]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame_incremental - arch/x86/kvm/vmx.c
projects / mirror_ubuntu-bionic-kernel.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
KVM: x86: Fix far-jump to non-canonical check
[mirror_ubuntu-bionic-kernel.git] / arch / x86 / kvm / vmx.c
... / ...
CommitLineData
1/*
2 * Kernel-based Virtual Machine driver for Linux
3 *
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
6 *
7 * Copyright (C) 2006 Qumranet, Inc.
8 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
9 *
10 * Authors:
11 * Avi Kivity <avi@qumranet.com>
12 * Yaniv Kamay <yaniv@qumranet.com>
13 *
14 * This work is licensed under the terms of the GNU GPL, version 2. See
15 * the COPYING file in the top-level directory.
16 *
17 */
18
19#include "irq.h"
20#include "mmu.h"
21#include "cpuid.h"
22
23#include <linux/kvm_host.h>
24#include <linux/module.h>
25#include <linux/kernel.h>
26#include <linux/mm.h>
27#include <linux/highmem.h>
28#include <linux/sched.h>
29#include <linux/moduleparam.h>
30#include <linux/mod_devicetable.h>
31#include <linux/ftrace_event.h>
32#include <linux/slab.h>
33#include <linux/tboot.h>
34#include <linux/hrtimer.h>
35#include "kvm_cache_regs.h"
36#include "x86.h"
37
38#include <asm/io.h>
39#include <asm/desc.h>
40#include <asm/vmx.h>
41#include <asm/virtext.h>
42#include <asm/mce.h>
43#include <asm/i387.h>
44#include <asm/xcr.h>
45#include <asm/perf_event.h>
46#include <asm/debugreg.h>
47#include <asm/kexec.h>
48
49#include "trace.h"
50
51#define __ex(x) __kvm_handle_fault_on_reboot(x)
52#define __ex_clear(x, reg) \
53 ____kvm_handle_fault_on_reboot(x, "xor " reg " , " reg)
54
55MODULE_AUTHOR("Qumranet");
56MODULE_LICENSE("GPL");
57
58static const struct x86_cpu_id vmx_cpu_id[] = {
59 X86_FEATURE_MATCH(X86_FEATURE_VMX),
60 {}
61};
62MODULE_DEVICE_TABLE(x86cpu, vmx_cpu_id);
63
64static bool __read_mostly enable_vpid = 1;
65module_param_named(vpid, enable_vpid, bool, 0444);
66
67static bool __read_mostly flexpriority_enabled = 1;
68module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
69
70static bool __read_mostly enable_ept = 1;
71module_param_named(ept, enable_ept, bool, S_IRUGO);
72
73static bool __read_mostly enable_unrestricted_guest = 1;
74module_param_named(unrestricted_guest,
75 enable_unrestricted_guest, bool, S_IRUGO);
76
77static bool __read_mostly enable_ept_ad_bits = 1;
78module_param_named(eptad, enable_ept_ad_bits, bool, S_IRUGO);
79
80static bool __read_mostly emulate_invalid_guest_state = true;
81module_param(emulate_invalid_guest_state, bool, S_IRUGO);
82
83static bool __read_mostly vmm_exclusive = 1;
84module_param(vmm_exclusive, bool, S_IRUGO);
85
86static bool __read_mostly fasteoi = 1;
87module_param(fasteoi, bool, S_IRUGO);
88
89static bool __read_mostly enable_apicv = 1;
90module_param(enable_apicv, bool, S_IRUGO);
91
92static bool __read_mostly enable_shadow_vmcs = 1;
93module_param_named(enable_shadow_vmcs, enable_shadow_vmcs, bool, S_IRUGO);
94/*
95 * If nested=1, nested virtualization is supported, i.e., guests may use
96 * VMX and be a hypervisor for its own guests. If nested=0, guests may not
97 * use VMX instructions.
98 */
99static bool __read_mostly nested = 0;
100module_param(nested, bool, S_IRUGO);
101
102#define KVM_GUEST_CR0_MASK (X86_CR0_NW | X86_CR0_CD)
103#define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST (X86_CR0_WP | X86_CR0_NE)
104#define KVM_VM_CR0_ALWAYS_ON \
105 (KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
106#define KVM_CR4_GUEST_OWNED_BITS \
107 (X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR \
108 | X86_CR4_OSXMMEXCPT)
109
110#define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
111#define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
112
113#define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
114
115#define VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE 5
116
117/*
118 * These 2 parameters are used to config the controls for Pause-Loop Exiting:
119 * ple_gap: upper bound on the amount of time between two successive
120 * executions of PAUSE in a loop. Also indicate if ple enabled.
121 * According to test, this time is usually smaller than 128 cycles.
122 * ple_window: upper bound on the amount of time a guest is allowed to execute
123 * in a PAUSE loop. Tests indicate that most spinlocks are held for
124 * less than 2^12 cycles
125 * Time is measured based on a counter that runs at the same rate as the TSC,
126 * refer SDM volume 3b section 21.6.13 & 22.1.3.
127 */
128#define KVM_VMX_DEFAULT_PLE_GAP 128
129#define KVM_VMX_DEFAULT_PLE_WINDOW 4096
130#define KVM_VMX_DEFAULT_PLE_WINDOW_GROW 2
131#define KVM_VMX_DEFAULT_PLE_WINDOW_SHRINK 0
132#define KVM_VMX_DEFAULT_PLE_WINDOW_MAX \
133 INT_MAX / KVM_VMX_DEFAULT_PLE_WINDOW_GROW
134
135static int ple_gap = KVM_VMX_DEFAULT_PLE_GAP;
136module_param(ple_gap, int, S_IRUGO);
137
138static int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
139module_param(ple_window, int, S_IRUGO);
140
141/* Default doubles per-vcpu window every exit. */
142static int ple_window_grow = KVM_VMX_DEFAULT_PLE_WINDOW_GROW;
143module_param(ple_window_grow, int, S_IRUGO);
144
145/* Default resets per-vcpu window every exit to ple_window. */
146static int ple_window_shrink = KVM_VMX_DEFAULT_PLE_WINDOW_SHRINK;
147module_param(ple_window_shrink, int, S_IRUGO);
148
149/* Default is to compute the maximum so we can never overflow. */
150static int ple_window_actual_max = KVM_VMX_DEFAULT_PLE_WINDOW_MAX;
151static int ple_window_max = KVM_VMX_DEFAULT_PLE_WINDOW_MAX;
152module_param(ple_window_max, int, S_IRUGO);
153
154extern const ulong vmx_return;
155
156#define NR_AUTOLOAD_MSRS 8
157#define VMCS02_POOL_SIZE 1
158
159struct vmcs {
160 u32 revision_id;
161 u32 abort;
162 char data[0];
163};
164
165/*
166 * Track a VMCS that may be loaded on a certain CPU. If it is (cpu!=-1), also
167 * remember whether it was VMLAUNCHed, and maintain a linked list of all VMCSs
168 * loaded on this CPU (so we can clear them if the CPU goes down).
169 */
170struct loaded_vmcs {
171 struct vmcs *vmcs;
172 int cpu;
173 int launched;
174 struct list_head loaded_vmcss_on_cpu_link;
175};
176
177struct shared_msr_entry {
178 unsigned index;
179 u64 data;
180 u64 mask;
181};
182
183/*
184 * struct vmcs12 describes the state that our guest hypervisor (L1) keeps for a
185 * single nested guest (L2), hence the name vmcs12. Any VMX implementation has
186 * a VMCS structure, and vmcs12 is our emulated VMX's VMCS. This structure is
187 * stored in guest memory specified by VMPTRLD, but is opaque to the guest,
188 * which must access it using VMREAD/VMWRITE/VMCLEAR instructions.
189 * More than one of these structures may exist, if L1 runs multiple L2 guests.
190 * nested_vmx_run() will use the data here to build a vmcs02: a VMCS for the
191 * underlying hardware which will be used to run L2.
192 * This structure is packed to ensure that its layout is identical across
193 * machines (necessary for live migration).
194 * If there are changes in this struct, VMCS12_REVISION must be changed.
195 */
196typedef u64 natural_width;
197struct __packed vmcs12 {
198 /* According to the Intel spec, a VMCS region must start with the
199 * following two fields. Then follow implementation-specific data.
200 */
201 u32 revision_id;
202 u32 abort;
203
204 u32 launch_state; /* set to 0 by VMCLEAR, to 1 by VMLAUNCH */
205 u32 padding[7]; /* room for future expansion */
206
207 u64 io_bitmap_a;
208 u64 io_bitmap_b;
209 u64 msr_bitmap;
210 u64 vm_exit_msr_store_addr;
211 u64 vm_exit_msr_load_addr;
212 u64 vm_entry_msr_load_addr;
213 u64 tsc_offset;
214 u64 virtual_apic_page_addr;
215 u64 apic_access_addr;
216 u64 ept_pointer;
217 u64 guest_physical_address;
218 u64 vmcs_link_pointer;
219 u64 guest_ia32_debugctl;
220 u64 guest_ia32_pat;
221 u64 guest_ia32_efer;
222 u64 guest_ia32_perf_global_ctrl;
223 u64 guest_pdptr0;
224 u64 guest_pdptr1;
225 u64 guest_pdptr2;
226 u64 guest_pdptr3;
227 u64 guest_bndcfgs;
228 u64 host_ia32_pat;
229 u64 host_ia32_efer;
230 u64 host_ia32_perf_global_ctrl;
231 u64 padding64[8]; /* room for future expansion */
232 /*
233 * To allow migration of L1 (complete with its L2 guests) between
234 * machines of different natural widths (32 or 64 bit), we cannot have
235 * unsigned long fields with no explict size. We use u64 (aliased
236 * natural_width) instead. Luckily, x86 is little-endian.
237 */
238 natural_width cr0_guest_host_mask;
239 natural_width cr4_guest_host_mask;
240 natural_width cr0_read_shadow;
241 natural_width cr4_read_shadow;
242 natural_width cr3_target_value0;
243 natural_width cr3_target_value1;
244 natural_width cr3_target_value2;
245 natural_width cr3_target_value3;
246 natural_width exit_qualification;
247 natural_width guest_linear_address;
248 natural_width guest_cr0;
249 natural_width guest_cr3;
250 natural_width guest_cr4;
251 natural_width guest_es_base;
252 natural_width guest_cs_base;
253 natural_width guest_ss_base;
254 natural_width guest_ds_base;
255 natural_width guest_fs_base;
256 natural_width guest_gs_base;
257 natural_width guest_ldtr_base;
258 natural_width guest_tr_base;
259 natural_width guest_gdtr_base;
260 natural_width guest_idtr_base;
261 natural_width guest_dr7;
262 natural_width guest_rsp;
263 natural_width guest_rip;
264 natural_width guest_rflags;
265 natural_width guest_pending_dbg_exceptions;
266 natural_width guest_sysenter_esp;
267 natural_width guest_sysenter_eip;
268 natural_width host_cr0;
269 natural_width host_cr3;
270 natural_width host_cr4;
271 natural_width host_fs_base;
272 natural_width host_gs_base;
273 natural_width host_tr_base;
274 natural_width host_gdtr_base;
275 natural_width host_idtr_base;
276 natural_width host_ia32_sysenter_esp;
277 natural_width host_ia32_sysenter_eip;
278 natural_width host_rsp;
279 natural_width host_rip;
280 natural_width paddingl[8]; /* room for future expansion */
281 u32 pin_based_vm_exec_control;
282 u32 cpu_based_vm_exec_control;
283 u32 exception_bitmap;
284 u32 page_fault_error_code_mask;
285 u32 page_fault_error_code_match;
286 u32 cr3_target_count;
287 u32 vm_exit_controls;
288 u32 vm_exit_msr_store_count;
289 u32 vm_exit_msr_load_count;
290 u32 vm_entry_controls;
291 u32 vm_entry_msr_load_count;
292 u32 vm_entry_intr_info_field;
293 u32 vm_entry_exception_error_code;
294 u32 vm_entry_instruction_len;
295 u32 tpr_threshold;
296 u32 secondary_vm_exec_control;
297 u32 vm_instruction_error;
298 u32 vm_exit_reason;
299 u32 vm_exit_intr_info;
300 u32 vm_exit_intr_error_code;
301 u32 idt_vectoring_info_field;
302 u32 idt_vectoring_error_code;
303 u32 vm_exit_instruction_len;
304 u32 vmx_instruction_info;
305 u32 guest_es_limit;
306 u32 guest_cs_limit;
307 u32 guest_ss_limit;
308 u32 guest_ds_limit;
309 u32 guest_fs_limit;
310 u32 guest_gs_limit;
311 u32 guest_ldtr_limit;
312 u32 guest_tr_limit;
313 u32 guest_gdtr_limit;
314 u32 guest_idtr_limit;
315 u32 guest_es_ar_bytes;
316 u32 guest_cs_ar_bytes;
317 u32 guest_ss_ar_bytes;
318 u32 guest_ds_ar_bytes;
319 u32 guest_fs_ar_bytes;
320 u32 guest_gs_ar_bytes;
321 u32 guest_ldtr_ar_bytes;
322 u32 guest_tr_ar_bytes;
323 u32 guest_interruptibility_info;
324 u32 guest_activity_state;
325 u32 guest_sysenter_cs;
326 u32 host_ia32_sysenter_cs;
327 u32 vmx_preemption_timer_value;
328 u32 padding32[7]; /* room for future expansion */
329 u16 virtual_processor_id;
330 u16 guest_es_selector;
331 u16 guest_cs_selector;
332 u16 guest_ss_selector;
333 u16 guest_ds_selector;
334 u16 guest_fs_selector;
335 u16 guest_gs_selector;
336 u16 guest_ldtr_selector;
337 u16 guest_tr_selector;
338 u16 host_es_selector;
339 u16 host_cs_selector;
340 u16 host_ss_selector;
341 u16 host_ds_selector;
342 u16 host_fs_selector;
343 u16 host_gs_selector;
344 u16 host_tr_selector;
345};
346
347/*
348 * VMCS12_REVISION is an arbitrary id that should be changed if the content or
349 * layout of struct vmcs12 is changed. MSR_IA32_VMX_BASIC returns this id, and
350 * VMPTRLD verifies that the VMCS region that L1 is loading contains this id.
351 */
352#define VMCS12_REVISION 0x11e57ed0
353
354/*
355 * VMCS12_SIZE is the number of bytes L1 should allocate for the VMXON region
356 * and any VMCS region. Although only sizeof(struct vmcs12) are used by the
357 * current implementation, 4K are reserved to avoid future complications.
358 */
359#define VMCS12_SIZE 0x1000
360
361/* Used to remember the last vmcs02 used for some recently used vmcs12s */
362struct vmcs02_list {
363 struct list_head list;
364 gpa_t vmptr;
365 struct loaded_vmcs vmcs02;
366};
367
368/*
369 * The nested_vmx structure is part of vcpu_vmx, and holds information we need
370 * for correct emulation of VMX (i.e., nested VMX) on this vcpu.
371 */
372struct nested_vmx {
373 /* Has the level1 guest done vmxon? */
374 bool vmxon;
375 gpa_t vmxon_ptr;
376
377 /* The guest-physical address of the current VMCS L1 keeps for L2 */
378 gpa_t current_vmptr;
379 /* The host-usable pointer to the above */
380 struct page *current_vmcs12_page;
381 struct vmcs12 *current_vmcs12;
382 struct vmcs *current_shadow_vmcs;
383 /*
384 * Indicates if the shadow vmcs must be updated with the
385 * data hold by vmcs12
386 */
387 bool sync_shadow_vmcs;
388
389 /* vmcs02_list cache of VMCSs recently used to run L2 guests */
390 struct list_head vmcs02_pool;
391 int vmcs02_num;
392 u64 vmcs01_tsc_offset;
393 /* L2 must run next, and mustn't decide to exit to L1. */
394 bool nested_run_pending;
395 /*
396 * Guest pages referred to in vmcs02 with host-physical pointers, so
397 * we must keep them pinned while L2 runs.
398 */
399 struct page *apic_access_page;
400 struct page *virtual_apic_page;
401 u64 msr_ia32_feature_control;
402
403 struct hrtimer preemption_timer;
404 bool preemption_timer_expired;
405
406 /* to migrate it to L2 if VM_ENTRY_LOAD_DEBUG_CONTROLS is off */
407 u64 vmcs01_debugctl;
408};
409
410#define POSTED_INTR_ON 0
411/* Posted-Interrupt Descriptor */
412struct pi_desc {
413 u32 pir[8]; /* Posted interrupt requested */
414 u32 control; /* bit 0 of control is outstanding notification bit */
415 u32 rsvd[7];
416} __aligned(64);
417
418static bool pi_test_and_set_on(struct pi_desc *pi_desc)
419{
420 return test_and_set_bit(POSTED_INTR_ON,
421 (unsigned long *)&pi_desc->control);
422}
423
424static bool pi_test_and_clear_on(struct pi_desc *pi_desc)
425{
426 return test_and_clear_bit(POSTED_INTR_ON,
427 (unsigned long *)&pi_desc->control);
428}
429
430static int pi_test_and_set_pir(int vector, struct pi_desc *pi_desc)
431{
432 return test_and_set_bit(vector, (unsigned long *)pi_desc->pir);
433}
434
435struct vcpu_vmx {
436 struct kvm_vcpu vcpu;
437 unsigned long host_rsp;
438 u8 fail;
439 bool nmi_known_unmasked;
440 u32 exit_intr_info;
441 u32 idt_vectoring_info;
442 ulong rflags;
443 struct shared_msr_entry *guest_msrs;
444 int nmsrs;
445 int save_nmsrs;
446 unsigned long host_idt_base;
447#ifdef CONFIG_X86_64
448 u64 msr_host_kernel_gs_base;
449 u64 msr_guest_kernel_gs_base;
450#endif
451 u32 vm_entry_controls_shadow;
452 u32 vm_exit_controls_shadow;
453 /*
454 * loaded_vmcs points to the VMCS currently used in this vcpu. For a
455 * non-nested (L1) guest, it always points to vmcs01. For a nested
456 * guest (L2), it points to a different VMCS.
457 */
458 struct loaded_vmcs vmcs01;
459 struct loaded_vmcs *loaded_vmcs;
460 bool __launched; /* temporary, used in vmx_vcpu_run */
461 struct msr_autoload {
462 unsigned nr;
463 struct vmx_msr_entry guest[NR_AUTOLOAD_MSRS];
464 struct vmx_msr_entry host[NR_AUTOLOAD_MSRS];
465 } msr_autoload;
466 struct {
467 int loaded;
468 u16 fs_sel, gs_sel, ldt_sel;
469#ifdef CONFIG_X86_64
470 u16 ds_sel, es_sel;
471#endif
472 int gs_ldt_reload_needed;
473 int fs_reload_needed;
474 u64 msr_host_bndcfgs;
475 unsigned long vmcs_host_cr4; /* May not match real cr4 */
476 } host_state;
477 struct {
478 int vm86_active;
479 ulong save_rflags;
480 struct kvm_segment segs[8];
481 } rmode;
482 struct {
483 u32 bitmask; /* 4 bits per segment (1 bit per field) */
484 struct kvm_save_segment {
485 u16 selector;
486 unsigned long base;
487 u32 limit;
488 u32 ar;
489 } seg[8];
490 } segment_cache;
491 int vpid;
492 bool emulation_required;
493
494 /* Support for vnmi-less CPUs */
495 int soft_vnmi_blocked;
496 ktime_t entry_time;
497 s64 vnmi_blocked_time;
498 u32 exit_reason;
499
500 bool rdtscp_enabled;
501
502 /* Posted interrupt descriptor */
503 struct pi_desc pi_desc;
504
505 /* Support for a guest hypervisor (nested VMX) */
506 struct nested_vmx nested;
507
508 /* Dynamic PLE window. */
509 int ple_window;
510 bool ple_window_dirty;
511};
512
513enum segment_cache_field {
514 SEG_FIELD_SEL = 0,
515 SEG_FIELD_BASE = 1,
516 SEG_FIELD_LIMIT = 2,
517 SEG_FIELD_AR = 3,
518
519 SEG_FIELD_NR = 4
520};
521
522static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
523{
524 return container_of(vcpu, struct vcpu_vmx, vcpu);
525}
526
527#define VMCS12_OFFSET(x) offsetof(struct vmcs12, x)
528#define FIELD(number, name) [number] = VMCS12_OFFSET(name)
529#define FIELD64(number, name) [number] = VMCS12_OFFSET(name), \
530 [number##_HIGH] = VMCS12_OFFSET(name)+4
531
532
533static unsigned long shadow_read_only_fields[] = {
534 /*
535 * We do NOT shadow fields that are modified when L0
536 * traps and emulates any vmx instruction (e.g. VMPTRLD,
537 * VMXON...) executed by L1.
538 * For example, VM_INSTRUCTION_ERROR is read
539 * by L1 if a vmx instruction fails (part of the error path).
540 * Note the code assumes this logic. If for some reason
541 * we start shadowing these fields then we need to
542 * force a shadow sync when L0 emulates vmx instructions
543 * (e.g. force a sync if VM_INSTRUCTION_ERROR is modified
544 * by nested_vmx_failValid)
545 */
546 VM_EXIT_REASON,
547 VM_EXIT_INTR_INFO,
548 VM_EXIT_INSTRUCTION_LEN,
549 IDT_VECTORING_INFO_FIELD,
550 IDT_VECTORING_ERROR_CODE,
551 VM_EXIT_INTR_ERROR_CODE,
552 EXIT_QUALIFICATION,
553 GUEST_LINEAR_ADDRESS,
554 GUEST_PHYSICAL_ADDRESS
555};
556static int max_shadow_read_only_fields =
557 ARRAY_SIZE(shadow_read_only_fields);
558
559static unsigned long shadow_read_write_fields[] = {
560 TPR_THRESHOLD,
561 GUEST_RIP,
562 GUEST_RSP,
563 GUEST_CR0,
564 GUEST_CR3,
565 GUEST_CR4,
566 GUEST_INTERRUPTIBILITY_INFO,
567 GUEST_RFLAGS,
568 GUEST_CS_SELECTOR,
569 GUEST_CS_AR_BYTES,
570 GUEST_CS_LIMIT,
571 GUEST_CS_BASE,
572 GUEST_ES_BASE,
573 GUEST_BNDCFGS,
574 CR0_GUEST_HOST_MASK,
575 CR0_READ_SHADOW,
576 CR4_READ_SHADOW,
577 TSC_OFFSET,
578 EXCEPTION_BITMAP,
579 CPU_BASED_VM_EXEC_CONTROL,
580 VM_ENTRY_EXCEPTION_ERROR_CODE,
581 VM_ENTRY_INTR_INFO_FIELD,
582 VM_ENTRY_INSTRUCTION_LEN,
583 VM_ENTRY_EXCEPTION_ERROR_CODE,
584 HOST_FS_BASE,
585 HOST_GS_BASE,
586 HOST_FS_SELECTOR,
587 HOST_GS_SELECTOR
588};
589static int max_shadow_read_write_fields =
590 ARRAY_SIZE(shadow_read_write_fields);
591
592static const unsigned short vmcs_field_to_offset_table[] = {
593 FIELD(VIRTUAL_PROCESSOR_ID, virtual_processor_id),
594 FIELD(GUEST_ES_SELECTOR, guest_es_selector),
595 FIELD(GUEST_CS_SELECTOR, guest_cs_selector),
596 FIELD(GUEST_SS_SELECTOR, guest_ss_selector),
597 FIELD(GUEST_DS_SELECTOR, guest_ds_selector),
598 FIELD(GUEST_FS_SELECTOR, guest_fs_selector),
599 FIELD(GUEST_GS_SELECTOR, guest_gs_selector),
600 FIELD(GUEST_LDTR_SELECTOR, guest_ldtr_selector),
601 FIELD(GUEST_TR_SELECTOR, guest_tr_selector),
602 FIELD(HOST_ES_SELECTOR, host_es_selector),
603 FIELD(HOST_CS_SELECTOR, host_cs_selector),
604 FIELD(HOST_SS_SELECTOR, host_ss_selector),
605 FIELD(HOST_DS_SELECTOR, host_ds_selector),
606 FIELD(HOST_FS_SELECTOR, host_fs_selector),
607 FIELD(HOST_GS_SELECTOR, host_gs_selector),
608 FIELD(HOST_TR_SELECTOR, host_tr_selector),
609 FIELD64(IO_BITMAP_A, io_bitmap_a),
610 FIELD64(IO_BITMAP_B, io_bitmap_b),
611 FIELD64(MSR_BITMAP, msr_bitmap),
612 FIELD64(VM_EXIT_MSR_STORE_ADDR, vm_exit_msr_store_addr),
613 FIELD64(VM_EXIT_MSR_LOAD_ADDR, vm_exit_msr_load_addr),
614 FIELD64(VM_ENTRY_MSR_LOAD_ADDR, vm_entry_msr_load_addr),
615 FIELD64(TSC_OFFSET, tsc_offset),
616 FIELD64(VIRTUAL_APIC_PAGE_ADDR, virtual_apic_page_addr),
617 FIELD64(APIC_ACCESS_ADDR, apic_access_addr),
618 FIELD64(EPT_POINTER, ept_pointer),
619 FIELD64(GUEST_PHYSICAL_ADDRESS, guest_physical_address),
620 FIELD64(VMCS_LINK_POINTER, vmcs_link_pointer),
621 FIELD64(GUEST_IA32_DEBUGCTL, guest_ia32_debugctl),
622 FIELD64(GUEST_IA32_PAT, guest_ia32_pat),
623 FIELD64(GUEST_IA32_EFER, guest_ia32_efer),
624 FIELD64(GUEST_IA32_PERF_GLOBAL_CTRL, guest_ia32_perf_global_ctrl),
625 FIELD64(GUEST_PDPTR0, guest_pdptr0),
626 FIELD64(GUEST_PDPTR1, guest_pdptr1),
627 FIELD64(GUEST_PDPTR2, guest_pdptr2),
628 FIELD64(GUEST_PDPTR3, guest_pdptr3),
629 FIELD64(GUEST_BNDCFGS, guest_bndcfgs),
630 FIELD64(HOST_IA32_PAT, host_ia32_pat),
631 FIELD64(HOST_IA32_EFER, host_ia32_efer),
632 FIELD64(HOST_IA32_PERF_GLOBAL_CTRL, host_ia32_perf_global_ctrl),
633 FIELD(PIN_BASED_VM_EXEC_CONTROL, pin_based_vm_exec_control),
634 FIELD(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control),
635 FIELD(EXCEPTION_BITMAP, exception_bitmap),
636 FIELD(PAGE_FAULT_ERROR_CODE_MASK, page_fault_error_code_mask),
637 FIELD(PAGE_FAULT_ERROR_CODE_MATCH, page_fault_error_code_match),
638 FIELD(CR3_TARGET_COUNT, cr3_target_count),
639 FIELD(VM_EXIT_CONTROLS, vm_exit_controls),
640 FIELD(VM_EXIT_MSR_STORE_COUNT, vm_exit_msr_store_count),
641 FIELD(VM_EXIT_MSR_LOAD_COUNT, vm_exit_msr_load_count),
642 FIELD(VM_ENTRY_CONTROLS, vm_entry_controls),
643 FIELD(VM_ENTRY_MSR_LOAD_COUNT, vm_entry_msr_load_count),
644 FIELD(VM_ENTRY_INTR_INFO_FIELD, vm_entry_intr_info_field),
645 FIELD(VM_ENTRY_EXCEPTION_ERROR_CODE, vm_entry_exception_error_code),
646 FIELD(VM_ENTRY_INSTRUCTION_LEN, vm_entry_instruction_len),
647 FIELD(TPR_THRESHOLD, tpr_threshold),
648 FIELD(SECONDARY_VM_EXEC_CONTROL, secondary_vm_exec_control),
649 FIELD(VM_INSTRUCTION_ERROR, vm_instruction_error),
650 FIELD(VM_EXIT_REASON, vm_exit_reason),
651 FIELD(VM_EXIT_INTR_INFO, vm_exit_intr_info),
652 FIELD(VM_EXIT_INTR_ERROR_CODE, vm_exit_intr_error_code),
653 FIELD(IDT_VECTORING_INFO_FIELD, idt_vectoring_info_field),
654 FIELD(IDT_VECTORING_ERROR_CODE, idt_vectoring_error_code),
655 FIELD(VM_EXIT_INSTRUCTION_LEN, vm_exit_instruction_len),
656 FIELD(VMX_INSTRUCTION_INFO, vmx_instruction_info),
657 FIELD(GUEST_ES_LIMIT, guest_es_limit),
658 FIELD(GUEST_CS_LIMIT, guest_cs_limit),
659 FIELD(GUEST_SS_LIMIT, guest_ss_limit),
660 FIELD(GUEST_DS_LIMIT, guest_ds_limit),
661 FIELD(GUEST_FS_LIMIT, guest_fs_limit),
662 FIELD(GUEST_GS_LIMIT, guest_gs_limit),
663 FIELD(GUEST_LDTR_LIMIT, guest_ldtr_limit),
664 FIELD(GUEST_TR_LIMIT, guest_tr_limit),
665 FIELD(GUEST_GDTR_LIMIT, guest_gdtr_limit),
666 FIELD(GUEST_IDTR_LIMIT, guest_idtr_limit),
667 FIELD(GUEST_ES_AR_BYTES, guest_es_ar_bytes),
668 FIELD(GUEST_CS_AR_BYTES, guest_cs_ar_bytes),
669 FIELD(GUEST_SS_AR_BYTES, guest_ss_ar_bytes),
670 FIELD(GUEST_DS_AR_BYTES, guest_ds_ar_bytes),
671 FIELD(GUEST_FS_AR_BYTES, guest_fs_ar_bytes),
672 FIELD(GUEST_GS_AR_BYTES, guest_gs_ar_bytes),
673 FIELD(GUEST_LDTR_AR_BYTES, guest_ldtr_ar_bytes),
674 FIELD(GUEST_TR_AR_BYTES, guest_tr_ar_bytes),
675 FIELD(GUEST_INTERRUPTIBILITY_INFO, guest_interruptibility_info),
676 FIELD(GUEST_ACTIVITY_STATE, guest_activity_state),
677 FIELD(GUEST_SYSENTER_CS, guest_sysenter_cs),
678 FIELD(HOST_IA32_SYSENTER_CS, host_ia32_sysenter_cs),
679 FIELD(VMX_PREEMPTION_TIMER_VALUE, vmx_preemption_timer_value),
680 FIELD(CR0_GUEST_HOST_MASK, cr0_guest_host_mask),
681 FIELD(CR4_GUEST_HOST_MASK, cr4_guest_host_mask),
682 FIELD(CR0_READ_SHADOW, cr0_read_shadow),
683 FIELD(CR4_READ_SHADOW, cr4_read_shadow),
684 FIELD(CR3_TARGET_VALUE0, cr3_target_value0),
685 FIELD(CR3_TARGET_VALUE1, cr3_target_value1),
686 FIELD(CR3_TARGET_VALUE2, cr3_target_value2),
687 FIELD(CR3_TARGET_VALUE3, cr3_target_value3),
688 FIELD(EXIT_QUALIFICATION, exit_qualification),
689 FIELD(GUEST_LINEAR_ADDRESS, guest_linear_address),
690 FIELD(GUEST_CR0, guest_cr0),
691 FIELD(GUEST_CR3, guest_cr3),
692 FIELD(GUEST_CR4, guest_cr4),
693 FIELD(GUEST_ES_BASE, guest_es_base),
694 FIELD(GUEST_CS_BASE, guest_cs_base),
695 FIELD(GUEST_SS_BASE, guest_ss_base),
696 FIELD(GUEST_DS_BASE, guest_ds_base),
697 FIELD(GUEST_FS_BASE, guest_fs_base),
698 FIELD(GUEST_GS_BASE, guest_gs_base),
699 FIELD(GUEST_LDTR_BASE, guest_ldtr_base),
700 FIELD(GUEST_TR_BASE, guest_tr_base),
701 FIELD(GUEST_GDTR_BASE, guest_gdtr_base),
702 FIELD(GUEST_IDTR_BASE, guest_idtr_base),
703 FIELD(GUEST_DR7, guest_dr7),
704 FIELD(GUEST_RSP, guest_rsp),
705 FIELD(GUEST_RIP, guest_rip),
706 FIELD(GUEST_RFLAGS, guest_rflags),
707 FIELD(GUEST_PENDING_DBG_EXCEPTIONS, guest_pending_dbg_exceptions),
708 FIELD(GUEST_SYSENTER_ESP, guest_sysenter_esp),
709 FIELD(GUEST_SYSENTER_EIP, guest_sysenter_eip),
710 FIELD(HOST_CR0, host_cr0),
711 FIELD(HOST_CR3, host_cr3),
712 FIELD(HOST_CR4, host_cr4),
713 FIELD(HOST_FS_BASE, host_fs_base),
714 FIELD(HOST_GS_BASE, host_gs_base),
715 FIELD(HOST_TR_BASE, host_tr_base),
716 FIELD(HOST_GDTR_BASE, host_gdtr_base),
717 FIELD(HOST_IDTR_BASE, host_idtr_base),
718 FIELD(HOST_IA32_SYSENTER_ESP, host_ia32_sysenter_esp),
719 FIELD(HOST_IA32_SYSENTER_EIP, host_ia32_sysenter_eip),
720 FIELD(HOST_RSP, host_rsp),
721 FIELD(HOST_RIP, host_rip),
722};
723static const int max_vmcs_field = ARRAY_SIZE(vmcs_field_to_offset_table);
724
725static inline short vmcs_field_to_offset(unsigned long field)
726{
727 if (field >= max_vmcs_field || vmcs_field_to_offset_table[field] == 0)
728 return -1;
729 return vmcs_field_to_offset_table[field];
730}
731
732static inline struct vmcs12 *get_vmcs12(struct kvm_vcpu *vcpu)
733{
734 return to_vmx(vcpu)->nested.current_vmcs12;
735}
736
737static struct page *nested_get_page(struct kvm_vcpu *vcpu, gpa_t addr)
738{
739 struct page *page = gfn_to_page(vcpu->kvm, addr >> PAGE_SHIFT);
740 if (is_error_page(page))
741 return NULL;
742
743 return page;
744}
745
746static void nested_release_page(struct page *page)
747{
748 kvm_release_page_dirty(page);
749}
750
751static void nested_release_page_clean(struct page *page)
752{
753 kvm_release_page_clean(page);
754}
755
756static unsigned long nested_ept_get_cr3(struct kvm_vcpu *vcpu);
757static u64 construct_eptp(unsigned long root_hpa);
758static void kvm_cpu_vmxon(u64 addr);
759static void kvm_cpu_vmxoff(void);
760static bool vmx_mpx_supported(void);
761static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr);
762static void vmx_set_segment(struct kvm_vcpu *vcpu,
763 struct kvm_segment *var, int seg);
764static void vmx_get_segment(struct kvm_vcpu *vcpu,
765 struct kvm_segment *var, int seg);
766static bool guest_state_valid(struct kvm_vcpu *vcpu);
767static u32 vmx_segment_access_rights(struct kvm_segment *var);
768static void vmx_sync_pir_to_irr_dummy(struct kvm_vcpu *vcpu);
769static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx);
770static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx);
771static int alloc_identity_pagetable(struct kvm *kvm);
772
773static DEFINE_PER_CPU(struct vmcs *, vmxarea);
774static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
775/*
776 * We maintain a per-CPU linked-list of VMCS loaded on that CPU. This is needed
777 * when a CPU is brought down, and we need to VMCLEAR all VMCSs loaded on it.
778 */
779static DEFINE_PER_CPU(struct list_head, loaded_vmcss_on_cpu);
780static DEFINE_PER_CPU(struct desc_ptr, host_gdt);
781
782static unsigned long *vmx_io_bitmap_a;
783static unsigned long *vmx_io_bitmap_b;
784static unsigned long *vmx_msr_bitmap_legacy;
785static unsigned long *vmx_msr_bitmap_longmode;
786static unsigned long *vmx_msr_bitmap_legacy_x2apic;
787static unsigned long *vmx_msr_bitmap_longmode_x2apic;
788static unsigned long *vmx_vmread_bitmap;
789static unsigned long *vmx_vmwrite_bitmap;
790
791static bool cpu_has_load_ia32_efer;
792static bool cpu_has_load_perf_global_ctrl;
793
794static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
795static DEFINE_SPINLOCK(vmx_vpid_lock);
796
797static struct vmcs_config {
798 int size;
799 int order;
800 u32 revision_id;
801 u32 pin_based_exec_ctrl;
802 u32 cpu_based_exec_ctrl;
803 u32 cpu_based_2nd_exec_ctrl;
804 u32 vmexit_ctrl;
805 u32 vmentry_ctrl;
806} vmcs_config;
807
808static struct vmx_capability {
809 u32 ept;
810 u32 vpid;
811} vmx_capability;
812
813#define VMX_SEGMENT_FIELD(seg) \
814 [VCPU_SREG_##seg] = { \
815 .selector = GUEST_##seg##_SELECTOR, \
816 .base = GUEST_##seg##_BASE, \
817 .limit = GUEST_##seg##_LIMIT, \
818 .ar_bytes = GUEST_##seg##_AR_BYTES, \
819 }
820
821static const struct kvm_vmx_segment_field {
822 unsigned selector;
823 unsigned base;
824 unsigned limit;
825 unsigned ar_bytes;
826} kvm_vmx_segment_fields[] = {
827 VMX_SEGMENT_FIELD(CS),
828 VMX_SEGMENT_FIELD(DS),
829 VMX_SEGMENT_FIELD(ES),
830 VMX_SEGMENT_FIELD(FS),
831 VMX_SEGMENT_FIELD(GS),
832 VMX_SEGMENT_FIELD(SS),
833 VMX_SEGMENT_FIELD(TR),
834 VMX_SEGMENT_FIELD(LDTR),
835};
836
837static u64 host_efer;
838
839static void ept_save_pdptrs(struct kvm_vcpu *vcpu);
840
841/*
842 * Keep MSR_STAR at the end, as setup_msrs() will try to optimize it
843 * away by decrementing the array size.
844 */
845static const u32 vmx_msr_index[] = {
846#ifdef CONFIG_X86_64
847 MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
848#endif
849 MSR_EFER, MSR_TSC_AUX, MSR_STAR,
850};
851
852static inline bool is_page_fault(u32 intr_info)
853{
854 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
855 INTR_INFO_VALID_MASK)) ==
856 (INTR_TYPE_HARD_EXCEPTION | PF_VECTOR | INTR_INFO_VALID_MASK);
857}
858
859static inline bool is_no_device(u32 intr_info)
860{
861 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
862 INTR_INFO_VALID_MASK)) ==
863 (INTR_TYPE_HARD_EXCEPTION | NM_VECTOR | INTR_INFO_VALID_MASK);
864}
865
866static inline bool is_invalid_opcode(u32 intr_info)
867{
868 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
869 INTR_INFO_VALID_MASK)) ==
870 (INTR_TYPE_HARD_EXCEPTION | UD_VECTOR | INTR_INFO_VALID_MASK);
871}
872
873static inline bool is_external_interrupt(u32 intr_info)
874{
875 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
876 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
877}
878
879static inline bool is_machine_check(u32 intr_info)
880{
881 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
882 INTR_INFO_VALID_MASK)) ==
883 (INTR_TYPE_HARD_EXCEPTION | MC_VECTOR | INTR_INFO_VALID_MASK);
884}
885
886static inline bool cpu_has_vmx_msr_bitmap(void)
887{
888 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS;
889}
890
891static inline bool cpu_has_vmx_tpr_shadow(void)
892{
893 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW;
894}
895
896static inline bool vm_need_tpr_shadow(struct kvm *kvm)
897{
898 return (cpu_has_vmx_tpr_shadow()) && (irqchip_in_kernel(kvm));
899}
900
901static inline bool cpu_has_secondary_exec_ctrls(void)
902{
903 return vmcs_config.cpu_based_exec_ctrl &
904 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
905}
906
907static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
908{
909 return vmcs_config.cpu_based_2nd_exec_ctrl &
910 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
911}
912
913static inline bool cpu_has_vmx_virtualize_x2apic_mode(void)
914{
915 return vmcs_config.cpu_based_2nd_exec_ctrl &
916 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
917}
918
919static inline bool cpu_has_vmx_apic_register_virt(void)
920{
921 return vmcs_config.cpu_based_2nd_exec_ctrl &
922 SECONDARY_EXEC_APIC_REGISTER_VIRT;
923}
924
925static inline bool cpu_has_vmx_virtual_intr_delivery(void)
926{
927 return vmcs_config.cpu_based_2nd_exec_ctrl &
928 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY;
929}
930
931static inline bool cpu_has_vmx_posted_intr(void)
932{
933 return vmcs_config.pin_based_exec_ctrl & PIN_BASED_POSTED_INTR;
934}
935
936static inline bool cpu_has_vmx_apicv(void)
937{
938 return cpu_has_vmx_apic_register_virt() &&
939 cpu_has_vmx_virtual_intr_delivery() &&
940 cpu_has_vmx_posted_intr();
941}
942
943static inline bool cpu_has_vmx_flexpriority(void)
944{
945 return cpu_has_vmx_tpr_shadow() &&
946 cpu_has_vmx_virtualize_apic_accesses();
947}
948
949static inline bool cpu_has_vmx_ept_execute_only(void)
950{
951 return vmx_capability.ept & VMX_EPT_EXECUTE_ONLY_BIT;
952}
953
954static inline bool cpu_has_vmx_eptp_uncacheable(void)
955{
956 return vmx_capability.ept & VMX_EPTP_UC_BIT;
957}
958
959static inline bool cpu_has_vmx_eptp_writeback(void)
960{
961 return vmx_capability.ept & VMX_EPTP_WB_BIT;
962}
963
964static inline bool cpu_has_vmx_ept_2m_page(void)
965{
966 return vmx_capability.ept & VMX_EPT_2MB_PAGE_BIT;
967}
968
969static inline bool cpu_has_vmx_ept_1g_page(void)
970{
971 return vmx_capability.ept & VMX_EPT_1GB_PAGE_BIT;
972}
973
974static inline bool cpu_has_vmx_ept_4levels(void)
975{
976 return vmx_capability.ept & VMX_EPT_PAGE_WALK_4_BIT;
977}
978
979static inline bool cpu_has_vmx_ept_ad_bits(void)
980{
981 return vmx_capability.ept & VMX_EPT_AD_BIT;
982}
983
984static inline bool cpu_has_vmx_invept_context(void)
985{
986 return vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT;
987}
988
989static inline bool cpu_has_vmx_invept_global(void)
990{
991 return vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT;
992}
993
994static inline bool cpu_has_vmx_invvpid_single(void)
995{
996 return vmx_capability.vpid & VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT;
997}
998
999static inline bool cpu_has_vmx_invvpid_global(void)
1000{
1001 return vmx_capability.vpid & VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT;
1002}
1003
1004static inline bool cpu_has_vmx_ept(void)
1005{
1006 return vmcs_config.cpu_based_2nd_exec_ctrl &
1007 SECONDARY_EXEC_ENABLE_EPT;
1008}
1009
1010static inline bool cpu_has_vmx_unrestricted_guest(void)
1011{
1012 return vmcs_config.cpu_based_2nd_exec_ctrl &
1013 SECONDARY_EXEC_UNRESTRICTED_GUEST;
1014}
1015
1016static inline bool cpu_has_vmx_ple(void)
1017{
1018 return vmcs_config.cpu_based_2nd_exec_ctrl &
1019 SECONDARY_EXEC_PAUSE_LOOP_EXITING;
1020}
1021
1022static inline bool vm_need_virtualize_apic_accesses(struct kvm *kvm)
1023{
1024 return flexpriority_enabled && irqchip_in_kernel(kvm);
1025}
1026
1027static inline bool cpu_has_vmx_vpid(void)
1028{
1029 return vmcs_config.cpu_based_2nd_exec_ctrl &
1030 SECONDARY_EXEC_ENABLE_VPID;
1031}
1032
1033static inline bool cpu_has_vmx_rdtscp(void)
1034{
1035 return vmcs_config.cpu_based_2nd_exec_ctrl &
1036 SECONDARY_EXEC_RDTSCP;
1037}
1038
1039static inline bool cpu_has_vmx_invpcid(void)
1040{
1041 return vmcs_config.cpu_based_2nd_exec_ctrl &
1042 SECONDARY_EXEC_ENABLE_INVPCID;
1043}
1044
1045static inline bool cpu_has_virtual_nmis(void)
1046{
1047 return vmcs_config.pin_based_exec_ctrl & PIN_BASED_VIRTUAL_NMIS;
1048}
1049
1050static inline bool cpu_has_vmx_wbinvd_exit(void)
1051{
1052 return vmcs_config.cpu_based_2nd_exec_ctrl &
1053 SECONDARY_EXEC_WBINVD_EXITING;
1054}
1055
1056static inline bool cpu_has_vmx_shadow_vmcs(void)
1057{
1058 u64 vmx_msr;
1059 rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
1060 /* check if the cpu supports writing r/o exit information fields */
1061 if (!(vmx_msr & MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS))
1062 return false;
1063
1064 return vmcs_config.cpu_based_2nd_exec_ctrl &
1065 SECONDARY_EXEC_SHADOW_VMCS;
1066}
1067
1068static inline bool report_flexpriority(void)
1069{
1070 return flexpriority_enabled;
1071}
1072
1073static inline bool nested_cpu_has(struct vmcs12 *vmcs12, u32 bit)
1074{
1075 return vmcs12->cpu_based_vm_exec_control & bit;
1076}
1077
1078static inline bool nested_cpu_has2(struct vmcs12 *vmcs12, u32 bit)
1079{
1080 return (vmcs12->cpu_based_vm_exec_control &
1081 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
1082 (vmcs12->secondary_vm_exec_control & bit);
1083}
1084
1085static inline bool nested_cpu_has_virtual_nmis(struct vmcs12 *vmcs12)
1086{
1087 return vmcs12->pin_based_vm_exec_control & PIN_BASED_VIRTUAL_NMIS;
1088}
1089
1090static inline bool nested_cpu_has_preemption_timer(struct vmcs12 *vmcs12)
1091{
1092 return vmcs12->pin_based_vm_exec_control &
1093 PIN_BASED_VMX_PREEMPTION_TIMER;
1094}
1095
1096static inline int nested_cpu_has_ept(struct vmcs12 *vmcs12)
1097{
1098 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_EPT);
1099}
1100
1101static inline bool is_exception(u32 intr_info)
1102{
1103 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
1104 == (INTR_TYPE_HARD_EXCEPTION | INTR_INFO_VALID_MASK);
1105}
1106
1107static void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
1108 u32 exit_intr_info,
1109 unsigned long exit_qualification);
1110static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
1111 struct vmcs12 *vmcs12,
1112 u32 reason, unsigned long qualification);
1113
1114static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
1115{
1116 int i;
1117
1118 for (i = 0; i < vmx->nmsrs; ++i)
1119 if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
1120 return i;
1121 return -1;
1122}
1123
1124static inline void __invvpid(int ext, u16 vpid, gva_t gva)
1125{
1126 struct {
1127 u64 vpid : 16;
1128 u64 rsvd : 48;
1129 u64 gva;
1130 } operand = { vpid, 0, gva };
1131
1132 asm volatile (__ex(ASM_VMX_INVVPID)
1133 /* CF==1 or ZF==1 --> rc = -1 */
1134 "; ja 1f ; ud2 ; 1:"
1135 : : "a"(&operand), "c"(ext) : "cc", "memory");
1136}
1137
1138static inline void __invept(int ext, u64 eptp, gpa_t gpa)
1139{
1140 struct {
1141 u64 eptp, gpa;
1142 } operand = {eptp, gpa};
1143
1144 asm volatile (__ex(ASM_VMX_INVEPT)
1145 /* CF==1 or ZF==1 --> rc = -1 */
1146 "; ja 1f ; ud2 ; 1:\n"
1147 : : "a" (&operand), "c" (ext) : "cc", "memory");
1148}
1149
1150static struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
1151{
1152 int i;
1153
1154 i = __find_msr_index(vmx, msr);
1155 if (i >= 0)
1156 return &vmx->guest_msrs[i];
1157 return NULL;
1158}
1159
1160static void vmcs_clear(struct vmcs *vmcs)
1161{
1162 u64 phys_addr = __pa(vmcs);
1163 u8 error;
1164
1165 asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0"
1166 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
1167 : "cc", "memory");
1168 if (error)
1169 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
1170 vmcs, phys_addr);
1171}
1172
1173static inline void loaded_vmcs_init(struct loaded_vmcs *loaded_vmcs)
1174{
1175 vmcs_clear(loaded_vmcs->vmcs);
1176 loaded_vmcs->cpu = -1;
1177 loaded_vmcs->launched = 0;
1178}
1179
1180static void vmcs_load(struct vmcs *vmcs)
1181{
1182 u64 phys_addr = __pa(vmcs);
1183 u8 error;
1184
1185 asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0"
1186 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
1187 : "cc", "memory");
1188 if (error)
1189 printk(KERN_ERR "kvm: vmptrld %p/%llx failed\n",
1190 vmcs, phys_addr);
1191}
1192
1193#ifdef CONFIG_KEXEC
1194/*
1195 * This bitmap is used to indicate whether the vmclear
1196 * operation is enabled on all cpus. All disabled by
1197 * default.
