) |
5389 (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask);
5390 if (kvm_set_cr4(vcpu, val))
5392 vmcs_writel(CR4_READ_SHADOW, orig_val);
5395 return kvm_set_cr4(vcpu, val);
5398 /* called to set cr0 as approriate for clts instruction exit. */
5399 static void handle_clts(struct kvm_vcpu *vcpu)
5401 if (is_guest_mode(vcpu)) {
5403 * We get here when L2 did CLTS, and L1 didn't shadow CR0.TS
5404 * but we did (!fpu_active). We need to keep GUEST_CR0.TS on,
5405 * just pretend it's off (also in arch.cr0 for fpu_activate).
5407 vmcs_writel(CR0_READ_SHADOW,
5408 vmcs_readl(CR0_READ_SHADOW) & ~X86_CR0_TS);
5409 vcpu->arch.cr0 &= ~X86_CR0_TS;
5411 vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
5414 static int handle_cr(struct kvm_vcpu *vcpu)
5416 unsigned long exit_qualification, val;
5421 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5422 cr = exit_qualification & 15;
5423 reg = (exit_qualification >> 8) & 15;
5424 switch ((exit_qualification >> 4) & 3) {
5425 case 0: /* mov to cr */
5426 val = kvm_register_readl(vcpu, reg);
5427 trace_kvm_cr_write(cr, val);
5430 err = handle_set_cr0(vcpu, val);
5431 kvm_complete_insn_gp(vcpu, err);
5434 err = kvm_set_cr3(vcpu, val);
5435 kvm_complete_insn_gp(vcpu, err);
5438 err = handle_set_cr4(vcpu, val);
5439 kvm_complete_insn_gp(vcpu, err);
5442 u8 cr8_prev = kvm_get_cr8(vcpu);
5444 err = kvm_set_cr8(vcpu, cr8);
5445 kvm_complete_insn_gp(vcpu, err);
5446 if (lapic_in_kernel(vcpu))
5448 if (cr8_prev <= cr8)
5450 vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
5457 trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
5458 skip_emulated_instruction(vcpu);
5459 vmx_fpu_activate(vcpu);
5461 case 1: /*mov from cr*/
5464 val = kvm_read_cr3(vcpu);
5465 kvm_register_write(vcpu, reg, val);
5466 trace_kvm_cr_read(cr, val);
5467 skip_emulated_instruction(vcpu);
5470 val = kvm_get_cr8(vcpu);
5471 kvm_register_write(vcpu, reg, val);
5472 trace_kvm_cr_read(cr, val);
5473 skip_emulated_instruction(vcpu);
5478 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
5479 trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
5480 kvm_lmsw(vcpu, val);
5482 skip_emulated_instruction(vcpu);
5487 vcpu->run->exit_reason = 0;
5488 vcpu_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
5489 (int)(exit_qualification >> 4) & 3, cr);
5493 static int handle_dr(struct kvm_vcpu *vcpu)
5495 unsigned long exit_qualification;
5498 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5499 dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
5501 /* First, if DR does not exist, trigger UD */
5502 if (!kvm_require_dr(vcpu, dr))
5505 /* Do not handle if the CPL > 0, will trigger GP on re-entry */
5506 if (!kvm_require_cpl(vcpu, 0))
5508 dr7 = vmcs_readl(GUEST_DR7);
5511 * As the vm-exit takes precedence over the debug trap, we
5512 * need to emulate the latter, either for the host or the
5513 * guest debugging itself.
5515 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
5516 vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
5517 vcpu->run->debug.arch.dr7 = dr7;
5518 vcpu->run->debug.arch.pc = kvm_get_linear_rip(vcpu);
5519 vcpu->run->debug.arch.exception = DB_VECTOR;
5520 vcpu->run->exit_reason = KVM_EXIT_DEBUG;
5523 vcpu->arch.dr6 &= ~15;
5524 vcpu->arch.dr6 |= DR6_BD | DR6_RTM;
5525 kvm_queue_exception(vcpu, DB_VECTOR);
5530 if (vcpu->guest_debug == 0) {
5531 u32 cpu_based_vm_exec_control;
5533 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5534 cpu_based_vm_exec_control &= ~CPU_BASED_MOV_DR_EXITING;
5535 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
5538 * No more DR vmexits; force a reload of the debug registers
5539 * and reenter on this instruction. The next vmexit will
5540 * retrieve the full state of the debug registers.
5542 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
5546 reg = DEBUG_REG_ACCESS_REG(exit_qualification);
5547 if (exit_qualification & TYPE_MOV_FROM_DR) {
5550 if (kvm_get_dr(vcpu, dr, &val))
5552 kvm_register_write(vcpu, reg, val);
5554 if (kvm_set_dr(vcpu, dr, kvm_register_readl(vcpu, reg)))
5557 skip_emulated_instruction(vcpu);
5561 static u64 vmx_get_dr6(struct kvm_vcpu *vcpu)
5563 return vcpu->arch.dr6;
5566 static void vmx_set_dr6(struct kvm_vcpu *vcpu, unsigned long val)
5570 static void vmx_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
5572 u32 cpu_based_vm_exec_control;
5574 get_debugreg(vcpu->arch.db[0], 0);
5575 get_debugreg(vcpu->arch.db[1], 1);
5576 get_debugreg(vcpu->arch.db[2], 2);
5577 get_debugreg(vcpu->arch.db[3], 3);
5578 get_debugreg(vcpu->arch.dr6, 6);
5579 vcpu->arch.dr7 = vmcs_readl(GUEST_DR7);
5581 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
5583 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5584 cpu_based_vm_exec_control |= CPU_BASED_MOV_DR_EXITING;
5585 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
5588 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
5590 vmcs_writel(GUEST_DR7, val);
5593 static int handle_cpuid(struct kvm_vcpu *vcpu)
5595 kvm_emulate_cpuid(vcpu);
5599 static int handle_rdmsr(struct kvm_vcpu *vcpu)
5601 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
5602 struct msr_data msr_info;
5604 msr_info.index = ecx;
5605 msr_info.host_initiated = false;
5606 if (vmx_get_msr(vcpu, &msr_info)) {
5607 trace_kvm_msr_read_ex(ecx);
5608 kvm_inject_gp(vcpu, 0);
5612 trace_kvm_msr_read(ecx, msr_info.data);
5614 /* FIXME: handling of bits 32:63 of rax, rdx */
5615 vcpu->arch.regs[VCPU_REGS_RAX] = msr_info.data & -1u;
5616 vcpu->arch.regs[VCPU_REGS_RDX] = (msr_info.data >> 32) & -1u;
5617 skip_emulated_instruction(vcpu);
5621 static int handle_wrmsr(struct kvm_vcpu *vcpu)
5623 struct msr_data msr;
5624 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
5625 u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
5626 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
5630 msr.host_initiated = false;
5631 if (kvm_set_msr(vcpu, &msr) != 0) {
5632 trace_kvm_msr_write_ex(ecx, data);
5633 kvm_inject_gp(vcpu, 0);
5637 trace_kvm_msr_write(ecx, data);
5638 skip_emulated_instruction(vcpu);
5642 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
5644 kvm_make_request(KVM_REQ_EVENT, vcpu);
5648 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
5650 u32 cpu_based_vm_exec_control;
5652 /* clear pending irq */
5653 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5654 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
5655 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
5657 kvm_make_request(KVM_REQ_EVENT, vcpu);
5659 ++vcpu->stat.irq_window_exits;
5663 static int handle_halt(struct kvm_vcpu *vcpu)
5665 return kvm_emulate_halt(vcpu);
5668 static int handle_vmcall(struct kvm_vcpu *vcpu)
5670 kvm_emulate_hypercall(vcpu);
5674 static int handle_invd(struct kvm_vcpu *vcpu)
5676 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
5679 static int handle_invlpg(struct kvm_vcpu *vcpu)
5681 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5683 kvm_mmu_invlpg(vcpu, exit_qualification);
5684 skip_emulated_instruction(vcpu);
5688 static int handle_rdpmc(struct kvm_vcpu *vcpu)
5692 err = kvm_rdpmc(vcpu);
5693 kvm_complete_insn_gp(vcpu, err);
5698 static int handle_wbinvd(struct kvm_vcpu *vcpu)
5700 kvm_emulate_wbinvd(vcpu);
5704 static int handle_xsetbv(struct kvm_vcpu *vcpu)
5706 u64 new_bv = kvm_read_edx_eax(vcpu);
5707 u32 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
5709 if (kvm_set_xcr(vcpu, index, new_bv) == 0)
5710 skip_emulated_instruction(vcpu);
5714 static int handle_xsaves(struct kvm_vcpu *vcpu)
5716 skip_emulated_instruction(vcpu);
5717 WARN(1, "this should never happen\n");
5721 static int handle_xrstors(struct kvm_vcpu *vcpu)
5723 skip_emulated_instruction(vcpu);
5724 WARN(1, "this should never happen\n");
5728 static int handle_apic_access(struct kvm_vcpu *vcpu)
5730 if (likely(fasteoi)) {
5731 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5732 int access_type, offset;
5734 access_type = exit_qualification & APIC_ACCESS_TYPE;
5735 offset = exit_qualification & APIC_ACCESS_OFFSET;
5737 * Sane guest uses MOV to write EOI, with written value
5738 * not cared. So make a short-circuit here by avoiding
5739 * heavy instruction emulation.
5741 if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
5742 (offset == APIC_EOI)) {
5743 kvm_lapic_set_eoi(vcpu);
5744 skip_emulated_instruction(vcpu);
5748 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
5751 static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu)
5753 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5754 int vector = exit_qualification & 0xff;
5756 /* EOI-induced VM exit is trap-like and thus no need to adjust IP */
5757 kvm_apic_set_eoi_accelerated(vcpu, vector);
5761 static int handle_apic_write(struct kvm_vcpu *vcpu)
5763 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5764 u32 offset = exit_qualification & 0xfff;
5766 /* APIC-write VM exit is trap-like and thus no need to adjust IP */
5767 kvm_apic_write_nodecode(vcpu, offset);
5771 static int handle_task_switch(struct kvm_vcpu *vcpu)
5773 struct vcpu_vmx *vmx = to_vmx(vcpu);
5774 unsigned long exit_qualification;
5775 bool has_error_code = false;
5778 int reason, type, idt_v, idt_index;
5780 idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
5781 idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK);
5782 type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
5784 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5786 reason = (u32)exit_qualification >> 30;
5787 if (reason == TASK_SWITCH_GATE && idt_v) {
5789 case INTR_TYPE_NMI_INTR:
5790 vcpu->arch.nmi_injected = false;
5791 vmx_set_nmi_mask(vcpu, true);
5793 case INTR_TYPE_EXT_INTR:
5794 case INTR_TYPE_SOFT_INTR:
5795 kvm_clear_interrupt_queue(vcpu);
5797 case INTR_TYPE_HARD_EXCEPTION:
5798 if (vmx->idt_vectoring_info &
5799 VECTORING_INFO_DELIVER_CODE_MASK) {
5800 has_error_code = true;
5802 vmcs_read32(IDT_VECTORING_ERROR_CODE);
5805 case INTR_TYPE_SOFT_EXCEPTION:
5806 kvm_clear_exception_queue(vcpu);
5812 tss_selector = exit_qualification;
5814 if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
5815 type != INTR_TYPE_EXT_INTR &&
5816 type != INTR_TYPE_NMI_INTR))
5817 skip_emulated_instruction(vcpu);
5819 if (kvm_task_switch(vcpu, tss_selector,
5820 type == INTR_TYPE_SOFT_INTR ? idt_index : -1, reason,
5821 has_error_code, error_code) == EMULATE_FAIL) {
5822 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5823 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5824 vcpu->run->internal.ndata = 0;
5829 * TODO: What about debug traps on tss switch?
5830 * Are we supposed to inject them and update dr6?
5836 static int handle_ept_violation(struct kvm_vcpu *vcpu)
5838 unsigned long exit_qualification;
5843 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5845 gla_validity = (exit_qualification >> 7) & 0x3;
5846 if (gla_validity != 0x3 && gla_validity != 0x1 && gla_validity != 0) {
5847 printk(KERN_ERR "EPT: Handling EPT violation failed!\n");
5848 printk(KERN_ERR "EPT: GPA: 0x%lx, GVA: 0x%lx\n",
5849 (long unsigned int)vmcs_read64(GUEST_PHYSICAL_ADDRESS),
5850 vmcs_readl(GUEST_LINEAR_ADDRESS));
5851 printk(KERN_ERR "EPT: Exit qualification is 0x%lx\n",
5852 (long unsigned int)exit_qualification);
5853 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
5854 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_VIOLATION;
5859 * EPT violation happened while executing iret from NMI,
5860 * "blocked by NMI" bit has to be set before next VM entry.
5861 * There are errata that may cause this bit to not be set:
5864 if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5865 cpu_has_virtual_nmis() &&
5866 (exit_qualification & INTR_INFO_UNBLOCK_NMI))
5867 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, GUEST_INTR_STATE_NMI);
5869 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5870 trace_kvm_page_fault(gpa, exit_qualification);
5872 /* It is a write fault? */
5873 error_code = exit_qualification & PFERR_WRITE_MASK;
5874 /* It is a fetch fault? */
5875 error_code |= (exit_qualification << 2) & PFERR_FETCH_MASK;
5876 /* ept page table is present? */
5877 error_code |= (exit_qualification >> 3) & PFERR_PRESENT_MASK;
5879 vcpu->arch.exit_qualification = exit_qualification;
5881 return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0);
5884 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
5889 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5890 if (!kvm_io_bus_write(vcpu, KVM_FAST_MMIO_BUS, gpa, 0, NULL)) {
5891 skip_emulated_instruction(vcpu);
5892 trace_kvm_fast_mmio(gpa);
5896 ret = handle_mmio_page_fault_common(vcpu, gpa, true);
5897 if (likely(ret == RET_MMIO_PF_EMULATE))
5898 return x86_emulate_instruction(vcpu, gpa, 0, NULL, 0) ==
5901 if (unlikely(ret == RET_MMIO_PF_INVALID))
5902 return kvm_mmu_page_fault(vcpu, gpa, 0, NULL, 0);
5904 if (unlikely(ret == RET_MMIO_PF_RETRY))
5907 /* It is the real ept misconfig */
5910 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
5911 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_MISCONFIG;
5916 static int handle_nmi_window(struct kvm_vcpu *vcpu)
5918 u32 cpu_based_vm_exec_control;
5920 /* clear pending NMI */
5921 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5922 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
5923 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
5924 ++vcpu->stat.nmi_window_exits;
5925 kvm_make_request(KVM_REQ_EVENT, vcpu);
5930 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
5932 struct vcpu_vmx *vmx = to_vmx(vcpu);
5933 enum emulation_result err = EMULATE_DONE;
5936 bool intr_window_requested;
5937 unsigned count = 130;
5939 cpu_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5940 intr_window_requested = cpu_exec_ctrl & CPU_BASED_VIRTUAL_INTR_PENDING;
5942 while (vmx->emulation_required && count-- != 0) {
5943 if (intr_window_requested && vmx_interrupt_allowed(vcpu))
5944 return handle_interrupt_window(&vmx->vcpu);
5946 if (test_bit(KVM_REQ_EVENT, &vcpu->requests))
5949 err = emulate_instruction(vcpu, EMULTYPE_NO_REEXECUTE);
5951 if (err == EMULATE_USER_EXIT) {
5952 ++vcpu->stat.mmio_exits;
5957 if (err != EMULATE_DONE) {
5958 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5959 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5960 vcpu->run->internal.ndata = 0;
5964 if (vcpu->arch.halt_request) {
5965 vcpu->arch.halt_request = 0;
5966 ret = kvm_vcpu_halt(vcpu);
5970 if (signal_pending(current))
5980 static int __grow_ple_window(int val)
5982 if (ple_window_grow < 1)
5985 val = min(val, ple_window_actual_max);
5987 if (ple_window_grow < ple_window)
5988 val *= ple_window_grow;
5990 val += ple_window_grow;
5995 static int __shrink_ple_window(int val, int modifier, int minimum)
6000 if (modifier < ple_window)
6005 return max(val, minimum);
6008 static void grow_ple_window(struct kvm_vcpu *vcpu)
6010 struct vcpu_vmx *vmx = to_vmx(vcpu);
6011 int old = vmx->ple_window;
6013 vmx->ple_window = __grow_ple_window(old);
6015 if (vmx->ple_window != old)
6016 vmx->ple_window_dirty = true;
6018 trace_kvm_ple_window_grow(vcpu->vcpu_id, vmx->ple_window, old);
6021 static void shrink_ple_window(struct kvm_vcpu *vcpu)
6023 struct vcpu_vmx *vmx = to_vmx(vcpu);
6024 int old = vmx->ple_window;
6026 vmx->ple_window = __shrink_ple_window(old,
6027 ple_window_shrink, ple_window);
6029 if (vmx->ple_window != old)
6030 vmx->ple_window_dirty = true;
6032 trace_kvm_ple_window_shrink(vcpu->vcpu_id, vmx->ple_window, old);
6036 * ple_window_actual_max is computed to be one grow_ple_window() below
6037 * ple_window_max. (See __grow_ple_window for the reason.)
6038 * This prevents overflows, because ple_window_max is int.
6039 * ple_window_max effectively rounded down to a multiple of ple_window_grow in
6041 * ple_window_max is also prevented from setting vmx->ple_window < ple_window.
6043 static void update_ple_window_actual_max(void)
6045 ple_window_actual_max =
6046 __shrink_ple_window(max(ple_window_max, ple_window),
6047 ple_window_grow, INT_MIN);
6051 * Handler for POSTED_INTERRUPT_WAKEUP_VECTOR.
6053 static void wakeup_handler(void)
6055 struct kvm_vcpu *vcpu;
6056 int cpu = smp_processor_id();
6058 spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
6059 list_for_each_entry(vcpu, &per_cpu(blocked_vcpu_on_cpu, cpu),
6060 blocked_vcpu_list) {
6061 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
6063 if (pi_test_on(pi_desc) == 1)
6064 kvm_vcpu_kick(vcpu);
6066 spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
6069 static __init int hardware_setup(void)
6071 int r = -ENOMEM, i, msr;
6073 rdmsrl_safe(MSR_EFER, &host_efer);
6075 for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i)
6076 kvm_define_shared_msr(i, vmx_msr_index[i]);
6078 vmx_io_bitmap_a = (unsigned long *)__get_free_page(GFP_KERNEL);
6079 if (!vmx_io_bitmap_a)
6082 vmx_io_bitmap_b = (unsigned long *)__get_free_page(GFP_KERNEL);
6083 if (!vmx_io_bitmap_b)
6086 vmx_msr_bitmap_legacy = (unsigned long *)__get_free_page(GFP_KERNEL);
6087 if (!vmx_msr_bitmap_legacy)
6090 vmx_msr_bitmap_legacy_x2apic =
6091 (unsigned long *)__get_free_page(GFP_KERNEL);
6092 if (!vmx_msr_bitmap_legacy_x2apic)
6095 vmx_msr_bitmap_longmode = (unsigned long *)__get_free_page(GFP_KERNEL);
6096 if (!vmx_msr_bitmap_longmode)
6099 vmx_msr_bitmap_longmode_x2apic =
6100 (unsigned long *)__get_free_page(GFP_KERNEL);
6101 if (!vmx_msr_bitmap_longmode_x2apic)
6105 vmx_msr_bitmap_nested =
6106 (unsigned long *)__get_free_page(GFP_KERNEL);
6107 if (!vmx_msr_bitmap_nested)
6111 vmx_vmread_bitmap = (unsigned long *)__get_free_page(GFP_KERNEL);
6112 if (!vmx_vmread_bitmap)
6115 vmx_vmwrite_bitmap = (unsigned long *)__get_free_page(GFP_KERNEL);
6116 if (!vmx_vmwrite_bitmap)
6119 memset(vmx_vmread_bitmap, 0xff, PAGE_SIZE);
6120 memset(vmx_vmwrite_bitmap, 0xff, PAGE_SIZE);
6123 * Allow direct access to the PC debug port (it is often used for I/O
6124 * delays, but the vmexits simply slow things down).
6126 memset(vmx_io_bitmap_a, 0xff, PAGE_SIZE);
6127 clear_bit(0x80, vmx_io_bitmap_a);
6129 memset(vmx_io_bitmap_b, 0xff, PAGE_SIZE);
6131 memset(vmx_msr_bitmap_legacy, 0xff, PAGE_SIZE);
6132 memset(vmx_msr_bitmap_longmode, 0xff, PAGE_SIZE);
6134 memset(vmx_msr_bitmap_nested, 0xff, PAGE_SIZE);
6136 if (setup_vmcs_config(&vmcs_config) < 0) {
6141 if (boot_cpu_has(X86_FEATURE_NX))
6142 kvm_enable_efer_bits(EFER_NX);
6144 if (!cpu_has_vmx_vpid())
6146 if (!cpu_has_vmx_shadow_vmcs())
6147 enable_shadow_vmcs = 0;
6148 if (enable_shadow_vmcs)
6149 init_vmcs_shadow_fields();
6151 if (!cpu_has_vmx_ept() ||
6152 !cpu_has_vmx_ept_4levels()) {
6154 enable_unrestricted_guest = 0;
6155 enable_ept_ad_bits = 0;
6158 if (!cpu_has_vmx_ept_ad_bits())
6159 enable_ept_ad_bits = 0;
6161 if (!cpu_has_vmx_unrestricted_guest())
6162 enable_unrestricted_guest = 0;
6164 if (!cpu_has_vmx_flexpriority())
6165 flexpriority_enabled = 0;
6168 * set_apic_access_page_addr() is used to reload apic access
6169 * page upon invalidation. No need to do anything if not
6170 * using the APIC_ACCESS_ADDR VMCS field.
6172 if (!flexpriority_enabled)
6173 kvm_x86_ops->set_apic_access_page_addr = NULL;
6175 if (!cpu_has_vmx_tpr_shadow())
6176 kvm_x86_ops->update_cr8_intercept = NULL;
6178 if (enable_ept && !cpu_has_vmx_ept_2m_page())
6179 kvm_disable_largepages();
6181 if (!cpu_has_vmx_ple())
6184 if (!cpu_has_vmx_apicv())
6188 kvm_x86_ops->update_cr8_intercept = NULL;
6190 kvm_x86_ops->hwapic_irr_update = NULL;
6191 kvm_x86_ops->hwapic_isr_update = NULL;
6192 kvm_x86_ops->deliver_posted_interrupt = NULL;
6193 kvm_x86_ops->sync_pir_to_irr = vmx_sync_pir_to_irr_dummy;
6196 vmx_disable_intercept_for_msr(MSR_FS_BASE, false);
6197 vmx_disable_intercept_for_msr(MSR_GS_BASE, false);
6198 vmx_disable_intercept_for_msr(MSR_KERNEL_GS_BASE, true);
6199 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_CS, false);
6200 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_ESP, false);
6201 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_EIP, false);
6202 vmx_disable_intercept_for_msr(MSR_IA32_BNDCFGS, true);
6204 memcpy(vmx_msr_bitmap_legacy_x2apic,
6205 vmx_msr_bitmap_legacy, PAGE_SIZE);
6206 memcpy(vmx_msr_bitmap_longmode_x2apic,
6207 vmx_msr_bitmap_longmode, PAGE_SIZE);
6209 set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
6212 for (msr = 0x800; msr <= 0x8ff; msr++)
6213 vmx_disable_intercept_msr_read_x2apic(msr);
6215 /* According SDM, in x2apic mode, the whole id reg is used.
