2 * Kernel-based Virtual Machine driver for Linux
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
7 * Copyright (C) 2006 Qumranet, Inc.
8 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
11 * Avi Kivity <avi@qumranet.com>
12 * Yaniv Kamay <yaniv@qumranet.com>
14 * This work is licensed under the terms of the GNU GPL, version 2. See
15 * the COPYING file in the top-level directory.
24 #include <linux/kvm_host.h>
25 #include <linux/module.h>
26 #include <linux/kernel.h>
28 #include <linux/highmem.h>
29 #include <linux/sched.h>
30 #include <linux/moduleparam.h>
31 #include <linux/mod_devicetable.h>
32 #include <linux/trace_events.h>
33 #include <linux/slab.h>
34 #include <linux/tboot.h>
35 #include <linux/hrtimer.h>
36 #include <linux/frame.h>
37 #include <linux/nospec.h>
38 #include "kvm_cache_regs.h"
45 #include <asm/virtext.h>
47 #include <asm/fpu/internal.h>
48 #include <asm/perf_event.h>
49 #include <asm/debugreg.h>
50 #include <asm/kexec.h>
52 #include <asm/irq_remapping.h>
53 #include <asm/mmu_context.h>
54 #include <asm/spec-ctrl.h>
55 #include <asm/mshyperv.h>
59 #include "vmx_evmcs.h"
61 #define __ex(x) __kvm_handle_fault_on_reboot(x)
62 #define __ex_clear(x, reg) \
63 ____kvm_handle_fault_on_reboot(x, "xor " reg " , " reg)
65 MODULE_AUTHOR("Qumranet");
66 MODULE_LICENSE("GPL");
68 static const struct x86_cpu_id vmx_cpu_id[] = {
69 X86_FEATURE_MATCH(X86_FEATURE_VMX),
72 MODULE_DEVICE_TABLE(x86cpu, vmx_cpu_id);
74 static bool __read_mostly enable_vpid = 1;
75 module_param_named(vpid, enable_vpid, bool, 0444);
77 static bool __read_mostly enable_vnmi = 1;
78 module_param_named(vnmi, enable_vnmi, bool, S_IRUGO);
80 static bool __read_mostly flexpriority_enabled = 1;
81 module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
83 static bool __read_mostly enable_ept = 1;
84 module_param_named(ept, enable_ept, bool, S_IRUGO);
86 static bool __read_mostly enable_unrestricted_guest = 1;
87 module_param_named(unrestricted_guest,
88 enable_unrestricted_guest, bool, S_IRUGO);
90 static bool __read_mostly enable_ept_ad_bits = 1;
91 module_param_named(eptad, enable_ept_ad_bits, bool, S_IRUGO);
93 static bool __read_mostly emulate_invalid_guest_state = true;
94 module_param(emulate_invalid_guest_state, bool, S_IRUGO);
96 static bool __read_mostly fasteoi = 1;
97 module_param(fasteoi, bool, S_IRUGO);
99 static bool __read_mostly enable_apicv = 1;
100 module_param(enable_apicv, bool, S_IRUGO);
102 static bool __read_mostly enable_shadow_vmcs = 1;
103 module_param_named(enable_shadow_vmcs, enable_shadow_vmcs, bool, S_IRUGO);
105 * If nested=1, nested virtualization is supported, i.e., guests may use
106 * VMX and be a hypervisor for its own guests. If nested=0, guests may not
107 * use VMX instructions.
109 static bool __read_mostly nested = 0;
110 module_param(nested, bool, S_IRUGO);
112 static u64 __read_mostly host_xss;
114 static bool __read_mostly enable_pml = 1;
115 module_param_named(pml, enable_pml, bool, S_IRUGO);
119 #define MSR_TYPE_RW 3
121 #define MSR_BITMAP_MODE_X2APIC 1
122 #define MSR_BITMAP_MODE_X2APIC_APICV 2
123 #define MSR_BITMAP_MODE_LM 4
125 #define KVM_VMX_TSC_MULTIPLIER_MAX 0xffffffffffffffffULL
127 /* Guest_tsc -> host_tsc conversion requires 64-bit division. */
128 static int __read_mostly cpu_preemption_timer_multi;
129 static bool __read_mostly enable_preemption_timer = 1;
131 module_param_named(preemption_timer, enable_preemption_timer, bool, S_IRUGO);
134 #define KVM_GUEST_CR0_MASK (X86_CR0_NW | X86_CR0_CD)
135 #define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST X86_CR0_NE
136 #define KVM_VM_CR0_ALWAYS_ON \
137 (KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | \
138 X86_CR0_WP | X86_CR0_PG | X86_CR0_PE)
139 #define KVM_CR4_GUEST_OWNED_BITS \
140 (X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR \
141 | X86_CR4_OSXMMEXCPT | X86_CR4_LA57 | X86_CR4_TSD)
143 #define KVM_VM_CR4_ALWAYS_ON_UNRESTRICTED_GUEST X86_CR4_VMXE
144 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
145 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
147 #define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
149 #define VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE 5
152 * Hyper-V requires all of these, so mark them as supported even though
153 * they are just treated the same as all-context.
155 #define VMX_VPID_EXTENT_SUPPORTED_MASK \
156 (VMX_VPID_EXTENT_INDIVIDUAL_ADDR_BIT | \
157 VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT | \
158 VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT | \
159 VMX_VPID_EXTENT_SINGLE_NON_GLOBAL_BIT)
162 * These 2 parameters are used to config the controls for Pause-Loop Exiting:
163 * ple_gap: upper bound on the amount of time between two successive
164 * executions of PAUSE in a loop. Also indicate if ple enabled.
165 * According to test, this time is usually smaller than 128 cycles.
166 * ple_window: upper bound on the amount of time a guest is allowed to execute
167 * in a PAUSE loop. Tests indicate that most spinlocks are held for
168 * less than 2^12 cycles
169 * Time is measured based on a counter that runs at the same rate as the TSC,
170 * refer SDM volume 3b section 21.6.13 & 22.1.3.
172 static unsigned int ple_gap = KVM_DEFAULT_PLE_GAP;
174 static unsigned int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
175 module_param(ple_window, uint, 0444);
177 /* Default doubles per-vcpu window every exit. */
178 static unsigned int ple_window_grow = KVM_DEFAULT_PLE_WINDOW_GROW;
179 module_param(ple_window_grow, uint, 0444);
181 /* Default resets per-vcpu window every exit to ple_window. */
182 static unsigned int ple_window_shrink = KVM_DEFAULT_PLE_WINDOW_SHRINK;
183 module_param(ple_window_shrink, uint, 0444);
185 /* Default is to compute the maximum so we can never overflow. */
186 static unsigned int ple_window_max = KVM_VMX_DEFAULT_PLE_WINDOW_MAX;
187 module_param(ple_window_max, uint, 0444);
189 extern const ulong vmx_return;
194 unsigned int tss_addr;
195 bool ept_identity_pagetable_done;
196 gpa_t ept_identity_map_addr;
199 #define NR_AUTOLOAD_MSRS 8
208 * Track a VMCS that may be loaded on a certain CPU. If it is (cpu!=-1), also
209 * remember whether it was VMLAUNCHed, and maintain a linked list of all VMCSs
210 * loaded on this CPU (so we can clear them if the CPU goes down).
214 struct vmcs *shadow_vmcs;
217 bool nmi_known_unmasked;
218 unsigned long vmcs_host_cr3; /* May not match real cr3 */
219 unsigned long vmcs_host_cr4; /* May not match real cr4 */
220 /* Support for vnmi-less CPUs */
221 int soft_vnmi_blocked;
223 s64 vnmi_blocked_time;
224 unsigned long *msr_bitmap;
225 struct list_head loaded_vmcss_on_cpu_link;
228 struct shared_msr_entry {
235 * struct vmcs12 describes the state that our guest hypervisor (L1) keeps for a
236 * single nested guest (L2), hence the name vmcs12. Any VMX implementation has
237 * a VMCS structure, and vmcs12 is our emulated VMX's VMCS. This structure is
238 * stored in guest memory specified by VMPTRLD, but is opaque to the guest,
239 * which must access it using VMREAD/VMWRITE/VMCLEAR instructions.
240 * More than one of these structures may exist, if L1 runs multiple L2 guests.
241 * nested_vmx_run() will use the data here to build the vmcs02: a VMCS for the
242 * underlying hardware which will be used to run L2.
243 * This structure is packed to ensure that its layout is identical across
244 * machines (necessary for live migration).
246 * IMPORTANT: Changing the layout of existing fields in this structure
247 * will break save/restore compatibility with older kvm releases. When
248 * adding new fields, either use space in the reserved padding* arrays
249 * or add the new fields to the end of the structure.
251 typedef u64 natural_width;
252 struct __packed vmcs12 {
253 /* According to the Intel spec, a VMCS region must start with the
254 * following two fields. Then follow implementation-specific data.
259 u32 launch_state; /* set to 0 by VMCLEAR, to 1 by VMLAUNCH */
260 u32 padding[7]; /* room for future expansion */
265 u64 vm_exit_msr_store_addr;
266 u64 vm_exit_msr_load_addr;
267 u64 vm_entry_msr_load_addr;
269 u64 virtual_apic_page_addr;
270 u64 apic_access_addr;
271 u64 posted_intr_desc_addr;
273 u64 eoi_exit_bitmap0;
274 u64 eoi_exit_bitmap1;
275 u64 eoi_exit_bitmap2;
276 u64 eoi_exit_bitmap3;
278 u64 guest_physical_address;
279 u64 vmcs_link_pointer;
280 u64 guest_ia32_debugctl;
283 u64 guest_ia32_perf_global_ctrl;
291 u64 host_ia32_perf_global_ctrl;
294 u64 vm_function_control;
295 u64 eptp_list_address;
297 u64 padding64[3]; /* room for future expansion */
299 * To allow migration of L1 (complete with its L2 guests) between
300 * machines of different natural widths (32 or 64 bit), we cannot have
301 * unsigned long fields with no explict size. We use u64 (aliased
302 * natural_width) instead. Luckily, x86 is little-endian.
304 natural_width cr0_guest_host_mask;
305 natural_width cr4_guest_host_mask;
306 natural_width cr0_read_shadow;
307 natural_width cr4_read_shadow;
308 natural_width cr3_target_value0;
309 natural_width cr3_target_value1;
310 natural_width cr3_target_value2;
311 natural_width cr3_target_value3;
312 natural_width exit_qualification;
313 natural_width guest_linear_address;
314 natural_width guest_cr0;
315 natural_width guest_cr3;
316 natural_width guest_cr4;
317 natural_width guest_es_base;
318 natural_width guest_cs_base;
319 natural_width guest_ss_base;
320 natural_width guest_ds_base;
321 natural_width guest_fs_base;
322 natural_width guest_gs_base;
323 natural_width guest_ldtr_base;
324 natural_width guest_tr_base;
325 natural_width guest_gdtr_base;
326 natural_width guest_idtr_base;
327 natural_width guest_dr7;
328 natural_width guest_rsp;
329 natural_width guest_rip;
330 natural_width guest_rflags;
331 natural_width guest_pending_dbg_exceptions;
332 natural_width guest_sysenter_esp;
333 natural_width guest_sysenter_eip;
334 natural_width host_cr0;
335 natural_width host_cr3;
336 natural_width host_cr4;
337 natural_width host_fs_base;
338 natural_width host_gs_base;
339 natural_width host_tr_base;
340 natural_width host_gdtr_base;
341 natural_width host_idtr_base;
342 natural_width host_ia32_sysenter_esp;
343 natural_width host_ia32_sysenter_eip;
344 natural_width host_rsp;
345 natural_width host_rip;
346 natural_width paddingl[8]; /* room for future expansion */
347 u32 pin_based_vm_exec_control;
348 u32 cpu_based_vm_exec_control;
349 u32 exception_bitmap;
350 u32 page_fault_error_code_mask;
351 u32 page_fault_error_code_match;
352 u32 cr3_target_count;
353 u32 vm_exit_controls;
354 u32 vm_exit_msr_store_count;
355 u32 vm_exit_msr_load_count;
356 u32 vm_entry_controls;
357 u32 vm_entry_msr_load_count;
358 u32 vm_entry_intr_info_field;
359 u32 vm_entry_exception_error_code;
360 u32 vm_entry_instruction_len;
362 u32 secondary_vm_exec_control;
363 u32 vm_instruction_error;
365 u32 vm_exit_intr_info;
366 u32 vm_exit_intr_error_code;
367 u32 idt_vectoring_info_field;
368 u32 idt_vectoring_error_code;
369 u32 vm_exit_instruction_len;
370 u32 vmx_instruction_info;
377 u32 guest_ldtr_limit;
379 u32 guest_gdtr_limit;
380 u32 guest_idtr_limit;
381 u32 guest_es_ar_bytes;
382 u32 guest_cs_ar_bytes;
383 u32 guest_ss_ar_bytes;
384 u32 guest_ds_ar_bytes;
385 u32 guest_fs_ar_bytes;
386 u32 guest_gs_ar_bytes;
387 u32 guest_ldtr_ar_bytes;
388 u32 guest_tr_ar_bytes;
389 u32 guest_interruptibility_info;
390 u32 guest_activity_state;
391 u32 guest_sysenter_cs;
392 u32 host_ia32_sysenter_cs;
393 u32 vmx_preemption_timer_value;
394 u32 padding32[7]; /* room for future expansion */
395 u16 virtual_processor_id;
397 u16 guest_es_selector;
398 u16 guest_cs_selector;
399 u16 guest_ss_selector;
400 u16 guest_ds_selector;
401 u16 guest_fs_selector;
402 u16 guest_gs_selector;
403 u16 guest_ldtr_selector;
404 u16 guest_tr_selector;
405 u16 guest_intr_status;
406 u16 host_es_selector;
407 u16 host_cs_selector;
408 u16 host_ss_selector;
409 u16 host_ds_selector;
410 u16 host_fs_selector;
411 u16 host_gs_selector;
412 u16 host_tr_selector;
417 * For save/restore compatibility, the vmcs12 field offsets must not change.
419 #define CHECK_OFFSET(field, loc) \
420 BUILD_BUG_ON_MSG(offsetof(struct vmcs12, field) != (loc), \
421 "Offset of " #field " in struct vmcs12 has changed.")
423 static inline void vmx_check_vmcs12_offsets(void) {
424 CHECK_OFFSET(revision_id, 0);
425 CHECK_OFFSET(abort, 4);
426 CHECK_OFFSET(launch_state, 8);
427 CHECK_OFFSET(io_bitmap_a, 40);
428 CHECK_OFFSET(io_bitmap_b, 48);
429 CHECK_OFFSET(msr_bitmap, 56);
430 CHECK_OFFSET(vm_exit_msr_store_addr, 64);
431 CHECK_OFFSET(vm_exit_msr_load_addr, 72);
432 CHECK_OFFSET(vm_entry_msr_load_addr, 80);
433 CHECK_OFFSET(tsc_offset, 88);
434 CHECK_OFFSET(virtual_apic_page_addr, 96);
435 CHECK_OFFSET(apic_access_addr, 104);
436 CHECK_OFFSET(posted_intr_desc_addr, 112);
437 CHECK_OFFSET(ept_pointer, 120);
438 CHECK_OFFSET(eoi_exit_bitmap0, 128);
439 CHECK_OFFSET(eoi_exit_bitmap1, 136);
440 CHECK_OFFSET(eoi_exit_bitmap2, 144);
441 CHECK_OFFSET(eoi_exit_bitmap3, 152);
442 CHECK_OFFSET(xss_exit_bitmap, 160);
443 CHECK_OFFSET(guest_physical_address, 168);
444 CHECK_OFFSET(vmcs_link_pointer, 176);
445 CHECK_OFFSET(guest_ia32_debugctl, 184);
446 CHECK_OFFSET(guest_ia32_pat, 192);
447 CHECK_OFFSET(guest_ia32_efer, 200);
448 CHECK_OFFSET(guest_ia32_perf_global_ctrl, 208);
449 CHECK_OFFSET(guest_pdptr0, 216);
450 CHECK_OFFSET(guest_pdptr1, 224);
451 CHECK_OFFSET(guest_pdptr2, 232);
452 CHECK_OFFSET(guest_pdptr3, 240);
453 CHECK_OFFSET(guest_bndcfgs, 248);
454 CHECK_OFFSET(host_ia32_pat, 256);
455 CHECK_OFFSET(host_ia32_efer, 264);
456 CHECK_OFFSET(host_ia32_perf_global_ctrl, 272);
457 CHECK_OFFSET(vmread_bitmap, 280);
458 CHECK_OFFSET(vmwrite_bitmap, 288);
459 CHECK_OFFSET(vm_function_control, 296);
460 CHECK_OFFSET(eptp_list_address, 304);
461 CHECK_OFFSET(pml_address, 312);
462 CHECK_OFFSET(cr0_guest_host_mask, 344);
463 CHECK_OFFSET(cr4_guest_host_mask, 352);
464 CHECK_OFFSET(cr0_read_shadow, 360);
465 CHECK_OFFSET(cr4_read_shadow, 368);
466 CHECK_OFFSET(cr3_target_value0, 376);
467 CHECK_OFFSET(cr3_target_value1, 384);
468 CHECK_OFFSET(cr3_target_value2, 392);
469 CHECK_OFFSET(cr3_target_value3, 400);
470 CHECK_OFFSET(exit_qualification, 408);
471 CHECK_OFFSET(guest_linear_address, 416);
472 CHECK_OFFSET(guest_cr0, 424);
473 CHECK_OFFSET(guest_cr3, 432);
474 CHECK_OFFSET(guest_cr4, 440);
475 CHECK_OFFSET(guest_es_base, 448);
476 CHECK_OFFSET(guest_cs_base, 456);
477 CHECK_OFFSET(guest_ss_base, 464);
478 CHECK_OFFSET(guest_ds_base, 472);
479 CHECK_OFFSET(guest_fs_base, 480);
480 CHECK_OFFSET(guest_gs_base, 488);
481 CHECK_OFFSET(guest_ldtr_base, 496);
482 CHECK_OFFSET(guest_tr_base, 504);
483 CHECK_OFFSET(guest_gdtr_base, 512);
484 CHECK_OFFSET(guest_idtr_base, 520);
485 CHECK_OFFSET(guest_dr7, 528);
486 CHECK_OFFSET(guest_rsp, 536);
487 CHECK_OFFSET(guest_rip, 544);
488 CHECK_OFFSET(guest_rflags, 552);
489 CHECK_OFFSET(guest_pending_dbg_exceptions, 560);
490 CHECK_OFFSET(guest_sysenter_esp, 568);
491 CHECK_OFFSET(guest_sysenter_eip, 576);
492 CHECK_OFFSET(host_cr0, 584);
493 CHECK_OFFSET(host_cr3, 592);
494 CHECK_OFFSET(host_cr4, 600);
495 CHECK_OFFSET(host_fs_base, 608);
496 CHECK_OFFSET(host_gs_base, 616);
497 CHECK_OFFSET(host_tr_base, 624);
498 CHECK_OFFSET(host_gdtr_base, 632);
499 CHECK_OFFSET(host_idtr_base, 640);
500 CHECK_OFFSET(host_ia32_sysenter_esp, 648);
501 CHECK_OFFSET(host_ia32_sysenter_eip, 656);
502 CHECK_OFFSET(host_rsp, 664);
503 CHECK_OFFSET(host_rip, 672);
504 CHECK_OFFSET(pin_based_vm_exec_control, 744);
505 CHECK_OFFSET(cpu_based_vm_exec_control, 748);
506 CHECK_OFFSET(exception_bitmap, 752);
507 CHECK_OFFSET(page_fault_error_code_mask, 756);
508 CHECK_OFFSET(page_fault_error_code_match, 760);
509 CHECK_OFFSET(cr3_target_count, 764);
510 CHECK_OFFSET(vm_exit_controls, 768);
511 CHECK_OFFSET(vm_exit_msr_store_count, 772);
512 CHECK_OFFSET(vm_exit_msr_load_count, 776);
513 CHECK_OFFSET(vm_entry_controls, 780);
514 CHECK_OFFSET(vm_entry_msr_load_count, 784);
515 CHECK_OFFSET(vm_entry_intr_info_field, 788);
516 CHECK_OFFSET(vm_entry_exception_error_code, 792);
517 CHECK_OFFSET(vm_entry_instruction_len, 796);
518 CHECK_OFFSET(tpr_threshold, 800);
519 CHECK_OFFSET(secondary_vm_exec_control, 804);
520 CHECK_OFFSET(vm_instruction_error, 808);
521 CHECK_OFFSET(vm_exit_reason, 812);
522 CHECK_OFFSET(vm_exit_intr_info, 816);
523 CHECK_OFFSET(vm_exit_intr_error_code, 820);
524 CHECK_OFFSET(idt_vectoring_info_field, 824);
525 CHECK_OFFSET(idt_vectoring_error_code, 828);
526 CHECK_OFFSET(vm_exit_instruction_len, 832);
527 CHECK_OFFSET(vmx_instruction_info, 836);
528 CHECK_OFFSET(guest_es_limit, 840);
529 CHECK_OFFSET(guest_cs_limit, 844);
530 CHECK_OFFSET(guest_ss_limit, 848);
531 CHECK_OFFSET(guest_ds_limit, 852);
532 CHECK_OFFSET(guest_fs_limit, 856);
533 CHECK_OFFSET(guest_gs_limit, 860);
534 CHECK_OFFSET(guest_ldtr_limit, 864);
535 CHECK_OFFSET(guest_tr_limit, 868);
536 CHECK_OFFSET(guest_gdtr_limit, 872);
537 CHECK_OFFSET(guest_idtr_limit, 876);
538 CHECK_OFFSET(guest_es_ar_bytes, 880);
539 CHECK_OFFSET(guest_cs_ar_bytes, 884);
540 CHECK_OFFSET(guest_ss_ar_bytes, 888);
541 CHECK_OFFSET(guest_ds_ar_bytes, 892);
542 CHECK_OFFSET(guest_fs_ar_bytes, 896);
543 CHECK_OFFSET(guest_gs_ar_bytes, 900);
544 CHECK_OFFSET(guest_ldtr_ar_bytes, 904);
545 CHECK_OFFSET(guest_tr_ar_bytes, 908);
546 CHECK_OFFSET(guest_interruptibility_info, 912);
547 CHECK_OFFSET(guest_activity_state, 916);
548 CHECK_OFFSET(guest_sysenter_cs, 920);
549 CHECK_OFFSET(host_ia32_sysenter_cs, 924);
550 CHECK_OFFSET(vmx_preemption_timer_value, 928);
551 CHECK_OFFSET(virtual_processor_id, 960);
552 CHECK_OFFSET(posted_intr_nv, 962);
553 CHECK_OFFSET(guest_es_selector, 964);
554 CHECK_OFFSET(guest_cs_selector, 966);
555 CHECK_OFFSET(guest_ss_selector, 968);
556 CHECK_OFFSET(guest_ds_selector, 970);
557 CHECK_OFFSET(guest_fs_selector, 972);
558 CHECK_OFFSET(guest_gs_selector, 974);
559 CHECK_OFFSET(guest_ldtr_selector, 976);
560 CHECK_OFFSET(guest_tr_selector, 978);
561 CHECK_OFFSET(guest_intr_status, 980);
562 CHECK_OFFSET(host_es_selector, 982);
563 CHECK_OFFSET(host_cs_selector, 984);
564 CHECK_OFFSET(host_ss_selector, 986);
565 CHECK_OFFSET(host_ds_selector, 988);
566 CHECK_OFFSET(host_fs_selector, 990);
567 CHECK_OFFSET(host_gs_selector, 992);
568 CHECK_OFFSET(host_tr_selector, 994);
569 CHECK_OFFSET(guest_pml_index, 996);
573 * VMCS12_REVISION is an arbitrary id that should be changed if the content or
574 * layout of struct vmcs12 is changed. MSR_IA32_VMX_BASIC returns this id, and
575 * VMPTRLD verifies that the VMCS region that L1 is loading contains this id.
577 * IMPORTANT: Changing this value will break save/restore compatibility with
578 * older kvm releases.
580 #define VMCS12_REVISION 0x11e57ed0
583 * VMCS12_SIZE is the number of bytes L1 should allocate for the VMXON region
584 * and any VMCS region. Although only sizeof(struct vmcs12) are used by the
585 * current implementation, 4K are reserved to avoid future complications.
587 #define VMCS12_SIZE 0x1000
590 * VMCS12_MAX_FIELD_INDEX is the highest index value used in any
591 * supported VMCS12 field encoding.
593 #define VMCS12_MAX_FIELD_INDEX 0x17
595 struct nested_vmx_msrs {
597 * We only store the "true" versions of the VMX capability MSRs. We
598 * generate the "non-true" versions by setting the must-be-1 bits
599 * according to the SDM.
601 u32 procbased_ctls_low;
602 u32 procbased_ctls_high;
603 u32 secondary_ctls_low;
604 u32 secondary_ctls_high;
605 u32 pinbased_ctls_low;
606 u32 pinbased_ctls_high;
625 * The nested_vmx structure is part of vcpu_vmx, and holds information we need
626 * for correct emulation of VMX (i.e., nested VMX) on this vcpu.
629 /* Has the level1 guest done vmxon? */
634 /* The guest-physical address of the current VMCS L1 keeps for L2 */
637 * Cache of the guest's VMCS, existing outside of guest memory.
638 * Loaded from guest memory during VMPTRLD. Flushed to guest
639 * memory during VMCLEAR and VMPTRLD.
641 struct vmcs12 *cached_vmcs12;
643 * Indicates if the shadow vmcs must be updated with the
644 * data hold by vmcs12
646 bool sync_shadow_vmcs;
649 bool change_vmcs01_virtual_apic_mode;
651 /* L2 must run next, and mustn't decide to exit to L1. */
652 bool nested_run_pending;
654 struct loaded_vmcs vmcs02;
657 * Guest pages referred to in the vmcs02 with host-physical
658 * pointers, so we must keep them pinned while L2 runs.
660 struct page *apic_access_page;
661 struct page *virtual_apic_page;
662 struct page *pi_desc_page;
663 struct pi_desc *pi_desc;
667 struct hrtimer preemption_timer;
668 bool preemption_timer_expired;
670 /* to migrate it to L2 if VM_ENTRY_LOAD_DEBUG_CONTROLS is off */
676 struct nested_vmx_msrs msrs;
678 /* SMM related state */
680 /* in VMX operation on SMM entry? */
682 /* in guest mode on SMM entry? */
687 #define POSTED_INTR_ON 0
688 #define POSTED_INTR_SN 1
690 /* Posted-Interrupt Descriptor */
692 u32 pir[8]; /* Posted interrupt requested */
695 /* bit 256 - Outstanding Notification */
697 /* bit 257 - Suppress Notification */
699 /* bit 271:258 - Reserved */
701 /* bit 279:272 - Notification Vector */
703 /* bit 287:280 - Reserved */
705 /* bit 319:288 - Notification Destination */
713 static bool pi_test_and_set_on(struct pi_desc *pi_desc)
715 return test_and_set_bit(POSTED_INTR_ON,
716 (unsigned long *)&pi_desc->control);
719 static bool pi_test_and_clear_on(struct pi_desc *pi_desc)
721 return test_and_clear_bit(POSTED_INTR_ON,
722 (unsigned long *)&pi_desc->control);
725 static int pi_test_and_set_pir(int vector, struct pi_desc *pi_desc)
727 return test_and_set_bit(vector, (unsigned long *)pi_desc->pir);
730 static inline void pi_clear_sn(struct pi_desc *pi_desc)
732 return clear_bit(POSTED_INTR_SN,
733 (unsigned long *)&pi_desc->control);
736 static inline void pi_set_sn(struct pi_desc *pi_desc)
738 return set_bit(POSTED_INTR_SN,
739 (unsigned long *)&pi_desc->control);
742 static inline void pi_clear_on(struct pi_desc *pi_desc)
744 clear_bit(POSTED_INTR_ON,
745 (unsigned long *)&pi_desc->control);
748 static inline int pi_test_on(struct pi_desc *pi_desc)
750 return test_bit(POSTED_INTR_ON,
751 (unsigned long *)&pi_desc->control);
754 static inline int pi_test_sn(struct pi_desc *pi_desc)
756 return test_bit(POSTED_INTR_SN,
757 (unsigned long *)&pi_desc->control);
761 struct kvm_vcpu vcpu;
762 unsigned long host_rsp;
766 u32 idt_vectoring_info;
768 struct shared_msr_entry *guest_msrs;
771 unsigned long host_idt_base;
773 u64 msr_host_kernel_gs_base;
774 u64 msr_guest_kernel_gs_base;
777 u64 arch_capabilities;
780 u32 vm_entry_controls_shadow;
781 u32 vm_exit_controls_shadow;
782 u32 secondary_exec_control;
785 * loaded_vmcs points to the VMCS currently used in this vcpu. For a
786 * non-nested (L1) guest, it always points to vmcs01. For a nested
787 * guest (L2), it points to a different VMCS.
789 struct loaded_vmcs vmcs01;
790 struct loaded_vmcs *loaded_vmcs;
791 bool __launched; /* temporary, used in vmx_vcpu_run */
792 struct msr_autoload {
794 struct vmx_msr_entry guest[NR_AUTOLOAD_MSRS];
795 struct vmx_msr_entry host[NR_AUTOLOAD_MSRS];
799 u16 fs_sel, gs_sel, ldt_sel;
803 int gs_ldt_reload_needed;
804 int fs_reload_needed;
809 struct kvm_segment segs[8];
812 u32 bitmask; /* 4 bits per segment (1 bit per field) */
813 struct kvm_save_segment {
821 bool emulation_required;
825 /* Posted interrupt descriptor */
826 struct pi_desc pi_desc;
828 /* Support for a guest hypervisor (nested VMX) */
829 struct nested_vmx nested;
831 /* Dynamic PLE window. */
833 bool ple_window_dirty;
835 /* Support for PML */
836 #define PML_ENTITY_NUM 512
839 /* apic deadline value in host tsc */
842 u64 current_tsc_ratio;
846 unsigned long host_debugctlmsr;
849 * Only bits masked by msr_ia32_feature_control_valid_bits can be set in
850 * msr_ia32_feature_control. FEATURE_CONTROL_LOCKED is always included
851 * in msr_ia32_feature_control_valid_bits.
853 u64 msr_ia32_feature_control;
854 u64 msr_ia32_feature_control_valid_bits;
857 enum segment_cache_field {
866 static inline struct kvm_vmx *to_kvm_vmx(struct kvm *kvm)
868 return container_of(kvm, struct kvm_vmx, kvm);
871 static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
873 return container_of(vcpu, struct vcpu_vmx, vcpu);
876 static struct pi_desc *vcpu_to_pi_desc(struct kvm_vcpu *vcpu)
878 return &(to_vmx(vcpu)->pi_desc);
881 #define ROL16(val, n) ((u16)(((u16)(val) << (n)) | ((u16)(val) >> (16 - (n)))))
882 #define VMCS12_OFFSET(x) offsetof(struct vmcs12, x)
883 #define FIELD(number, name) [ROL16(number, 6)] = VMCS12_OFFSET(name)
884 #define FIELD64(number, name) \
885 FIELD(number, name), \
886 [ROL16(number##_HIGH, 6)] = VMCS12_OFFSET(name) + sizeof(u32)
889 static u16 shadow_read_only_fields[] = {
890 #define SHADOW_FIELD_RO(x) x,
891 #include "vmx_shadow_fields.h"
893 static int max_shadow_read_only_fields =
894 ARRAY_SIZE(shadow_read_only_fields);
896 static u16 shadow_read_write_fields[] = {
897 #define SHADOW_FIELD_RW(x) x,
898 #include "vmx_shadow_fields.h"
900 static int max_shadow_read_write_fields =
901 ARRAY_SIZE(shadow_read_write_fields);
903 static const unsigned short vmcs_field_to_offset_table[] = {
904 FIELD(VIRTUAL_PROCESSOR_ID, virtual_processor_id),
905 FIELD(POSTED_INTR_NV, posted_intr_nv),
906 FIELD(GUEST_ES_SELECTOR, guest_es_selector),
907 FIELD(GUEST_CS_SELECTOR, guest_cs_selector),
908 FIELD(GUEST_SS_SELECTOR, guest_ss_selector),
909 FIELD(GUEST_DS_SELECTOR, guest_ds_selector),
910 FIELD(GUEST_FS_SELECTOR, guest_fs_selector),
911 FIELD(GUEST_GS_SELECTOR, guest_gs_selector),
912 FIELD(GUEST_LDTR_SELECTOR, guest_ldtr_selector),
913 FIELD(GUEST_TR_SELECTOR, guest_tr_selector),
914 FIELD(GUEST_INTR_STATUS, guest_intr_status),
915 FIELD(GUEST_PML_INDEX, guest_pml_index),
916 FIELD(HOST_ES_SELECTOR, host_es_selector),
917 FIELD(HOST_CS_SELECTOR, host_cs_selector),
918 FIELD(HOST_SS_SELECTOR, host_ss_selector),
919 FIELD(HOST_DS_SELECTOR, host_ds_selector),
920 FIELD(HOST_FS_SELECTOR, host_fs_selector),
921 FIELD(HOST_GS_SELECTOR, host_gs_selector),
922 FIELD(HOST_TR_SELECTOR, host_tr_selector),
923 FIELD64(IO_BITMAP_A, io_bitmap_a),
924 FIELD64(IO_BITMAP_B, io_bitmap_b),
925 FIELD64(MSR_BITMAP, msr_bitmap),
926 FIELD64(VM_EXIT_MSR_STORE_ADDR, vm_exit_msr_store_addr),
927 FIELD64(VM_EXIT_MSR_LOAD_ADDR, vm_exit_msr_load_addr),
928 FIELD64(VM_ENTRY_MSR_LOAD_ADDR, vm_entry_msr_load_addr),
929 FIELD64(PML_ADDRESS, pml_address),
930 FIELD64(TSC_OFFSET, tsc_offset),
931 FIELD64(VIRTUAL_APIC_PAGE_ADDR, virtual_apic_page_addr),
932 FIELD64(APIC_ACCESS_ADDR, apic_access_addr),
933 FIELD64(POSTED_INTR_DESC_ADDR, posted_intr_desc_addr),
934 FIELD64(VM_FUNCTION_CONTROL, vm_function_control),
935 FIELD64(EPT_POINTER, ept_pointer),
936 FIELD64(EOI_EXIT_BITMAP0, eoi_exit_bitmap0),
937 FIELD64(EOI_EXIT_BITMAP1, eoi_exit_bitmap1),
938 FIELD64(EOI_EXIT_BITMAP2, eoi_exit_bitmap2),
939 FIELD64(EOI_EXIT_BITMAP3, eoi_exit_bitmap3),
940 FIELD64(EPTP_LIST_ADDRESS, eptp_list_address),
941 FIELD64(VMREAD_BITMAP, vmread_bitmap),
942 FIELD64(VMWRITE_BITMAP, vmwrite_bitmap),
943 FIELD64(XSS_EXIT_BITMAP, xss_exit_bitmap),
944 FIELD64(GUEST_PHYSICAL_ADDRESS, guest_physical_address),
945 FIELD64(VMCS_LINK_POINTER, vmcs_link_pointer),
946 FIELD64(GUEST_IA32_DEBUGCTL, guest_ia32_debugctl),
947 FIELD64(GUEST_IA32_PAT, guest_ia32_pat),
948 FIELD64(GUEST_IA32_EFER, guest_ia32_efer),
949 FIELD64(GUEST_IA32_PERF_GLOBAL_CTRL, guest_ia32_perf_global_ctrl),
950 FIELD64(GUEST_PDPTR0, guest_pdptr0),
951 FIELD64(GUEST_PDPTR1, guest_pdptr1),
952 FIELD64(GUEST_PDPTR2, guest_pdptr2),
953 FIELD64(GUEST_PDPTR3, guest_pdptr3),
954 FIELD64(GUEST_BNDCFGS, guest_bndcfgs),
955 FIELD64(HOST_IA32_PAT, host_ia32_pat),
956 FIELD64(HOST_IA32_EFER, host_ia32_efer),
957 FIELD64(HOST_IA32_PERF_GLOBAL_CTRL, host_ia32_perf_global_ctrl),
958 FIELD(PIN_BASED_VM_EXEC_CONTROL, pin_based_vm_exec_control),
959 FIELD(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control),
960 FIELD(EXCEPTION_BITMAP, exception_bitmap),
961 FIELD(PAGE_FAULT_ERROR_CODE_MASK, page_fault_error_code_mask),
962 FIELD(PAGE_FAULT_ERROR_CODE_MATCH, page_fault_error_code_match),
963 FIELD(CR3_TARGET_COUNT, cr3_target_count),
964 FIELD(VM_EXIT_CONTROLS, vm_exit_controls),
965 FIELD(VM_EXIT_MSR_STORE_COUNT, vm_exit_msr_store_count),
966 FIELD(VM_EXIT_MSR_LOAD_COUNT, vm_exit_msr_load_count),
967 FIELD(VM_ENTRY_CONTROLS, vm_entry_controls),
968 FIELD(VM_ENTRY_MSR_LOAD_COUNT, vm_entry_msr_load_count),
969 FIELD(VM_ENTRY_INTR_INFO_FIELD, vm_entry_intr_info_field),
970 FIELD(VM_ENTRY_EXCEPTION_ERROR_CODE, vm_entry_exception_error_code),
971 FIELD(VM_ENTRY_INSTRUCTION_LEN, vm_entry_instruction_len),
972 FIELD(TPR_THRESHOLD, tpr_threshold),
973 FIELD(SECONDARY_VM_EXEC_CONTROL, secondary_vm_exec_control),
974 FIELD(VM_INSTRUCTION_ERROR, vm_instruction_error),
975 FIELD(VM_EXIT_REASON, vm_exit_reason),
976 FIELD(VM_EXIT_INTR_INFO, vm_exit_intr_info),
977 FIELD(VM_EXIT_INTR_ERROR_CODE, vm_exit_intr_error_code),
978 FIELD(IDT_VECTORING_INFO_FIELD, idt_vectoring_info_field),
979 FIELD(IDT_VECTORING_ERROR_CODE, idt_vectoring_error_code),
980 FIELD(VM_EXIT_INSTRUCTION_LEN, vm_exit_instruction_len),
981 FIELD(VMX_INSTRUCTION_INFO, vmx_instruction_info),
982 FIELD(GUEST_ES_LIMIT, guest_es_limit),
983 FIELD(GUEST_CS_LIMIT, guest_cs_limit),
984 FIELD(GUEST_SS_LIMIT, guest_ss_limit),
985 FIELD(GUEST_DS_LIMIT, guest_ds_limit),
986 FIELD(GUEST_FS_LIMIT, guest_fs_limit),
987 FIELD(GUEST_GS_LIMIT, guest_gs_limit),
988 FIELD(GUEST_LDTR_LIMIT, guest_ldtr_limit),
989 FIELD(GUEST_TR_LIMIT, guest_tr_limit),
990 FIELD(GUEST_GDTR_LIMIT, guest_gdtr_limit),
991 FIELD(GUEST_IDTR_LIMIT, guest_idtr_limit),
992 FIELD(GUEST_ES_AR_BYTES, guest_es_ar_bytes),
993 FIELD(GUEST_CS_AR_BYTES, guest_cs_ar_bytes),
994 FIELD(GUEST_SS_AR_BYTES, guest_ss_ar_bytes),
995 FIELD(GUEST_DS_AR_BYTES, guest_ds_ar_bytes),
996 FIELD(GUEST_FS_AR_BYTES, guest_fs_ar_bytes),
997 FIELD(GUEST_GS_AR_BYTES, guest_gs_ar_bytes),
998 FIELD(GUEST_LDTR_AR_BYTES, guest_ldtr_ar_bytes),
999 FIELD(GUEST_TR_AR_BYTES, guest_tr_ar_bytes),
1000 FIELD(GUEST_INTERRUPTIBILITY_INFO, guest_interruptibility_info),
1001 FIELD(GUEST_ACTIVITY_STATE, guest_activity_state),
1002 FIELD(GUEST_SYSENTER_CS, guest_sysenter_cs),
1003 FIELD(HOST_IA32_SYSENTER_CS, host_ia32_sysenter_cs),
1004 FIELD(VMX_PREEMPTION_TIMER_VALUE, vmx_preemption_timer_value),
1005 FIELD(CR0_GUEST_HOST_MASK, cr0_guest_host_mask),
1006 FIELD(CR4_GUEST_HOST_MASK, cr4_guest_host_mask),
1007 FIELD(CR0_READ_SHADOW, cr0_read_shadow),
1008 FIELD(CR4_READ_SHADOW, cr4_read_shadow),
1009 FIELD(CR3_TARGET_VALUE0, cr3_target_value0),
1010 FIELD(CR3_TARGET_VALUE1, cr3_target_value1),
1011 FIELD(CR3_TARGET_VALUE2, cr3_target_value2),
1012 FIELD(CR3_TARGET_VALUE3, cr3_target_value3),
1013 FIELD(EXIT_QUALIFICATION, exit_qualification),
1014 FIELD(GUEST_LINEAR_ADDRESS, guest_linear_address),
1015 FIELD(GUEST_CR0, guest_cr0),
1016 FIELD(GUEST_CR3, guest_cr3),
1017 FIELD(GUEST_CR4, guest_cr4),
1018 FIELD(GUEST_ES_BASE, guest_es_base),
1019 FIELD(GUEST_CS_BASE, guest_cs_base),
1020 FIELD(GUEST_SS_BASE, guest_ss_base),
1021 FIELD(GUEST_DS_BASE, guest_ds_base),
1022 FIELD(GUEST_FS_BASE, guest_fs_base),
1023 FIELD(GUEST_GS_BASE, guest_gs_base),
1024 FIELD(GUEST_LDTR_BASE, guest_ldtr_base),
1025 FIELD(GUEST_TR_BASE, guest_tr_base),
1026 FIELD(GUEST_GDTR_BASE, guest_gdtr_base),
1027 FIELD(GUEST_IDTR_BASE, guest_idtr_base),
1028 FIELD(GUEST_DR7, guest_dr7),
1029 FIELD(GUEST_RSP, guest_rsp),
1030 FIELD(GUEST_RIP, guest_rip),
1031 FIELD(GUEST_RFLAGS, guest_rflags),
1032 FIELD(GUEST_PENDING_DBG_EXCEPTIONS, guest_pending_dbg_exceptions),
1033 FIELD(GUEST_SYSENTER_ESP, guest_sysenter_esp),
1034 FIELD(GUEST_SYSENTER_EIP, guest_sysenter_eip),
1035 FIELD(HOST_CR0, host_cr0),
1036 FIELD(HOST_CR3, host_cr3),
1037 FIELD(HOST_CR4, host_cr4),
1038 FIELD(HOST_FS_BASE, host_fs_base),
1039 FIELD(HOST_GS_BASE, host_gs_base),
1040 FIELD(HOST_TR_BASE, host_tr_base),
1041 FIELD(HOST_GDTR_BASE, host_gdtr_base),
1042 FIELD(HOST_IDTR_BASE, host_idtr_base),
1043 FIELD(HOST_IA32_SYSENTER_ESP, host_ia32_sysenter_esp),
1044 FIELD(HOST_IA32_SYSENTER_EIP, host_ia32_sysenter_eip),
1045 FIELD(HOST_RSP, host_rsp),
1046 FIELD(HOST_RIP, host_rip),
1049 static inline short vmcs_field_to_offset(unsigned long field)
1051 const size_t size = ARRAY_SIZE(vmcs_field_to_offset_table);
1052 unsigned short offset;
1058 index = ROL16(field, 6);
1062 index = array_index_nospec(index, size);
1063 offset = vmcs_field_to_offset_table[index];
1069 static inline struct vmcs12 *get_vmcs12(struct kvm_vcpu *vcpu)
1071 return to_vmx(vcpu)->nested.cached_vmcs12;
1074 static bool nested_ept_ad_enabled(struct kvm_vcpu *vcpu);
1075 static unsigned long nested_ept_get_cr3(struct kvm_vcpu *vcpu);
1076 static u64 construct_eptp(struct kvm_vcpu *vcpu, unsigned long root_hpa);
1077 static bool vmx_xsaves_supported(void);
1078 static void vmx_set_segment(struct kvm_vcpu *vcpu,
1079 struct kvm_segment *var, int seg);
1080 static void vmx_get_segment(struct kvm_vcpu *vcpu,
1081 struct kvm_segment *var, int seg);
1082 static bool guest_state_valid(struct kvm_vcpu *vcpu);
1083 static u32 vmx_segment_access_rights(struct kvm_segment *var);
1084 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx);
1085 static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu);
1086 static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked);
1087 static bool nested_vmx_is_page_fault_vmexit(struct vmcs12 *vmcs12,
1089 static void vmx_update_msr_bitmap(struct kvm_vcpu *vcpu);
1090 static void __always_inline vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
1093 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
1094 static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
1096 * We maintain a per-CPU linked-list of VMCS loaded on that CPU. This is needed
1097 * when a CPU is brought down, and we need to VMCLEAR all VMCSs loaded on it.
1099 static DEFINE_PER_CPU(struct list_head, loaded_vmcss_on_cpu);
1102 * We maintian a per-CPU linked-list of vCPU, so in wakeup_handler() we
1103 * can find which vCPU should be waken up.
1105 static DEFINE_PER_CPU(struct list_head, blocked_vcpu_on_cpu);
1106 static DEFINE_PER_CPU(spinlock_t, blocked_vcpu_on_cpu_lock);
1114 static unsigned long *vmx_bitmap[VMX_BITMAP_NR];
1116 #define vmx_vmread_bitmap (vmx_bitmap[VMX_VMREAD_BITMAP])
1117 #define vmx_vmwrite_bitmap (vmx_bitmap[VMX_VMWRITE_BITMAP])
1119 static bool cpu_has_load_ia32_efer;
1120 static bool cpu_has_load_perf_global_ctrl;
1122 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
1123 static DEFINE_SPINLOCK(vmx_vpid_lock);
1125 static struct vmcs_config {
1130 u32 pin_based_exec_ctrl;
1131 u32 cpu_based_exec_ctrl;
1132 u32 cpu_based_2nd_exec_ctrl;
1135 struct nested_vmx_msrs nested;
1138 static struct vmx_capability {
1143 #define VMX_SEGMENT_FIELD(seg) \
1144 [VCPU_SREG_##seg] = { \
1145 .selector = GUEST_##seg##_SELECTOR, \
1146 .base = GUEST_##seg##_BASE, \
1147 .limit = GUEST_##seg##_LIMIT, \
1148 .ar_bytes = GUEST_##seg##_AR_BYTES, \
1151 static const struct kvm_vmx_segment_field {
1156 } kvm_vmx_segment_fields[] = {
1157 VMX_SEGMENT_FIELD(CS),
1158 VMX_SEGMENT_FIELD(DS),
1159 VMX_SEGMENT_FIELD(ES),
1160 VMX_SEGMENT_FIELD(FS),
1161 VMX_SEGMENT_FIELD(GS),
1162 VMX_SEGMENT_FIELD(SS),
1163 VMX_SEGMENT_FIELD(TR),
1164 VMX_SEGMENT_FIELD(LDTR),
1167 static u64 host_efer;
1169 static void ept_save_pdptrs(struct kvm_vcpu *vcpu);
1172 * Keep MSR_STAR at the end, as setup_msrs() will try to optimize it
1173 * away by decrementing the array size.
1175 static const u32 vmx_msr_index[] = {
1176 #ifdef CONFIG_X86_64
1177 MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
1179 MSR_EFER, MSR_TSC_AUX, MSR_STAR,
1182 DEFINE_STATIC_KEY_FALSE(enable_evmcs);
1184 #define current_evmcs ((struct hv_enlightened_vmcs *)this_cpu_read(current_vmcs))
1186 #define KVM_EVMCS_VERSION 1
1188 #if IS_ENABLED(CONFIG_HYPERV)
1189 static bool __read_mostly enlightened_vmcs = true;
1190 module_param(enlightened_vmcs, bool, 0444);
1192 static inline void evmcs_write64(unsigned long field, u64 value)
1195 int offset = get_evmcs_offset(field, &clean_field);
1200 *(u64 *)((char *)current_evmcs + offset) = value;
1202 current_evmcs->hv_clean_fields &= ~clean_field;
1205 static inline void evmcs_write32(unsigned long field, u32 value)
1208 int offset = get_evmcs_offset(field, &clean_field);
1213 *(u32 *)((char *)current_evmcs + offset) = value;
1214 current_evmcs->hv_clean_fields &= ~clean_field;
1217 static inline void evmcs_write16(unsigned long field, u16 value)
1220 int offset = get_evmcs_offset(field, &clean_field);
1225 *(u16 *)((char *)current_evmcs + offset) = value;
1226 current_evmcs->hv_clean_fields &= ~clean_field;
1229 static inline u64 evmcs_read64(unsigned long field)
1231 int offset = get_evmcs_offset(field, NULL);
1236 return *(u64 *)((char *)current_evmcs + offset);
1239 static inline u32 evmcs_read32(unsigned long field)
1241 int offset = get_evmcs_offset(field, NULL);
1246 return *(u32 *)((char *)current_evmcs + offset);
1249 static inline u16 evmcs_read16(unsigned long field)
1251 int offset = get_evmcs_offset(field, NULL);
1256 return *(u16 *)((char *)current_evmcs + offset);
1259 static inline void evmcs_touch_msr_bitmap(void)
1261 if (unlikely(!current_evmcs))
1264 if (current_evmcs->hv_enlightenments_control.msr_bitmap)
1265 current_evmcs->hv_clean_fields &=
1266 ~HV_VMX_ENLIGHTENED_CLEAN_FIELD_MSR_BITMAP;
1269 static void evmcs_load(u64 phys_addr)
1271 struct hv_vp_assist_page *vp_ap =
1272 hv_get_vp_assist_page(smp_processor_id());
1274 vp_ap->current_nested_vmcs = phys_addr;
1275 vp_ap->enlighten_vmentry = 1;
1278 static void evmcs_sanitize_exec_ctrls(struct vmcs_config *vmcs_conf)
1281 * Enlightened VMCSv1 doesn't support these:
1283 * POSTED_INTR_NV = 0x00000002,
1284 * GUEST_INTR_STATUS = 0x00000810,
1285 * APIC_ACCESS_ADDR = 0x00002014,
1286 * POSTED_INTR_DESC_ADDR = 0x00002016,
1287 * EOI_EXIT_BITMAP0 = 0x0000201c,
1288 * EOI_EXIT_BITMAP1 = 0x0000201e,
1289 * EOI_EXIT_BITMAP2 = 0x00002020,
1290 * EOI_EXIT_BITMAP3 = 0x00002022,
1292 vmcs_conf->pin_based_exec_ctrl &= ~PIN_BASED_POSTED_INTR;
1293 vmcs_conf->cpu_based_2nd_exec_ctrl &=
1294 ~SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY;
1295 vmcs_conf->cpu_based_2nd_exec_ctrl &=
1296 ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
1297 vmcs_conf->cpu_based_2nd_exec_ctrl &=
1298 ~SECONDARY_EXEC_APIC_REGISTER_VIRT;
1301 * GUEST_PML_INDEX = 0x00000812,
1302 * PML_ADDRESS = 0x0000200e,
1304 vmcs_conf->cpu_based_2nd_exec_ctrl &= ~SECONDARY_EXEC_ENABLE_PML;
1306 /* VM_FUNCTION_CONTROL = 0x00002018, */
1307 vmcs_conf->cpu_based_2nd_exec_ctrl &= ~SECONDARY_EXEC_ENABLE_VMFUNC;
1310 * EPTP_LIST_ADDRESS = 0x00002024,
1311 * VMREAD_BITMAP = 0x00002026,
1312 * VMWRITE_BITMAP = 0x00002028,
1314 vmcs_conf->cpu_based_2nd_exec_ctrl &= ~SECONDARY_EXEC_SHADOW_VMCS;
1317 * TSC_MULTIPLIER = 0x00002032,
1319 vmcs_conf->cpu_based_2nd_exec_ctrl &= ~SECONDARY_EXEC_TSC_SCALING;
1322 * PLE_GAP = 0x00004020,
1323 * PLE_WINDOW = 0x00004022,
1325 vmcs_conf->cpu_based_2nd_exec_ctrl &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
1328 * VMX_PREEMPTION_TIMER_VALUE = 0x0000482E,
1330 vmcs_conf->pin_based_exec_ctrl &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
1333 * GUEST_IA32_PERF_GLOBAL_CTRL = 0x00002808,
1334 * HOST_IA32_PERF_GLOBAL_CTRL = 0x00002c04,
1336 vmcs_conf->vmexit_ctrl &= ~VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL;
1337 vmcs_conf->vmentry_ctrl &= ~VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL;
1340 * Currently unsupported in KVM:
1341 * GUEST_IA32_RTIT_CTL = 0x00002814,
1344 #else /* !IS_ENABLED(CONFIG_HYPERV) */
1345 static inline void evmcs_write64(unsigned long field, u64 value) {}
1346 static inline void evmcs_write32(unsigned long field, u32 value) {}
1347 static inline void evmcs_write16(unsigned long field, u16 value) {}
1348 static inline u64 evmcs_read64(unsigned long field) { return 0; }
1349 static inline u32 evmcs_read32(unsigned long field) { return 0; }
1350 static inline u16 evmcs_read16(unsigned long field) { return 0; }
1351 static inline void evmcs_load(u64 phys_addr) {}
1352 static inline void evmcs_sanitize_exec_ctrls(struct vmcs_config *vmcs_conf) {}
1353 static inline void evmcs_touch_msr_bitmap(void) {}
1354 #endif /* IS_ENABLED(CONFIG_HYPERV) */
1356 static inline bool is_exception_n(u32 intr_info, u8 vector)
1358 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
1359 INTR_INFO_VALID_MASK)) ==
1360 (INTR_TYPE_HARD_EXCEPTION | vector | INTR_INFO_VALID_MASK);
1363 static inline bool is_debug(u32 intr_info)
1365 return is_exception_n(intr_info, DB_VECTOR);
1368 static inline bool is_breakpoint(u32 intr_info)
1370 return is_exception_n(intr_info, BP_VECTOR);
1373 static inline bool is_page_fault(u32 intr_info)
1375 return is_exception_n(intr_info, PF_VECTOR);
1378 static inline bool is_no_device(u32 intr_info)
1380 return is_exception_n(intr_info, NM_VECTOR);
1383 static inline bool is_invalid_opcode(u32 intr_info)
1385 return is_exception_n(intr_info, UD_VECTOR);
1388 static inline bool is_gp_fault(u32 intr_info)
1390 return is_exception_n(intr_info, GP_VECTOR);
1393 static inline bool is_external_interrupt(u32 intr_info)
1395 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
1396 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
1399 static inline bool is_machine_check(u32 intr_info)
1401 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
1402 INTR_INFO_VALID_MASK)) ==
1403 (INTR_TYPE_HARD_EXCEPTION | MC_VECTOR | INTR_INFO_VALID_MASK);
1406 /* Undocumented: icebp/int1 */
1407 static inline bool is_icebp(u32 intr_info)
1409 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
1410 == (INTR_TYPE_PRIV_SW_EXCEPTION | INTR_INFO_VALID_MASK);
1413 static inline bool cpu_has_vmx_msr_bitmap(void)
1415 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS;
1418 static inline bool cpu_has_vmx_tpr_shadow(void)
1420 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW;
1423 static inline bool cpu_need_tpr_shadow(struct kvm_vcpu *vcpu)
1425 return cpu_has_vmx_tpr_shadow() && lapic_in_kernel(vcpu);
1428 static inline bool cpu_has_secondary_exec_ctrls(void)
1430 return vmcs_config.cpu_based_exec_ctrl &
1431 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
1434 static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
1436 return vmcs_config.cpu_based_2nd_exec_ctrl &
1437 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
1440 static inline bool cpu_has_vmx_virtualize_x2apic_mode(void)
1442 return vmcs_config.cpu_based_2nd_exec_ctrl &
1443 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
1446 static inline bool cpu_has_vmx_apic_register_virt(void)
1448 return vmcs_config.cpu_based_2nd_exec_ctrl &
1449 SECONDARY_EXEC_APIC_REGISTER_VIRT;
1452 static inline bool cpu_has_vmx_virtual_intr_delivery(void)
1454 return vmcs_config.cpu_based_2nd_exec_ctrl &
1455 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY;
1459 * Comment's format: document - errata name - stepping - processor name.
