?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
KVM: VMX: Simplify saving guest rcx in vmx_vcpu_run
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / arch / x86 / kvm / vmx.c
blobda468c26639d55740b1a6623d931eca4d76a94ce
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
22 #include <linux/kvm_host.h>
23 #include <linux/module.h>
24 #include <linux/kernel.h>
25 #include <linux/mm.h>
26 #include <linux/highmem.h>
27 #include <linux/sched.h>
28 #include <linux/moduleparam.h>
29 #include <linux/ftrace_event.h>
30 #include <linux/slab.h>
31 #include <linux/tboot.h>
32 #include "kvm_cache_regs.h"
33 #include "x86.h"
34
35 #include <asm/io.h>
36 #include <asm/desc.h>
37 #include <asm/vmx.h>
38 #include <asm/virtext.h>
39 #include <asm/mce.h>
40 #include <asm/i387.h>
41 #include <asm/xcr.h>
42
43 #include "trace.h"
44
45 #define __ex(x) __kvm_handle_fault_on_reboot(x)
46
47 MODULE_AUTHOR("Qumranet");
48 MODULE_LICENSE("GPL");
49
50 static int __read_mostly bypass_guest_pf = 1;
51 module_param(bypass_guest_pf, bool, S_IRUGO);
52
53 static int __read_mostly enable_vpid = 1;
54 module_param_named(vpid, enable_vpid, bool, 0444);
55
56 static int __read_mostly flexpriority_enabled = 1;
57 module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
58
59 static int __read_mostly enable_ept = 1;
60 module_param_named(ept, enable_ept, bool, S_IRUGO);
61
62 static int __read_mostly enable_unrestricted_guest = 1;
63 module_param_named(unrestricted_guest,
64 enable_unrestricted_guest, bool, S_IRUGO);
65
66 static int __read_mostly emulate_invalid_guest_state = 0;
67 module_param(emulate_invalid_guest_state, bool, S_IRUGO);
68
69 static int __read_mostly vmm_exclusive = 1;
70 module_param(vmm_exclusive, bool, S_IRUGO);
71
72 static int __read_mostly yield_on_hlt = 1;
73 module_param(yield_on_hlt, bool, S_IRUGO);
74
75 #define KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST \
76 (X86_CR0_WP | X86_CR0_NE | X86_CR0_NW | X86_CR0_CD)
77 #define KVM_GUEST_CR0_MASK \
78 (KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
79 #define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST \
80 (X86_CR0_WP | X86_CR0_NE)
81 #define KVM_VM_CR0_ALWAYS_ON \
82 (KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
83 #define KVM_CR4_GUEST_OWNED_BITS \
84 (X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR \
85 | X86_CR4_OSXMMEXCPT)
86
87 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
88 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
89
90 #define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
91
92 /*
93 * These 2 parameters are used to config the controls for Pause-Loop Exiting:
94 * ple_gap: upper bound on the amount of time between two successive
95 * executions of PAUSE in a loop. Also indicate if ple enabled.
96 * According to test, this time is usually smaller than 128 cycles.
97 * ple_window: upper bound on the amount of time a guest is allowed to execute
98 * in a PAUSE loop. Tests indicate that most spinlocks are held for
99 * less than 2^12 cycles
100 * Time is measured based on a counter that runs at the same rate as the TSC,
101 * refer SDM volume 3b section 21.6.13 & 22.1.3.
102 */
103 #define KVM_VMX_DEFAULT_PLE_GAP 128
104 #define KVM_VMX_DEFAULT_PLE_WINDOW 4096
105 static int ple_gap = KVM_VMX_DEFAULT_PLE_GAP;
106 module_param(ple_gap, int, S_IRUGO);
107
108 static int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
109 module_param(ple_window, int, S_IRUGO);
110
111 #define NR_AUTOLOAD_MSRS 1
112
113 struct vmcs {
114 u32 revision_id;
115 u32 abort;
116 char data[0];
117 };
118
119 struct shared_msr_entry {
120 unsigned index;
121 u64 data;
122 u64 mask;
123 };
124
125 struct vcpu_vmx {
126 struct kvm_vcpu vcpu;
127 struct list_head local_vcpus_link;
128 unsigned long host_rsp;
129 int launched;
130 u8 fail;
131 u32 exit_intr_info;
132 u32 idt_vectoring_info;
133 struct shared_msr_entry *guest_msrs;
134 int nmsrs;
135 int save_nmsrs;
136 #ifdef CONFIG_X86_64
137 u64 msr_host_kernel_gs_base;
138 u64 msr_guest_kernel_gs_base;
139 #endif
140 struct vmcs *vmcs;
141 struct msr_autoload {
142 unsigned nr;
143 struct vmx_msr_entry guest[NR_AUTOLOAD_MSRS];
144 struct vmx_msr_entry host[NR_AUTOLOAD_MSRS];
145 } msr_autoload;
146 struct {
147 int loaded;
148 u16 fs_sel, gs_sel, ldt_sel;
149 int gs_ldt_reload_needed;
150 int fs_reload_needed;
151 } host_state;
152 struct {
153 int vm86_active;
154 ulong save_rflags;
155 struct kvm_save_segment {
156 u16 selector;
157 unsigned long base;
158 u32 limit;
159 u32 ar;
160 } tr, es, ds, fs, gs;
161 } rmode;
162 int vpid;
163 bool emulation_required;
164
165 /* Support for vnmi-less CPUs */
166 int soft_vnmi_blocked;
167 ktime_t entry_time;
168 s64 vnmi_blocked_time;
169 u32 exit_reason;
170
171 bool rdtscp_enabled;
172 };
173
174 static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
175 {
176 return container_of(vcpu, struct vcpu_vmx, vcpu);
177 }
178
179 static int init_rmode(struct kvm *kvm);
180 static u64 construct_eptp(unsigned long root_hpa);
181 static void kvm_cpu_vmxon(u64 addr);
182 static void kvm_cpu_vmxoff(void);
183 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3);
184
185 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
186 static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
187 static DEFINE_PER_CPU(struct list_head, vcpus_on_cpu);
188 static DEFINE_PER_CPU(struct desc_ptr, host_gdt);
189
190 static unsigned long *vmx_io_bitmap_a;
191 static unsigned long *vmx_io_bitmap_b;
192 static unsigned long *vmx_msr_bitmap_legacy;
193 static unsigned long *vmx_msr_bitmap_longmode;
194
195 static bool cpu_has_load_ia32_efer;
196
197 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
198 static DEFINE_SPINLOCK(vmx_vpid_lock);
199
200 static struct vmcs_config {
201 int size;
202 int order;
203 u32 revision_id;
204 u32 pin_based_exec_ctrl;
205 u32 cpu_based_exec_ctrl;
206 u32 cpu_based_2nd_exec_ctrl;
207 u32 vmexit_ctrl;
208 u32 vmentry_ctrl;
209 } vmcs_config;
210
211 static struct vmx_capability {
212 u32 ept;
213 u32 vpid;
214 } vmx_capability;
215
216 #define VMX_SEGMENT_FIELD(seg) \
217 [VCPU_SREG_##seg] = { \
218 .selector = GUEST_##seg##_SELECTOR, \
219 .base = GUEST_##seg##_BASE, \
220 .limit = GUEST_##seg##_LIMIT, \
221 .ar_bytes = GUEST_##seg##_AR_BYTES, \
222 }
223
224 static struct kvm_vmx_segment_field {
225 unsigned selector;
226 unsigned base;
227 unsigned limit;
228 unsigned ar_bytes;
229 } kvm_vmx_segment_fields[] = {
230 VMX_SEGMENT_FIELD(CS),
231 VMX_SEGMENT_FIELD(DS),
232 VMX_SEGMENT_FIELD(ES),
233 VMX_SEGMENT_FIELD(FS),
234 VMX_SEGMENT_FIELD(GS),
235 VMX_SEGMENT_FIELD(SS),
236 VMX_SEGMENT_FIELD(TR),
237 VMX_SEGMENT_FIELD(LDTR),
238 };
239
240 static u64 host_efer;
241
242 static void ept_save_pdptrs(struct kvm_vcpu *vcpu);
243
244 /*
245 * Keep MSR_STAR at the end, as setup_msrs() will try to optimize it
246 * away by decrementing the array size.