6216 * But in KVM, it only use the highest eight bits. Need to
6218 vmx_enable_intercept_msr_read_x2apic(0x802);
6220 vmx_enable_intercept_msr_read_x2apic(0x839);
6222 vmx_disable_intercept_msr_write_x2apic(0x808);
6224 vmx_disable_intercept_msr_write_x2apic(0x80b);
6226 vmx_disable_intercept_msr_write_x2apic(0x83f);
6230 kvm_mmu_set_mask_ptes(0ull,
6231 (enable_ept_ad_bits) ? VMX_EPT_ACCESS_BIT : 0ull,
6232 (enable_ept_ad_bits) ? VMX_EPT_DIRTY_BIT : 0ull,
6233 0ull, VMX_EPT_EXECUTABLE_MASK);
6234 ept_set_mmio_spte_mask();
6239 update_ple_window_actual_max();
6242 * Only enable PML when hardware supports PML feature, and both EPT
6243 * and EPT A/D bit features are enabled -- PML depends on them to work.
6245 if (!enable_ept || !enable_ept_ad_bits || !cpu_has_vmx_pml())
6249 kvm_x86_ops->slot_enable_log_dirty = NULL;
6250 kvm_x86_ops->slot_disable_log_dirty = NULL;
6251 kvm_x86_ops->flush_log_dirty = NULL;
6252 kvm_x86_ops->enable_log_dirty_pt_masked = NULL;
6255 kvm_set_posted_intr_wakeup_handler(wakeup_handler);
6257 return alloc_kvm_area();
6260 free_page((unsigned long)vmx_vmwrite_bitmap);
6262 free_page((unsigned long)vmx_vmread_bitmap);
6265 free_page((unsigned long)vmx_msr_bitmap_nested);
6267 free_page((unsigned long)vmx_msr_bitmap_longmode_x2apic);
6269 free_page((unsigned long)vmx_msr_bitmap_longmode);
6271 free_page((unsigned long)vmx_msr_bitmap_legacy_x2apic);
6273 free_page((unsigned long)vmx_msr_bitmap_legacy);
6275 free_page((unsigned long)vmx_io_bitmap_b);
6277 free_page((unsigned long)vmx_io_bitmap_a);
6282 static __exit void hardware_unsetup(void)
6284 free_page((unsigned long)vmx_msr_bitmap_legacy_x2apic);
6285 free_page((unsigned long)vmx_msr_bitmap_longmode_x2apic);
6286 free_page((unsigned long)vmx_msr_bitmap_legacy);
6287 free_page((unsigned long)vmx_msr_bitmap_longmode);
6288 free_page((unsigned long)vmx_io_bitmap_b);
6289 free_page((unsigned long)vmx_io_bitmap_a);
6290 free_page((unsigned long)vmx_vmwrite_bitmap);
6291 free_page((unsigned long)vmx_vmread_bitmap);
6293 free_page((unsigned long)vmx_msr_bitmap_nested);
6299 * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
6300 * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
6302 static int handle_pause(struct kvm_vcpu *vcpu)
6305 grow_ple_window(vcpu);
6307 skip_emulated_instruction(vcpu);
6308 kvm_vcpu_on_spin(vcpu);
6313 static int handle_nop(struct kvm_vcpu *vcpu)
6315 skip_emulated_instruction(vcpu);
6319 static int handle_mwait(struct kvm_vcpu *vcpu)
6321 printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
6322 return handle_nop(vcpu);
6325 static int handle_monitor_trap(struct kvm_vcpu *vcpu)
6330 static int handle_monitor(struct kvm_vcpu *vcpu)
6332 printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
6333 return handle_nop(vcpu);
6337 * To run an L2 guest, we need a vmcs02 based on the L1-specified vmcs12.
6338 * We could reuse a single VMCS for all the L2 guests, but we also want the
6339 * option to allocate a separate vmcs02 for each separate loaded vmcs12 - this
6340 * allows keeping them loaded on the processor, and in the future will allow
6341 * optimizations where prepare_vmcs02 doesn't need to set all the fields on
6342 * every entry if they never change.
6343 * So we keep, in vmx->nested.vmcs02_pool, a cache of size VMCS02_POOL_SIZE
6344 * (>=0) with a vmcs02 for each recently loaded vmcs12s, most recent first.
6346 * The following functions allocate and free a vmcs02 in this pool.
6349 /* Get a VMCS from the pool to use as vmcs02 for the current vmcs12. */
6350 static struct loaded_vmcs *nested_get_current_vmcs02(struct vcpu_vmx *vmx)
6352 struct vmcs02_list *item;
6353 list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
6354 if (item->vmptr == vmx->nested.current_vmptr) {
6355 list_move(&item->list, &vmx->nested.vmcs02_pool);
6356 return &item->vmcs02;
6359 if (vmx->nested.vmcs02_num >= max(VMCS02_POOL_SIZE, 1)) {
6360 /* Recycle the least recently used VMCS. */
6361 item = list_entry(vmx->nested.vmcs02_pool.prev,
6362 struct vmcs02_list, list);
6363 item->vmptr = vmx->nested.current_vmptr;
6364 list_move(&item->list, &vmx->nested.vmcs02_pool);
6365 return &item->vmcs02;
6368 /* Create a new VMCS */
6369 item = kmalloc(sizeof(struct vmcs02_list), GFP_KERNEL);
6372 item->vmcs02.vmcs = alloc_vmcs();
6373 if (!item->vmcs02.vmcs) {
6377 loaded_vmcs_init(&item->vmcs02);
6378 item->vmptr = vmx->nested.current_vmptr;
6379 list_add(&(item->list), &(vmx->nested.vmcs02_pool));
6380 vmx->nested.vmcs02_num++;
6381 return &item->vmcs02;
6384 /* Free and remove from pool a vmcs02 saved for a vmcs12 (if there is one) */
6385 static void nested_free_vmcs02(struct vcpu_vmx *vmx, gpa_t vmptr)
6387 struct vmcs02_list *item;
6388 list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
6389 if (item->vmptr == vmptr) {
6390 free_loaded_vmcs(&item->vmcs02);
6391 list_del(&item->list);
6393 vmx->nested.vmcs02_num--;
6399 * Free all VMCSs saved for this vcpu, except the one pointed by
6400 * vmx->loaded_vmcs. We must be running L1, so vmx->loaded_vmcs
6401 * must be &vmx->vmcs01.
6403 static void nested_free_all_saved_vmcss(struct vcpu_vmx *vmx)
6405 struct vmcs02_list *item, *n;
6407 WARN_ON(vmx->loaded_vmcs != &vmx->vmcs01);
6408 list_for_each_entry_safe(item, n, &vmx->nested.vmcs02_pool, list) {
6410 * Something will leak if the above WARN triggers. Better than
6413 if (vmx->loaded_vmcs == &item->vmcs02)
6416 free_loaded_vmcs(&item->vmcs02);
6417 list_del(&item->list);
6419 vmx->nested.vmcs02_num--;
6424 * The following 3 functions, nested_vmx_succeed()/failValid()/failInvalid(),
6425 * set the success or error code of an emulated VMX instruction, as specified
6426 * by Vol 2B, VMX Instruction Reference, "Conventions".
6428 static void nested_vmx_succeed(struct kvm_vcpu *vcpu)
6430 vmx_set_rflags(vcpu, vmx_get_rflags(vcpu)
6431 & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
6432 X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF));
6435 static void nested_vmx_failInvalid(struct kvm_vcpu *vcpu)
6437 vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
6438 & ~(X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
6439 X86_EFLAGS_SF | X86_EFLAGS_OF))
6443 static void nested_vmx_failValid(struct kvm_vcpu *vcpu,
6444 u32 vm_instruction_error)
6446 if (to_vmx(vcpu)->nested.current_vmptr == -1ull) {
6448 * failValid writes the error number to the current VMCS, which
6449 * can't be done there isn't a current VMCS.
6451 nested_vmx_failInvalid(vcpu);
6454 vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
6455 & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
6456 X86_EFLAGS_SF | X86_EFLAGS_OF))
6458 get_vmcs12(vcpu)->vm_instruction_error = vm_instruction_error;
6460 * We don't need to force a shadow sync because
6461 * VM_INSTRUCTION_ERROR is not shadowed
6465 static void nested_vmx_abort(struct kvm_vcpu *vcpu, u32 indicator)
6467 /* TODO: not to reset guest simply here. */
6468 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
6469 pr_warn("kvm: nested vmx abort, indicator %d\n", indicator);
6472 static enum hrtimer_restart vmx_preemption_timer_fn(struct hrtimer *timer)
6474 struct vcpu_vmx *vmx =
6475 container_of(timer, struct vcpu_vmx, nested.preemption_timer);
6477 vmx->nested.preemption_timer_expired = true;
6478 kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
6479 kvm_vcpu_kick(&vmx->vcpu);
6481 return HRTIMER_NORESTART;
6485 * Decode the memory-address operand of a vmx instruction, as recorded on an
6486 * exit caused by such an instruction (run by a guest hypervisor).
6487 * On success, returns 0. When the operand is invalid, returns 1 and throws
6490 static int get_vmx_mem_address(struct kvm_vcpu *vcpu,
6491 unsigned long exit_qualification,
6492 u32 vmx_instruction_info, bool wr, gva_t *ret)
6496 struct kvm_segment s;
6499 * According to Vol. 3B, "Information for VM Exits Due to Instruction
6500 * Execution", on an exit, vmx_instruction_info holds most of the
6501 * addressing components of the operand. Only the displacement part
6502 * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
6503 * For how an actual address is calculated from all these components,
6504 * refer to Vol. 1, "Operand Addressing".
6506 int scaling = vmx_instruction_info & 3;
6507 int addr_size = (vmx_instruction_info >> 7) & 7;
6508 bool is_reg = vmx_instruction_info & (1u << 10);
6509 int seg_reg = (vmx_instruction_info >> 15) & 7;
6510 int index_reg = (vmx_instruction_info >> 18) & 0xf;
6511 bool index_is_valid = !(vmx_instruction_info & (1u << 22));
6512 int base_reg = (vmx_instruction_info >> 23) & 0xf;
6513 bool base_is_valid = !(vmx_instruction_info & (1u << 27));
6516 kvm_queue_exception(vcpu, UD_VECTOR);
6520 /* Addr = segment_base + offset */
6521 /* offset = base + [index * scale] + displacement */
6522 off = exit_qualification; /* holds the displacement */
6524 off += kvm_register_read(vcpu, base_reg);
6526 off += kvm_register_read(vcpu, index_reg)<<scaling;
6527 vmx_get_segment(vcpu, &s, seg_reg);
6528 *ret = s.base + off;
6530 if (addr_size == 1) /* 32 bit */
6533 /* Checks for #GP/#SS exceptions. */
6535 if (is_protmode(vcpu)) {
6536 /* Protected mode: apply checks for segment validity in the
6538 * - segment type check (#GP(0) may be thrown)
6539 * - usability check (#GP(0)/#SS(0))
6540 * - limit check (#GP(0)/#SS(0))
6543 /* #GP(0) if the destination operand is located in a
6544 * read-only data segment or any code segment.
6546 exn = ((s.type & 0xa) == 0 || (s.type & 8));
6548 /* #GP(0) if the source operand is located in an
6549 * execute-only code segment
6551 exn = ((s.type & 0xa) == 8);
6554 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
6557 if (is_long_mode(vcpu)) {
6558 /* Long mode: #GP(0)/#SS(0) if the memory address is in a
6559 * non-canonical form. This is an only check for long mode.
6561 exn = is_noncanonical_address(*ret);
6562 } else if (is_protmode(vcpu)) {
6563 /* Protected mode: #GP(0)/#SS(0) if the segment is unusable.
6565 exn = (s.unusable != 0);
6566 /* Protected mode: #GP(0)/#SS(0) if the memory
6567 * operand is outside the segment limit.
6569 exn = exn || (off + sizeof(u64) > s.limit);
6572 kvm_queue_exception_e(vcpu,
6573 seg_reg == VCPU_SREG_SS ?
6574 SS_VECTOR : GP_VECTOR,
6583 * This function performs the various checks including
6584 * - if it's 4KB aligned
6585 * - No bits beyond the physical address width are set
6586 * - Returns 0 on success or else 1
6587 * (Intel SDM Section 30.3)
6589 static int nested_vmx_check_vmptr(struct kvm_vcpu *vcpu, int exit_reason,
6594 struct x86_exception e;
6596 struct vcpu_vmx *vmx = to_vmx(vcpu);
6597 int maxphyaddr = cpuid_maxphyaddr(vcpu);
6599 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
6600 vmcs_read32(VMX_INSTRUCTION_INFO), false, &gva))
6603 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &vmptr,
6604 sizeof(vmptr), &e)) {
6605 kvm_inject_page_fault(vcpu, &e);
6609 switch (exit_reason) {
6610 case EXIT_REASON_VMON:
6613 * The first 4 bytes of VMXON region contain the supported
6614 * VMCS revision identifier
6616 * Note - IA32_VMX_BASIC[48] will never be 1
6617 * for the nested case;
6618 * which replaces physical address width with 32
6621 if (!PAGE_ALIGNED(vmptr) || (vmptr >> maxphyaddr)) {
6622 nested_vmx_failInvalid(vcpu);
6623 skip_emulated_instruction(vcpu);
6627 page = nested_get_page(vcpu, vmptr);
6629 *(u32 *)kmap(page) != VMCS12_REVISION) {
6630 nested_vmx_failInvalid(vcpu);
6632 skip_emulated_instruction(vcpu);
6636 vmx->nested.vmxon_ptr = vmptr;
6638 case EXIT_REASON_VMCLEAR:
6639 if (!PAGE_ALIGNED(vmptr) || (vmptr >> maxphyaddr)) {
6640 nested_vmx_failValid(vcpu,
6641 VMXERR_VMCLEAR_INVALID_ADDRESS);
6642 skip_emulated_instruction(vcpu);
6646 if (vmptr == vmx->nested.vmxon_ptr) {
6647 nested_vmx_failValid(vcpu,
6648 VMXERR_VMCLEAR_VMXON_POINTER);
6649 skip_emulated_instruction(vcpu);
6653 case EXIT_REASON_VMPTRLD:
6654 if (!PAGE_ALIGNED(vmptr) || (vmptr >> maxphyaddr)) {
6655 nested_vmx_failValid(vcpu,
6656 VMXERR_VMPTRLD_INVALID_ADDRESS);
6657 skip_emulated_instruction(vcpu);
6661 if (vmptr == vmx->nested.vmxon_ptr) {
6662 nested_vmx_failValid(vcpu,
6663 VMXERR_VMCLEAR_VMXON_POINTER);
6664 skip_emulated_instruction(vcpu);
6669 return 1; /* shouldn't happen */
6678 * Emulate the VMXON instruction.
6679 * Currently, we just remember that VMX is active, and do not save or even
6680 * inspect the argument to VMXON (the so-called "VMXON pointer") because we
6681 * do not currently need to store anything in that guest-allocated memory
6682 * region. Consequently, VMCLEAR and VMPTRLD also do not verify that the their
6683 * argument is different from the VMXON pointer (which the spec says they do).
6685 static int handle_vmon(struct kvm_vcpu *vcpu)
6687 struct kvm_segment cs;
6688 struct vcpu_vmx *vmx = to_vmx(vcpu);
6689 struct vmcs *shadow_vmcs;
6690 const u64 VMXON_NEEDED_FEATURES = FEATURE_CONTROL_LOCKED
6691 | FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
6693 /* The Intel VMX Instruction Reference lists a bunch of bits that
6694 * are prerequisite to running VMXON, most notably cr4.VMXE must be
6695 * set to 1 (see vmx_set_cr4() for when we allow the guest to set this).
6696 * Otherwise, we should fail with #UD. We test these now:
6698 if (!kvm_read_cr4_bits(vcpu, X86_CR4_VMXE) ||
6699 !kvm_read_cr0_bits(vcpu, X86_CR0_PE) ||
6700 (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
6701 kvm_queue_exception(vcpu, UD_VECTOR);
6705 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
6706 if (is_long_mode(vcpu) && !cs.l) {
6707 kvm_queue_exception(vcpu, UD_VECTOR);
6711 if (vmx_get_cpl(vcpu)) {
6712 kvm_inject_gp(vcpu, 0);
6716 if (nested_vmx_check_vmptr(vcpu, EXIT_REASON_VMON, NULL))
6719 if (vmx->nested.vmxon) {
6720 nested_vmx_failValid(vcpu, VMXERR_VMXON_IN_VMX_ROOT_OPERATION);
6721 skip_emulated_instruction(vcpu);
6725 if ((vmx->nested.msr_ia32_feature_control & VMXON_NEEDED_FEATURES)
6726 != VMXON_NEEDED_FEATURES) {
6727 kvm_inject_gp(vcpu, 0);
6731 if (enable_shadow_vmcs) {
6732 shadow_vmcs = alloc_vmcs();
6735 /* mark vmcs as shadow */
6736 shadow_vmcs->revision_id |= (1u << 31);
6737 /* init shadow vmcs */
6738 vmcs_clear(shadow_vmcs);
6739 vmx->nested.current_shadow_vmcs = shadow_vmcs;
6742 INIT_LIST_HEAD(&(vmx->nested.vmcs02_pool));
6743 vmx->nested.vmcs02_num = 0;
6745 hrtimer_init(&vmx->nested.preemption_timer, CLOCK_MONOTONIC,
6747 vmx->nested.preemption_timer.function = vmx_preemption_timer_fn;
6749 vmx->nested.vmxon = true;
6751 skip_emulated_instruction(vcpu);
6752 nested_vmx_succeed(vcpu);
6757 * Intel's VMX Instruction Reference specifies a common set of prerequisites
6758 * for running VMX instructions (except VMXON, whose prerequisites are
6759 * slightly different). It also specifies what exception to inject otherwise.
6761 static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
6763 struct kvm_segment cs;
6764 struct vcpu_vmx *vmx = to_vmx(vcpu);
6766 if (!vmx->nested.vmxon) {
6767 kvm_queue_exception(vcpu, UD_VECTOR);
6771 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
6772 if ((vmx_get_rflags(vcpu) & X86_EFLAGS_VM) ||
6773 (is_long_mode(vcpu) && !cs.l)) {
6774 kvm_queue_exception(vcpu, UD_VECTOR);
6778 if (vmx_get_cpl(vcpu)) {
6779 kvm_inject_gp(vcpu, 0);
6786 static inline void nested_release_vmcs12(struct vcpu_vmx *vmx)
6788 if (vmx->nested.current_vmptr == -1ull)
6791 /* current_vmptr and current_vmcs12 are always set/reset together */
6792 if (WARN_ON(vmx->nested.current_vmcs12 == NULL))
6795 if (enable_shadow_vmcs) {
6796 /* copy to memory all shadowed fields in case
6797 they were modified */
6798 copy_shadow_to_vmcs12(vmx);
6799 vmx->nested.sync_shadow_vmcs = false;
6800 vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL,
6801 SECONDARY_EXEC_SHADOW_VMCS);
6802 vmcs_write64(VMCS_LINK_POINTER, -1ull);
6804 vmx->nested.posted_intr_nv = -1;
6805 kunmap(vmx->nested.current_vmcs12_page);
6806 nested_release_page(vmx->nested.current_vmcs12_page);
6807 vmx->nested.current_vmptr = -1ull;
6808 vmx->nested.current_vmcs12 = NULL;
6812 * Free whatever needs to be freed from vmx->nested when L1 goes down, or
6813 * just stops using VMX.
6815 static void free_nested(struct vcpu_vmx *vmx)
6817 if (!vmx->nested.vmxon)
6820 vmx->nested.vmxon = false;
6821 nested_release_vmcs12(vmx);
6822 if (enable_shadow_vmcs)
6823 free_vmcs(vmx->nested.current_shadow_vmcs);
6824 /* Unpin physical memory we referred to in current vmcs02 */
6825 if (vmx->nested.apic_access_page) {
6826 nested_release_page(vmx->nested.apic_access_page);
6827 vmx->nested.apic_access_page = NULL;
6829 if (vmx->nested.virtual_apic_page) {
6830 nested_release_page(vmx->nested.virtual_apic_page);
6831 vmx->nested.virtual_apic_page = NULL;
6833 if (vmx->nested.pi_desc_page) {
6834 kunmap(vmx->nested.pi_desc_page);
6835 nested_release_page(vmx->nested.pi_desc_page);
6836 vmx->nested.pi_desc_page = NULL;
6837 vmx->nested.pi_desc = NULL;
6840 nested_free_all_saved_vmcss(vmx);
6843 /* Emulate the VMXOFF instruction */
6844 static int handle_vmoff(struct kvm_vcpu *vcpu)
6846 if (!nested_vmx_check_permission(vcpu))
6848 free_nested(to_vmx(vcpu));
6849 skip_emulated_instruction(vcpu);
6850 nested_vmx_succeed(vcpu);
6854 /* Emulate the VMCLEAR instruction */
6855 static int handle_vmclear(struct kvm_vcpu *vcpu)
6857 struct vcpu_vmx *vmx = to_vmx(vcpu);
6859 struct vmcs12 *vmcs12;
6862 if (!nested_vmx_check_permission(vcpu))
6865 if (nested_vmx_check_vmptr(vcpu, EXIT_REASON_VMCLEAR, &vmptr))
6868 if (vmptr == vmx->nested.current_vmptr)
6869 nested_release_vmcs12(vmx);
6871 page = nested_get_page(vcpu, vmptr);
6874 * For accurate processor emulation, VMCLEAR beyond available
6875 * physical memory should do nothing at all. However, it is
6876 * possible that a nested vmx bug, not a guest hypervisor bug,
6877 * resulted in this case, so let's shut down before doing any
6880 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
6883 vmcs12 = kmap(page);
6884 vmcs12->launch_state = 0;
6886 nested_release_page(page);
6888 nested_free_vmcs02(vmx, vmptr);
6890 skip_emulated_instruction(vcpu);
6891 nested_vmx_succeed(vcpu);
6895 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch);
6897 /* Emulate the VMLAUNCH instruction */
6898 static int handle_vmlaunch(struct kvm_vcpu *vcpu)
6900 return nested_vmx_run(vcpu, true);
6903 /* Emulate the VMRESUME instruction */
6904 static int handle_vmresume(struct kvm_vcpu *vcpu)
6907 return nested_vmx_run(vcpu, false);
6910 enum vmcs_field_type {
6911 VMCS_FIELD_TYPE_U16 = 0,
6912 VMCS_FIELD_TYPE_U64 = 1,
6913 VMCS_FIELD_TYPE_U32 = 2,
6914 VMCS_FIELD_TYPE_NATURAL_WIDTH = 3
6917 static inline int vmcs_field_type(unsigned long field)
6919 if (0x1 & field) /* the *_HIGH fields are all 32 bit */
6920 return VMCS_FIELD_TYPE_U32;
6921 return (field >> 13) & 0x3 ;
6924 static inline int vmcs_field_readonly(unsigned long field)
6926 return (((field >> 10) & 0x3) == 1);
6930 * Read a vmcs12 field. Since these can have varying lengths and we return
6931 * one type, we chose the biggest type (u64) and zero-extend the return value
6932 * to that size. Note that the caller, handle_vmread, might need to use only
6933 * some of the bits we return here (e.g., on 32-bit guests, only 32 bits of
6934 * 64-bit fields are to be returned).