1461 * https://www.virtualbox.org/svn/vbox/trunk/src/VBox/VMM/VMMR0/HMR0.cpp
1463 static u32 vmx_preemption_cpu_tfms[] = {
1464 /* 323344.pdf - BA86 - D0 - Xeon 7500 Series */
1466 /* 323056.pdf - AAX65 - C2 - Xeon L3406 */
1467 /* 322814.pdf - AAT59 - C2 - i7-600, i5-500, i5-400 and i3-300 Mobile */
1468 /* 322911.pdf - AAU65 - C2 - i5-600, i3-500 Desktop and Pentium G6950 */
1470 /* 322911.pdf - AAU65 - K0 - i5-600, i3-500 Desktop and Pentium G6950 */
1472 /* 322373.pdf - AAO95 - B1 - Xeon 3400 Series */
1473 /* 322166.pdf - AAN92 - B1 - i7-800 and i5-700 Desktop */
1475 * 320767.pdf - AAP86 - B1 -
1476 * i7-900 Mobile Extreme, i7-800 and i7-700 Mobile
1479 /* 321333.pdf - AAM126 - C0 - Xeon 3500 */
1481 /* 321333.pdf - AAM126 - C1 - Xeon 3500 */
1483 /* 320836.pdf - AAJ124 - C0 - i7-900 Desktop Extreme and i7-900 Desktop */
1485 /* 321333.pdf - AAM126 - D0 - Xeon 3500 */
1486 /* 321324.pdf - AAK139 - D0 - Xeon 5500 */
1487 /* 320836.pdf - AAJ124 - D0 - i7-900 Extreme and i7-900 Desktop */
1491 static inline bool cpu_has_broken_vmx_preemption_timer(void)
1493 u32 eax = cpuid_eax(0x00000001), i;
1495 /* Clear the reserved bits */
1496 eax &= ~(0x3U << 14 | 0xfU << 28);
1497 for (i = 0; i < ARRAY_SIZE(vmx_preemption_cpu_tfms); i++)
1498 if (eax == vmx_preemption_cpu_tfms[i])
1504 static inline bool cpu_has_vmx_preemption_timer(void)
1506 return vmcs_config.pin_based_exec_ctrl &
1507 PIN_BASED_VMX_PREEMPTION_TIMER;
1510 static inline bool cpu_has_vmx_posted_intr(void)
1512 return IS_ENABLED(CONFIG_X86_LOCAL_APIC) &&
1513 vmcs_config.pin_based_exec_ctrl & PIN_BASED_POSTED_INTR;
1516 static inline bool cpu_has_vmx_apicv(void)
1518 return cpu_has_vmx_apic_register_virt() &&
1519 cpu_has_vmx_virtual_intr_delivery() &&
1520 cpu_has_vmx_posted_intr();
1523 static inline bool cpu_has_vmx_flexpriority(void)
1525 return cpu_has_vmx_tpr_shadow() &&
1526 cpu_has_vmx_virtualize_apic_accesses();
1529 static inline bool cpu_has_vmx_ept_execute_only(void)
1531 return vmx_capability.ept & VMX_EPT_EXECUTE_ONLY_BIT;
1534 static inline bool cpu_has_vmx_ept_2m_page(void)
1536 return vmx_capability.ept & VMX_EPT_2MB_PAGE_BIT;
1539 static inline bool cpu_has_vmx_ept_1g_page(void)
1541 return vmx_capability.ept & VMX_EPT_1GB_PAGE_BIT;
1544 static inline bool cpu_has_vmx_ept_4levels(void)
1546 return vmx_capability.ept & VMX_EPT_PAGE_WALK_4_BIT;
1549 static inline bool cpu_has_vmx_ept_mt_wb(void)
1551 return vmx_capability.ept & VMX_EPTP_WB_BIT;
1554 static inline bool cpu_has_vmx_ept_5levels(void)
1556 return vmx_capability.ept & VMX_EPT_PAGE_WALK_5_BIT;
1559 static inline bool cpu_has_vmx_ept_ad_bits(void)
1561 return vmx_capability.ept & VMX_EPT_AD_BIT;
1564 static inline bool cpu_has_vmx_invept_context(void)
1566 return vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT;
1569 static inline bool cpu_has_vmx_invept_global(void)
1571 return vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT;
1574 static inline bool cpu_has_vmx_invvpid_individual_addr(void)
1576 return vmx_capability.vpid & VMX_VPID_EXTENT_INDIVIDUAL_ADDR_BIT;
1579 static inline bool cpu_has_vmx_invvpid_single(void)
1581 return vmx_capability.vpid & VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT;
1584 static inline bool cpu_has_vmx_invvpid_global(void)
1586 return vmx_capability.vpid & VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT;
1589 static inline bool cpu_has_vmx_invvpid(void)
1591 return vmx_capability.vpid & VMX_VPID_INVVPID_BIT;
1594 static inline bool cpu_has_vmx_ept(void)
1596 return vmcs_config.cpu_based_2nd_exec_ctrl &
1597 SECONDARY_EXEC_ENABLE_EPT;
1600 static inline bool cpu_has_vmx_unrestricted_guest(void)
1602 return vmcs_config.cpu_based_2nd_exec_ctrl &
1603 SECONDARY_EXEC_UNRESTRICTED_GUEST;
1606 static inline bool cpu_has_vmx_ple(void)
1608 return vmcs_config.cpu_based_2nd_exec_ctrl &
1609 SECONDARY_EXEC_PAUSE_LOOP_EXITING;
1612 static inline bool cpu_has_vmx_basic_inout(void)
1614 return (((u64)vmcs_config.basic_cap << 32) & VMX_BASIC_INOUT);
1617 static inline bool cpu_need_virtualize_apic_accesses(struct kvm_vcpu *vcpu)
1619 return flexpriority_enabled && lapic_in_kernel(vcpu);
1622 static inline bool cpu_has_vmx_vpid(void)
1624 return vmcs_config.cpu_based_2nd_exec_ctrl &
1625 SECONDARY_EXEC_ENABLE_VPID;
1628 static inline bool cpu_has_vmx_rdtscp(void)
1630 return vmcs_config.cpu_based_2nd_exec_ctrl &
1631 SECONDARY_EXEC_RDTSCP;
1634 static inline bool cpu_has_vmx_invpcid(void)
1636 return vmcs_config.cpu_based_2nd_exec_ctrl &
1637 SECONDARY_EXEC_ENABLE_INVPCID;
1640 static inline bool cpu_has_virtual_nmis(void)
1642 return vmcs_config.pin_based_exec_ctrl & PIN_BASED_VIRTUAL_NMIS;
1645 static inline bool cpu_has_vmx_wbinvd_exit(void)
1647 return vmcs_config.cpu_based_2nd_exec_ctrl &
1648 SECONDARY_EXEC_WBINVD_EXITING;
1651 static inline bool cpu_has_vmx_shadow_vmcs(void)
1654 rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
1655 /* check if the cpu supports writing r/o exit information fields */
1656 if (!(vmx_msr & MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS))
1659 return vmcs_config.cpu_based_2nd_exec_ctrl &
1660 SECONDARY_EXEC_SHADOW_VMCS;
1663 static inline bool cpu_has_vmx_pml(void)
1665 return vmcs_config.cpu_based_2nd_exec_ctrl & SECONDARY_EXEC_ENABLE_PML;
1668 static inline bool cpu_has_vmx_tsc_scaling(void)
1670 return vmcs_config.cpu_based_2nd_exec_ctrl &
1671 SECONDARY_EXEC_TSC_SCALING;
1674 static inline bool cpu_has_vmx_vmfunc(void)
1676 return vmcs_config.cpu_based_2nd_exec_ctrl &
1677 SECONDARY_EXEC_ENABLE_VMFUNC;
1680 static bool vmx_umip_emulated(void)
1682 return vmcs_config.cpu_based_2nd_exec_ctrl &
1683 SECONDARY_EXEC_DESC;
1686 static inline bool report_flexpriority(void)
1688 return flexpriority_enabled;
1691 static inline unsigned nested_cpu_vmx_misc_cr3_count(struct kvm_vcpu *vcpu)
1693 return vmx_misc_cr3_count(to_vmx(vcpu)->nested.msrs.misc_low);
1697 * Do the virtual VMX capability MSRs specify that L1 can use VMWRITE
1698 * to modify any valid field of the VMCS, or are the VM-exit
1699 * information fields read-only?
1701 static inline bool nested_cpu_has_vmwrite_any_field(struct kvm_vcpu *vcpu)
1703 return to_vmx(vcpu)->nested.msrs.misc_low &
1704 MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS;
1707 static inline bool nested_cpu_has_zero_length_injection(struct kvm_vcpu *vcpu)
1709 return to_vmx(vcpu)->nested.msrs.misc_low & VMX_MISC_ZERO_LEN_INS;
1712 static inline bool nested_cpu_supports_monitor_trap_flag(struct kvm_vcpu *vcpu)
1714 return to_vmx(vcpu)->nested.msrs.procbased_ctls_high &
1715 CPU_BASED_MONITOR_TRAP_FLAG;
1718 static inline bool nested_cpu_has(struct vmcs12 *vmcs12, u32 bit)
1720 return vmcs12->cpu_based_vm_exec_control & bit;
1723 static inline bool nested_cpu_has2(struct vmcs12 *vmcs12, u32 bit)
1725 return (vmcs12->cpu_based_vm_exec_control &
1726 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
1727 (vmcs12->secondary_vm_exec_control & bit);
1730 static inline bool nested_cpu_has_preemption_timer(struct vmcs12 *vmcs12)
1732 return vmcs12->pin_based_vm_exec_control &
1733 PIN_BASED_VMX_PREEMPTION_TIMER;
1736 static inline bool nested_cpu_has_nmi_exiting(struct vmcs12 *vmcs12)
1738 return vmcs12->pin_based_vm_exec_control & PIN_BASED_NMI_EXITING;
1741 static inline bool nested_cpu_has_virtual_nmis(struct vmcs12 *vmcs12)
1743 return vmcs12->pin_based_vm_exec_control & PIN_BASED_VIRTUAL_NMIS;
1746 static inline int nested_cpu_has_ept(struct vmcs12 *vmcs12)
1748 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_EPT);
1751 static inline bool nested_cpu_has_xsaves(struct vmcs12 *vmcs12)
1753 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES);
1756 static inline bool nested_cpu_has_pml(struct vmcs12 *vmcs12)
1758 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_PML);
1761 static inline bool nested_cpu_has_virt_x2apic_mode(struct vmcs12 *vmcs12)
1763 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE);
1766 static inline bool nested_cpu_has_vpid(struct vmcs12 *vmcs12)
1768 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_VPID);
1771 static inline bool nested_cpu_has_apic_reg_virt(struct vmcs12 *vmcs12)
1773 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_APIC_REGISTER_VIRT);
1776 static inline bool nested_cpu_has_vid(struct vmcs12 *vmcs12)
1778 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
1781 static inline bool nested_cpu_has_posted_intr(struct vmcs12 *vmcs12)
1783 return vmcs12->pin_based_vm_exec_control & PIN_BASED_POSTED_INTR;
1786 static inline bool nested_cpu_has_vmfunc(struct vmcs12 *vmcs12)
1788 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_VMFUNC);
1791 static inline bool nested_cpu_has_eptp_switching(struct vmcs12 *vmcs12)
1793 return nested_cpu_has_vmfunc(vmcs12) &&
1794 (vmcs12->vm_function_control &
1795 VMX_VMFUNC_EPTP_SWITCHING);
1798 static inline bool is_nmi(u32 intr_info)
1800 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
1801 == (INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK);
1804 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
1806 unsigned long exit_qualification);
1807 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
1808 struct vmcs12 *vmcs12,
1809 u32 reason, unsigned long qualification);
1811 static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
1815 for (i = 0; i < vmx->nmsrs; ++i)
1816 if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
1821 static inline void __invvpid(int ext, u16 vpid, gva_t gva)
1827 } operand = { vpid, 0, gva };
1829 asm volatile (__ex(ASM_VMX_INVVPID)
1830 /* CF==1 or ZF==1 --> rc = -1 */
1831 "; ja 1f ; ud2 ; 1:"
1832 : : "a"(&operand), "c"(ext) : "cc", "memory");
1835 static inline void __invept(int ext, u64 eptp, gpa_t gpa)
1839 } operand = {eptp, gpa};
1841 asm volatile (__ex(ASM_VMX_INVEPT)
1842 /* CF==1 or ZF==1 --> rc = -1 */
1843 "; ja 1f ; ud2 ; 1:\n"
1844 : : "a" (&operand), "c" (ext) : "cc", "memory");
1847 static struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
1851 i = __find_msr_index(vmx, msr);
1853 return &vmx->guest_msrs[i];
1857 static void vmcs_clear(struct vmcs *vmcs)
1859 u64 phys_addr = __pa(vmcs);
1862 asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0"
1863 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
1866 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
1870 static inline void loaded_vmcs_init(struct loaded_vmcs *loaded_vmcs)
1872 vmcs_clear(loaded_vmcs->vmcs);
1873 if (loaded_vmcs->shadow_vmcs && loaded_vmcs->launched)
1874 vmcs_clear(loaded_vmcs->shadow_vmcs);
1875 loaded_vmcs->cpu = -1;
1876 loaded_vmcs->launched = 0;
1879 static void vmcs_load(struct vmcs *vmcs)
1881 u64 phys_addr = __pa(vmcs);
1884 if (static_branch_unlikely(&enable_evmcs))
1885 return evmcs_load(phys_addr);
1887 asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0"
1888 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
1891 printk(KERN_ERR "kvm: vmptrld %p/%llx failed\n",
1895 #ifdef CONFIG_KEXEC_CORE
1897 * This bitmap is used to indicate whether the vmclear
1898 * operation is enabled on all cpus. All disabled by
1901 static cpumask_t crash_vmclear_enabled_bitmap = CPU_MASK_NONE;
1903 static inline void crash_enable_local_vmclear(int cpu)
1905 cpumask_set_cpu(cpu, &crash_vmclear_enabled_bitmap);
1908 static inline void crash_disable_local_vmclear(int cpu)
1910 cpumask_clear_cpu(cpu, &crash_vmclear_enabled_bitmap);
1913 static inline int crash_local_vmclear_enabled(int cpu)
1915 return cpumask_test_cpu(cpu, &crash_vmclear_enabled_bitmap);
1918 static void crash_vmclear_local_loaded_vmcss(void)
1920 int cpu = raw_smp_processor_id();
1921 struct loaded_vmcs *v;
1923 if (!crash_local_vmclear_enabled(cpu))
1926 list_for_each_entry(v, &per_cpu(loaded_vmcss_on_cpu, cpu),
1927 loaded_vmcss_on_cpu_link)
1928 vmcs_clear(v->vmcs);
1931 static inline void crash_enable_local_vmclear(int cpu) { }
1932 static inline void crash_disable_local_vmclear(int cpu) { }
1933 #endif /* CONFIG_KEXEC_CORE */
1935 static void __loaded_vmcs_clear(void *arg)
1937 struct loaded_vmcs *loaded_vmcs = arg;
1938 int cpu = raw_smp_processor_id();
1940 if (loaded_vmcs->cpu != cpu)
1941 return; /* vcpu migration can race with cpu offline */
1942 if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
1943 per_cpu(current_vmcs, cpu) = NULL;
1944 crash_disable_local_vmclear(cpu);
1945 list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
1948 * we should ensure updating loaded_vmcs->loaded_vmcss_on_cpu_link
1949 * is before setting loaded_vmcs->vcpu to -1 which is done in
1950 * loaded_vmcs_init. Otherwise, other cpu can see vcpu = -1 fist
1951 * then adds the vmcs into percpu list before it is deleted.
1955 loaded_vmcs_init(loaded_vmcs);
1956 crash_enable_local_vmclear(cpu);
1959 static void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
1961 int cpu = loaded_vmcs->cpu;
1964 smp_call_function_single(cpu,
1965 __loaded_vmcs_clear, loaded_vmcs, 1);
1968 static inline void vpid_sync_vcpu_single(int vpid)
1973 if (cpu_has_vmx_invvpid_single())
1974 __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vpid, 0);
1977 static inline void vpid_sync_vcpu_global(void)
1979 if (cpu_has_vmx_invvpid_global())
1980 __invvpid(VMX_VPID_EXTENT_ALL_CONTEXT, 0, 0);
1983 static inline void vpid_sync_context(int vpid)
1985 if (cpu_has_vmx_invvpid_single())
1986 vpid_sync_vcpu_single(vpid);
1988 vpid_sync_vcpu_global();
1991 static inline void ept_sync_global(void)
1993 __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0);
1996 static inline void ept_sync_context(u64 eptp)
1998 if (cpu_has_vmx_invept_context())
1999 __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0);
2004 static __always_inline void vmcs_check16(unsigned long field)
2006 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2000,
2007 "16-bit accessor invalid for 64-bit field");
2008 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2001,
2009 "16-bit accessor invalid for 64-bit high field");
2010 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x4000,
2011 "16-bit accessor invalid for 32-bit high field");
2012 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x6000,
2013 "16-bit accessor invalid for natural width field");
2016 static __always_inline void vmcs_check32(unsigned long field)
2018 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0,
2019 "32-bit accessor invalid for 16-bit field");
2020 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x6000,
2021 "32-bit accessor invalid for natural width field");
2024 static __always_inline void vmcs_check64(unsigned long field)
2026 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0,
2027 "64-bit accessor invalid for 16-bit field");
2028 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2001,
2029 "64-bit accessor invalid for 64-bit high field");
2030 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x4000,
2031 "64-bit accessor invalid for 32-bit field");
2032 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x6000,
2033 "64-bit accessor invalid for natural width field");
2036 static __always_inline void vmcs_checkl(unsigned long field)
2038 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0,
2039 "Natural width accessor invalid for 16-bit field");
2040 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2000,
2041 "Natural width accessor invalid for 64-bit field");
2042 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2001,
2043 "Natural width accessor invalid for 64-bit high field");
2044 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x4000,
2045 "Natural width accessor invalid for 32-bit field");
2048 static __always_inline unsigned long __vmcs_readl(unsigned long field)
2050 unsigned long value;
2052 asm volatile (__ex_clear(ASM_VMX_VMREAD_RDX_RAX, "%0")
2053 : "=a"(value) : "d"(field) : "cc");
2057 static __always_inline u16 vmcs_read16(unsigned long field)
2059 vmcs_check16(field);
2060 if (static_branch_unlikely(&enable_evmcs))
2061 return evmcs_read16(field);
2062 return __vmcs_readl(field);
2065 static __always_inline u32 vmcs_read32(unsigned long field)
2067 vmcs_check32(field);
2068 if (static_branch_unlikely(&enable_evmcs))
2069 return evmcs_read32(field);
2070 return __vmcs_readl(field);
2073 static __always_inline u64 vmcs_read64(unsigned long field)
2075 vmcs_check64(field);
2076 if (static_branch_unlikely(&enable_evmcs))
2077 return evmcs_read64(field);
2078 #ifdef CONFIG_X86_64
2079 return __vmcs_readl(field);
2081 return __vmcs_readl(field) | ((u64)__vmcs_readl(field+1) << 32);
2085 static __always_inline unsigned long vmcs_readl(unsigned long field)
2088 if (static_branch_unlikely(&enable_evmcs))
2089 return evmcs_read64(field);
2090 return __vmcs_readl(field);
2093 static noinline void vmwrite_error(unsigned long field, unsigned long value)
2095 printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
2096 field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
2100 static __always_inline void __vmcs_writel(unsigned long field, unsigned long value)
2104 asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0"
2105 : "=q"(error) : "a"(value), "d"(field) : "cc");
2106 if (unlikely(error))
2107 vmwrite_error(field, value);
2110 static __always_inline void vmcs_write16(unsigned long field, u16 value)
2112 vmcs_check16(field);
2113 if (static_branch_unlikely(&enable_evmcs))
2114 return evmcs_write16(field, value);
2116 __vmcs_writel(field, value);
2119 static __always_inline void vmcs_write32(unsigned long field, u32 value)
2121 vmcs_check32(field);
2122 if (static_branch_unlikely(&enable_evmcs))
2123 return evmcs_write32(field, value);
2125 __vmcs_writel(field, value);
2128 static __always_inline void vmcs_write64(unsigned long field, u64 value)
2130 vmcs_check64(field);
2131 if (static_branch_unlikely(&enable_evmcs))
2132 return evmcs_write64(field, value);
2134 __vmcs_writel(field, value);
2135 #ifndef CONFIG_X86_64
2137 __vmcs_writel(field+1, value >> 32);
2141 static __always_inline void vmcs_writel(unsigned long field, unsigned long value)
2144 if (static_branch_unlikely(&enable_evmcs))
2145 return evmcs_write64(field, value);
2147 __vmcs_writel(field, value);
2150 static __always_inline void vmcs_clear_bits(unsigned long field, u32 mask)
2152 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x2000,
2153 "vmcs_clear_bits does not support 64-bit fields");
2154 if (static_branch_unlikely(&enable_evmcs))
2155 return evmcs_write32(field, evmcs_read32(field) & ~mask);
2157 __vmcs_writel(field, __vmcs_readl(field) & ~mask);
2160 static __always_inline void vmcs_set_bits(unsigned long field, u32 mask)
2162 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x2000,
2163 "vmcs_set_bits does not support 64-bit fields");
2164 if (static_branch_unlikely(&enable_evmcs))
2165 return evmcs_write32(field, evmcs_read32(field) | mask);
2167 __vmcs_writel(field, __vmcs_readl(field) | mask);
2170 static inline void vm_entry_controls_reset_shadow(struct vcpu_vmx *vmx)
2172 vmx->vm_entry_controls_shadow = vmcs_read32(VM_ENTRY_CONTROLS);
2175 static inline void vm_entry_controls_init(struct vcpu_vmx *vmx, u32 val)
2177 vmcs_write32(VM_ENTRY_CONTROLS, val);
2178 vmx->vm_entry_controls_shadow = val;
2181 static inline void vm_entry_controls_set(struct vcpu_vmx *vmx, u32 val)
2183 if (vmx->vm_entry_controls_shadow != val)
2184 vm_entry_controls_init(vmx, val);
2187 static inline u32 vm_entry_controls_get(struct vcpu_vmx *vmx)
2189 return vmx->vm_entry_controls_shadow;
2193 static inline void vm_entry_controls_setbit(struct vcpu_vmx *vmx, u32 val)
2195 vm_entry_controls_set(vmx, vm_entry_controls_get(vmx) | val);
2198 static inline void vm_entry_controls_clearbit(struct vcpu_vmx *vmx, u32 val)
2200 vm_entry_controls_set(vmx, vm_entry_controls_get(vmx) & ~val);
2203 static inline void vm_exit_controls_reset_shadow(struct vcpu_vmx *vmx)
2205 vmx->vm_exit_controls_shadow = vmcs_read32(VM_EXIT_CONTROLS);
2208 static inline void vm_exit_controls_init(struct vcpu_vmx *vmx, u32 val)
2210 vmcs_write32(VM_EXIT_CONTROLS, val);
2211 vmx->vm_exit_controls_shadow = val;
2214 static inline void vm_exit_controls_set(struct vcpu_vmx *vmx, u32 val)
2216 if (vmx->vm_exit_controls_shadow != val)
2217 vm_exit_controls_init(vmx, val);
2220 static inline u32 vm_exit_controls_get(struct vcpu_vmx *vmx)
2222 return vmx->vm_exit_controls_shadow;
2226 static inline void vm_exit_controls_setbit(struct vcpu_vmx *vmx, u32 val)
2228 vm_exit_controls_set(vmx, vm_exit_controls_get(vmx) | val);
2231 static inline void vm_exit_controls_clearbit(struct vcpu_vmx *vmx, u32 val)
2233 vm_exit_controls_set(vmx, vm_exit_controls_get(vmx) & ~val);
2236 static void vmx_segment_cache_clear(struct vcpu_vmx *vmx)
2238 vmx->segment_cache.bitmask = 0;
2241 static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
2245 u32 mask = 1 << (seg * SEG_FIELD_NR + field);
2247 if (!(vmx->vcpu.arch.regs_avail & (1 << VCPU_EXREG_SEGMENTS))) {
2248 vmx->vcpu.arch.regs_avail |= (1 << VCPU_EXREG_SEGMENTS);
2249 vmx->segment_cache.bitmask = 0;
2251 ret = vmx->segment_cache.bitmask & mask;
2252 vmx->segment_cache.bitmask |= mask;
2256 static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
2258 u16 *p = &vmx->segment_cache.seg[seg].selector;
2260 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
2261 *p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
2265 static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
2267 ulong *p = &vmx->segment_cache.seg[seg].base;
2269 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
2270 *p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
2274 static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
2276 u32 *p = &vmx->segment_cache.seg[seg].limit;
2278 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
2279 *p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
2283 static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
2285 u32 *p = &vmx->segment_cache.seg[seg].ar;
2287 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
2288 *p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
2292 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
2296 eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
2297 (1u << DB_VECTOR) | (1u << AC_VECTOR);
2299 * Guest access to VMware backdoor ports could legitimately
2300 * trigger #GP because of TSS I/O permission bitmap.
2301 * We intercept those #GP and allow access to them anyway
2304 if (enable_vmware_backdoor)
2305 eb |= (1u << GP_VECTOR);
2306 if ((vcpu->guest_debug &
2307 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
2308 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
2309 eb |= 1u << BP_VECTOR;
2310 if (to_vmx(vcpu)->rmode.vm86_active)
2313 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
2315 /* When we are running a nested L2 guest and L1 specified for it a
2316 * certain exception bitmap, we must trap the same exceptions and pass
2317 * them to L1. When running L2, we will only handle the exceptions
2318 * specified above if L1 did not want them.
2320 if (is_guest_mode(vcpu))
2321 eb |= get_vmcs12(vcpu)->exception_bitmap;
2323 vmcs_write32(EXCEPTION_BITMAP, eb);
2327 * Check if MSR is intercepted for currently loaded MSR bitmap.
2329 static bool msr_write_intercepted(struct kvm_vcpu *vcpu, u32 msr)
2331 unsigned long *msr_bitmap;
2332 int f = sizeof(unsigned long);
2334 if (!cpu_has_vmx_msr_bitmap())
2337 msr_bitmap = to_vmx(vcpu)->loaded_vmcs->msr_bitmap;
2339 if (msr <= 0x1fff) {
2340 return !!test_bit(msr, msr_bitmap + 0x800 / f);
2341 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
2343 return !!test_bit(msr, msr_bitmap + 0xc00 / f);
2350 * Check if MSR is intercepted for L01 MSR bitmap.
2352 static bool msr_write_intercepted_l01(struct kvm_vcpu *vcpu, u32 msr)
2354 unsigned long *msr_bitmap;
2355 int f = sizeof(unsigned long);
2357 if (!cpu_has_vmx_msr_bitmap())
2360 msr_bitmap = to_vmx(vcpu)->vmcs01.msr_bitmap;
2362 if (msr <= 0x1fff) {
2363 return !!test_bit(msr, msr_bitmap + 0x800 / f);
2364 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
2366 return !!test_bit(msr, msr_bitmap + 0xc00 / f);
2372 static void clear_atomic_switch_msr_special(struct vcpu_vmx *vmx,
2373 unsigned long entry, unsigned long exit)
2375 vm_entry_controls_clearbit(vmx, entry);
2376 vm_exit_controls_clearbit(vmx, exit);
2379 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
2382 struct msr_autoload *m = &vmx->msr_autoload;
2386 if (cpu_has_load_ia32_efer) {
2387 clear_atomic_switch_msr_special(vmx,
2388 VM_ENTRY_LOAD_IA32_EFER,
2389 VM_EXIT_LOAD_IA32_EFER);
2393 case MSR_CORE_PERF_GLOBAL_CTRL:
2394 if (cpu_has_load_perf_global_ctrl) {
2395 clear_atomic_switch_msr_special(vmx,
2396 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
2397 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
2403 for (i = 0; i < m->nr; ++i)
2404 if (m->guest[i].index == msr)
2410 m->guest[i] = m->guest[m->nr];
2411 m->host[i] = m->host[m->nr];
2412 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
2413 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
2416 static void add_atomic_switch_msr_special(struct vcpu_vmx *vmx,
2417 unsigned long entry, unsigned long exit,
2418 unsigned long guest_val_vmcs, unsigned long host_val_vmcs,
2419 u64 guest_val, u64 host_val)
2421 vmcs_write64(guest_val_vmcs, guest_val);
2422 vmcs_write64(host_val_vmcs, host_val);
2423 vm_entry_controls_setbit(vmx, entry);
2424 vm_exit_controls_setbit(vmx, exit);
2427 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
2428 u64 guest_val, u64 host_val)
2431 struct msr_autoload *m = &vmx->msr_autoload;
2435 if (cpu_has_load_ia32_efer) {
2436 add_atomic_switch_msr_special(vmx,
2437 VM_ENTRY_LOAD_IA32_EFER,
2438 VM_EXIT_LOAD_IA32_EFER,
2441 guest_val, host_val);
2445 case MSR_CORE_PERF_GLOBAL_CTRL:
2446 if (cpu_has_load_perf_global_ctrl) {
2447 add_atomic_switch_msr_special(vmx,
2448 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
2449 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL,
2450 GUEST_IA32_PERF_GLOBAL_CTRL,
2451 HOST_IA32_PERF_GLOBAL_CTRL,
2452 guest_val, host_val);
2456 case MSR_IA32_PEBS_ENABLE:
2457 /* PEBS needs a quiescent period after being disabled (to write
2458 * a record). Disabling PEBS through VMX MSR swapping doesn't
2459 * provide that period, so a CPU could write host's record into
2462 wrmsrl(MSR_IA32_PEBS_ENABLE, 0);
2465 for (i = 0; i < m->nr; ++i)
2466 if (m->guest[i].index == msr)
2469 if (i == NR_AUTOLOAD_MSRS) {
2470 printk_once(KERN_WARNING "Not enough msr switch entries. "
2471 "Can't add msr %x\n", msr);
2473 } else if (i == m->nr) {
2475 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
2476 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
2479 m->guest[i].index = msr;
2480 m->guest[i].value = guest_val;
2481 m->host[i].index = msr;
2482 m->host[i].value = host_val;
2485 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
2487 u64 guest_efer = vmx->vcpu.arch.efer;
2488 u64 ignore_bits = 0;
2492 * NX is needed to handle CR0.WP=1, CR4.SMEP=1. Testing
2493 * host CPUID is more efficient than testing guest CPUID
2494 * or CR4. Host SMEP is anyway a requirement for guest SMEP.
2496 if (boot_cpu_has(X86_FEATURE_SMEP))
2497 guest_efer |= EFER_NX;
2498 else if (!(guest_efer & EFER_NX))
2499 ignore_bits |= EFER_NX;
2503 * LMA and LME handled by hardware; SCE meaningless outside long mode.
2505 ignore_bits |= EFER_SCE;
2506 #ifdef CONFIG_X86_64
2507 ignore_bits |= EFER_LMA | EFER_LME;
2508 /* SCE is meaningful only in long mode on Intel */
2509 if (guest_efer & EFER_LMA)
2510 ignore_bits &= ~(u64)EFER_SCE;
2513 clear_atomic_switch_msr(vmx, MSR_EFER);
2516 * On EPT, we can't emulate NX, so we must switch EFER atomically.
2517 * On CPUs that support "load IA32_EFER", always switch EFER
2518 * atomically, since it's faster than switching it manually.
2520 if (cpu_has_load_ia32_efer ||
2521 (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX))) {
2522 if (!(guest_efer & EFER_LMA))
2523 guest_efer &= ~EFER_LME;
2524 if (guest_efer != host_efer)
2525 add_atomic_switch_msr(vmx, MSR_EFER,
2526 guest_efer, host_efer);
2529 guest_efer &= ~ignore_bits;
2530 guest_efer |= host_efer & ignore_bits;
2532 vmx->guest_msrs[efer_offset].data = guest_efer;
2533 vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
2539 #ifdef CONFIG_X86_32
2541 * On 32-bit kernels, VM exits still load the FS and GS bases from the
2542 * VMCS rather than the segment table. KVM uses this helper to figure
2543 * out the current bases to poke them into the VMCS before entry.
2545 static unsigned long segment_base(u16 selector)
2547 struct desc_struct *table;
2550 if (!(selector & ~SEGMENT_RPL_MASK))
2553 table = get_current_gdt_ro();
2555 if ((selector & SEGMENT_TI_MASK) == SEGMENT_LDT) {
2556 u16 ldt_selector = kvm_read_ldt();
2558 if (!(ldt_selector & ~SEGMENT_RPL_MASK))
2561 table = (struct desc_struct *)segment_base(ldt_selector);
2563 v = get_desc_base(&table[selector >> 3]);
2568 static void vmx_save_host_state(struct kvm_vcpu *vcpu)
2570 struct vcpu_vmx *vmx = to_vmx(vcpu);
2571 #ifdef CONFIG_X86_64
2572 int cpu = raw_smp_processor_id();
2573 unsigned long fs_base, kernel_gs_base;
2577 if (vmx->host_state.loaded)
2580 vmx->host_state.loaded = 1;
2582 * Set host fs and gs selectors. Unfortunately, 22.2.3 does not
2583 * allow segment selectors with cpl > 0 or ti == 1.
2585 vmx->host_state.ldt_sel = kvm_read_ldt();
2586 vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
2588 #ifdef CONFIG_X86_64
2589 if (likely(is_64bit_mm(current->mm))) {
2590 save_fsgs_for_kvm();
2591 vmx->host_state.fs_sel = current->thread.fsindex;
2592 vmx->host_state.gs_sel = current->thread.gsindex;
2593 fs_base = current->thread.fsbase;
2594 kernel_gs_base = current->thread.gsbase;
2597 savesegment(fs, vmx->host_state.fs_sel);
2598 savesegment(gs, vmx->host_state.gs_sel);
2599 #ifdef CONFIG_X86_64
2600 fs_base = read_msr(MSR_FS_BASE);
2601 kernel_gs_base = read_msr(MSR_KERNEL_GS_BASE);
2604 if (!(vmx->host_state.fs_sel & 7)) {
2605 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
2606 vmx->host_state.fs_reload_needed = 0;
2608 vmcs_write16(HOST_FS_SELECTOR, 0);
2609 vmx->host_state.fs_reload_needed = 1;
2611 if (!(vmx->host_state.gs_sel & 7))
2612 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
2614 vmcs_write16(HOST_GS_SELECTOR, 0);
2615 vmx->host_state.gs_ldt_reload_needed = 1;
2618 #ifdef CONFIG_X86_64
2619 savesegment(ds, vmx->host_state.ds_sel);
2620 savesegment(es, vmx->host_state.es_sel);
2622 vmcs_writel(HOST_FS_BASE, fs_base);
2623 vmcs_writel(HOST_GS_BASE, cpu_kernelmode_gs_base(cpu));
2625 vmx->msr_host_kernel_gs_base = kernel_gs_base;
2626 if (is_long_mode(&vmx->vcpu))
2627 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
2629 vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
2630 vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
2632 for (i = 0; i < vmx->save_nmsrs; ++i)
2633 kvm_set_shared_msr(vmx->guest_msrs[i].index,
2634 vmx->guest_msrs[i].data,
2635 vmx->guest_msrs[i].mask);
2638 static void __vmx_load_host_state(struct vcpu_vmx *vmx)
2640 if (!vmx->host_state.loaded)
2643 ++vmx->vcpu.stat.host_state_reload;
2644 vmx->host_state.loaded = 0;
2645 #ifdef CONFIG_X86_64
2646 if (is_long_mode(&vmx->vcpu))
2647 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
2649 if (vmx->host_state.gs_ldt_reload_needed) {
2650 kvm_load_ldt(vmx->host_state.ldt_sel);
2651 #ifdef CONFIG_X86_64
2652 load_gs_index(vmx->host_state.gs_sel);
2654 loadsegment(gs, vmx->host_state.gs_sel);
2657 if (vmx->host_state.fs_reload_needed)
2658 loadsegment(fs, vmx->host_state.fs_sel);
2659 #ifdef CONFIG_X86_64
2660 if (unlikely(vmx->host_state.ds_sel | vmx->host_state.es_sel)) {
2661 loadsegment(ds, vmx->host_state.ds_sel);
2662 loadsegment(es, vmx->host_state.es_sel);
2665 invalidate_tss_limit();
2666 #ifdef CONFIG_X86_64
2667 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
2669 load_fixmap_gdt(raw_smp_processor_id());
2672 static void vmx_load_host_state(struct vcpu_vmx *vmx)
2675 __vmx_load_host_state(vmx);
2679 static void vmx_vcpu_pi_load(struct kvm_vcpu *vcpu, int cpu)
2681 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
2682 struct pi_desc old, new;
2686 * In case of hot-plug or hot-unplug, we may have to undo
2687 * vmx_vcpu_pi_put even if there is no assigned device. And we
2688 * always keep PI.NDST up to date for simplicity: it makes the
2689 * code easier, and CPU migration is not a fast path.
2691 if (!pi_test_sn(pi_desc) && vcpu->cpu == cpu)
2695 * First handle the simple case where no cmpxchg is necessary; just
2696 * allow posting non-urgent interrupts.
2698 * If the 'nv' field is POSTED_INTR_WAKEUP_VECTOR, do not change
2699 * PI.NDST: pi_post_block will do it for us and the wakeup_handler
2700 * expects the VCPU to be on the blocked_vcpu_list that matches
2703 if (pi_desc->nv == POSTED_INTR_WAKEUP_VECTOR ||
2705 pi_clear_sn(pi_desc);
2709 /* The full case. */
2711 old.control = new.control = pi_desc->control;
2713 dest = cpu_physical_id(cpu);
2715 if (x2apic_enabled())
2718 new.ndst = (dest << 8) & 0xFF00;
2721 } while (cmpxchg64(&pi_desc->control, old.control,
2722 new.control) != old.control);
2725 static void decache_tsc_multiplier(struct vcpu_vmx *vmx)
2727 vmx->current_tsc_ratio = vmx->vcpu.arch.tsc_scaling_ratio;
2728 vmcs_write64(TSC_MULTIPLIER, vmx->current_tsc_ratio);
2732 * Switches to specified vcpu, until a matching vcpu_put(), but assumes
2733 * vcpu mutex is already taken.
2735 static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2737 struct vcpu_vmx *vmx = to_vmx(vcpu);
2738 bool already_loaded = vmx->loaded_vmcs->cpu == cpu;
2740 if (!already_loaded) {
2741 loaded_vmcs_clear(vmx->loaded_vmcs);
2742 local_irq_disable();
2743 crash_disable_local_vmclear(cpu);
2746 * Read loaded_vmcs->cpu should be before fetching
2747 * loaded_vmcs->loaded_vmcss_on_cpu_link.
2748 * See the comments in __loaded_vmcs_clear().
2752 list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
2753 &per_cpu(loaded_vmcss_on_cpu, cpu));
2754 crash_enable_local_vmclear(cpu);
2758 if (per_cpu(current_vmcs, cpu) != vmx->loaded_vmcs->vmcs) {
2759 per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
2760 vmcs_load(vmx->loaded_vmcs->vmcs);
2761 indirect_branch_prediction_barrier();
2764 if (!already_loaded) {
2765 void *gdt = get_current_gdt_ro();
2766 unsigned long sysenter_esp;
2768 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
2771 * Linux uses per-cpu TSS and GDT, so set these when switching
2772 * processors. See 22.2.4.
2774 vmcs_writel(HOST_TR_BASE,
2775 (unsigned long)&get_cpu_entry_area(cpu)->tss.x86_tss);
2776 vmcs_writel(HOST_GDTR_BASE, (unsigned long)gdt); /* 22.2.4 */
2779 * VM exits change the host TR limit to 0x67 after a VM
2780 * exit. This is okay, since 0x67 covers everything except
2781 * the IO bitmap and have have code to handle the IO bitmap
2782 * being lost after a VM exit.
2784 BUILD_BUG_ON(IO_BITMAP_OFFSET - 1 != 0x67);
2786 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
2787 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
2789 vmx->loaded_vmcs->cpu = cpu;
2792 /* Setup TSC multiplier */
2793 if (kvm_has_tsc_control &&
2794 vmx->current_tsc_ratio != vcpu->arch.tsc_scaling_ratio)
2795 decache_tsc_multiplier(vmx);
2797 vmx_vcpu_pi_load(vcpu, cpu);
2798 vmx->host_pkru = read_pkru();
2799 vmx->host_debugctlmsr = get_debugctlmsr();
2802 static void vmx_vcpu_pi_put(struct kvm_vcpu *vcpu)
2804 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
2806 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
2807 !irq_remapping_cap(IRQ_POSTING_CAP) ||
2808 !kvm_vcpu_apicv_active(vcpu))
2811 /* Set SN when the vCPU is preempted */
2812 if (vcpu->preempted)
2816 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
2818 vmx_vcpu_pi_put(vcpu);
2820 __vmx_load_host_state(to_vmx(vcpu));
2823 static bool emulation_required(struct kvm_vcpu *vcpu)
2825 return emulate_invalid_guest_state && !guest_state_valid(vcpu);
2828 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu);
2831 * Return the cr0 value that a nested guest would read. This is a combination
2832 * of the real cr0 used to run the guest (guest_cr0), and the bits shadowed by
2833 * its hypervisor (cr0_read_shadow).
2835 static inline unsigned long nested_read_cr0(struct vmcs12 *fields)
2837 return (fields->guest_cr0 & ~fields->cr0_guest_host_mask) |
2838 (fields->cr0_read_shadow & fields->cr0_guest_host_mask);
2840 static inline unsigned long nested_read_cr4(struct vmcs12 *fields)
2842 return (fields->guest_cr4 & ~fields->cr4_guest_host_mask) |
2843 (fields->cr4_read_shadow & fields->cr4_guest_host_mask);
2846 static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
2848 unsigned long rflags, save_rflags;
2850 if (!test_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail)) {
2851 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
2852 rflags = vmcs_readl(GUEST_RFLAGS);
2853 if (to_vmx(vcpu)->rmode.vm86_active) {
2854 rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
2855 save_rflags = to_vmx(vcpu)->rmode.save_rflags;
2856 rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
2858 to_vmx(vcpu)->rflags = rflags;
2860 return to_vmx(vcpu)->rflags;
2863 static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
2865 unsigned long old_rflags = vmx_get_rflags(vcpu);
2867 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
2868 to_vmx(vcpu)->rflags = rflags;
2869 if (to_vmx(vcpu)->rmode.vm86_active) {
2870 to_vmx(vcpu)->rmode.save_rflags = rflags;
2871 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
2873 vmcs_writel(GUEST_RFLAGS, rflags);
2875 if ((old_rflags ^ to_vmx(vcpu)->rflags) & X86_EFLAGS_VM)
2876 to_vmx(vcpu)->emulation_required = emulation_required(vcpu);
2879 static u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu)
2881 u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
2884 if (interruptibility & GUEST_INTR_STATE_STI)
2885 ret |= KVM_X86_SHADOW_INT_STI;
2886 if (interruptibility & GUEST_INTR_STATE_MOV_SS)
2887 ret |= KVM_X86_SHADOW_INT_MOV_SS;
2892 static void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
2894 u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
2895 u32 interruptibility = interruptibility_old;
2897 interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
2899 if (mask & KVM_X86_SHADOW_INT_MOV_SS)
2900 interruptibility |= GUEST_INTR_STATE_MOV_SS;
2901 else if (mask & KVM_X86_SHADOW_INT_STI)
2902 interruptibility |= GUEST_INTR_STATE_STI;
2904 if ((interruptibility != interruptibility_old))
2905 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
2908 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
2912 rip = kvm_rip_read(vcpu);
2913 rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
2914 kvm_rip_write(vcpu, rip);
2916 /* skipping an emulated instruction also counts */
2917 vmx_set_interrupt_shadow(vcpu, 0);
2920 static void nested_vmx_inject_exception_vmexit(struct kvm_vcpu *vcpu,
2921 unsigned long exit_qual)
2923 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2924 unsigned int nr = vcpu->arch.exception.nr;
2925 u32 intr_info = nr | INTR_INFO_VALID_MASK;
2927 if (vcpu->arch.exception.has_error_code) {
2928 vmcs12->vm_exit_intr_error_code = vcpu->arch.exception.error_code;
2929 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
2932 if (kvm_exception_is_soft(nr))
2933 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
2935 intr_info |= INTR_TYPE_HARD_EXCEPTION;
2937 if (!(vmcs12->idt_vectoring_info_field & VECTORING_INFO_VALID_MASK) &&
2938 vmx_get_nmi_mask(vcpu))
2939 intr_info |= INTR_INFO_UNBLOCK_NMI;
2941 nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI, intr_info, exit_qual);
2945 * KVM wants to inject page-faults which it got to the guest. This function
2946 * checks whether in a nested guest, we need to inject them to L1 or L2.
2948 static int nested_vmx_check_exception(struct kvm_vcpu *vcpu, unsigned long *exit_qual)
2950 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2951 unsigned int nr = vcpu->arch.exception.nr;
2953 if (nr == PF_VECTOR) {
2954 if (vcpu->arch.exception.nested_apf) {
2955 *exit_qual = vcpu->arch.apf.nested_apf_token;
2959 * FIXME: we must not write CR2 when L1 intercepts an L2 #PF exception.
2960 * The fix is to add the ancillary datum (CR2 or DR6) to structs
2961 * kvm_queued_exception and kvm_vcpu_events, so that CR2 and DR6
2962 * can be written only when inject_pending_event runs. This should be
2963 * conditional on a new capability---if the capability is disabled,
2964 * kvm_multiple_exception would write the ancillary information to
2965 * CR2 or DR6, for backwards ABI-compatibility.
2967 if (nested_vmx_is_page_fault_vmexit(vmcs12,
2968 vcpu->arch.exception.error_code)) {
2969 *exit_qual = vcpu->arch.cr2;
2973 if (vmcs12->exception_bitmap & (1u << nr)) {
2974 if (nr == DB_VECTOR)
2975 *exit_qual = vcpu->arch.dr6;
2985 static void vmx_clear_hlt(struct kvm_vcpu *vcpu)
2988 * Ensure that we clear the HLT state in the VMCS. We don't need to
2989 * explicitly skip the instruction because if the HLT state is set,
2990 * then the instruction is already executing and RIP has already been
2993 if (kvm_hlt_in_guest(vcpu->kvm) &&
2994 vmcs_read32(GUEST_ACTIVITY_STATE) == GUEST_ACTIVITY_HLT)
2995 vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
2998 static void vmx_queue_exception(struct kvm_vcpu *vcpu)
3000 struct vcpu_vmx *vmx = to_vmx(vcpu);
3001 unsigned nr = vcpu->arch.exception.nr;
3002 bool has_error_code = vcpu->arch.exception.has_error_code;
3003 u32 error_code = vcpu->arch.exception.error_code;
3004 u32 intr_info = nr | INTR_INFO_VALID_MASK;
3006 if (has_error_code) {
3007 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
3008 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
3011 if (vmx->rmode.vm86_active) {
3013 if (kvm_exception_is_soft(nr))
3014 inc_eip = vcpu->arch.event_exit_inst_len;
3015 if (kvm_inject_realmode_interrupt(vcpu, nr, inc_eip) != EMULATE_DONE)
3016 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3020 WARN_ON_ONCE(vmx->emulation_required);
3022 if (kvm_exception_is_soft(nr)) {
3023 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
3024 vmx->vcpu.arch.event_exit_inst_len);
3025 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
3027 intr_info |= INTR_TYPE_HARD_EXCEPTION;
3029 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
3031 vmx_clear_hlt(vcpu);
3034 static bool vmx_rdtscp_supported(void)
3036 return cpu_has_vmx_rdtscp();
3039 static bool vmx_invpcid_supported(void)
3041 return cpu_has_vmx_invpcid() && enable_ept;
3045 * Swap MSR entry in host/guest MSR entry array.
3047 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
3049 struct shared_msr_entry tmp;
3051 tmp = vmx->guest_msrs[to];
3052 vmx->guest_msrs[to] = vmx->guest_msrs[from];
3053 vmx->guest_msrs[from] = tmp;
3057 * Set up the vmcs to automatically save and restore system
3058 * msrs. Don't touch the 64-bit msrs if the guest is in legacy
3059 * mode, as fiddling with msrs is very expensive.
3061 static void setup_msrs(struct vcpu_vmx *vmx)
3063 int save_nmsrs, index;
3066 #ifdef CONFIG_X86_64
3067 if (is_long_mode(&vmx->vcpu)) {
3068 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
3070 move_msr_up(vmx, index, save_nmsrs++);
3071 index = __find_msr_index(vmx, MSR_LSTAR);
3073 move_msr_up(vmx, index, save_nmsrs++);
3074 index = __find_msr_index(vmx, MSR_CSTAR);
3076 move_msr_up(vmx, index, save_nmsrs++);
3077 index = __find_msr_index(vmx, MSR_TSC_AUX);
3078 if (index >= 0 && guest_cpuid_has(&vmx->vcpu, X86_FEATURE_RDTSCP))
3079 move_msr_up(vmx, index, save_nmsrs++);
3081 * MSR_STAR is only needed on long mode guests, and only
3082 * if efer.sce is enabled.
3084 index = __find_msr_index(vmx, MSR_STAR);
3085 if ((index >= 0) && (vmx->vcpu.arch.efer & EFER_SCE))
3086 move_msr_up(vmx, index, save_nmsrs++);
3089 index = __find_msr_index(vmx, MSR_EFER);
3090 if (index >= 0 && update_transition_efer(vmx, index))
3091 move_msr_up(vmx, index, save_nmsrs++);
3093 vmx->save_nmsrs = save_nmsrs;
3095 if (cpu_has_vmx_msr_bitmap())
3096 vmx_update_msr_bitmap(&vmx->vcpu);
3099 static u64 vmx_read_l1_tsc_offset(struct kvm_vcpu *vcpu)
3101 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3103 if (is_guest_mode(vcpu) &&
3104 (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING))
3105 return vcpu->arch.tsc_offset - vmcs12->tsc_offset;
3107 return vcpu->arch.tsc_offset;
3111 * writes 'offset' into guest's timestamp counter offset register
3113 static void vmx_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
3115 if (is_guest_mode(vcpu)) {
3117 * We're here if L1 chose not to trap WRMSR to TSC. According
3118 * to the spec, this should set L1's TSC; The offset that L1
3119 * set for L2 remains unchanged, and still needs to be added
3120 * to the newly set TSC to get L2's TSC.
3122 struct vmcs12 *vmcs12;
3123 /* recalculate vmcs02.TSC_OFFSET: */
3124 vmcs12 = get_vmcs12(vcpu);
3125 vmcs_write64(TSC_OFFSET, offset +
3126 (nested_cpu_has(vmcs12, CPU_BASED_USE_TSC_OFFSETING) ?
3127 vmcs12->tsc_offset : 0));
3129 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
3130 vmcs_read64(TSC_OFFSET), offset);
3131 vmcs_write64(TSC_OFFSET, offset);
3136 * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
3137 * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
3138 * all guests if the "nested" module option is off, and can also be disabled
3139 * for a single guest by disabling its VMX cpuid bit.
3141 static inline bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
3143 return nested && guest_cpuid_has(vcpu, X86_FEATURE_VMX);
3147 * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be
3148 * returned for the various VMX controls MSRs when nested VMX is enabled.
3149 * The same values should also be used to verify that vmcs12 control fields are
3150 * valid during nested entry from L1 to L2.
3151 * Each of these control msrs has a low and high 32-bit half: A low bit is on
3152 * if the corresponding bit in the (32-bit) control field *must* be on, and a
3153 * bit in the high half is on if the corresponding bit in the control field
3154 * may be on. See also vmx_control_verify().
3156 static void nested_vmx_setup_ctls_msrs(struct nested_vmx_msrs *msrs, bool apicv)
3159 memset(msrs, 0, sizeof(*msrs));
3164 * Note that as a general rule, the high half of the MSRs (bits in
3165 * the control fields which may be 1) should be initialized by the
3166 * intersection of the underlying hardware's MSR (i.e., features which
3167 * can be supported) and the list of features we want to expose -
3168 * because they are known to be properly supported in our code.
3169 * Also, usually, the low half of the MSRs (bits which must be 1) can
3170 * be set to 0, meaning that L1 may turn off any of these bits. The
3171 * reason is that if one of these bits is necessary, it will appear
3172 * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control
3173 * fields of vmcs01 and vmcs02, will turn these bits off - and
3174 * nested_vmx_exit_reflected() will not pass related exits to L1.
3175 * These rules have exceptions below.
3178 /* pin-based controls */
3179 rdmsr(MSR_IA32_VMX_PINBASED_CTLS,
3180 msrs->pinbased_ctls_low,
3181 msrs->pinbased_ctls_high);
3182 msrs->pinbased_ctls_low |=
3183 PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
3184 msrs->pinbased_ctls_high &=
3185 PIN_BASED_EXT_INTR_MASK |
3186 PIN_BASED_NMI_EXITING |
3187 PIN_BASED_VIRTUAL_NMIS |
3188 (apicv ? PIN_BASED_POSTED_INTR : 0);
3189 msrs->pinbased_ctls_high |=
3190 PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
3191 PIN_BASED_VMX_PREEMPTION_TIMER;
3194 rdmsr(MSR_IA32_VMX_EXIT_CTLS,
3195 msrs->exit_ctls_low,
3196 msrs->exit_ctls_high);
3197 msrs->exit_ctls_low =
3198 VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
3200 msrs->exit_ctls_high &=
3201 #ifdef CONFIG_X86_64
3202 VM_EXIT_HOST_ADDR_SPACE_SIZE |
3204 VM_EXIT_LOAD_IA32_PAT | VM_EXIT_SAVE_IA32_PAT;
3205 msrs->exit_ctls_high |=
3206 VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR |
3207 VM_EXIT_LOAD_IA32_EFER | VM_EXIT_SAVE_IA32_EFER |
3208 VM_EXIT_SAVE_VMX_PREEMPTION_TIMER | VM_EXIT_ACK_INTR_ON_EXIT;
3210 if (kvm_mpx_supported())
3211 msrs->exit_ctls_high |= VM_EXIT_CLEAR_BNDCFGS;
3213 /* We support free control of debug control saving. */
3214 msrs->exit_ctls_low &= ~VM_EXIT_SAVE_DEBUG_CONTROLS;
3216 /* entry controls */
3217 rdmsr(MSR_IA32_VMX_ENTRY_CTLS,
3218 msrs->entry_ctls_low,
3219 msrs->entry_ctls_high);
3220 msrs->entry_ctls_low =
3221 VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
3222 msrs->entry_ctls_high &=
3223 #ifdef CONFIG_X86_64
3224 VM_ENTRY_IA32E_MODE |
3226 VM_ENTRY_LOAD_IA32_PAT;
3227 msrs->entry_ctls_high |=
3228 (VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR | VM_ENTRY_LOAD_IA32_EFER);
3229 if (kvm_mpx_supported())
3230 msrs->entry_ctls_high |= VM_ENTRY_LOAD_BNDCFGS;
3232 /* We support free control of debug control loading. */
3233 msrs->entry_ctls_low &= ~VM_ENTRY_LOAD_DEBUG_CONTROLS;
3235 /* cpu-based controls */
3236 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS,
3237 msrs->procbased_ctls_low,
3238 msrs->procbased_ctls_high);
3239 msrs->procbased_ctls_low =
3240 CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
3241 msrs->procbased_ctls_high &=
3242 CPU_BASED_VIRTUAL_INTR_PENDING |
3243 CPU_BASED_VIRTUAL_NMI_PENDING | CPU_BASED_USE_TSC_OFFSETING |
3244 CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING |
3245 CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING |
3246 CPU_BASED_CR3_STORE_EXITING |
3247 #ifdef CONFIG_X86_64
3248 CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING |
3250 CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING |
3251 CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_TRAP_FLAG |
3252 CPU_BASED_MONITOR_EXITING | CPU_BASED_RDPMC_EXITING |
3253 CPU_BASED_RDTSC_EXITING | CPU_BASED_PAUSE_EXITING |
3254 CPU_BASED_TPR_SHADOW | CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
3256 * We can allow some features even when not supported by the
3257 * hardware. For example, L1 can specify an MSR bitmap - and we
3258 * can use it to avoid exits to L1 - even when L0 runs L2
3259 * without MSR bitmaps.