247 */
248 static const u32 vmx_msr_index[] = {
249 #ifdef CONFIG_X86_64
250 MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
251 #endif
252 MSR_EFER, MSR_TSC_AUX, MSR_STAR,
253 };
254 #define NR_VMX_MSR ARRAY_SIZE(vmx_msr_index)
255
256 static inline bool is_page_fault(u32 intr_info)
257 {
258 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
259 INTR_INFO_VALID_MASK)) ==
260 (INTR_TYPE_HARD_EXCEPTION | PF_VECTOR | INTR_INFO_VALID_MASK);
261 }
262
263 static inline bool is_no_device(u32 intr_info)
264 {
265 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
266 INTR_INFO_VALID_MASK)) ==
267 (INTR_TYPE_HARD_EXCEPTION | NM_VECTOR | INTR_INFO_VALID_MASK);
268 }
269
270 static inline bool is_invalid_opcode(u32 intr_info)
271 {
272 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
273 INTR_INFO_VALID_MASK)) ==
274 (INTR_TYPE_HARD_EXCEPTION | UD_VECTOR | INTR_INFO_VALID_MASK);
275 }
276
277 static inline bool is_external_interrupt(u32 intr_info)
278 {
279 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
280 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
281 }
282
283 static inline bool is_machine_check(u32 intr_info)
284 {
285 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
286 INTR_INFO_VALID_MASK)) ==
287 (INTR_TYPE_HARD_EXCEPTION | MC_VECTOR | INTR_INFO_VALID_MASK);
288 }
289
290 static inline bool cpu_has_vmx_msr_bitmap(void)
291 {
292 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS;
293 }
294
295 static inline bool cpu_has_vmx_tpr_shadow(void)
296 {
297 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW;
298 }
299
300 static inline bool vm_need_tpr_shadow(struct kvm *kvm)
301 {
302 return (cpu_has_vmx_tpr_shadow()) && (irqchip_in_kernel(kvm));
303 }
304
305 static inline bool cpu_has_secondary_exec_ctrls(void)
306 {
307 return vmcs_config.cpu_based_exec_ctrl &
308 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
309 }
310
311 static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
312 {
313 return vmcs_config.cpu_based_2nd_exec_ctrl &
314 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
315 }
316
317 static inline bool cpu_has_vmx_flexpriority(void)
318 {
319 return cpu_has_vmx_tpr_shadow() &&
320 cpu_has_vmx_virtualize_apic_accesses();
321 }
322
323 static inline bool cpu_has_vmx_ept_execute_only(void)
324 {
325 return vmx_capability.ept & VMX_EPT_EXECUTE_ONLY_BIT;
326 }
327
328 static inline bool cpu_has_vmx_eptp_uncacheable(void)
329 {
330 return vmx_capability.ept & VMX_EPTP_UC_BIT;
331 }
332
333 static inline bool cpu_has_vmx_eptp_writeback(void)
334 {
335 return vmx_capability.ept & VMX_EPTP_WB_BIT;
336 }
337
338 static inline bool cpu_has_vmx_ept_2m_page(void)
339 {
340 return vmx_capability.ept & VMX_EPT_2MB_PAGE_BIT;
341 }
342
343 static inline bool cpu_has_vmx_ept_1g_page(void)
344 {
345 return vmx_capability.ept & VMX_EPT_1GB_PAGE_BIT;
346 }
347
348 static inline bool cpu_has_vmx_ept_4levels(void)
349 {
350 return vmx_capability.ept & VMX_EPT_PAGE_WALK_4_BIT;
351 }
352
353 static inline bool cpu_has_vmx_invept_individual_addr(void)
354 {
355 return vmx_capability.ept & VMX_EPT_EXTENT_INDIVIDUAL_BIT;
356 }
357
358 static inline bool cpu_has_vmx_invept_context(void)
359 {
360 return vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT;
361 }
362
363 static inline bool cpu_has_vmx_invept_global(void)
364 {
365 return vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT;
366 }
367
368 static inline bool cpu_has_vmx_invvpid_single(void)
369 {
370 return vmx_capability.vpid & VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT;
371 }
372
373 static inline bool cpu_has_vmx_invvpid_global(void)
374 {
375 return vmx_capability.vpid & VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT;
376 }
377
378 static inline bool cpu_has_vmx_ept(void)
379 {
380 return vmcs_config.cpu_based_2nd_exec_ctrl &
381 SECONDARY_EXEC_ENABLE_EPT;
382 }
383
384 static inline bool cpu_has_vmx_unrestricted_guest(void)
385 {
386 return vmcs_config.cpu_based_2nd_exec_ctrl &
387 SECONDARY_EXEC_UNRESTRICTED_GUEST;
388 }
389
390 static inline bool cpu_has_vmx_ple(void)
391 {
392 return vmcs_config.cpu_based_2nd_exec_ctrl &
393 SECONDARY_EXEC_PAUSE_LOOP_EXITING;
394 }
395
396 static inline bool vm_need_virtualize_apic_accesses(struct kvm *kvm)
397 {
398 return flexpriority_enabled && irqchip_in_kernel(kvm);
399 }
400
401 static inline bool cpu_has_vmx_vpid(void)
402 {
403 return vmcs_config.cpu_based_2nd_exec_ctrl &
404 SECONDARY_EXEC_ENABLE_VPID;
405 }
406
407 static inline bool cpu_has_vmx_rdtscp(void)
408 {
409 return vmcs_config.cpu_based_2nd_exec_ctrl &
410 SECONDARY_EXEC_RDTSCP;
411 }
412
413 static inline bool cpu_has_virtual_nmis(void)
414 {
415 return vmcs_config.pin_based_exec_ctrl & PIN_BASED_VIRTUAL_NMIS;
416 }
417
418 static inline bool cpu_has_vmx_wbinvd_exit(void)
419 {
420 return vmcs_config.cpu_based_2nd_exec_ctrl &
421 SECONDARY_EXEC_WBINVD_EXITING;
422 }
423
424 static inline bool report_flexpriority(void)
425 {
426 return flexpriority_enabled;
427 }
428
429 static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
430 {
431 int i;
432
433 for (i = 0; i < vmx->nmsrs; ++i)
434 if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
435 return i;
436 return -1;
437 }
438
439 static inline void __invvpid(int ext, u16 vpid, gva_t gva)
440 {
441 struct {
442 u64 vpid : 16;
443 u64 rsvd : 48;
444 u64 gva;
445 } operand = { vpid, 0, gva };
446
447 asm volatile (__ex(ASM_VMX_INVVPID)
448 /* CF==1 or ZF==1 --> rc = -1 */
449 "; ja 1f ; ud2 ; 1:"
450 : : "a"(&operand), "c"(ext) : "cc", "memory");
451 }
452
453 static inline void __invept(int ext, u64 eptp, gpa_t gpa)
454 {
455 struct {
456 u64 eptp, gpa;
457 } operand = {eptp, gpa};
458
459 asm volatile (__ex(ASM_VMX_INVEPT)
460 /* CF==1 or ZF==1 --> rc = -1 */
461 "; ja 1f ; ud2 ; 1:\n"
462 : : "a" (&operand), "c" (ext) : "cc", "memory");
463 }
464
465 static struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
466 {
467 int i;
468
469 i = __find_msr_index(vmx, msr);
470 if (i >= 0)
471 return &vmx->guest_msrs[i];
472 return NULL;
473 }
474
475 static void vmcs_clear(struct vmcs *vmcs)
476 {
477 u64 phys_addr = __pa(vmcs);
478 u8 error;
479
480 asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0"
481 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
482 : "cc", "memory");
483 if (error)
484 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
485 vmcs, phys_addr);
486 }
487
488 static void vmcs_load(struct vmcs *vmcs)
489 {
490 u64 phys_addr = __pa(vmcs);
491 u8 error;
492
493 asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0"
494 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
495 : "cc", "memory");
496 if (error)
497 printk(KERN_ERR "kvm: vmptrld %p/%llx fail\n",
498 vmcs, phys_addr);
499 }
500
501 static void __vcpu_clear(void *arg)
502 {
503 struct vcpu_vmx *vmx = arg;
504 int cpu = raw_smp_processor_id();
505
506 if (vmx->vcpu.cpu == cpu)
507 vmcs_clear(vmx->vmcs);
508 if (per_cpu(current_vmcs, cpu) == vmx->vmcs)
509 per_cpu(current_vmcs, cpu) = NULL;
510 list_del(&vmx->local_vcpus_link);
511 vmx->vcpu.cpu = -1;
512 vmx->launched = 0;
513 }
514
515 static void vcpu_clear(struct vcpu_vmx *vmx)
516 {
517 if (vmx->vcpu.cpu == -1)
518 return;
519 smp_call_function_single(vmx->vcpu.cpu, __vcpu_clear, vmx, 1);
520 }
521
522 static inline void vpid_sync_vcpu_single(struct vcpu_vmx *vmx)
523 {
524 if (vmx->vpid == 0)
525 return;
526
527 if (cpu_has_vmx_invvpid_single())
528 __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vmx->vpid, 0);
529 }
530
531 static inline void vpid_sync_vcpu_global(void)
532 {
533 if (cpu_has_vmx_invvpid_global())
534 __invvpid(VMX_VPID_EXTENT_ALL_CONTEXT, 0, 0);
535 }
536
537 static inline void vpid_sync_context(struct vcpu_vmx *vmx)
538 {
539 if (cpu_has_vmx_invvpid_single())
540 vpid_sync_vcpu_single(vmx);
541 else
542 vpid_sync_vcpu_global();
543 }
544
545 static inline void ept_sync_global(void)
546 {
547 if (cpu_has_vmx_invept_global())
548 __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0);
549 }
550
551 static inline void ept_sync_context(u64 eptp)
552 {
553 if (enable_ept) {
554 if (cpu_has_vmx_invept_context())
555 __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0);
556 else
557 ept_sync_global();
558 }
559 }
560
561 static inline void ept_sync_individual_addr(u64 eptp, gpa_t gpa)
562 {
563 if (enable_ept) {
564 if (cpu_has_vmx_invept_individual_addr())
565 __invept(VMX_EPT_EXTENT_INDIVIDUAL_ADDR,
566 eptp, gpa);
567 else
568 ept_sync_context(eptp);
569 }
570 }
571
572 static unsigned long vmcs_readl(unsigned long field)
573 {
574 unsigned long value = 0;
575
576 asm volatile (__ex(ASM_VMX_VMREAD_RDX_RAX)
577 : "+a"(value) : "d"(field) : "cc");
578 return value;
579 }
580
581 static u16 vmcs_read16(unsigned long field)
582 {
583 return vmcs_readl(field);
584 }
585
586 static u32 vmcs_read32(unsigned long field)
587 {
588 return vmcs_readl(field);
589 }
590
591 static u64 vmcs_read64(unsigned long field)
592 {
593 #ifdef CONFIG_X86_64
594 return vmcs_readl(field);
595 #else
596 return vmcs_readl(field) | ((u64)vmcs_readl(field+1) << 32);
597 #endif
598 }
599
600 static noinline void vmwrite_error(unsigned long field, unsigned long value)
601 {
602 printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
603 field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
604 dump_stack();
605 }
606
607 static void vmcs_writel(unsigned long field, unsigned long value)
608 {
609 u8 error;
610
611 asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0"
612 : "=q"(error) : "a"(value), "d"(field) : "cc");
613 if (unlikely(error))
614 vmwrite_error(field, value);
615 }
616
617 static void vmcs_write16(unsigned long field, u16 value)
618 {
619 vmcs_writel(field, value);
620 }
621
622 static void vmcs_write32(unsigned long field, u32 value)
623 {
624 vmcs_writel(field, value);
625 }
626
627 static void vmcs_write64(unsigned long field, u64 value)
628 {
629 vmcs_writel(field, value);
630 #ifndef CONFIG_X86_64
631 asm volatile ("");
632 vmcs_writel(field+1, value >> 32);
633 #endif
634 }
635
636 static void vmcs_clear_bits(unsigned long field, u32 mask)
637 {
638 vmcs_writel(field, vmcs_readl(field) & ~mask);
639 }
640
641 static void vmcs_set_bits(unsigned long field, u32 mask)
642 {
643 vmcs_writel(field, vmcs_readl(field) | mask);
644 }
645
646 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
647 {
648 u32 eb;
649
650 eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
651 (1u << NM_VECTOR) | (1u << DB_VECTOR);
652 if ((vcpu->guest_debug &
653 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
654 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
655 eb |= 1u << BP_VECTOR;
656 if (to_vmx(vcpu)->rmode.vm86_active)
657 eb = ~0;
658 if (enable_ept)
659 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
660 if (vcpu->fpu_active)
661 eb &= ~(1u << NM_VECTOR);
662 vmcs_write32(EXCEPTION_BITMAP, eb);
663 }
664
665 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
666 {
667 unsigned i;
668 struct msr_autoload *m = &vmx->msr_autoload;
669
670 if (msr == MSR_EFER && cpu_has_load_ia32_efer) {
671 vmcs_clear_bits(VM_ENTRY_CONTROLS, VM_ENTRY_LOAD_IA32_EFER);
672 vmcs_clear_bits(VM_EXIT_CONTROLS, VM_EXIT_LOAD_IA32_EFER);
673 return;
674 }
675
676 for (i = 0; i < m->nr; ++i)
677 if (m->guest[i].index == msr)
678 break;
679
680 if (i == m->nr)
681 return;
682 --m->nr;
683 m->guest[i] = m->guest[m->nr];
684 m->host[i] = m->host[m->nr];
685 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
686 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
687 }
688
689 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
690 u64 guest_val, u64 host_val)
691 {
692 unsigned i;
693 struct msr_autoload *m = &vmx->msr_autoload;
694
695 if (msr == MSR_EFER && cpu_has_load_ia32_efer) {
696 vmcs_write64(GUEST_IA32_EFER, guest_val);
697 vmcs_write64(HOST_IA32_EFER, host_val);
698 vmcs_set_bits(VM_ENTRY_CONTROLS, VM_ENTRY_LOAD_IA32_EFER);
699 vmcs_set_bits(VM_EXIT_CONTROLS, VM_EXIT_LOAD_IA32_EFER);
700 return;
701 }
702
703 for (i = 0; i < m->nr; ++i)
704 if (m->guest[i].index == msr)
705 break;
706
707 if (i == m->nr) {
708 ++m->nr;
709 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
710 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
711 }
712
713 m->guest[i].index = msr;
714 m->guest[i].value = guest_val;
715 m->host[i].index = msr;
716 m->host[i].value = host_val;
717 }
718
719 static void reload_tss(void)
720 {
721 /*
722 * VT restores TR but not its size. Useless.
