projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
KVM: nVMX: Avoid reading VM_EXIT_INTR_ERROR_CODE needlessly on nested exits
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / arch / x86 / kvm / vmx.c
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 "kvm_cache_regs.h"
35 #include "x86.h"
36
37 #include <asm/io.h>
38 #include <asm/desc.h>
39 #include <asm/vmx.h>
40 #include <asm/virtext.h>
41 #include <asm/mce.h>
42 #include <asm/i387.h>
43 #include <asm/xcr.h>
44 #include <asm/perf_event.h>
45 #include <asm/kexec.h>
46
47 #include "trace.h"
48
49 #define __ex(x) __kvm_handle_fault_on_reboot(x)
50 #define __ex_clear(x, reg) \
51 ____kvm_handle_fault_on_reboot(x, "xor " reg " , " reg)
52
53 MODULE_AUTHOR("Qumranet");
54 MODULE_LICENSE("GPL");
55
56 static const struct x86_cpu_id vmx_cpu_id[] = {
57 X86_FEATURE_MATCH(X86_FEATURE_VMX),
58 {}
59 };
60 MODULE_DEVICE_TABLE(x86cpu, vmx_cpu_id);
61
62 static bool __read_mostly enable_vpid = 1;
63 module_param_named(vpid, enable_vpid, bool, 0444);
64
65 static bool __read_mostly flexpriority_enabled = 1;
66 module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
67
68 static bool __read_mostly enable_ept = 1;
69 module_param_named(ept, enable_ept, bool, S_IRUGO);
70
71 static bool __read_mostly enable_unrestricted_guest = 1;
72 module_param_named(unrestricted_guest,
73 enable_unrestricted_guest, bool, S_IRUGO);
74
75 static bool __read_mostly enable_ept_ad_bits = 1;
76 module_param_named(eptad, enable_ept_ad_bits, bool, S_IRUGO);
77
78 static bool __read_mostly emulate_invalid_guest_state = true;
79 module_param(emulate_invalid_guest_state, bool, S_IRUGO);
80
81 static bool __read_mostly vmm_exclusive = 1;
82 module_param(vmm_exclusive, bool, S_IRUGO);
83
84 static bool __read_mostly fasteoi = 1;
85 module_param(fasteoi, bool, S_IRUGO);
86
87 static bool __read_mostly enable_apicv_reg_vid;
88
89 /*
90 * If nested=1, nested virtualization is supported, i.e., guests may use
91 * VMX and be a hypervisor for its own guests. If nested=0, guests may not
92 * use VMX instructions.
93 */
94 static bool __read_mostly nested = 0;
95 module_param(nested, bool, S_IRUGO);
96
97 #define KVM_GUEST_CR0_MASK (X86_CR0_NW | X86_CR0_CD)
98 #define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST (X86_CR0_WP | X86_CR0_NE)
99 #define KVM_VM_CR0_ALWAYS_ON \
100 (KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
101 #define KVM_CR4_GUEST_OWNED_BITS \
102 (X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR \
103 | X86_CR4_OSXMMEXCPT)
104
105 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
106 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
107
108 #define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
109
110 /*
111 * These 2 parameters are used to config the controls for Pause-Loop Exiting:
112 * ple_gap: upper bound on the amount of time between two successive
113 * executions of PAUSE in a loop. Also indicate if ple enabled.
114 * According to test, this time is usually smaller than 128 cycles.
115 * ple_window: upper bound on the amount of time a guest is allowed to execute
116 * in a PAUSE loop. Tests indicate that most spinlocks are held for
117 * less than 2^12 cycles
118 * Time is measured based on a counter that runs at the same rate as the TSC,
119 * refer SDM volume 3b section 21.6.13 & 22.1.3.
120 */
121 #define KVM_VMX_DEFAULT_PLE_GAP 128
122 #define KVM_VMX_DEFAULT_PLE_WINDOW 4096
123 static int ple_gap = KVM_VMX_DEFAULT_PLE_GAP;
124 module_param(ple_gap, int, S_IRUGO);
125
126 static int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
127 module_param(ple_window, int, S_IRUGO);
128
129 extern const ulong vmx_return;
130
131 #define NR_AUTOLOAD_MSRS 8
132 #define VMCS02_POOL_SIZE 1
133
134 struct vmcs {
135 u32 revision_id;
136 u32 abort;
137 char data[0];
138 };
139
140 /*
141 * Track a VMCS that may be loaded on a certain CPU. If it is (cpu!=-1), also
142 * remember whether it was VMLAUNCHed, and maintain a linked list of all VMCSs
143 * loaded on this CPU (so we can clear them if the CPU goes down).
144 */
145 struct loaded_vmcs {
146 struct vmcs *vmcs;
147 int cpu;
148 int launched;
149 struct list_head loaded_vmcss_on_cpu_link;
150 };
151
152 struct shared_msr_entry {
153 unsigned index;
154 u64 data;
155 u64 mask;
156 };
157
158 /*
159 * struct vmcs12 describes the state that our guest hypervisor (L1) keeps for a
160 * single nested guest (L2), hence the name vmcs12. Any VMX implementation has
161 * a VMCS structure, and vmcs12 is our emulated VMX's VMCS. This structure is
162 * stored in guest memory specified by VMPTRLD, but is opaque to the guest,
163 * which must access it using VMREAD/VMWRITE/VMCLEAR instructions.
164 * More than one of these structures may exist, if L1 runs multiple L2 guests.
165 * nested_vmx_run() will use the data here to build a vmcs02: a VMCS for the
166 * underlying hardware which will be used to run L2.
167 * This structure is packed to ensure that its layout is identical across
168 * machines (necessary for live migration).
169 * If there are changes in this struct, VMCS12_REVISION must be changed.
170 */
171 typedef u64 natural_width;
172 struct __packed vmcs12 {
173 /* According to the Intel spec, a VMCS region must start with the
174 * following two fields. Then follow implementation-specific data.
175 */
176 u32 revision_id;
177 u32 abort;
178
179 u32 launch_state; /* set to 0 by VMCLEAR, to 1 by VMLAUNCH */
180 u32 padding[7]; /* room for future expansion */
181
182 u64 io_bitmap_a;
183 u64 io_bitmap_b;
184 u64 msr_bitmap;
185 u64 vm_exit_msr_store_addr;
186 u64 vm_exit_msr_load_addr;
187 u64 vm_entry_msr_load_addr;
188 u64 tsc_offset;
189 u64 virtual_apic_page_addr;
190 u64 apic_access_addr;
191 u64 ept_pointer;
192 u64 guest_physical_address;
193 u64 vmcs_link_pointer;
194 u64 guest_ia32_debugctl;
195 u64 guest_ia32_pat;
196 u64 guest_ia32_efer;
197 u64 guest_ia32_perf_global_ctrl;
198 u64 guest_pdptr0;
199 u64 guest_pdptr1;
200 u64 guest_pdptr2;
201 u64 guest_pdptr3;
202 u64 host_ia32_pat;
203 u64 host_ia32_efer;
204 u64 host_ia32_perf_global_ctrl;
205 u64 padding64[8]; /* room for future expansion */
206 /*
207 * To allow migration of L1 (complete with its L2 guests) between
208 * machines of different natural widths (32 or 64 bit), we cannot have
209 * unsigned long fields with no explict size. We use u64 (aliased
210 * natural_width) instead. Luckily, x86 is little-endian.
211 */
212 natural_width cr0_guest_host_mask;
213 natural_width cr4_guest_host_mask;
214 natural_width cr0_read_shadow;
215 natural_width cr4_read_shadow;
216 natural_width cr3_target_value0;
217 natural_width cr3_target_value1;
218 natural_width cr3_target_value2;
219 natural_width cr3_target_value3;
220 natural_width exit_qualification;
221 natural_width guest_linear_address;
222 natural_width guest_cr0;
223 natural_width guest_cr3;
224 natural_width guest_cr4;
225 natural_width guest_es_base;
226 natural_width guest_cs_base;
227 natural_width guest_ss_base;
228 natural_width guest_ds_base;
229 natural_width guest_fs_base;
230 natural_width guest_gs_base;
231 natural_width guest_ldtr_base;
232 natural_width guest_tr_base;
233 natural_width guest_gdtr_base;
234 natural_width guest_idtr_base;
235 natural_width guest_dr7;
236 natural_width guest_rsp;
237 natural_width guest_rip;
238 natural_width guest_rflags;
239 natural_width guest_pending_dbg_exceptions;
240 natural_width guest_sysenter_esp;
241 natural_width guest_sysenter_eip;
242 natural_width host_cr0;
243 natural_width host_cr3;
244 natural_width host_cr4;
245 natural_width host_fs_base;
246 natural_width host_gs_base;
247 natural_width host_tr_base;
248 natural_width host_gdtr_base;
249 natural_width host_idtr_base;
250 natural_width host_ia32_sysenter_esp;
251 natural_width host_ia32_sysenter_eip;
252 natural_width host_rsp;
253 natural_width host_rip;
254 natural_width paddingl[8]; /* room for future expansion */
255 u32 pin_based_vm_exec_control;
256 u32 cpu_based_vm_exec_control;
257 u32 exception_bitmap;
258 u32 page_fault_error_code_mask;
259 u32 page_fault_error_code_match;
260 u32 cr3_target_count;
261 u32 vm_exit_controls;
262 u32 vm_exit_msr_store_count;
263 u32 vm_exit_msr_load_count;
264 u32 vm_entry_controls;
265 u32 vm_entry_msr_load_count;
266 u32 vm_entry_intr_info_field;
267 u32 vm_entry_exception_error_code;
268 u32 vm_entry_instruction_len;
269 u32 tpr_threshold;
270 u32 secondary_vm_exec_control;
271 u32 vm_instruction_error;
272 u32 vm_exit_reason;
273 u32 vm_exit_intr_info;
274 u32 vm_exit_intr_error_code;
275 u32 idt_vectoring_info_field;
276 u32 idt_vectoring_error_code;
277 u32 vm_exit_instruction_len;
278 u32 vmx_instruction_info;
279 u32 guest_es_limit;
280 u32 guest_cs_limit;
281 u32 guest_ss_limit;
282 u32 guest_ds_limit;
283 u32 guest_fs_limit;
284 u32 guest_gs_limit;
285 u32 guest_ldtr_limit;
286 u32 guest_tr_limit;
287 u32 guest_gdtr_limit;
288 u32 guest_idtr_limit;
289 u32 guest_es_ar_bytes;
290 u32 guest_cs_ar_bytes;
291 u32 guest_ss_ar_bytes;
292 u32 guest_ds_ar_bytes;
293 u32 guest_fs_ar_bytes;
294 u32 guest_gs_ar_bytes;
295 u32 guest_ldtr_ar_bytes;
296 u32 guest_tr_ar_bytes;
297 u32 guest_interruptibility_info;
298 u32 guest_activity_state;
299 u32 guest_sysenter_cs;
300 u32 host_ia32_sysenter_cs;
301 u32 vmx_preemption_timer_value;
302 u32 padding32[7]; /* room for future expansion */
303 u16 virtual_processor_id;
304 u16 guest_es_selector;
305 u16 guest_cs_selector;
306 u16 guest_ss_selector;
307 u16 guest_ds_selector;
308 u16 guest_fs_selector;
309 u16 guest_gs_selector;
310 u16 guest_ldtr_selector;
311 u16 guest_tr_selector;
312 u16 host_es_selector;
313 u16 host_cs_selector;
314 u16 host_ss_selector;
315 u16 host_ds_selector;
316 u16 host_fs_selector;
317 u16 host_gs_selector;
318 u16 host_tr_selector;
319 };
320
321 /*
322 * VMCS12_REVISION is an arbitrary id that should be changed if the content or
323 * layout of struct vmcs12 is changed. MSR_IA32_VMX_BASIC returns this id, and
324 * VMPTRLD verifies that the VMCS region that L1 is loading contains this id.
325 */
326 #define VMCS12_REVISION 0x11e57ed0
327
328 /*
329 * VMCS12_SIZE is the number of bytes L1 should allocate for the VMXON region
330 * and any VMCS region. Although only sizeof(struct vmcs12) are used by the
331 * current implementation, 4K are reserved to avoid future complications.
332 */
333 #define VMCS12_SIZE 0x1000
334
335 /* Used to remember the last vmcs02 used for some recently used vmcs12s */
336 struct vmcs02_list {
337 struct list_head list;
338 gpa_t vmptr;
339 struct loaded_vmcs vmcs02;
340 };
341
342 /*
343 * The nested_vmx structure is part of vcpu_vmx, and holds information we need
344 * for correct emulation of VMX (i.e., nested VMX) on this vcpu.
345 */
346 struct nested_vmx {
347 /* Has the level1 guest done vmxon? */
348 bool vmxon;
349
350 /* The guest-physical address of the current VMCS L1 keeps for L2 */
351 gpa_t current_vmptr;
352 /* The host-usable pointer to the above */
353 struct page *current_vmcs12_page;
354 struct vmcs12 *current_vmcs12;
355
356 /* vmcs02_list cache of VMCSs recently used to run L2 guests */
357 struct list_head vmcs02_pool;
358 int vmcs02_num;
359 u64 vmcs01_tsc_offset;
360 /* L2 must run next, and mustn't decide to exit to L1. */
361 bool nested_run_pending;
362 /*
363 * Guest pages referred to in vmcs02 with host-physical pointers, so
364 * we must keep them pinned while L2 runs.
365 */
366 struct page *apic_access_page;
367 };
368
369 struct vcpu_vmx {
370 struct kvm_vcpu vcpu;
371 unsigned long host_rsp;
372 u8 fail;
373 u8 cpl;
374 bool nmi_known_unmasked;
375 u32 exit_intr_info;
376 u32 idt_vectoring_info;
377 ulong rflags;
378 struct shared_msr_entry *guest_msrs;
379 int nmsrs;
380 int save_nmsrs;
381 #ifdef CONFIG_X86_64
382 u64 msr_host_kernel_gs_base;
383 u64 msr_guest_kernel_gs_base;
384 #endif
385 /*
386 * loaded_vmcs points to the VMCS currently used in this vcpu. For a
387 * non-nested (L1) guest, it always points to vmcs01. For a nested
388 * guest (L2), it points to a different VMCS.
389 */
390 struct loaded_vmcs vmcs01;
391 struct loaded_vmcs *loaded_vmcs;
392 bool __launched; /* temporary, used in vmx_vcpu_run */
393 struct msr_autoload {
394 unsigned nr;
395 struct vmx_msr_entry guest[NR_AUTOLOAD_MSRS];
396 struct vmx_msr_entry host[NR_AUTOLOAD_MSRS];
397 } msr_autoload;
398 struct {
399 int loaded;
400 u16 fs_sel, gs_sel, ldt_sel;
401 #ifdef CONFIG_X86_64
402 u16 ds_sel, es_sel;
403 #endif
404 int gs_ldt_reload_needed;
405 int fs_reload_needed;
406 } host_state;
407 struct {
408 int vm86_active;
409 ulong save_rflags;
410 struct kvm_segment segs[8];
411 } rmode;
412 struct {
413 u32 bitmask; /* 4 bits per segment (1 bit per field) */
414 struct kvm_save_segment {
415 u16 selector;
416 unsigned long base;
417 u32 limit;
418 u32 ar;
419 } seg[8];
420 } segment_cache;
421 int vpid;
422 bool emulation_required;
423
424 /* Support for vnmi-less CPUs */
425 int soft_vnmi_blocked;
426 ktime_t entry_time;
427 s64 vnmi_blocked_time;
428 u32 exit_reason;
429
430 bool rdtscp_enabled;
431
432 /* Support for a guest hypervisor (nested VMX) */
433 struct nested_vmx nested;
434 };
435
436 enum segment_cache_field {
437 SEG_FIELD_SEL = 0,
438 SEG_FIELD_BASE = 1,
439 SEG_FIELD_LIMIT = 2,
440 SEG_FIELD_AR = 3,
441
442 SEG_FIELD_NR = 4
443 };
444
445 static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
446 {
447 return container_of(vcpu, struct vcpu_vmx, vcpu);
448 }
449
450 #define VMCS12_OFFSET(x) offsetof(struct vmcs12, x)
451 #define FIELD(number, name) [number] = VMCS12_OFFSET(name)
452 #define FIELD64(number, name) [number] = VMCS12_OFFSET(name), \
453 [number##_HIGH] = VMCS12_OFFSET(name)+4
454
455 static const unsigned short vmcs_field_to_offset_table[] = {
456 FIELD(VIRTUAL_PROCESSOR_ID, virtual_processor_id),
457 FIELD(GUEST_ES_SELECTOR, guest_es_selector),
458 FIELD(GUEST_CS_SELECTOR, guest_cs_selector),
459 FIELD(GUEST_SS_SELECTOR, guest_ss_selector),
460 FIELD(GUEST_DS_SELECTOR, guest_ds_selector),
461 FIELD(GUEST_FS_SELECTOR, guest_fs_selector),
462 FIELD(GUEST_GS_SELECTOR, guest_gs_selector),
463 FIELD(GUEST_LDTR_SELECTOR, guest_ldtr_selector),
464 FIELD(GUEST_TR_SELECTOR, guest_tr_selector),
465 FIELD(HOST_ES_SELECTOR, host_es_selector),
466 FIELD(HOST_CS_SELECTOR, host_cs_selector),
467 FIELD(HOST_SS_SELECTOR, host_ss_selector),
468 FIELD(HOST_DS_SELECTOR, host_ds_selector),
469 FIELD(HOST_FS_SELECTOR, host_fs_selector),
470 FIELD(HOST_GS_SELECTOR, host_gs_selector),
471 FIELD(HOST_TR_SELECTOR, host_tr_selector),
472 FIELD64(IO_BITMAP_A, io_bitmap_a),
473 FIELD64(IO_BITMAP_B, io_bitmap_b),
474 FIELD64(MSR_BITMAP, msr_bitmap),
475 FIELD64(VM_EXIT_MSR_STORE_ADDR, vm_exit_msr_store_addr),
476 FIELD64(VM_EXIT_MSR_LOAD_ADDR, vm_exit_msr_load_addr),
477 FIELD64(VM_ENTRY_MSR_LOAD_ADDR, vm_entry_msr_load_addr),
478 FIELD64(TSC_OFFSET, tsc_offset),
479 FIELD64(VIRTUAL_APIC_PAGE_ADDR, virtual_apic_page_addr),
480 FIELD64(APIC_ACCESS_ADDR, apic_access_addr),
481 FIELD64(EPT_POINTER, ept_pointer),
482 FIELD64(GUEST_PHYSICAL_ADDRESS, guest_physical_address),
483 FIELD64(VMCS_LINK_POINTER, vmcs_link_pointer),
484 FIELD64(GUEST_IA32_DEBUGCTL, guest_ia32_debugctl),
485 FIELD64(GUEST_IA32_PAT, guest_ia32_pat),
486 FIELD64(GUEST_IA32_EFER, guest_ia32_efer),
487 FIELD64(GUEST_IA32_PERF_GLOBAL_CTRL, guest_ia32_perf_global_ctrl),
488 FIELD64(GUEST_PDPTR0, guest_pdptr0),
489 FIELD64(GUEST_PDPTR1, guest_pdptr1),
490 FIELD64(GUEST_PDPTR2, guest_pdptr2),
491 FIELD64(GUEST_PDPTR3, guest_pdptr3),
492 FIELD64(HOST_IA32_PAT, host_ia32_pat),
493 FIELD64(HOST_IA32_EFER, host_ia32_efer),
494 FIELD64(HOST_IA32_PERF_GLOBAL_CTRL, host_ia32_perf_global_ctrl),
495 FIELD(PIN_BASED_VM_EXEC_CONTROL, pin_based_vm_exec_control),
496 FIELD(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control),
497 FIELD(EXCEPTION_BITMAP, exception_bitmap),
498 FIELD(PAGE_FAULT_ERROR_CODE_MASK, page_fault_error_code_mask),
499 FIELD(PAGE_FAULT_ERROR_CODE_MATCH, page_fault_error_code_match),
500 FIELD(CR3_TARGET_COUNT, cr3_target_count),
501 FIELD(VM_EXIT_CONTROLS, vm_exit_controls),
502 FIELD(VM_EXIT_MSR_STORE_COUNT, vm_exit_msr_store_count),
503 FIELD(VM_EXIT_MSR_LOAD_COUNT, vm_exit_msr_load_count),
504 FIELD(VM_ENTRY_CONTROLS, vm_entry_controls),
505 FIELD(VM_ENTRY_MSR_LOAD_COUNT, vm_entry_msr_load_count),
506 FIELD(VM_ENTRY_INTR_INFO_FIELD, vm_entry_intr_info_field),
507 FIELD(VM_ENTRY_EXCEPTION_ERROR_CODE, vm_entry_exception_error_code),
508 FIELD(VM_ENTRY_INSTRUCTION_LEN, vm_entry_instruction_len),
509 FIELD(TPR_THRESHOLD, tpr_threshold),
510 FIELD(SECONDARY_VM_EXEC_CONTROL, secondary_vm_exec_control),
511 FIELD(VM_INSTRUCTION_ERROR, vm_instruction_error),
512 FIELD(VM_EXIT_REASON, vm_exit_reason),
513 FIELD(VM_EXIT_INTR_INFO, vm_exit_intr_info),
514 FIELD(VM_EXIT_INTR_ERROR_CODE, vm_exit_intr_error_code),
515 FIELD(IDT_VECTORING_INFO_FIELD, idt_vectoring_info_field),
516 FIELD(IDT_VECTORING_ERROR_CODE, idt_vectoring_error_code),
517 FIELD(VM_EXIT_INSTRUCTION_LEN, vm_exit_instruction_len),
518 FIELD(VMX_INSTRUCTION_INFO, vmx_instruction_info),
519 FIELD(GUEST_ES_LIMIT, guest_es_limit),
520 FIELD(GUEST_CS_LIMIT, guest_cs_limit),
521 FIELD(GUEST_SS_LIMIT, guest_ss_limit),
522 FIELD(GUEST_DS_LIMIT, guest_ds_limit),
523 FIELD(GUEST_FS_LIMIT, guest_fs_limit),
524 FIELD(GUEST_GS_LIMIT, guest_gs_limit),
525 FIELD(GUEST_LDTR_LIMIT, guest_ldtr_limit),
526 FIELD(GUEST_TR_LIMIT, guest_tr_limit),
527 FIELD(GUEST_GDTR_LIMIT, guest_gdtr_limit),
528 FIELD(GUEST_IDTR_LIMIT, guest_idtr_limit),
529 FIELD(GUEST_ES_AR_BYTES, guest_es_ar_bytes),
530 FIELD(GUEST_CS_AR_BYTES, guest_cs_ar_bytes),
531 FIELD(GUEST_SS_AR_BYTES, guest_ss_ar_bytes),
532 FIELD(GUEST_DS_AR_BYTES, guest_ds_ar_bytes),
533 FIELD(GUEST_FS_AR_BYTES, guest_fs_ar_bytes),
534 FIELD(GUEST_GS_AR_BYTES, guest_gs_ar_bytes),
535 FIELD(GUEST_LDTR_AR_BYTES, guest_ldtr_ar_bytes),
536 FIELD(GUEST_TR_AR_BYTES, guest_tr_ar_bytes),
537 FIELD(GUEST_INTERRUPTIBILITY_INFO, guest_interruptibility_info),
538 FIELD(GUEST_ACTIVITY_STATE, guest_activity_state),
539 FIELD(GUEST_SYSENTER_CS, guest_sysenter_cs),
540 FIELD(HOST_IA32_SYSENTER_CS, host_ia32_sysenter_cs),
541 FIELD(VMX_PREEMPTION_TIMER_VALUE, vmx_preemption_timer_value),
542 FIELD(CR0_GUEST_HOST_MASK, cr0_guest_host_mask),
543 FIELD(CR4_GUEST_HOST_MASK, cr4_guest_host_mask),
544 FIELD(CR0_READ_SHADOW, cr0_read_shadow),
545 FIELD(CR4_READ_SHADOW, cr4_read_shadow),
546 FIELD(CR3_TARGET_VALUE0, cr3_target_value0),
547 FIELD(CR3_TARGET_VALUE1, cr3_target_value1),
548 FIELD(CR3_TARGET_VALUE2, cr3_target_value2),
549 FIELD(CR3_TARGET_VALUE3, cr3_target_value3),
550 FIELD(EXIT_QUALIFICATION, exit_qualification),
551 FIELD(GUEST_LINEAR_ADDRESS, guest_linear_address),
552 FIELD(GUEST_CR0, guest_cr0),
553 FIELD(GUEST_CR3, guest_cr3),
554 FIELD(GUEST_CR4, guest_cr4),
555 FIELD(GUEST_ES_BASE, guest_es_base),
556 FIELD(GUEST_CS_BASE, guest_cs_base),
557 FIELD(GUEST_SS_BASE, guest_ss_base),
558 FIELD(GUEST_DS_BASE, guest_ds_base),
559 FIELD(GUEST_FS_BASE, guest_fs_base),
560 FIELD(GUEST_GS_BASE, guest_gs_base),
561 FIELD(GUEST_LDTR_BASE, guest_ldtr_base),
562 FIELD(GUEST_TR_BASE, guest_tr_base),
563 FIELD(GUEST_GDTR_BASE, guest_gdtr_base),
564 FIELD(GUEST_IDTR_BASE, guest_idtr_base),
565 FIELD(GUEST_DR7, guest_dr7),
566 FIELD(GUEST_RSP, guest_rsp),
567 FIELD(GUEST_RIP, guest_rip),
568 FIELD(GUEST_RFLAGS, guest_rflags),
569 FIELD(GUEST_PENDING_DBG_EXCEPTIONS, guest_pending_dbg_exceptions),
570 FIELD(GUEST_SYSENTER_ESP, guest_sysenter_esp),
571 FIELD(GUEST_SYSENTER_EIP, guest_sysenter_eip),
572 FIELD(HOST_CR0, host_cr0),
573 FIELD(HOST_CR3, host_cr3),
574 FIELD(HOST_CR4, host_cr4),
575 FIELD(HOST_FS_BASE, host_fs_base),
576 FIELD(HOST_GS_BASE, host_gs_base),
577 FIELD(HOST_TR_BASE, host_tr_base),
578 FIELD(HOST_GDTR_BASE, host_gdtr_base),
579 FIELD(HOST_IDTR_BASE, host_idtr_base),
580 FIELD(HOST_IA32_SYSENTER_ESP, host_ia32_sysenter_esp),
581 FIELD(HOST_IA32_SYSENTER_EIP, host_ia32_sysenter_eip),
582 FIELD(HOST_RSP, host_rsp),
583 FIELD(HOST_RIP, host_rip),
584 };
585 static const int max_vmcs_field = ARRAY_SIZE(vmcs_field_to_offset_table);
586
587 static inline short vmcs_field_to_offset(unsigned long field)
588 {
589 if (field >= max_vmcs_field || vmcs_field_to_offset_table[field] == 0)
590 return -1;
591 return vmcs_field_to_offset_table[field];
592 }
593
594 static inline struct vmcs12 *get_vmcs12(struct kvm_vcpu *vcpu)
595 {
596 return to_vmx(vcpu)->nested.current_vmcs12;
597 }
598
599 static struct page *nested_get_page(struct kvm_vcpu *vcpu, gpa_t addr)
600 {
601 struct page *page = gfn_to_page(vcpu->kvm, addr >> PAGE_SHIFT);
602 if (is_error_page(page))
603 return NULL;
604
605 return page;
606 }
607
608 static void nested_release_page(struct page *page)
609 {
610 kvm_release_page_dirty(page);
611 }
612
613 static void nested_release_page_clean(struct page *page)
614 {
615 kvm_release_page_clean(page);
616 }
617
618 static u64 construct_eptp(unsigned long root_hpa);
619 static void kvm_cpu_vmxon(u64 addr);
620 static void kvm_cpu_vmxoff(void);
621 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3);
622 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr);
623 static void vmx_set_segment(struct kvm_vcpu *vcpu,
624 struct kvm_segment *var, int seg);
625 static void vmx_get_segment(struct kvm_vcpu *vcpu,
626 struct kvm_segment *var, int seg);
627 static bool guest_state_valid(struct kvm_vcpu *vcpu);
628 static u32 vmx_segment_access_rights(struct kvm_segment *var);
629
630 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
631 static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
632 /*
633 * We maintain a per-CPU linked-list of VMCS loaded on that CPU. This is needed
634 * when a CPU is brought down, and we need to VMCLEAR all VMCSs loaded on it.
635 */
636 static DEFINE_PER_CPU(struct list_head, loaded_vmcss_on_cpu);
637 static DEFINE_PER_CPU(struct desc_ptr, host_gdt);
638
639 static unsigned long *vmx_io_bitmap_a;
640 static unsigned long *vmx_io_bitmap_b;
641 static unsigned long *vmx_msr_bitmap_legacy;
642 static unsigned long *vmx_msr_bitmap_longmode;
643 static unsigned long *vmx_msr_bitmap_legacy_x2apic;
644 static unsigned long *vmx_msr_bitmap_longmode_x2apic;
645
646 static bool cpu_has_load_ia32_efer;
647 static bool cpu_has_load_perf_global_ctrl;
648
649 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
650 static DEFINE_SPINLOCK(vmx_vpid_lock);
651
652 static struct vmcs_config {
653 int size;
654 int order;
655 u32 revision_id;
656 u32 pin_based_exec_ctrl;
657 u32 cpu_based_exec_ctrl;
658 u32 cpu_based_2nd_exec_ctrl;
659 u32 vmexit_ctrl;
660 u32 vmentry_ctrl;
661 } vmcs_config;
662
663 static struct vmx_capability {
664 u32 ept;
665 u32 vpid;
666 } vmx_capability;
667
668 #define VMX_SEGMENT_FIELD(seg) \
669 [VCPU_SREG_##seg] = { \
670 .selector = GUEST_##seg##_SELECTOR, \
671 .base = GUEST_##seg##_BASE, \
672 .limit = GUEST_##seg##_LIMIT, \
673 .ar_bytes = GUEST_##seg##_AR_BYTES, \
674 }
675
676 static const struct kvm_vmx_segment_field {
677 unsigned selector;
678 unsigned base;
679 unsigned limit;
680 unsigned ar_bytes;
681 } kvm_vmx_segment_fields[] = {
682 VMX_SEGMENT_FIELD(CS),
683 VMX_SEGMENT_FIELD(DS),
684 VMX_SEGMENT_FIELD(ES),
685 VMX_SEGMENT_FIELD(FS),
686 VMX_SEGMENT_FIELD(GS),
687 VMX_SEGMENT_FIELD(SS),
688 VMX_SEGMENT_FIELD(TR),
689 VMX_SEGMENT_FIELD(LDTR),
690 };
691
692 static u64 host_efer;
693
694 static void ept_save_pdptrs(struct kvm_vcpu *vcpu);
695
696 /*
697 * Keep MSR_STAR at the end, as setup_msrs() will try to optimize it
698 * away by decrementing the array size.
699 */
700 static const u32 vmx_msr_index[] = {
701 #ifdef CONFIG_X86_64
702 MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
703 #endif
704 MSR_EFER, MSR_TSC_AUX, MSR_STAR,
705 };
706 #define NR_VMX_MSR ARRAY_SIZE(vmx_msr_index)
707
708 static inline bool is_page_fault(u32 intr_info)
709 {
710 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
711 INTR_INFO_VALID_MASK)) ==
712 (INTR_TYPE_HARD_EXCEPTION | PF_VECTOR | INTR_INFO_VALID_MASK);
713 }
714
715 static inline bool is_no_device(u32 intr_info)
716 {
717 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
718 INTR_INFO_VALID_MASK)) ==
719 (INTR_TYPE_HARD_EXCEPTION | NM_VECTOR | INTR_INFO_VALID_MASK);
720 }
721
722 static inline bool is_invalid_opcode(u32 intr_info)
723 {
724 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
725 INTR_INFO_VALID_MASK)) ==
726 (INTR_TYPE_HARD_EXCEPTION | UD_VECTOR | INTR_INFO_VALID_MASK);
727 }
728
729 static inline bool is_external_interrupt(u32 intr_info)
730 {
731 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
732 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
733 }
734
735 static inline bool is_machine_check(u32 intr_info)
736 {
737 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
738 INTR_INFO_VALID_MASK)) ==
739 (INTR_TYPE_HARD_EXCEPTION | MC_VECTOR | INTR_INFO_VALID_MASK);
740 }
741
742 static inline bool cpu_has_vmx_msr_bitmap(void)
743 {
744 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS;
745 }
746
747 static inline bool cpu_has_vmx_tpr_shadow(void)
748 {
749 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW;
750 }
751
752 static inline bool vm_need_tpr_shadow(struct kvm *kvm)
753 {
754 return (cpu_has_vmx_tpr_shadow()) && (irqchip_in_kernel(kvm));
755 }
756
757 static inline bool cpu_has_secondary_exec_ctrls(void)
758 {
759 return vmcs_config.cpu_based_exec_ctrl &
760 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
761 }
762
763 static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
764 {
765 return vmcs_config.cpu_based_2nd_exec_ctrl &
766 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
767 }
768
769 static inline bool cpu_has_vmx_virtualize_x2apic_mode(void)
770 {
771 return vmcs_config.cpu_based_2nd_exec_ctrl &
772 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
773 }
774
775 static inline bool cpu_has_vmx_apic_register_virt(void)
776 {
777 return vmcs_config.cpu_based_2nd_exec_ctrl &
778 SECONDARY_EXEC_APIC_REGISTER_VIRT;
779 }
780
781 static inline bool cpu_has_vmx_virtual_intr_delivery(void)
782 {
783 return vmcs_config.cpu_based_2nd_exec_ctrl &
784 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY;
785 }
786
787 static inline bool cpu_has_vmx_flexpriority(void)
788 {
789 return cpu_has_vmx_tpr_shadow() &&
790 cpu_has_vmx_virtualize_apic_accesses();
791 }
792
793 static inline bool cpu_has_vmx_ept_execute_only(void)
794 {
795 return vmx_capability.ept & VMX_EPT_EXECUTE_ONLY_BIT;
796 }
797
798 static inline bool cpu_has_vmx_eptp_uncacheable(void)
799 {
800 return vmx_capability.ept & VMX_EPTP_UC_BIT;
801 }
802
803 static inline bool cpu_has_vmx_eptp_writeback(void)
804 {
805 return vmx_capability.ept & VMX_EPTP_WB_BIT;
806 }
807
808 static inline bool cpu_has_vmx_ept_2m_page(void)
809 {
810 return vmx_capability.ept & VMX_EPT_2MB_PAGE_BIT;
811 }
812
813 static inline bool cpu_has_vmx_ept_1g_page(void)
814 {
815 return vmx_capability.ept & VMX_EPT_1GB_PAGE_BIT;
816 }
817
818 static inline bool cpu_has_vmx_ept_4levels(void)
819 {
820 return vmx_capability.ept & VMX_EPT_PAGE_WALK_4_BIT;
821 }
822
823 static inline bool cpu_has_vmx_ept_ad_bits(void)
824 {
825 return vmx_capability.ept & VMX_EPT_AD_BIT;
826 }
827
828 static inline bool cpu_has_vmx_invept_context(void)
829 {
830 return vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT;
831 }
832
833 static inline bool cpu_has_vmx_invept_global(void)
834 {
835 return vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT;
836 }
837
838 static inline bool cpu_has_vmx_invvpid_single(void)
839 {
840 return vmx_capability.vpid & VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT;
841 }
842
843 static inline bool cpu_has_vmx_invvpid_global(void)
844 {
845 return vmx_capability.vpid & VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT;
846 }
847
848 static inline bool cpu_has_vmx_ept(void)
849 {
850 return vmcs_config.cpu_based_2nd_exec_ctrl &
851 SECONDARY_EXEC_ENABLE_EPT;
852 }
853
854 static inline bool cpu_has_vmx_unrestricted_guest(void)
855 {
856 return vmcs_config.cpu_based_2nd_exec_ctrl &
857 SECONDARY_EXEC_UNRESTRICTED_GUEST;
858 }
859
860 static inline bool cpu_has_vmx_ple(void)
861 {
862 return vmcs_config.cpu_based_2nd_exec_ctrl &
863 SECONDARY_EXEC_PAUSE_LOOP_EXITING;
864 }
865
866 static inline bool vm_need_virtualize_apic_accesses(struct kvm *kvm)
867 {
868 return flexpriority_enabled && irqchip_in_kernel(kvm);
869 }
870
871 static inline bool cpu_has_vmx_vpid(void)
872 {
873 return vmcs_config.cpu_based_2nd_exec_ctrl &
874 SECONDARY_EXEC_ENABLE_VPID;
875 }
876
877 static inline bool cpu_has_vmx_rdtscp(void)
878 {
879 return vmcs_config.cpu_based_2nd_exec_ctrl &
880 SECONDARY_EXEC_RDTSCP;
881 }
882
883 static inline bool cpu_has_vmx_invpcid(void)
884 {
885 return vmcs_config.cpu_based_2nd_exec_ctrl &
886 SECONDARY_EXEC_ENABLE_INVPCID;
887 }
888
889 static inline bool cpu_has_virtual_nmis(void)
890 {
891 return vmcs_config.pin_based_exec_ctrl & PIN_BASED_VIRTUAL_NMIS;
892 }
893
894 static inline bool cpu_has_vmx_wbinvd_exit(void)
895 {
896 return vmcs_config.cpu_based_2nd_exec_ctrl &
897 SECONDARY_EXEC_WBINVD_EXITING;
898 }
899
900 static inline bool report_flexpriority(void)
901 {
902 return flexpriority_enabled;
903 }
904
905 static inline bool nested_cpu_has(struct vmcs12 *vmcs12, u32 bit)
906 {
907 return vmcs12->cpu_based_vm_exec_control & bit;
908 }
909
910 static inline bool nested_cpu_has2(struct vmcs12 *vmcs12, u32 bit)
911 {
912 return (vmcs12->cpu_based_vm_exec_control &
913 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
914 (vmcs12->secondary_vm_exec_control & bit);
915 }
916
917 static inline bool nested_cpu_has_virtual_nmis(struct vmcs12 *vmcs12,
918 struct kvm_vcpu *vcpu)
919 {
920 return vmcs12->pin_based_vm_exec_control & PIN_BASED_VIRTUAL_NMIS;
921 }
922
923 static inline bool is_exception(u32 intr_info)
924 {
925 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
926 == (INTR_TYPE_HARD_EXCEPTION | INTR_INFO_VALID_MASK);
927 }
928
929 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu);
930 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
931 struct vmcs12 *vmcs12,
932 u32 reason, unsigned long qualification);
933
934 static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
935 {
936 int i;
937
938 for (i = 0; i < vmx->nmsrs; ++i)
939 if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
940 return i;
941 return -1;
942 }
943
944 static inline void __invvpid(int ext, u16 vpid, gva_t gva)
945 {
946 struct {
947 u64 vpid : 16;
948 u64 rsvd : 48;
949 u64 gva;
950 } operand = { vpid, 0, gva };
951
952 asm volatile (__ex(ASM_VMX_INVVPID)
953 /* CF==1 or ZF==1 --> rc = -1 */
954 "; ja 1f ; ud2 ; 1:"
955 : : "a"(&operand), "c"(ext) : "cc", "memory");
956 }
957
958 static inline void __invept(int ext, u64 eptp, gpa_t gpa)
959 {
960 struct {
961 u64 eptp, gpa;
962 } operand = {eptp, gpa};
963
964 asm volatile (__ex(ASM_VMX_INVEPT)
965 /* CF==1 or ZF==1 --> rc = -1 */
966 "; ja 1f ; ud2 ; 1:\n"
967 : : "a" (&operand), "c" (ext) : "cc", "memory");
968 }
969
970 static struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
971 {
972 int i;
973
974 i = __find_msr_index(vmx, msr);
975 if (i >= 0)
976 return &vmx->guest_msrs[i];
977 return NULL;
978 }
979
980 static void vmcs_clear(struct vmcs *vmcs)
981 {
982 u64 phys_addr = __pa(vmcs);
983 u8 error;
984
985 asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0"
986 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
987 : "cc", "memory");
988 if (error)
989 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
990 vmcs, phys_addr);
991 }
992
993 static inline void loaded_vmcs_init(struct loaded_vmcs *loaded_vmcs)
994 {
995 vmcs_clear(loaded_vmcs->vmcs);
996 loaded_vmcs->cpu = -1;
997 loaded_vmcs->launched = 0;
998 }
999
1000 static void vmcs_load(struct vmcs *vmcs)
1001 {
1002 u64 phys_addr = __pa(vmcs);
1003 u8 error;
1004
1005 asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0"
1006 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
1007 : "cc", "memory");
1008 if (error)
1009 printk(KERN_ERR "kvm: vmptrld %p/%llx failed\n",
1010 vmcs, phys_addr);
1011 }
1012
1013 #ifdef CONFIG_KEXEC
1014 /*
1015 * This bitmap is used to indicate whether the vmclear
1016 * operation is enabled on all cpus. All disabled by
1017 * default.