6936 static inline int vmcs12_read_any(struct kvm_vcpu *vcpu,
6937 unsigned long field, u64 *ret)
6939 short offset = vmcs_field_to_offset(field);
6945 p = ((char *)(get_vmcs12(vcpu))) + offset;
6947 switch (vmcs_field_type(field)) {
6948 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
6949 *ret = *((natural_width *)p);
6951 case VMCS_FIELD_TYPE_U16:
6954 case VMCS_FIELD_TYPE_U32:
6957 case VMCS_FIELD_TYPE_U64:
6967 static inline int vmcs12_write_any(struct kvm_vcpu *vcpu,
6968 unsigned long field, u64 field_value){
6969 short offset = vmcs_field_to_offset(field);
6970 char *p = ((char *) get_vmcs12(vcpu)) + offset;
6974 switch (vmcs_field_type(field)) {
6975 case VMCS_FIELD_TYPE_U16:
6976 *(u16 *)p = field_value;
6978 case VMCS_FIELD_TYPE_U32:
6979 *(u32 *)p = field_value;
6981 case VMCS_FIELD_TYPE_U64:
6982 *(u64 *)p = field_value;
6984 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
6985 *(natural_width *)p = field_value;
6994 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx)
6997 unsigned long field;
6999 struct vmcs *shadow_vmcs = vmx->nested.current_shadow_vmcs;
7000 const unsigned long *fields = shadow_read_write_fields;
7001 const int num_fields = max_shadow_read_write_fields;
7005 vmcs_load(shadow_vmcs);
7007 for (i = 0; i < num_fields; i++) {
7009 switch (vmcs_field_type(field)) {
7010 case VMCS_FIELD_TYPE_U16:
7011 field_value = vmcs_read16(field);
7013 case VMCS_FIELD_TYPE_U32:
7014 field_value = vmcs_read32(field);
7016 case VMCS_FIELD_TYPE_U64:
7017 field_value = vmcs_read64(field);
7019 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7020 field_value = vmcs_readl(field);
7026 vmcs12_write_any(&vmx->vcpu, field, field_value);
7029 vmcs_clear(shadow_vmcs);
7030 vmcs_load(vmx->loaded_vmcs->vmcs);
7035 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx)
7037 const unsigned long *fields[] = {
7038 shadow_read_write_fields,
7039 shadow_read_only_fields
7041 const int max_fields[] = {
7042 max_shadow_read_write_fields,
7043 max_shadow_read_only_fields
7046 unsigned long field;
7047 u64 field_value = 0;
7048 struct vmcs *shadow_vmcs = vmx->nested.current_shadow_vmcs;
7050 vmcs_load(shadow_vmcs);
7052 for (q = 0; q < ARRAY_SIZE(fields); q++) {
7053 for (i = 0; i < max_fields[q]; i++) {
7054 field = fields[q][i];
7055 vmcs12_read_any(&vmx->vcpu, field, &field_value);
7057 switch (vmcs_field_type(field)) {
7058 case VMCS_FIELD_TYPE_U16:
7059 vmcs_write16(field, (u16)field_value);
7061 case VMCS_FIELD_TYPE_U32:
7062 vmcs_write32(field, (u32)field_value);
7064 case VMCS_FIELD_TYPE_U64:
7065 vmcs_write64(field, (u64)field_value);
7067 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7068 vmcs_writel(field, (long)field_value);
7077 vmcs_clear(shadow_vmcs);
7078 vmcs_load(vmx->loaded_vmcs->vmcs);
7082 * VMX instructions which assume a current vmcs12 (i.e., that VMPTRLD was
7083 * used before) all generate the same failure when it is missing.
7085 static int nested_vmx_check_vmcs12(struct kvm_vcpu *vcpu)
7087 struct vcpu_vmx *vmx = to_vmx(vcpu);
7088 if (vmx->nested.current_vmptr == -1ull) {
7089 nested_vmx_failInvalid(vcpu);
7090 skip_emulated_instruction(vcpu);
7096 static int handle_vmread(struct kvm_vcpu *vcpu)
7098 unsigned long field;
7100 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7101 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7104 if (!nested_vmx_check_permission(vcpu) ||
7105 !nested_vmx_check_vmcs12(vcpu))
7108 /* Decode instruction info and find the field to read */
7109 field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
7110 /* Read the field, zero-extended to a u64 field_value */
7111 if (vmcs12_read_any(vcpu, field, &field_value) < 0) {
7112 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
7113 skip_emulated_instruction(vcpu);
7117 * Now copy part of this value to register or memory, as requested.
7118 * Note that the number of bits actually copied is 32 or 64 depending
7119 * on the guest's mode (32 or 64 bit), not on the given field's length.
7121 if (vmx_instruction_info & (1u << 10)) {
7122 kvm_register_writel(vcpu, (((vmx_instruction_info) >> 3) & 0xf),
7125 if (get_vmx_mem_address(vcpu, exit_qualification,
7126 vmx_instruction_info, true, &gva))
7128 /* _system ok, as nested_vmx_check_permission verified cpl=0 */
7129 kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, gva,
7130 &field_value, (is_long_mode(vcpu) ? 8 : 4), NULL);
7133 nested_vmx_succeed(vcpu);
7134 skip_emulated_instruction(vcpu);
7139 static int handle_vmwrite(struct kvm_vcpu *vcpu)
7141 unsigned long field;
7143 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7144 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7145 /* The value to write might be 32 or 64 bits, depending on L1's long
7146 * mode, and eventually we need to write that into a field of several
7147 * possible lengths. The code below first zero-extends the value to 64
7148 * bit (field_value), and then copies only the approriate number of
7149 * bits into the vmcs12 field.
7151 u64 field_value = 0;
7152 struct x86_exception e;
7154 if (!nested_vmx_check_permission(vcpu) ||
7155 !nested_vmx_check_vmcs12(vcpu))
7158 if (vmx_instruction_info & (1u << 10))
7159 field_value = kvm_register_readl(vcpu,
7160 (((vmx_instruction_info) >> 3) & 0xf));
7162 if (get_vmx_mem_address(vcpu, exit_qualification,
7163 vmx_instruction_info, false, &gva))
7165 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva,
7166 &field_value, (is_64_bit_mode(vcpu) ? 8 : 4), &e)) {
7167 kvm_inject_page_fault(vcpu, &e);
7173 field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
7174 if (vmcs_field_readonly(field)) {
7175 nested_vmx_failValid(vcpu,
7176 VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
7177 skip_emulated_instruction(vcpu);
7181 if (vmcs12_write_any(vcpu, field, field_value) < 0) {
7182 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
7183 skip_emulated_instruction(vcpu);
7187 nested_vmx_succeed(vcpu);
7188 skip_emulated_instruction(vcpu);
7192 /* Emulate the VMPTRLD instruction */
7193 static int handle_vmptrld(struct kvm_vcpu *vcpu)
7195 struct vcpu_vmx *vmx = to_vmx(vcpu);
7198 if (!nested_vmx_check_permission(vcpu))
7201 if (nested_vmx_check_vmptr(vcpu, EXIT_REASON_VMPTRLD, &vmptr))
7204 if (vmx->nested.current_vmptr != vmptr) {
7205 struct vmcs12 *new_vmcs12;
7207 page = nested_get_page(vcpu, vmptr);
7209 nested_vmx_failInvalid(vcpu);
7210 skip_emulated_instruction(vcpu);
7213 new_vmcs12 = kmap(page);
7214 if (new_vmcs12->revision_id != VMCS12_REVISION) {
7216 nested_release_page_clean(page);
7217 nested_vmx_failValid(vcpu,
7218 VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
7219 skip_emulated_instruction(vcpu);
7223 nested_release_vmcs12(vmx);
7224 vmx->nested.current_vmptr = vmptr;
7225 vmx->nested.current_vmcs12 = new_vmcs12;
7226 vmx->nested.current_vmcs12_page = page;
7227 if (enable_shadow_vmcs) {
7228 vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
7229 SECONDARY_EXEC_SHADOW_VMCS);
7230 vmcs_write64(VMCS_LINK_POINTER,
7231 __pa(vmx->nested.current_shadow_vmcs));
7232 vmx->nested.sync_shadow_vmcs = true;
7236 nested_vmx_succeed(vcpu);
7237 skip_emulated_instruction(vcpu);
7241 /* Emulate the VMPTRST instruction */
7242 static int handle_vmptrst(struct kvm_vcpu *vcpu)
7244 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7245 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7247 struct x86_exception e;
7249 if (!nested_vmx_check_permission(vcpu))
7252 if (get_vmx_mem_address(vcpu, exit_qualification,
7253 vmx_instruction_info, true, &vmcs_gva))
7255 /* ok to use *_system, as nested_vmx_check_permission verified cpl=0 */
7256 if (kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, vmcs_gva,
7257 (void *)&to_vmx(vcpu)->nested.current_vmptr,
7259 kvm_inject_page_fault(vcpu, &e);
7262 nested_vmx_succeed(vcpu);
7263 skip_emulated_instruction(vcpu);
7267 /* Emulate the INVEPT instruction */
7268 static int handle_invept(struct kvm_vcpu *vcpu)
7270 struct vcpu_vmx *vmx = to_vmx(vcpu);
7271 u32 vmx_instruction_info, types;
7274 struct x86_exception e;
7279 if (!(vmx->nested.nested_vmx_secondary_ctls_high &
7280 SECONDARY_EXEC_ENABLE_EPT) ||
7281 !(vmx->nested.nested_vmx_ept_caps & VMX_EPT_INVEPT_BIT)) {
7282 kvm_queue_exception(vcpu, UD_VECTOR);
7286 if (!nested_vmx_check_permission(vcpu))
7289 if (!kvm_read_cr0_bits(vcpu, X86_CR0_PE)) {
7290 kvm_queue_exception(vcpu, UD_VECTOR);
7294 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7295 type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
7297 types = (vmx->nested.nested_vmx_ept_caps >> VMX_EPT_EXTENT_SHIFT) & 6;
7299 if (!(types & (1UL << type))) {
7300 nested_vmx_failValid(vcpu,
7301 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
7305 /* According to the Intel VMX instruction reference, the memory
7306 * operand is read even if it isn't needed (e.g., for type==global)
7308 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
7309 vmx_instruction_info, false, &gva))
7311 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &operand,
7312 sizeof(operand), &e)) {
7313 kvm_inject_page_fault(vcpu, &e);
7318 case VMX_EPT_EXTENT_GLOBAL:
7319 kvm_mmu_sync_roots(vcpu);
7320 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
7321 nested_vmx_succeed(vcpu);
7324 /* Trap single context invalidation invept calls */
7329 skip_emulated_instruction(vcpu);
7333 static int handle_invvpid(struct kvm_vcpu *vcpu)
7335 struct vcpu_vmx *vmx = to_vmx(vcpu);
7336 u32 vmx_instruction_info;
7337 unsigned long type, types;
7339 struct x86_exception e;
7342 if (!(vmx->nested.nested_vmx_secondary_ctls_high &
7343 SECONDARY_EXEC_ENABLE_VPID) ||
7344 !(vmx->nested.nested_vmx_vpid_caps & VMX_VPID_INVVPID_BIT)) {
7345 kvm_queue_exception(vcpu, UD_VECTOR);
7349 if (!nested_vmx_check_permission(vcpu))
7352 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7353 type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
7355 types = (vmx->nested.nested_vmx_vpid_caps >> 8) & 0x7;
7357 if (!(types & (1UL << type))) {
7358 nested_vmx_failValid(vcpu,
7359 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
7363 /* according to the intel vmx instruction reference, the memory
7364 * operand is read even if it isn't needed (e.g., for type==global)
7366 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
7367 vmx_instruction_info, false, &gva))
7369 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &vpid,
7371 kvm_inject_page_fault(vcpu, &e);
7376 case VMX_VPID_EXTENT_ALL_CONTEXT:
7377 if (get_vmcs12(vcpu)->virtual_processor_id == 0) {
7378 nested_vmx_failValid(vcpu,
7379 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
7382 vmx_flush_tlb(vcpu);
7383 nested_vmx_succeed(vcpu);
7386 /* Trap single context invalidation invvpid calls */
7391 skip_emulated_instruction(vcpu);
7395 static int handle_pml_full(struct kvm_vcpu *vcpu)
7397 unsigned long exit_qualification;
7399 trace_kvm_pml_full(vcpu->vcpu_id);
7401 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7404 * PML buffer FULL happened while executing iret from NMI,
7405 * "blocked by NMI" bit has to be set before next VM entry.
7407 if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
7408 cpu_has_virtual_nmis() &&
7409 (exit_qualification & INTR_INFO_UNBLOCK_NMI))
7410 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
7411 GUEST_INTR_STATE_NMI);
7414 * PML buffer already flushed at beginning of VMEXIT. Nothing to do
7415 * here.., and there's no userspace involvement needed for PML.
7420 static int handle_pcommit(struct kvm_vcpu *vcpu)
7422 /* we never catch pcommit instruct for L1 guest. */
7428 * The exit handlers return 1 if the exit was handled fully and guest execution
7429 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
7430 * to be done to userspace and return 0.
7432 static int (*const kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
7433 [EXIT_REASON_EXCEPTION_NMI] = handle_exception,
7434 [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt,
7435 [EXIT_REASON_TRIPLE_FAULT] = handle_triple_fault,
7436 [EXIT_REASON_NMI_WINDOW] = handle_nmi_window,
7437 [EXIT_REASON_IO_INSTRUCTION] = handle_io,
7438 [EXIT_REASON_CR_ACCESS] = handle_cr,
7439 [EXIT_REASON_DR_ACCESS] = handle_dr,
7440 [EXIT_REASON_CPUID] = handle_cpuid,
7441 [EXIT_REASON_MSR_READ] = handle_rdmsr,
7442 [EXIT_REASON_MSR_WRITE] = handle_wrmsr,
7443 [EXIT_REASON_PENDING_INTERRUPT] = handle_interrupt_window,
7444 [EXIT_REASON_HLT] = handle_halt,
7445 [EXIT_REASON_INVD] = handle_invd,
7446 [EXIT_REASON_INVLPG] = handle_invlpg,
7447 [EXIT_REASON_RDPMC] = handle_rdpmc,
7448 [EXIT_REASON_VMCALL] = handle_vmcall,
7449 [EXIT_REASON_VMCLEAR] = handle_vmclear,
7450 [EXIT_REASON_VMLAUNCH] = handle_vmlaunch,
7451 [EXIT_REASON_VMPTRLD] = handle_vmptrld,
7452 [EXIT_REASON_VMPTRST] = handle_vmptrst,
7453 [EXIT_REASON_VMREAD] = handle_vmread,
7454 [EXIT_REASON_VMRESUME] = handle_vmresume,
7455 [EXIT_REASON_VMWRITE] = handle_vmwrite,
7456 [EXIT_REASON_VMOFF] = handle_vmoff,
7457 [EXIT_REASON_VMON] = handle_vmon,
7458 [EXIT_REASON_TPR_BELOW_THRESHOLD] = handle_tpr_below_threshold,
7459 [EXIT_REASON_APIC_ACCESS] = handle_apic_access,
7460 [EXIT_REASON_APIC_WRITE] = handle_apic_write,
7461 [EXIT_REASON_EOI_INDUCED] = handle_apic_eoi_induced,
7462 [EXIT_REASON_WBINVD] = handle_wbinvd,
7463 [EXIT_REASON_XSETBV] = handle_xsetbv,
7464 [EXIT_REASON_TASK_SWITCH] = handle_task_switch,
7465 [EXIT_REASON_MCE_DURING_VMENTRY] = handle_machine_check,
7466 [EXIT_REASON_EPT_VIOLATION] = handle_ept_violation,
7467 [EXIT_REASON_EPT_MISCONFIG] = handle_ept_misconfig,
7468 [EXIT_REASON_PAUSE_INSTRUCTION] = handle_pause,
7469 [EXIT_REASON_MWAIT_INSTRUCTION] = handle_mwait,
7470 [EXIT_REASON_MONITOR_TRAP_FLAG] = handle_monitor_trap,
7471 [EXIT_REASON_MONITOR_INSTRUCTION] = handle_monitor,
7472 [EXIT_REASON_INVEPT] = handle_invept,
7473 [EXIT_REASON_INVVPID] = handle_invvpid,
7474 [EXIT_REASON_XSAVES] = handle_xsaves,
7475 [EXIT_REASON_XRSTORS] = handle_xrstors,
7476 [EXIT_REASON_PML_FULL] = handle_pml_full,
7477 [EXIT_REASON_PCOMMIT] = handle_pcommit,
7480 static const int kvm_vmx_max_exit_handlers =
7481 ARRAY_SIZE(kvm_vmx_exit_handlers);
7483 static bool nested_vmx_exit_handled_io(struct kvm_vcpu *vcpu,
7484 struct vmcs12 *vmcs12)
7486 unsigned long exit_qualification;
7487 gpa_t bitmap, last_bitmap;
7492 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
7493 return nested_cpu_has(vmcs12, CPU_BASED_UNCOND_IO_EXITING);
7495 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7497 port = exit_qualification >> 16;
7498 size = (exit_qualification & 7) + 1;
7500 last_bitmap = (gpa_t)-1;
7505 bitmap = vmcs12->io_bitmap_a;
7506 else if (port < 0x10000)
7507 bitmap = vmcs12->io_bitmap_b;
7510 bitmap += (port & 0x7fff) / 8;
7512 if (last_bitmap != bitmap)
7513 if (kvm_vcpu_read_guest(vcpu, bitmap, &b, 1))
7515 if (b & (1 << (port & 7)))
7520 last_bitmap = bitmap;
7527 * Return 1 if we should exit from L2 to L1 to handle an MSR access access,
7528 * rather than handle it ourselves in L0. I.e., check whether L1 expressed
7529 * disinterest in the current event (read or write a specific MSR) by using an
7530 * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps.
7532 static bool nested_vmx_exit_handled_msr(struct kvm_vcpu *vcpu,
7533 struct vmcs12 *vmcs12, u32 exit_reason)
7535 u32 msr_index = vcpu->arch.regs[VCPU_REGS_RCX];
7538 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
7542 * The MSR_BITMAP page is divided into four 1024-byte bitmaps,
7543 * for the four combinations of read/write and low/high MSR numbers.
7544 * First we need to figure out which of the four to use:
7546 bitmap = vmcs12->msr_bitmap;
7547 if (exit_reason == EXIT_REASON_MSR_WRITE)
7549 if (msr_index >= 0xc0000000) {
7550 msr_index -= 0xc0000000;
7554 /* Then read the msr_index'th bit from this bitmap: */
7555 if (msr_index < 1024*8) {
7557 if (kvm_vcpu_read_guest(vcpu, bitmap + msr_index/8, &b, 1))
7559 return 1 & (b >> (msr_index & 7));
7561 return true; /* let L1 handle the wrong parameter */
7565 * Return 1 if we should exit from L2 to L1 to handle a CR access exit,
7566 * rather than handle it ourselves in L0. I.e., check if L1 wanted to
7567 * intercept (via guest_host_mask etc.) the current event.
7569 static bool nested_vmx_exit_handled_cr(struct kvm_vcpu *vcpu,
7570 struct vmcs12 *vmcs12)
7572 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7573 int cr = exit_qualification & 15;
7574 int reg = (exit_qualification >> 8) & 15;
7575 unsigned long val = kvm_register_readl(vcpu, reg);
7577 switch ((exit_qualification >> 4) & 3) {
7578 case 0: /* mov to cr */
7581 if (vmcs12->cr0_guest_host_mask &
7582 (val ^ vmcs12->cr0_read_shadow))
7586 if ((vmcs12->cr3_target_count >= 1 &&
7587 vmcs12->cr3_target_value0 == val) ||
7588 (vmcs12->cr3_target_count >= 2 &&
7589 vmcs12->cr3_target_value1 == val) ||
7590 (vmcs12->cr3_target_count >= 3 &&
7591 vmcs12->cr3_target_value2 == val) ||
7592 (vmcs12->cr3_target_count >= 4 &&
7593 vmcs12->cr3_target_value3 == val))
7595 if (nested_cpu_has(vmcs12, CPU_BASED_CR3_LOAD_EXITING))
7599 if (vmcs12->cr4_guest_host_mask &
7600 (vmcs12->cr4_read_shadow ^ val))
7604 if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING))
7610 if ((vmcs12->cr0_guest_host_mask & X86_CR0_TS) &&
7611 (vmcs12->cr0_read_shadow & X86_CR0_TS))
7614 case 1: /* mov from cr */
7617 if (vmcs12->cpu_based_vm_exec_control &
7618 CPU_BASED_CR3_STORE_EXITING)
7622 if (vmcs12->cpu_based_vm_exec_control &
7623 CPU_BASED_CR8_STORE_EXITING)
7630 * lmsw can change bits 1..3 of cr0, and only set bit 0 of
7631 * cr0. Other attempted changes are ignored, with no exit.
7633 if (vmcs12->cr0_guest_host_mask & 0xe &
7634 (val ^ vmcs12->cr0_read_shadow))
7636 if ((vmcs12->cr0_guest_host_mask & 0x1) &&
7637 !(vmcs12->cr0_read_shadow & 0x1) &&
7646 * Return 1 if we should exit from L2 to L1 to handle an exit, or 0 if we
7647 * should handle it ourselves in L0 (and then continue L2). Only call this
7648 * when in is_guest_mode (L2).
7650 static bool nested_vmx_exit_handled(struct kvm_vcpu *vcpu)
7652 u32 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
7653 struct vcpu_vmx *vmx = to_vmx(vcpu);
7654 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7655 u32 exit_reason = vmx->exit_reason;
7657 trace_kvm_nested_vmexit(kvm_rip_read(vcpu), exit_reason,
7658 vmcs_readl(EXIT_QUALIFICATION),
7659 vmx->idt_vectoring_info,
7661 vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
7664 if (vmx->nested.nested_run_pending)
7667 if (unlikely(vmx->fail)) {
7668 pr_info_ratelimited("%s failed vm entry %x\n", __func__,
7669 vmcs_read32(VM_INSTRUCTION_ERROR));
7673 switch (exit_reason) {
7674 case EXIT_REASON_EXCEPTION_NMI:
7675 if (!is_exception(intr_info))
7677 else if (is_page_fault(intr_info))
7679 else if (is_no_device(intr_info) &&
7680 !(vmcs12->guest_cr0 & X86_CR0_TS))
7682 return vmcs12->exception_bitmap &
7683 (1u << (intr_info & INTR_INFO_VECTOR_MASK));
7684 case EXIT_REASON_EXTERNAL_INTERRUPT:
7686 case EXIT_REASON_TRIPLE_FAULT:
7688 case EXIT_REASON_PENDING_INTERRUPT:
7689 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_INTR_PENDING);
7690 case EXIT_REASON_NMI_WINDOW:
7691 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_NMI_PENDING);
7692 case EXIT_REASON_TASK_SWITCH:
7694 case EXIT_REASON_CPUID:
7695 if (kvm_register_read(vcpu, VCPU_REGS_RAX) == 0xa)
7698 case EXIT_REASON_HLT:
7699 return nested_cpu_has(vmcs12, CPU_BASED_HLT_EXITING);
7700 case EXIT_REASON_INVD:
7702 case EXIT_REASON_INVLPG:
7703 return nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
7704 case EXIT_REASON_RDPMC:
7705 return nested_cpu_has(vmcs12, CPU_BASED_RDPMC_EXITING);
7706 case EXIT_REASON_RDTSC: case EXIT_REASON_RDTSCP:
7707 return nested_cpu_has(vmcs12, CPU_BASED_RDTSC_EXITING);
7708 case EXIT_REASON_VMCALL: case EXIT_REASON_VMCLEAR:
7709 case EXIT_REASON_VMLAUNCH: case EXIT_REASON_VMPTRLD:
7710 case EXIT_REASON_VMPTRST: case EXIT_REASON_VMREAD:
7711 case EXIT_REASON_VMRESUME: case EXIT_REASON_VMWRITE:
7712 case EXIT_REASON_VMOFF: case EXIT_REASON_VMON:
7713 case EXIT_REASON_INVEPT: case EXIT_REASON_INVVPID:
7715 * VMX instructions trap unconditionally. This allows L1 to
7716 * emulate them for its L2 guest, i.e., allows 3-level nesting!