1198 */
1199static cpumask_t crash_vmclear_enabled_bitmap = CPU_MASK_NONE;
1200
1201static inline void crash_enable_local_vmclear(int cpu)
1202{
1203 cpumask_set_cpu(cpu, &crash_vmclear_enabled_bitmap);
1204}
1205
1206static inline void crash_disable_local_vmclear(int cpu)
1207{
1208 cpumask_clear_cpu(cpu, &crash_vmclear_enabled_bitmap);
1209}
1210
1211static inline int crash_local_vmclear_enabled(int cpu)
1212{
1213 return cpumask_test_cpu(cpu, &crash_vmclear_enabled_bitmap);
1214}
1215
1216static void crash_vmclear_local_loaded_vmcss(void)
1217{
1218 int cpu = raw_smp_processor_id();
1219 struct loaded_vmcs *v;
1220
1221 if (!crash_local_vmclear_enabled(cpu))
1222 return;
1223
1224 list_for_each_entry(v, &per_cpu(loaded_vmcss_on_cpu, cpu),
1225 loaded_vmcss_on_cpu_link)
1226 vmcs_clear(v->vmcs);
1227}
1228#else
1229static inline void crash_enable_local_vmclear(int cpu) { }
1230static inline void crash_disable_local_vmclear(int cpu) { }
1231#endif /* CONFIG_KEXEC */
1232
1233static void __loaded_vmcs_clear(void *arg)
1234{
1235 struct loaded_vmcs *loaded_vmcs = arg;
1236 int cpu = raw_smp_processor_id();
1237
1238 if (loaded_vmcs->cpu != cpu)
1239 return; /* vcpu migration can race with cpu offline */
1240 if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
1241 per_cpu(current_vmcs, cpu) = NULL;
1242 crash_disable_local_vmclear(cpu);
1243 list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
1244
1245 /*
1246 * we should ensure updating loaded_vmcs->loaded_vmcss_on_cpu_link
1247 * is before setting loaded_vmcs->vcpu to -1 which is done in
1248 * loaded_vmcs_init. Otherwise, other cpu can see vcpu = -1 fist
1249 * then adds the vmcs into percpu list before it is deleted.
1250 */
1251 smp_wmb();
1252
1253 loaded_vmcs_init(loaded_vmcs);
1254 crash_enable_local_vmclear(cpu);
1255}
1256
1257static void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
1258{
1259 int cpu = loaded_vmcs->cpu;
1260
1261 if (cpu != -1)
1262 smp_call_function_single(cpu,
1263 __loaded_vmcs_clear, loaded_vmcs, 1);
1264}
1265
1266static inline void vpid_sync_vcpu_single(struct vcpu_vmx *vmx)
1267{
1268 if (vmx->vpid == 0)
1269 return;
1270
1271 if (cpu_has_vmx_invvpid_single())
1272 __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vmx->vpid, 0);
1273}
1274
1275static inline void vpid_sync_vcpu_global(void)
1276{
1277 if (cpu_has_vmx_invvpid_global())
1278 __invvpid(VMX_VPID_EXTENT_ALL_CONTEXT, 0, 0);
1279}
1280
1281static inline void vpid_sync_context(struct vcpu_vmx *vmx)
1282{
1283 if (cpu_has_vmx_invvpid_single())
1284 vpid_sync_vcpu_single(vmx);
1285 else
1286 vpid_sync_vcpu_global();
1287}
1288
1289static inline void ept_sync_global(void)
1290{
1291 if (cpu_has_vmx_invept_global())
1292 __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0);
1293}
1294
1295static inline void ept_sync_context(u64 eptp)
1296{
1297 if (enable_ept) {
1298 if (cpu_has_vmx_invept_context())
1299 __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0);
1300 else
1301 ept_sync_global();
1302 }
1303}
1304
1305static __always_inline unsigned long vmcs_readl(unsigned long field)
1306{
1307 unsigned long value;
1308
1309 asm volatile (__ex_clear(ASM_VMX_VMREAD_RDX_RAX, "%0")
1310 : "=a"(value) : "d"(field) : "cc");
1311 return value;
1312}
1313
1314static __always_inline u16 vmcs_read16(unsigned long field)
1315{
1316 return vmcs_readl(field);
1317}
1318
1319static __always_inline u32 vmcs_read32(unsigned long field)
1320{
1321 return vmcs_readl(field);
1322}
1323
1324static __always_inline u64 vmcs_read64(unsigned long field)
1325{
1326#ifdef CONFIG_X86_64
1327 return vmcs_readl(field);
1328#else
1329 return vmcs_readl(field) | ((u64)vmcs_readl(field+1) << 32);
1330#endif
1331}
1332
1333static noinline void vmwrite_error(unsigned long field, unsigned long value)
1334{
1335 printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
1336 field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
1337 dump_stack();
1338}
1339
1340static void vmcs_writel(unsigned long field, unsigned long value)
1341{
1342 u8 error;
1343
1344 asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0"
1345 : "=q"(error) : "a"(value), "d"(field) : "cc");
1346 if (unlikely(error))
1347 vmwrite_error(field, value);
1348}
1349
1350static void vmcs_write16(unsigned long field, u16 value)
1351{
1352 vmcs_writel(field, value);
1353}
1354
1355static void vmcs_write32(unsigned long field, u32 value)
1356{
1357 vmcs_writel(field, value);
1358}
1359
1360static void vmcs_write64(unsigned long field, u64 value)
1361{
1362 vmcs_writel(field, value);
1363#ifndef CONFIG_X86_64
1364 asm volatile ("");
1365 vmcs_writel(field+1, value >> 32);
1366#endif
1367}
1368
1369static void vmcs_clear_bits(unsigned long field, u32 mask)
1370{
1371 vmcs_writel(field, vmcs_readl(field) & ~mask);
1372}
1373
1374static void vmcs_set_bits(unsigned long field, u32 mask)
1375{
1376 vmcs_writel(field, vmcs_readl(field) | mask);
1377}
1378
1379static inline void vm_entry_controls_init(struct vcpu_vmx *vmx, u32 val)
1380{
1381 vmcs_write32(VM_ENTRY_CONTROLS, val);
1382 vmx->vm_entry_controls_shadow = val;
1383}
1384
1385static inline void vm_entry_controls_set(struct vcpu_vmx *vmx, u32 val)
1386{
1387 if (vmx->vm_entry_controls_shadow != val)
1388 vm_entry_controls_init(vmx, val);
1389}
1390
1391static inline u32 vm_entry_controls_get(struct vcpu_vmx *vmx)
1392{
1393 return vmx->vm_entry_controls_shadow;
1394}
1395
1396
1397static inline void vm_entry_controls_setbit(struct vcpu_vmx *vmx, u32 val)
1398{
1399 vm_entry_controls_set(vmx, vm_entry_controls_get(vmx) | val);
1400}
1401
1402static inline void vm_entry_controls_clearbit(struct vcpu_vmx *vmx, u32 val)
1403{
1404 vm_entry_controls_set(vmx, vm_entry_controls_get(vmx) & ~val);
1405}
1406
1407static inline void vm_exit_controls_init(struct vcpu_vmx *vmx, u32 val)
1408{
1409 vmcs_write32(VM_EXIT_CONTROLS, val);
1410 vmx->vm_exit_controls_shadow = val;
1411}
1412
1413static inline void vm_exit_controls_set(struct vcpu_vmx *vmx, u32 val)
1414{
1415 if (vmx->vm_exit_controls_shadow != val)
1416 vm_exit_controls_init(vmx, val);
1417}
1418
1419static inline u32 vm_exit_controls_get(struct vcpu_vmx *vmx)
1420{
1421 return vmx->vm_exit_controls_shadow;
1422}
1423
1424
1425static inline void vm_exit_controls_setbit(struct vcpu_vmx *vmx, u32 val)
1426{
1427 vm_exit_controls_set(vmx, vm_exit_controls_get(vmx) | val);
1428}
1429
1430static inline void vm_exit_controls_clearbit(struct vcpu_vmx *vmx, u32 val)
1431{
1432 vm_exit_controls_set(vmx, vm_exit_controls_get(vmx) & ~val);
1433}
1434
1435static void vmx_segment_cache_clear(struct vcpu_vmx *vmx)
1436{
1437 vmx->segment_cache.bitmask = 0;
1438}
1439
1440static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
1441 unsigned field)
1442{
1443 bool ret;
1444 u32 mask = 1 << (seg * SEG_FIELD_NR + field);
1445
1446 if (!(vmx->vcpu.arch.regs_avail & (1 << VCPU_EXREG_SEGMENTS))) {
1447 vmx->vcpu.arch.regs_avail |= (1 << VCPU_EXREG_SEGMENTS);
1448 vmx->segment_cache.bitmask = 0;
1449 }
1450 ret = vmx->segment_cache.bitmask & mask;
1451 vmx->segment_cache.bitmask |= mask;
1452 return ret;
1453}
1454
1455static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
1456{
1457 u16 *p = &vmx->segment_cache.seg[seg].selector;
1458
1459 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
1460 *p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
1461 return *p;
1462}
1463
1464static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
1465{
1466 ulong *p = &vmx->segment_cache.seg[seg].base;
1467
1468 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
1469 *p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
1470 return *p;
1471}
1472
1473static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
1474{
1475 u32 *p = &vmx->segment_cache.seg[seg].limit;
1476
1477 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
1478 *p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
1479 return *p;
1480}
1481
1482static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
1483{
1484 u32 *p = &vmx->segment_cache.seg[seg].ar;
1485
1486 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
1487 *p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
1488 return *p;
1489}
1490
1491static void update_exception_bitmap(struct kvm_vcpu *vcpu)
1492{
1493 u32 eb;
1494
1495 eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
1496 (1u << NM_VECTOR) | (1u << DB_VECTOR);
1497 if ((vcpu->guest_debug &
1498 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
1499 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
1500 eb |= 1u << BP_VECTOR;
1501 if (to_vmx(vcpu)->rmode.vm86_active)
1502 eb = ~0;
1503 if (enable_ept)
1504 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
1505 if (vcpu->fpu_active)
1506 eb &= ~(1u << NM_VECTOR);
1507
1508 /* When we are running a nested L2 guest and L1 specified for it a
1509 * certain exception bitmap, we must trap the same exceptions and pass
1510 * them to L1. When running L2, we will only handle the exceptions
1511 * specified above if L1 did not want them.
1512 */
1513 if (is_guest_mode(vcpu))
1514 eb |= get_vmcs12(vcpu)->exception_bitmap;
1515
1516 vmcs_write32(EXCEPTION_BITMAP, eb);
1517}
1518
1519static void clear_atomic_switch_msr_special(struct vcpu_vmx *vmx,
1520 unsigned long entry, unsigned long exit)
1521{
1522 vm_entry_controls_clearbit(vmx, entry);
1523 vm_exit_controls_clearbit(vmx, exit);
1524}
1525
1526static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
1527{
1528 unsigned i;
1529 struct msr_autoload *m = &vmx->msr_autoload;
1530
1531 switch (msr) {
1532 case MSR_EFER:
1533 if (cpu_has_load_ia32_efer) {
1534 clear_atomic_switch_msr_special(vmx,
1535 VM_ENTRY_LOAD_IA32_EFER,
1536 VM_EXIT_LOAD_IA32_EFER);
1537 return;
1538 }
1539 break;
1540 case MSR_CORE_PERF_GLOBAL_CTRL:
1541 if (cpu_has_load_perf_global_ctrl) {
1542 clear_atomic_switch_msr_special(vmx,
1543 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1544 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
1545 return;
1546 }
1547 break;
1548 }
1549
1550 for (i = 0; i < m->nr; ++i)
1551 if (m->guest[i].index == msr)
1552 break;
1553
1554 if (i == m->nr)
1555 return;
1556 --m->nr;
1557 m->guest[i] = m->guest[m->nr];
1558 m->host[i] = m->host[m->nr];
1559 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1560 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1561}
1562
1563static void add_atomic_switch_msr_special(struct vcpu_vmx *vmx,
1564 unsigned long entry, unsigned long exit,
1565 unsigned long guest_val_vmcs, unsigned long host_val_vmcs,
1566 u64 guest_val, u64 host_val)
1567{
1568 vmcs_write64(guest_val_vmcs, guest_val);
1569 vmcs_write64(host_val_vmcs, host_val);
1570 vm_entry_controls_setbit(vmx, entry);
1571 vm_exit_controls_setbit(vmx, exit);
1572}
1573
1574static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
1575 u64 guest_val, u64 host_val)
1576{
1577 unsigned i;
1578 struct msr_autoload *m = &vmx->msr_autoload;
1579
1580 switch (msr) {
1581 case MSR_EFER:
1582 if (cpu_has_load_ia32_efer) {
1583 add_atomic_switch_msr_special(vmx,
1584 VM_ENTRY_LOAD_IA32_EFER,
1585 VM_EXIT_LOAD_IA32_EFER,
1586 GUEST_IA32_EFER,
1587 HOST_IA32_EFER,
1588 guest_val, host_val);
1589 return;
1590 }
1591 break;
1592 case MSR_CORE_PERF_GLOBAL_CTRL:
1593 if (cpu_has_load_perf_global_ctrl) {
1594 add_atomic_switch_msr_special(vmx,
1595 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1596 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL,
1597 GUEST_IA32_PERF_GLOBAL_CTRL,
1598 HOST_IA32_PERF_GLOBAL_CTRL,
1599 guest_val, host_val);
1600 return;
1601 }
1602 break;
1603 }
1604
1605 for (i = 0; i < m->nr; ++i)
1606 if (m->guest[i].index == msr)
1607 break;
1608
1609 if (i == NR_AUTOLOAD_MSRS) {
1610 printk_once(KERN_WARNING "Not enough msr switch entries. "
1611 "Can't add msr %x\n", msr);
1612 return;
1613 } else if (i == m->nr) {
1614 ++m->nr;
1615 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1616 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1617 }
1618
1619 m->guest[i].index = msr;
1620 m->guest[i].value = guest_val;
1621 m->host[i].index = msr;
1622 m->host[i].value = host_val;
1623}
1624
1625static void reload_tss(void)
1626{
1627 /*
1628 * VT restores TR but not its size. Useless.
1629 */
1630 struct desc_ptr *gdt = this_cpu_ptr(&host_gdt);
1631 struct desc_struct *descs;
1632
1633 descs = (void *)gdt->address;
1634 descs[GDT_ENTRY_TSS].type = 9; /* available TSS */
1635 load_TR_desc();
1636}
1637
1638static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
1639{
1640 u64 guest_efer;
1641 u64 ignore_bits;
1642
1643 guest_efer = vmx->vcpu.arch.efer;
1644
1645 /*
1646 * NX is emulated; LMA and LME handled by hardware; SCE meaningless
1647 * outside long mode
1648 */
1649 ignore_bits = EFER_NX | EFER_SCE;
1650#ifdef CONFIG_X86_64
1651 ignore_bits |= EFER_LMA | EFER_LME;
1652 /* SCE is meaningful only in long mode on Intel */
1653 if (guest_efer & EFER_LMA)
1654 ignore_bits &= ~(u64)EFER_SCE;
1655#endif
1656 guest_efer &= ~ignore_bits;
1657 guest_efer |= host_efer & ignore_bits;
1658 vmx->guest_msrs[efer_offset].data = guest_efer;
1659 vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
1660
1661 clear_atomic_switch_msr(vmx, MSR_EFER);
1662 /* On ept, can't emulate nx, and must switch nx atomically */
1663 if (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX)) {
1664 guest_efer = vmx->vcpu.arch.efer;
1665 if (!(guest_efer & EFER_LMA))
1666 guest_efer &= ~EFER_LME;
1667 add_atomic_switch_msr(vmx, MSR_EFER, guest_efer, host_efer);
1668 return false;
1669 }
1670
1671 return true;
1672}
1673
1674static unsigned long segment_base(u16 selector)
1675{
1676 struct desc_ptr *gdt = this_cpu_ptr(&host_gdt);
1677 struct desc_struct *d;
1678 unsigned long table_base;
1679 unsigned long v;
1680
1681 if (!(selector & ~3))
1682 return 0;
1683
1684 table_base = gdt->address;
1685
1686 if (selector & 4) { /* from ldt */
1687 u16 ldt_selector = kvm_read_ldt();
1688
1689 if (!(ldt_selector & ~3))
1690 return 0;
1691
1692 table_base = segment_base(ldt_selector);
1693 }
1694 d = (struct desc_struct *)(table_base + (selector & ~7));
1695 v = get_desc_base(d);
1696#ifdef CONFIG_X86_64
1697 if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
1698 v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
1699#endif
1700 return v;
1701}
1702
1703static inline unsigned long kvm_read_tr_base(void)
1704{
1705 u16 tr;
1706 asm("str %0" : "=g"(tr));
1707 return segment_base(tr);
1708}
1709
1710static void vmx_save_host_state(struct kvm_vcpu *vcpu)
1711{
1712 struct vcpu_vmx *vmx = to_vmx(vcpu);
1713 int i;
1714
1715 if (vmx->host_state.loaded)
1716 return;
1717
1718 vmx->host_state.loaded = 1;
1719 /*
1720 * Set host fs and gs selectors. Unfortunately, 22.2.3 does not
1721 * allow segment selectors with cpl > 0 or ti == 1.
1722 */
1723 vmx->host_state.ldt_sel = kvm_read_ldt();
1724 vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
1725 savesegment(fs, vmx->host_state.fs_sel);
1726 if (!(vmx->host_state.fs_sel & 7)) {
1727 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
1728 vmx->host_state.fs_reload_needed = 0;
1729 } else {
1730 vmcs_write16(HOST_FS_SELECTOR, 0);
1731 vmx->host_state.fs_reload_needed = 1;
1732 }
1733 savesegment(gs, vmx->host_state.gs_sel);
1734 if (!(vmx->host_state.gs_sel & 7))
1735 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
1736 else {
1737 vmcs_write16(HOST_GS_SELECTOR, 0);
1738 vmx->host_state.gs_ldt_reload_needed = 1;
1739 }
1740
1741#ifdef CONFIG_X86_64
1742 savesegment(ds, vmx->host_state.ds_sel);
1743 savesegment(es, vmx->host_state.es_sel);
1744#endif
1745
1746#ifdef CONFIG_X86_64
1747 vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
1748 vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
1749#else
1750 vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
1751 vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
1752#endif
1753
1754#ifdef CONFIG_X86_64
1755 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1756 if (is_long_mode(&vmx->vcpu))
1757 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1758#endif
1759 if (boot_cpu_has(X86_FEATURE_MPX))
1760 rdmsrl(MSR_IA32_BNDCFGS, vmx->host_state.msr_host_bndcfgs);
1761 for (i = 0; i < vmx->save_nmsrs; ++i)
1762 kvm_set_shared_msr(vmx->guest_msrs[i].index,
1763 vmx->guest_msrs[i].data,
1764 vmx->guest_msrs[i].mask);
1765}
1766
1767static void __vmx_load_host_state(struct vcpu_vmx *vmx)
1768{
1769 if (!vmx->host_state.loaded)
1770 return;
1771
1772 ++vmx->vcpu.stat.host_state_reload;
1773 vmx->host_state.loaded = 0;
1774#ifdef CONFIG_X86_64
1775 if (is_long_mode(&vmx->vcpu))
1776 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1777#endif
1778 if (vmx->host_state.gs_ldt_reload_needed) {
1779 kvm_load_ldt(vmx->host_state.ldt_sel);
1780#ifdef CONFIG_X86_64
1781 load_gs_index(vmx->host_state.gs_sel);
1782#else
1783 loadsegment(gs, vmx->host_state.gs_sel);
1784#endif
1785 }
1786 if (vmx->host_state.fs_reload_needed)
1787 loadsegment(fs, vmx->host_state.fs_sel);
1788#ifdef CONFIG_X86_64
1789 if (unlikely(vmx->host_state.ds_sel | vmx->host_state.es_sel)) {
1790 loadsegment(ds, vmx->host_state.ds_sel);
1791 loadsegment(es, vmx->host_state.es_sel);
1792 }
1793#endif
1794 reload_tss();
1795#ifdef CONFIG_X86_64
1796 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1797#endif
1798 if (vmx->host_state.msr_host_bndcfgs)
1799 wrmsrl(MSR_IA32_BNDCFGS, vmx->host_state.msr_host_bndcfgs);
1800 /*
1801 * If the FPU is not active (through the host task or
1802 * the guest vcpu), then restore the cr0.TS bit.
1803 */
1804 if (!user_has_fpu() && !vmx->vcpu.guest_fpu_loaded)
1805 stts();
1806 load_gdt(this_cpu_ptr(&host_gdt));
1807}
1808
1809static void vmx_load_host_state(struct vcpu_vmx *vmx)
1810{
1811 preempt_disable();
1812 __vmx_load_host_state(vmx);
1813 preempt_enable();
1814}
1815
1816/*
1817 * Switches to specified vcpu, until a matching vcpu_put(), but assumes
1818 * vcpu mutex is already taken.
1819 */
1820static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1821{
1822 struct vcpu_vmx *vmx = to_vmx(vcpu);
1823 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
1824
1825 if (!vmm_exclusive)
1826 kvm_cpu_vmxon(phys_addr);
1827 else if (vmx->loaded_vmcs->cpu != cpu)
1828 loaded_vmcs_clear(vmx->loaded_vmcs);
1829
1830 if (per_cpu(current_vmcs, cpu) != vmx->loaded_vmcs->vmcs) {
1831 per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
1832 vmcs_load(vmx->loaded_vmcs->vmcs);
1833 }
1834
1835 if (vmx->loaded_vmcs->cpu != cpu) {
1836 struct desc_ptr *gdt = this_cpu_ptr(&host_gdt);
1837 unsigned long sysenter_esp;
1838
1839 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1840 local_irq_disable();
1841 crash_disable_local_vmclear(cpu);
1842
1843 /*
1844 * Read loaded_vmcs->cpu should be before fetching
1845 * loaded_vmcs->loaded_vmcss_on_cpu_link.
1846 * See the comments in __loaded_vmcs_clear().
1847 */
1848 smp_rmb();
1849
1850 list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
1851 &per_cpu(loaded_vmcss_on_cpu, cpu));
1852 crash_enable_local_vmclear(cpu);
1853 local_irq_enable();
1854
1855 /*
1856 * Linux uses per-cpu TSS and GDT, so set these when switching
1857 * processors.
1858 */
1859 vmcs_writel(HOST_TR_BASE, kvm_read_tr_base()); /* 22.2.4 */
1860 vmcs_writel(HOST_GDTR_BASE, gdt->address); /* 22.2.4 */
1861
1862 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
1863 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
1864 vmx->loaded_vmcs->cpu = cpu;
1865 }
1866}
1867
1868static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
1869{
1870 __vmx_load_host_state(to_vmx(vcpu));
1871 if (!vmm_exclusive) {
1872 __loaded_vmcs_clear(to_vmx(vcpu)->loaded_vmcs);
1873 vcpu->cpu = -1;
1874 kvm_cpu_vmxoff();
1875 }
1876}
1877
1878static void vmx_fpu_activate(struct kvm_vcpu *vcpu)
1879{
1880 ulong cr0;
1881
1882 if (vcpu->fpu_active)
1883 return;
1884 vcpu->fpu_active = 1;
1885 cr0 = vmcs_readl(GUEST_CR0);
1886 cr0 &= ~(X86_CR0_TS | X86_CR0_MP);
1887 cr0 |= kvm_read_cr0_bits(vcpu, X86_CR0_TS | X86_CR0_MP);
1888 vmcs_writel(GUEST_CR0, cr0);
1889 update_exception_bitmap(vcpu);
1890 vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
1891 if (is_guest_mode(vcpu))
1892 vcpu->arch.cr0_guest_owned_bits &=
1893 ~get_vmcs12(vcpu)->cr0_guest_host_mask;
1894 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
1895}
1896
1897static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu);
1898
1899/*
1900 * Return the cr0 value that a nested guest would read. This is a combination
1901 * of the real cr0 used to run the guest (guest_cr0), and the bits shadowed by
1902 * its hypervisor (cr0_read_shadow).
1903 */
1904static inline unsigned long nested_read_cr0(struct vmcs12 *fields)
1905{
1906 return (fields->guest_cr0 & ~fields->cr0_guest_host_mask) |
1907 (fields->cr0_read_shadow & fields->cr0_guest_host_mask);
1908}
1909static inline unsigned long nested_read_cr4(struct vmcs12 *fields)
1910{
1911 return (fields->guest_cr4 & ~fields->cr4_guest_host_mask) |
1912 (fields->cr4_read_shadow & fields->cr4_guest_host_mask);
1913}
1914
1915static void vmx_fpu_deactivate(struct kvm_vcpu *vcpu)
1916{
1917 /* Note that there is no vcpu->fpu_active = 0 here. The caller must
1918 * set this *before* calling this function.
1919 */
1920 vmx_decache_cr0_guest_bits(vcpu);
1921 vmcs_set_bits(GUEST_CR0, X86_CR0_TS | X86_CR0_MP);
1922 update_exception_bitmap(vcpu);
1923 vcpu->arch.cr0_guest_owned_bits = 0;
1924 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
1925 if (is_guest_mode(vcpu)) {
1926 /*
1927 * L1's specified read shadow might not contain the TS bit,
1928 * so now that we turned on shadowing of this bit, we need to
1929 * set this bit of the shadow. Like in nested_vmx_run we need
1930 * nested_read_cr0(vmcs12), but vmcs12->guest_cr0 is not yet
1931 * up-to-date here because we just decached cr0.TS (and we'll
1932 * only update vmcs12->guest_cr0 on nested exit).
1933 */
1934 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1935 vmcs12->guest_cr0 = (vmcs12->guest_cr0 & ~X86_CR0_TS) |
1936 (vcpu->arch.cr0 & X86_CR0_TS);
1937 vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
1938 } else
1939 vmcs_writel(CR0_READ_SHADOW, vcpu->arch.cr0);
1940}
1941
1942static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
1943{
1944 unsigned long rflags, save_rflags;
1945
1946 if (!test_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail)) {
1947 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
1948 rflags = vmcs_readl(GUEST_RFLAGS);
1949 if (to_vmx(vcpu)->rmode.vm86_active) {
1950 rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
1951 save_rflags = to_vmx(vcpu)->rmode.save_rflags;
1952 rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
1953 }
1954 to_vmx(vcpu)->rflags = rflags;
1955 }
1956 return to_vmx(vcpu)->rflags;
1957}
1958
1959static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1960{
1961 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
1962 to_vmx(vcpu)->rflags = rflags;
1963 if (to_vmx(vcpu)->rmode.vm86_active) {
1964 to_vmx(vcpu)->rmode.save_rflags = rflags;
1965 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1966 }
1967 vmcs_writel(GUEST_RFLAGS, rflags);
1968}
1969
1970static u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu)
1971{
1972 u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1973 int ret = 0;
1974
1975 if (interruptibility & GUEST_INTR_STATE_STI)
1976 ret |= KVM_X86_SHADOW_INT_STI;
1977 if (interruptibility & GUEST_INTR_STATE_MOV_SS)
1978 ret |= KVM_X86_SHADOW_INT_MOV_SS;
1979
1980 return ret;
1981}
1982
1983static void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1984{
1985 u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1986 u32 interruptibility = interruptibility_old;
1987
1988 interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
1989
1990 if (mask & KVM_X86_SHADOW_INT_MOV_SS)
1991 interruptibility |= GUEST_INTR_STATE_MOV_SS;
1992 else if (mask & KVM_X86_SHADOW_INT_STI)
1993 interruptibility |= GUEST_INTR_STATE_STI;
1994
1995 if ((interruptibility != interruptibility_old))
1996 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
1997}
1998
1999static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
2000{
2001 unsigned long rip;
2002
2003 rip = kvm_rip_read(vcpu);
2004 rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
2005 kvm_rip_write(vcpu, rip);
2006
2007 /* skipping an emulated instruction also counts */
2008 vmx_set_interrupt_shadow(vcpu, 0);
2009}
2010
2011/*
2012 * KVM wants to inject page-faults which it got to the guest. This function
2013 * checks whether in a nested guest, we need to inject them to L1 or L2.
2014 */
2015static int nested_vmx_check_exception(struct kvm_vcpu *vcpu, unsigned nr)
2016{
2017 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2018
2019 if (!(vmcs12->exception_bitmap & (1u << nr)))
2020 return 0;
2021
2022 nested_vmx_vmexit(vcpu, to_vmx(vcpu)->exit_reason,
2023 vmcs_read32(VM_EXIT_INTR_INFO),
2024 vmcs_readl(EXIT_QUALIFICATION));
2025 return 1;
2026}
2027
2028static void vmx_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
2029 bool has_error_code, u32 error_code,
2030 bool reinject)
2031{
2032 struct vcpu_vmx *vmx = to_vmx(vcpu);
2033 u32 intr_info = nr | INTR_INFO_VALID_MASK;
2034
2035 if (!reinject && is_guest_mode(vcpu) &&
2036 nested_vmx_check_exception(vcpu, nr))
2037 return;
2038
2039 if (has_error_code) {
2040 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
2041 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
2042 }
2043
2044 if (vmx->rmode.vm86_active) {
2045 int inc_eip = 0;
2046 if (kvm_exception_is_soft(nr))
2047 inc_eip = vcpu->arch.event_exit_inst_len;
2048 if (kvm_inject_realmode_interrupt(vcpu, nr, inc_eip) != EMULATE_DONE)
2049 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2050 return;
2051 }
2052
2053 if (kvm_exception_is_soft(nr)) {
2054 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2055 vmx->vcpu.arch.event_exit_inst_len);
2056 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
2057 } else
2058 intr_info |= INTR_TYPE_HARD_EXCEPTION;
2059
2060 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
2061}
2062
2063static bool vmx_rdtscp_supported(void)
2064{
2065 return cpu_has_vmx_rdtscp();
2066}
2067
2068static bool vmx_invpcid_supported(void)
2069{
2070 return cpu_has_vmx_invpcid() && enable_ept;
2071}
2072
2073/*
2074 * Swap MSR entry in host/guest MSR entry array.
2075 */
2076static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
2077{
2078 struct shared_msr_entry tmp;
2079
2080 tmp = vmx->guest_msrs[to];
2081 vmx->guest_msrs[to] = vmx->guest_msrs[from];
2082 vmx->guest_msrs[from] = tmp;
2083}
2084
2085static void vmx_set_msr_bitmap(struct kvm_vcpu *vcpu)
2086{
2087 unsigned long *msr_bitmap;
2088
2089 if (irqchip_in_kernel(vcpu->kvm) && apic_x2apic_mode(vcpu->arch.apic)) {
2090 if (is_long_mode(vcpu))
2091 msr_bitmap = vmx_msr_bitmap_longmode_x2apic;
2092 else
2093 msr_bitmap = vmx_msr_bitmap_legacy_x2apic;
2094 } else {
2095 if (is_long_mode(vcpu))
2096 msr_bitmap = vmx_msr_bitmap_longmode;
2097 else
2098 msr_bitmap = vmx_msr_bitmap_legacy;
2099 }
2100
2101 vmcs_write64(MSR_BITMAP, __pa(msr_bitmap));
2102}
2103
2104/*
2105 * Set up the vmcs to automatically save and restore system
2106 * msrs. Don't touch the 64-bit msrs if the guest is in legacy
2107 * mode, as fiddling with msrs is very expensive.
2108 */
2109static void setup_msrs(struct vcpu_vmx *vmx)
2110{
2111 int save_nmsrs, index;
2112
2113 save_nmsrs = 0;
2114#ifdef CONFIG_X86_64
2115 if (is_long_mode(&vmx->vcpu)) {
2116 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
2117 if (index >= 0)
2118 move_msr_up(vmx, index, save_nmsrs++);
2119 index = __find_msr_index(vmx, MSR_LSTAR);
2120 if (index >= 0)
2121 move_msr_up(vmx, index, save_nmsrs++);
2122 index = __find_msr_index(vmx, MSR_CSTAR);
2123 if (index >= 0)
2124 move_msr_up(vmx, index, save_nmsrs++);
2125 index = __find_msr_index(vmx, MSR_TSC_AUX);
2126 if (index >= 0 && vmx->rdtscp_enabled)
2127 move_msr_up(vmx, index, save_nmsrs++);
2128 /*
2129 * MSR_STAR is only needed on long mode guests, and only
2130 * if efer.sce is enabled.
2131 */
2132 index = __find_msr_index(vmx, MSR_STAR);
2133 if ((index >= 0) && (vmx->vcpu.arch.efer & EFER_SCE))
2134 move_msr_up(vmx, index, save_nmsrs++);
2135 }
2136#endif
2137 index = __find_msr_index(vmx, MSR_EFER);
2138 if (index >= 0 && update_transition_efer(vmx, index))
2139 move_msr_up(vmx, index, save_nmsrs++);
2140
2141 vmx->save_nmsrs = save_nmsrs;
2142
2143 if (cpu_has_vmx_msr_bitmap())
2144 vmx_set_msr_bitmap(&vmx->vcpu);
2145}
2146
2147/*
2148 * reads and returns guest's timestamp counter "register"
2149 * guest_tsc = host_tsc + tsc_offset -- 21.3
2150 */
2151static u64 guest_read_tsc(void)
2152{
2153 u64 host_tsc, tsc_offset;
2154
2155 rdtscll(host_tsc);
2156 tsc_offset = vmcs_read64(TSC_OFFSET);
2157 return host_tsc + tsc_offset;
2158}
2159
2160/*
2161 * Like guest_read_tsc, but always returns L1's notion of the timestamp
2162 * counter, even if a nested guest (L2) is currently running.
2163 */
2164static u64 vmx_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
2165{
2166 u64 tsc_offset;
2167
2168 tsc_offset = is_guest_mode(vcpu) ?
2169 to_vmx(vcpu)->nested.vmcs01_tsc_offset :
2170 vmcs_read64(TSC_OFFSET);
2171 return host_tsc + tsc_offset;
2172}
2173
2174/*
2175 * Engage any workarounds for mis-matched TSC rates. Currently limited to
2176 * software catchup for faster rates on slower CPUs.
2177 */
2178static void vmx_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
2179{
2180 if (!scale)
2181 return;
2182
2183 if (user_tsc_khz > tsc_khz) {
2184 vcpu->arch.tsc_catchup = 1;
2185 vcpu->arch.tsc_always_catchup = 1;
2186 } else
2187 WARN(1, "user requested TSC rate below hardware speed\n");
2188}
2189
2190static u64 vmx_read_tsc_offset(struct kvm_vcpu *vcpu)
2191{
2192 return vmcs_read64(TSC_OFFSET);
2193}
2194
2195/*
2196 * writes 'offset' into guest's timestamp counter offset register
2197 */
2198static void vmx_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
2199{
2200 if (is_guest_mode(vcpu)) {
2201 /*
2202 * We're here if L1 chose not to trap WRMSR to TSC. According
2203 * to the spec, this should set L1's TSC; The offset that L1
2204 * set for L2 remains unchanged, and still needs to be added
2205 * to the newly set TSC to get L2's TSC.
2206 */
2207 struct vmcs12 *vmcs12;
2208 to_vmx(vcpu)->nested.vmcs01_tsc_offset = offset;
2209 /* recalculate vmcs02.TSC_OFFSET: */
2210 vmcs12 = get_vmcs12(vcpu);
2211 vmcs_write64(TSC_OFFSET, offset +
2212 (nested_cpu_has(vmcs12, CPU_BASED_USE_TSC_OFFSETING) ?
2213 vmcs12->tsc_offset : 0));
2214 } else {
2215 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
2216 vmcs_read64(TSC_OFFSET), offset);
2217 vmcs_write64(TSC_OFFSET, offset);
2218 }
2219}
2220
2221static void vmx_adjust_tsc_offset(struct kvm_vcpu *vcpu, s64 adjustment, bool host)
2222{
2223 u64 offset = vmcs_read64(TSC_OFFSET);
2224
2225 vmcs_write64(TSC_OFFSET, offset + adjustment);
2226 if (is_guest_mode(vcpu)) {
2227 /* Even when running L2, the adjustment needs to apply to L1 */
2228 to_vmx(vcpu)->nested.vmcs01_tsc_offset += adjustment;
2229 } else
2230 trace_kvm_write_tsc_offset(vcpu->vcpu_id, offset,
2231 offset + adjustment);
2232}
2233
2234static u64 vmx_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
2235{
2236 return target_tsc - native_read_tsc();
2237}
2238
2239static bool guest_cpuid_has_vmx(struct kvm_vcpu *vcpu)
2240{
2241 struct kvm_cpuid_entry2 *best = kvm_find_cpuid_entry(vcpu, 1, 0);
2242 return best && (best->ecx & (1 << (X86_FEATURE_VMX & 31)));
2243}
2244
2245/*
2246 * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
2247 * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
2248 * all guests if the "nested" module option is off, and can also be disabled
2249 * for a single guest by disabling its VMX cpuid bit.
2250 */
2251static inline bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
2252{
2253 return nested && guest_cpuid_has_vmx(vcpu);
2254}
2255
2256/*
2257 * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be
2258 * returned for the various VMX controls MSRs when nested VMX is enabled.
2259 * The same values should also be used to verify that vmcs12 control fields are
2260 * valid during nested entry from L1 to L2.
2261 * Each of these control msrs has a low and high 32-bit half: A low bit is on
2262 * if the corresponding bit in the (32-bit) control field *must* be on, and a
2263 * bit in the high half is on if the corresponding bit in the control field
2264 * may be on. See also vmx_control_verify().
2265 * TODO: allow these variables to be modified (downgraded) by module options
2266 * or other means.
2267 */
2268static u32 nested_vmx_procbased_ctls_low, nested_vmx_procbased_ctls_high;
2269static u32 nested_vmx_true_procbased_ctls_low;
2270static u32 nested_vmx_secondary_ctls_low, nested_vmx_secondary_ctls_high;
2271static u32 nested_vmx_pinbased_ctls_low, nested_vmx_pinbased_ctls_high;
2272static u32 nested_vmx_exit_ctls_low, nested_vmx_exit_ctls_high;
2273static u32 nested_vmx_true_exit_ctls_low;
2274static u32 nested_vmx_entry_ctls_low, nested_vmx_entry_ctls_high;
2275static u32 nested_vmx_true_entry_ctls_low;
2276static u32 nested_vmx_misc_low, nested_vmx_misc_high;
2277static u32 nested_vmx_ept_caps;
2278static __init void nested_vmx_setup_ctls_msrs(void)
2279{
2280 /*
2281 * Note that as a general rule, the high half of the MSRs (bits in
2282 * the control fields which may be 1) should be initialized by the
2283 * intersection of the underlying hardware's MSR (i.e., features which
2284 * can be supported) and the list of features we want to expose -
2285 * because they are known to be properly supported in our code.
2286 * Also, usually, the low half of the MSRs (bits which must be 1) can
2287 * be set to 0, meaning that L1 may turn off any of these bits. The
2288 * reason is that if one of these bits is necessary, it will appear
2289 * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control
2290 * fields of vmcs01 and vmcs02, will turn these bits off - and
2291 * nested_vmx_exit_handled() will not pass related exits to L1.
2292 * These rules have exceptions below.