723 */
724 struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
725 struct desc_struct *descs;
726
727 descs = (void *)gdt->address;
728 descs[GDT_ENTRY_TSS].type = 9; /* available TSS */
729 load_TR_desc();
730 }
731
732 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
733 {
734 u64 guest_efer;
735 u64 ignore_bits;
736
737 guest_efer = vmx->vcpu.arch.efer;
738
739 /*
740 * NX is emulated; LMA and LME handled by hardware; SCE meaninless
741 * outside long mode
742 */
743 ignore_bits = EFER_NX | EFER_SCE;
744 #ifdef CONFIG_X86_64
745 ignore_bits |= EFER_LMA | EFER_LME;
746 /* SCE is meaningful only in long mode on Intel */
747 if (guest_efer & EFER_LMA)
748 ignore_bits &= ~(u64)EFER_SCE;
749 #endif
750 guest_efer &= ~ignore_bits;
751 guest_efer |= host_efer & ignore_bits;
752 vmx->guest_msrs[efer_offset].data = guest_efer;
753 vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
754
755 clear_atomic_switch_msr(vmx, MSR_EFER);
756 /* On ept, can't emulate nx, and must switch nx atomically */
757 if (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX)) {
758 guest_efer = vmx->vcpu.arch.efer;
759 if (!(guest_efer & EFER_LMA))
760 guest_efer &= ~EFER_LME;
761 add_atomic_switch_msr(vmx, MSR_EFER, guest_efer, host_efer);
762 return false;
763 }
764
765 return true;
766 }
767
768 static unsigned long segment_base(u16 selector)
769 {
770 struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
771 struct desc_struct *d;
772 unsigned long table_base;
773 unsigned long v;
774
775 if (!(selector & ~3))
776 return 0;
777
778 table_base = gdt->address;
779
780 if (selector & 4) { /* from ldt */
781 u16 ldt_selector = kvm_read_ldt();
782
783 if (!(ldt_selector & ~3))
784 return 0;
785
786 table_base = segment_base(ldt_selector);
787 }
788 d = (struct desc_struct *)(table_base + (selector & ~7));
789 v = get_desc_base(d);
790 #ifdef CONFIG_X86_64
791 if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
792 v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
793 #endif
794 return v;
795 }
796
797 static inline unsigned long kvm_read_tr_base(void)
798 {
799 u16 tr;
800 asm("str %0" : "=g"(tr));
801 return segment_base(tr);
802 }
803
804 static void vmx_save_host_state(struct kvm_vcpu *vcpu)
805 {
806 struct vcpu_vmx *vmx = to_vmx(vcpu);
807 int i;
808
809 if (vmx->host_state.loaded)
810 return;
811
812 vmx->host_state.loaded = 1;
813 /*
814 * Set host fs and gs selectors. Unfortunately, 22.2.3 does not
815 * allow segment selectors with cpl > 0 or ti == 1.
816 */
817 vmx->host_state.ldt_sel = kvm_read_ldt();
818 vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
819 savesegment(fs, vmx->host_state.fs_sel);
820 if (!(vmx->host_state.fs_sel & 7)) {
821 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
822 vmx->host_state.fs_reload_needed = 0;
823 } else {
824 vmcs_write16(HOST_FS_SELECTOR, 0);
825 vmx->host_state.fs_reload_needed = 1;
826 }
827 savesegment(gs, vmx->host_state.gs_sel);
828 if (!(vmx->host_state.gs_sel & 7))
829 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
830 else {
831 vmcs_write16(HOST_GS_SELECTOR, 0);
832 vmx->host_state.gs_ldt_reload_needed = 1;
833 }
834
835 #ifdef CONFIG_X86_64
836 vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
837 vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
838 #else
839 vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
840 vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
841 #endif
842
843 #ifdef CONFIG_X86_64
844 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
845 if (is_long_mode(&vmx->vcpu))
846 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
847 #endif
848 for (i = 0; i < vmx->save_nmsrs; ++i)
849 kvm_set_shared_msr(vmx->guest_msrs[i].index,
850 vmx->guest_msrs[i].data,
851 vmx->guest_msrs[i].mask);
852 }
853
854 static void __vmx_load_host_state(struct vcpu_vmx *vmx)
855 {
856 if (!vmx->host_state.loaded)
857 return;
858
859 ++vmx->vcpu.stat.host_state_reload;
860 vmx->host_state.loaded = 0;
861 #ifdef CONFIG_X86_64
862 if (is_long_mode(&vmx->vcpu))
863 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
864 #endif
865 if (vmx->host_state.gs_ldt_reload_needed) {
866 kvm_load_ldt(vmx->host_state.ldt_sel);
867 #ifdef CONFIG_X86_64
868 load_gs_index(vmx->host_state.gs_sel);
869 #else
870 loadsegment(gs, vmx->host_state.gs_sel);
871 #endif
872 }
873 if (vmx->host_state.fs_reload_needed)
874 loadsegment(fs, vmx->host_state.fs_sel);
875 reload_tss();
876 #ifdef CONFIG_X86_64
877 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
878 #endif
879 if (current_thread_info()->status & TS_USEDFPU)
880 clts();
881 load_gdt(&__get_cpu_var(host_gdt));
882 }
883
884 static void vmx_load_host_state(struct vcpu_vmx *vmx)
885 {
886 preempt_disable();
887 __vmx_load_host_state(vmx);
888 preempt_enable();
889 }
890
891 /*
892 * Switches to specified vcpu, until a matching vcpu_put(), but assumes
893 * vcpu mutex is already taken.
894 */
895 static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
896 {
897 struct vcpu_vmx *vmx = to_vmx(vcpu);
898 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
899
900 if (!vmm_exclusive)
901 kvm_cpu_vmxon(phys_addr);
902 else if (vcpu->cpu != cpu)
903 vcpu_clear(vmx);
904
905 if (per_cpu(current_vmcs, cpu) != vmx->vmcs) {
906 per_cpu(current_vmcs, cpu) = vmx->vmcs;
907 vmcs_load(vmx->vmcs);
908 }
909
910 if (vcpu->cpu != cpu) {
911 struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
912 unsigned long sysenter_esp;
913
914 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
915 local_irq_disable();
916 list_add(&vmx->local_vcpus_link,
917 &per_cpu(vcpus_on_cpu, cpu));
918 local_irq_enable();
919
920 /*
921 * Linux uses per-cpu TSS and GDT, so set these when switching
922 * processors.