7719 case EXIT_REASON_CR_ACCESS:
7720 return nested_vmx_exit_handled_cr(vcpu, vmcs12);
7721 case EXIT_REASON_DR_ACCESS:
7722 return nested_cpu_has(vmcs12, CPU_BASED_MOV_DR_EXITING);
7723 case EXIT_REASON_IO_INSTRUCTION:
7724 return nested_vmx_exit_handled_io(vcpu, vmcs12);
7725 case EXIT_REASON_MSR_READ:
7726 case EXIT_REASON_MSR_WRITE:
7727 return nested_vmx_exit_handled_msr(vcpu, vmcs12, exit_reason);
7728 case EXIT_REASON_INVALID_STATE:
7730 case EXIT_REASON_MWAIT_INSTRUCTION:
7731 return nested_cpu_has(vmcs12, CPU_BASED_MWAIT_EXITING);
7732 case EXIT_REASON_MONITOR_TRAP_FLAG:
7733 return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_TRAP_FLAG);
7734 case EXIT_REASON_MONITOR_INSTRUCTION:
7735 return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_EXITING);
7736 case EXIT_REASON_PAUSE_INSTRUCTION:
7737 return nested_cpu_has(vmcs12, CPU_BASED_PAUSE_EXITING) ||
7738 nested_cpu_has2(vmcs12,
7739 SECONDARY_EXEC_PAUSE_LOOP_EXITING);
7740 case EXIT_REASON_MCE_DURING_VMENTRY:
7742 case EXIT_REASON_TPR_BELOW_THRESHOLD:
7743 return nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW);
7744 case EXIT_REASON_APIC_ACCESS:
7745 return nested_cpu_has2(vmcs12,
7746 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
7747 case EXIT_REASON_APIC_WRITE:
7748 case EXIT_REASON_EOI_INDUCED:
7749 /* apic_write and eoi_induced should exit unconditionally. */
7751 case EXIT_REASON_EPT_VIOLATION:
7753 * L0 always deals with the EPT violation. If nested EPT is
7754 * used, and the nested mmu code discovers that the address is
7755 * missing in the guest EPT table (EPT12), the EPT violation
7756 * will be injected with nested_ept_inject_page_fault()
7759 case EXIT_REASON_EPT_MISCONFIG:
7761 * L2 never uses directly L1's EPT, but rather L0's own EPT
7762 * table (shadow on EPT) or a merged EPT table that L0 built
7763 * (EPT on EPT). So any problems with the structure of the
7764 * table is L0's fault.
7767 case EXIT_REASON_WBINVD:
7768 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_WBINVD_EXITING);
7769 case EXIT_REASON_XSETBV:
7771 case EXIT_REASON_XSAVES: case EXIT_REASON_XRSTORS:
7773 * This should never happen, since it is not possible to
7774 * set XSS to a non-zero value---neither in L1 nor in L2.
7775 * If if it were, XSS would have to be checked against
7776 * the XSS exit bitmap in vmcs12.
7778 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES);
7779 case EXIT_REASON_PCOMMIT:
7780 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_PCOMMIT);
7786 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
7788 *info1 = vmcs_readl(EXIT_QUALIFICATION);
7789 *info2 = vmcs_read32(VM_EXIT_INTR_INFO);
7792 static int vmx_enable_pml(struct vcpu_vmx *vmx)
7794 struct page *pml_pg;
7796 pml_pg = alloc_page(GFP_KERNEL | __GFP_ZERO);
7800 vmx->pml_pg = pml_pg;
7802 vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
7803 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
7805 vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL, SECONDARY_EXEC_ENABLE_PML);
7810 static void vmx_disable_pml(struct vcpu_vmx *vmx)
7812 ASSERT(vmx->pml_pg);
7813 __free_page(vmx->pml_pg);
7816 vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL, SECONDARY_EXEC_ENABLE_PML);
7819 static void vmx_flush_pml_buffer(struct kvm_vcpu *vcpu)
7821 struct vcpu_vmx *vmx = to_vmx(vcpu);
7825 pml_idx = vmcs_read16(GUEST_PML_INDEX);
7827 /* Do nothing if PML buffer is empty */
7828 if (pml_idx == (PML_ENTITY_NUM - 1))
7831 /* PML index always points to next available PML buffer entity */
7832 if (pml_idx >= PML_ENTITY_NUM)
7837 pml_buf = page_address(vmx->pml_pg);
7838 for (; pml_idx < PML_ENTITY_NUM; pml_idx++) {
7841 gpa = pml_buf[pml_idx];
7842 WARN_ON(gpa & (PAGE_SIZE - 1));
7843 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
7846 /* reset PML index */
7847 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
7851 * Flush all vcpus' PML buffer and update logged GPAs to dirty_bitmap.
7852 * Called before reporting dirty_bitmap to userspace.
7854 static void kvm_flush_pml_buffers(struct kvm *kvm)
7857 struct kvm_vcpu *vcpu;
7859 * We only need to kick vcpu out of guest mode here, as PML buffer
7860 * is flushed at beginning of all VMEXITs, and it's obvious that only
7861 * vcpus running in guest are possible to have unflushed GPAs in PML
7864 kvm_for_each_vcpu(i, vcpu, kvm)
7865 kvm_vcpu_kick(vcpu);
7868 static void vmx_dump_sel(char *name, uint32_t sel)
7870 pr_err("%s sel=0x%04x, attr=0x%05x, limit=0x%08x, base=0x%016lx\n",
7871 name, vmcs_read32(sel),
7872 vmcs_read32(sel + GUEST_ES_AR_BYTES - GUEST_ES_SELECTOR),
7873 vmcs_read32(sel + GUEST_ES_LIMIT - GUEST_ES_SELECTOR),
7874 vmcs_readl(sel + GUEST_ES_BASE - GUEST_ES_SELECTOR));
7877 static void vmx_dump_dtsel(char *name, uint32_t limit)
7879 pr_err("%s limit=0x%08x, base=0x%016lx\n",
7880 name, vmcs_read32(limit),
7881 vmcs_readl(limit + GUEST_GDTR_BASE - GUEST_GDTR_LIMIT));
7884 static void dump_vmcs(void)
7886 u32 vmentry_ctl = vmcs_read32(VM_ENTRY_CONTROLS);
7887 u32 vmexit_ctl = vmcs_read32(VM_EXIT_CONTROLS);
7888 u32 cpu_based_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
7889 u32 pin_based_exec_ctrl = vmcs_read32(PIN_BASED_VM_EXEC_CONTROL);
7890 u32 secondary_exec_control = 0;
7891 unsigned long cr4 = vmcs_readl(GUEST_CR4);
7892 u64 efer = vmcs_readl(GUEST_IA32_EFER);
7895 if (cpu_has_secondary_exec_ctrls())
7896 secondary_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
7898 pr_err("*** Guest State ***\n");
7899 pr_err("CR0: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
7900 vmcs_readl(GUEST_CR0), vmcs_readl(CR0_READ_SHADOW),
7901 vmcs_readl(CR0_GUEST_HOST_MASK));
7902 pr_err("CR4: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
7903 cr4, vmcs_readl(CR4_READ_SHADOW), vmcs_readl(CR4_GUEST_HOST_MASK));
7904 pr_err("CR3 = 0x%016lx\n", vmcs_readl(GUEST_CR3));
7905 if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT) &&
7906 (cr4 & X86_CR4_PAE) && !(efer & EFER_LMA))
7908 pr_err("PDPTR0 = 0x%016lx PDPTR1 = 0x%016lx\n",
7909 vmcs_readl(GUEST_PDPTR0), vmcs_readl(GUEST_PDPTR1));
7910 pr_err("PDPTR2 = 0x%016lx PDPTR3 = 0x%016lx\n",
7911 vmcs_readl(GUEST_PDPTR2), vmcs_readl(GUEST_PDPTR3));
7913 pr_err("RSP = 0x%016lx RIP = 0x%016lx\n",
7914 vmcs_readl(GUEST_RSP), vmcs_readl(GUEST_RIP));
7915 pr_err("RFLAGS=0x%08lx DR7 = 0x%016lx\n",
7916 vmcs_readl(GUEST_RFLAGS), vmcs_readl(GUEST_DR7));
7917 pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
7918 vmcs_readl(GUEST_SYSENTER_ESP),
7919 vmcs_read32(GUEST_SYSENTER_CS), vmcs_readl(GUEST_SYSENTER_EIP));
7920 vmx_dump_sel("CS: ", GUEST_CS_SELECTOR);
7921 vmx_dump_sel("DS: ", GUEST_DS_SELECTOR);
7922 vmx_dump_sel("SS: ", GUEST_SS_SELECTOR);
7923 vmx_dump_sel("ES: ", GUEST_ES_SELECTOR);
7924 vmx_dump_sel("FS: ", GUEST_FS_SELECTOR);
7925 vmx_dump_sel("GS: ", GUEST_GS_SELECTOR);
7926 vmx_dump_dtsel("GDTR:", GUEST_GDTR_LIMIT);
7927 vmx_dump_sel("LDTR:", GUEST_LDTR_SELECTOR);
7928 vmx_dump_dtsel("IDTR:", GUEST_IDTR_LIMIT);
7929 vmx_dump_sel("TR: ", GUEST_TR_SELECTOR);
7930 if ((vmexit_ctl & (VM_EXIT_SAVE_IA32_PAT | VM_EXIT_SAVE_IA32_EFER)) ||
7931 (vmentry_ctl & (VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_IA32_EFER)))
7932 pr_err("EFER = 0x%016llx PAT = 0x%016lx\n",
7933 efer, vmcs_readl(GUEST_IA32_PAT));
7934 pr_err("DebugCtl = 0x%016lx DebugExceptions = 0x%016lx\n",
7935 vmcs_readl(GUEST_IA32_DEBUGCTL),
7936 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS));
7937 if (vmentry_ctl & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
7938 pr_err("PerfGlobCtl = 0x%016lx\n",
7939 vmcs_readl(GUEST_IA32_PERF_GLOBAL_CTRL));
7940 if (vmentry_ctl & VM_ENTRY_LOAD_BNDCFGS)
7941 pr_err("BndCfgS = 0x%016lx\n", vmcs_readl(GUEST_BNDCFGS));
7942 pr_err("Interruptibility = %08x ActivityState = %08x\n",
7943 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO),
7944 vmcs_read32(GUEST_ACTIVITY_STATE));
7945 if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
7946 pr_err("InterruptStatus = %04x\n",
7947 vmcs_read16(GUEST_INTR_STATUS));
7949 pr_err("*** Host State ***\n");
7950 pr_err("RIP = 0x%016lx RSP = 0x%016lx\n",
7951 vmcs_readl(HOST_RIP), vmcs_readl(HOST_RSP));
7952 pr_err("CS=%04x SS=%04x DS=%04x ES=%04x FS=%04x GS=%04x TR=%04x\n",
7953 vmcs_read16(HOST_CS_SELECTOR), vmcs_read16(HOST_SS_SELECTOR),
7954 vmcs_read16(HOST_DS_SELECTOR), vmcs_read16(HOST_ES_SELECTOR),
7955 vmcs_read16(HOST_FS_SELECTOR), vmcs_read16(HOST_GS_SELECTOR),
7956 vmcs_read16(HOST_TR_SELECTOR));
7957 pr_err("FSBase=%016lx GSBase=%016lx TRBase=%016lx\n",
7958 vmcs_readl(HOST_FS_BASE), vmcs_readl(HOST_GS_BASE),
7959 vmcs_readl(HOST_TR_BASE));
7960 pr_err("GDTBase=%016lx IDTBase=%016lx\n",
7961 vmcs_readl(HOST_GDTR_BASE), vmcs_readl(HOST_IDTR_BASE));
7962 pr_err("CR0=%016lx CR3=%016lx CR4=%016lx\n",
7963 vmcs_readl(HOST_CR0), vmcs_readl(HOST_CR3),
7964 vmcs_readl(HOST_CR4));
7965 pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
7966 vmcs_readl(HOST_IA32_SYSENTER_ESP),
7967 vmcs_read32(HOST_IA32_SYSENTER_CS),
7968 vmcs_readl(HOST_IA32_SYSENTER_EIP));
7969 if (vmexit_ctl & (VM_EXIT_LOAD_IA32_PAT | VM_EXIT_LOAD_IA32_EFER))
7970 pr_err("EFER = 0x%016lx PAT = 0x%016lx\n",
7971 vmcs_readl(HOST_IA32_EFER), vmcs_readl(HOST_IA32_PAT));
7972 if (vmexit_ctl & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
7973 pr_err("PerfGlobCtl = 0x%016lx\n",
7974 vmcs_readl(HOST_IA32_PERF_GLOBAL_CTRL));
7976 pr_err("*** Control State ***\n");
7977 pr_err("PinBased=%08x CPUBased=%08x SecondaryExec=%08x\n",
7978 pin_based_exec_ctrl, cpu_based_exec_ctrl, secondary_exec_control);
7979 pr_err("EntryControls=%08x ExitControls=%08x\n", vmentry_ctl, vmexit_ctl);
7980 pr_err("ExceptionBitmap=%08x PFECmask=%08x PFECmatch=%08x\n",
7981 vmcs_read32(EXCEPTION_BITMAP),
7982 vmcs_read32(PAGE_FAULT_ERROR_CODE_MASK),
7983 vmcs_read32(PAGE_FAULT_ERROR_CODE_MATCH));
7984 pr_err("VMEntry: intr_info=%08x errcode=%08x ilen=%08x\n",
7985 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
7986 vmcs_read32(VM_ENTRY_EXCEPTION_ERROR_CODE),
7987 vmcs_read32(VM_ENTRY_INSTRUCTION_LEN));
7988 pr_err("VMExit: intr_info=%08x errcode=%08x ilen=%08x\n",
7989 vmcs_read32(VM_EXIT_INTR_INFO),
7990 vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
7991 vmcs_read32(VM_EXIT_INSTRUCTION_LEN));
7992 pr_err(" reason=%08x qualification=%016lx\n",
7993 vmcs_read32(VM_EXIT_REASON), vmcs_readl(EXIT_QUALIFICATION));
7994 pr_err("IDTVectoring: info=%08x errcode=%08x\n",
7995 vmcs_read32(IDT_VECTORING_INFO_FIELD),
7996 vmcs_read32(IDT_VECTORING_ERROR_CODE));
7997 pr_err("TSC Offset = 0x%016lx\n", vmcs_readl(TSC_OFFSET));
7998 if (cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW)
7999 pr_err("TPR Threshold = 0x%02x\n", vmcs_read32(TPR_THRESHOLD));
8000 if (pin_based_exec_ctrl & PIN_BASED_POSTED_INTR)
8001 pr_err("PostedIntrVec = 0x%02x\n", vmcs_read16(POSTED_INTR_NV));
8002 if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT))
8003 pr_err("EPT pointer = 0x%016lx\n", vmcs_readl(EPT_POINTER));
8004 n = vmcs_read32(CR3_TARGET_COUNT);
8005 for (i = 0; i + 1 < n; i += 4)
8006 pr_err("CR3 target%u=%016lx target%u=%016lx\n",
8007 i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2),
8008 i + 1, vmcs_readl(CR3_TARGET_VALUE0 + i * 2 + 2));
8010 pr_err("CR3 target%u=%016lx\n",
8011 i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2));
8012 if (secondary_exec_control & SECONDARY_EXEC_PAUSE_LOOP_EXITING)
8013 pr_err("PLE Gap=%08x Window=%08x\n",
8014 vmcs_read32(PLE_GAP), vmcs_read32(PLE_WINDOW));
8015 if (secondary_exec_control & SECONDARY_EXEC_ENABLE_VPID)
8016 pr_err("Virtual processor ID = 0x%04x\n",
8017 vmcs_read16(VIRTUAL_PROCESSOR_ID));
8021 * The guest has exited. See if we can fix it or if we need userspace
8024 static int vmx_handle_exit(struct kvm_vcpu *vcpu)
8026 struct vcpu_vmx *vmx = to_vmx(vcpu);
8027 u32 exit_reason = vmx->exit_reason;
8028 u32 vectoring_info = vmx->idt_vectoring_info;
8031 * Flush logged GPAs PML buffer, this will make dirty_bitmap more
8032 * updated. Another good is, in kvm_vm_ioctl_get_dirty_log, before
8033 * querying dirty_bitmap, we only need to kick all vcpus out of guest
8034 * mode as if vcpus is in root mode, the PML buffer must has been
8038 vmx_flush_pml_buffer(vcpu);
8040 /* If guest state is invalid, start emulating */
8041 if (vmx->emulation_required)
8042 return handle_invalid_guest_state(vcpu);
8044 if (is_guest_mode(vcpu) && nested_vmx_exit_handled(vcpu)) {
8045 nested_vmx_vmexit(vcpu, exit_reason,
8046 vmcs_read32(VM_EXIT_INTR_INFO),
8047 vmcs_readl(EXIT_QUALIFICATION));
8051 if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
8053 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
8054 vcpu->run->fail_entry.hardware_entry_failure_reason
8059 if (unlikely(vmx->fail)) {
8060 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
8061 vcpu->run->fail_entry.hardware_entry_failure_reason
8062 = vmcs_read32(VM_INSTRUCTION_ERROR);
8068 * Do not try to fix EXIT_REASON_EPT_MISCONFIG if it caused by
8069 * delivery event since it indicates guest is accessing MMIO.
8070 * The vm-exit can be triggered again after return to guest that
8071 * will cause infinite loop.
8073 if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
8074 (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
8075 exit_reason != EXIT_REASON_EPT_VIOLATION &&
8076 exit_reason != EXIT_REASON_TASK_SWITCH)) {
8077 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
8078 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_DELIVERY_EV;
8079 vcpu->run->internal.ndata = 2;
8080 vcpu->run->internal.data[0] = vectoring_info;
8081 vcpu->run->internal.data[1] = exit_reason;
8085 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked &&
8086 !(is_guest_mode(vcpu) && nested_cpu_has_virtual_nmis(
8087 get_vmcs12(vcpu))))) {
8088 if (vmx_interrupt_allowed(vcpu)) {
8089 vmx->soft_vnmi_blocked = 0;
8090 } else if (vmx->vnmi_blocked_time > 1000000000LL &&
8091 vcpu->arch.nmi_pending) {
8093 * This CPU don't support us in finding the end of an
8094 * NMI-blocked window if the guest runs with IRQs
8095 * disabled. So we pull the trigger after 1 s of
8096 * futile waiting, but inform the user about this.
8098 printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
8099 "state on VCPU %d after 1 s timeout\n",
8100 __func__, vcpu->vcpu_id);
8101 vmx->soft_vnmi_blocked = 0;
8105 if (exit_reason < kvm_vmx_max_exit_handlers
8106 && kvm_vmx_exit_handlers[exit_reason])
8107 return kvm_vmx_exit_handlers[exit_reason](vcpu);
8109 WARN_ONCE(1, "vmx: unexpected exit reason 0x%x\n", exit_reason);
8110 kvm_queue_exception(vcpu, UD_VECTOR);
8115 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
8117 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
8119 if (is_guest_mode(vcpu) &&
8120 nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
8123 if (irr == -1 || tpr < irr) {
8124 vmcs_write32(TPR_THRESHOLD, 0);
8128 vmcs_write32(TPR_THRESHOLD, irr);
8131 static void vmx_set_virtual_x2apic_mode(struct kvm_vcpu *vcpu, bool set)
8133 u32 sec_exec_control;
8136 * There is not point to enable virtualize x2apic without enable
8139 if (!cpu_has_vmx_virtualize_x2apic_mode() ||
8140 !vmx_cpu_uses_apicv(vcpu))
8143 if (!cpu_need_tpr_shadow(vcpu))
8146 sec_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
8149 sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
8150 sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
8152 sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
8153 sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
8155 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, sec_exec_control);
8157 vmx_set_msr_bitmap(vcpu);
8160 static void vmx_set_apic_access_page_addr(struct kvm_vcpu *vcpu, hpa_t hpa)
8162 struct vcpu_vmx *vmx = to_vmx(vcpu);
8165 * Currently we do not handle the nested case where L2 has an
8166 * APIC access page of its own; that page is still pinned.
8167 * Hence, we skip the case where the VCPU is in guest mode _and_
8168 * L1 prepared an APIC access page for L2.
8170 * For the case where L1 and L2 share the same APIC access page
8171 * (flexpriority=Y but SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES clear
8172 * in the vmcs12), this function will only update either the vmcs01
8173 * or the vmcs02. If the former, the vmcs02 will be updated by
8174 * prepare_vmcs02. If the latter, the vmcs01 will be updated in
8175 * the next L2->L1 exit.
8177 if (!is_guest_mode(vcpu) ||
8178 !nested_cpu_has2(vmx->nested.current_vmcs12,
8179 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
8180 vmcs_write64(APIC_ACCESS_ADDR, hpa);
8183 static void vmx_hwapic_isr_update(struct kvm *kvm, int isr)
8191 status = vmcs_read16(GUEST_INTR_STATUS);
8196 vmcs_write16(GUEST_INTR_STATUS, status);
8200 static void vmx_set_rvi(int vector)
8208 status = vmcs_read16(GUEST_INTR_STATUS);
8209 old = (u8)status & 0xff;
8210 if ((u8)vector != old) {
8212 status |= (u8)vector;
8213 vmcs_write16(GUEST_INTR_STATUS, status);
8217 static void vmx_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
8219 if (!is_guest_mode(vcpu)) {
8220 vmx_set_rvi(max_irr);
8228 * In guest mode. If a vmexit is needed, vmx_check_nested_events
8231 if (nested_exit_on_intr(vcpu))
8235 * Else, fall back to pre-APICv interrupt injection since L2
8236 * is run without virtual interrupt delivery.
8238 if (!kvm_event_needs_reinjection(vcpu) &&
8239 vmx_interrupt_allowed(vcpu)) {
8240 kvm_queue_interrupt(vcpu, max_irr, false);
8241 vmx_inject_irq(vcpu);
8245 static void vmx_load_eoi_exitmap(struct kvm_vcpu *vcpu)
8247 u64 *eoi_exit_bitmap = vcpu->arch.eoi_exit_bitmap;
8248 if (!vmx_cpu_uses_apicv(vcpu))
8251 vmcs_write64(EOI_EXIT_BITMAP0, eoi_exit_bitmap[0]);
8252 vmcs_write64(EOI_EXIT_BITMAP1, eoi_exit_bitmap[1]);
8253 vmcs_write64(EOI_EXIT_BITMAP2, eoi_exit_bitmap[2]);
8254 vmcs_write64(EOI_EXIT_BITMAP3, eoi_exit_bitmap[3]);
8257 static void vmx_complete_atomic_exit(struct vcpu_vmx *vmx)
8261 if (!(vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY
8262 || vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI))
8265 vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8266 exit_intr_info = vmx->exit_intr_info;
8268 /* Handle machine checks before interrupts are enabled */
8269 if (is_machine_check(exit_intr_info))
8270 kvm_machine_check();
8272 /* We need to handle NMIs before interrupts are enabled */
8273 if ((exit_intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR &&
8274 (exit_intr_info & INTR_INFO_VALID_MASK)) {
8275 kvm_before_handle_nmi(&vmx->vcpu);
8277 kvm_after_handle_nmi(&vmx->vcpu);
8281 static void vmx_handle_external_intr(struct kvm_vcpu *vcpu)
8283 u32 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8286 * If external interrupt exists, IF bit is set in rflags/eflags on the
8287 * interrupt stack frame, and interrupt will be enabled on a return
8288 * from interrupt handler.
8290 if ((exit_intr_info & (INTR_INFO_VALID_MASK | INTR_INFO_INTR_TYPE_MASK))
8291 == (INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR)) {
8292 unsigned int vector;
8293 unsigned long entry;
8295 struct vcpu_vmx *vmx = to_vmx(vcpu);
8296 #ifdef CONFIG_X86_64
8300 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
8301 desc = (gate_desc *)vmx->host_idt_base + vector;
8302 entry = gate_offset(*desc);
8304 #ifdef CONFIG_X86_64
8305 "mov %%" _ASM_SP ", %[sp]\n\t"
8306 "and $0xfffffffffffffff0, %%" _ASM_SP "\n\t"
8311 "orl $0x200, (%%" _ASM_SP ")\n\t"
8312 __ASM_SIZE(push) " $%c[cs]\n\t"
8313 "call *%[entry]\n\t"
8315 #ifdef CONFIG_X86_64
8320 [ss]"i"(__KERNEL_DS),
8321 [cs]"i"(__KERNEL_CS)
8327 static bool vmx_has_high_real_mode_segbase(void)
8329 return enable_unrestricted_guest || emulate_invalid_guest_state;
8332 static bool vmx_mpx_supported(void)
8334 return (vmcs_config.vmexit_ctrl & VM_EXIT_CLEAR_BNDCFGS) &&
8335 (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_BNDCFGS);
8338 static bool vmx_xsaves_supported(void)
8340 return vmcs_config.cpu_based_2nd_exec_ctrl &
8341 SECONDARY_EXEC_XSAVES;
8344 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
8349 bool idtv_info_valid;
8351 idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
8353 if (cpu_has_virtual_nmis()) {
8354 if (vmx->nmi_known_unmasked)
8357 * Can't use vmx->exit_intr_info since we're not sure what
8358 * the exit reason is.