1018 */
1019 static cpumask_t crash_vmclear_enabled_bitmap = CPU_MASK_NONE;
1020
1021 static inline void crash_enable_local_vmclear(int cpu)
1022 {
1023 cpumask_set_cpu(cpu, &crash_vmclear_enabled_bitmap);
1024 }
1025
1026 static inline void crash_disable_local_vmclear(int cpu)
1027 {
1028 cpumask_clear_cpu(cpu, &crash_vmclear_enabled_bitmap);
1029 }
1030
1031 static inline int crash_local_vmclear_enabled(int cpu)
1032 {
1033 return cpumask_test_cpu(cpu, &crash_vmclear_enabled_bitmap);
1034 }
1035
1036 static void crash_vmclear_local_loaded_vmcss(void)
1037 {
1038 int cpu = raw_smp_processor_id();
1039 struct loaded_vmcs *v;
1040
1041 if (!crash_local_vmclear_enabled(cpu))
1042 return;
1043
1044 list_for_each_entry(v, &per_cpu(loaded_vmcss_on_cpu, cpu),
1045 loaded_vmcss_on_cpu_link)
1046 vmcs_clear(v->vmcs);
1047 }
1048 #else
1049 static inline void crash_enable_local_vmclear(int cpu) { }
1050 static inline void crash_disable_local_vmclear(int cpu) { }
1051 #endif /* CONFIG_KEXEC */
1052
1053 static void __loaded_vmcs_clear(void *arg)
1054 {
1055 struct loaded_vmcs *loaded_vmcs = arg;
1056 int cpu = raw_smp_processor_id();
1057
1058 if (loaded_vmcs->cpu != cpu)
1059 return; /* vcpu migration can race with cpu offline */
1060 if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
1061 per_cpu(current_vmcs, cpu) = NULL;
1062 crash_disable_local_vmclear(cpu);
1063 list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
1064
1065 /*
1066 * we should ensure updating loaded_vmcs->loaded_vmcss_on_cpu_link
1067 * is before setting loaded_vmcs->vcpu to -1 which is done in
1068 * loaded_vmcs_init. Otherwise, other cpu can see vcpu = -1 fist
1069 * then adds the vmcs into percpu list before it is deleted.
1070 */
1071 smp_wmb();
1072
1073 loaded_vmcs_init(loaded_vmcs);
1074 crash_enable_local_vmclear(cpu);
1075 }
1076
1077 static void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
1078 {
1079 int cpu = loaded_vmcs->cpu;
1080
1081 if (cpu != -1)
1082 smp_call_function_single(cpu,
1083 __loaded_vmcs_clear, loaded_vmcs, 1);
1084 }
1085
1086 static inline void vpid_sync_vcpu_single(struct vcpu_vmx *vmx)
1087 {
1088 if (vmx->vpid == 0)
1089 return;
1090
1091 if (cpu_has_vmx_invvpid_single())
1092 __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vmx->vpid, 0);
1093 }
1094
1095 static inline void vpid_sync_vcpu_global(void)
1096 {
1097 if (cpu_has_vmx_invvpid_global())
1098 __invvpid(VMX_VPID_EXTENT_ALL_CONTEXT, 0, 0);
1099 }
1100
1101 static inline void vpid_sync_context(struct vcpu_vmx *vmx)
1102 {
1103 if (cpu_has_vmx_invvpid_single())
1104 vpid_sync_vcpu_single(vmx);
1105 else
1106 vpid_sync_vcpu_global();
1107 }
1108
1109 static inline void ept_sync_global(void)
1110 {
1111 if (cpu_has_vmx_invept_global())
1112 __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0);
1113 }
1114
1115 static inline void ept_sync_context(u64 eptp)
1116 {
1117 if (enable_ept) {
1118 if (cpu_has_vmx_invept_context())
1119 __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0);
1120 else
1121 ept_sync_global();
1122 }
1123 }
1124
1125 static __always_inline unsigned long vmcs_readl(unsigned long field)
1126 {
1127 unsigned long value;
1128
1129 asm volatile (__ex_clear(ASM_VMX_VMREAD_RDX_RAX, "%0")
1130 : "=a"(value) : "d"(field) : "cc");
1131 return value;
1132 }
1133
1134 static __always_inline u16 vmcs_read16(unsigned long field)
1135 {
1136 return vmcs_readl(field);
1137 }
1138
1139 static __always_inline u32 vmcs_read32(unsigned long field)
1140 {
1141 return vmcs_readl(field);
1142 }
1143
1144 static __always_inline u64 vmcs_read64(unsigned long field)
1145 {
1146 #ifdef CONFIG_X86_64
1147 return vmcs_readl(field);
1148 #else
1149 return vmcs_readl(field) | ((u64)vmcs_readl(field+1) << 32);
1150 #endif
1151 }
1152
1153 static noinline void vmwrite_error(unsigned long field, unsigned long value)
1154 {
1155 printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
1156 field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
1157 dump_stack();
1158 }
1159
1160 static void vmcs_writel(unsigned long field, unsigned long value)
1161 {
1162 u8 error;
1163
1164 asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0"
1165 : "=q"(error) : "a"(value), "d"(field) : "cc");
1166 if (unlikely(error))
1167 vmwrite_error(field, value);
1168 }
1169
1170 static void vmcs_write16(unsigned long field, u16 value)
1171 {
1172 vmcs_writel(field, value);
1173 }
1174
1175 static void vmcs_write32(unsigned long field, u32 value)
1176 {
1177 vmcs_writel(field, value);
1178 }
1179
1180 static void vmcs_write64(unsigned long field, u64 value)
1181 {
1182 vmcs_writel(field, value);
1183 #ifndef CONFIG_X86_64
1184 asm volatile ("");
1185 vmcs_writel(field+1, value >> 32);
1186 #endif
1187 }
1188
1189 static void vmcs_clear_bits(unsigned long field, u32 mask)
1190 {
1191 vmcs_writel(field, vmcs_readl(field) & ~mask);
1192 }
1193
1194 static void vmcs_set_bits(unsigned long field, u32 mask)
1195 {
1196 vmcs_writel(field, vmcs_readl(field) | mask);
1197 }
1198
1199 static void vmx_segment_cache_clear(struct vcpu_vmx *vmx)
1200 {
1201 vmx->segment_cache.bitmask = 0;
1202 }
1203
1204 static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
1205 unsigned field)
1206 {
1207 bool ret;
1208 u32 mask = 1 << (seg * SEG_FIELD_NR + field);
1209
1210 if (!(vmx->vcpu.arch.regs_avail & (1 << VCPU_EXREG_SEGMENTS))) {
1211 vmx->vcpu.arch.regs_avail |= (1 << VCPU_EXREG_SEGMENTS);
1212 vmx->segment_cache.bitmask = 0;
1213 }
1214 ret = vmx->segment_cache.bitmask & mask;
1215 vmx->segment_cache.bitmask |= mask;
1216 return ret;
1217 }
1218
1219 static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
1220 {
1221 u16 *p = &vmx->segment_cache.seg[seg].selector;
1222
1223 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
1224 *p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
1225 return *p;
1226 }
1227
1228 static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
1229 {
1230 ulong *p = &vmx->segment_cache.seg[seg].base;
1231
1232 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
1233 *p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
1234 return *p;
1235 }
1236
1237 static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
1238 {
1239 u32 *p = &vmx->segment_cache.seg[seg].limit;
1240
1241 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
1242 *p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
1243 return *p;
1244 }
1245
1246 static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
1247 {
1248 u32 *p = &vmx->segment_cache.seg[seg].ar;
1249
1250 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
1251 *p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
1252 return *p;
1253 }
1254
1255 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
1256 {
1257 u32 eb;
1258
1259 eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
1260 (1u << NM_VECTOR) | (1u << DB_VECTOR);
1261 if ((vcpu->guest_debug &
1262 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
1263 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
1264 eb |= 1u << BP_VECTOR;
1265 if (to_vmx(vcpu)->rmode.vm86_active)
1266 eb = ~0;
1267 if (enable_ept)
1268 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
1269 if (vcpu->fpu_active)
1270 eb &= ~(1u << NM_VECTOR);
1271
1272 /* When we are running a nested L2 guest and L1 specified for it a
1273 * certain exception bitmap, we must trap the same exceptions and pass
1274 * them to L1. When running L2, we will only handle the exceptions
1275 * specified above if L1 did not want them.
1276 */
1277 if (is_guest_mode(vcpu))
1278 eb |= get_vmcs12(vcpu)->exception_bitmap;
1279
1280 vmcs_write32(EXCEPTION_BITMAP, eb);
1281 }
1282
1283 static void clear_atomic_switch_msr_special(unsigned long entry,
1284 unsigned long exit)
1285 {
1286 vmcs_clear_bits(VM_ENTRY_CONTROLS, entry);
1287 vmcs_clear_bits(VM_EXIT_CONTROLS, exit);
1288 }
1289
1290 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
1291 {
1292 unsigned i;
1293 struct msr_autoload *m = &vmx->msr_autoload;
1294
1295 switch (msr) {
1296 case MSR_EFER:
1297 if (cpu_has_load_ia32_efer) {
1298 clear_atomic_switch_msr_special(VM_ENTRY_LOAD_IA32_EFER,
1299 VM_EXIT_LOAD_IA32_EFER);
1300 return;
1301 }
1302 break;
1303 case MSR_CORE_PERF_GLOBAL_CTRL:
1304 if (cpu_has_load_perf_global_ctrl) {
1305 clear_atomic_switch_msr_special(
1306 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1307 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
1308 return;
1309 }
1310 break;
1311 }
1312
1313 for (i = 0; i < m->nr; ++i)
1314 if (m->guest[i].index == msr)
1315 break;
1316
1317 if (i == m->nr)
1318 return;
1319 --m->nr;
1320 m->guest[i] = m->guest[m->nr];
1321 m->host[i] = m->host[m->nr];
1322 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1323 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1324 }
1325
1326 static void add_atomic_switch_msr_special(unsigned long entry,
1327 unsigned long exit, unsigned long guest_val_vmcs,
1328 unsigned long host_val_vmcs, u64 guest_val, u64 host_val)
1329 {
1330 vmcs_write64(guest_val_vmcs, guest_val);
1331 vmcs_write64(host_val_vmcs, host_val);
1332 vmcs_set_bits(VM_ENTRY_CONTROLS, entry);
1333 vmcs_set_bits(VM_EXIT_CONTROLS, exit);
1334 }
1335
1336 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
1337 u64 guest_val, u64 host_val)
1338 {
1339 unsigned i;
1340 struct msr_autoload *m = &vmx->msr_autoload;
1341
1342 switch (msr) {
1343 case MSR_EFER:
1344 if (cpu_has_load_ia32_efer) {
1345 add_atomic_switch_msr_special(VM_ENTRY_LOAD_IA32_EFER,
1346 VM_EXIT_LOAD_IA32_EFER,
1347 GUEST_IA32_EFER,
1348 HOST_IA32_EFER,
1349 guest_val, host_val);
1350 return;
1351 }
1352 break;
1353 case MSR_CORE_PERF_GLOBAL_CTRL:
1354 if (cpu_has_load_perf_global_ctrl) {
1355 add_atomic_switch_msr_special(
1356 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1357 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL,
1358 GUEST_IA32_PERF_GLOBAL_CTRL,
1359 HOST_IA32_PERF_GLOBAL_CTRL,
1360 guest_val, host_val);
1361 return;
1362 }
1363 break;
1364 }
1365
1366 for (i = 0; i < m->nr; ++i)
1367 if (m->guest[i].index == msr)
1368 break;
1369
1370 if (i == NR_AUTOLOAD_MSRS) {
1371 printk_once(KERN_WARNING"Not enough mst switch entries. "
1372 "Can't add msr %x\n", msr);
1373 return;
1374 } else if (i == m->nr) {
1375 ++m->nr;
1376 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1377 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1378 }
1379
1380 m->guest[i].index = msr;
1381 m->guest[i].value = guest_val;
1382 m->host[i].index = msr;
1383 m->host[i].value = host_val;
1384 }
1385
1386 static void reload_tss(void)
1387 {
1388 /*
1389 * VT restores TR but not its size. Useless.
1390 */
1391 struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
1392 struct desc_struct *descs;
1393
1394 descs = (void *)gdt->address;
1395 descs[GDT_ENTRY_TSS].type = 9; /* available TSS */
1396 load_TR_desc();
1397 }
1398
1399 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
1400 {
1401 u64 guest_efer;
1402 u64 ignore_bits;
1403
1404 guest_efer = vmx->vcpu.arch.efer;
1405
1406 /*
1407 * NX is emulated; LMA and LME handled by hardware; SCE meaningless
1408 * outside long mode
1409 */
1410 ignore_bits = EFER_NX | EFER_SCE;
1411 #ifdef CONFIG_X86_64
1412 ignore_bits |= EFER_LMA | EFER_LME;
1413 /* SCE is meaningful only in long mode on Intel */
1414 if (guest_efer & EFER_LMA)
1415 ignore_bits &= ~(u64)EFER_SCE;
1416 #endif
1417 guest_efer &= ~ignore_bits;
1418 guest_efer |= host_efer & ignore_bits;
1419 vmx->guest_msrs[efer_offset].data = guest_efer;
1420 vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
1421
1422 clear_atomic_switch_msr(vmx, MSR_EFER);
1423 /* On ept, can't emulate nx, and must switch nx atomically */
1424 if (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX)) {
1425 guest_efer = vmx->vcpu.arch.efer;
1426 if (!(guest_efer & EFER_LMA))
1427 guest_efer &= ~EFER_LME;
1428 add_atomic_switch_msr(vmx, MSR_EFER, guest_efer, host_efer);
1429 return false;
1430 }
1431
1432 return true;
1433 }
1434
1435 static unsigned long segment_base(u16 selector)
1436 {
1437 struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
1438 struct desc_struct *d;
1439 unsigned long table_base;
1440 unsigned long v;
1441
1442 if (!(selector & ~3))
1443 return 0;
1444
1445 table_base = gdt->address;
1446
1447 if (selector & 4) { /* from ldt */
1448 u16 ldt_selector = kvm_read_ldt();
1449
1450 if (!(ldt_selector & ~3))
1451 return 0;
1452
1453 table_base = segment_base(ldt_selector);
1454 }
1455 d = (struct desc_struct *)(table_base + (selector & ~7));
1456 v = get_desc_base(d);
1457 #ifdef CONFIG_X86_64
1458 if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
1459 v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
1460 #endif
1461 return v;
1462 }
1463
1464 static inline unsigned long kvm_read_tr_base(void)
1465 {
1466 u16 tr;
1467 asm("str %0" : "=g"(tr));
1468 return segment_base(tr);
1469 }
1470
1471 static void vmx_save_host_state(struct kvm_vcpu *vcpu)
1472 {
1473 struct vcpu_vmx *vmx = to_vmx(vcpu);
1474 int i;
1475
1476 if (vmx->host_state.loaded)
1477 return;
1478
1479 vmx->host_state.loaded = 1;
1480 /*
1481 * Set host fs and gs selectors. Unfortunately, 22.2.3 does not
1482 * allow segment selectors with cpl > 0 or ti == 1.
1483 */
1484 vmx->host_state.ldt_sel = kvm_read_ldt();
1485 vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
1486 savesegment(fs, vmx->host_state.fs_sel);
1487 if (!(vmx->host_state.fs_sel & 7)) {
1488 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
1489 vmx->host_state.fs_reload_needed = 0;
1490 } else {
1491 vmcs_write16(HOST_FS_SELECTOR, 0);
1492 vmx->host_state.fs_reload_needed = 1;
1493 }
1494 savesegment(gs, vmx->host_state.gs_sel);
1495 if (!(vmx->host_state.gs_sel & 7))
1496 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
1497 else {
1498 vmcs_write16(HOST_GS_SELECTOR, 0);
1499 vmx->host_state.gs_ldt_reload_needed = 1;
1500 }
1501
1502 #ifdef CONFIG_X86_64
1503 savesegment(ds, vmx->host_state.ds_sel);
1504 savesegment(es, vmx->host_state.es_sel);
1505 #endif
1506
1507 #ifdef CONFIG_X86_64
1508 vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
1509 vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
1510 #else
1511 vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
1512 vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
1513 #endif
1514
1515 #ifdef CONFIG_X86_64
1516 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1517 if (is_long_mode(&vmx->vcpu))
1518 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1519 #endif
1520 for (i = 0; i < vmx->save_nmsrs; ++i)
1521 kvm_set_shared_msr(vmx->guest_msrs[i].index,
1522 vmx->guest_msrs[i].data,
1523 vmx->guest_msrs[i].mask);
1524 }
1525
1526 static void __vmx_load_host_state(struct vcpu_vmx *vmx)
1527 {
1528 if (!vmx->host_state.loaded)
1529 return;
1530
1531 ++vmx->vcpu.stat.host_state_reload;
1532 vmx->host_state.loaded = 0;
1533 #ifdef CONFIG_X86_64
1534 if (is_long_mode(&vmx->vcpu))
1535 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1536 #endif
1537 if (vmx->host_state.gs_ldt_reload_needed) {
1538 kvm_load_ldt(vmx->host_state.ldt_sel);
1539 #ifdef CONFIG_X86_64
1540 load_gs_index(vmx->host_state.gs_sel);
1541 #else
1542 loadsegment(gs, vmx->host_state.gs_sel);
1543 #endif
1544 }
1545 if (vmx->host_state.fs_reload_needed)
1546 loadsegment(fs, vmx->host_state.fs_sel);
1547 #ifdef CONFIG_X86_64
1548 if (unlikely(vmx->host_state.ds_sel | vmx->host_state.es_sel)) {
1549 loadsegment(ds, vmx->host_state.ds_sel);
1550 loadsegment(es, vmx->host_state.es_sel);
1551 }
1552 #endif
1553 reload_tss();
1554 #ifdef CONFIG_X86_64
1555 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1556 #endif
1557 /*
1558 * If the FPU is not active (through the host task or
1559 * the guest vcpu), then restore the cr0.TS bit.
1560 */
1561 if (!user_has_fpu() && !vmx->vcpu.guest_fpu_loaded)
1562 stts();
1563 load_gdt(&__get_cpu_var(host_gdt));
1564 }
1565
1566 static void vmx_load_host_state(struct vcpu_vmx *vmx)
1567 {
1568 preempt_disable();
1569 __vmx_load_host_state(vmx);
1570 preempt_enable();
1571 }
1572
1573 /*
1574 * Switches to specified vcpu, until a matching vcpu_put(), but assumes
1575 * vcpu mutex is already taken.
1576 */
1577 static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1578 {
1579 struct vcpu_vmx *vmx = to_vmx(vcpu);
1580 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
1581
1582 if (!vmm_exclusive)
1583 kvm_cpu_vmxon(phys_addr);
1584 else if (vmx->loaded_vmcs->cpu != cpu)
1585 loaded_vmcs_clear(vmx->loaded_vmcs);
1586
1587 if (per_cpu(current_vmcs, cpu) != vmx->loaded_vmcs->vmcs) {
1588 per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
1589 vmcs_load(vmx->loaded_vmcs->vmcs);
1590 }
1591
1592 if (vmx->loaded_vmcs->cpu != cpu) {
1593 struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
1594 unsigned long sysenter_esp;
1595
1596 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1597 local_irq_disable();
1598 crash_disable_local_vmclear(cpu);
1599
1600 /*
1601 * Read loaded_vmcs->cpu should be before fetching
1602 * loaded_vmcs->loaded_vmcss_on_cpu_link.
1603 * See the comments in __loaded_vmcs_clear().
1604 */
1605 smp_rmb();
1606
1607 list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
1608 &per_cpu(loaded_vmcss_on_cpu, cpu));
1609 crash_enable_local_vmclear(cpu);
1610 local_irq_enable();
1611
1612 /*
1613 * Linux uses per-cpu TSS and GDT, so set these when switching
1614 * processors.
1615 */
1616 vmcs_writel(HOST_TR_BASE, kvm_read_tr_base()); /* 22.2.4 */
1617 vmcs_writel(HOST_GDTR_BASE, gdt->address); /* 22.2.4 */
1618
1619 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
1620 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
1621 vmx->loaded_vmcs->cpu = cpu;
1622 }
1623 }
1624
1625 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
1626 {
1627 __vmx_load_host_state(to_vmx(vcpu));
1628 if (!vmm_exclusive) {
1629 __loaded_vmcs_clear(to_vmx(vcpu)->loaded_vmcs);
1630 vcpu->cpu = -1;
1631 kvm_cpu_vmxoff();
1632 }
1633 }
1634
1635 static void vmx_fpu_activate(struct kvm_vcpu *vcpu)
1636 {
1637 ulong cr0;
1638
1639 if (vcpu->fpu_active)
1640 return;
1641 vcpu->fpu_active = 1;
1642 cr0 = vmcs_readl(GUEST_CR0);
1643 cr0 &= ~(X86_CR0_TS | X86_CR0_MP);
1644 cr0 |= kvm_read_cr0_bits(vcpu, X86_CR0_TS | X86_CR0_MP);
1645 vmcs_writel(GUEST_CR0, cr0);
1646 update_exception_bitmap(vcpu);
1647 vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
1648 if (is_guest_mode(vcpu))
1649 vcpu->arch.cr0_guest_owned_bits &=
1650 ~get_vmcs12(vcpu)->cr0_guest_host_mask;
1651 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
1652 }
1653
1654 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu);
1655
1656 /*
1657 * Return the cr0 value that a nested guest would read. This is a combination
1658 * of the real cr0 used to run the guest (guest_cr0), and the bits shadowed by
1659 * its hypervisor (cr0_read_shadow).
1660 */
1661 static inline unsigned long nested_read_cr0(struct vmcs12 *fields)
1662 {
1663 return (fields->guest_cr0 & ~fields->cr0_guest_host_mask) |
1664 (fields->cr0_read_shadow & fields->cr0_guest_host_mask);
1665 }
1666 static inline unsigned long nested_read_cr4(struct vmcs12 *fields)
1667 {
1668 return (fields->guest_cr4 & ~fields->cr4_guest_host_mask) |
1669 (fields->cr4_read_shadow & fields->cr4_guest_host_mask);
1670 }
1671
1672 static void vmx_fpu_deactivate(struct kvm_vcpu *vcpu)
1673 {
1674 /* Note that there is no vcpu->fpu_active = 0 here. The caller must
1675 * set this *before* calling this function.
1676 */
1677 vmx_decache_cr0_guest_bits(vcpu);
1678 vmcs_set_bits(GUEST_CR0, X86_CR0_TS | X86_CR0_MP);
1679 update_exception_bitmap(vcpu);
1680 vcpu->arch.cr0_guest_owned_bits = 0;
1681 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
1682 if (is_guest_mode(vcpu)) {
1683 /*
1684 * L1's specified read shadow might not contain the TS bit,
1685 * so now that we turned on shadowing of this bit, we need to
1686 * set this bit of the shadow. Like in nested_vmx_run we need
1687 * nested_read_cr0(vmcs12), but vmcs12->guest_cr0 is not yet
1688 * up-to-date here because we just decached cr0.TS (and we'll
1689 * only update vmcs12->guest_cr0 on nested exit).
1690 */
1691 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1692 vmcs12->guest_cr0 = (vmcs12->guest_cr0 & ~X86_CR0_TS) |
1693 (vcpu->arch.cr0 & X86_CR0_TS);
1694 vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
1695 } else
1696 vmcs_writel(CR0_READ_SHADOW, vcpu->arch.cr0);
1697 }
1698
1699 static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
1700 {
1701 unsigned long rflags, save_rflags;
1702
1703 if (!test_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail)) {
1704 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
1705 rflags = vmcs_readl(GUEST_RFLAGS);
1706 if (to_vmx(vcpu)->rmode.vm86_active) {
1707 rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
1708 save_rflags = to_vmx(vcpu)->rmode.save_rflags;
1709 rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
1710 }
1711 to_vmx(vcpu)->rflags = rflags;
1712 }
1713 return to_vmx(vcpu)->rflags;
1714 }
1715
1716 static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1717 {
1718 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
1719 to_vmx(vcpu)->rflags = rflags;
1720 if (to_vmx(vcpu)->rmode.vm86_active) {
1721 to_vmx(vcpu)->rmode.save_rflags = rflags;
1722 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1723 }
1724 vmcs_writel(GUEST_RFLAGS, rflags);
1725 }
1726
1727 static u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1728 {
1729 u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1730 int ret = 0;
1731
1732 if (interruptibility & GUEST_INTR_STATE_STI)
1733 ret |= KVM_X86_SHADOW_INT_STI;
1734 if (interruptibility & GUEST_INTR_STATE_MOV_SS)
1735 ret |= KVM_X86_SHADOW_INT_MOV_SS;
1736
1737 return ret & mask;
1738 }
1739
1740 static void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1741 {
1742 u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1743 u32 interruptibility = interruptibility_old;
1744
1745 interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
1746
1747 if (mask & KVM_X86_SHADOW_INT_MOV_SS)
1748 interruptibility |= GUEST_INTR_STATE_MOV_SS;
1749 else if (mask & KVM_X86_SHADOW_INT_STI)
1750 interruptibility |= GUEST_INTR_STATE_STI;
1751
1752 if ((interruptibility != interruptibility_old))
1753 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
1754 }
1755
1756 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
1757 {
1758 unsigned long rip;
1759
1760 rip = kvm_rip_read(vcpu);
1761 rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
1762 kvm_rip_write(vcpu, rip);
1763
1764 /* skipping an emulated instruction also counts */
1765 vmx_set_interrupt_shadow(vcpu, 0);
1766 }
1767
1768 /*
1769 * KVM wants to inject page-faults which it got to the guest. This function
1770 * checks whether in a nested guest, we need to inject them to L1 or L2.
1771 * This function assumes it is called with the exit reason in vmcs02 being
1772 * a #PF exception (this is the only case in which KVM injects a #PF when L2
1773 * is running).
1774 */
1775 static int nested_pf_handled(struct kvm_vcpu *vcpu)
1776 {
1777 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1778
1779 /* TODO: also check PFEC_MATCH/MASK, not just EB.PF. */
1780 if (!(vmcs12->exception_bitmap & (1u << PF_VECTOR)))
1781 return 0;
1782
1783 nested_vmx_vmexit(vcpu);
1784 return 1;
1785 }
1786
1787 static void vmx_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
1788 bool has_error_code, u32 error_code,
1789 bool reinject)
1790 {
1791 struct vcpu_vmx *vmx = to_vmx(vcpu);
1792 u32 intr_info = nr | INTR_INFO_VALID_MASK;
1793
1794 if (nr == PF_VECTOR && is_guest_mode(vcpu) &&
1795 nested_pf_handled(vcpu))
1796 return;
1797
1798 if (has_error_code) {
1799 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
1800 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
1801 }
1802
1803 if (vmx->rmode.vm86_active) {
1804 int inc_eip = 0;
1805 if (kvm_exception_is_soft(nr))
1806 inc_eip = vcpu->arch.event_exit_inst_len;
1807 if (kvm_inject_realmode_interrupt(vcpu, nr, inc_eip) != EMULATE_DONE)
1808 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
1809 return;
1810 }
1811
1812 if (kvm_exception_is_soft(nr)) {
1813 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
1814 vmx->vcpu.arch.event_exit_inst_len);
1815 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
1816 } else
1817 intr_info |= INTR_TYPE_HARD_EXCEPTION;
1818
1819 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
1820 }
1821
1822 static bool vmx_rdtscp_supported(void)
1823 {
1824 return cpu_has_vmx_rdtscp();
1825 }
1826
1827 static bool vmx_invpcid_supported(void)
1828 {
1829 return cpu_has_vmx_invpcid() && enable_ept;
1830 }
1831
1832 /*
1833 * Swap MSR entry in host/guest MSR entry array.
1834 */
1835 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
1836 {
1837 struct shared_msr_entry tmp;
1838
1839 tmp = vmx->guest_msrs[to];
1840 vmx->guest_msrs[to] = vmx->guest_msrs[from];
1841 vmx->guest_msrs[from] = tmp;
1842 }
1843
1844 static void vmx_set_msr_bitmap(struct kvm_vcpu *vcpu)
1845 {
1846 unsigned long *msr_bitmap;
1847
1848 if (irqchip_in_kernel(vcpu->kvm) && apic_x2apic_mode(vcpu->arch.apic)) {
1849 if (is_long_mode(vcpu))
1850 msr_bitmap = vmx_msr_bitmap_longmode_x2apic;
1851 else
1852 msr_bitmap = vmx_msr_bitmap_legacy_x2apic;
1853 } else {
1854 if (is_long_mode(vcpu))
1855 msr_bitmap = vmx_msr_bitmap_longmode;
1856 else
1857 msr_bitmap = vmx_msr_bitmap_legacy;
1858 }
1859
1860 vmcs_write64(MSR_BITMAP, __pa(msr_bitmap));
1861 }
1862
1863 /*
1864 * Set up the vmcs to automatically save and restore system
1865 * msrs. Don't touch the 64-bit msrs if the guest is in legacy
1866 * mode, as fiddling with msrs is very expensive.
1867 */
1868 static void setup_msrs(struct vcpu_vmx *vmx)
1869 {
1870 int save_nmsrs, index;
1871
1872 save_nmsrs = 0;
1873 #ifdef CONFIG_X86_64
1874 if (is_long_mode(&vmx->vcpu)) {
1875 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
1876 if (index >= 0)
1877 move_msr_up(vmx, index, save_nmsrs++);
1878 index = __find_msr_index(vmx, MSR_LSTAR);
1879 if (index >= 0)
1880 move_msr_up(vmx, index, save_nmsrs++);
1881 index = __find_msr_index(vmx, MSR_CSTAR);
1882 if (index >= 0)
1883 move_msr_up(vmx, index, save_nmsrs++);
1884 index = __find_msr_index(vmx, MSR_TSC_AUX);
1885 if (index >= 0 && vmx->rdtscp_enabled)
1886 move_msr_up(vmx, index, save_nmsrs++);
1887 /*
1888 * MSR_STAR is only needed on long mode guests, and only
1889 * if efer.sce is enabled.
1890 */
1891 index = __find_msr_index(vmx, MSR_STAR);
1892 if ((index >= 0) && (vmx->vcpu.arch.efer & EFER_SCE))
1893 move_msr_up(vmx, index, save_nmsrs++);
1894 }
1895 #endif
1896 index = __find_msr_index(vmx, MSR_EFER);
1897 if (index >= 0 && update_transition_efer(vmx, index))
1898 move_msr_up(vmx, index, save_nmsrs++);
1899
1900 vmx->save_nmsrs = save_nmsrs;
1901
1902 if (cpu_has_vmx_msr_bitmap())
1903 vmx_set_msr_bitmap(&vmx->vcpu);
1904 }
1905
1906 /*
1907 * reads and returns guest's timestamp counter "register"
1908 * guest_tsc = host_tsc + tsc_offset -- 21.3
1909 */
1910 static u64 guest_read_tsc(void)
1911 {
1912 u64 host_tsc, tsc_offset;
1913
1914 rdtscll(host_tsc);
1915 tsc_offset = vmcs_read64(TSC_OFFSET);
1916 return host_tsc + tsc_offset;
1917 }
1918
1919 /*
1920 * Like guest_read_tsc, but always returns L1's notion of the timestamp
1921 * counter, even if a nested guest (L2) is currently running.
1922 */
1923 u64 vmx_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1924 {
1925 u64 tsc_offset;
1926
1927 tsc_offset = is_guest_mode(vcpu) ?
1928 to_vmx(vcpu)->nested.vmcs01_tsc_offset :
1929 vmcs_read64(TSC_OFFSET);
1930 return host_tsc + tsc_offset;
1931 }
1932
1933 /*
1934 * Engage any workarounds for mis-matched TSC rates. Currently limited to
1935 * software catchup for faster rates on slower CPUs.
1936 */
1937 static void vmx_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1938 {
1939 if (!scale)
1940 return;
1941
1942 if (user_tsc_khz > tsc_khz) {
1943 vcpu->arch.tsc_catchup = 1;
1944 vcpu->arch.tsc_always_catchup = 1;
1945 } else
1946 WARN(1, "user requested TSC rate below hardware speed\n");
1947 }
1948
1949 static u64 vmx_read_tsc_offset(struct kvm_vcpu *vcpu)
1950 {
1951 return vmcs_read64(TSC_OFFSET);
1952 }
1953
1954 /*
1955 * writes 'offset' into guest's timestamp counter offset register
1956 */
1957 static void vmx_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1958 {
1959 if (is_guest_mode(vcpu)) {
1960 /*
1961 * We're here if L1 chose not to trap WRMSR to TSC. According
1962 * to the spec, this should set L1's TSC; The offset that L1
1963 * set for L2 remains unchanged, and still needs to be added
1964 * to the newly set TSC to get L2's TSC.
1965 */
1966 struct vmcs12 *vmcs12;
1967 to_vmx(vcpu)->nested.vmcs01_tsc_offset = offset;
1968 /* recalculate vmcs02.TSC_OFFSET: */
1969 vmcs12 = get_vmcs12(vcpu);
1970 vmcs_write64(TSC_OFFSET, offset +
1971 (nested_cpu_has(vmcs12, CPU_BASED_USE_TSC_OFFSETING) ?