2293 */
2294
2295 /* pin-based controls */
2296 rdmsr(MSR_IA32_VMX_PINBASED_CTLS,
2297 nested_vmx_pinbased_ctls_low, nested_vmx_pinbased_ctls_high);
2298 nested_vmx_pinbased_ctls_low |= PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
2299 nested_vmx_pinbased_ctls_high &= PIN_BASED_EXT_INTR_MASK |
2300 PIN_BASED_NMI_EXITING | PIN_BASED_VIRTUAL_NMIS;
2301 nested_vmx_pinbased_ctls_high |= PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
2302 PIN_BASED_VMX_PREEMPTION_TIMER;
2303
2304 /* exit controls */
2305 rdmsr(MSR_IA32_VMX_EXIT_CTLS,
2306 nested_vmx_exit_ctls_low, nested_vmx_exit_ctls_high);
2307 nested_vmx_exit_ctls_low = VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
2308
2309 nested_vmx_exit_ctls_high &=
2310#ifdef CONFIG_X86_64
2311 VM_EXIT_HOST_ADDR_SPACE_SIZE |
2312#endif
2313 VM_EXIT_LOAD_IA32_PAT | VM_EXIT_SAVE_IA32_PAT;
2314 nested_vmx_exit_ctls_high |= VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR |
2315 VM_EXIT_LOAD_IA32_EFER | VM_EXIT_SAVE_IA32_EFER |
2316 VM_EXIT_SAVE_VMX_PREEMPTION_TIMER | VM_EXIT_ACK_INTR_ON_EXIT;
2317
2318 if (vmx_mpx_supported())
2319 nested_vmx_exit_ctls_high |= VM_EXIT_CLEAR_BNDCFGS;
2320
2321 /* We support free control of debug control saving. */
2322 nested_vmx_true_exit_ctls_low = nested_vmx_exit_ctls_low &
2323 ~VM_EXIT_SAVE_DEBUG_CONTROLS;
2324
2325 /* entry controls */
2326 rdmsr(MSR_IA32_VMX_ENTRY_CTLS,
2327 nested_vmx_entry_ctls_low, nested_vmx_entry_ctls_high);
2328 nested_vmx_entry_ctls_low = VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
2329 nested_vmx_entry_ctls_high &=
2330#ifdef CONFIG_X86_64
2331 VM_ENTRY_IA32E_MODE |
2332#endif
2333 VM_ENTRY_LOAD_IA32_PAT;
2334 nested_vmx_entry_ctls_high |= (VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR |
2335 VM_ENTRY_LOAD_IA32_EFER);
2336 if (vmx_mpx_supported())
2337 nested_vmx_entry_ctls_high |= VM_ENTRY_LOAD_BNDCFGS;
2338
2339 /* We support free control of debug control loading. */
2340 nested_vmx_true_entry_ctls_low = nested_vmx_entry_ctls_low &
2341 ~VM_ENTRY_LOAD_DEBUG_CONTROLS;
2342
2343 /* cpu-based controls */
2344 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS,
2345 nested_vmx_procbased_ctls_low, nested_vmx_procbased_ctls_high);
2346 nested_vmx_procbased_ctls_low = CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
2347 nested_vmx_procbased_ctls_high &=
2348 CPU_BASED_VIRTUAL_INTR_PENDING |
2349 CPU_BASED_VIRTUAL_NMI_PENDING | CPU_BASED_USE_TSC_OFFSETING |
2350 CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING |
2351 CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING |
2352 CPU_BASED_CR3_STORE_EXITING |
2353#ifdef CONFIG_X86_64
2354 CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING |
2355#endif
2356 CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING |
2357 CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_EXITING |
2358 CPU_BASED_RDPMC_EXITING | CPU_BASED_RDTSC_EXITING |
2359 CPU_BASED_PAUSE_EXITING | CPU_BASED_TPR_SHADOW |
2360 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2361 /*
2362 * We can allow some features even when not supported by the
2363 * hardware. For example, L1 can specify an MSR bitmap - and we
2364 * can use it to avoid exits to L1 - even when L0 runs L2
2365 * without MSR bitmaps.
2366 */
2367 nested_vmx_procbased_ctls_high |= CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
2368 CPU_BASED_USE_MSR_BITMAPS;
2369
2370 /* We support free control of CR3 access interception. */
2371 nested_vmx_true_procbased_ctls_low = nested_vmx_procbased_ctls_low &
2372 ~(CPU_BASED_CR3_LOAD_EXITING | CPU_BASED_CR3_STORE_EXITING);
2373
2374 /* secondary cpu-based controls */
2375 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
2376 nested_vmx_secondary_ctls_low, nested_vmx_secondary_ctls_high);
2377 nested_vmx_secondary_ctls_low = 0;
2378 nested_vmx_secondary_ctls_high &=
2379 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2380 SECONDARY_EXEC_UNRESTRICTED_GUEST |
2381 SECONDARY_EXEC_WBINVD_EXITING;
2382
2383 if (enable_ept) {
2384 /* nested EPT: emulate EPT also to L1 */
2385 nested_vmx_secondary_ctls_high |= SECONDARY_EXEC_ENABLE_EPT;
2386 nested_vmx_ept_caps = VMX_EPT_PAGE_WALK_4_BIT |
2387 VMX_EPTP_WB_BIT | VMX_EPT_2MB_PAGE_BIT |
2388 VMX_EPT_INVEPT_BIT;
2389 nested_vmx_ept_caps &= vmx_capability.ept;
2390 /*
2391 * For nested guests, we don't do anything specific
2392 * for single context invalidation. Hence, only advertise
2393 * support for global context invalidation.
2394 */
2395 nested_vmx_ept_caps |= VMX_EPT_EXTENT_GLOBAL_BIT;
2396 } else
2397 nested_vmx_ept_caps = 0;
2398
2399 /* miscellaneous data */
2400 rdmsr(MSR_IA32_VMX_MISC, nested_vmx_misc_low, nested_vmx_misc_high);
2401 nested_vmx_misc_low &= VMX_MISC_SAVE_EFER_LMA;
2402 nested_vmx_misc_low |= VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE |
2403 VMX_MISC_ACTIVITY_HLT;
2404 nested_vmx_misc_high = 0;
2405}
2406
2407static inline bool vmx_control_verify(u32 control, u32 low, u32 high)
2408{
2409 /*
2410 * Bits 0 in high must be 0, and bits 1 in low must be 1.
2411 */
2412 return ((control & high) | low) == control;
2413}
2414
2415static inline u64 vmx_control_msr(u32 low, u32 high)
2416{
2417 return low | ((u64)high << 32);
2418}
2419
2420/* Returns 0 on success, non-0 otherwise. */
2421static int vmx_get_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
2422{
2423 switch (msr_index) {
2424 case MSR_IA32_VMX_BASIC:
2425 /*
2426 * This MSR reports some information about VMX support. We
2427 * should return information about the VMX we emulate for the
2428 * guest, and the VMCS structure we give it - not about the
2429 * VMX support of the underlying hardware.
2430 */
2431 *pdata = VMCS12_REVISION | VMX_BASIC_TRUE_CTLS |
2432 ((u64)VMCS12_SIZE << VMX_BASIC_VMCS_SIZE_SHIFT) |
2433 (VMX_BASIC_MEM_TYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT);
2434 break;
2435 case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
2436 case MSR_IA32_VMX_PINBASED_CTLS:
2437 *pdata = vmx_control_msr(nested_vmx_pinbased_ctls_low,
2438 nested_vmx_pinbased_ctls_high);
2439 break;
2440 case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
2441 *pdata = vmx_control_msr(nested_vmx_true_procbased_ctls_low,
2442 nested_vmx_procbased_ctls_high);
2443 break;
2444 case MSR_IA32_VMX_PROCBASED_CTLS:
2445 *pdata = vmx_control_msr(nested_vmx_procbased_ctls_low,
2446 nested_vmx_procbased_ctls_high);
2447 break;
2448 case MSR_IA32_VMX_TRUE_EXIT_CTLS:
2449 *pdata = vmx_control_msr(nested_vmx_true_exit_ctls_low,
2450 nested_vmx_exit_ctls_high);
2451 break;
2452 case MSR_IA32_VMX_EXIT_CTLS:
2453 *pdata = vmx_control_msr(nested_vmx_exit_ctls_low,
2454 nested_vmx_exit_ctls_high);
2455 break;
2456 case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
2457 *pdata = vmx_control_msr(nested_vmx_true_entry_ctls_low,
2458 nested_vmx_entry_ctls_high);
2459 break;
2460 case MSR_IA32_VMX_ENTRY_CTLS:
2461 *pdata = vmx_control_msr(nested_vmx_entry_ctls_low,
2462 nested_vmx_entry_ctls_high);
2463 break;
2464 case MSR_IA32_VMX_MISC:
2465 *pdata = vmx_control_msr(nested_vmx_misc_low,
2466 nested_vmx_misc_high);
2467 break;
2468 /*
2469 * These MSRs specify bits which the guest must keep fixed (on or off)
2470 * while L1 is in VMXON mode (in L1's root mode, or running an L2).
2471 * We picked the standard core2 setting.
2472 */
2473#define VMXON_CR0_ALWAYSON (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
2474#define VMXON_CR4_ALWAYSON X86_CR4_VMXE
2475 case MSR_IA32_VMX_CR0_FIXED0:
2476 *pdata = VMXON_CR0_ALWAYSON;
2477 break;
2478 case MSR_IA32_VMX_CR0_FIXED1:
2479 *pdata = -1ULL;
2480 break;
2481 case MSR_IA32_VMX_CR4_FIXED0:
2482 *pdata = VMXON_CR4_ALWAYSON;
2483 break;
2484 case MSR_IA32_VMX_CR4_FIXED1:
2485 *pdata = -1ULL;
2486 break;
2487 case MSR_IA32_VMX_VMCS_ENUM:
2488 *pdata = 0x2e; /* highest index: VMX_PREEMPTION_TIMER_VALUE */
2489 break;
2490 case MSR_IA32_VMX_PROCBASED_CTLS2:
2491 *pdata = vmx_control_msr(nested_vmx_secondary_ctls_low,
2492 nested_vmx_secondary_ctls_high);
2493 break;
2494 case MSR_IA32_VMX_EPT_VPID_CAP:
2495 /* Currently, no nested vpid support */
2496 *pdata = nested_vmx_ept_caps;
2497 break;
2498 default:
2499 return 1;
2500 }
2501
2502 return 0;
2503}
2504
2505/*
2506 * Reads an msr value (of 'msr_index') into 'pdata'.
2507 * Returns 0 on success, non-0 otherwise.
2508 * Assumes vcpu_load() was already called.
2509 */
2510static int vmx_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
2511{
2512 u64 data;
2513 struct shared_msr_entry *msr;
2514
2515 if (!pdata) {
2516 printk(KERN_ERR "BUG: get_msr called with NULL pdata\n");
2517 return -EINVAL;
2518 }
2519
2520 switch (msr_index) {
2521#ifdef CONFIG_X86_64
2522 case MSR_FS_BASE:
2523 data = vmcs_readl(GUEST_FS_BASE);
2524 break;
2525 case MSR_GS_BASE:
2526 data = vmcs_readl(GUEST_GS_BASE);
2527 break;
2528 case MSR_KERNEL_GS_BASE:
2529 vmx_load_host_state(to_vmx(vcpu));
2530 data = to_vmx(vcpu)->msr_guest_kernel_gs_base;
2531 break;
2532#endif
2533 case MSR_EFER:
2534 return kvm_get_msr_common(vcpu, msr_index, pdata);
2535 case MSR_IA32_TSC:
2536 data = guest_read_tsc();
2537 break;
2538 case MSR_IA32_SYSENTER_CS:
2539 data = vmcs_read32(GUEST_SYSENTER_CS);
2540 break;
2541 case MSR_IA32_SYSENTER_EIP:
2542 data = vmcs_readl(GUEST_SYSENTER_EIP);
2543 break;
2544 case MSR_IA32_SYSENTER_ESP:
2545 data = vmcs_readl(GUEST_SYSENTER_ESP);
2546 break;
2547 case MSR_IA32_BNDCFGS:
2548 if (!vmx_mpx_supported())
2549 return 1;
2550 data = vmcs_read64(GUEST_BNDCFGS);
2551 break;
2552 case MSR_IA32_FEATURE_CONTROL:
2553 if (!nested_vmx_allowed(vcpu))
2554 return 1;
2555 data = to_vmx(vcpu)->nested.msr_ia32_feature_control;
2556 break;
2557 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
2558 if (!nested_vmx_allowed(vcpu))
2559 return 1;
2560 return vmx_get_vmx_msr(vcpu, msr_index, pdata);
2561 case MSR_TSC_AUX:
2562 if (!to_vmx(vcpu)->rdtscp_enabled)
2563 return 1;
2564 /* Otherwise falls through */
2565 default:
2566 msr = find_msr_entry(to_vmx(vcpu), msr_index);
2567 if (msr) {
2568 data = msr->data;
2569 break;
2570 }
2571 return kvm_get_msr_common(vcpu, msr_index, pdata);
2572 }
2573
2574 *pdata = data;
2575 return 0;
2576}
2577
2578static void vmx_leave_nested(struct kvm_vcpu *vcpu);
2579
2580/*
2581 * Writes msr value into into the appropriate "register".
2582 * Returns 0 on success, non-0 otherwise.
2583 * Assumes vcpu_load() was already called.
2584 */
2585static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2586{
2587 struct vcpu_vmx *vmx = to_vmx(vcpu);
2588 struct shared_msr_entry *msr;
2589 int ret = 0;
2590 u32 msr_index = msr_info->index;
2591 u64 data = msr_info->data;
2592
2593 switch (msr_index) {
2594 case MSR_EFER:
2595 ret = kvm_set_msr_common(vcpu, msr_info);
2596 break;
2597#ifdef CONFIG_X86_64
2598 case MSR_FS_BASE:
2599 vmx_segment_cache_clear(vmx);
2600 vmcs_writel(GUEST_FS_BASE, data);
2601 break;
2602 case MSR_GS_BASE:
2603 vmx_segment_cache_clear(vmx);
2604 vmcs_writel(GUEST_GS_BASE, data);
2605 break;
2606 case MSR_KERNEL_GS_BASE:
2607 vmx_load_host_state(vmx);
2608 vmx->msr_guest_kernel_gs_base = data;
2609 break;
2610#endif
2611 case MSR_IA32_SYSENTER_CS:
2612 vmcs_write32(GUEST_SYSENTER_CS, data);
2613 break;
2614 case MSR_IA32_SYSENTER_EIP:
2615 vmcs_writel(GUEST_SYSENTER_EIP, data);
2616 break;
2617 case MSR_IA32_SYSENTER_ESP:
2618 vmcs_writel(GUEST_SYSENTER_ESP, data);
2619 break;
2620 case MSR_IA32_BNDCFGS:
2621 if (!vmx_mpx_supported())
2622 return 1;
2623 vmcs_write64(GUEST_BNDCFGS, data);
2624 break;
2625 case MSR_IA32_TSC:
2626 kvm_write_tsc(vcpu, msr_info);
2627 break;
2628 case MSR_IA32_CR_PAT:
2629 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2630 if (!kvm_mtrr_valid(vcpu, MSR_IA32_CR_PAT, data))
2631 return 1;
2632 vmcs_write64(GUEST_IA32_PAT, data);
2633 vcpu->arch.pat = data;
2634 break;
2635 }
2636 ret = kvm_set_msr_common(vcpu, msr_info);
2637 break;
2638 case MSR_IA32_TSC_ADJUST:
2639 ret = kvm_set_msr_common(vcpu, msr_info);
2640 break;
2641 case MSR_IA32_FEATURE_CONTROL:
2642 if (!nested_vmx_allowed(vcpu) ||
2643 (to_vmx(vcpu)->nested.msr_ia32_feature_control &
2644 FEATURE_CONTROL_LOCKED && !msr_info->host_initiated))
2645 return 1;
2646 vmx->nested.msr_ia32_feature_control = data;
2647 if (msr_info->host_initiated && data == 0)
2648 vmx_leave_nested(vcpu);
2649 break;
2650 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
2651 return 1; /* they are read-only */
2652 case MSR_TSC_AUX:
2653 if (!vmx->rdtscp_enabled)
2654 return 1;
2655 /* Check reserved bit, higher 32 bits should be zero */
2656 if ((data >> 32) != 0)
2657 return 1;
2658 /* Otherwise falls through */
2659 default:
2660 msr = find_msr_entry(vmx, msr_index);
2661 if (msr) {
2662 u64 old_msr_data = msr->data;
2663 msr->data = data;
2664 if (msr - vmx->guest_msrs < vmx->save_nmsrs) {
2665 preempt_disable();
2666 ret = kvm_set_shared_msr(msr->index, msr->data,
2667 msr->mask);
2668 preempt_enable();
2669 if (ret)
2670 msr->data = old_msr_data;
2671 }
2672 break;
2673 }
2674 ret = kvm_set_msr_common(vcpu, msr_info);
2675 }
2676
2677 return ret;
2678}
2679
2680static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
2681{
2682 __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
2683 switch (reg) {
2684 case VCPU_REGS_RSP:
2685 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
2686 break;
2687 case VCPU_REGS_RIP:
2688 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
2689 break;
2690 case VCPU_EXREG_PDPTR:
2691 if (enable_ept)
2692 ept_save_pdptrs(vcpu);
2693 break;
2694 default:
2695 break;
2696 }
2697}
2698
2699static __init int cpu_has_kvm_support(void)
2700{
2701 return cpu_has_vmx();
2702}
2703
2704static __init int vmx_disabled_by_bios(void)
2705{
2706 u64 msr;
2707
2708 rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
2709 if (msr & FEATURE_CONTROL_LOCKED) {
2710 /* launched w/ TXT and VMX disabled */
2711 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
2712 && tboot_enabled())
2713 return 1;
2714 /* launched w/o TXT and VMX only enabled w/ TXT */
2715 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
2716 && (msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
2717 && !tboot_enabled()) {
2718 printk(KERN_WARNING "kvm: disable TXT in the BIOS or "
2719 "activate TXT before enabling KVM\n");
2720 return 1;
2721 }
2722 /* launched w/o TXT and VMX disabled */
2723 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
2724 && !tboot_enabled())
2725 return 1;
2726 }
2727
2728 return 0;
2729}
2730
2731static void kvm_cpu_vmxon(u64 addr)
2732{
2733 asm volatile (ASM_VMX_VMXON_RAX
2734 : : "a"(&addr), "m"(addr)
2735 : "memory", "cc");
2736}
2737
2738static int hardware_enable(void)
2739{
2740 int cpu = raw_smp_processor_id();
2741 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
2742 u64 old, test_bits;
2743
2744 if (read_cr4() & X86_CR4_VMXE)
2745 return -EBUSY;
2746
2747 INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
2748
2749 /*
2750 * Now we can enable the vmclear operation in kdump
2751 * since the loaded_vmcss_on_cpu list on this cpu
2752 * has been initialized.
2753 *
2754 * Though the cpu is not in VMX operation now, there
2755 * is no problem to enable the vmclear operation
2756 * for the loaded_vmcss_on_cpu list is empty!
2757 */
2758 crash_enable_local_vmclear(cpu);
2759
2760 rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
2761
2762 test_bits = FEATURE_CONTROL_LOCKED;
2763 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
2764 if (tboot_enabled())
2765 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX;
2766
2767 if ((old & test_bits) != test_bits) {
2768 /* enable and lock */
2769 wrmsrl(MSR_IA32_FEATURE_CONTROL, old | test_bits);
2770 }
2771 write_cr4(read_cr4() | X86_CR4_VMXE); /* FIXME: not cpu hotplug safe */
2772
2773 if (vmm_exclusive) {
2774 kvm_cpu_vmxon(phys_addr);
2775 ept_sync_global();
2776 }
2777
2778 native_store_gdt(this_cpu_ptr(&host_gdt));
2779
2780 return 0;
2781}
2782
2783static void vmclear_local_loaded_vmcss(void)
2784{
2785 int cpu = raw_smp_processor_id();
2786 struct loaded_vmcs *v, *n;
2787
2788 list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
2789 loaded_vmcss_on_cpu_link)
2790 __loaded_vmcs_clear(v);
2791}
2792
2793
2794/* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
2795 * tricks.
2796 */
2797static void kvm_cpu_vmxoff(void)
2798{
2799 asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
2800}
2801
2802static void hardware_disable(void)
2803{
2804 if (vmm_exclusive) {
2805 vmclear_local_loaded_vmcss();
2806 kvm_cpu_vmxoff();
2807 }
2808 write_cr4(read_cr4() & ~X86_CR4_VMXE);
2809}
2810
2811static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
2812 u32 msr, u32 *result)
2813{
2814 u32 vmx_msr_low, vmx_msr_high;
2815 u32 ctl = ctl_min | ctl_opt;
2816
2817 rdmsr(msr, vmx_msr_low, vmx_msr_high);
2818
2819 ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
2820 ctl |= vmx_msr_low; /* bit == 1 in low word ==> must be one */
2821
2822 /* Ensure minimum (required) set of control bits are supported. */
2823 if (ctl_min & ~ctl)
2824 return -EIO;
2825
2826 *result = ctl;
2827 return 0;
2828}
2829
2830static __init bool allow_1_setting(u32 msr, u32 ctl)
2831{
2832 u32 vmx_msr_low, vmx_msr_high;
2833
2834 rdmsr(msr, vmx_msr_low, vmx_msr_high);
2835 return vmx_msr_high & ctl;
2836}
2837
2838static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
2839{
2840 u32 vmx_msr_low, vmx_msr_high;
2841 u32 min, opt, min2, opt2;
2842 u32 _pin_based_exec_control = 0;
2843 u32 _cpu_based_exec_control = 0;
2844 u32 _cpu_based_2nd_exec_control = 0;
2845 u32 _vmexit_control = 0;
2846 u32 _vmentry_control = 0;
2847
2848 min = CPU_BASED_HLT_EXITING |
2849#ifdef CONFIG_X86_64
2850 CPU_BASED_CR8_LOAD_EXITING |
2851 CPU_BASED_CR8_STORE_EXITING |
2852#endif
2853 CPU_BASED_CR3_LOAD_EXITING |
2854 CPU_BASED_CR3_STORE_EXITING |
2855 CPU_BASED_USE_IO_BITMAPS |
2856 CPU_BASED_MOV_DR_EXITING |
2857 CPU_BASED_USE_TSC_OFFSETING |
2858 CPU_BASED_MWAIT_EXITING |
2859 CPU_BASED_MONITOR_EXITING |
2860 CPU_BASED_INVLPG_EXITING |
2861 CPU_BASED_RDPMC_EXITING;
2862
2863 opt = CPU_BASED_TPR_SHADOW |
2864 CPU_BASED_USE_MSR_BITMAPS |
2865 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2866 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
2867 &_cpu_based_exec_control) < 0)
2868 return -EIO;
2869#ifdef CONFIG_X86_64
2870 if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2871 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
2872 ~CPU_BASED_CR8_STORE_EXITING;
2873#endif
2874 if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
2875 min2 = 0;
2876 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2877 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2878 SECONDARY_EXEC_WBINVD_EXITING |
2879 SECONDARY_EXEC_ENABLE_VPID |
2880 SECONDARY_EXEC_ENABLE_EPT |
2881 SECONDARY_EXEC_UNRESTRICTED_GUEST |
2882 SECONDARY_EXEC_PAUSE_LOOP_EXITING |
2883 SECONDARY_EXEC_RDTSCP |
2884 SECONDARY_EXEC_ENABLE_INVPCID |
2885 SECONDARY_EXEC_APIC_REGISTER_VIRT |
2886 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2887 SECONDARY_EXEC_SHADOW_VMCS;
2888 if (adjust_vmx_controls(min2, opt2,
2889 MSR_IA32_VMX_PROCBASED_CTLS2,
2890 &_cpu_based_2nd_exec_control) < 0)
2891 return -EIO;
2892 }
2893#ifndef CONFIG_X86_64
2894 if (!(_cpu_based_2nd_exec_control &
2895 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
2896 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
2897#endif
2898
2899 if (!(_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2900 _cpu_based_2nd_exec_control &= ~(
2901 SECONDARY_EXEC_APIC_REGISTER_VIRT |
2902 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2903 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
2904
2905 if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
2906 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
2907 enabled */
2908 _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
2909 CPU_BASED_CR3_STORE_EXITING |
2910 CPU_BASED_INVLPG_EXITING);
2911 rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
2912 vmx_capability.ept, vmx_capability.vpid);
2913 }
2914
2915 min = VM_EXIT_SAVE_DEBUG_CONTROLS;
2916#ifdef CONFIG_X86_64
2917 min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
2918#endif
2919 opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT |
2920 VM_EXIT_ACK_INTR_ON_EXIT | VM_EXIT_CLEAR_BNDCFGS;
2921 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
2922 &_vmexit_control) < 0)
2923 return -EIO;
2924
2925 min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
2926 opt = PIN_BASED_VIRTUAL_NMIS | PIN_BASED_POSTED_INTR;
2927 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
2928 &_pin_based_exec_control) < 0)
2929 return -EIO;
2930
2931 if (!(_cpu_based_2nd_exec_control &
2932 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) ||
2933 !(_vmexit_control & VM_EXIT_ACK_INTR_ON_EXIT))
2934 _pin_based_exec_control &= ~PIN_BASED_POSTED_INTR;
2935
2936 min = VM_ENTRY_LOAD_DEBUG_CONTROLS;
2937 opt = VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_BNDCFGS;
2938 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
2939 &_vmentry_control) < 0)
2940 return -EIO;
2941
2942 rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
2943
2944 /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
2945 if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
2946 return -EIO;
2947
2948#ifdef CONFIG_X86_64
2949 /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
2950 if (vmx_msr_high & (1u<<16))
2951 return -EIO;
2952#endif
2953
2954 /* Require Write-Back (WB) memory type for VMCS accesses. */
2955 if (((vmx_msr_high >> 18) & 15) != 6)
2956 return -EIO;
2957
2958 vmcs_conf->size = vmx_msr_high & 0x1fff;
2959 vmcs_conf->order = get_order(vmcs_config.size);
2960 vmcs_conf->revision_id = vmx_msr_low;
2961
2962 vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
2963 vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
2964 vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
2965 vmcs_conf->vmexit_ctrl = _vmexit_control;
2966 vmcs_conf->vmentry_ctrl = _vmentry_control;
2967
2968 cpu_has_load_ia32_efer =
2969 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
2970 VM_ENTRY_LOAD_IA32_EFER)
2971 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
2972 VM_EXIT_LOAD_IA32_EFER);
2973
2974 cpu_has_load_perf_global_ctrl =
2975 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
2976 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
2977 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
2978 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
2979
2980 /*
2981 * Some cpus support VM_ENTRY_(LOAD|SAVE)_IA32_PERF_GLOBAL_CTRL
2982 * but due to arrata below it can't be used. Workaround is to use
2983 * msr load mechanism to switch IA32_PERF_GLOBAL_CTRL.
2984 *
2985 * VM Exit May Incorrectly Clear IA32_PERF_GLOBAL_CTRL [34:32]
2986 *
2987 * AAK155 (model 26)
2988 * AAP115 (model 30)
2989 * AAT100 (model 37)
2990 * BC86,AAY89,BD102 (model 44)
2991 * BA97 (model 46)
2992 *
2993 */
2994 if (cpu_has_load_perf_global_ctrl && boot_cpu_data.x86 == 0x6) {
2995 switch (boot_cpu_data.x86_model) {
2996 case 26:
2997 case 30:
2998 case 37:
2999 case 44:
3000 case 46:
3001 cpu_has_load_perf_global_ctrl = false;
3002 printk_once(KERN_WARNING"kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
3003 "does not work properly. Using workaround\n");
3004 break;
3005 default:
3006 break;
3007 }
3008 }
3009
3010 return 0;
3011}
3012
3013static struct vmcs *alloc_vmcs_cpu(int cpu)
3014{
3015 int node = cpu_to_node(cpu);
3016 struct page *pages;
3017 struct vmcs *vmcs;
3018
3019 pages = alloc_pages_exact_node(node, GFP_KERNEL, vmcs_config.order);
3020 if (!pages)
3021 return NULL;
3022 vmcs = page_address(pages);
3023 memset(vmcs, 0, vmcs_config.size);
3024 vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
3025 return vmcs;
3026}
3027
3028static struct vmcs *alloc_vmcs(void)
3029{
3030 return alloc_vmcs_cpu(raw_smp_processor_id());
3031}
3032
3033static void free_vmcs(struct vmcs *vmcs)
3034{
3035 free_pages((unsigned long)vmcs, vmcs_config.order);
3036}
3037
3038/*
3039 * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
3040 */
3041static void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
3042{
3043 if (!loaded_vmcs->vmcs)
3044 return;
3045 loaded_vmcs_clear(loaded_vmcs);
3046 free_vmcs(loaded_vmcs->vmcs);
3047 loaded_vmcs->vmcs = NULL;
3048}
3049
3050static void free_kvm_area(void)
3051{
3052 int cpu;
3053
3054 for_each_possible_cpu(cpu) {
3055 free_vmcs(per_cpu(vmxarea, cpu));
3056 per_cpu(vmxarea, cpu) = NULL;
3057 }
3058}
3059
3060static void init_vmcs_shadow_fields(void)
3061{
3062 int i, j;
3063
3064 /* No checks for read only fields yet */
3065
3066 for (i = j = 0; i < max_shadow_read_write_fields; i++) {
3067 switch (shadow_read_write_fields[i]) {
3068 case GUEST_BNDCFGS:
3069 if (!vmx_mpx_supported())
3070 continue;
3071 break;
3072 default:
3073 break;
3074 }
3075
3076 if (j < i)
3077 shadow_read_write_fields[j] =
3078 shadow_read_write_fields[i];
3079 j++;
3080 }
3081 max_shadow_read_write_fields = j;
3082
3083 /* shadowed fields guest access without vmexit */
3084 for (i = 0; i < max_shadow_read_write_fields; i++) {
3085 clear_bit(shadow_read_write_fields[i],
3086 vmx_vmwrite_bitmap);
3087 clear_bit(shadow_read_write_fields[i],
3088 vmx_vmread_bitmap);
3089 }
3090 for (i = 0; i < max_shadow_read_only_fields; i++)
3091 clear_bit(shadow_read_only_fields[i],
3092 vmx_vmread_bitmap);
3093}
3094
3095static __init int alloc_kvm_area(void)
3096{
3097 int cpu;
3098
3099 for_each_possible_cpu(cpu) {
3100 struct vmcs *vmcs;
3101
3102 vmcs = alloc_vmcs_cpu(cpu);
3103 if (!vmcs) {
3104 free_kvm_area();
3105 return -ENOMEM;
3106 }
3107
3108 per_cpu(vmxarea, cpu) = vmcs;
3109 }
3110 return 0;
3111}
3112
3113static __init int hardware_setup(void)
3114{
3115 if (setup_vmcs_config(&vmcs_config) < 0)
3116 return -EIO;
3117
3118 if (boot_cpu_has(X86_FEATURE_NX))
3119 kvm_enable_efer_bits(EFER_NX);
3120
3121 if (!cpu_has_vmx_vpid())
3122 enable_vpid = 0;
3123 if (!cpu_has_vmx_shadow_vmcs())
3124 enable_shadow_vmcs = 0;
3125 if (enable_shadow_vmcs)
3126 init_vmcs_shadow_fields();
3127
3128 if (!cpu_has_vmx_ept() ||
3129 !cpu_has_vmx_ept_4levels()) {
3130 enable_ept = 0;
3131 enable_unrestricted_guest = 0;
3132 enable_ept_ad_bits = 0;
3133 }
3134
3135 if (!cpu_has_vmx_ept_ad_bits())
3136 enable_ept_ad_bits = 0;
3137
3138 if (!cpu_has_vmx_unrestricted_guest())
3139 enable_unrestricted_guest = 0;
3140
3141 if (!cpu_has_vmx_flexpriority()) {
3142 flexpriority_enabled = 0;
3143
3144 /*
3145 * set_apic_access_page_addr() is used to reload apic access
3146 * page upon invalidation. No need to do anything if the
3147 * processor does not have the APIC_ACCESS_ADDR VMCS field.
3148 */
3149 kvm_x86_ops->set_apic_access_page_addr = NULL;
3150 }
3151
3152 if (!cpu_has_vmx_tpr_shadow())
3153 kvm_x86_ops->update_cr8_intercept = NULL;
3154
3155 if (enable_ept && !cpu_has_vmx_ept_2m_page())
3156 kvm_disable_largepages();
3157
3158 if (!cpu_has_vmx_ple())
3159 ple_gap = 0;
3160
3161 if (!cpu_has_vmx_apicv())
3162 enable_apicv = 0;
3163
3164 if (enable_apicv)
3165 kvm_x86_ops->update_cr8_intercept = NULL;
3166 else {
3167 kvm_x86_ops->hwapic_irr_update = NULL;
3168 kvm_x86_ops->deliver_posted_interrupt = NULL;
3169 kvm_x86_ops->sync_pir_to_irr = vmx_sync_pir_to_irr_dummy;
3170 }
3171
3172 if (nested)
3173 nested_vmx_setup_ctls_msrs();
3174
3175 return alloc_kvm_area();
3176}
3177
3178static __exit void hardware_unsetup(void)
3179{
3180 free_kvm_area();
3181}
3182
3183static bool emulation_required(struct kvm_vcpu *vcpu)
3184{
3185 return emulate_invalid_guest_state && !guest_state_valid(vcpu);
3186}
3187
3188static void fix_pmode_seg(struct kvm_vcpu *vcpu, int seg,
3189 struct kvm_segment *save)
3190{
3191 if (!emulate_invalid_guest_state) {
3192 /*
3193 * CS and SS RPL should be equal during guest entry according
3194 * to VMX spec, but in reality it is not always so. Since vcpu
3195 * is in the middle of the transition from real mode to
3196 * protected mode it is safe to assume that RPL 0 is a good
3197 * default value.
3198 */
3199 if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
3200 save->selector &= ~SELECTOR_RPL_MASK;
3201 save->dpl = save->selector & SELECTOR_RPL_MASK;
3202 save->s = 1;
3203 }
3204 vmx_set_segment(vcpu, save, seg);
3205}
3206
3207static void enter_pmode(struct kvm_vcpu *vcpu)
3208{
3209 unsigned long flags;
3210 struct vcpu_vmx *vmx = to_vmx(vcpu);
3211
3212 /*
3213 * Update real mode segment cache. It may be not up-to-date if sement
3214 * register was written while vcpu was in a guest mode.
3215 */
3216 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
3217 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
3218 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
3219 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
3220 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
3221 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
3222
3223 vmx->rmode.vm86_active = 0;
3224
3225 vmx_segment_cache_clear(vmx);
3226
3227 vmx_set_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
3228
3229 flags = vmcs_readl(GUEST_RFLAGS);
3230 flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
3231 flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
3232 vmcs_writel(GUEST_RFLAGS, flags);
3233
3234 vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
3235 (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
3236
3237 update_exception_bitmap(vcpu);
3238
3239 fix_pmode_seg(vcpu, VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
3240 fix_pmode_seg(vcpu, VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
3241 fix_pmode_seg(vcpu, VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
3242 fix_pmode_seg(vcpu, VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
3243 fix_pmode_seg(vcpu, VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
3244 fix_pmode_seg(vcpu, VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
3245}
3246
3247static void fix_rmode_seg(int seg, struct kvm_segment *save)
3248{
3249 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3250 struct kvm_segment var = *save;
3251
3252 var.dpl = 0x3;
3253 if (seg == VCPU_SREG_CS)
3254 var.type = 0x3;
3255
3256 if (!emulate_invalid_guest_state) {
3257 var.selector = var.base >> 4;
3258 var.base = var.base & 0xffff0;
3259 var.limit = 0xffff;
3260 var.g = 0;
3261 var.db = 0;
3262 var.present = 1;
3263 var.s = 1;
3264 var.l = 0;
3265 var.unusable = 0;
3266 var.type = 0x3;
3267 var.avl = 0;
3268 if (save->base & 0xf)
3269 printk_once(KERN_WARNING "kvm: segment base is not "
3270 "paragraph aligned when entering "
3271 "protected mode (seg=%d)", seg);
3272 }
3273
3274 vmcs_write16(sf->selector, var.selector);
3275 vmcs_write32(sf->base, var.base);
3276 vmcs_write32(sf->limit, var.limit);
3277 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(&var));
3278}
3279
3280static void enter_rmode(struct kvm_vcpu *vcpu)
3281{
3282 unsigned long flags;
3283 struct vcpu_vmx *vmx = to_vmx(vcpu);
3284
3285 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
3286 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
3287 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
3288 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
3289 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
3290 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
3291 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
3292
3293 vmx->rmode.vm86_active = 1;
3294
3295 /*
3296 * Very old userspace does not call KVM_SET_TSS_ADDR before entering
3297 * vcpu. Warn the user that an update is overdue.
3298 */
3299 if (!vcpu->kvm->arch.tss_addr)
3300 printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
3301 "called before entering vcpu\n");
3302
3303 vmx_segment_cache_clear(vmx);
3304
3305 vmcs_writel(GUEST_TR_BASE, vcpu->kvm->arch.tss_addr);
3306 vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
3307 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
3308
3309 flags = vmcs_readl(GUEST_RFLAGS);
3310 vmx->rmode.save_rflags = flags;
3311
3312 flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
3313
3314 vmcs_writel(GUEST_RFLAGS, flags);
3315 vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
3316 update_exception_bitmap(vcpu);
3317
3318 fix_rmode_seg(VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
3319 fix_rmode_seg(VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
3320 fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
3321 fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
3322 fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
3323 fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
3324
3325 kvm_mmu_reset_context(vcpu);
3326}
3327
3328static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
3329{
3330 struct vcpu_vmx *vmx = to_vmx(vcpu);
3331 struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
3332
3333 if (!msr)
3334 return;
3335
3336 /*
3337 * Force kernel_gs_base reloading before EFER changes, as control
3338 * of this msr depends on is_long_mode().
3339 */
3340 vmx_load_host_state(to_vmx(vcpu));
3341 vcpu->arch.efer = efer;
3342 if (efer & EFER_LMA) {
3343 vm_entry_controls_setbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
3344 msr->data = efer;
3345 } else {
3346 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
3347
3348 msr->data = efer & ~EFER_LME;
3349 }
3350 setup_msrs(vmx);
3351}
3352
3353#ifdef CONFIG_X86_64
3354
3355static void enter_lmode(struct kvm_vcpu *vcpu)
3356{
3357 u32 guest_tr_ar;
3358
3359 vmx_segment_cache_clear(to_vmx(vcpu));
3360
3361 guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
3362 if ((guest_tr_ar & AR_TYPE_MASK) != AR_TYPE_BUSY_64_TSS) {
3363 pr_debug_ratelimited("%s: tss fixup for long mode. \n",
3364 __func__);
3365 vmcs_write32(GUEST_TR_AR_BYTES,
3366 (guest_tr_ar & ~AR_TYPE_MASK)
3367 | AR_TYPE_BUSY_64_TSS);
3368 }
3369 vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
3370}
3371
3372static void exit_lmode(struct kvm_vcpu *vcpu)
3373{
3374 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
3375 vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
3376}
3377
3378#endif
3379
3380static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
3381{
3382 vpid_sync_context(to_vmx(vcpu));
3383 if (enable_ept) {
3384 if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
3385 return;
3386 ept_sync_context(construct_eptp(vcpu->arch.mmu.root_hpa));
3387 }
3388}
3389
3390static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
3391{
3392 ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
3393
3394 vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
3395 vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
3396}
3397
3398static void vmx_decache_cr3(struct kvm_vcpu *vcpu)
3399{
3400 if (enable_ept && is_paging(vcpu))
3401 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
3402 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
3403}
3404
3405static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
3406{
3407 ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
3408
3409 vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
3410 vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
3411}
3412
3413static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
3414{
3415 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
3416
3417 if (!test_bit(VCPU_EXREG_PDPTR,
3418 (unsigned long *)&vcpu->arch.regs_dirty))
3419 return;
3420
3421 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
3422 vmcs_write64(GUEST_PDPTR0, mmu->pdptrs[0]);
3423 vmcs_write64(GUEST_PDPTR1, mmu->pdptrs[1]);
3424 vmcs_write64(GUEST_PDPTR2, mmu->pdptrs[2]);
3425 vmcs_write64(GUEST_PDPTR3, mmu->pdptrs[3]);
3426 }
3427}
3428
3429static void ept_save_pdptrs(struct kvm_vcpu *vcpu)
3430{
3431 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
3432
3433 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
3434 mmu->pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
3435 mmu->pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
3436 mmu->pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
3437 mmu->pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
3438 }
3439
3440 __set_bit(VCPU_EXREG_PDPTR,
3441 (unsigned long *)&vcpu->arch.regs_avail);
3442 __set_bit(VCPU_EXREG_PDPTR,
3443 (unsigned long *)&vcpu->arch.regs_dirty);
3444}
3445
3446static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
3447
3448static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
3449 unsigned long cr0,
3450 struct kvm_vcpu *vcpu)
3451{
3452 if (!test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
3453 vmx_decache_cr3(vcpu);
3454 if (!(cr0 & X86_CR0_PG)) {
3455 /* From paging/starting to nonpaging */
3456 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
3457 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
3458 (CPU_BASED_CR3_LOAD_EXITING |
3459 CPU_BASED_CR3_STORE_EXITING));
3460 vcpu->arch.cr0 = cr0;
3461 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
3462 } else if (!is_paging(vcpu)) {
3463 /* From nonpaging to paging */
3464 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
3465 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
3466 ~(CPU_BASED_CR3_LOAD_EXITING |
3467 CPU_BASED_CR3_STORE_EXITING));
3468 vcpu->arch.cr0 = cr0;
3469 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
3470 }
3471
3472 if (!(cr0 & X86_CR0_WP))
3473 *hw_cr0 &= ~X86_CR0_WP;
3474}
3475
3476static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
3477{
3478 struct vcpu_vmx *vmx = to_vmx(vcpu);
3479 unsigned long hw_cr0;
3480
3481 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK);
3482 if (enable_unrestricted_guest)
3483 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
3484 else {
3485 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON;
3486
3487 if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
3488 enter_pmode(vcpu);
3489
3490 if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
3491 enter_rmode(vcpu);
3492 }
3493
3494#ifdef CONFIG_X86_64
3495 if (vcpu->arch.efer & EFER_LME) {
3496 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
3497 enter_lmode(vcpu);
3498 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
3499 exit_lmode(vcpu);
3500 }
3501#endif
3502
3503 if (enable_ept)
3504 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
3505
3506 if (!vcpu->fpu_active)
3507 hw_cr0 |= X86_CR0_TS | X86_CR0_MP;
3508
3509 vmcs_writel(CR0_READ_SHADOW, cr0);
3510 vmcs_writel(GUEST_CR0, hw_cr0);
3511 vcpu->arch.cr0 = cr0;
3512
3513 /* depends on vcpu->arch.cr0 to be set to a new value */
3514 vmx->emulation_required = emulation_required(vcpu);
3515}
3516
3517static u64 construct_eptp(unsigned long root_hpa)
3518{
3519 u64 eptp;
3520
3521 /* TODO write the value reading from MSR */
3522 eptp = VMX_EPT_DEFAULT_MT |
3523 VMX_EPT_DEFAULT_GAW << VMX_EPT_GAW_EPTP_SHIFT;
3524 if (enable_ept_ad_bits)
3525 eptp |= VMX_EPT_AD_ENABLE_BIT;
3526 eptp |= (root_hpa & PAGE_MASK);
3527
3528 return eptp;
3529}
3530
3531static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
3532{
3533 unsigned long guest_cr3;
3534 u64 eptp;
3535
3536 guest_cr3 = cr3;
3537 if (enable_ept) {
3538 eptp = construct_eptp(cr3);
3539 vmcs_write64(EPT_POINTER, eptp);
3540 if (is_paging(vcpu) || is_guest_mode(vcpu))
3541 guest_cr3 = kvm_read_cr3(vcpu);
3542 else
3543 guest_cr3 = vcpu->kvm->arch.ept_identity_map_addr;
3544 ept_load_pdptrs(vcpu);
3545 }
3546
3547 vmx_flush_tlb(vcpu);
3548 vmcs_writel(GUEST_CR3, guest_cr3);
3549}
3550
3551static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
3552{
3553 unsigned long hw_cr4 = cr4 | (to_vmx(vcpu)->rmode.vm86_active ?