8360 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8361 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
8362 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
8364 * SDM 3: 27.7.1.2 (September 2008)
8365 * Re-set bit "block by NMI" before VM entry if vmexit caused by
8366 * a guest IRET fault.
8367 * SDM 3: 23.2.2 (September 2008)
8368 * Bit 12 is undefined in any of the following cases:
8369 * If the VM exit sets the valid bit in the IDT-vectoring
8370 * information field.
8371 * If the VM exit is due to a double fault.
8373 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
8374 vector != DF_VECTOR && !idtv_info_valid)
8375 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
8376 GUEST_INTR_STATE_NMI);
8378 vmx->nmi_known_unmasked =
8379 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
8380 & GUEST_INTR_STATE_NMI);
8381 } else if (unlikely(vmx->soft_vnmi_blocked))
8382 vmx->vnmi_blocked_time +=
8383 ktime_to_ns(ktime_sub(ktime_get(), vmx->entry_time));
8386 static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu,
8387 u32 idt_vectoring_info,
8388 int instr_len_field,
8389 int error_code_field)
8393 bool idtv_info_valid;
8395 idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
8397 vcpu->arch.nmi_injected = false;
8398 kvm_clear_exception_queue(vcpu);
8399 kvm_clear_interrupt_queue(vcpu);
8401 if (!idtv_info_valid)
8404 kvm_make_request(KVM_REQ_EVENT, vcpu);
8406 vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
8407 type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
8410 case INTR_TYPE_NMI_INTR:
8411 vcpu->arch.nmi_injected = true;
8413 * SDM 3: 27.7.1.2 (September 2008)
8414 * Clear bit "block by NMI" before VM entry if a NMI
8417 vmx_set_nmi_mask(vcpu, false);
8419 case INTR_TYPE_SOFT_EXCEPTION:
8420 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
8422 case INTR_TYPE_HARD_EXCEPTION:
8423 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
8424 u32 err = vmcs_read32(error_code_field);
8425 kvm_requeue_exception_e(vcpu, vector, err);
8427 kvm_requeue_exception(vcpu, vector);
8429 case INTR_TYPE_SOFT_INTR:
8430 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
8432 case INTR_TYPE_EXT_INTR:
8433 kvm_queue_interrupt(vcpu, vector, type == INTR_TYPE_SOFT_INTR);
8440 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
8442 __vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info,
8443 VM_EXIT_INSTRUCTION_LEN,
8444 IDT_VECTORING_ERROR_CODE);
8447 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
8449 __vmx_complete_interrupts(vcpu,
8450 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
8451 VM_ENTRY_INSTRUCTION_LEN,
8452 VM_ENTRY_EXCEPTION_ERROR_CODE);
8454 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
8457 static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
8460 struct perf_guest_switch_msr *msrs;
8462 msrs = perf_guest_get_msrs(&nr_msrs);
8467 for (i = 0; i < nr_msrs; i++)
8468 if (msrs[i].host == msrs[i].guest)
8469 clear_atomic_switch_msr(vmx, msrs[i].msr);
8471 add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
8475 static void __noclone vmx_vcpu_run(struct kvm_vcpu *vcpu)
8477 struct vcpu_vmx *vmx = to_vmx(vcpu);
8478 unsigned long debugctlmsr, cr4;
8480 /* Record the guest's net vcpu time for enforced NMI injections. */
8481 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked))
8482 vmx->entry_time = ktime_get();
8484 /* Don't enter VMX if guest state is invalid, let the exit handler
8485 start emulation until we arrive back to a valid state */
8486 if (vmx->emulation_required)
8489 if (vmx->ple_window_dirty) {
8490 vmx->ple_window_dirty = false;
8491 vmcs_write32(PLE_WINDOW, vmx->ple_window);
8494 if (vmx->nested.sync_shadow_vmcs) {
8495 copy_vmcs12_to_shadow(vmx);
8496 vmx->nested.sync_shadow_vmcs = false;
8499 if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
8500 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
8501 if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
8502 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
8504 cr4 = cr4_read_shadow();
8505 if (unlikely(cr4 != vmx->host_state.vmcs_host_cr4)) {
8506 vmcs_writel(HOST_CR4, cr4);
8507 vmx->host_state.vmcs_host_cr4 = cr4;
8510 /* When single-stepping over STI and MOV SS, we must clear the
8511 * corresponding interruptibility bits in the guest state. Otherwise
8512 * vmentry fails as it then expects bit 14 (BS) in pending debug
8513 * exceptions being set, but that's not correct for the guest debugging
8515 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8516 vmx_set_interrupt_shadow(vcpu, 0);
8518 atomic_switch_perf_msrs(vmx);
8519 debugctlmsr = get_debugctlmsr();
8521 vmx->__launched = vmx->loaded_vmcs->launched;
8523 /* Store host registers */
8524 "push %%" _ASM_DX "; push %%" _ASM_BP ";"
8525 "push %%" _ASM_CX " \n\t" /* placeholder for guest rcx */
8526 "push %%" _ASM_CX " \n\t"
8527 "cmp %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
8529 "mov %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
8530 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
8532 /* Reload cr2 if changed */
8533 "mov %c[cr2](%0), %%" _ASM_AX " \n\t"
8534 "mov %%cr2, %%" _ASM_DX " \n\t"
8535 "cmp %%" _ASM_AX ", %%" _ASM_DX " \n\t"
8537 "mov %%" _ASM_AX", %%cr2 \n\t"
8539 /* Check if vmlaunch of vmresume is needed */
8540 "cmpl $0, %c[launched](%0) \n\t"
8541 /* Load guest registers. Don't clobber flags. */
8542 "mov %c[rax](%0), %%" _ASM_AX " \n\t"
8543 "mov %c[rbx](%0), %%" _ASM_BX " \n\t"
8544 "mov %c[rdx](%0), %%" _ASM_DX " \n\t"
8545 "mov %c[rsi](%0), %%" _ASM_SI " \n\t"
8546 "mov %c[rdi](%0), %%" _ASM_DI " \n\t"
8547 "mov %c[rbp](%0), %%" _ASM_BP " \n\t"
8548 #ifdef CONFIG_X86_64
8549 "mov %c[r8](%0), %%r8 \n\t"
8550 "mov %c[r9](%0), %%r9 \n\t"
8551 "mov %c[r10](%0), %%r10 \n\t"
8552 "mov %c[r11](%0), %%r11 \n\t"
8553 "mov %c[r12](%0), %%r12 \n\t"
8554 "mov %c[r13](%0), %%r13 \n\t"
8555 "mov %c[r14](%0), %%r14 \n\t"
8556 "mov %c[r15](%0), %%r15 \n\t"
8558 "mov %c[rcx](%0), %%" _ASM_CX " \n\t" /* kills %0 (ecx) */
8560 /* Enter guest mode */
8562 __ex(ASM_VMX_VMLAUNCH) "\n\t"
8564 "1: " __ex(ASM_VMX_VMRESUME) "\n\t"
8566 /* Save guest registers, load host registers, keep flags */
8567 "mov %0, %c[wordsize](%%" _ASM_SP ") \n\t"
8569 "mov %%" _ASM_AX ", %c[rax](%0) \n\t"
8570 "mov %%" _ASM_BX ", %c[rbx](%0) \n\t"
8571 __ASM_SIZE(pop) " %c[rcx](%0) \n\t"
8572 "mov %%" _ASM_DX ", %c[rdx](%0) \n\t"
8573 "mov %%" _ASM_SI ", %c[rsi](%0) \n\t"
8574 "mov %%" _ASM_DI ", %c[rdi](%0) \n\t"
8575 "mov %%" _ASM_BP ", %c[rbp](%0) \n\t"
8576 #ifdef CONFIG_X86_64
8577 "mov %%r8, %c[r8](%0) \n\t"
8578 "mov %%r9, %c[r9](%0) \n\t"
8579 "mov %%r10, %c[r10](%0) \n\t"
8580 "mov %%r11, %c[r11](%0) \n\t"
8581 "mov %%r12, %c[r12](%0) \n\t"
8582 "mov %%r13, %c[r13](%0) \n\t"
8583 "mov %%r14, %c[r14](%0) \n\t"
8584 "mov %%r15, %c[r15](%0) \n\t"
8586 "mov %%cr2, %%" _ASM_AX " \n\t"
8587 "mov %%" _ASM_AX ", %c[cr2](%0) \n\t"
8589 "pop %%" _ASM_BP "; pop %%" _ASM_DX " \n\t"
8590 "setbe %c[fail](%0) \n\t"
8591 ".pushsection .rodata \n\t"
8592 ".global vmx_return \n\t"
8593 "vmx_return: " _ASM_PTR " 2b \n\t"
8595 : : "c"(vmx), "d"((unsigned long)HOST_RSP),
8596 [launched]"i"(offsetof(struct vcpu_vmx, __launched)),
8597 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
8598 [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
8599 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
8600 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
8601 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
8602 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
8603 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
8604 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
8605 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
8606 #ifdef CONFIG_X86_64
8607 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
8608 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
8609 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
8610 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
8611 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
8612 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
8613 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
8614 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
8616 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2)),
8617 [wordsize]"i"(sizeof(ulong))
8619 #ifdef CONFIG_X86_64
8620 , "rax", "rbx", "rdi", "rsi"
8621 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
8623 , "eax", "ebx", "edi", "esi"
8627 /* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
8629 update_debugctlmsr(debugctlmsr);
8631 #ifndef CONFIG_X86_64
8633 * The sysexit path does not restore ds/es, so we must set them to
8634 * a reasonable value ourselves.
8636 * We can't defer this to vmx_load_host_state() since that function
8637 * may be executed in interrupt context, which saves and restore segments
8638 * around it, nullifying its effect.
8640 loadsegment(ds, __USER_DS);
8641 loadsegment(es, __USER_DS);
8644 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
8645 | (1 << VCPU_EXREG_RFLAGS)
8646 | (1 << VCPU_EXREG_PDPTR)
8647 | (1 << VCPU_EXREG_SEGMENTS)
8648 | (1 << VCPU_EXREG_CR3));
8649 vcpu->arch.regs_dirty = 0;
8651 vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
8653 vmx->loaded_vmcs->launched = 1;
8655 vmx->exit_reason = vmcs_read32(VM_EXIT_REASON);
8656 trace_kvm_exit(vmx->exit_reason, vcpu, KVM_ISA_VMX);
8659 * the KVM_REQ_EVENT optimization bit is only on for one entry, and if
8660 * we did not inject a still-pending event to L1 now because of
8661 * nested_run_pending, we need to re-enable this bit.
8663 if (vmx->nested.nested_run_pending)
8664 kvm_make_request(KVM_REQ_EVENT, vcpu);
8666 vmx->nested.nested_run_pending = 0;
8668 vmx_complete_atomic_exit(vmx);
8669 vmx_recover_nmi_blocking(vmx);
8670 vmx_complete_interrupts(vmx);
8673 static void vmx_load_vmcs01(struct kvm_vcpu *vcpu)
8675 struct vcpu_vmx *vmx = to_vmx(vcpu);
8678 if (vmx->loaded_vmcs == &vmx->vmcs01)
8682 vmx->loaded_vmcs = &vmx->vmcs01;
8684 vmx_vcpu_load(vcpu, cpu);
8689 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
8691 struct vcpu_vmx *vmx = to_vmx(vcpu);
8694 vmx_disable_pml(vmx);
8695 free_vpid(vmx->vpid);
8696 leave_guest_mode(vcpu);
8697 vmx_load_vmcs01(vcpu);
8699 free_loaded_vmcs(vmx->loaded_vmcs);
8700 kfree(vmx->guest_msrs);
8701 kvm_vcpu_uninit(vcpu);
8702 kmem_cache_free(kvm_vcpu_cache, vmx);
8705 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
8708 struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
8712 return ERR_PTR(-ENOMEM);
8714 vmx->vpid = allocate_vpid();
8716 err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
8720 vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
8721 BUILD_BUG_ON(ARRAY_SIZE(vmx_msr_index) * sizeof(vmx->guest_msrs[0])
8725 if (!vmx->guest_msrs) {
8729 vmx->loaded_vmcs = &vmx->vmcs01;
8730 vmx->loaded_vmcs->vmcs = alloc_vmcs();
8731 if (!vmx->loaded_vmcs->vmcs)
8734 kvm_cpu_vmxon(__pa(per_cpu(vmxarea, raw_smp_processor_id())));
8735 loaded_vmcs_init(vmx->loaded_vmcs);
8740 vmx_vcpu_load(&vmx->vcpu, cpu);
8741 vmx->vcpu.cpu = cpu;
8742 err = vmx_vcpu_setup(vmx);
8743 vmx_vcpu_put(&vmx->vcpu);
8747 if (cpu_need_virtualize_apic_accesses(&vmx->vcpu)) {
8748 err = alloc_apic_access_page(kvm);
8754 if (!kvm->arch.ept_identity_map_addr)
8755 kvm->arch.ept_identity_map_addr =
8756 VMX_EPT_IDENTITY_PAGETABLE_ADDR;
8757 err = init_rmode_identity_map(kvm);
8763 nested_vmx_setup_ctls_msrs(vmx);
8765 vmx->nested.posted_intr_nv = -1;
8766 vmx->nested.current_vmptr = -1ull;
8767 vmx->nested.current_vmcs12 = NULL;
8770 * If PML is turned on, failure on enabling PML just results in failure
8771 * of creating the vcpu, therefore we can simplify PML logic (by
8772 * avoiding dealing with cases, such as enabling PML partially on vcpus
8773 * for the guest, etc.
8776 err = vmx_enable_pml(vmx);
8784 free_loaded_vmcs(vmx->loaded_vmcs);
8786 kfree(vmx->guest_msrs);
8788 kvm_vcpu_uninit(&vmx->vcpu);
8790 free_vpid(vmx->vpid);
8791 kmem_cache_free(kvm_vcpu_cache, vmx);
8792 return ERR_PTR(err);
8795 static void __init vmx_check_processor_compat(void *rtn)
8797 struct vmcs_config vmcs_conf;
8800 if (setup_vmcs_config(&vmcs_conf) < 0)
8802 if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
8803 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
8804 smp_processor_id());
8809 static int get_ept_level(void)
8811 return VMX_EPT_DEFAULT_GAW + 1;
8814 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
8819 /* For VT-d and EPT combination
8820 * 1. MMIO: always map as UC
8822 * a. VT-d without snooping control feature: can't guarantee the
8823 * result, try to trust guest.
8824 * b. VT-d with snooping control feature: snooping control feature of
8825 * VT-d engine can guarantee the cache correctness. Just set it
8826 * to WB to keep consistent with host. So the same as item 3.
8827 * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
8828 * consistent with host MTRR
8831 cache = MTRR_TYPE_UNCACHABLE;
8835 if (!kvm_arch_has_noncoherent_dma(vcpu->kvm)) {
8836 ipat = VMX_EPT_IPAT_BIT;
8837 cache = MTRR_TYPE_WRBACK;
8841 if (kvm_read_cr0(vcpu) & X86_CR0_CD) {
8842 ipat = VMX_EPT_IPAT_BIT;
8843 if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
8844 cache = MTRR_TYPE_WRBACK;
8846 cache = MTRR_TYPE_UNCACHABLE;
8850 cache = kvm_mtrr_get_guest_memory_type(vcpu, gfn);
8853 return (cache << VMX_EPT_MT_EPTE_SHIFT) | ipat;
8856 static int vmx_get_lpage_level(void)
8858 if (enable_ept && !cpu_has_vmx_ept_1g_page())
8859 return PT_DIRECTORY_LEVEL;
8861 /* For shadow and EPT supported 1GB page */
8862 return PT_PDPE_LEVEL;
8865 static void vmcs_set_secondary_exec_control(u32 new_ctl)
8868 * These bits in the secondary execution controls field
8869 * are dynamic, the others are mostly based on the hypervisor
8870 * architecture and the guest's CPUID. Do not touch the
8874 SECONDARY_EXEC_SHADOW_VMCS |
8875 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
8876 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
8878 u32 cur_ctl = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
8880 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
8881 (new_ctl & ~mask) | (cur_ctl & mask));
8884 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
8886 struct kvm_cpuid_entry2 *best;
8887 struct vcpu_vmx *vmx = to_vmx(vcpu);
8888 u32 secondary_exec_ctl = vmx_secondary_exec_control(vmx);
8890 if (vmx_rdtscp_supported()) {
8891 bool rdtscp_enabled = guest_cpuid_has_rdtscp(vcpu);
8892 if (!rdtscp_enabled)
8893 secondary_exec_ctl &= ~SECONDARY_EXEC_RDTSCP;
8897 vmx->nested.nested_vmx_secondary_ctls_high |=
8898 SECONDARY_EXEC_RDTSCP;
8900 vmx->nested.nested_vmx_secondary_ctls_high &=
8901 ~SECONDARY_EXEC_RDTSCP;
8905 /* Exposing INVPCID only when PCID is exposed */
8906 best = kvm_find_cpuid_entry(vcpu, 0x7, 0);
8907 if (vmx_invpcid_supported() &&
8908 (!best || !(best->ebx & bit(X86_FEATURE_INVPCID)) ||
8909 !guest_cpuid_has_pcid(vcpu))) {
8910 secondary_exec_ctl &= ~SECONDARY_EXEC_ENABLE_INVPCID;
8913 best->ebx &= ~bit(X86_FEATURE_INVPCID);
8916 vmcs_set_secondary_exec_control(secondary_exec_ctl);
8918 if (static_cpu_has(X86_FEATURE_PCOMMIT) && nested) {
8919 if (guest_cpuid_has_pcommit(vcpu))
8920 vmx->nested.nested_vmx_secondary_ctls_high |=
8921 SECONDARY_EXEC_PCOMMIT;
8923 vmx->nested.nested_vmx_secondary_ctls_high &=
8924 ~SECONDARY_EXEC_PCOMMIT;
8928 static void vmx_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
8930 if (func == 1 && nested)
8931 entry->ecx |= bit(X86_FEATURE_VMX);
8934 static void nested_ept_inject_page_fault(struct kvm_vcpu *vcpu,
8935 struct x86_exception *fault)
8937 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
8940 if (fault->error_code & PFERR_RSVD_MASK)
8941 exit_reason = EXIT_REASON_EPT_MISCONFIG;
8943 exit_reason = EXIT_REASON_EPT_VIOLATION;
8944 nested_vmx_vmexit(vcpu, exit_reason, 0, vcpu->arch.exit_qualification);
8945 vmcs12->guest_physical_address = fault->address;
8948 /* Callbacks for nested_ept_init_mmu_context: */
8950 static unsigned long nested_ept_get_cr3(struct kvm_vcpu *vcpu)
8952 /* return the page table to be shadowed - in our case, EPT12 */
8953 return get_vmcs12(vcpu)->ept_pointer;
8956 static void nested_ept_init_mmu_context(struct kvm_vcpu *vcpu)
8958 WARN_ON(mmu_is_nested(vcpu));
8959 kvm_init_shadow_ept_mmu(vcpu,
8960 to_vmx(vcpu)->nested.nested_vmx_ept_caps &
8961 VMX_EPT_EXECUTE_ONLY_BIT);
8962 vcpu->arch.mmu.set_cr3 = vmx_set_cr3;
8963 vcpu->arch.mmu.get_cr3 = nested_ept_get_cr3;
8964 vcpu->arch.mmu.inject_page_fault = nested_ept_inject_page_fault;
8966 vcpu->arch.walk_mmu = &vcpu->arch.nested_mmu;
8969 static void nested_ept_uninit_mmu_context(struct kvm_vcpu *vcpu)
8971 vcpu->arch.walk_mmu = &vcpu->arch.mmu;
8974 static bool nested_vmx_is_page_fault_vmexit(struct vmcs12 *vmcs12,
8977 bool inequality, bit;
8979 bit = (vmcs12->exception_bitmap & (1u << PF_VECTOR)) != 0;
8981 (error_code & vmcs12->page_fault_error_code_mask) !=
8982 vmcs12->page_fault_error_code_match;
8983 return inequality ^ bit;
8986 static void vmx_inject_page_fault_nested(struct kvm_vcpu *vcpu,
8987 struct x86_exception *fault)
8989 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
8991 WARN_ON(!is_guest_mode(vcpu));
8993 if (nested_vmx_is_page_fault_vmexit(vmcs12, fault->error_code))
8994 nested_vmx_vmexit(vcpu, to_vmx(vcpu)->exit_reason,
8995 vmcs_read32(VM_EXIT_INTR_INFO),
8996 vmcs_readl(EXIT_QUALIFICATION));
8998 kvm_inject_page_fault(vcpu, fault);
9001 static bool nested_get_vmcs12_pages(struct kvm_vcpu *vcpu,
9002 struct vmcs12 *vmcs12)
9004 struct vcpu_vmx *vmx = to_vmx(vcpu);
9005 int maxphyaddr = cpuid_maxphyaddr(vcpu);
9007 if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
9008 if (!PAGE_ALIGNED(vmcs12->apic_access_addr) ||
9009 vmcs12->apic_access_addr >> maxphyaddr)
9013 * Translate L1 physical address to host physical
9014 * address for vmcs02. Keep the page pinned, so this
9015 * physical address remains valid. We keep a reference
9016 * to it so we can release it later.
9018 if (vmx->nested.apic_access_page) /* shouldn't happen */
9019 nested_release_page(vmx->nested.apic_access_page);
9020 vmx->nested.apic_access_page =
9021 nested_get_page(vcpu, vmcs12->apic_access_addr);
9024 if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
9025 if (!PAGE_ALIGNED(vmcs12->virtual_apic_page_addr) ||
9026 vmcs12->virtual_apic_page_addr >> maxphyaddr)
9029 if (vmx->nested.virtual_apic_page) /* shouldn't happen */
9030 nested_release_page(vmx->nested.virtual_apic_page);
9031 vmx->nested.virtual_apic_page =
9032 nested_get_page(vcpu, vmcs12->virtual_apic_page_addr);
9035 * Failing the vm entry is _not_ what the processor does
9036 * but it's basically the only possibility we have.
9037 * We could still enter the guest if CR8 load exits are
9038 * enabled, CR8 store exits are enabled, and virtualize APIC
9039 * access is disabled; in this case the processor would never
9040 * use the TPR shadow and we could simply clear the bit from
9041 * the execution control. But such a configuration is useless,
9042 * so let's keep the code simple.
9044 if (!vmx->nested.virtual_apic_page)
9048 if (nested_cpu_has_posted_intr(vmcs12)) {
9049 if (!IS_ALIGNED(vmcs12->posted_intr_desc_addr, 64) ||
9050 vmcs12->posted_intr_desc_addr >> maxphyaddr)
9053 if (vmx->nested.pi_desc_page) { /* shouldn't happen */
9054 kunmap(vmx->nested.pi_desc_page);
9055 nested_release_page(vmx->nested.pi_desc_page);
9057 vmx->nested.pi_desc_page =
9058 nested_get_page(vcpu, vmcs12->posted_intr_desc_addr);
9059 if (!vmx->nested.pi_desc_page)
9062 vmx->nested.pi_desc =
9063 (struct pi_desc *)kmap(vmx->nested.pi_desc_page);
9064 if (!vmx->nested.pi_desc) {
9065 nested_release_page_clean(vmx->nested.pi_desc_page);
9068 vmx->nested.pi_desc =
9069 (struct pi_desc *)((void *)vmx->nested.pi_desc +
9070 (unsigned long)(vmcs12->posted_intr_desc_addr &
9077 static void vmx_start_preemption_timer(struct kvm_vcpu *vcpu)
9079 u64 preemption_timeout = get_vmcs12(vcpu)->vmx_preemption_timer_value;
9080 struct vcpu_vmx *vmx = to_vmx(vcpu);
9082 if (vcpu->arch.virtual_tsc_khz == 0)
9085 /* Make sure short timeouts reliably trigger an immediate vmexit.