923 */
924 vmcs_writel(HOST_TR_BASE, kvm_read_tr_base()); /* 22.2.4 */
925 vmcs_writel(HOST_GDTR_BASE, gdt->address); /* 22.2.4 */
926
927 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
928 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
929 }
930 }
931
932 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
933 {
934 __vmx_load_host_state(to_vmx(vcpu));
935 if (!vmm_exclusive) {
936 __vcpu_clear(to_vmx(vcpu));
937 kvm_cpu_vmxoff();
938 }
939 }
940
941 static void vmx_fpu_activate(struct kvm_vcpu *vcpu)
942 {
943 ulong cr0;
944
945 if (vcpu->fpu_active)
946 return;
947 vcpu->fpu_active = 1;
948 cr0 = vmcs_readl(GUEST_CR0);
949 cr0 &= ~(X86_CR0_TS | X86_CR0_MP);
950 cr0 |= kvm_read_cr0_bits(vcpu, X86_CR0_TS | X86_CR0_MP);
951 vmcs_writel(GUEST_CR0, cr0);
952 update_exception_bitmap(vcpu);
953 vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
954 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
955 }
956
957 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu);
958
959 static void vmx_fpu_deactivate(struct kvm_vcpu *vcpu)
960 {
961 vmx_decache_cr0_guest_bits(vcpu);
962 vmcs_set_bits(GUEST_CR0, X86_CR0_TS | X86_CR0_MP);
963 update_exception_bitmap(vcpu);
964 vcpu->arch.cr0_guest_owned_bits = 0;
965 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
966 vmcs_writel(CR0_READ_SHADOW, vcpu->arch.cr0);
967 }
968
969 static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
970 {
971 unsigned long rflags, save_rflags;
972
973 rflags = vmcs_readl(GUEST_RFLAGS);
974 if (to_vmx(vcpu)->rmode.vm86_active) {
975 rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
976 save_rflags = to_vmx(vcpu)->rmode.save_rflags;
977 rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
978 }
979 return rflags;
980 }
981
982 static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
983 {
984 if (to_vmx(vcpu)->rmode.vm86_active) {
985 to_vmx(vcpu)->rmode.save_rflags = rflags;
986 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
987 }
988 vmcs_writel(GUEST_RFLAGS, rflags);
989 }
990
991 static u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
992 {
993 u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
994 int ret = 0;
995
996 if (interruptibility & GUEST_INTR_STATE_STI)
997 ret |= KVM_X86_SHADOW_INT_STI;
998 if (interruptibility & GUEST_INTR_STATE_MOV_SS)
999 ret |= KVM_X86_SHADOW_INT_MOV_SS;
1000
1001 return ret & mask;
1002 }
1003
1004 static void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1005 {
1006 u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1007 u32 interruptibility = interruptibility_old;
1008
1009 interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
1010
1011 if (mask & KVM_X86_SHADOW_INT_MOV_SS)
1012 interruptibility |= GUEST_INTR_STATE_MOV_SS;
1013 else if (mask & KVM_X86_SHADOW_INT_STI)
1014 interruptibility |= GUEST_INTR_STATE_STI;
1015
1016 if ((interruptibility != interruptibility_old))
1017 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
1018 }
1019
1020 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
1021 {
1022 unsigned long rip;
1023
1024 rip = kvm_rip_read(vcpu);
1025 rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
1026 kvm_rip_write(vcpu, rip);
1027
1028 /* skipping an emulated instruction also counts */
1029 vmx_set_interrupt_shadow(vcpu, 0);
1030 }
1031
1032 static void vmx_clear_hlt(struct kvm_vcpu *vcpu)
1033 {
1034 /* Ensure that we clear the HLT state in the VMCS. We don't need to
1035 * explicitly skip the instruction because if the HLT state is set, then
1036 * the instruction is already executing and RIP has already been
1037 * advanced. */
1038 if (!yield_on_hlt &&
1039 vmcs_read32(GUEST_ACTIVITY_STATE) == GUEST_ACTIVITY_HLT)
1040 vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
1041 }
1042
1043 static void vmx_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
1044 bool has_error_code, u32 error_code,
1045 bool reinject)
1046 {
1047 struct vcpu_vmx *vmx = to_vmx(vcpu);
1048 u32 intr_info = nr | INTR_INFO_VALID_MASK;
1049
1050 if (has_error_code) {
1051 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
1052 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
1053 }
1054
1055 if (vmx->rmode.vm86_active) {
1056 if (kvm_inject_realmode_interrupt(vcpu, nr) != EMULATE_DONE)
1057 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
1058 return;
1059 }
1060
1061 if (kvm_exception_is_soft(nr)) {
1062 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
1063 vmx->vcpu.arch.event_exit_inst_len);
1064 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
1065 } else
1066 intr_info |= INTR_TYPE_HARD_EXCEPTION;
1067
1068 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
1069 vmx_clear_hlt(vcpu);
1070 }
1071
1072 static bool vmx_rdtscp_supported(void)
1073 {
1074 return cpu_has_vmx_rdtscp();
1075 }
1076
1077 /*
1078 * Swap MSR entry in host/guest MSR entry array.
1079 */
1080 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
1081 {
1082 struct shared_msr_entry tmp;
1083
1084 tmp = vmx->guest_msrs[to];
1085 vmx->guest_msrs[to] = vmx->guest_msrs[from];
1086 vmx->guest_msrs[from] = tmp;
1087 }
1088
1089 /*
1090 * Set up the vmcs to automatically save and restore system
1091 * msrs. Don't touch the 64-bit msrs if the guest is in legacy
1092 * mode, as fiddling with msrs is very expensive.
1093 */
1094 static void setup_msrs(struct vcpu_vmx *vmx)
1095 {
1096 int save_nmsrs, index;
1097 unsigned long *msr_bitmap;
1098
1099 vmx_load_host_state(vmx);
1100 save_nmsrs = 0;
1101 #ifdef CONFIG_X86_64
1102 if (is_long_mode(&vmx->vcpu)) {
1103 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
1104 if (index >= 0)
1105 move_msr_up(vmx, index, save_nmsrs++);
1106 index = __find_msr_index(vmx, MSR_LSTAR);
1107 if (index >= 0)
1108 move_msr_up(vmx, index, save_nmsrs++);
1109 index = __find_msr_index(vmx, MSR_CSTAR);
1110 if (index >= 0)
1111 move_msr_up(vmx, index, save_nmsrs++);
1112 index = __find_msr_index(vmx, MSR_TSC_AUX);
1113 if (index >= 0 && vmx->rdtscp_enabled)
1114 move_msr_up(vmx, index, save_nmsrs++);
1115 /*
1116 * MSR_STAR is only needed on long mode guests, and only
1117 * if efer.sce is enabled.
1118 */
1119 index = __find_msr_index(vmx, MSR_STAR);
1120 if ((index >= 0) && (vmx->vcpu.arch.efer & EFER_SCE))
1121 move_msr_up(vmx, index, save_nmsrs++);
1122 }
1123 #endif
1124 index = __find_msr_index(vmx, MSR_EFER);
1125 if (index >= 0 && update_transition_efer(vmx, index))
1126 move_msr_up(vmx, index, save_nmsrs++);
1127
1128 vmx->save_nmsrs = save_nmsrs;
1129
1130 if (cpu_has_vmx_msr_bitmap()) {
1131 if (is_long_mode(&vmx->vcpu))
1132 msr_bitmap = vmx_msr_bitmap_longmode;
1133 else
1134 msr_bitmap = vmx_msr_bitmap_legacy;
1135
1136 vmcs_write64(MSR_BITMAP, __pa(msr_bitmap));
1137 }
1138 }
1139
1140 /*
1141 * reads and returns guest's timestamp counter "register"
1142 * guest_tsc = host_tsc + tsc_offset -- 21.3
1143 */
1144 static u64 guest_read_tsc(void)
1145 {
1146 u64 host_tsc, tsc_offset;
1147
1148 rdtscll(host_tsc);
1149 tsc_offset = vmcs_read64(TSC_OFFSET);
1150 return host_tsc + tsc_offset;
1151 }
1152
1153 /*
1154 * writes 'offset' into guest's timestamp counter offset register
1155 */
1156 static void vmx_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1157 {
1158 vmcs_write64(TSC_OFFSET, offset);
1159 }
1160
1161 static void vmx_adjust_tsc_offset(struct kvm_vcpu *vcpu, s64 adjustment)
1162 {
1163 u64 offset = vmcs_read64(TSC_OFFSET);
1164 vmcs_write64(TSC_OFFSET, offset + adjustment);
1165 }
1166
1167 /*
1168 * Reads an msr value (of 'msr_index') into 'pdata'.
1169 * Returns 0 on success, non-0 otherwise.
1170 * Assumes vcpu_load() was already called.
1171 */
1172 static int vmx_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1173 {
1174 u64 data;
1175 struct shared_msr_entry *msr;
1176
1177 if (!pdata) {
1178 printk(KERN_ERR "BUG: get_msr called with NULL pdata\n");
1179 return -EINVAL;
1180 }
1181
1182 switch (msr_index) {
1183 #ifdef CONFIG_X86_64
1184 case MSR_FS_BASE:
1185 data = vmcs_readl(GUEST_FS_BASE);
1186 break;
1187 case MSR_GS_BASE:
1188 data = vmcs_readl(GUEST_GS_BASE);
1189 break;
1190 case MSR_KERNEL_GS_BASE:
1191 vmx_load_host_state(to_vmx(vcpu));
1192 data = to_vmx(vcpu)->msr_guest_kernel_gs_base;
1193 break;
1194 #endif
1195 case MSR_EFER:
1196 return kvm_get_msr_common(vcpu, msr_index, pdata);
1197 case MSR_IA32_TSC:
1198 data = guest_read_tsc();
1199 break;
1200 case MSR_IA32_SYSENTER_CS:
1201 data = vmcs_read32(GUEST_SYSENTER_CS);
1202 break;
1203 case MSR_IA32_SYSENTER_EIP:
1204 data = vmcs_readl(GUEST_SYSENTER_EIP);
1205 break;
1206 case MSR_IA32_SYSENTER_ESP:
1207 data = vmcs_readl(GUEST_SYSENTER_ESP);
1208 break;
1209 case MSR_TSC_AUX:
1210 if (!to_vmx(vcpu)->rdtscp_enabled)
1211 return 1;
1212 /* Otherwise falls through */
1213 default:
1214 vmx_load_host_state(to_vmx(vcpu));
1215 msr = find_msr_entry(to_vmx(vcpu), msr_index);
1216 if (msr) {
1217 vmx_load_host_state(to_vmx(vcpu));
1218 data = msr->data;
1219 break;
1220 }
1221 return kvm_get_msr_common(vcpu, msr_index, pdata);
1222 }
1223
1224 *pdata = data;
1225 return 0;
1226 }
1227
1228 /*
1229 * Writes msr value into into the appropriate "register".