3554 KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON);
3555
3556 if (cr4 & X86_CR4_VMXE) {
3557 /*
3558 * To use VMXON (and later other VMX instructions), a guest
3559 * must first be able to turn on cr4.VMXE (see handle_vmon()).
3560 * So basically the check on whether to allow nested VMX
3561 * is here.
3562 */
3563 if (!nested_vmx_allowed(vcpu))
3564 return 1;
3565 }
3566 if (to_vmx(vcpu)->nested.vmxon &&
3567 ((cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON))
3568 return 1;
3569
3570 vcpu->arch.cr4 = cr4;
3571 if (enable_ept) {
3572 if (!is_paging(vcpu)) {
3573 hw_cr4 &= ~X86_CR4_PAE;
3574 hw_cr4 |= X86_CR4_PSE;
3575 /*
3576 * SMEP/SMAP is disabled if CPU is in non-paging mode
3577 * in hardware. However KVM always uses paging mode to
3578 * emulate guest non-paging mode with TDP.
3579 * To emulate this behavior, SMEP/SMAP needs to be
3580 * manually disabled when guest switches to non-paging
3581 * mode.
3582 */
3583 hw_cr4 &= ~(X86_CR4_SMEP | X86_CR4_SMAP);
3584 } else if (!(cr4 & X86_CR4_PAE)) {
3585 hw_cr4 &= ~X86_CR4_PAE;
3586 }
3587 }
3588
3589 vmcs_writel(CR4_READ_SHADOW, cr4);
3590 vmcs_writel(GUEST_CR4, hw_cr4);
3591 return 0;
3592}
3593
3594static void vmx_get_segment(struct kvm_vcpu *vcpu,
3595 struct kvm_segment *var, int seg)
3596{
3597 struct vcpu_vmx *vmx = to_vmx(vcpu);
3598 u32 ar;
3599
3600 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3601 *var = vmx->rmode.segs[seg];
3602 if (seg == VCPU_SREG_TR
3603 || var->selector == vmx_read_guest_seg_selector(vmx, seg))
3604 return;
3605 var->base = vmx_read_guest_seg_base(vmx, seg);
3606 var->selector = vmx_read_guest_seg_selector(vmx, seg);
3607 return;
3608 }
3609 var->base = vmx_read_guest_seg_base(vmx, seg);
3610 var->limit = vmx_read_guest_seg_limit(vmx, seg);
3611 var->selector = vmx_read_guest_seg_selector(vmx, seg);
3612 ar = vmx_read_guest_seg_ar(vmx, seg);
3613 var->unusable = (ar >> 16) & 1;
3614 var->type = ar & 15;
3615 var->s = (ar >> 4) & 1;
3616 var->dpl = (ar >> 5) & 3;
3617 /*
3618 * Some userspaces do not preserve unusable property. Since usable
3619 * segment has to be present according to VMX spec we can use present
3620 * property to amend userspace bug by making unusable segment always
3621 * nonpresent. vmx_segment_access_rights() already marks nonpresent
3622 * segment as unusable.
3623 */
3624 var->present = !var->unusable;
3625 var->avl = (ar >> 12) & 1;
3626 var->l = (ar >> 13) & 1;
3627 var->db = (ar >> 14) & 1;
3628 var->g = (ar >> 15) & 1;
3629}
3630
3631static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
3632{
3633 struct kvm_segment s;
3634
3635 if (to_vmx(vcpu)->rmode.vm86_active) {
3636 vmx_get_segment(vcpu, &s, seg);
3637 return s.base;
3638 }
3639 return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
3640}
3641
3642static int vmx_get_cpl(struct kvm_vcpu *vcpu)
3643{
3644 struct vcpu_vmx *vmx = to_vmx(vcpu);
3645
3646 if (unlikely(vmx->rmode.vm86_active))
3647 return 0;
3648 else {
3649 int ar = vmx_read_guest_seg_ar(vmx, VCPU_SREG_SS);
3650 return AR_DPL(ar);
3651 }
3652}
3653
3654static u32 vmx_segment_access_rights(struct kvm_segment *var)
3655{
3656 u32 ar;
3657
3658 if (var->unusable || !var->present)
3659 ar = 1 << 16;
3660 else {
3661 ar = var->type & 15;
3662 ar |= (var->s & 1) << 4;
3663 ar |= (var->dpl & 3) << 5;
3664 ar |= (var->present & 1) << 7;
3665 ar |= (var->avl & 1) << 12;
3666 ar |= (var->l & 1) << 13;
3667 ar |= (var->db & 1) << 14;
3668 ar |= (var->g & 1) << 15;
3669 }
3670
3671 return ar;
3672}
3673
3674static void vmx_set_segment(struct kvm_vcpu *vcpu,
3675 struct kvm_segment *var, int seg)
3676{
3677 struct vcpu_vmx *vmx = to_vmx(vcpu);
3678 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3679
3680 vmx_segment_cache_clear(vmx);
3681
3682 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3683 vmx->rmode.segs[seg] = *var;
3684 if (seg == VCPU_SREG_TR)
3685 vmcs_write16(sf->selector, var->selector);
3686 else if (var->s)
3687 fix_rmode_seg(seg, &vmx->rmode.segs[seg]);
3688 goto out;
3689 }
3690
3691 vmcs_writel(sf->base, var->base);
3692 vmcs_write32(sf->limit, var->limit);
3693 vmcs_write16(sf->selector, var->selector);
3694
3695 /*
3696 * Fix the "Accessed" bit in AR field of segment registers for older
3697 * qemu binaries.
3698 * IA32 arch specifies that at the time of processor reset the
3699 * "Accessed" bit in the AR field of segment registers is 1. And qemu
3700 * is setting it to 0 in the userland code. This causes invalid guest
3701 * state vmexit when "unrestricted guest" mode is turned on.
3702 * Fix for this setup issue in cpu_reset is being pushed in the qemu
3703 * tree. Newer qemu binaries with that qemu fix would not need this
3704 * kvm hack.
3705 */
3706 if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
3707 var->type |= 0x1; /* Accessed */
3708
3709 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(var));
3710
3711out:
3712 vmx->emulation_required = emulation_required(vcpu);
3713}
3714
3715static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3716{
3717 u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
3718
3719 *db = (ar >> 14) & 1;
3720 *l = (ar >> 13) & 1;
3721}
3722
3723static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3724{
3725 dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
3726 dt->address = vmcs_readl(GUEST_IDTR_BASE);
3727}
3728
3729static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3730{
3731 vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
3732 vmcs_writel(GUEST_IDTR_BASE, dt->address);
3733}
3734
3735static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3736{
3737 dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
3738 dt->address = vmcs_readl(GUEST_GDTR_BASE);
3739}
3740
3741static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3742{
3743 vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
3744 vmcs_writel(GUEST_GDTR_BASE, dt->address);
3745}
3746
3747static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
3748{
3749 struct kvm_segment var;
3750 u32 ar;
3751
3752 vmx_get_segment(vcpu, &var, seg);
3753 var.dpl = 0x3;
3754 if (seg == VCPU_SREG_CS)
3755 var.type = 0x3;
3756 ar = vmx_segment_access_rights(&var);
3757
3758 if (var.base != (var.selector << 4))
3759 return false;
3760 if (var.limit != 0xffff)
3761 return false;
3762 if (ar != 0xf3)
3763 return false;
3764
3765 return true;
3766}
3767
3768static bool code_segment_valid(struct kvm_vcpu *vcpu)
3769{
3770 struct kvm_segment cs;
3771 unsigned int cs_rpl;
3772
3773 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3774 cs_rpl = cs.selector & SELECTOR_RPL_MASK;
3775
3776 if (cs.unusable)
3777 return false;
3778 if (~cs.type & (AR_TYPE_CODE_MASK|AR_TYPE_ACCESSES_MASK))
3779 return false;
3780 if (!cs.s)
3781 return false;
3782 if (cs.type & AR_TYPE_WRITEABLE_MASK) {
3783 if (cs.dpl > cs_rpl)
3784 return false;
3785 } else {
3786 if (cs.dpl != cs_rpl)
3787 return false;
3788 }
3789 if (!cs.present)
3790 return false;
3791
3792 /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
3793 return true;
3794}
3795
3796static bool stack_segment_valid(struct kvm_vcpu *vcpu)
3797{
3798 struct kvm_segment ss;
3799 unsigned int ss_rpl;
3800
3801 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3802 ss_rpl = ss.selector & SELECTOR_RPL_MASK;
3803
3804 if (ss.unusable)
3805 return true;
3806 if (ss.type != 3 && ss.type != 7)
3807 return false;
3808 if (!ss.s)
3809 return false;
3810 if (ss.dpl != ss_rpl) /* DPL != RPL */
3811 return false;
3812 if (!ss.present)
3813 return false;
3814
3815 return true;
3816}
3817
3818static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
3819{
3820 struct kvm_segment var;
3821 unsigned int rpl;
3822
3823 vmx_get_segment(vcpu, &var, seg);
3824 rpl = var.selector & SELECTOR_RPL_MASK;
3825
3826 if (var.unusable)
3827 return true;
3828 if (!var.s)
3829 return false;
3830 if (!var.present)
3831 return false;
3832 if (~var.type & (AR_TYPE_CODE_MASK|AR_TYPE_WRITEABLE_MASK)) {
3833 if (var.dpl < rpl) /* DPL < RPL */
3834 return false;
3835 }
3836
3837 /* TODO: Add other members to kvm_segment_field to allow checking for other access
3838 * rights flags
3839 */
3840 return true;
3841}
3842
3843static bool tr_valid(struct kvm_vcpu *vcpu)
3844{
3845 struct kvm_segment tr;
3846
3847 vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
3848
3849 if (tr.unusable)
3850 return false;
3851 if (tr.selector & SELECTOR_TI_MASK) /* TI = 1 */
3852 return false;
3853 if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
3854 return false;
3855 if (!tr.present)
3856 return false;
3857
3858 return true;
3859}
3860
3861static bool ldtr_valid(struct kvm_vcpu *vcpu)
3862{
3863 struct kvm_segment ldtr;
3864
3865 vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
3866
3867 if (ldtr.unusable)
3868 return true;
3869 if (ldtr.selector & SELECTOR_TI_MASK) /* TI = 1 */
3870 return false;
3871 if (ldtr.type != 2)
3872 return false;
3873 if (!ldtr.present)
3874 return false;
3875
3876 return true;
3877}
3878
3879static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
3880{
3881 struct kvm_segment cs, ss;
3882
3883 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3884 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3885
3886 return ((cs.selector & SELECTOR_RPL_MASK) ==
3887 (ss.selector & SELECTOR_RPL_MASK));
3888}
3889
3890/*
3891 * Check if guest state is valid. Returns true if valid, false if
3892 * not.
3893 * We assume that registers are always usable
3894 */
3895static bool guest_state_valid(struct kvm_vcpu *vcpu)
3896{
3897 if (enable_unrestricted_guest)
3898 return true;
3899
3900 /* real mode guest state checks */
3901 if (!is_protmode(vcpu) || (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
3902 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
3903 return false;
3904 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
3905 return false;
3906 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
3907 return false;
3908 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
3909 return false;
3910 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
3911 return false;
3912 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
3913 return false;
3914 } else {
3915 /* protected mode guest state checks */
3916 if (!cs_ss_rpl_check(vcpu))
3917 return false;
3918 if (!code_segment_valid(vcpu))
3919 return false;
3920 if (!stack_segment_valid(vcpu))
3921 return false;
3922 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
3923 return false;
3924 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
3925 return false;
3926 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
3927 return false;
3928 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
3929 return false;
3930 if (!tr_valid(vcpu))
3931 return false;
3932 if (!ldtr_valid(vcpu))
3933 return false;
3934 }
3935 /* TODO:
3936 * - Add checks on RIP
3937 * - Add checks on RFLAGS
3938 */
3939
3940 return true;
3941}
3942
3943static int init_rmode_tss(struct kvm *kvm)
3944{
3945 gfn_t fn;
3946 u16 data = 0;
3947 int idx, r;
3948
3949 idx = srcu_read_lock(&kvm->srcu);
3950 fn = kvm->arch.tss_addr >> PAGE_SHIFT;
3951 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3952 if (r < 0)
3953 goto out;
3954 data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
3955 r = kvm_write_guest_page(kvm, fn++, &data,
3956 TSS_IOPB_BASE_OFFSET, sizeof(u16));
3957 if (r < 0)
3958 goto out;
3959 r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
3960 if (r < 0)
3961 goto out;
3962 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3963 if (r < 0)
3964 goto out;
3965 data = ~0;
3966 r = kvm_write_guest_page(kvm, fn, &data,
3967 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
3968 sizeof(u8));
3969out:
3970 srcu_read_unlock(&kvm->srcu, idx);
3971 return r;
3972}
3973
3974static int init_rmode_identity_map(struct kvm *kvm)
3975{
3976 int i, idx, r = 0;
3977 pfn_t identity_map_pfn;
3978 u32 tmp;
3979
3980 if (!enable_ept)
3981 return 0;
3982
3983 /* Protect kvm->arch.ept_identity_pagetable_done. */
3984 mutex_lock(&kvm->slots_lock);
3985
3986 if (likely(kvm->arch.ept_identity_pagetable_done))
3987 goto out2;
3988
3989 identity_map_pfn = kvm->arch.ept_identity_map_addr >> PAGE_SHIFT;
3990
3991 r = alloc_identity_pagetable(kvm);
3992 if (r < 0)
3993 goto out2;
3994
3995 idx = srcu_read_lock(&kvm->srcu);
3996 r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
3997 if (r < 0)
3998 goto out;
3999 /* Set up identity-mapping pagetable for EPT in real mode */
4000 for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
4001 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
4002 _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
4003 r = kvm_write_guest_page(kvm, identity_map_pfn,
4004 &tmp, i * sizeof(tmp), sizeof(tmp));
4005 if (r < 0)
4006 goto out;
4007 }
4008 kvm->arch.ept_identity_pagetable_done = true;
4009
4010out:
4011 srcu_read_unlock(&kvm->srcu, idx);
4012
4013out2:
4014 mutex_unlock(&kvm->slots_lock);
4015 return r;
4016}
4017
4018static void seg_setup(int seg)
4019{
4020 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
4021 unsigned int ar;
4022
4023 vmcs_write16(sf->selector, 0);
4024 vmcs_writel(sf->base, 0);
4025 vmcs_write32(sf->limit, 0xffff);
4026 ar = 0x93;
4027 if (seg == VCPU_SREG_CS)
4028 ar |= 0x08; /* code segment */
4029
4030 vmcs_write32(sf->ar_bytes, ar);
4031}
4032
4033static int alloc_apic_access_page(struct kvm *kvm)
4034{
4035 struct page *page;
4036 struct kvm_userspace_memory_region kvm_userspace_mem;
4037 int r = 0;
4038
4039 mutex_lock(&kvm->slots_lock);
4040 if (kvm->arch.apic_access_page_done)
4041 goto out;
4042 kvm_userspace_mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
4043 kvm_userspace_mem.flags = 0;
4044 kvm_userspace_mem.guest_phys_addr = APIC_DEFAULT_PHYS_BASE;
4045 kvm_userspace_mem.memory_size = PAGE_SIZE;
4046 r = __kvm_set_memory_region(kvm, &kvm_userspace_mem);
4047 if (r)
4048 goto out;
4049
4050 page = gfn_to_page(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
4051 if (is_error_page(page)) {
4052 r = -EFAULT;
4053 goto out;
4054 }
4055
4056 /*
4057 * Do not pin the page in memory, so that memory hot-unplug
4058 * is able to migrate it.
4059 */
4060 put_page(page);
4061 kvm->arch.apic_access_page_done = true;
4062out:
4063 mutex_unlock(&kvm->slots_lock);
4064 return r;
4065}
4066
4067static int alloc_identity_pagetable(struct kvm *kvm)
4068{
4069 /* Called with kvm->slots_lock held. */
4070
4071 struct kvm_userspace_memory_region kvm_userspace_mem;
4072 int r = 0;
4073
4074 BUG_ON(kvm->arch.ept_identity_pagetable_done);
4075
4076 kvm_userspace_mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
4077 kvm_userspace_mem.flags = 0;
4078 kvm_userspace_mem.guest_phys_addr =
4079 kvm->arch.ept_identity_map_addr;
4080 kvm_userspace_mem.memory_size = PAGE_SIZE;
4081 r = __kvm_set_memory_region(kvm, &kvm_userspace_mem);
4082
4083 return r;
4084}
4085
4086static void allocate_vpid(struct vcpu_vmx *vmx)
4087{
4088 int vpid;
4089
4090 vmx->vpid = 0;
4091 if (!enable_vpid)
4092 return;
4093 spin_lock(&vmx_vpid_lock);
4094 vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
4095 if (vpid < VMX_NR_VPIDS) {
4096 vmx->vpid = vpid;
4097 __set_bit(vpid, vmx_vpid_bitmap);
4098 }
4099 spin_unlock(&vmx_vpid_lock);
4100}
4101
4102static void free_vpid(struct vcpu_vmx *vmx)
4103{
4104 if (!enable_vpid)
4105 return;
4106 spin_lock(&vmx_vpid_lock);
4107 if (vmx->vpid != 0)
4108 __clear_bit(vmx->vpid, vmx_vpid_bitmap);
4109 spin_unlock(&vmx_vpid_lock);
4110}
4111
4112#define MSR_TYPE_R 1
4113#define MSR_TYPE_W 2
4114static void __vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
4115 u32 msr, int type)
4116{
4117 int f = sizeof(unsigned long);
4118
4119 if (!cpu_has_vmx_msr_bitmap())
4120 return;
4121
4122 /*
4123 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
4124 * have the write-low and read-high bitmap offsets the wrong way round.
4125 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
4126 */
4127 if (msr <= 0x1fff) {
4128 if (type & MSR_TYPE_R)
4129 /* read-low */
4130 __clear_bit(msr, msr_bitmap + 0x000 / f);
4131
4132 if (type & MSR_TYPE_W)
4133 /* write-low */
4134 __clear_bit(msr, msr_bitmap + 0x800 / f);
4135
4136 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
4137 msr &= 0x1fff;
4138 if (type & MSR_TYPE_R)
4139 /* read-high */
4140 __clear_bit(msr, msr_bitmap + 0x400 / f);
4141
4142 if (type & MSR_TYPE_W)
4143 /* write-high */
4144 __clear_bit(msr, msr_bitmap + 0xc00 / f);
4145
4146 }
4147}
4148
4149static void __vmx_enable_intercept_for_msr(unsigned long *msr_bitmap,
4150 u32 msr, int type)
4151{
4152 int f = sizeof(unsigned long);
4153
4154 if (!cpu_has_vmx_msr_bitmap())
4155 return;
4156
4157 /*
4158 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
4159 * have the write-low and read-high bitmap offsets the wrong way round.
4160 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
4161 */
4162 if (msr <= 0x1fff) {
4163 if (type & MSR_TYPE_R)
4164 /* read-low */
4165 __set_bit(msr, msr_bitmap + 0x000 / f);
4166
4167 if (type & MSR_TYPE_W)
4168 /* write-low */
4169 __set_bit(msr, msr_bitmap + 0x800 / f);
4170
4171 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
4172 msr &= 0x1fff;
4173 if (type & MSR_TYPE_R)
4174 /* read-high */
4175 __set_bit(msr, msr_bitmap + 0x400 / f);
4176
4177 if (type & MSR_TYPE_W)
4178 /* write-high */
4179 __set_bit(msr, msr_bitmap + 0xc00 / f);
4180
4181 }
4182}
4183
4184static void vmx_disable_intercept_for_msr(u32 msr, bool longmode_only)
4185{
4186 if (!longmode_only)
4187 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy,
4188 msr, MSR_TYPE_R | MSR_TYPE_W);
4189 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode,
4190 msr, MSR_TYPE_R | MSR_TYPE_W);
4191}
4192
4193static void vmx_enable_intercept_msr_read_x2apic(u32 msr)
4194{
4195 __vmx_enable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic,
4196 msr, MSR_TYPE_R);
4197 __vmx_enable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic,
4198 msr, MSR_TYPE_R);
4199}
4200
4201static void vmx_disable_intercept_msr_read_x2apic(u32 msr)
4202{
4203 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic,
4204 msr, MSR_TYPE_R);
4205 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic,
4206 msr, MSR_TYPE_R);
4207}
4208
4209static void vmx_disable_intercept_msr_write_x2apic(u32 msr)
4210{
4211 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic,
4212 msr, MSR_TYPE_W);
4213 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic,
4214 msr, MSR_TYPE_W);
4215}
4216
4217static int vmx_vm_has_apicv(struct kvm *kvm)
4218{
4219 return enable_apicv && irqchip_in_kernel(kvm);
4220}
4221
4222/*
4223 * Send interrupt to vcpu via posted interrupt way.
4224 * 1. If target vcpu is running(non-root mode), send posted interrupt
4225 * notification to vcpu and hardware will sync PIR to vIRR atomically.
4226 * 2. If target vcpu isn't running(root mode), kick it to pick up the
4227 * interrupt from PIR in next vmentry.
4228 */
4229static void vmx_deliver_posted_interrupt(struct kvm_vcpu *vcpu, int vector)
4230{
4231 struct vcpu_vmx *vmx = to_vmx(vcpu);
4232 int r;
4233
4234 if (pi_test_and_set_pir(vector, &vmx->pi_desc))
4235 return;
4236
4237 r = pi_test_and_set_on(&vmx->pi_desc);
4238 kvm_make_request(KVM_REQ_EVENT, vcpu);
4239#ifdef CONFIG_SMP
4240 if (!r && (vcpu->mode == IN_GUEST_MODE))
4241 apic->send_IPI_mask(get_cpu_mask(vcpu->cpu),
4242 POSTED_INTR_VECTOR);
4243 else
4244#endif
4245 kvm_vcpu_kick(vcpu);
4246}
4247
4248static void vmx_sync_pir_to_irr(struct kvm_vcpu *vcpu)
4249{
4250 struct vcpu_vmx *vmx = to_vmx(vcpu);
4251
4252 if (!pi_test_and_clear_on(&vmx->pi_desc))
4253 return;
4254
4255 kvm_apic_update_irr(vcpu, vmx->pi_desc.pir);
4256}
4257
4258static void vmx_sync_pir_to_irr_dummy(struct kvm_vcpu *vcpu)
4259{
4260 return;
4261}
4262
4263/*
4264 * Set up the vmcs's constant host-state fields, i.e., host-state fields that
4265 * will not change in the lifetime of the guest.
4266 * Note that host-state that does change is set elsewhere. E.g., host-state
4267 * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
4268 */
4269static void vmx_set_constant_host_state(struct vcpu_vmx *vmx)
4270{
4271 u32 low32, high32;
4272 unsigned long tmpl;
4273 struct desc_ptr dt;
4274 unsigned long cr4;
4275
4276 vmcs_writel(HOST_CR0, read_cr0() & ~X86_CR0_TS); /* 22.2.3 */
4277 vmcs_writel(HOST_CR3, read_cr3()); /* 22.2.3 FIXME: shadow tables */
4278
4279 /* Save the most likely value for this task's CR4 in the VMCS. */
4280 cr4 = read_cr4();
4281 vmcs_writel(HOST_CR4, cr4); /* 22.2.3, 22.2.5 */
4282 vmx->host_state.vmcs_host_cr4 = cr4;
4283
4284 vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */
4285#ifdef CONFIG_X86_64
4286 /*
4287 * Load null selectors, so we can avoid reloading them in
4288 * __vmx_load_host_state(), in case userspace uses the null selectors
4289 * too (the expected case).
4290 */
4291 vmcs_write16(HOST_DS_SELECTOR, 0);
4292 vmcs_write16(HOST_ES_SELECTOR, 0);
4293#else
4294 vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
4295 vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS); /* 22.2.4 */
4296#endif
4297 vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
4298 vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8); /* 22.2.4 */
4299
4300 native_store_idt(&dt);
4301 vmcs_writel(HOST_IDTR_BASE, dt.address); /* 22.2.4 */
4302 vmx->host_idt_base = dt.address;
4303
4304 vmcs_writel(HOST_RIP, vmx_return); /* 22.2.5 */
4305
4306 rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
4307 vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
4308 rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
4309 vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl); /* 22.2.3 */
4310
4311 if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
4312 rdmsr(MSR_IA32_CR_PAT, low32, high32);
4313 vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
4314 }
4315}
4316
4317static void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
4318{
4319 vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
4320 if (enable_ept)
4321 vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
4322 if (is_guest_mode(&vmx->vcpu))
4323 vmx->vcpu.arch.cr4_guest_owned_bits &=
4324 ~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask;
4325 vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
4326}
4327
4328static u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx)
4329{
4330 u32 pin_based_exec_ctrl = vmcs_config.pin_based_exec_ctrl;
4331
4332 if (!vmx_vm_has_apicv(vmx->vcpu.kvm))
4333 pin_based_exec_ctrl &= ~PIN_BASED_POSTED_INTR;
4334 return pin_based_exec_ctrl;
4335}
4336
4337static u32 vmx_exec_control(struct vcpu_vmx *vmx)
4338{
4339 u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
4340
4341 if (vmx->vcpu.arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)
4342 exec_control &= ~CPU_BASED_MOV_DR_EXITING;
4343
4344 if (!vm_need_tpr_shadow(vmx->vcpu.kvm)) {
4345 exec_control &= ~CPU_BASED_TPR_SHADOW;
4346#ifdef CONFIG_X86_64
4347 exec_control |= CPU_BASED_CR8_STORE_EXITING |
4348 CPU_BASED_CR8_LOAD_EXITING;
4349#endif
4350 }
4351 if (!enable_ept)
4352 exec_control |= CPU_BASED_CR3_STORE_EXITING |
4353 CPU_BASED_CR3_LOAD_EXITING |
4354 CPU_BASED_INVLPG_EXITING;
4355 return exec_control;
4356}
4357
4358static u32 vmx_secondary_exec_control(struct vcpu_vmx *vmx)
4359{
4360 u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
4361 if (!vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
4362 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
4363 if (vmx->vpid == 0)
4364 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
4365 if (!enable_ept) {
4366 exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
4367 enable_unrestricted_guest = 0;
4368 /* Enable INVPCID for non-ept guests may cause performance regression. */
4369 exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
4370 }
4371 if (!enable_unrestricted_guest)
4372 exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
4373 if (!ple_gap)
4374 exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
4375 if (!vmx_vm_has_apicv(vmx->vcpu.kvm))
4376 exec_control &= ~(SECONDARY_EXEC_APIC_REGISTER_VIRT |
4377 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4378 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
4379 /* SECONDARY_EXEC_SHADOW_VMCS is enabled when L1 executes VMPTRLD
4380 (handle_vmptrld).
4381 We can NOT enable shadow_vmcs here because we don't have yet
4382 a current VMCS12
4383 */
4384 exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
4385 return exec_control;
4386}
4387
4388static void ept_set_mmio_spte_mask(void)
4389{
4390 /*
4391 * EPT Misconfigurations can be generated if the value of bits 2:0
4392 * of an EPT paging-structure entry is 110b (write/execute).
4393 * Also, magic bits (0x3ull << 62) is set to quickly identify mmio
4394 * spte.
4395 */
4396 kvm_mmu_set_mmio_spte_mask((0x3ull << 62) | 0x6ull);
4397}
4398
4399/*
4400 * Sets up the vmcs for emulated real mode.
4401 */
4402static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
4403{
4404#ifdef CONFIG_X86_64
4405 unsigned long a;
4406#endif
4407 int i;
4408
4409 /* I/O */
4410 vmcs_write64(IO_BITMAP_A, __pa(vmx_io_bitmap_a));
4411 vmcs_write64(IO_BITMAP_B, __pa(vmx_io_bitmap_b));
4412
4413 if (enable_shadow_vmcs) {
4414 vmcs_write64(VMREAD_BITMAP, __pa(vmx_vmread_bitmap));
4415 vmcs_write64(VMWRITE_BITMAP, __pa(vmx_vmwrite_bitmap));
4416 }
4417 if (cpu_has_vmx_msr_bitmap())
4418 vmcs_write64(MSR_BITMAP, __pa(vmx_msr_bitmap_legacy));
4419
4420 vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
4421
4422 /* Control */
4423 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, vmx_pin_based_exec_ctrl(vmx));
4424
4425 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, vmx_exec_control(vmx));
4426
4427 if (cpu_has_secondary_exec_ctrls()) {
4428 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
4429 vmx_secondary_exec_control(vmx));
4430 }
4431
4432 if (vmx_vm_has_apicv(vmx->vcpu.kvm)) {
4433 vmcs_write64(EOI_EXIT_BITMAP0, 0);
4434 vmcs_write64(EOI_EXIT_BITMAP1, 0);
4435 vmcs_write64(EOI_EXIT_BITMAP2, 0);
4436 vmcs_write64(EOI_EXIT_BITMAP3, 0);
4437
4438 vmcs_write16(GUEST_INTR_STATUS, 0);
4439
4440 vmcs_write64(POSTED_INTR_NV, POSTED_INTR_VECTOR);
4441 vmcs_write64(POSTED_INTR_DESC_ADDR, __pa((&vmx->pi_desc)));
4442 }
4443
4444 if (ple_gap) {
4445 vmcs_write32(PLE_GAP, ple_gap);
4446 vmx->ple_window = ple_window;
4447 vmx->ple_window_dirty = true;
4448 }
4449
4450 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
4451 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
4452 vmcs_write32(CR3_TARGET_COUNT, 0); /* 22.2.1 */
4453
4454 vmcs_write16(HOST_FS_SELECTOR, 0); /* 22.2.4 */
4455 vmcs_write16(HOST_GS_SELECTOR, 0); /* 22.2.4 */
4456 vmx_set_constant_host_state(vmx);
4457#ifdef CONFIG_X86_64
4458 rdmsrl(MSR_FS_BASE, a);
4459 vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
4460 rdmsrl(MSR_GS_BASE, a);
4461 vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
4462#else
4463 vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
4464 vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
4465#endif
4466
4467 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
4468 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
4469 vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host));
4470 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
4471 vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest));
4472
4473 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
4474 u32 msr_low, msr_high;
4475 u64 host_pat;
4476 rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high);
4477 host_pat = msr_low | ((u64) msr_high << 32);
4478 /* Write the default value follow host pat */
4479 vmcs_write64(GUEST_IA32_PAT, host_pat);
4480 /* Keep arch.pat sync with GUEST_IA32_PAT */
4481 vmx->vcpu.arch.pat = host_pat;
4482 }
4483
4484 for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i) {
4485 u32 index = vmx_msr_index[i];
4486 u32 data_low, data_high;
4487 int j = vmx->nmsrs;
4488
4489 if (rdmsr_safe(index, &data_low, &data_high) < 0)
4490 continue;
4491 if (wrmsr_safe(index, data_low, data_high) < 0)
4492 continue;
4493 vmx->guest_msrs[j].index = i;
4494 vmx->guest_msrs[j].data = 0;
4495 vmx->guest_msrs[j].mask = -1ull;
4496 ++vmx->nmsrs;
4497 }
4498
4499
4500 vm_exit_controls_init(vmx, vmcs_config.vmexit_ctrl);
4501
4502 /* 22.2.1, 20.8.1 */
4503 vm_entry_controls_init(vmx, vmcs_config.vmentry_ctrl);
4504
4505 vmcs_writel(CR0_GUEST_HOST_MASK, ~0UL);
4506 set_cr4_guest_host_mask(vmx);
4507
4508 return 0;
4509}
4510
4511static void vmx_vcpu_reset(struct kvm_vcpu *vcpu)
4512{
4513 struct vcpu_vmx *vmx = to_vmx(vcpu);
4514 struct msr_data apic_base_msr;
4515
4516 vmx->rmode.vm86_active = 0;
4517
4518 vmx->soft_vnmi_blocked = 0;
4519
4520 vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
4521 kvm_set_cr8(&vmx->vcpu, 0);
4522 apic_base_msr.data = APIC_DEFAULT_PHYS_BASE | MSR_IA32_APICBASE_ENABLE;
4523 if (kvm_vcpu_is_bsp(&vmx->vcpu))
4524 apic_base_msr.data |= MSR_IA32_APICBASE_BSP;
4525 apic_base_msr.host_initiated = true;
4526 kvm_set_apic_base(&vmx->vcpu, &apic_base_msr);
4527
4528 vmx_segment_cache_clear(vmx);
4529
4530 seg_setup(VCPU_SREG_CS);
4531 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
4532 vmcs_write32(GUEST_CS_BASE, 0xffff0000);
4533
4534 seg_setup(VCPU_SREG_DS);
4535 seg_setup(VCPU_SREG_ES);
4536 seg_setup(VCPU_SREG_FS);
4537 seg_setup(VCPU_SREG_GS);
4538 seg_setup(VCPU_SREG_SS);
4539
4540 vmcs_write16(GUEST_TR_SELECTOR, 0);
4541 vmcs_writel(GUEST_TR_BASE, 0);
4542 vmcs_write32(GUEST_TR_LIMIT, 0xffff);
4543 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
4544
4545 vmcs_write16(GUEST_LDTR_SELECTOR, 0);
4546 vmcs_writel(GUEST_LDTR_BASE, 0);
4547 vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
4548 vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
4549
4550 vmcs_write32(GUEST_SYSENTER_CS, 0);
4551 vmcs_writel(GUEST_SYSENTER_ESP, 0);
4552 vmcs_writel(GUEST_SYSENTER_EIP, 0);
4553
4554 vmcs_writel(GUEST_RFLAGS, 0x02);
4555 kvm_rip_write(vcpu, 0xfff0);
4556
4557 vmcs_writel(GUEST_GDTR_BASE, 0);
4558 vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
4559
4560 vmcs_writel(GUEST_IDTR_BASE, 0);
4561 vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
4562
4563 vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
4564 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
4565 vmcs_write32(GUEST_PENDING_DBG_EXCEPTIONS, 0);
4566
4567 /* Special registers */
4568 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
4569
4570 setup_msrs(vmx);
4571
4572 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */
4573
4574 if (cpu_has_vmx_tpr_shadow()) {
4575 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
4576 if (vm_need_tpr_shadow(vmx->vcpu.kvm))
4577 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
4578 __pa(vmx->vcpu.arch.apic->regs));
4579 vmcs_write32(TPR_THRESHOLD, 0);
4580 }
4581
4582 kvm_vcpu_reload_apic_access_page(vcpu);
4583
4584 if (vmx_vm_has_apicv(vcpu->kvm))
4585 memset(&vmx->pi_desc, 0, sizeof(struct pi_desc));
4586
4587 if (vmx->vpid != 0)
4588 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
4589
4590 vmx->vcpu.arch.cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
4591 vmx_set_cr0(&vmx->vcpu, kvm_read_cr0(vcpu)); /* enter rmode */
4592 vmx_set_cr4(&vmx->vcpu, 0);
4593 vmx_set_efer(&vmx->vcpu, 0);
4594 vmx_fpu_activate(&vmx->vcpu);
4595 update_exception_bitmap(&vmx->vcpu);
4596
4597 vpid_sync_context(vmx);
4598}
4599
4600/*
4601 * In nested virtualization, check if L1 asked to exit on external interrupts.
4602 * For most existing hypervisors, this will always return true.
4603 */
4604static bool nested_exit_on_intr(struct kvm_vcpu *vcpu)
4605{
4606 return get_vmcs12(vcpu)->pin_based_vm_exec_control &
4607 PIN_BASED_EXT_INTR_MASK;
4608}
4609
4610/*
4611 * In nested virtualization, check if L1 has set
4612 * VM_EXIT_ACK_INTR_ON_EXIT
4613 */
4614static bool nested_exit_intr_ack_set(struct kvm_vcpu *vcpu)
4615{
4616 return get_vmcs12(vcpu)->vm_exit_controls &
4617 VM_EXIT_ACK_INTR_ON_EXIT;
4618}
4619
4620static bool nested_exit_on_nmi(struct kvm_vcpu *vcpu)
4621{
4622 return get_vmcs12(vcpu)->pin_based_vm_exec_control &
4623 PIN_BASED_NMI_EXITING;
4624}
4625
4626static void enable_irq_window(struct kvm_vcpu *vcpu)
4627{
4628 u32 cpu_based_vm_exec_control;
4629
4630 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
4631 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
4632 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
4633}
4634
4635static void enable_nmi_window(struct kvm_vcpu *vcpu)
4636{
4637 u32 cpu_based_vm_exec_control;
4638
4639 if (!cpu_has_virtual_nmis() ||
4640 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
4641 enable_irq_window(vcpu);
4642 return;
4643 }
4644
4645 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
4646 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_NMI_PENDING;
4647 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
4648}
4649
4650static void vmx_inject_irq(struct kvm_vcpu *vcpu)
4651{
4652 struct vcpu_vmx *vmx = to_vmx(vcpu);
4653 uint32_t intr;
4654 int irq = vcpu->arch.interrupt.nr;
4655
4656 trace_kvm_inj_virq(irq);
4657
4658 ++vcpu->stat.irq_injections;
4659 if (vmx->rmode.vm86_active) {
4660 int inc_eip = 0;
4661 if (vcpu->arch.interrupt.soft)
4662 inc_eip = vcpu->arch.event_exit_inst_len;
4663 if (kvm_inject_realmode_interrupt(vcpu, irq, inc_eip) != EMULATE_DONE)
4664 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4665 return;
4666 }
4667 intr = irq | INTR_INFO_VALID_MASK;
4668 if (vcpu->arch.interrupt.soft) {
4669 intr |= INTR_TYPE_SOFT_INTR;
4670 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
4671 vmx->vcpu.arch.event_exit_inst_len);
4672 } else
4673 intr |= INTR_TYPE_EXT_INTR;
4674 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
4675}
4676
4677static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
4678{
4679 struct vcpu_vmx *vmx = to_vmx(vcpu);
4680
4681 if (is_guest_mode(vcpu))
4682 return;
4683
4684 if (!cpu_has_virtual_nmis()) {
4685 /*
4686 * Tracking the NMI-blocked state in software is built upon
4687 * finding the next open IRQ window. This, in turn, depends on
4688 * well-behaving guests: They have to keep IRQs disabled at
4689 * least as long as the NMI handler runs. Otherwise we may
4690 * cause NMI nesting, maybe breaking the guest. But as this is
4691 * highly unlikely, we can live with the residual risk.
4692 */
4693 vmx->soft_vnmi_blocked = 1;
4694 vmx->vnmi_blocked_time = 0;
4695 }
4696
4697 ++vcpu->stat.nmi_injections;
4698 vmx->nmi_known_unmasked = false;
4699 if (vmx->rmode.vm86_active) {
4700 if (kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0) != EMULATE_DONE)
4701 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4702 return;
4703 }
4704 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
4705 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
4706}
4707
4708static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
4709{
4710 if (!cpu_has_virtual_nmis())
4711 return to_vmx(vcpu)->soft_vnmi_blocked;
4712 if (to_vmx(vcpu)->nmi_known_unmasked)
4713 return false;
4714 return vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
4715}
4716
4717static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
4718{
4719 struct vcpu_vmx *vmx = to_vmx(vcpu);
4720
4721 if (!cpu_has_virtual_nmis()) {
4722 if (vmx->soft_vnmi_blocked != masked) {
4723 vmx->soft_vnmi_blocked = masked;
4724 vmx->vnmi_blocked_time = 0;
4725 }
4726 } else {
4727 vmx->nmi_known_unmasked = !masked;
4728 if (masked)
4729 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
4730 GUEST_INTR_STATE_NMI);
4731 else
4732 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
4733 GUEST_INTR_STATE_NMI);
4734 }
4735}
4736
4737static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
4738{
4739 if (to_vmx(vcpu)->nested.nested_run_pending)
4740 return 0;
4741
4742 if (!cpu_has_virtual_nmis() && to_vmx(vcpu)->soft_vnmi_blocked)
4743 return 0;
4744
4745 return !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4746 (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
4747 | GUEST_INTR_STATE_NMI));
4748}
4749
4750static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
4751{
4752 return (!to_vmx(vcpu)->nested.nested_run_pending &&
4753 vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
4754 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4755 (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
4756}
4757
4758static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
4759{
4760 int ret;
4761 struct kvm_userspace_memory_region tss_mem = {
4762 .slot = TSS_PRIVATE_MEMSLOT,
4763 .guest_phys_addr = addr,
4764 .memory_size = PAGE_SIZE * 3,
4765 .flags = 0,
4766 };
4767
4768 ret = kvm_set_memory_region(kvm, &tss_mem);
4769 if (ret)
4770 return ret;
4771 kvm->arch.tss_addr = addr;
4772 return init_rmode_tss(kvm);
4773}
4774
4775static bool rmode_exception(struct kvm_vcpu *vcpu, int vec)
4776{
4777 switch (vec) {
4778 case BP_VECTOR:
4779 /*
4780 * Update instruction length as we may reinject the exception
4781 * from user space while in guest debugging mode.
4782 */
4783 to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
4784 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4785 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
4786 return false;
4787 /* fall through */
4788 case DB_VECTOR:
4789 if (vcpu->guest_debug &
4790 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
4791 return false;
4792 /* fall through */
4793 case DE_VECTOR:
4794 case OF_VECTOR:
4795 case BR_VECTOR:
4796 case UD_VECTOR:
4797 case DF_VECTOR:
4798 case SS_VECTOR:
4799 case GP_VECTOR:
4800 case MF_VECTOR:
4801 return true;
4802 break;
4803 }
4804 return false;
4805}
4806
4807static int handle_rmode_exception(struct kvm_vcpu *vcpu,
4808 int vec, u32 err_code)
4809{
4810 /*
4811 * Instruction with address size override prefix opcode 0x67
4812 * Cause the #SS fault with 0 error code in VM86 mode.
4813 */
4814 if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0) {
4815 if (emulate_instruction(vcpu, 0) == EMULATE_DONE) {
4816 if (vcpu->arch.halt_request) {
4817 vcpu->arch.halt_request = 0;
4818 return kvm_emulate_halt(vcpu);
4819 }
4820 return 1;
4821 }
4822 return 0;
4823 }
4824
4825 /*
4826 * Forward all other exceptions that are valid in real mode.
4827 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
4828 * the required debugging infrastructure rework.
4829 */
4830 kvm_queue_exception(vcpu, vec);
4831 return 1;
4832}
4833
4834/*
4835 * Trigger machine check on the host. We assume all the MSRs are already set up
4836 * by the CPU and that we still run on the same CPU as the MCE occurred on.
4837 * We pass a fake environment to the machine check handler because we want
4838 * the guest to be always treated like user space, no matter what context
4839 * it used internally.
4840 */
4841static void kvm_machine_check(void)
4842{
4843#if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
4844 struct pt_regs regs = {
4845 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
4846 .flags = X86_EFLAGS_IF,
4847 };
4848
4849 do_machine_check(&regs, 0);
4850#endif
4851}
4852
4853static int handle_machine_check(struct kvm_vcpu *vcpu)
4854{
4855 /* already handled by vcpu_run */
4856 return 1;
4857}
4858
4859static int handle_exception(struct kvm_vcpu *vcpu)
4860{
4861 struct vcpu_vmx *vmx = to_vmx(vcpu);
4862 struct kvm_run *kvm_run = vcpu->run;
4863 u32 intr_info, ex_no, error_code;
4864 unsigned long cr2, rip, dr6;
4865 u32 vect_info;
4866 enum emulation_result er;
4867
4868 vect_info = vmx->idt_vectoring_info;
4869 intr_info = vmx->exit_intr_info;
4870
4871 if (is_machine_check(intr_info))
4872 return handle_machine_check(vcpu);
4873
4874 if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR)
4875 return 1; /* already handled by vmx_vcpu_run() */
4876
4877 if (is_no_device(intr_info)) {
4878 vmx_fpu_activate(vcpu);
4879 return 1;
4880 }
4881
4882 if (is_invalid_opcode(intr_info)) {
4883 er = emulate_instruction(vcpu, EMULTYPE_TRAP_UD);
4884 if (er != EMULATE_DONE)
4885 kvm_queue_exception(vcpu, UD_VECTOR);
4886 return 1;
4887 }
4888
4889 error_code = 0;
4890 if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
4891 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
4892
4893 /*
4894 * The #PF with PFEC.RSVD = 1 indicates the guest is accessing
4895 * MMIO, it is better to report an internal error.