9086 * hrtimer_start does not guarantee this. */
9087 if (preemption_timeout <= 1) {
9088 vmx_preemption_timer_fn(&vmx->nested.preemption_timer);
9092 preemption_timeout <<= VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
9093 preemption_timeout *= 1000000;
9094 do_div(preemption_timeout, vcpu->arch.virtual_tsc_khz);
9095 hrtimer_start(&vmx->nested.preemption_timer,
9096 ns_to_ktime(preemption_timeout), HRTIMER_MODE_REL);
9099 static int nested_vmx_check_msr_bitmap_controls(struct kvm_vcpu *vcpu,
9100 struct vmcs12 *vmcs12)
9105 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
9108 if (vmcs12_read_any(vcpu, MSR_BITMAP, &addr)) {
9112 maxphyaddr = cpuid_maxphyaddr(vcpu);
9114 if (!PAGE_ALIGNED(vmcs12->msr_bitmap) ||
9115 ((addr + PAGE_SIZE) >> maxphyaddr))
9122 * Merge L0's and L1's MSR bitmap, return false to indicate that
9123 * we do not use the hardware.
9125 static inline bool nested_vmx_merge_msr_bitmap(struct kvm_vcpu *vcpu,
9126 struct vmcs12 *vmcs12)
9130 unsigned long *msr_bitmap;
9132 if (!nested_cpu_has_virt_x2apic_mode(vmcs12))
9135 page = nested_get_page(vcpu, vmcs12->msr_bitmap);
9140 msr_bitmap = (unsigned long *)kmap(page);
9142 nested_release_page_clean(page);
9147 if (nested_cpu_has_virt_x2apic_mode(vmcs12)) {
9148 if (nested_cpu_has_apic_reg_virt(vmcs12))
9149 for (msr = 0x800; msr <= 0x8ff; msr++)
9150 nested_vmx_disable_intercept_for_msr(
9152 vmx_msr_bitmap_nested,
9154 /* TPR is allowed */
9155 nested_vmx_disable_intercept_for_msr(msr_bitmap,
9156 vmx_msr_bitmap_nested,
9157 APIC_BASE_MSR + (APIC_TASKPRI >> 4),
9158 MSR_TYPE_R | MSR_TYPE_W);
9159 if (nested_cpu_has_vid(vmcs12)) {
9160 /* EOI and self-IPI are allowed */
9161 nested_vmx_disable_intercept_for_msr(
9163 vmx_msr_bitmap_nested,
9164 APIC_BASE_MSR + (APIC_EOI >> 4),
9166 nested_vmx_disable_intercept_for_msr(
9168 vmx_msr_bitmap_nested,
9169 APIC_BASE_MSR + (APIC_SELF_IPI >> 4),
9174 * Enable reading intercept of all the x2apic
9175 * MSRs. We should not rely on vmcs12 to do any
9176 * optimizations here, it may have been modified
9179 for (msr = 0x800; msr <= 0x8ff; msr++)
9180 __vmx_enable_intercept_for_msr(
9181 vmx_msr_bitmap_nested,
9185 __vmx_enable_intercept_for_msr(
9186 vmx_msr_bitmap_nested,
9187 APIC_BASE_MSR + (APIC_TASKPRI >> 4),
9189 __vmx_enable_intercept_for_msr(
9190 vmx_msr_bitmap_nested,
9191 APIC_BASE_MSR + (APIC_EOI >> 4),
9193 __vmx_enable_intercept_for_msr(
9194 vmx_msr_bitmap_nested,
9195 APIC_BASE_MSR + (APIC_SELF_IPI >> 4),
9199 nested_release_page_clean(page);
9204 static int nested_vmx_check_apicv_controls(struct kvm_vcpu *vcpu,
9205 struct vmcs12 *vmcs12)
9207 if (!nested_cpu_has_virt_x2apic_mode(vmcs12) &&
9208 !nested_cpu_has_apic_reg_virt(vmcs12) &&
9209 !nested_cpu_has_vid(vmcs12) &&
9210 !nested_cpu_has_posted_intr(vmcs12))
9214 * If virtualize x2apic mode is enabled,
9215 * virtualize apic access must be disabled.
9217 if (nested_cpu_has_virt_x2apic_mode(vmcs12) &&
9218 nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
9222 * If virtual interrupt delivery is enabled,
9223 * we must exit on external interrupts.
9225 if (nested_cpu_has_vid(vmcs12) &&
9226 !nested_exit_on_intr(vcpu))
9230 * bits 15:8 should be zero in posted_intr_nv,
9231 * the descriptor address has been already checked
9232 * in nested_get_vmcs12_pages.
9234 if (nested_cpu_has_posted_intr(vmcs12) &&
9235 (!nested_cpu_has_vid(vmcs12) ||
9236 !nested_exit_intr_ack_set(vcpu) ||
9237 vmcs12->posted_intr_nv & 0xff00))
9240 /* tpr shadow is needed by all apicv features. */
9241 if (!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
9247 static int nested_vmx_check_msr_switch(struct kvm_vcpu *vcpu,
9248 unsigned long count_field,
9249 unsigned long addr_field)
9254 if (vmcs12_read_any(vcpu, count_field, &count) ||
9255 vmcs12_read_any(vcpu, addr_field, &addr)) {
9261 maxphyaddr = cpuid_maxphyaddr(vcpu);
9262 if (!IS_ALIGNED(addr, 16) || addr >> maxphyaddr ||
9263 (addr + count * sizeof(struct vmx_msr_entry) - 1) >> maxphyaddr) {
9264 pr_warn_ratelimited(
9265 "nVMX: invalid MSR switch (0x%lx, %d, %llu, 0x%08llx)",
9266 addr_field, maxphyaddr, count, addr);
9272 static int nested_vmx_check_msr_switch_controls(struct kvm_vcpu *vcpu,
9273 struct vmcs12 *vmcs12)
9275 if (vmcs12->vm_exit_msr_load_count == 0 &&
9276 vmcs12->vm_exit_msr_store_count == 0 &&
9277 vmcs12->vm_entry_msr_load_count == 0)
9278 return 0; /* Fast path */
9279 if (nested_vmx_check_msr_switch(vcpu, VM_EXIT_MSR_LOAD_COUNT,
9280 VM_EXIT_MSR_LOAD_ADDR) ||
9281 nested_vmx_check_msr_switch(vcpu, VM_EXIT_MSR_STORE_COUNT,
9282 VM_EXIT_MSR_STORE_ADDR) ||
9283 nested_vmx_check_msr_switch(vcpu, VM_ENTRY_MSR_LOAD_COUNT,
9284 VM_ENTRY_MSR_LOAD_ADDR))
9289 static int nested_vmx_msr_check_common(struct kvm_vcpu *vcpu,
9290 struct vmx_msr_entry *e)
9292 /* x2APIC MSR accesses are not allowed */
9293 if (vcpu->arch.apic_base & X2APIC_ENABLE && e->index >> 8 == 0x8)
9295 if (e->index == MSR_IA32_UCODE_WRITE || /* SDM Table 35-2 */
9296 e->index == MSR_IA32_UCODE_REV)
9298 if (e->reserved != 0)
9303 static int nested_vmx_load_msr_check(struct kvm_vcpu *vcpu,
9304 struct vmx_msr_entry *e)
9306 if (e->index == MSR_FS_BASE ||
9307 e->index == MSR_GS_BASE ||
9308 e->index == MSR_IA32_SMM_MONITOR_CTL || /* SMM is not supported */
9309 nested_vmx_msr_check_common(vcpu, e))
9314 static int nested_vmx_store_msr_check(struct kvm_vcpu *vcpu,
9315 struct vmx_msr_entry *e)
9317 if (e->index == MSR_IA32_SMBASE || /* SMM is not supported */
9318 nested_vmx_msr_check_common(vcpu, e))
9324 * Load guest's/host's msr at nested entry/exit.
9325 * return 0 for success, entry index for failure.
9327 static u32 nested_vmx_load_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
9330 struct vmx_msr_entry e;
9331 struct msr_data msr;
9333 msr.host_initiated = false;
9334 for (i = 0; i < count; i++) {
9335 if (kvm_vcpu_read_guest(vcpu, gpa + i * sizeof(e),
9337 pr_warn_ratelimited(
9338 "%s cannot read MSR entry (%u, 0x%08llx)\n",
9339 __func__, i, gpa + i * sizeof(e));
9342 if (nested_vmx_load_msr_check(vcpu, &e)) {
9343 pr_warn_ratelimited(
9344 "%s check failed (%u, 0x%x, 0x%x)\n",
9345 __func__, i, e.index, e.reserved);
9348 msr.index = e.index;
9350 if (kvm_set_msr(vcpu, &msr)) {
9351 pr_warn_ratelimited(
9352 "%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
9353 __func__, i, e.index, e.value);
9362 static int nested_vmx_store_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
9365 struct vmx_msr_entry e;
9367 for (i = 0; i < count; i++) {
9368 struct msr_data msr_info;
9369 if (kvm_vcpu_read_guest(vcpu,
9370 gpa + i * sizeof(e),
9371 &e, 2 * sizeof(u32))) {
9372 pr_warn_ratelimited(
9373 "%s cannot read MSR entry (%u, 0x%08llx)\n",
9374 __func__, i, gpa + i * sizeof(e));
9377 if (nested_vmx_store_msr_check(vcpu, &e)) {
9378 pr_warn_ratelimited(
9379 "%s check failed (%u, 0x%x, 0x%x)\n",
9380 __func__, i, e.index, e.reserved);
9383 msr_info.host_initiated = false;
9384 msr_info.index = e.index;
9385 if (kvm_get_msr(vcpu, &msr_info)) {
9386 pr_warn_ratelimited(
9387 "%s cannot read MSR (%u, 0x%x)\n",
9388 __func__, i, e.index);
9391 if (kvm_vcpu_write_guest(vcpu,
9392 gpa + i * sizeof(e) +
9393 offsetof(struct vmx_msr_entry, value),
9394 &msr_info.data, sizeof(msr_info.data))) {
9395 pr_warn_ratelimited(
9396 "%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
9397 __func__, i, e.index, msr_info.data);
9405 * prepare_vmcs02 is called when the L1 guest hypervisor runs its nested
9406 * L2 guest. L1 has a vmcs for L2 (vmcs12), and this function "merges" it
9407 * with L0's requirements for its guest (a.k.a. vmcs01), so we can run the L2
9408 * guest in a way that will both be appropriate to L1's requests, and our
9409 * needs. In addition to modifying the active vmcs (which is vmcs02), this
9410 * function also has additional necessary side-effects, like setting various
9411 * vcpu->arch fields.
9413 static void prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
9415 struct vcpu_vmx *vmx = to_vmx(vcpu);
9418 vmcs_write16(GUEST_ES_SELECTOR, vmcs12->guest_es_selector);
9419 vmcs_write16(GUEST_CS_SELECTOR, vmcs12->guest_cs_selector);
9420 vmcs_write16(GUEST_SS_SELECTOR, vmcs12->guest_ss_selector);
9421 vmcs_write16(GUEST_DS_SELECTOR, vmcs12->guest_ds_selector);
9422 vmcs_write16(GUEST_FS_SELECTOR, vmcs12->guest_fs_selector);
9423 vmcs_write16(GUEST_GS_SELECTOR, vmcs12->guest_gs_selector);
9424 vmcs_write16(GUEST_LDTR_SELECTOR, vmcs12->guest_ldtr_selector);
9425 vmcs_write16(GUEST_TR_SELECTOR, vmcs12->guest_tr_selector);
9426 vmcs_write32(GUEST_ES_LIMIT, vmcs12->guest_es_limit);
9427 vmcs_write32(GUEST_CS_LIMIT, vmcs12->guest_cs_limit);
9428 vmcs_write32(GUEST_SS_LIMIT, vmcs12->guest_ss_limit);
9429 vmcs_write32(GUEST_DS_LIMIT, vmcs12->guest_ds_limit);
9430 vmcs_write32(GUEST_FS_LIMIT, vmcs12->guest_fs_limit);
9431 vmcs_write32(GUEST_GS_LIMIT, vmcs12->guest_gs_limit);
9432 vmcs_write32(GUEST_LDTR_LIMIT, vmcs12->guest_ldtr_limit);
9433 vmcs_write32(GUEST_TR_LIMIT, vmcs12->guest_tr_limit);
9434 vmcs_write32(GUEST_GDTR_LIMIT, vmcs12->guest_gdtr_limit);
9435 vmcs_write32(GUEST_IDTR_LIMIT, vmcs12->guest_idtr_limit);
9436 vmcs_write32(GUEST_ES_AR_BYTES, vmcs12->guest_es_ar_bytes);
9437 vmcs_write32(GUEST_CS_AR_BYTES, vmcs12->guest_cs_ar_bytes);
9438 vmcs_write32(GUEST_SS_AR_BYTES, vmcs12->guest_ss_ar_bytes);
9439 vmcs_write32(GUEST_DS_AR_BYTES, vmcs12->guest_ds_ar_bytes);
9440 vmcs_write32(GUEST_FS_AR_BYTES, vmcs12->guest_fs_ar_bytes);
9441 vmcs_write32(GUEST_GS_AR_BYTES, vmcs12->guest_gs_ar_bytes);
9442 vmcs_write32(GUEST_LDTR_AR_BYTES, vmcs12->guest_ldtr_ar_bytes);
9443 vmcs_write32(GUEST_TR_AR_BYTES, vmcs12->guest_tr_ar_bytes);
9444 vmcs_writel(GUEST_ES_BASE, vmcs12->guest_es_base);
9445 vmcs_writel(GUEST_CS_BASE, vmcs12->guest_cs_base);
9446 vmcs_writel(GUEST_SS_BASE, vmcs12->guest_ss_base);
9447 vmcs_writel(GUEST_DS_BASE, vmcs12->guest_ds_base);
9448 vmcs_writel(GUEST_FS_BASE, vmcs12->guest_fs_base);
9449 vmcs_writel(GUEST_GS_BASE, vmcs12->guest_gs_base);
9450 vmcs_writel(GUEST_LDTR_BASE, vmcs12->guest_ldtr_base);
9451 vmcs_writel(GUEST_TR_BASE, vmcs12->guest_tr_base);
9452 vmcs_writel(GUEST_GDTR_BASE, vmcs12->guest_gdtr_base);
9453 vmcs_writel(GUEST_IDTR_BASE, vmcs12->guest_idtr_base);
9455 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS) {
9456 kvm_set_dr(vcpu, 7, vmcs12->guest_dr7);
9457 vmcs_write64(GUEST_IA32_DEBUGCTL, vmcs12->guest_ia32_debugctl);
9459 kvm_set_dr(vcpu, 7, vcpu->arch.dr7);
9460 vmcs_write64(GUEST_IA32_DEBUGCTL, vmx->nested.vmcs01_debugctl);
9462 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
9463 vmcs12->vm_entry_intr_info_field);
9464 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
9465 vmcs12->vm_entry_exception_error_code);
9466 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
9467 vmcs12->vm_entry_instruction_len);
9468 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
9469 vmcs12->guest_interruptibility_info);
9470 vmcs_write32(GUEST_SYSENTER_CS, vmcs12->guest_sysenter_cs);
9471 vmx_set_rflags(vcpu, vmcs12->guest_rflags);
9472 vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
9473 vmcs12->guest_pending_dbg_exceptions);
9474 vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->guest_sysenter_esp);
9475 vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->guest_sysenter_eip);
9477 if (nested_cpu_has_xsaves(vmcs12))
9478 vmcs_write64(XSS_EXIT_BITMAP, vmcs12->xss_exit_bitmap);
9479 vmcs_write64(VMCS_LINK_POINTER, -1ull);
9481 exec_control = vmcs12->pin_based_vm_exec_control;
9482 exec_control |= vmcs_config.pin_based_exec_ctrl;
9483 exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
9485 if (nested_cpu_has_posted_intr(vmcs12)) {
9487 * Note that we use L0's vector here and in
9488 * vmx_deliver_nested_posted_interrupt.
9490 vmx->nested.posted_intr_nv = vmcs12->posted_intr_nv;
9491 vmx->nested.pi_pending = false;
9492 vmcs_write64(POSTED_INTR_NV, POSTED_INTR_VECTOR);
9493 vmcs_write64(POSTED_INTR_DESC_ADDR,
9494 page_to_phys(vmx->nested.pi_desc_page) +
9495 (unsigned long)(vmcs12->posted_intr_desc_addr &
9498 exec_control &= ~PIN_BASED_POSTED_INTR;
9500 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, exec_control);
9502 vmx->nested.preemption_timer_expired = false;
9503 if (nested_cpu_has_preemption_timer(vmcs12))
9504 vmx_start_preemption_timer(vcpu);
9507 * Whether page-faults are trapped is determined by a combination of
9508 * 3 settings: PFEC_MASK, PFEC_MATCH and EXCEPTION_BITMAP.PF.
9509 * If enable_ept, L0 doesn't care about page faults and we should
9510 * set all of these to L1's desires. However, if !enable_ept, L0 does
9511 * care about (at least some) page faults, and because it is not easy
9512 * (if at all possible?) to merge L0 and L1's desires, we simply ask
9513 * to exit on each and every L2 page fault. This is done by setting
9514 * MASK=MATCH=0 and (see below) EB.PF=1.
9515 * Note that below we don't need special code to set EB.PF beyond the
9516 * "or"ing of the EB of vmcs01 and vmcs12, because when enable_ept,
9517 * vmcs01's EB.PF is 0 so the "or" will take vmcs12's value, and when
9518 * !enable_ept, EB.PF is 1, so the "or" will always be 1.
9520 * A problem with this approach (when !enable_ept) is that L1 may be
9521 * injected with more page faults than it asked for. This could have
9522 * caused problems, but in practice existing hypervisors don't care.
9523 * To fix this, we will need to emulate the PFEC checking (on the L1
9524 * page tables), using walk_addr(), when injecting PFs to L1.
9526 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK,
9527 enable_ept ? vmcs12->page_fault_error_code_mask : 0);
9528 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH,
9529 enable_ept ? vmcs12->page_fault_error_code_match : 0);
9531 if (cpu_has_secondary_exec_ctrls()) {
9532 exec_control = vmx_secondary_exec_control(vmx);
9534 /* Take the following fields only from vmcs12 */
9535 exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
9536 SECONDARY_EXEC_RDTSCP |
9537 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
9538 SECONDARY_EXEC_APIC_REGISTER_VIRT |
9539 SECONDARY_EXEC_PCOMMIT);
9540 if (nested_cpu_has(vmcs12,
9541 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS))
9542 exec_control |= vmcs12->secondary_vm_exec_control;
9544 if (exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES) {
9546 * If translation failed, no matter: This feature asks
9547 * to exit when accessing the given address, and if it
9548 * can never be accessed, this feature won't do
9551 if (!vmx->nested.apic_access_page)
9553 ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
9555 vmcs_write64(APIC_ACCESS_ADDR,
9556 page_to_phys(vmx->nested.apic_access_page));
9557 } else if (!(nested_cpu_has_virt_x2apic_mode(vmcs12)) &&
9558 cpu_need_virtualize_apic_accesses(&vmx->vcpu)) {
9560 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
9561 kvm_vcpu_reload_apic_access_page(vcpu);
9564 if (exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) {
9565 vmcs_write64(EOI_EXIT_BITMAP0,
9566 vmcs12->eoi_exit_bitmap0);
9567 vmcs_write64(EOI_EXIT_BITMAP1,
9568 vmcs12->eoi_exit_bitmap1);
9569 vmcs_write64(EOI_EXIT_BITMAP2,
9570 vmcs12->eoi_exit_bitmap2);
9571 vmcs_write64(EOI_EXIT_BITMAP3,
9572 vmcs12->eoi_exit_bitmap3);
9573 vmcs_write16(GUEST_INTR_STATUS,
9574 vmcs12->guest_intr_status);
9577 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
9582 * Set host-state according to L0's settings (vmcs12 is irrelevant here)
9583 * Some constant fields are set here by vmx_set_constant_host_state().
9584 * Other fields are different per CPU, and will be set later when
9585 * vmx_vcpu_load() is called, and when vmx_save_host_state() is called.
9587 vmx_set_constant_host_state(vmx);
9590 * HOST_RSP is normally set correctly in vmx_vcpu_run() just before
9591 * entry, but only if the current (host) sp changed from the value
9592 * we wrote last (vmx->host_rsp). This cache is no longer relevant
9593 * if we switch vmcs, and rather than hold a separate cache per vmcs,
9594 * here we just force the write to happen on entry.
9598 exec_control = vmx_exec_control(vmx); /* L0's desires */
9599 exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
9600 exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
9601 exec_control &= ~CPU_BASED_TPR_SHADOW;
9602 exec_control |= vmcs12->cpu_based_vm_exec_control;
9604 if (exec_control & CPU_BASED_TPR_SHADOW) {
9605 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
9606 page_to_phys(vmx->nested.virtual_apic_page));
9607 vmcs_write32(TPR_THRESHOLD, vmcs12->tpr_threshold);
9610 if (cpu_has_vmx_msr_bitmap() &&
9611 exec_control & CPU_BASED_USE_MSR_BITMAPS) {
9612 nested_vmx_merge_msr_bitmap(vcpu, vmcs12);
9613 /* MSR_BITMAP will be set by following vmx_set_efer. */
9615 exec_control &= ~CPU_BASED_USE_MSR_BITMAPS;
9618 * Merging of IO bitmap not currently supported.
9619 * Rather, exit every time.
9621 exec_control &= ~CPU_BASED_USE_IO_BITMAPS;
9622 exec_control |= CPU_BASED_UNCOND_IO_EXITING;
9624 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
9626 /* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the
9627 * bitwise-or of what L1 wants to trap for L2, and what we want to
9628 * trap. Note that CR0.TS also needs updating - we do this later.
9630 update_exception_bitmap(vcpu);
9631 vcpu->arch.cr0_guest_owned_bits &= ~vmcs12->cr0_guest_host_mask;
9632 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
9634 /* L2->L1 exit controls are emulated - the hardware exit is to L0 so
9635 * we should use its exit controls. Note that VM_EXIT_LOAD_IA32_EFER
9636 * bits are further modified by vmx_set_efer() below.
9638 vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
9640 /* vmcs12's VM_ENTRY_LOAD_IA32_EFER and VM_ENTRY_IA32E_MODE are
9641 * emulated by vmx_set_efer(), below.
9643 vm_entry_controls_init(vmx,
9644 (vmcs12->vm_entry_controls & ~VM_ENTRY_LOAD_IA32_EFER &
9645 ~VM_ENTRY_IA32E_MODE) |
9646 (vmcs_config.vmentry_ctrl & ~VM_ENTRY_IA32E_MODE));
9648 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT) {
9649 vmcs_write64(GUEST_IA32_PAT, vmcs12->guest_ia32_pat);
9650 vcpu->arch.pat = vmcs12->guest_ia32_pat;
9651 } else if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
9652 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
9655 set_cr4_guest_host_mask(vmx);
9657 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS)
9658 vmcs_write64(GUEST_BNDCFGS, vmcs12->guest_bndcfgs);
9660 if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING)
9661 vmcs_write64(TSC_OFFSET,
9662 vmx->nested.vmcs01_tsc_offset + vmcs12->tsc_offset);
9664 vmcs_write64(TSC_OFFSET, vmx->nested.vmcs01_tsc_offset);
9668 * Trivially support vpid by letting L2s share their parent
9669 * L1's vpid. TODO: move to a more elaborate solution, giving
9670 * each L2 its own vpid and exposing the vpid feature to L1.