1230 * Returns 0 on success, non-0 otherwise.
1231 * Assumes vcpu_load() was already called.
1232 */
1233 static int vmx_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1234 {
1235 struct vcpu_vmx *vmx = to_vmx(vcpu);
1236 struct shared_msr_entry *msr;
1237 int ret = 0;
1238
1239 switch (msr_index) {
1240 case MSR_EFER:
1241 vmx_load_host_state(vmx);
1242 ret = kvm_set_msr_common(vcpu, msr_index, data);
1243 break;
1244 #ifdef CONFIG_X86_64
1245 case MSR_FS_BASE:
1246 vmcs_writel(GUEST_FS_BASE, data);
1247 break;
1248 case MSR_GS_BASE:
1249 vmcs_writel(GUEST_GS_BASE, data);
1250 break;
1251 case MSR_KERNEL_GS_BASE:
1252 vmx_load_host_state(vmx);
1253 vmx->msr_guest_kernel_gs_base = data;
1254 break;
1255 #endif
1256 case MSR_IA32_SYSENTER_CS:
1257 vmcs_write32(GUEST_SYSENTER_CS, data);
1258 break;
1259 case MSR_IA32_SYSENTER_EIP:
1260 vmcs_writel(GUEST_SYSENTER_EIP, data);
1261 break;
1262 case MSR_IA32_SYSENTER_ESP:
1263 vmcs_writel(GUEST_SYSENTER_ESP, data);
1264 break;
1265 case MSR_IA32_TSC:
1266 kvm_write_tsc(vcpu, data);
1267 break;
1268 case MSR_IA32_CR_PAT:
1269 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
1270 vmcs_write64(GUEST_IA32_PAT, data);
1271 vcpu->arch.pat = data;
1272 break;
1273 }
1274 ret = kvm_set_msr_common(vcpu, msr_index, data);
1275 break;
1276 case MSR_TSC_AUX:
1277 if (!vmx->rdtscp_enabled)
1278 return 1;
1279 /* Check reserved bit, higher 32 bits should be zero */
1280 if ((data >> 32) != 0)
1281 return 1;
1282 /* Otherwise falls through */
1283 default:
1284 msr = find_msr_entry(vmx, msr_index);
1285 if (msr) {
1286 vmx_load_host_state(vmx);
1287 msr->data = data;
1288 break;
1289 }
1290 ret = kvm_set_msr_common(vcpu, msr_index, data);
1291 }
1292
1293 return ret;
1294 }
1295
1296 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
1297 {
1298 __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
1299 switch (reg) {
1300 case VCPU_REGS_RSP:
1301 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
1302 break;
1303 case VCPU_REGS_RIP:
1304 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
1305 break;
1306 case VCPU_EXREG_PDPTR:
1307 if (enable_ept)
1308 ept_save_pdptrs(vcpu);
1309 break;
1310 default:
1311 break;
1312 }
1313 }
1314
1315 static void set_guest_debug(struct kvm_vcpu *vcpu, struct kvm_guest_debug *dbg)
1316 {
1317 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1318 vmcs_writel(GUEST_DR7, dbg->arch.debugreg[7]);
1319 else
1320 vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
1321
1322 update_exception_bitmap(vcpu);
1323 }
1324
1325 static __init int cpu_has_kvm_support(void)
1326 {
1327 return cpu_has_vmx();
1328 }
1329
1330 static __init int vmx_disabled_by_bios(void)
1331 {
1332 u64 msr;
1333
1334 rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
1335 if (msr & FEATURE_CONTROL_LOCKED) {
1336 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
1337 && tboot_enabled())
1338 return 1;
1339 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
1340 && !tboot_enabled()) {
1341 printk(KERN_WARNING "kvm: disable TXT in the BIOS or "
1342 " activate TXT before enabling KVM\n");
1343 return 1;
1344 }
1345 }
1346
1347 return 0;
1348 /* locked but not enabled */
1349 }
1350
1351 static void kvm_cpu_vmxon(u64 addr)
1352 {
1353 asm volatile (ASM_VMX_VMXON_RAX
1354 : : "a"(&addr), "m"(addr)
1355 : "memory", "cc");
1356 }
1357
1358 static int hardware_enable(void *garbage)
1359 {
1360 int cpu = raw_smp_processor_id();
1361 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
1362 u64 old, test_bits;
1363
1364 if (read_cr4() & X86_CR4_VMXE)
1365 return -EBUSY;
1366
1367 INIT_LIST_HEAD(&per_cpu(vcpus_on_cpu, cpu));
1368 rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
1369
1370 test_bits = FEATURE_CONTROL_LOCKED;
1371 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
1372 if (tboot_enabled())
1373 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX;
1374
1375 if ((old & test_bits) != test_bits) {
1376 /* enable and lock */
1377 wrmsrl(MSR_IA32_FEATURE_CONTROL, old | test_bits);
1378 }
1379 write_cr4(read_cr4() | X86_CR4_VMXE); /* FIXME: not cpu hotplug safe */
1380
1381 if (vmm_exclusive) {
1382 kvm_cpu_vmxon(phys_addr);
1383 ept_sync_global();
1384 }
1385
1386 store_gdt(&__get_cpu_var(host_gdt));
1387
1388 return 0;
1389 }
1390
1391 static void vmclear_local_vcpus(void)
1392 {
1393 int cpu = raw_smp_processor_id();
1394 struct vcpu_vmx *vmx, *n;
1395
1396 list_for_each_entry_safe(vmx, n, &per_cpu(vcpus_on_cpu, cpu),
1397 local_vcpus_link)
1398 __vcpu_clear(vmx);
1399 }
1400
1401
1402 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
1403 * tricks.
1404 */
1405 static void kvm_cpu_vmxoff(void)
1406 {
1407 asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
1408 }
1409
1410 static void hardware_disable(void *garbage)
1411 {
1412 if (vmm_exclusive) {
1413 vmclear_local_vcpus();
1414 kvm_cpu_vmxoff();
1415 }
1416 write_cr4(read_cr4() & ~X86_CR4_VMXE);
1417 }
1418
1419 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
1420 u32 msr, u32 *result)
1421 {
1422 u32 vmx_msr_low, vmx_msr_high;
1423 u32 ctl = ctl_min | ctl_opt;
1424
1425 rdmsr(msr, vmx_msr_low, vmx_msr_high);
1426
1427 ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
1428 ctl |= vmx_msr_low; /* bit == 1 in low word ==> must be one */
1429
1430 /* Ensure minimum (required) set of control bits are supported. */
1431 if (ctl_min & ~ctl)
1432 return -EIO;
1433
1434 *result = ctl;
1435 return 0;
1436 }
1437
1438 static __init bool allow_1_setting(u32 msr, u32 ctl)
1439 {
1440 u32 vmx_msr_low, vmx_msr_high;
1441
1442 rdmsr(msr, vmx_msr_low, vmx_msr_high);
1443 return vmx_msr_high & ctl;
1444 }
1445
1446 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
1447 {
1448 u32 vmx_msr_low, vmx_msr_high;
1449 u32 min, opt, min2, opt2;
1450 u32 _pin_based_exec_control = 0;
1451 u32 _cpu_based_exec_control = 0;
1452 u32 _cpu_based_2nd_exec_control = 0;
1453 u32 _vmexit_control = 0;
1454 u32 _vmentry_control = 0;
1455
1456 min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
1457 opt = PIN_BASED_VIRTUAL_NMIS;
1458 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
1459 &_pin_based_exec_control) < 0)
1460 return -EIO;
1461
1462 min =
1463 #ifdef CONFIG_X86_64
1464 CPU_BASED_CR8_LOAD_EXITING |
1465 CPU_BASED_CR8_STORE_EXITING |
1466 #endif
1467 CPU_BASED_CR3_LOAD_EXITING |
1468 CPU_BASED_CR3_STORE_EXITING |
1469 CPU_BASED_USE_IO_BITMAPS |
1470 CPU_BASED_MOV_DR_EXITING |
1471 CPU_BASED_USE_TSC_OFFSETING |
1472 CPU_BASED_MWAIT_EXITING |
1473 CPU_BASED_MONITOR_EXITING |
1474 CPU_BASED_INVLPG_EXITING;
1475
1476 if (yield_on_hlt)
1477 min |= CPU_BASED_HLT_EXITING;
1478
1479 opt = CPU_BASED_TPR_SHADOW |
1480 CPU_BASED_USE_MSR_BITMAPS |
1481 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
1482 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
1483 &_cpu_based_exec_control) < 0)
1484 return -EIO;
1485 #ifdef CONFIG_X86_64
1486 if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
1487 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
1488 ~CPU_BASED_CR8_STORE_EXITING;
1489 #endif
1490 if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
1491 min2 = 0;
1492 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
1493 SECONDARY_EXEC_WBINVD_EXITING |
1494 SECONDARY_EXEC_ENABLE_VPID |
1495 SECONDARY_EXEC_ENABLE_EPT |
1496 SECONDARY_EXEC_UNRESTRICTED_GUEST |
1497 SECONDARY_EXEC_PAUSE_LOOP_EXITING |
1498 SECONDARY_EXEC_RDTSCP;
1499 if (adjust_vmx_controls(min2, opt2,
1500 MSR_IA32_VMX_PROCBASED_CTLS2,
1501 &_cpu_based_2nd_exec_control) < 0)
1502 return -EIO;
1503 }
1504 #ifndef CONFIG_X86_64
1505 if (!(_cpu_based_2nd_exec_control &
1506 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
1507 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
1508 #endif
1509 if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
1510 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
1511 enabled */
1512 _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
1513 CPU_BASED_CR3_STORE_EXITING |
1514 CPU_BASED_INVLPG_EXITING);
1515 rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
1516 vmx_capability.ept, vmx_capability.vpid);
1517 }
1518
1519 min = 0;
1520 #ifdef CONFIG_X86_64
1521 min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
1522 #endif
1523 opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT;
1524 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
1525 &_vmexit_control) < 0)
1526 return -EIO;
1527