3261 msrs->procbased_ctls_high |=
3262 CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
3263 CPU_BASED_USE_MSR_BITMAPS;
3265 /* We support free control of CR3 access interception. */
3266 msrs->procbased_ctls_low &=
3267 ~(CPU_BASED_CR3_LOAD_EXITING | CPU_BASED_CR3_STORE_EXITING);
3270 * secondary cpu-based controls. Do not include those that
3271 * depend on CPUID bits, they are added later by vmx_cpuid_update.
3273 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
3274 msrs->secondary_ctls_low,
3275 msrs->secondary_ctls_high);
3276 msrs->secondary_ctls_low = 0;
3277 msrs->secondary_ctls_high &=
3278 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
3279 SECONDARY_EXEC_DESC |
3280 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
3281 SECONDARY_EXEC_APIC_REGISTER_VIRT |
3282 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
3283 SECONDARY_EXEC_WBINVD_EXITING;
3286 /* nested EPT: emulate EPT also to L1 */
3287 msrs->secondary_ctls_high |=
3288 SECONDARY_EXEC_ENABLE_EPT;
3289 msrs->ept_caps = VMX_EPT_PAGE_WALK_4_BIT |
3290 VMX_EPTP_WB_BIT | VMX_EPT_INVEPT_BIT;
3291 if (cpu_has_vmx_ept_execute_only())
3293 VMX_EPT_EXECUTE_ONLY_BIT;
3294 msrs->ept_caps &= vmx_capability.ept;
3295 msrs->ept_caps |= VMX_EPT_EXTENT_GLOBAL_BIT |
3296 VMX_EPT_EXTENT_CONTEXT_BIT | VMX_EPT_2MB_PAGE_BIT |
3297 VMX_EPT_1GB_PAGE_BIT;
3298 if (enable_ept_ad_bits) {
3299 msrs->secondary_ctls_high |=
3300 SECONDARY_EXEC_ENABLE_PML;
3301 msrs->ept_caps |= VMX_EPT_AD_BIT;
3305 if (cpu_has_vmx_vmfunc()) {
3306 msrs->secondary_ctls_high |=
3307 SECONDARY_EXEC_ENABLE_VMFUNC;
3309 * Advertise EPTP switching unconditionally
3310 * since we emulate it
3313 msrs->vmfunc_controls =
3314 VMX_VMFUNC_EPTP_SWITCHING;
3318 * Old versions of KVM use the single-context version without
3319 * checking for support, so declare that it is supported even
3320 * though it is treated as global context. The alternative is
3321 * not failing the single-context invvpid, and it is worse.
3324 msrs->secondary_ctls_high |=
3325 SECONDARY_EXEC_ENABLE_VPID;
3326 msrs->vpid_caps = VMX_VPID_INVVPID_BIT |
3327 VMX_VPID_EXTENT_SUPPORTED_MASK;
3330 if (enable_unrestricted_guest)
3331 msrs->secondary_ctls_high |=
3332 SECONDARY_EXEC_UNRESTRICTED_GUEST;
3334 /* miscellaneous data */
3335 rdmsr(MSR_IA32_VMX_MISC,
3338 msrs->misc_low &= VMX_MISC_SAVE_EFER_LMA;
3340 MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS |
3341 VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE |
3342 VMX_MISC_ACTIVITY_HLT;
3343 msrs->misc_high = 0;
3346 * This MSR reports some information about VMX support. We
3347 * should return information about the VMX we emulate for the
3348 * guest, and the VMCS structure we give it - not about the
3349 * VMX support of the underlying hardware.
3353 VMX_BASIC_TRUE_CTLS |
3354 ((u64)VMCS12_SIZE << VMX_BASIC_VMCS_SIZE_SHIFT) |
3355 (VMX_BASIC_MEM_TYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT);
3357 if (cpu_has_vmx_basic_inout())
3358 msrs->basic |= VMX_BASIC_INOUT;
3361 * These MSRs specify bits which the guest must keep fixed on
3362 * while L1 is in VMXON mode (in L1's root mode, or running an L2).
3363 * We picked the standard core2 setting.
3365 #define VMXON_CR0_ALWAYSON (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
3366 #define VMXON_CR4_ALWAYSON X86_CR4_VMXE
3367 msrs->cr0_fixed0 = VMXON_CR0_ALWAYSON;
3368 msrs->cr4_fixed0 = VMXON_CR4_ALWAYSON;
3370 /* These MSRs specify bits which the guest must keep fixed off. */
3371 rdmsrl(MSR_IA32_VMX_CR0_FIXED1, msrs->cr0_fixed1);
3372 rdmsrl(MSR_IA32_VMX_CR4_FIXED1, msrs->cr4_fixed1);
3374 /* highest index: VMX_PREEMPTION_TIMER_VALUE */
3375 msrs->vmcs_enum = VMCS12_MAX_FIELD_INDEX << 1;
3379 * if fixed0[i] == 1: val[i] must be 1
3380 * if fixed1[i] == 0: val[i] must be 0
3382 static inline bool fixed_bits_valid(u64 val, u64 fixed0, u64 fixed1)
3384 return ((val & fixed1) | fixed0) == val;
3387 static inline bool vmx_control_verify(u32 control, u32 low, u32 high)
3389 return fixed_bits_valid(control, low, high);
3392 static inline u64 vmx_control_msr(u32 low, u32 high)
3394 return low | ((u64)high << 32);
3397 static bool is_bitwise_subset(u64 superset, u64 subset, u64 mask)
3402 return (superset | subset) == superset;
3405 static int vmx_restore_vmx_basic(struct vcpu_vmx *vmx, u64 data)
3407 const u64 feature_and_reserved =
3408 /* feature (except bit 48; see below) */
3409 BIT_ULL(49) | BIT_ULL(54) | BIT_ULL(55) |
3411 BIT_ULL(31) | GENMASK_ULL(47, 45) | GENMASK_ULL(63, 56);
3412 u64 vmx_basic = vmx->nested.msrs.basic;
3414 if (!is_bitwise_subset(vmx_basic, data, feature_and_reserved))
3418 * KVM does not emulate a version of VMX that constrains physical
3419 * addresses of VMX structures (e.g. VMCS) to 32-bits.
3421 if (data & BIT_ULL(48))
3424 if (vmx_basic_vmcs_revision_id(vmx_basic) !=
3425 vmx_basic_vmcs_revision_id(data))
3428 if (vmx_basic_vmcs_size(vmx_basic) > vmx_basic_vmcs_size(data))
3431 vmx->nested.msrs.basic = data;
3436 vmx_restore_control_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
3441 switch (msr_index) {
3442 case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
3443 lowp = &vmx->nested.msrs.pinbased_ctls_low;
3444 highp = &vmx->nested.msrs.pinbased_ctls_high;
3446 case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
3447 lowp = &vmx->nested.msrs.procbased_ctls_low;
3448 highp = &vmx->nested.msrs.procbased_ctls_high;
3450 case MSR_IA32_VMX_TRUE_EXIT_CTLS:
3451 lowp = &vmx->nested.msrs.exit_ctls_low;
3452 highp = &vmx->nested.msrs.exit_ctls_high;
3454 case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
3455 lowp = &vmx->nested.msrs.entry_ctls_low;
3456 highp = &vmx->nested.msrs.entry_ctls_high;
3458 case MSR_IA32_VMX_PROCBASED_CTLS2:
3459 lowp = &vmx->nested.msrs.secondary_ctls_low;
3460 highp = &vmx->nested.msrs.secondary_ctls_high;
3466 supported = vmx_control_msr(*lowp, *highp);
3468 /* Check must-be-1 bits are still 1. */
3469 if (!is_bitwise_subset(data, supported, GENMASK_ULL(31, 0)))
3472 /* Check must-be-0 bits are still 0. */
3473 if (!is_bitwise_subset(supported, data, GENMASK_ULL(63, 32)))
3477 *highp = data >> 32;
3481 static int vmx_restore_vmx_misc(struct vcpu_vmx *vmx, u64 data)
3483 const u64 feature_and_reserved_bits =
3485 BIT_ULL(5) | GENMASK_ULL(8, 6) | BIT_ULL(14) | BIT_ULL(15) |
3486 BIT_ULL(28) | BIT_ULL(29) | BIT_ULL(30) |
3488 GENMASK_ULL(13, 9) | BIT_ULL(31);
3491 vmx_misc = vmx_control_msr(vmx->nested.msrs.misc_low,
3492 vmx->nested.msrs.misc_high);
3494 if (!is_bitwise_subset(vmx_misc, data, feature_and_reserved_bits))
3497 if ((vmx->nested.msrs.pinbased_ctls_high &
3498 PIN_BASED_VMX_PREEMPTION_TIMER) &&
3499 vmx_misc_preemption_timer_rate(data) !=
3500 vmx_misc_preemption_timer_rate(vmx_misc))
3503 if (vmx_misc_cr3_count(data) > vmx_misc_cr3_count(vmx_misc))
3506 if (vmx_misc_max_msr(data) > vmx_misc_max_msr(vmx_misc))
3509 if (vmx_misc_mseg_revid(data) != vmx_misc_mseg_revid(vmx_misc))
3512 vmx->nested.msrs.misc_low = data;
3513 vmx->nested.msrs.misc_high = data >> 32;
3516 * If L1 has read-only VM-exit information fields, use the
3517 * less permissive vmx_vmwrite_bitmap to specify write
3518 * permissions for the shadow VMCS.
3520 if (enable_shadow_vmcs && !nested_cpu_has_vmwrite_any_field(&vmx->vcpu))
3521 vmcs_write64(VMWRITE_BITMAP, __pa(vmx_vmwrite_bitmap));
3526 static int vmx_restore_vmx_ept_vpid_cap(struct vcpu_vmx *vmx, u64 data)
3528 u64 vmx_ept_vpid_cap;
3530 vmx_ept_vpid_cap = vmx_control_msr(vmx->nested.msrs.ept_caps,
3531 vmx->nested.msrs.vpid_caps);
3533 /* Every bit is either reserved or a feature bit. */
3534 if (!is_bitwise_subset(vmx_ept_vpid_cap, data, -1ULL))
3537 vmx->nested.msrs.ept_caps = data;
3538 vmx->nested.msrs.vpid_caps = data >> 32;
3542 static int vmx_restore_fixed0_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
3546 switch (msr_index) {
3547 case MSR_IA32_VMX_CR0_FIXED0:
3548 msr = &vmx->nested.msrs.cr0_fixed0;
3550 case MSR_IA32_VMX_CR4_FIXED0:
3551 msr = &vmx->nested.msrs.cr4_fixed0;
3558 * 1 bits (which indicates bits which "must-be-1" during VMX operation)
3559 * must be 1 in the restored value.
3561 if (!is_bitwise_subset(data, *msr, -1ULL))
3569 * Called when userspace is restoring VMX MSRs.
3571 * Returns 0 on success, non-0 otherwise.
3573 static int vmx_set_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
3575 struct vcpu_vmx *vmx = to_vmx(vcpu);
3578 * Don't allow changes to the VMX capability MSRs while the vCPU
3579 * is in VMX operation.
3581 if (vmx->nested.vmxon)
3584 switch (msr_index) {
3585 case MSR_IA32_VMX_BASIC:
3586 return vmx_restore_vmx_basic(vmx, data);
3587 case MSR_IA32_VMX_PINBASED_CTLS:
3588 case MSR_IA32_VMX_PROCBASED_CTLS:
3589 case MSR_IA32_VMX_EXIT_CTLS:
3590 case MSR_IA32_VMX_ENTRY_CTLS:
3592 * The "non-true" VMX capability MSRs are generated from the
3593 * "true" MSRs, so we do not support restoring them directly.
3595 * If userspace wants to emulate VMX_BASIC[55]=0, userspace
3596 * should restore the "true" MSRs with the must-be-1 bits
3597 * set according to the SDM Vol 3. A.2 "RESERVED CONTROLS AND
3598 * DEFAULT SETTINGS".
3601 case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
3602 case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
3603 case MSR_IA32_VMX_TRUE_EXIT_CTLS:
3604 case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
3605 case MSR_IA32_VMX_PROCBASED_CTLS2:
3606 return vmx_restore_control_msr(vmx, msr_index, data);
3607 case MSR_IA32_VMX_MISC:
3608 return vmx_restore_vmx_misc(vmx, data);
3609 case MSR_IA32_VMX_CR0_FIXED0:
3610 case MSR_IA32_VMX_CR4_FIXED0:
3611 return vmx_restore_fixed0_msr(vmx, msr_index, data);
3612 case MSR_IA32_VMX_CR0_FIXED1:
3613 case MSR_IA32_VMX_CR4_FIXED1:
3615 * These MSRs are generated based on the vCPU's CPUID, so we
3616 * do not support restoring them directly.
3619 case MSR_IA32_VMX_EPT_VPID_CAP:
3620 return vmx_restore_vmx_ept_vpid_cap(vmx, data);
3621 case MSR_IA32_VMX_VMCS_ENUM:
3622 vmx->nested.msrs.vmcs_enum = data;
3626 * The rest of the VMX capability MSRs do not support restore.
3632 /* Returns 0 on success, non-0 otherwise. */
3633 static int vmx_get_vmx_msr(struct nested_vmx_msrs *msrs, u32 msr_index, u64 *pdata)
3635 switch (msr_index) {
3636 case MSR_IA32_VMX_BASIC:
3637 *pdata = msrs->basic;
3639 case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
3640 case MSR_IA32_VMX_PINBASED_CTLS:
3641 *pdata = vmx_control_msr(
3642 msrs->pinbased_ctls_low,
3643 msrs->pinbased_ctls_high);
3644 if (msr_index == MSR_IA32_VMX_PINBASED_CTLS)
3645 *pdata |= PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
3647 case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
3648 case MSR_IA32_VMX_PROCBASED_CTLS:
3649 *pdata = vmx_control_msr(
3650 msrs->procbased_ctls_low,
3651 msrs->procbased_ctls_high);
3652 if (msr_index == MSR_IA32_VMX_PROCBASED_CTLS)
3653 *pdata |= CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
3655 case MSR_IA32_VMX_TRUE_EXIT_CTLS:
3656 case MSR_IA32_VMX_EXIT_CTLS:
3657 *pdata = vmx_control_msr(
3658 msrs->exit_ctls_low,
3659 msrs->exit_ctls_high);
3660 if (msr_index == MSR_IA32_VMX_EXIT_CTLS)
3661 *pdata |= VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
3663 case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
3664 case MSR_IA32_VMX_ENTRY_CTLS:
3665 *pdata = vmx_control_msr(
3666 msrs->entry_ctls_low,
3667 msrs->entry_ctls_high);
3668 if (msr_index == MSR_IA32_VMX_ENTRY_CTLS)
3669 *pdata |= VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
3671 case MSR_IA32_VMX_MISC:
3672 *pdata = vmx_control_msr(
3676 case MSR_IA32_VMX_CR0_FIXED0:
3677 *pdata = msrs->cr0_fixed0;
3679 case MSR_IA32_VMX_CR0_FIXED1:
3680 *pdata = msrs->cr0_fixed1;
3682 case MSR_IA32_VMX_CR4_FIXED0:
3683 *pdata = msrs->cr4_fixed0;
3685 case MSR_IA32_VMX_CR4_FIXED1:
3686 *pdata = msrs->cr4_fixed1;
3688 case MSR_IA32_VMX_VMCS_ENUM:
3689 *pdata = msrs->vmcs_enum;
3691 case MSR_IA32_VMX_PROCBASED_CTLS2:
3692 *pdata = vmx_control_msr(
3693 msrs->secondary_ctls_low,
3694 msrs->secondary_ctls_high);
3696 case MSR_IA32_VMX_EPT_VPID_CAP:
3697 *pdata = msrs->ept_caps |
3698 ((u64)msrs->vpid_caps << 32);
3700 case MSR_IA32_VMX_VMFUNC:
3701 *pdata = msrs->vmfunc_controls;
3710 static inline bool vmx_feature_control_msr_valid(struct kvm_vcpu *vcpu,
3713 uint64_t valid_bits = to_vmx(vcpu)->msr_ia32_feature_control_valid_bits;
3715 return !(val & ~valid_bits);
3718 static int vmx_get_msr_feature(struct kvm_msr_entry *msr)
3720 switch (msr->index) {
3721 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
3724 return vmx_get_vmx_msr(&vmcs_config.nested, msr->index, &msr->data);
3733 * Reads an msr value (of 'msr_index') into 'pdata'.
3734 * Returns 0 on success, non-0 otherwise.
3735 * Assumes vcpu_load() was already called.
3737 static int vmx_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3739 struct vcpu_vmx *vmx = to_vmx(vcpu);
3740 struct shared_msr_entry *msr;
3742 switch (msr_info->index) {
3743 #ifdef CONFIG_X86_64
3745 msr_info->data = vmcs_readl(GUEST_FS_BASE);
3748 msr_info->data = vmcs_readl(GUEST_GS_BASE);
3750 case MSR_KERNEL_GS_BASE:
3751 vmx_load_host_state(vmx);
3752 msr_info->data = vmx->msr_guest_kernel_gs_base;
3756 return kvm_get_msr_common(vcpu, msr_info);
3757 case MSR_IA32_SPEC_CTRL:
3758 if (!msr_info->host_initiated &&
3759 !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
3762 msr_info->data = to_vmx(vcpu)->spec_ctrl;
3764 case MSR_IA32_ARCH_CAPABILITIES:
3765 if (!msr_info->host_initiated &&
3766 !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
3768 msr_info->data = to_vmx(vcpu)->arch_capabilities;
3770 case MSR_IA32_SYSENTER_CS:
3771 msr_info->data = vmcs_read32(GUEST_SYSENTER_CS);
3773 case MSR_IA32_SYSENTER_EIP:
3774 msr_info->data = vmcs_readl(GUEST_SYSENTER_EIP);
3776 case MSR_IA32_SYSENTER_ESP:
3777 msr_info->data = vmcs_readl(GUEST_SYSENTER_ESP);
3779 case MSR_IA32_BNDCFGS:
3780 if (!kvm_mpx_supported() ||
3781 (!msr_info->host_initiated &&
3782 !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
3784 msr_info->data = vmcs_read64(GUEST_BNDCFGS);
3786 case MSR_IA32_MCG_EXT_CTL:
3787 if (!msr_info->host_initiated &&
3788 !(vmx->msr_ia32_feature_control &
3789 FEATURE_CONTROL_LMCE))
3791 msr_info->data = vcpu->arch.mcg_ext_ctl;
3793 case MSR_IA32_FEATURE_CONTROL:
3794 msr_info->data = vmx->msr_ia32_feature_control;
3796 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
3797 if (!nested_vmx_allowed(vcpu))
3799 return vmx_get_vmx_msr(&vmx->nested.msrs, msr_info->index,
3802 if (!vmx_xsaves_supported())
3804 msr_info->data = vcpu->arch.ia32_xss;
3807 if (!msr_info->host_initiated &&
3808 !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
3810 /* Otherwise falls through */
3812 msr = find_msr_entry(vmx, msr_info->index);
3814 msr_info->data = msr->data;
3817 return kvm_get_msr_common(vcpu, msr_info);
3823 static void vmx_leave_nested(struct kvm_vcpu *vcpu);
3826 * Writes msr value into into the appropriate "register".
3827 * Returns 0 on success, non-0 otherwise.
3828 * Assumes vcpu_load() was already called.
3830 static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3832 struct vcpu_vmx *vmx = to_vmx(vcpu);
3833 struct shared_msr_entry *msr;
3835 u32 msr_index = msr_info->index;
3836 u64 data = msr_info->data;
3838 switch (msr_index) {
3840 ret = kvm_set_msr_common(vcpu, msr_info);
3842 #ifdef CONFIG_X86_64
3844 vmx_segment_cache_clear(vmx);
3845 vmcs_writel(GUEST_FS_BASE, data);
3848 vmx_segment_cache_clear(vmx);
3849 vmcs_writel(GUEST_GS_BASE, data);
3851 case MSR_KERNEL_GS_BASE:
3852 vmx_load_host_state(vmx);
3853 vmx->msr_guest_kernel_gs_base = data;
3856 case MSR_IA32_SYSENTER_CS:
3857 vmcs_write32(GUEST_SYSENTER_CS, data);
3859 case MSR_IA32_SYSENTER_EIP:
3860 vmcs_writel(GUEST_SYSENTER_EIP, data);
3862 case MSR_IA32_SYSENTER_ESP:
3863 vmcs_writel(GUEST_SYSENTER_ESP, data);
3865 case MSR_IA32_BNDCFGS:
3866 if (!kvm_mpx_supported() ||
3867 (!msr_info->host_initiated &&
3868 !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
3870 if (is_noncanonical_address(data & PAGE_MASK, vcpu) ||
3871 (data & MSR_IA32_BNDCFGS_RSVD))
3873 vmcs_write64(GUEST_BNDCFGS, data);
3875 case MSR_IA32_SPEC_CTRL:
3876 if (!msr_info->host_initiated &&
3877 !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
3880 /* The STIBP bit doesn't fault even if it's not advertised */
3881 if (data & ~(SPEC_CTRL_IBRS | SPEC_CTRL_STIBP | SPEC_CTRL_SSBD))
3884 vmx->spec_ctrl = data;
3891 * When it's written (to non-zero) for the first time, pass
3895 * The handling of the MSR bitmap for L2 guests is done in
3896 * nested_vmx_merge_msr_bitmap. We should not touch the
3897 * vmcs02.msr_bitmap here since it gets completely overwritten
3898 * in the merging. We update the vmcs01 here for L1 as well
3899 * since it will end up touching the MSR anyway now.
3901 vmx_disable_intercept_for_msr(vmx->vmcs01.msr_bitmap,
3905 case MSR_IA32_PRED_CMD:
3906 if (!msr_info->host_initiated &&
3907 !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
3910 if (data & ~PRED_CMD_IBPB)
3916 wrmsrl(MSR_IA32_PRED_CMD, PRED_CMD_IBPB);
3920 * When it's written (to non-zero) for the first time, pass
3924 * The handling of the MSR bitmap for L2 guests is done in
3925 * nested_vmx_merge_msr_bitmap. We should not touch the
3926 * vmcs02.msr_bitmap here since it gets completely overwritten
3929 vmx_disable_intercept_for_msr(vmx->vmcs01.msr_bitmap, MSR_IA32_PRED_CMD,
3932 case MSR_IA32_ARCH_CAPABILITIES:
3933 if (!msr_info->host_initiated)
3935 vmx->arch_capabilities = data;
3937 case MSR_IA32_CR_PAT:
3938 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
3939 if (!kvm_mtrr_valid(vcpu, MSR_IA32_CR_PAT, data))
3941 vmcs_write64(GUEST_IA32_PAT, data);
3942 vcpu->arch.pat = data;
3945 ret = kvm_set_msr_common(vcpu, msr_info);
3947 case MSR_IA32_TSC_ADJUST:
3948 ret = kvm_set_msr_common(vcpu, msr_info);
3950 case MSR_IA32_MCG_EXT_CTL:
3951 if ((!msr_info->host_initiated &&
3952 !(to_vmx(vcpu)->msr_ia32_feature_control &
3953 FEATURE_CONTROL_LMCE)) ||
3954 (data & ~MCG_EXT_CTL_LMCE_EN))
3956 vcpu->arch.mcg_ext_ctl = data;
3958 case MSR_IA32_FEATURE_CONTROL:
3959 if (!vmx_feature_control_msr_valid(vcpu, data) ||
3960 (to_vmx(vcpu)->msr_ia32_feature_control &
3961 FEATURE_CONTROL_LOCKED && !msr_info->host_initiated))
3963 vmx->msr_ia32_feature_control = data;
3964 if (msr_info->host_initiated && data == 0)
3965 vmx_leave_nested(vcpu);
3967 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
3968 if (!msr_info->host_initiated)
3969 return 1; /* they are read-only */
3970 if (!nested_vmx_allowed(vcpu))
3972 return vmx_set_vmx_msr(vcpu, msr_index, data);
3974 if (!vmx_xsaves_supported())
3977 * The only supported bit as of Skylake is bit 8, but
3978 * it is not supported on KVM.
3982 vcpu->arch.ia32_xss = data;
3983 if (vcpu->arch.ia32_xss != host_xss)
3984 add_atomic_switch_msr(vmx, MSR_IA32_XSS,
3985 vcpu->arch.ia32_xss, host_xss);
3987 clear_atomic_switch_msr(vmx, MSR_IA32_XSS);
3990 if (!msr_info->host_initiated &&
3991 !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
3993 /* Check reserved bit, higher 32 bits should be zero */
3994 if ((data >> 32) != 0)
3996 /* Otherwise falls through */
3998 msr = find_msr_entry(vmx, msr_index);
4000 u64 old_msr_data = msr->data;
4002 if (msr - vmx->guest_msrs < vmx->save_nmsrs) {
4004 ret = kvm_set_shared_msr(msr->index, msr->data,
4008 msr->data = old_msr_data;
4012 ret = kvm_set_msr_common(vcpu, msr_info);
4018 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
4020 __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
4023 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
4026 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
4028 case VCPU_EXREG_PDPTR:
4030 ept_save_pdptrs(vcpu);
4037 static __init int cpu_has_kvm_support(void)
4039 return cpu_has_vmx();
4042 static __init int vmx_disabled_by_bios(void)
4046 rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
4047 if (msr & FEATURE_CONTROL_LOCKED) {
4048 /* launched w/ TXT and VMX disabled */
4049 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
4052 /* launched w/o TXT and VMX only enabled w/ TXT */
4053 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
4054 && (msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
4055 && !tboot_enabled()) {
4056 printk(KERN_WARNING "kvm: disable TXT in the BIOS or "
4057 "activate TXT before enabling KVM\n");
4060 /* launched w/o TXT and VMX disabled */
4061 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
4062 && !tboot_enabled())
4069 static void kvm_cpu_vmxon(u64 addr)
4071 cr4_set_bits(X86_CR4_VMXE);
4072 intel_pt_handle_vmx(1);
4074 asm volatile (ASM_VMX_VMXON_RAX
4075 : : "a"(&addr), "m"(addr)
4079 static int hardware_enable(void)
4081 int cpu = raw_smp_processor_id();
4082 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
4085 if (cr4_read_shadow() & X86_CR4_VMXE)
4089 * This can happen if we hot-added a CPU but failed to allocate
4090 * VP assist page for it.
4092 if (static_branch_unlikely(&enable_evmcs) &&
4093 !hv_get_vp_assist_page(cpu))
4096 INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
4097 INIT_LIST_HEAD(&per_cpu(blocked_vcpu_on_cpu, cpu));
4098 spin_lock_init(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
4101 * Now we can enable the vmclear operation in kdump
4102 * since the loaded_vmcss_on_cpu list on this cpu
4103 * has been initialized.
4105 * Though the cpu is not in VMX operation now, there
4106 * is no problem to enable the vmclear operation
4107 * for the loaded_vmcss_on_cpu list is empty!
4109 crash_enable_local_vmclear(cpu);
4111 rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
4113 test_bits = FEATURE_CONTROL_LOCKED;
4114 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
4115 if (tboot_enabled())
4116 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX;
4118 if ((old & test_bits) != test_bits) {
4119 /* enable and lock */
4120 wrmsrl(MSR_IA32_FEATURE_CONTROL, old | test_bits);
4122 kvm_cpu_vmxon(phys_addr);
4129 static void vmclear_local_loaded_vmcss(void)
4131 int cpu = raw_smp_processor_id();
4132 struct loaded_vmcs *v, *n;
4134 list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
4135 loaded_vmcss_on_cpu_link)
4136 __loaded_vmcs_clear(v);
4140 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
4143 static void kvm_cpu_vmxoff(void)
4145 asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
4147 intel_pt_handle_vmx(0);
4148 cr4_clear_bits(X86_CR4_VMXE);
4151 static void hardware_disable(void)
4153 vmclear_local_loaded_vmcss();
4157 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
4158 u32 msr, u32 *result)
4160 u32 vmx_msr_low, vmx_msr_high;
4161 u32 ctl = ctl_min | ctl_opt;
4163 rdmsr(msr, vmx_msr_low, vmx_msr_high);
4165 ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
4166 ctl |= vmx_msr_low; /* bit == 1 in low word ==> must be one */
4168 /* Ensure minimum (required) set of control bits are supported. */
4176 static __init bool allow_1_setting(u32 msr, u32 ctl)
4178 u32 vmx_msr_low, vmx_msr_high;
4180 rdmsr(msr, vmx_msr_low, vmx_msr_high);
4181 return vmx_msr_high & ctl;
4184 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
4186 u32 vmx_msr_low, vmx_msr_high;
4187 u32 min, opt, min2, opt2;
4188 u32 _pin_based_exec_control = 0;
4189 u32 _cpu_based_exec_control = 0;
4190 u32 _cpu_based_2nd_exec_control = 0;
4191 u32 _vmexit_control = 0;
4192 u32 _vmentry_control = 0;
4194 memset(vmcs_conf, 0, sizeof(*vmcs_conf));
4195 min = CPU_BASED_HLT_EXITING |
4196 #ifdef CONFIG_X86_64
4197 CPU_BASED_CR8_LOAD_EXITING |
4198 CPU_BASED_CR8_STORE_EXITING |
4200 CPU_BASED_CR3_LOAD_EXITING |
4201 CPU_BASED_CR3_STORE_EXITING |
4202 CPU_BASED_UNCOND_IO_EXITING |
4203 CPU_BASED_MOV_DR_EXITING |
4204 CPU_BASED_USE_TSC_OFFSETING |
4205 CPU_BASED_MWAIT_EXITING |
4206 CPU_BASED_MONITOR_EXITING |
4207 CPU_BASED_INVLPG_EXITING |
4208 CPU_BASED_RDPMC_EXITING;
4210 opt = CPU_BASED_TPR_SHADOW |
4211 CPU_BASED_USE_MSR_BITMAPS |
4212 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
4213 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
4214 &_cpu_based_exec_control) < 0)
4216 #ifdef CONFIG_X86_64
4217 if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
4218 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
4219 ~CPU_BASED_CR8_STORE_EXITING;
4221 if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
4223 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
4224 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
4225 SECONDARY_EXEC_WBINVD_EXITING |
4226 SECONDARY_EXEC_ENABLE_VPID |
4227 SECONDARY_EXEC_ENABLE_EPT |
4228 SECONDARY_EXEC_UNRESTRICTED_GUEST |
4229 SECONDARY_EXEC_PAUSE_LOOP_EXITING |
4230 SECONDARY_EXEC_DESC |
4231 SECONDARY_EXEC_RDTSCP |
4232 SECONDARY_EXEC_ENABLE_INVPCID |
4233 SECONDARY_EXEC_APIC_REGISTER_VIRT |
4234 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
4235 SECONDARY_EXEC_SHADOW_VMCS |
4236 SECONDARY_EXEC_XSAVES |
4237 SECONDARY_EXEC_RDSEED_EXITING |
4238 SECONDARY_EXEC_RDRAND_EXITING |
4239 SECONDARY_EXEC_ENABLE_PML |
4240 SECONDARY_EXEC_TSC_SCALING |
4241 SECONDARY_EXEC_ENABLE_VMFUNC;
4242 if (adjust_vmx_controls(min2, opt2,
4243 MSR_IA32_VMX_PROCBASED_CTLS2,
4244 &_cpu_based_2nd_exec_control) < 0)
4247 #ifndef CONFIG_X86_64
4248 if (!(_cpu_based_2nd_exec_control &
4249 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
4250 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
4253 if (!(_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
4254 _cpu_based_2nd_exec_control &= ~(
4255 SECONDARY_EXEC_APIC_REGISTER_VIRT |
4256 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
4257 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4259 rdmsr_safe(MSR_IA32_VMX_EPT_VPID_CAP,
4260 &vmx_capability.ept, &vmx_capability.vpid);
4262 if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
4263 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
4265 _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
4266 CPU_BASED_CR3_STORE_EXITING |
4267 CPU_BASED_INVLPG_EXITING);
4268 } else if (vmx_capability.ept) {
4269 vmx_capability.ept = 0;
4270 pr_warn_once("EPT CAP should not exist if not support "
4271 "1-setting enable EPT VM-execution control\n");
4273 if (!(_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_VPID) &&
4274 vmx_capability.vpid) {
4275 vmx_capability.vpid = 0;
4276 pr_warn_once("VPID CAP should not exist if not support "
4277 "1-setting enable VPID VM-execution control\n");
4280 min = VM_EXIT_SAVE_DEBUG_CONTROLS | VM_EXIT_ACK_INTR_ON_EXIT;
4281 #ifdef CONFIG_X86_64
4282 min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
4284 opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT |
4285 VM_EXIT_CLEAR_BNDCFGS;
4286 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
4287 &_vmexit_control) < 0)
4290 min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
4291 opt = PIN_BASED_VIRTUAL_NMIS | PIN_BASED_POSTED_INTR |
4292 PIN_BASED_VMX_PREEMPTION_TIMER;
4293 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
4294 &_pin_based_exec_control) < 0)
4297 if (cpu_has_broken_vmx_preemption_timer())
4298 _pin_based_exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
4299 if (!(_cpu_based_2nd_exec_control &
4300 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY))
4301 _pin_based_exec_control &= ~PIN_BASED_POSTED_INTR;
4303 min = VM_ENTRY_LOAD_DEBUG_CONTROLS;
4304 opt = VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_BNDCFGS;
4305 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
4306 &_vmentry_control) < 0)
4309 rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
4311 /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
4312 if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
4315 #ifdef CONFIG_X86_64
4316 /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
4317 if (vmx_msr_high & (1u<<16))
4321 /* Require Write-Back (WB) memory type for VMCS accesses. */
4322 if (((vmx_msr_high >> 18) & 15) != 6)
4325 vmcs_conf->size = vmx_msr_high & 0x1fff;
4326 vmcs_conf->order = get_order(vmcs_conf->size);
4327 vmcs_conf->basic_cap = vmx_msr_high & ~0x1fff;
4329 vmcs_conf->revision_id = vmx_msr_low;
4331 vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
4332 vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
4333 vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
4334 vmcs_conf->vmexit_ctrl = _vmexit_control;
4335 vmcs_conf->vmentry_ctrl = _vmentry_control;
4337 if (static_branch_unlikely(&enable_evmcs))
4338 evmcs_sanitize_exec_ctrls(vmcs_conf);
4340 cpu_has_load_ia32_efer =
4341 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
4342 VM_ENTRY_LOAD_IA32_EFER)
4343 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
4344 VM_EXIT_LOAD_IA32_EFER);
4346 cpu_has_load_perf_global_ctrl =
4347 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
4348 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
4349 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
4350 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
4353 * Some cpus support VM_ENTRY_(LOAD|SAVE)_IA32_PERF_GLOBAL_CTRL
4354 * but due to errata below it can't be used. Workaround is to use
4355 * msr load mechanism to switch IA32_PERF_GLOBAL_CTRL.
4357 * VM Exit May Incorrectly Clear IA32_PERF_GLOBAL_CTRL [34:32]
4362 * BC86,AAY89,BD102 (model 44)
4366 if (cpu_has_load_perf_global_ctrl && boot_cpu_data.x86 == 0x6) {
4367 switch (boot_cpu_data.x86_model) {
4373 cpu_has_load_perf_global_ctrl = false;
4374 printk_once(KERN_WARNING"kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
4375 "does not work properly. Using workaround\n");
4382 if (boot_cpu_has(X86_FEATURE_XSAVES))
4383 rdmsrl(MSR_IA32_XSS, host_xss);
4388 static struct vmcs *alloc_vmcs_cpu(int cpu)
4390 int node = cpu_to_node(cpu);
4394 pages = __alloc_pages_node(node, GFP_KERNEL, vmcs_config.order);
4397 vmcs = page_address(pages);
4398 memset(vmcs, 0, vmcs_config.size);
4400 /* KVM supports Enlightened VMCS v1 only */
4401 if (static_branch_unlikely(&enable_evmcs))
4402 vmcs->revision_id = KVM_EVMCS_VERSION;
4404 vmcs->revision_id = vmcs_config.revision_id;
4409 static void free_vmcs(struct vmcs *vmcs)
4411 free_pages((unsigned long)vmcs, vmcs_config.order);
4415 * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
4417 static void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
4419 if (!loaded_vmcs->vmcs)
4421 loaded_vmcs_clear(loaded_vmcs);
4422 free_vmcs(loaded_vmcs->vmcs);
4423 loaded_vmcs->vmcs = NULL;
4424 if (loaded_vmcs->msr_bitmap)
4425 free_page((unsigned long)loaded_vmcs->msr_bitmap);
4426 WARN_ON(loaded_vmcs->shadow_vmcs != NULL);
4429 static struct vmcs *alloc_vmcs(void)
4431 return alloc_vmcs_cpu(raw_smp_processor_id());
4434 static int alloc_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
4436 loaded_vmcs->vmcs = alloc_vmcs();
4437 if (!loaded_vmcs->vmcs)
4440 loaded_vmcs->shadow_vmcs = NULL;
4441 loaded_vmcs_init(loaded_vmcs);
4443 if (cpu_has_vmx_msr_bitmap()) {
4444 loaded_vmcs->msr_bitmap = (unsigned long *)__get_free_page(GFP_KERNEL);
4445 if (!loaded_vmcs->msr_bitmap)
4447 memset(loaded_vmcs->msr_bitmap, 0xff, PAGE_SIZE);
4449 if (IS_ENABLED(CONFIG_HYPERV) &&
4450 static_branch_unlikely(&enable_evmcs) &&
4451 (ms_hyperv.nested_features & HV_X64_NESTED_MSR_BITMAP)) {
4452 struct hv_enlightened_vmcs *evmcs =
4453 (struct hv_enlightened_vmcs *)loaded_vmcs->vmcs;
4455 evmcs->hv_enlightenments_control.msr_bitmap = 1;
4461 free_loaded_vmcs(loaded_vmcs);
4465 static void free_kvm_area(void)
4469 for_each_possible_cpu(cpu) {
4470 free_vmcs(per_cpu(vmxarea, cpu));
4471 per_cpu(vmxarea, cpu) = NULL;
4475 enum vmcs_field_width {
4476 VMCS_FIELD_WIDTH_U16 = 0,
4477 VMCS_FIELD_WIDTH_U64 = 1,
4478 VMCS_FIELD_WIDTH_U32 = 2,
4479 VMCS_FIELD_WIDTH_NATURAL_WIDTH = 3
4482 static inline int vmcs_field_width(unsigned long field)
4484 if (0x1 & field) /* the *_HIGH fields are all 32 bit */
4485 return VMCS_FIELD_WIDTH_U32;
4486 return (field >> 13) & 0x3 ;
4489 static inline int vmcs_field_readonly(unsigned long field)
4491 return (((field >> 10) & 0x3) == 1);
4494 static void init_vmcs_shadow_fields(void)
4498 for (i = j = 0; i < max_shadow_read_only_fields; i++) {
4499 u16 field = shadow_read_only_fields[i];
4500 if (vmcs_field_width(field) == VMCS_FIELD_WIDTH_U64 &&
4501 (i + 1 == max_shadow_read_only_fields ||
4502 shadow_read_only_fields[i + 1] != field + 1))
4503 pr_err("Missing field from shadow_read_only_field %x\n",
4506 clear_bit(field, vmx_vmread_bitmap);
4507 #ifdef CONFIG_X86_64
4512 shadow_read_only_fields[j] = field;
4515 max_shadow_read_only_fields = j;
4517 for (i = j = 0; i < max_shadow_read_write_fields; i++) {
4518 u16 field = shadow_read_write_fields[i];
4519 if (vmcs_field_width(field) == VMCS_FIELD_WIDTH_U64 &&
4520 (i + 1 == max_shadow_read_write_fields ||
4521 shadow_read_write_fields[i + 1] != field + 1))
4522 pr_err("Missing field from shadow_read_write_field %x\n",
4526 * PML and the preemption timer can be emulated, but the
4527 * processor cannot vmwrite to fields that don't exist
4531 case GUEST_PML_INDEX:
4532 if (!cpu_has_vmx_pml())
4535 case VMX_PREEMPTION_TIMER_VALUE:
4536 if (!cpu_has_vmx_preemption_timer())
4539 case GUEST_INTR_STATUS:
4540 if (!cpu_has_vmx_apicv())
4547 clear_bit(field, vmx_vmwrite_bitmap);
4548 clear_bit(field, vmx_vmread_bitmap);
4549 #ifdef CONFIG_X86_64
4554 shadow_read_write_fields[j] = field;
4557 max_shadow_read_write_fields = j;
4560 static __init int alloc_kvm_area(void)
4564 for_each_possible_cpu(cpu) {
4567 vmcs = alloc_vmcs_cpu(cpu);
4574 * When eVMCS is enabled, alloc_vmcs_cpu() sets
4575 * vmcs->revision_id to KVM_EVMCS_VERSION instead of
4576 * revision_id reported by MSR_IA32_VMX_BASIC.
4578 * However, even though not explictly documented by
4579 * TLFS, VMXArea passed as VMXON argument should
4580 * still be marked with revision_id reported by
4583 if (static_branch_unlikely(&enable_evmcs))
4584 vmcs->revision_id = vmcs_config.revision_id;
4586 per_cpu(vmxarea, cpu) = vmcs;
4591 static void fix_pmode_seg(struct kvm_vcpu *vcpu, int seg,
4592 struct kvm_segment *save)
4594 if (!emulate_invalid_guest_state) {
4596 * CS and SS RPL should be equal during guest entry according
4597 * to VMX spec, but in reality it is not always so. Since vcpu
4598 * is in the middle of the transition from real mode to
4599 * protected mode it is safe to assume that RPL 0 is a good
4602 if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
4603 save->selector &= ~SEGMENT_RPL_MASK;
4604 save->dpl = save->selector & SEGMENT_RPL_MASK;
4607 vmx_set_segment(vcpu, save, seg);
4610 static void enter_pmode(struct kvm_vcpu *vcpu)
4612 unsigned long flags;
4613 struct vcpu_vmx *vmx = to_vmx(vcpu);
4616 * Update real mode segment cache. It may be not up-to-date if sement
4617 * register was written while vcpu was in a guest mode.
4619 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
4620 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
4621 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
4622 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
4623 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
4624 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
4626 vmx->rmode.vm86_active = 0;
4628 vmx_segment_cache_clear(vmx);
4630 vmx_set_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
4632 flags = vmcs_readl(GUEST_RFLAGS);
4633 flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
4634 flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
4635 vmcs_writel(GUEST_RFLAGS, flags);
4637 vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
4638 (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
4640 update_exception_bitmap(vcpu);
4642 fix_pmode_seg(vcpu, VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
4643 fix_pmode_seg(vcpu, VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
4644 fix_pmode_seg(vcpu, VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
4645 fix_pmode_seg(vcpu, VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
4646 fix_pmode_seg(vcpu, VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
4647 fix_pmode_seg(vcpu, VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
4650 static void fix_rmode_seg(int seg, struct kvm_segment *save)
4652 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
4653 struct kvm_segment var = *save;
4656 if (seg == VCPU_SREG_CS)
4659 if (!emulate_invalid_guest_state) {
4660 var.selector = var.base >> 4;
4661 var.base = var.base & 0xffff0;
4671 if (save->base & 0xf)
4672 printk_once(KERN_WARNING "kvm: segment base is not "
4673 "paragraph aligned when entering "
4674 "protected mode (seg=%d)", seg);
4677 vmcs_write16(sf->selector, var.selector);
4678 vmcs_writel(sf->base, var.base);
4679 vmcs_write32(sf->limit, var.limit);
4680 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(&var));
4683 static void enter_rmode(struct kvm_vcpu *vcpu)
4685 unsigned long flags;
4686 struct vcpu_vmx *vmx = to_vmx(vcpu);
4687 struct kvm_vmx *kvm_vmx = to_kvm_vmx(vcpu->kvm);
4689 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
4690 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
4691 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
4692 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
4693 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
4694 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
4695 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
4697 vmx->rmode.vm86_active = 1;
4700 * Very old userspace does not call KVM_SET_TSS_ADDR before entering
4701 * vcpu. Warn the user that an update is overdue.
4703 if (!kvm_vmx->tss_addr)
4704 printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
4705 "called before entering vcpu\n");
4707 vmx_segment_cache_clear(vmx);
4709 vmcs_writel(GUEST_TR_BASE, kvm_vmx->tss_addr);
4710 vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
4711 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
4713 flags = vmcs_readl(GUEST_RFLAGS);
4714 vmx->rmode.save_rflags = flags;
4716 flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
4718 vmcs_writel(GUEST_RFLAGS, flags);
4719 vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
4720 update_exception_bitmap(vcpu);
4722 fix_rmode_seg(VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
4723 fix_rmode_seg(VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
4724 fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
4725 fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
4726 fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
4727 fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
4729 kvm_mmu_reset_context(vcpu);
4732 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
4734 struct vcpu_vmx *vmx = to_vmx(vcpu);
4735 struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
4741 * Force kernel_gs_base reloading before EFER changes, as control
4742 * of this msr depends on is_long_mode().
4744 vmx_load_host_state(to_vmx(vcpu));
4745 vcpu->arch.efer = efer;
4746 if (efer & EFER_LMA) {
4747 vm_entry_controls_setbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
4750 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
4752 msr->data = efer & ~EFER_LME;
4757 #ifdef CONFIG_X86_64
4759 static void enter_lmode(struct kvm_vcpu *vcpu)
4763 vmx_segment_cache_clear(to_vmx(vcpu));
4765 guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
4766 if ((guest_tr_ar & VMX_AR_TYPE_MASK) != VMX_AR_TYPE_BUSY_64_TSS) {
4767 pr_debug_ratelimited("%s: tss fixup for long mode. \n",
4769 vmcs_write32(GUEST_TR_AR_BYTES,
4770 (guest_tr_ar & ~VMX_AR_TYPE_MASK)
4771 | VMX_AR_TYPE_BUSY_64_TSS);
4773 vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
4776 static void exit_lmode(struct kvm_vcpu *vcpu)
4778 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
4779 vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
4784 static inline void __vmx_flush_tlb(struct kvm_vcpu *vcpu, int vpid,
4785 bool invalidate_gpa)
4787 if (enable_ept && (invalidate_gpa || !enable_vpid)) {
4788 if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
4790 ept_sync_context(construct_eptp(vcpu, vcpu->arch.mmu.root_hpa));
4792 vpid_sync_context(vpid);
4796 static void vmx_flush_tlb(struct kvm_vcpu *vcpu, bool invalidate_gpa)
4798 __vmx_flush_tlb(vcpu, to_vmx(vcpu)->vpid, invalidate_gpa);
4801 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
4803 ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
4805 vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
4806 vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
4809 static void vmx_decache_cr3(struct kvm_vcpu *vcpu)
4811 if (enable_unrestricted_guest || (enable_ept && is_paging(vcpu)))
4812 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
4813 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
4816 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
4818 ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
4820 vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
4821 vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
4824 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
4826 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
4828 if (!test_bit(VCPU_EXREG_PDPTR,
4829 (unsigned long *)&vcpu->arch.regs_dirty))
4832 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
4833 vmcs_write64(GUEST_PDPTR0, mmu->pdptrs[0]);
4834 vmcs_write64(GUEST_PDPTR1, mmu->pdptrs[1]);
4835 vmcs_write64(GUEST_PDPTR2, mmu->pdptrs[2]);
4836 vmcs_write64(GUEST_PDPTR3, mmu->pdptrs[3]);
4840 static void ept_save_pdptrs(struct kvm_vcpu *vcpu)
4842 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
4844 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
4845 mmu->pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
4846 mmu->pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
4847 mmu->pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
4848 mmu->pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
4851 __set_bit(VCPU_EXREG_PDPTR,
4852 (unsigned long *)&vcpu->arch.regs_avail);
4853 __set_bit(VCPU_EXREG_PDPTR,
4854 (unsigned long *)&vcpu->arch.regs_dirty);
4857 static bool nested_guest_cr0_valid(struct kvm_vcpu *vcpu, unsigned long val)
4859 u64 fixed0 = to_vmx(vcpu)->nested.msrs.cr0_fixed0;
4860 u64 fixed1 = to_vmx(vcpu)->nested.msrs.cr0_fixed1;
4861 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4863 if (to_vmx(vcpu)->nested.msrs.secondary_ctls_high &
4864 SECONDARY_EXEC_UNRESTRICTED_GUEST &&
4865 nested_cpu_has2(vmcs12, SECONDARY_EXEC_UNRESTRICTED_GUEST))
4866 fixed0 &= ~(X86_CR0_PE | X86_CR0_PG);
4868 return fixed_bits_valid(val, fixed0, fixed1);
4871 static bool nested_host_cr0_valid(struct kvm_vcpu *vcpu, unsigned long val)
4873 u64 fixed0 = to_vmx(vcpu)->nested.msrs.cr0_fixed0;
4874 u64 fixed1 = to_vmx(vcpu)->nested.msrs.cr0_fixed1;
4876 return fixed_bits_valid(val, fixed0, fixed1);
4879 static bool nested_cr4_valid(struct kvm_vcpu *vcpu, unsigned long val)
4881 u64 fixed0 = to_vmx(vcpu)->nested.msrs.cr4_fixed0;
4882 u64 fixed1 = to_vmx(vcpu)->nested.msrs.cr4_fixed1;
4884 return fixed_bits_valid(val, fixed0, fixed1);
4887 /* No difference in the restrictions on guest and host CR4 in VMX operation. */
4888 #define nested_guest_cr4_valid nested_cr4_valid
4889 #define nested_host_cr4_valid nested_cr4_valid
4891 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
4893 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
4895 struct kvm_vcpu *vcpu)
4897 if (!test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
4898 vmx_decache_cr3(vcpu);
4899 if (!(cr0 & X86_CR0_PG)) {
4900 /* From paging/starting to nonpaging */
4901 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
4902 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
4903 (CPU_BASED_CR3_LOAD_EXITING |
4904 CPU_BASED_CR3_STORE_EXITING));
4905 vcpu->arch.cr0 = cr0;
4906 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
4907 } else if (!is_paging(vcpu)) {
4908 /* From nonpaging to paging */
4909 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
4910 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
4911 ~(CPU_BASED_CR3_LOAD_EXITING |
4912 CPU_BASED_CR3_STORE_EXITING));
4913 vcpu->arch.cr0 = cr0;
4914 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
4917 if (!(cr0 & X86_CR0_WP))
4918 *hw_cr0 &= ~X86_CR0_WP;
4921 static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
4923 struct vcpu_vmx *vmx = to_vmx(vcpu);
4924 unsigned long hw_cr0;
4926 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK);
4927 if (enable_unrestricted_guest)
4928 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
4930 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON;
4932 if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
4935 if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
4939 #ifdef CONFIG_X86_64
4940 if (vcpu->arch.efer & EFER_LME) {
4941 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
4943 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
4948 if (enable_ept && !enable_unrestricted_guest)
4949 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
4951 vmcs_writel(CR0_READ_SHADOW, cr0);
4952 vmcs_writel(GUEST_CR0, hw_cr0);
4953 vcpu->arch.cr0 = cr0;
4955 /* depends on vcpu->arch.cr0 to be set to a new value */
4956 vmx->emulation_required = emulation_required(vcpu);
4959 static int get_ept_level(struct kvm_vcpu *vcpu)
4961 if (cpu_has_vmx_ept_5levels() && (cpuid_maxphyaddr(vcpu) > 48))
4966 static u64 construct_eptp(struct kvm_vcpu *vcpu, unsigned long root_hpa)
4968 u64 eptp = VMX_EPTP_MT_WB;
4970 eptp |= (get_ept_level(vcpu) == 5) ? VMX_EPTP_PWL_5 : VMX_EPTP_PWL_4;
4972 if (enable_ept_ad_bits &&
4973 (!is_guest_mode(vcpu) || nested_ept_ad_enabled(vcpu)))
4974 eptp |= VMX_EPTP_AD_ENABLE_BIT;
4975 eptp |= (root_hpa & PAGE_MASK);
4980 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
4982 unsigned long guest_cr3;
4987 eptp = construct_eptp(vcpu, cr3);
4988 vmcs_write64(EPT_POINTER, eptp);
4989 if (enable_unrestricted_guest || is_paging(vcpu) ||
4990 is_guest_mode(vcpu))
4991 guest_cr3 = kvm_read_cr3(vcpu);
4993 guest_cr3 = to_kvm_vmx(vcpu->kvm)->ept_identity_map_addr;
4994 ept_load_pdptrs(vcpu);
4997 vmx_flush_tlb(vcpu, true);
4998 vmcs_writel(GUEST_CR3, guest_cr3);
5001 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
5004 * Pass through host's Machine Check Enable value to hw_cr4, which
5005 * is in force while we are in guest mode. Do not let guests control
5006 * this bit, even if host CR4.MCE == 0.
5008 unsigned long hw_cr4;
5010 hw_cr4 = (cr4_read_shadow() & X86_CR4_MCE) | (cr4 & ~X86_CR4_MCE);
5011 if (enable_unrestricted_guest)
5012 hw_cr4 |= KVM_VM_CR4_ALWAYS_ON_UNRESTRICTED_GUEST;
5013 else if (to_vmx(vcpu)->rmode.vm86_active)
5014 hw_cr4 |= KVM_RMODE_VM_CR4_ALWAYS_ON;
5016 hw_cr4 |= KVM_PMODE_VM_CR4_ALWAYS_ON;
5018 if (!boot_cpu_has(X86_FEATURE_UMIP) && vmx_umip_emulated()) {
5019 if (cr4 & X86_CR4_UMIP) {
5020 vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
5021 SECONDARY_EXEC_DESC);
5022 hw_cr4 &= ~X86_CR4_UMIP;
5023 } else if (!is_guest_mode(vcpu) ||
5024 !nested_cpu_has2(get_vmcs12(vcpu), SECONDARY_EXEC_DESC))
5025 vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL,
5026 SECONDARY_EXEC_DESC);
5029 if (cr4 & X86_CR4_VMXE) {
5031 * To use VMXON (and later other VMX instructions), a guest
5032 * must first be able to turn on cr4.VMXE (see handle_vmon()).