9672 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
9673 vmx_flush_tlb(vcpu);
9676 if (nested_cpu_has_ept(vmcs12)) {
9677 kvm_mmu_unload(vcpu);
9678 nested_ept_init_mmu_context(vcpu);
9681 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER)
9682 vcpu->arch.efer = vmcs12->guest_ia32_efer;
9683 else if (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE)
9684 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
9686 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
9687 /* Note: modifies VM_ENTRY/EXIT_CONTROLS and GUEST/HOST_IA32_EFER */
9688 vmx_set_efer(vcpu, vcpu->arch.efer);
9691 * This sets GUEST_CR0 to vmcs12->guest_cr0, with possibly a modified
9692 * TS bit (for lazy fpu) and bits which we consider mandatory enabled.
9693 * The CR0_READ_SHADOW is what L2 should have expected to read given
9694 * the specifications by L1; It's not enough to take
9695 * vmcs12->cr0_read_shadow because on our cr0_guest_host_mask we we
9696 * have more bits than L1 expected.
9698 vmx_set_cr0(vcpu, vmcs12->guest_cr0);
9699 vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
9701 vmx_set_cr4(vcpu, vmcs12->guest_cr4);
9702 vmcs_writel(CR4_READ_SHADOW, nested_read_cr4(vmcs12));
9704 /* shadow page tables on either EPT or shadow page tables */
9705 kvm_set_cr3(vcpu, vmcs12->guest_cr3);
9706 kvm_mmu_reset_context(vcpu);
9709 vcpu->arch.walk_mmu->inject_page_fault = vmx_inject_page_fault_nested;
9712 * L1 may access the L2's PDPTR, so save them to construct vmcs12
9715 vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
9716 vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
9717 vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
9718 vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
9721 kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->guest_rsp);
9722 kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->guest_rip);
9726 * nested_vmx_run() handles a nested entry, i.e., a VMLAUNCH or VMRESUME on L1
9727 * for running an L2 nested guest.
9729 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
9731 struct vmcs12 *vmcs12;
9732 struct vcpu_vmx *vmx = to_vmx(vcpu);
9734 struct loaded_vmcs *vmcs02;
9738 if (!nested_vmx_check_permission(vcpu) ||
9739 !nested_vmx_check_vmcs12(vcpu))
9742 skip_emulated_instruction(vcpu);
9743 vmcs12 = get_vmcs12(vcpu);
9745 if (enable_shadow_vmcs)
9746 copy_shadow_to_vmcs12(vmx);
9749 * The nested entry process starts with enforcing various prerequisites
9750 * on vmcs12 as required by the Intel SDM, and act appropriately when
9751 * they fail: As the SDM explains, some conditions should cause the
9752 * instruction to fail, while others will cause the instruction to seem
9753 * to succeed, but return an EXIT_REASON_INVALID_STATE.
9754 * To speed up the normal (success) code path, we should avoid checking
9755 * for misconfigurations which will anyway be caught by the processor
9756 * when using the merged vmcs02.
9758 if (vmcs12->launch_state == launch) {
9759 nested_vmx_failValid(vcpu,
9760 launch ? VMXERR_VMLAUNCH_NONCLEAR_VMCS
9761 : VMXERR_VMRESUME_NONLAUNCHED_VMCS);
9765 if (vmcs12->guest_activity_state != GUEST_ACTIVITY_ACTIVE &&
9766 vmcs12->guest_activity_state != GUEST_ACTIVITY_HLT) {
9767 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
9771 if (!nested_get_vmcs12_pages(vcpu, vmcs12)) {
9772 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
9776 if (nested_vmx_check_msr_bitmap_controls(vcpu, vmcs12)) {
9777 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
9781 if (nested_vmx_check_apicv_controls(vcpu, vmcs12)) {
9782 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
9786 if (nested_vmx_check_msr_switch_controls(vcpu, vmcs12)) {
9787 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
9791 if (!vmx_control_verify(vmcs12->cpu_based_vm_exec_control,
9792 vmx->nested.nested_vmx_true_procbased_ctls_low,
9793 vmx->nested.nested_vmx_procbased_ctls_high) ||
9794 !vmx_control_verify(vmcs12->secondary_vm_exec_control,
9795 vmx->nested.nested_vmx_secondary_ctls_low,
9796 vmx->nested.nested_vmx_secondary_ctls_high) ||
9797 !vmx_control_verify(vmcs12->pin_based_vm_exec_control,
9798 vmx->nested.nested_vmx_pinbased_ctls_low,
9799 vmx->nested.nested_vmx_pinbased_ctls_high) ||
9800 !vmx_control_verify(vmcs12->vm_exit_controls,
9801 vmx->nested.nested_vmx_true_exit_ctls_low,
9802 vmx->nested.nested_vmx_exit_ctls_high) ||
9803 !vmx_control_verify(vmcs12->vm_entry_controls,
9804 vmx->nested.nested_vmx_true_entry_ctls_low,
9805 vmx->nested.nested_vmx_entry_ctls_high))
9807 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
9811 if (((vmcs12->host_cr0 & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON) ||
9812 ((vmcs12->host_cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON)) {
9813 nested_vmx_failValid(vcpu,
9814 VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
9818 if (!nested_cr0_valid(vcpu, vmcs12->guest_cr0) ||
9819 ((vmcs12->guest_cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON)) {
9820 nested_vmx_entry_failure(vcpu, vmcs12,
9821 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
9824 if (vmcs12->vmcs_link_pointer != -1ull) {
9825 nested_vmx_entry_failure(vcpu, vmcs12,
9826 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_VMCS_LINK_PTR);
9831 * If the load IA32_EFER VM-entry control is 1, the following checks
9832 * are performed on the field for the IA32_EFER MSR:
9833 * - Bits reserved in the IA32_EFER MSR must be 0.
9834 * - Bit 10 (corresponding to IA32_EFER.LMA) must equal the value of
9835 * the IA-32e mode guest VM-exit control. It must also be identical
9836 * to bit 8 (LME) if bit 31 in the CR0 field (corresponding to
9839 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER) {
9840 ia32e = (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) != 0;
9841 if (!kvm_valid_efer(vcpu, vmcs12->guest_ia32_efer) ||
9842 ia32e != !!(vmcs12->guest_ia32_efer & EFER_LMA) ||
9843 ((vmcs12->guest_cr0 & X86_CR0_PG) &&
9844 ia32e != !!(vmcs12->guest_ia32_efer & EFER_LME))) {
9845 nested_vmx_entry_failure(vcpu, vmcs12,
9846 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
9852 * If the load IA32_EFER VM-exit control is 1, bits reserved in the
9853 * IA32_EFER MSR must be 0 in the field for that register. In addition,
9854 * the values of the LMA and LME bits in the field must each be that of
9855 * the host address-space size VM-exit control.
9857 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER) {
9858 ia32e = (vmcs12->vm_exit_controls &
9859 VM_EXIT_HOST_ADDR_SPACE_SIZE) != 0;
9860 if (!kvm_valid_efer(vcpu, vmcs12->host_ia32_efer) ||
9861 ia32e != !!(vmcs12->host_ia32_efer & EFER_LMA) ||
9862 ia32e != !!(vmcs12->host_ia32_efer & EFER_LME)) {
9863 nested_vmx_entry_failure(vcpu, vmcs12,
9864 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
9870 * We're finally done with prerequisite checking, and can start with
9874 vmcs02 = nested_get_current_vmcs02(vmx);
9878 enter_guest_mode(vcpu);
9880 vmx->nested.vmcs01_tsc_offset = vmcs_read64(TSC_OFFSET);
9882 if (!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS))
9883 vmx->nested.vmcs01_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
9886 vmx->loaded_vmcs = vmcs02;
9888 vmx_vcpu_load(vcpu, cpu);
9892 vmx_segment_cache_clear(vmx);
9894 prepare_vmcs02(vcpu, vmcs12);
9896 msr_entry_idx = nested_vmx_load_msr(vcpu,
9897 vmcs12->vm_entry_msr_load_addr,
9898 vmcs12->vm_entry_msr_load_count);
9899 if (msr_entry_idx) {
9900 leave_guest_mode(vcpu);
9901 vmx_load_vmcs01(vcpu);
9902 nested_vmx_entry_failure(vcpu, vmcs12,
9903 EXIT_REASON_MSR_LOAD_FAIL, msr_entry_idx);
9907 vmcs12->launch_state = 1;
9909 if (vmcs12->guest_activity_state == GUEST_ACTIVITY_HLT)
9910 return kvm_vcpu_halt(vcpu);
9912 vmx->nested.nested_run_pending = 1;
9915 * Note no nested_vmx_succeed or nested_vmx_fail here. At this point
9916 * we are no longer running L1, and VMLAUNCH/VMRESUME has not yet
9917 * returned as far as L1 is concerned. It will only return (and set
9918 * the success flag) when L2 exits (see nested_vmx_vmexit()).
9924 * On a nested exit from L2 to L1, vmcs12.guest_cr0 might not be up-to-date
9925 * because L2 may have changed some cr0 bits directly (CRO_GUEST_HOST_MASK).
9926 * This function returns the new value we should put in vmcs12.guest_cr0.
9927 * It's not enough to just return the vmcs02 GUEST_CR0. Rather,
9928 * 1. Bits that neither L0 nor L1 trapped, were set directly by L2 and are now
9929 * available in vmcs02 GUEST_CR0. (Note: It's enough to check that L0
9930 * didn't trap the bit, because if L1 did, so would L0).
9931 * 2. Bits that L1 asked to trap (and therefore L0 also did) could not have
9932 * been modified by L2, and L1 knows it. So just leave the old value of
9933 * the bit from vmcs12.guest_cr0. Note that the bit from vmcs02 GUEST_CR0
9934 * isn't relevant, because if L0 traps this bit it can set it to anything.
9935 * 3. Bits that L1 didn't trap, but L0 did. L1 believes the guest could have
9936 * changed these bits, and therefore they need to be updated, but L0
9937 * didn't necessarily allow them to be changed in GUEST_CR0 - and rather
9938 * put them in vmcs02 CR0_READ_SHADOW. So take these bits from there.
9940 static inline unsigned long
9941 vmcs12_guest_cr0(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
9944 /*1*/ (vmcs_readl(GUEST_CR0) & vcpu->arch.cr0_guest_owned_bits) |
9945 /*2*/ (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask) |
9946 /*3*/ (vmcs_readl(CR0_READ_SHADOW) & ~(vmcs12->cr0_guest_host_mask |
9947 vcpu->arch.cr0_guest_owned_bits));
9950 static inline unsigned long
9951 vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
9954 /*1*/ (vmcs_readl(GUEST_CR4) & vcpu->arch.cr4_guest_owned_bits) |
9955 /*2*/ (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask) |
9956 /*3*/ (vmcs_readl(CR4_READ_SHADOW) & ~(vmcs12->cr4_guest_host_mask |
9957 vcpu->arch.cr4_guest_owned_bits));
9960 static void vmcs12_save_pending_event(struct kvm_vcpu *vcpu,
9961 struct vmcs12 *vmcs12)
9966 if (vcpu->arch.exception.pending && vcpu->arch.exception.reinject) {
9967 nr = vcpu->arch.exception.nr;
9968 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
9970 if (kvm_exception_is_soft(nr)) {
9971 vmcs12->vm_exit_instruction_len =
9972 vcpu->arch.event_exit_inst_len;
9973 idt_vectoring |= INTR_TYPE_SOFT_EXCEPTION;
9975 idt_vectoring |= INTR_TYPE_HARD_EXCEPTION;
9977 if (vcpu->arch.exception.has_error_code) {
9978 idt_vectoring |= VECTORING_INFO_DELIVER_CODE_MASK;
9979 vmcs12->idt_vectoring_error_code =
9980 vcpu->arch.exception.error_code;
9983 vmcs12->idt_vectoring_info_field = idt_vectoring;
9984 } else if (vcpu->arch.nmi_injected) {
9985 vmcs12->idt_vectoring_info_field =
9986 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR;
9987 } else if (vcpu->arch.interrupt.pending) {
9988 nr = vcpu->arch.interrupt.nr;
9989 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
9991 if (vcpu->arch.interrupt.soft) {
9992 idt_vectoring |= INTR_TYPE_SOFT_INTR;
9993 vmcs12->vm_entry_instruction_len =
9994 vcpu->arch.event_exit_inst_len;
9996 idt_vectoring |= INTR_TYPE_EXT_INTR;
9998 vmcs12->idt_vectoring_info_field = idt_vectoring;
10002 static int vmx_check_nested_events(struct kvm_vcpu *vcpu, bool external_intr)
10004 struct vcpu_vmx *vmx = to_vmx(vcpu);
10006 if (nested_cpu_has_preemption_timer(get_vmcs12(vcpu)) &&
10007 vmx->nested.preemption_timer_expired) {
10008 if (vmx->nested.nested_run_pending)
10010 nested_vmx_vmexit(vcpu, EXIT_REASON_PREEMPTION_TIMER, 0, 0);
10014 if (vcpu->arch.nmi_pending && nested_exit_on_nmi(vcpu)) {
10015 if (vmx->nested.nested_run_pending ||
10016 vcpu->arch.interrupt.pending)
10018 nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
10019 NMI_VECTOR | INTR_TYPE_NMI_INTR |
10020 INTR_INFO_VALID_MASK, 0);
10022 * The NMI-triggered VM exit counts as injection:
10023 * clear this one and block further NMIs.
10025 vcpu->arch.nmi_pending = 0;
10026 vmx_set_nmi_mask(vcpu, true);
10030 if ((kvm_cpu_has_interrupt(vcpu) || external_intr) &&
10031 nested_exit_on_intr(vcpu)) {
10032 if (vmx->nested.nested_run_pending)
10034 nested_vmx_vmexit(vcpu, EXIT_REASON_EXTERNAL_INTERRUPT, 0, 0);
10038 return vmx_complete_nested_posted_interrupt(vcpu);
10041 static u32 vmx_get_preemption_timer_value(struct kvm_vcpu *vcpu)
10043 ktime_t remaining =
10044 hrtimer_get_remaining(&to_vmx(vcpu)->nested.preemption_timer);
10047 if (ktime_to_ns(remaining) <= 0)
10050 value = ktime_to_ns(remaining) * vcpu->arch.virtual_tsc_khz;
10051 do_div(value, 1000000);
10052 return value >> VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
10056 * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
10057 * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),
10058 * and this function updates it to reflect the changes to the guest state while
10059 * L2 was running (and perhaps made some exits which were handled directly by L0
10060 * without going back to L1), and to reflect the exit reason.
10061 * Note that we do not have to copy here all VMCS fields, just those that
10062 * could have changed by the L2 guest or the exit - i.e., the guest-state and
10063 * exit-information fields only. Other fields are modified by L1 with VMWRITE,
10064 * which already writes to vmcs12 directly.
10066 static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
10067 u32 exit_reason, u32 exit_intr_info,
10068 unsigned long exit_qualification)
10070 /* update guest state fields: */
10071 vmcs12->guest_cr0 = vmcs12_guest_cr0(vcpu, vmcs12);
10072 vmcs12->guest_cr4 = vmcs12_guest_cr4(vcpu, vmcs12);
10074 vmcs12->guest_rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
10075 vmcs12->guest_rip = kvm_register_read(vcpu, VCPU_REGS_RIP);
10076 vmcs12->guest_rflags = vmcs_readl(GUEST_RFLAGS);
10078 vmcs12->guest_es_selector = vmcs_read16(GUEST_ES_SELECTOR);
10079 vmcs12->guest_cs_selector = vmcs_read16(GUEST_CS_SELECTOR);
10080 vmcs12->guest_ss_selector = vmcs_read16(GUEST_SS_SELECTOR);
10081 vmcs12->guest_ds_selector = vmcs_read16(GUEST_DS_SELECTOR);
10082 vmcs12->guest_fs_selector = vmcs_read16(GUEST_FS_SELECTOR);
10083 vmcs12->guest_gs_selector = vmcs_read16(GUEST_GS_SELECTOR);
10084 vmcs12->guest_ldtr_selector = vmcs_read16(GUEST_LDTR_SELECTOR);
10085 vmcs12->guest_tr_selector = vmcs_read16(GUEST_TR_SELECTOR);
10086 vmcs12->guest_es_limit = vmcs_read32(GUEST_ES_LIMIT);
10087 vmcs12->guest_cs_limit = vmcs_read32(GUEST_CS_LIMIT);
10088 vmcs12->guest_ss_limit = vmcs_read32(GUEST_SS_LIMIT);
10089 vmcs12->guest_ds_limit = vmcs_read32(GUEST_DS_LIMIT);
10090 vmcs12->guest_fs_limit = vmcs_read32(GUEST_FS_LIMIT);
10091 vmcs12->guest_gs_limit = vmcs_read32(GUEST_GS_LIMIT);
10092 vmcs12->guest_ldtr_limit = vmcs_read32(GUEST_LDTR_LIMIT);
10093 vmcs12->guest_tr_limit = vmcs_read32(GUEST_TR_LIMIT);
10094 vmcs12->guest_gdtr_limit = vmcs_read32(GUEST_GDTR_LIMIT);
10095 vmcs12->guest_idtr_limit = vmcs_read32(GUEST_IDTR_LIMIT);
10096 vmcs12->guest_es_ar_bytes = vmcs_read32(GUEST_ES_AR_BYTES);
10097 vmcs12->guest_cs_ar_bytes = vmcs_read32(GUEST_CS_AR_BYTES);
10098 vmcs12->guest_ss_ar_bytes = vmcs_read32(GUEST_SS_AR_BYTES);
10099 vmcs12->guest_ds_ar_bytes = vmcs_read32(GUEST_DS_AR_BYTES);
10100 vmcs12->guest_fs_ar_bytes = vmcs_read32(GUEST_FS_AR_BYTES);
10101 vmcs12->guest_gs_ar_bytes = vmcs_read32(GUEST_GS_AR_BYTES);
10102 vmcs12->guest_ldtr_ar_bytes = vmcs_read32(GUEST_LDTR_AR_BYTES);
10103 vmcs12->guest_tr_ar_bytes = vmcs_read32(GUEST_TR_AR_BYTES);
10104 vmcs12->guest_es_base = vmcs_readl(GUEST_ES_BASE);
10105 vmcs12->guest_cs_base = vmcs_readl(GUEST_CS_BASE);
10106 vmcs12->guest_ss_base = vmcs_readl(GUEST_SS_BASE);
10107 vmcs12->guest_ds_base = vmcs_readl(GUEST_DS_BASE);
10108 vmcs12->guest_fs_base = vmcs_readl(GUEST_FS_BASE);
10109 vmcs12->guest_gs_base = vmcs_readl(GUEST_GS_BASE);
10110 vmcs12->guest_ldtr_base = vmcs_readl(GUEST_LDTR_BASE);
10111 vmcs12->guest_tr_base = vmcs_readl(GUEST_TR_BASE);
10112 vmcs12->guest_gdtr_base = vmcs_readl(GUEST_GDTR_BASE);
10113 vmcs12->guest_idtr_base = vmcs_readl(GUEST_IDTR_BASE);
10115 vmcs12->guest_interruptibility_info =
10116 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
10117 vmcs12->guest_pending_dbg_exceptions =
10118 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS);
10119 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
10120 vmcs12->guest_activity_state = GUEST_ACTIVITY_HLT;
10122 vmcs12->guest_activity_state = GUEST_ACTIVITY_ACTIVE;
10124 if (nested_cpu_has_preemption_timer(vmcs12)) {
10125 if (vmcs12->vm_exit_controls &
10126 VM_EXIT_SAVE_VMX_PREEMPTION_TIMER)
10127 vmcs12->vmx_preemption_timer_value =
10128 vmx_get_preemption_timer_value(vcpu);
10129 hrtimer_cancel(&to_vmx(vcpu)->nested.preemption_timer);
10133 * In some cases (usually, nested EPT), L2 is allowed to change its
10134 * own CR3 without exiting. If it has changed it, we must keep it.
10135 * Of course, if L0 is using shadow page tables, GUEST_CR3 was defined
10136 * by L0, not L1 or L2, so we mustn't unconditionally copy it to vmcs12.
10138 * Additionally, restore L2's PDPTR to vmcs12.
10141 vmcs12->guest_cr3 = vmcs_read64(GUEST_CR3);
10142 vmcs12->guest_pdptr0 = vmcs_read64(GUEST_PDPTR0);
10143 vmcs12->guest_pdptr1 = vmcs_read64(GUEST_PDPTR1);
10144 vmcs12->guest_pdptr2 = vmcs_read64(GUEST_PDPTR2);
10145 vmcs12->guest_pdptr3 = vmcs_read64(GUEST_PDPTR3);
10148 if (nested_cpu_has_vid(vmcs12))
10149 vmcs12->guest_intr_status = vmcs_read16(GUEST_INTR_STATUS);
10151 vmcs12->vm_entry_controls =
10152 (vmcs12->vm_entry_controls & ~VM_ENTRY_IA32E_MODE) |
10153 (vm_entry_controls_get(to_vmx(vcpu)) & VM_ENTRY_IA32E_MODE);
10155 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_DEBUG_CONTROLS) {
10156 kvm_get_dr(vcpu, 7, (unsigned long *)&vmcs12->guest_dr7);
10157 vmcs12->guest_ia32_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
10160 /* TODO: These cannot have changed unless we have MSR bitmaps and
10161 * the relevant bit asks not to trap the change */
10162 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_PAT)
10163 vmcs12->guest_ia32_pat = vmcs_read64(GUEST_IA32_PAT);
10164 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_EFER)
10165 vmcs12->guest_ia32_efer = vcpu->arch.efer;
10166 vmcs12->guest_sysenter_cs = vmcs_read32(GUEST_SYSENTER_CS);
10167 vmcs12->guest_sysenter_esp = vmcs_readl(GUEST_SYSENTER_ESP);
10168 vmcs12->guest_sysenter_eip = vmcs_readl(GUEST_SYSENTER_EIP);
10169 if (vmx_mpx_supported())
10170 vmcs12->guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
10171 if (nested_cpu_has_xsaves(vmcs12))
10172 vmcs12->xss_exit_bitmap = vmcs_read64(XSS_EXIT_BITMAP);
10174 /* update exit information fields: */
10176 vmcs12->vm_exit_reason = exit_reason;
10177 vmcs12->exit_qualification = exit_qualification;
10179 vmcs12->vm_exit_intr_info = exit_intr_info;
10180 if ((vmcs12->vm_exit_intr_info &
10181 (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK)) ==
10182 (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK))
10183 vmcs12->vm_exit_intr_error_code =
10184 vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
10185 vmcs12->idt_vectoring_info_field = 0;
10186 vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
10187 vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
10189 if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY)) {
10190 /* vm_entry_intr_info_field is cleared on exit. Emulate this
10191 * instead of reading the real value. */
10192 vmcs12->vm_entry_intr_info_field &= ~INTR_INFO_VALID_MASK;
10195 * Transfer the event that L0 or L1 may wanted to inject into
10196 * L2 to IDT_VECTORING_INFO_FIELD.
10198 vmcs12_save_pending_event(vcpu, vmcs12);
10202 * Drop what we picked up for L2 via vmx_complete_interrupts. It is
10203 * preserved above and would only end up incorrectly in L1.
10205 vcpu->arch.nmi_injected = false;
10206 kvm_clear_exception_queue(vcpu);
10207 kvm_clear_interrupt_queue(vcpu);
10211 * A part of what we need to when the nested L2 guest exits and we want to
10212 * run its L1 parent, is to reset L1's guest state to the host state specified
10214 * This function is to be called not only on normal nested exit, but also on
10215 * a nested entry failure, as explained in Intel's spec, 3B.23.7 ("VM-Entry
10216 * Failures During or After Loading Guest State").
10217 * This function should be called when the active VMCS is L1's (vmcs01).