1972 vmcs12->tsc_offset : 0));
1973 } else {
1974 vmcs_write64(TSC_OFFSET, offset);
1975 }
1976 }
1977
1978 static void vmx_adjust_tsc_offset(struct kvm_vcpu *vcpu, s64 adjustment, bool host)
1979 {
1980 u64 offset = vmcs_read64(TSC_OFFSET);
1981 vmcs_write64(TSC_OFFSET, offset + adjustment);
1982 if (is_guest_mode(vcpu)) {
1983 /* Even when running L2, the adjustment needs to apply to L1 */
1984 to_vmx(vcpu)->nested.vmcs01_tsc_offset += adjustment;
1985 }
1986 }
1987
1988 static u64 vmx_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1989 {
1990 return target_tsc - native_read_tsc();
1991 }
1992
1993 static bool guest_cpuid_has_vmx(struct kvm_vcpu *vcpu)
1994 {
1995 struct kvm_cpuid_entry2 *best = kvm_find_cpuid_entry(vcpu, 1, 0);
1996 return best && (best->ecx & (1 << (X86_FEATURE_VMX & 31)));
1997 }
1998
1999 /*
2000 * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
2001 * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
2002 * all guests if the "nested" module option is off, and can also be disabled
2003 * for a single guest by disabling its VMX cpuid bit.
2004 */
2005 static inline bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
2006 {
2007 return nested && guest_cpuid_has_vmx(vcpu);
2008 }
2009
2010 /*
2011 * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be
2012 * returned for the various VMX controls MSRs when nested VMX is enabled.
2013 * The same values should also be used to verify that vmcs12 control fields are
2014 * valid during nested entry from L1 to L2.
2015 * Each of these control msrs has a low and high 32-bit half: A low bit is on
2016 * if the corresponding bit in the (32-bit) control field *must* be on, and a
2017 * bit in the high half is on if the corresponding bit in the control field
2018 * may be on. See also vmx_control_verify().
2019 * TODO: allow these variables to be modified (downgraded) by module options
2020 * or other means.
2021 */
2022 static u32 nested_vmx_procbased_ctls_low, nested_vmx_procbased_ctls_high;
2023 static u32 nested_vmx_secondary_ctls_low, nested_vmx_secondary_ctls_high;
2024 static u32 nested_vmx_pinbased_ctls_low, nested_vmx_pinbased_ctls_high;
2025 static u32 nested_vmx_exit_ctls_low, nested_vmx_exit_ctls_high;
2026 static u32 nested_vmx_entry_ctls_low, nested_vmx_entry_ctls_high;
2027 static u32 nested_vmx_misc_low, nested_vmx_misc_high;
2028 static __init void nested_vmx_setup_ctls_msrs(void)
2029 {
2030 /*
2031 * Note that as a general rule, the high half of the MSRs (bits in
2032 * the control fields which may be 1) should be initialized by the
2033 * intersection of the underlying hardware's MSR (i.e., features which
2034 * can be supported) and the list of features we want to expose -
2035 * because they are known to be properly supported in our code.
2036 * Also, usually, the low half of the MSRs (bits which must be 1) can
2037 * be set to 0, meaning that L1 may turn off any of these bits. The
2038 * reason is that if one of these bits is necessary, it will appear
2039 * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control
2040 * fields of vmcs01 and vmcs02, will turn these bits off - and
2041 * nested_vmx_exit_handled() will not pass related exits to L1.
2042 * These rules have exceptions below.
2043 */
2044
2045 /* pin-based controls */
2046 rdmsr(MSR_IA32_VMX_PINBASED_CTLS,
2047 nested_vmx_pinbased_ctls_low, nested_vmx_pinbased_ctls_high);
2048 /*
2049 * According to the Intel spec, if bit 55 of VMX_BASIC is off (as it is
2050 * in our case), bits 1, 2 and 4 (i.e., 0x16) must be 1 in this MSR.
2051 */
2052 nested_vmx_pinbased_ctls_low |= PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
2053 nested_vmx_pinbased_ctls_high &= PIN_BASED_EXT_INTR_MASK |
2054 PIN_BASED_NMI_EXITING | PIN_BASED_VIRTUAL_NMIS |
2055 PIN_BASED_VMX_PREEMPTION_TIMER;
2056 nested_vmx_pinbased_ctls_high |= PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
2057
2058 /*
2059 * Exit controls
2060 * If bit 55 of VMX_BASIC is off, bits 0-8 and 10, 11, 13, 14, 16 and
2061 * 17 must be 1.
2062 */
2063 nested_vmx_exit_ctls_low = VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
2064 /* Note that guest use of VM_EXIT_ACK_INTR_ON_EXIT is not supported. */
2065 #ifdef CONFIG_X86_64
2066 nested_vmx_exit_ctls_high = VM_EXIT_HOST_ADDR_SPACE_SIZE;
2067 #else
2068 nested_vmx_exit_ctls_high = 0;
2069 #endif
2070 nested_vmx_exit_ctls_high |= VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
2071
2072 /* entry controls */
2073 rdmsr(MSR_IA32_VMX_ENTRY_CTLS,
2074 nested_vmx_entry_ctls_low, nested_vmx_entry_ctls_high);
2075 /* If bit 55 of VMX_BASIC is off, bits 0-8 and 12 must be 1. */
2076 nested_vmx_entry_ctls_low = VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
2077 nested_vmx_entry_ctls_high &=
2078 VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_IA32E_MODE;
2079 nested_vmx_entry_ctls_high |= VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
2080
2081 /* cpu-based controls */
2082 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS,
2083 nested_vmx_procbased_ctls_low, nested_vmx_procbased_ctls_high);
2084 nested_vmx_procbased_ctls_low = 0;
2085 nested_vmx_procbased_ctls_high &=
2086 CPU_BASED_VIRTUAL_INTR_PENDING | CPU_BASED_USE_TSC_OFFSETING |
2087 CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING |
2088 CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING |
2089 CPU_BASED_CR3_STORE_EXITING |
2090 #ifdef CONFIG_X86_64
2091 CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING |
2092 #endif
2093 CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING |
2094 CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_EXITING |
2095 CPU_BASED_RDPMC_EXITING | CPU_BASED_RDTSC_EXITING |
2096 CPU_BASED_PAUSE_EXITING |
2097 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2098 /*
2099 * We can allow some features even when not supported by the
2100 * hardware. For example, L1 can specify an MSR bitmap - and we
2101 * can use it to avoid exits to L1 - even when L0 runs L2
2102 * without MSR bitmaps.
2103 */
2104 nested_vmx_procbased_ctls_high |= CPU_BASED_USE_MSR_BITMAPS;
2105
2106 /* secondary cpu-based controls */
2107 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
2108 nested_vmx_secondary_ctls_low, nested_vmx_secondary_ctls_high);
2109 nested_vmx_secondary_ctls_low = 0;
2110 nested_vmx_secondary_ctls_high &=
2111 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2112 SECONDARY_EXEC_WBINVD_EXITING;
2113
2114 /* miscellaneous data */
2115 rdmsr(MSR_IA32_VMX_MISC, nested_vmx_misc_low, nested_vmx_misc_high);
2116 nested_vmx_misc_low &= VMX_MISC_PREEMPTION_TIMER_RATE_MASK |
2117 VMX_MISC_SAVE_EFER_LMA;
2118 nested_vmx_misc_high = 0;
2119 }
2120
2121 static inline bool vmx_control_verify(u32 control, u32 low, u32 high)
2122 {
2123 /*
2124 * Bits 0 in high must be 0, and bits 1 in low must be 1.
2125 */
2126 return ((control & high) | low) == control;
2127 }
2128
2129 static inline u64 vmx_control_msr(u32 low, u32 high)
2130 {
2131 return low | ((u64)high << 32);
2132 }
2133
2134 /*
2135 * If we allow our guest to use VMX instructions (i.e., nested VMX), we should
2136 * also let it use VMX-specific MSRs.
2137 * vmx_get_vmx_msr() and vmx_set_vmx_msr() return 1 when we handled a
2138 * VMX-specific MSR, or 0 when we haven't (and the caller should handle it
2139 * like all other MSRs).
2140 */
2141 static int vmx_get_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
2142 {
2143 if (!nested_vmx_allowed(vcpu) && msr_index >= MSR_IA32_VMX_BASIC &&
2144 msr_index <= MSR_IA32_VMX_TRUE_ENTRY_CTLS) {
2145 /*
2146 * According to the spec, processors which do not support VMX
2147 * should throw a #GP(0) when VMX capability MSRs are read.
2148 */
2149 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
2150 return 1;
2151 }
2152
2153 switch (msr_index) {
2154 case MSR_IA32_FEATURE_CONTROL:
2155 *pdata = 0;
2156 break;
2157 case MSR_IA32_VMX_BASIC:
2158 /*
2159 * This MSR reports some information about VMX support. We
2160 * should return information about the VMX we emulate for the
2161 * guest, and the VMCS structure we give it - not about the
2162 * VMX support of the underlying hardware.
2163 */
2164 *pdata = VMCS12_REVISION |
2165 ((u64)VMCS12_SIZE << VMX_BASIC_VMCS_SIZE_SHIFT) |
2166 (VMX_BASIC_MEM_TYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT);
2167 break;
2168 case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
2169 case MSR_IA32_VMX_PINBASED_CTLS:
2170 *pdata = vmx_control_msr(nested_vmx_pinbased_ctls_low,
2171 nested_vmx_pinbased_ctls_high);
2172 break;
2173 case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
2174 case MSR_IA32_VMX_PROCBASED_CTLS:
2175 *pdata = vmx_control_msr(nested_vmx_procbased_ctls_low,
2176 nested_vmx_procbased_ctls_high);
2177 break;
2178 case MSR_IA32_VMX_TRUE_EXIT_CTLS:
2179 case MSR_IA32_VMX_EXIT_CTLS:
2180 *pdata = vmx_control_msr(nested_vmx_exit_ctls_low,
2181 nested_vmx_exit_ctls_high);
2182 break;
2183 case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
2184 case MSR_IA32_VMX_ENTRY_CTLS:
2185 *pdata = vmx_control_msr(nested_vmx_entry_ctls_low,
2186 nested_vmx_entry_ctls_high);
2187 break;
2188 case MSR_IA32_VMX_MISC:
2189 *pdata = vmx_control_msr(nested_vmx_misc_low,
2190 nested_vmx_misc_high);
2191 break;
2192 /*
2193 * These MSRs specify bits which the guest must keep fixed (on or off)
2194 * while L1 is in VMXON mode (in L1's root mode, or running an L2).
2195 * We picked the standard core2 setting.
2196 */
2197 #define VMXON_CR0_ALWAYSON (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
2198 #define VMXON_CR4_ALWAYSON X86_CR4_VMXE
2199 case MSR_IA32_VMX_CR0_FIXED0:
2200 *pdata = VMXON_CR0_ALWAYSON;
2201 break;
2202 case MSR_IA32_VMX_CR0_FIXED1:
2203 *pdata = -1ULL;
2204 break;
2205 case MSR_IA32_VMX_CR4_FIXED0:
2206 *pdata = VMXON_CR4_ALWAYSON;
2207 break;
2208 case MSR_IA32_VMX_CR4_FIXED1:
2209 *pdata = -1ULL;
2210 break;
2211 case MSR_IA32_VMX_VMCS_ENUM:
2212 *pdata = 0x1f;
2213 break;
2214 case MSR_IA32_VMX_PROCBASED_CTLS2:
2215 *pdata = vmx_control_msr(nested_vmx_secondary_ctls_low,
2216 nested_vmx_secondary_ctls_high);
2217 break;
2218 case MSR_IA32_VMX_EPT_VPID_CAP:
2219 /* Currently, no nested ept or nested vpid */
2220 *pdata = 0;
2221 break;
2222 default:
2223 return 0;
2224 }
2225
2226 return 1;
2227 }
2228
2229 static int vmx_set_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
2230 {
2231 if (!nested_vmx_allowed(vcpu))
2232 return 0;
2233
2234 if (msr_index == MSR_IA32_FEATURE_CONTROL)
2235 /* TODO: the right thing. */
2236 return 1;
2237 /*
2238 * No need to treat VMX capability MSRs specially: If we don't handle
2239 * them, handle_wrmsr will #GP(0), which is correct (they are readonly)
2240 */
2241 return 0;
2242 }
2243
2244 /*
2245 * Reads an msr value (of 'msr_index') into 'pdata'.
2246 * Returns 0 on success, non-0 otherwise.
2247 * Assumes vcpu_load() was already called.
2248 */
2249 static int vmx_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
2250 {
2251 u64 data;
2252 struct shared_msr_entry *msr;
2253
2254 if (!pdata) {
2255 printk(KERN_ERR "BUG: get_msr called with NULL pdata\n");
2256 return -EINVAL;
2257 }
2258
2259 switch (msr_index) {
2260 #ifdef CONFIG_X86_64
2261 case MSR_FS_BASE:
2262 data = vmcs_readl(GUEST_FS_BASE);
2263 break;
2264 case MSR_GS_BASE:
2265 data = vmcs_readl(GUEST_GS_BASE);
2266 break;
2267 case MSR_KERNEL_GS_BASE:
2268 vmx_load_host_state(to_vmx(vcpu));
2269 data = to_vmx(vcpu)->msr_guest_kernel_gs_base;
2270 break;
2271 #endif
2272 case MSR_EFER:
2273 return kvm_get_msr_common(vcpu, msr_index, pdata);
2274 case MSR_IA32_TSC:
2275 data = guest_read_tsc();
2276 break;
2277 case MSR_IA32_SYSENTER_CS:
2278 data = vmcs_read32(GUEST_SYSENTER_CS);
2279 break;
2280 case MSR_IA32_SYSENTER_EIP:
2281 data = vmcs_readl(GUEST_SYSENTER_EIP);
2282 break;
2283 case MSR_IA32_SYSENTER_ESP:
2284 data = vmcs_readl(GUEST_SYSENTER_ESP);
2285 break;
2286 case MSR_TSC_AUX:
2287 if (!to_vmx(vcpu)->rdtscp_enabled)
2288 return 1;
2289 /* Otherwise falls through */
2290 default:
2291 if (vmx_get_vmx_msr(vcpu, msr_index, pdata))
2292 return 0;
2293 msr = find_msr_entry(to_vmx(vcpu), msr_index);
2294 if (msr) {
2295 data = msr->data;
2296 break;
2297 }
2298 return kvm_get_msr_common(vcpu, msr_index, pdata);
2299 }
2300
2301 *pdata = data;
2302 return 0;
2303 }
2304
2305 /*
2306 * Writes msr value into into the appropriate "register".
2307 * Returns 0 on success, non-0 otherwise.
2308 * Assumes vcpu_load() was already called.
2309 */
2310 static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2311 {
2312 struct vcpu_vmx *vmx = to_vmx(vcpu);
2313 struct shared_msr_entry *msr;
2314 int ret = 0;
2315 u32 msr_index = msr_info->index;
2316 u64 data = msr_info->data;
2317
2318 switch (msr_index) {
2319 case MSR_EFER:
2320 ret = kvm_set_msr_common(vcpu, msr_info);
2321 break;
2322 #ifdef CONFIG_X86_64
2323 case MSR_FS_BASE:
2324 vmx_segment_cache_clear(vmx);
2325 vmcs_writel(GUEST_FS_BASE, data);
2326 break;
2327 case MSR_GS_BASE:
2328 vmx_segment_cache_clear(vmx);
2329 vmcs_writel(GUEST_GS_BASE, data);
2330 break;
2331 case MSR_KERNEL_GS_BASE:
2332 vmx_load_host_state(vmx);
2333 vmx->msr_guest_kernel_gs_base = data;
2334 break;
2335 #endif
2336 case MSR_IA32_SYSENTER_CS:
2337 vmcs_write32(GUEST_SYSENTER_CS, data);
2338 break;
2339 case MSR_IA32_SYSENTER_EIP:
2340 vmcs_writel(GUEST_SYSENTER_EIP, data);
2341 break;
2342 case MSR_IA32_SYSENTER_ESP:
2343 vmcs_writel(GUEST_SYSENTER_ESP, data);
2344 break;
2345 case MSR_IA32_TSC:
2346 kvm_write_tsc(vcpu, msr_info);
2347 break;
2348 case MSR_IA32_CR_PAT:
2349 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2350 vmcs_write64(GUEST_IA32_PAT, data);
2351 vcpu->arch.pat = data;
2352 break;
2353 }
2354 ret = kvm_set_msr_common(vcpu, msr_info);
2355 break;
2356 case MSR_IA32_TSC_ADJUST:
2357 ret = kvm_set_msr_common(vcpu, msr_info);
2358 break;
2359 case MSR_TSC_AUX:
2360 if (!vmx->rdtscp_enabled)
2361 return 1;
2362 /* Check reserved bit, higher 32 bits should be zero */
2363 if ((data >> 32) != 0)
2364 return 1;
2365 /* Otherwise falls through */
2366 default:
2367 if (vmx_set_vmx_msr(vcpu, msr_index, data))
2368 break;
2369 msr = find_msr_entry(vmx, msr_index);
2370 if (msr) {
2371 msr->data = data;
2372 if (msr - vmx->guest_msrs < vmx->save_nmsrs) {
2373 preempt_disable();
2374 kvm_set_shared_msr(msr->index, msr->data,
2375 msr->mask);
2376 preempt_enable();
2377 }
2378 break;
2379 }
2380 ret = kvm_set_msr_common(vcpu, msr_info);
2381 }
2382
2383 return ret;
2384 }
2385
2386 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
2387 {
2388 __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
2389 switch (reg) {
2390 case VCPU_REGS_RSP:
2391 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
2392 break;
2393 case VCPU_REGS_RIP:
2394 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
2395 break;
2396 case VCPU_EXREG_PDPTR:
2397 if (enable_ept)
2398 ept_save_pdptrs(vcpu);
2399 break;
2400 default:
2401 break;
2402 }
2403 }
2404
2405 static __init int cpu_has_kvm_support(void)
2406 {
2407 return cpu_has_vmx();
2408 }
2409
2410 static __init int vmx_disabled_by_bios(void)
2411 {
2412 u64 msr;
2413
2414 rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
2415 if (msr & FEATURE_CONTROL_LOCKED) {
2416 /* launched w/ TXT and VMX disabled */
2417 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
2418 && tboot_enabled())
2419 return 1;
2420 /* launched w/o TXT and VMX only enabled w/ TXT */
2421 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
2422 && (msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
2423 && !tboot_enabled()) {
2424 printk(KERN_WARNING "kvm: disable TXT in the BIOS or "
2425 "activate TXT before enabling KVM\n");
2426 return 1;
2427 }
2428 /* launched w/o TXT and VMX disabled */
2429 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
2430 && !tboot_enabled())
2431 return 1;
2432 }
2433
2434 return 0;
2435 }
2436
2437 static void kvm_cpu_vmxon(u64 addr)
2438 {
2439 asm volatile (ASM_VMX_VMXON_RAX
2440 : : "a"(&addr), "m"(addr)
2441 : "memory", "cc");
2442 }
2443
2444 static int hardware_enable(void *garbage)
2445 {
2446 int cpu = raw_smp_processor_id();
2447 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
2448 u64 old, test_bits;
2449
2450 if (read_cr4() & X86_CR4_VMXE)
2451 return -EBUSY;
2452
2453 INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
2454
2455 /*
2456 * Now we can enable the vmclear operation in kdump
2457 * since the loaded_vmcss_on_cpu list on this cpu
2458 * has been initialized.
2459 *
2460 * Though the cpu is not in VMX operation now, there
2461 * is no problem to enable the vmclear operation
2462 * for the loaded_vmcss_on_cpu list is empty!
2463 */
2464 crash_enable_local_vmclear(cpu);
2465
2466 rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
2467
2468 test_bits = FEATURE_CONTROL_LOCKED;
2469 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
2470 if (tboot_enabled())
2471 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX;
2472
2473 if ((old & test_bits) != test_bits) {
2474 /* enable and lock */
2475 wrmsrl(MSR_IA32_FEATURE_CONTROL, old | test_bits);
2476 }
2477 write_cr4(read_cr4() | X86_CR4_VMXE); /* FIXME: not cpu hotplug safe */
2478
2479 if (vmm_exclusive) {
2480 kvm_cpu_vmxon(phys_addr);
2481 ept_sync_global();
2482 }
2483
2484 store_gdt(&__get_cpu_var(host_gdt));
2485
2486 return 0;
2487 }
2488
2489 static void vmclear_local_loaded_vmcss(void)
2490 {
2491 int cpu = raw_smp_processor_id();
2492 struct loaded_vmcs *v, *n;
2493
2494 list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
2495 loaded_vmcss_on_cpu_link)
2496 __loaded_vmcs_clear(v);
2497 }
2498
2499
2500 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
2501 * tricks.
2502 */
2503 static void kvm_cpu_vmxoff(void)
2504 {
2505 asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
2506 }
2507
2508 static void hardware_disable(void *garbage)
2509 {
2510 if (vmm_exclusive) {
2511 vmclear_local_loaded_vmcss();
2512 kvm_cpu_vmxoff();
2513 }
2514 write_cr4(read_cr4() & ~X86_CR4_VMXE);
2515 }
2516
2517 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
2518 u32 msr, u32 *result)
2519 {
2520 u32 vmx_msr_low, vmx_msr_high;
2521 u32 ctl = ctl_min | ctl_opt;
2522
2523 rdmsr(msr, vmx_msr_low, vmx_msr_high);
2524
2525 ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
2526 ctl |= vmx_msr_low; /* bit == 1 in low word ==> must be one */
2527
2528 /* Ensure minimum (required) set of control bits are supported. */
2529 if (ctl_min & ~ctl)
2530 return -EIO;
2531
2532 *result = ctl;
2533 return 0;
2534 }
2535
2536 static __init bool allow_1_setting(u32 msr, u32 ctl)
2537 {
2538 u32 vmx_msr_low, vmx_msr_high;
2539
2540 rdmsr(msr, vmx_msr_low, vmx_msr_high);
2541 return vmx_msr_high & ctl;
2542 }
2543
2544 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
2545 {
2546 u32 vmx_msr_low, vmx_msr_high;
2547 u32 min, opt, min2, opt2;
2548 u32 _pin_based_exec_control = 0;
2549 u32 _cpu_based_exec_control = 0;
2550 u32 _cpu_based_2nd_exec_control = 0;
2551 u32 _vmexit_control = 0;
2552 u32 _vmentry_control = 0;
2553
2554 min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
2555 opt = PIN_BASED_VIRTUAL_NMIS;
2556 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
2557 &_pin_based_exec_control) < 0)
2558 return -EIO;
2559
2560 min = CPU_BASED_HLT_EXITING |
2561 #ifdef CONFIG_X86_64
2562 CPU_BASED_CR8_LOAD_EXITING |
2563 CPU_BASED_CR8_STORE_EXITING |
2564 #endif
2565 CPU_BASED_CR3_LOAD_EXITING |
2566 CPU_BASED_CR3_STORE_EXITING |
2567 CPU_BASED_USE_IO_BITMAPS |
2568 CPU_BASED_MOV_DR_EXITING |
2569 CPU_BASED_USE_TSC_OFFSETING |
2570 CPU_BASED_MWAIT_EXITING |
2571 CPU_BASED_MONITOR_EXITING |
2572 CPU_BASED_INVLPG_EXITING |
2573 CPU_BASED_RDPMC_EXITING;
2574
2575 opt = CPU_BASED_TPR_SHADOW |
2576 CPU_BASED_USE_MSR_BITMAPS |
2577 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2578 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
2579 &_cpu_based_exec_control) < 0)
2580 return -EIO;
2581 #ifdef CONFIG_X86_64
2582 if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2583 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
2584 ~CPU_BASED_CR8_STORE_EXITING;
2585 #endif
2586 if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
2587 min2 = 0;
2588 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2589 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2590 SECONDARY_EXEC_WBINVD_EXITING |
2591 SECONDARY_EXEC_ENABLE_VPID |
2592 SECONDARY_EXEC_ENABLE_EPT |
2593 SECONDARY_EXEC_UNRESTRICTED_GUEST |
2594 SECONDARY_EXEC_PAUSE_LOOP_EXITING |
2595 SECONDARY_EXEC_RDTSCP |
2596 SECONDARY_EXEC_ENABLE_INVPCID |
2597 SECONDARY_EXEC_APIC_REGISTER_VIRT |
2598 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY;
2599 if (adjust_vmx_controls(min2, opt2,
2600 MSR_IA32_VMX_PROCBASED_CTLS2,
2601 &_cpu_based_2nd_exec_control) < 0)
2602 return -EIO;
2603 }
2604 #ifndef CONFIG_X86_64
2605 if (!(_cpu_based_2nd_exec_control &
2606 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
2607 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
2608 #endif
2609
2610 if (!(_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2611 _cpu_based_2nd_exec_control &= ~(
2612 SECONDARY_EXEC_APIC_REGISTER_VIRT |
2613 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2614 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
2615
2616 if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
2617 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
2618 enabled */
2619 _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
2620 CPU_BASED_CR3_STORE_EXITING |
2621 CPU_BASED_INVLPG_EXITING);
2622 rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
2623 vmx_capability.ept, vmx_capability.vpid);
2624 }
2625
2626 min = 0;
2627 #ifdef CONFIG_X86_64
2628 min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
2629 #endif
2630 opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT;
2631 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
2632 &_vmexit_control) < 0)
2633 return -EIO;
2634
2635 min = 0;
2636 opt = VM_ENTRY_LOAD_IA32_PAT;
2637 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
2638 &_vmentry_control) < 0)
2639 return -EIO;
2640
2641 rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
2642
2643 /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
2644 if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
2645 return -EIO;
2646
2647 #ifdef CONFIG_X86_64
2648 /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
2649 if (vmx_msr_high & (1u<<16))
2650 return -EIO;
2651 #endif
2652
2653 /* Require Write-Back (WB) memory type for VMCS accesses. */
2654 if (((vmx_msr_high >> 18) & 15) != 6)
2655 return -EIO;
2656
2657 vmcs_conf->size = vmx_msr_high & 0x1fff;
2658 vmcs_conf->order = get_order(vmcs_config.size);
2659 vmcs_conf->revision_id = vmx_msr_low;
2660
2661 vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
2662 vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
2663 vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
2664 vmcs_conf->vmexit_ctrl = _vmexit_control;
2665 vmcs_conf->vmentry_ctrl = _vmentry_control;
2666
2667 cpu_has_load_ia32_efer =
2668 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
2669 VM_ENTRY_LOAD_IA32_EFER)
2670 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
2671 VM_EXIT_LOAD_IA32_EFER);
2672
2673 cpu_has_load_perf_global_ctrl =
2674 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
2675 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
2676 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
2677 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
2678
2679 /*
2680 * Some cpus support VM_ENTRY_(LOAD|SAVE)_IA32_PERF_GLOBAL_CTRL
2681 * but due to arrata below it can't be used. Workaround is to use
2682 * msr load mechanism to switch IA32_PERF_GLOBAL_CTRL.
2683 *
2684 * VM Exit May Incorrectly Clear IA32_PERF_GLOBAL_CTRL [34:32]
2685 *
2686 * AAK155 (model 26)
2687 * AAP115 (model 30)
2688 * AAT100 (model 37)
2689 * BC86,AAY89,BD102 (model 44)
2690 * BA97 (model 46)
2691 *
2692 */
2693 if (cpu_has_load_perf_global_ctrl && boot_cpu_data.x86 == 0x6) {
2694 switch (boot_cpu_data.x86_model) {
2695 case 26:
2696 case 30:
2697 case 37:
2698 case 44:
2699 case 46:
2700 cpu_has_load_perf_global_ctrl = false;
2701 printk_once(KERN_WARNING"kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
2702 "does not work properly. Using workaround\n");
2703 break;
2704 default:
2705 break;
2706 }
2707 }
2708
2709 return 0;
2710 }
2711
2712 static struct vmcs *alloc_vmcs_cpu(int cpu)
2713 {
2714 int node = cpu_to_node(cpu);
2715 struct page *pages;
2716 struct vmcs *vmcs;
2717
2718 pages = alloc_pages_exact_node(node, GFP_KERNEL, vmcs_config.order);
2719 if (!pages)
2720 return NULL;
2721 vmcs = page_address(pages);
2722 memset(vmcs, 0, vmcs_config.size);
2723 vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
2724 return vmcs;
2725 }
2726
2727 static struct vmcs *alloc_vmcs(void)
2728 {
2729 return alloc_vmcs_cpu(raw_smp_processor_id());
2730 }
2731
2732 static void free_vmcs(struct vmcs *vmcs)
2733 {
2734 free_pages((unsigned long)vmcs, vmcs_config.order);
2735 }
2736
2737 /*
2738 * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
2739 */
2740 static void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2741 {
2742 if (!loaded_vmcs->vmcs)
2743 return;
2744 loaded_vmcs_clear(loaded_vmcs);
2745 free_vmcs(loaded_vmcs->vmcs);
2746 loaded_vmcs->vmcs = NULL;
2747 }
2748
2749 static void free_kvm_area(void)
2750 {
2751 int cpu;
2752
2753 for_each_possible_cpu(cpu) {
2754 free_vmcs(per_cpu(vmxarea, cpu));
2755 per_cpu(vmxarea, cpu) = NULL;
2756 }
2757 }
2758
2759 static __init int alloc_kvm_area(void)
2760 {
2761 int cpu;
2762
2763 for_each_possible_cpu(cpu) {
2764 struct vmcs *vmcs;
2765
2766 vmcs = alloc_vmcs_cpu(cpu);
2767 if (!vmcs) {
2768 free_kvm_area();
2769 return -ENOMEM;
2770 }
2771
2772 per_cpu(vmxarea, cpu) = vmcs;
2773 }
2774 return 0;
2775 }
2776
2777 static __init int hardware_setup(void)
2778 {
2779 if (setup_vmcs_config(&vmcs_config) < 0)
2780 return -EIO;
2781
2782 if (boot_cpu_has(X86_FEATURE_NX))
2783 kvm_enable_efer_bits(EFER_NX);
2784
2785 if (!cpu_has_vmx_vpid())
2786 enable_vpid = 0;
2787
2788 if (!cpu_has_vmx_ept() ||
2789 !cpu_has_vmx_ept_4levels()) {
2790 enable_ept = 0;
2791 enable_unrestricted_guest = 0;
2792 enable_ept_ad_bits = 0;
2793 }
2794
2795 if (!cpu_has_vmx_ept_ad_bits())
2796 enable_ept_ad_bits = 0;
2797
2798 if (!cpu_has_vmx_unrestricted_guest())
2799 enable_unrestricted_guest = 0;
2800
2801 if (!cpu_has_vmx_flexpriority())
2802 flexpriority_enabled = 0;
2803
2804 if (!cpu_has_vmx_tpr_shadow())
2805 kvm_x86_ops->update_cr8_intercept = NULL;
2806
2807 if (enable_ept && !cpu_has_vmx_ept_2m_page())
2808 kvm_disable_largepages();
2809
2810 if (!cpu_has_vmx_ple())
2811 ple_gap = 0;
2812
2813 if (!cpu_has_vmx_apic_register_virt() ||
2814 !cpu_has_vmx_virtual_intr_delivery())
2815 enable_apicv_reg_vid = 0;
2816
2817 if (enable_apicv_reg_vid)
2818 kvm_x86_ops->update_cr8_intercept = NULL;
2819 else
2820 kvm_x86_ops->hwapic_irr_update = NULL;
2821
2822 if (nested)
2823 nested_vmx_setup_ctls_msrs();
2824
2825 return alloc_kvm_area();
2826 }
2827
2828 static __exit void hardware_unsetup(void)
2829 {
2830 free_kvm_area();
2831 }
2832
2833 static bool emulation_required(struct kvm_vcpu *vcpu)
2834 {
2835 return emulate_invalid_guest_state && !guest_state_valid(vcpu);
2836 }
2837
2838 static void fix_pmode_seg(struct kvm_vcpu *vcpu, int seg,
2839 struct kvm_segment *save)
2840 {
2841 if (!emulate_invalid_guest_state) {
2842 /*
2843 * CS and SS RPL should be equal during guest entry according
2844 * to VMX spec, but in reality it is not always so. Since vcpu
2845 * is in the middle of the transition from real mode to
2846 * protected mode it is safe to assume that RPL 0 is a good
2847 * default value.
2848 */
2849 if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
2850 save->selector &= ~SELECTOR_RPL_MASK;
2851 save->dpl = save->selector & SELECTOR_RPL_MASK;
2852 save->s = 1;
2853 }
2854 vmx_set_segment(vcpu, save, seg);
2855 }
2856
2857 static void enter_pmode(struct kvm_vcpu *vcpu)
2858 {
2859 unsigned long flags;
2860 struct vcpu_vmx *vmx = to_vmx(vcpu);
2861
2862 /*
2863 * Update real mode segment cache. It may be not up-to-date if sement
2864 * register was written while vcpu was in a guest mode.
2865 */
2866 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
2867 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
2868 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
2869 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
2870 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
2871 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
2872
2873 vmx->rmode.vm86_active = 0;
2874
2875 vmx_segment_cache_clear(vmx);
2876
2877 vmx_set_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
2878
2879 flags = vmcs_readl(GUEST_RFLAGS);
2880 flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
2881 flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
2882 vmcs_writel(GUEST_RFLAGS, flags);
2883
2884 vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
2885 (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
2886
2887 update_exception_bitmap(vcpu);
2888
2889 fix_pmode_seg(vcpu, VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
2890 fix_pmode_seg(vcpu, VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
2891 fix_pmode_seg(vcpu, VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
2892 fix_pmode_seg(vcpu, VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
2893 fix_pmode_seg(vcpu, VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
2894 fix_pmode_seg(vcpu, VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
2895
2896 /* CPL is always 0 when CPU enters protected mode */
2897 __set_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail);
2898 vmx->cpl = 0;
2899 }
2900
2901 static void fix_rmode_seg(int seg, struct kvm_segment *save)
2902 {
2903 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2904 struct kvm_segment var = *save;
2905
2906 var.dpl = 0x3;
2907 if (seg == VCPU_SREG_CS)
2908 var.type = 0x3;
2909
2910 if (!emulate_invalid_guest_state) {
2911 var.selector = var.base >> 4;
2912 var.base = var.base & 0xffff0;
2913 var.limit = 0xffff;
2914 var.g = 0;
2915 var.db = 0;
2916 var.present = 1;
2917 var.s = 1;
2918 var.l = 0;
2919 var.unusable = 0;
2920 var.type = 0x3;
2921 var.avl = 0;
2922 if (save->base & 0xf)
2923 printk_once(KERN_WARNING "kvm: segment base is not "
2924 "paragraph aligned when entering "
2925 "protected mode (seg=%d)", seg);
2926 }
2927
2928 vmcs_write16(sf->selector, var.selector);
2929 vmcs_write32(sf->base, var.base);
2930 vmcs_write32(sf->limit, var.limit);
2931 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(&var));
2932 }
2933
2934 static void enter_rmode(struct kvm_vcpu *vcpu)
2935 {
2936 unsigned long flags;
2937 struct vcpu_vmx *vmx = to_vmx(vcpu);
2938
2939 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
2940 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
2941 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
2942 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
2943 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
2944 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
2945 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
2946
2947 vmx->rmode.vm86_active = 1;
2948
2949 /*
2950 * Very old userspace does not call KVM_SET_TSS_ADDR before entering
2951 * vcpu. Warn the user that an update is overdue.
2952 */
2953 if (!vcpu->kvm->arch.tss_addr)
2954 printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
2955 "called before entering vcpu\n");
2956
2957 vmx_segment_cache_clear(vmx);
2958
2959 vmcs_writel(GUEST_TR_BASE, vcpu->kvm->arch.tss_addr);
2960 vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
2961 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
2962
2963 flags = vmcs_readl(GUEST_RFLAGS);
2964 vmx->rmode.save_rflags = flags;
2965
2966 flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
2967
2968 vmcs_writel(GUEST_RFLAGS, flags);
2969 vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
2970 update_exception_bitmap(vcpu);
2971
2972 fix_rmode_seg(VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
2973 fix_rmode_seg(VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
2974 fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
2975 fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
2976 fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
2977 fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
2978
2979 kvm_mmu_reset_context(vcpu);
2980 }
2981
2982 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
2983 {
2984 struct vcpu_vmx *vmx = to_vmx(vcpu);
2985 struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
2986
2987 if (!msr)
2988 return;
2989
2990 /*
2991 * Force kernel_gs_base reloading before EFER changes, as control
2992 * of this msr depends on is_long_mode().