1528 min = 0;
1529 opt = VM_ENTRY_LOAD_IA32_PAT;
1530 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
1531 &_vmentry_control) < 0)
1532 return -EIO;
1533
1534 rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
1535
1536 /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
1537 if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
1538 return -EIO;
1539
1540 #ifdef CONFIG_X86_64
1541 /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
1542 if (vmx_msr_high & (1u<<16))
1543 return -EIO;
1544 #endif
1545
1546 /* Require Write-Back (WB) memory type for VMCS accesses. */
1547 if (((vmx_msr_high >> 18) & 15) != 6)
1548 return -EIO;
1549
1550 vmcs_conf->size = vmx_msr_high & 0x1fff;
1551 vmcs_conf->order = get_order(vmcs_config.size);
1552 vmcs_conf->revision_id = vmx_msr_low;
1553
1554 vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
1555 vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
1556 vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
1557 vmcs_conf->vmexit_ctrl = _vmexit_control;
1558 vmcs_conf->vmentry_ctrl = _vmentry_control;
1559
1560 cpu_has_load_ia32_efer =
1561 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
1562 VM_ENTRY_LOAD_IA32_EFER)
1563 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
1564 VM_EXIT_LOAD_IA32_EFER);
1565
1566 return 0;
1567 }
1568
1569 static struct vmcs *alloc_vmcs_cpu(int cpu)
1570 {
1571 int node = cpu_to_node(cpu);
1572 struct page *pages;
1573 struct vmcs *vmcs;
1574
1575 pages = alloc_pages_exact_node(node, GFP_KERNEL, vmcs_config.order);
1576 if (!pages)
1577 return NULL;
1578 vmcs = page_address(pages);
1579 memset(vmcs, 0, vmcs_config.size);
1580 vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
1581 return vmcs;
1582 }
1583
1584 static struct vmcs *alloc_vmcs(void)
1585 {
1586 return alloc_vmcs_cpu(raw_smp_processor_id());
1587 }
1588
1589 static void free_vmcs(struct vmcs *vmcs)
1590 {
1591 free_pages((unsigned long)vmcs, vmcs_config.order);
1592 }
1593
1594 static void free_kvm_area(void)
1595 {
1596 int cpu;
1597
1598 for_each_possible_cpu(cpu) {
1599 free_vmcs(per_cpu(vmxarea, cpu));
1600 per_cpu(vmxarea, cpu) = NULL;
1601 }
1602 }
1603
1604 static __init int alloc_kvm_area(void)
1605 {
1606 int cpu;
1607
1608 for_each_possible_cpu(cpu) {
1609 struct vmcs *vmcs;
1610
1611 vmcs = alloc_vmcs_cpu(cpu);
1612 if (!vmcs) {
1613 free_kvm_area();
1614 return -ENOMEM;
1615 }
1616
1617 per_cpu(vmxarea, cpu) = vmcs;
1618 }
1619 return 0;
1620 }
1621
1622 static __init int hardware_setup(void)
1623 {
1624 if (setup_vmcs_config(&vmcs_config) < 0)
1625 return -EIO;
1626
1627 if (boot_cpu_has(X86_FEATURE_NX))
1628 kvm_enable_efer_bits(EFER_NX);
1629
1630 if (!cpu_has_vmx_vpid())
1631 enable_vpid = 0;
1632
1633 if (!cpu_has_vmx_ept() ||
1634 !cpu_has_vmx_ept_4levels()) {
1635 enable_ept = 0;
1636 enable_unrestricted_guest = 0;
1637 }
1638
1639 if (!cpu_has_vmx_unrestricted_guest())
1640 enable_unrestricted_guest = 0;
1641
1642 if (!cpu_has_vmx_flexpriority())
1643 flexpriority_enabled = 0;
1644
1645 if (!cpu_has_vmx_tpr_shadow())
1646 kvm_x86_ops->update_cr8_intercept = NULL;
1647
1648 if (enable_ept && !cpu_has_vmx_ept_2m_page())
1649 kvm_disable_largepages();
1650
1651 if (!cpu_has_vmx_ple())
1652 ple_gap = 0;
1653
1654 return alloc_kvm_area();
1655 }
1656
1657 static __exit void hardware_unsetup(void)
1658 {
1659 free_kvm_area();
1660 }
1661
1662 static void fix_pmode_dataseg(int seg, struct kvm_save_segment *save)
1663 {
1664 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1665
1666 if (vmcs_readl(sf->base) == save->base && (save->base & AR_S_MASK)) {
1667 vmcs_write16(sf->selector, save->selector);
1668 vmcs_writel(sf->base, save->base);
1669 vmcs_write32(sf->limit, save->limit);
1670 vmcs_write32(sf->ar_bytes, save->ar);
1671 } else {
1672 u32 dpl = (vmcs_read16(sf->selector) & SELECTOR_RPL_MASK)
1673 << AR_DPL_SHIFT;
1674 vmcs_write32(sf->ar_bytes, 0x93 | dpl);
1675 }
1676 }
1677
1678 static void enter_pmode(struct kvm_vcpu *vcpu)
1679 {
1680 unsigned long flags;
1681 struct vcpu_vmx *vmx = to_vmx(vcpu);
1682
1683 vmx->emulation_required = 1;
1684 vmx->rmode.vm86_active = 0;
1685
1686 vmcs_write16(GUEST_TR_SELECTOR, vmx->rmode.tr.selector);
1687 vmcs_writel(GUEST_TR_BASE, vmx->rmode.tr.base);
1688 vmcs_write32(GUEST_TR_LIMIT, vmx->rmode.tr.limit);
1689 vmcs_write32(GUEST_TR_AR_BYTES, vmx->rmode.tr.ar);
1690
1691 flags = vmcs_readl(GUEST_RFLAGS);
1692 flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
1693 flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
1694 vmcs_writel(GUEST_RFLAGS, flags);
1695
1696 vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
1697 (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
1698
1699 update_exception_bitmap(vcpu);
1700
1701 if (emulate_invalid_guest_state)
1702 return;
1703
1704 fix_pmode_dataseg(VCPU_SREG_ES, &vmx->rmode.es);
1705 fix_pmode_dataseg(VCPU_SREG_DS, &vmx->rmode.ds);
1706 fix_pmode_dataseg(VCPU_SREG_GS, &vmx->rmode.gs);
1707 fix_pmode_dataseg(VCPU_SREG_FS, &vmx->rmode.fs);
1708
1709 vmcs_write16(GUEST_SS_SELECTOR, 0);
1710 vmcs_write32(GUEST_SS_AR_BYTES, 0x93);
1711
1712 vmcs_write16(GUEST_CS_SELECTOR,
1713 vmcs_read16(GUEST_CS_SELECTOR) & ~SELECTOR_RPL_MASK);
1714 vmcs_write32(GUEST_CS_AR_BYTES, 0x9b);
1715 }
1716
1717 static gva_t rmode_tss_base(struct kvm *kvm)
1718 {
1719 if (!kvm->arch.tss_addr) {
1720 struct kvm_memslots *slots;
1721 gfn_t base_gfn;
1722
1723 slots = kvm_memslots(kvm);
1724 base_gfn = slots->memslots[0].base_gfn +
1725 kvm->memslots->memslots[0].npages - 3;
1726 return base_gfn << PAGE_SHIFT;
1727 }
1728 return kvm->arch.tss_addr;
1729 }
1730
1731 static void fix_rmode_seg(int seg, struct kvm_save_segment *save)
1732 {
1733 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1734
1735 save->selector = vmcs_read16(sf->selector);
1736 save->base = vmcs_readl(sf->base);
1737 save->limit = vmcs_read32(sf->limit);
1738 save->ar = vmcs_read32(sf->ar_bytes);
1739 vmcs_write16(sf->selector, save->base >> 4);
1740 vmcs_write32(sf->base, save->base & 0xffff0);
1741 vmcs_write32(sf->limit, 0xffff);
1742 vmcs_write32(sf->ar_bytes, 0xf3);
1743 if (save->base & 0xf)
1744 printk_once(KERN_WARNING "kvm: segment base is not paragraph"
1745 " aligned when entering protected mode (seg=%d)",
1746 seg);
1747 }
1748
1749 static void enter_rmode(struct kvm_vcpu *vcpu)
1750 {
1751 unsigned long flags;
1752 struct vcpu_vmx *vmx = to_vmx(vcpu);
1753
1754 if (enable_unrestricted_guest)
1755 return;
1756
1757 vmx->emulation_required = 1;
1758 vmx->rmode.vm86_active = 1;
1759
1760 vmx->rmode.tr.selector = vmcs_read16(GUEST_TR_SELECTOR);
1761 vmx->rmode.tr.base = vmcs_readl(GUEST_TR_BASE);
1762 vmcs_writel(GUEST_TR_BASE, rmode_tss_base(vcpu->kvm));
1763
1764 vmx->rmode.tr.limit = vmcs_read32(GUEST_TR_LIMIT);
1765 vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
1766
1767 vmx->rmode.tr.ar = vmcs_read32(GUEST_TR_AR_BYTES);
1768 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
1769
1770 flags = vmcs_readl(GUEST_RFLAGS);
1771 vmx->rmode.save_rflags = flags;
1772
1773 flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1774
1775 vmcs_writel(GUEST_RFLAGS, flags);
1776 vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
1777 update_exception_bitmap(vcpu);
1778
1779 if (emulate_invalid_guest_state)
1780 goto continue_rmode;
1781
1782 vmcs_write16(GUEST_SS_SELECTOR, vmcs_readl(GUEST_SS_BASE) >> 4);
1783 vmcs_write32(GUEST_SS_LIMIT, 0xffff);
1784 vmcs_write32(GUEST_SS_AR_BYTES, 0xf3);
1785
1786 vmcs_write32(GUEST_CS_AR_BYTES, 0xf3);
1787 vmcs_write32(GUEST_CS_LIMIT, 0xffff);
1788 if (vmcs_readl(GUEST_CS_BASE) == 0xffff0000)
1789 vmcs_writel(GUEST_CS_BASE, 0xf0000);
1790 vmcs_write16(GUEST_CS_SELECTOR, vmcs_readl(GUEST_CS_BASE) >> 4);
1791
1792 fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.es);
1793 fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.ds);
1794 fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.gs);
1795 fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.fs);
1796
1797 continue_rmode:
1798 kvm_mmu_reset_context(vcpu);
1799 init_rmode(vcpu->kvm);
1800 }
1801
1802 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
1803 {
1804 struct vcpu_vmx *vmx = to_vmx(vcpu);
1805 struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
1806
1807 if (!msr)
1808 return;
1809
1810 /*
1811 * Force kernel_gs_base reloading before EFER changes, as control
1812 * of this msr depends on is_long_mode().