10219 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
10220 struct vmcs12 *vmcs12)
10222 struct kvm_segment seg;
10224 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER)
10225 vcpu->arch.efer = vmcs12->host_ia32_efer;
10226 else if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
10227 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
10229 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
10230 vmx_set_efer(vcpu, vcpu->arch.efer);
10232 kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->host_rsp);
10233 kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->host_rip);
10234 vmx_set_rflags(vcpu, X86_EFLAGS_FIXED);
10236 * Note that calling vmx_set_cr0 is important, even if cr0 hasn't
10237 * actually changed, because it depends on the current state of
10238 * fpu_active (which may have changed).
10239 * Note that vmx_set_cr0 refers to efer set above.
10241 vmx_set_cr0(vcpu, vmcs12->host_cr0);
10243 * If we did fpu_activate()/fpu_deactivate() during L2's run, we need
10244 * to apply the same changes to L1's vmcs. We just set cr0 correctly,
10245 * but we also need to update cr0_guest_host_mask and exception_bitmap.
10247 update_exception_bitmap(vcpu);
10248 vcpu->arch.cr0_guest_owned_bits = (vcpu->fpu_active ? X86_CR0_TS : 0);
10249 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
10252 * Note that CR4_GUEST_HOST_MASK is already set in the original vmcs01
10253 * (KVM doesn't change it)- no reason to call set_cr4_guest_host_mask();
10255 vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
10256 kvm_set_cr4(vcpu, vmcs12->host_cr4);
10258 nested_ept_uninit_mmu_context(vcpu);
10260 kvm_set_cr3(vcpu, vmcs12->host_cr3);
10261 kvm_mmu_reset_context(vcpu);
10264 vcpu->arch.walk_mmu->inject_page_fault = kvm_inject_page_fault;
10268 * Trivially support vpid by letting L2s share their parent
10269 * L1's vpid. TODO: move to a more elaborate solution, giving
10270 * each L2 its own vpid and exposing the vpid feature to L1.
10272 vmx_flush_tlb(vcpu);
10276 vmcs_write32(GUEST_SYSENTER_CS, vmcs12->host_ia32_sysenter_cs);
10277 vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->host_ia32_sysenter_esp);
10278 vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->host_ia32_sysenter_eip);
10279 vmcs_writel(GUEST_IDTR_BASE, vmcs12->host_idtr_base);
10280 vmcs_writel(GUEST_GDTR_BASE, vmcs12->host_gdtr_base);
10282 /* If not VM_EXIT_CLEAR_BNDCFGS, the L2 value propagates to L1. */
10283 if (vmcs12->vm_exit_controls & VM_EXIT_CLEAR_BNDCFGS)
10284 vmcs_write64(GUEST_BNDCFGS, 0);
10286 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) {
10287 vmcs_write64(GUEST_IA32_PAT, vmcs12->host_ia32_pat);
10288 vcpu->arch.pat = vmcs12->host_ia32_pat;
10290 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
10291 vmcs_write64(GUEST_IA32_PERF_GLOBAL_CTRL,
10292 vmcs12->host_ia32_perf_global_ctrl);
10294 /* Set L1 segment info according to Intel SDM
10295 27.5.2 Loading Host Segment and Descriptor-Table Registers */
10296 seg = (struct kvm_segment) {
10298 .limit = 0xFFFFFFFF,
10299 .selector = vmcs12->host_cs_selector,
10305 if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
10309 vmx_set_segment(vcpu, &seg, VCPU_SREG_CS);
10310 seg = (struct kvm_segment) {
10312 .limit = 0xFFFFFFFF,
10319 seg.selector = vmcs12->host_ds_selector;
10320 vmx_set_segment(vcpu, &seg, VCPU_SREG_DS);
10321 seg.selector = vmcs12->host_es_selector;
10322 vmx_set_segment(vcpu, &seg, VCPU_SREG_ES);
10323 seg.selector = vmcs12->host_ss_selector;
10324 vmx_set_segment(vcpu, &seg, VCPU_SREG_SS);
10325 seg.selector = vmcs12->host_fs_selector;
10326 seg.base = vmcs12->host_fs_base;
10327 vmx_set_segment(vcpu, &seg, VCPU_SREG_FS);
10328 seg.selector = vmcs12->host_gs_selector;
10329 seg.base = vmcs12->host_gs_base;
10330 vmx_set_segment(vcpu, &seg, VCPU_SREG_GS);
10331 seg = (struct kvm_segment) {
10332 .base = vmcs12->host_tr_base,
10334 .selector = vmcs12->host_tr_selector,
10338 vmx_set_segment(vcpu, &seg, VCPU_SREG_TR);
10340 kvm_set_dr(vcpu, 7, 0x400);
10341 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
10343 if (cpu_has_vmx_msr_bitmap())
10344 vmx_set_msr_bitmap(vcpu);
10346 if (nested_vmx_load_msr(vcpu, vmcs12->vm_exit_msr_load_addr,
10347 vmcs12->vm_exit_msr_load_count))
10348 nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
10352 * Emulate an exit from nested guest (L2) to L1, i.e., prepare to run L1
10353 * and modify vmcs12 to make it see what it would expect to see there if
10354 * L2 was its real guest. Must only be called when in L2 (is_guest_mode())
10356 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
10357 u32 exit_intr_info,
10358 unsigned long exit_qualification)
10360 struct vcpu_vmx *vmx = to_vmx(vcpu);
10361 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
10363 /* trying to cancel vmlaunch/vmresume is a bug */
10364 WARN_ON_ONCE(vmx->nested.nested_run_pending);
10366 leave_guest_mode(vcpu);
10367 prepare_vmcs12(vcpu, vmcs12, exit_reason, exit_intr_info,
10368 exit_qualification);
10370 if (nested_vmx_store_msr(vcpu, vmcs12->vm_exit_msr_store_addr,
10371 vmcs12->vm_exit_msr_store_count))
10372 nested_vmx_abort(vcpu, VMX_ABORT_SAVE_GUEST_MSR_FAIL);
10374 vmx_load_vmcs01(vcpu);
10376 if ((exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT)
10377 && nested_exit_intr_ack_set(vcpu)) {
10378 int irq = kvm_cpu_get_interrupt(vcpu);
10380 vmcs12->vm_exit_intr_info = irq |
10381 INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR;
10384 trace_kvm_nested_vmexit_inject(vmcs12->vm_exit_reason,
10385 vmcs12->exit_qualification,
10386 vmcs12->idt_vectoring_info_field,
10387 vmcs12->vm_exit_intr_info,
10388 vmcs12->vm_exit_intr_error_code,
10391 vm_entry_controls_init(vmx, vmcs_read32(VM_ENTRY_CONTROLS));
10392 vm_exit_controls_init(vmx, vmcs_read32(VM_EXIT_CONTROLS));
10393 vmx_segment_cache_clear(vmx);
10395 /* if no vmcs02 cache requested, remove the one we used */
10396 if (VMCS02_POOL_SIZE == 0)
10397 nested_free_vmcs02(vmx, vmx->nested.current_vmptr);
10399 load_vmcs12_host_state(vcpu, vmcs12);
10401 /* Update TSC_OFFSET if TSC was changed while L2 ran */
10402 vmcs_write64(TSC_OFFSET, vmx->nested.vmcs01_tsc_offset);
10404 /* This is needed for same reason as it was needed in prepare_vmcs02 */
10407 /* Unpin physical memory we referred to in vmcs02 */
10408 if (vmx->nested.apic_access_page) {
10409 nested_release_page(vmx->nested.apic_access_page);
10410 vmx->nested.apic_access_page = NULL;
10412 if (vmx->nested.virtual_apic_page) {
10413 nested_release_page(vmx->nested.virtual_apic_page);
10414 vmx->nested.virtual_apic_page = NULL;
10416 if (vmx->nested.pi_desc_page) {
10417 kunmap(vmx->nested.pi_desc_page);
10418 nested_release_page(vmx->nested.pi_desc_page);
10419 vmx->nested.pi_desc_page = NULL;
10420 vmx->nested.pi_desc = NULL;
10424 * We are now running in L2, mmu_notifier will force to reload the
10425 * page's hpa for L2 vmcs. Need to reload it for L1 before entering L1.
10427 kvm_vcpu_reload_apic_access_page(vcpu);
10430 * Exiting from L2 to L1, we're now back to L1 which thinks it just
10431 * finished a VMLAUNCH or VMRESUME instruction, so we need to set the
10432 * success or failure flag accordingly.
10434 if (unlikely(vmx->fail)) {
10436 nested_vmx_failValid(vcpu, vmcs_read32(VM_INSTRUCTION_ERROR));
10438 nested_vmx_succeed(vcpu);
10439 if (enable_shadow_vmcs)
10440 vmx->nested.sync_shadow_vmcs = true;
10442 /* in case we halted in L2 */
10443 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10447 * Forcibly leave nested mode in order to be able to reset the VCPU later on.
10449 static void vmx_leave_nested(struct kvm_vcpu *vcpu)
10451 if (is_guest_mode(vcpu))
10452 nested_vmx_vmexit(vcpu, -1, 0, 0);
10453 free_nested(to_vmx(vcpu));
10457 * L1's failure to enter L2 is a subset of a normal exit, as explained in
10458 * 23.7 "VM-entry failures during or after loading guest state" (this also
10459 * lists the acceptable exit-reason and exit-qualification parameters).
10460 * It should only be called before L2 actually succeeded to run, and when
10461 * vmcs01 is current (it doesn't leave_guest_mode() or switch vmcss).
10463 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
10464 struct vmcs12 *vmcs12,
10465 u32 reason, unsigned long qualification)
10467 load_vmcs12_host_state(vcpu, vmcs12);
10468 vmcs12->vm_exit_reason = reason | VMX_EXIT_REASONS_FAILED_VMENTRY;
10469 vmcs12->exit_qualification = qualification;
10470 nested_vmx_succeed(vcpu);
10471 if (enable_shadow_vmcs)
10472 to_vmx(vcpu)->nested.sync_shadow_vmcs = true;
10475 static int vmx_check_intercept(struct kvm_vcpu *vcpu,
10476 struct x86_instruction_info *info,
10477 enum x86_intercept_stage stage)
10479 return X86EMUL_CONTINUE;
10482 static void vmx_sched_in(struct kvm_vcpu *vcpu, int cpu)
10485 shrink_ple_window(vcpu);
10488 static void vmx_slot_enable_log_dirty(struct kvm *kvm,
10489 struct kvm_memory_slot *slot)
10491 kvm_mmu_slot_leaf_clear_dirty(kvm, slot);
10492 kvm_mmu_slot_largepage_remove_write_access(kvm, slot);
10495 static void vmx_slot_disable_log_dirty(struct kvm *kvm,
10496 struct kvm_memory_slot *slot)
10498 kvm_mmu_slot_set_dirty(kvm, slot);
10501 static void vmx_flush_log_dirty(struct kvm *kvm)
10503 kvm_flush_pml_buffers(kvm);
10506 static void vmx_enable_log_dirty_pt_masked(struct kvm *kvm,
10507 struct kvm_memory_slot *memslot,
10508 gfn_t offset, unsigned long mask)
10510 kvm_mmu_clear_dirty_pt_masked(kvm, memslot, offset, mask);
10514 * This routine does the following things for vCPU which is going
10515 * to be blocked if VT-d PI is enabled.
10516 * - Store the vCPU to the wakeup list, so when interrupts happen
10517 * we can find the right vCPU to wake up.
10518 * - Change the Posted-interrupt descriptor as below:
10519 * 'NDST' <-- vcpu->pre_pcpu
10520 * 'NV' <-- POSTED_INTR_WAKEUP_VECTOR
10521 * - If 'ON' is set during this process, which means at least one
10522 * interrupt is posted for this vCPU, we cannot block it, in
10523 * this case, return 1, otherwise, return 0.
10526 static int vmx_pre_block(struct kvm_vcpu *vcpu)
10528 unsigned long flags;
10530 struct pi_desc old, new;
10531 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
10533 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
10534 !irq_remapping_cap(IRQ_POSTING_CAP))
10537 vcpu->pre_pcpu = vcpu->cpu;
10538 spin_lock_irqsave(&per_cpu(blocked_vcpu_on_cpu_lock,
10539 vcpu->pre_pcpu), flags);
10540 list_add_tail(&vcpu->blocked_vcpu_list,
10541 &per_cpu(blocked_vcpu_on_cpu,
10543 spin_unlock_irqrestore(&per_cpu(blocked_vcpu_on_cpu_lock,
10544 vcpu->pre_pcpu), flags);
10547 old.control = new.control = pi_desc->control;
10550 * We should not block the vCPU if
10551 * an interrupt is posted for it.
10553 if (pi_test_on(pi_desc) == 1) {
10554 spin_lock_irqsave(&per_cpu(blocked_vcpu_on_cpu_lock,
10555 vcpu->pre_pcpu), flags);
10556 list_del(&vcpu->blocked_vcpu_list);
10557 spin_unlock_irqrestore(
10558 &per_cpu(blocked_vcpu_on_cpu_lock,
10559 vcpu->pre_pcpu), flags);
10560 vcpu->pre_pcpu = -1;
10565 WARN((pi_desc->sn == 1),
10566 "Warning: SN field of posted-interrupts "
10567 "is set before blocking\n");
10570 * Since vCPU can be preempted during this process,
10571 * vcpu->cpu could be different with pre_pcpu, we
10572 * need to set pre_pcpu as the destination of wakeup
10573 * notification event, then we can find the right vCPU
10574 * to wakeup in wakeup handler if interrupts happen
10575 * when the vCPU is in blocked state.
10577 dest = cpu_physical_id(vcpu->pre_pcpu);
10579 if (x2apic_enabled())
10582 new.ndst = (dest << 8) & 0xFF00;
10584 /* set 'NV' to 'wakeup vector' */
10585 new.nv = POSTED_INTR_WAKEUP_VECTOR;
10586 } while (cmpxchg(&pi_desc->control, old.control,
10587 new.control) != old.control);
10592 static void vmx_post_block(struct kvm_vcpu *vcpu)
10594 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
10595 struct pi_desc old, new;
10597 unsigned long flags;
10599 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
10600 !irq_remapping_cap(IRQ_POSTING_CAP))
10604 old.control = new.control = pi_desc->control;
10606 dest = cpu_physical_id(vcpu->cpu);
10608 if (x2apic_enabled())
10611 new.ndst = (dest << 8) & 0xFF00;
10613 /* Allow posting non-urgent interrupts */
10616 /* set 'NV' to 'notification vector' */
10617 new.nv = POSTED_INTR_VECTOR;
10618 } while (cmpxchg(&pi_desc->control, old.control,
10619 new.control) != old.control);
10621 if(vcpu->pre_pcpu != -1) {
10623 &per_cpu(blocked_vcpu_on_cpu_lock,
10624 vcpu->pre_pcpu), flags);
10625 list_del(&vcpu->blocked_vcpu_list);
10626 spin_unlock_irqrestore(
10627 &per_cpu(blocked_vcpu_on_cpu_lock,
10628 vcpu->pre_pcpu), flags);
10629 vcpu->pre_pcpu = -1;
10634 * vmx_update_pi_irte - set IRTE for Posted-Interrupts
10637 * @host_irq: host irq of the interrupt
10638 * @guest_irq: gsi of the interrupt
10639 * @set: set or unset PI
10640 * returns 0 on success, < 0 on failure
10642 static int vmx_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
10643 uint32_t guest_irq, bool set)
10645 struct kvm_kernel_irq_routing_entry *e;
10646 struct kvm_irq_routing_table *irq_rt;
10647 struct kvm_lapic_irq irq;
10648 struct kvm_vcpu *vcpu;
10649 struct vcpu_data vcpu_info;
10650 int idx, ret = -EINVAL;
10652 if (!kvm_arch_has_assigned_device(kvm) ||
10653 !irq_remapping_cap(IRQ_POSTING_CAP))
10656 idx = srcu_read_lock(&kvm->irq_srcu);
10657 irq_rt = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu);
10658 BUG_ON(guest_irq >= irq_rt->nr_rt_entries);
10660 hlist_for_each_entry(e, &irq_rt->map[guest_irq], link) {
10661 if (e->type != KVM_IRQ_ROUTING_MSI)
10664 * VT-d PI cannot support posting multicast/broadcast
10665 * interrupts to a vCPU, we still use interrupt remapping
10666 * for these kind of interrupts.
10668 * For lowest-priority interrupts, we only support
10669 * those with single CPU as the destination, e.g. user
10670 * configures the interrupts via /proc/irq or uses
10671 * irqbalance to make the interrupts single-CPU.
10673 * We will support full lowest-priority interrupt later.
10676 kvm_set_msi_irq(e, &irq);
10677 if (!kvm_intr_is_single_vcpu(kvm, &irq, &vcpu))
10680 vcpu_info.pi_desc_addr = __pa(vcpu_to_pi_desc(vcpu));
10681 vcpu_info.vector = irq.vector;
10683 trace_kvm_pi_irte_update(vcpu->vcpu_id, e->gsi,
10684 vcpu_info.vector, vcpu_info.pi_desc_addr, set);
10687 ret = irq_set_vcpu_affinity(host_irq, &vcpu_info);
10689 /* suppress notification event before unposting */
10690 pi_set_sn(vcpu_to_pi_desc(vcpu));
10691 ret = irq_set_vcpu_affinity(host_irq, NULL);
10692 pi_clear_sn(vcpu_to_pi_desc(vcpu));
10696 printk(KERN_INFO "%s: failed to update PI IRTE\n",
10704 srcu_read_unlock(&kvm->irq_srcu, idx);
10708 static struct kvm_x86_ops vmx_x86_ops = {
10709 .cpu_has_kvm_support = cpu_has_kvm_support,
10710 .disabled_by_bios = vmx_disabled_by_bios,
10711 .hardware_setup = hardware_setup,
10712 .hardware_unsetup = hardware_unsetup,
10713 .check_processor_compatibility = vmx_check_processor_compat,
10714 .hardware_enable = hardware_enable,
10715 .hardware_disable = hardware_disable,
10716 .cpu_has_accelerated_tpr = report_flexpriority,
10717 .cpu_has_high_real_mode_segbase = vmx_has_high_real_mode_segbase,
10719 .vcpu_create = vmx_create_vcpu,
10720 .vcpu_free = vmx_free_vcpu,
10721 .vcpu_reset = vmx_vcpu_reset,
10723 .prepare_guest_switch = vmx_save_host_state,
10724 .vcpu_load = vmx_vcpu_load,
10725 .vcpu_put = vmx_vcpu_put,
10727 .update_db_bp_intercept = update_exception_bitmap,
10728 .get_msr = vmx_get_msr,
10729 .set_msr = vmx_set_msr,
10730 .get_segment_base = vmx_get_segment_base,
10731 .get_segment = vmx_get_segment,
10732 .set_segment = vmx_set_segment,
10733 .get_cpl = vmx_get_cpl,
10734 .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
10735 .decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
10736 .decache_cr3 = vmx_decache_cr3,
10737 .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
10738 .set_cr0 = vmx_set_cr0,
10739 .set_cr3 = vmx_set_cr3,
10740 .set_cr4 = vmx_set_cr4,
10741 .set_efer = vmx_set_efer,
10742 .get_idt = vmx_get_idt,
10743 .set_idt = vmx_set_idt,
10744 .get_gdt = vmx_get_gdt,
10745 .set_gdt = vmx_set_gdt,
10746 .get_dr6 = vmx_get_dr6,
10747 .set_dr6 = vmx_set_dr6,
10748 .set_dr7 = vmx_set_dr7,
10749 .sync_dirty_debug_regs = vmx_sync_dirty_debug_regs,
10750 .cache_reg = vmx_cache_reg,
10751 .get_rflags = vmx_get_rflags,
10752 .set_rflags = vmx_set_rflags,
10753 .fpu_activate = vmx_fpu_activate,
10754 .fpu_deactivate = vmx_fpu_deactivate,
10756 .tlb_flush = vmx_flush_tlb,
10758 .run = vmx_vcpu_run,
10759 .handle_exit = vmx_handle_exit,
10760 .skip_emulated_instruction = skip_emulated_instruction,
10761 .set_interrupt_shadow = vmx_set_interrupt_shadow,
10762 .get_interrupt_shadow = vmx_get_interrupt_shadow,
10763 .patch_hypercall = vmx_patch_hypercall,
10764 .set_irq = vmx_inject_irq,
10765 .set_nmi = vmx_inject_nmi,
10766 .queue_exception = vmx_queue_exception,
10767 .cancel_injection = vmx_cancel_injection,
10768 .interrupt_allowed = vmx_interrupt_allowed,
10769 .nmi_allowed = vmx_nmi_allowed,
10770 .get_nmi_mask = vmx_get_nmi_mask,
10771 .set_nmi_mask = vmx_set_nmi_mask,
10772 .enable_nmi_window = enable_nmi_window,
10773 .enable_irq_window = enable_irq_window,
10774 .update_cr8_intercept = update_cr8_intercept,
10775 .set_virtual_x2apic_mode = vmx_set_virtual_x2apic_mode,
10776 .set_apic_access_page_addr = vmx_set_apic_access_page_addr,
10777 .cpu_uses_apicv = vmx_cpu_uses_apicv,
10778 .load_eoi_exitmap = vmx_load_eoi_exitmap,
10779 .hwapic_irr_update = vmx_hwapic_irr_update,
10780 .hwapic_isr_update = vmx_hwapic_isr_update,
10781 .sync_pir_to_irr = vmx_sync_pir_to_irr,
10782 .deliver_posted_interrupt = vmx_deliver_posted_interrupt,
10784 .set_tss_addr = vmx_set_tss_addr,
10785 .get_tdp_level = get_ept_level,
10786 .get_mt_mask = vmx_get_mt_mask,
10788 .get_exit_info = vmx_get_exit_info,
10790 .get_lpage_level = vmx_get_lpage_level,
10792 .cpuid_update = vmx_cpuid_update,
10794 .rdtscp_supported = vmx_rdtscp_supported,
10795 .invpcid_supported = vmx_invpcid_supported,
10797 .set_supported_cpuid = vmx_set_supported_cpuid,
10799 .has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
10801 .set_tsc_khz = vmx_set_tsc_khz,
10802 .read_tsc_offset = vmx_read_tsc_offset,
10803 .write_tsc_offset = vmx_write_tsc_offset,
10804 .adjust_tsc_offset = vmx_adjust_tsc_offset,
10805 .compute_tsc_offset = vmx_compute_tsc_offset,
10806 .read_l1_tsc = vmx_read_l1_tsc,
10808 .set_tdp_cr3 = vmx_set_cr3,
10810 .check_intercept = vmx_check_intercept,
10811 .handle_external_intr = vmx_handle_external_intr,
10812 .mpx_supported = vmx_mpx_supported,
10813 .xsaves_supported = vmx_xsaves_supported,
10815 .check_nested_events = vmx_check_nested_events,
10817 .sched_in = vmx_sched_in,
10819 .slot_enable_log_dirty = vmx_slot_enable_log_dirty,
10820 .slot_disable_log_dirty = vmx_slot_disable_log_dirty,
10821 .flush_log_dirty = vmx_flush_log_dirty,
10822 .enable_log_dirty_pt_masked = vmx_enable_log_dirty_pt_masked,
10824 .pre_block = vmx_pre_block,
10825 .post_block = vmx_post_block,
10827 .pmu_ops = &intel_pmu_ops,
10829 .update_pi_irte = vmx_update_pi_irte,
10832 static int __init vmx_init(void)
10834 int r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx),
10835 __alignof__(struct vcpu_vmx), THIS_MODULE);
10839 #ifdef CONFIG_KEXEC_CORE
10840 rcu_assign_pointer(crash_vmclear_loaded_vmcss,
10841 crash_vmclear_local_loaded_vmcss);
10847 static void __exit vmx_exit(void)
10849 #ifdef CONFIG_KEXEC_CORE
10850 RCU_INIT_POINTER(crash_vmclear_loaded_vmcss, NULL);
10857 module_init(vmx_init)
10858 module_exit(vmx_exit)