5033 * So basically the check on whether to allow nested VMX
5036 if (!nested_vmx_allowed(vcpu))
5040 if (to_vmx(vcpu)->nested.vmxon && !nested_cr4_valid(vcpu, cr4))
5043 vcpu->arch.cr4 = cr4;
5045 if (!enable_unrestricted_guest) {
5047 if (!is_paging(vcpu)) {
5048 hw_cr4 &= ~X86_CR4_PAE;
5049 hw_cr4 |= X86_CR4_PSE;
5050 } else if (!(cr4 & X86_CR4_PAE)) {
5051 hw_cr4 &= ~X86_CR4_PAE;
5056 * SMEP/SMAP/PKU is disabled if CPU is in non-paging mode in
5057 * hardware. To emulate this behavior, SMEP/SMAP/PKU needs
5058 * to be manually disabled when guest switches to non-paging
5061 * If !enable_unrestricted_guest, the CPU is always running
5062 * with CR0.PG=1 and CR4 needs to be modified.
5063 * If enable_unrestricted_guest, the CPU automatically
5064 * disables SMEP/SMAP/PKU when the guest sets CR0.PG=0.
5066 if (!is_paging(vcpu))
5067 hw_cr4 &= ~(X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE);
5070 vmcs_writel(CR4_READ_SHADOW, cr4);
5071 vmcs_writel(GUEST_CR4, hw_cr4);
5075 static void vmx_get_segment(struct kvm_vcpu *vcpu,
5076 struct kvm_segment *var, int seg)
5078 struct vcpu_vmx *vmx = to_vmx(vcpu);
5081 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
5082 *var = vmx->rmode.segs[seg];
5083 if (seg == VCPU_SREG_TR
5084 || var->selector == vmx_read_guest_seg_selector(vmx, seg))
5086 var->base = vmx_read_guest_seg_base(vmx, seg);
5087 var->selector = vmx_read_guest_seg_selector(vmx, seg);
5090 var->base = vmx_read_guest_seg_base(vmx, seg);
5091 var->limit = vmx_read_guest_seg_limit(vmx, seg);
5092 var->selector = vmx_read_guest_seg_selector(vmx, seg);
5093 ar = vmx_read_guest_seg_ar(vmx, seg);
5094 var->unusable = (ar >> 16) & 1;
5095 var->type = ar & 15;
5096 var->s = (ar >> 4) & 1;
5097 var->dpl = (ar >> 5) & 3;
5099 * Some userspaces do not preserve unusable property. Since usable
5100 * segment has to be present according to VMX spec we can use present
5101 * property to amend userspace bug by making unusable segment always
5102 * nonpresent. vmx_segment_access_rights() already marks nonpresent
5103 * segment as unusable.
5105 var->present = !var->unusable;
5106 var->avl = (ar >> 12) & 1;
5107 var->l = (ar >> 13) & 1;
5108 var->db = (ar >> 14) & 1;
5109 var->g = (ar >> 15) & 1;
5112 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
5114 struct kvm_segment s;
5116 if (to_vmx(vcpu)->rmode.vm86_active) {
5117 vmx_get_segment(vcpu, &s, seg);
5120 return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
5123 static int vmx_get_cpl(struct kvm_vcpu *vcpu)
5125 struct vcpu_vmx *vmx = to_vmx(vcpu);
5127 if (unlikely(vmx->rmode.vm86_active))
5130 int ar = vmx_read_guest_seg_ar(vmx, VCPU_SREG_SS);
5131 return VMX_AR_DPL(ar);
5135 static u32 vmx_segment_access_rights(struct kvm_segment *var)
5139 if (var->unusable || !var->present)
5142 ar = var->type & 15;
5143 ar |= (var->s & 1) << 4;
5144 ar |= (var->dpl & 3) << 5;
5145 ar |= (var->present & 1) << 7;
5146 ar |= (var->avl & 1) << 12;
5147 ar |= (var->l & 1) << 13;
5148 ar |= (var->db & 1) << 14;
5149 ar |= (var->g & 1) << 15;
5155 static void vmx_set_segment(struct kvm_vcpu *vcpu,
5156 struct kvm_segment *var, int seg)
5158 struct vcpu_vmx *vmx = to_vmx(vcpu);
5159 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
5161 vmx_segment_cache_clear(vmx);
5163 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
5164 vmx->rmode.segs[seg] = *var;
5165 if (seg == VCPU_SREG_TR)
5166 vmcs_write16(sf->selector, var->selector);
5168 fix_rmode_seg(seg, &vmx->rmode.segs[seg]);
5172 vmcs_writel(sf->base, var->base);
5173 vmcs_write32(sf->limit, var->limit);
5174 vmcs_write16(sf->selector, var->selector);
5177 * Fix the "Accessed" bit in AR field of segment registers for older
5179 * IA32 arch specifies that at the time of processor reset the
5180 * "Accessed" bit in the AR field of segment registers is 1. And qemu
5181 * is setting it to 0 in the userland code. This causes invalid guest
5182 * state vmexit when "unrestricted guest" mode is turned on.
5183 * Fix for this setup issue in cpu_reset is being pushed in the qemu
5184 * tree. Newer qemu binaries with that qemu fix would not need this
5187 if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
5188 var->type |= 0x1; /* Accessed */
5190 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(var));
5193 vmx->emulation_required = emulation_required(vcpu);
5196 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
5198 u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
5200 *db = (ar >> 14) & 1;
5201 *l = (ar >> 13) & 1;
5204 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
5206 dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
5207 dt->address = vmcs_readl(GUEST_IDTR_BASE);
5210 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
5212 vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
5213 vmcs_writel(GUEST_IDTR_BASE, dt->address);
5216 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
5218 dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
5219 dt->address = vmcs_readl(GUEST_GDTR_BASE);
5222 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
5224 vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
5225 vmcs_writel(GUEST_GDTR_BASE, dt->address);
5228 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
5230 struct kvm_segment var;
5233 vmx_get_segment(vcpu, &var, seg);
5235 if (seg == VCPU_SREG_CS)
5237 ar = vmx_segment_access_rights(&var);
5239 if (var.base != (var.selector << 4))
5241 if (var.limit != 0xffff)
5249 static bool code_segment_valid(struct kvm_vcpu *vcpu)
5251 struct kvm_segment cs;
5252 unsigned int cs_rpl;
5254 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
5255 cs_rpl = cs.selector & SEGMENT_RPL_MASK;
5259 if (~cs.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_ACCESSES_MASK))
5263 if (cs.type & VMX_AR_TYPE_WRITEABLE_MASK) {
5264 if (cs.dpl > cs_rpl)
5267 if (cs.dpl != cs_rpl)
5273 /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
5277 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
5279 struct kvm_segment ss;
5280 unsigned int ss_rpl;
5282 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
5283 ss_rpl = ss.selector & SEGMENT_RPL_MASK;
5287 if (ss.type != 3 && ss.type != 7)
5291 if (ss.dpl != ss_rpl) /* DPL != RPL */
5299 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
5301 struct kvm_segment var;
5304 vmx_get_segment(vcpu, &var, seg);
5305 rpl = var.selector & SEGMENT_RPL_MASK;
5313 if (~var.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_WRITEABLE_MASK)) {
5314 if (var.dpl < rpl) /* DPL < RPL */
5318 /* TODO: Add other members to kvm_segment_field to allow checking for other access
5324 static bool tr_valid(struct kvm_vcpu *vcpu)
5326 struct kvm_segment tr;
5328 vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
5332 if (tr.selector & SEGMENT_TI_MASK) /* TI = 1 */
5334 if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
5342 static bool ldtr_valid(struct kvm_vcpu *vcpu)
5344 struct kvm_segment ldtr;
5346 vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
5350 if (ldtr.selector & SEGMENT_TI_MASK) /* TI = 1 */
5360 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
5362 struct kvm_segment cs, ss;
5364 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
5365 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
5367 return ((cs.selector & SEGMENT_RPL_MASK) ==
5368 (ss.selector & SEGMENT_RPL_MASK));
5372 * Check if guest state is valid. Returns true if valid, false if
5374 * We assume that registers are always usable
5376 static bool guest_state_valid(struct kvm_vcpu *vcpu)
5378 if (enable_unrestricted_guest)
5381 /* real mode guest state checks */
5382 if (!is_protmode(vcpu) || (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
5383 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
5385 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
5387 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
5389 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
5391 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
5393 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
5396 /* protected mode guest state checks */
5397 if (!cs_ss_rpl_check(vcpu))
5399 if (!code_segment_valid(vcpu))
5401 if (!stack_segment_valid(vcpu))
5403 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
5405 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
5407 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
5409 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
5411 if (!tr_valid(vcpu))
5413 if (!ldtr_valid(vcpu))
5417 * - Add checks on RIP
5418 * - Add checks on RFLAGS
5424 static bool page_address_valid(struct kvm_vcpu *vcpu, gpa_t gpa)
5426 return PAGE_ALIGNED(gpa) && !(gpa >> cpuid_maxphyaddr(vcpu));
5429 static int init_rmode_tss(struct kvm *kvm)
5435 idx = srcu_read_lock(&kvm->srcu);
5436 fn = to_kvm_vmx(kvm)->tss_addr >> PAGE_SHIFT;
5437 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
5440 data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
5441 r = kvm_write_guest_page(kvm, fn++, &data,
5442 TSS_IOPB_BASE_OFFSET, sizeof(u16));
5445 r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
5448 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
5452 r = kvm_write_guest_page(kvm, fn, &data,
5453 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
5456 srcu_read_unlock(&kvm->srcu, idx);
5460 static int init_rmode_identity_map(struct kvm *kvm)
5462 struct kvm_vmx *kvm_vmx = to_kvm_vmx(kvm);
5464 kvm_pfn_t identity_map_pfn;
5467 /* Protect kvm_vmx->ept_identity_pagetable_done. */
5468 mutex_lock(&kvm->slots_lock);
5470 if (likely(kvm_vmx->ept_identity_pagetable_done))
5473 if (!kvm_vmx->ept_identity_map_addr)
5474 kvm_vmx->ept_identity_map_addr = VMX_EPT_IDENTITY_PAGETABLE_ADDR;
5475 identity_map_pfn = kvm_vmx->ept_identity_map_addr >> PAGE_SHIFT;
5477 r = __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
5478 kvm_vmx->ept_identity_map_addr, PAGE_SIZE);
5482 idx = srcu_read_lock(&kvm->srcu);
5483 r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
5486 /* Set up identity-mapping pagetable for EPT in real mode */
5487 for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
5488 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
5489 _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
5490 r = kvm_write_guest_page(kvm, identity_map_pfn,
5491 &tmp, i * sizeof(tmp), sizeof(tmp));
5495 kvm_vmx->ept_identity_pagetable_done = true;
5498 srcu_read_unlock(&kvm->srcu, idx);
5501 mutex_unlock(&kvm->slots_lock);
5505 static void seg_setup(int seg)
5507 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
5510 vmcs_write16(sf->selector, 0);
5511 vmcs_writel(sf->base, 0);
5512 vmcs_write32(sf->limit, 0xffff);
5514 if (seg == VCPU_SREG_CS)
5515 ar |= 0x08; /* code segment */
5517 vmcs_write32(sf->ar_bytes, ar);
5520 static int alloc_apic_access_page(struct kvm *kvm)
5525 mutex_lock(&kvm->slots_lock);
5526 if (kvm->arch.apic_access_page_done)
5528 r = __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
5529 APIC_DEFAULT_PHYS_BASE, PAGE_SIZE);
5533 page = gfn_to_page(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
5534 if (is_error_page(page)) {
5540 * Do not pin the page in memory, so that memory hot-unplug
5541 * is able to migrate it.
5544 kvm->arch.apic_access_page_done = true;
5546 mutex_unlock(&kvm->slots_lock);
5550 static int allocate_vpid(void)
5556 spin_lock(&vmx_vpid_lock);
5557 vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
5558 if (vpid < VMX_NR_VPIDS)
5559 __set_bit(vpid, vmx_vpid_bitmap);
5562 spin_unlock(&vmx_vpid_lock);
5566 static void free_vpid(int vpid)
5568 if (!enable_vpid || vpid == 0)
5570 spin_lock(&vmx_vpid_lock);
5571 __clear_bit(vpid, vmx_vpid_bitmap);
5572 spin_unlock(&vmx_vpid_lock);
5575 static void __always_inline vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
5578 int f = sizeof(unsigned long);
5580 if (!cpu_has_vmx_msr_bitmap())
5583 if (static_branch_unlikely(&enable_evmcs))
5584 evmcs_touch_msr_bitmap();
5587 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
5588 * have the write-low and read-high bitmap offsets the wrong way round.
5589 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
5591 if (msr <= 0x1fff) {
5592 if (type & MSR_TYPE_R)
5594 __clear_bit(msr, msr_bitmap + 0x000 / f);
5596 if (type & MSR_TYPE_W)
5598 __clear_bit(msr, msr_bitmap + 0x800 / f);
5600 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
5602 if (type & MSR_TYPE_R)
5604 __clear_bit(msr, msr_bitmap + 0x400 / f);
5606 if (type & MSR_TYPE_W)
5608 __clear_bit(msr, msr_bitmap + 0xc00 / f);
5613 static void __always_inline vmx_enable_intercept_for_msr(unsigned long *msr_bitmap,
5616 int f = sizeof(unsigned long);
5618 if (!cpu_has_vmx_msr_bitmap())
5621 if (static_branch_unlikely(&enable_evmcs))
5622 evmcs_touch_msr_bitmap();
5625 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
5626 * have the write-low and read-high bitmap offsets the wrong way round.
5627 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
5629 if (msr <= 0x1fff) {
5630 if (type & MSR_TYPE_R)
5632 __set_bit(msr, msr_bitmap + 0x000 / f);
5634 if (type & MSR_TYPE_W)
5636 __set_bit(msr, msr_bitmap + 0x800 / f);
5638 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
5640 if (type & MSR_TYPE_R)
5642 __set_bit(msr, msr_bitmap + 0x400 / f);
5644 if (type & MSR_TYPE_W)
5646 __set_bit(msr, msr_bitmap + 0xc00 / f);
5651 static void __always_inline vmx_set_intercept_for_msr(unsigned long *msr_bitmap,
5652 u32 msr, int type, bool value)
5655 vmx_enable_intercept_for_msr(msr_bitmap, msr, type);
5657 vmx_disable_intercept_for_msr(msr_bitmap, msr, type);
5661 * If a msr is allowed by L0, we should check whether it is allowed by L1.
5662 * The corresponding bit will be cleared unless both of L0 and L1 allow it.
5664 static void nested_vmx_disable_intercept_for_msr(unsigned long *msr_bitmap_l1,
5665 unsigned long *msr_bitmap_nested,
5668 int f = sizeof(unsigned long);
5671 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
5672 * have the write-low and read-high bitmap offsets the wrong way round.
5673 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
5675 if (msr <= 0x1fff) {
5676 if (type & MSR_TYPE_R &&
5677 !test_bit(msr, msr_bitmap_l1 + 0x000 / f))
5679 __clear_bit(msr, msr_bitmap_nested + 0x000 / f);
5681 if (type & MSR_TYPE_W &&
5682 !test_bit(msr, msr_bitmap_l1 + 0x800 / f))
5684 __clear_bit(msr, msr_bitmap_nested + 0x800 / f);
5686 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
5688 if (type & MSR_TYPE_R &&
5689 !test_bit(msr, msr_bitmap_l1 + 0x400 / f))
5691 __clear_bit(msr, msr_bitmap_nested + 0x400 / f);
5693 if (type & MSR_TYPE_W &&
5694 !test_bit(msr, msr_bitmap_l1 + 0xc00 / f))
5696 __clear_bit(msr, msr_bitmap_nested + 0xc00 / f);
5701 static u8 vmx_msr_bitmap_mode(struct kvm_vcpu *vcpu)
5705 if (cpu_has_secondary_exec_ctrls() &&
5706 (vmcs_read32(SECONDARY_VM_EXEC_CONTROL) &
5707 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE)) {
5708 mode |= MSR_BITMAP_MODE_X2APIC;
5709 if (enable_apicv && kvm_vcpu_apicv_active(vcpu))
5710 mode |= MSR_BITMAP_MODE_X2APIC_APICV;
5713 if (is_long_mode(vcpu))
5714 mode |= MSR_BITMAP_MODE_LM;
5719 #define X2APIC_MSR(r) (APIC_BASE_MSR + ((r) >> 4))
5721 static void vmx_update_msr_bitmap_x2apic(unsigned long *msr_bitmap,
5726 for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
5727 unsigned word = msr / BITS_PER_LONG;
5728 msr_bitmap[word] = (mode & MSR_BITMAP_MODE_X2APIC_APICV) ? 0 : ~0;
5729 msr_bitmap[word + (0x800 / sizeof(long))] = ~0;
5732 if (mode & MSR_BITMAP_MODE_X2APIC) {
5734 * TPR reads and writes can be virtualized even if virtual interrupt
5735 * delivery is not in use.
5737 vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_TASKPRI), MSR_TYPE_RW);
5738 if (mode & MSR_BITMAP_MODE_X2APIC_APICV) {
5739 vmx_enable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_TMCCT), MSR_TYPE_R);
5740 vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_EOI), MSR_TYPE_W);
5741 vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_SELF_IPI), MSR_TYPE_W);
5746 static void vmx_update_msr_bitmap(struct kvm_vcpu *vcpu)
5748 struct vcpu_vmx *vmx = to_vmx(vcpu);
5749 unsigned long *msr_bitmap = vmx->vmcs01.msr_bitmap;
5750 u8 mode = vmx_msr_bitmap_mode(vcpu);
5751 u8 changed = mode ^ vmx->msr_bitmap_mode;
5756 vmx_set_intercept_for_msr(msr_bitmap, MSR_KERNEL_GS_BASE, MSR_TYPE_RW,
5757 !(mode & MSR_BITMAP_MODE_LM));
5759 if (changed & (MSR_BITMAP_MODE_X2APIC | MSR_BITMAP_MODE_X2APIC_APICV))
5760 vmx_update_msr_bitmap_x2apic(msr_bitmap, mode);
5762 vmx->msr_bitmap_mode = mode;
5765 static bool vmx_get_enable_apicv(struct kvm_vcpu *vcpu)
5767 return enable_apicv;
5770 static void nested_mark_vmcs12_pages_dirty(struct kvm_vcpu *vcpu)
5772 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5776 * Don't need to mark the APIC access page dirty; it is never
5777 * written to by the CPU during APIC virtualization.
5780 if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
5781 gfn = vmcs12->virtual_apic_page_addr >> PAGE_SHIFT;
5782 kvm_vcpu_mark_page_dirty(vcpu, gfn);
5785 if (nested_cpu_has_posted_intr(vmcs12)) {
5786 gfn = vmcs12->posted_intr_desc_addr >> PAGE_SHIFT;
5787 kvm_vcpu_mark_page_dirty(vcpu, gfn);
5792 static void vmx_complete_nested_posted_interrupt(struct kvm_vcpu *vcpu)
5794 struct vcpu_vmx *vmx = to_vmx(vcpu);
5799 if (!vmx->nested.pi_desc || !vmx->nested.pi_pending)
5802 vmx->nested.pi_pending = false;
5803 if (!pi_test_and_clear_on(vmx->nested.pi_desc))
5806 max_irr = find_last_bit((unsigned long *)vmx->nested.pi_desc->pir, 256);
5807 if (max_irr != 256) {
5808 vapic_page = kmap(vmx->nested.virtual_apic_page);
5809 __kvm_apic_update_irr(vmx->nested.pi_desc->pir,
5810 vapic_page, &max_irr);
5811 kunmap(vmx->nested.virtual_apic_page);
5813 status = vmcs_read16(GUEST_INTR_STATUS);
5814 if ((u8)max_irr > ((u8)status & 0xff)) {
5816 status |= (u8)max_irr;
5817 vmcs_write16(GUEST_INTR_STATUS, status);
5821 nested_mark_vmcs12_pages_dirty(vcpu);
5824 static inline bool kvm_vcpu_trigger_posted_interrupt(struct kvm_vcpu *vcpu,
5828 int pi_vec = nested ? POSTED_INTR_NESTED_VECTOR : POSTED_INTR_VECTOR;
5830 if (vcpu->mode == IN_GUEST_MODE) {
5832 * The vector of interrupt to be delivered to vcpu had
5833 * been set in PIR before this function.
5835 * Following cases will be reached in this block, and
5836 * we always send a notification event in all cases as
5839 * Case 1: vcpu keeps in non-root mode. Sending a
5840 * notification event posts the interrupt to vcpu.
5842 * Case 2: vcpu exits to root mode and is still
5843 * runnable. PIR will be synced to vIRR before the
5844 * next vcpu entry. Sending a notification event in
5845 * this case has no effect, as vcpu is not in root
5848 * Case 3: vcpu exits to root mode and is blocked.
5849 * vcpu_block() has already synced PIR to vIRR and
5850 * never blocks vcpu if vIRR is not cleared. Therefore,
5851 * a blocked vcpu here does not wait for any requested
5852 * interrupts in PIR, and sending a notification event
5853 * which has no effect is safe here.
5856 apic->send_IPI_mask(get_cpu_mask(vcpu->cpu), pi_vec);
5863 static int vmx_deliver_nested_posted_interrupt(struct kvm_vcpu *vcpu,
5866 struct vcpu_vmx *vmx = to_vmx(vcpu);
5868 if (is_guest_mode(vcpu) &&
5869 vector == vmx->nested.posted_intr_nv) {
5871 * If a posted intr is not recognized by hardware,
5872 * we will accomplish it in the next vmentry.
5874 vmx->nested.pi_pending = true;
5875 kvm_make_request(KVM_REQ_EVENT, vcpu);
5876 /* the PIR and ON have been set by L1. */
5877 if (!kvm_vcpu_trigger_posted_interrupt(vcpu, true))
5878 kvm_vcpu_kick(vcpu);
5884 * Send interrupt to vcpu via posted interrupt way.
5885 * 1. If target vcpu is running(non-root mode), send posted interrupt
5886 * notification to vcpu and hardware will sync PIR to vIRR atomically.
5887 * 2. If target vcpu isn't running(root mode), kick it to pick up the
5888 * interrupt from PIR in next vmentry.
5890 static void vmx_deliver_posted_interrupt(struct kvm_vcpu *vcpu, int vector)
5892 struct vcpu_vmx *vmx = to_vmx(vcpu);
5895 r = vmx_deliver_nested_posted_interrupt(vcpu, vector);
5899 if (pi_test_and_set_pir(vector, &vmx->pi_desc))
5902 /* If a previous notification has sent the IPI, nothing to do. */
5903 if (pi_test_and_set_on(&vmx->pi_desc))
5906 if (!kvm_vcpu_trigger_posted_interrupt(vcpu, false))
5907 kvm_vcpu_kick(vcpu);
5911 * Set up the vmcs's constant host-state fields, i.e., host-state fields that
5912 * will not change in the lifetime of the guest.
5913 * Note that host-state that does change is set elsewhere. E.g., host-state
5914 * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
5916 static void vmx_set_constant_host_state(struct vcpu_vmx *vmx)
5921 unsigned long cr0, cr3, cr4;
5924 WARN_ON(cr0 & X86_CR0_TS);
5925 vmcs_writel(HOST_CR0, cr0); /* 22.2.3 */
5928 * Save the most likely value for this task's CR3 in the VMCS.
5929 * We can't use __get_current_cr3_fast() because we're not atomic.
5932 vmcs_writel(HOST_CR3, cr3); /* 22.2.3 FIXME: shadow tables */
5933 vmx->loaded_vmcs->vmcs_host_cr3 = cr3;
5935 /* Save the most likely value for this task's CR4 in the VMCS. */
5936 cr4 = cr4_read_shadow();
5937 vmcs_writel(HOST_CR4, cr4); /* 22.2.3, 22.2.5 */
5938 vmx->loaded_vmcs->vmcs_host_cr4 = cr4;
5940 vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */
5941 #ifdef CONFIG_X86_64
5943 * Load null selectors, so we can avoid reloading them in
5944 * __vmx_load_host_state(), in case userspace uses the null selectors
5945 * too (the expected case).
5947 vmcs_write16(HOST_DS_SELECTOR, 0);
5948 vmcs_write16(HOST_ES_SELECTOR, 0);
5950 vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
5951 vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS); /* 22.2.4 */
5953 vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
5954 vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8); /* 22.2.4 */
5957 vmcs_writel(HOST_IDTR_BASE, dt.address); /* 22.2.4 */
5958 vmx->host_idt_base = dt.address;
5960 vmcs_writel(HOST_RIP, vmx_return); /* 22.2.5 */
5962 rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
5963 vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
5964 rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
5965 vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl); /* 22.2.3 */
5967 if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
5968 rdmsr(MSR_IA32_CR_PAT, low32, high32);
5969 vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
5973 static void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
5975 vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
5977 vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
5978 if (is_guest_mode(&vmx->vcpu))
5979 vmx->vcpu.arch.cr4_guest_owned_bits &=
5980 ~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask;
5981 vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
5984 static u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx)
5986 u32 pin_based_exec_ctrl = vmcs_config.pin_based_exec_ctrl;
5988 if (!kvm_vcpu_apicv_active(&vmx->vcpu))
5989 pin_based_exec_ctrl &= ~PIN_BASED_POSTED_INTR;
5992 pin_based_exec_ctrl &= ~PIN_BASED_VIRTUAL_NMIS;
5994 /* Enable the preemption timer dynamically */
5995 pin_based_exec_ctrl &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
5996 return pin_based_exec_ctrl;
5999 static void vmx_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu)
6001 struct vcpu_vmx *vmx = to_vmx(vcpu);
6003 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, vmx_pin_based_exec_ctrl(vmx));
6004 if (cpu_has_secondary_exec_ctrls()) {
6005 if (kvm_vcpu_apicv_active(vcpu))
6006 vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
6007 SECONDARY_EXEC_APIC_REGISTER_VIRT |
6008 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
6010 vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL,
6011 SECONDARY_EXEC_APIC_REGISTER_VIRT |
6012 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
6015 if (cpu_has_vmx_msr_bitmap())
6016 vmx_update_msr_bitmap(vcpu);
6019 static u32 vmx_exec_control(struct vcpu_vmx *vmx)
6021 u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
6023 if (vmx->vcpu.arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)
6024 exec_control &= ~CPU_BASED_MOV_DR_EXITING;
6026 if (!cpu_need_tpr_shadow(&vmx->vcpu)) {
6027 exec_control &= ~CPU_BASED_TPR_SHADOW;
6028 #ifdef CONFIG_X86_64
6029 exec_control |= CPU_BASED_CR8_STORE_EXITING |
6030 CPU_BASED_CR8_LOAD_EXITING;
6034 exec_control |= CPU_BASED_CR3_STORE_EXITING |
6035 CPU_BASED_CR3_LOAD_EXITING |
6036 CPU_BASED_INVLPG_EXITING;
6037 if (kvm_mwait_in_guest(vmx->vcpu.kvm))
6038 exec_control &= ~(CPU_BASED_MWAIT_EXITING |
6039 CPU_BASED_MONITOR_EXITING);
6040 if (kvm_hlt_in_guest(vmx->vcpu.kvm))
6041 exec_control &= ~CPU_BASED_HLT_EXITING;
6042 return exec_control;
6045 static bool vmx_rdrand_supported(void)
6047 return vmcs_config.cpu_based_2nd_exec_ctrl &
6048 SECONDARY_EXEC_RDRAND_EXITING;
6051 static bool vmx_rdseed_supported(void)
6053 return vmcs_config.cpu_based_2nd_exec_ctrl &
6054 SECONDARY_EXEC_RDSEED_EXITING;
6057 static void vmx_compute_secondary_exec_control(struct vcpu_vmx *vmx)
6059 struct kvm_vcpu *vcpu = &vmx->vcpu;
6061 u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
6063 if (!cpu_need_virtualize_apic_accesses(vcpu))
6064 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
6066 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
6068 exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
6069 enable_unrestricted_guest = 0;
6070 /* Enable INVPCID for non-ept guests may cause performance regression. */
6071 exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
6073 if (!enable_unrestricted_guest)
6074 exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
6075 if (kvm_pause_in_guest(vmx->vcpu.kvm))
6076 exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
6077 if (!kvm_vcpu_apicv_active(vcpu))
6078 exec_control &= ~(SECONDARY_EXEC_APIC_REGISTER_VIRT |
6079 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
6080 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
6082 /* SECONDARY_EXEC_DESC is enabled/disabled on writes to CR4.UMIP,
6083 * in vmx_set_cr4. */
6084 exec_control &= ~SECONDARY_EXEC_DESC;
6086 /* SECONDARY_EXEC_SHADOW_VMCS is enabled when L1 executes VMPTRLD
6088 We can NOT enable shadow_vmcs here because we don't have yet
6091 exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
6094 exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
6096 if (vmx_xsaves_supported()) {
6097 /* Exposing XSAVES only when XSAVE is exposed */
6098 bool xsaves_enabled =
6099 guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
6100 guest_cpuid_has(vcpu, X86_FEATURE_XSAVES);
6102 if (!xsaves_enabled)
6103 exec_control &= ~SECONDARY_EXEC_XSAVES;
6107 vmx->nested.msrs.secondary_ctls_high |=
6108 SECONDARY_EXEC_XSAVES;
6110 vmx->nested.msrs.secondary_ctls_high &=
6111 ~SECONDARY_EXEC_XSAVES;
6115 if (vmx_rdtscp_supported()) {
6116 bool rdtscp_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP);
6117 if (!rdtscp_enabled)
6118 exec_control &= ~SECONDARY_EXEC_RDTSCP;
6122 vmx->nested.msrs.secondary_ctls_high |=
6123 SECONDARY_EXEC_RDTSCP;
6125 vmx->nested.msrs.secondary_ctls_high &=
6126 ~SECONDARY_EXEC_RDTSCP;
6130 if (vmx_invpcid_supported()) {
6131 /* Exposing INVPCID only when PCID is exposed */
6132 bool invpcid_enabled =
6133 guest_cpuid_has(vcpu, X86_FEATURE_INVPCID) &&
6134 guest_cpuid_has(vcpu, X86_FEATURE_PCID);
6136 if (!invpcid_enabled) {
6137 exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
6138 guest_cpuid_clear(vcpu, X86_FEATURE_INVPCID);
6142 if (invpcid_enabled)
6143 vmx->nested.msrs.secondary_ctls_high |=
6144 SECONDARY_EXEC_ENABLE_INVPCID;
6146 vmx->nested.msrs.secondary_ctls_high &=
6147 ~SECONDARY_EXEC_ENABLE_INVPCID;
6151 if (vmx_rdrand_supported()) {
6152 bool rdrand_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDRAND);
6154 exec_control &= ~SECONDARY_EXEC_RDRAND_EXITING;
6158 vmx->nested.msrs.secondary_ctls_high |=
6159 SECONDARY_EXEC_RDRAND_EXITING;
6161 vmx->nested.msrs.secondary_ctls_high &=
6162 ~SECONDARY_EXEC_RDRAND_EXITING;
6166 if (vmx_rdseed_supported()) {
6167 bool rdseed_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDSEED);
6169 exec_control &= ~SECONDARY_EXEC_RDSEED_EXITING;
6173 vmx->nested.msrs.secondary_ctls_high |=
6174 SECONDARY_EXEC_RDSEED_EXITING;
6176 vmx->nested.msrs.secondary_ctls_high &=
6177 ~SECONDARY_EXEC_RDSEED_EXITING;
6181 vmx->secondary_exec_control = exec_control;
6184 static void ept_set_mmio_spte_mask(void)
6187 * EPT Misconfigurations can be generated if the value of bits 2:0
6188 * of an EPT paging-structure entry is 110b (write/execute).
6190 kvm_mmu_set_mmio_spte_mask(VMX_EPT_RWX_MASK,
6191 VMX_EPT_MISCONFIG_WX_VALUE);
6194 #define VMX_XSS_EXIT_BITMAP 0
6196 * Sets up the vmcs for emulated real mode.
6198 static void vmx_vcpu_setup(struct vcpu_vmx *vmx)
6200 #ifdef CONFIG_X86_64
6205 if (enable_shadow_vmcs) {
6207 * At vCPU creation, "VMWRITE to any supported field
6208 * in the VMCS" is supported, so use the more
6209 * permissive vmx_vmread_bitmap to specify both read
6210 * and write permissions for the shadow VMCS.
6212 vmcs_write64(VMREAD_BITMAP, __pa(vmx_vmread_bitmap));
6213 vmcs_write64(VMWRITE_BITMAP, __pa(vmx_vmread_bitmap));
6215 if (cpu_has_vmx_msr_bitmap())
6216 vmcs_write64(MSR_BITMAP, __pa(vmx->vmcs01.msr_bitmap));
6218 vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
6221 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, vmx_pin_based_exec_ctrl(vmx));
6222 vmx->hv_deadline_tsc = -1;
6224 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, vmx_exec_control(vmx));
6226 if (cpu_has_secondary_exec_ctrls()) {
6227 vmx_compute_secondary_exec_control(vmx);
6228 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
6229 vmx->secondary_exec_control);
6232 if (kvm_vcpu_apicv_active(&vmx->vcpu)) {
6233 vmcs_write64(EOI_EXIT_BITMAP0, 0);
6234 vmcs_write64(EOI_EXIT_BITMAP1, 0);
6235 vmcs_write64(EOI_EXIT_BITMAP2, 0);
6236 vmcs_write64(EOI_EXIT_BITMAP3, 0);
6238 vmcs_write16(GUEST_INTR_STATUS, 0);
6240 vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR);
6241 vmcs_write64(POSTED_INTR_DESC_ADDR, __pa((&vmx->pi_desc)));
6244 if (!kvm_pause_in_guest(vmx->vcpu.kvm)) {
6245 vmcs_write32(PLE_GAP, ple_gap);
6246 vmx->ple_window = ple_window;
6247 vmx->ple_window_dirty = true;
6250 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
6251 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
6252 vmcs_write32(CR3_TARGET_COUNT, 0); /* 22.2.1 */
6254 vmcs_write16(HOST_FS_SELECTOR, 0); /* 22.2.4 */
6255 vmcs_write16(HOST_GS_SELECTOR, 0); /* 22.2.4 */
6256 vmx_set_constant_host_state(vmx);
6257 #ifdef CONFIG_X86_64
6258 rdmsrl(MSR_FS_BASE, a);
6259 vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
6260 rdmsrl(MSR_GS_BASE, a);
6261 vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
6263 vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
6264 vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
6267 if (cpu_has_vmx_vmfunc())
6268 vmcs_write64(VM_FUNCTION_CONTROL, 0);
6270 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
6271 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
6272 vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host));
6273 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
6274 vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest));
6276 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
6277 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
6279 for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i) {
6280 u32 index = vmx_msr_index[i];
6281 u32 data_low, data_high;
6284 if (rdmsr_safe(index, &data_low, &data_high) < 0)
6286 if (wrmsr_safe(index, data_low, data_high) < 0)
6288 vmx->guest_msrs[j].index = i;
6289 vmx->guest_msrs[j].data = 0;
6290 vmx->guest_msrs[j].mask = -1ull;
6294 if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
6295 rdmsrl(MSR_IA32_ARCH_CAPABILITIES, vmx->arch_capabilities);
6297 vm_exit_controls_init(vmx, vmcs_config.vmexit_ctrl);
6299 /* 22.2.1, 20.8.1 */
6300 vm_entry_controls_init(vmx, vmcs_config.vmentry_ctrl);
6302 vmx->vcpu.arch.cr0_guest_owned_bits = X86_CR0_TS;
6303 vmcs_writel(CR0_GUEST_HOST_MASK, ~X86_CR0_TS);
6305 set_cr4_guest_host_mask(vmx);
6307 if (vmx_xsaves_supported())
6308 vmcs_write64(XSS_EXIT_BITMAP, VMX_XSS_EXIT_BITMAP);
6311 ASSERT(vmx->pml_pg);
6312 vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
6313 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
6317 static void vmx_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
6319 struct vcpu_vmx *vmx = to_vmx(vcpu);
6320 struct msr_data apic_base_msr;
6323 vmx->rmode.vm86_active = 0;
6326 vcpu->arch.microcode_version = 0x100000000ULL;
6327 vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
6328 kvm_set_cr8(vcpu, 0);
6331 apic_base_msr.data = APIC_DEFAULT_PHYS_BASE |
6332 MSR_IA32_APICBASE_ENABLE;
6333 if (kvm_vcpu_is_reset_bsp(vcpu))
6334 apic_base_msr.data |= MSR_IA32_APICBASE_BSP;
6335 apic_base_msr.host_initiated = true;
6336 kvm_set_apic_base(vcpu, &apic_base_msr);
6339 vmx_segment_cache_clear(vmx);
6341 seg_setup(VCPU_SREG_CS);
6342 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
6343 vmcs_writel(GUEST_CS_BASE, 0xffff0000ul);
6345 seg_setup(VCPU_SREG_DS);
6346 seg_setup(VCPU_SREG_ES);
6347 seg_setup(VCPU_SREG_FS);
6348 seg_setup(VCPU_SREG_GS);
6349 seg_setup(VCPU_SREG_SS);
6351 vmcs_write16(GUEST_TR_SELECTOR, 0);
6352 vmcs_writel(GUEST_TR_BASE, 0);
6353 vmcs_write32(GUEST_TR_LIMIT, 0xffff);
6354 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
6356 vmcs_write16(GUEST_LDTR_SELECTOR, 0);
6357 vmcs_writel(GUEST_LDTR_BASE, 0);
6358 vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
6359 vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
6362 vmcs_write32(GUEST_SYSENTER_CS, 0);
6363 vmcs_writel(GUEST_SYSENTER_ESP, 0);
6364 vmcs_writel(GUEST_SYSENTER_EIP, 0);
6365 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
6368 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
6369 kvm_rip_write(vcpu, 0xfff0);
6371 vmcs_writel(GUEST_GDTR_BASE, 0);
6372 vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
6374 vmcs_writel(GUEST_IDTR_BASE, 0);
6375 vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
6377 vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
6378 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
6379 vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS, 0);
6380 if (kvm_mpx_supported())
6381 vmcs_write64(GUEST_BNDCFGS, 0);
6385 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */
6387 if (cpu_has_vmx_tpr_shadow() && !init_event) {
6388 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
6389 if (cpu_need_tpr_shadow(vcpu))
6390 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
6391 __pa(vcpu->arch.apic->regs));
6392 vmcs_write32(TPR_THRESHOLD, 0);
6395 kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
6398 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
6400 cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
6401 vmx->vcpu.arch.cr0 = cr0;
6402 vmx_set_cr0(vcpu, cr0); /* enter rmode */
6403 vmx_set_cr4(vcpu, 0);
6404 vmx_set_efer(vcpu, 0);
6406 update_exception_bitmap(vcpu);
6408 vpid_sync_context(vmx->vpid);
6410 vmx_clear_hlt(vcpu);
6414 * In nested virtualization, check if L1 asked to exit on external interrupts.
6415 * For most existing hypervisors, this will always return true.
6417 static bool nested_exit_on_intr(struct kvm_vcpu *vcpu)
6419 return get_vmcs12(vcpu)->pin_based_vm_exec_control &
6420 PIN_BASED_EXT_INTR_MASK;
6424 * In nested virtualization, check if L1 has set
6425 * VM_EXIT_ACK_INTR_ON_EXIT
6427 static bool nested_exit_intr_ack_set(struct kvm_vcpu *vcpu)
6429 return get_vmcs12(vcpu)->vm_exit_controls &
6430 VM_EXIT_ACK_INTR_ON_EXIT;
6433 static bool nested_exit_on_nmi(struct kvm_vcpu *vcpu)
6435 return nested_cpu_has_nmi_exiting(get_vmcs12(vcpu));
6438 static void enable_irq_window(struct kvm_vcpu *vcpu)
6440 vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL,
6441 CPU_BASED_VIRTUAL_INTR_PENDING);
6444 static void enable_nmi_window(struct kvm_vcpu *vcpu)
6447 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
6448 enable_irq_window(vcpu);
6452 vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL,
6453 CPU_BASED_VIRTUAL_NMI_PENDING);
6456 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
6458 struct vcpu_vmx *vmx = to_vmx(vcpu);
6460 int irq = vcpu->arch.interrupt.nr;
6462 trace_kvm_inj_virq(irq);
6464 ++vcpu->stat.irq_injections;
6465 if (vmx->rmode.vm86_active) {
6467 if (vcpu->arch.interrupt.soft)
6468 inc_eip = vcpu->arch.event_exit_inst_len;
6469 if (kvm_inject_realmode_interrupt(vcpu, irq, inc_eip) != EMULATE_DONE)
6470 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
6473 intr = irq | INTR_INFO_VALID_MASK;
6474 if (vcpu->arch.interrupt.soft) {
6475 intr |= INTR_TYPE_SOFT_INTR;
6476 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
6477 vmx->vcpu.arch.event_exit_inst_len);
6479 intr |= INTR_TYPE_EXT_INTR;
6480 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
6482 vmx_clear_hlt(vcpu);
6485 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
6487 struct vcpu_vmx *vmx = to_vmx(vcpu);
6491 * Tracking the NMI-blocked state in software is built upon
6492 * finding the next open IRQ window. This, in turn, depends on
6493 * well-behaving guests: They have to keep IRQs disabled at
6494 * least as long as the NMI handler runs. Otherwise we may
6495 * cause NMI nesting, maybe breaking the guest. But as this is
6496 * highly unlikely, we can live with the residual risk.
6498 vmx->loaded_vmcs->soft_vnmi_blocked = 1;
6499 vmx->loaded_vmcs->vnmi_blocked_time = 0;
6502 ++vcpu->stat.nmi_injections;
6503 vmx->loaded_vmcs->nmi_known_unmasked = false;
6505 if (vmx->rmode.vm86_active) {
6506 if (kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0) != EMULATE_DONE)
6507 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
6511 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
6512 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
6514 vmx_clear_hlt(vcpu);
6517 static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
6519 struct vcpu_vmx *vmx = to_vmx(vcpu);
6523 return vmx->loaded_vmcs->soft_vnmi_blocked;
6524 if (vmx->loaded_vmcs->nmi_known_unmasked)
6526 masked = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
6527 vmx->loaded_vmcs->nmi_known_unmasked = !masked;
6531 static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
6533 struct vcpu_vmx *vmx = to_vmx(vcpu);
6536 if (vmx->loaded_vmcs->soft_vnmi_blocked != masked) {
6537 vmx->loaded_vmcs->soft_vnmi_blocked = masked;
6538 vmx->loaded_vmcs->vnmi_blocked_time = 0;
6541 vmx->loaded_vmcs->nmi_known_unmasked = !masked;
6543 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
6544 GUEST_INTR_STATE_NMI);
6546 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
6547 GUEST_INTR_STATE_NMI);
6551 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
6553 if (to_vmx(vcpu)->nested.nested_run_pending)
6557 to_vmx(vcpu)->loaded_vmcs->soft_vnmi_blocked)
6560 return !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
6561 (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
6562 | GUEST_INTR_STATE_NMI));
6565 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
6567 return (!to_vmx(vcpu)->nested.nested_run_pending &&
6568 vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
6569 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
6570 (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
6573 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
6577 if (enable_unrestricted_guest)
6580 ret = x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, addr,
6584 to_kvm_vmx(kvm)->tss_addr = addr;
6585 return init_rmode_tss(kvm);
6588 static int vmx_set_identity_map_addr(struct kvm *kvm, u64 ident_addr)
6590 to_kvm_vmx(kvm)->ept_identity_map_addr = ident_addr;
6594 static bool rmode_exception(struct kvm_vcpu *vcpu, int vec)
6599 * Update instruction length as we may reinject the exception
6600 * from user space while in guest debugging mode.
6602 to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
6603 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
6604 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
6608 if (vcpu->guest_debug &
6609 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
6626 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
6627 int vec, u32 err_code)
6630 * Instruction with address size override prefix opcode 0x67
6631 * Cause the #SS fault with 0 error code in VM86 mode.
6633 if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0) {
6634 if (emulate_instruction(vcpu, 0) == EMULATE_DONE) {
6635 if (vcpu->arch.halt_request) {
6636 vcpu->arch.halt_request = 0;
6637 return kvm_vcpu_halt(vcpu);
6645 * Forward all other exceptions that are valid in real mode.
6646 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
6647 * the required debugging infrastructure rework.
6649 kvm_queue_exception(vcpu, vec);
6654 * Trigger machine check on the host. We assume all the MSRs are already set up
6655 * by the CPU and that we still run on the same CPU as the MCE occurred on.
6656 * We pass a fake environment to the machine check handler because we want
6657 * the guest to be always treated like user space, no matter what context
6658 * it used internally.
6660 static void kvm_machine_check(void)
6662 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
6663 struct pt_regs regs = {
6664 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
6665 .flags = X86_EFLAGS_IF,
6668 do_machine_check(®s, 0);
6672 static int handle_machine_check(struct kvm_vcpu *vcpu)
6674 /* already handled by vcpu_run */
6678 static int handle_exception(struct kvm_vcpu *vcpu)
6680 struct vcpu_vmx *vmx = to_vmx(vcpu);
6681 struct kvm_run *kvm_run = vcpu->run;
6682 u32 intr_info, ex_no, error_code;
6683 unsigned long cr2, rip, dr6;
6685 enum emulation_result er;
6687 vect_info = vmx->idt_vectoring_info;
6688 intr_info = vmx->exit_intr_info;
6690 if (is_machine_check(intr_info))
6691 return handle_machine_check(vcpu);
6693 if (is_nmi(intr_info))
6694 return 1; /* already handled by vmx_vcpu_run() */
6696 if (is_invalid_opcode(intr_info))
6697 return handle_ud(vcpu);
6700 if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
6701 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
6703 if (!vmx->rmode.vm86_active && is_gp_fault(intr_info)) {
6704 WARN_ON_ONCE(!enable_vmware_backdoor);
6705 er = emulate_instruction(vcpu,
6706 EMULTYPE_VMWARE | EMULTYPE_NO_UD_ON_FAIL);
6707 if (er == EMULATE_USER_EXIT)
6709 else if (er != EMULATE_DONE)
6710 kvm_queue_exception_e(vcpu, GP_VECTOR, error_code);
6715 * The #PF with PFEC.RSVD = 1 indicates the guest is accessing
6716 * MMIO, it is better to report an internal error.
6717 * See the comments in vmx_handle_exit.
6719 if ((vect_info & VECTORING_INFO_VALID_MASK) &&
6720 !(is_page_fault(intr_info) && !(error_code & PFERR_RSVD_MASK))) {
6721 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6722 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
6723 vcpu->run->internal.ndata = 3;
6724 vcpu->run->internal.data[0] = vect_info;
6725 vcpu->run->internal.data[1] = intr_info;
6726 vcpu->run->internal.data[2] = error_code;
6730 if (is_page_fault(intr_info)) {
6731 cr2 = vmcs_readl(EXIT_QUALIFICATION);
6732 /* EPT won't cause page fault directly */
6733 WARN_ON_ONCE(!vcpu->arch.apf.host_apf_reason && enable_ept);
6734 return kvm_handle_page_fault(vcpu, error_code, cr2, NULL, 0);
6737 ex_no = intr_info & INTR_INFO_VECTOR_MASK;
6739 if (vmx->rmode.vm86_active && rmode_exception(vcpu, ex_no))
6740 return handle_rmode_exception(vcpu, ex_no, error_code);
6744 kvm_queue_exception_e(vcpu, AC_VECTOR, error_code);
6747 dr6 = vmcs_readl(EXIT_QUALIFICATION);
6748 if (!(vcpu->guest_debug &
6749 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
6750 vcpu->arch.dr6 &= ~15;
6751 vcpu->arch.dr6 |= dr6 | DR6_RTM;
6752 if (is_icebp(intr_info))
6753 skip_emulated_instruction(vcpu);
6755 kvm_queue_exception(vcpu, DB_VECTOR);
6758 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
6759 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
6763 * Update instruction length as we may reinject #BP from
6764 * user space while in guest debugging mode. Reading it for
6765 * #DB as well causes no harm, it is not used in that case.
6767 vmx->vcpu.arch.event_exit_inst_len =
6768 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
6769 kvm_run->exit_reason = KVM_EXIT_DEBUG;
6770 rip = kvm_rip_read(vcpu);
6771 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
6772 kvm_run->debug.arch.exception = ex_no;
6775 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
6776 kvm_run->ex.exception = ex_no;
6777 kvm_run->ex.error_code = error_code;
6783 static int handle_external_interrupt(struct kvm_vcpu *vcpu)
6785 ++vcpu->stat.irq_exits;
6789 static int handle_triple_fault(struct kvm_vcpu *vcpu)
6791 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
6792 vcpu->mmio_needed = 0;
6796 static int handle_io(struct kvm_vcpu *vcpu)
6798 unsigned long exit_qualification;
6799 int size, in, string;
6802 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6803 string = (exit_qualification & 16) != 0;
6805 ++vcpu->stat.io_exits;
6808 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
6810 port = exit_qualification >> 16;
6811 size = (exit_qualification & 7) + 1;
6812 in = (exit_qualification & 8) != 0;
6814 return kvm_fast_pio(vcpu, size, port, in);
6818 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
6821 * Patch in the VMCALL instruction:
6823 hypercall[0] = 0x0f;
6824 hypercall[1] = 0x01;
6825 hypercall[2] = 0xc1;
6828 /* called to set cr0 as appropriate for a mov-to-cr0 exit. */
6829 static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
6831 if (is_guest_mode(vcpu)) {
6832 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6833 unsigned long orig_val = val;
6836 * We get here when L2 changed cr0 in a way that did not change
6837 * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
6838 * but did change L0 shadowed bits. So we first calculate the
6839 * effective cr0 value that L1 would like to write into the
6840 * hardware. It consists of the L2-owned bits from the new
6841 * value combined with the L1-owned bits from L1's guest_cr0.
6843 val = (val & ~vmcs12->cr0_guest_host_mask) |
6844 (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask);
6846 if (!nested_guest_cr0_valid(vcpu, val))
6849 if (kvm_set_cr0(vcpu, val))
6851 vmcs_writel(CR0_READ_SHADOW, orig_val);
6854 if (to_vmx(vcpu)->nested.vmxon &&
6855 !nested_host_cr0_valid(vcpu, val))
6858 return kvm_set_cr0(vcpu, val);
6862 static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
6864 if (is_guest_mode(vcpu)) {
6865 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6866 unsigned long orig_val = val;
6868 /* analogously to handle_set_cr0 */
6869 val = (val & ~vmcs12->cr4_guest_host_mask) |
6870 (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask);
6871 if (kvm_set_cr4(vcpu, val))
6873 vmcs_writel(CR4_READ_SHADOW, orig_val);
6876 return kvm_set_cr4(vcpu, val);
6879 static int handle_desc(struct kvm_vcpu *vcpu)
6881 WARN_ON(!(vcpu->arch.cr4 & X86_CR4_UMIP));
6882 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
6885 static int handle_cr(struct kvm_vcpu *vcpu)
6887 unsigned long exit_qualification, val;
6893 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6894 cr = exit_qualification & 15;
6895 reg = (exit_qualification >> 8) & 15;
6896 switch ((exit_qualification >> 4) & 3) {
6897 case 0: /* mov to cr */
6898 val = kvm_register_readl(vcpu, reg);
6899 trace_kvm_cr_write(cr, val);
6902 err = handle_set_cr0(vcpu, val);
6903 return kvm_complete_insn_gp(vcpu, err);
6905 WARN_ON_ONCE(enable_unrestricted_guest);
6906 err = kvm_set_cr3(vcpu, val);
6907 return kvm_complete_insn_gp(vcpu, err);
6909 err = handle_set_cr4(vcpu, val);
6910 return kvm_complete_insn_gp(vcpu, err);
6912 u8 cr8_prev = kvm_get_cr8(vcpu);
6914 err = kvm_set_cr8(vcpu, cr8);
6915 ret = kvm_complete_insn_gp(vcpu, err);
6916 if (lapic_in_kernel(vcpu))
6918 if (cr8_prev <= cr8)
6921 * TODO: we might be squashing a
6922 * KVM_GUESTDBG_SINGLESTEP-triggered
6923 * KVM_EXIT_DEBUG here.
6925 vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
6931 WARN_ONCE(1, "Guest should always own CR0.TS");
6932 vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
6933 trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
6934 return kvm_skip_emulated_instruction(vcpu);
6935 case 1: /*mov from cr*/
6938 WARN_ON_ONCE(enable_unrestricted_guest);
6939 val = kvm_read_cr3(vcpu);
6940 kvm_register_write(vcpu, reg, val);
6941 trace_kvm_cr_read(cr, val);
6942 return kvm_skip_emulated_instruction(vcpu);
6944 val = kvm_get_cr8(vcpu);
6945 kvm_register_write(vcpu, reg, val);
6946 trace_kvm_cr_read(cr, val);
6947 return kvm_skip_emulated_instruction(vcpu);
6951 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
6952 trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
6953 kvm_lmsw(vcpu, val);
6955 return kvm_skip_emulated_instruction(vcpu);
6959 vcpu->run->exit_reason = 0;
6960 vcpu_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
6961 (int)(exit_qualification >> 4) & 3, cr);
6965 static int handle_dr(struct kvm_vcpu *vcpu)
6967 unsigned long exit_qualification;
6970 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6971 dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
6973 /* First, if DR does not exist, trigger UD */
6974 if (!kvm_require_dr(vcpu, dr))
6977 /* Do not handle if the CPL > 0, will trigger GP on re-entry */
6978 if (!kvm_require_cpl(vcpu, 0))
6980 dr7 = vmcs_readl(GUEST_DR7);
6983 * As the vm-exit takes precedence over the debug trap, we
6984 * need to emulate the latter, either for the host or the
6985 * guest debugging itself.
6987 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
6988 vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
6989 vcpu->run->debug.arch.dr7 = dr7;
6990 vcpu->run->debug.arch.pc = kvm_get_linear_rip(vcpu);
6991 vcpu->run->debug.arch.exception = DB_VECTOR;
6992 vcpu->run->exit_reason = KVM_EXIT_DEBUG;
6995 vcpu->arch.dr6 &= ~15;
6996 vcpu->arch.dr6 |= DR6_BD | DR6_RTM;
6997 kvm_queue_exception(vcpu, DB_VECTOR);
7002 if (vcpu->guest_debug == 0) {
7003 vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
7004 CPU_BASED_MOV_DR_EXITING);
7007 * No more DR vmexits; force a reload of the debug registers
7008 * and reenter on this instruction. The next vmexit will
7009 * retrieve the full state of the debug registers.