1813 */
1814 vmx_load_host_state(to_vmx(vcpu));
1815 vcpu->arch.efer = efer;
1816 if (efer & EFER_LMA) {
1817 vmcs_write32(VM_ENTRY_CONTROLS,
1818 vmcs_read32(VM_ENTRY_CONTROLS) |
1819 VM_ENTRY_IA32E_MODE);
1820 msr->data = efer;
1821 } else {
1822 vmcs_write32(VM_ENTRY_CONTROLS,
1823 vmcs_read32(VM_ENTRY_CONTROLS) &
1824 ~VM_ENTRY_IA32E_MODE);
1825
1826 msr->data = efer & ~EFER_LME;
1827 }
1828 setup_msrs(vmx);
1829 }
1830
1831 #ifdef CONFIG_X86_64
1832
1833 static void enter_lmode(struct kvm_vcpu *vcpu)
1834 {
1835 u32 guest_tr_ar;
1836
1837 guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
1838 if ((guest_tr_ar & AR_TYPE_MASK) != AR_TYPE_BUSY_64_TSS) {
1839 printk(KERN_DEBUG "%s: tss fixup for long mode. \n",
1840 __func__);
1841 vmcs_write32(GUEST_TR_AR_BYTES,
1842 (guest_tr_ar & ~AR_TYPE_MASK)
1843 | AR_TYPE_BUSY_64_TSS);
1844 }
1845 vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
1846 }
1847
1848 static void exit_lmode(struct kvm_vcpu *vcpu)
1849 {
1850 vmcs_write32(VM_ENTRY_CONTROLS,
1851 vmcs_read32(VM_ENTRY_CONTROLS)
1852 & ~VM_ENTRY_IA32E_MODE);
1853 vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
1854 }
1855
1856 #endif
1857
1858 static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
1859 {
1860 vpid_sync_context(to_vmx(vcpu));
1861 if (enable_ept) {
1862 if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
1863 return;
1864 ept_sync_context(construct_eptp(vcpu->arch.mmu.root_hpa));
1865 }
1866 }
1867
1868 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
1869 {
1870 ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
1871
1872 vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
1873 vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
1874 }
1875
1876 static void vmx_decache_cr3(struct kvm_vcpu *vcpu)
1877 {
1878 if (enable_ept && is_paging(vcpu))
1879 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
1880 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
1881 }
1882
1883 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
1884 {
1885 ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
1886
1887 vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
1888 vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
1889 }
1890
1891 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
1892 {
1893 if (!test_bit(VCPU_EXREG_PDPTR,
1894 (unsigned long *)&vcpu->arch.regs_dirty))
1895 return;
1896
1897 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
1898 vmcs_write64(GUEST_PDPTR0, vcpu->arch.mmu.pdptrs[0]);
1899 vmcs_write64(GUEST_PDPTR1, vcpu->arch.mmu.pdptrs[1]);
1900 vmcs_write64(GUEST_PDPTR2, vcpu->arch.mmu.pdptrs[2]);
1901 vmcs_write64(GUEST_PDPTR3, vcpu->arch.mmu.pdptrs[3]);
1902 }
1903 }
1904
1905 static void ept_save_pdptrs(struct kvm_vcpu *vcpu)
1906 {
1907 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
1908 vcpu->arch.mmu.pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
1909 vcpu->arch.mmu.pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
1910 vcpu->arch.mmu.pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
1911 vcpu->arch.mmu.pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
1912 }
1913
1914 __set_bit(VCPU_EXREG_PDPTR,
1915 (unsigned long *)&vcpu->arch.regs_avail);
1916 __set_bit(VCPU_EXREG_PDPTR,
1917 (unsigned long *)&vcpu->arch.regs_dirty);
1918 }
1919
1920 static void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
1921
1922 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
1923 unsigned long cr0,
1924 struct kvm_vcpu *vcpu)
1925 {
1926 vmx_decache_cr3(vcpu);
1927 if (!(cr0 & X86_CR0_PG)) {
1928 /* From paging/starting to nonpaging */
1929 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
1930 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
1931 (CPU_BASED_CR3_LOAD_EXITING |
1932 CPU_BASED_CR3_STORE_EXITING));
1933 vcpu->arch.cr0 = cr0;
1934 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
1935 } else if (!is_paging(vcpu)) {
1936 /* From nonpaging to paging */
1937 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
1938 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
1939 ~(CPU_BASED_CR3_LOAD_EXITING |
1940 CPU_BASED_CR3_STORE_EXITING));
1941 vcpu->arch.cr0 = cr0;
1942 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
1943 }
1944
1945 if (!(cr0 & X86_CR0_WP))
1946 *hw_cr0 &= ~X86_CR0_WP;
1947 }
1948
1949 static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
1950 {
1951 struct vcpu_vmx *vmx = to_vmx(vcpu);
1952 unsigned long hw_cr0;
1953
1954 if (enable_unrestricted_guest)
1955 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST)
1956 | KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
1957 else
1958 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK) | KVM_VM_CR0_ALWAYS_ON;
1959
1960 if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
1961 enter_pmode(vcpu);
1962
1963 if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
1964 enter_rmode(vcpu);
1965
1966 #ifdef CONFIG_X86_64
1967 if (vcpu->arch.efer & EFER_LME) {
1968 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
1969 enter_lmode(vcpu);
1970 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
1971 exit_lmode(vcpu);
1972 }
1973 #endif
1974
1975 if (enable_ept)
1976 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
1977
1978 if (!vcpu->fpu_active)
1979 hw_cr0 |= X86_CR0_TS | X86_CR0_MP;
1980
1981 vmcs_writel(CR0_READ_SHADOW, cr0);
1982 vmcs_writel(GUEST_CR0, hw_cr0);
1983 vcpu->arch.cr0 = cr0;
1984 }
1985
1986 static u64 construct_eptp(unsigned long root_hpa)
1987 {
1988 u64 eptp;
1989
1990 /* TODO write the value reading from MSR */
1991 eptp = VMX_EPT_DEFAULT_MT |
1992 VMX_EPT_DEFAULT_GAW << VMX_EPT_GAW_EPTP_SHIFT;
1993 eptp |= (root_hpa & PAGE_MASK);
1994
1995 return eptp;
1996 }
1997
1998 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1999 {
2000 unsigned long guest_cr3;
2001 u64 eptp;
2002
2003 guest_cr3 = cr3;
2004 if (enable_ept) {
2005 eptp = construct_eptp(cr3);
2006 vmcs_write64(EPT_POINTER, eptp);
2007 guest_cr3 = is_paging(vcpu) ? kvm_read_cr3(vcpu) :
2008 vcpu->kvm->arch.ept_identity_map_addr;
2009 ept_load_pdptrs(vcpu);
2010 }
2011
2012 vmx_flush_tlb(vcpu);
2013 vmcs_writel(GUEST_CR3, guest_cr3);
2014 }
2015
2016 static void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
2017 {
2018 unsigned long hw_cr4 = cr4 | (to_vmx(vcpu)->rmode.vm86_active ?