4896 * See the comments in vmx_handle_exit.
4897 */
4898 if ((vect_info & VECTORING_INFO_VALID_MASK) &&
4899 !(is_page_fault(intr_info) && !(error_code & PFERR_RSVD_MASK))) {
4900 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4901 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
4902 vcpu->run->internal.ndata = 2;
4903 vcpu->run->internal.data[0] = vect_info;
4904 vcpu->run->internal.data[1] = intr_info;
4905 return 0;
4906 }
4907
4908 if (is_page_fault(intr_info)) {
4909 /* EPT won't cause page fault directly */
4910 BUG_ON(enable_ept);
4911 cr2 = vmcs_readl(EXIT_QUALIFICATION);
4912 trace_kvm_page_fault(cr2, error_code);
4913
4914 if (kvm_event_needs_reinjection(vcpu))
4915 kvm_mmu_unprotect_page_virt(vcpu, cr2);
4916 return kvm_mmu_page_fault(vcpu, cr2, error_code, NULL, 0);
4917 }
4918
4919 ex_no = intr_info & INTR_INFO_VECTOR_MASK;
4920
4921 if (vmx->rmode.vm86_active && rmode_exception(vcpu, ex_no))
4922 return handle_rmode_exception(vcpu, ex_no, error_code);
4923
4924 switch (ex_no) {
4925 case DB_VECTOR:
4926 dr6 = vmcs_readl(EXIT_QUALIFICATION);
4927 if (!(vcpu->guest_debug &
4928 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
4929 vcpu->arch.dr6 &= ~15;
4930 vcpu->arch.dr6 |= dr6 | DR6_RTM;
4931 if (!(dr6 & ~DR6_RESERVED)) /* icebp */
4932 skip_emulated_instruction(vcpu);
4933
4934 kvm_queue_exception(vcpu, DB_VECTOR);
4935 return 1;
4936 }
4937 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
4938 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
4939 /* fall through */
4940 case BP_VECTOR:
4941 /*
4942 * Update instruction length as we may reinject #BP from
4943 * user space while in guest debugging mode. Reading it for
4944 * #DB as well causes no harm, it is not used in that case.
4945 */
4946 vmx->vcpu.arch.event_exit_inst_len =
4947 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4948 kvm_run->exit_reason = KVM_EXIT_DEBUG;
4949 rip = kvm_rip_read(vcpu);
4950 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
4951 kvm_run->debug.arch.exception = ex_no;
4952 break;
4953 default:
4954 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
4955 kvm_run->ex.exception = ex_no;
4956 kvm_run->ex.error_code = error_code;
4957 break;
4958 }
4959 return 0;
4960}
4961
4962static int handle_external_interrupt(struct kvm_vcpu *vcpu)
4963{
4964 ++vcpu->stat.irq_exits;
4965 return 1;
4966}
4967
4968static int handle_triple_fault(struct kvm_vcpu *vcpu)
4969{
4970 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
4971 return 0;
4972}
4973
4974static int handle_io(struct kvm_vcpu *vcpu)
4975{
4976 unsigned long exit_qualification;
4977 int size, in, string;
4978 unsigned port;
4979
4980 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4981 string = (exit_qualification & 16) != 0;
4982 in = (exit_qualification & 8) != 0;
4983
4984 ++vcpu->stat.io_exits;
4985
4986 if (string || in)
4987 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
4988
4989 port = exit_qualification >> 16;
4990 size = (exit_qualification & 7) + 1;
4991 skip_emulated_instruction(vcpu);
4992
4993 return kvm_fast_pio_out(vcpu, size, port);
4994}
4995
4996static void
4997vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
4998{
4999 /*
5000 * Patch in the VMCALL instruction:
5001 */
5002 hypercall[0] = 0x0f;
5003 hypercall[1] = 0x01;
5004 hypercall[2] = 0xc1;
5005}
5006
5007static bool nested_cr0_valid(struct vmcs12 *vmcs12, unsigned long val)
5008{
5009 unsigned long always_on = VMXON_CR0_ALWAYSON;
5010
5011 if (nested_vmx_secondary_ctls_high &
5012 SECONDARY_EXEC_UNRESTRICTED_GUEST &&
5013 nested_cpu_has2(vmcs12, SECONDARY_EXEC_UNRESTRICTED_GUEST))
5014 always_on &= ~(X86_CR0_PE | X86_CR0_PG);
5015 return (val & always_on) == always_on;
5016}
5017
5018/* called to set cr0 as appropriate for a mov-to-cr0 exit. */
5019static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
5020{
5021 if (is_guest_mode(vcpu)) {
5022 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5023 unsigned long orig_val = val;
5024
5025 /*
5026 * We get here when L2 changed cr0 in a way that did not change
5027 * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
5028 * but did change L0 shadowed bits. So we first calculate the
5029 * effective cr0 value that L1 would like to write into the
5030 * hardware. It consists of the L2-owned bits from the new
5031 * value combined with the L1-owned bits from L1's guest_cr0.
5032 */
5033 val = (val & ~vmcs12->cr0_guest_host_mask) |
5034 (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask);
5035
5036 if (!nested_cr0_valid(vmcs12, val))
5037 return 1;
5038
5039 if (kvm_set_cr0(vcpu, val))
5040 return 1;
5041 vmcs_writel(CR0_READ_SHADOW, orig_val);
5042 return 0;
5043 } else {
5044 if (to_vmx(vcpu)->nested.vmxon &&
5045 ((val & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON))
5046 return 1;
5047 return kvm_set_cr0(vcpu, val);
5048 }
5049}
5050
5051static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
5052{
5053 if (is_guest_mode(vcpu)) {
5054 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5055 unsigned long orig_val = val;
5056
5057 /* analogously to handle_set_cr0 */
5058 val = (val & ~vmcs12->cr4_guest_host_mask) |
5059 (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask);
5060 if (kvm_set_cr4(vcpu, val))
5061 return 1;
5062 vmcs_writel(CR4_READ_SHADOW, orig_val);
5063 return 0;
5064 } else
5065 return kvm_set_cr4(vcpu, val);
5066}
5067
5068/* called to set cr0 as approriate for clts instruction exit. */
5069static void handle_clts(struct kvm_vcpu *vcpu)
5070{
5071 if (is_guest_mode(vcpu)) {
5072 /*
5073 * We get here when L2 did CLTS, and L1 didn't shadow CR0.TS
5074 * but we did (!fpu_active). We need to keep GUEST_CR0.TS on,
5075 * just pretend it's off (also in arch.cr0 for fpu_activate).
5076 */
5077 vmcs_writel(CR0_READ_SHADOW,
5078 vmcs_readl(CR0_READ_SHADOW) & ~X86_CR0_TS);
5079 vcpu->arch.cr0 &= ~X86_CR0_TS;
5080 } else
5081 vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
5082}
5083
5084static int handle_cr(struct kvm_vcpu *vcpu)
5085{
5086 unsigned long exit_qualification, val;
5087 int cr;
5088 int reg;
5089 int err;
5090
5091 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5092 cr = exit_qualification & 15;
5093 reg = (exit_qualification >> 8) & 15;
5094 switch ((exit_qualification >> 4) & 3) {
5095 case 0: /* mov to cr */
5096 val = kvm_register_readl(vcpu, reg);
5097 trace_kvm_cr_write(cr, val);
5098 switch (cr) {
5099 case 0:
5100 err = handle_set_cr0(vcpu, val);
5101 kvm_complete_insn_gp(vcpu, err);
5102 return 1;
5103 case 3:
5104 err = kvm_set_cr3(vcpu, val);
5105 kvm_complete_insn_gp(vcpu, err);
5106 return 1;
5107 case 4:
5108 err = handle_set_cr4(vcpu, val);
5109 kvm_complete_insn_gp(vcpu, err);
5110 return 1;
5111 case 8: {
5112 u8 cr8_prev = kvm_get_cr8(vcpu);
5113 u8 cr8 = (u8)val;
5114 err = kvm_set_cr8(vcpu, cr8);
5115 kvm_complete_insn_gp(vcpu, err);
5116 if (irqchip_in_kernel(vcpu->kvm))
5117 return 1;
5118 if (cr8_prev <= cr8)
5119 return 1;
5120 vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
5121 return 0;
5122 }
5123 }
5124 break;
5125 case 2: /* clts */
5126 handle_clts(vcpu);
5127 trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
5128 skip_emulated_instruction(vcpu);
5129 vmx_fpu_activate(vcpu);
5130 return 1;
5131 case 1: /*mov from cr*/
5132 switch (cr) {
5133 case 3:
5134 val = kvm_read_cr3(vcpu);
5135 kvm_register_write(vcpu, reg, val);
5136 trace_kvm_cr_read(cr, val);
5137 skip_emulated_instruction(vcpu);
5138 return 1;
5139 case 8:
5140 val = kvm_get_cr8(vcpu);
5141 kvm_register_write(vcpu, reg, val);
5142 trace_kvm_cr_read(cr, val);
5143 skip_emulated_instruction(vcpu);
5144 return 1;
5145 }
5146 break;
5147 case 3: /* lmsw */
5148 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
5149 trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
5150 kvm_lmsw(vcpu, val);
5151
5152 skip_emulated_instruction(vcpu);
5153 return 1;
5154 default:
5155 break;
5156 }
5157 vcpu->run->exit_reason = 0;
5158 vcpu_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
5159 (int)(exit_qualification >> 4) & 3, cr);
5160 return 0;
5161}
5162
5163static int handle_dr(struct kvm_vcpu *vcpu)
5164{
5165 unsigned long exit_qualification;
5166 int dr, reg;
5167
5168 /* Do not handle if the CPL > 0, will trigger GP on re-entry */
5169 if (!kvm_require_cpl(vcpu, 0))
5170 return 1;
5171 dr = vmcs_readl(GUEST_DR7);
5172 if (dr & DR7_GD) {
5173 /*
5174 * As the vm-exit takes precedence over the debug trap, we
5175 * need to emulate the latter, either for the host or the
5176 * guest debugging itself.
5177 */
5178 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
5179 vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
5180 vcpu->run->debug.arch.dr7 = dr;
5181 vcpu->run->debug.arch.pc =
5182 vmcs_readl(GUEST_CS_BASE) +
5183 vmcs_readl(GUEST_RIP);
5184 vcpu->run->debug.arch.exception = DB_VECTOR;
5185 vcpu->run->exit_reason = KVM_EXIT_DEBUG;
5186 return 0;
5187 } else {
5188 vcpu->arch.dr7 &= ~DR7_GD;
5189 vcpu->arch.dr6 |= DR6_BD | DR6_RTM;
5190 vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
5191 kvm_queue_exception(vcpu, DB_VECTOR);
5192 return 1;
5193 }
5194 }
5195
5196 if (vcpu->guest_debug == 0) {
5197 u32 cpu_based_vm_exec_control;
5198
5199 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5200 cpu_based_vm_exec_control &= ~CPU_BASED_MOV_DR_EXITING;
5201 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
5202
5203 /*
5204 * No more DR vmexits; force a reload of the debug registers
5205 * and reenter on this instruction. The next vmexit will
5206 * retrieve the full state of the debug registers.
5207 */
5208 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
5209 return 1;
5210 }
5211
5212 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5213 dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
5214 reg = DEBUG_REG_ACCESS_REG(exit_qualification);
5215 if (exit_qualification & TYPE_MOV_FROM_DR) {
5216 unsigned long val;
5217
5218 if (kvm_get_dr(vcpu, dr, &val))
5219 return 1;
5220 kvm_register_write(vcpu, reg, val);
5221 } else
5222 if (kvm_set_dr(vcpu, dr, kvm_register_readl(vcpu, reg)))
5223 return 1;
5224
5225 skip_emulated_instruction(vcpu);
5226 return 1;
5227}
5228
5229static u64 vmx_get_dr6(struct kvm_vcpu *vcpu)
5230{
5231 return vcpu->arch.dr6;
5232}
5233
5234static void vmx_set_dr6(struct kvm_vcpu *vcpu, unsigned long val)
5235{
5236}
5237
5238static void vmx_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
5239{
5240 u32 cpu_based_vm_exec_control;
5241
5242 get_debugreg(vcpu->arch.db[0], 0);
5243 get_debugreg(vcpu->arch.db[1], 1);
5244 get_debugreg(vcpu->arch.db[2], 2);
5245 get_debugreg(vcpu->arch.db[3], 3);
5246 get_debugreg(vcpu->arch.dr6, 6);
5247 vcpu->arch.dr7 = vmcs_readl(GUEST_DR7);
5248
5249 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
5250
5251 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5252 cpu_based_vm_exec_control |= CPU_BASED_MOV_DR_EXITING;
5253 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
5254}
5255
5256static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
5257{
5258 vmcs_writel(GUEST_DR7, val);
5259}
5260
5261static int handle_cpuid(struct kvm_vcpu *vcpu)
5262{
5263 kvm_emulate_cpuid(vcpu);
5264 return 1;
5265}
5266
5267static int handle_rdmsr(struct kvm_vcpu *vcpu)
5268{
5269 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
5270 u64 data;
5271
5272 if (vmx_get_msr(vcpu, ecx, &data)) {
5273 trace_kvm_msr_read_ex(ecx);
5274 kvm_inject_gp(vcpu, 0);
5275 return 1;
5276 }
5277
5278 trace_kvm_msr_read(ecx, data);
5279
5280 /* FIXME: handling of bits 32:63 of rax, rdx */
5281 vcpu->arch.regs[VCPU_REGS_RAX] = data & -1u;
5282 vcpu->arch.regs[VCPU_REGS_RDX] = (data >> 32) & -1u;
5283 skip_emulated_instruction(vcpu);
5284 return 1;
5285}
5286
5287static int handle_wrmsr(struct kvm_vcpu *vcpu)
5288{
5289 struct msr_data msr;
5290 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
5291 u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
5292 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
5293
5294 msr.data = data;
5295 msr.index = ecx;
5296 msr.host_initiated = false;
5297 if (kvm_set_msr(vcpu, &msr) != 0) {
5298 trace_kvm_msr_write_ex(ecx, data);
5299 kvm_inject_gp(vcpu, 0);
5300 return 1;
5301 }
5302
5303 trace_kvm_msr_write(ecx, data);
5304 skip_emulated_instruction(vcpu);
5305 return 1;
5306}
5307
5308static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
5309{
5310 kvm_make_request(KVM_REQ_EVENT, vcpu);
5311 return 1;
5312}
5313
5314static int handle_interrupt_window(struct kvm_vcpu *vcpu)
5315{
5316 u32 cpu_based_vm_exec_control;
5317
5318 /* clear pending irq */
5319 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5320 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
5321 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
5322
5323 kvm_make_request(KVM_REQ_EVENT, vcpu);
5324
5325 ++vcpu->stat.irq_window_exits;
5326
5327 /*
5328 * If the user space waits to inject interrupts, exit as soon as
5329 * possible
5330 */
5331 if (!irqchip_in_kernel(vcpu->kvm) &&
5332 vcpu->run->request_interrupt_window &&
5333 !kvm_cpu_has_interrupt(vcpu)) {
5334 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
5335 return 0;
5336 }
5337 return 1;
5338}
5339
5340static int handle_halt(struct kvm_vcpu *vcpu)
5341{
5342 skip_emulated_instruction(vcpu);
5343 return kvm_emulate_halt(vcpu);
5344}
5345
5346static int handle_vmcall(struct kvm_vcpu *vcpu)
5347{
5348 skip_emulated_instruction(vcpu);
5349 kvm_emulate_hypercall(vcpu);
5350 return 1;
5351}
5352
5353static int handle_invd(struct kvm_vcpu *vcpu)
5354{
5355 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
5356}
5357
5358static int handle_invlpg(struct kvm_vcpu *vcpu)
5359{
5360 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5361
5362 kvm_mmu_invlpg(vcpu, exit_qualification);
5363 skip_emulated_instruction(vcpu);
5364 return 1;
5365}
5366
5367static int handle_rdpmc(struct kvm_vcpu *vcpu)
5368{
5369 int err;
5370
5371 err = kvm_rdpmc(vcpu);
5372 kvm_complete_insn_gp(vcpu, err);
5373
5374 return 1;
5375}
5376
5377static int handle_wbinvd(struct kvm_vcpu *vcpu)
5378{
5379 skip_emulated_instruction(vcpu);
5380 kvm_emulate_wbinvd(vcpu);
5381 return 1;
5382}
5383
5384static int handle_xsetbv(struct kvm_vcpu *vcpu)
5385{
5386 u64 new_bv = kvm_read_edx_eax(vcpu);
5387 u32 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
5388
5389 if (kvm_set_xcr(vcpu, index, new_bv) == 0)
5390 skip_emulated_instruction(vcpu);
5391 return 1;
5392}
5393
5394static int handle_apic_access(struct kvm_vcpu *vcpu)
5395{
5396 if (likely(fasteoi)) {
5397 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5398 int access_type, offset;
5399
5400 access_type = exit_qualification & APIC_ACCESS_TYPE;
5401 offset = exit_qualification & APIC_ACCESS_OFFSET;
5402 /*
5403 * Sane guest uses MOV to write EOI, with written value
5404 * not cared. So make a short-circuit here by avoiding
5405 * heavy instruction emulation.
5406 */
5407 if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
5408 (offset == APIC_EOI)) {
5409 kvm_lapic_set_eoi(vcpu);
5410 skip_emulated_instruction(vcpu);
5411 return 1;
5412 }
5413 }
5414 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
5415}
5416
5417static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu)
5418{
5419 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5420 int vector = exit_qualification & 0xff;
5421
5422 /* EOI-induced VM exit is trap-like and thus no need to adjust IP */
5423 kvm_apic_set_eoi_accelerated(vcpu, vector);
5424 return 1;
5425}
5426
5427static int handle_apic_write(struct kvm_vcpu *vcpu)
5428{
5429 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5430 u32 offset = exit_qualification & 0xfff;
5431
5432 /* APIC-write VM exit is trap-like and thus no need to adjust IP */
5433 kvm_apic_write_nodecode(vcpu, offset);
5434 return 1;
5435}
5436
5437static int handle_task_switch(struct kvm_vcpu *vcpu)
5438{
5439 struct vcpu_vmx *vmx = to_vmx(vcpu);
5440 unsigned long exit_qualification;
5441 bool has_error_code = false;
5442 u32 error_code = 0;
5443 u16 tss_selector;
5444 int reason, type, idt_v, idt_index;
5445
5446 idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
5447 idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK);
5448 type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
5449
5450 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5451
5452 reason = (u32)exit_qualification >> 30;
5453 if (reason == TASK_SWITCH_GATE && idt_v) {
5454 switch (type) {
5455 case INTR_TYPE_NMI_INTR:
5456 vcpu->arch.nmi_injected = false;
5457 vmx_set_nmi_mask(vcpu, true);
5458 break;
5459 case INTR_TYPE_EXT_INTR:
5460 case INTR_TYPE_SOFT_INTR:
5461 kvm_clear_interrupt_queue(vcpu);
5462 break;
5463 case INTR_TYPE_HARD_EXCEPTION:
5464 if (vmx->idt_vectoring_info &
5465 VECTORING_INFO_DELIVER_CODE_MASK) {
5466 has_error_code = true;
5467 error_code =
5468 vmcs_read32(IDT_VECTORING_ERROR_CODE);
5469 }
5470 /* fall through */
5471 case INTR_TYPE_SOFT_EXCEPTION:
5472 kvm_clear_exception_queue(vcpu);
5473 break;
5474 default:
5475 break;
5476 }
5477 }
5478 tss_selector = exit_qualification;
5479
5480 if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
5481 type != INTR_TYPE_EXT_INTR &&
5482 type != INTR_TYPE_NMI_INTR))
5483 skip_emulated_instruction(vcpu);
5484
5485 if (kvm_task_switch(vcpu, tss_selector,
5486 type == INTR_TYPE_SOFT_INTR ? idt_index : -1, reason,
5487 has_error_code, error_code) == EMULATE_FAIL) {
5488 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5489 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5490 vcpu->run->internal.ndata = 0;
5491 return 0;
5492 }
5493
5494 /* clear all local breakpoint enable flags */
5495 vmcs_writel(GUEST_DR7, vmcs_readl(GUEST_DR7) & ~0x55);
5496
5497 /*
5498 * TODO: What about debug traps on tss switch?
5499 * Are we supposed to inject them and update dr6?
5500 */
5501
5502 return 1;
5503}
5504
5505static int handle_ept_violation(struct kvm_vcpu *vcpu)
5506{
5507 unsigned long exit_qualification;
5508 gpa_t gpa;
5509 u32 error_code;
5510 int gla_validity;
5511
5512 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5513
5514 gla_validity = (exit_qualification >> 7) & 0x3;
5515 if (gla_validity != 0x3 && gla_validity != 0x1 && gla_validity != 0) {
5516 printk(KERN_ERR "EPT: Handling EPT violation failed!\n");
5517 printk(KERN_ERR "EPT: GPA: 0x%lx, GVA: 0x%lx\n",
5518 (long unsigned int)vmcs_read64(GUEST_PHYSICAL_ADDRESS),
5519 vmcs_readl(GUEST_LINEAR_ADDRESS));
5520 printk(KERN_ERR "EPT: Exit qualification is 0x%lx\n",
5521 (long unsigned int)exit_qualification);
5522 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
5523 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_VIOLATION;
5524 return 0;
5525 }
5526
5527 /*
5528 * EPT violation happened while executing iret from NMI,
5529 * "blocked by NMI" bit has to be set before next VM entry.
5530 * There are errata that may cause this bit to not be set:
5531 * AAK134, BY25.
5532 */
5533 if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5534 cpu_has_virtual_nmis() &&
5535 (exit_qualification & INTR_INFO_UNBLOCK_NMI))
5536 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, GUEST_INTR_STATE_NMI);
5537
5538 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5539 trace_kvm_page_fault(gpa, exit_qualification);
5540
5541 /* It is a write fault? */
5542 error_code = exit_qualification & (1U << 1);
5543 /* It is a fetch fault? */
5544 error_code |= (exit_qualification & (1U << 2)) << 2;
5545 /* ept page table is present? */
5546 error_code |= (exit_qualification >> 3) & 0x1;
5547
5548 vcpu->arch.exit_qualification = exit_qualification;
5549
5550 return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0);
5551}
5552
5553static u64 ept_rsvd_mask(u64 spte, int level)
5554{
5555 int i;
5556 u64 mask = 0;
5557
5558 for (i = 51; i > boot_cpu_data.x86_phys_bits; i--)
5559 mask |= (1ULL << i);
5560
5561 if (level == 4)
5562 /* bits 7:3 reserved */
5563 mask |= 0xf8;
5564 else if (spte & (1ULL << 7))
5565 /*
5566 * 1GB/2MB page, bits 29:12 or 20:12 reserved respectively,
5567 * level == 1 if the hypervisor is using the ignored bit 7.
5568 */
5569 mask |= (PAGE_SIZE << ((level - 1) * 9)) - PAGE_SIZE;
5570 else if (level > 1)
5571 /* bits 6:3 reserved */
5572 mask |= 0x78;
5573
5574 return mask;
5575}
5576
5577static void ept_misconfig_inspect_spte(struct kvm_vcpu *vcpu, u64 spte,
5578 int level)
5579{
5580 printk(KERN_ERR "%s: spte 0x%llx level %d\n", __func__, spte, level);
5581
5582 /* 010b (write-only) */
5583 WARN_ON((spte & 0x7) == 0x2);
5584
5585 /* 110b (write/execute) */
5586 WARN_ON((spte & 0x7) == 0x6);
5587
5588 /* 100b (execute-only) and value not supported by logical processor */
5589 if (!cpu_has_vmx_ept_execute_only())
5590 WARN_ON((spte & 0x7) == 0x4);
5591
5592 /* not 000b */
5593 if ((spte & 0x7)) {
5594 u64 rsvd_bits = spte & ept_rsvd_mask(spte, level);
5595
5596 if (rsvd_bits != 0) {
5597 printk(KERN_ERR "%s: rsvd_bits = 0x%llx\n",
5598 __func__, rsvd_bits);
5599 WARN_ON(1);
5600 }
5601
5602 /* bits 5:3 are _not_ reserved for large page or leaf page */
5603 if ((rsvd_bits & 0x38) == 0) {
5604 u64 ept_mem_type = (spte & 0x38) >> 3;
5605
5606 if (ept_mem_type == 2 || ept_mem_type == 3 ||
5607 ept_mem_type == 7) {
5608 printk(KERN_ERR "%s: ept_mem_type=0x%llx\n",
5609 __func__, ept_mem_type);
5610 WARN_ON(1);
5611 }
5612 }
5613 }
5614}
5615
5616static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
5617{
5618 u64 sptes[4];
5619 int nr_sptes, i, ret;
5620 gpa_t gpa;
5621
5622 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5623 if (!kvm_io_bus_write(vcpu->kvm, KVM_FAST_MMIO_BUS, gpa, 0, NULL)) {
5624 skip_emulated_instruction(vcpu);
5625 return 1;
5626 }
5627
5628 ret = handle_mmio_page_fault_common(vcpu, gpa, true);
5629 if (likely(ret == RET_MMIO_PF_EMULATE))
5630 return x86_emulate_instruction(vcpu, gpa, 0, NULL, 0) ==
5631 EMULATE_DONE;
5632
5633 if (unlikely(ret == RET_MMIO_PF_INVALID))
5634 return kvm_mmu_page_fault(vcpu, gpa, 0, NULL, 0);
5635
5636 if (unlikely(ret == RET_MMIO_PF_RETRY))
5637 return 1;
5638
5639 /* It is the real ept misconfig */
5640 printk(KERN_ERR "EPT: Misconfiguration.\n");
5641 printk(KERN_ERR "EPT: GPA: 0x%llx\n", gpa);
5642
5643 nr_sptes = kvm_mmu_get_spte_hierarchy(vcpu, gpa, sptes);
5644
5645 for (i = PT64_ROOT_LEVEL; i > PT64_ROOT_LEVEL - nr_sptes; --i)
5646 ept_misconfig_inspect_spte(vcpu, sptes[i-1], i);
5647
5648 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
5649 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_MISCONFIG;
5650
5651 return 0;
5652}
5653
5654static int handle_nmi_window(struct kvm_vcpu *vcpu)
5655{
5656 u32 cpu_based_vm_exec_control;
5657
5658 /* clear pending NMI */
5659 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5660 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
5661 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
5662 ++vcpu->stat.nmi_window_exits;
5663 kvm_make_request(KVM_REQ_EVENT, vcpu);
5664
5665 return 1;
5666}
5667
5668static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
5669{
5670 struct vcpu_vmx *vmx = to_vmx(vcpu);
5671 enum emulation_result err = EMULATE_DONE;
5672 int ret = 1;
5673 u32 cpu_exec_ctrl;
5674 bool intr_window_requested;
5675 unsigned count = 130;
5676
5677 cpu_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5678 intr_window_requested = cpu_exec_ctrl & CPU_BASED_VIRTUAL_INTR_PENDING;
5679
5680 while (vmx->emulation_required && count-- != 0) {
5681 if (intr_window_requested && vmx_interrupt_allowed(vcpu))
5682 return handle_interrupt_window(&vmx->vcpu);
5683
5684 if (test_bit(KVM_REQ_EVENT, &vcpu->requests))
5685 return 1;
5686
5687 err = emulate_instruction(vcpu, EMULTYPE_NO_REEXECUTE);
5688
5689 if (err == EMULATE_USER_EXIT) {
5690 ++vcpu->stat.mmio_exits;
5691 ret = 0;
5692 goto out;
5693 }
5694
5695 if (err != EMULATE_DONE) {
5696 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5697 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5698 vcpu->run->internal.ndata = 0;
5699 return 0;
5700 }
5701
5702 if (vcpu->arch.halt_request) {
5703 vcpu->arch.halt_request = 0;
5704 ret = kvm_emulate_halt(vcpu);
5705 goto out;
5706 }
5707
5708 if (signal_pending(current))
5709 goto out;
5710 if (need_resched())
5711 schedule();
5712 }
5713
5714out:
5715 return ret;
5716}
5717
5718static int __grow_ple_window(int val)
5719{
5720 if (ple_window_grow < 1)
5721 return ple_window;
5722
5723 val = min(val, ple_window_actual_max);
5724
5725 if (ple_window_grow < ple_window)
5726 val *= ple_window_grow;
5727 else
5728 val += ple_window_grow;
5729
5730 return val;
5731}
5732
5733static int __shrink_ple_window(int val, int modifier, int minimum)
5734{
5735 if (modifier < 1)
5736 return ple_window;
5737
5738 if (modifier < ple_window)
5739 val /= modifier;
5740 else
5741 val -= modifier;
5742
5743 return max(val, minimum);
5744}
5745
5746static void grow_ple_window(struct kvm_vcpu *vcpu)
5747{
5748 struct vcpu_vmx *vmx = to_vmx(vcpu);
5749 int old = vmx->ple_window;
5750
5751 vmx->ple_window = __grow_ple_window(old);
5752
5753 if (vmx->ple_window != old)
5754 vmx->ple_window_dirty = true;
5755
5756 trace_kvm_ple_window_grow(vcpu->vcpu_id, vmx->ple_window, old);
5757}
5758
5759static void shrink_ple_window(struct kvm_vcpu *vcpu)
5760{
5761 struct vcpu_vmx *vmx = to_vmx(vcpu);
5762 int old = vmx->ple_window;
5763
5764 vmx->ple_window = __shrink_ple_window(old,
5765 ple_window_shrink, ple_window);
5766
5767 if (vmx->ple_window != old)
5768 vmx->ple_window_dirty = true;
5769
5770 trace_kvm_ple_window_shrink(vcpu->vcpu_id, vmx->ple_window, old);
5771}
5772
5773/*
5774 * ple_window_actual_max is computed to be one grow_ple_window() below
5775 * ple_window_max. (See __grow_ple_window for the reason.)
5776 * This prevents overflows, because ple_window_max is int.
5777 * ple_window_max effectively rounded down to a multiple of ple_window_grow in
5778 * this process.
5779 * ple_window_max is also prevented from setting vmx->ple_window < ple_window.
5780 */
5781static void update_ple_window_actual_max(void)
5782{
5783 ple_window_actual_max =
5784 __shrink_ple_window(max(ple_window_max, ple_window),
5785 ple_window_grow, INT_MIN);
5786}
5787
5788/*
5789 * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
5790 * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
5791 */
5792static int handle_pause(struct kvm_vcpu *vcpu)
5793{
5794 if (ple_gap)
5795 grow_ple_window(vcpu);
5796
5797 skip_emulated_instruction(vcpu);
5798 kvm_vcpu_on_spin(vcpu);
5799
5800 return 1;
5801}
5802
5803static int handle_nop(struct kvm_vcpu *vcpu)
5804{
5805 skip_emulated_instruction(vcpu);
5806 return 1;
5807}
5808
5809static int handle_mwait(struct kvm_vcpu *vcpu)
5810{
5811 printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
5812 return handle_nop(vcpu);
5813}
5814
5815static int handle_monitor(struct kvm_vcpu *vcpu)
5816{
5817 printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
5818 return handle_nop(vcpu);
5819}
5820
5821/*
5822 * To run an L2 guest, we need a vmcs02 based on the L1-specified vmcs12.
5823 * We could reuse a single VMCS for all the L2 guests, but we also want the
5824 * option to allocate a separate vmcs02 for each separate loaded vmcs12 - this
5825 * allows keeping them loaded on the processor, and in the future will allow
5826 * optimizations where prepare_vmcs02 doesn't need to set all the fields on
5827 * every entry if they never change.
5828 * So we keep, in vmx->nested.vmcs02_pool, a cache of size VMCS02_POOL_SIZE
5829 * (>=0) with a vmcs02 for each recently loaded vmcs12s, most recent first.
5830 *
5831 * The following functions allocate and free a vmcs02 in this pool.
5832 */
5833
5834/* Get a VMCS from the pool to use as vmcs02 for the current vmcs12. */
5835static struct loaded_vmcs *nested_get_current_vmcs02(struct vcpu_vmx *vmx)
5836{
5837 struct vmcs02_list *item;
5838 list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
5839 if (item->vmptr == vmx->nested.current_vmptr) {
5840 list_move(&item->list, &vmx->nested.vmcs02_pool);
5841 return &item->vmcs02;
5842 }
5843
5844 if (vmx->nested.vmcs02_num >= max(VMCS02_POOL_SIZE, 1)) {
5845 /* Recycle the least recently used VMCS. */
5846 item = list_entry(vmx->nested.vmcs02_pool.prev,
5847 struct vmcs02_list, list);
5848 item->vmptr = vmx->nested.current_vmptr;
5849 list_move(&item->list, &vmx->nested.vmcs02_pool);
5850 return &item->vmcs02;
5851 }
5852
5853 /* Create a new VMCS */
5854 item = kmalloc(sizeof(struct vmcs02_list), GFP_KERNEL);
5855 if (!item)
5856 return NULL;
5857 item->vmcs02.vmcs = alloc_vmcs();
5858 if (!item->vmcs02.vmcs) {
5859 kfree(item);
5860 return NULL;
5861 }
5862 loaded_vmcs_init(&item->vmcs02);
5863 item->vmptr = vmx->nested.current_vmptr;
5864 list_add(&(item->list), &(vmx->nested.vmcs02_pool));
5865 vmx->nested.vmcs02_num++;
5866 return &item->vmcs02;
5867}
5868
5869/* Free and remove from pool a vmcs02 saved for a vmcs12 (if there is one) */
5870static void nested_free_vmcs02(struct vcpu_vmx *vmx, gpa_t vmptr)
5871{
5872 struct vmcs02_list *item;
5873 list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
5874 if (item->vmptr == vmptr) {
5875 free_loaded_vmcs(&item->vmcs02);
5876 list_del(&item->list);
5877 kfree(item);
5878 vmx->nested.vmcs02_num--;
5879 return;
5880 }
5881}
5882
5883/*
5884 * Free all VMCSs saved for this vcpu, except the one pointed by
5885 * vmx->loaded_vmcs. We must be running L1, so vmx->loaded_vmcs
5886 * must be &vmx->vmcs01.
5887 */
5888static void nested_free_all_saved_vmcss(struct vcpu_vmx *vmx)
5889{
5890 struct vmcs02_list *item, *n;
5891
5892 WARN_ON(vmx->loaded_vmcs != &vmx->vmcs01);
5893 list_for_each_entry_safe(item, n, &vmx->nested.vmcs02_pool, list) {
5894 /*
5895 * Something will leak if the above WARN triggers. Better than
5896 * a use-after-free.
5897 */
5898 if (vmx->loaded_vmcs == &item->vmcs02)
5899 continue;
5900
5901 free_loaded_vmcs(&item->vmcs02);
5902 list_del(&item->list);
5903 kfree(item);
5904 vmx->nested.vmcs02_num--;
5905 }
5906}
5907
5908/*
5909 * The following 3 functions, nested_vmx_succeed()/failValid()/failInvalid(),
5910 * set the success or error code of an emulated VMX instruction, as specified
5911 * by Vol 2B, VMX Instruction Reference, "Conventions".
5912 */
5913static void nested_vmx_succeed(struct kvm_vcpu *vcpu)
5914{
5915 vmx_set_rflags(vcpu, vmx_get_rflags(vcpu)
5916 & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
5917 X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF));
5918}
5919
5920static void nested_vmx_failInvalid(struct kvm_vcpu *vcpu)
5921{
5922 vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
5923 & ~(X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
5924 X86_EFLAGS_SF | X86_EFLAGS_OF))
5925 | X86_EFLAGS_CF);
5926}
5927
5928static void nested_vmx_failValid(struct kvm_vcpu *vcpu,
5929 u32 vm_instruction_error)
5930{
5931 if (to_vmx(vcpu)->nested.current_vmptr == -1ull) {
5932 /*
5933 * failValid writes the error number to the current VMCS, which
5934 * can't be done there isn't a current VMCS.
5935 */
5936 nested_vmx_failInvalid(vcpu);
5937 return;
5938 }
5939 vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
5940 & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
5941 X86_EFLAGS_SF | X86_EFLAGS_OF))
5942 | X86_EFLAGS_ZF);
5943 get_vmcs12(vcpu)->vm_instruction_error = vm_instruction_error;
5944 /*
5945 * We don't need to force a shadow sync because
5946 * VM_INSTRUCTION_ERROR is not shadowed
5947 */
5948}
5949
5950static enum hrtimer_restart vmx_preemption_timer_fn(struct hrtimer *timer)
5951{
5952 struct vcpu_vmx *vmx =
5953 container_of(timer, struct vcpu_vmx, nested.preemption_timer);
5954
5955 vmx->nested.preemption_timer_expired = true;
5956 kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
5957 kvm_vcpu_kick(&vmx->vcpu);
5958
5959 return HRTIMER_NORESTART;
5960}
5961
5962/*
5963 * Decode the memory-address operand of a vmx instruction, as recorded on an
5964 * exit caused by such an instruction (run by a guest hypervisor).
5965 * On success, returns 0. When the operand is invalid, returns 1 and throws
5966 * #UD or #GP.
5967 */
5968static int get_vmx_mem_address(struct kvm_vcpu *vcpu,
5969 unsigned long exit_qualification,
5970 u32 vmx_instruction_info, gva_t *ret)
5971{
5972 /*
5973 * According to Vol. 3B, "Information for VM Exits Due to Instruction
5974 * Execution", on an exit, vmx_instruction_info holds most of the
5975 * addressing components of the operand. Only the displacement part
5976 * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
5977 * For how an actual address is calculated from all these components,
5978 * refer to Vol. 1, "Operand Addressing".
5979 */
5980 int scaling = vmx_instruction_info & 3;
5981 int addr_size = (vmx_instruction_info >> 7) & 7;
5982 bool is_reg = vmx_instruction_info & (1u << 10);
5983 int seg_reg = (vmx_instruction_info >> 15) & 7;
5984 int index_reg = (vmx_instruction_info >> 18) & 0xf;
5985 bool index_is_valid = !(vmx_instruction_info & (1u << 22));
5986 int base_reg = (vmx_instruction_info >> 23) & 0xf;
5987 bool base_is_valid = !(vmx_instruction_info & (1u << 27));
5988
5989 if (is_reg) {
5990 kvm_queue_exception(vcpu, UD_VECTOR);
5991 return 1;
5992 }
5993
5994 /* Addr = segment_base + offset */
5995 /* offset = base + [index * scale] + displacement */
5996 *ret = vmx_get_segment_base(vcpu, seg_reg);
5997 if (base_is_valid)
5998 *ret += kvm_register_read(vcpu, base_reg);
5999 if (index_is_valid)
6000 *ret += kvm_register_read(vcpu, index_reg)<<scaling;
6001 *ret += exit_qualification; /* holds the displacement */
6002
6003 if (addr_size == 1) /* 32 bit */
6004 *ret &= 0xffffffff;
6005
6006 /*
6007 * TODO: throw #GP (and return 1) in various cases that the VM*
6008 * instructions require it - e.g., offset beyond segment limit,
6009 * unusable or unreadable/unwritable segment, non-canonical 64-bit
6010 * address, and so on. Currently these are not checked.
6011 */
6012 return 0;
6013}
6014
6015/*
6016 * This function performs the various checks including
6017 * - if it's 4KB aligned
6018 * - No bits beyond the physical address width are set
6019 * - Returns 0 on success or else 1
6020 * (Intel SDM Section 30.3)
6021 */
6022static int nested_vmx_check_vmptr(struct kvm_vcpu *vcpu, int exit_reason,
6023 gpa_t *vmpointer)
6024{
6025 gva_t gva;
6026 gpa_t vmptr;
6027 struct x86_exception e;
6028 struct page *page;
6029 struct vcpu_vmx *vmx = to_vmx(vcpu);
6030 int maxphyaddr = cpuid_maxphyaddr(vcpu);
6031
6032 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
6033 vmcs_read32(VMX_INSTRUCTION_INFO), &gva))
6034 return 1;
6035
6036 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &vmptr,
6037 sizeof(vmptr), &e)) {
6038 kvm_inject_page_fault(vcpu, &e);
6039 return 1;
6040 }
6041
6042 switch (exit_reason) {
6043 case EXIT_REASON_VMON:
6044 /*
6045 * SDM 3: 24.11.5
6046 * The first 4 bytes of VMXON region contain the supported
6047 * VMCS revision identifier
6048 *
6049 * Note - IA32_VMX_BASIC[48] will never be 1
6050 * for the nested case;
6051 * which replaces physical address width with 32
6052 *
6053 */
6054 if (!PAGE_ALIGNED(vmptr) || (vmptr >> maxphyaddr)) {
6055 nested_vmx_failInvalid(vcpu);
6056 skip_emulated_instruction(vcpu);
6057 return 1;
6058 }
6059
6060 page = nested_get_page(vcpu, vmptr);
6061 if (page == NULL ||
6062 *(u32 *)kmap(page) != VMCS12_REVISION) {
6063 nested_vmx_failInvalid(vcpu);
6064 kunmap(page);
6065 skip_emulated_instruction(vcpu);
6066 return 1;
6067 }
6068 kunmap(page);
6069 vmx->nested.vmxon_ptr = vmptr;
6070 break;
6071 case EXIT_REASON_VMCLEAR:
6072 if (!PAGE_ALIGNED(vmptr) || (vmptr >> maxphyaddr)) {
6073 nested_vmx_failValid(vcpu,
6074 VMXERR_VMCLEAR_INVALID_ADDRESS);
6075 skip_emulated_instruction(vcpu);
6076 return 1;
6077 }
6078
6079 if (vmptr == vmx->nested.vmxon_ptr) {
6080 nested_vmx_failValid(vcpu,
6081 VMXERR_VMCLEAR_VMXON_POINTER);
6082 skip_emulated_instruction(vcpu);
6083 return 1;
6084 }
6085 break;
6086 case EXIT_REASON_VMPTRLD:
6087 if (!PAGE_ALIGNED(vmptr) || (vmptr >> maxphyaddr)) {
6088 nested_vmx_failValid(vcpu,
6089 VMXERR_VMPTRLD_INVALID_ADDRESS);
6090 skip_emulated_instruction(vcpu);
6091 return 1;
6092 }
6093
6094 if (vmptr == vmx->nested.vmxon_ptr) {
6095 nested_vmx_failValid(vcpu,
6096 VMXERR_VMCLEAR_VMXON_POINTER);
6097 skip_emulated_instruction(vcpu);
6098 return 1;
6099 }
6100 break;
6101 default:
6102 return 1; /* shouldn't happen */
6103 }
6104
6105 if (vmpointer)
6106 *vmpointer = vmptr;
6107 return 0;
6108}
6109
6110/*
6111 * Emulate the VMXON instruction.
6112 * Currently, we just remember that VMX is active, and do not save or even
6113 * inspect the argument to VMXON (the so-called "VMXON pointer") because we
6114 * do not currently need to store anything in that guest-allocated memory
6115 * region. Consequently, VMCLEAR and VMPTRLD also do not verify that the their
6116 * argument is different from the VMXON pointer (which the spec says they do).
6117 */
6118static int handle_vmon(struct kvm_vcpu *vcpu)
6119{
6120 struct kvm_segment cs;
6121 struct vcpu_vmx *vmx = to_vmx(vcpu);
6122 struct vmcs *shadow_vmcs;
6123 const u64 VMXON_NEEDED_FEATURES = FEATURE_CONTROL_LOCKED
6124 | FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
6125
6126 /* The Intel VMX Instruction Reference lists a bunch of bits that
6127 * are prerequisite to running VMXON, most notably cr4.VMXE must be
6128 * set to 1 (see vmx_set_cr4() for when we allow the guest to set this).