2993 */
2994 vmx_load_host_state(to_vmx(vcpu));
2995 vcpu->arch.efer = efer;
2996 if (efer & EFER_LMA) {
2997 vmcs_write32(VM_ENTRY_CONTROLS,
2998 vmcs_read32(VM_ENTRY_CONTROLS) |
2999 VM_ENTRY_IA32E_MODE);
3000 msr->data = efer;
3001 } else {
3002 vmcs_write32(VM_ENTRY_CONTROLS,
3003 vmcs_read32(VM_ENTRY_CONTROLS) &
3004 ~VM_ENTRY_IA32E_MODE);
3005
3006 msr->data = efer & ~EFER_LME;
3007 }
3008 setup_msrs(vmx);
3009 }
3010
3011 #ifdef CONFIG_X86_64
3012
3013 static void enter_lmode(struct kvm_vcpu *vcpu)
3014 {
3015 u32 guest_tr_ar;
3016
3017 vmx_segment_cache_clear(to_vmx(vcpu));
3018
3019 guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
3020 if ((guest_tr_ar & AR_TYPE_MASK) != AR_TYPE_BUSY_64_TSS) {
3021 pr_debug_ratelimited("%s: tss fixup for long mode. \n",
3022 __func__);
3023 vmcs_write32(GUEST_TR_AR_BYTES,
3024 (guest_tr_ar & ~AR_TYPE_MASK)
3025 | AR_TYPE_BUSY_64_TSS);
3026 }
3027 vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
3028 }
3029
3030 static void exit_lmode(struct kvm_vcpu *vcpu)
3031 {
3032 vmcs_write32(VM_ENTRY_CONTROLS,
3033 vmcs_read32(VM_ENTRY_CONTROLS)
3034 & ~VM_ENTRY_IA32E_MODE);
3035 vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
3036 }
3037
3038 #endif
3039
3040 static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
3041 {
3042 vpid_sync_context(to_vmx(vcpu));
3043 if (enable_ept) {
3044 if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
3045 return;
3046 ept_sync_context(construct_eptp(vcpu->arch.mmu.root_hpa));
3047 }
3048 }
3049
3050 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
3051 {
3052 ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
3053
3054 vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
3055 vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
3056 }
3057
3058 static void vmx_decache_cr3(struct kvm_vcpu *vcpu)
3059 {
3060 if (enable_ept && is_paging(vcpu))
3061 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
3062 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
3063 }
3064
3065 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
3066 {
3067 ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
3068
3069 vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
3070 vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
3071 }
3072
3073 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
3074 {
3075 if (!test_bit(VCPU_EXREG_PDPTR,
3076 (unsigned long *)&vcpu->arch.regs_dirty))
3077 return;
3078
3079 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
3080 vmcs_write64(GUEST_PDPTR0, vcpu->arch.mmu.pdptrs[0]);
3081 vmcs_write64(GUEST_PDPTR1, vcpu->arch.mmu.pdptrs[1]);
3082 vmcs_write64(GUEST_PDPTR2, vcpu->arch.mmu.pdptrs[2]);
3083 vmcs_write64(GUEST_PDPTR3, vcpu->arch.mmu.pdptrs[3]);
3084 }
3085 }
3086
3087 static void ept_save_pdptrs(struct kvm_vcpu *vcpu)
3088 {
3089 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
3090 vcpu->arch.mmu.pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
3091 vcpu->arch.mmu.pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
3092 vcpu->arch.mmu.pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
3093 vcpu->arch.mmu.pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
3094 }
3095
3096 __set_bit(VCPU_EXREG_PDPTR,
3097 (unsigned long *)&vcpu->arch.regs_avail);
3098 __set_bit(VCPU_EXREG_PDPTR,
3099 (unsigned long *)&vcpu->arch.regs_dirty);
3100 }
3101
3102 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
3103
3104 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
3105 unsigned long cr0,
3106 struct kvm_vcpu *vcpu)
3107 {
3108 if (!test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
3109 vmx_decache_cr3(vcpu);
3110 if (!(cr0 & X86_CR0_PG)) {
3111 /* From paging/starting to nonpaging */
3112 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
3113 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
3114 (CPU_BASED_CR3_LOAD_EXITING |
3115 CPU_BASED_CR3_STORE_EXITING));
3116 vcpu->arch.cr0 = cr0;
3117 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
3118 } else if (!is_paging(vcpu)) {
3119 /* From nonpaging to paging */
3120 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
3121 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
3122 ~(CPU_BASED_CR3_LOAD_EXITING |
3123 CPU_BASED_CR3_STORE_EXITING));
3124 vcpu->arch.cr0 = cr0;
3125 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
3126 }
3127
3128 if (!(cr0 & X86_CR0_WP))
3129 *hw_cr0 &= ~X86_CR0_WP;
3130 }
3131
3132 static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
3133 {
3134 struct vcpu_vmx *vmx = to_vmx(vcpu);
3135 unsigned long hw_cr0;
3136
3137 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK);
3138 if (enable_unrestricted_guest)
3139 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
3140 else {
3141 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON;
3142
3143 if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
3144 enter_pmode(vcpu);
3145
3146 if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
3147 enter_rmode(vcpu);
3148 }
3149
3150 #ifdef CONFIG_X86_64
3151 if (vcpu->arch.efer & EFER_LME) {
3152 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
3153 enter_lmode(vcpu);
3154 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
3155 exit_lmode(vcpu);
3156 }
3157 #endif
3158
3159 if (enable_ept)
3160 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
3161
3162 if (!vcpu->fpu_active)
3163 hw_cr0 |= X86_CR0_TS | X86_CR0_MP;
3164
3165 vmcs_writel(CR0_READ_SHADOW, cr0);
3166 vmcs_writel(GUEST_CR0, hw_cr0);
3167 vcpu->arch.cr0 = cr0;
3168
3169 /* depends on vcpu->arch.cr0 to be set to a new value */
3170 vmx->emulation_required = emulation_required(vcpu);
3171 }
3172
3173 static u64 construct_eptp(unsigned long root_hpa)
3174 {
3175 u64 eptp;
3176
3177 /* TODO write the value reading from MSR */
3178 eptp = VMX_EPT_DEFAULT_MT |
3179 VMX_EPT_DEFAULT_GAW << VMX_EPT_GAW_EPTP_SHIFT;
3180 if (enable_ept_ad_bits)
3181 eptp |= VMX_EPT_AD_ENABLE_BIT;
3182 eptp |= (root_hpa & PAGE_MASK);
3183
3184 return eptp;
3185 }
3186
3187 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
3188 {
3189 unsigned long guest_cr3;
3190 u64 eptp;
3191
3192 guest_cr3 = cr3;
3193 if (enable_ept) {
3194 eptp = construct_eptp(cr3);
3195 vmcs_write64(EPT_POINTER, eptp);
3196 guest_cr3 = is_paging(vcpu) ? kvm_read_cr3(vcpu) :
3197 vcpu->kvm->arch.ept_identity_map_addr;
3198 ept_load_pdptrs(vcpu);
3199 }
3200
3201 vmx_flush_tlb(vcpu);
3202 vmcs_writel(GUEST_CR3, guest_cr3);
3203 }
3204
3205 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
3206 {
3207 unsigned long hw_cr4 = cr4 | (to_vmx(vcpu)->rmode.vm86_active ?
3208 KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON);
3209
3210 if (cr4 & X86_CR4_VMXE) {
3211 /*
3212 * To use VMXON (and later other VMX instructions), a guest
3213 * must first be able to turn on cr4.VMXE (see handle_vmon()).
3214 * So basically the check on whether to allow nested VMX
3215 * is here.
3216 */
3217 if (!nested_vmx_allowed(vcpu))
3218 return 1;
3219 }
3220 if (to_vmx(vcpu)->nested.vmxon &&
3221 ((cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON))
3222 return 1;
3223
3224 vcpu->arch.cr4 = cr4;
3225 if (enable_ept) {
3226 if (!is_paging(vcpu)) {
3227 hw_cr4 &= ~X86_CR4_PAE;
3228 hw_cr4 |= X86_CR4_PSE;
3229 /*
3230 * SMEP is disabled if CPU is in non-paging mode in
3231 * hardware. However KVM always uses paging mode to
3232 * emulate guest non-paging mode with TDP.
3233 * To emulate this behavior, SMEP needs to be manually
3234 * disabled when guest switches to non-paging mode.
3235 */
3236 hw_cr4 &= ~X86_CR4_SMEP;
3237 } else if (!(cr4 & X86_CR4_PAE)) {
3238 hw_cr4 &= ~X86_CR4_PAE;
3239 }
3240 }
3241
3242 vmcs_writel(CR4_READ_SHADOW, cr4);
3243 vmcs_writel(GUEST_CR4, hw_cr4);
3244 return 0;
3245 }
3246
3247 static void vmx_get_segment(struct kvm_vcpu *vcpu,
3248 struct kvm_segment *var, int seg)
3249 {
3250 struct vcpu_vmx *vmx = to_vmx(vcpu);
3251 u32 ar;
3252
3253 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3254 *var = vmx->rmode.segs[seg];
3255 if (seg == VCPU_SREG_TR
3256 || var->selector == vmx_read_guest_seg_selector(vmx, seg))
3257 return;
3258 var->base = vmx_read_guest_seg_base(vmx, seg);
3259 var->selector = vmx_read_guest_seg_selector(vmx, seg);
3260 return;
3261 }
3262 var->base = vmx_read_guest_seg_base(vmx, seg);
3263 var->limit = vmx_read_guest_seg_limit(vmx, seg);
3264 var->selector = vmx_read_guest_seg_selector(vmx, seg);
3265 ar = vmx_read_guest_seg_ar(vmx, seg);
3266 var->type = ar & 15;
3267 var->s = (ar >> 4) & 1;
3268 var->dpl = (ar >> 5) & 3;
3269 var->present = (ar >> 7) & 1;
3270 var->avl = (ar >> 12) & 1;
3271 var->l = (ar >> 13) & 1;
3272 var->db = (ar >> 14) & 1;
3273 var->g = (ar >> 15) & 1;
3274 var->unusable = (ar >> 16) & 1;
3275 }
3276
3277 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
3278 {
3279 struct kvm_segment s;
3280
3281 if (to_vmx(vcpu)->rmode.vm86_active) {
3282 vmx_get_segment(vcpu, &s, seg);
3283 return s.base;
3284 }
3285 return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
3286 }
3287
3288 static int vmx_get_cpl(struct kvm_vcpu *vcpu)
3289 {
3290 struct vcpu_vmx *vmx = to_vmx(vcpu);
3291
3292 if (!is_protmode(vcpu))
3293 return 0;
3294
3295 if (!is_long_mode(vcpu)
3296 && (kvm_get_rflags(vcpu) & X86_EFLAGS_VM)) /* if virtual 8086 */
3297 return 3;
3298
3299 if (!test_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail)) {
3300 __set_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail);
3301 vmx->cpl = vmx_read_guest_seg_selector(vmx, VCPU_SREG_CS) & 3;
3302 }
3303
3304 return vmx->cpl;
3305 }
3306
3307
3308 static u32 vmx_segment_access_rights(struct kvm_segment *var)
3309 {
3310 u32 ar;
3311
3312 if (var->unusable || !var->present)
3313 ar = 1 << 16;
3314 else {
3315 ar = var->type & 15;
3316 ar |= (var->s & 1) << 4;
3317 ar |= (var->dpl & 3) << 5;
3318 ar |= (var->present & 1) << 7;
3319 ar |= (var->avl & 1) << 12;
3320 ar |= (var->l & 1) << 13;
3321 ar |= (var->db & 1) << 14;
3322 ar |= (var->g & 1) << 15;
3323 }
3324
3325 return ar;
3326 }
3327
3328 static void vmx_set_segment(struct kvm_vcpu *vcpu,
3329 struct kvm_segment *var, int seg)
3330 {
3331 struct vcpu_vmx *vmx = to_vmx(vcpu);
3332 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3333
3334 vmx_segment_cache_clear(vmx);
3335 if (seg == VCPU_SREG_CS)
3336 __clear_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail);
3337
3338 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3339 vmx->rmode.segs[seg] = *var;
3340 if (seg == VCPU_SREG_TR)
3341 vmcs_write16(sf->selector, var->selector);
3342 else if (var->s)
3343 fix_rmode_seg(seg, &vmx->rmode.segs[seg]);
3344 goto out;
3345 }
3346
3347 vmcs_writel(sf->base, var->base);
3348 vmcs_write32(sf->limit, var->limit);
3349 vmcs_write16(sf->selector, var->selector);
3350
3351 /*
3352 * Fix the "Accessed" bit in AR field of segment registers for older
3353 * qemu binaries.
3354 * IA32 arch specifies that at the time of processor reset the
3355 * "Accessed" bit in the AR field of segment registers is 1. And qemu
3356 * is setting it to 0 in the userland code. This causes invalid guest
3357 * state vmexit when "unrestricted guest" mode is turned on.
3358 * Fix for this setup issue in cpu_reset is being pushed in the qemu
3359 * tree. Newer qemu binaries with that qemu fix would not need this
3360 * kvm hack.
3361 */
3362 if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
3363 var->type |= 0x1; /* Accessed */
3364
3365 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(var));
3366
3367 out:
3368 vmx->emulation_required |= emulation_required(vcpu);
3369 }
3370
3371 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3372 {
3373 u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
3374
3375 *db = (ar >> 14) & 1;
3376 *l = (ar >> 13) & 1;
3377 }
3378
3379 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3380 {
3381 dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
3382 dt->address = vmcs_readl(GUEST_IDTR_BASE);
3383 }
3384
3385 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3386 {
3387 vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
3388 vmcs_writel(GUEST_IDTR_BASE, dt->address);
3389 }
3390
3391 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3392 {
3393 dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
3394 dt->address = vmcs_readl(GUEST_GDTR_BASE);
3395 }
3396
3397 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3398 {
3399 vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
3400 vmcs_writel(GUEST_GDTR_BASE, dt->address);
3401 }
3402
3403 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
3404 {
3405 struct kvm_segment var;
3406 u32 ar;
3407
3408 vmx_get_segment(vcpu, &var, seg);
3409 var.dpl = 0x3;
3410 if (seg == VCPU_SREG_CS)
3411 var.type = 0x3;
3412 ar = vmx_segment_access_rights(&var);
3413
3414 if (var.base != (var.selector << 4))
3415 return false;
3416 if (var.limit != 0xffff)
3417 return false;
3418 if (ar != 0xf3)
3419 return false;
3420
3421 return true;
3422 }
3423
3424 static bool code_segment_valid(struct kvm_vcpu *vcpu)
3425 {
3426 struct kvm_segment cs;
3427 unsigned int cs_rpl;
3428
3429 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3430 cs_rpl = cs.selector & SELECTOR_RPL_MASK;
3431
3432 if (cs.unusable)
3433 return false;
3434 if (~cs.type & (AR_TYPE_CODE_MASK|AR_TYPE_ACCESSES_MASK))
3435 return false;
3436 if (!cs.s)
3437 return false;
3438 if (cs.type & AR_TYPE_WRITEABLE_MASK) {
3439 if (cs.dpl > cs_rpl)
3440 return false;
3441 } else {
3442 if (cs.dpl != cs_rpl)
3443 return false;
3444 }
3445 if (!cs.present)
3446 return false;
3447
3448 /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
3449 return true;
3450 }
3451
3452 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
3453 {
3454 struct kvm_segment ss;
3455 unsigned int ss_rpl;
3456
3457 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3458 ss_rpl = ss.selector & SELECTOR_RPL_MASK;
3459
3460 if (ss.unusable)
3461 return true;
3462 if (ss.type != 3 && ss.type != 7)
3463 return false;
3464 if (!ss.s)
3465 return false;
3466 if (ss.dpl != ss_rpl) /* DPL != RPL */
3467 return false;
3468 if (!ss.present)
3469 return false;
3470
3471 return true;
3472 }
3473
3474 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
3475 {
3476 struct kvm_segment var;
3477 unsigned int rpl;
3478
3479 vmx_get_segment(vcpu, &var, seg);
3480 rpl = var.selector & SELECTOR_RPL_MASK;
3481
3482 if (var.unusable)
3483 return true;
3484 if (!var.s)
3485 return false;
3486 if (!var.present)
3487 return false;
3488 if (~var.type & (AR_TYPE_CODE_MASK|AR_TYPE_WRITEABLE_MASK)) {
3489 if (var.dpl < rpl) /* DPL < RPL */
3490 return false;
3491 }
3492
3493 /* TODO: Add other members to kvm_segment_field to allow checking for other access
3494 * rights flags
3495 */
3496 return true;
3497 }
3498
3499 static bool tr_valid(struct kvm_vcpu *vcpu)
3500 {
3501 struct kvm_segment tr;
3502
3503 vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
3504
3505 if (tr.unusable)
3506 return false;
3507 if (tr.selector & SELECTOR_TI_MASK) /* TI = 1 */
3508 return false;
3509 if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
3510 return false;
3511 if (!tr.present)
3512 return false;
3513
3514 return true;
3515 }
3516
3517 static bool ldtr_valid(struct kvm_vcpu *vcpu)
3518 {
3519 struct kvm_segment ldtr;
3520
3521 vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
3522
3523 if (ldtr.unusable)
3524 return true;
3525 if (ldtr.selector & SELECTOR_TI_MASK) /* TI = 1 */
3526 return false;
3527 if (ldtr.type != 2)
3528 return false;
3529 if (!ldtr.present)
3530 return false;
3531
3532 return true;
3533 }
3534
3535 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
3536 {
3537 struct kvm_segment cs, ss;
3538
3539 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3540 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3541
3542 return ((cs.selector & SELECTOR_RPL_MASK) ==
3543 (ss.selector & SELECTOR_RPL_MASK));
3544 }
3545
3546 /*
3547 * Check if guest state is valid. Returns true if valid, false if
3548 * not.
3549 * We assume that registers are always usable
3550 */
3551 static bool guest_state_valid(struct kvm_vcpu *vcpu)
3552 {
3553 if (enable_unrestricted_guest)
3554 return true;
3555
3556 /* real mode guest state checks */
3557 if (!is_protmode(vcpu)) {
3558 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
3559 return false;
3560 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
3561 return false;
3562 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
3563 return false;
3564 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
3565 return false;
3566 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
3567 return false;
3568 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
3569 return false;
3570 } else {
3571 /* protected mode guest state checks */
3572 if (!cs_ss_rpl_check(vcpu))
3573 return false;
3574 if (!code_segment_valid(vcpu))
3575 return false;
3576 if (!stack_segment_valid(vcpu))
3577 return false;
3578 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
3579 return false;
3580 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
3581 return false;
3582 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
3583 return false;
3584 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
3585 return false;
3586 if (!tr_valid(vcpu))
3587 return false;
3588 if (!ldtr_valid(vcpu))
3589 return false;
3590 }
3591 /* TODO:
3592 * - Add checks on RIP
3593 * - Add checks on RFLAGS
3594 */
3595
3596 return true;
3597 }
3598
3599 static int init_rmode_tss(struct kvm *kvm)
3600 {
3601 gfn_t fn;
3602 u16 data = 0;
3603 int r, idx, ret = 0;
3604
3605 idx = srcu_read_lock(&kvm->srcu);
3606 fn = kvm->arch.tss_addr >> PAGE_SHIFT;
3607 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3608 if (r < 0)
3609 goto out;
3610 data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
3611 r = kvm_write_guest_page(kvm, fn++, &data,
3612 TSS_IOPB_BASE_OFFSET, sizeof(u16));
3613 if (r < 0)
3614 goto out;
3615 r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
3616 if (r < 0)
3617 goto out;
3618 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3619 if (r < 0)
3620 goto out;
3621 data = ~0;
3622 r = kvm_write_guest_page(kvm, fn, &data,
3623 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
3624 sizeof(u8));
3625 if (r < 0)
3626 goto out;
3627
3628 ret = 1;
3629 out:
3630 srcu_read_unlock(&kvm->srcu, idx);
3631 return ret;
3632 }
3633
3634 static int init_rmode_identity_map(struct kvm *kvm)
3635 {
3636 int i, idx, r, ret;
3637 pfn_t identity_map_pfn;
3638 u32 tmp;
3639
3640 if (!enable_ept)
3641 return 1;
3642 if (unlikely(!kvm->arch.ept_identity_pagetable)) {
3643 printk(KERN_ERR "EPT: identity-mapping pagetable "
3644 "haven't been allocated!\n");
3645 return 0;
3646 }
3647 if (likely(kvm->arch.ept_identity_pagetable_done))
3648 return 1;
3649 ret = 0;
3650 identity_map_pfn = kvm->arch.ept_identity_map_addr >> PAGE_SHIFT;
3651 idx = srcu_read_lock(&kvm->srcu);
3652 r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
3653 if (r < 0)
3654 goto out;
3655 /* Set up identity-mapping pagetable for EPT in real mode */
3656 for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
3657 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
3658 _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
3659 r = kvm_write_guest_page(kvm, identity_map_pfn,
3660 &tmp, i * sizeof(tmp), sizeof(tmp));
3661 if (r < 0)
3662 goto out;
3663 }
3664 kvm->arch.ept_identity_pagetable_done = true;
3665 ret = 1;
3666 out:
3667 srcu_read_unlock(&kvm->srcu, idx);
3668 return ret;
3669 }
3670
3671 static void seg_setup(int seg)
3672 {
3673 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3674 unsigned int ar;
3675
3676 vmcs_write16(sf->selector, 0);
3677 vmcs_writel(sf->base, 0);
3678 vmcs_write32(sf->limit, 0xffff);
3679 ar = 0x93;
3680 if (seg == VCPU_SREG_CS)
3681 ar |= 0x08; /* code segment */
3682
3683 vmcs_write32(sf->ar_bytes, ar);
3684 }
3685
3686 static int alloc_apic_access_page(struct kvm *kvm)
3687 {
3688 struct page *page;
3689 struct kvm_userspace_memory_region kvm_userspace_mem;
3690 int r = 0;
3691
3692 mutex_lock(&kvm->slots_lock);
3693 if (kvm->arch.apic_access_page)
3694 goto out;
3695 kvm_userspace_mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
3696 kvm_userspace_mem.flags = 0;
3697 kvm_userspace_mem.guest_phys_addr = 0xfee00000ULL;
3698 kvm_userspace_mem.memory_size = PAGE_SIZE;
3699 r = __kvm_set_memory_region(kvm, &kvm_userspace_mem);
3700 if (r)
3701 goto out;
3702
3703 page = gfn_to_page(kvm, 0xfee00);
3704 if (is_error_page(page)) {
3705 r = -EFAULT;
3706 goto out;
3707 }
3708
3709 kvm->arch.apic_access_page = page;
3710 out:
3711 mutex_unlock(&kvm->slots_lock);
3712 return r;
3713 }
3714
3715 static int alloc_identity_pagetable(struct kvm *kvm)
3716 {
3717 struct page *page;
3718 struct kvm_userspace_memory_region kvm_userspace_mem;
3719 int r = 0;
3720
3721 mutex_lock(&kvm->slots_lock);
3722 if (kvm->arch.ept_identity_pagetable)
3723 goto out;
3724 kvm_userspace_mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
3725 kvm_userspace_mem.flags = 0;
3726 kvm_userspace_mem.guest_phys_addr =
3727 kvm->arch.ept_identity_map_addr;
3728 kvm_userspace_mem.memory_size = PAGE_SIZE;
3729 r = __kvm_set_memory_region(kvm, &kvm_userspace_mem);
3730 if (r)
3731 goto out;
3732
3733 page = gfn_to_page(kvm, kvm->arch.ept_identity_map_addr >> PAGE_SHIFT);
3734 if (is_error_page(page)) {
3735 r = -EFAULT;
3736 goto out;
3737 }
3738
3739 kvm->arch.ept_identity_pagetable = page;
3740 out:
3741 mutex_unlock(&kvm->slots_lock);
3742 return r;
3743 }
3744
3745 static void allocate_vpid(struct vcpu_vmx *vmx)
3746 {
3747 int vpid;
3748
3749 vmx->vpid = 0;
3750 if (!enable_vpid)
3751 return;
3752 spin_lock(&vmx_vpid_lock);
3753 vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
3754 if (vpid < VMX_NR_VPIDS) {
3755 vmx->vpid = vpid;
3756 __set_bit(vpid, vmx_vpid_bitmap);
3757 }
3758 spin_unlock(&vmx_vpid_lock);
3759 }
3760
3761 static void free_vpid(struct vcpu_vmx *vmx)
3762 {
3763 if (!enable_vpid)
3764 return;
3765 spin_lock(&vmx_vpid_lock);
3766 if (vmx->vpid != 0)
3767 __clear_bit(vmx->vpid, vmx_vpid_bitmap);
3768 spin_unlock(&vmx_vpid_lock);
3769 }
3770
3771 #define MSR_TYPE_R 1
3772 #define MSR_TYPE_W 2
3773 static void __vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
3774 u32 msr, int type)
3775 {
3776 int f = sizeof(unsigned long);
3777
3778 if (!cpu_has_vmx_msr_bitmap())
3779 return;
3780
3781 /*
3782 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
3783 * have the write-low and read-high bitmap offsets the wrong way round.
3784 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
3785 */
3786 if (msr <= 0x1fff) {
3787 if (type & MSR_TYPE_R)
3788 /* read-low */
3789 __clear_bit(msr, msr_bitmap + 0x000 / f);
3790
3791 if (type & MSR_TYPE_W)
3792 /* write-low */
3793 __clear_bit(msr, msr_bitmap + 0x800 / f);
3794
3795 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
3796 msr &= 0x1fff;
3797 if (type & MSR_TYPE_R)
3798 /* read-high */
3799 __clear_bit(msr, msr_bitmap + 0x400 / f);
3800
3801 if (type & MSR_TYPE_W)
3802 /* write-high */
3803 __clear_bit(msr, msr_bitmap + 0xc00 / f);
3804
3805 }
3806 }
3807
3808 static void __vmx_enable_intercept_for_msr(unsigned long *msr_bitmap,
3809 u32 msr, int type)
3810 {
3811 int f = sizeof(unsigned long);
3812
3813 if (!cpu_has_vmx_msr_bitmap())
3814 return;
3815
3816 /*
3817 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
3818 * have the write-low and read-high bitmap offsets the wrong way round.
3819 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
3820 */
3821 if (msr <= 0x1fff) {
3822 if (type & MSR_TYPE_R)
3823 /* read-low */
3824 __set_bit(msr, msr_bitmap + 0x000 / f);
3825
3826 if (type & MSR_TYPE_W)
3827 /* write-low */
3828 __set_bit(msr, msr_bitmap + 0x800 / f);
3829
3830 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
3831 msr &= 0x1fff;
3832 if (type & MSR_TYPE_R)
3833 /* read-high */
3834 __set_bit(msr, msr_bitmap + 0x400 / f);
3835
3836 if (type & MSR_TYPE_W)
3837 /* write-high */
3838 __set_bit(msr, msr_bitmap + 0xc00 / f);
3839
3840 }
3841 }
3842
3843 static void vmx_disable_intercept_for_msr(u32 msr, bool longmode_only)
3844 {
3845 if (!longmode_only)
3846 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy,
3847 msr, MSR_TYPE_R | MSR_TYPE_W);
3848 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode,
3849 msr, MSR_TYPE_R | MSR_TYPE_W);
3850 }
3851
3852 static void vmx_enable_intercept_msr_read_x2apic(u32 msr)
3853 {
3854 __vmx_enable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic,
3855 msr, MSR_TYPE_R);
3856 __vmx_enable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic,
3857 msr, MSR_TYPE_R);
3858 }
3859
3860 static void vmx_disable_intercept_msr_read_x2apic(u32 msr)
3861 {
3862 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic,
3863 msr, MSR_TYPE_R);
3864 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic,
3865 msr, MSR_TYPE_R);
3866 }
3867
3868 static void vmx_disable_intercept_msr_write_x2apic(u32 msr)
3869 {
3870 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic,
3871 msr, MSR_TYPE_W);
3872 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic,
3873 msr, MSR_TYPE_W);
3874 }
3875
3876 /*
3877 * Set up the vmcs's constant host-state fields, i.e., host-state fields that
3878 * will not change in the lifetime of the guest.
3879 * Note that host-state that does change is set elsewhere. E.g., host-state
3880 * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
3881 */
3882 static void vmx_set_constant_host_state(void)
3883 {
3884 u32 low32, high32;
3885 unsigned long tmpl;
3886 struct desc_ptr dt;
3887
3888 vmcs_writel(HOST_CR0, read_cr0() & ~X86_CR0_TS); /* 22.2.3 */
3889 vmcs_writel(HOST_CR4, read_cr4()); /* 22.2.3, 22.2.5 */
3890 vmcs_writel(HOST_CR3, read_cr3()); /* 22.2.3 FIXME: shadow tables */
3891
3892 vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */
3893 #ifdef CONFIG_X86_64
3894 /*
3895 * Load null selectors, so we can avoid reloading them in
3896 * __vmx_load_host_state(), in case userspace uses the null selectors
3897 * too (the expected case).
3898 */
3899 vmcs_write16(HOST_DS_SELECTOR, 0);
3900 vmcs_write16(HOST_ES_SELECTOR, 0);
3901 #else
3902 vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
3903 vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS); /* 22.2.4 */
3904 #endif
3905 vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
3906 vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8); /* 22.2.4 */
3907
3908 native_store_idt(&dt);
3909 vmcs_writel(HOST_IDTR_BASE, dt.address); /* 22.2.4 */
3910
3911 vmcs_writel(HOST_RIP, vmx_return); /* 22.2.5 */
3912
3913 rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
3914 vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
3915 rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
3916 vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl); /* 22.2.3 */
3917
3918 if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
3919 rdmsr(MSR_IA32_CR_PAT, low32, high32);
3920 vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
3921 }
3922 }
3923
3924 static void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
3925 {
3926 vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
3927 if (enable_ept)
3928 vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
3929 if (is_guest_mode(&vmx->vcpu))
3930 vmx->vcpu.arch.cr4_guest_owned_bits &=
3931 ~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask;
3932 vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
3933 }
3934
3935 static u32 vmx_exec_control(struct vcpu_vmx *vmx)
3936 {
3937 u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
3938 if (!vm_need_tpr_shadow(vmx->vcpu.kvm)) {
3939 exec_control &= ~CPU_BASED_TPR_SHADOW;
3940 #ifdef CONFIG_X86_64
3941 exec_control |= CPU_BASED_CR8_STORE_EXITING |
3942 CPU_BASED_CR8_LOAD_EXITING;
3943 #endif
3944 }
3945 if (!enable_ept)
3946 exec_control |= CPU_BASED_CR3_STORE_EXITING |
3947 CPU_BASED_CR3_LOAD_EXITING |
3948 CPU_BASED_INVLPG_EXITING;
3949 return exec_control;
3950 }
3951
3952 static int vmx_vm_has_apicv(struct kvm *kvm)
3953 {
3954 return enable_apicv_reg_vid && irqchip_in_kernel(kvm);
3955 }
3956
3957 static u32 vmx_secondary_exec_control(struct vcpu_vmx *vmx)
3958 {
3959 u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
3960 if (!vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
3961 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
3962 if (vmx->vpid == 0)
3963 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
3964 if (!enable_ept) {
3965 exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
3966 enable_unrestricted_guest = 0;
3967 /* Enable INVPCID for non-ept guests may cause performance regression. */
3968 exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
3969 }
3970 if (!enable_unrestricted_guest)
3971 exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
3972 if (!ple_gap)
3973 exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
3974 if (!vmx_vm_has_apicv(vmx->vcpu.kvm))
3975 exec_control &= ~(SECONDARY_EXEC_APIC_REGISTER_VIRT |
3976 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
3977 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
3978 return exec_control;
3979 }
3980
3981 static void ept_set_mmio_spte_mask(void)
3982 {
3983 /*
3984 * EPT Misconfigurations can be generated if the value of bits 2:0
3985 * of an EPT paging-structure entry is 110b (write/execute).
3986 * Also, magic bits (0xffull << 49) is set to quickly identify mmio
3987 * spte.
3988 */
3989 kvm_mmu_set_mmio_spte_mask(0xffull << 49 | 0x6ull);
3990 }
3991
3992 /*
3993 * Sets up the vmcs for emulated real mode.
3994 */
3995 static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
3996 {
3997 #ifdef CONFIG_X86_64
3998 unsigned long a;
3999 #endif
4000 int i;
4001
4002 /* I/O */
4003 vmcs_write64(IO_BITMAP_A, __pa(vmx_io_bitmap_a));
4004 vmcs_write64(IO_BITMAP_B, __pa(vmx_io_bitmap_b));
4005
4006 if (cpu_has_vmx_msr_bitmap())
4007 vmcs_write64(MSR_BITMAP, __pa(vmx_msr_bitmap_legacy));
4008
4009 vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
4010
4011 /* Control */
4012 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL,
4013 vmcs_config.pin_based_exec_ctrl);
4014
4015 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, vmx_exec_control(vmx));
4016
4017 if (cpu_has_secondary_exec_ctrls()) {
4018 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
4019 vmx_secondary_exec_control(vmx));
4020 }
4021
4022 if (enable_apicv_reg_vid) {
4023 vmcs_write64(EOI_EXIT_BITMAP0, 0);
4024 vmcs_write64(EOI_EXIT_BITMAP1, 0);
4025 vmcs_write64(EOI_EXIT_BITMAP2, 0);
4026 vmcs_write64(EOI_EXIT_BITMAP3, 0);
4027
4028 vmcs_write16(GUEST_INTR_STATUS, 0);
4029 }
4030
4031 if (ple_gap) {
4032 vmcs_write32(PLE_GAP, ple_gap);
4033 vmcs_write32(PLE_WINDOW, ple_window);
4034 }
4035
4036 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
4037 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
4038 vmcs_write32(CR3_TARGET_COUNT, 0); /* 22.2.1 */
4039
4040 vmcs_write16(HOST_FS_SELECTOR, 0); /* 22.2.4 */
4041 vmcs_write16(HOST_GS_SELECTOR, 0); /* 22.2.4 */
4042 vmx_set_constant_host_state();
4043 #ifdef CONFIG_X86_64
4044 rdmsrl(MSR_FS_BASE, a);
4045 vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
4046 rdmsrl(MSR_GS_BASE, a);
4047 vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
4048 #else
4049 vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
4050 vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
4051 #endif
4052
4053 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
4054 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
4055 vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host));
4056 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
4057 vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest));
4058
4059 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
4060 u32 msr_low, msr_high;
4061 u64 host_pat;
4062 rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high);
4063 host_pat = msr_low | ((u64) msr_high << 32);
4064 /* Write the default value follow host pat */
4065 vmcs_write64(GUEST_IA32_PAT, host_pat);
4066 /* Keep arch.pat sync with GUEST_IA32_PAT */
4067 vmx->vcpu.arch.pat = host_pat;
4068 }
4069
4070 for (i = 0; i < NR_VMX_MSR; ++i) {
4071 u32 index = vmx_msr_index[i];
4072 u32 data_low, data_high;
4073 int j = vmx->nmsrs;
4074
4075 if (rdmsr_safe(index, &data_low, &data_high) < 0)
4076 continue;
4077 if (wrmsr_safe(index, data_low, data_high) < 0)
4078 continue;
4079 vmx->guest_msrs[j].index = i;
4080 vmx->guest_msrs[j].data = 0;
4081 vmx->guest_msrs[j].mask = -1ull;
4082 ++vmx->nmsrs;
4083 }
4084
4085 vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
4086
4087 /* 22.2.1, 20.8.1 */
4088 vmcs_write32(VM_ENTRY_CONTROLS, vmcs_config.vmentry_ctrl);
4089
4090 vmcs_writel(CR0_GUEST_HOST_MASK, ~0UL);
4091 set_cr4_guest_host_mask(vmx);
4092
4093 return 0;
4094 }
4095
4096 static void vmx_vcpu_reset(struct kvm_vcpu *vcpu)
4097 {
4098 struct vcpu_vmx *vmx = to_vmx(vcpu);
4099 u64 msr;
4100
4101 vmx->rmode.vm86_active = 0;
4102
4103 vmx->soft_vnmi_blocked = 0;
4104
4105 vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
4106 kvm_set_cr8(&vmx->vcpu, 0);
4107 msr = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
4108 if (kvm_vcpu_is_bsp(&vmx->vcpu))
4109 msr |= MSR_IA32_APICBASE_BSP;
4110 kvm_set_apic_base(&vmx->vcpu, msr);
4111
4112 vmx_segment_cache_clear(vmx);
4113
4114 seg_setup(VCPU_SREG_CS);
4115 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
4116 vmcs_write32(GUEST_CS_BASE, 0xffff0000);
4117
4118 seg_setup(VCPU_SREG_DS);
4119 seg_setup(VCPU_SREG_ES);
4120 seg_setup(VCPU_SREG_FS);
4121 seg_setup(VCPU_SREG_GS);
4122 seg_setup(VCPU_SREG_SS);
4123
4124 vmcs_write16(GUEST_TR_SELECTOR, 0);
4125 vmcs_writel(GUEST_TR_BASE, 0);
4126 vmcs_write32(GUEST_TR_LIMIT, 0xffff);
4127 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
4128
4129 vmcs_write16(GUEST_LDTR_SELECTOR, 0);
4130 vmcs_writel(GUEST_LDTR_BASE, 0);
4131 vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
4132 vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
4133
4134 vmcs_write32(GUEST_SYSENTER_CS, 0);
4135 vmcs_writel(GUEST_SYSENTER_ESP, 0);
4136 vmcs_writel(GUEST_SYSENTER_EIP, 0);
4137
4138 vmcs_writel(GUEST_RFLAGS, 0x02);
4139 kvm_rip_write(vcpu, 0xfff0);
4140
4141 vmcs_writel(GUEST_GDTR_BASE, 0);
4142 vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
4143
4144 vmcs_writel(GUEST_IDTR_BASE, 0);
4145 vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
4146
4147 vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
4148 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
4149 vmcs_write32(GUEST_PENDING_DBG_EXCEPTIONS, 0);
4150
4151 /* Special registers */
4152 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
4153
4154 setup_msrs(vmx);
4155
4156 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */
4157
4158 if (cpu_has_vmx_tpr_shadow()) {
4159 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
4160 if (vm_need_tpr_shadow(vmx->vcpu.kvm))
4161 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
4162 __pa(vmx->vcpu.arch.apic->regs));
4163 vmcs_write32(TPR_THRESHOLD, 0);
4164 }
4165
4166 if (vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
4167 vmcs_write64(APIC_ACCESS_ADDR,
4168 page_to_phys(vmx->vcpu.kvm->arch.apic_access_page));
4169
4170 if (vmx->vpid != 0)
4171 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
4172
4173 vmx->vcpu.arch.cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
4174 vmx_set_cr0(&vmx->vcpu, kvm_read_cr0(vcpu)); /* enter rmode */
4175 vmx_set_cr4(&vmx->vcpu, 0);
4176 vmx_set_efer(&vmx->vcpu, 0);
4177 vmx_fpu_activate(&vmx->vcpu);
4178 update_exception_bitmap(&vmx->vcpu);
4179
4180 vpid_sync_context(vmx);
4181 }
4182
4183 /*
4184 * In nested virtualization, check if L1 asked to exit on external interrupts.