7011 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
7015 reg = DEBUG_REG_ACCESS_REG(exit_qualification);
7016 if (exit_qualification & TYPE_MOV_FROM_DR) {
7019 if (kvm_get_dr(vcpu, dr, &val))
7021 kvm_register_write(vcpu, reg, val);
7023 if (kvm_set_dr(vcpu, dr, kvm_register_readl(vcpu, reg)))
7026 return kvm_skip_emulated_instruction(vcpu);
7029 static u64 vmx_get_dr6(struct kvm_vcpu *vcpu)
7031 return vcpu->arch.dr6;
7034 static void vmx_set_dr6(struct kvm_vcpu *vcpu, unsigned long val)
7038 static void vmx_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
7040 get_debugreg(vcpu->arch.db[0], 0);
7041 get_debugreg(vcpu->arch.db[1], 1);
7042 get_debugreg(vcpu->arch.db[2], 2);
7043 get_debugreg(vcpu->arch.db[3], 3);
7044 get_debugreg(vcpu->arch.dr6, 6);
7045 vcpu->arch.dr7 = vmcs_readl(GUEST_DR7);
7047 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
7048 vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL, CPU_BASED_MOV_DR_EXITING);
7051 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
7053 vmcs_writel(GUEST_DR7, val);
7056 static int handle_cpuid(struct kvm_vcpu *vcpu)
7058 return kvm_emulate_cpuid(vcpu);
7061 static int handle_rdmsr(struct kvm_vcpu *vcpu)
7063 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
7064 struct msr_data msr_info;
7066 msr_info.index = ecx;
7067 msr_info.host_initiated = false;
7068 if (vmx_get_msr(vcpu, &msr_info)) {
7069 trace_kvm_msr_read_ex(ecx);
7070 kvm_inject_gp(vcpu, 0);
7074 trace_kvm_msr_read(ecx, msr_info.data);
7076 /* FIXME: handling of bits 32:63 of rax, rdx */
7077 vcpu->arch.regs[VCPU_REGS_RAX] = msr_info.data & -1u;
7078 vcpu->arch.regs[VCPU_REGS_RDX] = (msr_info.data >> 32) & -1u;
7079 return kvm_skip_emulated_instruction(vcpu);
7082 static int handle_wrmsr(struct kvm_vcpu *vcpu)
7084 struct msr_data msr;
7085 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
7086 u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
7087 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
7091 msr.host_initiated = false;
7092 if (kvm_set_msr(vcpu, &msr) != 0) {
7093 trace_kvm_msr_write_ex(ecx, data);
7094 kvm_inject_gp(vcpu, 0);
7098 trace_kvm_msr_write(ecx, data);
7099 return kvm_skip_emulated_instruction(vcpu);
7102 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
7104 kvm_apic_update_ppr(vcpu);
7108 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
7110 vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
7111 CPU_BASED_VIRTUAL_INTR_PENDING);
7113 kvm_make_request(KVM_REQ_EVENT, vcpu);
7115 ++vcpu->stat.irq_window_exits;
7119 static int handle_halt(struct kvm_vcpu *vcpu)
7121 return kvm_emulate_halt(vcpu);
7124 static int handle_vmcall(struct kvm_vcpu *vcpu)
7126 return kvm_emulate_hypercall(vcpu);
7129 static int handle_invd(struct kvm_vcpu *vcpu)
7131 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
7134 static int handle_invlpg(struct kvm_vcpu *vcpu)
7136 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7138 kvm_mmu_invlpg(vcpu, exit_qualification);
7139 return kvm_skip_emulated_instruction(vcpu);
7142 static int handle_rdpmc(struct kvm_vcpu *vcpu)
7146 err = kvm_rdpmc(vcpu);
7147 return kvm_complete_insn_gp(vcpu, err);
7150 static int handle_wbinvd(struct kvm_vcpu *vcpu)
7152 return kvm_emulate_wbinvd(vcpu);
7155 static int handle_xsetbv(struct kvm_vcpu *vcpu)
7157 u64 new_bv = kvm_read_edx_eax(vcpu);
7158 u32 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
7160 if (kvm_set_xcr(vcpu, index, new_bv) == 0)
7161 return kvm_skip_emulated_instruction(vcpu);
7165 static int handle_xsaves(struct kvm_vcpu *vcpu)
7167 kvm_skip_emulated_instruction(vcpu);
7168 WARN(1, "this should never happen\n");
7172 static int handle_xrstors(struct kvm_vcpu *vcpu)
7174 kvm_skip_emulated_instruction(vcpu);
7175 WARN(1, "this should never happen\n");
7179 static int handle_apic_access(struct kvm_vcpu *vcpu)
7181 if (likely(fasteoi)) {
7182 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7183 int access_type, offset;
7185 access_type = exit_qualification & APIC_ACCESS_TYPE;
7186 offset = exit_qualification & APIC_ACCESS_OFFSET;
7188 * Sane guest uses MOV to write EOI, with written value
7189 * not cared. So make a short-circuit here by avoiding
7190 * heavy instruction emulation.
7192 if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
7193 (offset == APIC_EOI)) {
7194 kvm_lapic_set_eoi(vcpu);
7195 return kvm_skip_emulated_instruction(vcpu);
7198 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
7201 static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu)
7203 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7204 int vector = exit_qualification & 0xff;
7206 /* EOI-induced VM exit is trap-like and thus no need to adjust IP */
7207 kvm_apic_set_eoi_accelerated(vcpu, vector);
7211 static int handle_apic_write(struct kvm_vcpu *vcpu)
7213 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7214 u32 offset = exit_qualification & 0xfff;
7216 /* APIC-write VM exit is trap-like and thus no need to adjust IP */
7217 kvm_apic_write_nodecode(vcpu, offset);
7221 static int handle_task_switch(struct kvm_vcpu *vcpu)
7223 struct vcpu_vmx *vmx = to_vmx(vcpu);
7224 unsigned long exit_qualification;
7225 bool has_error_code = false;
7228 int reason, type, idt_v, idt_index;
7230 idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
7231 idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK);
7232 type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
7234 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7236 reason = (u32)exit_qualification >> 30;
7237 if (reason == TASK_SWITCH_GATE && idt_v) {
7239 case INTR_TYPE_NMI_INTR:
7240 vcpu->arch.nmi_injected = false;
7241 vmx_set_nmi_mask(vcpu, true);
7243 case INTR_TYPE_EXT_INTR:
7244 case INTR_TYPE_SOFT_INTR:
7245 kvm_clear_interrupt_queue(vcpu);
7247 case INTR_TYPE_HARD_EXCEPTION:
7248 if (vmx->idt_vectoring_info &
7249 VECTORING_INFO_DELIVER_CODE_MASK) {
7250 has_error_code = true;
7252 vmcs_read32(IDT_VECTORING_ERROR_CODE);
7255 case INTR_TYPE_SOFT_EXCEPTION:
7256 kvm_clear_exception_queue(vcpu);
7262 tss_selector = exit_qualification;
7264 if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
7265 type != INTR_TYPE_EXT_INTR &&
7266 type != INTR_TYPE_NMI_INTR))
7267 skip_emulated_instruction(vcpu);
7269 if (kvm_task_switch(vcpu, tss_selector,
7270 type == INTR_TYPE_SOFT_INTR ? idt_index : -1, reason,
7271 has_error_code, error_code) == EMULATE_FAIL) {
7272 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
7273 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
7274 vcpu->run->internal.ndata = 0;
7279 * TODO: What about debug traps on tss switch?
7280 * Are we supposed to inject them and update dr6?
7286 static int handle_ept_violation(struct kvm_vcpu *vcpu)
7288 unsigned long exit_qualification;
7292 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7295 * EPT violation happened while executing iret from NMI,
7296 * "blocked by NMI" bit has to be set before next VM entry.
7297 * There are errata that may cause this bit to not be set:
7300 if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
7302 (exit_qualification & INTR_INFO_UNBLOCK_NMI))
7303 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, GUEST_INTR_STATE_NMI);
7305 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
7306 trace_kvm_page_fault(gpa, exit_qualification);
7308 /* Is it a read fault? */
7309 error_code = (exit_qualification & EPT_VIOLATION_ACC_READ)
7310 ? PFERR_USER_MASK : 0;
7311 /* Is it a write fault? */
7312 error_code |= (exit_qualification & EPT_VIOLATION_ACC_WRITE)
7313 ? PFERR_WRITE_MASK : 0;
7314 /* Is it a fetch fault? */
7315 error_code |= (exit_qualification & EPT_VIOLATION_ACC_INSTR)
7316 ? PFERR_FETCH_MASK : 0;
7317 /* ept page table entry is present? */
7318 error_code |= (exit_qualification &
7319 (EPT_VIOLATION_READABLE | EPT_VIOLATION_WRITABLE |
7320 EPT_VIOLATION_EXECUTABLE))
7321 ? PFERR_PRESENT_MASK : 0;
7323 error_code |= (exit_qualification & 0x100) != 0 ?
7324 PFERR_GUEST_FINAL_MASK : PFERR_GUEST_PAGE_MASK;
7326 vcpu->arch.exit_qualification = exit_qualification;
7327 return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0);
7330 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
7335 * A nested guest cannot optimize MMIO vmexits, because we have an
7336 * nGPA here instead of the required GPA.
7338 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
7339 if (!is_guest_mode(vcpu) &&
7340 !kvm_io_bus_write(vcpu, KVM_FAST_MMIO_BUS, gpa, 0, NULL)) {
7341 trace_kvm_fast_mmio(gpa);
7343 * Doing kvm_skip_emulated_instruction() depends on undefined
7344 * behavior: Intel's manual doesn't mandate
7345 * VM_EXIT_INSTRUCTION_LEN to be set in VMCS when EPT MISCONFIG
7346 * occurs and while on real hardware it was observed to be set,
7347 * other hypervisors (namely Hyper-V) don't set it, we end up
7348 * advancing IP with some random value. Disable fast mmio when
7349 * running nested and keep it for real hardware in hope that
7350 * VM_EXIT_INSTRUCTION_LEN will always be set correctly.
7352 if (!static_cpu_has(X86_FEATURE_HYPERVISOR))
7353 return kvm_skip_emulated_instruction(vcpu);
7355 return x86_emulate_instruction(vcpu, gpa, EMULTYPE_SKIP,
7356 NULL, 0) == EMULATE_DONE;
7359 return kvm_mmu_page_fault(vcpu, gpa, PFERR_RSVD_MASK, NULL, 0);
7362 static int handle_nmi_window(struct kvm_vcpu *vcpu)
7364 WARN_ON_ONCE(!enable_vnmi);
7365 vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
7366 CPU_BASED_VIRTUAL_NMI_PENDING);
7367 ++vcpu->stat.nmi_window_exits;
7368 kvm_make_request(KVM_REQ_EVENT, vcpu);
7373 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
7375 struct vcpu_vmx *vmx = to_vmx(vcpu);
7376 enum emulation_result err = EMULATE_DONE;
7379 bool intr_window_requested;
7380 unsigned count = 130;
7383 * We should never reach the point where we are emulating L2
7384 * due to invalid guest state as that means we incorrectly
7385 * allowed a nested VMEntry with an invalid vmcs12.
7387 WARN_ON_ONCE(vmx->emulation_required && vmx->nested.nested_run_pending);
7389 cpu_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
7390 intr_window_requested = cpu_exec_ctrl & CPU_BASED_VIRTUAL_INTR_PENDING;
7392 while (vmx->emulation_required && count-- != 0) {
7393 if (intr_window_requested && vmx_interrupt_allowed(vcpu))
7394 return handle_interrupt_window(&vmx->vcpu);
7396 if (kvm_test_request(KVM_REQ_EVENT, vcpu))
7399 err = emulate_instruction(vcpu, 0);
7401 if (err == EMULATE_USER_EXIT) {
7402 ++vcpu->stat.mmio_exits;
7407 if (err != EMULATE_DONE)
7408 goto emulation_error;
7410 if (vmx->emulation_required && !vmx->rmode.vm86_active &&
7411 vcpu->arch.exception.pending)
7412 goto emulation_error;
7414 if (vcpu->arch.halt_request) {
7415 vcpu->arch.halt_request = 0;
7416 ret = kvm_vcpu_halt(vcpu);
7420 if (signal_pending(current))
7430 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
7431 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
7432 vcpu->run->internal.ndata = 0;
7436 static void grow_ple_window(struct kvm_vcpu *vcpu)
7438 struct vcpu_vmx *vmx = to_vmx(vcpu);
7439 int old = vmx->ple_window;
7441 vmx->ple_window = __grow_ple_window(old, ple_window,
7445 if (vmx->ple_window != old)
7446 vmx->ple_window_dirty = true;
7448 trace_kvm_ple_window_grow(vcpu->vcpu_id, vmx->ple_window, old);
7451 static void shrink_ple_window(struct kvm_vcpu *vcpu)
7453 struct vcpu_vmx *vmx = to_vmx(vcpu);
7454 int old = vmx->ple_window;
7456 vmx->ple_window = __shrink_ple_window(old, ple_window,
7460 if (vmx->ple_window != old)
7461 vmx->ple_window_dirty = true;
7463 trace_kvm_ple_window_shrink(vcpu->vcpu_id, vmx->ple_window, old);
7467 * Handler for POSTED_INTERRUPT_WAKEUP_VECTOR.
7469 static void wakeup_handler(void)
7471 struct kvm_vcpu *vcpu;
7472 int cpu = smp_processor_id();
7474 spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
7475 list_for_each_entry(vcpu, &per_cpu(blocked_vcpu_on_cpu, cpu),
7476 blocked_vcpu_list) {
7477 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
7479 if (pi_test_on(pi_desc) == 1)
7480 kvm_vcpu_kick(vcpu);
7482 spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
7485 static void vmx_enable_tdp(void)
7487 kvm_mmu_set_mask_ptes(VMX_EPT_READABLE_MASK,
7488 enable_ept_ad_bits ? VMX_EPT_ACCESS_BIT : 0ull,
7489 enable_ept_ad_bits ? VMX_EPT_DIRTY_BIT : 0ull,
7490 0ull, VMX_EPT_EXECUTABLE_MASK,
7491 cpu_has_vmx_ept_execute_only() ? 0ull : VMX_EPT_READABLE_MASK,
7492 VMX_EPT_RWX_MASK, 0ull);
7494 ept_set_mmio_spte_mask();
7498 static __init int hardware_setup(void)
7500 unsigned long host_bndcfgs;
7503 rdmsrl_safe(MSR_EFER, &host_efer);
7505 for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i)
7506 kvm_define_shared_msr(i, vmx_msr_index[i]);
7508 for (i = 0; i < VMX_BITMAP_NR; i++) {
7509 vmx_bitmap[i] = (unsigned long *)__get_free_page(GFP_KERNEL);
7514 memset(vmx_vmread_bitmap, 0xff, PAGE_SIZE);
7515 memset(vmx_vmwrite_bitmap, 0xff, PAGE_SIZE);
7517 if (setup_vmcs_config(&vmcs_config) < 0) {
7522 if (boot_cpu_has(X86_FEATURE_NX))
7523 kvm_enable_efer_bits(EFER_NX);
7525 if (boot_cpu_has(X86_FEATURE_MPX)) {
7526 rdmsrl(MSR_IA32_BNDCFGS, host_bndcfgs);
7527 WARN_ONCE(host_bndcfgs, "KVM: BNDCFGS in host will be lost");
7530 if (!cpu_has_vmx_vpid() || !cpu_has_vmx_invvpid() ||
7531 !(cpu_has_vmx_invvpid_single() || cpu_has_vmx_invvpid_global()))
7534 if (!cpu_has_vmx_ept() ||
7535 !cpu_has_vmx_ept_4levels() ||
7536 !cpu_has_vmx_ept_mt_wb() ||
7537 !cpu_has_vmx_invept_global())
7540 if (!cpu_has_vmx_ept_ad_bits() || !enable_ept)
7541 enable_ept_ad_bits = 0;
7543 if (!cpu_has_vmx_unrestricted_guest() || !enable_ept)
7544 enable_unrestricted_guest = 0;
7546 if (!cpu_has_vmx_flexpriority())
7547 flexpriority_enabled = 0;
7549 if (!cpu_has_virtual_nmis())
7553 * set_apic_access_page_addr() is used to reload apic access
7554 * page upon invalidation. No need to do anything if not
7555 * using the APIC_ACCESS_ADDR VMCS field.
7557 if (!flexpriority_enabled)
7558 kvm_x86_ops->set_apic_access_page_addr = NULL;
7560 if (!cpu_has_vmx_tpr_shadow())
7561 kvm_x86_ops->update_cr8_intercept = NULL;
7563 if (enable_ept && !cpu_has_vmx_ept_2m_page())
7564 kvm_disable_largepages();
7566 if (!cpu_has_vmx_ple()) {
7569 ple_window_grow = 0;
7571 ple_window_shrink = 0;
7574 if (!cpu_has_vmx_apicv()) {
7576 kvm_x86_ops->sync_pir_to_irr = NULL;
7579 if (cpu_has_vmx_tsc_scaling()) {
7580 kvm_has_tsc_control = true;
7581 kvm_max_tsc_scaling_ratio = KVM_VMX_TSC_MULTIPLIER_MAX;
7582 kvm_tsc_scaling_ratio_frac_bits = 48;
7585 set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
7593 * Only enable PML when hardware supports PML feature, and both EPT
7594 * and EPT A/D bit features are enabled -- PML depends on them to work.
7596 if (!enable_ept || !enable_ept_ad_bits || !cpu_has_vmx_pml())
7600 kvm_x86_ops->slot_enable_log_dirty = NULL;
7601 kvm_x86_ops->slot_disable_log_dirty = NULL;
7602 kvm_x86_ops->flush_log_dirty = NULL;
7603 kvm_x86_ops->enable_log_dirty_pt_masked = NULL;
7606 if (cpu_has_vmx_preemption_timer() && enable_preemption_timer) {
7609 rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
7610 cpu_preemption_timer_multi =
7611 vmx_msr & VMX_MISC_PREEMPTION_TIMER_RATE_MASK;
7613 kvm_x86_ops->set_hv_timer = NULL;
7614 kvm_x86_ops->cancel_hv_timer = NULL;
7617 if (!cpu_has_vmx_shadow_vmcs())
7618 enable_shadow_vmcs = 0;
7619 if (enable_shadow_vmcs)
7620 init_vmcs_shadow_fields();
7622 kvm_set_posted_intr_wakeup_handler(wakeup_handler);
7623 nested_vmx_setup_ctls_msrs(&vmcs_config.nested, enable_apicv);
7625 kvm_mce_cap_supported |= MCG_LMCE_P;
7627 return alloc_kvm_area();
7630 for (i = 0; i < VMX_BITMAP_NR; i++)
7631 free_page((unsigned long)vmx_bitmap[i]);
7636 static __exit void hardware_unsetup(void)
7640 for (i = 0; i < VMX_BITMAP_NR; i++)
7641 free_page((unsigned long)vmx_bitmap[i]);
7647 * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
7648 * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
7650 static int handle_pause(struct kvm_vcpu *vcpu)
7652 if (!kvm_pause_in_guest(vcpu->kvm))
7653 grow_ple_window(vcpu);
7656 * Intel sdm vol3 ch-25.1.3 says: The "PAUSE-loop exiting"
7657 * VM-execution control is ignored if CPL > 0. OTOH, KVM
7658 * never set PAUSE_EXITING and just set PLE if supported,
7659 * so the vcpu must be CPL=0 if it gets a PAUSE exit.
7661 kvm_vcpu_on_spin(vcpu, true);
7662 return kvm_skip_emulated_instruction(vcpu);
7665 static int handle_nop(struct kvm_vcpu *vcpu)
7667 return kvm_skip_emulated_instruction(vcpu);
7670 static int handle_mwait(struct kvm_vcpu *vcpu)
7672 printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
7673 return handle_nop(vcpu);
7676 static int handle_invalid_op(struct kvm_vcpu *vcpu)
7678 kvm_queue_exception(vcpu, UD_VECTOR);
7682 static int handle_monitor_trap(struct kvm_vcpu *vcpu)
7687 static int handle_monitor(struct kvm_vcpu *vcpu)
7689 printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
7690 return handle_nop(vcpu);
7694 * The following 3 functions, nested_vmx_succeed()/failValid()/failInvalid(),
7695 * set the success or error code of an emulated VMX instruction, as specified
7696 * by Vol 2B, VMX Instruction Reference, "Conventions".
7698 static void nested_vmx_succeed(struct kvm_vcpu *vcpu)
7700 vmx_set_rflags(vcpu, vmx_get_rflags(vcpu)
7701 & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
7702 X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF));
7705 static void nested_vmx_failInvalid(struct kvm_vcpu *vcpu)
7707 vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
7708 & ~(X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
7709 X86_EFLAGS_SF | X86_EFLAGS_OF))
7713 static void nested_vmx_failValid(struct kvm_vcpu *vcpu,
7714 u32 vm_instruction_error)
7716 if (to_vmx(vcpu)->nested.current_vmptr == -1ull) {
7718 * failValid writes the error number to the current VMCS, which
7719 * can't be done there isn't a current VMCS.
7721 nested_vmx_failInvalid(vcpu);
7724 vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
7725 & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
7726 X86_EFLAGS_SF | X86_EFLAGS_OF))
7728 get_vmcs12(vcpu)->vm_instruction_error = vm_instruction_error;
7730 * We don't need to force a shadow sync because
7731 * VM_INSTRUCTION_ERROR is not shadowed
7735 static void nested_vmx_abort(struct kvm_vcpu *vcpu, u32 indicator)
7737 /* TODO: not to reset guest simply here. */
7738 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
7739 pr_debug_ratelimited("kvm: nested vmx abort, indicator %d\n", indicator);
7742 static enum hrtimer_restart vmx_preemption_timer_fn(struct hrtimer *timer)
7744 struct vcpu_vmx *vmx =
7745 container_of(timer, struct vcpu_vmx, nested.preemption_timer);
7747 vmx->nested.preemption_timer_expired = true;
7748 kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
7749 kvm_vcpu_kick(&vmx->vcpu);
7751 return HRTIMER_NORESTART;
7755 * Decode the memory-address operand of a vmx instruction, as recorded on an
7756 * exit caused by such an instruction (run by a guest hypervisor).
7757 * On success, returns 0. When the operand is invalid, returns 1 and throws
7760 static int get_vmx_mem_address(struct kvm_vcpu *vcpu,
7761 unsigned long exit_qualification,
7762 u32 vmx_instruction_info, bool wr, gva_t *ret)
7766 struct kvm_segment s;
7769 * According to Vol. 3B, "Information for VM Exits Due to Instruction
7770 * Execution", on an exit, vmx_instruction_info holds most of the
7771 * addressing components of the operand. Only the displacement part
7772 * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
7773 * For how an actual address is calculated from all these components,
7774 * refer to Vol. 1, "Operand Addressing".
7776 int scaling = vmx_instruction_info & 3;
7777 int addr_size = (vmx_instruction_info >> 7) & 7;
7778 bool is_reg = vmx_instruction_info & (1u << 10);
7779 int seg_reg = (vmx_instruction_info >> 15) & 7;
7780 int index_reg = (vmx_instruction_info >> 18) & 0xf;
7781 bool index_is_valid = !(vmx_instruction_info & (1u << 22));
7782 int base_reg = (vmx_instruction_info >> 23) & 0xf;
7783 bool base_is_valid = !(vmx_instruction_info & (1u << 27));
7786 kvm_queue_exception(vcpu, UD_VECTOR);
7790 /* Addr = segment_base + offset */
7791 /* offset = base + [index * scale] + displacement */
7792 off = exit_qualification; /* holds the displacement */
7794 off += kvm_register_read(vcpu, base_reg);
7796 off += kvm_register_read(vcpu, index_reg)<<scaling;
7797 vmx_get_segment(vcpu, &s, seg_reg);
7798 *ret = s.base + off;
7800 if (addr_size == 1) /* 32 bit */
7803 /* Checks for #GP/#SS exceptions. */
7805 if (is_long_mode(vcpu)) {
7806 /* Long mode: #GP(0)/#SS(0) if the memory address is in a
7807 * non-canonical form. This is the only check on the memory
7808 * destination for long mode!
7810 exn = is_noncanonical_address(*ret, vcpu);
7811 } else if (is_protmode(vcpu)) {
7812 /* Protected mode: apply checks for segment validity in the
7814 * - segment type check (#GP(0) may be thrown)
7815 * - usability check (#GP(0)/#SS(0))
7816 * - limit check (#GP(0)/#SS(0))
7819 /* #GP(0) if the destination operand is located in a
7820 * read-only data segment or any code segment.
7822 exn = ((s.type & 0xa) == 0 || (s.type & 8));
7824 /* #GP(0) if the source operand is located in an
7825 * execute-only code segment
7827 exn = ((s.type & 0xa) == 8);
7829 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
7832 /* Protected mode: #GP(0)/#SS(0) if the segment is unusable.
7834 exn = (s.unusable != 0);
7835 /* Protected mode: #GP(0)/#SS(0) if the memory
7836 * operand is outside the segment limit.
7838 exn = exn || (off + sizeof(u64) > s.limit);
7841 kvm_queue_exception_e(vcpu,
7842 seg_reg == VCPU_SREG_SS ?
7843 SS_VECTOR : GP_VECTOR,
7851 static int nested_vmx_get_vmptr(struct kvm_vcpu *vcpu, gpa_t *vmpointer)
7854 struct x86_exception e;
7856 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
7857 vmcs_read32(VMX_INSTRUCTION_INFO), false, &gva))
7860 if (kvm_read_guest_virt(vcpu, gva, vmpointer, sizeof(*vmpointer), &e)) {
7861 kvm_inject_page_fault(vcpu, &e);
7868 static int enter_vmx_operation(struct kvm_vcpu *vcpu)
7870 struct vcpu_vmx *vmx = to_vmx(vcpu);
7871 struct vmcs *shadow_vmcs;
7874 r = alloc_loaded_vmcs(&vmx->nested.vmcs02);
7878 vmx->nested.cached_vmcs12 = kmalloc(VMCS12_SIZE, GFP_KERNEL);
7879 if (!vmx->nested.cached_vmcs12)
7880 goto out_cached_vmcs12;
7882 if (enable_shadow_vmcs) {
7883 shadow_vmcs = alloc_vmcs();
7885 goto out_shadow_vmcs;
7886 /* mark vmcs as shadow */
7887 shadow_vmcs->revision_id |= (1u << 31);
7888 /* init shadow vmcs */
7889 vmcs_clear(shadow_vmcs);
7890 vmx->vmcs01.shadow_vmcs = shadow_vmcs;
7893 hrtimer_init(&vmx->nested.preemption_timer, CLOCK_MONOTONIC,
7894 HRTIMER_MODE_REL_PINNED);
7895 vmx->nested.preemption_timer.function = vmx_preemption_timer_fn;
7897 vmx->nested.vmxon = true;
7901 kfree(vmx->nested.cached_vmcs12);
7904 free_loaded_vmcs(&vmx->nested.vmcs02);
7911 * Emulate the VMXON instruction.
7912 * Currently, we just remember that VMX is active, and do not save or even
7913 * inspect the argument to VMXON (the so-called "VMXON pointer") because we
7914 * do not currently need to store anything in that guest-allocated memory
7915 * region. Consequently, VMCLEAR and VMPTRLD also do not verify that the their
7916 * argument is different from the VMXON pointer (which the spec says they do).
7918 static int handle_vmon(struct kvm_vcpu *vcpu)
7923 struct vcpu_vmx *vmx = to_vmx(vcpu);
7924 const u64 VMXON_NEEDED_FEATURES = FEATURE_CONTROL_LOCKED
7925 | FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
7928 * The Intel VMX Instruction Reference lists a bunch of bits that are
7929 * prerequisite to running VMXON, most notably cr4.VMXE must be set to
7930 * 1 (see vmx_set_cr4() for when we allow the guest to set this).
7931 * Otherwise, we should fail with #UD. But most faulting conditions
7932 * have already been checked by hardware, prior to the VM-exit for
7933 * VMXON. We do test guest cr4.VMXE because processor CR4 always has
7934 * that bit set to 1 in non-root mode.
7936 if (!kvm_read_cr4_bits(vcpu, X86_CR4_VMXE)) {
7937 kvm_queue_exception(vcpu, UD_VECTOR);
7941 /* CPL=0 must be checked manually. */
7942 if (vmx_get_cpl(vcpu)) {
7943 kvm_queue_exception(vcpu, UD_VECTOR);
7947 if (vmx->nested.vmxon) {
7948 nested_vmx_failValid(vcpu, VMXERR_VMXON_IN_VMX_ROOT_OPERATION);
7949 return kvm_skip_emulated_instruction(vcpu);
7952 if ((vmx->msr_ia32_feature_control & VMXON_NEEDED_FEATURES)
7953 != VMXON_NEEDED_FEATURES) {
7954 kvm_inject_gp(vcpu, 0);
7958 if (nested_vmx_get_vmptr(vcpu, &vmptr))
7963 * The first 4 bytes of VMXON region contain the supported
7964 * VMCS revision identifier
7966 * Note - IA32_VMX_BASIC[48] will never be 1 for the nested case;
7967 * which replaces physical address width with 32
7969 if (!PAGE_ALIGNED(vmptr) || (vmptr >> cpuid_maxphyaddr(vcpu))) {
7970 nested_vmx_failInvalid(vcpu);
7971 return kvm_skip_emulated_instruction(vcpu);
7974 page = kvm_vcpu_gpa_to_page(vcpu, vmptr);
7975 if (is_error_page(page)) {
7976 nested_vmx_failInvalid(vcpu);
7977 return kvm_skip_emulated_instruction(vcpu);
7979 if (*(u32 *)kmap(page) != VMCS12_REVISION) {
7981 kvm_release_page_clean(page);
7982 nested_vmx_failInvalid(vcpu);
7983 return kvm_skip_emulated_instruction(vcpu);
7986 kvm_release_page_clean(page);
7988 vmx->nested.vmxon_ptr = vmptr;
7989 ret = enter_vmx_operation(vcpu);
7993 nested_vmx_succeed(vcpu);
7994 return kvm_skip_emulated_instruction(vcpu);
7998 * Intel's VMX Instruction Reference specifies a common set of prerequisites
7999 * for running VMX instructions (except VMXON, whose prerequisites are
8000 * slightly different). It also specifies what exception to inject otherwise.
8001 * Note that many of these exceptions have priority over VM exits, so they
8002 * don't have to be checked again here.
8004 static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
8006 if (vmx_get_cpl(vcpu)) {
8007 kvm_queue_exception(vcpu, UD_VECTOR);
8011 if (!to_vmx(vcpu)->nested.vmxon) {
8012 kvm_queue_exception(vcpu, UD_VECTOR);
8018 static void vmx_disable_shadow_vmcs(struct vcpu_vmx *vmx)
8020 vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL, SECONDARY_EXEC_SHADOW_VMCS);
8021 vmcs_write64(VMCS_LINK_POINTER, -1ull);
8024 static inline void nested_release_vmcs12(struct vcpu_vmx *vmx)
8026 if (vmx->nested.current_vmptr == -1ull)
8029 if (enable_shadow_vmcs) {
8030 /* copy to memory all shadowed fields in case
8031 they were modified */
8032 copy_shadow_to_vmcs12(vmx);
8033 vmx->nested.sync_shadow_vmcs = false;
8034 vmx_disable_shadow_vmcs(vmx);
8036 vmx->nested.posted_intr_nv = -1;
8038 /* Flush VMCS12 to guest memory */
8039 kvm_vcpu_write_guest_page(&vmx->vcpu,
8040 vmx->nested.current_vmptr >> PAGE_SHIFT,
8041 vmx->nested.cached_vmcs12, 0, VMCS12_SIZE);
8043 vmx->nested.current_vmptr = -1ull;
8047 * Free whatever needs to be freed from vmx->nested when L1 goes down, or
8048 * just stops using VMX.
8050 static void free_nested(struct vcpu_vmx *vmx)
8052 if (!vmx->nested.vmxon && !vmx->nested.smm.vmxon)
8055 vmx->nested.vmxon = false;
8056 vmx->nested.smm.vmxon = false;
8057 free_vpid(vmx->nested.vpid02);
8058 vmx->nested.posted_intr_nv = -1;
8059 vmx->nested.current_vmptr = -1ull;
8060 if (enable_shadow_vmcs) {
8061 vmx_disable_shadow_vmcs(vmx);
8062 vmcs_clear(vmx->vmcs01.shadow_vmcs);
8063 free_vmcs(vmx->vmcs01.shadow_vmcs);
8064 vmx->vmcs01.shadow_vmcs = NULL;
8066 kfree(vmx->nested.cached_vmcs12);
8067 /* Unpin physical memory we referred to in the vmcs02 */
8068 if (vmx->nested.apic_access_page) {
8069 kvm_release_page_dirty(vmx->nested.apic_access_page);
8070 vmx->nested.apic_access_page = NULL;
8072 if (vmx->nested.virtual_apic_page) {
8073 kvm_release_page_dirty(vmx->nested.virtual_apic_page);
8074 vmx->nested.virtual_apic_page = NULL;
8076 if (vmx->nested.pi_desc_page) {
8077 kunmap(vmx->nested.pi_desc_page);
8078 kvm_release_page_dirty(vmx->nested.pi_desc_page);
8079 vmx->nested.pi_desc_page = NULL;
8080 vmx->nested.pi_desc = NULL;
8083 free_loaded_vmcs(&vmx->nested.vmcs02);
8086 /* Emulate the VMXOFF instruction */
8087 static int handle_vmoff(struct kvm_vcpu *vcpu)
8089 if (!nested_vmx_check_permission(vcpu))
8091 free_nested(to_vmx(vcpu));
8092 nested_vmx_succeed(vcpu);
8093 return kvm_skip_emulated_instruction(vcpu);
8096 /* Emulate the VMCLEAR instruction */
8097 static int handle_vmclear(struct kvm_vcpu *vcpu)
8099 struct vcpu_vmx *vmx = to_vmx(vcpu);
8103 if (!nested_vmx_check_permission(vcpu))
8106 if (nested_vmx_get_vmptr(vcpu, &vmptr))
8109 if (!PAGE_ALIGNED(vmptr) || (vmptr >> cpuid_maxphyaddr(vcpu))) {
8110 nested_vmx_failValid(vcpu, VMXERR_VMCLEAR_INVALID_ADDRESS);
8111 return kvm_skip_emulated_instruction(vcpu);
8114 if (vmptr == vmx->nested.vmxon_ptr) {
8115 nested_vmx_failValid(vcpu, VMXERR_VMCLEAR_VMXON_POINTER);
8116 return kvm_skip_emulated_instruction(vcpu);
8119 if (vmptr == vmx->nested.current_vmptr)
8120 nested_release_vmcs12(vmx);
8122 kvm_vcpu_write_guest(vcpu,
8123 vmptr + offsetof(struct vmcs12, launch_state),
8124 &zero, sizeof(zero));
8126 nested_vmx_succeed(vcpu);
8127 return kvm_skip_emulated_instruction(vcpu);
8130 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch);
8132 /* Emulate the VMLAUNCH instruction */
8133 static int handle_vmlaunch(struct kvm_vcpu *vcpu)
8135 return nested_vmx_run(vcpu, true);
8138 /* Emulate the VMRESUME instruction */
8139 static int handle_vmresume(struct kvm_vcpu *vcpu)
8142 return nested_vmx_run(vcpu, false);
8146 * Read a vmcs12 field. Since these can have varying lengths and we return
8147 * one type, we chose the biggest type (u64) and zero-extend the return value
8148 * to that size. Note that the caller, handle_vmread, might need to use only
8149 * some of the bits we return here (e.g., on 32-bit guests, only 32 bits of
8150 * 64-bit fields are to be returned).
8152 static inline int vmcs12_read_any(struct kvm_vcpu *vcpu,
8153 unsigned long field, u64 *ret)
8155 short offset = vmcs_field_to_offset(field);
8161 p = ((char *)(get_vmcs12(vcpu))) + offset;
8163 switch (vmcs_field_width(field)) {
8164 case VMCS_FIELD_WIDTH_NATURAL_WIDTH:
8165 *ret = *((natural_width *)p);
8167 case VMCS_FIELD_WIDTH_U16:
8170 case VMCS_FIELD_WIDTH_U32:
8173 case VMCS_FIELD_WIDTH_U64:
8183 static inline int vmcs12_write_any(struct kvm_vcpu *vcpu,
8184 unsigned long field, u64 field_value){
8185 short offset = vmcs_field_to_offset(field);
8186 char *p = ((char *) get_vmcs12(vcpu)) + offset;
8190 switch (vmcs_field_width(field)) {
8191 case VMCS_FIELD_WIDTH_U16:
8192 *(u16 *)p = field_value;
8194 case VMCS_FIELD_WIDTH_U32:
8195 *(u32 *)p = field_value;
8197 case VMCS_FIELD_WIDTH_U64:
8198 *(u64 *)p = field_value;
8200 case VMCS_FIELD_WIDTH_NATURAL_WIDTH:
8201 *(natural_width *)p = field_value;
8211 * Copy the writable VMCS shadow fields back to the VMCS12, in case
8212 * they have been modified by the L1 guest. Note that the "read-only"
8213 * VM-exit information fields are actually writable if the vCPU is
8214 * configured to support "VMWRITE to any supported field in the VMCS."
8216 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx)
8218 const u16 *fields[] = {
8219 shadow_read_write_fields,
8220 shadow_read_only_fields
8222 const int max_fields[] = {
8223 max_shadow_read_write_fields,
8224 max_shadow_read_only_fields
8227 unsigned long field;
8229 struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
8233 vmcs_load(shadow_vmcs);
8235 for (q = 0; q < ARRAY_SIZE(fields); q++) {
8236 for (i = 0; i < max_fields[q]; i++) {
8237 field = fields[q][i];
8238 field_value = __vmcs_readl(field);
8239 vmcs12_write_any(&vmx->vcpu, field, field_value);
8242 * Skip the VM-exit information fields if they are read-only.
8244 if (!nested_cpu_has_vmwrite_any_field(&vmx->vcpu))
8248 vmcs_clear(shadow_vmcs);
8249 vmcs_load(vmx->loaded_vmcs->vmcs);
8254 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx)
8256 const u16 *fields[] = {
8257 shadow_read_write_fields,
8258 shadow_read_only_fields
8260 const int max_fields[] = {
8261 max_shadow_read_write_fields,
8262 max_shadow_read_only_fields
8265 unsigned long field;
8266 u64 field_value = 0;
8267 struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
8269 vmcs_load(shadow_vmcs);
8271 for (q = 0; q < ARRAY_SIZE(fields); q++) {
8272 for (i = 0; i < max_fields[q]; i++) {
8273 field = fields[q][i];
8274 vmcs12_read_any(&vmx->vcpu, field, &field_value);
8275 __vmcs_writel(field, field_value);
8279 vmcs_clear(shadow_vmcs);
8280 vmcs_load(vmx->loaded_vmcs->vmcs);
8284 * VMX instructions which assume a current vmcs12 (i.e., that VMPTRLD was
8285 * used before) all generate the same failure when it is missing.
8287 static int nested_vmx_check_vmcs12(struct kvm_vcpu *vcpu)
8289 struct vcpu_vmx *vmx = to_vmx(vcpu);
8290 if (vmx->nested.current_vmptr == -1ull) {
8291 nested_vmx_failInvalid(vcpu);
8297 static int handle_vmread(struct kvm_vcpu *vcpu)
8299 unsigned long field;
8301 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
8302 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
8305 if (!nested_vmx_check_permission(vcpu))
8308 if (!nested_vmx_check_vmcs12(vcpu))
8309 return kvm_skip_emulated_instruction(vcpu);
8311 /* Decode instruction info and find the field to read */
8312 field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
8313 /* Read the field, zero-extended to a u64 field_value */
8314 if (vmcs12_read_any(vcpu, field, &field_value) < 0) {
8315 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
8316 return kvm_skip_emulated_instruction(vcpu);
8319 * Now copy part of this value to register or memory, as requested.
8320 * Note that the number of bits actually copied is 32 or 64 depending
8321 * on the guest's mode (32 or 64 bit), not on the given field's length.
8323 if (vmx_instruction_info & (1u << 10)) {
8324 kvm_register_writel(vcpu, (((vmx_instruction_info) >> 3) & 0xf),
8327 if (get_vmx_mem_address(vcpu, exit_qualification,
8328 vmx_instruction_info, true, &gva))
8330 /* _system ok, nested_vmx_check_permission has verified cpl=0 */
8331 kvm_write_guest_virt_system(vcpu, gva, &field_value,
8332 (is_long_mode(vcpu) ? 8 : 4), NULL);
8335 nested_vmx_succeed(vcpu);
8336 return kvm_skip_emulated_instruction(vcpu);
8340 static int handle_vmwrite(struct kvm_vcpu *vcpu)
8342 unsigned long field;
8344 struct vcpu_vmx *vmx = to_vmx(vcpu);
8345 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
8346 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
8348 /* The value to write might be 32 or 64 bits, depending on L1's long
8349 * mode, and eventually we need to write that into a field of several
8350 * possible lengths. The code below first zero-extends the value to 64
8351 * bit (field_value), and then copies only the appropriate number of
8352 * bits into the vmcs12 field.
8354 u64 field_value = 0;
8355 struct x86_exception e;
8357 if (!nested_vmx_check_permission(vcpu))
8360 if (!nested_vmx_check_vmcs12(vcpu))
8361 return kvm_skip_emulated_instruction(vcpu);
8363 if (vmx_instruction_info & (1u << 10))
8364 field_value = kvm_register_readl(vcpu,
8365 (((vmx_instruction_info) >> 3) & 0xf));
8367 if (get_vmx_mem_address(vcpu, exit_qualification,
8368 vmx_instruction_info, false, &gva))
8370 if (kvm_read_guest_virt(vcpu, gva, &field_value,
8371 (is_64_bit_mode(vcpu) ? 8 : 4), &e)) {
8372 kvm_inject_page_fault(vcpu, &e);
8378 field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
8380 * If the vCPU supports "VMWRITE to any supported field in the
8381 * VMCS," then the "read-only" fields are actually read/write.
8383 if (vmcs_field_readonly(field) &&
8384 !nested_cpu_has_vmwrite_any_field(vcpu)) {
8385 nested_vmx_failValid(vcpu,
8386 VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
8387 return kvm_skip_emulated_instruction(vcpu);
8390 if (vmcs12_write_any(vcpu, field, field_value) < 0) {
8391 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
8392 return kvm_skip_emulated_instruction(vcpu);
8396 #define SHADOW_FIELD_RW(x) case x:
8397 #include "vmx_shadow_fields.h"
8399 * The fields that can be updated by L1 without a vmexit are
8400 * always updated in the vmcs02, the others go down the slow
8401 * path of prepare_vmcs02.
8405 vmx->nested.dirty_vmcs12 = true;
8409 nested_vmx_succeed(vcpu);
8410 return kvm_skip_emulated_instruction(vcpu);
8413 static void set_current_vmptr(struct vcpu_vmx *vmx, gpa_t vmptr)
8415 vmx->nested.current_vmptr = vmptr;
8416 if (enable_shadow_vmcs) {
8417 vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
8418 SECONDARY_EXEC_SHADOW_VMCS);
8419 vmcs_write64(VMCS_LINK_POINTER,
8420 __pa(vmx->vmcs01.shadow_vmcs));
8421 vmx->nested.sync_shadow_vmcs = true;
8423 vmx->nested.dirty_vmcs12 = true;
8426 /* Emulate the VMPTRLD instruction */
8427 static int handle_vmptrld(struct kvm_vcpu *vcpu)
8429 struct vcpu_vmx *vmx = to_vmx(vcpu);
8432 if (!nested_vmx_check_permission(vcpu))
8435 if (nested_vmx_get_vmptr(vcpu, &vmptr))
8438 if (!PAGE_ALIGNED(vmptr) || (vmptr >> cpuid_maxphyaddr(vcpu))) {
8439 nested_vmx_failValid(vcpu, VMXERR_VMPTRLD_INVALID_ADDRESS);
8440 return kvm_skip_emulated_instruction(vcpu);
8443 if (vmptr == vmx->nested.vmxon_ptr) {
8444 nested_vmx_failValid(vcpu, VMXERR_VMPTRLD_VMXON_POINTER);
8445 return kvm_skip_emulated_instruction(vcpu);
8448 if (vmx->nested.current_vmptr != vmptr) {
8449 struct vmcs12 *new_vmcs12;
8451 page = kvm_vcpu_gpa_to_page(vcpu, vmptr);
8452 if (is_error_page(page)) {
8453 nested_vmx_failInvalid(vcpu);
8454 return kvm_skip_emulated_instruction(vcpu);
8456 new_vmcs12 = kmap(page);
8457 if (new_vmcs12->revision_id != VMCS12_REVISION) {
8459 kvm_release_page_clean(page);
8460 nested_vmx_failValid(vcpu,
8461 VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
8462 return kvm_skip_emulated_instruction(vcpu);
8465 nested_release_vmcs12(vmx);
8467 * Load VMCS12 from guest memory since it is not already
8470 memcpy(vmx->nested.cached_vmcs12, new_vmcs12, VMCS12_SIZE);
8472 kvm_release_page_clean(page);
8474 set_current_vmptr(vmx, vmptr);
8477 nested_vmx_succeed(vcpu);
8478 return kvm_skip_emulated_instruction(vcpu);
8481 /* Emulate the VMPTRST instruction */
8482 static int handle_vmptrst(struct kvm_vcpu *vcpu)
8484 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
8485 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
8487 struct x86_exception e;
8489 if (!nested_vmx_check_permission(vcpu))
8492 if (get_vmx_mem_address(vcpu, exit_qualification,
8493 vmx_instruction_info, true, &vmcs_gva))
8495 /* *_system ok, nested_vmx_check_permission has verified cpl=0 */
8496 if (kvm_write_guest_virt_system(vcpu, vmcs_gva,
8497 (void *)&to_vmx(vcpu)->nested.current_vmptr,
8499 kvm_inject_page_fault(vcpu, &e);
8502 nested_vmx_succeed(vcpu);
8503 return kvm_skip_emulated_instruction(vcpu);
8506 /* Emulate the INVEPT instruction */
8507 static int handle_invept(struct kvm_vcpu *vcpu)
8509 struct vcpu_vmx *vmx = to_vmx(vcpu);
8510 u32 vmx_instruction_info, types;
8513 struct x86_exception e;
8518 if (!(vmx->nested.msrs.secondary_ctls_high &
8519 SECONDARY_EXEC_ENABLE_EPT) ||
8520 !(vmx->nested.msrs.ept_caps & VMX_EPT_INVEPT_BIT)) {
8521 kvm_queue_exception(vcpu, UD_VECTOR);
8525 if (!nested_vmx_check_permission(vcpu))
8528 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
8529 type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
8531 types = (vmx->nested.msrs.ept_caps >> VMX_EPT_EXTENT_SHIFT) & 6;
8533 if (type >= 32 || !(types & (1 << type))) {
8534 nested_vmx_failValid(vcpu,
8535 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
8536 return kvm_skip_emulated_instruction(vcpu);
8539 /* According to the Intel VMX instruction reference, the memory
8540 * operand is read even if it isn't needed (e.g., for type==global)
8542 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
8543 vmx_instruction_info, false, &gva))
8545 if (kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e)) {
8546 kvm_inject_page_fault(vcpu, &e);
8551 case VMX_EPT_EXTENT_GLOBAL:
8553 * TODO: track mappings and invalidate
8554 * single context requests appropriately
8556 case VMX_EPT_EXTENT_CONTEXT:
8557 kvm_mmu_sync_roots(vcpu);
8558 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
8559 nested_vmx_succeed(vcpu);
8566 return kvm_skip_emulated_instruction(vcpu);
8569 static int handle_invvpid(struct kvm_vcpu *vcpu)
8571 struct vcpu_vmx *vmx = to_vmx(vcpu);
8572 u32 vmx_instruction_info;
8573 unsigned long type, types;
8575 struct x86_exception e;
8581 if (!(vmx->nested.msrs.secondary_ctls_high &
8582 SECONDARY_EXEC_ENABLE_VPID) ||
8583 !(vmx->nested.msrs.vpid_caps & VMX_VPID_INVVPID_BIT)) {
8584 kvm_queue_exception(vcpu, UD_VECTOR);
8588 if (!nested_vmx_check_permission(vcpu))
8591 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
8592 type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
8594 types = (vmx->nested.msrs.vpid_caps &
8595 VMX_VPID_EXTENT_SUPPORTED_MASK) >> 8;
8597 if (type >= 32 || !(types & (1 << type))) {
8598 nested_vmx_failValid(vcpu,
8599 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
8600 return kvm_skip_emulated_instruction(vcpu);
8603 /* according to the intel vmx instruction reference, the memory
8604 * operand is read even if it isn't needed (e.g., for type==global)
8606 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
8607 vmx_instruction_info, false, &gva))
8609 if (kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e)) {
8610 kvm_inject_page_fault(vcpu, &e);
8613 if (operand.vpid >> 16) {
8614 nested_vmx_failValid(vcpu,
8615 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
8616 return kvm_skip_emulated_instruction(vcpu);
8620 case VMX_VPID_EXTENT_INDIVIDUAL_ADDR:
8621 if (!operand.vpid ||
8622 is_noncanonical_address(operand.gla, vcpu)) {
8623 nested_vmx_failValid(vcpu,
8624 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
8625 return kvm_skip_emulated_instruction(vcpu);
8627 if (cpu_has_vmx_invvpid_individual_addr() &&
8628 vmx->nested.vpid02) {
8629 __invvpid(VMX_VPID_EXTENT_INDIVIDUAL_ADDR,
8630 vmx->nested.vpid02, operand.gla);
8632 __vmx_flush_tlb(vcpu, vmx->nested.vpid02, true);
8634 case VMX_VPID_EXTENT_SINGLE_CONTEXT:
8635 case VMX_VPID_EXTENT_SINGLE_NON_GLOBAL:
8636 if (!operand.vpid) {
8637 nested_vmx_failValid(vcpu,
8638 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
8639 return kvm_skip_emulated_instruction(vcpu);
8641 __vmx_flush_tlb(vcpu, vmx->nested.vpid02, true);
8643 case VMX_VPID_EXTENT_ALL_CONTEXT:
8644 __vmx_flush_tlb(vcpu, vmx->nested.vpid02, true);
8648 return kvm_skip_emulated_instruction(vcpu);
8651 nested_vmx_succeed(vcpu);
8653 return kvm_skip_emulated_instruction(vcpu);
8656 static int handle_pml_full(struct kvm_vcpu *vcpu)
8658 unsigned long exit_qualification;
8660 trace_kvm_pml_full(vcpu->vcpu_id);
8662 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
8665 * PML buffer FULL happened while executing iret from NMI,
8666 * "blocked by NMI" bit has to be set before next VM entry.
8668 if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
8670 (exit_qualification & INTR_INFO_UNBLOCK_NMI))
8671 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
8672 GUEST_INTR_STATE_NMI);
8675 * PML buffer already flushed at beginning of VMEXIT. Nothing to do
8676 * here.., and there's no userspace involvement needed for PML.
8681 static int handle_preemption_timer(struct kvm_vcpu *vcpu)
8683 kvm_lapic_expired_hv_timer(vcpu);
8687 static bool valid_ept_address(struct kvm_vcpu *vcpu, u64 address)
8689 struct vcpu_vmx *vmx = to_vmx(vcpu);
8690 int maxphyaddr = cpuid_maxphyaddr(vcpu);
8692 /* Check for memory type validity */
8693 switch (address & VMX_EPTP_MT_MASK) {
8694 case VMX_EPTP_MT_UC:
8695 if (!(vmx->nested.msrs.ept_caps & VMX_EPTP_UC_BIT))
8698 case VMX_EPTP_MT_WB:
8699 if (!(vmx->nested.msrs.ept_caps & VMX_EPTP_WB_BIT))
8706 /* only 4 levels page-walk length are valid */
8707 if ((address & VMX_EPTP_PWL_MASK) != VMX_EPTP_PWL_4)
8710 /* Reserved bits should not be set */
8711 if (address >> maxphyaddr || ((address >> 7) & 0x1f))
8714 /* AD, if set, should be supported */
8715 if (address & VMX_EPTP_AD_ENABLE_BIT) {
8716 if (!(vmx->nested.msrs.ept_caps & VMX_EPT_AD_BIT))
8723 static int nested_vmx_eptp_switching(struct kvm_vcpu *vcpu,
8724 struct vmcs12 *vmcs12)
8726 u32 index = vcpu->arch.regs[VCPU_REGS_RCX];
8728 bool accessed_dirty;
8729 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
8731 if (!nested_cpu_has_eptp_switching(vmcs12) ||
8732 !nested_cpu_has_ept(vmcs12))
8735 if (index >= VMFUNC_EPTP_ENTRIES)
8739 if (kvm_vcpu_read_guest_page(vcpu, vmcs12->eptp_list_address >> PAGE_SHIFT,
8740 &address, index * 8, 8))
8743 accessed_dirty = !!(address & VMX_EPTP_AD_ENABLE_BIT);
8746 * If the (L2) guest does a vmfunc to the currently
8747 * active ept pointer, we don't have to do anything else
8749 if (vmcs12->ept_pointer != address) {
8750 if (!valid_ept_address(vcpu, address))
8753 kvm_mmu_unload(vcpu);
8754 mmu->ept_ad = accessed_dirty;
8755 mmu->base_role.ad_disabled = !accessed_dirty;
8756 vmcs12->ept_pointer = address;
8758 * TODO: Check what's the correct approach in case
8759 * mmu reload fails. Currently, we just let the next
8760 * reload potentially fail
8762 kvm_mmu_reload(vcpu);
8768 static int handle_vmfunc(struct kvm_vcpu *vcpu)
8770 struct vcpu_vmx *vmx = to_vmx(vcpu);
8771 struct vmcs12 *vmcs12;
8772 u32 function = vcpu->arch.regs[VCPU_REGS_RAX];
8775 * VMFUNC is only supported for nested guests, but we always enable the
8776 * secondary control for simplicity; for non-nested mode, fake that we
8777 * didn't by injecting #UD.
8779 if (!is_guest_mode(vcpu)) {
8780 kvm_queue_exception(vcpu, UD_VECTOR);
8784 vmcs12 = get_vmcs12(vcpu);
8785 if ((vmcs12->vm_function_control & (1 << function)) == 0)
8790 if (nested_vmx_eptp_switching(vcpu, vmcs12))
8796 return kvm_skip_emulated_instruction(vcpu);
8799 nested_vmx_vmexit(vcpu, vmx->exit_reason,
8800 vmcs_read32(VM_EXIT_INTR_INFO),
8801 vmcs_readl(EXIT_QUALIFICATION));
8806 * The exit handlers return 1 if the exit was handled fully and guest execution
8807 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
8808 * to be done to userspace and return 0.