6129 * Otherwise, we should fail with #UD. We test these now:
6130 */
6131 if (!kvm_read_cr4_bits(vcpu, X86_CR4_VMXE) ||
6132 !kvm_read_cr0_bits(vcpu, X86_CR0_PE) ||
6133 (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
6134 kvm_queue_exception(vcpu, UD_VECTOR);
6135 return 1;
6136 }
6137
6138 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
6139 if (is_long_mode(vcpu) && !cs.l) {
6140 kvm_queue_exception(vcpu, UD_VECTOR);
6141 return 1;
6142 }
6143
6144 if (vmx_get_cpl(vcpu)) {
6145 kvm_inject_gp(vcpu, 0);
6146 return 1;
6147 }
6148
6149 if (nested_vmx_check_vmptr(vcpu, EXIT_REASON_VMON, NULL))
6150 return 1;
6151
6152 if (vmx->nested.vmxon) {
6153 nested_vmx_failValid(vcpu, VMXERR_VMXON_IN_VMX_ROOT_OPERATION);
6154 skip_emulated_instruction(vcpu);
6155 return 1;
6156 }
6157
6158 if ((vmx->nested.msr_ia32_feature_control & VMXON_NEEDED_FEATURES)
6159 != VMXON_NEEDED_FEATURES) {
6160 kvm_inject_gp(vcpu, 0);
6161 return 1;
6162 }
6163
6164 if (enable_shadow_vmcs) {
6165 shadow_vmcs = alloc_vmcs();
6166 if (!shadow_vmcs)
6167 return -ENOMEM;
6168 /* mark vmcs as shadow */
6169 shadow_vmcs->revision_id |= (1u << 31);
6170 /* init shadow vmcs */
6171 vmcs_clear(shadow_vmcs);
6172 vmx->nested.current_shadow_vmcs = shadow_vmcs;
6173 }
6174
6175 INIT_LIST_HEAD(&(vmx->nested.vmcs02_pool));
6176 vmx->nested.vmcs02_num = 0;
6177
6178 hrtimer_init(&vmx->nested.preemption_timer, CLOCK_MONOTONIC,
6179 HRTIMER_MODE_REL);
6180 vmx->nested.preemption_timer.function = vmx_preemption_timer_fn;
6181
6182 vmx->nested.vmxon = true;
6183
6184 skip_emulated_instruction(vcpu);
6185 nested_vmx_succeed(vcpu);
6186 return 1;
6187}
6188
6189/*
6190 * Intel's VMX Instruction Reference specifies a common set of prerequisites
6191 * for running VMX instructions (except VMXON, whose prerequisites are
6192 * slightly different). It also specifies what exception to inject otherwise.
6193 */
6194static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
6195{
6196 struct kvm_segment cs;
6197 struct vcpu_vmx *vmx = to_vmx(vcpu);
6198
6199 if (!vmx->nested.vmxon) {
6200 kvm_queue_exception(vcpu, UD_VECTOR);
6201 return 0;
6202 }
6203
6204 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
6205 if ((vmx_get_rflags(vcpu) & X86_EFLAGS_VM) ||
6206 (is_long_mode(vcpu) && !cs.l)) {
6207 kvm_queue_exception(vcpu, UD_VECTOR);
6208 return 0;
6209 }
6210
6211 if (vmx_get_cpl(vcpu)) {
6212 kvm_inject_gp(vcpu, 0);
6213 return 0;
6214 }
6215
6216 return 1;
6217}
6218
6219static inline void nested_release_vmcs12(struct vcpu_vmx *vmx)
6220{
6221 u32 exec_control;
6222 if (vmx->nested.current_vmptr == -1ull)
6223 return;
6224
6225 /* current_vmptr and current_vmcs12 are always set/reset together */
6226 if (WARN_ON(vmx->nested.current_vmcs12 == NULL))
6227 return;
6228
6229 if (enable_shadow_vmcs) {
6230 /* copy to memory all shadowed fields in case
6231 they were modified */
6232 copy_shadow_to_vmcs12(vmx);
6233 vmx->nested.sync_shadow_vmcs = false;
6234 exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
6235 exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
6236 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
6237 vmcs_write64(VMCS_LINK_POINTER, -1ull);
6238 }
6239 kunmap(vmx->nested.current_vmcs12_page);
6240 nested_release_page(vmx->nested.current_vmcs12_page);
6241 vmx->nested.current_vmptr = -1ull;
6242 vmx->nested.current_vmcs12 = NULL;
6243}
6244
6245/*
6246 * Free whatever needs to be freed from vmx->nested when L1 goes down, or
6247 * just stops using VMX.
6248 */
6249static void free_nested(struct vcpu_vmx *vmx)
6250{
6251 if (!vmx->nested.vmxon)
6252 return;
6253
6254 vmx->nested.vmxon = false;
6255 nested_release_vmcs12(vmx);
6256 if (enable_shadow_vmcs)
6257 free_vmcs(vmx->nested.current_shadow_vmcs);
6258 /* Unpin physical memory we referred to in current vmcs02 */
6259 if (vmx->nested.apic_access_page) {
6260 nested_release_page(vmx->nested.apic_access_page);
6261 vmx->nested.apic_access_page = NULL;
6262 }
6263 if (vmx->nested.virtual_apic_page) {
6264 nested_release_page(vmx->nested.virtual_apic_page);
6265 vmx->nested.virtual_apic_page = NULL;
6266 }
6267
6268 nested_free_all_saved_vmcss(vmx);
6269}
6270
6271/* Emulate the VMXOFF instruction */
6272static int handle_vmoff(struct kvm_vcpu *vcpu)
6273{
6274 if (!nested_vmx_check_permission(vcpu))
6275 return 1;
6276 free_nested(to_vmx(vcpu));
6277 skip_emulated_instruction(vcpu);
6278 nested_vmx_succeed(vcpu);
6279 return 1;
6280}
6281
6282/* Emulate the VMCLEAR instruction */
6283static int handle_vmclear(struct kvm_vcpu *vcpu)
6284{
6285 struct vcpu_vmx *vmx = to_vmx(vcpu);
6286 gpa_t vmptr;
6287 struct vmcs12 *vmcs12;
6288 struct page *page;
6289
6290 if (!nested_vmx_check_permission(vcpu))
6291 return 1;
6292
6293 if (nested_vmx_check_vmptr(vcpu, EXIT_REASON_VMCLEAR, &vmptr))
6294 return 1;
6295
6296 if (vmptr == vmx->nested.current_vmptr)
6297 nested_release_vmcs12(vmx);
6298
6299 page = nested_get_page(vcpu, vmptr);
6300 if (page == NULL) {
6301 /*
6302 * For accurate processor emulation, VMCLEAR beyond available
6303 * physical memory should do nothing at all. However, it is
6304 * possible that a nested vmx bug, not a guest hypervisor bug,
6305 * resulted in this case, so let's shut down before doing any
6306 * more damage:
6307 */
6308 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
6309 return 1;
6310 }
6311 vmcs12 = kmap(page);
6312 vmcs12->launch_state = 0;
6313 kunmap(page);
6314 nested_release_page(page);
6315
6316 nested_free_vmcs02(vmx, vmptr);
6317
6318 skip_emulated_instruction(vcpu);
6319 nested_vmx_succeed(vcpu);
6320 return 1;
6321}
6322
6323static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch);
6324
6325/* Emulate the VMLAUNCH instruction */
6326static int handle_vmlaunch(struct kvm_vcpu *vcpu)
6327{
6328 return nested_vmx_run(vcpu, true);
6329}
6330
6331/* Emulate the VMRESUME instruction */
6332static int handle_vmresume(struct kvm_vcpu *vcpu)
6333{
6334
6335 return nested_vmx_run(vcpu, false);
6336}
6337
6338enum vmcs_field_type {
6339 VMCS_FIELD_TYPE_U16 = 0,
6340 VMCS_FIELD_TYPE_U64 = 1,
6341 VMCS_FIELD_TYPE_U32 = 2,
6342 VMCS_FIELD_TYPE_NATURAL_WIDTH = 3
6343};
6344
6345static inline int vmcs_field_type(unsigned long field)
6346{
6347 if (0x1 & field) /* the *_HIGH fields are all 32 bit */
6348 return VMCS_FIELD_TYPE_U32;
6349 return (field >> 13) & 0x3 ;
6350}
6351
6352static inline int vmcs_field_readonly(unsigned long field)
6353{
6354 return (((field >> 10) & 0x3) == 1);
6355}
6356
6357/*
6358 * Read a vmcs12 field. Since these can have varying lengths and we return
6359 * one type, we chose the biggest type (u64) and zero-extend the return value
6360 * to that size. Note that the caller, handle_vmread, might need to use only
6361 * some of the bits we return here (e.g., on 32-bit guests, only 32 bits of
6362 * 64-bit fields are to be returned).
6363 */
6364static inline bool vmcs12_read_any(struct kvm_vcpu *vcpu,
6365 unsigned long field, u64 *ret)
6366{
6367 short offset = vmcs_field_to_offset(field);
6368 char *p;
6369
6370 if (offset < 0)
6371 return 0;
6372
6373 p = ((char *)(get_vmcs12(vcpu))) + offset;
6374
6375 switch (vmcs_field_type(field)) {
6376 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
6377 *ret = *((natural_width *)p);
6378 return 1;
6379 case VMCS_FIELD_TYPE_U16:
6380 *ret = *((u16 *)p);
6381 return 1;
6382 case VMCS_FIELD_TYPE_U32:
6383 *ret = *((u32 *)p);
6384 return 1;
6385 case VMCS_FIELD_TYPE_U64:
6386 *ret = *((u64 *)p);
6387 return 1;
6388 default:
6389 return 0; /* can never happen. */
6390 }
6391}
6392
6393
6394static inline bool vmcs12_write_any(struct kvm_vcpu *vcpu,
6395 unsigned long field, u64 field_value){
6396 short offset = vmcs_field_to_offset(field);
6397 char *p = ((char *) get_vmcs12(vcpu)) + offset;
6398 if (offset < 0)
6399 return false;
6400
6401 switch (vmcs_field_type(field)) {
6402 case VMCS_FIELD_TYPE_U16:
6403 *(u16 *)p = field_value;
6404 return true;
6405 case VMCS_FIELD_TYPE_U32:
6406 *(u32 *)p = field_value;
6407 return true;
6408 case VMCS_FIELD_TYPE_U64:
6409 *(u64 *)p = field_value;
6410 return true;
6411 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
6412 *(natural_width *)p = field_value;
6413 return true;
6414 default:
6415 return false; /* can never happen. */
6416 }
6417
6418}
6419
6420static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx)
6421{
6422 int i;
6423 unsigned long field;
6424 u64 field_value;
6425 struct vmcs *shadow_vmcs = vmx->nested.current_shadow_vmcs;
6426 const unsigned long *fields = shadow_read_write_fields;
6427 const int num_fields = max_shadow_read_write_fields;
6428
6429 vmcs_load(shadow_vmcs);
6430
6431 for (i = 0; i < num_fields; i++) {
6432 field = fields[i];
6433 switch (vmcs_field_type(field)) {
6434 case VMCS_FIELD_TYPE_U16:
6435 field_value = vmcs_read16(field);
6436 break;
6437 case VMCS_FIELD_TYPE_U32:
6438 field_value = vmcs_read32(field);
6439 break;
6440 case VMCS_FIELD_TYPE_U64:
6441 field_value = vmcs_read64(field);
6442 break;
6443 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
6444 field_value = vmcs_readl(field);
6445 break;
6446 }
6447 vmcs12_write_any(&vmx->vcpu, field, field_value);
6448 }
6449
6450 vmcs_clear(shadow_vmcs);
6451 vmcs_load(vmx->loaded_vmcs->vmcs);
6452}
6453
6454static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx)
6455{
6456 const unsigned long *fields[] = {
6457 shadow_read_write_fields,
6458 shadow_read_only_fields
6459 };
6460 const int max_fields[] = {
6461 max_shadow_read_write_fields,
6462 max_shadow_read_only_fields
6463 };
6464 int i, q;
6465 unsigned long field;
6466 u64 field_value = 0;
6467 struct vmcs *shadow_vmcs = vmx->nested.current_shadow_vmcs;
6468
6469 vmcs_load(shadow_vmcs);
6470
6471 for (q = 0; q < ARRAY_SIZE(fields); q++) {
6472 for (i = 0; i < max_fields[q]; i++) {
6473 field = fields[q][i];
6474 vmcs12_read_any(&vmx->vcpu, field, &field_value);
6475
6476 switch (vmcs_field_type(field)) {
6477 case VMCS_FIELD_TYPE_U16:
6478 vmcs_write16(field, (u16)field_value);
6479 break;
6480 case VMCS_FIELD_TYPE_U32:
6481 vmcs_write32(field, (u32)field_value);
6482 break;
6483 case VMCS_FIELD_TYPE_U64:
6484 vmcs_write64(field, (u64)field_value);
6485 break;
6486 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
6487 vmcs_writel(field, (long)field_value);
6488 break;
6489 }
6490 }
6491 }
6492
6493 vmcs_clear(shadow_vmcs);
6494 vmcs_load(vmx->loaded_vmcs->vmcs);
6495}
6496
6497/*
6498 * VMX instructions which assume a current vmcs12 (i.e., that VMPTRLD was
6499 * used before) all generate the same failure when it is missing.
6500 */
6501static int nested_vmx_check_vmcs12(struct kvm_vcpu *vcpu)
6502{
6503 struct vcpu_vmx *vmx = to_vmx(vcpu);
6504 if (vmx->nested.current_vmptr == -1ull) {
6505 nested_vmx_failInvalid(vcpu);
6506 skip_emulated_instruction(vcpu);
6507 return 0;
6508 }
6509 return 1;
6510}
6511
6512static int handle_vmread(struct kvm_vcpu *vcpu)
6513{
6514 unsigned long field;
6515 u64 field_value;
6516 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6517 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
6518 gva_t gva = 0;
6519
6520 if (!nested_vmx_check_permission(vcpu) ||
6521 !nested_vmx_check_vmcs12(vcpu))
6522 return 1;
6523
6524 /* Decode instruction info and find the field to read */
6525 field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
6526 /* Read the field, zero-extended to a u64 field_value */
6527 if (!vmcs12_read_any(vcpu, field, &field_value)) {
6528 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
6529 skip_emulated_instruction(vcpu);
6530 return 1;
6531 }
6532 /*
6533 * Now copy part of this value to register or memory, as requested.
6534 * Note that the number of bits actually copied is 32 or 64 depending
6535 * on the guest's mode (32 or 64 bit), not on the given field's length.
6536 */
6537 if (vmx_instruction_info & (1u << 10)) {
6538 kvm_register_writel(vcpu, (((vmx_instruction_info) >> 3) & 0xf),
6539 field_value);
6540 } else {
6541 if (get_vmx_mem_address(vcpu, exit_qualification,
6542 vmx_instruction_info, &gva))
6543 return 1;
6544 /* _system ok, as nested_vmx_check_permission verified cpl=0 */
6545 kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, gva,
6546 &field_value, (is_long_mode(vcpu) ? 8 : 4), NULL);
6547 }
6548
6549 nested_vmx_succeed(vcpu);
6550 skip_emulated_instruction(vcpu);
6551 return 1;
6552}
6553
6554
6555static int handle_vmwrite(struct kvm_vcpu *vcpu)
6556{
6557 unsigned long field;
6558 gva_t gva;
6559 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6560 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
6561 /* The value to write might be 32 or 64 bits, depending on L1's long
6562 * mode, and eventually we need to write that into a field of several
6563 * possible lengths. The code below first zero-extends the value to 64
6564 * bit (field_value), and then copies only the approriate number of
6565 * bits into the vmcs12 field.
6566 */
6567 u64 field_value = 0;
6568 struct x86_exception e;
6569
6570 if (!nested_vmx_check_permission(vcpu) ||
6571 !nested_vmx_check_vmcs12(vcpu))
6572 return 1;
6573
6574 if (vmx_instruction_info & (1u << 10))
6575 field_value = kvm_register_readl(vcpu,
6576 (((vmx_instruction_info) >> 3) & 0xf));
6577 else {
6578 if (get_vmx_mem_address(vcpu, exit_qualification,
6579 vmx_instruction_info, &gva))
6580 return 1;
6581 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva,
6582 &field_value, (is_64_bit_mode(vcpu) ? 8 : 4), &e)) {
6583 kvm_inject_page_fault(vcpu, &e);
6584 return 1;
6585 }
6586 }
6587
6588
6589 field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
6590 if (vmcs_field_readonly(field)) {
6591 nested_vmx_failValid(vcpu,
6592 VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
6593 skip_emulated_instruction(vcpu);
6594 return 1;
6595 }
6596
6597 if (!vmcs12_write_any(vcpu, field, field_value)) {
6598 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
6599 skip_emulated_instruction(vcpu);
6600 return 1;
6601 }
6602
6603 nested_vmx_succeed(vcpu);
6604 skip_emulated_instruction(vcpu);
6605 return 1;
6606}
6607
6608/* Emulate the VMPTRLD instruction */
6609static int handle_vmptrld(struct kvm_vcpu *vcpu)
6610{
6611 struct vcpu_vmx *vmx = to_vmx(vcpu);
6612 gpa_t vmptr;
6613 u32 exec_control;
6614
6615 if (!nested_vmx_check_permission(vcpu))
6616 return 1;
6617
6618 if (nested_vmx_check_vmptr(vcpu, EXIT_REASON_VMPTRLD, &vmptr))
6619 return 1;
6620
6621 if (vmx->nested.current_vmptr != vmptr) {
6622 struct vmcs12 *new_vmcs12;
6623 struct page *page;
6624 page = nested_get_page(vcpu, vmptr);
6625 if (page == NULL) {
6626 nested_vmx_failInvalid(vcpu);
6627 skip_emulated_instruction(vcpu);
6628 return 1;
6629 }
6630 new_vmcs12 = kmap(page);
6631 if (new_vmcs12->revision_id != VMCS12_REVISION) {
6632 kunmap(page);
6633 nested_release_page_clean(page);
6634 nested_vmx_failValid(vcpu,
6635 VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
6636 skip_emulated_instruction(vcpu);
6637 return 1;
6638 }
6639
6640 nested_release_vmcs12(vmx);
6641 vmx->nested.current_vmptr = vmptr;
6642 vmx->nested.current_vmcs12 = new_vmcs12;
6643 vmx->nested.current_vmcs12_page = page;
6644 if (enable_shadow_vmcs) {
6645 exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
6646 exec_control |= SECONDARY_EXEC_SHADOW_VMCS;
6647 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
6648 vmcs_write64(VMCS_LINK_POINTER,
6649 __pa(vmx->nested.current_shadow_vmcs));
6650 vmx->nested.sync_shadow_vmcs = true;
6651 }
6652 }
6653
6654 nested_vmx_succeed(vcpu);
6655 skip_emulated_instruction(vcpu);
6656 return 1;
6657}
6658
6659/* Emulate the VMPTRST instruction */
6660static int handle_vmptrst(struct kvm_vcpu *vcpu)
6661{
6662 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6663 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
6664 gva_t vmcs_gva;
6665 struct x86_exception e;
6666
6667 if (!nested_vmx_check_permission(vcpu))
6668 return 1;
6669
6670 if (get_vmx_mem_address(vcpu, exit_qualification,
6671 vmx_instruction_info, &vmcs_gva))
6672 return 1;
6673 /* ok to use *_system, as nested_vmx_check_permission verified cpl=0 */
6674 if (kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, vmcs_gva,
6675 (void *)&to_vmx(vcpu)->nested.current_vmptr,
6676 sizeof(u64), &e)) {
6677 kvm_inject_page_fault(vcpu, &e);
6678 return 1;
6679 }
6680 nested_vmx_succeed(vcpu);
6681 skip_emulated_instruction(vcpu);
6682 return 1;
6683}
6684
6685/* Emulate the INVEPT instruction */
6686static int handle_invept(struct kvm_vcpu *vcpu)
6687{
6688 u32 vmx_instruction_info, types;
6689 unsigned long type;
6690 gva_t gva;
6691 struct x86_exception e;
6692 struct {
6693 u64 eptp, gpa;
6694 } operand;
6695
6696 if (!(nested_vmx_secondary_ctls_high & SECONDARY_EXEC_ENABLE_EPT) ||
6697 !(nested_vmx_ept_caps & VMX_EPT_INVEPT_BIT)) {
6698 kvm_queue_exception(vcpu, UD_VECTOR);
6699 return 1;
6700 }
6701
6702 if (!nested_vmx_check_permission(vcpu))
6703 return 1;
6704
6705 if (!kvm_read_cr0_bits(vcpu, X86_CR0_PE)) {
6706 kvm_queue_exception(vcpu, UD_VECTOR);
6707 return 1;
6708 }
6709
6710 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
6711 type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
6712
6713 types = (nested_vmx_ept_caps >> VMX_EPT_EXTENT_SHIFT) & 6;
6714
6715 if (!(types & (1UL << type))) {
6716 nested_vmx_failValid(vcpu,
6717 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
6718 return 1;
6719 }
6720
6721 /* According to the Intel VMX instruction reference, the memory
6722 * operand is read even if it isn't needed (e.g., for type==global)
6723 */
6724 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
6725 vmx_instruction_info, &gva))
6726 return 1;
6727 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &operand,
6728 sizeof(operand), &e)) {
6729 kvm_inject_page_fault(vcpu, &e);
6730 return 1;
6731 }
6732
6733 switch (type) {
6734 case VMX_EPT_EXTENT_GLOBAL:
6735 kvm_mmu_sync_roots(vcpu);
6736 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
6737 nested_vmx_succeed(vcpu);
6738 break;
6739 default:
6740 /* Trap single context invalidation invept calls */
6741 BUG_ON(1);
6742 break;
6743 }
6744
6745 skip_emulated_instruction(vcpu);
6746 return 1;
6747}
6748
6749static int handle_invvpid(struct kvm_vcpu *vcpu)
6750{
6751 kvm_queue_exception(vcpu, UD_VECTOR);
6752 return 1;
6753}
6754
6755/*
6756 * The exit handlers return 1 if the exit was handled fully and guest execution
6757 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
6758 * to be done to userspace and return 0.
6759 */
6760static int (*const kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
6761 [EXIT_REASON_EXCEPTION_NMI] = handle_exception,
6762 [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt,
6763 [EXIT_REASON_TRIPLE_FAULT] = handle_triple_fault,
6764 [EXIT_REASON_NMI_WINDOW] = handle_nmi_window,
6765 [EXIT_REASON_IO_INSTRUCTION] = handle_io,
6766 [EXIT_REASON_CR_ACCESS] = handle_cr,
6767 [EXIT_REASON_DR_ACCESS] = handle_dr,
6768 [EXIT_REASON_CPUID] = handle_cpuid,
6769 [EXIT_REASON_MSR_READ] = handle_rdmsr,
6770 [EXIT_REASON_MSR_WRITE] = handle_wrmsr,
6771 [EXIT_REASON_PENDING_INTERRUPT] = handle_interrupt_window,
6772 [EXIT_REASON_HLT] = handle_halt,
6773 [EXIT_REASON_INVD] = handle_invd,
6774 [EXIT_REASON_INVLPG] = handle_invlpg,
6775 [EXIT_REASON_RDPMC] = handle_rdpmc,
6776 [EXIT_REASON_VMCALL] = handle_vmcall,
6777 [EXIT_REASON_VMCLEAR] = handle_vmclear,
6778 [EXIT_REASON_VMLAUNCH] = handle_vmlaunch,
6779 [EXIT_REASON_VMPTRLD] = handle_vmptrld,
6780 [EXIT_REASON_VMPTRST] = handle_vmptrst,
6781 [EXIT_REASON_VMREAD] = handle_vmread,
6782 [EXIT_REASON_VMRESUME] = handle_vmresume,
6783 [EXIT_REASON_VMWRITE] = handle_vmwrite,
6784 [EXIT_REASON_VMOFF] = handle_vmoff,
6785 [EXIT_REASON_VMON] = handle_vmon,
6786 [EXIT_REASON_TPR_BELOW_THRESHOLD] = handle_tpr_below_threshold,
6787 [EXIT_REASON_APIC_ACCESS] = handle_apic_access,
6788 [EXIT_REASON_APIC_WRITE] = handle_apic_write,
6789 [EXIT_REASON_EOI_INDUCED] = handle_apic_eoi_induced,
6790 [EXIT_REASON_WBINVD] = handle_wbinvd,
6791 [EXIT_REASON_XSETBV] = handle_xsetbv,
6792 [EXIT_REASON_TASK_SWITCH] = handle_task_switch,
6793 [EXIT_REASON_MCE_DURING_VMENTRY] = handle_machine_check,
6794 [EXIT_REASON_EPT_VIOLATION] = handle_ept_violation,
6795 [EXIT_REASON_EPT_MISCONFIG] = handle_ept_misconfig,
6796 [EXIT_REASON_PAUSE_INSTRUCTION] = handle_pause,
6797 [EXIT_REASON_MWAIT_INSTRUCTION] = handle_mwait,
6798 [EXIT_REASON_MONITOR_INSTRUCTION] = handle_monitor,
6799 [EXIT_REASON_INVEPT] = handle_invept,
6800 [EXIT_REASON_INVVPID] = handle_invvpid,
6801};
6802
6803static const int kvm_vmx_max_exit_handlers =
6804 ARRAY_SIZE(kvm_vmx_exit_handlers);
6805
6806static bool nested_vmx_exit_handled_io(struct kvm_vcpu *vcpu,
6807 struct vmcs12 *vmcs12)
6808{
6809 unsigned long exit_qualification;
6810 gpa_t bitmap, last_bitmap;
6811 unsigned int port;
6812 int size;
6813 u8 b;
6814
6815 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
6816 return nested_cpu_has(vmcs12, CPU_BASED_UNCOND_IO_EXITING);
6817
6818 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6819
6820 port = exit_qualification >> 16;
6821 size = (exit_qualification & 7) + 1;
6822
6823 last_bitmap = (gpa_t)-1;
6824 b = -1;
6825
6826 while (size > 0) {
6827 if (port < 0x8000)
6828 bitmap = vmcs12->io_bitmap_a;
6829 else if (port < 0x10000)
6830 bitmap = vmcs12->io_bitmap_b;
6831 else
6832 return 1;
6833 bitmap += (port & 0x7fff) / 8;
6834
6835 if (last_bitmap != bitmap)
6836 if (kvm_read_guest(vcpu->kvm, bitmap, &b, 1))
6837 return 1;
6838 if (b & (1 << (port & 7)))
6839 return 1;
6840
6841 port++;
6842 size--;
6843 last_bitmap = bitmap;
6844 }
6845
6846 return 0;
6847}
6848
6849/*
6850 * Return 1 if we should exit from L2 to L1 to handle an MSR access access,
6851 * rather than handle it ourselves in L0. I.e., check whether L1 expressed
6852 * disinterest in the current event (read or write a specific MSR) by using an
6853 * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps.
6854 */
6855static bool nested_vmx_exit_handled_msr(struct kvm_vcpu *vcpu,
6856 struct vmcs12 *vmcs12, u32 exit_reason)
6857{
6858 u32 msr_index = vcpu->arch.regs[VCPU_REGS_RCX];
6859 gpa_t bitmap;
6860
6861 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
6862 return 1;
6863
6864 /*
6865 * The MSR_BITMAP page is divided into four 1024-byte bitmaps,
6866 * for the four combinations of read/write and low/high MSR numbers.
6867 * First we need to figure out which of the four to use:
6868 */
6869 bitmap = vmcs12->msr_bitmap;
6870 if (exit_reason == EXIT_REASON_MSR_WRITE)
6871 bitmap += 2048;
6872 if (msr_index >= 0xc0000000) {
6873 msr_index -= 0xc0000000;
6874 bitmap += 1024;
6875 }
6876
6877 /* Then read the msr_index'th bit from this bitmap: */
6878 if (msr_index < 1024*8) {
6879 unsigned char b;
6880 if (kvm_read_guest(vcpu->kvm, bitmap + msr_index/8, &b, 1))
6881 return 1;
6882 return 1 & (b >> (msr_index & 7));
6883 } else
6884 return 1; /* let L1 handle the wrong parameter */
6885}
6886
6887/*
6888 * Return 1 if we should exit from L2 to L1 to handle a CR access exit,
6889 * rather than handle it ourselves in L0. I.e., check if L1 wanted to
6890 * intercept (via guest_host_mask etc.) the current event.
6891 */
6892static bool nested_vmx_exit_handled_cr(struct kvm_vcpu *vcpu,
6893 struct vmcs12 *vmcs12)
6894{
6895 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6896 int cr = exit_qualification & 15;
6897 int reg = (exit_qualification >> 8) & 15;
6898 unsigned long val = kvm_register_readl(vcpu, reg);
6899
6900 switch ((exit_qualification >> 4) & 3) {
6901 case 0: /* mov to cr */
6902 switch (cr) {
6903 case 0:
6904 if (vmcs12->cr0_guest_host_mask &
6905 (val ^ vmcs12->cr0_read_shadow))
6906 return 1;
6907 break;
6908 case 3:
6909 if ((vmcs12->cr3_target_count >= 1 &&
6910 vmcs12->cr3_target_value0 == val) ||
6911 (vmcs12->cr3_target_count >= 2 &&
6912 vmcs12->cr3_target_value1 == val) ||
6913 (vmcs12->cr3_target_count >= 3 &&
6914 vmcs12->cr3_target_value2 == val) ||
6915 (vmcs12->cr3_target_count >= 4 &&
6916 vmcs12->cr3_target_value3 == val))
6917 return 0;
6918 if (nested_cpu_has(vmcs12, CPU_BASED_CR3_LOAD_EXITING))
6919 return 1;
6920 break;
6921 case 4:
6922 if (vmcs12->cr4_guest_host_mask &
6923 (vmcs12->cr4_read_shadow ^ val))
6924 return 1;
6925 break;
6926 case 8:
6927 if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING))
6928 return 1;
6929 break;
6930 }
6931 break;
6932 case 2: /* clts */
6933 if ((vmcs12->cr0_guest_host_mask & X86_CR0_TS) &&
6934 (vmcs12->cr0_read_shadow & X86_CR0_TS))
6935 return 1;
6936 break;
6937 case 1: /* mov from cr */
6938 switch (cr) {
6939 case 3:
6940 if (vmcs12->cpu_based_vm_exec_control &
6941 CPU_BASED_CR3_STORE_EXITING)
6942 return 1;
6943 break;
6944 case 8:
6945 if (vmcs12->cpu_based_vm_exec_control &
6946 CPU_BASED_CR8_STORE_EXITING)
6947 return 1;
6948 break;
6949 }
6950 break;
6951 case 3: /* lmsw */
6952 /*
6953 * lmsw can change bits 1..3 of cr0, and only set bit 0 of
6954 * cr0. Other attempted changes are ignored, with no exit.
6955 */
6956 if (vmcs12->cr0_guest_host_mask & 0xe &
6957 (val ^ vmcs12->cr0_read_shadow))
6958 return 1;
6959 if ((vmcs12->cr0_guest_host_mask & 0x1) &&
6960 !(vmcs12->cr0_read_shadow & 0x1) &&
6961 (val & 0x1))
6962 return 1;
6963 break;
6964 }
6965 return 0;
6966}
6967
6968/*
6969 * Return 1 if we should exit from L2 to L1 to handle an exit, or 0 if we
6970 * should handle it ourselves in L0 (and then continue L2). Only call this
6971 * when in is_guest_mode (L2).
6972 */
6973static bool nested_vmx_exit_handled(struct kvm_vcpu *vcpu)
6974{
6975 u32 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6976 struct vcpu_vmx *vmx = to_vmx(vcpu);
6977 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6978 u32 exit_reason = vmx->exit_reason;
6979
6980 trace_kvm_nested_vmexit(kvm_rip_read(vcpu), exit_reason,
6981 vmcs_readl(EXIT_QUALIFICATION),
6982 vmx->idt_vectoring_info,
6983 intr_info,
6984 vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
6985 KVM_ISA_VMX);
6986
6987 if (vmx->nested.nested_run_pending)
6988 return 0;
6989
6990 if (unlikely(vmx->fail)) {
6991 pr_info_ratelimited("%s failed vm entry %x\n", __func__,
6992 vmcs_read32(VM_INSTRUCTION_ERROR));
6993 return 1;
6994 }
6995
6996 switch (exit_reason) {
6997 case EXIT_REASON_EXCEPTION_NMI:
6998 if (!is_exception(intr_info))
6999 return 0;
7000 else if (is_page_fault(intr_info))
7001 return enable_ept;
7002 else if (is_no_device(intr_info) &&
7003 !(vmcs12->guest_cr0 & X86_CR0_TS))
7004 return 0;
7005 return vmcs12->exception_bitmap &
7006 (1u << (intr_info & INTR_INFO_VECTOR_MASK));
7007 case EXIT_REASON_EXTERNAL_INTERRUPT:
7008 return 0;
7009 case EXIT_REASON_TRIPLE_FAULT:
7010 return 1;
7011 case EXIT_REASON_PENDING_INTERRUPT:
7012 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_INTR_PENDING);
7013 case EXIT_REASON_NMI_WINDOW:
7014 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_NMI_PENDING);
7015 case EXIT_REASON_TASK_SWITCH:
7016 return 1;
7017 case EXIT_REASON_CPUID:
7018 if (kvm_register_read(vcpu, VCPU_REGS_RAX) == 0xa)
7019 return 0;
7020 return 1;
7021 case EXIT_REASON_HLT:
7022 return nested_cpu_has(vmcs12, CPU_BASED_HLT_EXITING);
7023 case EXIT_REASON_INVD:
7024 return 1;
7025 case EXIT_REASON_INVLPG:
7026 return nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
7027 case EXIT_REASON_RDPMC:
7028 return nested_cpu_has(vmcs12, CPU_BASED_RDPMC_EXITING);
7029 case EXIT_REASON_RDTSC:
7030 return nested_cpu_has(vmcs12, CPU_BASED_RDTSC_EXITING);
7031 case EXIT_REASON_VMCALL: case EXIT_REASON_VMCLEAR:
7032 case EXIT_REASON_VMLAUNCH: case EXIT_REASON_VMPTRLD:
7033 case EXIT_REASON_VMPTRST: case EXIT_REASON_VMREAD:
7034 case EXIT_REASON_VMRESUME: case EXIT_REASON_VMWRITE:
7035 case EXIT_REASON_VMOFF: case EXIT_REASON_VMON:
7036 case EXIT_REASON_INVEPT: case EXIT_REASON_INVVPID:
7037 /*
7038 * VMX instructions trap unconditionally. This allows L1 to
7039 * emulate them for its L2 guest, i.e., allows 3-level nesting!
7040 */
7041 return 1;
7042 case EXIT_REASON_CR_ACCESS:
7043 return nested_vmx_exit_handled_cr(vcpu, vmcs12);
7044 case EXIT_REASON_DR_ACCESS:
7045 return nested_cpu_has(vmcs12, CPU_BASED_MOV_DR_EXITING);
7046 case EXIT_REASON_IO_INSTRUCTION:
7047 return nested_vmx_exit_handled_io(vcpu, vmcs12);
7048 case EXIT_REASON_MSR_READ:
7049 case EXIT_REASON_MSR_WRITE:
7050 return nested_vmx_exit_handled_msr(vcpu, vmcs12, exit_reason);
7051 case EXIT_REASON_INVALID_STATE:
7052 return 1;
7053 case EXIT_REASON_MWAIT_INSTRUCTION:
7054 return nested_cpu_has(vmcs12, CPU_BASED_MWAIT_EXITING);
7055 case EXIT_REASON_MONITOR_INSTRUCTION:
7056 return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_EXITING);
7057 case EXIT_REASON_PAUSE_INSTRUCTION:
7058 return nested_cpu_has(vmcs12, CPU_BASED_PAUSE_EXITING) ||
7059 nested_cpu_has2(vmcs12,
7060 SECONDARY_EXEC_PAUSE_LOOP_EXITING);
7061 case EXIT_REASON_MCE_DURING_VMENTRY:
7062 return 0;
7063 case EXIT_REASON_TPR_BELOW_THRESHOLD:
7064 return nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW);
7065 case EXIT_REASON_APIC_ACCESS:
7066 return nested_cpu_has2(vmcs12,
7067 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
7068 case EXIT_REASON_EPT_VIOLATION:
7069 /*
7070 * L0 always deals with the EPT violation. If nested EPT is
7071 * used, and the nested mmu code discovers that the address is
7072 * missing in the guest EPT table (EPT12), the EPT violation
7073 * will be injected with nested_ept_inject_page_fault()
7074 */
7075 return 0;
7076 case EXIT_REASON_EPT_MISCONFIG:
7077 /*
7078 * L2 never uses directly L1's EPT, but rather L0's own EPT
7079 * table (shadow on EPT) or a merged EPT table that L0 built
7080 * (EPT on EPT). So any problems with the structure of the
7081 * table is L0's fault.
7082 */
7083 return 0;
7084 case EXIT_REASON_WBINVD:
7085 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_WBINVD_EXITING);
7086 case EXIT_REASON_XSETBV:
7087 return 1;
7088 default:
7089 return 1;
7090 }
7091}
7092
7093static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
7094{
7095 *info1 = vmcs_readl(EXIT_QUALIFICATION);
7096 *info2 = vmcs_read32(VM_EXIT_INTR_INFO);
7097}
7098
7099/*
7100 * The guest has exited. See if we can fix it or if we need userspace
7101 * assistance.
7102 */
7103static int vmx_handle_exit(struct kvm_vcpu *vcpu)
7104{
7105 struct vcpu_vmx *vmx = to_vmx(vcpu);
7106 u32 exit_reason = vmx->exit_reason;
7107 u32 vectoring_info = vmx->idt_vectoring_info;
7108
7109 /* If guest state is invalid, start emulating */
7110 if (vmx->emulation_required)
7111 return handle_invalid_guest_state(vcpu);
7112
7113 if (is_guest_mode(vcpu) && nested_vmx_exit_handled(vcpu)) {
7114 nested_vmx_vmexit(vcpu, exit_reason,
7115 vmcs_read32(VM_EXIT_INTR_INFO),
7116 vmcs_readl(EXIT_QUALIFICATION));
7117 return 1;
7118 }
7119
7120 if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
7121 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
7122 vcpu->run->fail_entry.hardware_entry_failure_reason
7123 = exit_reason;
7124 return 0;
7125 }
7126
7127 if (unlikely(vmx->fail)) {
7128 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
7129 vcpu->run->fail_entry.hardware_entry_failure_reason
7130 = vmcs_read32(VM_INSTRUCTION_ERROR);
7131 return 0;
7132 }
7133
7134 /*
7135 * Note:
7136 * Do not try to fix EXIT_REASON_EPT_MISCONFIG if it caused by
7137 * delivery event since it indicates guest is accessing MMIO.
7138 * The vm-exit can be triggered again after return to guest that
7139 * will cause infinite loop.
7140 */
7141 if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
7142 (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
7143 exit_reason != EXIT_REASON_EPT_VIOLATION &&
7144 exit_reason != EXIT_REASON_TASK_SWITCH)) {
7145 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
7146 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_DELIVERY_EV;
7147 vcpu->run->internal.ndata = 2;
7148 vcpu->run->internal.data[0] = vectoring_info;
7149 vcpu->run->internal.data[1] = exit_reason;
7150 return 0;
7151 }
7152
7153 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked &&
7154 !(is_guest_mode(vcpu) && nested_cpu_has_virtual_nmis(
7155 get_vmcs12(vcpu))))) {
7156 if (vmx_interrupt_allowed(vcpu)) {
7157 vmx->soft_vnmi_blocked = 0;
7158 } else if (vmx->vnmi_blocked_time > 1000000000LL &&
7159 vcpu->arch.nmi_pending) {
7160 /*
7161 * This CPU don't support us in finding the end of an
7162 * NMI-blocked window if the guest runs with IRQs
7163 * disabled. So we pull the trigger after 1 s of
7164 * futile waiting, but inform the user about this.
7165 */
7166 printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
7167 "state on VCPU %d after 1 s timeout\n",
7168 __func__, vcpu->vcpu_id);
7169 vmx->soft_vnmi_blocked = 0;
7170 }
7171 }
7172
7173 if (exit_reason < kvm_vmx_max_exit_handlers
7174 && kvm_vmx_exit_handlers[exit_reason])
7175 return kvm_vmx_exit_handlers[exit_reason](vcpu);
7176 else {
7177 WARN_ONCE(1, "vmx: unexpected exit reason 0x%x\n", exit_reason);
7178 kvm_queue_exception(vcpu, UD_VECTOR);
7179 return 1;
7180 }
7181}
7182
7183static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
7184{
7185 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7186
7187 if (is_guest_mode(vcpu) &&
7188 nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
7189 return;
7190
7191 if (irr == -1 || tpr < irr) {
7192 vmcs_write32(TPR_THRESHOLD, 0);
7193 return;
7194 }
7195
7196 vmcs_write32(TPR_THRESHOLD, irr);
7197}
7198
7199static void vmx_set_virtual_x2apic_mode(struct kvm_vcpu *vcpu, bool set)
7200{
7201 u32 sec_exec_control;
7202
7203 /*
7204 * There is not point to enable virtualize x2apic without enable
7205 * apicv
7206 */
7207 if (!cpu_has_vmx_virtualize_x2apic_mode() ||
7208 !vmx_vm_has_apicv(vcpu->kvm))
7209 return;
7210
7211 if (!vm_need_tpr_shadow(vcpu->kvm))
7212 return;
7213
7214 sec_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
7215
7216 if (set) {
7217 sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
7218 sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
7219 } else {
7220 sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
7221 sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
7222 }
7223 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, sec_exec_control);
7224
7225 vmx_set_msr_bitmap(vcpu);
7226}
7227
7228static void vmx_set_apic_access_page_addr(struct kvm_vcpu *vcpu, hpa_t hpa)
7229{
7230 struct vcpu_vmx *vmx = to_vmx(vcpu);
7231
7232 /*
7233 * Currently we do not handle the nested case where L2 has an
7234 * APIC access page of its own; that page is still pinned.
7235 * Hence, we skip the case where the VCPU is in guest mode _and_
7236 * L1 prepared an APIC access page for L2.
7237 *
7238 * For the case where L1 and L2 share the same APIC access page
7239 * (flexpriority=Y but SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES clear
7240 * in the vmcs12), this function will only update either the vmcs01
7241 * or the vmcs02. If the former, the vmcs02 will be updated by
7242 * prepare_vmcs02. If the latter, the vmcs01 will be updated in
7243 * the next L2->L1 exit.
7244 */
7245 if (!is_guest_mode(vcpu) ||
7246 !nested_cpu_has2(vmx->nested.current_vmcs12,
7247 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
7248 vmcs_write64(APIC_ACCESS_ADDR, hpa);
7249}
7250
7251static void vmx_hwapic_isr_update(struct kvm *kvm, int isr)
7252{
7253 u16 status;
7254 u8 old;
7255
7256 if (!vmx_vm_has_apicv(kvm))
7257 return;
7258
7259 if (isr == -1)
7260 isr = 0;
7261
7262 status = vmcs_read16(GUEST_INTR_STATUS);
7263 old = status >> 8;
7264 if (isr != old) {
7265 status &= 0xff;
7266 status |= isr << 8;
7267 vmcs_write16(GUEST_INTR_STATUS, status);
7268 }
7269}
7270
7271static void vmx_set_rvi(int vector)
7272{
7273 u16 status;
7274 u8 old;
7275
7276 status = vmcs_read16(GUEST_INTR_STATUS);
7277 old = (u8)status & 0xff;
7278 if ((u8)vector != old) {
7279 status &= ~0xff;
7280 status |= (u8)vector;
7281 vmcs_write16(GUEST_INTR_STATUS, status);
7282 }
7283}
7284
7285static void vmx_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
7286{
7287 if (max_irr == -1)
7288 return;
7289
7290 /*
7291 * If a vmexit is needed, vmx_check_nested_events handles it.
7292 */
7293 if (is_guest_mode(vcpu) && nested_exit_on_intr(vcpu))
7294 return;
7295
7296 if (!is_guest_mode(vcpu)) {
7297 vmx_set_rvi(max_irr);
7298 return;
7299 }
7300
7301 /*
7302 * Fall back to pre-APICv interrupt injection since L2
7303 * is run without virtual interrupt delivery.