4185 * For most existing hypervisors, this will always return true.
4186 */
4187 static bool nested_exit_on_intr(struct kvm_vcpu *vcpu)
4188 {
4189 return get_vmcs12(vcpu)->pin_based_vm_exec_control &
4190 PIN_BASED_EXT_INTR_MASK;
4191 }
4192
4193 static void enable_irq_window(struct kvm_vcpu *vcpu)
4194 {
4195 u32 cpu_based_vm_exec_control;
4196 if (is_guest_mode(vcpu) && nested_exit_on_intr(vcpu)) {
4197 /*
4198 * We get here if vmx_interrupt_allowed() said we can't
4199 * inject to L1 now because L2 must run. Ask L2 to exit
4200 * right after entry, so we can inject to L1 more promptly.
4201 */
4202 kvm_make_request(KVM_REQ_IMMEDIATE_EXIT, vcpu);
4203 return;
4204 }
4205
4206 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
4207 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
4208 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
4209 }
4210
4211 static void enable_nmi_window(struct kvm_vcpu *vcpu)
4212 {
4213 u32 cpu_based_vm_exec_control;
4214
4215 if (!cpu_has_virtual_nmis()) {
4216 enable_irq_window(vcpu);
4217 return;
4218 }
4219
4220 if (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
4221 enable_irq_window(vcpu);
4222 return;
4223 }
4224 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
4225 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_NMI_PENDING;
4226 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
4227 }
4228
4229 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
4230 {
4231 struct vcpu_vmx *vmx = to_vmx(vcpu);
4232 uint32_t intr;
4233 int irq = vcpu->arch.interrupt.nr;
4234
4235 trace_kvm_inj_virq(irq);
4236
4237 ++vcpu->stat.irq_injections;
4238 if (vmx->rmode.vm86_active) {
4239 int inc_eip = 0;
4240 if (vcpu->arch.interrupt.soft)
4241 inc_eip = vcpu->arch.event_exit_inst_len;
4242 if (kvm_inject_realmode_interrupt(vcpu, irq, inc_eip) != EMULATE_DONE)
4243 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4244 return;
4245 }
4246 intr = irq | INTR_INFO_VALID_MASK;
4247 if (vcpu->arch.interrupt.soft) {
4248 intr |= INTR_TYPE_SOFT_INTR;
4249 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
4250 vmx->vcpu.arch.event_exit_inst_len);
4251 } else
4252 intr |= INTR_TYPE_EXT_INTR;
4253 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
4254 }
4255
4256 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
4257 {
4258 struct vcpu_vmx *vmx = to_vmx(vcpu);
4259
4260 if (is_guest_mode(vcpu))
4261 return;
4262
4263 if (!cpu_has_virtual_nmis()) {
4264 /*
4265 * Tracking the NMI-blocked state in software is built upon
4266 * finding the next open IRQ window. This, in turn, depends on
4267 * well-behaving guests: They have to keep IRQs disabled at
4268 * least as long as the NMI handler runs. Otherwise we may
4269 * cause NMI nesting, maybe breaking the guest. But as this is
4270 * highly unlikely, we can live with the residual risk.
4271 */
4272 vmx->soft_vnmi_blocked = 1;
4273 vmx->vnmi_blocked_time = 0;
4274 }
4275
4276 ++vcpu->stat.nmi_injections;
4277 vmx->nmi_known_unmasked = false;
4278 if (vmx->rmode.vm86_active) {
4279 if (kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0) != EMULATE_DONE)
4280 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4281 return;
4282 }
4283 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
4284 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
4285 }
4286
4287 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
4288 {
4289 if (!cpu_has_virtual_nmis() && to_vmx(vcpu)->soft_vnmi_blocked)
4290 return 0;
4291
4292 return !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4293 (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
4294 | GUEST_INTR_STATE_NMI));
4295 }
4296
4297 static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
4298 {
4299 if (!cpu_has_virtual_nmis())
4300 return to_vmx(vcpu)->soft_vnmi_blocked;
4301 if (to_vmx(vcpu)->nmi_known_unmasked)
4302 return false;
4303 return vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
4304 }
4305
4306 static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
4307 {
4308 struct vcpu_vmx *vmx = to_vmx(vcpu);
4309
4310 if (!cpu_has_virtual_nmis()) {
4311 if (vmx->soft_vnmi_blocked != masked) {
4312 vmx->soft_vnmi_blocked = masked;
4313 vmx->vnmi_blocked_time = 0;
4314 }
4315 } else {
4316 vmx->nmi_known_unmasked = !masked;
4317 if (masked)
4318 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
4319 GUEST_INTR_STATE_NMI);
4320 else
4321 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
4322 GUEST_INTR_STATE_NMI);
4323 }
4324 }
4325
4326 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
4327 {
4328 if (is_guest_mode(vcpu)) {
4329 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4330
4331 if (to_vmx(vcpu)->nested.nested_run_pending)
4332 return 0;
4333 if (nested_exit_on_intr(vcpu)) {
4334 nested_vmx_vmexit(vcpu);
4335 vmcs12->vm_exit_reason =
4336 EXIT_REASON_EXTERNAL_INTERRUPT;
4337 vmcs12->vm_exit_intr_info = 0;
4338 /*
4339 * fall through to normal code, but now in L1, not L2
4340 */
4341 }
4342 }
4343
4344 return (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
4345 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4346 (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
4347 }
4348
4349 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
4350 {
4351 int ret;
4352 struct kvm_userspace_memory_region tss_mem = {
4353 .slot = TSS_PRIVATE_MEMSLOT,
4354 .guest_phys_addr = addr,
4355 .memory_size = PAGE_SIZE * 3,
4356 .flags = 0,
4357 };
4358
4359 ret = kvm_set_memory_region(kvm, &tss_mem);
4360 if (ret)
4361 return ret;
4362 kvm->arch.tss_addr = addr;
4363 if (!init_rmode_tss(kvm))
4364 return -ENOMEM;
4365
4366 return 0;
4367 }
4368
4369 static bool rmode_exception(struct kvm_vcpu *vcpu, int vec)
4370 {
4371 switch (vec) {
4372 case BP_VECTOR:
4373 /*
4374 * Update instruction length as we may reinject the exception
4375 * from user space while in guest debugging mode.
4376 */
4377 to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
4378 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4379 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
4380 return false;
4381 /* fall through */
4382 case DB_VECTOR:
4383 if (vcpu->guest_debug &
4384 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
4385 return false;
4386 /* fall through */
4387 case DE_VECTOR:
4388 case OF_VECTOR:
4389 case BR_VECTOR:
4390 case UD_VECTOR:
4391 case DF_VECTOR:
4392 case SS_VECTOR:
4393 case GP_VECTOR:
4394 case MF_VECTOR:
4395 return true;
4396 break;
4397 }
4398 return false;
4399 }
4400
4401 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
4402 int vec, u32 err_code)
4403 {
4404 /*
4405 * Instruction with address size override prefix opcode 0x67
4406 * Cause the #SS fault with 0 error code in VM86 mode.
4407 */
4408 if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0) {
4409 if (emulate_instruction(vcpu, 0) == EMULATE_DONE) {
4410 if (vcpu->arch.halt_request) {
4411 vcpu->arch.halt_request = 0;
4412 return kvm_emulate_halt(vcpu);
4413 }
4414 return 1;
4415 }
4416 return 0;
4417 }
4418
4419 /*
4420 * Forward all other exceptions that are valid in real mode.
4421 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
4422 * the required debugging infrastructure rework.
4423 */
4424 kvm_queue_exception(vcpu, vec);
4425 return 1;
4426 }
4427
4428 /*
4429 * Trigger machine check on the host. We assume all the MSRs are already set up
4430 * by the CPU and that we still run on the same CPU as the MCE occurred on.
4431 * We pass a fake environment to the machine check handler because we want
4432 * the guest to be always treated like user space, no matter what context
4433 * it used internally.
4434 */
4435 static void kvm_machine_check(void)
4436 {
4437 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
4438 struct pt_regs regs = {
4439 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
4440 .flags = X86_EFLAGS_IF,
4441 };
4442
4443 do_machine_check(&regs, 0);
4444 #endif
4445 }
4446
4447 static int handle_machine_check(struct kvm_vcpu *vcpu)
4448 {
4449 /* already handled by vcpu_run */
4450 return 1;
4451 }
4452
4453 static int handle_exception(struct kvm_vcpu *vcpu)
4454 {
4455 struct vcpu_vmx *vmx = to_vmx(vcpu);
4456 struct kvm_run *kvm_run = vcpu->run;
4457 u32 intr_info, ex_no, error_code;
4458 unsigned long cr2, rip, dr6;
4459 u32 vect_info;
4460 enum emulation_result er;
4461
4462 vect_info = vmx->idt_vectoring_info;
4463 intr_info = vmx->exit_intr_info;
4464
4465 if (is_machine_check(intr_info))
4466 return handle_machine_check(vcpu);
4467
4468 if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR)
4469 return 1; /* already handled by vmx_vcpu_run() */
4470
4471 if (is_no_device(intr_info)) {
4472 vmx_fpu_activate(vcpu);
4473 return 1;
4474 }
4475
4476 if (is_invalid_opcode(intr_info)) {
4477 er = emulate_instruction(vcpu, EMULTYPE_TRAP_UD);
4478 if (er != EMULATE_DONE)
4479 kvm_queue_exception(vcpu, UD_VECTOR);
4480 return 1;
4481 }
4482
4483 error_code = 0;
4484 if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
4485 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
4486
4487 /*
4488 * The #PF with PFEC.RSVD = 1 indicates the guest is accessing
4489 * MMIO, it is better to report an internal error.
4490 * See the comments in vmx_handle_exit.
4491 */
4492 if ((vect_info & VECTORING_INFO_VALID_MASK) &&
4493 !(is_page_fault(intr_info) && !(error_code & PFERR_RSVD_MASK))) {
4494 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4495 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
4496 vcpu->run->internal.ndata = 2;
4497 vcpu->run->internal.data[0] = vect_info;
4498 vcpu->run->internal.data[1] = intr_info;
4499 return 0;
4500 }
4501
4502 if (is_page_fault(intr_info)) {
4503 /* EPT won't cause page fault directly */
4504 BUG_ON(enable_ept);
4505 cr2 = vmcs_readl(EXIT_QUALIFICATION);
4506 trace_kvm_page_fault(cr2, error_code);
4507
4508 if (kvm_event_needs_reinjection(vcpu))
4509 kvm_mmu_unprotect_page_virt(vcpu, cr2);
4510 return kvm_mmu_page_fault(vcpu, cr2, error_code, NULL, 0);
4511 }
4512
4513 ex_no = intr_info & INTR_INFO_VECTOR_MASK;
4514
4515 if (vmx->rmode.vm86_active && rmode_exception(vcpu, ex_no))
4516 return handle_rmode_exception(vcpu, ex_no, error_code);
4517
4518 switch (ex_no) {
4519 case DB_VECTOR:
4520 dr6 = vmcs_readl(EXIT_QUALIFICATION);
4521 if (!(vcpu->guest_debug &
4522 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
4523 vcpu->arch.dr6 = dr6 | DR6_FIXED_1;
4524 kvm_queue_exception(vcpu, DB_VECTOR);
4525 return 1;
4526 }
4527 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
4528 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
4529 /* fall through */
4530 case BP_VECTOR:
4531 /*
4532 * Update instruction length as we may reinject #BP from
4533 * user space while in guest debugging mode. Reading it for
4534 * #DB as well causes no harm, it is not used in that case.
4535 */
4536 vmx->vcpu.arch.event_exit_inst_len =
4537 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4538 kvm_run->exit_reason = KVM_EXIT_DEBUG;
4539 rip = kvm_rip_read(vcpu);
4540 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
4541 kvm_run->debug.arch.exception = ex_no;
4542 break;
4543 default:
4544 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
4545 kvm_run->ex.exception = ex_no;
4546 kvm_run->ex.error_code = error_code;
4547 break;
4548 }
4549 return 0;
4550 }
4551
4552 static int handle_external_interrupt(struct kvm_vcpu *vcpu)
4553 {
4554 ++vcpu->stat.irq_exits;
4555 return 1;
4556 }
4557
4558 static int handle_triple_fault(struct kvm_vcpu *vcpu)
4559 {
4560 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
4561 return 0;
4562 }
4563
4564 static int handle_io(struct kvm_vcpu *vcpu)
4565 {
4566 unsigned long exit_qualification;
4567 int size, in, string;
4568 unsigned port;
4569
4570 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4571 string = (exit_qualification & 16) != 0;
4572 in = (exit_qualification & 8) != 0;
4573
4574 ++vcpu->stat.io_exits;
4575
4576 if (string || in)
4577 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
4578
4579 port = exit_qualification >> 16;
4580 size = (exit_qualification & 7) + 1;
4581 skip_emulated_instruction(vcpu);
4582
4583 return kvm_fast_pio_out(vcpu, size, port);
4584 }
4585
4586 static void
4587 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
4588 {
4589 /*
4590 * Patch in the VMCALL instruction:
4591 */
4592 hypercall[0] = 0x0f;
4593 hypercall[1] = 0x01;
4594 hypercall[2] = 0xc1;
4595 }
4596
4597 /* called to set cr0 as appropriate for a mov-to-cr0 exit. */
4598 static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
4599 {
4600 if (is_guest_mode(vcpu)) {
4601 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4602 unsigned long orig_val = val;
4603
4604 /*
4605 * We get here when L2 changed cr0 in a way that did not change
4606 * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
4607 * but did change L0 shadowed bits. So we first calculate the
4608 * effective cr0 value that L1 would like to write into the
4609 * hardware. It consists of the L2-owned bits from the new
4610 * value combined with the L1-owned bits from L1's guest_cr0.
4611 */
4612 val = (val & ~vmcs12->cr0_guest_host_mask) |
4613 (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask);
4614
4615 /* TODO: will have to take unrestricted guest mode into
4616 * account */
4617 if ((val & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON)
4618 return 1;
4619
4620 if (kvm_set_cr0(vcpu, val))
4621 return 1;
4622 vmcs_writel(CR0_READ_SHADOW, orig_val);
4623 return 0;
4624 } else {
4625 if (to_vmx(vcpu)->nested.vmxon &&
4626 ((val & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON))
4627 return 1;
4628 return kvm_set_cr0(vcpu, val);
4629 }
4630 }
4631
4632 static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
4633 {
4634 if (is_guest_mode(vcpu)) {
4635 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4636 unsigned long orig_val = val;
4637
4638 /* analogously to handle_set_cr0 */
4639 val = (val & ~vmcs12->cr4_guest_host_mask) |
4640 (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask);
4641 if (kvm_set_cr4(vcpu, val))
4642 return 1;
4643 vmcs_writel(CR4_READ_SHADOW, orig_val);
4644 return 0;
4645 } else
4646 return kvm_set_cr4(vcpu, val);
4647 }
4648
4649 /* called to set cr0 as approriate for clts instruction exit. */
4650 static void handle_clts(struct kvm_vcpu *vcpu)
4651 {
4652 if (is_guest_mode(vcpu)) {
4653 /*
4654 * We get here when L2 did CLTS, and L1 didn't shadow CR0.TS
4655 * but we did (!fpu_active). We need to keep GUEST_CR0.TS on,
4656 * just pretend it's off (also in arch.cr0 for fpu_activate).
4657 */
4658 vmcs_writel(CR0_READ_SHADOW,
4659 vmcs_readl(CR0_READ_SHADOW) & ~X86_CR0_TS);
4660 vcpu->arch.cr0 &= ~X86_CR0_TS;
4661 } else
4662 vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
4663 }
4664
4665 static int handle_cr(struct kvm_vcpu *vcpu)
4666 {
4667 unsigned long exit_qualification, val;
4668 int cr;
4669 int reg;
4670 int err;
4671
4672 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4673 cr = exit_qualification & 15;
4674 reg = (exit_qualification >> 8) & 15;
4675 switch ((exit_qualification >> 4) & 3) {
4676 case 0: /* mov to cr */
4677 val = kvm_register_read(vcpu, reg);
4678 trace_kvm_cr_write(cr, val);
4679 switch (cr) {
4680 case 0:
4681 err = handle_set_cr0(vcpu, val);
4682 kvm_complete_insn_gp(vcpu, err);
4683 return 1;
4684 case 3:
4685 err = kvm_set_cr3(vcpu, val);
4686 kvm_complete_insn_gp(vcpu, err);
4687 return 1;
4688 case 4:
4689 err = handle_set_cr4(vcpu, val);
4690 kvm_complete_insn_gp(vcpu, err);
4691 return 1;
4692 case 8: {
4693 u8 cr8_prev = kvm_get_cr8(vcpu);
4694 u8 cr8 = kvm_register_read(vcpu, reg);
4695 err = kvm_set_cr8(vcpu, cr8);
4696 kvm_complete_insn_gp(vcpu, err);
4697 if (irqchip_in_kernel(vcpu->kvm))
4698 return 1;
4699 if (cr8_prev <= cr8)
4700 return 1;
4701 vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
4702 return 0;
4703 }
4704 }
4705 break;
4706 case 2: /* clts */
4707 handle_clts(vcpu);
4708 trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
4709 skip_emulated_instruction(vcpu);
4710 vmx_fpu_activate(vcpu);
4711 return 1;
4712 case 1: /*mov from cr*/
4713 switch (cr) {
4714 case 3:
4715 val = kvm_read_cr3(vcpu);
4716 kvm_register_write(vcpu, reg, val);
4717 trace_kvm_cr_read(cr, val);
4718 skip_emulated_instruction(vcpu);
4719 return 1;
4720 case 8:
4721 val = kvm_get_cr8(vcpu);
4722 kvm_register_write(vcpu, reg, val);
4723 trace_kvm_cr_read(cr, val);
4724 skip_emulated_instruction(vcpu);
4725 return 1;
4726 }
4727 break;
4728 case 3: /* lmsw */
4729 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
4730 trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
4731 kvm_lmsw(vcpu, val);
4732
4733 skip_emulated_instruction(vcpu);
4734 return 1;
4735 default:
4736 break;
4737 }
4738 vcpu->run->exit_reason = 0;
4739 vcpu_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
4740 (int)(exit_qualification >> 4) & 3, cr);
4741 return 0;
4742 }
4743
4744 static int handle_dr(struct kvm_vcpu *vcpu)
4745 {
4746 unsigned long exit_qualification;
4747 int dr, reg;
4748
4749 /* Do not handle if the CPL > 0, will trigger GP on re-entry */
4750 if (!kvm_require_cpl(vcpu, 0))
4751 return 1;
4752 dr = vmcs_readl(GUEST_DR7);
4753 if (dr & DR7_GD) {
4754 /*
4755 * As the vm-exit takes precedence over the debug trap, we
4756 * need to emulate the latter, either for the host or the
4757 * guest debugging itself.
4758 */
4759 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
4760 vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
4761 vcpu->run->debug.arch.dr7 = dr;
4762 vcpu->run->debug.arch.pc =
4763 vmcs_readl(GUEST_CS_BASE) +
4764 vmcs_readl(GUEST_RIP);
4765 vcpu->run->debug.arch.exception = DB_VECTOR;
4766 vcpu->run->exit_reason = KVM_EXIT_DEBUG;
4767 return 0;
4768 } else {
4769 vcpu->arch.dr7 &= ~DR7_GD;
4770 vcpu->arch.dr6 |= DR6_BD;
4771 vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
4772 kvm_queue_exception(vcpu, DB_VECTOR);
4773 return 1;
4774 }
4775 }
4776
4777 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4778 dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
4779 reg = DEBUG_REG_ACCESS_REG(exit_qualification);
4780 if (exit_qualification & TYPE_MOV_FROM_DR) {
4781 unsigned long val;
4782 if (!kvm_get_dr(vcpu, dr, &val))
4783 kvm_register_write(vcpu, reg, val);
4784 } else
4785 kvm_set_dr(vcpu, dr, vcpu->arch.regs[reg]);
4786 skip_emulated_instruction(vcpu);
4787 return 1;
4788 }
4789
4790 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
4791 {
4792 vmcs_writel(GUEST_DR7, val);
4793 }
4794
4795 static int handle_cpuid(struct kvm_vcpu *vcpu)
4796 {
4797 kvm_emulate_cpuid(vcpu);
4798 return 1;
4799 }
4800
4801 static int handle_rdmsr(struct kvm_vcpu *vcpu)
4802 {
4803 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
4804 u64 data;
4805
4806 if (vmx_get_msr(vcpu, ecx, &data)) {
4807 trace_kvm_msr_read_ex(ecx);
4808 kvm_inject_gp(vcpu, 0);
4809 return 1;
4810 }
4811
4812 trace_kvm_msr_read(ecx, data);
4813
4814 /* FIXME: handling of bits 32:63 of rax, rdx */
4815 vcpu->arch.regs[VCPU_REGS_RAX] = data & -1u;
4816 vcpu->arch.regs[VCPU_REGS_RDX] = (data >> 32) & -1u;
4817 skip_emulated_instruction(vcpu);
4818 return 1;
4819 }
4820
4821 static int handle_wrmsr(struct kvm_vcpu *vcpu)
4822 {
4823 struct msr_data msr;
4824 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
4825 u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
4826 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
4827
4828 msr.data = data;
4829 msr.index = ecx;
4830 msr.host_initiated = false;
4831 if (vmx_set_msr(vcpu, &msr) != 0) {
4832 trace_kvm_msr_write_ex(ecx, data);
4833 kvm_inject_gp(vcpu, 0);
4834 return 1;
4835 }
4836
4837 trace_kvm_msr_write(ecx, data);
4838 skip_emulated_instruction(vcpu);
4839 return 1;
4840 }
4841
4842 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
4843 {
4844 kvm_make_request(KVM_REQ_EVENT, vcpu);
4845 return 1;
4846 }
4847
4848 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
4849 {
4850 u32 cpu_based_vm_exec_control;
4851
4852 /* clear pending irq */
4853 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
4854 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
4855 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
4856
4857 kvm_make_request(KVM_REQ_EVENT, vcpu);
4858
4859 ++vcpu->stat.irq_window_exits;
4860
4861 /*
4862 * If the user space waits to inject interrupts, exit as soon as
4863 * possible
4864 */
4865 if (!irqchip_in_kernel(vcpu->kvm) &&
4866 vcpu->run->request_interrupt_window &&
4867 !kvm_cpu_has_interrupt(vcpu)) {
4868 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
4869 return 0;
4870 }
4871 return 1;
4872 }
4873
4874 static int handle_halt(struct kvm_vcpu *vcpu)
4875 {
4876 skip_emulated_instruction(vcpu);
4877 return kvm_emulate_halt(vcpu);
4878 }
4879
4880 static int handle_vmcall(struct kvm_vcpu *vcpu)
4881 {
4882 skip_emulated_instruction(vcpu);
4883 kvm_emulate_hypercall(vcpu);
4884 return 1;
4885 }
4886
4887 static int handle_invd(struct kvm_vcpu *vcpu)
4888 {
4889 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
4890 }
4891
4892 static int handle_invlpg(struct kvm_vcpu *vcpu)
4893 {
4894 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4895
4896 kvm_mmu_invlpg(vcpu, exit_qualification);
4897 skip_emulated_instruction(vcpu);
4898 return 1;
4899 }
4900
4901 static int handle_rdpmc(struct kvm_vcpu *vcpu)
4902 {
4903 int err;
4904
4905 err = kvm_rdpmc(vcpu);
4906 kvm_complete_insn_gp(vcpu, err);
4907
4908 return 1;
4909 }
4910
4911 static int handle_wbinvd(struct kvm_vcpu *vcpu)
4912 {
4913 skip_emulated_instruction(vcpu);
4914 kvm_emulate_wbinvd(vcpu);
4915 return 1;
4916 }
4917
4918 static int handle_xsetbv(struct kvm_vcpu *vcpu)
4919 {
4920 u64 new_bv = kvm_read_edx_eax(vcpu);
4921 u32 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
4922
4923 if (kvm_set_xcr(vcpu, index, new_bv) == 0)
4924 skip_emulated_instruction(vcpu);
4925 return 1;
4926 }
4927
4928 static int handle_apic_access(struct kvm_vcpu *vcpu)
4929 {
4930 if (likely(fasteoi)) {
4931 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4932 int access_type, offset;
4933
4934 access_type = exit_qualification & APIC_ACCESS_TYPE;
4935 offset = exit_qualification & APIC_ACCESS_OFFSET;
4936 /*
4937 * Sane guest uses MOV to write EOI, with written value
4938 * not cared. So make a short-circuit here by avoiding
4939 * heavy instruction emulation.
4940 */
4941 if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
4942 (offset == APIC_EOI)) {
4943 kvm_lapic_set_eoi(vcpu);
4944 skip_emulated_instruction(vcpu);
4945 return 1;
4946 }
4947 }
4948 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
4949 }
4950
4951 static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu)
4952 {
4953 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4954 int vector = exit_qualification & 0xff;
4955
4956 /* EOI-induced VM exit is trap-like and thus no need to adjust IP */
4957 kvm_apic_set_eoi_accelerated(vcpu, vector);
4958 return 1;
4959 }
4960
4961 static int handle_apic_write(struct kvm_vcpu *vcpu)
4962 {
4963 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4964 u32 offset = exit_qualification & 0xfff;
4965
4966 /* APIC-write VM exit is trap-like and thus no need to adjust IP */
4967 kvm_apic_write_nodecode(vcpu, offset);
4968 return 1;
4969 }
4970
4971 static int handle_task_switch(struct kvm_vcpu *vcpu)
4972 {
4973 struct vcpu_vmx *vmx = to_vmx(vcpu);
4974 unsigned long exit_qualification;
4975 bool has_error_code = false;
4976 u32 error_code = 0;
4977 u16 tss_selector;
4978 int reason, type, idt_v, idt_index;
4979
4980 idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
4981 idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK);
4982 type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
4983
4984 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4985
4986 reason = (u32)exit_qualification >> 30;
4987 if (reason == TASK_SWITCH_GATE && idt_v) {
4988 switch (type) {
4989 case INTR_TYPE_NMI_INTR:
4990 vcpu->arch.nmi_injected = false;
4991 vmx_set_nmi_mask(vcpu, true);
4992 break;
4993 case INTR_TYPE_EXT_INTR:
4994 case INTR_TYPE_SOFT_INTR:
4995 kvm_clear_interrupt_queue(vcpu);
4996 break;
4997 case INTR_TYPE_HARD_EXCEPTION:
4998 if (vmx->idt_vectoring_info &
4999 VECTORING_INFO_DELIVER_CODE_MASK) {
5000 has_error_code = true;
5001 error_code =
5002 vmcs_read32(IDT_VECTORING_ERROR_CODE);
5003 }
5004 /* fall through */
5005 case INTR_TYPE_SOFT_EXCEPTION:
5006 kvm_clear_exception_queue(vcpu);
5007 break;
5008 default:
5009 break;
5010 }
5011 }
5012 tss_selector = exit_qualification;
5013
5014 if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
5015 type != INTR_TYPE_EXT_INTR &&
5016 type != INTR_TYPE_NMI_INTR))
5017 skip_emulated_instruction(vcpu);
5018
5019 if (kvm_task_switch(vcpu, tss_selector,
5020 type == INTR_TYPE_SOFT_INTR ? idt_index : -1, reason,
5021 has_error_code, error_code) == EMULATE_FAIL) {
5022 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5023 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5024 vcpu->run->internal.ndata = 0;
5025 return 0;
5026 }
5027
5028 /* clear all local breakpoint enable flags */
5029 vmcs_writel(GUEST_DR7, vmcs_readl(GUEST_DR7) & ~55);
5030
5031 /*
5032 * TODO: What about debug traps on tss switch?
5033 * Are we supposed to inject them and update dr6?
5034 */
5035
5036 return 1;
5037 }
5038
5039 static int handle_ept_violation(struct kvm_vcpu *vcpu)
5040 {
5041 unsigned long exit_qualification;
5042 gpa_t gpa;
5043 u32 error_code;
5044 int gla_validity;
5045
5046 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5047
5048 gla_validity = (exit_qualification >> 7) & 0x3;
5049 if (gla_validity != 0x3 && gla_validity != 0x1 && gla_validity != 0) {
5050 printk(KERN_ERR "EPT: Handling EPT violation failed!\n");
5051 printk(KERN_ERR "EPT: GPA: 0x%lx, GVA: 0x%lx\n",
5052 (long unsigned int)vmcs_read64(GUEST_PHYSICAL_ADDRESS),
5053 vmcs_readl(GUEST_LINEAR_ADDRESS));
5054 printk(KERN_ERR "EPT: Exit qualification is 0x%lx\n",
5055 (long unsigned int)exit_qualification);
5056 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
5057 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_VIOLATION;
5058 return 0;
5059 }
5060
5061 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5062 trace_kvm_page_fault(gpa, exit_qualification);
5063
5064 /* It is a write fault? */
5065 error_code = exit_qualification & (1U << 1);
5066 /* ept page table is present? */
5067 error_code |= (exit_qualification >> 3) & 0x1;
5068
5069 return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0);
5070 }
5071
5072 static u64 ept_rsvd_mask(u64 spte, int level)
5073 {
5074 int i;
5075 u64 mask = 0;
5076
5077 for (i = 51; i > boot_cpu_data.x86_phys_bits; i--)
5078 mask |= (1ULL << i);
5079
5080 if (level > 2)
5081 /* bits 7:3 reserved */
5082 mask |= 0xf8;
5083 else if (level == 2) {
5084 if (spte & (1ULL << 7))
5085 /* 2MB ref, bits 20:12 reserved */
5086 mask |= 0x1ff000;
5087 else
5088 /* bits 6:3 reserved */
5089 mask |= 0x78;
5090 }
5091
5092 return mask;
5093 }
5094
5095 static void ept_misconfig_inspect_spte(struct kvm_vcpu *vcpu, u64 spte,
5096 int level)
5097 {
5098 printk(KERN_ERR "%s: spte 0x%llx level %d\n", __func__, spte, level);
5099
5100 /* 010b (write-only) */
5101 WARN_ON((spte & 0x7) == 0x2);
5102
5103 /* 110b (write/execute) */
5104 WARN_ON((spte & 0x7) == 0x6);
5105
5106 /* 100b (execute-only) and value not supported by logical processor */
5107 if (!cpu_has_vmx_ept_execute_only())
5108 WARN_ON((spte & 0x7) == 0x4);
5109
5110 /* not 000b */
5111 if ((spte & 0x7)) {
5112 u64 rsvd_bits = spte & ept_rsvd_mask(spte, level);
5113
5114 if (rsvd_bits != 0) {
5115 printk(KERN_ERR "%s: rsvd_bits = 0x%llx\n",
5116 __func__, rsvd_bits);
5117 WARN_ON(1);
5118 }
5119
5120 if (level == 1 || (level == 2 && (spte & (1ULL << 7)))) {
5121 u64 ept_mem_type = (spte & 0x38) >> 3;
5122
5123 if (ept_mem_type == 2 || ept_mem_type == 3 ||
5124 ept_mem_type == 7) {
5125 printk(KERN_ERR "%s: ept_mem_type=0x%llx\n",
5126 __func__, ept_mem_type);
5127 WARN_ON(1);
5128 }
5129 }
5130 }
5131 }
5132
5133 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
5134 {
5135 u64 sptes[4];
5136 int nr_sptes, i, ret;
5137 gpa_t gpa;
5138
5139 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5140
5141 ret = handle_mmio_page_fault_common(vcpu, gpa, true);
5142 if (likely(ret == 1))
5143 return x86_emulate_instruction(vcpu, gpa, 0, NULL, 0) ==
5144 EMULATE_DONE;
5145 if (unlikely(!ret))
5146 return 1;
5147
5148 /* It is the real ept misconfig */
5149 printk(KERN_ERR "EPT: Misconfiguration.\n");
5150 printk(KERN_ERR "EPT: GPA: 0x%llx\n", gpa);
5151
5152 nr_sptes = kvm_mmu_get_spte_hierarchy(vcpu, gpa, sptes);
5153
5154 for (i = PT64_ROOT_LEVEL; i > PT64_ROOT_LEVEL - nr_sptes; --i)
5155 ept_misconfig_inspect_spte(vcpu, sptes[i-1], i);
5156
5157 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
5158 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_MISCONFIG;
5159
5160 return 0;
5161 }
5162
5163 static int handle_nmi_window(struct kvm_vcpu *vcpu)
5164 {
5165 u32 cpu_based_vm_exec_control;
5166
5167 /* clear pending NMI */
5168 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5169 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
5170 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
5171 ++vcpu->stat.nmi_window_exits;
5172 kvm_make_request(KVM_REQ_EVENT, vcpu);
5173
5174 return 1;
5175 }
5176
5177 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
5178 {
5179 struct vcpu_vmx *vmx = to_vmx(vcpu);
5180 enum emulation_result err = EMULATE_DONE;
5181 int ret = 1;
5182 u32 cpu_exec_ctrl;
5183 bool intr_window_requested;
5184 unsigned count = 130;
5185
5186 cpu_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5187 intr_window_requested = cpu_exec_ctrl & CPU_BASED_VIRTUAL_INTR_PENDING;
5188
5189 while (!guest_state_valid(vcpu) && count-- != 0) {
5190 if (intr_window_requested && vmx_interrupt_allowed(vcpu))
5191 return handle_interrupt_window(&vmx->vcpu);
5192
5193 if (test_bit(KVM_REQ_EVENT, &vcpu->requests))
5194 return 1;
5195
5196 err = emulate_instruction(vcpu, EMULTYPE_NO_REEXECUTE);
5197
5198 if (err == EMULATE_DO_MMIO) {
5199 ret = 0;
5200 goto out;
5201 }
5202
5203 if (err != EMULATE_DONE) {
5204 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5205 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5206 vcpu->run->internal.ndata = 0;
5207 return 0;
5208 }
5209
5210 if (signal_pending(current))
5211 goto out;
5212 if (need_resched())
5213 schedule();
5214 }
5215
5216 vmx->emulation_required = emulation_required(vcpu);
5217 out:
5218 return ret;
5219 }
5220
5221 /*
5222 * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
5223 * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
5224 */
5225 static int handle_pause(struct kvm_vcpu *vcpu)
5226 {
5227 skip_emulated_instruction(vcpu);
5228 kvm_vcpu_on_spin(vcpu);
5229
5230 return 1;
5231 }
5232
5233 static int handle_invalid_op(struct kvm_vcpu *vcpu)
5234 {
5235 kvm_queue_exception(vcpu, UD_VECTOR);
5236 return 1;
5237 }
5238
5239 /*
5240 * To run an L2 guest, we need a vmcs02 based on the L1-specified vmcs12.
5241 * We could reuse a single VMCS for all the L2 guests, but we also want the
5242 * option to allocate a separate vmcs02 for each separate loaded vmcs12 - this
5243 * allows keeping them loaded on the processor, and in the future will allow
5244 * optimizations where prepare_vmcs02 doesn't need to set all the fields on
5245 * every entry if they never change.