8810 static int (*const kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
8811 [EXIT_REASON_EXCEPTION_NMI] = handle_exception,
8812 [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt,
8813 [EXIT_REASON_TRIPLE_FAULT] = handle_triple_fault,
8814 [EXIT_REASON_NMI_WINDOW] = handle_nmi_window,
8815 [EXIT_REASON_IO_INSTRUCTION] = handle_io,
8816 [EXIT_REASON_CR_ACCESS] = handle_cr,
8817 [EXIT_REASON_DR_ACCESS] = handle_dr,
8818 [EXIT_REASON_CPUID] = handle_cpuid,
8819 [EXIT_REASON_MSR_READ] = handle_rdmsr,
8820 [EXIT_REASON_MSR_WRITE] = handle_wrmsr,
8821 [EXIT_REASON_PENDING_INTERRUPT] = handle_interrupt_window,
8822 [EXIT_REASON_HLT] = handle_halt,
8823 [EXIT_REASON_INVD] = handle_invd,
8824 [EXIT_REASON_INVLPG] = handle_invlpg,
8825 [EXIT_REASON_RDPMC] = handle_rdpmc,
8826 [EXIT_REASON_VMCALL] = handle_vmcall,
8827 [EXIT_REASON_VMCLEAR] = handle_vmclear,
8828 [EXIT_REASON_VMLAUNCH] = handle_vmlaunch,
8829 [EXIT_REASON_VMPTRLD] = handle_vmptrld,
8830 [EXIT_REASON_VMPTRST] = handle_vmptrst,
8831 [EXIT_REASON_VMREAD] = handle_vmread,
8832 [EXIT_REASON_VMRESUME] = handle_vmresume,
8833 [EXIT_REASON_VMWRITE] = handle_vmwrite,
8834 [EXIT_REASON_VMOFF] = handle_vmoff,
8835 [EXIT_REASON_VMON] = handle_vmon,
8836 [EXIT_REASON_TPR_BELOW_THRESHOLD] = handle_tpr_below_threshold,
8837 [EXIT_REASON_APIC_ACCESS] = handle_apic_access,
8838 [EXIT_REASON_APIC_WRITE] = handle_apic_write,
8839 [EXIT_REASON_EOI_INDUCED] = handle_apic_eoi_induced,
8840 [EXIT_REASON_WBINVD] = handle_wbinvd,
8841 [EXIT_REASON_XSETBV] = handle_xsetbv,
8842 [EXIT_REASON_TASK_SWITCH] = handle_task_switch,
8843 [EXIT_REASON_MCE_DURING_VMENTRY] = handle_machine_check,
8844 [EXIT_REASON_GDTR_IDTR] = handle_desc,
8845 [EXIT_REASON_LDTR_TR] = handle_desc,
8846 [EXIT_REASON_EPT_VIOLATION] = handle_ept_violation,
8847 [EXIT_REASON_EPT_MISCONFIG] = handle_ept_misconfig,
8848 [EXIT_REASON_PAUSE_INSTRUCTION] = handle_pause,
8849 [EXIT_REASON_MWAIT_INSTRUCTION] = handle_mwait,
8850 [EXIT_REASON_MONITOR_TRAP_FLAG] = handle_monitor_trap,
8851 [EXIT_REASON_MONITOR_INSTRUCTION] = handle_monitor,
8852 [EXIT_REASON_INVEPT] = handle_invept,
8853 [EXIT_REASON_INVVPID] = handle_invvpid,
8854 [EXIT_REASON_RDRAND] = handle_invalid_op,
8855 [EXIT_REASON_RDSEED] = handle_invalid_op,
8856 [EXIT_REASON_XSAVES] = handle_xsaves,
8857 [EXIT_REASON_XRSTORS] = handle_xrstors,
8858 [EXIT_REASON_PML_FULL] = handle_pml_full,
8859 [EXIT_REASON_VMFUNC] = handle_vmfunc,
8860 [EXIT_REASON_PREEMPTION_TIMER] = handle_preemption_timer,
8863 static const int kvm_vmx_max_exit_handlers =
8864 ARRAY_SIZE(kvm_vmx_exit_handlers);
8866 static bool nested_vmx_exit_handled_io(struct kvm_vcpu *vcpu,
8867 struct vmcs12 *vmcs12)
8869 unsigned long exit_qualification;
8870 gpa_t bitmap, last_bitmap;
8875 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
8876 return nested_cpu_has(vmcs12, CPU_BASED_UNCOND_IO_EXITING);
8878 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
8880 port = exit_qualification >> 16;
8881 size = (exit_qualification & 7) + 1;
8883 last_bitmap = (gpa_t)-1;
8888 bitmap = vmcs12->io_bitmap_a;
8889 else if (port < 0x10000)
8890 bitmap = vmcs12->io_bitmap_b;
8893 bitmap += (port & 0x7fff) / 8;
8895 if (last_bitmap != bitmap)
8896 if (kvm_vcpu_read_guest(vcpu, bitmap, &b, 1))
8898 if (b & (1 << (port & 7)))
8903 last_bitmap = bitmap;
8910 * Return 1 if we should exit from L2 to L1 to handle an MSR access access,
8911 * rather than handle it ourselves in L0. I.e., check whether L1 expressed
8912 * disinterest in the current event (read or write a specific MSR) by using an
8913 * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps.
8915 static bool nested_vmx_exit_handled_msr(struct kvm_vcpu *vcpu,
8916 struct vmcs12 *vmcs12, u32 exit_reason)
8918 u32 msr_index = vcpu->arch.regs[VCPU_REGS_RCX];
8921 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
8925 * The MSR_BITMAP page is divided into four 1024-byte bitmaps,
8926 * for the four combinations of read/write and low/high MSR numbers.
8927 * First we need to figure out which of the four to use:
8929 bitmap = vmcs12->msr_bitmap;
8930 if (exit_reason == EXIT_REASON_MSR_WRITE)
8932 if (msr_index >= 0xc0000000) {
8933 msr_index -= 0xc0000000;
8937 /* Then read the msr_index'th bit from this bitmap: */
8938 if (msr_index < 1024*8) {
8940 if (kvm_vcpu_read_guest(vcpu, bitmap + msr_index/8, &b, 1))
8942 return 1 & (b >> (msr_index & 7));
8944 return true; /* let L1 handle the wrong parameter */
8948 * Return 1 if we should exit from L2 to L1 to handle a CR access exit,
8949 * rather than handle it ourselves in L0. I.e., check if L1 wanted to
8950 * intercept (via guest_host_mask etc.) the current event.
8952 static bool nested_vmx_exit_handled_cr(struct kvm_vcpu *vcpu,
8953 struct vmcs12 *vmcs12)
8955 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
8956 int cr = exit_qualification & 15;
8960 switch ((exit_qualification >> 4) & 3) {
8961 case 0: /* mov to cr */
8962 reg = (exit_qualification >> 8) & 15;
8963 val = kvm_register_readl(vcpu, reg);
8966 if (vmcs12->cr0_guest_host_mask &
8967 (val ^ vmcs12->cr0_read_shadow))
8971 if ((vmcs12->cr3_target_count >= 1 &&
8972 vmcs12->cr3_target_value0 == val) ||
8973 (vmcs12->cr3_target_count >= 2 &&
8974 vmcs12->cr3_target_value1 == val) ||
8975 (vmcs12->cr3_target_count >= 3 &&
8976 vmcs12->cr3_target_value2 == val) ||
8977 (vmcs12->cr3_target_count >= 4 &&
8978 vmcs12->cr3_target_value3 == val))
8980 if (nested_cpu_has(vmcs12, CPU_BASED_CR3_LOAD_EXITING))
8984 if (vmcs12->cr4_guest_host_mask &
8985 (vmcs12->cr4_read_shadow ^ val))
8989 if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING))
8995 if ((vmcs12->cr0_guest_host_mask & X86_CR0_TS) &&
8996 (vmcs12->cr0_read_shadow & X86_CR0_TS))
8999 case 1: /* mov from cr */
9002 if (vmcs12->cpu_based_vm_exec_control &
9003 CPU_BASED_CR3_STORE_EXITING)
9007 if (vmcs12->cpu_based_vm_exec_control &
9008 CPU_BASED_CR8_STORE_EXITING)
9015 * lmsw can change bits 1..3 of cr0, and only set bit 0 of
9016 * cr0. Other attempted changes are ignored, with no exit.
9018 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
9019 if (vmcs12->cr0_guest_host_mask & 0xe &
9020 (val ^ vmcs12->cr0_read_shadow))
9022 if ((vmcs12->cr0_guest_host_mask & 0x1) &&
9023 !(vmcs12->cr0_read_shadow & 0x1) &&
9032 * Return 1 if we should exit from L2 to L1 to handle an exit, or 0 if we
9033 * should handle it ourselves in L0 (and then continue L2). Only call this
9034 * when in is_guest_mode (L2).
9036 static bool nested_vmx_exit_reflected(struct kvm_vcpu *vcpu, u32 exit_reason)
9038 u32 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
9039 struct vcpu_vmx *vmx = to_vmx(vcpu);
9040 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
9042 if (vmx->nested.nested_run_pending)
9045 if (unlikely(vmx->fail)) {
9046 pr_info_ratelimited("%s failed vm entry %x\n", __func__,
9047 vmcs_read32(VM_INSTRUCTION_ERROR));
9052 * The host physical addresses of some pages of guest memory
9053 * are loaded into the vmcs02 (e.g. vmcs12's Virtual APIC
9054 * Page). The CPU may write to these pages via their host
9055 * physical address while L2 is running, bypassing any
9056 * address-translation-based dirty tracking (e.g. EPT write
9059 * Mark them dirty on every exit from L2 to prevent them from
9060 * getting out of sync with dirty tracking.
9062 nested_mark_vmcs12_pages_dirty(vcpu);
9064 trace_kvm_nested_vmexit(kvm_rip_read(vcpu), exit_reason,
9065 vmcs_readl(EXIT_QUALIFICATION),
9066 vmx->idt_vectoring_info,
9068 vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
9071 switch (exit_reason) {
9072 case EXIT_REASON_EXCEPTION_NMI:
9073 if (is_nmi(intr_info))
9075 else if (is_page_fault(intr_info))
9076 return !vmx->vcpu.arch.apf.host_apf_reason && enable_ept;
9077 else if (is_no_device(intr_info) &&
9078 !(vmcs12->guest_cr0 & X86_CR0_TS))
9080 else if (is_debug(intr_info) &&
9082 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
9084 else if (is_breakpoint(intr_info) &&
9085 vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
9087 return vmcs12->exception_bitmap &
9088 (1u << (intr_info & INTR_INFO_VECTOR_MASK));
9089 case EXIT_REASON_EXTERNAL_INTERRUPT:
9091 case EXIT_REASON_TRIPLE_FAULT:
9093 case EXIT_REASON_PENDING_INTERRUPT:
9094 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_INTR_PENDING);
9095 case EXIT_REASON_NMI_WINDOW:
9096 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_NMI_PENDING);
9097 case EXIT_REASON_TASK_SWITCH:
9099 case EXIT_REASON_CPUID:
9101 case EXIT_REASON_HLT:
9102 return nested_cpu_has(vmcs12, CPU_BASED_HLT_EXITING);
9103 case EXIT_REASON_INVD:
9105 case EXIT_REASON_INVLPG:
9106 return nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
9107 case EXIT_REASON_RDPMC:
9108 return nested_cpu_has(vmcs12, CPU_BASED_RDPMC_EXITING);
9109 case EXIT_REASON_RDRAND:
9110 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDRAND_EXITING);
9111 case EXIT_REASON_RDSEED:
9112 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDSEED_EXITING);
9113 case EXIT_REASON_RDTSC: case EXIT_REASON_RDTSCP:
9114 return nested_cpu_has(vmcs12, CPU_BASED_RDTSC_EXITING);
9115 case EXIT_REASON_VMCALL: case EXIT_REASON_VMCLEAR:
9116 case EXIT_REASON_VMLAUNCH: case EXIT_REASON_VMPTRLD:
9117 case EXIT_REASON_VMPTRST: case EXIT_REASON_VMREAD:
9118 case EXIT_REASON_VMRESUME: case EXIT_REASON_VMWRITE:
9119 case EXIT_REASON_VMOFF: case EXIT_REASON_VMON:
9120 case EXIT_REASON_INVEPT: case EXIT_REASON_INVVPID:
9122 * VMX instructions trap unconditionally. This allows L1 to
9123 * emulate them for its L2 guest, i.e., allows 3-level nesting!
9126 case EXIT_REASON_CR_ACCESS:
9127 return nested_vmx_exit_handled_cr(vcpu, vmcs12);
9128 case EXIT_REASON_DR_ACCESS:
9129 return nested_cpu_has(vmcs12, CPU_BASED_MOV_DR_EXITING);
9130 case EXIT_REASON_IO_INSTRUCTION:
9131 return nested_vmx_exit_handled_io(vcpu, vmcs12);
9132 case EXIT_REASON_GDTR_IDTR: case EXIT_REASON_LDTR_TR:
9133 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_DESC);
9134 case EXIT_REASON_MSR_READ:
9135 case EXIT_REASON_MSR_WRITE:
9136 return nested_vmx_exit_handled_msr(vcpu, vmcs12, exit_reason);
9137 case EXIT_REASON_INVALID_STATE:
9139 case EXIT_REASON_MWAIT_INSTRUCTION:
9140 return nested_cpu_has(vmcs12, CPU_BASED_MWAIT_EXITING);
9141 case EXIT_REASON_MONITOR_TRAP_FLAG:
9142 return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_TRAP_FLAG);
9143 case EXIT_REASON_MONITOR_INSTRUCTION:
9144 return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_EXITING);
9145 case EXIT_REASON_PAUSE_INSTRUCTION:
9146 return nested_cpu_has(vmcs12, CPU_BASED_PAUSE_EXITING) ||
9147 nested_cpu_has2(vmcs12,
9148 SECONDARY_EXEC_PAUSE_LOOP_EXITING);
9149 case EXIT_REASON_MCE_DURING_VMENTRY:
9151 case EXIT_REASON_TPR_BELOW_THRESHOLD:
9152 return nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW);
9153 case EXIT_REASON_APIC_ACCESS:
9154 case EXIT_REASON_APIC_WRITE:
9155 case EXIT_REASON_EOI_INDUCED:
9157 * The controls for "virtualize APIC accesses," "APIC-
9158 * register virtualization," and "virtual-interrupt
9159 * delivery" only come from vmcs12.
9162 case EXIT_REASON_EPT_VIOLATION:
9164 * L0 always deals with the EPT violation. If nested EPT is
9165 * used, and the nested mmu code discovers that the address is
9166 * missing in the guest EPT table (EPT12), the EPT violation
9167 * will be injected with nested_ept_inject_page_fault()
9170 case EXIT_REASON_EPT_MISCONFIG:
9172 * L2 never uses directly L1's EPT, but rather L0's own EPT
9173 * table (shadow on EPT) or a merged EPT table that L0 built
9174 * (EPT on EPT). So any problems with the structure of the
9175 * table is L0's fault.
9178 case EXIT_REASON_INVPCID:
9180 nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_INVPCID) &&
9181 nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
9182 case EXIT_REASON_WBINVD:
9183 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_WBINVD_EXITING);
9184 case EXIT_REASON_XSETBV:
9186 case EXIT_REASON_XSAVES: case EXIT_REASON_XRSTORS:
9188 * This should never happen, since it is not possible to
9189 * set XSS to a non-zero value---neither in L1 nor in L2.
9190 * If if it were, XSS would have to be checked against
9191 * the XSS exit bitmap in vmcs12.
9193 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES);
9194 case EXIT_REASON_PREEMPTION_TIMER:
9196 case EXIT_REASON_PML_FULL:
9197 /* We emulate PML support to L1. */
9199 case EXIT_REASON_VMFUNC:
9200 /* VM functions are emulated through L2->L0 vmexits. */
9207 static int nested_vmx_reflect_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason)
9209 u32 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
9212 * At this point, the exit interruption info in exit_intr_info
9213 * is only valid for EXCEPTION_NMI exits. For EXTERNAL_INTERRUPT
9214 * we need to query the in-kernel LAPIC.
9216 WARN_ON(exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT);
9217 if ((exit_intr_info &
9218 (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK)) ==
9219 (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK)) {
9220 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
9221 vmcs12->vm_exit_intr_error_code =
9222 vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
9225 nested_vmx_vmexit(vcpu, exit_reason, exit_intr_info,
9226 vmcs_readl(EXIT_QUALIFICATION));
9230 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
9232 *info1 = vmcs_readl(EXIT_QUALIFICATION);
9233 *info2 = vmcs_read32(VM_EXIT_INTR_INFO);
9236 static void vmx_destroy_pml_buffer(struct vcpu_vmx *vmx)
9239 __free_page(vmx->pml_pg);
9244 static void vmx_flush_pml_buffer(struct kvm_vcpu *vcpu)
9246 struct vcpu_vmx *vmx = to_vmx(vcpu);
9250 pml_idx = vmcs_read16(GUEST_PML_INDEX);
9252 /* Do nothing if PML buffer is empty */
9253 if (pml_idx == (PML_ENTITY_NUM - 1))
9256 /* PML index always points to next available PML buffer entity */
9257 if (pml_idx >= PML_ENTITY_NUM)
9262 pml_buf = page_address(vmx->pml_pg);
9263 for (; pml_idx < PML_ENTITY_NUM; pml_idx++) {
9266 gpa = pml_buf[pml_idx];
9267 WARN_ON(gpa & (PAGE_SIZE - 1));
9268 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
9271 /* reset PML index */
9272 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
9276 * Flush all vcpus' PML buffer and update logged GPAs to dirty_bitmap.
9277 * Called before reporting dirty_bitmap to userspace.
9279 static void kvm_flush_pml_buffers(struct kvm *kvm)
9282 struct kvm_vcpu *vcpu;
9284 * We only need to kick vcpu out of guest mode here, as PML buffer
9285 * is flushed at beginning of all VMEXITs, and it's obvious that only
9286 * vcpus running in guest are possible to have unflushed GPAs in PML
9289 kvm_for_each_vcpu(i, vcpu, kvm)
9290 kvm_vcpu_kick(vcpu);
9293 static void vmx_dump_sel(char *name, uint32_t sel)
9295 pr_err("%s sel=0x%04x, attr=0x%05x, limit=0x%08x, base=0x%016lx\n",
9296 name, vmcs_read16(sel),
9297 vmcs_read32(sel + GUEST_ES_AR_BYTES - GUEST_ES_SELECTOR),
9298 vmcs_read32(sel + GUEST_ES_LIMIT - GUEST_ES_SELECTOR),
9299 vmcs_readl(sel + GUEST_ES_BASE - GUEST_ES_SELECTOR));
9302 static void vmx_dump_dtsel(char *name, uint32_t limit)
9304 pr_err("%s limit=0x%08x, base=0x%016lx\n",
9305 name, vmcs_read32(limit),
9306 vmcs_readl(limit + GUEST_GDTR_BASE - GUEST_GDTR_LIMIT));
9309 static void dump_vmcs(void)
9311 u32 vmentry_ctl = vmcs_read32(VM_ENTRY_CONTROLS);
9312 u32 vmexit_ctl = vmcs_read32(VM_EXIT_CONTROLS);
9313 u32 cpu_based_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
9314 u32 pin_based_exec_ctrl = vmcs_read32(PIN_BASED_VM_EXEC_CONTROL);
9315 u32 secondary_exec_control = 0;
9316 unsigned long cr4 = vmcs_readl(GUEST_CR4);
9317 u64 efer = vmcs_read64(GUEST_IA32_EFER);
9320 if (cpu_has_secondary_exec_ctrls())
9321 secondary_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
9323 pr_err("*** Guest State ***\n");
9324 pr_err("CR0: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
9325 vmcs_readl(GUEST_CR0), vmcs_readl(CR0_READ_SHADOW),
9326 vmcs_readl(CR0_GUEST_HOST_MASK));
9327 pr_err("CR4: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
9328 cr4, vmcs_readl(CR4_READ_SHADOW), vmcs_readl(CR4_GUEST_HOST_MASK));
9329 pr_err("CR3 = 0x%016lx\n", vmcs_readl(GUEST_CR3));
9330 if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT) &&
9331 (cr4 & X86_CR4_PAE) && !(efer & EFER_LMA))
9333 pr_err("PDPTR0 = 0x%016llx PDPTR1 = 0x%016llx\n",
9334 vmcs_read64(GUEST_PDPTR0), vmcs_read64(GUEST_PDPTR1));
9335 pr_err("PDPTR2 = 0x%016llx PDPTR3 = 0x%016llx\n",
9336 vmcs_read64(GUEST_PDPTR2), vmcs_read64(GUEST_PDPTR3));
9338 pr_err("RSP = 0x%016lx RIP = 0x%016lx\n",
9339 vmcs_readl(GUEST_RSP), vmcs_readl(GUEST_RIP));
9340 pr_err("RFLAGS=0x%08lx DR7 = 0x%016lx\n",
9341 vmcs_readl(GUEST_RFLAGS), vmcs_readl(GUEST_DR7));
9342 pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
9343 vmcs_readl(GUEST_SYSENTER_ESP),
9344 vmcs_read32(GUEST_SYSENTER_CS), vmcs_readl(GUEST_SYSENTER_EIP));
9345 vmx_dump_sel("CS: ", GUEST_CS_SELECTOR);
9346 vmx_dump_sel("DS: ", GUEST_DS_SELECTOR);
9347 vmx_dump_sel("SS: ", GUEST_SS_SELECTOR);
9348 vmx_dump_sel("ES: ", GUEST_ES_SELECTOR);
9349 vmx_dump_sel("FS: ", GUEST_FS_SELECTOR);
9350 vmx_dump_sel("GS: ", GUEST_GS_SELECTOR);
9351 vmx_dump_dtsel("GDTR:", GUEST_GDTR_LIMIT);
9352 vmx_dump_sel("LDTR:", GUEST_LDTR_SELECTOR);
9353 vmx_dump_dtsel("IDTR:", GUEST_IDTR_LIMIT);
9354 vmx_dump_sel("TR: ", GUEST_TR_SELECTOR);
9355 if ((vmexit_ctl & (VM_EXIT_SAVE_IA32_PAT | VM_EXIT_SAVE_IA32_EFER)) ||
9356 (vmentry_ctl & (VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_IA32_EFER)))
9357 pr_err("EFER = 0x%016llx PAT = 0x%016llx\n",
9358 efer, vmcs_read64(GUEST_IA32_PAT));
9359 pr_err("DebugCtl = 0x%016llx DebugExceptions = 0x%016lx\n",
9360 vmcs_read64(GUEST_IA32_DEBUGCTL),
9361 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS));
9362 if (cpu_has_load_perf_global_ctrl &&
9363 vmentry_ctl & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
9364 pr_err("PerfGlobCtl = 0x%016llx\n",
9365 vmcs_read64(GUEST_IA32_PERF_GLOBAL_CTRL));
9366 if (vmentry_ctl & VM_ENTRY_LOAD_BNDCFGS)
9367 pr_err("BndCfgS = 0x%016llx\n", vmcs_read64(GUEST_BNDCFGS));
9368 pr_err("Interruptibility = %08x ActivityState = %08x\n",
9369 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO),
9370 vmcs_read32(GUEST_ACTIVITY_STATE));
9371 if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
9372 pr_err("InterruptStatus = %04x\n",
9373 vmcs_read16(GUEST_INTR_STATUS));
9375 pr_err("*** Host State ***\n");
9376 pr_err("RIP = 0x%016lx RSP = 0x%016lx\n",
9377 vmcs_readl(HOST_RIP), vmcs_readl(HOST_RSP));
9378 pr_err("CS=%04x SS=%04x DS=%04x ES=%04x FS=%04x GS=%04x TR=%04x\n",
9379 vmcs_read16(HOST_CS_SELECTOR), vmcs_read16(HOST_SS_SELECTOR),
9380 vmcs_read16(HOST_DS_SELECTOR), vmcs_read16(HOST_ES_SELECTOR),
9381 vmcs_read16(HOST_FS_SELECTOR), vmcs_read16(HOST_GS_SELECTOR),
9382 vmcs_read16(HOST_TR_SELECTOR));
9383 pr_err("FSBase=%016lx GSBase=%016lx TRBase=%016lx\n",
9384 vmcs_readl(HOST_FS_BASE), vmcs_readl(HOST_GS_BASE),
9385 vmcs_readl(HOST_TR_BASE));
9386 pr_err("GDTBase=%016lx IDTBase=%016lx\n",
9387 vmcs_readl(HOST_GDTR_BASE), vmcs_readl(HOST_IDTR_BASE));
9388 pr_err("CR0=%016lx CR3=%016lx CR4=%016lx\n",
9389 vmcs_readl(HOST_CR0), vmcs_readl(HOST_CR3),
9390 vmcs_readl(HOST_CR4));
9391 pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
9392 vmcs_readl(HOST_IA32_SYSENTER_ESP),
9393 vmcs_read32(HOST_IA32_SYSENTER_CS),
9394 vmcs_readl(HOST_IA32_SYSENTER_EIP));
9395 if (vmexit_ctl & (VM_EXIT_LOAD_IA32_PAT | VM_EXIT_LOAD_IA32_EFER))
9396 pr_err("EFER = 0x%016llx PAT = 0x%016llx\n",
9397 vmcs_read64(HOST_IA32_EFER),
9398 vmcs_read64(HOST_IA32_PAT));
9399 if (cpu_has_load_perf_global_ctrl &&
9400 vmexit_ctl & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
9401 pr_err("PerfGlobCtl = 0x%016llx\n",
9402 vmcs_read64(HOST_IA32_PERF_GLOBAL_CTRL));
9404 pr_err("*** Control State ***\n");
9405 pr_err("PinBased=%08x CPUBased=%08x SecondaryExec=%08x\n",
9406 pin_based_exec_ctrl, cpu_based_exec_ctrl, secondary_exec_control);
9407 pr_err("EntryControls=%08x ExitControls=%08x\n", vmentry_ctl, vmexit_ctl);
9408 pr_err("ExceptionBitmap=%08x PFECmask=%08x PFECmatch=%08x\n",
9409 vmcs_read32(EXCEPTION_BITMAP),
9410 vmcs_read32(PAGE_FAULT_ERROR_CODE_MASK),
9411 vmcs_read32(PAGE_FAULT_ERROR_CODE_MATCH));
9412 pr_err("VMEntry: intr_info=%08x errcode=%08x ilen=%08x\n",
9413 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
9414 vmcs_read32(VM_ENTRY_EXCEPTION_ERROR_CODE),
9415 vmcs_read32(VM_ENTRY_INSTRUCTION_LEN));
9416 pr_err("VMExit: intr_info=%08x errcode=%08x ilen=%08x\n",
9417 vmcs_read32(VM_EXIT_INTR_INFO),
9418 vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
9419 vmcs_read32(VM_EXIT_INSTRUCTION_LEN));
9420 pr_err(" reason=%08x qualification=%016lx\n",
9421 vmcs_read32(VM_EXIT_REASON), vmcs_readl(EXIT_QUALIFICATION));
9422 pr_err("IDTVectoring: info=%08x errcode=%08x\n",
9423 vmcs_read32(IDT_VECTORING_INFO_FIELD),
9424 vmcs_read32(IDT_VECTORING_ERROR_CODE));
9425 pr_err("TSC Offset = 0x%016llx\n", vmcs_read64(TSC_OFFSET));
9426 if (secondary_exec_control & SECONDARY_EXEC_TSC_SCALING)
9427 pr_err("TSC Multiplier = 0x%016llx\n",
9428 vmcs_read64(TSC_MULTIPLIER));
9429 if (cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW)
9430 pr_err("TPR Threshold = 0x%02x\n", vmcs_read32(TPR_THRESHOLD));
9431 if (pin_based_exec_ctrl & PIN_BASED_POSTED_INTR)
9432 pr_err("PostedIntrVec = 0x%02x\n", vmcs_read16(POSTED_INTR_NV));
9433 if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT))
9434 pr_err("EPT pointer = 0x%016llx\n", vmcs_read64(EPT_POINTER));
9435 n = vmcs_read32(CR3_TARGET_COUNT);
9436 for (i = 0; i + 1 < n; i += 4)
9437 pr_err("CR3 target%u=%016lx target%u=%016lx\n",
9438 i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2),
9439 i + 1, vmcs_readl(CR3_TARGET_VALUE0 + i * 2 + 2));
9441 pr_err("CR3 target%u=%016lx\n",
9442 i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2));
9443 if (secondary_exec_control & SECONDARY_EXEC_PAUSE_LOOP_EXITING)
9444 pr_err("PLE Gap=%08x Window=%08x\n",
9445 vmcs_read32(PLE_GAP), vmcs_read32(PLE_WINDOW));
9446 if (secondary_exec_control & SECONDARY_EXEC_ENABLE_VPID)
9447 pr_err("Virtual processor ID = 0x%04x\n",
9448 vmcs_read16(VIRTUAL_PROCESSOR_ID));
9452 * The guest has exited. See if we can fix it or if we need userspace
9455 static int vmx_handle_exit(struct kvm_vcpu *vcpu)
9457 struct vcpu_vmx *vmx = to_vmx(vcpu);
9458 u32 exit_reason = vmx->exit_reason;
9459 u32 vectoring_info = vmx->idt_vectoring_info;
9461 trace_kvm_exit(exit_reason, vcpu, KVM_ISA_VMX);
9464 * Flush logged GPAs PML buffer, this will make dirty_bitmap more
9465 * updated. Another good is, in kvm_vm_ioctl_get_dirty_log, before
9466 * querying dirty_bitmap, we only need to kick all vcpus out of guest
9467 * mode as if vcpus is in root mode, the PML buffer must has been
9471 vmx_flush_pml_buffer(vcpu);
9473 /* If guest state is invalid, start emulating */
9474 if (vmx->emulation_required)
9475 return handle_invalid_guest_state(vcpu);
9477 if (is_guest_mode(vcpu) && nested_vmx_exit_reflected(vcpu, exit_reason))
9478 return nested_vmx_reflect_vmexit(vcpu, exit_reason);
9480 if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
9482 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
9483 vcpu->run->fail_entry.hardware_entry_failure_reason
9488 if (unlikely(vmx->fail)) {
9489 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
9490 vcpu->run->fail_entry.hardware_entry_failure_reason
9491 = vmcs_read32(VM_INSTRUCTION_ERROR);
9497 * Do not try to fix EXIT_REASON_EPT_MISCONFIG if it caused by
9498 * delivery event since it indicates guest is accessing MMIO.
9499 * The vm-exit can be triggered again after return to guest that
9500 * will cause infinite loop.
9502 if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
9503 (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
9504 exit_reason != EXIT_REASON_EPT_VIOLATION &&
9505 exit_reason != EXIT_REASON_PML_FULL &&
9506 exit_reason != EXIT_REASON_TASK_SWITCH)) {
9507 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
9508 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_DELIVERY_EV;
9509 vcpu->run->internal.ndata = 3;
9510 vcpu->run->internal.data[0] = vectoring_info;
9511 vcpu->run->internal.data[1] = exit_reason;
9512 vcpu->run->internal.data[2] = vcpu->arch.exit_qualification;
9513 if (exit_reason == EXIT_REASON_EPT_MISCONFIG) {
9514 vcpu->run->internal.ndata++;
9515 vcpu->run->internal.data[3] =
9516 vmcs_read64(GUEST_PHYSICAL_ADDRESS);
9521 if (unlikely(!enable_vnmi &&
9522 vmx->loaded_vmcs->soft_vnmi_blocked)) {
9523 if (vmx_interrupt_allowed(vcpu)) {
9524 vmx->loaded_vmcs->soft_vnmi_blocked = 0;
9525 } else if (vmx->loaded_vmcs->vnmi_blocked_time > 1000000000LL &&
9526 vcpu->arch.nmi_pending) {
9528 * This CPU don't support us in finding the end of an
9529 * NMI-blocked window if the guest runs with IRQs
9530 * disabled. So we pull the trigger after 1 s of
9531 * futile waiting, but inform the user about this.
9533 printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
9534 "state on VCPU %d after 1 s timeout\n",
9535 __func__, vcpu->vcpu_id);
9536 vmx->loaded_vmcs->soft_vnmi_blocked = 0;
9540 if (exit_reason < kvm_vmx_max_exit_handlers
9541 && kvm_vmx_exit_handlers[exit_reason])
9542 return kvm_vmx_exit_handlers[exit_reason](vcpu);
9544 vcpu_unimpl(vcpu, "vmx: unexpected exit reason 0x%x\n",
9546 kvm_queue_exception(vcpu, UD_VECTOR);
9551 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
9553 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
9555 if (is_guest_mode(vcpu) &&
9556 nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
9559 if (irr == -1 || tpr < irr) {
9560 vmcs_write32(TPR_THRESHOLD, 0);
9564 vmcs_write32(TPR_THRESHOLD, irr);
9567 static void vmx_set_virtual_apic_mode(struct kvm_vcpu *vcpu)
9569 u32 sec_exec_control;
9571 if (!lapic_in_kernel(vcpu))
9574 /* Postpone execution until vmcs01 is the current VMCS. */
9575 if (is_guest_mode(vcpu)) {
9576 to_vmx(vcpu)->nested.change_vmcs01_virtual_apic_mode = true;
9580 if (!cpu_need_tpr_shadow(vcpu))
9583 sec_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
9584 sec_exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
9585 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE);
9587 switch (kvm_get_apic_mode(vcpu)) {
9588 case LAPIC_MODE_INVALID:
9589 WARN_ONCE(true, "Invalid local APIC state");
9590 case LAPIC_MODE_DISABLED:
9592 case LAPIC_MODE_XAPIC:
9593 if (flexpriority_enabled) {
9595 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
9596 vmx_flush_tlb(vcpu, true);
9599 case LAPIC_MODE_X2APIC:
9600 if (cpu_has_vmx_virtualize_x2apic_mode())
9602 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
9605 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, sec_exec_control);
9607 vmx_update_msr_bitmap(vcpu);
9610 static void vmx_set_apic_access_page_addr(struct kvm_vcpu *vcpu, hpa_t hpa)
9612 if (!is_guest_mode(vcpu)) {
9613 vmcs_write64(APIC_ACCESS_ADDR, hpa);
9614 vmx_flush_tlb(vcpu, true);
9618 static void vmx_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr)
9626 status = vmcs_read16(GUEST_INTR_STATUS);
9628 if (max_isr != old) {
9630 status |= max_isr << 8;
9631 vmcs_write16(GUEST_INTR_STATUS, status);
9635 static void vmx_set_rvi(int vector)
9643 status = vmcs_read16(GUEST_INTR_STATUS);
9644 old = (u8)status & 0xff;
9645 if ((u8)vector != old) {
9647 status |= (u8)vector;
9648 vmcs_write16(GUEST_INTR_STATUS, status);
9652 static void vmx_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
9655 * When running L2, updating RVI is only relevant when
9656 * vmcs12 virtual-interrupt-delivery enabled.
9657 * However, it can be enabled only when L1 also
9658 * intercepts external-interrupts and in that case
9659 * we should not update vmcs02 RVI but instead intercept
9660 * interrupt. Therefore, do nothing when running L2.
9662 if (!is_guest_mode(vcpu))
9663 vmx_set_rvi(max_irr);
9666 static int vmx_sync_pir_to_irr(struct kvm_vcpu *vcpu)
9668 struct vcpu_vmx *vmx = to_vmx(vcpu);
9670 bool max_irr_updated;
9672 WARN_ON(!vcpu->arch.apicv_active);
9673 if (pi_test_on(&vmx->pi_desc)) {
9674 pi_clear_on(&vmx->pi_desc);
9676 * IOMMU can write to PIR.ON, so the barrier matters even on UP.
9677 * But on x86 this is just a compiler barrier anyway.
9679 smp_mb__after_atomic();
9681 kvm_apic_update_irr(vcpu, vmx->pi_desc.pir, &max_irr);
9684 * If we are running L2 and L1 has a new pending interrupt
9685 * which can be injected, we should re-evaluate
9686 * what should be done with this new L1 interrupt.
9687 * If L1 intercepts external-interrupts, we should
9688 * exit from L2 to L1. Otherwise, interrupt should be
9689 * delivered directly to L2.
9691 if (is_guest_mode(vcpu) && max_irr_updated) {
9692 if (nested_exit_on_intr(vcpu))
9693 kvm_vcpu_exiting_guest_mode(vcpu);
9695 kvm_make_request(KVM_REQ_EVENT, vcpu);
9698 max_irr = kvm_lapic_find_highest_irr(vcpu);
9700 vmx_hwapic_irr_update(vcpu, max_irr);
9704 static void vmx_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
9706 if (!kvm_vcpu_apicv_active(vcpu))
9709 vmcs_write64(EOI_EXIT_BITMAP0, eoi_exit_bitmap[0]);
9710 vmcs_write64(EOI_EXIT_BITMAP1, eoi_exit_bitmap[1]);
9711 vmcs_write64(EOI_EXIT_BITMAP2, eoi_exit_bitmap[2]);
9712 vmcs_write64(EOI_EXIT_BITMAP3, eoi_exit_bitmap[3]);
9715 static void vmx_apicv_post_state_restore(struct kvm_vcpu *vcpu)
9717 struct vcpu_vmx *vmx = to_vmx(vcpu);
9719 pi_clear_on(&vmx->pi_desc);
9720 memset(vmx->pi_desc.pir, 0, sizeof(vmx->pi_desc.pir));
9723 static void vmx_complete_atomic_exit(struct vcpu_vmx *vmx)
9725 u32 exit_intr_info = 0;
9726 u16 basic_exit_reason = (u16)vmx->exit_reason;
9728 if (!(basic_exit_reason == EXIT_REASON_MCE_DURING_VMENTRY
9729 || basic_exit_reason == EXIT_REASON_EXCEPTION_NMI))
9732 if (!(vmx->exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY))
9733 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
9734 vmx->exit_intr_info = exit_intr_info;
9736 /* if exit due to PF check for async PF */
9737 if (is_page_fault(exit_intr_info))
9738 vmx->vcpu.arch.apf.host_apf_reason = kvm_read_and_reset_pf_reason();
9740 /* Handle machine checks before interrupts are enabled */
9741 if (basic_exit_reason == EXIT_REASON_MCE_DURING_VMENTRY ||
9742 is_machine_check(exit_intr_info))
9743 kvm_machine_check();
9745 /* We need to handle NMIs before interrupts are enabled */
9746 if (is_nmi(exit_intr_info)) {
9747 kvm_before_interrupt(&vmx->vcpu);
9749 kvm_after_interrupt(&vmx->vcpu);
9753 static void vmx_handle_external_intr(struct kvm_vcpu *vcpu)
9755 u32 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
9757 if ((exit_intr_info & (INTR_INFO_VALID_MASK | INTR_INFO_INTR_TYPE_MASK))
9758 == (INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR)) {
9759 unsigned int vector;
9760 unsigned long entry;
9762 struct vcpu_vmx *vmx = to_vmx(vcpu);
9763 #ifdef CONFIG_X86_64
9767 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
9768 desc = (gate_desc *)vmx->host_idt_base + vector;
9769 entry = gate_offset(desc);
9771 #ifdef CONFIG_X86_64
9772 "mov %%" _ASM_SP ", %[sp]\n\t"
9773 "and $0xfffffffffffffff0, %%" _ASM_SP "\n\t"
9778 __ASM_SIZE(push) " $%c[cs]\n\t"
9781 #ifdef CONFIG_X86_64
9786 THUNK_TARGET(entry),
9787 [ss]"i"(__KERNEL_DS),
9788 [cs]"i"(__KERNEL_CS)
9792 STACK_FRAME_NON_STANDARD(vmx_handle_external_intr);
9794 static bool vmx_has_emulated_msr(int index)
9797 case MSR_IA32_SMBASE:
9799 * We cannot do SMM unless we can run the guest in big
9802 return enable_unrestricted_guest || emulate_invalid_guest_state;
9803 case MSR_AMD64_VIRT_SPEC_CTRL:
9804 /* This is AMD only. */
9811 static bool vmx_mpx_supported(void)
9813 return (vmcs_config.vmexit_ctrl & VM_EXIT_CLEAR_BNDCFGS) &&
9814 (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_BNDCFGS);
9817 static bool vmx_xsaves_supported(void)
9819 return vmcs_config.cpu_based_2nd_exec_ctrl &
9820 SECONDARY_EXEC_XSAVES;
9823 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
9828 bool idtv_info_valid;
9830 idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
9833 if (vmx->loaded_vmcs->nmi_known_unmasked)
9836 * Can't use vmx->exit_intr_info since we're not sure what
9837 * the exit reason is.
9839 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
9840 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
9841 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
9843 * SDM 3: 27.7.1.2 (September 2008)
9844 * Re-set bit "block by NMI" before VM entry if vmexit caused by
9845 * a guest IRET fault.
9846 * SDM 3: 23.2.2 (September 2008)
9847 * Bit 12 is undefined in any of the following cases:
9848 * If the VM exit sets the valid bit in the IDT-vectoring
9849 * information field.
9850 * If the VM exit is due to a double fault.
9852 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
9853 vector != DF_VECTOR && !idtv_info_valid)
9854 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
9855 GUEST_INTR_STATE_NMI);
9857 vmx->loaded_vmcs->nmi_known_unmasked =
9858 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
9859 & GUEST_INTR_STATE_NMI);
9860 } else if (unlikely(vmx->loaded_vmcs->soft_vnmi_blocked))
9861 vmx->loaded_vmcs->vnmi_blocked_time +=
9862 ktime_to_ns(ktime_sub(ktime_get(),
9863 vmx->loaded_vmcs->entry_time));
9866 static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu,
9867 u32 idt_vectoring_info,
9868 int instr_len_field,
9869 int error_code_field)
9873 bool idtv_info_valid;
9875 idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
9877 vcpu->arch.nmi_injected = false;
9878 kvm_clear_exception_queue(vcpu);
9879 kvm_clear_interrupt_queue(vcpu);
9881 if (!idtv_info_valid)
9884 kvm_make_request(KVM_REQ_EVENT, vcpu);
9886 vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
9887 type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
9890 case INTR_TYPE_NMI_INTR:
9891 vcpu->arch.nmi_injected = true;
9893 * SDM 3: 27.7.1.2 (September 2008)
9894 * Clear bit "block by NMI" before VM entry if a NMI
9897 vmx_set_nmi_mask(vcpu, false);
9899 case INTR_TYPE_SOFT_EXCEPTION:
9900 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
9902 case INTR_TYPE_HARD_EXCEPTION:
9903 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
9904 u32 err = vmcs_read32(error_code_field);
9905 kvm_requeue_exception_e(vcpu, vector, err);
9907 kvm_requeue_exception(vcpu, vector);
9909 case INTR_TYPE_SOFT_INTR:
9910 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
9912 case INTR_TYPE_EXT_INTR:
9913 kvm_queue_interrupt(vcpu, vector, type == INTR_TYPE_SOFT_INTR);
9920 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
9922 __vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info,
9923 VM_EXIT_INSTRUCTION_LEN,
9924 IDT_VECTORING_ERROR_CODE);
9927 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
9929 __vmx_complete_interrupts(vcpu,
9930 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
9931 VM_ENTRY_INSTRUCTION_LEN,
9932 VM_ENTRY_EXCEPTION_ERROR_CODE);
9934 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
9937 static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
9940 struct perf_guest_switch_msr *msrs;
9942 msrs = perf_guest_get_msrs(&nr_msrs);
9947 for (i = 0; i < nr_msrs; i++)
9948 if (msrs[i].host == msrs[i].guest)
9949 clear_atomic_switch_msr(vmx, msrs[i].msr);
9951 add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
9955 static void vmx_arm_hv_timer(struct kvm_vcpu *vcpu)
9957 struct vcpu_vmx *vmx = to_vmx(vcpu);
9961 if (vmx->hv_deadline_tsc == -1)
9965 if (vmx->hv_deadline_tsc > tscl)
9966 /* sure to be 32 bit only because checked on set_hv_timer */
9967 delta_tsc = (u32)((vmx->hv_deadline_tsc - tscl) >>
9968 cpu_preemption_timer_multi);
9972 vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, delta_tsc);
9975 static void __noclone vmx_vcpu_run(struct kvm_vcpu *vcpu)
9977 struct vcpu_vmx *vmx = to_vmx(vcpu);
9978 unsigned long cr3, cr4, evmcs_rsp;
9980 /* Record the guest's net vcpu time for enforced NMI injections. */
9981 if (unlikely(!enable_vnmi &&
9982 vmx->loaded_vmcs->soft_vnmi_blocked))
9983 vmx->loaded_vmcs->entry_time = ktime_get();
9985 /* Don't enter VMX if guest state is invalid, let the exit handler
9986 start emulation until we arrive back to a valid state */
9987 if (vmx->emulation_required)
9990 if (vmx->ple_window_dirty) {
9991 vmx->ple_window_dirty = false;
9992 vmcs_write32(PLE_WINDOW, vmx->ple_window);
9995 if (vmx->nested.sync_shadow_vmcs) {
9996 copy_vmcs12_to_shadow(vmx);
9997 vmx->nested.sync_shadow_vmcs = false;
10000 if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
10001 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
10002 if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
10003 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
10005 cr3 = __get_current_cr3_fast();
10006 if (unlikely(cr3 != vmx->loaded_vmcs->vmcs_host_cr3)) {
10007 vmcs_writel(HOST_CR3, cr3);
10008 vmx->loaded_vmcs->vmcs_host_cr3 = cr3;
10011 cr4 = cr4_read_shadow();
10012 if (unlikely(cr4 != vmx->loaded_vmcs->vmcs_host_cr4)) {
10013 vmcs_writel(HOST_CR4, cr4);
10014 vmx->loaded_vmcs->vmcs_host_cr4 = cr4;
10017 /* When single-stepping over STI and MOV SS, we must clear the
10018 * corresponding interruptibility bits in the guest state. Otherwise
10019 * vmentry fails as it then expects bit 14 (BS) in pending debug
10020 * exceptions being set, but that's not correct for the guest debugging
10022 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
10023 vmx_set_interrupt_shadow(vcpu, 0);
10025 if (static_cpu_has(X86_FEATURE_PKU) &&
10026 kvm_read_cr4_bits(vcpu, X86_CR4_PKE) &&
10027 vcpu->arch.pkru != vmx->host_pkru)
10028 __write_pkru(vcpu->arch.pkru);
10030 atomic_switch_perf_msrs(vmx);
10032 vmx_arm_hv_timer(vcpu);
10035 * If this vCPU has touched SPEC_CTRL, restore the guest's value if
10036 * it's non-zero. Since vmentry is serialising on affected CPUs, there
10037 * is no need to worry about the conditional branch over the wrmsr
10038 * being speculatively taken.
10040 x86_spec_ctrl_set_guest(vmx->spec_ctrl, 0);
10042 vmx->__launched = vmx->loaded_vmcs->launched;
10044 evmcs_rsp = static_branch_unlikely(&enable_evmcs) ?
10045 (unsigned long)¤t_evmcs->host_rsp : 0;
10048 /* Store host registers */
10049 "push %%" _ASM_DX "; push %%" _ASM_BP ";"
10050 "push %%" _ASM_CX " \n\t" /* placeholder for guest rcx */
10051 "push %%" _ASM_CX " \n\t"
10052 "cmp %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
10054 "mov %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
10055 /* Avoid VMWRITE when Enlightened VMCS is in use */
10056 "test %%" _ASM_SI ", %%" _ASM_SI " \n\t"
10058 "mov %%" _ASM_SP ", (%%" _ASM_SI ") \n\t"
10061 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
10063 /* Reload cr2 if changed */
10064 "mov %c[cr2](%0), %%" _ASM_AX " \n\t"
10065 "mov %%cr2, %%" _ASM_DX " \n\t"
10066 "cmp %%" _ASM_AX ", %%" _ASM_DX " \n\t"
10068 "mov %%" _ASM_AX", %%cr2 \n\t"
10070 /* Check if vmlaunch of vmresume is needed */
10071 "cmpl $0, %c[launched](%0) \n\t"
10072 /* Load guest registers. Don't clobber flags. */
10073 "mov %c[rax](%0), %%" _ASM_AX " \n\t"
10074 "mov %c[rbx](%0), %%" _ASM_BX " \n\t"
10075 "mov %c[rdx](%0), %%" _ASM_DX " \n\t"
10076 "mov %c[rsi](%0), %%" _ASM_SI " \n\t"
10077 "mov %c[rdi](%0), %%" _ASM_DI " \n\t"
10078 "mov %c[rbp](%0), %%" _ASM_BP " \n\t"
10079 #ifdef CONFIG_X86_64
10080 "mov %c[r8](%0), %%r8 \n\t"
10081 "mov %c[r9](%0), %%r9 \n\t"
10082 "mov %c[r10](%0), %%r10 \n\t"
10083 "mov %c[r11](%0), %%r11 \n\t"
10084 "mov %c[r12](%0), %%r12 \n\t"
10085 "mov %c[r13](%0), %%r13 \n\t"
10086 "mov %c[r14](%0), %%r14 \n\t"
10087 "mov %c[r15](%0), %%r15 \n\t"
10089 "mov %c[rcx](%0), %%" _ASM_CX " \n\t" /* kills %0 (ecx) */
10091 /* Enter guest mode */
10093 __ex(ASM_VMX_VMLAUNCH) "\n\t"
10095 "1: " __ex(ASM_VMX_VMRESUME) "\n\t"
10097 /* Save guest registers, load host registers, keep flags */
10098 "mov %0, %c[wordsize](%%" _ASM_SP ") \n\t"
10100 "setbe %c[fail](%0)\n\t"
10101 "mov %%" _ASM_AX ", %c[rax](%0) \n\t"
10102 "mov %%" _ASM_BX ", %c[rbx](%0) \n\t"
10103 __ASM_SIZE(pop) " %c[rcx](%0) \n\t"
10104 "mov %%" _ASM_DX ", %c[rdx](%0) \n\t"
10105 "mov %%" _ASM_SI ", %c[rsi](%0) \n\t"
10106 "mov %%" _ASM_DI ", %c[rdi](%0) \n\t"
10107 "mov %%" _ASM_BP ", %c[rbp](%0) \n\t"
10108 #ifdef CONFIG_X86_64
10109 "mov %%r8, %c[r8](%0) \n\t"
10110 "mov %%r9, %c[r9](%0) \n\t"
10111 "mov %%r10, %c[r10](%0) \n\t"
10112 "mov %%r11, %c[r11](%0) \n\t"
10113 "mov %%r12, %c[r12](%0) \n\t"
10114 "mov %%r13, %c[r13](%0) \n\t"
10115 "mov %%r14, %c[r14](%0) \n\t"
10116 "mov %%r15, %c[r15](%0) \n\t"
10117 "xor %%r8d, %%r8d \n\t"
10118 "xor %%r9d, %%r9d \n\t"
10119 "xor %%r10d, %%r10d \n\t"
10120 "xor %%r11d, %%r11d \n\t"
10121 "xor %%r12d, %%r12d \n\t"
10122 "xor %%r13d, %%r13d \n\t"
10123 "xor %%r14d, %%r14d \n\t"
10124 "xor %%r15d, %%r15d \n\t"
10126 "mov %%cr2, %%" _ASM_AX " \n\t"
10127 "mov %%" _ASM_AX ", %c[cr2](%0) \n\t"
10129 "xor %%eax, %%eax \n\t"
10130 "xor %%ebx, %%ebx \n\t"
10131 "xor %%esi, %%esi \n\t"
10132 "xor %%edi, %%edi \n\t"
10133 "pop %%" _ASM_BP "; pop %%" _ASM_DX " \n\t"
10134 ".pushsection .rodata \n\t"
10135 ".global vmx_return \n\t"
10136 "vmx_return: " _ASM_PTR " 2b \n\t"
10138 : : "c"(vmx), "d"((unsigned long)HOST_RSP), "S"(evmcs_rsp),
10139 [launched]"i"(offsetof(struct vcpu_vmx, __launched)),
10140 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
10141 [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
10142 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
10143 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
10144 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
10145 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
10146 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
10147 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
10148 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
10149 #ifdef CONFIG_X86_64
10150 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
10151 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
10152 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
10153 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
10154 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
10155 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
10156 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
10157 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
10159 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2)),
10160 [wordsize]"i"(sizeof(ulong))
10162 #ifdef CONFIG_X86_64
10163 , "rax", "rbx", "rdi"
10164 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
10166 , "eax", "ebx", "edi"
10171 * We do not use IBRS in the kernel. If this vCPU has used the
10172 * SPEC_CTRL MSR it may have left it on; save the value and
10173 * turn it off. This is much more efficient than blindly adding
10174 * it to the atomic save/restore list. Especially as the former
10175 * (Saving guest MSRs on vmexit) doesn't even exist in KVM.
10177 * For non-nested case:
10178 * If the L01 MSR bitmap does not intercept the MSR, then we need to
10182 * If the L02 MSR bitmap does not intercept the MSR, then we need to
10185 if (unlikely(!msr_write_intercepted(vcpu, MSR_IA32_SPEC_CTRL)))
10186 vmx->spec_ctrl = native_read_msr(MSR_IA32_SPEC_CTRL);
10188 x86_spec_ctrl_restore_host(vmx->spec_ctrl, 0);
10190 /* Eliminate branch target predictions from guest mode */
10193 /* All fields are clean at this point */
10194 if (static_branch_unlikely(&enable_evmcs))
10195 current_evmcs->hv_clean_fields |=
10196 HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
10198 /* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
10199 if (vmx->host_debugctlmsr)
10200 update_debugctlmsr(vmx->host_debugctlmsr);
10202 #ifndef CONFIG_X86_64
10204 * The sysexit path does not restore ds/es, so we must set them to
10205 * a reasonable value ourselves.
10207 * We can't defer this to vmx_load_host_state() since that function
10208 * may be executed in interrupt context, which saves and restore segments
10209 * around it, nullifying its effect.
10211 loadsegment(ds, __USER_DS);
10212 loadsegment(es, __USER_DS);
10215 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
10216 | (1 << VCPU_EXREG_RFLAGS)
10217 | (1 << VCPU_EXREG_PDPTR)
10218 | (1 << VCPU_EXREG_SEGMENTS)
10219 | (1 << VCPU_EXREG_CR3));
10220 vcpu->arch.regs_dirty = 0;
10223 * eager fpu is enabled if PKEY is supported and CR4 is switched
10224 * back on host, so it is safe to read guest PKRU from current
10227 if (static_cpu_has(X86_FEATURE_PKU) &&
10228 kvm_read_cr4_bits(vcpu, X86_CR4_PKE)) {
10229 vcpu->arch.pkru = __read_pkru();
10230 if (vcpu->arch.pkru != vmx->host_pkru)
10231 __write_pkru(vmx->host_pkru);
10234 vmx->nested.nested_run_pending = 0;
10235 vmx->idt_vectoring_info = 0;
10237 vmx->exit_reason = vmx->fail ? 0xdead : vmcs_read32(VM_EXIT_REASON);
10238 if (vmx->fail || (vmx->exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY))
10241 vmx->loaded_vmcs->launched = 1;
10242 vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
10244 vmx_complete_atomic_exit(vmx);
10245 vmx_recover_nmi_blocking(vmx);
10246 vmx_complete_interrupts(vmx);
10248 STACK_FRAME_NON_STANDARD(vmx_vcpu_run);
10250 static struct kvm *vmx_vm_alloc(void)
10252 struct kvm_vmx *kvm_vmx = vzalloc(sizeof(struct kvm_vmx));
10253 return &kvm_vmx->kvm;
10256 static void vmx_vm_free(struct kvm *kvm)
10258 vfree(to_kvm_vmx(kvm));
10261 static void vmx_switch_vmcs(struct kvm_vcpu *vcpu, struct loaded_vmcs *vmcs)
10263 struct vcpu_vmx *vmx = to_vmx(vcpu);
10266 if (vmx->loaded_vmcs == vmcs)
10270 vmx->loaded_vmcs = vmcs;
10271 vmx_vcpu_put(vcpu);
10272 vmx_vcpu_load(vcpu, cpu);
10277 * Ensure that the current vmcs of the logical processor is the
10278 * vmcs01 of the vcpu before calling free_nested().
10280 static void vmx_free_vcpu_nested(struct kvm_vcpu *vcpu)
10282 struct vcpu_vmx *vmx = to_vmx(vcpu);
10285 vmx_switch_vmcs(vcpu, &vmx->vmcs01);
10290 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
10292 struct vcpu_vmx *vmx = to_vmx(vcpu);
10295 vmx_destroy_pml_buffer(vmx);
10296 free_vpid(vmx->vpid);
10297 leave_guest_mode(vcpu);
10298 vmx_free_vcpu_nested(vcpu);
10299 free_loaded_vmcs(vmx->loaded_vmcs);
10300 kfree(vmx->guest_msrs);
10301 kvm_vcpu_uninit(vcpu);
10302 kmem_cache_free(kvm_vcpu_cache, vmx);
10305 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
10308 struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
10309 unsigned long *msr_bitmap;
10313 return ERR_PTR(-ENOMEM);
10315 vmx->vpid = allocate_vpid();
10317 err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
10324 * If PML is turned on, failure on enabling PML just results in failure
10325 * of creating the vcpu, therefore we can simplify PML logic (by
10326 * avoiding dealing with cases, such as enabling PML partially on vcpus
10327 * for the guest, etc.