7304 */
7305 if (!kvm_event_needs_reinjection(vcpu) &&
7306 vmx_interrupt_allowed(vcpu)) {
7307 kvm_queue_interrupt(vcpu, max_irr, false);
7308 vmx_inject_irq(vcpu);
7309 }
7310}
7311
7312static void vmx_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
7313{
7314 if (!vmx_vm_has_apicv(vcpu->kvm))
7315 return;
7316
7317 vmcs_write64(EOI_EXIT_BITMAP0, eoi_exit_bitmap[0]);
7318 vmcs_write64(EOI_EXIT_BITMAP1, eoi_exit_bitmap[1]);
7319 vmcs_write64(EOI_EXIT_BITMAP2, eoi_exit_bitmap[2]);
7320 vmcs_write64(EOI_EXIT_BITMAP3, eoi_exit_bitmap[3]);
7321}
7322
7323static void vmx_complete_atomic_exit(struct vcpu_vmx *vmx)
7324{
7325 u32 exit_intr_info;
7326
7327 if (!(vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY
7328 || vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI))
7329 return;
7330
7331 vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
7332 exit_intr_info = vmx->exit_intr_info;
7333
7334 /* Handle machine checks before interrupts are enabled */
7335 if (is_machine_check(exit_intr_info))
7336 kvm_machine_check();
7337
7338 /* We need to handle NMIs before interrupts are enabled */
7339 if ((exit_intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR &&
7340 (exit_intr_info & INTR_INFO_VALID_MASK)) {
7341 kvm_before_handle_nmi(&vmx->vcpu);
7342 asm("int $2");
7343 kvm_after_handle_nmi(&vmx->vcpu);
7344 }
7345}
7346
7347static void vmx_handle_external_intr(struct kvm_vcpu *vcpu)
7348{
7349 u32 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
7350
7351 /*
7352 * If external interrupt exists, IF bit is set in rflags/eflags on the
7353 * interrupt stack frame, and interrupt will be enabled on a return
7354 * from interrupt handler.
7355 */
7356 if ((exit_intr_info & (INTR_INFO_VALID_MASK | INTR_INFO_INTR_TYPE_MASK))
7357 == (INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR)) {
7358 unsigned int vector;
7359 unsigned long entry;
7360 gate_desc *desc;
7361 struct vcpu_vmx *vmx = to_vmx(vcpu);
7362#ifdef CONFIG_X86_64
7363 unsigned long tmp;
7364#endif
7365
7366 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
7367 desc = (gate_desc *)vmx->host_idt_base + vector;
7368 entry = gate_offset(*desc);
7369 asm volatile(
7370#ifdef CONFIG_X86_64
7371 "mov %%" _ASM_SP ", %[sp]\n\t"
7372 "and $0xfffffffffffffff0, %%" _ASM_SP "\n\t"
7373 "push $%c[ss]\n\t"
7374 "push %[sp]\n\t"
7375#endif
7376 "pushf\n\t"
7377 "orl $0x200, (%%" _ASM_SP ")\n\t"
7378 __ASM_SIZE(push) " $%c[cs]\n\t"
7379 "call *%[entry]\n\t"
7380 :
7381#ifdef CONFIG_X86_64
7382 [sp]"=&r"(tmp)
7383#endif
7384 :
7385 [entry]"r"(entry),
7386 [ss]"i"(__KERNEL_DS),
7387 [cs]"i"(__KERNEL_CS)
7388 );
7389 } else
7390 local_irq_enable();
7391}
7392
7393static bool vmx_mpx_supported(void)
7394{
7395 return (vmcs_config.vmexit_ctrl & VM_EXIT_CLEAR_BNDCFGS) &&
7396 (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_BNDCFGS);
7397}
7398
7399static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
7400{
7401 u32 exit_intr_info;
7402 bool unblock_nmi;
7403 u8 vector;
7404 bool idtv_info_valid;
7405
7406 idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
7407
7408 if (cpu_has_virtual_nmis()) {
7409 if (vmx->nmi_known_unmasked)
7410 return;
7411 /*
7412 * Can't use vmx->exit_intr_info since we're not sure what
7413 * the exit reason is.
7414 */
7415 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
7416 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
7417 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
7418 /*
7419 * SDM 3: 27.7.1.2 (September 2008)
7420 * Re-set bit "block by NMI" before VM entry if vmexit caused by
7421 * a guest IRET fault.
7422 * SDM 3: 23.2.2 (September 2008)
7423 * Bit 12 is undefined in any of the following cases:
7424 * If the VM exit sets the valid bit in the IDT-vectoring
7425 * information field.
7426 * If the VM exit is due to a double fault.
7427 */
7428 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
7429 vector != DF_VECTOR && !idtv_info_valid)
7430 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
7431 GUEST_INTR_STATE_NMI);
7432 else
7433 vmx->nmi_known_unmasked =
7434 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
7435 & GUEST_INTR_STATE_NMI);
7436 } else if (unlikely(vmx->soft_vnmi_blocked))
7437 vmx->vnmi_blocked_time +=
7438 ktime_to_ns(ktime_sub(ktime_get(), vmx->entry_time));
7439}
7440
7441static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu,
7442 u32 idt_vectoring_info,
7443 int instr_len_field,
7444 int error_code_field)
7445{
7446 u8 vector;
7447 int type;
7448 bool idtv_info_valid;
7449
7450 idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
7451
7452 vcpu->arch.nmi_injected = false;
7453 kvm_clear_exception_queue(vcpu);
7454 kvm_clear_interrupt_queue(vcpu);
7455
7456 if (!idtv_info_valid)
7457 return;
7458
7459 kvm_make_request(KVM_REQ_EVENT, vcpu);
7460
7461 vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
7462 type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
7463
7464 switch (type) {
7465 case INTR_TYPE_NMI_INTR:
7466 vcpu->arch.nmi_injected = true;
7467 /*
7468 * SDM 3: 27.7.1.2 (September 2008)
7469 * Clear bit "block by NMI" before VM entry if a NMI
7470 * delivery faulted.
7471 */
7472 vmx_set_nmi_mask(vcpu, false);
7473 break;
7474 case INTR_TYPE_SOFT_EXCEPTION:
7475 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
7476 /* fall through */
7477 case INTR_TYPE_HARD_EXCEPTION:
7478 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
7479 u32 err = vmcs_read32(error_code_field);
7480 kvm_requeue_exception_e(vcpu, vector, err);
7481 } else
7482 kvm_requeue_exception(vcpu, vector);
7483 break;
7484 case INTR_TYPE_SOFT_INTR:
7485 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
7486 /* fall through */
7487 case INTR_TYPE_EXT_INTR:
7488 kvm_queue_interrupt(vcpu, vector, type == INTR_TYPE_SOFT_INTR);
7489 break;
7490 default:
7491 break;
7492 }
7493}
7494
7495static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
7496{
7497 __vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info,
7498 VM_EXIT_INSTRUCTION_LEN,
7499 IDT_VECTORING_ERROR_CODE);
7500}
7501
7502static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
7503{
7504 __vmx_complete_interrupts(vcpu,
7505 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
7506 VM_ENTRY_INSTRUCTION_LEN,
7507 VM_ENTRY_EXCEPTION_ERROR_CODE);
7508
7509 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
7510}
7511
7512static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
7513{
7514 int i, nr_msrs;
7515 struct perf_guest_switch_msr *msrs;
7516
7517 msrs = perf_guest_get_msrs(&nr_msrs);
7518
7519 if (!msrs)
7520 return;
7521
7522 for (i = 0; i < nr_msrs; i++)
7523 if (msrs[i].host == msrs[i].guest)
7524 clear_atomic_switch_msr(vmx, msrs[i].msr);
7525 else
7526 add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
7527 msrs[i].host);
7528}
7529
7530static void __noclone vmx_vcpu_run(struct kvm_vcpu *vcpu)
7531{
7532 struct vcpu_vmx *vmx = to_vmx(vcpu);
7533 unsigned long debugctlmsr, cr4;
7534
7535 /* Record the guest's net vcpu time for enforced NMI injections. */
7536 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked))
7537 vmx->entry_time = ktime_get();
7538
7539 /* Don't enter VMX if guest state is invalid, let the exit handler
7540 start emulation until we arrive back to a valid state */
7541 if (vmx->emulation_required)
7542 return;
7543
7544 if (vmx->ple_window_dirty) {
7545 vmx->ple_window_dirty = false;
7546 vmcs_write32(PLE_WINDOW, vmx->ple_window);
7547 }
7548
7549 if (vmx->nested.sync_shadow_vmcs) {
7550 copy_vmcs12_to_shadow(vmx);
7551 vmx->nested.sync_shadow_vmcs = false;
7552 }
7553
7554 if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
7555 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
7556 if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
7557 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
7558
7559 cr4 = read_cr4();
7560 if (unlikely(cr4 != vmx->host_state.vmcs_host_cr4)) {
7561 vmcs_writel(HOST_CR4, cr4);
7562 vmx->host_state.vmcs_host_cr4 = cr4;
7563 }
7564
7565 /* When single-stepping over STI and MOV SS, we must clear the
7566 * corresponding interruptibility bits in the guest state. Otherwise
7567 * vmentry fails as it then expects bit 14 (BS) in pending debug
7568 * exceptions being set, but that's not correct for the guest debugging
7569 * case. */
7570 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7571 vmx_set_interrupt_shadow(vcpu, 0);
7572
7573 atomic_switch_perf_msrs(vmx);
7574 debugctlmsr = get_debugctlmsr();
7575
7576 vmx->__launched = vmx->loaded_vmcs->launched;
7577 asm(
7578 /* Store host registers */
7579 "push %%" _ASM_DX "; push %%" _ASM_BP ";"
7580 "push %%" _ASM_CX " \n\t" /* placeholder for guest rcx */
7581 "push %%" _ASM_CX " \n\t"
7582 "cmp %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
7583 "je 1f \n\t"
7584 "mov %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
7585 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
7586 "1: \n\t"
7587 /* Reload cr2 if changed */
7588 "mov %c[cr2](%0), %%" _ASM_AX " \n\t"
7589 "mov %%cr2, %%" _ASM_DX " \n\t"
7590 "cmp %%" _ASM_AX ", %%" _ASM_DX " \n\t"
7591 "je 2f \n\t"
7592 "mov %%" _ASM_AX", %%cr2 \n\t"
7593 "2: \n\t"
7594 /* Check if vmlaunch of vmresume is needed */
7595 "cmpl $0, %c[launched](%0) \n\t"
7596 /* Load guest registers. Don't clobber flags. */
7597 "mov %c[rax](%0), %%" _ASM_AX " \n\t"
7598 "mov %c[rbx](%0), %%" _ASM_BX " \n\t"
7599 "mov %c[rdx](%0), %%" _ASM_DX " \n\t"
7600 "mov %c[rsi](%0), %%" _ASM_SI " \n\t"
7601 "mov %c[rdi](%0), %%" _ASM_DI " \n\t"
7602 "mov %c[rbp](%0), %%" _ASM_BP " \n\t"
7603#ifdef CONFIG_X86_64
7604 "mov %c[r8](%0), %%r8 \n\t"
7605 "mov %c[r9](%0), %%r9 \n\t"
7606 "mov %c[r10](%0), %%r10 \n\t"
7607 "mov %c[r11](%0), %%r11 \n\t"
7608 "mov %c[r12](%0), %%r12 \n\t"
7609 "mov %c[r13](%0), %%r13 \n\t"
7610 "mov %c[r14](%0), %%r14 \n\t"
7611 "mov %c[r15](%0), %%r15 \n\t"
7612#endif
7613 "mov %c[rcx](%0), %%" _ASM_CX " \n\t" /* kills %0 (ecx) */
7614
7615 /* Enter guest mode */
7616 "jne 1f \n\t"
7617 __ex(ASM_VMX_VMLAUNCH) "\n\t"
7618 "jmp 2f \n\t"
7619 "1: " __ex(ASM_VMX_VMRESUME) "\n\t"
7620 "2: "
7621 /* Save guest registers, load host registers, keep flags */
7622 "mov %0, %c[wordsize](%%" _ASM_SP ") \n\t"
7623 "pop %0 \n\t"
7624 "mov %%" _ASM_AX ", %c[rax](%0) \n\t"
7625 "mov %%" _ASM_BX ", %c[rbx](%0) \n\t"
7626 __ASM_SIZE(pop) " %c[rcx](%0) \n\t"
7627 "mov %%" _ASM_DX ", %c[rdx](%0) \n\t"
7628 "mov %%" _ASM_SI ", %c[rsi](%0) \n\t"
7629 "mov %%" _ASM_DI ", %c[rdi](%0) \n\t"
7630 "mov %%" _ASM_BP ", %c[rbp](%0) \n\t"
7631#ifdef CONFIG_X86_64
7632 "mov %%r8, %c[r8](%0) \n\t"
7633 "mov %%r9, %c[r9](%0) \n\t"
7634 "mov %%r10, %c[r10](%0) \n\t"
7635 "mov %%r11, %c[r11](%0) \n\t"
7636 "mov %%r12, %c[r12](%0) \n\t"
7637 "mov %%r13, %c[r13](%0) \n\t"
7638 "mov %%r14, %c[r14](%0) \n\t"
7639 "mov %%r15, %c[r15](%0) \n\t"
7640#endif
7641 "mov %%cr2, %%" _ASM_AX " \n\t"
7642 "mov %%" _ASM_AX ", %c[cr2](%0) \n\t"
7643
7644 "pop %%" _ASM_BP "; pop %%" _ASM_DX " \n\t"
7645 "setbe %c[fail](%0) \n\t"
7646 ".pushsection .rodata \n\t"
7647 ".global vmx_return \n\t"
7648 "vmx_return: " _ASM_PTR " 2b \n\t"
7649 ".popsection"
7650 : : "c"(vmx), "d"((unsigned long)HOST_RSP),
7651 [launched]"i"(offsetof(struct vcpu_vmx, __launched)),
7652 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
7653 [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
7654 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
7655 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
7656 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
7657 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
7658 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
7659 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
7660 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
7661#ifdef CONFIG_X86_64
7662 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
7663 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
7664 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
7665 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
7666 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
7667 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
7668 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
7669 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
7670#endif
7671 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2)),
7672 [wordsize]"i"(sizeof(ulong))
7673 : "cc", "memory"
7674#ifdef CONFIG_X86_64
7675 , "rax", "rbx", "rdi", "rsi"
7676 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
7677#else
7678 , "eax", "ebx", "edi", "esi"
7679#endif
7680 );
7681
7682 /* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
7683 if (debugctlmsr)
7684 update_debugctlmsr(debugctlmsr);
7685
7686#ifndef CONFIG_X86_64
7687 /*
7688 * The sysexit path does not restore ds/es, so we must set them to
7689 * a reasonable value ourselves.
7690 *
7691 * We can't defer this to vmx_load_host_state() since that function
7692 * may be executed in interrupt context, which saves and restore segments
7693 * around it, nullifying its effect.
7694 */
7695 loadsegment(ds, __USER_DS);
7696 loadsegment(es, __USER_DS);
7697#endif
7698
7699 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
7700 | (1 << VCPU_EXREG_RFLAGS)
7701 | (1 << VCPU_EXREG_PDPTR)
7702 | (1 << VCPU_EXREG_SEGMENTS)
7703 | (1 << VCPU_EXREG_CR3));
7704 vcpu->arch.regs_dirty = 0;
7705
7706 vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
7707
7708 vmx->loaded_vmcs->launched = 1;
7709
7710 vmx->exit_reason = vmcs_read32(VM_EXIT_REASON);
7711 trace_kvm_exit(vmx->exit_reason, vcpu, KVM_ISA_VMX);
7712
7713 /*
7714 * the KVM_REQ_EVENT optimization bit is only on for one entry, and if
7715 * we did not inject a still-pending event to L1 now because of
7716 * nested_run_pending, we need to re-enable this bit.
7717 */
7718 if (vmx->nested.nested_run_pending)
7719 kvm_make_request(KVM_REQ_EVENT, vcpu);
7720
7721 vmx->nested.nested_run_pending = 0;
7722
7723 vmx_complete_atomic_exit(vmx);
7724 vmx_recover_nmi_blocking(vmx);
7725 vmx_complete_interrupts(vmx);
7726}
7727
7728static void vmx_load_vmcs01(struct kvm_vcpu *vcpu)
7729{
7730 struct vcpu_vmx *vmx = to_vmx(vcpu);
7731 int cpu;
7732
7733 if (vmx->loaded_vmcs == &vmx->vmcs01)
7734 return;
7735
7736 cpu = get_cpu();
7737 vmx->loaded_vmcs = &vmx->vmcs01;
7738 vmx_vcpu_put(vcpu);
7739 vmx_vcpu_load(vcpu, cpu);
7740 vcpu->cpu = cpu;
7741 put_cpu();
7742}
7743
7744static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
7745{
7746 struct vcpu_vmx *vmx = to_vmx(vcpu);
7747
7748 free_vpid(vmx);
7749 leave_guest_mode(vcpu);
7750 vmx_load_vmcs01(vcpu);
7751 free_nested(vmx);
7752 free_loaded_vmcs(vmx->loaded_vmcs);
7753 kfree(vmx->guest_msrs);
7754 kvm_vcpu_uninit(vcpu);
7755 kmem_cache_free(kvm_vcpu_cache, vmx);
7756}
7757
7758static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
7759{
7760 int err;
7761 struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
7762 int cpu;
7763
7764 if (!vmx)
7765 return ERR_PTR(-ENOMEM);
7766
7767 allocate_vpid(vmx);
7768
7769 err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
7770 if (err)
7771 goto free_vcpu;
7772
7773 vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
7774 BUILD_BUG_ON(ARRAY_SIZE(vmx_msr_index) * sizeof(vmx->guest_msrs[0])
7775 > PAGE_SIZE);
7776
7777 err = -ENOMEM;
7778 if (!vmx->guest_msrs) {
7779 goto uninit_vcpu;
7780 }
7781
7782 vmx->loaded_vmcs = &vmx->vmcs01;
7783 vmx->loaded_vmcs->vmcs = alloc_vmcs();
7784 if (!vmx->loaded_vmcs->vmcs)
7785 goto free_msrs;
7786 if (!vmm_exclusive)
7787 kvm_cpu_vmxon(__pa(per_cpu(vmxarea, raw_smp_processor_id())));
7788 loaded_vmcs_init(vmx->loaded_vmcs);
7789 if (!vmm_exclusive)
7790 kvm_cpu_vmxoff();
7791
7792 cpu = get_cpu();
7793 vmx_vcpu_load(&vmx->vcpu, cpu);
7794 vmx->vcpu.cpu = cpu;
7795 err = vmx_vcpu_setup(vmx);
7796 vmx_vcpu_put(&vmx->vcpu);
7797 put_cpu();
7798 if (err)
7799 goto free_vmcs;
7800 if (vm_need_virtualize_apic_accesses(kvm)) {
7801 err = alloc_apic_access_page(kvm);
7802 if (err)
7803 goto free_vmcs;
7804 }
7805
7806 if (enable_ept) {
7807 if (!kvm->arch.ept_identity_map_addr)
7808 kvm->arch.ept_identity_map_addr =
7809 VMX_EPT_IDENTITY_PAGETABLE_ADDR;
7810 err = init_rmode_identity_map(kvm);
7811 if (err)
7812 goto free_vmcs;
7813 }
7814
7815 vmx->nested.current_vmptr = -1ull;
7816 vmx->nested.current_vmcs12 = NULL;
7817
7818 return &vmx->vcpu;
7819
7820free_vmcs:
7821 free_loaded_vmcs(vmx->loaded_vmcs);
7822free_msrs:
7823 kfree(vmx->guest_msrs);
7824uninit_vcpu:
7825 kvm_vcpu_uninit(&vmx->vcpu);
7826free_vcpu:
7827 free_vpid(vmx);
7828 kmem_cache_free(kvm_vcpu_cache, vmx);
7829 return ERR_PTR(err);
7830}
7831
7832static void __init vmx_check_processor_compat(void *rtn)
7833{
7834 struct vmcs_config vmcs_conf;
7835
7836 *(int *)rtn = 0;
7837 if (setup_vmcs_config(&vmcs_conf) < 0)
7838 *(int *)rtn = -EIO;
7839 if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
7840 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
7841 smp_processor_id());
7842 *(int *)rtn = -EIO;
7843 }
7844}
7845
7846static int get_ept_level(void)
7847{
7848 return VMX_EPT_DEFAULT_GAW + 1;
7849}
7850
7851static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
7852{
7853 u64 ret;
7854
7855 /* For VT-d and EPT combination
7856 * 1. MMIO: always map as UC
7857 * 2. EPT with VT-d:
7858 * a. VT-d without snooping control feature: can't guarantee the
7859 * result, try to trust guest.
7860 * b. VT-d with snooping control feature: snooping control feature of
7861 * VT-d engine can guarantee the cache correctness. Just set it
7862 * to WB to keep consistent with host. So the same as item 3.
7863 * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
7864 * consistent with host MTRR
7865 */
7866 if (is_mmio)
7867 ret = MTRR_TYPE_UNCACHABLE << VMX_EPT_MT_EPTE_SHIFT;
7868 else if (kvm_arch_has_noncoherent_dma(vcpu->kvm))
7869 ret = kvm_get_guest_memory_type(vcpu, gfn) <<
7870 VMX_EPT_MT_EPTE_SHIFT;
7871 else
7872 ret = (MTRR_TYPE_WRBACK << VMX_EPT_MT_EPTE_SHIFT)
7873 | VMX_EPT_IPAT_BIT;
7874
7875 return ret;
7876}
7877
7878static int vmx_get_lpage_level(void)
7879{
7880 if (enable_ept && !cpu_has_vmx_ept_1g_page())
7881 return PT_DIRECTORY_LEVEL;
7882 else
7883 /* For shadow and EPT supported 1GB page */
7884 return PT_PDPE_LEVEL;
7885}
7886
7887static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
7888{
7889 struct kvm_cpuid_entry2 *best;
7890 struct vcpu_vmx *vmx = to_vmx(vcpu);
7891 u32 exec_control;
7892
7893 vmx->rdtscp_enabled = false;
7894 if (vmx_rdtscp_supported()) {
7895 exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
7896 if (exec_control & SECONDARY_EXEC_RDTSCP) {
7897 best = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
7898 if (best && (best->edx & bit(X86_FEATURE_RDTSCP)))
7899 vmx->rdtscp_enabled = true;
7900 else {
7901 exec_control &= ~SECONDARY_EXEC_RDTSCP;
7902 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
7903 exec_control);
7904 }
7905 }
7906 }
7907
7908 /* Exposing INVPCID only when PCID is exposed */
7909 best = kvm_find_cpuid_entry(vcpu, 0x7, 0);
7910 if (vmx_invpcid_supported() &&
7911 best && (best->ebx & bit(X86_FEATURE_INVPCID)) &&
7912 guest_cpuid_has_pcid(vcpu)) {
7913 exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
7914 exec_control |= SECONDARY_EXEC_ENABLE_INVPCID;
7915 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
7916 exec_control);
7917 } else {
7918 if (cpu_has_secondary_exec_ctrls()) {
7919 exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
7920 exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
7921 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
7922 exec_control);
7923 }
7924 if (best)
7925 best->ebx &= ~bit(X86_FEATURE_INVPCID);
7926 }
7927}
7928
7929static void vmx_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
7930{
7931 if (func == 1 && nested)
7932 entry->ecx |= bit(X86_FEATURE_VMX);
7933}
7934
7935static void nested_ept_inject_page_fault(struct kvm_vcpu *vcpu,
7936 struct x86_exception *fault)
7937{
7938 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7939 u32 exit_reason;
7940
7941 if (fault->error_code & PFERR_RSVD_MASK)
7942 exit_reason = EXIT_REASON_EPT_MISCONFIG;
7943 else
7944 exit_reason = EXIT_REASON_EPT_VIOLATION;
7945 nested_vmx_vmexit(vcpu, exit_reason, 0, vcpu->arch.exit_qualification);
7946 vmcs12->guest_physical_address = fault->address;
7947}
7948
7949/* Callbacks for nested_ept_init_mmu_context: */
7950
7951static unsigned long nested_ept_get_cr3(struct kvm_vcpu *vcpu)
7952{
7953 /* return the page table to be shadowed - in our case, EPT12 */
7954 return get_vmcs12(vcpu)->ept_pointer;
7955}
7956
7957static void nested_ept_init_mmu_context(struct kvm_vcpu *vcpu)
7958{
7959 kvm_init_shadow_ept_mmu(vcpu, &vcpu->arch.mmu,
7960 nested_vmx_ept_caps & VMX_EPT_EXECUTE_ONLY_BIT);
7961
7962 vcpu->arch.mmu.set_cr3 = vmx_set_cr3;
7963 vcpu->arch.mmu.get_cr3 = nested_ept_get_cr3;
7964 vcpu->arch.mmu.inject_page_fault = nested_ept_inject_page_fault;
7965
7966 vcpu->arch.walk_mmu = &vcpu->arch.nested_mmu;
7967}
7968
7969static void nested_ept_uninit_mmu_context(struct kvm_vcpu *vcpu)
7970{
7971 vcpu->arch.walk_mmu = &vcpu->arch.mmu;
7972}
7973
7974static void vmx_inject_page_fault_nested(struct kvm_vcpu *vcpu,
7975 struct x86_exception *fault)
7976{
7977 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7978
7979 WARN_ON(!is_guest_mode(vcpu));
7980
7981 /* TODO: also check PFEC_MATCH/MASK, not just EB.PF. */
7982 if (vmcs12->exception_bitmap & (1u << PF_VECTOR))
7983 nested_vmx_vmexit(vcpu, to_vmx(vcpu)->exit_reason,
7984 vmcs_read32(VM_EXIT_INTR_INFO),
7985 vmcs_readl(EXIT_QUALIFICATION));
7986 else
7987 kvm_inject_page_fault(vcpu, fault);
7988}
7989
7990static bool nested_get_vmcs12_pages(struct kvm_vcpu *vcpu,
7991 struct vmcs12 *vmcs12)
7992{
7993 struct vcpu_vmx *vmx = to_vmx(vcpu);
7994
7995 if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
7996 /* TODO: Also verify bits beyond physical address width are 0 */
7997 if (!PAGE_ALIGNED(vmcs12->apic_access_addr))
7998 return false;
7999
8000 /*
8001 * Translate L1 physical address to host physical
8002 * address for vmcs02. Keep the page pinned, so this
8003 * physical address remains valid. We keep a reference
8004 * to it so we can release it later.
8005 */
8006 if (vmx->nested.apic_access_page) /* shouldn't happen */
8007 nested_release_page(vmx->nested.apic_access_page);
8008 vmx->nested.apic_access_page =
8009 nested_get_page(vcpu, vmcs12->apic_access_addr);
8010 }
8011
8012 if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
8013 /* TODO: Also verify bits beyond physical address width are 0 */
8014 if (!PAGE_ALIGNED(vmcs12->virtual_apic_page_addr))
8015 return false;
8016
8017 if (vmx->nested.virtual_apic_page) /* shouldn't happen */
8018 nested_release_page(vmx->nested.virtual_apic_page);
8019 vmx->nested.virtual_apic_page =
8020 nested_get_page(vcpu, vmcs12->virtual_apic_page_addr);
8021
8022 /*
8023 * Failing the vm entry is _not_ what the processor does
8024 * but it's basically the only possibility we have.
8025 * We could still enter the guest if CR8 load exits are
8026 * enabled, CR8 store exits are enabled, and virtualize APIC
8027 * access is disabled; in this case the processor would never
8028 * use the TPR shadow and we could simply clear the bit from
8029 * the execution control. But such a configuration is useless,
8030 * so let's keep the code simple.
8031 */
8032 if (!vmx->nested.virtual_apic_page)
8033 return false;
8034 }
8035
8036 return true;
8037}
8038
8039static void vmx_start_preemption_timer(struct kvm_vcpu *vcpu)
8040{
8041 u64 preemption_timeout = get_vmcs12(vcpu)->vmx_preemption_timer_value;
8042 struct vcpu_vmx *vmx = to_vmx(vcpu);
8043
8044 if (vcpu->arch.virtual_tsc_khz == 0)
8045 return;
8046
8047 /* Make sure short timeouts reliably trigger an immediate vmexit.
8048 * hrtimer_start does not guarantee this. */
8049 if (preemption_timeout <= 1) {
8050 vmx_preemption_timer_fn(&vmx->nested.preemption_timer);
8051 return;
8052 }
8053
8054 preemption_timeout <<= VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
8055 preemption_timeout *= 1000000;
8056 do_div(preemption_timeout, vcpu->arch.virtual_tsc_khz);
8057 hrtimer_start(&vmx->nested.preemption_timer,
8058 ns_to_ktime(preemption_timeout), HRTIMER_MODE_REL);
8059}
8060
8061/*
8062 * prepare_vmcs02 is called when the L1 guest hypervisor runs its nested
8063 * L2 guest. L1 has a vmcs for L2 (vmcs12), and this function "merges" it
8064 * with L0's requirements for its guest (a.k.a. vmcs01), so we can run the L2
8065 * guest in a way that will both be appropriate to L1's requests, and our
8066 * needs. In addition to modifying the active vmcs (which is vmcs02), this
8067 * function also has additional necessary side-effects, like setting various
8068 * vcpu->arch fields.
8069 */
8070static void prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
8071{
8072 struct vcpu_vmx *vmx = to_vmx(vcpu);
8073 u32 exec_control;
8074
8075 vmcs_write16(GUEST_ES_SELECTOR, vmcs12->guest_es_selector);
8076 vmcs_write16(GUEST_CS_SELECTOR, vmcs12->guest_cs_selector);
8077 vmcs_write16(GUEST_SS_SELECTOR, vmcs12->guest_ss_selector);
8078 vmcs_write16(GUEST_DS_SELECTOR, vmcs12->guest_ds_selector);
8079 vmcs_write16(GUEST_FS_SELECTOR, vmcs12->guest_fs_selector);
8080 vmcs_write16(GUEST_GS_SELECTOR, vmcs12->guest_gs_selector);
8081 vmcs_write16(GUEST_LDTR_SELECTOR, vmcs12->guest_ldtr_selector);
8082 vmcs_write16(GUEST_TR_SELECTOR, vmcs12->guest_tr_selector);
8083 vmcs_write32(GUEST_ES_LIMIT, vmcs12->guest_es_limit);
8084 vmcs_write32(GUEST_CS_LIMIT, vmcs12->guest_cs_limit);
8085 vmcs_write32(GUEST_SS_LIMIT, vmcs12->guest_ss_limit);
8086 vmcs_write32(GUEST_DS_LIMIT, vmcs12->guest_ds_limit);
8087 vmcs_write32(GUEST_FS_LIMIT, vmcs12->guest_fs_limit);
8088 vmcs_write32(GUEST_GS_LIMIT, vmcs12->guest_gs_limit);
8089 vmcs_write32(GUEST_LDTR_LIMIT, vmcs12->guest_ldtr_limit);
8090 vmcs_write32(GUEST_TR_LIMIT, vmcs12->guest_tr_limit);
8091 vmcs_write32(GUEST_GDTR_LIMIT, vmcs12->guest_gdtr_limit);
8092 vmcs_write32(GUEST_IDTR_LIMIT, vmcs12->guest_idtr_limit);
8093 vmcs_write32(GUEST_ES_AR_BYTES, vmcs12->guest_es_ar_bytes);
8094 vmcs_write32(GUEST_CS_AR_BYTES, vmcs12->guest_cs_ar_bytes);
8095 vmcs_write32(GUEST_SS_AR_BYTES, vmcs12->guest_ss_ar_bytes);
8096 vmcs_write32(GUEST_DS_AR_BYTES, vmcs12->guest_ds_ar_bytes);
8097 vmcs_write32(GUEST_FS_AR_BYTES, vmcs12->guest_fs_ar_bytes);
8098 vmcs_write32(GUEST_GS_AR_BYTES, vmcs12->guest_gs_ar_bytes);
8099 vmcs_write32(GUEST_LDTR_AR_BYTES, vmcs12->guest_ldtr_ar_bytes);
8100 vmcs_write32(GUEST_TR_AR_BYTES, vmcs12->guest_tr_ar_bytes);
8101 vmcs_writel(GUEST_ES_BASE, vmcs12->guest_es_base);
8102 vmcs_writel(GUEST_CS_BASE, vmcs12->guest_cs_base);
8103 vmcs_writel(GUEST_SS_BASE, vmcs12->guest_ss_base);
8104 vmcs_writel(GUEST_DS_BASE, vmcs12->guest_ds_base);
8105 vmcs_writel(GUEST_FS_BASE, vmcs12->guest_fs_base);
8106 vmcs_writel(GUEST_GS_BASE, vmcs12->guest_gs_base);
8107 vmcs_writel(GUEST_LDTR_BASE, vmcs12->guest_ldtr_base);
8108 vmcs_writel(GUEST_TR_BASE, vmcs12->guest_tr_base);
8109 vmcs_writel(GUEST_GDTR_BASE, vmcs12->guest_gdtr_base);
8110 vmcs_writel(GUEST_IDTR_BASE, vmcs12->guest_idtr_base);
8111
8112 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS) {
8113 kvm_set_dr(vcpu, 7, vmcs12->guest_dr7);
8114 vmcs_write64(GUEST_IA32_DEBUGCTL, vmcs12->guest_ia32_debugctl);
8115 } else {
8116 kvm_set_dr(vcpu, 7, vcpu->arch.dr7);
8117 vmcs_write64(GUEST_IA32_DEBUGCTL, vmx->nested.vmcs01_debugctl);
8118 }
8119 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
8120 vmcs12->vm_entry_intr_info_field);
8121 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
8122 vmcs12->vm_entry_exception_error_code);
8123 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
8124 vmcs12->vm_entry_instruction_len);
8125 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
8126 vmcs12->guest_interruptibility_info);
8127 vmcs_write32(GUEST_SYSENTER_CS, vmcs12->guest_sysenter_cs);
8128 vmx_set_rflags(vcpu, vmcs12->guest_rflags);
8129 vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
8130 vmcs12->guest_pending_dbg_exceptions);
8131 vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->guest_sysenter_esp);
8132 vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->guest_sysenter_eip);
8133
8134 vmcs_write64(VMCS_LINK_POINTER, -1ull);
8135
8136 exec_control = vmcs12->pin_based_vm_exec_control;
8137 exec_control |= vmcs_config.pin_based_exec_ctrl;
8138 exec_control &= ~(PIN_BASED_VMX_PREEMPTION_TIMER |
8139 PIN_BASED_POSTED_INTR);
8140 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, exec_control);
8141
8142 vmx->nested.preemption_timer_expired = false;
8143 if (nested_cpu_has_preemption_timer(vmcs12))
8144 vmx_start_preemption_timer(vcpu);
8145
8146 /*
8147 * Whether page-faults are trapped is determined by a combination of
8148 * 3 settings: PFEC_MASK, PFEC_MATCH and EXCEPTION_BITMAP.PF.
8149 * If enable_ept, L0 doesn't care about page faults and we should
8150 * set all of these to L1's desires. However, if !enable_ept, L0 does
8151 * care about (at least some) page faults, and because it is not easy
8152 * (if at all possible?) to merge L0 and L1's desires, we simply ask
8153 * to exit on each and every L2 page fault. This is done by setting
8154 * MASK=MATCH=0 and (see below) EB.PF=1.
8155 * Note that below we don't need special code to set EB.PF beyond the
8156 * "or"ing of the EB of vmcs01 and vmcs12, because when enable_ept,
8157 * vmcs01's EB.PF is 0 so the "or" will take vmcs12's value, and when
8158 * !enable_ept, EB.PF is 1, so the "or" will always be 1.
8159 *
8160 * A problem with this approach (when !enable_ept) is that L1 may be
8161 * injected with more page faults than it asked for. This could have
8162 * caused problems, but in practice existing hypervisors don't care.
8163 * To fix this, we will need to emulate the PFEC checking (on the L1
8164 * page tables), using walk_addr(), when injecting PFs to L1.
8165 */
8166 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK,
8167 enable_ept ? vmcs12->page_fault_error_code_mask : 0);
8168 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH,
8169 enable_ept ? vmcs12->page_fault_error_code_match : 0);
8170
8171 if (cpu_has_secondary_exec_ctrls()) {
8172 exec_control = vmx_secondary_exec_control(vmx);
8173 if (!vmx->rdtscp_enabled)
8174 exec_control &= ~SECONDARY_EXEC_RDTSCP;
8175 /* Take the following fields only from vmcs12 */
8176 exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
8177 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
8178 SECONDARY_EXEC_APIC_REGISTER_VIRT);
8179 if (nested_cpu_has(vmcs12,
8180 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS))
8181 exec_control |= vmcs12->secondary_vm_exec_control;
8182
8183 if (exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES) {
8184 /*
8185 * If translation failed, no matter: This feature asks
8186 * to exit when accessing the given address, and if it
8187 * can never be accessed, this feature won't do
8188 * anything anyway.
8189 */
8190 if (!vmx->nested.apic_access_page)
8191 exec_control &=
8192 ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
8193 else
8194 vmcs_write64(APIC_ACCESS_ADDR,
8195 page_to_phys(vmx->nested.apic_access_page));
8196 } else if (vm_need_virtualize_apic_accesses(vmx->vcpu.kvm)) {
8197 exec_control |=
8198 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
8199 kvm_vcpu_reload_apic_access_page(vcpu);
8200 }
8201
8202 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
8203 }
8204
8205
8206 /*
8207 * Set host-state according to L0's settings (vmcs12 is irrelevant here)
8208 * Some constant fields are set here by vmx_set_constant_host_state().
8209 * Other fields are different per CPU, and will be set later when
8210 * vmx_vcpu_load() is called, and when vmx_save_host_state() is called.
8211 */
8212 vmx_set_constant_host_state(vmx);
8213
8214 /*
8215 * HOST_RSP is normally set correctly in vmx_vcpu_run() just before
8216 * entry, but only if the current (host) sp changed from the value
8217 * we wrote last (vmx->host_rsp). This cache is no longer relevant
8218 * if we switch vmcs, and rather than hold a separate cache per vmcs,
8219 * here we just force the write to happen on entry.
8220 */
8221 vmx->host_rsp = 0;
8222
8223 exec_control = vmx_exec_control(vmx); /* L0's desires */
8224 exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
8225 exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
8226 exec_control &= ~CPU_BASED_TPR_SHADOW;
8227 exec_control |= vmcs12->cpu_based_vm_exec_control;
8228
8229 if (exec_control & CPU_BASED_TPR_SHADOW) {
8230 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
8231 page_to_phys(vmx->nested.virtual_apic_page));
8232 vmcs_write32(TPR_THRESHOLD, vmcs12->tpr_threshold);
8233 }
8234
8235 /*
8236 * Merging of IO and MSR bitmaps not currently supported.
8237 * Rather, exit every time.
8238 */
8239 exec_control &= ~CPU_BASED_USE_MSR_BITMAPS;
8240 exec_control &= ~CPU_BASED_USE_IO_BITMAPS;
8241 exec_control |= CPU_BASED_UNCOND_IO_EXITING;
8242
8243 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
8244
8245 /* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the
8246 * bitwise-or of what L1 wants to trap for L2, and what we want to
8247 * trap. Note that CR0.TS also needs updating - we do this later.
8248 */
8249 update_exception_bitmap(vcpu);
8250 vcpu->arch.cr0_guest_owned_bits &= ~vmcs12->cr0_guest_host_mask;
8251 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
8252
8253 /* L2->L1 exit controls are emulated - the hardware exit is to L0 so
8254 * we should use its exit controls. Note that VM_EXIT_LOAD_IA32_EFER
8255 * bits are further modified by vmx_set_efer() below.
8256 */
8257 vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
8258
8259 /* vmcs12's VM_ENTRY_LOAD_IA32_EFER and VM_ENTRY_IA32E_MODE are
8260 * emulated by vmx_set_efer(), below.
8261 */
8262 vm_entry_controls_init(vmx,
8263 (vmcs12->vm_entry_controls & ~VM_ENTRY_LOAD_IA32_EFER &
8264 ~VM_ENTRY_IA32E_MODE) |
8265 (vmcs_config.vmentry_ctrl & ~VM_ENTRY_IA32E_MODE));
8266
8267 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT) {
8268 vmcs_write64(GUEST_IA32_PAT, vmcs12->guest_ia32_pat);
8269 vcpu->arch.pat = vmcs12->guest_ia32_pat;
8270 } else if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
8271 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
8272
8273
8274 set_cr4_guest_host_mask(vmx);
8275
8276 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS)
8277 vmcs_write64(GUEST_BNDCFGS, vmcs12->guest_bndcfgs);
8278
8279 if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING)
8280 vmcs_write64(TSC_OFFSET,
8281 vmx->nested.vmcs01_tsc_offset + vmcs12->tsc_offset);
8282 else
8283 vmcs_write64(TSC_OFFSET, vmx->nested.vmcs01_tsc_offset);
8284
8285 if (enable_vpid) {
8286 /*
8287 * Trivially support vpid by letting L2s share their parent
8288 * L1's vpid. TODO: move to a more elaborate solution, giving
8289 * each L2 its own vpid and exposing the vpid feature to L1.
8290 */
8291 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
8292 vmx_flush_tlb(vcpu);
8293 }
8294
8295 if (nested_cpu_has_ept(vmcs12)) {
8296 kvm_mmu_unload(vcpu);
8297 nested_ept_init_mmu_context(vcpu);
8298 }
8299
8300 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER)
8301 vcpu->arch.efer = vmcs12->guest_ia32_efer;
8302 else if (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE)
8303 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
8304 else
8305 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
8306 /* Note: modifies VM_ENTRY/EXIT_CONTROLS and GUEST/HOST_IA32_EFER */
8307 vmx_set_efer(vcpu, vcpu->arch.efer);
8308
8309 /*
8310 * This sets GUEST_CR0 to vmcs12->guest_cr0, with possibly a modified
8311 * TS bit (for lazy fpu) and bits which we consider mandatory enabled.
8312 * The CR0_READ_SHADOW is what L2 should have expected to read given
8313 * the specifications by L1; It's not enough to take
8314 * vmcs12->cr0_read_shadow because on our cr0_guest_host_mask we we
8315 * have more bits than L1 expected.
8316 */
8317 vmx_set_cr0(vcpu, vmcs12->guest_cr0);
8318 vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
8319
8320 vmx_set_cr4(vcpu, vmcs12->guest_cr4);
8321 vmcs_writel(CR4_READ_SHADOW, nested_read_cr4(vmcs12));
8322
8323 /* shadow page tables on either EPT or shadow page tables */
8324 kvm_set_cr3(vcpu, vmcs12->guest_cr3);
8325 kvm_mmu_reset_context(vcpu);
8326
8327 if (!enable_ept)
8328 vcpu->arch.walk_mmu->inject_page_fault = vmx_inject_page_fault_nested;
8329
8330 /*
8331 * L1 may access the L2's PDPTR, so save them to construct vmcs12
8332 */
8333 if (enable_ept) {
8334 vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
8335 vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
8336 vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
8337 vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
8338 }
8339
8340 kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->guest_rsp);
8341 kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->guest_rip);
8342}
8343
8344/*
8345 * nested_vmx_run() handles a nested entry, i.e., a VMLAUNCH or VMRESUME on L1
8346 * for running an L2 nested guest.
8347 */
8348static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
8349{
8350 struct vmcs12 *vmcs12;
8351 struct vcpu_vmx *vmx = to_vmx(vcpu);
8352 int cpu;
8353 struct loaded_vmcs *vmcs02;
8354 bool ia32e;
8355
8356 if (!nested_vmx_check_permission(vcpu) ||
8357 !nested_vmx_check_vmcs12(vcpu))
8358 return 1;
8359
8360 skip_emulated_instruction(vcpu);
8361 vmcs12 = get_vmcs12(vcpu);
8362
8363 if (enable_shadow_vmcs)
8364 copy_shadow_to_vmcs12(vmx);
8365
8366 /*
8367 * The nested entry process starts with enforcing various prerequisites
8368 * on vmcs12 as required by the Intel SDM, and act appropriately when
8369 * they fail: As the SDM explains, some conditions should cause the
8370 * instruction to fail, while others will cause the instruction to seem
8371 * to succeed, but return an EXIT_REASON_INVALID_STATE.
8372 * To speed up the normal (success) code path, we should avoid checking
8373 * for misconfigurations which will anyway be caught by the processor
8374 * when using the merged vmcs02.