5246 * So we keep, in vmx->nested.vmcs02_pool, a cache of size VMCS02_POOL_SIZE
5247 * (>=0) with a vmcs02 for each recently loaded vmcs12s, most recent first.
5248 *
5249 * The following functions allocate and free a vmcs02 in this pool.
5250 */
5251
5252 /* Get a VMCS from the pool to use as vmcs02 for the current vmcs12. */
5253 static struct loaded_vmcs *nested_get_current_vmcs02(struct vcpu_vmx *vmx)
5254 {
5255 struct vmcs02_list *item;
5256 list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
5257 if (item->vmptr == vmx->nested.current_vmptr) {
5258 list_move(&item->list, &vmx->nested.vmcs02_pool);
5259 return &item->vmcs02;
5260 }
5261
5262 if (vmx->nested.vmcs02_num >= max(VMCS02_POOL_SIZE, 1)) {
5263 /* Recycle the least recently used VMCS. */
5264 item = list_entry(vmx->nested.vmcs02_pool.prev,
5265 struct vmcs02_list, list);
5266 item->vmptr = vmx->nested.current_vmptr;
5267 list_move(&item->list, &vmx->nested.vmcs02_pool);
5268 return &item->vmcs02;
5269 }
5270
5271 /* Create a new VMCS */
5272 item = kmalloc(sizeof(struct vmcs02_list), GFP_KERNEL);
5273 if (!item)
5274 return NULL;
5275 item->vmcs02.vmcs = alloc_vmcs();
5276 if (!item->vmcs02.vmcs) {
5277 kfree(item);
5278 return NULL;
5279 }
5280 loaded_vmcs_init(&item->vmcs02);
5281 item->vmptr = vmx->nested.current_vmptr;
5282 list_add(&(item->list), &(vmx->nested.vmcs02_pool));
5283 vmx->nested.vmcs02_num++;
5284 return &item->vmcs02;
5285 }
5286
5287 /* Free and remove from pool a vmcs02 saved for a vmcs12 (if there is one) */
5288 static void nested_free_vmcs02(struct vcpu_vmx *vmx, gpa_t vmptr)
5289 {
5290 struct vmcs02_list *item;
5291 list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
5292 if (item->vmptr == vmptr) {
5293 free_loaded_vmcs(&item->vmcs02);
5294 list_del(&item->list);
5295 kfree(item);
5296 vmx->nested.vmcs02_num--;
5297 return;
5298 }
5299 }
5300
5301 /*
5302 * Free all VMCSs saved for this vcpu, except the one pointed by
5303 * vmx->loaded_vmcs. These include the VMCSs in vmcs02_pool (except the one
5304 * currently used, if running L2), and vmcs01 when running L2.
5305 */
5306 static void nested_free_all_saved_vmcss(struct vcpu_vmx *vmx)
5307 {
5308 struct vmcs02_list *item, *n;
5309 list_for_each_entry_safe(item, n, &vmx->nested.vmcs02_pool, list) {
5310 if (vmx->loaded_vmcs != &item->vmcs02)
5311 free_loaded_vmcs(&item->vmcs02);
5312 list_del(&item->list);
5313 kfree(item);
5314 }
5315 vmx->nested.vmcs02_num = 0;
5316
5317 if (vmx->loaded_vmcs != &vmx->vmcs01)
5318 free_loaded_vmcs(&vmx->vmcs01);
5319 }
5320
5321 /*
5322 * Emulate the VMXON instruction.
5323 * Currently, we just remember that VMX is active, and do not save or even
5324 * inspect the argument to VMXON (the so-called "VMXON pointer") because we
5325 * do not currently need to store anything in that guest-allocated memory
5326 * region. Consequently, VMCLEAR and VMPTRLD also do not verify that the their
5327 * argument is different from the VMXON pointer (which the spec says they do).
5328 */
5329 static int handle_vmon(struct kvm_vcpu *vcpu)
5330 {
5331 struct kvm_segment cs;
5332 struct vcpu_vmx *vmx = to_vmx(vcpu);
5333
5334 /* The Intel VMX Instruction Reference lists a bunch of bits that
5335 * are prerequisite to running VMXON, most notably cr4.VMXE must be
5336 * set to 1 (see vmx_set_cr4() for when we allow the guest to set this).
5337 * Otherwise, we should fail with #UD. We test these now:
5338 */
5339 if (!kvm_read_cr4_bits(vcpu, X86_CR4_VMXE) ||
5340 !kvm_read_cr0_bits(vcpu, X86_CR0_PE) ||
5341 (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
5342 kvm_queue_exception(vcpu, UD_VECTOR);
5343 return 1;
5344 }
5345
5346 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
5347 if (is_long_mode(vcpu) && !cs.l) {
5348 kvm_queue_exception(vcpu, UD_VECTOR);
5349 return 1;
5350 }
5351
5352 if (vmx_get_cpl(vcpu)) {
5353 kvm_inject_gp(vcpu, 0);
5354 return 1;
5355 }
5356
5357 INIT_LIST_HEAD(&(vmx->nested.vmcs02_pool));
5358 vmx->nested.vmcs02_num = 0;
5359
5360 vmx->nested.vmxon = true;
5361
5362 skip_emulated_instruction(vcpu);
5363 return 1;
5364 }
5365
5366 /*
5367 * Intel's VMX Instruction Reference specifies a common set of prerequisites
5368 * for running VMX instructions (except VMXON, whose prerequisites are
5369 * slightly different). It also specifies what exception to inject otherwise.
5370 */
5371 static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
5372 {
5373 struct kvm_segment cs;
5374 struct vcpu_vmx *vmx = to_vmx(vcpu);
5375
5376 if (!vmx->nested.vmxon) {
5377 kvm_queue_exception(vcpu, UD_VECTOR);
5378 return 0;
5379 }
5380
5381 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
5382 if ((vmx_get_rflags(vcpu) & X86_EFLAGS_VM) ||
5383 (is_long_mode(vcpu) && !cs.l)) {
5384 kvm_queue_exception(vcpu, UD_VECTOR);
5385 return 0;
5386 }
5387
5388 if (vmx_get_cpl(vcpu)) {
5389 kvm_inject_gp(vcpu, 0);
5390 return 0;
5391 }
5392
5393 return 1;
5394 }
5395
5396 /*
5397 * Free whatever needs to be freed from vmx->nested when L1 goes down, or
5398 * just stops using VMX.
5399 */
5400 static void free_nested(struct vcpu_vmx *vmx)
5401 {
5402 if (!vmx->nested.vmxon)
5403 return;
5404 vmx->nested.vmxon = false;
5405 if (vmx->nested.current_vmptr != -1ull) {
5406 kunmap(vmx->nested.current_vmcs12_page);
5407 nested_release_page(vmx->nested.current_vmcs12_page);
5408 vmx->nested.current_vmptr = -1ull;
5409 vmx->nested.current_vmcs12 = NULL;
5410 }
5411 /* Unpin physical memory we referred to in current vmcs02 */
5412 if (vmx->nested.apic_access_page) {
5413 nested_release_page(vmx->nested.apic_access_page);
5414 vmx->nested.apic_access_page = 0;
5415 }
5416
5417 nested_free_all_saved_vmcss(vmx);
5418 }
5419
5420 /* Emulate the VMXOFF instruction */
5421 static int handle_vmoff(struct kvm_vcpu *vcpu)
5422 {
5423 if (!nested_vmx_check_permission(vcpu))
5424 return 1;
5425 free_nested(to_vmx(vcpu));
5426 skip_emulated_instruction(vcpu);
5427 return 1;
5428 }
5429
5430 /*
5431 * Decode the memory-address operand of a vmx instruction, as recorded on an
5432 * exit caused by such an instruction (run by a guest hypervisor).
5433 * On success, returns 0. When the operand is invalid, returns 1 and throws
5434 * #UD or #GP.
5435 */
5436 static int get_vmx_mem_address(struct kvm_vcpu *vcpu,
5437 unsigned long exit_qualification,
5438 u32 vmx_instruction_info, gva_t *ret)
5439 {
5440 /*
5441 * According to Vol. 3B, "Information for VM Exits Due to Instruction
5442 * Execution", on an exit, vmx_instruction_info holds most of the
5443 * addressing components of the operand. Only the displacement part
5444 * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
5445 * For how an actual address is calculated from all these components,
5446 * refer to Vol. 1, "Operand Addressing".
5447 */
5448 int scaling = vmx_instruction_info & 3;
5449 int addr_size = (vmx_instruction_info >> 7) & 7;
5450 bool is_reg = vmx_instruction_info & (1u << 10);
5451 int seg_reg = (vmx_instruction_info >> 15) & 7;
5452 int index_reg = (vmx_instruction_info >> 18) & 0xf;
5453 bool index_is_valid = !(vmx_instruction_info & (1u << 22));
5454 int base_reg = (vmx_instruction_info >> 23) & 0xf;
5455 bool base_is_valid = !(vmx_instruction_info & (1u << 27));
5456
5457 if (is_reg) {
5458 kvm_queue_exception(vcpu, UD_VECTOR);
5459 return 1;
5460 }
5461
5462 /* Addr = segment_base + offset */
5463 /* offset = base + [index * scale] + displacement */
5464 *ret = vmx_get_segment_base(vcpu, seg_reg);
5465 if (base_is_valid)
5466 *ret += kvm_register_read(vcpu, base_reg);
5467 if (index_is_valid)
5468 *ret += kvm_register_read(vcpu, index_reg)<<scaling;
5469 *ret += exit_qualification; /* holds the displacement */
5470
5471 if (addr_size == 1) /* 32 bit */
5472 *ret &= 0xffffffff;
5473
5474 /*
5475 * TODO: throw #GP (and return 1) in various cases that the VM*
5476 * instructions require it - e.g., offset beyond segment limit,
5477 * unusable or unreadable/unwritable segment, non-canonical 64-bit
5478 * address, and so on. Currently these are not checked.
5479 */
5480 return 0;
5481 }
5482
5483 /*
5484 * The following 3 functions, nested_vmx_succeed()/failValid()/failInvalid(),
5485 * set the success or error code of an emulated VMX instruction, as specified
5486 * by Vol 2B, VMX Instruction Reference, "Conventions".
5487 */
5488 static void nested_vmx_succeed(struct kvm_vcpu *vcpu)
5489 {
5490 vmx_set_rflags(vcpu, vmx_get_rflags(vcpu)
5491 & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
5492 X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF));
5493 }
5494
5495 static void nested_vmx_failInvalid(struct kvm_vcpu *vcpu)
5496 {
5497 vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
5498 & ~(X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
5499 X86_EFLAGS_SF | X86_EFLAGS_OF))
5500 | X86_EFLAGS_CF);
5501 }
5502
5503 static void nested_vmx_failValid(struct kvm_vcpu *vcpu,
5504 u32 vm_instruction_error)
5505 {
5506 if (to_vmx(vcpu)->nested.current_vmptr == -1ull) {
5507 /*
5508 * failValid writes the error number to the current VMCS, which
5509 * can't be done there isn't a current VMCS.
5510 */
5511 nested_vmx_failInvalid(vcpu);
5512 return;
5513 }
5514 vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
5515 & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
5516 X86_EFLAGS_SF | X86_EFLAGS_OF))
5517 | X86_EFLAGS_ZF);
5518 get_vmcs12(vcpu)->vm_instruction_error = vm_instruction_error;
5519 }
5520
5521 /* Emulate the VMCLEAR instruction */
5522 static int handle_vmclear(struct kvm_vcpu *vcpu)
5523 {
5524 struct vcpu_vmx *vmx = to_vmx(vcpu);
5525 gva_t gva;
5526 gpa_t vmptr;
5527 struct vmcs12 *vmcs12;
5528 struct page *page;
5529 struct x86_exception e;
5530
5531 if (!nested_vmx_check_permission(vcpu))
5532 return 1;
5533
5534 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
5535 vmcs_read32(VMX_INSTRUCTION_INFO), &gva))
5536 return 1;
5537
5538 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &vmptr,
5539 sizeof(vmptr), &e)) {
5540 kvm_inject_page_fault(vcpu, &e);
5541 return 1;
5542 }
5543
5544 if (!IS_ALIGNED(vmptr, PAGE_SIZE)) {
5545 nested_vmx_failValid(vcpu, VMXERR_VMCLEAR_INVALID_ADDRESS);
5546 skip_emulated_instruction(vcpu);
5547 return 1;
5548 }
5549
5550 if (vmptr == vmx->nested.current_vmptr) {
5551 kunmap(vmx->nested.current_vmcs12_page);
5552 nested_release_page(vmx->nested.current_vmcs12_page);
5553 vmx->nested.current_vmptr = -1ull;
5554 vmx->nested.current_vmcs12 = NULL;
5555 }
5556
5557 page = nested_get_page(vcpu, vmptr);
5558 if (page == NULL) {
5559 /*
5560 * For accurate processor emulation, VMCLEAR beyond available
5561 * physical memory should do nothing at all. However, it is
5562 * possible that a nested vmx bug, not a guest hypervisor bug,
5563 * resulted in this case, so let's shut down before doing any
5564 * more damage:
5565 */
5566 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5567 return 1;
5568 }
5569 vmcs12 = kmap(page);
5570 vmcs12->launch_state = 0;
5571 kunmap(page);
5572 nested_release_page(page);
5573
5574 nested_free_vmcs02(vmx, vmptr);
5575
5576 skip_emulated_instruction(vcpu);
5577 nested_vmx_succeed(vcpu);
5578 return 1;
5579 }
5580
5581 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch);
5582
5583 /* Emulate the VMLAUNCH instruction */
5584 static int handle_vmlaunch(struct kvm_vcpu *vcpu)
5585 {
5586 return nested_vmx_run(vcpu, true);
5587 }
5588
5589 /* Emulate the VMRESUME instruction */
5590 static int handle_vmresume(struct kvm_vcpu *vcpu)
5591 {
5592
5593 return nested_vmx_run(vcpu, false);
5594 }
5595
5596 enum vmcs_field_type {
5597 VMCS_FIELD_TYPE_U16 = 0,
5598 VMCS_FIELD_TYPE_U64 = 1,
5599 VMCS_FIELD_TYPE_U32 = 2,
5600 VMCS_FIELD_TYPE_NATURAL_WIDTH = 3
5601 };
5602
5603 static inline int vmcs_field_type(unsigned long field)
5604 {
5605 if (0x1 & field) /* the *_HIGH fields are all 32 bit */
5606 return VMCS_FIELD_TYPE_U32;
5607 return (field >> 13) & 0x3 ;
5608 }
5609
5610 static inline int vmcs_field_readonly(unsigned long field)
5611 {
5612 return (((field >> 10) & 0x3) == 1);
5613 }
5614
5615 /*
5616 * Read a vmcs12 field. Since these can have varying lengths and we return
5617 * one type, we chose the biggest type (u64) and zero-extend the return value
5618 * to that size. Note that the caller, handle_vmread, might need to use only
5619 * some of the bits we return here (e.g., on 32-bit guests, only 32 bits of
5620 * 64-bit fields are to be returned).
5621 */
5622 static inline bool vmcs12_read_any(struct kvm_vcpu *vcpu,
5623 unsigned long field, u64 *ret)
5624 {
5625 short offset = vmcs_field_to_offset(field);
5626 char *p;
5627
5628 if (offset < 0)
5629 return 0;
5630
5631 p = ((char *)(get_vmcs12(vcpu))) + offset;
5632
5633 switch (vmcs_field_type(field)) {
5634 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
5635 *ret = *((natural_width *)p);
5636 return 1;
5637 case VMCS_FIELD_TYPE_U16:
5638 *ret = *((u16 *)p);
5639 return 1;
5640 case VMCS_FIELD_TYPE_U32:
5641 *ret = *((u32 *)p);
5642 return 1;
5643 case VMCS_FIELD_TYPE_U64:
5644 *ret = *((u64 *)p);
5645 return 1;
5646 default:
5647 return 0; /* can never happen. */
5648 }
5649 }
5650
5651 /*
5652 * VMX instructions which assume a current vmcs12 (i.e., that VMPTRLD was
5653 * used before) all generate the same failure when it is missing.
5654 */
5655 static int nested_vmx_check_vmcs12(struct kvm_vcpu *vcpu)
5656 {
5657 struct vcpu_vmx *vmx = to_vmx(vcpu);
5658 if (vmx->nested.current_vmptr == -1ull) {
5659 nested_vmx_failInvalid(vcpu);
5660 skip_emulated_instruction(vcpu);
5661 return 0;
5662 }
5663 return 1;
5664 }
5665
5666 static int handle_vmread(struct kvm_vcpu *vcpu)
5667 {
5668 unsigned long field;
5669 u64 field_value;
5670 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5671 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5672 gva_t gva = 0;
5673
5674 if (!nested_vmx_check_permission(vcpu) ||
5675 !nested_vmx_check_vmcs12(vcpu))
5676 return 1;
5677
5678 /* Decode instruction info and find the field to read */
5679 field = kvm_register_read(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
5680 /* Read the field, zero-extended to a u64 field_value */
5681 if (!vmcs12_read_any(vcpu, field, &field_value)) {
5682 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5683 skip_emulated_instruction(vcpu);
5684 return 1;
5685 }
5686 /*
5687 * Now copy part of this value to register or memory, as requested.
5688 * Note that the number of bits actually copied is 32 or 64 depending
5689 * on the guest's mode (32 or 64 bit), not on the given field's length.
5690 */
5691 if (vmx_instruction_info & (1u << 10)) {
5692 kvm_register_write(vcpu, (((vmx_instruction_info) >> 3) & 0xf),
5693 field_value);
5694 } else {
5695 if (get_vmx_mem_address(vcpu, exit_qualification,
5696 vmx_instruction_info, &gva))
5697 return 1;
5698 /* _system ok, as nested_vmx_check_permission verified cpl=0 */
5699 kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, gva,
5700 &field_value, (is_long_mode(vcpu) ? 8 : 4), NULL);
5701 }
5702
5703 nested_vmx_succeed(vcpu);
5704 skip_emulated_instruction(vcpu);
5705 return 1;
5706 }
5707
5708
5709 static int handle_vmwrite(struct kvm_vcpu *vcpu)
5710 {
5711 unsigned long field;
5712 gva_t gva;
5713 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5714 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5715 char *p;
5716 short offset;
5717 /* The value to write might be 32 or 64 bits, depending on L1's long
5718 * mode, and eventually we need to write that into a field of several
5719 * possible lengths. The code below first zero-extends the value to 64
5720 * bit (field_value), and then copies only the approriate number of
5721 * bits into the vmcs12 field.
5722 */
5723 u64 field_value = 0;
5724 struct x86_exception e;
5725
5726 if (!nested_vmx_check_permission(vcpu) ||
5727 !nested_vmx_check_vmcs12(vcpu))
5728 return 1;
5729
5730 if (vmx_instruction_info & (1u << 10))
5731 field_value = kvm_register_read(vcpu,
5732 (((vmx_instruction_info) >> 3) & 0xf));
5733 else {
5734 if (get_vmx_mem_address(vcpu, exit_qualification,
5735 vmx_instruction_info, &gva))
5736 return 1;
5737 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva,
5738 &field_value, (is_long_mode(vcpu) ? 8 : 4), &e)) {
5739 kvm_inject_page_fault(vcpu, &e);
5740 return 1;
5741 }
5742 }
5743
5744
5745 field = kvm_register_read(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
5746 if (vmcs_field_readonly(field)) {
5747 nested_vmx_failValid(vcpu,
5748 VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
5749 skip_emulated_instruction(vcpu);
5750 return 1;
5751 }
5752
5753 offset = vmcs_field_to_offset(field);
5754 if (offset < 0) {
5755 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5756 skip_emulated_instruction(vcpu);
5757 return 1;
5758 }
5759 p = ((char *) get_vmcs12(vcpu)) + offset;
5760
5761 switch (vmcs_field_type(field)) {
5762 case VMCS_FIELD_TYPE_U16:
5763 *(u16 *)p = field_value;
5764 break;
5765 case VMCS_FIELD_TYPE_U32:
5766 *(u32 *)p = field_value;
5767 break;
5768 case VMCS_FIELD_TYPE_U64:
5769 *(u64 *)p = field_value;
5770 break;
5771 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
5772 *(natural_width *)p = field_value;
5773 break;
5774 default:
5775 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5776 skip_emulated_instruction(vcpu);
5777 return 1;
5778 }
5779
5780 nested_vmx_succeed(vcpu);
5781 skip_emulated_instruction(vcpu);
5782 return 1;
5783 }
5784
5785 /* Emulate the VMPTRLD instruction */
5786 static int handle_vmptrld(struct kvm_vcpu *vcpu)
5787 {
5788 struct vcpu_vmx *vmx = to_vmx(vcpu);
5789 gva_t gva;
5790 gpa_t vmptr;
5791 struct x86_exception e;
5792
5793 if (!nested_vmx_check_permission(vcpu))
5794 return 1;
5795
5796 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
5797 vmcs_read32(VMX_INSTRUCTION_INFO), &gva))
5798 return 1;
5799
5800 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &vmptr,
5801 sizeof(vmptr), &e)) {
5802 kvm_inject_page_fault(vcpu, &e);
5803 return 1;
5804 }
5805
5806 if (!IS_ALIGNED(vmptr, PAGE_SIZE)) {
5807 nested_vmx_failValid(vcpu, VMXERR_VMPTRLD_INVALID_ADDRESS);
5808 skip_emulated_instruction(vcpu);
5809 return 1;
5810 }
5811
5812 if (vmx->nested.current_vmptr != vmptr) {
5813 struct vmcs12 *new_vmcs12;
5814 struct page *page;
5815 page = nested_get_page(vcpu, vmptr);
5816 if (page == NULL) {
5817 nested_vmx_failInvalid(vcpu);
5818 skip_emulated_instruction(vcpu);
5819 return 1;
5820 }
5821 new_vmcs12 = kmap(page);
5822 if (new_vmcs12->revision_id != VMCS12_REVISION) {
5823 kunmap(page);
5824 nested_release_page_clean(page);
5825 nested_vmx_failValid(vcpu,
5826 VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
5827 skip_emulated_instruction(vcpu);
5828 return 1;
5829 }
5830 if (vmx->nested.current_vmptr != -1ull) {
5831 kunmap(vmx->nested.current_vmcs12_page);
5832 nested_release_page(vmx->nested.current_vmcs12_page);
5833 }
5834
5835 vmx->nested.current_vmptr = vmptr;
5836 vmx->nested.current_vmcs12 = new_vmcs12;
5837 vmx->nested.current_vmcs12_page = page;
5838 }
5839
5840 nested_vmx_succeed(vcpu);
5841 skip_emulated_instruction(vcpu);
5842 return 1;
5843 }
5844
5845 /* Emulate the VMPTRST instruction */
5846 static int handle_vmptrst(struct kvm_vcpu *vcpu)
5847 {
5848 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5849 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5850 gva_t vmcs_gva;
5851 struct x86_exception e;
5852
5853 if (!nested_vmx_check_permission(vcpu))
5854 return 1;
5855
5856 if (get_vmx_mem_address(vcpu, exit_qualification,
5857 vmx_instruction_info, &vmcs_gva))
5858 return 1;
5859 /* ok to use *_system, as nested_vmx_check_permission verified cpl=0 */
5860 if (kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, vmcs_gva,
5861 (void *)&to_vmx(vcpu)->nested.current_vmptr,
5862 sizeof(u64), &e)) {
5863 kvm_inject_page_fault(vcpu, &e);
5864 return 1;
5865 }
5866 nested_vmx_succeed(vcpu);
5867 skip_emulated_instruction(vcpu);
5868 return 1;
5869 }
5870
5871 /*
5872 * The exit handlers return 1 if the exit was handled fully and guest execution
5873 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
5874 * to be done to userspace and return 0.
5875 */
5876 static int (*const kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
5877 [EXIT_REASON_EXCEPTION_NMI] = handle_exception,
5878 [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt,
5879 [EXIT_REASON_TRIPLE_FAULT] = handle_triple_fault,
5880 [EXIT_REASON_NMI_WINDOW] = handle_nmi_window,
5881 [EXIT_REASON_IO_INSTRUCTION] = handle_io,
5882 [EXIT_REASON_CR_ACCESS] = handle_cr,
5883 [EXIT_REASON_DR_ACCESS] = handle_dr,
5884 [EXIT_REASON_CPUID] = handle_cpuid,
5885 [EXIT_REASON_MSR_READ] = handle_rdmsr,
5886 [EXIT_REASON_MSR_WRITE] = handle_wrmsr,
5887 [EXIT_REASON_PENDING_INTERRUPT] = handle_interrupt_window,
5888 [EXIT_REASON_HLT] = handle_halt,
5889 [EXIT_REASON_INVD] = handle_invd,
5890 [EXIT_REASON_INVLPG] = handle_invlpg,
5891 [EXIT_REASON_RDPMC] = handle_rdpmc,
5892 [EXIT_REASON_VMCALL] = handle_vmcall,
5893 [EXIT_REASON_VMCLEAR] = handle_vmclear,
5894 [EXIT_REASON_VMLAUNCH] = handle_vmlaunch,
5895 [EXIT_REASON_VMPTRLD] = handle_vmptrld,
5896 [EXIT_REASON_VMPTRST] = handle_vmptrst,
5897 [EXIT_REASON_VMREAD] = handle_vmread,
5898 [EXIT_REASON_VMRESUME] = handle_vmresume,
5899 [EXIT_REASON_VMWRITE] = handle_vmwrite,
5900 [EXIT_REASON_VMOFF] = handle_vmoff,
5901 [EXIT_REASON_VMON] = handle_vmon,
5902 [EXIT_REASON_TPR_BELOW_THRESHOLD] = handle_tpr_below_threshold,
5903 [EXIT_REASON_APIC_ACCESS] = handle_apic_access,
5904 [EXIT_REASON_APIC_WRITE] = handle_apic_write,
5905 [EXIT_REASON_EOI_INDUCED] = handle_apic_eoi_induced,
5906 [EXIT_REASON_WBINVD] = handle_wbinvd,
5907 [EXIT_REASON_XSETBV] = handle_xsetbv,
5908 [EXIT_REASON_TASK_SWITCH] = handle_task_switch,
5909 [EXIT_REASON_MCE_DURING_VMENTRY] = handle_machine_check,
5910 [EXIT_REASON_EPT_VIOLATION] = handle_ept_violation,
5911 [EXIT_REASON_EPT_MISCONFIG] = handle_ept_misconfig,
5912 [EXIT_REASON_PAUSE_INSTRUCTION] = handle_pause,
5913 [EXIT_REASON_MWAIT_INSTRUCTION] = handle_invalid_op,
5914 [EXIT_REASON_MONITOR_INSTRUCTION] = handle_invalid_op,
5915 };
5916
5917 static const int kvm_vmx_max_exit_handlers =
5918 ARRAY_SIZE(kvm_vmx_exit_handlers);
5919
5920 static bool nested_vmx_exit_handled_io(struct kvm_vcpu *vcpu,
5921 struct vmcs12 *vmcs12)
5922 {
5923 unsigned long exit_qualification;
5924 gpa_t bitmap, last_bitmap;
5925 unsigned int port;
5926 int size;
5927 u8 b;
5928
5929 if (nested_cpu_has(vmcs12, CPU_BASED_UNCOND_IO_EXITING))
5930 return 1;
5931
5932 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
5933 return 0;
5934
5935 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5936
5937 port = exit_qualification >> 16;
5938 size = (exit_qualification & 7) + 1;
5939
5940 last_bitmap = (gpa_t)-1;
5941 b = -1;
5942
5943 while (size > 0) {
5944 if (port < 0x8000)
5945 bitmap = vmcs12->io_bitmap_a;
5946 else if (port < 0x10000)
5947 bitmap = vmcs12->io_bitmap_b;
5948 else
5949 return 1;
5950 bitmap += (port & 0x7fff) / 8;
5951
5952 if (last_bitmap != bitmap)
5953 if (kvm_read_guest(vcpu->kvm, bitmap, &b, 1))
5954 return 1;
5955 if (b & (1 << (port & 7)))
5956 return 1;
5957
5958 port++;
5959 size--;
5960 last_bitmap = bitmap;
5961 }
5962
5963 return 0;
5964 }
5965
5966 /*
5967 * Return 1 if we should exit from L2 to L1 to handle an MSR access access,
5968 * rather than handle it ourselves in L0. I.e., check whether L1 expressed
5969 * disinterest in the current event (read or write a specific MSR) by using an
5970 * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps.
5971 */
5972 static bool nested_vmx_exit_handled_msr(struct kvm_vcpu *vcpu,
5973 struct vmcs12 *vmcs12, u32 exit_reason)
5974 {
5975 u32 msr_index = vcpu->arch.regs[VCPU_REGS_RCX];
5976 gpa_t bitmap;
5977
5978 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
5979 return 1;
5980
5981 /*
5982 * The MSR_BITMAP page is divided into four 1024-byte bitmaps,
5983 * for the four combinations of read/write and low/high MSR numbers.
5984 * First we need to figure out which of the four to use:
5985 */
5986 bitmap = vmcs12->msr_bitmap;
5987 if (exit_reason == EXIT_REASON_MSR_WRITE)
5988 bitmap += 2048;
5989 if (msr_index >= 0xc0000000) {
5990 msr_index -= 0xc0000000;
5991 bitmap += 1024;
5992 }
5993
5994 /* Then read the msr_index'th bit from this bitmap: */
5995 if (msr_index < 1024*8) {
5996 unsigned char b;
5997 if (kvm_read_guest(vcpu->kvm, bitmap + msr_index/8, &b, 1))
5998 return 1;
5999 return 1 & (b >> (msr_index & 7));
6000 } else
6001 return 1; /* let L1 handle the wrong parameter */
6002 }
6003
6004 /*
6005 * Return 1 if we should exit from L2 to L1 to handle a CR access exit,
6006 * rather than handle it ourselves in L0. I.e., check if L1 wanted to
6007 * intercept (via guest_host_mask etc.) the current event.
6008 */
6009 static bool nested_vmx_exit_handled_cr(struct kvm_vcpu *vcpu,
6010 struct vmcs12 *vmcs12)
6011 {
6012 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6013 int cr = exit_qualification & 15;
6014 int reg = (exit_qualification >> 8) & 15;
6015 unsigned long val = kvm_register_read(vcpu, reg);
6016
6017 switch ((exit_qualification >> 4) & 3) {
6018 case 0: /* mov to cr */
6019 switch (cr) {
6020 case 0:
6021 if (vmcs12->cr0_guest_host_mask &
6022 (val ^ vmcs12->cr0_read_shadow))
6023 return 1;
6024 break;
6025 case 3:
6026 if ((vmcs12->cr3_target_count >= 1 &&
6027 vmcs12->cr3_target_value0 == val) ||
6028 (vmcs12->cr3_target_count >= 2 &&
6029 vmcs12->cr3_target_value1 == val) ||
6030 (vmcs12->cr3_target_count >= 3 &&
6031 vmcs12->cr3_target_value2 == val) ||
6032 (vmcs12->cr3_target_count >= 4 &&
6033 vmcs12->cr3_target_value3 == val))
6034 return 0;
6035 if (nested_cpu_has(vmcs12, CPU_BASED_CR3_LOAD_EXITING))
6036 return 1;
6037 break;
6038 case 4:
6039 if (vmcs12->cr4_guest_host_mask &
6040 (vmcs12->cr4_read_shadow ^ val))
6041 return 1;
6042 break;
6043 case 8:
6044 if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING))
6045 return 1;
6046 break;
6047 }
6048 break;
6049 case 2: /* clts */
6050 if ((vmcs12->cr0_guest_host_mask & X86_CR0_TS) &&
6051 (vmcs12->cr0_read_shadow & X86_CR0_TS))
6052 return 1;
6053 break;
6054 case 1: /* mov from cr */
6055 switch (cr) {
6056 case 3:
6057 if (vmcs12->cpu_based_vm_exec_control &
6058 CPU_BASED_CR3_STORE_EXITING)
6059 return 1;
6060 break;
6061 case 8:
6062 if (vmcs12->cpu_based_vm_exec_control &
6063 CPU_BASED_CR8_STORE_EXITING)
6064 return 1;
6065 break;
6066 }
6067 break;
6068 case 3: /* lmsw */
6069 /*
6070 * lmsw can change bits 1..3 of cr0, and only set bit 0 of
6071 * cr0. Other attempted changes are ignored, with no exit.
6072 */
6073 if (vmcs12->cr0_guest_host_mask & 0xe &
6074 (val ^ vmcs12->cr0_read_shadow))
6075 return 1;
6076 if ((vmcs12->cr0_guest_host_mask & 0x1) &&
6077 !(vmcs12->cr0_read_shadow & 0x1) &&
6078 (val & 0x1))
6079 return 1;
6080 break;
6081 }
6082 return 0;
6083 }
6084
6085 /*
6086 * Return 1 if we should exit from L2 to L1 to handle an exit, or 0 if we
6087 * should handle it ourselves in L0 (and then continue L2). Only call this
6088 * when in is_guest_mode (L2).
6089 */
6090 static bool nested_vmx_exit_handled(struct kvm_vcpu *vcpu)
6091 {
6092 u32 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6093 struct vcpu_vmx *vmx = to_vmx(vcpu);
6094 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6095 u32 exit_reason = vmx->exit_reason;
6096
6097 if (vmx->nested.nested_run_pending)
6098 return 0;
6099
6100 if (unlikely(vmx->fail)) {
6101 pr_info_ratelimited("%s failed vm entry %x\n", __func__,
6102 vmcs_read32(VM_INSTRUCTION_ERROR));
6103 return 1;
6104 }
6105
6106 switch (exit_reason) {
6107 case EXIT_REASON_EXCEPTION_NMI:
6108 if (!is_exception(intr_info))
6109 return 0;
6110 else if (is_page_fault(intr_info))
6111 return enable_ept;
6112 return vmcs12->exception_bitmap &
6113 (1u << (intr_info & INTR_INFO_VECTOR_MASK));
6114 case EXIT_REASON_EXTERNAL_INTERRUPT:
6115 return 0;
6116 case EXIT_REASON_TRIPLE_FAULT:
6117 return 1;
6118 case EXIT_REASON_PENDING_INTERRUPT:
6119 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_INTR_PENDING);
6120 case EXIT_REASON_NMI_WINDOW:
6121 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_NMI_PENDING);
6122 case EXIT_REASON_TASK_SWITCH:
6123 return 1;
6124 case EXIT_REASON_CPUID:
6125 return 1;
6126 case EXIT_REASON_HLT:
6127 return nested_cpu_has(vmcs12, CPU_BASED_HLT_EXITING);
6128 case EXIT_REASON_INVD:
6129 return 1;
6130 case EXIT_REASON_INVLPG:
6131 return nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
6132 case EXIT_REASON_RDPMC:
6133 return nested_cpu_has(vmcs12, CPU_BASED_RDPMC_EXITING);
6134 case EXIT_REASON_RDTSC:
6135 return nested_cpu_has(vmcs12, CPU_BASED_RDTSC_EXITING);
6136 case EXIT_REASON_VMCALL: case EXIT_REASON_VMCLEAR:
6137 case EXIT_REASON_VMLAUNCH: case EXIT_REASON_VMPTRLD:
6138 case EXIT_REASON_VMPTRST: case EXIT_REASON_VMREAD:
6139 case EXIT_REASON_VMRESUME: case EXIT_REASON_VMWRITE:
6140 case EXIT_REASON_VMOFF: case EXIT_REASON_VMON:
6141 /*
6142 * VMX instructions trap unconditionally. This allows L1 to
6143 * emulate them for its L2 guest, i.e., allows 3-level nesting!
6144 */
6145 return 1;
6146 case EXIT_REASON_CR_ACCESS:
6147 return nested_vmx_exit_handled_cr(vcpu, vmcs12);
6148 case EXIT_REASON_DR_ACCESS:
6149 return nested_cpu_has(vmcs12, CPU_BASED_MOV_DR_EXITING);
6150 case EXIT_REASON_IO_INSTRUCTION:
6151 return nested_vmx_exit_handled_io(vcpu, vmcs12);
6152 case EXIT_REASON_MSR_READ:
6153 case EXIT_REASON_MSR_WRITE:
6154 return nested_vmx_exit_handled_msr(vcpu, vmcs12, exit_reason);
6155 case EXIT_REASON_INVALID_STATE:
6156 return 1;
6157 case EXIT_REASON_MWAIT_INSTRUCTION:
6158 return nested_cpu_has(vmcs12, CPU_BASED_MWAIT_EXITING);
6159 case EXIT_REASON_MONITOR_INSTRUCTION:
6160 return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_EXITING);
6161 case EXIT_REASON_PAUSE_INSTRUCTION:
6162 return nested_cpu_has(vmcs12, CPU_BASED_PAUSE_EXITING) ||
6163 nested_cpu_has2(vmcs12,
6164 SECONDARY_EXEC_PAUSE_LOOP_EXITING);
6165 case EXIT_REASON_MCE_DURING_VMENTRY:
6166 return 0;
6167 case EXIT_REASON_TPR_BELOW_THRESHOLD:
6168 return 1;
6169 case EXIT_REASON_APIC_ACCESS:
6170 return nested_cpu_has2(vmcs12,
6171 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
6172 case EXIT_REASON_EPT_VIOLATION:
6173 case EXIT_REASON_EPT_MISCONFIG:
6174 return 0;
6175 case EXIT_REASON_PREEMPTION_TIMER:
6176 return vmcs12->pin_based_vm_exec_control &
6177 PIN_BASED_VMX_PREEMPTION_TIMER;
6178 case EXIT_REASON_WBINVD:
6179 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_WBINVD_EXITING);
6180 case EXIT_REASON_XSETBV:
6181 return 1;
6182 default:
6183 return 1;
6184 }
6185 }
6186
6187 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
6188 {
6189 *info1 = vmcs_readl(EXIT_QUALIFICATION);
6190 *info2 = vmcs_read32(VM_EXIT_INTR_INFO);
6191 }
6192
6193 /*
6194 * The guest has exited. See if we can fix it or if we need userspace
6195 * assistance.