10330 vmx->pml_pg = alloc_page(GFP_KERNEL | __GFP_ZERO);
10335 vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
10336 BUILD_BUG_ON(ARRAY_SIZE(vmx_msr_index) * sizeof(vmx->guest_msrs[0])
10339 if (!vmx->guest_msrs)
10342 err = alloc_loaded_vmcs(&vmx->vmcs01);
10346 msr_bitmap = vmx->vmcs01.msr_bitmap;
10347 vmx_disable_intercept_for_msr(msr_bitmap, MSR_FS_BASE, MSR_TYPE_RW);
10348 vmx_disable_intercept_for_msr(msr_bitmap, MSR_GS_BASE, MSR_TYPE_RW);
10349 vmx_disable_intercept_for_msr(msr_bitmap, MSR_KERNEL_GS_BASE, MSR_TYPE_RW);
10350 vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_CS, MSR_TYPE_RW);
10351 vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_ESP, MSR_TYPE_RW);
10352 vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_EIP, MSR_TYPE_RW);
10353 vmx->msr_bitmap_mode = 0;
10355 vmx->loaded_vmcs = &vmx->vmcs01;
10357 vmx_vcpu_load(&vmx->vcpu, cpu);
10358 vmx->vcpu.cpu = cpu;
10359 vmx_vcpu_setup(vmx);
10360 vmx_vcpu_put(&vmx->vcpu);
10362 if (cpu_need_virtualize_apic_accesses(&vmx->vcpu)) {
10363 err = alloc_apic_access_page(kvm);
10368 if (enable_ept && !enable_unrestricted_guest) {
10369 err = init_rmode_identity_map(kvm);
10375 nested_vmx_setup_ctls_msrs(&vmx->nested.msrs,
10376 kvm_vcpu_apicv_active(&vmx->vcpu));
10377 vmx->nested.vpid02 = allocate_vpid();
10380 vmx->nested.posted_intr_nv = -1;
10381 vmx->nested.current_vmptr = -1ull;
10383 vmx->msr_ia32_feature_control_valid_bits = FEATURE_CONTROL_LOCKED;
10386 * Enforce invariant: pi_desc.nv is always either POSTED_INTR_VECTOR
10387 * or POSTED_INTR_WAKEUP_VECTOR.
10389 vmx->pi_desc.nv = POSTED_INTR_VECTOR;
10390 vmx->pi_desc.sn = 1;
10395 free_vpid(vmx->nested.vpid02);
10396 free_loaded_vmcs(vmx->loaded_vmcs);
10398 kfree(vmx->guest_msrs);
10400 vmx_destroy_pml_buffer(vmx);
10402 kvm_vcpu_uninit(&vmx->vcpu);
10404 free_vpid(vmx->vpid);
10405 kmem_cache_free(kvm_vcpu_cache, vmx);
10406 return ERR_PTR(err);
10409 static int vmx_vm_init(struct kvm *kvm)
10412 kvm->arch.pause_in_guest = true;
10416 static void __init vmx_check_processor_compat(void *rtn)
10418 struct vmcs_config vmcs_conf;
10421 if (setup_vmcs_config(&vmcs_conf) < 0)
10422 *(int *)rtn = -EIO;
10423 nested_vmx_setup_ctls_msrs(&vmcs_conf.nested, enable_apicv);
10424 if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
10425 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
10426 smp_processor_id());
10427 *(int *)rtn = -EIO;
10431 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
10436 /* For VT-d and EPT combination
10437 * 1. MMIO: always map as UC
10438 * 2. EPT with VT-d:
10439 * a. VT-d without snooping control feature: can't guarantee the
10440 * result, try to trust guest.
10441 * b. VT-d with snooping control feature: snooping control feature of
10442 * VT-d engine can guarantee the cache correctness. Just set it
10443 * to WB to keep consistent with host. So the same as item 3.
10444 * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
10445 * consistent with host MTRR
10448 cache = MTRR_TYPE_UNCACHABLE;
10452 if (!kvm_arch_has_noncoherent_dma(vcpu->kvm)) {
10453 ipat = VMX_EPT_IPAT_BIT;
10454 cache = MTRR_TYPE_WRBACK;
10458 if (kvm_read_cr0(vcpu) & X86_CR0_CD) {
10459 ipat = VMX_EPT_IPAT_BIT;
10460 if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
10461 cache = MTRR_TYPE_WRBACK;
10463 cache = MTRR_TYPE_UNCACHABLE;
10467 cache = kvm_mtrr_get_guest_memory_type(vcpu, gfn);
10470 return (cache << VMX_EPT_MT_EPTE_SHIFT) | ipat;
10473 static int vmx_get_lpage_level(void)
10475 if (enable_ept && !cpu_has_vmx_ept_1g_page())
10476 return PT_DIRECTORY_LEVEL;
10478 /* For shadow and EPT supported 1GB page */
10479 return PT_PDPE_LEVEL;
10482 static void vmcs_set_secondary_exec_control(u32 new_ctl)
10485 * These bits in the secondary execution controls field
10486 * are dynamic, the others are mostly based on the hypervisor
10487 * architecture and the guest's CPUID. Do not touch the
10491 SECONDARY_EXEC_SHADOW_VMCS |
10492 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
10493 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
10494 SECONDARY_EXEC_DESC;
10496 u32 cur_ctl = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
10498 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
10499 (new_ctl & ~mask) | (cur_ctl & mask));
10503 * Generate MSR_IA32_VMX_CR{0,4}_FIXED1 according to CPUID. Only set bits
10504 * (indicating "allowed-1") if they are supported in the guest's CPUID.
10506 static void nested_vmx_cr_fixed1_bits_update(struct kvm_vcpu *vcpu)
10508 struct vcpu_vmx *vmx = to_vmx(vcpu);
10509 struct kvm_cpuid_entry2 *entry;
10511 vmx->nested.msrs.cr0_fixed1 = 0xffffffff;
10512 vmx->nested.msrs.cr4_fixed1 = X86_CR4_PCE;
10514 #define cr4_fixed1_update(_cr4_mask, _reg, _cpuid_mask) do { \
10515 if (entry && (entry->_reg & (_cpuid_mask))) \
10516 vmx->nested.msrs.cr4_fixed1 |= (_cr4_mask); \
10519 entry = kvm_find_cpuid_entry(vcpu, 0x1, 0);
10520 cr4_fixed1_update(X86_CR4_VME, edx, bit(X86_FEATURE_VME));
10521 cr4_fixed1_update(X86_CR4_PVI, edx, bit(X86_FEATURE_VME));
10522 cr4_fixed1_update(X86_CR4_TSD, edx, bit(X86_FEATURE_TSC));
10523 cr4_fixed1_update(X86_CR4_DE, edx, bit(X86_FEATURE_DE));
10524 cr4_fixed1_update(X86_CR4_PSE, edx, bit(X86_FEATURE_PSE));
10525 cr4_fixed1_update(X86_CR4_PAE, edx, bit(X86_FEATURE_PAE));
10526 cr4_fixed1_update(X86_CR4_MCE, edx, bit(X86_FEATURE_MCE));
10527 cr4_fixed1_update(X86_CR4_PGE, edx, bit(X86_FEATURE_PGE));
10528 cr4_fixed1_update(X86_CR4_OSFXSR, edx, bit(X86_FEATURE_FXSR));
10529 cr4_fixed1_update(X86_CR4_OSXMMEXCPT, edx, bit(X86_FEATURE_XMM));
10530 cr4_fixed1_update(X86_CR4_VMXE, ecx, bit(X86_FEATURE_VMX));
10531 cr4_fixed1_update(X86_CR4_SMXE, ecx, bit(X86_FEATURE_SMX));
10532 cr4_fixed1_update(X86_CR4_PCIDE, ecx, bit(X86_FEATURE_PCID));
10533 cr4_fixed1_update(X86_CR4_OSXSAVE, ecx, bit(X86_FEATURE_XSAVE));
10535 entry = kvm_find_cpuid_entry(vcpu, 0x7, 0);
10536 cr4_fixed1_update(X86_CR4_FSGSBASE, ebx, bit(X86_FEATURE_FSGSBASE));
10537 cr4_fixed1_update(X86_CR4_SMEP, ebx, bit(X86_FEATURE_SMEP));
10538 cr4_fixed1_update(X86_CR4_SMAP, ebx, bit(X86_FEATURE_SMAP));
10539 cr4_fixed1_update(X86_CR4_PKE, ecx, bit(X86_FEATURE_PKU));
10540 cr4_fixed1_update(X86_CR4_UMIP, ecx, bit(X86_FEATURE_UMIP));
10542 #undef cr4_fixed1_update
10545 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
10547 struct vcpu_vmx *vmx = to_vmx(vcpu);
10549 if (cpu_has_secondary_exec_ctrls()) {
10550 vmx_compute_secondary_exec_control(vmx);
10551 vmcs_set_secondary_exec_control(vmx->secondary_exec_control);
10554 if (nested_vmx_allowed(vcpu))
10555 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
10556 FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
10558 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
10559 ~FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
10561 if (nested_vmx_allowed(vcpu))
10562 nested_vmx_cr_fixed1_bits_update(vcpu);
10565 static void vmx_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
10567 if (func == 1 && nested)
10568 entry->ecx |= bit(X86_FEATURE_VMX);
10571 static void nested_ept_inject_page_fault(struct kvm_vcpu *vcpu,
10572 struct x86_exception *fault)
10574 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
10575 struct vcpu_vmx *vmx = to_vmx(vcpu);
10577 unsigned long exit_qualification = vcpu->arch.exit_qualification;
10579 if (vmx->nested.pml_full) {
10580 exit_reason = EXIT_REASON_PML_FULL;
10581 vmx->nested.pml_full = false;
10582 exit_qualification &= INTR_INFO_UNBLOCK_NMI;
10583 } else if (fault->error_code & PFERR_RSVD_MASK)
10584 exit_reason = EXIT_REASON_EPT_MISCONFIG;
10586 exit_reason = EXIT_REASON_EPT_VIOLATION;
10588 nested_vmx_vmexit(vcpu, exit_reason, 0, exit_qualification);
10589 vmcs12->guest_physical_address = fault->address;
10592 static bool nested_ept_ad_enabled(struct kvm_vcpu *vcpu)
10594 return nested_ept_get_cr3(vcpu) & VMX_EPTP_AD_ENABLE_BIT;
10597 /* Callbacks for nested_ept_init_mmu_context: */
10599 static unsigned long nested_ept_get_cr3(struct kvm_vcpu *vcpu)
10601 /* return the page table to be shadowed - in our case, EPT12 */
10602 return get_vmcs12(vcpu)->ept_pointer;
10605 static int nested_ept_init_mmu_context(struct kvm_vcpu *vcpu)
10607 WARN_ON(mmu_is_nested(vcpu));
10608 if (!valid_ept_address(vcpu, nested_ept_get_cr3(vcpu)))
10611 kvm_mmu_unload(vcpu);
10612 kvm_init_shadow_ept_mmu(vcpu,
10613 to_vmx(vcpu)->nested.msrs.ept_caps &
10614 VMX_EPT_EXECUTE_ONLY_BIT,
10615 nested_ept_ad_enabled(vcpu));
10616 vcpu->arch.mmu.set_cr3 = vmx_set_cr3;
10617 vcpu->arch.mmu.get_cr3 = nested_ept_get_cr3;
10618 vcpu->arch.mmu.inject_page_fault = nested_ept_inject_page_fault;
10620 vcpu->arch.walk_mmu = &vcpu->arch.nested_mmu;
10624 static void nested_ept_uninit_mmu_context(struct kvm_vcpu *vcpu)
10626 vcpu->arch.walk_mmu = &vcpu->arch.mmu;
10629 static bool nested_vmx_is_page_fault_vmexit(struct vmcs12 *vmcs12,
10632 bool inequality, bit;
10634 bit = (vmcs12->exception_bitmap & (1u << PF_VECTOR)) != 0;
10636 (error_code & vmcs12->page_fault_error_code_mask) !=
10637 vmcs12->page_fault_error_code_match;
10638 return inequality ^ bit;
10641 static void vmx_inject_page_fault_nested(struct kvm_vcpu *vcpu,
10642 struct x86_exception *fault)
10644 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
10646 WARN_ON(!is_guest_mode(vcpu));
10648 if (nested_vmx_is_page_fault_vmexit(vmcs12, fault->error_code) &&
10649 !to_vmx(vcpu)->nested.nested_run_pending) {
10650 vmcs12->vm_exit_intr_error_code = fault->error_code;
10651 nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
10652 PF_VECTOR | INTR_TYPE_HARD_EXCEPTION |
10653 INTR_INFO_DELIVER_CODE_MASK | INTR_INFO_VALID_MASK,
10656 kvm_inject_page_fault(vcpu, fault);
10660 static inline bool nested_vmx_prepare_msr_bitmap(struct kvm_vcpu *vcpu,
10661 struct vmcs12 *vmcs12);
10663 static void nested_get_vmcs12_pages(struct kvm_vcpu *vcpu)
10665 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
10666 struct vcpu_vmx *vmx = to_vmx(vcpu);
10670 if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
10672 * Translate L1 physical address to host physical
10673 * address for vmcs02. Keep the page pinned, so this
10674 * physical address remains valid. We keep a reference
10675 * to it so we can release it later.
10677 if (vmx->nested.apic_access_page) { /* shouldn't happen */
10678 kvm_release_page_dirty(vmx->nested.apic_access_page);
10679 vmx->nested.apic_access_page = NULL;
10681 page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->apic_access_addr);
10683 * If translation failed, no matter: This feature asks
10684 * to exit when accessing the given address, and if it
10685 * can never be accessed, this feature won't do
10688 if (!is_error_page(page)) {
10689 vmx->nested.apic_access_page = page;
10690 hpa = page_to_phys(vmx->nested.apic_access_page);
10691 vmcs_write64(APIC_ACCESS_ADDR, hpa);
10693 vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL,
10694 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
10698 if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
10699 if (vmx->nested.virtual_apic_page) { /* shouldn't happen */
10700 kvm_release_page_dirty(vmx->nested.virtual_apic_page);
10701 vmx->nested.virtual_apic_page = NULL;
10703 page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->virtual_apic_page_addr);
10706 * If translation failed, VM entry will fail because
10707 * prepare_vmcs02 set VIRTUAL_APIC_PAGE_ADDR to -1ull.
10708 * Failing the vm entry is _not_ what the processor
10709 * does but it's basically the only possibility we
10710 * have. We could still enter the guest if CR8 load
10711 * exits are enabled, CR8 store exits are enabled, and
10712 * virtualize APIC access is disabled; in this case
10713 * the processor would never use the TPR shadow and we
10714 * could simply clear the bit from the execution
10715 * control. But such a configuration is useless, so
10716 * let's keep the code simple.
10718 if (!is_error_page(page)) {
10719 vmx->nested.virtual_apic_page = page;
10720 hpa = page_to_phys(vmx->nested.virtual_apic_page);
10721 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, hpa);
10725 if (nested_cpu_has_posted_intr(vmcs12)) {
10726 if (vmx->nested.pi_desc_page) { /* shouldn't happen */
10727 kunmap(vmx->nested.pi_desc_page);
10728 kvm_release_page_dirty(vmx->nested.pi_desc_page);
10729 vmx->nested.pi_desc_page = NULL;
10731 page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->posted_intr_desc_addr);
10732 if (is_error_page(page))
10734 vmx->nested.pi_desc_page = page;
10735 vmx->nested.pi_desc = kmap(vmx->nested.pi_desc_page);
10736 vmx->nested.pi_desc =
10737 (struct pi_desc *)((void *)vmx->nested.pi_desc +
10738 (unsigned long)(vmcs12->posted_intr_desc_addr &
10740 vmcs_write64(POSTED_INTR_DESC_ADDR,
10741 page_to_phys(vmx->nested.pi_desc_page) +
10742 (unsigned long)(vmcs12->posted_intr_desc_addr &
10745 if (nested_vmx_prepare_msr_bitmap(vcpu, vmcs12))
10746 vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL,
10747 CPU_BASED_USE_MSR_BITMAPS);
10749 vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
10750 CPU_BASED_USE_MSR_BITMAPS);
10753 static void vmx_start_preemption_timer(struct kvm_vcpu *vcpu)
10755 u64 preemption_timeout = get_vmcs12(vcpu)->vmx_preemption_timer_value;
10756 struct vcpu_vmx *vmx = to_vmx(vcpu);
10758 if (vcpu->arch.virtual_tsc_khz == 0)
10761 /* Make sure short timeouts reliably trigger an immediate vmexit.
10762 * hrtimer_start does not guarantee this. */
10763 if (preemption_timeout <= 1) {
10764 vmx_preemption_timer_fn(&vmx->nested.preemption_timer);
10768 preemption_timeout <<= VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
10769 preemption_timeout *= 1000000;
10770 do_div(preemption_timeout, vcpu->arch.virtual_tsc_khz);
10771 hrtimer_start(&vmx->nested.preemption_timer,
10772 ns_to_ktime(preemption_timeout), HRTIMER_MODE_REL);
10775 static int nested_vmx_check_io_bitmap_controls(struct kvm_vcpu *vcpu,
10776 struct vmcs12 *vmcs12)
10778 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
10781 if (!page_address_valid(vcpu, vmcs12->io_bitmap_a) ||
10782 !page_address_valid(vcpu, vmcs12->io_bitmap_b))
10788 static int nested_vmx_check_msr_bitmap_controls(struct kvm_vcpu *vcpu,
10789 struct vmcs12 *vmcs12)
10791 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
10794 if (!page_address_valid(vcpu, vmcs12->msr_bitmap))
10800 static int nested_vmx_check_tpr_shadow_controls(struct kvm_vcpu *vcpu,
10801 struct vmcs12 *vmcs12)
10803 if (!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
10806 if (!page_address_valid(vcpu, vmcs12->virtual_apic_page_addr))
10813 * Merge L0's and L1's MSR bitmap, return false to indicate that
10814 * we do not use the hardware.
10816 static inline bool nested_vmx_prepare_msr_bitmap(struct kvm_vcpu *vcpu,
10817 struct vmcs12 *vmcs12)
10821 unsigned long *msr_bitmap_l1;
10822 unsigned long *msr_bitmap_l0 = to_vmx(vcpu)->nested.vmcs02.msr_bitmap;
10824 * pred_cmd & spec_ctrl are trying to verify two things:
10826 * 1. L0 gave a permission to L1 to actually passthrough the MSR. This
10827 * ensures that we do not accidentally generate an L02 MSR bitmap
10828 * from the L12 MSR bitmap that is too permissive.
10829 * 2. That L1 or L2s have actually used the MSR. This avoids
10830 * unnecessarily merging of the bitmap if the MSR is unused. This
10831 * works properly because we only update the L01 MSR bitmap lazily.
10832 * So even if L0 should pass L1 these MSRs, the L01 bitmap is only
10833 * updated to reflect this when L1 (or its L2s) actually write to
10836 bool pred_cmd = !msr_write_intercepted_l01(vcpu, MSR_IA32_PRED_CMD);
10837 bool spec_ctrl = !msr_write_intercepted_l01(vcpu, MSR_IA32_SPEC_CTRL);
10839 /* Nothing to do if the MSR bitmap is not in use. */
10840 if (!cpu_has_vmx_msr_bitmap() ||
10841 !nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
10844 if (!nested_cpu_has_virt_x2apic_mode(vmcs12) &&
10845 !pred_cmd && !spec_ctrl)
10848 page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->msr_bitmap);
10849 if (is_error_page(page))
10852 msr_bitmap_l1 = (unsigned long *)kmap(page);
10853 if (nested_cpu_has_apic_reg_virt(vmcs12)) {
10855 * L0 need not intercept reads for MSRs between 0x800 and 0x8ff, it
10856 * just lets the processor take the value from the virtual-APIC page;
10857 * take those 256 bits directly from the L1 bitmap.
10859 for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
10860 unsigned word = msr / BITS_PER_LONG;
10861 msr_bitmap_l0[word] = msr_bitmap_l1[word];
10862 msr_bitmap_l0[word + (0x800 / sizeof(long))] = ~0;
10865 for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
10866 unsigned word = msr / BITS_PER_LONG;
10867 msr_bitmap_l0[word] = ~0;
10868 msr_bitmap_l0[word + (0x800 / sizeof(long))] = ~0;
10872 nested_vmx_disable_intercept_for_msr(
10873 msr_bitmap_l1, msr_bitmap_l0,
10874 X2APIC_MSR(APIC_TASKPRI),
10877 if (nested_cpu_has_vid(vmcs12)) {
10878 nested_vmx_disable_intercept_for_msr(
10879 msr_bitmap_l1, msr_bitmap_l0,
10880 X2APIC_MSR(APIC_EOI),
10882 nested_vmx_disable_intercept_for_msr(
10883 msr_bitmap_l1, msr_bitmap_l0,
10884 X2APIC_MSR(APIC_SELF_IPI),
10889 nested_vmx_disable_intercept_for_msr(
10890 msr_bitmap_l1, msr_bitmap_l0,
10891 MSR_IA32_SPEC_CTRL,
10892 MSR_TYPE_R | MSR_TYPE_W);
10895 nested_vmx_disable_intercept_for_msr(
10896 msr_bitmap_l1, msr_bitmap_l0,
10901 kvm_release_page_clean(page);
10906 static int nested_vmx_check_apic_access_controls(struct kvm_vcpu *vcpu,
10907 struct vmcs12 *vmcs12)
10909 if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES) &&
10910 !page_address_valid(vcpu, vmcs12->apic_access_addr))
10916 static int nested_vmx_check_apicv_controls(struct kvm_vcpu *vcpu,
10917 struct vmcs12 *vmcs12)
10919 if (!nested_cpu_has_virt_x2apic_mode(vmcs12) &&
10920 !nested_cpu_has_apic_reg_virt(vmcs12) &&
10921 !nested_cpu_has_vid(vmcs12) &&
10922 !nested_cpu_has_posted_intr(vmcs12))
10926 * If virtualize x2apic mode is enabled,
10927 * virtualize apic access must be disabled.
10929 if (nested_cpu_has_virt_x2apic_mode(vmcs12) &&
10930 nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
10934 * If virtual interrupt delivery is enabled,
10935 * we must exit on external interrupts.
10937 if (nested_cpu_has_vid(vmcs12) &&
10938 !nested_exit_on_intr(vcpu))
10942 * bits 15:8 should be zero in posted_intr_nv,
10943 * the descriptor address has been already checked
10944 * in nested_get_vmcs12_pages.
10946 if (nested_cpu_has_posted_intr(vmcs12) &&
10947 (!nested_cpu_has_vid(vmcs12) ||
10948 !nested_exit_intr_ack_set(vcpu) ||
10949 vmcs12->posted_intr_nv & 0xff00))
10952 /* tpr shadow is needed by all apicv features. */
10953 if (!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
10959 static int nested_vmx_check_msr_switch(struct kvm_vcpu *vcpu,
10960 unsigned long count_field,
10961 unsigned long addr_field)
10966 if (vmcs12_read_any(vcpu, count_field, &count) ||
10967 vmcs12_read_any(vcpu, addr_field, &addr)) {
10973 maxphyaddr = cpuid_maxphyaddr(vcpu);
10974 if (!IS_ALIGNED(addr, 16) || addr >> maxphyaddr ||
10975 (addr + count * sizeof(struct vmx_msr_entry) - 1) >> maxphyaddr) {
10976 pr_debug_ratelimited(
10977 "nVMX: invalid MSR switch (0x%lx, %d, %llu, 0x%08llx)",
10978 addr_field, maxphyaddr, count, addr);
10984 static int nested_vmx_check_msr_switch_controls(struct kvm_vcpu *vcpu,
10985 struct vmcs12 *vmcs12)
10987 if (vmcs12->vm_exit_msr_load_count == 0 &&
10988 vmcs12->vm_exit_msr_store_count == 0 &&
10989 vmcs12->vm_entry_msr_load_count == 0)
10990 return 0; /* Fast path */
10991 if (nested_vmx_check_msr_switch(vcpu, VM_EXIT_MSR_LOAD_COUNT,
10992 VM_EXIT_MSR_LOAD_ADDR) ||
10993 nested_vmx_check_msr_switch(vcpu, VM_EXIT_MSR_STORE_COUNT,
10994 VM_EXIT_MSR_STORE_ADDR) ||
10995 nested_vmx_check_msr_switch(vcpu, VM_ENTRY_MSR_LOAD_COUNT,
10996 VM_ENTRY_MSR_LOAD_ADDR))
11001 static int nested_vmx_check_pml_controls(struct kvm_vcpu *vcpu,
11002 struct vmcs12 *vmcs12)
11004 u64 address = vmcs12->pml_address;
11005 int maxphyaddr = cpuid_maxphyaddr(vcpu);
11007 if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_PML)) {
11008 if (!nested_cpu_has_ept(vmcs12) ||
11009 !IS_ALIGNED(address, 4096) ||
11010 address >> maxphyaddr)
11017 static int nested_vmx_msr_check_common(struct kvm_vcpu *vcpu,
11018 struct vmx_msr_entry *e)
11020 /* x2APIC MSR accesses are not allowed */
11021 if (vcpu->arch.apic_base & X2APIC_ENABLE && e->index >> 8 == 0x8)
11023 if (e->index == MSR_IA32_UCODE_WRITE || /* SDM Table 35-2 */
11024 e->index == MSR_IA32_UCODE_REV)
11026 if (e->reserved != 0)
11031 static int nested_vmx_load_msr_check(struct kvm_vcpu *vcpu,
11032 struct vmx_msr_entry *e)
11034 if (e->index == MSR_FS_BASE ||
11035 e->index == MSR_GS_BASE ||
11036 e->index == MSR_IA32_SMM_MONITOR_CTL || /* SMM is not supported */
11037 nested_vmx_msr_check_common(vcpu, e))
11042 static int nested_vmx_store_msr_check(struct kvm_vcpu *vcpu,
11043 struct vmx_msr_entry *e)
11045 if (e->index == MSR_IA32_SMBASE || /* SMM is not supported */
11046 nested_vmx_msr_check_common(vcpu, e))
11052 * Load guest's/host's msr at nested entry/exit.
11053 * return 0 for success, entry index for failure.
11055 static u32 nested_vmx_load_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
11058 struct vmx_msr_entry e;
11059 struct msr_data msr;
11061 msr.host_initiated = false;
11062 for (i = 0; i < count; i++) {
11063 if (kvm_vcpu_read_guest(vcpu, gpa + i * sizeof(e),
11065 pr_debug_ratelimited(
11066 "%s cannot read MSR entry (%u, 0x%08llx)\n",
11067 __func__, i, gpa + i * sizeof(e));
11070 if (nested_vmx_load_msr_check(vcpu, &e)) {
11071 pr_debug_ratelimited(
11072 "%s check failed (%u, 0x%x, 0x%x)\n",
11073 __func__, i, e.index, e.reserved);
11076 msr.index = e.index;
11077 msr.data = e.value;
11078 if (kvm_set_msr(vcpu, &msr)) {
11079 pr_debug_ratelimited(
11080 "%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
11081 __func__, i, e.index, e.value);
11090 static int nested_vmx_store_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
11093 struct vmx_msr_entry e;
11095 for (i = 0; i < count; i++) {
11096 struct msr_data msr_info;
11097 if (kvm_vcpu_read_guest(vcpu,
11098 gpa + i * sizeof(e),
11099 &e, 2 * sizeof(u32))) {
11100 pr_debug_ratelimited(
11101 "%s cannot read MSR entry (%u, 0x%08llx)\n",
11102 __func__, i, gpa + i * sizeof(e));
11105 if (nested_vmx_store_msr_check(vcpu, &e)) {
11106 pr_debug_ratelimited(
11107 "%s check failed (%u, 0x%x, 0x%x)\n",
11108 __func__, i, e.index, e.reserved);
11111 msr_info.host_initiated = false;
11112 msr_info.index = e.index;
11113 if (kvm_get_msr(vcpu, &msr_info)) {
11114 pr_debug_ratelimited(
11115 "%s cannot read MSR (%u, 0x%x)\n",
11116 __func__, i, e.index);
11119 if (kvm_vcpu_write_guest(vcpu,
11120 gpa + i * sizeof(e) +
11121 offsetof(struct vmx_msr_entry, value),
11122 &msr_info.data, sizeof(msr_info.data))) {
11123 pr_debug_ratelimited(
11124 "%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
11125 __func__, i, e.index, msr_info.data);
11132 static bool nested_cr3_valid(struct kvm_vcpu *vcpu, unsigned long val)
11134 unsigned long invalid_mask;
11136 invalid_mask = (~0ULL) << cpuid_maxphyaddr(vcpu);
11137 return (val & invalid_mask) == 0;
11141 * Load guest's/host's cr3 at nested entry/exit. nested_ept is true if we are
11142 * emulating VM entry into a guest with EPT enabled.
11143 * Returns 0 on success, 1 on failure. Invalid state exit qualification code
11144 * is assigned to entry_failure_code on failure.
11146 static int nested_vmx_load_cr3(struct kvm_vcpu *vcpu, unsigned long cr3, bool nested_ept,
11147 u32 *entry_failure_code)
11149 if (cr3 != kvm_read_cr3(vcpu) || (!nested_ept && pdptrs_changed(vcpu))) {
11150 if (!nested_cr3_valid(vcpu, cr3)) {
11151 *entry_failure_code = ENTRY_FAIL_DEFAULT;
11156 * If PAE paging and EPT are both on, CR3 is not used by the CPU and
11157 * must not be dereferenced.
11159 if (!is_long_mode(vcpu) && is_pae(vcpu) && is_paging(vcpu) &&
11161 if (!load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3)) {
11162 *entry_failure_code = ENTRY_FAIL_PDPTE;
11167 vcpu->arch.cr3 = cr3;
11168 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
11171 kvm_mmu_reset_context(vcpu);
11175 static void prepare_vmcs02_full(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
11177 struct vcpu_vmx *vmx = to_vmx(vcpu);
11179 vmcs_write16(GUEST_ES_SELECTOR, vmcs12->guest_es_selector);
11180 vmcs_write16(GUEST_SS_SELECTOR, vmcs12->guest_ss_selector);
11181 vmcs_write16(GUEST_DS_SELECTOR, vmcs12->guest_ds_selector);
11182 vmcs_write16(GUEST_FS_SELECTOR, vmcs12->guest_fs_selector);
11183 vmcs_write16(GUEST_GS_SELECTOR, vmcs12->guest_gs_selector);
11184 vmcs_write16(GUEST_LDTR_SELECTOR, vmcs12->guest_ldtr_selector);
11185 vmcs_write16(GUEST_TR_SELECTOR, vmcs12->guest_tr_selector);
11186 vmcs_write32(GUEST_ES_LIMIT, vmcs12->guest_es_limit);
11187 vmcs_write32(GUEST_SS_LIMIT, vmcs12->guest_ss_limit);
11188 vmcs_write32(GUEST_DS_LIMIT, vmcs12->guest_ds_limit);
11189 vmcs_write32(GUEST_FS_LIMIT, vmcs12->guest_fs_limit);
11190 vmcs_write32(GUEST_GS_LIMIT, vmcs12->guest_gs_limit);
11191 vmcs_write32(GUEST_LDTR_LIMIT, vmcs12->guest_ldtr_limit);
11192 vmcs_write32(GUEST_TR_LIMIT, vmcs12->guest_tr_limit);
11193 vmcs_write32(GUEST_GDTR_LIMIT, vmcs12->guest_gdtr_limit);
11194 vmcs_write32(GUEST_IDTR_LIMIT, vmcs12->guest_idtr_limit);
11195 vmcs_write32(GUEST_ES_AR_BYTES, vmcs12->guest_es_ar_bytes);
11196 vmcs_write32(GUEST_SS_AR_BYTES, vmcs12->guest_ss_ar_bytes);
11197 vmcs_write32(GUEST_DS_AR_BYTES, vmcs12->guest_ds_ar_bytes);
11198 vmcs_write32(GUEST_FS_AR_BYTES, vmcs12->guest_fs_ar_bytes);
11199 vmcs_write32(GUEST_GS_AR_BYTES, vmcs12->guest_gs_ar_bytes);
11200 vmcs_write32(GUEST_LDTR_AR_BYTES, vmcs12->guest_ldtr_ar_bytes);
11201 vmcs_write32(GUEST_TR_AR_BYTES, vmcs12->guest_tr_ar_bytes);
11202 vmcs_writel(GUEST_SS_BASE, vmcs12->guest_ss_base);
11203 vmcs_writel(GUEST_DS_BASE, vmcs12->guest_ds_base);
11204 vmcs_writel(GUEST_FS_BASE, vmcs12->guest_fs_base);
11205 vmcs_writel(GUEST_GS_BASE, vmcs12->guest_gs_base);
11206 vmcs_writel(GUEST_LDTR_BASE, vmcs12->guest_ldtr_base);
11207 vmcs_writel(GUEST_TR_BASE, vmcs12->guest_tr_base);
11208 vmcs_writel(GUEST_GDTR_BASE, vmcs12->guest_gdtr_base);
11209 vmcs_writel(GUEST_IDTR_BASE, vmcs12->guest_idtr_base);
11211 vmcs_write32(GUEST_SYSENTER_CS, vmcs12->guest_sysenter_cs);
11212 vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
11213 vmcs12->guest_pending_dbg_exceptions);
11214 vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->guest_sysenter_esp);
11215 vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->guest_sysenter_eip);
11217 if (nested_cpu_has_xsaves(vmcs12))
11218 vmcs_write64(XSS_EXIT_BITMAP, vmcs12->xss_exit_bitmap);
11219 vmcs_write64(VMCS_LINK_POINTER, -1ull);
11221 if (cpu_has_vmx_posted_intr())
11222 vmcs_write16(POSTED_INTR_NV, POSTED_INTR_NESTED_VECTOR);
11225 * Whether page-faults are trapped is determined by a combination of
11226 * 3 settings: PFEC_MASK, PFEC_MATCH and EXCEPTION_BITMAP.PF.
11227 * If enable_ept, L0 doesn't care about page faults and we should
11228 * set all of these to L1's desires. However, if !enable_ept, L0 does
11229 * care about (at least some) page faults, and because it is not easy
11230 * (if at all possible?) to merge L0 and L1's desires, we simply ask
11231 * to exit on each and every L2 page fault. This is done by setting
11232 * MASK=MATCH=0 and (see below) EB.PF=1.
11233 * Note that below we don't need special code to set EB.PF beyond the
11234 * "or"ing of the EB of vmcs01 and vmcs12, because when enable_ept,
11235 * vmcs01's EB.PF is 0 so the "or" will take vmcs12's value, and when
11236 * !enable_ept, EB.PF is 1, so the "or" will always be 1.
11238 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK,
11239 enable_ept ? vmcs12->page_fault_error_code_mask : 0);
11240 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH,
11241 enable_ept ? vmcs12->page_fault_error_code_match : 0);
11243 /* All VMFUNCs are currently emulated through L0 vmexits. */
11244 if (cpu_has_vmx_vmfunc())
11245 vmcs_write64(VM_FUNCTION_CONTROL, 0);
11247 if (cpu_has_vmx_apicv()) {
11248 vmcs_write64(EOI_EXIT_BITMAP0, vmcs12->eoi_exit_bitmap0);
11249 vmcs_write64(EOI_EXIT_BITMAP1, vmcs12->eoi_exit_bitmap1);
11250 vmcs_write64(EOI_EXIT_BITMAP2, vmcs12->eoi_exit_bitmap2);
11251 vmcs_write64(EOI_EXIT_BITMAP3, vmcs12->eoi_exit_bitmap3);
11255 * Set host-state according to L0's settings (vmcs12 is irrelevant here)
11256 * Some constant fields are set here by vmx_set_constant_host_state().
11257 * Other fields are different per CPU, and will be set later when
11258 * vmx_vcpu_load() is called, and when vmx_save_host_state() is called.
11260 vmx_set_constant_host_state(vmx);
11263 * Set the MSR load/store lists to match L0's settings.
11265 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
11266 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.nr);
11267 vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host));
11268 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.nr);
11269 vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest));
11271 set_cr4_guest_host_mask(vmx);
11273 if (vmx_mpx_supported())
11274 vmcs_write64(GUEST_BNDCFGS, vmcs12->guest_bndcfgs);
11277 if (nested_cpu_has_vpid(vmcs12) && vmx->nested.vpid02)
11278 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->nested.vpid02);
11280 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
11284 * L1 may access the L2's PDPTR, so save them to construct vmcs12
11287 vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
11288 vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
11289 vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
11290 vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
11293 if (cpu_has_vmx_msr_bitmap())
11294 vmcs_write64(MSR_BITMAP, __pa(vmx->nested.vmcs02.msr_bitmap));
11298 * prepare_vmcs02 is called when the L1 guest hypervisor runs its nested
11299 * L2 guest. L1 has a vmcs for L2 (vmcs12), and this function "merges" it
11300 * with L0's requirements for its guest (a.k.a. vmcs01), so we can run the L2
11301 * guest in a way that will both be appropriate to L1's requests, and our
11302 * needs. In addition to modifying the active vmcs (which is vmcs02), this
11303 * function also has additional necessary side-effects, like setting various
11304 * vcpu->arch fields.
11305 * Returns 0 on success, 1 on failure. Invalid state exit qualification code
11306 * is assigned to entry_failure_code on failure.
11308 static int prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
11309 u32 *entry_failure_code)
11311 struct vcpu_vmx *vmx = to_vmx(vcpu);
11312 u32 exec_control, vmcs12_exec_ctrl;
11314 if (vmx->nested.dirty_vmcs12) {
11315 prepare_vmcs02_full(vcpu, vmcs12);
11316 vmx->nested.dirty_vmcs12 = false;
11320 * First, the fields that are shadowed. This must be kept in sync
11321 * with vmx_shadow_fields.h.
11324 vmcs_write16(GUEST_CS_SELECTOR, vmcs12->guest_cs_selector);
11325 vmcs_write32(GUEST_CS_LIMIT, vmcs12->guest_cs_limit);
11326 vmcs_write32(GUEST_CS_AR_BYTES, vmcs12->guest_cs_ar_bytes);
11327 vmcs_writel(GUEST_ES_BASE, vmcs12->guest_es_base);
11328 vmcs_writel(GUEST_CS_BASE, vmcs12->guest_cs_base);
11331 * Not in vmcs02: GUEST_PML_INDEX, HOST_FS_SELECTOR, HOST_GS_SELECTOR,
11332 * HOST_FS_BASE, HOST_GS_BASE.
11335 if (vmx->nested.nested_run_pending &&
11336 (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS)) {
11337 kvm_set_dr(vcpu, 7, vmcs12->guest_dr7);
11338 vmcs_write64(GUEST_IA32_DEBUGCTL, vmcs12->guest_ia32_debugctl);
11340 kvm_set_dr(vcpu, 7, vcpu->arch.dr7);
11341 vmcs_write64(GUEST_IA32_DEBUGCTL, vmx->nested.vmcs01_debugctl);
11343 if (vmx->nested.nested_run_pending) {
11344 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
11345 vmcs12->vm_entry_intr_info_field);
11346 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
11347 vmcs12->vm_entry_exception_error_code);
11348 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
11349 vmcs12->vm_entry_instruction_len);
11350 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
11351 vmcs12->guest_interruptibility_info);
11352 vmx->loaded_vmcs->nmi_known_unmasked =
11353 !(vmcs12->guest_interruptibility_info & GUEST_INTR_STATE_NMI);
11355 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
11357 vmx_set_rflags(vcpu, vmcs12->guest_rflags);
11359 exec_control = vmcs12->pin_based_vm_exec_control;
11361 /* Preemption timer setting is only taken from vmcs01. */
11362 exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
11363 exec_control |= vmcs_config.pin_based_exec_ctrl;
11364 if (vmx->hv_deadline_tsc == -1)
11365 exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
11367 /* Posted interrupts setting is only taken from vmcs12. */
11368 if (nested_cpu_has_posted_intr(vmcs12)) {
11369 vmx->nested.posted_intr_nv = vmcs12->posted_intr_nv;
11370 vmx->nested.pi_pending = false;
11372 exec_control &= ~PIN_BASED_POSTED_INTR;
11375 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, exec_control);
11377 vmx->nested.preemption_timer_expired = false;
11378 if (nested_cpu_has_preemption_timer(vmcs12))
11379 vmx_start_preemption_timer(vcpu);
11381 if (cpu_has_secondary_exec_ctrls()) {
11382 exec_control = vmx->secondary_exec_control;
11384 /* Take the following fields only from vmcs12 */
11385 exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
11386 SECONDARY_EXEC_ENABLE_INVPCID |
11387 SECONDARY_EXEC_RDTSCP |
11388 SECONDARY_EXEC_XSAVES |
11389 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
11390 SECONDARY_EXEC_APIC_REGISTER_VIRT |
11391 SECONDARY_EXEC_ENABLE_VMFUNC);
11392 if (nested_cpu_has(vmcs12,
11393 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS)) {
11394 vmcs12_exec_ctrl = vmcs12->secondary_vm_exec_control &
11395 ~SECONDARY_EXEC_ENABLE_PML;
11396 exec_control |= vmcs12_exec_ctrl;
11399 if (exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
11400 vmcs_write16(GUEST_INTR_STATUS,
11401 vmcs12->guest_intr_status);
11404 * Write an illegal value to APIC_ACCESS_ADDR. Later,
11405 * nested_get_vmcs12_pages will either fix it up or
11406 * remove the VM execution control.
11408 if (exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)
11409 vmcs_write64(APIC_ACCESS_ADDR, -1ull);
11411 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
11415 * HOST_RSP is normally set correctly in vmx_vcpu_run() just before
11416 * entry, but only if the current (host) sp changed from the value
11417 * we wrote last (vmx->host_rsp). This cache is no longer relevant
11418 * if we switch vmcs, and rather than hold a separate cache per vmcs,
11419 * here we just force the write to happen on entry.
11423 exec_control = vmx_exec_control(vmx); /* L0's desires */
11424 exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
11425 exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
11426 exec_control &= ~CPU_BASED_TPR_SHADOW;
11427 exec_control |= vmcs12->cpu_based_vm_exec_control;
11430 * Write an illegal value to VIRTUAL_APIC_PAGE_ADDR. Later, if
11431 * nested_get_vmcs12_pages can't fix it up, the illegal value
11432 * will result in a VM entry failure.
11434 if (exec_control & CPU_BASED_TPR_SHADOW) {
11435 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, -1ull);
11436 vmcs_write32(TPR_THRESHOLD, vmcs12->tpr_threshold);
11438 #ifdef CONFIG_X86_64
11439 exec_control |= CPU_BASED_CR8_LOAD_EXITING |
11440 CPU_BASED_CR8_STORE_EXITING;
11445 * A vmexit (to either L1 hypervisor or L0 userspace) is always needed
11446 * for I/O port accesses.
11448 exec_control &= ~CPU_BASED_USE_IO_BITMAPS;
11449 exec_control |= CPU_BASED_UNCOND_IO_EXITING;
11451 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
11453 /* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the
11454 * bitwise-or of what L1 wants to trap for L2, and what we want to
11455 * trap. Note that CR0.TS also needs updating - we do this later.
11457 update_exception_bitmap(vcpu);
11458 vcpu->arch.cr0_guest_owned_bits &= ~vmcs12->cr0_guest_host_mask;
11459 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
11461 /* L2->L1 exit controls are emulated - the hardware exit is to L0 so
11462 * we should use its exit controls. Note that VM_EXIT_LOAD_IA32_EFER
11463 * bits are further modified by vmx_set_efer() below.
11465 vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
11467 /* vmcs12's VM_ENTRY_LOAD_IA32_EFER and VM_ENTRY_IA32E_MODE are
11468 * emulated by vmx_set_efer(), below.
11470 vm_entry_controls_init(vmx,
11471 (vmcs12->vm_entry_controls & ~VM_ENTRY_LOAD_IA32_EFER &
11472 ~VM_ENTRY_IA32E_MODE) |
11473 (vmcs_config.vmentry_ctrl & ~VM_ENTRY_IA32E_MODE));
11475 if (vmx->nested.nested_run_pending &&
11476 (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT)) {
11477 vmcs_write64(GUEST_IA32_PAT, vmcs12->guest_ia32_pat);
11478 vcpu->arch.pat = vmcs12->guest_ia32_pat;
11479 } else if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
11480 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
11483 vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
11485 if (kvm_has_tsc_control)
11486 decache_tsc_multiplier(vmx);
11490 * There is no direct mapping between vpid02 and vpid12, the
11491 * vpid02 is per-vCPU for L0 and reused while the value of
11492 * vpid12 is changed w/ one invvpid during nested vmentry.
11493 * The vpid12 is allocated by L1 for L2, so it will not
11494 * influence global bitmap(for vpid01 and vpid02 allocation)
11495 * even if spawn a lot of nested vCPUs.
11497 if (nested_cpu_has_vpid(vmcs12) && vmx->nested.vpid02) {
11498 if (vmcs12->virtual_processor_id != vmx->nested.last_vpid) {
11499 vmx->nested.last_vpid = vmcs12->virtual_processor_id;
11500 __vmx_flush_tlb(vcpu, vmx->nested.vpid02, true);
11503 vmx_flush_tlb(vcpu, true);
11509 * Conceptually we want to copy the PML address and index from
11510 * vmcs01 here, and then back to vmcs01 on nested vmexit. But,
11511 * since we always flush the log on each vmexit, this happens
11512 * to be equivalent to simply resetting the fields in vmcs02.
11514 ASSERT(vmx->pml_pg);
11515 vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
11516 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
11519 if (nested_cpu_has_ept(vmcs12)) {
11520 if (nested_ept_init_mmu_context(vcpu)) {
11521 *entry_failure_code = ENTRY_FAIL_DEFAULT;
11524 } else if (nested_cpu_has2(vmcs12,
11525 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
11526 vmx_flush_tlb(vcpu, true);
11530 * This sets GUEST_CR0 to vmcs12->guest_cr0, possibly modifying those
11531 * bits which we consider mandatory enabled.
11532 * The CR0_READ_SHADOW is what L2 should have expected to read given
11533 * the specifications by L1; It's not enough to take
11534 * vmcs12->cr0_read_shadow because on our cr0_guest_host_mask we we
11535 * have more bits than L1 expected.
11537 vmx_set_cr0(vcpu, vmcs12->guest_cr0);
11538 vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
11540 vmx_set_cr4(vcpu, vmcs12->guest_cr4);
11541 vmcs_writel(CR4_READ_SHADOW, nested_read_cr4(vmcs12));
11543 if (vmx->nested.nested_run_pending &&
11544 (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER))
11545 vcpu->arch.efer = vmcs12->guest_ia32_efer;
11546 else if (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE)
11547 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
11549 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
11550 /* Note: modifies VM_ENTRY/EXIT_CONTROLS and GUEST/HOST_IA32_EFER */
11551 vmx_set_efer(vcpu, vcpu->arch.efer);
11554 * Guest state is invalid and unrestricted guest is disabled,
11555 * which means L1 attempted VMEntry to L2 with invalid state.
11556 * Fail the VMEntry.
11558 if (vmx->emulation_required) {
11559 *entry_failure_code = ENTRY_FAIL_DEFAULT;
11563 /* Shadow page tables on either EPT or shadow page tables. */
11564 if (nested_vmx_load_cr3(vcpu, vmcs12->guest_cr3, nested_cpu_has_ept(vmcs12),
11565 entry_failure_code))
11569 vcpu->arch.walk_mmu->inject_page_fault = vmx_inject_page_fault_nested;
11571 kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->guest_rsp);
11572 kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->guest_rip);
11576 static int nested_vmx_check_nmi_controls(struct vmcs12 *vmcs12)
11578 if (!nested_cpu_has_nmi_exiting(vmcs12) &&
11579 nested_cpu_has_virtual_nmis(vmcs12))
11582 if (!nested_cpu_has_virtual_nmis(vmcs12) &&
11583 nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_NMI_PENDING))
11589 static int check_vmentry_prereqs(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
11591 struct vcpu_vmx *vmx = to_vmx(vcpu);
11593 if (vmcs12->guest_activity_state != GUEST_ACTIVITY_ACTIVE &&
11594 vmcs12->guest_activity_state != GUEST_ACTIVITY_HLT)
11595 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
11597 if (nested_vmx_check_io_bitmap_controls(vcpu, vmcs12))
11598 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
11600 if (nested_vmx_check_msr_bitmap_controls(vcpu, vmcs12))
11601 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
11603 if (nested_vmx_check_apic_access_controls(vcpu, vmcs12))
11604 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
11606 if (nested_vmx_check_tpr_shadow_controls(vcpu, vmcs12))
11607 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
11609 if (nested_vmx_check_apicv_controls(vcpu, vmcs12))
11610 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
11612 if (nested_vmx_check_msr_switch_controls(vcpu, vmcs12))
11613 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
11615 if (nested_vmx_check_pml_controls(vcpu, vmcs12))
11616 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
11618 if (!vmx_control_verify(vmcs12->cpu_based_vm_exec_control,
11619 vmx->nested.msrs.procbased_ctls_low,
11620 vmx->nested.msrs.procbased_ctls_high) ||
11621 (nested_cpu_has(vmcs12, CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
11622 !vmx_control_verify(vmcs12->secondary_vm_exec_control,
11623 vmx->nested.msrs.secondary_ctls_low,
11624 vmx->nested.msrs.secondary_ctls_high)) ||
11625 !vmx_control_verify(vmcs12->pin_based_vm_exec_control,
11626 vmx->nested.msrs.pinbased_ctls_low,
11627 vmx->nested.msrs.pinbased_ctls_high) ||
11628 !vmx_control_verify(vmcs12->vm_exit_controls,
11629 vmx->nested.msrs.exit_ctls_low,
11630 vmx->nested.msrs.exit_ctls_high) ||
11631 !vmx_control_verify(vmcs12->vm_entry_controls,
11632 vmx->nested.msrs.entry_ctls_low,
11633 vmx->nested.msrs.entry_ctls_high))
11634 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
11636 if (nested_vmx_check_nmi_controls(vmcs12))
11637 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
11639 if (nested_cpu_has_vmfunc(vmcs12)) {
11640 if (vmcs12->vm_function_control &
11641 ~vmx->nested.msrs.vmfunc_controls)
11642 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
11644 if (nested_cpu_has_eptp_switching(vmcs12)) {
11645 if (!nested_cpu_has_ept(vmcs12) ||
11646 !page_address_valid(vcpu, vmcs12->eptp_list_address))
11647 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
11651 if (vmcs12->cr3_target_count > nested_cpu_vmx_misc_cr3_count(vcpu))
11652 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
11654 if (!nested_host_cr0_valid(vcpu, vmcs12->host_cr0) ||
11655 !nested_host_cr4_valid(vcpu, vmcs12->host_cr4) ||
11656 !nested_cr3_valid(vcpu, vmcs12->host_cr3))
11657 return VMXERR_ENTRY_INVALID_HOST_STATE_FIELD;
11660 * From the Intel SDM, volume 3:
11661 * Fields relevant to VM-entry event injection must be set properly.
11662 * These fields are the VM-entry interruption-information field, the
11663 * VM-entry exception error code, and the VM-entry instruction length.
11665 if (vmcs12->vm_entry_intr_info_field & INTR_INFO_VALID_MASK) {
11666 u32 intr_info = vmcs12->vm_entry_intr_info_field;
11667 u8 vector = intr_info & INTR_INFO_VECTOR_MASK;
11668 u32 intr_type = intr_info & INTR_INFO_INTR_TYPE_MASK;
11669 bool has_error_code = intr_info & INTR_INFO_DELIVER_CODE_MASK;
11670 bool should_have_error_code;
11671 bool urg = nested_cpu_has2(vmcs12,
11672 SECONDARY_EXEC_UNRESTRICTED_GUEST);
11673 bool prot_mode = !urg || vmcs12->guest_cr0 & X86_CR0_PE;
11675 /* VM-entry interruption-info field: interruption type */
11676 if (intr_type == INTR_TYPE_RESERVED ||
11677 (intr_type == INTR_TYPE_OTHER_EVENT &&
11678 !nested_cpu_supports_monitor_trap_flag(vcpu)))
11679 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
11681 /* VM-entry interruption-info field: vector */
11682 if ((intr_type == INTR_TYPE_NMI_INTR && vector != NMI_VECTOR) ||
11683 (intr_type == INTR_TYPE_HARD_EXCEPTION && vector > 31) ||
11684 (intr_type == INTR_TYPE_OTHER_EVENT && vector != 0))
11685 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
11687 /* VM-entry interruption-info field: deliver error code */
11688 should_have_error_code =
11689 intr_type == INTR_TYPE_HARD_EXCEPTION && prot_mode &&
11690 x86_exception_has_error_code(vector);
11691 if (has_error_code != should_have_error_code)
11692 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
11694 /* VM-entry exception error code */
11695 if (has_error_code &&
11696 vmcs12->vm_entry_exception_error_code & GENMASK(31, 15))
11697 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
11699 /* VM-entry interruption-info field: reserved bits */
11700 if (intr_info & INTR_INFO_RESVD_BITS_MASK)
11701 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
11703 /* VM-entry instruction length */
11704 switch (intr_type) {
11705 case INTR_TYPE_SOFT_EXCEPTION:
11706 case INTR_TYPE_SOFT_INTR:
11707 case INTR_TYPE_PRIV_SW_EXCEPTION:
11708 if ((vmcs12->vm_entry_instruction_len > 15) ||
11709 (vmcs12->vm_entry_instruction_len == 0 &&
11710 !nested_cpu_has_zero_length_injection(vcpu)))
11711 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
11718 static int check_vmentry_postreqs(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
11723 *exit_qual = ENTRY_FAIL_DEFAULT;
11725 if (!nested_guest_cr0_valid(vcpu, vmcs12->guest_cr0) ||
11726 !nested_guest_cr4_valid(vcpu, vmcs12->guest_cr4))
11729 if (!nested_cpu_has2(vmcs12, SECONDARY_EXEC_SHADOW_VMCS) &&
11730 vmcs12->vmcs_link_pointer != -1ull) {
11731 *exit_qual = ENTRY_FAIL_VMCS_LINK_PTR;
11736 * If the load IA32_EFER VM-entry control is 1, the following checks
11737 * are performed on the field for the IA32_EFER MSR:
11738 * - Bits reserved in the IA32_EFER MSR must be 0.
11739 * - Bit 10 (corresponding to IA32_EFER.LMA) must equal the value of
11740 * the IA-32e mode guest VM-exit control. It must also be identical
11741 * to bit 8 (LME) if bit 31 in the CR0 field (corresponding to
11744 if (to_vmx(vcpu)->nested.nested_run_pending &&
11745 (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER)) {
11746 ia32e = (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) != 0;
11747 if (!kvm_valid_efer(vcpu, vmcs12->guest_ia32_efer) ||
11748 ia32e != !!(vmcs12->guest_ia32_efer & EFER_LMA) ||
11749 ((vmcs12->guest_cr0 & X86_CR0_PG) &&
11750 ia32e != !!(vmcs12->guest_ia32_efer & EFER_LME)))
11755 * If the load IA32_EFER VM-exit control is 1, bits reserved in the
11756 * IA32_EFER MSR must be 0 in the field for that register. In addition,
11757 * the values of the LMA and LME bits in the field must each be that of
11758 * the host address-space size VM-exit control.
11760 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER) {
11761 ia32e = (vmcs12->vm_exit_controls &
11762 VM_EXIT_HOST_ADDR_SPACE_SIZE) != 0;
11763 if (!kvm_valid_efer(vcpu, vmcs12->host_ia32_efer) ||
11764 ia32e != !!(vmcs12->host_ia32_efer & EFER_LMA) ||
11765 ia32e != !!(vmcs12->host_ia32_efer & EFER_LME))
11769 if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS) &&
11770 (is_noncanonical_address(vmcs12->guest_bndcfgs & PAGE_MASK, vcpu) ||
11771 (vmcs12->guest_bndcfgs & MSR_IA32_BNDCFGS_RSVD)))
11778 * If exit_qual is NULL, this is being called from RSM.