8375 */
8376 if (vmcs12->launch_state == launch) {
8377 nested_vmx_failValid(vcpu,
8378 launch ? VMXERR_VMLAUNCH_NONCLEAR_VMCS
8379 : VMXERR_VMRESUME_NONLAUNCHED_VMCS);
8380 return 1;
8381 }
8382
8383 if (vmcs12->guest_activity_state != GUEST_ACTIVITY_ACTIVE &&
8384 vmcs12->guest_activity_state != GUEST_ACTIVITY_HLT) {
8385 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
8386 return 1;
8387 }
8388
8389 if ((vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_MSR_BITMAPS) &&
8390 !PAGE_ALIGNED(vmcs12->msr_bitmap)) {
8391 /*TODO: Also verify bits beyond physical address width are 0*/
8392 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
8393 return 1;
8394 }
8395
8396 if (!nested_get_vmcs12_pages(vcpu, vmcs12)) {
8397 /*TODO: Also verify bits beyond physical address width are 0*/
8398 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
8399 return 1;
8400 }
8401
8402 if (vmcs12->vm_entry_msr_load_count > 0 ||
8403 vmcs12->vm_exit_msr_load_count > 0 ||
8404 vmcs12->vm_exit_msr_store_count > 0) {
8405 pr_warn_ratelimited("%s: VMCS MSR_{LOAD,STORE} unsupported\n",
8406 __func__);
8407 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
8408 return 1;
8409 }
8410
8411 if (!vmx_control_verify(vmcs12->cpu_based_vm_exec_control,
8412 nested_vmx_true_procbased_ctls_low,
8413 nested_vmx_procbased_ctls_high) ||
8414 !vmx_control_verify(vmcs12->secondary_vm_exec_control,
8415 nested_vmx_secondary_ctls_low, nested_vmx_secondary_ctls_high) ||
8416 !vmx_control_verify(vmcs12->pin_based_vm_exec_control,
8417 nested_vmx_pinbased_ctls_low, nested_vmx_pinbased_ctls_high) ||
8418 !vmx_control_verify(vmcs12->vm_exit_controls,
8419 nested_vmx_true_exit_ctls_low,
8420 nested_vmx_exit_ctls_high) ||
8421 !vmx_control_verify(vmcs12->vm_entry_controls,
8422 nested_vmx_true_entry_ctls_low,
8423 nested_vmx_entry_ctls_high))
8424 {
8425 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
8426 return 1;
8427 }
8428
8429 if (((vmcs12->host_cr0 & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON) ||
8430 ((vmcs12->host_cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON)) {
8431 nested_vmx_failValid(vcpu,
8432 VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
8433 return 1;
8434 }
8435
8436 if (!nested_cr0_valid(vmcs12, vmcs12->guest_cr0) ||
8437 ((vmcs12->guest_cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON)) {
8438 nested_vmx_entry_failure(vcpu, vmcs12,
8439 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
8440 return 1;
8441 }
8442 if (vmcs12->vmcs_link_pointer != -1ull) {
8443 nested_vmx_entry_failure(vcpu, vmcs12,
8444 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_VMCS_LINK_PTR);
8445 return 1;
8446 }
8447
8448 /*
8449 * If the load IA32_EFER VM-entry control is 1, the following checks
8450 * are performed on the field for the IA32_EFER MSR:
8451 * - Bits reserved in the IA32_EFER MSR must be 0.
8452 * - Bit 10 (corresponding to IA32_EFER.LMA) must equal the value of
8453 * the IA-32e mode guest VM-exit control. It must also be identical
8454 * to bit 8 (LME) if bit 31 in the CR0 field (corresponding to
8455 * CR0.PG) is 1.
8456 */
8457 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER) {
8458 ia32e = (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) != 0;
8459 if (!kvm_valid_efer(vcpu, vmcs12->guest_ia32_efer) ||
8460 ia32e != !!(vmcs12->guest_ia32_efer & EFER_LMA) ||
8461 ((vmcs12->guest_cr0 & X86_CR0_PG) &&
8462 ia32e != !!(vmcs12->guest_ia32_efer & EFER_LME))) {
8463 nested_vmx_entry_failure(vcpu, vmcs12,
8464 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
8465 return 1;
8466 }
8467 }
8468
8469 /*
8470 * If the load IA32_EFER VM-exit control is 1, bits reserved in the
8471 * IA32_EFER MSR must be 0 in the field for that register. In addition,
8472 * the values of the LMA and LME bits in the field must each be that of
8473 * the host address-space size VM-exit control.
8474 */
8475 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER) {
8476 ia32e = (vmcs12->vm_exit_controls &
8477 VM_EXIT_HOST_ADDR_SPACE_SIZE) != 0;
8478 if (!kvm_valid_efer(vcpu, vmcs12->host_ia32_efer) ||
8479 ia32e != !!(vmcs12->host_ia32_efer & EFER_LMA) ||
8480 ia32e != !!(vmcs12->host_ia32_efer & EFER_LME)) {
8481 nested_vmx_entry_failure(vcpu, vmcs12,
8482 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
8483 return 1;
8484 }
8485 }
8486
8487 /*
8488 * We're finally done with prerequisite checking, and can start with
8489 * the nested entry.
8490 */
8491
8492 vmcs02 = nested_get_current_vmcs02(vmx);
8493 if (!vmcs02)
8494 return -ENOMEM;
8495
8496 enter_guest_mode(vcpu);
8497
8498 vmx->nested.vmcs01_tsc_offset = vmcs_read64(TSC_OFFSET);
8499
8500 if (!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS))
8501 vmx->nested.vmcs01_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
8502
8503 cpu = get_cpu();
8504 vmx->loaded_vmcs = vmcs02;
8505 vmx_vcpu_put(vcpu);
8506 vmx_vcpu_load(vcpu, cpu);
8507 vcpu->cpu = cpu;
8508 put_cpu();
8509
8510 vmx_segment_cache_clear(vmx);
8511
8512 vmcs12->launch_state = 1;
8513
8514 prepare_vmcs02(vcpu, vmcs12);
8515
8516 if (vmcs12->guest_activity_state == GUEST_ACTIVITY_HLT)
8517 return kvm_emulate_halt(vcpu);
8518
8519 vmx->nested.nested_run_pending = 1;
8520
8521 /*
8522 * Note no nested_vmx_succeed or nested_vmx_fail here. At this point
8523 * we are no longer running L1, and VMLAUNCH/VMRESUME has not yet
8524 * returned as far as L1 is concerned. It will only return (and set
8525 * the success flag) when L2 exits (see nested_vmx_vmexit()).
8526 */
8527 return 1;
8528}
8529
8530/*
8531 * On a nested exit from L2 to L1, vmcs12.guest_cr0 might not be up-to-date
8532 * because L2 may have changed some cr0 bits directly (CRO_GUEST_HOST_MASK).
8533 * This function returns the new value we should put in vmcs12.guest_cr0.
8534 * It's not enough to just return the vmcs02 GUEST_CR0. Rather,
8535 * 1. Bits that neither L0 nor L1 trapped, were set directly by L2 and are now
8536 * available in vmcs02 GUEST_CR0. (Note: It's enough to check that L0
8537 * didn't trap the bit, because if L1 did, so would L0).
8538 * 2. Bits that L1 asked to trap (and therefore L0 also did) could not have
8539 * been modified by L2, and L1 knows it. So just leave the old value of
8540 * the bit from vmcs12.guest_cr0. Note that the bit from vmcs02 GUEST_CR0
8541 * isn't relevant, because if L0 traps this bit it can set it to anything.
8542 * 3. Bits that L1 didn't trap, but L0 did. L1 believes the guest could have
8543 * changed these bits, and therefore they need to be updated, but L0
8544 * didn't necessarily allow them to be changed in GUEST_CR0 - and rather
8545 * put them in vmcs02 CR0_READ_SHADOW. So take these bits from there.
8546 */
8547static inline unsigned long
8548vmcs12_guest_cr0(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
8549{
8550 return
8551 /*1*/ (vmcs_readl(GUEST_CR0) & vcpu->arch.cr0_guest_owned_bits) |
8552 /*2*/ (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask) |
8553 /*3*/ (vmcs_readl(CR0_READ_SHADOW) & ~(vmcs12->cr0_guest_host_mask |
8554 vcpu->arch.cr0_guest_owned_bits));
8555}
8556
8557static inline unsigned long
8558vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
8559{
8560 return
8561 /*1*/ (vmcs_readl(GUEST_CR4) & vcpu->arch.cr4_guest_owned_bits) |
8562 /*2*/ (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask) |
8563 /*3*/ (vmcs_readl(CR4_READ_SHADOW) & ~(vmcs12->cr4_guest_host_mask |
8564 vcpu->arch.cr4_guest_owned_bits));
8565}
8566
8567static void vmcs12_save_pending_event(struct kvm_vcpu *vcpu,
8568 struct vmcs12 *vmcs12)
8569{
8570 u32 idt_vectoring;
8571 unsigned int nr;
8572
8573 if (vcpu->arch.exception.pending && vcpu->arch.exception.reinject) {
8574 nr = vcpu->arch.exception.nr;
8575 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
8576
8577 if (kvm_exception_is_soft(nr)) {
8578 vmcs12->vm_exit_instruction_len =
8579 vcpu->arch.event_exit_inst_len;
8580 idt_vectoring |= INTR_TYPE_SOFT_EXCEPTION;
8581 } else
8582 idt_vectoring |= INTR_TYPE_HARD_EXCEPTION;
8583
8584 if (vcpu->arch.exception.has_error_code) {
8585 idt_vectoring |= VECTORING_INFO_DELIVER_CODE_MASK;
8586 vmcs12->idt_vectoring_error_code =
8587 vcpu->arch.exception.error_code;
8588 }
8589
8590 vmcs12->idt_vectoring_info_field = idt_vectoring;
8591 } else if (vcpu->arch.nmi_injected) {
8592 vmcs12->idt_vectoring_info_field =
8593 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR;
8594 } else if (vcpu->arch.interrupt.pending) {
8595 nr = vcpu->arch.interrupt.nr;
8596 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
8597
8598 if (vcpu->arch.interrupt.soft) {
8599 idt_vectoring |= INTR_TYPE_SOFT_INTR;
8600 vmcs12->vm_entry_instruction_len =
8601 vcpu->arch.event_exit_inst_len;
8602 } else
8603 idt_vectoring |= INTR_TYPE_EXT_INTR;
8604
8605 vmcs12->idt_vectoring_info_field = idt_vectoring;
8606 }
8607}
8608
8609static int vmx_check_nested_events(struct kvm_vcpu *vcpu, bool external_intr)
8610{
8611 struct vcpu_vmx *vmx = to_vmx(vcpu);
8612
8613 if (nested_cpu_has_preemption_timer(get_vmcs12(vcpu)) &&
8614 vmx->nested.preemption_timer_expired) {
8615 if (vmx->nested.nested_run_pending)
8616 return -EBUSY;
8617 nested_vmx_vmexit(vcpu, EXIT_REASON_PREEMPTION_TIMER, 0, 0);
8618 return 0;
8619 }
8620
8621 if (vcpu->arch.nmi_pending && nested_exit_on_nmi(vcpu)) {
8622 if (vmx->nested.nested_run_pending ||
8623 vcpu->arch.interrupt.pending)
8624 return -EBUSY;
8625 nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
8626 NMI_VECTOR | INTR_TYPE_NMI_INTR |
8627 INTR_INFO_VALID_MASK, 0);
8628 /*
8629 * The NMI-triggered VM exit counts as injection:
8630 * clear this one and block further NMIs.
8631 */
8632 vcpu->arch.nmi_pending = 0;
8633 vmx_set_nmi_mask(vcpu, true);
8634 return 0;
8635 }
8636
8637 if ((kvm_cpu_has_interrupt(vcpu) || external_intr) &&
8638 nested_exit_on_intr(vcpu)) {
8639 if (vmx->nested.nested_run_pending)
8640 return -EBUSY;
8641 nested_vmx_vmexit(vcpu, EXIT_REASON_EXTERNAL_INTERRUPT, 0, 0);
8642 }
8643
8644 return 0;
8645}
8646
8647static u32 vmx_get_preemption_timer_value(struct kvm_vcpu *vcpu)
8648{
8649 ktime_t remaining =
8650 hrtimer_get_remaining(&to_vmx(vcpu)->nested.preemption_timer);
8651 u64 value;
8652
8653 if (ktime_to_ns(remaining) <= 0)
8654 return 0;
8655
8656 value = ktime_to_ns(remaining) * vcpu->arch.virtual_tsc_khz;
8657 do_div(value, 1000000);
8658 return value >> VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
8659}
8660
8661/*
8662 * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
8663 * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),
8664 * and this function updates it to reflect the changes to the guest state while
8665 * L2 was running (and perhaps made some exits which were handled directly by L0
8666 * without going back to L1), and to reflect the exit reason.
8667 * Note that we do not have to copy here all VMCS fields, just those that
8668 * could have changed by the L2 guest or the exit - i.e., the guest-state and
8669 * exit-information fields only. Other fields are modified by L1 with VMWRITE,
8670 * which already writes to vmcs12 directly.
8671 */
8672static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
8673 u32 exit_reason, u32 exit_intr_info,
8674 unsigned long exit_qualification)
8675{
8676 /* update guest state fields: */
8677 vmcs12->guest_cr0 = vmcs12_guest_cr0(vcpu, vmcs12);
8678 vmcs12->guest_cr4 = vmcs12_guest_cr4(vcpu, vmcs12);
8679
8680 vmcs12->guest_rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
8681 vmcs12->guest_rip = kvm_register_read(vcpu, VCPU_REGS_RIP);
8682 vmcs12->guest_rflags = vmcs_readl(GUEST_RFLAGS);
8683
8684 vmcs12->guest_es_selector = vmcs_read16(GUEST_ES_SELECTOR);
8685 vmcs12->guest_cs_selector = vmcs_read16(GUEST_CS_SELECTOR);
8686 vmcs12->guest_ss_selector = vmcs_read16(GUEST_SS_SELECTOR);
8687 vmcs12->guest_ds_selector = vmcs_read16(GUEST_DS_SELECTOR);
8688 vmcs12->guest_fs_selector = vmcs_read16(GUEST_FS_SELECTOR);
8689 vmcs12->guest_gs_selector = vmcs_read16(GUEST_GS_SELECTOR);
8690 vmcs12->guest_ldtr_selector = vmcs_read16(GUEST_LDTR_SELECTOR);
8691 vmcs12->guest_tr_selector = vmcs_read16(GUEST_TR_SELECTOR);
8692 vmcs12->guest_es_limit = vmcs_read32(GUEST_ES_LIMIT);
8693 vmcs12->guest_cs_limit = vmcs_read32(GUEST_CS_LIMIT);
8694 vmcs12->guest_ss_limit = vmcs_read32(GUEST_SS_LIMIT);
8695 vmcs12->guest_ds_limit = vmcs_read32(GUEST_DS_LIMIT);
8696 vmcs12->guest_fs_limit = vmcs_read32(GUEST_FS_LIMIT);
8697 vmcs12->guest_gs_limit = vmcs_read32(GUEST_GS_LIMIT);
8698 vmcs12->guest_ldtr_limit = vmcs_read32(GUEST_LDTR_LIMIT);
8699 vmcs12->guest_tr_limit = vmcs_read32(GUEST_TR_LIMIT);
8700 vmcs12->guest_gdtr_limit = vmcs_read32(GUEST_GDTR_LIMIT);
8701 vmcs12->guest_idtr_limit = vmcs_read32(GUEST_IDTR_LIMIT);
8702 vmcs12->guest_es_ar_bytes = vmcs_read32(GUEST_ES_AR_BYTES);
8703 vmcs12->guest_cs_ar_bytes = vmcs_read32(GUEST_CS_AR_BYTES);
8704 vmcs12->guest_ss_ar_bytes = vmcs_read32(GUEST_SS_AR_BYTES);
8705 vmcs12->guest_ds_ar_bytes = vmcs_read32(GUEST_DS_AR_BYTES);
8706 vmcs12->guest_fs_ar_bytes = vmcs_read32(GUEST_FS_AR_BYTES);
8707 vmcs12->guest_gs_ar_bytes = vmcs_read32(GUEST_GS_AR_BYTES);
8708 vmcs12->guest_ldtr_ar_bytes = vmcs_read32(GUEST_LDTR_AR_BYTES);
8709 vmcs12->guest_tr_ar_bytes = vmcs_read32(GUEST_TR_AR_BYTES);
8710 vmcs12->guest_es_base = vmcs_readl(GUEST_ES_BASE);
8711 vmcs12->guest_cs_base = vmcs_readl(GUEST_CS_BASE);
8712 vmcs12->guest_ss_base = vmcs_readl(GUEST_SS_BASE);
8713 vmcs12->guest_ds_base = vmcs_readl(GUEST_DS_BASE);
8714 vmcs12->guest_fs_base = vmcs_readl(GUEST_FS_BASE);
8715 vmcs12->guest_gs_base = vmcs_readl(GUEST_GS_BASE);
8716 vmcs12->guest_ldtr_base = vmcs_readl(GUEST_LDTR_BASE);
8717 vmcs12->guest_tr_base = vmcs_readl(GUEST_TR_BASE);
8718 vmcs12->guest_gdtr_base = vmcs_readl(GUEST_GDTR_BASE);
8719 vmcs12->guest_idtr_base = vmcs_readl(GUEST_IDTR_BASE);
8720
8721 vmcs12->guest_interruptibility_info =
8722 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
8723 vmcs12->guest_pending_dbg_exceptions =
8724 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS);
8725 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
8726 vmcs12->guest_activity_state = GUEST_ACTIVITY_HLT;
8727 else
8728 vmcs12->guest_activity_state = GUEST_ACTIVITY_ACTIVE;
8729
8730 if (nested_cpu_has_preemption_timer(vmcs12)) {
8731 if (vmcs12->vm_exit_controls &
8732 VM_EXIT_SAVE_VMX_PREEMPTION_TIMER)
8733 vmcs12->vmx_preemption_timer_value =
8734 vmx_get_preemption_timer_value(vcpu);
8735 hrtimer_cancel(&to_vmx(vcpu)->nested.preemption_timer);
8736 }
8737
8738 /*
8739 * In some cases (usually, nested EPT), L2 is allowed to change its
8740 * own CR3 without exiting. If it has changed it, we must keep it.
8741 * Of course, if L0 is using shadow page tables, GUEST_CR3 was defined
8742 * by L0, not L1 or L2, so we mustn't unconditionally copy it to vmcs12.
8743 *
8744 * Additionally, restore L2's PDPTR to vmcs12.
8745 */
8746 if (enable_ept) {
8747 vmcs12->guest_cr3 = vmcs_read64(GUEST_CR3);
8748 vmcs12->guest_pdptr0 = vmcs_read64(GUEST_PDPTR0);
8749 vmcs12->guest_pdptr1 = vmcs_read64(GUEST_PDPTR1);
8750 vmcs12->guest_pdptr2 = vmcs_read64(GUEST_PDPTR2);
8751 vmcs12->guest_pdptr3 = vmcs_read64(GUEST_PDPTR3);
8752 }
8753
8754 vmcs12->vm_entry_controls =
8755 (vmcs12->vm_entry_controls & ~VM_ENTRY_IA32E_MODE) |
8756 (vm_entry_controls_get(to_vmx(vcpu)) & VM_ENTRY_IA32E_MODE);
8757
8758 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_DEBUG_CONTROLS) {
8759 kvm_get_dr(vcpu, 7, (unsigned long *)&vmcs12->guest_dr7);
8760 vmcs12->guest_ia32_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
8761 }
8762
8763 /* TODO: These cannot have changed unless we have MSR bitmaps and
8764 * the relevant bit asks not to trap the change */
8765 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_PAT)
8766 vmcs12->guest_ia32_pat = vmcs_read64(GUEST_IA32_PAT);
8767 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_EFER)
8768 vmcs12->guest_ia32_efer = vcpu->arch.efer;
8769 vmcs12->guest_sysenter_cs = vmcs_read32(GUEST_SYSENTER_CS);
8770 vmcs12->guest_sysenter_esp = vmcs_readl(GUEST_SYSENTER_ESP);
8771 vmcs12->guest_sysenter_eip = vmcs_readl(GUEST_SYSENTER_EIP);
8772 if (vmx_mpx_supported())
8773 vmcs12->guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
8774
8775 /* update exit information fields: */
8776
8777 vmcs12->vm_exit_reason = exit_reason;
8778 vmcs12->exit_qualification = exit_qualification;
8779
8780 vmcs12->vm_exit_intr_info = exit_intr_info;
8781 if ((vmcs12->vm_exit_intr_info &
8782 (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK)) ==
8783 (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK))
8784 vmcs12->vm_exit_intr_error_code =
8785 vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
8786 vmcs12->idt_vectoring_info_field = 0;
8787 vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
8788 vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
8789
8790 if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY)) {
8791 /* vm_entry_intr_info_field is cleared on exit. Emulate this
8792 * instead of reading the real value. */
8793 vmcs12->vm_entry_intr_info_field &= ~INTR_INFO_VALID_MASK;
8794
8795 /*
8796 * Transfer the event that L0 or L1 may wanted to inject into
8797 * L2 to IDT_VECTORING_INFO_FIELD.
8798 */
8799 vmcs12_save_pending_event(vcpu, vmcs12);
8800 }
8801
8802 /*
8803 * Drop what we picked up for L2 via vmx_complete_interrupts. It is
8804 * preserved above and would only end up incorrectly in L1.
8805 */
8806 vcpu->arch.nmi_injected = false;
8807 kvm_clear_exception_queue(vcpu);
8808 kvm_clear_interrupt_queue(vcpu);
8809}
8810
8811/*
8812 * A part of what we need to when the nested L2 guest exits and we want to
8813 * run its L1 parent, is to reset L1's guest state to the host state specified
8814 * in vmcs12.
8815 * This function is to be called not only on normal nested exit, but also on
8816 * a nested entry failure, as explained in Intel's spec, 3B.23.7 ("VM-Entry
8817 * Failures During or After Loading Guest State").
8818 * This function should be called when the active VMCS is L1's (vmcs01).
8819 */
8820static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
8821 struct vmcs12 *vmcs12)
8822{
8823 struct kvm_segment seg;
8824
8825 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER)
8826 vcpu->arch.efer = vmcs12->host_ia32_efer;
8827 else if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
8828 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
8829 else
8830 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
8831 vmx_set_efer(vcpu, vcpu->arch.efer);
8832
8833 kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->host_rsp);
8834 kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->host_rip);
8835 vmx_set_rflags(vcpu, X86_EFLAGS_FIXED);
8836 /*
8837 * Note that calling vmx_set_cr0 is important, even if cr0 hasn't
8838 * actually changed, because it depends on the current state of
8839 * fpu_active (which may have changed).
8840 * Note that vmx_set_cr0 refers to efer set above.
8841 */
8842 vmx_set_cr0(vcpu, vmcs12->host_cr0);
8843 /*
8844 * If we did fpu_activate()/fpu_deactivate() during L2's run, we need
8845 * to apply the same changes to L1's vmcs. We just set cr0 correctly,
8846 * but we also need to update cr0_guest_host_mask and exception_bitmap.
8847 */
8848 update_exception_bitmap(vcpu);
8849 vcpu->arch.cr0_guest_owned_bits = (vcpu->fpu_active ? X86_CR0_TS : 0);
8850 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
8851
8852 /*
8853 * Note that CR4_GUEST_HOST_MASK is already set in the original vmcs01
8854 * (KVM doesn't change it)- no reason to call set_cr4_guest_host_mask();
8855 */
8856 vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
8857 kvm_set_cr4(vcpu, vmcs12->host_cr4);
8858
8859 nested_ept_uninit_mmu_context(vcpu);
8860
8861 kvm_set_cr3(vcpu, vmcs12->host_cr3);
8862 kvm_mmu_reset_context(vcpu);
8863
8864 if (!enable_ept)
8865 vcpu->arch.walk_mmu->inject_page_fault = kvm_inject_page_fault;
8866
8867 if (enable_vpid) {
8868 /*
8869 * Trivially support vpid by letting L2s share their parent
8870 * L1's vpid. TODO: move to a more elaborate solution, giving
8871 * each L2 its own vpid and exposing the vpid feature to L1.
8872 */
8873 vmx_flush_tlb(vcpu);
8874 }
8875
8876
8877 vmcs_write32(GUEST_SYSENTER_CS, vmcs12->host_ia32_sysenter_cs);
8878 vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->host_ia32_sysenter_esp);
8879 vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->host_ia32_sysenter_eip);
8880 vmcs_writel(GUEST_IDTR_BASE, vmcs12->host_idtr_base);
8881 vmcs_writel(GUEST_GDTR_BASE, vmcs12->host_gdtr_base);
8882
8883 /* If not VM_EXIT_CLEAR_BNDCFGS, the L2 value propagates to L1. */
8884 if (vmcs12->vm_exit_controls & VM_EXIT_CLEAR_BNDCFGS)
8885 vmcs_write64(GUEST_BNDCFGS, 0);
8886
8887 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) {
8888 vmcs_write64(GUEST_IA32_PAT, vmcs12->host_ia32_pat);
8889 vcpu->arch.pat = vmcs12->host_ia32_pat;
8890 }
8891 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
8892 vmcs_write64(GUEST_IA32_PERF_GLOBAL_CTRL,
8893 vmcs12->host_ia32_perf_global_ctrl);
8894
8895 /* Set L1 segment info according to Intel SDM
8896 27.5.2 Loading Host Segment and Descriptor-Table Registers */
8897 seg = (struct kvm_segment) {
8898 .base = 0,
8899 .limit = 0xFFFFFFFF,
8900 .selector = vmcs12->host_cs_selector,
8901 .type = 11,
8902 .present = 1,
8903 .s = 1,
8904 .g = 1
8905 };
8906 if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
8907 seg.l = 1;
8908 else
8909 seg.db = 1;
8910 vmx_set_segment(vcpu, &seg, VCPU_SREG_CS);
8911 seg = (struct kvm_segment) {
8912 .base = 0,
8913 .limit = 0xFFFFFFFF,
8914 .type = 3,
8915 .present = 1,
8916 .s = 1,
8917 .db = 1,
8918 .g = 1
8919 };
8920 seg.selector = vmcs12->host_ds_selector;
8921 vmx_set_segment(vcpu, &seg, VCPU_SREG_DS);
8922 seg.selector = vmcs12->host_es_selector;
8923 vmx_set_segment(vcpu, &seg, VCPU_SREG_ES);
8924 seg.selector = vmcs12->host_ss_selector;
8925 vmx_set_segment(vcpu, &seg, VCPU_SREG_SS);
8926 seg.selector = vmcs12->host_fs_selector;
8927 seg.base = vmcs12->host_fs_base;
8928 vmx_set_segment(vcpu, &seg, VCPU_SREG_FS);
8929 seg.selector = vmcs12->host_gs_selector;
8930 seg.base = vmcs12->host_gs_base;
8931 vmx_set_segment(vcpu, &seg, VCPU_SREG_GS);
8932 seg = (struct kvm_segment) {
8933 .base = vmcs12->host_tr_base,
8934 .limit = 0x67,
8935 .selector = vmcs12->host_tr_selector,
8936 .type = 11,
8937 .present = 1
8938 };
8939 vmx_set_segment(vcpu, &seg, VCPU_SREG_TR);
8940
8941 kvm_set_dr(vcpu, 7, 0x400);
8942 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
8943}
8944
8945/*
8946 * Emulate an exit from nested guest (L2) to L1, i.e., prepare to run L1
8947 * and modify vmcs12 to make it see what it would expect to see there if
8948 * L2 was its real guest. Must only be called when in L2 (is_guest_mode())
8949 */
8950static void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
8951 u32 exit_intr_info,
8952 unsigned long exit_qualification)
8953{
8954 struct vcpu_vmx *vmx = to_vmx(vcpu);
8955 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
8956
8957 /* trying to cancel vmlaunch/vmresume is a bug */
8958 WARN_ON_ONCE(vmx->nested.nested_run_pending);
8959
8960 leave_guest_mode(vcpu);
8961 prepare_vmcs12(vcpu, vmcs12, exit_reason, exit_intr_info,
8962 exit_qualification);
8963
8964 vmx_load_vmcs01(vcpu);
8965
8966 if ((exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT)
8967 && nested_exit_intr_ack_set(vcpu)) {
8968 int irq = kvm_cpu_get_interrupt(vcpu);
8969 WARN_ON(irq < 0);
8970 vmcs12->vm_exit_intr_info = irq |
8971 INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR;
8972 }
8973
8974 trace_kvm_nested_vmexit_inject(vmcs12->vm_exit_reason,
8975 vmcs12->exit_qualification,
8976 vmcs12->idt_vectoring_info_field,
8977 vmcs12->vm_exit_intr_info,
8978 vmcs12->vm_exit_intr_error_code,
8979 KVM_ISA_VMX);
8980
8981 vm_entry_controls_init(vmx, vmcs_read32(VM_ENTRY_CONTROLS));
8982 vm_exit_controls_init(vmx, vmcs_read32(VM_EXIT_CONTROLS));
8983 vmx_segment_cache_clear(vmx);
8984
8985 /* if no vmcs02 cache requested, remove the one we used */
8986 if (VMCS02_POOL_SIZE == 0)
8987 nested_free_vmcs02(vmx, vmx->nested.current_vmptr);
8988
8989 load_vmcs12_host_state(vcpu, vmcs12);
8990
8991 /* Update TSC_OFFSET if TSC was changed while L2 ran */
8992 vmcs_write64(TSC_OFFSET, vmx->nested.vmcs01_tsc_offset);
8993
8994 /* This is needed for same reason as it was needed in prepare_vmcs02 */
8995 vmx->host_rsp = 0;
8996
8997 /* Unpin physical memory we referred to in vmcs02 */
8998 if (vmx->nested.apic_access_page) {
8999 nested_release_page(vmx->nested.apic_access_page);
9000 vmx->nested.apic_access_page = NULL;
9001 }
9002 if (vmx->nested.virtual_apic_page) {
9003 nested_release_page(vmx->nested.virtual_apic_page);
9004 vmx->nested.virtual_apic_page = NULL;
9005 }
9006
9007 /*
9008 * We are now running in L2, mmu_notifier will force to reload the
9009 * page's hpa for L2 vmcs. Need to reload it for L1 before entering L1.
9010 */
9011 kvm_vcpu_reload_apic_access_page(vcpu);
9012
9013 /*
9014 * Exiting from L2 to L1, we're now back to L1 which thinks it just
9015 * finished a VMLAUNCH or VMRESUME instruction, so we need to set the
9016 * success or failure flag accordingly.
9017 */
9018 if (unlikely(vmx->fail)) {
9019 vmx->fail = 0;
9020 nested_vmx_failValid(vcpu, vmcs_read32(VM_INSTRUCTION_ERROR));
9021 } else
9022 nested_vmx_succeed(vcpu);
9023 if (enable_shadow_vmcs)
9024 vmx->nested.sync_shadow_vmcs = true;
9025
9026 /* in case we halted in L2 */
9027 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9028}
9029
9030/*
9031 * Forcibly leave nested mode in order to be able to reset the VCPU later on.
9032 */
9033static void vmx_leave_nested(struct kvm_vcpu *vcpu)
9034{
9035 if (is_guest_mode(vcpu))
9036 nested_vmx_vmexit(vcpu, -1, 0, 0);
9037 free_nested(to_vmx(vcpu));
9038}
9039
9040/*
9041 * L1's failure to enter L2 is a subset of a normal exit, as explained in
9042 * 23.7 "VM-entry failures during or after loading guest state" (this also
9043 * lists the acceptable exit-reason and exit-qualification parameters).
9044 * It should only be called before L2 actually succeeded to run, and when
9045 * vmcs01 is current (it doesn't leave_guest_mode() or switch vmcss).
9046 */
9047static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
9048 struct vmcs12 *vmcs12,
9049 u32 reason, unsigned long qualification)
9050{
9051 load_vmcs12_host_state(vcpu, vmcs12);
9052 vmcs12->vm_exit_reason = reason | VMX_EXIT_REASONS_FAILED_VMENTRY;
9053 vmcs12->exit_qualification = qualification;
9054 nested_vmx_succeed(vcpu);
9055 if (enable_shadow_vmcs)
9056 to_vmx(vcpu)->nested.sync_shadow_vmcs = true;
9057}
9058
9059static int vmx_check_intercept(struct kvm_vcpu *vcpu,
9060 struct x86_instruction_info *info,
9061 enum x86_intercept_stage stage)
9062{
9063 return X86EMUL_CONTINUE;
9064}
9065
9066static void vmx_sched_in(struct kvm_vcpu *vcpu, int cpu)
9067{
9068 if (ple_gap)
9069 shrink_ple_window(vcpu);
9070}
9071
9072static struct kvm_x86_ops vmx_x86_ops = {
9073 .cpu_has_kvm_support = cpu_has_kvm_support,
9074 .disabled_by_bios = vmx_disabled_by_bios,
9075 .hardware_setup = hardware_setup,
9076 .hardware_unsetup = hardware_unsetup,
9077 .check_processor_compatibility = vmx_check_processor_compat,
9078 .hardware_enable = hardware_enable,
9079 .hardware_disable = hardware_disable,
9080 .cpu_has_accelerated_tpr = report_flexpriority,
9081
9082 .vcpu_create = vmx_create_vcpu,
9083 .vcpu_free = vmx_free_vcpu,
9084 .vcpu_reset = vmx_vcpu_reset,
9085
9086 .prepare_guest_switch = vmx_save_host_state,
9087 .vcpu_load = vmx_vcpu_load,
9088 .vcpu_put = vmx_vcpu_put,
9089
9090 .update_db_bp_intercept = update_exception_bitmap,
9091 .get_msr = vmx_get_msr,
9092 .set_msr = vmx_set_msr,
9093 .get_segment_base = vmx_get_segment_base,
9094 .get_segment = vmx_get_segment,
9095 .set_segment = vmx_set_segment,
9096 .get_cpl = vmx_get_cpl,
9097 .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
9098 .decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
9099 .decache_cr3 = vmx_decache_cr3,
9100 .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
9101 .set_cr0 = vmx_set_cr0,
9102 .set_cr3 = vmx_set_cr3,
9103 .set_cr4 = vmx_set_cr4,
9104 .set_efer = vmx_set_efer,
9105 .get_idt = vmx_get_idt,
9106 .set_idt = vmx_set_idt,
9107 .get_gdt = vmx_get_gdt,
9108 .set_gdt = vmx_set_gdt,
9109 .get_dr6 = vmx_get_dr6,
9110 .set_dr6 = vmx_set_dr6,
9111 .set_dr7 = vmx_set_dr7,
9112 .sync_dirty_debug_regs = vmx_sync_dirty_debug_regs,
9113 .cache_reg = vmx_cache_reg,
9114 .get_rflags = vmx_get_rflags,
9115 .set_rflags = vmx_set_rflags,
9116 .fpu_deactivate = vmx_fpu_deactivate,
9117
9118 .tlb_flush = vmx_flush_tlb,
9119
9120 .run = vmx_vcpu_run,
9121 .handle_exit = vmx_handle_exit,
9122 .skip_emulated_instruction = skip_emulated_instruction,
9123 .set_interrupt_shadow = vmx_set_interrupt_shadow,
9124 .get_interrupt_shadow = vmx_get_interrupt_shadow,
9125 .patch_hypercall = vmx_patch_hypercall,
9126 .set_irq = vmx_inject_irq,
9127 .set_nmi = vmx_inject_nmi,
9128 .queue_exception = vmx_queue_exception,
9129 .cancel_injection = vmx_cancel_injection,
9130 .interrupt_allowed = vmx_interrupt_allowed,
9131 .nmi_allowed = vmx_nmi_allowed,
9132 .get_nmi_mask = vmx_get_nmi_mask,
9133 .set_nmi_mask = vmx_set_nmi_mask,
9134 .enable_nmi_window = enable_nmi_window,
9135 .enable_irq_window = enable_irq_window,
9136 .update_cr8_intercept = update_cr8_intercept,
9137 .set_virtual_x2apic_mode = vmx_set_virtual_x2apic_mode,
9138 .set_apic_access_page_addr = vmx_set_apic_access_page_addr,
9139 .vm_has_apicv = vmx_vm_has_apicv,
9140 .load_eoi_exitmap = vmx_load_eoi_exitmap,
9141 .hwapic_irr_update = vmx_hwapic_irr_update,
9142 .hwapic_isr_update = vmx_hwapic_isr_update,
9143 .sync_pir_to_irr = vmx_sync_pir_to_irr,
9144 .deliver_posted_interrupt = vmx_deliver_posted_interrupt,
9145
9146 .set_tss_addr = vmx_set_tss_addr,
9147 .get_tdp_level = get_ept_level,
9148 .get_mt_mask = vmx_get_mt_mask,
9149
9150 .get_exit_info = vmx_get_exit_info,
9151
9152 .get_lpage_level = vmx_get_lpage_level,
9153
9154 .cpuid_update = vmx_cpuid_update,
9155
9156 .rdtscp_supported = vmx_rdtscp_supported,
9157 .invpcid_supported = vmx_invpcid_supported,
9158
9159 .set_supported_cpuid = vmx_set_supported_cpuid,
9160
9161 .has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
9162
9163 .set_tsc_khz = vmx_set_tsc_khz,
9164 .read_tsc_offset = vmx_read_tsc_offset,
9165 .write_tsc_offset = vmx_write_tsc_offset,
9166 .adjust_tsc_offset = vmx_adjust_tsc_offset,
9167 .compute_tsc_offset = vmx_compute_tsc_offset,
9168 .read_l1_tsc = vmx_read_l1_tsc,
9169
9170 .set_tdp_cr3 = vmx_set_cr3,
9171
9172 .check_intercept = vmx_check_intercept,
9173 .handle_external_intr = vmx_handle_external_intr,
9174 .mpx_supported = vmx_mpx_supported,
9175
9176 .check_nested_events = vmx_check_nested_events,
9177
9178 .sched_in = vmx_sched_in,
9179};
9180
9181static int __init vmx_init(void)
9182{
9183 int r, i, msr;
9184
9185 rdmsrl_safe(MSR_EFER, &host_efer);
9186
9187 for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i)
9188 kvm_define_shared_msr(i, vmx_msr_index[i]);
9189
9190 vmx_io_bitmap_a = (unsigned long *)__get_free_page(GFP_KERNEL);
9191 if (!vmx_io_bitmap_a)
9192 return -ENOMEM;
9193
9194 r = -ENOMEM;
9195
9196 vmx_io_bitmap_b = (unsigned long *)__get_free_page(GFP_KERNEL);
9197 if (!vmx_io_bitmap_b)
9198 goto out;
9199
9200 vmx_msr_bitmap_legacy = (unsigned long *)__get_free_page(GFP_KERNEL);
9201 if (!vmx_msr_bitmap_legacy)
9202 goto out1;
9203
9204 vmx_msr_bitmap_legacy_x2apic =
9205 (unsigned long *)__get_free_page(GFP_KERNEL);
9206 if (!vmx_msr_bitmap_legacy_x2apic)
9207 goto out2;
9208
9209 vmx_msr_bitmap_longmode = (unsigned long *)__get_free_page(GFP_KERNEL);
9210 if (!vmx_msr_bitmap_longmode)
9211 goto out3;
9212
9213 vmx_msr_bitmap_longmode_x2apic =
9214 (unsigned long *)__get_free_page(GFP_KERNEL);
9215 if (!vmx_msr_bitmap_longmode_x2apic)
9216 goto out4;
9217 vmx_vmread_bitmap = (unsigned long *)__get_free_page(GFP_KERNEL);
9218 if (!vmx_vmread_bitmap)
9219 goto out5;
9220
9221 vmx_vmwrite_bitmap = (unsigned long *)__get_free_page(GFP_KERNEL);
9222 if (!vmx_vmwrite_bitmap)
9223 goto out6;
9224
9225 memset(vmx_vmread_bitmap, 0xff, PAGE_SIZE);
9226 memset(vmx_vmwrite_bitmap, 0xff, PAGE_SIZE);
9227
9228 /*
9229 * Allow direct access to the PC debug port (it is often used for I/O
9230 * delays, but the vmexits simply slow things down).
9231 */
9232 memset(vmx_io_bitmap_a, 0xff, PAGE_SIZE);
9233 clear_bit(0x80, vmx_io_bitmap_a);
9234
9235 memset(vmx_io_bitmap_b, 0xff, PAGE_SIZE);
9236
9237 memset(vmx_msr_bitmap_legacy, 0xff, PAGE_SIZE);
9238 memset(vmx_msr_bitmap_longmode, 0xff, PAGE_SIZE);
9239
9240 set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
9241
9242 r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx),
9243 __alignof__(struct vcpu_vmx), THIS_MODULE);
9244 if (r)
9245 goto out7;
9246
9247#ifdef CONFIG_KEXEC
9248 rcu_assign_pointer(crash_vmclear_loaded_vmcss,
9249 crash_vmclear_local_loaded_vmcss);
9250#endif
9251
9252 vmx_disable_intercept_for_msr(MSR_FS_BASE, false);
9253 vmx_disable_intercept_for_msr(MSR_GS_BASE, false);
9254 vmx_disable_intercept_for_msr(MSR_KERNEL_GS_BASE, true);
9255 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_CS, false);
9256 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_ESP, false);
9257 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_EIP, false);
9258 vmx_disable_intercept_for_msr(MSR_IA32_BNDCFGS, true);
9259
9260 memcpy(vmx_msr_bitmap_legacy_x2apic,
9261 vmx_msr_bitmap_legacy, PAGE_SIZE);
9262 memcpy(vmx_msr_bitmap_longmode_x2apic,
9263 vmx_msr_bitmap_longmode, PAGE_SIZE);
9264
9265 if (enable_apicv) {
9266 for (msr = 0x800; msr <= 0x8ff; msr++)
9267 vmx_disable_intercept_msr_read_x2apic(msr);
9268
9269 /* According SDM, in x2apic mode, the whole id reg is used.
9270 * But in KVM, it only use the highest eight bits. Need to
9271 * intercept it */
9272 vmx_enable_intercept_msr_read_x2apic(0x802);
9273 /* TMCCT */
9274 vmx_enable_intercept_msr_read_x2apic(0x839);
9275 /* TPR */
9276 vmx_disable_intercept_msr_write_x2apic(0x808);
9277 /* EOI */
9278 vmx_disable_intercept_msr_write_x2apic(0x80b);
9279 /* SELF-IPI */
9280 vmx_disable_intercept_msr_write_x2apic(0x83f);
9281 }
9282
9283 if (enable_ept) {
9284 kvm_mmu_set_mask_ptes(0ull,
9285 (enable_ept_ad_bits) ? VMX_EPT_ACCESS_BIT : 0ull,
9286 (enable_ept_ad_bits) ? VMX_EPT_DIRTY_BIT : 0ull,
9287 0ull, VMX_EPT_EXECUTABLE_MASK);
9288 ept_set_mmio_spte_mask();
9289 kvm_enable_tdp();
9290 } else
9291 kvm_disable_tdp();
9292
9293 update_ple_window_actual_max();
9294
9295 return 0;
9296
9297out7:
9298 free_page((unsigned long)vmx_vmwrite_bitmap);
9299out6:
9300 free_page((unsigned long)vmx_vmread_bitmap);
9301out5:
9302 free_page((unsigned long)vmx_msr_bitmap_longmode_x2apic);
9303out4:
9304 free_page((unsigned long)vmx_msr_bitmap_longmode);
9305out3:
9306 free_page((unsigned long)vmx_msr_bitmap_legacy_x2apic);
9307out2:
9308 free_page((unsigned long)vmx_msr_bitmap_legacy);
9309out1:
9310 free_page((unsigned long)vmx_io_bitmap_b);
9311out:
9312 free_page((unsigned long)vmx_io_bitmap_a);
9313 return r;
9314}
9315
9316static void __exit vmx_exit(void)
9317{
9318 free_page((unsigned long)vmx_msr_bitmap_legacy_x2apic);
9319 free_page((unsigned long)vmx_msr_bitmap_longmode_x2apic);
9320 free_page((unsigned long)vmx_msr_bitmap_legacy);
9321 free_page((unsigned long)vmx_msr_bitmap_longmode);
9322 free_page((unsigned long)vmx_io_bitmap_b);
9323 free_page((unsigned long)vmx_io_bitmap_a);
9324 free_page((unsigned long)vmx_vmwrite_bitmap);
9325 free_page((unsigned long)vmx_vmread_bitmap);
9326
9327#ifdef CONFIG_KEXEC
9328 RCU_INIT_POINTER(crash_vmclear_loaded_vmcss, NULL);
9329 synchronize_rcu();
9330#endif
9331
9332 kvm_exit();
9333}
9334
9335module_init(vmx_init)
9336module_exit(vmx_exit)
ubuntu-bionic-kernel mirror