6196 */
6197 static int vmx_handle_exit(struct kvm_vcpu *vcpu)
6198 {
6199 struct vcpu_vmx *vmx = to_vmx(vcpu);
6200 u32 exit_reason = vmx->exit_reason;
6201 u32 vectoring_info = vmx->idt_vectoring_info;
6202
6203 /* If guest state is invalid, start emulating */
6204 if (vmx->emulation_required)
6205 return handle_invalid_guest_state(vcpu);
6206
6207 /*
6208 * the KVM_REQ_EVENT optimization bit is only on for one entry, and if
6209 * we did not inject a still-pending event to L1 now because of
6210 * nested_run_pending, we need to re-enable this bit.
6211 */
6212 if (vmx->nested.nested_run_pending)
6213 kvm_make_request(KVM_REQ_EVENT, vcpu);
6214
6215 if (!is_guest_mode(vcpu) && (exit_reason == EXIT_REASON_VMLAUNCH ||
6216 exit_reason == EXIT_REASON_VMRESUME))
6217 vmx->nested.nested_run_pending = 1;
6218 else
6219 vmx->nested.nested_run_pending = 0;
6220
6221 if (is_guest_mode(vcpu) && nested_vmx_exit_handled(vcpu)) {
6222 nested_vmx_vmexit(vcpu);
6223 return 1;
6224 }
6225
6226 if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
6227 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
6228 vcpu->run->fail_entry.hardware_entry_failure_reason
6229 = exit_reason;
6230 return 0;
6231 }
6232
6233 if (unlikely(vmx->fail)) {
6234 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
6235 vcpu->run->fail_entry.hardware_entry_failure_reason
6236 = vmcs_read32(VM_INSTRUCTION_ERROR);
6237 return 0;
6238 }
6239
6240 /*
6241 * Note:
6242 * Do not try to fix EXIT_REASON_EPT_MISCONFIG if it caused by
6243 * delivery event since it indicates guest is accessing MMIO.
6244 * The vm-exit can be triggered again after return to guest that
6245 * will cause infinite loop.
6246 */
6247 if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
6248 (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
6249 exit_reason != EXIT_REASON_EPT_VIOLATION &&
6250 exit_reason != EXIT_REASON_TASK_SWITCH)) {
6251 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6252 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_DELIVERY_EV;
6253 vcpu->run->internal.ndata = 2;
6254 vcpu->run->internal.data[0] = vectoring_info;
6255 vcpu->run->internal.data[1] = exit_reason;
6256 return 0;
6257 }
6258
6259 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked &&
6260 !(is_guest_mode(vcpu) && nested_cpu_has_virtual_nmis(
6261 get_vmcs12(vcpu), vcpu)))) {
6262 if (vmx_interrupt_allowed(vcpu)) {
6263 vmx->soft_vnmi_blocked = 0;
6264 } else if (vmx->vnmi_blocked_time > 1000000000LL &&
6265 vcpu->arch.nmi_pending) {
6266 /*
6267 * This CPU don't support us in finding the end of an
6268 * NMI-blocked window if the guest runs with IRQs
6269 * disabled. So we pull the trigger after 1 s of
6270 * futile waiting, but inform the user about this.
6271 */
6272 printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
6273 "state on VCPU %d after 1 s timeout\n",
6274 __func__, vcpu->vcpu_id);
6275 vmx->soft_vnmi_blocked = 0;
6276 }
6277 }
6278
6279 if (exit_reason < kvm_vmx_max_exit_handlers
6280 && kvm_vmx_exit_handlers[exit_reason])
6281 return kvm_vmx_exit_handlers[exit_reason](vcpu);
6282 else {
6283 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
6284 vcpu->run->hw.hardware_exit_reason = exit_reason;
6285 }
6286 return 0;
6287 }
6288
6289 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
6290 {
6291 if (irr == -1 || tpr < irr) {
6292 vmcs_write32(TPR_THRESHOLD, 0);
6293 return;
6294 }
6295
6296 vmcs_write32(TPR_THRESHOLD, irr);
6297 }
6298
6299 static void vmx_set_virtual_x2apic_mode(struct kvm_vcpu *vcpu, bool set)
6300 {
6301 u32 sec_exec_control;
6302
6303 /*
6304 * There is not point to enable virtualize x2apic without enable
6305 * apicv
6306 */
6307 if (!cpu_has_vmx_virtualize_x2apic_mode() ||
6308 !vmx_vm_has_apicv(vcpu->kvm))
6309 return;
6310
6311 if (!vm_need_tpr_shadow(vcpu->kvm))
6312 return;
6313
6314 sec_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
6315
6316 if (set) {
6317 sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
6318 sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
6319 } else {
6320 sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
6321 sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
6322 }
6323 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, sec_exec_control);
6324
6325 vmx_set_msr_bitmap(vcpu);
6326 }
6327
6328 static void vmx_hwapic_isr_update(struct kvm *kvm, int isr)
6329 {
6330 u16 status;
6331 u8 old;
6332
6333 if (!vmx_vm_has_apicv(kvm))
6334 return;
6335
6336 if (isr == -1)
6337 isr = 0;
6338
6339 status = vmcs_read16(GUEST_INTR_STATUS);
6340 old = status >> 8;
6341 if (isr != old) {
6342 status &= 0xff;
6343 status |= isr << 8;
6344 vmcs_write16(GUEST_INTR_STATUS, status);
6345 }
6346 }
6347
6348 static void vmx_set_rvi(int vector)
6349 {
6350 u16 status;
6351 u8 old;
6352
6353 status = vmcs_read16(GUEST_INTR_STATUS);
6354 old = (u8)status & 0xff;
6355 if ((u8)vector != old) {
6356 status &= ~0xff;
6357 status |= (u8)vector;
6358 vmcs_write16(GUEST_INTR_STATUS, status);
6359 }
6360 }
6361
6362 static void vmx_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
6363 {
6364 if (max_irr == -1)
6365 return;
6366
6367 vmx_set_rvi(max_irr);
6368 }
6369
6370 static void vmx_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
6371 {
6372 vmcs_write64(EOI_EXIT_BITMAP0, eoi_exit_bitmap[0]);
6373 vmcs_write64(EOI_EXIT_BITMAP1, eoi_exit_bitmap[1]);
6374 vmcs_write64(EOI_EXIT_BITMAP2, eoi_exit_bitmap[2]);
6375 vmcs_write64(EOI_EXIT_BITMAP3, eoi_exit_bitmap[3]);
6376 }
6377
6378 static void vmx_complete_atomic_exit(struct vcpu_vmx *vmx)
6379 {
6380 u32 exit_intr_info;
6381
6382 if (!(vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY
6383 || vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI))
6384 return;
6385
6386 vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6387 exit_intr_info = vmx->exit_intr_info;
6388
6389 /* Handle machine checks before interrupts are enabled */
6390 if (is_machine_check(exit_intr_info))
6391 kvm_machine_check();
6392
6393 /* We need to handle NMIs before interrupts are enabled */
6394 if ((exit_intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR &&
6395 (exit_intr_info & INTR_INFO_VALID_MASK)) {
6396 kvm_before_handle_nmi(&vmx->vcpu);
6397 asm("int $2");
6398 kvm_after_handle_nmi(&vmx->vcpu);
6399 }
6400 }
6401
6402 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
6403 {
6404 u32 exit_intr_info;
6405 bool unblock_nmi;
6406 u8 vector;
6407 bool idtv_info_valid;
6408
6409 idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
6410
6411 if (cpu_has_virtual_nmis()) {
6412 if (vmx->nmi_known_unmasked)
6413 return;
6414 /*
6415 * Can't use vmx->exit_intr_info since we're not sure what
6416 * the exit reason is.
6417 */
6418 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6419 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
6420 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
6421 /*
6422 * SDM 3: 27.7.1.2 (September 2008)
6423 * Re-set bit "block by NMI" before VM entry if vmexit caused by
6424 * a guest IRET fault.
6425 * SDM 3: 23.2.2 (September 2008)
6426 * Bit 12 is undefined in any of the following cases:
6427 * If the VM exit sets the valid bit in the IDT-vectoring
6428 * information field.
6429 * If the VM exit is due to a double fault.
6430 */
6431 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
6432 vector != DF_VECTOR && !idtv_info_valid)
6433 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
6434 GUEST_INTR_STATE_NMI);
6435 else
6436 vmx->nmi_known_unmasked =
6437 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
6438 & GUEST_INTR_STATE_NMI);
6439 } else if (unlikely(vmx->soft_vnmi_blocked))
6440 vmx->vnmi_blocked_time +=
6441 ktime_to_ns(ktime_sub(ktime_get(), vmx->entry_time));
6442 }
6443
6444 static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu,
6445 u32 idt_vectoring_info,
6446 int instr_len_field,
6447 int error_code_field)
6448 {
6449 u8 vector;
6450 int type;
6451 bool idtv_info_valid;
6452
6453 idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
6454
6455 vcpu->arch.nmi_injected = false;
6456 kvm_clear_exception_queue(vcpu);
6457 kvm_clear_interrupt_queue(vcpu);
6458
6459 if (!idtv_info_valid)
6460 return;
6461
6462 kvm_make_request(KVM_REQ_EVENT, vcpu);
6463
6464 vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
6465 type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
6466
6467 switch (type) {
6468 case INTR_TYPE_NMI_INTR:
6469 vcpu->arch.nmi_injected = true;
6470 /*
6471 * SDM 3: 27.7.1.2 (September 2008)
6472 * Clear bit "block by NMI" before VM entry if a NMI
6473 * delivery faulted.
6474 */
6475 vmx_set_nmi_mask(vcpu, false);
6476 break;
6477 case INTR_TYPE_SOFT_EXCEPTION:
6478 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
6479 /* fall through */
6480 case INTR_TYPE_HARD_EXCEPTION:
6481 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
6482 u32 err = vmcs_read32(error_code_field);
6483 kvm_queue_exception_e(vcpu, vector, err);
6484 } else
6485 kvm_queue_exception(vcpu, vector);
6486 break;
6487 case INTR_TYPE_SOFT_INTR:
6488 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
6489 /* fall through */
6490 case INTR_TYPE_EXT_INTR:
6491 kvm_queue_interrupt(vcpu, vector, type == INTR_TYPE_SOFT_INTR);
6492 break;
6493 default:
6494 break;
6495 }
6496 }
6497
6498 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
6499 {
6500 __vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info,
6501 VM_EXIT_INSTRUCTION_LEN,
6502 IDT_VECTORING_ERROR_CODE);
6503 }
6504
6505 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
6506 {
6507 __vmx_complete_interrupts(vcpu,
6508 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
6509 VM_ENTRY_INSTRUCTION_LEN,
6510 VM_ENTRY_EXCEPTION_ERROR_CODE);
6511
6512 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
6513 }
6514
6515 static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
6516 {
6517 int i, nr_msrs;
6518 struct perf_guest_switch_msr *msrs;
6519
6520 msrs = perf_guest_get_msrs(&nr_msrs);
6521
6522 if (!msrs)
6523 return;
6524
6525 for (i = 0; i < nr_msrs; i++)
6526 if (msrs[i].host == msrs[i].guest)
6527 clear_atomic_switch_msr(vmx, msrs[i].msr);
6528 else
6529 add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
6530 msrs[i].host);
6531 }
6532
6533 static void __noclone vmx_vcpu_run(struct kvm_vcpu *vcpu)
6534 {
6535 struct vcpu_vmx *vmx = to_vmx(vcpu);
6536 unsigned long debugctlmsr;
6537
6538 /* Record the guest's net vcpu time for enforced NMI injections. */
6539 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked))
6540 vmx->entry_time = ktime_get();
6541
6542 /* Don't enter VMX if guest state is invalid, let the exit handler
6543 start emulation until we arrive back to a valid state */
6544 if (vmx->emulation_required)
6545 return;
6546
6547 if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
6548 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
6549 if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
6550 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
6551
6552 /* When single-stepping over STI and MOV SS, we must clear the
6553 * corresponding interruptibility bits in the guest state. Otherwise
6554 * vmentry fails as it then expects bit 14 (BS) in pending debug
6555 * exceptions being set, but that's not correct for the guest debugging
6556 * case. */
6557 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6558 vmx_set_interrupt_shadow(vcpu, 0);
6559
6560 atomic_switch_perf_msrs(vmx);
6561 debugctlmsr = get_debugctlmsr();
6562
6563 vmx->__launched = vmx->loaded_vmcs->launched;
6564 asm(
6565 /* Store host registers */
6566 "push %%" _ASM_DX "; push %%" _ASM_BP ";"
6567 "push %%" _ASM_CX " \n\t" /* placeholder for guest rcx */
6568 "push %%" _ASM_CX " \n\t"
6569 "cmp %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
6570 "je 1f \n\t"
6571 "mov %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
6572 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
6573 "1: \n\t"
6574 /* Reload cr2 if changed */
6575 "mov %c[cr2](%0), %%" _ASM_AX " \n\t"
6576 "mov %%cr2, %%" _ASM_DX " \n\t"
6577 "cmp %%" _ASM_AX ", %%" _ASM_DX " \n\t"
6578 "je 2f \n\t"
6579 "mov %%" _ASM_AX", %%cr2 \n\t"
6580 "2: \n\t"
6581 /* Check if vmlaunch of vmresume is needed */
6582 "cmpl $0, %c[launched](%0) \n\t"
6583 /* Load guest registers. Don't clobber flags. */
6584 "mov %c[rax](%0), %%" _ASM_AX " \n\t"
6585 "mov %c[rbx](%0), %%" _ASM_BX " \n\t"
6586 "mov %c[rdx](%0), %%" _ASM_DX " \n\t"
6587 "mov %c[rsi](%0), %%" _ASM_SI " \n\t"
6588 "mov %c[rdi](%0), %%" _ASM_DI " \n\t"
6589 "mov %c[rbp](%0), %%" _ASM_BP " \n\t"
6590 #ifdef CONFIG_X86_64
6591 "mov %c[r8](%0), %%r8 \n\t"
6592 "mov %c[r9](%0), %%r9 \n\t"
6593 "mov %c[r10](%0), %%r10 \n\t"
6594 "mov %c[r11](%0), %%r11 \n\t"
6595 "mov %c[r12](%0), %%r12 \n\t"
6596 "mov %c[r13](%0), %%r13 \n\t"
6597 "mov %c[r14](%0), %%r14 \n\t"
6598 "mov %c[r15](%0), %%r15 \n\t"
6599 #endif
6600 "mov %c[rcx](%0), %%" _ASM_CX " \n\t" /* kills %0 (ecx) */
6601
6602 /* Enter guest mode */
6603 "jne 1f \n\t"
6604 __ex(ASM_VMX_VMLAUNCH) "\n\t"
6605 "jmp 2f \n\t"
6606 "1: " __ex(ASM_VMX_VMRESUME) "\n\t"
6607 "2: "
6608 /* Save guest registers, load host registers, keep flags */
6609 "mov %0, %c[wordsize](%%" _ASM_SP ") \n\t"
6610 "pop %0 \n\t"
6611 "mov %%" _ASM_AX ", %c[rax](%0) \n\t"
6612 "mov %%" _ASM_BX ", %c[rbx](%0) \n\t"
6613 __ASM_SIZE(pop) " %c[rcx](%0) \n\t"
6614 "mov %%" _ASM_DX ", %c[rdx](%0) \n\t"
6615 "mov %%" _ASM_SI ", %c[rsi](%0) \n\t"
6616 "mov %%" _ASM_DI ", %c[rdi](%0) \n\t"
6617 "mov %%" _ASM_BP ", %c[rbp](%0) \n\t"
6618 #ifdef CONFIG_X86_64
6619 "mov %%r8, %c[r8](%0) \n\t"
6620 "mov %%r9, %c[r9](%0) \n\t"
6621 "mov %%r10, %c[r10](%0) \n\t"
6622 "mov %%r11, %c[r11](%0) \n\t"
6623 "mov %%r12, %c[r12](%0) \n\t"
6624 "mov %%r13, %c[r13](%0) \n\t"
6625 "mov %%r14, %c[r14](%0) \n\t"
6626 "mov %%r15, %c[r15](%0) \n\t"
6627 #endif
6628 "mov %%cr2, %%" _ASM_AX " \n\t"
6629 "mov %%" _ASM_AX ", %c[cr2](%0) \n\t"
6630
6631 "pop %%" _ASM_BP "; pop %%" _ASM_DX " \n\t"
6632 "setbe %c[fail](%0) \n\t"
6633 ".pushsection .rodata \n\t"
6634 ".global vmx_return \n\t"
6635 "vmx_return: " _ASM_PTR " 2b \n\t"
6636 ".popsection"
6637 : : "c"(vmx), "d"((unsigned long)HOST_RSP),
6638 [launched]"i"(offsetof(struct vcpu_vmx, __launched)),
6639 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
6640 [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
6641 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
6642 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
6643 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
6644 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
6645 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
6646 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
6647 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
6648 #ifdef CONFIG_X86_64
6649 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
6650 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
6651 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
6652 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
6653 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
6654 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
6655 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
6656 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
6657 #endif
6658 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2)),
6659 [wordsize]"i"(sizeof(ulong))
6660 : "cc", "memory"
6661 #ifdef CONFIG_X86_64
6662 , "rax", "rbx", "rdi", "rsi"
6663 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
6664 #else
6665 , "eax", "ebx", "edi", "esi"
6666 #endif
6667 );
6668
6669 /* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
6670 if (debugctlmsr)
6671 update_debugctlmsr(debugctlmsr);
6672
6673 #ifndef CONFIG_X86_64
6674 /*
6675 * The sysexit path does not restore ds/es, so we must set them to
6676 * a reasonable value ourselves.
6677 *
6678 * We can't defer this to vmx_load_host_state() since that function
6679 * may be executed in interrupt context, which saves and restore segments
6680 * around it, nullifying its effect.
6681 */
6682 loadsegment(ds, __USER_DS);
6683 loadsegment(es, __USER_DS);
6684 #endif
6685
6686 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
6687 | (1 << VCPU_EXREG_RFLAGS)
6688 | (1 << VCPU_EXREG_CPL)
6689 | (1 << VCPU_EXREG_PDPTR)
6690 | (1 << VCPU_EXREG_SEGMENTS)
6691 | (1 << VCPU_EXREG_CR3));
6692 vcpu->arch.regs_dirty = 0;
6693
6694 vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
6695
6696 vmx->loaded_vmcs->launched = 1;
6697
6698 vmx->exit_reason = vmcs_read32(VM_EXIT_REASON);
6699 trace_kvm_exit(vmx->exit_reason, vcpu, KVM_ISA_VMX);
6700
6701 vmx_complete_atomic_exit(vmx);
6702 vmx_recover_nmi_blocking(vmx);
6703 vmx_complete_interrupts(vmx);
6704 }
6705
6706 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
6707 {
6708 struct vcpu_vmx *vmx = to_vmx(vcpu);
6709
6710 free_vpid(vmx);
6711 free_nested(vmx);
6712 free_loaded_vmcs(vmx->loaded_vmcs);
6713 kfree(vmx->guest_msrs);
6714 kvm_vcpu_uninit(vcpu);
6715 kmem_cache_free(kvm_vcpu_cache, vmx);
6716 }
6717
6718 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
6719 {
6720 int err;
6721 struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
6722 int cpu;
6723
6724 if (!vmx)
6725 return ERR_PTR(-ENOMEM);
6726
6727 allocate_vpid(vmx);
6728
6729 err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
6730 if (err)
6731 goto free_vcpu;
6732
6733 vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
6734 err = -ENOMEM;
6735 if (!vmx->guest_msrs) {
6736 goto uninit_vcpu;
6737 }
6738
6739 vmx->loaded_vmcs = &vmx->vmcs01;
6740 vmx->loaded_vmcs->vmcs = alloc_vmcs();
6741 if (!vmx->loaded_vmcs->vmcs)
6742 goto free_msrs;
6743 if (!vmm_exclusive)
6744 kvm_cpu_vmxon(__pa(per_cpu(vmxarea, raw_smp_processor_id())));
6745 loaded_vmcs_init(vmx->loaded_vmcs);
6746 if (!vmm_exclusive)
6747 kvm_cpu_vmxoff();
6748
6749 cpu = get_cpu();
6750 vmx_vcpu_load(&vmx->vcpu, cpu);
6751 vmx->vcpu.cpu = cpu;
6752 err = vmx_vcpu_setup(vmx);
6753 vmx_vcpu_put(&vmx->vcpu);
6754 put_cpu();
6755 if (err)
6756 goto free_vmcs;
6757 if (vm_need_virtualize_apic_accesses(kvm)) {
6758 err = alloc_apic_access_page(kvm);
6759 if (err)
6760 goto free_vmcs;
6761 }
6762
6763 if (enable_ept) {
6764 if (!kvm->arch.ept_identity_map_addr)
6765 kvm->arch.ept_identity_map_addr =
6766 VMX_EPT_IDENTITY_PAGETABLE_ADDR;
6767 err = -ENOMEM;
6768 if (alloc_identity_pagetable(kvm) != 0)
6769 goto free_vmcs;
6770 if (!init_rmode_identity_map(kvm))
6771 goto free_vmcs;
6772 }
6773
6774 vmx->nested.current_vmptr = -1ull;
6775 vmx->nested.current_vmcs12 = NULL;
6776
6777 return &vmx->vcpu;
6778
6779 free_vmcs:
6780 free_loaded_vmcs(vmx->loaded_vmcs);
6781 free_msrs:
6782 kfree(vmx->guest_msrs);
6783 uninit_vcpu:
6784 kvm_vcpu_uninit(&vmx->vcpu);
6785 free_vcpu:
6786 free_vpid(vmx);
6787 kmem_cache_free(kvm_vcpu_cache, vmx);
6788 return ERR_PTR(err);
6789 }
6790
6791 static void __init vmx_check_processor_compat(void *rtn)
6792 {
6793 struct vmcs_config vmcs_conf;
6794
6795 *(int *)rtn = 0;
6796 if (setup_vmcs_config(&vmcs_conf) < 0)
6797 *(int *)rtn = -EIO;
6798 if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
6799 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
6800 smp_processor_id());
6801 *(int *)rtn = -EIO;
6802 }
6803 }
6804
6805 static int get_ept_level(void)
6806 {
6807 return VMX_EPT_DEFAULT_GAW + 1;
6808 }
6809
6810 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
6811 {
6812 u64 ret;
6813
6814 /* For VT-d and EPT combination
6815 * 1. MMIO: always map as UC
6816 * 2. EPT with VT-d:
6817 * a. VT-d without snooping control feature: can't guarantee the
6818 * result, try to trust guest.
6819 * b. VT-d with snooping control feature: snooping control feature of
6820 * VT-d engine can guarantee the cache correctness. Just set it
6821 * to WB to keep consistent with host. So the same as item 3.
6822 * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
6823 * consistent with host MTRR
6824 */
6825 if (is_mmio)
6826 ret = MTRR_TYPE_UNCACHABLE << VMX_EPT_MT_EPTE_SHIFT;
6827 else if (vcpu->kvm->arch.iommu_domain &&
6828 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY))
6829 ret = kvm_get_guest_memory_type(vcpu, gfn) <<
6830 VMX_EPT_MT_EPTE_SHIFT;
6831 else
6832 ret = (MTRR_TYPE_WRBACK << VMX_EPT_MT_EPTE_SHIFT)
6833 | VMX_EPT_IPAT_BIT;
6834
6835 return ret;
6836 }
6837
6838 static int vmx_get_lpage_level(void)
6839 {
6840 if (enable_ept && !cpu_has_vmx_ept_1g_page())
6841 return PT_DIRECTORY_LEVEL;
6842 else
6843 /* For shadow and EPT supported 1GB page */
6844 return PT_PDPE_LEVEL;
6845 }
6846
6847 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
6848 {
6849 struct kvm_cpuid_entry2 *best;
6850 struct vcpu_vmx *vmx = to_vmx(vcpu);
6851 u32 exec_control;
6852
6853 vmx->rdtscp_enabled = false;
6854 if (vmx_rdtscp_supported()) {
6855 exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
6856 if (exec_control & SECONDARY_EXEC_RDTSCP) {
6857 best = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
6858 if (best && (best->edx & bit(X86_FEATURE_RDTSCP)))
6859 vmx->rdtscp_enabled = true;
6860 else {
6861 exec_control &= ~SECONDARY_EXEC_RDTSCP;
6862 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
6863 exec_control);
6864 }
6865 }
6866 }
6867
6868 /* Exposing INVPCID only when PCID is exposed */
6869 best = kvm_find_cpuid_entry(vcpu, 0x7, 0);
6870 if (vmx_invpcid_supported() &&
6871 best && (best->ebx & bit(X86_FEATURE_INVPCID)) &&
6872 guest_cpuid_has_pcid(vcpu)) {
6873 exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
6874 exec_control |= SECONDARY_EXEC_ENABLE_INVPCID;
6875 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
6876 exec_control);
6877 } else {
6878 if (cpu_has_secondary_exec_ctrls()) {
6879 exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
6880 exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
6881 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
6882 exec_control);
6883 }
6884 if (best)
6885 best->ebx &= ~bit(X86_FEATURE_INVPCID);
6886 }
6887 }
6888
6889 static void vmx_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
6890 {
6891 if (func == 1 && nested)
6892 entry->ecx |= bit(X86_FEATURE_VMX);
6893 }
6894
6895 /*
6896 * prepare_vmcs02 is called when the L1 guest hypervisor runs its nested
6897 * L2 guest. L1 has a vmcs for L2 (vmcs12), and this function "merges" it
6898 * with L0's requirements for its guest (a.k.a. vmsc01), so we can run the L2
6899 * guest in a way that will both be appropriate to L1's requests, and our
6900 * needs. In addition to modifying the active vmcs (which is vmcs02), this
6901 * function also has additional necessary side-effects, like setting various
6902 * vcpu->arch fields.
6903 */
6904 static void prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
6905 {
6906 struct vcpu_vmx *vmx = to_vmx(vcpu);
6907 u32 exec_control;
6908
6909 vmcs_write16(GUEST_ES_SELECTOR, vmcs12->guest_es_selector);
6910 vmcs_write16(GUEST_CS_SELECTOR, vmcs12->guest_cs_selector);
6911 vmcs_write16(GUEST_SS_SELECTOR, vmcs12->guest_ss_selector);
6912 vmcs_write16(GUEST_DS_SELECTOR, vmcs12->guest_ds_selector);
6913 vmcs_write16(GUEST_FS_SELECTOR, vmcs12->guest_fs_selector);
6914 vmcs_write16(GUEST_GS_SELECTOR, vmcs12->guest_gs_selector);
6915 vmcs_write16(GUEST_LDTR_SELECTOR, vmcs12->guest_ldtr_selector);
6916 vmcs_write16(GUEST_TR_SELECTOR, vmcs12->guest_tr_selector);
6917 vmcs_write32(GUEST_ES_LIMIT, vmcs12->guest_es_limit);
6918 vmcs_write32(GUEST_CS_LIMIT, vmcs12->guest_cs_limit);
6919 vmcs_write32(GUEST_SS_LIMIT, vmcs12->guest_ss_limit);
6920 vmcs_write32(GUEST_DS_LIMIT, vmcs12->guest_ds_limit);
6921 vmcs_write32(GUEST_FS_LIMIT, vmcs12->guest_fs_limit);
6922 vmcs_write32(GUEST_GS_LIMIT, vmcs12->guest_gs_limit);
6923 vmcs_write32(GUEST_LDTR_LIMIT, vmcs12->guest_ldtr_limit);
6924 vmcs_write32(GUEST_TR_LIMIT, vmcs12->guest_tr_limit);
6925 vmcs_write32(GUEST_GDTR_LIMIT, vmcs12->guest_gdtr_limit);
6926 vmcs_write32(GUEST_IDTR_LIMIT, vmcs12->guest_idtr_limit);
6927 vmcs_write32(GUEST_ES_AR_BYTES, vmcs12->guest_es_ar_bytes);
6928 vmcs_write32(GUEST_CS_AR_BYTES, vmcs12->guest_cs_ar_bytes);
6929 vmcs_write32(GUEST_SS_AR_BYTES, vmcs12->guest_ss_ar_bytes);
6930 vmcs_write32(GUEST_DS_AR_BYTES, vmcs12->guest_ds_ar_bytes);
6931 vmcs_write32(GUEST_FS_AR_BYTES, vmcs12->guest_fs_ar_bytes);
6932 vmcs_write32(GUEST_GS_AR_BYTES, vmcs12->guest_gs_ar_bytes);
6933 vmcs_write32(GUEST_LDTR_AR_BYTES, vmcs12->guest_ldtr_ar_bytes);
6934 vmcs_write32(GUEST_TR_AR_BYTES, vmcs12->guest_tr_ar_bytes);
6935 vmcs_writel(GUEST_ES_BASE, vmcs12->guest_es_base);
6936 vmcs_writel(GUEST_CS_BASE, vmcs12->guest_cs_base);
6937 vmcs_writel(GUEST_SS_BASE, vmcs12->guest_ss_base);
6938 vmcs_writel(GUEST_DS_BASE, vmcs12->guest_ds_base);
6939 vmcs_writel(GUEST_FS_BASE, vmcs12->guest_fs_base);
6940 vmcs_writel(GUEST_GS_BASE, vmcs12->guest_gs_base);
6941 vmcs_writel(GUEST_LDTR_BASE, vmcs12->guest_ldtr_base);
6942 vmcs_writel(GUEST_TR_BASE, vmcs12->guest_tr_base);
6943 vmcs_writel(GUEST_GDTR_BASE, vmcs12->guest_gdtr_base);
6944 vmcs_writel(GUEST_IDTR_BASE, vmcs12->guest_idtr_base);
6945
6946 vmcs_write64(GUEST_IA32_DEBUGCTL, vmcs12->guest_ia32_debugctl);
6947 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
6948 vmcs12->vm_entry_intr_info_field);
6949 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
6950 vmcs12->vm_entry_exception_error_code);
6951 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
6952 vmcs12->vm_entry_instruction_len);
6953 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
6954 vmcs12->guest_interruptibility_info);
6955 vmcs_write32(GUEST_ACTIVITY_STATE, vmcs12->guest_activity_state);
6956 vmcs_write32(GUEST_SYSENTER_CS, vmcs12->guest_sysenter_cs);
6957 kvm_set_dr(vcpu, 7, vmcs12->guest_dr7);
6958 vmcs_writel(GUEST_RFLAGS, vmcs12->guest_rflags);
6959 vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
6960 vmcs12->guest_pending_dbg_exceptions);
6961 vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->guest_sysenter_esp);
6962 vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->guest_sysenter_eip);
6963
6964 vmcs_write64(VMCS_LINK_POINTER, -1ull);
6965
6966 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL,
6967 (vmcs_config.pin_based_exec_ctrl |
6968 vmcs12->pin_based_vm_exec_control));
6969
6970 if (vmcs12->pin_based_vm_exec_control & PIN_BASED_VMX_PREEMPTION_TIMER)
6971 vmcs_write32(VMX_PREEMPTION_TIMER_VALUE,
6972 vmcs12->vmx_preemption_timer_value);
6973
6974 /*
6975 * Whether page-faults are trapped is determined by a combination of
6976 * 3 settings: PFEC_MASK, PFEC_MATCH and EXCEPTION_BITMAP.PF.
6977 * If enable_ept, L0 doesn't care about page faults and we should
6978 * set all of these to L1's desires. However, if !enable_ept, L0 does
6979 * care about (at least some) page faults, and because it is not easy
6980 * (if at all possible?) to merge L0 and L1's desires, we simply ask
6981 * to exit on each and every L2 page fault. This is done by setting
6982 * MASK=MATCH=0 and (see below) EB.PF=1.
6983 * Note that below we don't need special code to set EB.PF beyond the
6984 * "or"ing of the EB of vmcs01 and vmcs12, because when enable_ept,
6985 * vmcs01's EB.PF is 0 so the "or" will take vmcs12's value, and when
6986 * !enable_ept, EB.PF is 1, so the "or" will always be 1.
6987 *
6988 * A problem with this approach (when !enable_ept) is that L1 may be
6989 * injected with more page faults than it asked for. This could have
6990 * caused problems, but in practice existing hypervisors don't care.
6991 * To fix this, we will need to emulate the PFEC checking (on the L1
6992 * page tables), using walk_addr(), when injecting PFs to L1.
6993 */
6994 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK,
6995 enable_ept ? vmcs12->page_fault_error_code_mask : 0);
6996 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH,
6997 enable_ept ? vmcs12->page_fault_error_code_match : 0);
6998
6999 if (cpu_has_secondary_exec_ctrls()) {
7000 u32 exec_control = vmx_secondary_exec_control(vmx);
7001 if (!vmx->rdtscp_enabled)
7002 exec_control &= ~SECONDARY_EXEC_RDTSCP;
7003 /* Take the following fields only from vmcs12 */
7004 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
7005 if (nested_cpu_has(vmcs12,
7006 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS))
7007 exec_control |= vmcs12->secondary_vm_exec_control;
7008
7009 if (exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES) {
7010 /*
7011 * Translate L1 physical address to host physical
7012 * address for vmcs02. Keep the page pinned, so this
7013 * physical address remains valid. We keep a reference
7014 * to it so we can release it later.
7015 */
7016 if (vmx->nested.apic_access_page) /* shouldn't happen */
7017 nested_release_page(vmx->nested.apic_access_page);
7018 vmx->nested.apic_access_page =
7019 nested_get_page(vcpu, vmcs12->apic_access_addr);
7020 /*
7021 * If translation failed, no matter: This feature asks
7022 * to exit when accessing the given address, and if it
7023 * can never be accessed, this feature won't do
7024 * anything anyway.
7025 */
7026 if (!vmx->nested.apic_access_page)
7027 exec_control &=
7028 ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
7029 else
7030 vmcs_write64(APIC_ACCESS_ADDR,
7031 page_to_phys(vmx->nested.apic_access_page));
7032 }
7033
7034 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
7035 }
7036
7037
7038 /*
7039 * Set host-state according to L0's settings (vmcs12 is irrelevant here)
7040 * Some constant fields are set here by vmx_set_constant_host_state().
7041 * Other fields are different per CPU, and will be set later when
7042 * vmx_vcpu_load() is called, and when vmx_save_host_state() is called.
7043 */
7044 vmx_set_constant_host_state();
7045
7046 /*
7047 * HOST_RSP is normally set correctly in vmx_vcpu_run() just before
7048 * entry, but only if the current (host) sp changed from the value
7049 * we wrote last (vmx->host_rsp). This cache is no longer relevant
7050 * if we switch vmcs, and rather than hold a separate cache per vmcs,
7051 * here we just force the write to happen on entry.
7052 */
7053 vmx->host_rsp = 0;
7054
7055 exec_control = vmx_exec_control(vmx); /* L0's desires */
7056 exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
7057 exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
7058 exec_control &= ~CPU_BASED_TPR_SHADOW;
7059 exec_control |= vmcs12->cpu_based_vm_exec_control;
7060 /*
7061 * Merging of IO and MSR bitmaps not currently supported.
7062 * Rather, exit every time.
7063 */
7064 exec_control &= ~CPU_BASED_USE_MSR_BITMAPS;
7065 exec_control &= ~CPU_BASED_USE_IO_BITMAPS;
7066 exec_control |= CPU_BASED_UNCOND_IO_EXITING;
7067
7068 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
7069
7070 /* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the
7071 * bitwise-or of what L1 wants to trap for L2, and what we want to
7072 * trap. Note that CR0.TS also needs updating - we do this later.