11779 * Otherwise it's called from vmlaunch/vmresume.
11781 static int enter_vmx_non_root_mode(struct kvm_vcpu *vcpu, u32 *exit_qual)
11783 struct vcpu_vmx *vmx = to_vmx(vcpu);
11784 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
11785 bool from_vmentry = !!exit_qual;
11786 u32 dummy_exit_qual;
11789 enter_guest_mode(vcpu);
11791 if (!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS))
11792 vmx->nested.vmcs01_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
11794 vmx_switch_vmcs(vcpu, &vmx->nested.vmcs02);
11795 vmx_segment_cache_clear(vmx);
11797 if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING)
11798 vcpu->arch.tsc_offset += vmcs12->tsc_offset;
11800 r = EXIT_REASON_INVALID_STATE;
11801 if (prepare_vmcs02(vcpu, vmcs12, from_vmentry ? exit_qual : &dummy_exit_qual))
11804 if (from_vmentry) {
11805 nested_get_vmcs12_pages(vcpu);
11807 r = EXIT_REASON_MSR_LOAD_FAIL;
11808 *exit_qual = nested_vmx_load_msr(vcpu,
11809 vmcs12->vm_entry_msr_load_addr,
11810 vmcs12->vm_entry_msr_load_count);
11815 * The MMU is not initialized to point at the right entities yet and
11816 * "get pages" would need to read data from the guest (i.e. we will
11817 * need to perform gpa to hpa translation). Request a call
11818 * to nested_get_vmcs12_pages before the next VM-entry. The MSRs
11819 * have already been set at vmentry time and should not be reset.
11821 kvm_make_request(KVM_REQ_GET_VMCS12_PAGES, vcpu);
11825 * Note no nested_vmx_succeed or nested_vmx_fail here. At this point
11826 * we are no longer running L1, and VMLAUNCH/VMRESUME has not yet
11827 * returned as far as L1 is concerned. It will only return (and set
11828 * the success flag) when L2 exits (see nested_vmx_vmexit()).
11833 if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING)
11834 vcpu->arch.tsc_offset -= vmcs12->tsc_offset;
11835 leave_guest_mode(vcpu);
11836 vmx_switch_vmcs(vcpu, &vmx->vmcs01);
11841 * nested_vmx_run() handles a nested entry, i.e., a VMLAUNCH or VMRESUME on L1
11842 * for running an L2 nested guest.
11844 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
11846 struct vmcs12 *vmcs12;
11847 struct vcpu_vmx *vmx = to_vmx(vcpu);
11848 u32 interrupt_shadow = vmx_get_interrupt_shadow(vcpu);
11852 if (!nested_vmx_check_permission(vcpu))
11855 if (!nested_vmx_check_vmcs12(vcpu))
11858 vmcs12 = get_vmcs12(vcpu);
11860 if (enable_shadow_vmcs)
11861 copy_shadow_to_vmcs12(vmx);
11864 * The nested entry process starts with enforcing various prerequisites
11865 * on vmcs12 as required by the Intel SDM, and act appropriately when
11866 * they fail: As the SDM explains, some conditions should cause the
11867 * instruction to fail, while others will cause the instruction to seem
11868 * to succeed, but return an EXIT_REASON_INVALID_STATE.
11869 * To speed up the normal (success) code path, we should avoid checking
11870 * for misconfigurations which will anyway be caught by the processor
11871 * when using the merged vmcs02.
11873 if (interrupt_shadow & KVM_X86_SHADOW_INT_MOV_SS) {
11874 nested_vmx_failValid(vcpu,
11875 VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS);
11879 if (vmcs12->launch_state == launch) {
11880 nested_vmx_failValid(vcpu,
11881 launch ? VMXERR_VMLAUNCH_NONCLEAR_VMCS
11882 : VMXERR_VMRESUME_NONLAUNCHED_VMCS);
11886 ret = check_vmentry_prereqs(vcpu, vmcs12);
11888 nested_vmx_failValid(vcpu, ret);
11893 * After this point, the trap flag no longer triggers a singlestep trap
11894 * on the vm entry instructions; don't call kvm_skip_emulated_instruction.
11895 * This is not 100% correct; for performance reasons, we delegate most
11896 * of the checks on host state to the processor. If those fail,
11897 * the singlestep trap is missed.
11899 skip_emulated_instruction(vcpu);
11901 ret = check_vmentry_postreqs(vcpu, vmcs12, &exit_qual);
11903 nested_vmx_entry_failure(vcpu, vmcs12,
11904 EXIT_REASON_INVALID_STATE, exit_qual);
11909 * We're finally done with prerequisite checking, and can start with
11910 * the nested entry.
11913 vmx->nested.nested_run_pending = 1;
11914 ret = enter_vmx_non_root_mode(vcpu, &exit_qual);
11916 nested_vmx_entry_failure(vcpu, vmcs12, ret, exit_qual);
11917 vmx->nested.nested_run_pending = 0;
11922 * If we're entering a halted L2 vcpu and the L2 vcpu won't be woken
11923 * by event injection, halt vcpu.
11925 if ((vmcs12->guest_activity_state == GUEST_ACTIVITY_HLT) &&
11926 !(vmcs12->vm_entry_intr_info_field & INTR_INFO_VALID_MASK)) {
11927 vmx->nested.nested_run_pending = 0;
11928 return kvm_vcpu_halt(vcpu);
11933 return kvm_skip_emulated_instruction(vcpu);
11937 * On a nested exit from L2 to L1, vmcs12.guest_cr0 might not be up-to-date
11938 * because L2 may have changed some cr0 bits directly (CRO_GUEST_HOST_MASK).
11939 * This function returns the new value we should put in vmcs12.guest_cr0.
11940 * It's not enough to just return the vmcs02 GUEST_CR0. Rather,
11941 * 1. Bits that neither L0 nor L1 trapped, were set directly by L2 and are now
11942 * available in vmcs02 GUEST_CR0. (Note: It's enough to check that L0
11943 * didn't trap the bit, because if L1 did, so would L0).
11944 * 2. Bits that L1 asked to trap (and therefore L0 also did) could not have
11945 * been modified by L2, and L1 knows it. So just leave the old value of
11946 * the bit from vmcs12.guest_cr0. Note that the bit from vmcs02 GUEST_CR0
11947 * isn't relevant, because if L0 traps this bit it can set it to anything.
11948 * 3. Bits that L1 didn't trap, but L0 did. L1 believes the guest could have
11949 * changed these bits, and therefore they need to be updated, but L0
11950 * didn't necessarily allow them to be changed in GUEST_CR0 - and rather
11951 * put them in vmcs02 CR0_READ_SHADOW. So take these bits from there.
11953 static inline unsigned long
11954 vmcs12_guest_cr0(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
11957 /*1*/ (vmcs_readl(GUEST_CR0) & vcpu->arch.cr0_guest_owned_bits) |
11958 /*2*/ (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask) |
11959 /*3*/ (vmcs_readl(CR0_READ_SHADOW) & ~(vmcs12->cr0_guest_host_mask |
11960 vcpu->arch.cr0_guest_owned_bits));
11963 static inline unsigned long
11964 vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
11967 /*1*/ (vmcs_readl(GUEST_CR4) & vcpu->arch.cr4_guest_owned_bits) |
11968 /*2*/ (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask) |
11969 /*3*/ (vmcs_readl(CR4_READ_SHADOW) & ~(vmcs12->cr4_guest_host_mask |
11970 vcpu->arch.cr4_guest_owned_bits));
11973 static void vmcs12_save_pending_event(struct kvm_vcpu *vcpu,
11974 struct vmcs12 *vmcs12)
11979 if (vcpu->arch.exception.injected) {
11980 nr = vcpu->arch.exception.nr;
11981 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
11983 if (kvm_exception_is_soft(nr)) {
11984 vmcs12->vm_exit_instruction_len =
11985 vcpu->arch.event_exit_inst_len;
11986 idt_vectoring |= INTR_TYPE_SOFT_EXCEPTION;
11988 idt_vectoring |= INTR_TYPE_HARD_EXCEPTION;
11990 if (vcpu->arch.exception.has_error_code) {
11991 idt_vectoring |= VECTORING_INFO_DELIVER_CODE_MASK;
11992 vmcs12->idt_vectoring_error_code =
11993 vcpu->arch.exception.error_code;
11996 vmcs12->idt_vectoring_info_field = idt_vectoring;
11997 } else if (vcpu->arch.nmi_injected) {
11998 vmcs12->idt_vectoring_info_field =
11999 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR;
12000 } else if (vcpu->arch.interrupt.injected) {
12001 nr = vcpu->arch.interrupt.nr;
12002 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
12004 if (vcpu->arch.interrupt.soft) {
12005 idt_vectoring |= INTR_TYPE_SOFT_INTR;
12006 vmcs12->vm_entry_instruction_len =
12007 vcpu->arch.event_exit_inst_len;
12009 idt_vectoring |= INTR_TYPE_EXT_INTR;
12011 vmcs12->idt_vectoring_info_field = idt_vectoring;
12015 static int vmx_check_nested_events(struct kvm_vcpu *vcpu, bool external_intr)
12017 struct vcpu_vmx *vmx = to_vmx(vcpu);
12018 unsigned long exit_qual;
12019 bool block_nested_events =
12020 vmx->nested.nested_run_pending || kvm_event_needs_reinjection(vcpu);
12022 if (vcpu->arch.exception.pending &&
12023 nested_vmx_check_exception(vcpu, &exit_qual)) {
12024 if (block_nested_events)
12026 nested_vmx_inject_exception_vmexit(vcpu, exit_qual);
12030 if (nested_cpu_has_preemption_timer(get_vmcs12(vcpu)) &&
12031 vmx->nested.preemption_timer_expired) {
12032 if (block_nested_events)
12034 nested_vmx_vmexit(vcpu, EXIT_REASON_PREEMPTION_TIMER, 0, 0);
12038 if (vcpu->arch.nmi_pending && nested_exit_on_nmi(vcpu)) {
12039 if (block_nested_events)
12041 nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
12042 NMI_VECTOR | INTR_TYPE_NMI_INTR |
12043 INTR_INFO_VALID_MASK, 0);
12045 * The NMI-triggered VM exit counts as injection:
12046 * clear this one and block further NMIs.
12048 vcpu->arch.nmi_pending = 0;
12049 vmx_set_nmi_mask(vcpu, true);
12053 if ((kvm_cpu_has_interrupt(vcpu) || external_intr) &&
12054 nested_exit_on_intr(vcpu)) {
12055 if (block_nested_events)
12057 nested_vmx_vmexit(vcpu, EXIT_REASON_EXTERNAL_INTERRUPT, 0, 0);
12061 vmx_complete_nested_posted_interrupt(vcpu);
12065 static u32 vmx_get_preemption_timer_value(struct kvm_vcpu *vcpu)
12067 ktime_t remaining =
12068 hrtimer_get_remaining(&to_vmx(vcpu)->nested.preemption_timer);
12071 if (ktime_to_ns(remaining) <= 0)
12074 value = ktime_to_ns(remaining) * vcpu->arch.virtual_tsc_khz;
12075 do_div(value, 1000000);
12076 return value >> VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
12080 * Update the guest state fields of vmcs12 to reflect changes that
12081 * occurred while L2 was running. (The "IA-32e mode guest" bit of the
12082 * VM-entry controls is also updated, since this is really a guest
12085 static void sync_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
12087 vmcs12->guest_cr0 = vmcs12_guest_cr0(vcpu, vmcs12);
12088 vmcs12->guest_cr4 = vmcs12_guest_cr4(vcpu, vmcs12);
12090 vmcs12->guest_rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
12091 vmcs12->guest_rip = kvm_register_read(vcpu, VCPU_REGS_RIP);
12092 vmcs12->guest_rflags = vmcs_readl(GUEST_RFLAGS);
12094 vmcs12->guest_es_selector = vmcs_read16(GUEST_ES_SELECTOR);
12095 vmcs12->guest_cs_selector = vmcs_read16(GUEST_CS_SELECTOR);
12096 vmcs12->guest_ss_selector = vmcs_read16(GUEST_SS_SELECTOR);
12097 vmcs12->guest_ds_selector = vmcs_read16(GUEST_DS_SELECTOR);
12098 vmcs12->guest_fs_selector = vmcs_read16(GUEST_FS_SELECTOR);
12099 vmcs12->guest_gs_selector = vmcs_read16(GUEST_GS_SELECTOR);
12100 vmcs12->guest_ldtr_selector = vmcs_read16(GUEST_LDTR_SELECTOR);
12101 vmcs12->guest_tr_selector = vmcs_read16(GUEST_TR_SELECTOR);
12102 vmcs12->guest_es_limit = vmcs_read32(GUEST_ES_LIMIT);
12103 vmcs12->guest_cs_limit = vmcs_read32(GUEST_CS_LIMIT);
12104 vmcs12->guest_ss_limit = vmcs_read32(GUEST_SS_LIMIT);
12105 vmcs12->guest_ds_limit = vmcs_read32(GUEST_DS_LIMIT);
12106 vmcs12->guest_fs_limit = vmcs_read32(GUEST_FS_LIMIT);
12107 vmcs12->guest_gs_limit = vmcs_read32(GUEST_GS_LIMIT);
12108 vmcs12->guest_ldtr_limit = vmcs_read32(GUEST_LDTR_LIMIT);
12109 vmcs12->guest_tr_limit = vmcs_read32(GUEST_TR_LIMIT);
12110 vmcs12->guest_gdtr_limit = vmcs_read32(GUEST_GDTR_LIMIT);
12111 vmcs12->guest_idtr_limit = vmcs_read32(GUEST_IDTR_LIMIT);
12112 vmcs12->guest_es_ar_bytes = vmcs_read32(GUEST_ES_AR_BYTES);
12113 vmcs12->guest_cs_ar_bytes = vmcs_read32(GUEST_CS_AR_BYTES);
12114 vmcs12->guest_ss_ar_bytes = vmcs_read32(GUEST_SS_AR_BYTES);
12115 vmcs12->guest_ds_ar_bytes = vmcs_read32(GUEST_DS_AR_BYTES);
12116 vmcs12->guest_fs_ar_bytes = vmcs_read32(GUEST_FS_AR_BYTES);
12117 vmcs12->guest_gs_ar_bytes = vmcs_read32(GUEST_GS_AR_BYTES);
12118 vmcs12->guest_ldtr_ar_bytes = vmcs_read32(GUEST_LDTR_AR_BYTES);
12119 vmcs12->guest_tr_ar_bytes = vmcs_read32(GUEST_TR_AR_BYTES);
12120 vmcs12->guest_es_base = vmcs_readl(GUEST_ES_BASE);
12121 vmcs12->guest_cs_base = vmcs_readl(GUEST_CS_BASE);
12122 vmcs12->guest_ss_base = vmcs_readl(GUEST_SS_BASE);
12123 vmcs12->guest_ds_base = vmcs_readl(GUEST_DS_BASE);
12124 vmcs12->guest_fs_base = vmcs_readl(GUEST_FS_BASE);
12125 vmcs12->guest_gs_base = vmcs_readl(GUEST_GS_BASE);
12126 vmcs12->guest_ldtr_base = vmcs_readl(GUEST_LDTR_BASE);
12127 vmcs12->guest_tr_base = vmcs_readl(GUEST_TR_BASE);
12128 vmcs12->guest_gdtr_base = vmcs_readl(GUEST_GDTR_BASE);
12129 vmcs12->guest_idtr_base = vmcs_readl(GUEST_IDTR_BASE);
12131 vmcs12->guest_interruptibility_info =
12132 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
12133 vmcs12->guest_pending_dbg_exceptions =
12134 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS);
12135 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
12136 vmcs12->guest_activity_state = GUEST_ACTIVITY_HLT;
12138 vmcs12->guest_activity_state = GUEST_ACTIVITY_ACTIVE;
12140 if (nested_cpu_has_preemption_timer(vmcs12)) {
12141 if (vmcs12->vm_exit_controls &
12142 VM_EXIT_SAVE_VMX_PREEMPTION_TIMER)
12143 vmcs12->vmx_preemption_timer_value =
12144 vmx_get_preemption_timer_value(vcpu);
12145 hrtimer_cancel(&to_vmx(vcpu)->nested.preemption_timer);
12149 * In some cases (usually, nested EPT), L2 is allowed to change its
12150 * own CR3 without exiting. If it has changed it, we must keep it.
12151 * Of course, if L0 is using shadow page tables, GUEST_CR3 was defined
12152 * by L0, not L1 or L2, so we mustn't unconditionally copy it to vmcs12.
12154 * Additionally, restore L2's PDPTR to vmcs12.
12157 vmcs12->guest_cr3 = vmcs_readl(GUEST_CR3);
12158 vmcs12->guest_pdptr0 = vmcs_read64(GUEST_PDPTR0);
12159 vmcs12->guest_pdptr1 = vmcs_read64(GUEST_PDPTR1);
12160 vmcs12->guest_pdptr2 = vmcs_read64(GUEST_PDPTR2);
12161 vmcs12->guest_pdptr3 = vmcs_read64(GUEST_PDPTR3);
12164 vmcs12->guest_linear_address = vmcs_readl(GUEST_LINEAR_ADDRESS);
12166 if (nested_cpu_has_vid(vmcs12))
12167 vmcs12->guest_intr_status = vmcs_read16(GUEST_INTR_STATUS);
12169 vmcs12->vm_entry_controls =
12170 (vmcs12->vm_entry_controls & ~VM_ENTRY_IA32E_MODE) |
12171 (vm_entry_controls_get(to_vmx(vcpu)) & VM_ENTRY_IA32E_MODE);
12173 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_DEBUG_CONTROLS) {
12174 kvm_get_dr(vcpu, 7, (unsigned long *)&vmcs12->guest_dr7);
12175 vmcs12->guest_ia32_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
12178 /* TODO: These cannot have changed unless we have MSR bitmaps and
12179 * the relevant bit asks not to trap the change */
12180 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_PAT)
12181 vmcs12->guest_ia32_pat = vmcs_read64(GUEST_IA32_PAT);
12182 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_EFER)
12183 vmcs12->guest_ia32_efer = vcpu->arch.efer;
12184 vmcs12->guest_sysenter_cs = vmcs_read32(GUEST_SYSENTER_CS);
12185 vmcs12->guest_sysenter_esp = vmcs_readl(GUEST_SYSENTER_ESP);
12186 vmcs12->guest_sysenter_eip = vmcs_readl(GUEST_SYSENTER_EIP);
12187 if (kvm_mpx_supported())
12188 vmcs12->guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
12192 * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
12193 * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),
12194 * and this function updates it to reflect the changes to the guest state while
12195 * L2 was running (and perhaps made some exits which were handled directly by L0
12196 * without going back to L1), and to reflect the exit reason.
12197 * Note that we do not have to copy here all VMCS fields, just those that
12198 * could have changed by the L2 guest or the exit - i.e., the guest-state and
12199 * exit-information fields only. Other fields are modified by L1 with VMWRITE,
12200 * which already writes to vmcs12 directly.
12202 static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
12203 u32 exit_reason, u32 exit_intr_info,
12204 unsigned long exit_qualification)
12206 /* update guest state fields: */
12207 sync_vmcs12(vcpu, vmcs12);
12209 /* update exit information fields: */
12211 vmcs12->vm_exit_reason = exit_reason;
12212 vmcs12->exit_qualification = exit_qualification;
12213 vmcs12->vm_exit_intr_info = exit_intr_info;
12215 vmcs12->idt_vectoring_info_field = 0;
12216 vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
12217 vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
12219 if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY)) {
12220 vmcs12->launch_state = 1;
12222 /* vm_entry_intr_info_field is cleared on exit. Emulate this
12223 * instead of reading the real value. */
12224 vmcs12->vm_entry_intr_info_field &= ~INTR_INFO_VALID_MASK;
12227 * Transfer the event that L0 or L1 may wanted to inject into
12228 * L2 to IDT_VECTORING_INFO_FIELD.
12230 vmcs12_save_pending_event(vcpu, vmcs12);
12234 * Drop what we picked up for L2 via vmx_complete_interrupts. It is
12235 * preserved above and would only end up incorrectly in L1.
12237 vcpu->arch.nmi_injected = false;
12238 kvm_clear_exception_queue(vcpu);
12239 kvm_clear_interrupt_queue(vcpu);
12242 static void load_vmcs12_mmu_host_state(struct kvm_vcpu *vcpu,
12243 struct vmcs12 *vmcs12)
12245 u32 entry_failure_code;
12247 nested_ept_uninit_mmu_context(vcpu);
12250 * Only PDPTE load can fail as the value of cr3 was checked on entry and
12251 * couldn't have changed.
12253 if (nested_vmx_load_cr3(vcpu, vmcs12->host_cr3, false, &entry_failure_code))
12254 nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_PDPTE_FAIL);
12257 vcpu->arch.walk_mmu->inject_page_fault = kvm_inject_page_fault;
12261 * A part of what we need to when the nested L2 guest exits and we want to
12262 * run its L1 parent, is to reset L1's guest state to the host state specified
12264 * This function is to be called not only on normal nested exit, but also on
12265 * a nested entry failure, as explained in Intel's spec, 3B.23.7 ("VM-Entry
12266 * Failures During or After Loading Guest State").
12267 * This function should be called when the active VMCS is L1's (vmcs01).
12269 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
12270 struct vmcs12 *vmcs12)
12272 struct kvm_segment seg;
12274 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER)
12275 vcpu->arch.efer = vmcs12->host_ia32_efer;
12276 else if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
12277 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
12279 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
12280 vmx_set_efer(vcpu, vcpu->arch.efer);
12282 kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->host_rsp);
12283 kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->host_rip);
12284 vmx_set_rflags(vcpu, X86_EFLAGS_FIXED);
12286 * Note that calling vmx_set_cr0 is important, even if cr0 hasn't
12287 * actually changed, because vmx_set_cr0 refers to efer set above.
12289 * CR0_GUEST_HOST_MASK is already set in the original vmcs01
12290 * (KVM doesn't change it);
12292 vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
12293 vmx_set_cr0(vcpu, vmcs12->host_cr0);
12295 /* Same as above - no reason to call set_cr4_guest_host_mask(). */
12296 vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
12297 vmx_set_cr4(vcpu, vmcs12->host_cr4);
12299 load_vmcs12_mmu_host_state(vcpu, vmcs12);
12302 * If vmcs01 don't use VPID, CPU flushes TLB on every
12303 * VMEntry/VMExit. Thus, no need to flush TLB.
12305 * If vmcs12 uses VPID, TLB entries populated by L2 are
12306 * tagged with vmx->nested.vpid02 while L1 entries are tagged
12307 * with vmx->vpid. Thus, no need to flush TLB.
12309 * Therefore, flush TLB only in case vmcs01 uses VPID and
12310 * vmcs12 don't use VPID as in this case L1 & L2 TLB entries
12311 * are both tagged with vmx->vpid.
12314 !(nested_cpu_has_vpid(vmcs12) && to_vmx(vcpu)->nested.vpid02)) {
12315 vmx_flush_tlb(vcpu, true);
12318 vmcs_write32(GUEST_SYSENTER_CS, vmcs12->host_ia32_sysenter_cs);
12319 vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->host_ia32_sysenter_esp);
12320 vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->host_ia32_sysenter_eip);
12321 vmcs_writel(GUEST_IDTR_BASE, vmcs12->host_idtr_base);
12322 vmcs_writel(GUEST_GDTR_BASE, vmcs12->host_gdtr_base);
12323 vmcs_write32(GUEST_IDTR_LIMIT, 0xFFFF);
12324 vmcs_write32(GUEST_GDTR_LIMIT, 0xFFFF);
12326 /* If not VM_EXIT_CLEAR_BNDCFGS, the L2 value propagates to L1. */
12327 if (vmcs12->vm_exit_controls & VM_EXIT_CLEAR_BNDCFGS)
12328 vmcs_write64(GUEST_BNDCFGS, 0);
12330 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) {
12331 vmcs_write64(GUEST_IA32_PAT, vmcs12->host_ia32_pat);
12332 vcpu->arch.pat = vmcs12->host_ia32_pat;
12334 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
12335 vmcs_write64(GUEST_IA32_PERF_GLOBAL_CTRL,
12336 vmcs12->host_ia32_perf_global_ctrl);
12338 /* Set L1 segment info according to Intel SDM
12339 27.5.2 Loading Host Segment and Descriptor-Table Registers */
12340 seg = (struct kvm_segment) {
12342 .limit = 0xFFFFFFFF,
12343 .selector = vmcs12->host_cs_selector,
12349 if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
12353 vmx_set_segment(vcpu, &seg, VCPU_SREG_CS);
12354 seg = (struct kvm_segment) {
12356 .limit = 0xFFFFFFFF,
12363 seg.selector = vmcs12->host_ds_selector;
12364 vmx_set_segment(vcpu, &seg, VCPU_SREG_DS);
12365 seg.selector = vmcs12->host_es_selector;
12366 vmx_set_segment(vcpu, &seg, VCPU_SREG_ES);
12367 seg.selector = vmcs12->host_ss_selector;
12368 vmx_set_segment(vcpu, &seg, VCPU_SREG_SS);
12369 seg.selector = vmcs12->host_fs_selector;
12370 seg.base = vmcs12->host_fs_base;
12371 vmx_set_segment(vcpu, &seg, VCPU_SREG_FS);
12372 seg.selector = vmcs12->host_gs_selector;
12373 seg.base = vmcs12->host_gs_base;
12374 vmx_set_segment(vcpu, &seg, VCPU_SREG_GS);
12375 seg = (struct kvm_segment) {
12376 .base = vmcs12->host_tr_base,
12378 .selector = vmcs12->host_tr_selector,
12382 vmx_set_segment(vcpu, &seg, VCPU_SREG_TR);
12384 kvm_set_dr(vcpu, 7, 0x400);
12385 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
12387 if (cpu_has_vmx_msr_bitmap())
12388 vmx_update_msr_bitmap(vcpu);
12390 if (nested_vmx_load_msr(vcpu, vmcs12->vm_exit_msr_load_addr,
12391 vmcs12->vm_exit_msr_load_count))
12392 nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
12396 * Emulate an exit from nested guest (L2) to L1, i.e., prepare to run L1
12397 * and modify vmcs12 to make it see what it would expect to see there if
12398 * L2 was its real guest. Must only be called when in L2 (is_guest_mode())
12400 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
12401 u32 exit_intr_info,
12402 unsigned long exit_qualification)
12404 struct vcpu_vmx *vmx = to_vmx(vcpu);
12405 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
12407 /* trying to cancel vmlaunch/vmresume is a bug */
12408 WARN_ON_ONCE(vmx->nested.nested_run_pending);
12411 * The only expected VM-instruction error is "VM entry with
12412 * invalid control field(s)." Anything else indicates a
12415 WARN_ON_ONCE(vmx->fail && (vmcs_read32(VM_INSTRUCTION_ERROR) !=
12416 VMXERR_ENTRY_INVALID_CONTROL_FIELD));
12418 leave_guest_mode(vcpu);
12420 if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING)
12421 vcpu->arch.tsc_offset -= vmcs12->tsc_offset;
12423 if (likely(!vmx->fail)) {
12424 if (exit_reason == -1)
12425 sync_vmcs12(vcpu, vmcs12);
12427 prepare_vmcs12(vcpu, vmcs12, exit_reason, exit_intr_info,
12428 exit_qualification);
12430 if (nested_vmx_store_msr(vcpu, vmcs12->vm_exit_msr_store_addr,
12431 vmcs12->vm_exit_msr_store_count))
12432 nested_vmx_abort(vcpu, VMX_ABORT_SAVE_GUEST_MSR_FAIL);
12435 vmx_switch_vmcs(vcpu, &vmx->vmcs01);
12436 vm_entry_controls_reset_shadow(vmx);
12437 vm_exit_controls_reset_shadow(vmx);
12438 vmx_segment_cache_clear(vmx);
12440 /* Update any VMCS fields that might have changed while L2 ran */
12441 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.nr);
12442 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.nr);
12443 vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
12444 if (vmx->hv_deadline_tsc == -1)
12445 vmcs_clear_bits(PIN_BASED_VM_EXEC_CONTROL,
12446 PIN_BASED_VMX_PREEMPTION_TIMER);
12448 vmcs_set_bits(PIN_BASED_VM_EXEC_CONTROL,
12449 PIN_BASED_VMX_PREEMPTION_TIMER);
12450 if (kvm_has_tsc_control)
12451 decache_tsc_multiplier(vmx);
12453 if (vmx->nested.change_vmcs01_virtual_apic_mode) {
12454 vmx->nested.change_vmcs01_virtual_apic_mode = false;
12455 vmx_set_virtual_apic_mode(vcpu);
12456 } else if (!nested_cpu_has_ept(vmcs12) &&
12457 nested_cpu_has2(vmcs12,
12458 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
12459 vmx_flush_tlb(vcpu, true);
12462 /* This is needed for same reason as it was needed in prepare_vmcs02 */
12465 /* Unpin physical memory we referred to in vmcs02 */
12466 if (vmx->nested.apic_access_page) {
12467 kvm_release_page_dirty(vmx->nested.apic_access_page);
12468 vmx->nested.apic_access_page = NULL;
12470 if (vmx->nested.virtual_apic_page) {
12471 kvm_release_page_dirty(vmx->nested.virtual_apic_page);
12472 vmx->nested.virtual_apic_page = NULL;
12474 if (vmx->nested.pi_desc_page) {
12475 kunmap(vmx->nested.pi_desc_page);
12476 kvm_release_page_dirty(vmx->nested.pi_desc_page);
12477 vmx->nested.pi_desc_page = NULL;
12478 vmx->nested.pi_desc = NULL;
12482 * We are now running in L2, mmu_notifier will force to reload the
12483 * page's hpa for L2 vmcs. Need to reload it for L1 before entering L1.
12485 kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
12487 if (enable_shadow_vmcs && exit_reason != -1)
12488 vmx->nested.sync_shadow_vmcs = true;
12490 /* in case we halted in L2 */
12491 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
12493 if (likely(!vmx->fail)) {
12495 * TODO: SDM says that with acknowledge interrupt on
12496 * exit, bit 31 of the VM-exit interrupt information
12497 * (valid interrupt) is always set to 1 on
12498 * EXIT_REASON_EXTERNAL_INTERRUPT, so we shouldn't
12499 * need kvm_cpu_has_interrupt(). See the commit
12500 * message for details.
12502 if (nested_exit_intr_ack_set(vcpu) &&
12503 exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT &&
12504 kvm_cpu_has_interrupt(vcpu)) {
12505 int irq = kvm_cpu_get_interrupt(vcpu);
12507 vmcs12->vm_exit_intr_info = irq |
12508 INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR;
12511 if (exit_reason != -1)
12512 trace_kvm_nested_vmexit_inject(vmcs12->vm_exit_reason,
12513 vmcs12->exit_qualification,
12514 vmcs12->idt_vectoring_info_field,
12515 vmcs12->vm_exit_intr_info,
12516 vmcs12->vm_exit_intr_error_code,
12519 load_vmcs12_host_state(vcpu, vmcs12);
12525 * After an early L2 VM-entry failure, we're now back
12526 * in L1 which thinks it just finished a VMLAUNCH or
12527 * VMRESUME instruction, so we need to set the failure
12528 * flag and the VM-instruction error field of the VMCS
12531 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
12533 load_vmcs12_mmu_host_state(vcpu, vmcs12);
12536 * The emulated instruction was already skipped in
12537 * nested_vmx_run, but the updated RIP was never
12538 * written back to the vmcs01.
12540 skip_emulated_instruction(vcpu);
12545 * Forcibly leave nested mode in order to be able to reset the VCPU later on.
12547 static void vmx_leave_nested(struct kvm_vcpu *vcpu)
12549 if (is_guest_mode(vcpu)) {
12550 to_vmx(vcpu)->nested.nested_run_pending = 0;
12551 nested_vmx_vmexit(vcpu, -1, 0, 0);
12553 free_nested(to_vmx(vcpu));
12557 * L1's failure to enter L2 is a subset of a normal exit, as explained in
12558 * 23.7 "VM-entry failures during or after loading guest state" (this also
12559 * lists the acceptable exit-reason and exit-qualification parameters).
12560 * It should only be called before L2 actually succeeded to run, and when
12561 * vmcs01 is current (it doesn't leave_guest_mode() or switch vmcss).
12563 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
12564 struct vmcs12 *vmcs12,
12565 u32 reason, unsigned long qualification)
12567 load_vmcs12_host_state(vcpu, vmcs12);
12568 vmcs12->vm_exit_reason = reason | VMX_EXIT_REASONS_FAILED_VMENTRY;
12569 vmcs12->exit_qualification = qualification;
12570 nested_vmx_succeed(vcpu);
12571 if (enable_shadow_vmcs)
12572 to_vmx(vcpu)->nested.sync_shadow_vmcs = true;
12575 static int vmx_check_intercept(struct kvm_vcpu *vcpu,
12576 struct x86_instruction_info *info,
12577 enum x86_intercept_stage stage)
12579 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
12580 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
12583 * RDPID causes #UD if disabled through secondary execution controls.
12584 * Because it is marked as EmulateOnUD, we need to intercept it here.
12586 if (info->intercept == x86_intercept_rdtscp &&
12587 !nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDTSCP)) {
12588 ctxt->exception.vector = UD_VECTOR;
12589 ctxt->exception.error_code_valid = false;
12590 return X86EMUL_PROPAGATE_FAULT;
12593 /* TODO: check more intercepts... */
12594 return X86EMUL_CONTINUE;
12597 #ifdef CONFIG_X86_64
12598 /* (a << shift) / divisor, return 1 if overflow otherwise 0 */
12599 static inline int u64_shl_div_u64(u64 a, unsigned int shift,
12600 u64 divisor, u64 *result)
12602 u64 low = a << shift, high = a >> (64 - shift);
12604 /* To avoid the overflow on divq */
12605 if (high >= divisor)
12608 /* Low hold the result, high hold rem which is discarded */
12609 asm("divq %2\n\t" : "=a" (low), "=d" (high) :
12610 "rm" (divisor), "0" (low), "1" (high));
12616 static int vmx_set_hv_timer(struct kvm_vcpu *vcpu, u64 guest_deadline_tsc)
12618 struct vcpu_vmx *vmx;
12619 u64 tscl, guest_tscl, delta_tsc, lapic_timer_advance_cycles;
12621 if (kvm_mwait_in_guest(vcpu->kvm))
12622 return -EOPNOTSUPP;
12624 vmx = to_vmx(vcpu);
12626 guest_tscl = kvm_read_l1_tsc(vcpu, tscl);
12627 delta_tsc = max(guest_deadline_tsc, guest_tscl) - guest_tscl;
12628 lapic_timer_advance_cycles = nsec_to_cycles(vcpu, lapic_timer_advance_ns);
12630 if (delta_tsc > lapic_timer_advance_cycles)
12631 delta_tsc -= lapic_timer_advance_cycles;
12635 /* Convert to host delta tsc if tsc scaling is enabled */
12636 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio &&
12637 u64_shl_div_u64(delta_tsc,
12638 kvm_tsc_scaling_ratio_frac_bits,
12639 vcpu->arch.tsc_scaling_ratio,
12644 * If the delta tsc can't fit in the 32 bit after the multi shift,
12645 * we can't use the preemption timer.
12646 * It's possible that it fits on later vmentries, but checking
12647 * on every vmentry is costly so we just use an hrtimer.
12649 if (delta_tsc >> (cpu_preemption_timer_multi + 32))
12652 vmx->hv_deadline_tsc = tscl + delta_tsc;
12653 vmcs_set_bits(PIN_BASED_VM_EXEC_CONTROL,
12654 PIN_BASED_VMX_PREEMPTION_TIMER);
12656 return delta_tsc == 0;
12659 static void vmx_cancel_hv_timer(struct kvm_vcpu *vcpu)
12661 struct vcpu_vmx *vmx = to_vmx(vcpu);
12662 vmx->hv_deadline_tsc = -1;
12663 vmcs_clear_bits(PIN_BASED_VM_EXEC_CONTROL,
12664 PIN_BASED_VMX_PREEMPTION_TIMER);
12668 static void vmx_sched_in(struct kvm_vcpu *vcpu, int cpu)
12670 if (!kvm_pause_in_guest(vcpu->kvm))
12671 shrink_ple_window(vcpu);
12674 static void vmx_slot_enable_log_dirty(struct kvm *kvm,
12675 struct kvm_memory_slot *slot)
12677 kvm_mmu_slot_leaf_clear_dirty(kvm, slot);
12678 kvm_mmu_slot_largepage_remove_write_access(kvm, slot);
12681 static void vmx_slot_disable_log_dirty(struct kvm *kvm,
12682 struct kvm_memory_slot *slot)
12684 kvm_mmu_slot_set_dirty(kvm, slot);
12687 static void vmx_flush_log_dirty(struct kvm *kvm)
12689 kvm_flush_pml_buffers(kvm);
12692 static int vmx_write_pml_buffer(struct kvm_vcpu *vcpu)
12694 struct vmcs12 *vmcs12;
12695 struct vcpu_vmx *vmx = to_vmx(vcpu);
12697 struct page *page = NULL;
12700 if (is_guest_mode(vcpu)) {
12701 WARN_ON_ONCE(vmx->nested.pml_full);
12704 * Check if PML is enabled for the nested guest.
12705 * Whether eptp bit 6 is set is already checked
12706 * as part of A/D emulation.
12708 vmcs12 = get_vmcs12(vcpu);
12709 if (!nested_cpu_has_pml(vmcs12))
12712 if (vmcs12->guest_pml_index >= PML_ENTITY_NUM) {
12713 vmx->nested.pml_full = true;
12717 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS) & ~0xFFFull;
12719 page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->pml_address);
12720 if (is_error_page(page))
12723 pml_address = kmap(page);
12724 pml_address[vmcs12->guest_pml_index--] = gpa;
12726 kvm_release_page_clean(page);
12732 static void vmx_enable_log_dirty_pt_masked(struct kvm *kvm,
12733 struct kvm_memory_slot *memslot,
12734 gfn_t offset, unsigned long mask)
12736 kvm_mmu_clear_dirty_pt_masked(kvm, memslot, offset, mask);
12739 static void __pi_post_block(struct kvm_vcpu *vcpu)
12741 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
12742 struct pi_desc old, new;
12746 old.control = new.control = pi_desc->control;
12747 WARN(old.nv != POSTED_INTR_WAKEUP_VECTOR,
12748 "Wakeup handler not enabled while the VCPU is blocked\n");
12750 dest = cpu_physical_id(vcpu->cpu);
12752 if (x2apic_enabled())
12755 new.ndst = (dest << 8) & 0xFF00;
12757 /* set 'NV' to 'notification vector' */
12758 new.nv = POSTED_INTR_VECTOR;
12759 } while (cmpxchg64(&pi_desc->control, old.control,
12760 new.control) != old.control);
12762 if (!WARN_ON_ONCE(vcpu->pre_pcpu == -1)) {
12763 spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
12764 list_del(&vcpu->blocked_vcpu_list);
12765 spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
12766 vcpu->pre_pcpu = -1;
12771 * This routine does the following things for vCPU which is going
12772 * to be blocked if VT-d PI is enabled.
12773 * - Store the vCPU to the wakeup list, so when interrupts happen
12774 * we can find the right vCPU to wake up.
12775 * - Change the Posted-interrupt descriptor as below:
12776 * 'NDST' <-- vcpu->pre_pcpu
12777 * 'NV' <-- POSTED_INTR_WAKEUP_VECTOR
12778 * - If 'ON' is set during this process, which means at least one
12779 * interrupt is posted for this vCPU, we cannot block it, in
12780 * this case, return 1, otherwise, return 0.
12783 static int pi_pre_block(struct kvm_vcpu *vcpu)
12786 struct pi_desc old, new;
12787 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
12789 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
12790 !irq_remapping_cap(IRQ_POSTING_CAP) ||
12791 !kvm_vcpu_apicv_active(vcpu))
12794 WARN_ON(irqs_disabled());
12795 local_irq_disable();
12796 if (!WARN_ON_ONCE(vcpu->pre_pcpu != -1)) {
12797 vcpu->pre_pcpu = vcpu->cpu;
12798 spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
12799 list_add_tail(&vcpu->blocked_vcpu_list,
12800 &per_cpu(blocked_vcpu_on_cpu,
12802 spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
12806 old.control = new.control = pi_desc->control;
12808 WARN((pi_desc->sn == 1),
12809 "Warning: SN field of posted-interrupts "
12810 "is set before blocking\n");
12813 * Since vCPU can be preempted during this process,
12814 * vcpu->cpu could be different with pre_pcpu, we
12815 * need to set pre_pcpu as the destination of wakeup
12816 * notification event, then we can find the right vCPU
12817 * to wakeup in wakeup handler if interrupts happen
12818 * when the vCPU is in blocked state.
12820 dest = cpu_physical_id(vcpu->pre_pcpu);
12822 if (x2apic_enabled())
12825 new.ndst = (dest << 8) & 0xFF00;
12827 /* set 'NV' to 'wakeup vector' */
12828 new.nv = POSTED_INTR_WAKEUP_VECTOR;
12829 } while (cmpxchg64(&pi_desc->control, old.control,
12830 new.control) != old.control);
12832 /* We should not block the vCPU if an interrupt is posted for it. */
12833 if (pi_test_on(pi_desc) == 1)
12834 __pi_post_block(vcpu);
12836 local_irq_enable();
12837 return (vcpu->pre_pcpu == -1);
12840 static int vmx_pre_block(struct kvm_vcpu *vcpu)
12842 if (pi_pre_block(vcpu))
12845 if (kvm_lapic_hv_timer_in_use(vcpu))
12846 kvm_lapic_switch_to_sw_timer(vcpu);
12851 static void pi_post_block(struct kvm_vcpu *vcpu)
12853 if (vcpu->pre_pcpu == -1)
12856 WARN_ON(irqs_disabled());
12857 local_irq_disable();
12858 __pi_post_block(vcpu);
12859 local_irq_enable();
12862 static void vmx_post_block(struct kvm_vcpu *vcpu)
12864 if (kvm_x86_ops->set_hv_timer)
12865 kvm_lapic_switch_to_hv_timer(vcpu);
12867 pi_post_block(vcpu);
12871 * vmx_update_pi_irte - set IRTE for Posted-Interrupts
12874 * @host_irq: host irq of the interrupt
12875 * @guest_irq: gsi of the interrupt
12876 * @set: set or unset PI
12877 * returns 0 on success, < 0 on failure
12879 static int vmx_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
12880 uint32_t guest_irq, bool set)
12882 struct kvm_kernel_irq_routing_entry *e;
12883 struct kvm_irq_routing_table *irq_rt;
12884 struct kvm_lapic_irq irq;
12885 struct kvm_vcpu *vcpu;
12886 struct vcpu_data vcpu_info;
12889 if (!kvm_arch_has_assigned_device(kvm) ||
12890 !irq_remapping_cap(IRQ_POSTING_CAP) ||
12891 !kvm_vcpu_apicv_active(kvm->vcpus[0]))
12894 idx = srcu_read_lock(&kvm->irq_srcu);
12895 irq_rt = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu);
12896 if (guest_irq >= irq_rt->nr_rt_entries ||
12897 hlist_empty(&irq_rt->map[guest_irq])) {
12898 pr_warn_once("no route for guest_irq %u/%u (broken user space?)\n",
12899 guest_irq, irq_rt->nr_rt_entries);
12903 hlist_for_each_entry(e, &irq_rt->map[guest_irq], link) {
12904 if (e->type != KVM_IRQ_ROUTING_MSI)
12907 * VT-d PI cannot support posting multicast/broadcast
12908 * interrupts to a vCPU, we still use interrupt remapping
12909 * for these kind of interrupts.
12911 * For lowest-priority interrupts, we only support
12912 * those with single CPU as the destination, e.g. user
12913 * configures the interrupts via /proc/irq or uses
12914 * irqbalance to make the interrupts single-CPU.
12916 * We will support full lowest-priority interrupt later.
12919 kvm_set_msi_irq(kvm, e, &irq);
12920 if (!kvm_intr_is_single_vcpu(kvm, &irq, &vcpu)) {
12922 * Make sure the IRTE is in remapped mode if
12923 * we don't handle it in posted mode.
12925 ret = irq_set_vcpu_affinity(host_irq, NULL);
12928 "failed to back to remapped mode, irq: %u\n",
12936 vcpu_info.pi_desc_addr = __pa(vcpu_to_pi_desc(vcpu));
12937 vcpu_info.vector = irq.vector;
12939 trace_kvm_pi_irte_update(host_irq, vcpu->vcpu_id, e->gsi,
12940 vcpu_info.vector, vcpu_info.pi_desc_addr, set);
12943 ret = irq_set_vcpu_affinity(host_irq, &vcpu_info);
12945 ret = irq_set_vcpu_affinity(host_irq, NULL);
12948 printk(KERN_INFO "%s: failed to update PI IRTE\n",
12956 srcu_read_unlock(&kvm->irq_srcu, idx);
12960 static void vmx_setup_mce(struct kvm_vcpu *vcpu)
12962 if (vcpu->arch.mcg_cap & MCG_LMCE_P)
12963 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
12964 FEATURE_CONTROL_LMCE;
12966 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
12967 ~FEATURE_CONTROL_LMCE;
12970 static int vmx_smi_allowed(struct kvm_vcpu *vcpu)
12972 /* we need a nested vmexit to enter SMM, postpone if run is pending */
12973 if (to_vmx(vcpu)->nested.nested_run_pending)
12978 static int vmx_pre_enter_smm(struct kvm_vcpu *vcpu, char *smstate)
12980 struct vcpu_vmx *vmx = to_vmx(vcpu);
12982 vmx->nested.smm.guest_mode = is_guest_mode(vcpu);
12983 if (vmx->nested.smm.guest_mode)
12984 nested_vmx_vmexit(vcpu, -1, 0, 0);
12986 vmx->nested.smm.vmxon = vmx->nested.vmxon;
12987 vmx->nested.vmxon = false;
12988 vmx_clear_hlt(vcpu);
12992 static int vmx_pre_leave_smm(struct kvm_vcpu *vcpu, u64 smbase)
12994 struct vcpu_vmx *vmx = to_vmx(vcpu);
12997 if (vmx->nested.smm.vmxon) {
12998 vmx->nested.vmxon = true;
12999 vmx->nested.smm.vmxon = false;
13002 if (vmx->nested.smm.guest_mode) {
13003 vcpu->arch.hflags &= ~HF_SMM_MASK;
13004 ret = enter_vmx_non_root_mode(vcpu, NULL);
13005 vcpu->arch.hflags |= HF_SMM_MASK;
13009 vmx->nested.smm.guest_mode = false;
13014 static int enable_smi_window(struct kvm_vcpu *vcpu)
13019 static struct kvm_x86_ops vmx_x86_ops __ro_after_init = {
13020 .cpu_has_kvm_support = cpu_has_kvm_support,
13021 .disabled_by_bios = vmx_disabled_by_bios,
13022 .hardware_setup = hardware_setup,
13023 .hardware_unsetup = hardware_unsetup,
13024 .check_processor_compatibility = vmx_check_processor_compat,
13025 .hardware_enable = hardware_enable,
13026 .hardware_disable = hardware_disable,
13027 .cpu_has_accelerated_tpr = report_flexpriority,
13028 .has_emulated_msr = vmx_has_emulated_msr,
13030 .vm_init = vmx_vm_init,
13031 .vm_alloc = vmx_vm_alloc,
13032 .vm_free = vmx_vm_free,
13034 .vcpu_create = vmx_create_vcpu,
13035 .vcpu_free = vmx_free_vcpu,
13036 .vcpu_reset = vmx_vcpu_reset,
13038 .prepare_guest_switch = vmx_save_host_state,
13039 .vcpu_load = vmx_vcpu_load,
13040 .vcpu_put = vmx_vcpu_put,
13042 .update_bp_intercept = update_exception_bitmap,
13043 .get_msr_feature = vmx_get_msr_feature,
13044 .get_msr = vmx_get_msr,
13045 .set_msr = vmx_set_msr,
13046 .get_segment_base = vmx_get_segment_base,
13047 .get_segment = vmx_get_segment,
13048 .set_segment = vmx_set_segment,
13049 .get_cpl = vmx_get_cpl,
13050 .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
13051 .decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
13052 .decache_cr3 = vmx_decache_cr3,
13053 .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
13054 .set_cr0 = vmx_set_cr0,
13055 .set_cr3 = vmx_set_cr3,
13056 .set_cr4 = vmx_set_cr4,
13057 .set_efer = vmx_set_efer,
13058 .get_idt = vmx_get_idt,
13059 .set_idt = vmx_set_idt,
13060 .get_gdt = vmx_get_gdt,
13061 .set_gdt = vmx_set_gdt,
13062 .get_dr6 = vmx_get_dr6,
13063 .set_dr6 = vmx_set_dr6,
13064 .set_dr7 = vmx_set_dr7,
13065 .sync_dirty_debug_regs = vmx_sync_dirty_debug_regs,
13066 .cache_reg = vmx_cache_reg,
13067 .get_rflags = vmx_get_rflags,
13068 .set_rflags = vmx_set_rflags,
13070 .tlb_flush = vmx_flush_tlb,
13072 .run = vmx_vcpu_run,
13073 .handle_exit = vmx_handle_exit,
13074 .skip_emulated_instruction = skip_emulated_instruction,
13075 .set_interrupt_shadow = vmx_set_interrupt_shadow,
13076 .get_interrupt_shadow = vmx_get_interrupt_shadow,
13077 .patch_hypercall = vmx_patch_hypercall,
13078 .set_irq = vmx_inject_irq,
13079 .set_nmi = vmx_inject_nmi,
13080 .queue_exception = vmx_queue_exception,
13081 .cancel_injection = vmx_cancel_injection,
13082 .interrupt_allowed = vmx_interrupt_allowed,
13083 .nmi_allowed = vmx_nmi_allowed,
13084 .get_nmi_mask = vmx_get_nmi_mask,
13085 .set_nmi_mask = vmx_set_nmi_mask,
13086 .enable_nmi_window = enable_nmi_window,
13087 .enable_irq_window = enable_irq_window,
13088 .update_cr8_intercept = update_cr8_intercept,
13089 .set_virtual_apic_mode = vmx_set_virtual_apic_mode,
13090 .set_apic_access_page_addr = vmx_set_apic_access_page_addr,
13091 .get_enable_apicv = vmx_get_enable_apicv,
13092 .refresh_apicv_exec_ctrl = vmx_refresh_apicv_exec_ctrl,
13093 .load_eoi_exitmap = vmx_load_eoi_exitmap,
13094 .apicv_post_state_restore = vmx_apicv_post_state_restore,
13095 .hwapic_irr_update = vmx_hwapic_irr_update,
13096 .hwapic_isr_update = vmx_hwapic_isr_update,
13097 .sync_pir_to_irr = vmx_sync_pir_to_irr,
13098 .deliver_posted_interrupt = vmx_deliver_posted_interrupt,
13100 .set_tss_addr = vmx_set_tss_addr,
13101 .set_identity_map_addr = vmx_set_identity_map_addr,
13102 .get_tdp_level = get_ept_level,
13103 .get_mt_mask = vmx_get_mt_mask,
13105 .get_exit_info = vmx_get_exit_info,
13107 .get_lpage_level = vmx_get_lpage_level,
13109 .cpuid_update = vmx_cpuid_update,
13111 .rdtscp_supported = vmx_rdtscp_supported,
13112 .invpcid_supported = vmx_invpcid_supported,
13114 .set_supported_cpuid = vmx_set_supported_cpuid,
13116 .has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
13118 .read_l1_tsc_offset = vmx_read_l1_tsc_offset,
13119 .write_tsc_offset = vmx_write_tsc_offset,
13121 .set_tdp_cr3 = vmx_set_cr3,
13123 .check_intercept = vmx_check_intercept,
13124 .handle_external_intr = vmx_handle_external_intr,
13125 .mpx_supported = vmx_mpx_supported,
13126 .xsaves_supported = vmx_xsaves_supported,
13127 .umip_emulated = vmx_umip_emulated,
13129 .check_nested_events = vmx_check_nested_events,
13131 .sched_in = vmx_sched_in,
13133 .slot_enable_log_dirty = vmx_slot_enable_log_dirty,
13134 .slot_disable_log_dirty = vmx_slot_disable_log_dirty,
13135 .flush_log_dirty = vmx_flush_log_dirty,
13136 .enable_log_dirty_pt_masked = vmx_enable_log_dirty_pt_masked,
13137 .write_log_dirty = vmx_write_pml_buffer,
13139 .pre_block = vmx_pre_block,
13140 .post_block = vmx_post_block,
13142 .pmu_ops = &intel_pmu_ops,
13144 .update_pi_irte = vmx_update_pi_irte,
13146 #ifdef CONFIG_X86_64
13147 .set_hv_timer = vmx_set_hv_timer,
13148 .cancel_hv_timer = vmx_cancel_hv_timer,
13151 .setup_mce = vmx_setup_mce,
13153 .get_vmcs12_pages = nested_get_vmcs12_pages,
13155 .smi_allowed = vmx_smi_allowed,
13156 .pre_enter_smm = vmx_pre_enter_smm,
13157 .pre_leave_smm = vmx_pre_leave_smm,
13158 .enable_smi_window = enable_smi_window,
13161 static int __init vmx_init(void)
13165 #if IS_ENABLED(CONFIG_HYPERV)
13167 * Enlightened VMCS usage should be recommended and the host needs
13168 * to support eVMCS v1 or above. We can also disable eVMCS support
13169 * with module parameter.
13171 if (enlightened_vmcs &&
13172 ms_hyperv.hints & HV_X64_ENLIGHTENED_VMCS_RECOMMENDED &&
13173 (ms_hyperv.nested_features & HV_X64_ENLIGHTENED_VMCS_VERSION) >=
13174 KVM_EVMCS_VERSION) {
13177 /* Check that we have assist pages on all online CPUs */
13178 for_each_online_cpu(cpu) {
13179 if (!hv_get_vp_assist_page(cpu)) {
13180 enlightened_vmcs = false;
13185 if (enlightened_vmcs) {
13186 pr_info("KVM: vmx: using Hyper-V Enlightened VMCS\n");
13187 static_branch_enable(&enable_evmcs);
13190 enlightened_vmcs = false;
13194 r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx),
13195 __alignof__(struct vcpu_vmx), THIS_MODULE);
13199 #ifdef CONFIG_KEXEC_CORE
13200 rcu_assign_pointer(crash_vmclear_loaded_vmcss,
13201 crash_vmclear_local_loaded_vmcss);
13203 vmx_check_vmcs12_offsets();
13208 static void __exit vmx_exit(void)
13210 #ifdef CONFIG_KEXEC_CORE
13211 RCU_INIT_POINTER(crash_vmclear_loaded_vmcss, NULL);
13217 #if IS_ENABLED(CONFIG_HYPERV)
13218 if (static_branch_unlikely(&enable_evmcs)) {
13220 struct hv_vp_assist_page *vp_ap;
13222 * Reset everything to support using non-enlightened VMCS
13223 * access later (e.g. when we reload the module with
13224 * enlightened_vmcs=0)
13226 for_each_online_cpu(cpu) {
13227 vp_ap = hv_get_vp_assist_page(cpu);
13232 vp_ap->current_nested_vmcs = 0;
13233 vp_ap->enlighten_vmentry = 0;
13236 static_branch_disable(&enable_evmcs);
13241 module_init(vmx_init)
13242 module_exit(vmx_exit)