7073 */
7074 update_exception_bitmap(vcpu);
7075 vcpu->arch.cr0_guest_owned_bits &= ~vmcs12->cr0_guest_host_mask;
7076 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
7077
7078 /* Note: IA32_MODE, LOAD_IA32_EFER are modified by vmx_set_efer below */
7079 vmcs_write32(VM_EXIT_CONTROLS,
7080 vmcs12->vm_exit_controls | vmcs_config.vmexit_ctrl);
7081 vmcs_write32(VM_ENTRY_CONTROLS, vmcs12->vm_entry_controls |
7082 (vmcs_config.vmentry_ctrl & ~VM_ENTRY_IA32E_MODE));
7083
7084 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT)
7085 vmcs_write64(GUEST_IA32_PAT, vmcs12->guest_ia32_pat);
7086 else if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
7087 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
7088
7089
7090 set_cr4_guest_host_mask(vmx);
7091
7092 if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING)
7093 vmcs_write64(TSC_OFFSET,
7094 vmx->nested.vmcs01_tsc_offset + vmcs12->tsc_offset);
7095 else
7096 vmcs_write64(TSC_OFFSET, vmx->nested.vmcs01_tsc_offset);
7097
7098 if (enable_vpid) {
7099 /*
7100 * Trivially support vpid by letting L2s share their parent
7101 * L1's vpid. TODO: move to a more elaborate solution, giving
7102 * each L2 its own vpid and exposing the vpid feature to L1.
7103 */
7104 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
7105 vmx_flush_tlb(vcpu);
7106 }
7107
7108 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER)
7109 vcpu->arch.efer = vmcs12->guest_ia32_efer;
7110 if (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE)
7111 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
7112 else
7113 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
7114 /* Note: modifies VM_ENTRY/EXIT_CONTROLS and GUEST/HOST_IA32_EFER */
7115 vmx_set_efer(vcpu, vcpu->arch.efer);
7116
7117 /*
7118 * This sets GUEST_CR0 to vmcs12->guest_cr0, with possibly a modified
7119 * TS bit (for lazy fpu) and bits which we consider mandatory enabled.
7120 * The CR0_READ_SHADOW is what L2 should have expected to read given
7121 * the specifications by L1; It's not enough to take
7122 * vmcs12->cr0_read_shadow because on our cr0_guest_host_mask we we
7123 * have more bits than L1 expected.
7124 */
7125 vmx_set_cr0(vcpu, vmcs12->guest_cr0);
7126 vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
7127
7128 vmx_set_cr4(vcpu, vmcs12->guest_cr4);
7129 vmcs_writel(CR4_READ_SHADOW, nested_read_cr4(vmcs12));
7130
7131 /* shadow page tables on either EPT or shadow page tables */
7132 kvm_set_cr3(vcpu, vmcs12->guest_cr3);
7133 kvm_mmu_reset_context(vcpu);
7134
7135 kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->guest_rsp);
7136 kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->guest_rip);
7137 }
7138
7139 /*
7140 * nested_vmx_run() handles a nested entry, i.e., a VMLAUNCH or VMRESUME on L1
7141 * for running an L2 nested guest.
7142 */
7143 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
7144 {
7145 struct vmcs12 *vmcs12;
7146 struct vcpu_vmx *vmx = to_vmx(vcpu);
7147 int cpu;
7148 struct loaded_vmcs *vmcs02;
7149
7150 if (!nested_vmx_check_permission(vcpu) ||
7151 !nested_vmx_check_vmcs12(vcpu))
7152 return 1;
7153
7154 skip_emulated_instruction(vcpu);
7155 vmcs12 = get_vmcs12(vcpu);
7156
7157 /*
7158 * The nested entry process starts with enforcing various prerequisites
7159 * on vmcs12 as required by the Intel SDM, and act appropriately when
7160 * they fail: As the SDM explains, some conditions should cause the
7161 * instruction to fail, while others will cause the instruction to seem
7162 * to succeed, but return an EXIT_REASON_INVALID_STATE.
7163 * To speed up the normal (success) code path, we should avoid checking
7164 * for misconfigurations which will anyway be caught by the processor
7165 * when using the merged vmcs02.
7166 */
7167 if (vmcs12->launch_state == launch) {
7168 nested_vmx_failValid(vcpu,
7169 launch ? VMXERR_VMLAUNCH_NONCLEAR_VMCS
7170 : VMXERR_VMRESUME_NONLAUNCHED_VMCS);
7171 return 1;
7172 }
7173
7174 if ((vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_MSR_BITMAPS) &&
7175 !IS_ALIGNED(vmcs12->msr_bitmap, PAGE_SIZE)) {
7176 /*TODO: Also verify bits beyond physical address width are 0*/
7177 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
7178 return 1;
7179 }
7180
7181 if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES) &&
7182 !IS_ALIGNED(vmcs12->apic_access_addr, PAGE_SIZE)) {
7183 /*TODO: Also verify bits beyond physical address width are 0*/
7184 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
7185 return 1;
7186 }
7187
7188 if (vmcs12->vm_entry_msr_load_count > 0 ||
7189 vmcs12->vm_exit_msr_load_count > 0 ||
7190 vmcs12->vm_exit_msr_store_count > 0) {
7191 pr_warn_ratelimited("%s: VMCS MSR_{LOAD,STORE} unsupported\n",
7192 __func__);
7193 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
7194 return 1;
7195 }
7196
7197 if (!vmx_control_verify(vmcs12->cpu_based_vm_exec_control,
7198 nested_vmx_procbased_ctls_low, nested_vmx_procbased_ctls_high) ||
7199 !vmx_control_verify(vmcs12->secondary_vm_exec_control,
7200 nested_vmx_secondary_ctls_low, nested_vmx_secondary_ctls_high) ||
7201 !vmx_control_verify(vmcs12->pin_based_vm_exec_control,
7202 nested_vmx_pinbased_ctls_low, nested_vmx_pinbased_ctls_high) ||
7203 !vmx_control_verify(vmcs12->vm_exit_controls,
7204 nested_vmx_exit_ctls_low, nested_vmx_exit_ctls_high) ||
7205 !vmx_control_verify(vmcs12->vm_entry_controls,
7206 nested_vmx_entry_ctls_low, nested_vmx_entry_ctls_high))
7207 {
7208 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
7209 return 1;
7210 }
7211
7212 if (((vmcs12->host_cr0 & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON) ||
7213 ((vmcs12->host_cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON)) {
7214 nested_vmx_failValid(vcpu,
7215 VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7216 return 1;
7217 }
7218
7219 if (((vmcs12->guest_cr0 & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON) ||
7220 ((vmcs12->guest_cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON)) {
7221 nested_vmx_entry_failure(vcpu, vmcs12,
7222 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
7223 return 1;
7224 }
7225 if (vmcs12->vmcs_link_pointer != -1ull) {
7226 nested_vmx_entry_failure(vcpu, vmcs12,
7227 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_VMCS_LINK_PTR);
7228 return 1;
7229 }
7230
7231 /*
7232 * We're finally done with prerequisite checking, and can start with
7233 * the nested entry.
7234 */
7235
7236 vmcs02 = nested_get_current_vmcs02(vmx);
7237 if (!vmcs02)
7238 return -ENOMEM;
7239
7240 enter_guest_mode(vcpu);
7241
7242 vmx->nested.vmcs01_tsc_offset = vmcs_read64(TSC_OFFSET);
7243
7244 cpu = get_cpu();
7245 vmx->loaded_vmcs = vmcs02;
7246 vmx_vcpu_put(vcpu);
7247 vmx_vcpu_load(vcpu, cpu);
7248 vcpu->cpu = cpu;
7249 put_cpu();
7250
7251 vmx_segment_cache_clear(vmx);
7252
7253 vmcs12->launch_state = 1;
7254
7255 prepare_vmcs02(vcpu, vmcs12);
7256
7257 /*
7258 * Note no nested_vmx_succeed or nested_vmx_fail here. At this point
7259 * we are no longer running L1, and VMLAUNCH/VMRESUME has not yet
7260 * returned as far as L1 is concerned. It will only return (and set
7261 * the success flag) when L2 exits (see nested_vmx_vmexit()).
7262 */
7263 return 1;
7264 }
7265
7266 /*
7267 * On a nested exit from L2 to L1, vmcs12.guest_cr0 might not be up-to-date
7268 * because L2 may have changed some cr0 bits directly (CRO_GUEST_HOST_MASK).
7269 * This function returns the new value we should put in vmcs12.guest_cr0.
7270 * It's not enough to just return the vmcs02 GUEST_CR0. Rather,
7271 * 1. Bits that neither L0 nor L1 trapped, were set directly by L2 and are now
7272 * available in vmcs02 GUEST_CR0. (Note: It's enough to check that L0
7273 * didn't trap the bit, because if L1 did, so would L0).
7274 * 2. Bits that L1 asked to trap (and therefore L0 also did) could not have
7275 * been modified by L2, and L1 knows it. So just leave the old value of
7276 * the bit from vmcs12.guest_cr0. Note that the bit from vmcs02 GUEST_CR0
7277 * isn't relevant, because if L0 traps this bit it can set it to anything.
7278 * 3. Bits that L1 didn't trap, but L0 did. L1 believes the guest could have
7279 * changed these bits, and therefore they need to be updated, but L0
7280 * didn't necessarily allow them to be changed in GUEST_CR0 - and rather
7281 * put them in vmcs02 CR0_READ_SHADOW. So take these bits from there.
7282 */
7283 static inline unsigned long
7284 vmcs12_guest_cr0(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
7285 {
7286 return
7287 /*1*/ (vmcs_readl(GUEST_CR0) & vcpu->arch.cr0_guest_owned_bits) |
7288 /*2*/ (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask) |
7289 /*3*/ (vmcs_readl(CR0_READ_SHADOW) & ~(vmcs12->cr0_guest_host_mask |
7290 vcpu->arch.cr0_guest_owned_bits));
7291 }
7292
7293 static inline unsigned long
7294 vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
7295 {
7296 return
7297 /*1*/ (vmcs_readl(GUEST_CR4) & vcpu->arch.cr4_guest_owned_bits) |
7298 /*2*/ (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask) |
7299 /*3*/ (vmcs_readl(CR4_READ_SHADOW) & ~(vmcs12->cr4_guest_host_mask |
7300 vcpu->arch.cr4_guest_owned_bits));
7301 }
7302
7303 static void vmcs12_save_pending_event(struct kvm_vcpu *vcpu,
7304 struct vmcs12 *vmcs12)
7305 {
7306 u32 idt_vectoring;
7307 unsigned int nr;
7308
7309 if (vcpu->arch.exception.pending) {
7310 nr = vcpu->arch.exception.nr;
7311 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
7312
7313 if (kvm_exception_is_soft(nr)) {
7314 vmcs12->vm_exit_instruction_len =
7315 vcpu->arch.event_exit_inst_len;
7316 idt_vectoring |= INTR_TYPE_SOFT_EXCEPTION;
7317 } else
7318 idt_vectoring |= INTR_TYPE_HARD_EXCEPTION;
7319
7320 if (vcpu->arch.exception.has_error_code) {
7321 idt_vectoring |= VECTORING_INFO_DELIVER_CODE_MASK;
7322 vmcs12->idt_vectoring_error_code =
7323 vcpu->arch.exception.error_code;
7324 }
7325
7326 vmcs12->idt_vectoring_info_field = idt_vectoring;
7327 } else if (vcpu->arch.nmi_pending) {
7328 vmcs12->idt_vectoring_info_field =
7329 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR;
7330 } else if (vcpu->arch.interrupt.pending) {
7331 nr = vcpu->arch.interrupt.nr;
7332 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
7333
7334 if (vcpu->arch.interrupt.soft) {
7335 idt_vectoring |= INTR_TYPE_SOFT_INTR;
7336 vmcs12->vm_entry_instruction_len =
7337 vcpu->arch.event_exit_inst_len;
7338 } else
7339 idt_vectoring |= INTR_TYPE_EXT_INTR;
7340
7341 vmcs12->idt_vectoring_info_field = idt_vectoring;
7342 }
7343 }
7344
7345 /*
7346 * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
7347 * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),
7348 * and this function updates it to reflect the changes to the guest state while
7349 * L2 was running (and perhaps made some exits which were handled directly by L0
7350 * without going back to L1), and to reflect the exit reason.
7351 * Note that we do not have to copy here all VMCS fields, just those that
7352 * could have changed by the L2 guest or the exit - i.e., the guest-state and
7353 * exit-information fields only. Other fields are modified by L1 with VMWRITE,
7354 * which already writes to vmcs12 directly.
7355 */
7356 static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
7357 {
7358 /* update guest state fields: */
7359 vmcs12->guest_cr0 = vmcs12_guest_cr0(vcpu, vmcs12);
7360 vmcs12->guest_cr4 = vmcs12_guest_cr4(vcpu, vmcs12);
7361
7362 kvm_get_dr(vcpu, 7, (unsigned long *)&vmcs12->guest_dr7);
7363 vmcs12->guest_rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
7364 vmcs12->guest_rip = kvm_register_read(vcpu, VCPU_REGS_RIP);
7365 vmcs12->guest_rflags = vmcs_readl(GUEST_RFLAGS);
7366
7367 vmcs12->guest_es_selector = vmcs_read16(GUEST_ES_SELECTOR);
7368 vmcs12->guest_cs_selector = vmcs_read16(GUEST_CS_SELECTOR);
7369 vmcs12->guest_ss_selector = vmcs_read16(GUEST_SS_SELECTOR);
7370 vmcs12->guest_ds_selector = vmcs_read16(GUEST_DS_SELECTOR);
7371 vmcs12->guest_fs_selector = vmcs_read16(GUEST_FS_SELECTOR);
7372 vmcs12->guest_gs_selector = vmcs_read16(GUEST_GS_SELECTOR);
7373 vmcs12->guest_ldtr_selector = vmcs_read16(GUEST_LDTR_SELECTOR);
7374 vmcs12->guest_tr_selector = vmcs_read16(GUEST_TR_SELECTOR);
7375 vmcs12->guest_es_limit = vmcs_read32(GUEST_ES_LIMIT);
7376 vmcs12->guest_cs_limit = vmcs_read32(GUEST_CS_LIMIT);
7377 vmcs12->guest_ss_limit = vmcs_read32(GUEST_SS_LIMIT);
7378 vmcs12->guest_ds_limit = vmcs_read32(GUEST_DS_LIMIT);
7379 vmcs12->guest_fs_limit = vmcs_read32(GUEST_FS_LIMIT);
7380 vmcs12->guest_gs_limit = vmcs_read32(GUEST_GS_LIMIT);
7381 vmcs12->guest_ldtr_limit = vmcs_read32(GUEST_LDTR_LIMIT);
7382 vmcs12->guest_tr_limit = vmcs_read32(GUEST_TR_LIMIT);
7383 vmcs12->guest_gdtr_limit = vmcs_read32(GUEST_GDTR_LIMIT);
7384 vmcs12->guest_idtr_limit = vmcs_read32(GUEST_IDTR_LIMIT);
7385 vmcs12->guest_es_ar_bytes = vmcs_read32(GUEST_ES_AR_BYTES);
7386 vmcs12->guest_cs_ar_bytes = vmcs_read32(GUEST_CS_AR_BYTES);
7387 vmcs12->guest_ss_ar_bytes = vmcs_read32(GUEST_SS_AR_BYTES);
7388 vmcs12->guest_ds_ar_bytes = vmcs_read32(GUEST_DS_AR_BYTES);
7389 vmcs12->guest_fs_ar_bytes = vmcs_read32(GUEST_FS_AR_BYTES);
7390 vmcs12->guest_gs_ar_bytes = vmcs_read32(GUEST_GS_AR_BYTES);
7391 vmcs12->guest_ldtr_ar_bytes = vmcs_read32(GUEST_LDTR_AR_BYTES);
7392 vmcs12->guest_tr_ar_bytes = vmcs_read32(GUEST_TR_AR_BYTES);
7393 vmcs12->guest_es_base = vmcs_readl(GUEST_ES_BASE);
7394 vmcs12->guest_cs_base = vmcs_readl(GUEST_CS_BASE);
7395 vmcs12->guest_ss_base = vmcs_readl(GUEST_SS_BASE);
7396 vmcs12->guest_ds_base = vmcs_readl(GUEST_DS_BASE);
7397 vmcs12->guest_fs_base = vmcs_readl(GUEST_FS_BASE);
7398 vmcs12->guest_gs_base = vmcs_readl(GUEST_GS_BASE);
7399 vmcs12->guest_ldtr_base = vmcs_readl(GUEST_LDTR_BASE);
7400 vmcs12->guest_tr_base = vmcs_readl(GUEST_TR_BASE);
7401 vmcs12->guest_gdtr_base = vmcs_readl(GUEST_GDTR_BASE);
7402 vmcs12->guest_idtr_base = vmcs_readl(GUEST_IDTR_BASE);
7403
7404 vmcs12->guest_activity_state = vmcs_read32(GUEST_ACTIVITY_STATE);
7405 vmcs12->guest_interruptibility_info =
7406 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
7407 vmcs12->guest_pending_dbg_exceptions =
7408 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS);
7409
7410 vmcs12->vm_entry_controls =
7411 (vmcs12->vm_entry_controls & ~VM_ENTRY_IA32E_MODE) |
7412 (vmcs_read32(VM_ENTRY_CONTROLS) & VM_ENTRY_IA32E_MODE);
7413
7414 /* TODO: These cannot have changed unless we have MSR bitmaps and
7415 * the relevant bit asks not to trap the change */
7416 vmcs12->guest_ia32_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
7417 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_PAT)
7418 vmcs12->guest_ia32_pat = vmcs_read64(GUEST_IA32_PAT);
7419 vmcs12->guest_sysenter_cs = vmcs_read32(GUEST_SYSENTER_CS);
7420 vmcs12->guest_sysenter_esp = vmcs_readl(GUEST_SYSENTER_ESP);
7421 vmcs12->guest_sysenter_eip = vmcs_readl(GUEST_SYSENTER_EIP);
7422
7423 /* update exit information fields: */
7424
7425 vmcs12->vm_exit_reason = to_vmx(vcpu)->exit_reason;
7426 vmcs12->exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7427
7428 vmcs12->vm_exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
7429 if ((vmcs12->vm_exit_intr_info &
7430 (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK)) ==
7431 (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK))
7432 vmcs12->vm_exit_intr_error_code =
7433 vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
7434 vmcs12->idt_vectoring_info_field = 0;
7435 vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
7436 vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7437
7438 if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY)) {
7439 /* vm_entry_intr_info_field is cleared on exit. Emulate this
7440 * instead of reading the real value. */
7441 vmcs12->vm_entry_intr_info_field &= ~INTR_INFO_VALID_MASK;
7442
7443 /*
7444 * Transfer the event that L0 or L1 may wanted to inject into
7445 * L2 to IDT_VECTORING_INFO_FIELD.
7446 */
7447 vmcs12_save_pending_event(vcpu, vmcs12);
7448 }
7449
7450 /*
7451 * Drop what we picked up for L2 via vmx_complete_interrupts. It is
7452 * preserved above and would only end up incorrectly in L1.
7453 */
7454 vcpu->arch.nmi_injected = false;
7455 kvm_clear_exception_queue(vcpu);
7456 kvm_clear_interrupt_queue(vcpu);
7457 }
7458
7459 /*
7460 * A part of what we need to when the nested L2 guest exits and we want to
7461 * run its L1 parent, is to reset L1's guest state to the host state specified
7462 * in vmcs12.
7463 * This function is to be called not only on normal nested exit, but also on
7464 * a nested entry failure, as explained in Intel's spec, 3B.23.7 ("VM-Entry
7465 * Failures During or After Loading Guest State").
7466 * This function should be called when the active VMCS is L1's (vmcs01).
7467 */
7468 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
7469 struct vmcs12 *vmcs12)
7470 {
7471 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER)
7472 vcpu->arch.efer = vmcs12->host_ia32_efer;
7473 if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
7474 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
7475 else
7476 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
7477 vmx_set_efer(vcpu, vcpu->arch.efer);
7478
7479 kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->host_rsp);
7480 kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->host_rip);
7481 vmx_set_rflags(vcpu, X86_EFLAGS_BIT1);
7482 /*
7483 * Note that calling vmx_set_cr0 is important, even if cr0 hasn't
7484 * actually changed, because it depends on the current state of
7485 * fpu_active (which may have changed).
7486 * Note that vmx_set_cr0 refers to efer set above.
7487 */
7488 kvm_set_cr0(vcpu, vmcs12->host_cr0);
7489 /*
7490 * If we did fpu_activate()/fpu_deactivate() during L2's run, we need
7491 * to apply the same changes to L1's vmcs. We just set cr0 correctly,
7492 * but we also need to update cr0_guest_host_mask and exception_bitmap.
7493 */
7494 update_exception_bitmap(vcpu);
7495 vcpu->arch.cr0_guest_owned_bits = (vcpu->fpu_active ? X86_CR0_TS : 0);
7496 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
7497
7498 /*
7499 * Note that CR4_GUEST_HOST_MASK is already set in the original vmcs01
7500 * (KVM doesn't change it)- no reason to call set_cr4_guest_host_mask();
7501 */
7502 vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
7503 kvm_set_cr4(vcpu, vmcs12->host_cr4);
7504
7505 /* shadow page tables on either EPT or shadow page tables */
7506 kvm_set_cr3(vcpu, vmcs12->host_cr3);
7507 kvm_mmu_reset_context(vcpu);
7508
7509 if (enable_vpid) {
7510 /*
7511 * Trivially support vpid by letting L2s share their parent
7512 * L1's vpid. TODO: move to a more elaborate solution, giving
7513 * each L2 its own vpid and exposing the vpid feature to L1.
7514 */
7515 vmx_flush_tlb(vcpu);
7516 }
7517
7518
7519 vmcs_write32(GUEST_SYSENTER_CS, vmcs12->host_ia32_sysenter_cs);
7520 vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->host_ia32_sysenter_esp);
7521 vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->host_ia32_sysenter_eip);
7522 vmcs_writel(GUEST_IDTR_BASE, vmcs12->host_idtr_base);
7523 vmcs_writel(GUEST_GDTR_BASE, vmcs12->host_gdtr_base);
7524 vmcs_writel(GUEST_TR_BASE, vmcs12->host_tr_base);
7525 vmcs_writel(GUEST_GS_BASE, vmcs12->host_gs_base);
7526 vmcs_writel(GUEST_FS_BASE, vmcs12->host_fs_base);
7527 vmcs_write16(GUEST_ES_SELECTOR, vmcs12->host_es_selector);
7528 vmcs_write16(GUEST_CS_SELECTOR, vmcs12->host_cs_selector);
7529 vmcs_write16(GUEST_SS_SELECTOR, vmcs12->host_ss_selector);
7530 vmcs_write16(GUEST_DS_SELECTOR, vmcs12->host_ds_selector);
7531 vmcs_write16(GUEST_FS_SELECTOR, vmcs12->host_fs_selector);
7532 vmcs_write16(GUEST_GS_SELECTOR, vmcs12->host_gs_selector);
7533 vmcs_write16(GUEST_TR_SELECTOR, vmcs12->host_tr_selector);
7534
7535 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT)
7536 vmcs_write64(GUEST_IA32_PAT, vmcs12->host_ia32_pat);
7537 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
7538 vmcs_write64(GUEST_IA32_PERF_GLOBAL_CTRL,
7539 vmcs12->host_ia32_perf_global_ctrl);
7540
7541 kvm_set_dr(vcpu, 7, 0x400);
7542 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
7543 }
7544
7545 /*
7546 * Emulate an exit from nested guest (L2) to L1, i.e., prepare to run L1
7547 * and modify vmcs12 to make it see what it would expect to see there if
7548 * L2 was its real guest. Must only be called when in L2 (is_guest_mode())
7549 */
7550 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu)
7551 {
7552 struct vcpu_vmx *vmx = to_vmx(vcpu);
7553 int cpu;
7554 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7555
7556 /* trying to cancel vmlaunch/vmresume is a bug */
7557 WARN_ON_ONCE(vmx->nested.nested_run_pending);
7558
7559 leave_guest_mode(vcpu);
7560 prepare_vmcs12(vcpu, vmcs12);
7561
7562 cpu = get_cpu();
7563 vmx->loaded_vmcs = &vmx->vmcs01;
7564 vmx_vcpu_put(vcpu);
7565 vmx_vcpu_load(vcpu, cpu);
7566 vcpu->cpu = cpu;
7567 put_cpu();
7568
7569 vmx_segment_cache_clear(vmx);
7570
7571 /* if no vmcs02 cache requested, remove the one we used */
7572 if (VMCS02_POOL_SIZE == 0)
7573 nested_free_vmcs02(vmx, vmx->nested.current_vmptr);
7574
7575 load_vmcs12_host_state(vcpu, vmcs12);
7576
7577 /* Update TSC_OFFSET if TSC was changed while L2 ran */
7578 vmcs_write64(TSC_OFFSET, vmx->nested.vmcs01_tsc_offset);
7579
7580 /* This is needed for same reason as it was needed in prepare_vmcs02 */
7581 vmx->host_rsp = 0;
7582
7583 /* Unpin physical memory we referred to in vmcs02 */
7584 if (vmx->nested.apic_access_page) {
7585 nested_release_page(vmx->nested.apic_access_page);
7586 vmx->nested.apic_access_page = 0;
7587 }
7588
7589 /*
7590 * Exiting from L2 to L1, we're now back to L1 which thinks it just
7591 * finished a VMLAUNCH or VMRESUME instruction, so we need to set the
7592 * success or failure flag accordingly.
7593 */
7594 if (unlikely(vmx->fail)) {
7595 vmx->fail = 0;
7596 nested_vmx_failValid(vcpu, vmcs_read32(VM_INSTRUCTION_ERROR));
7597 } else
7598 nested_vmx_succeed(vcpu);
7599 }
7600
7601 /*
7602 * L1's failure to enter L2 is a subset of a normal exit, as explained in
7603 * 23.7 "VM-entry failures during or after loading guest state" (this also
7604 * lists the acceptable exit-reason and exit-qualification parameters).
7605 * It should only be called before L2 actually succeeded to run, and when
7606 * vmcs01 is current (it doesn't leave_guest_mode() or switch vmcss).
7607 */
7608 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
7609 struct vmcs12 *vmcs12,
7610 u32 reason, unsigned long qualification)
7611 {
7612 load_vmcs12_host_state(vcpu, vmcs12);
7613 vmcs12->vm_exit_reason = reason | VMX_EXIT_REASONS_FAILED_VMENTRY;
7614 vmcs12->exit_qualification = qualification;
7615 nested_vmx_succeed(vcpu);
7616 }
7617
7618 static int vmx_check_intercept(struct kvm_vcpu *vcpu,
7619 struct x86_instruction_info *info,
7620 enum x86_intercept_stage stage)
7621 {
7622 return X86EMUL_CONTINUE;
7623 }
7624
7625 static struct kvm_x86_ops vmx_x86_ops = {
7626 .cpu_has_kvm_support = cpu_has_kvm_support,
7627 .disabled_by_bios = vmx_disabled_by_bios,
7628 .hardware_setup = hardware_setup,
7629 .hardware_unsetup = hardware_unsetup,
7630 .check_processor_compatibility = vmx_check_processor_compat,
7631 .hardware_enable = hardware_enable,
7632 .hardware_disable = hardware_disable,
7633 .cpu_has_accelerated_tpr = report_flexpriority,
7634
7635 .vcpu_create = vmx_create_vcpu,
7636 .vcpu_free = vmx_free_vcpu,
7637 .vcpu_reset = vmx_vcpu_reset,
7638
7639 .prepare_guest_switch = vmx_save_host_state,
7640 .vcpu_load = vmx_vcpu_load,
7641 .vcpu_put = vmx_vcpu_put,
7642
7643 .update_db_bp_intercept = update_exception_bitmap,
7644 .get_msr = vmx_get_msr,
7645 .set_msr = vmx_set_msr,
7646 .get_segment_base = vmx_get_segment_base,
7647 .get_segment = vmx_get_segment,
7648 .set_segment = vmx_set_segment,
7649 .get_cpl = vmx_get_cpl,
7650 .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
7651 .decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
7652 .decache_cr3 = vmx_decache_cr3,
7653 .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
7654 .set_cr0 = vmx_set_cr0,
7655 .set_cr3 = vmx_set_cr3,
7656 .set_cr4 = vmx_set_cr4,
7657 .set_efer = vmx_set_efer,
7658 .get_idt = vmx_get_idt,
7659 .set_idt = vmx_set_idt,
7660 .get_gdt = vmx_get_gdt,
7661 .set_gdt = vmx_set_gdt,
7662 .set_dr7 = vmx_set_dr7,
7663 .cache_reg = vmx_cache_reg,
7664 .get_rflags = vmx_get_rflags,
7665 .set_rflags = vmx_set_rflags,
7666 .fpu_activate = vmx_fpu_activate,
7667 .fpu_deactivate = vmx_fpu_deactivate,
7668
7669 .tlb_flush = vmx_flush_tlb,
7670
7671 .run = vmx_vcpu_run,
7672 .handle_exit = vmx_handle_exit,
7673 .skip_emulated_instruction = skip_emulated_instruction,
7674 .set_interrupt_shadow = vmx_set_interrupt_shadow,
7675 .get_interrupt_shadow = vmx_get_interrupt_shadow,
7676 .patch_hypercall = vmx_patch_hypercall,
7677 .set_irq = vmx_inject_irq,
7678 .set_nmi = vmx_inject_nmi,
7679 .queue_exception = vmx_queue_exception,
7680 .cancel_injection = vmx_cancel_injection,
7681 .interrupt_allowed = vmx_interrupt_allowed,
7682 .nmi_allowed = vmx_nmi_allowed,
7683 .get_nmi_mask = vmx_get_nmi_mask,
7684 .set_nmi_mask = vmx_set_nmi_mask,
7685 .enable_nmi_window = enable_nmi_window,
7686 .enable_irq_window = enable_irq_window,
7687 .update_cr8_intercept = update_cr8_intercept,
7688 .set_virtual_x2apic_mode = vmx_set_virtual_x2apic_mode,
7689 .vm_has_apicv = vmx_vm_has_apicv,
7690 .load_eoi_exitmap = vmx_load_eoi_exitmap,
7691 .hwapic_irr_update = vmx_hwapic_irr_update,
7692 .hwapic_isr_update = vmx_hwapic_isr_update,
7693
7694 .set_tss_addr = vmx_set_tss_addr,
7695 .get_tdp_level = get_ept_level,
7696 .get_mt_mask = vmx_get_mt_mask,
7697
7698 .get_exit_info = vmx_get_exit_info,
7699
7700 .get_lpage_level = vmx_get_lpage_level,
7701
7702 .cpuid_update = vmx_cpuid_update,
7703
7704 .rdtscp_supported = vmx_rdtscp_supported,
7705 .invpcid_supported = vmx_invpcid_supported,
7706
7707 .set_supported_cpuid = vmx_set_supported_cpuid,
7708
7709 .has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
7710
7711 .set_tsc_khz = vmx_set_tsc_khz,
7712 .read_tsc_offset = vmx_read_tsc_offset,
7713 .write_tsc_offset = vmx_write_tsc_offset,
7714 .adjust_tsc_offset = vmx_adjust_tsc_offset,
7715 .compute_tsc_offset = vmx_compute_tsc_offset,
7716 .read_l1_tsc = vmx_read_l1_tsc,
7717
7718 .set_tdp_cr3 = vmx_set_cr3,
7719
7720 .check_intercept = vmx_check_intercept,
7721 };
7722
7723 static int __init vmx_init(void)
7724 {
7725 int r, i, msr;
7726
7727 rdmsrl_safe(MSR_EFER, &host_efer);
7728
7729 for (i = 0; i < NR_VMX_MSR; ++i)
7730 kvm_define_shared_msr(i, vmx_msr_index[i]);
7731
7732 vmx_io_bitmap_a = (unsigned long *)__get_free_page(GFP_KERNEL);
7733 if (!vmx_io_bitmap_a)
7734 return -ENOMEM;
7735
7736 r = -ENOMEM;
7737
7738 vmx_io_bitmap_b = (unsigned long *)__get_free_page(GFP_KERNEL);
7739 if (!vmx_io_bitmap_b)
7740 goto out;
7741
7742 vmx_msr_bitmap_legacy = (unsigned long *)__get_free_page(GFP_KERNEL);
7743 if (!vmx_msr_bitmap_legacy)
7744 goto out1;
7745
7746 vmx_msr_bitmap_legacy_x2apic =
7747 (unsigned long *)__get_free_page(GFP_KERNEL);
7748 if (!vmx_msr_bitmap_legacy_x2apic)
7749 goto out2;
7750
7751 vmx_msr_bitmap_longmode = (unsigned long *)__get_free_page(GFP_KERNEL);
7752 if (!vmx_msr_bitmap_longmode)
7753 goto out3;
7754
7755 vmx_msr_bitmap_longmode_x2apic =
7756 (unsigned long *)__get_free_page(GFP_KERNEL);
7757 if (!vmx_msr_bitmap_longmode_x2apic)
7758 goto out4;
7759
7760 /*
7761 * Allow direct access to the PC debug port (it is often used for I/O
7762 * delays, but the vmexits simply slow things down).
7763 */
7764 memset(vmx_io_bitmap_a, 0xff, PAGE_SIZE);
7765 clear_bit(0x80, vmx_io_bitmap_a);
7766
7767 memset(vmx_io_bitmap_b, 0xff, PAGE_SIZE);
7768
7769 memset(vmx_msr_bitmap_legacy, 0xff, PAGE_SIZE);
7770 memset(vmx_msr_bitmap_longmode, 0xff, PAGE_SIZE);
7771
7772 set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
7773
7774 r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx),
7775 __alignof__(struct vcpu_vmx), THIS_MODULE);
7776 if (r)
7777 goto out5;
7778
7779 #ifdef CONFIG_KEXEC
7780 rcu_assign_pointer(crash_vmclear_loaded_vmcss,
7781 crash_vmclear_local_loaded_vmcss);
7782 #endif
7783
7784 vmx_disable_intercept_for_msr(MSR_FS_BASE, false);
7785 vmx_disable_intercept_for_msr(MSR_GS_BASE, false);
7786 vmx_disable_intercept_for_msr(MSR_KERNEL_GS_BASE, true);
7787 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_CS, false);
7788 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_ESP, false);
7789 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_EIP, false);
7790 memcpy(vmx_msr_bitmap_legacy_x2apic,
7791 vmx_msr_bitmap_legacy, PAGE_SIZE);
7792 memcpy(vmx_msr_bitmap_longmode_x2apic,
7793 vmx_msr_bitmap_longmode, PAGE_SIZE);
7794
7795 if (enable_apicv_reg_vid) {
7796 for (msr = 0x800; msr <= 0x8ff; msr++)
7797 vmx_disable_intercept_msr_read_x2apic(msr);
7798
7799 /* According SDM, in x2apic mode, the whole id reg is used.
7800 * But in KVM, it only use the highest eight bits. Need to
7801 * intercept it */
7802 vmx_enable_intercept_msr_read_x2apic(0x802);
7803 /* TMCCT */
7804 vmx_enable_intercept_msr_read_x2apic(0x839);
7805 /* TPR */
7806 vmx_disable_intercept_msr_write_x2apic(0x808);
7807 /* EOI */
7808 vmx_disable_intercept_msr_write_x2apic(0x80b);
7809 /* SELF-IPI */
7810 vmx_disable_intercept_msr_write_x2apic(0x83f);
7811 }
7812
7813 if (enable_ept) {
7814 kvm_mmu_set_mask_ptes(0ull,
7815 (enable_ept_ad_bits) ? VMX_EPT_ACCESS_BIT : 0ull,
7816 (enable_ept_ad_bits) ? VMX_EPT_DIRTY_BIT : 0ull,
7817 0ull, VMX_EPT_EXECUTABLE_MASK);
7818 ept_set_mmio_spte_mask();
7819 kvm_enable_tdp();
7820 } else
7821 kvm_disable_tdp();
7822
7823 return 0;
7824
7825 out5:
7826 free_page((unsigned long)vmx_msr_bitmap_longmode_x2apic);
7827 out4:
7828 free_page((unsigned long)vmx_msr_bitmap_longmode);
7829 out3:
7830 free_page((unsigned long)vmx_msr_bitmap_legacy_x2apic);
7831 out2:
7832 free_page((unsigned long)vmx_msr_bitmap_legacy);
7833 out1:
7834 free_page((unsigned long)vmx_io_bitmap_b);
7835 out:
7836 free_page((unsigned long)vmx_io_bitmap_a);
7837 return r;
7838 }
7839
7840 static void __exit vmx_exit(void)
7841 {
7842 free_page((unsigned long)vmx_msr_bitmap_legacy_x2apic);
7843 free_page((unsigned long)vmx_msr_bitmap_longmode_x2apic);
7844 free_page((unsigned long)vmx_msr_bitmap_legacy);
7845 free_page((unsigned long)vmx_msr_bitmap_longmode);
7846 free_page((unsigned long)vmx_io_bitmap_b);
7847 free_page((unsigned long)vmx_io_bitmap_a);
7848
7849 #ifdef CONFIG_KEXEC
7850 rcu_assign_pointer(crash_vmclear_loaded_vmcss, NULL);
7851 synchronize_rcu();
7852 #endif
7853
7854 kvm_exit();
7855 }
7856
7857 module_init(vmx_init)
7858 module_exit(vmx_exit)
kernel source for telekom puls (alcatel/mediatek)