2019 KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON);
2020
2021 vcpu->arch.cr4 = cr4;
2022 if (enable_ept) {
2023 if (!is_paging(vcpu)) {
2024 hw_cr4 &= ~X86_CR4_PAE;
2025 hw_cr4 |= X86_CR4_PSE;
2026 } else if (!(cr4 & X86_CR4_PAE)) {
2027 hw_cr4 &= ~X86_CR4_PAE;
2028 }
2029 }
2030
2031 vmcs_writel(CR4_READ_SHADOW, cr4);
2032 vmcs_writel(GUEST_CR4, hw_cr4);
2033 }
2034
2035 static void vmx_get_segment(struct kvm_vcpu *vcpu,
2036 struct kvm_segment *var, int seg)
2037 {
2038 struct vcpu_vmx *vmx = to_vmx(vcpu);
2039 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2040 struct kvm_save_segment *save;
2041 u32 ar;
2042
2043 if (vmx->rmode.vm86_active
2044 && (seg == VCPU_SREG_TR || seg == VCPU_SREG_ES
2045 || seg == VCPU_SREG_DS || seg == VCPU_SREG_FS
2046 || seg == VCPU_SREG_GS)
2047 && !emulate_invalid_guest_state) {
2048 switch (seg) {
2049 case VCPU_SREG_TR: save = &vmx->rmode.tr; break;
2050 case VCPU_SREG_ES: save = &vmx->rmode.es; break;
2051 case VCPU_SREG_DS: save = &vmx->rmode.ds; break;
2052 case VCPU_SREG_FS: save = &vmx->rmode.fs; break;
2053 case VCPU_SREG_GS: save = &vmx->rmode.gs; break;
2054 default: BUG();
2055 }
2056 var->selector = save->selector;
2057 var->base = save->base;
2058 var->limit = save->limit;
2059 ar = save->ar;
2060 if (seg == VCPU_SREG_TR
2061 || var->selector == vmcs_read16(sf->selector))
2062 goto use_saved_rmode_seg;
2063 }
2064 var->base = vmcs_readl(sf->base);
2065 var->limit = vmcs_read32(sf->limit);
2066 var->selector = vmcs_read16(sf->selector);
2067 ar = vmcs_read32(sf->ar_bytes);
2068 use_saved_rmode_seg:
2069 if ((ar & AR_UNUSABLE_MASK) && !emulate_invalid_guest_state)
2070 ar = 0;
2071 var->type = ar & 15;
2072 var->s = (ar >> 4) & 1;
2073 var->dpl = (ar >> 5) & 3;
2074 var->present = (ar >> 7) & 1;
2075 var->avl = (ar >> 12) & 1;
2076 var->l = (ar >> 13) & 1;
2077 var->db = (ar >> 14) & 1;
2078 var->g = (ar >> 15) & 1;
2079 var->unusable = (ar >> 16) & 1;
2080 }
2081
2082 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
2083 {
2084 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2085 struct kvm_segment s;
2086
2087 if (to_vmx(vcpu)->rmode.vm86_active) {
2088 vmx_get_segment(vcpu, &s, seg);
2089 return s.base;
2090 }
2091 return vmcs_readl(sf->base);
2092 }
2093
2094 static int vmx_get_cpl(struct kvm_vcpu *vcpu)
2095 {
2096 if (!is_protmode(vcpu))
2097 return 0;
2098
2099 if (vmx_get_rflags(vcpu) & X86_EFLAGS_VM) /* if virtual 8086 */
2100 return 3;
2101
2102 return vmcs_read16(GUEST_CS_SELECTOR) & 3;
2103 }
2104
2105 static u32 vmx_segment_access_rights(struct kvm_segment *var)
2106 {
2107 u32 ar;
2108
2109 if (var->unusable)
2110 ar = 1 << 16;
2111 else {
2112 ar = var->type & 15;
2113 ar |= (var->s & 1) << 4;
2114 ar |= (var->dpl & 3) << 5;
2115 ar |= (var->present & 1) << 7;
2116 ar |= (var->avl & 1) << 12;
2117 ar |= (var->l & 1) << 13;
2118 ar |= (var->db & 1) << 14;
2119 ar |= (var->g & 1) << 15;
2120 }
2121 if (ar == 0) /* a 0 value means unusable */
2122 ar = AR_UNUSABLE_MASK;
2123
2124 return ar;
2125 }
2126
2127 static void vmx_set_segment(struct kvm_vcpu *vcpu,
2128 struct kvm_segment *var, int seg)
2129 {
2130 struct vcpu_vmx *vmx = to_vmx(vcpu);
2131 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2132 u32 ar;
2133
2134 if (vmx->rmode.vm86_active && seg == VCPU_SREG_TR) {
2135 vmx->rmode.tr.selector = var->selector;
2136 vmx->rmode.tr.base = var->base;
2137 vmx->rmode.tr.limit = var->limit;
2138 vmx->rmode.tr.ar = vmx_segment_access_rights(var);
2139 return;
2140 }
2141 vmcs_writel(sf->base, var->base);
2142 vmcs_write32(sf->limit, var->limit);
2143 vmcs_write16(sf->selector, var->selector);
2144 if (vmx->rmode.vm86_active && var->s) {
2145 /*
2146 * Hack real-mode segments into vm86 compatibility.
2147 */
2148 if (var->base == 0xffff0000 && var->selector == 0xf000)
2149 vmcs_writel(sf->base, 0xf0000);
2150 ar = 0xf3;
2151 } else
2152 ar = vmx_segment_access_rights(var);
2153
2154 /*
2155 * Fix the "Accessed" bit in AR field of segment registers for older
2156 * qemu binaries.
2157 * IA32 arch specifies that at the time of processor reset the
2158 * "Accessed" bit in the AR field of segment registers is 1. And qemu
2159 * is setting it to 0 in the usedland code. This causes invalid guest
2160 * state vmexit when "unrestricted guest" mode is turned on.
2161 * Fix for this setup issue in cpu_reset is being pushed in the qemu
2162 * tree. Newer qemu binaries with that qemu fix would not need this
2163 * kvm hack.
2164 */
2165 if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
2166 ar |= 0x1; /* Accessed */
2167
2168 vmcs_write32(sf->ar_bytes, ar);
2169 }
2170
2171 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
2172 {
2173 u32 ar = vmcs_read32(GUEST_CS_AR_BYTES);
2174
2175 *db = (ar >> 14) & 1;
2176 *l = (ar >> 13) & 1;
2177 }
2178
2179 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
2180 {
2181 dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
2182 dt->address = vmcs_readl(GUEST_IDTR_BASE);
2183 }
2184
2185 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
2186 {
2187 vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
2188 vmcs_writel(GUEST_IDTR_BASE, dt->address);
2189 }
2190
2191 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
2192 {
2193 dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
2194 dt->address = vmcs_readl(GUEST_GDTR_BASE);
2195 }
2196
2197 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
2198 {
2199 vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
2200 vmcs_writel(GUEST_GDTR_BASE, dt->address);
2201 }
2202
2203 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
2204 {
2205 struct kvm_segment var;
2206 u32 ar;
2207
2208 vmx_get_segment(vcpu, &var, seg);
2209 ar = vmx_segment_access_rights(&var);
2210
2211 if (var.base != (var.selector << 4))
2212 return false;
2213 if (var.limit != 0xffff)
2214 return false;
2215 if (ar != 0xf3)
2216 return false;
2217
2218 return true;
2219 }
2220
2221 static bool code_segment_valid(struct kvm_vcpu *vcpu)
2222 {
2223 struct kvm_segment cs;
2224 unsigned int cs_rpl;
2225
2226 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
2227 cs_rpl = cs.selector & SELECTOR_RPL_MASK;
2228
2229 if (cs.unusable)
2230 return false;
2231 if (~cs.type & (AR_TYPE_CODE_MASK|AR_TYPE_ACCESSES_MASK))
2232 return false;
2233 if (!cs.s)
2234 return false;
2235 if (cs.type & AR_TYPE_WRITEABLE_MASK) {
2236 if (cs.dpl > cs_rpl)
2237 return false;
2238 } else {
2239 if (cs.dpl != cs_rpl)
2240 return false;
2241 }
2242 if (!cs.present)
2243 return false;
2244
2245 /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
2246 return true;
2247 }
2248
2249 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
2250 {
2251 struct kvm_segment ss;
2252 unsigned int ss_rpl;
2253
2254 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
2255 ss_rpl = ss.selector & SELECTOR_RPL_MASK;
2256
2257 if (ss.unusable)
2258 return true;
2259 if (ss.type != 3 && ss.type != 7)
2260 return false;
2261 if (!ss.s)
2262 return false;
2263 if (ss.dpl != ss_rpl) /* DPL != RPL */
2264 return false;
2265 if (!ss.present)
2266 return false;
2267
2268 return true;
2269 }
2270
2271 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
2272 {
2273 struct kvm_segment var;
2274 unsigned int rpl;
2275
2276 vmx_get_segment(vcpu, &var, seg);
2277 rpl = var.selector & SELECTOR_RPL_MASK;
2278
2279 if (var.unusable)
2280 return true;
2281 if (!var.s)
2282 return false;
2283 if (!var.present)
2284 return false;
2285 if (~var.type & (AR_TYPE_CODE_MASK|AR_TYPE_WRITEABLE_MASK)) {
2286 if (var.dpl < rpl) /* DPL < RPL */
2287 return false;
2288 }
2289
2290 /* TODO: Add other members to kvm_segment_field to allow checking for other access
2291 * rights flags
2292 */
2293 return true;
2294 }
2295
2296 static bool tr_valid(struct kvm_vcpu *vcpu)
2297 {
2298 struct kvm_segment tr;
2299
2300 vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
2301
2302 if (tr.unusable)
2303 return false;
2304 if (tr.selector & SELECTOR_TI_MASK) /* TI = 1 */
2305 return false;
2306 if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
2307 return false;
2308 if (!tr.present)
2309 return false;
2310
2311 return true;
2312 }
2313
2314 static bool ldtr_valid(struct kvm_vcpu *vcpu)
2315 {
2316 struct kvm_segment ldtr;
2317
2318 vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
2319
2320 if (ldtr.unusable)
2321 return true;
2322 if (ldtr.selector & SELECTOR_TI_MASK) /* TI = 1 */
2323 return false;
2324 if (ldtr.type != 2)
2325 return false;
2326 if (!ldtr.present)
2327 return false;
2328
2329 return true;
2330 }
2331
2332 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
2333 {
2334 struct kvm_segment cs, ss;
2335
2336 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
2337 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
2338
2339 return ((cs.selector & SELECTOR_RPL_MASK) ==
2340 (ss.selector & SELECTOR_RPL_MASK));
2341 }
2342
2343 /*
2344 * Check if guest state is valid. Returns true if valid, false if
2345 * not.
2346 * We assume that registers are always usable
2347 */
2348 static bool guest_state_valid(struct kvm_vcpu *vcpu)
2349 {
2350 /* real mode guest state checks */
2351 if (!is_protmode(vcpu)) {
2352 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
2353 return false;
2354 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
2355 return false;
2356 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
2357 return false;
2358 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
2359 return false;
2360 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
2361 return false;
2362 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
2363 return false;
2364 } else {
2365 /* protected mode guest state checks */
2366 if (!cs_ss_rpl_check(vcpu))
2367 return false;
2368 if (!code_segment_valid(vcpu))
2369 return false;
2370 if (!stack_segment_valid(vcpu))
2371 return false;
2372 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
2373 return false;
2374 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
2375 return false;
2376 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
2377 return false;
2378 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
2379 return false;
2380 if (!tr_valid(vcpu))
2381 return false;
2382 if (!ldtr_valid(vcpu))
2383 return false;
2384 }
2385 /* TODO:
2386 * - Add checks on RIP
2387 * - Add checks on RFLAGS
2388 */
2389
2390 return true;
2391 }
2392
2393 static int init_rmode_tss(struct kvm *kvm)
2394 {
2395 gfn_t fn = rmode_tss_base(kvm) >> PAGE_SHIFT;
2396 u16 data = 0;
2397 int ret = 0;
2398 int r;
2399
2400 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
2401 if (r < 0)
2402 goto out;
2403 data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
2404 r = kvm_write_guest_page(kvm, fn++, &data,
2405 TSS_IOPB_BASE_OFFSET, sizeof(u16));
2406 if (r < 0)
2407 goto out;
2408 r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
2409 if (r < 0)
2410 goto out;
2411 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
2412 if (r < 0)
2413 goto out;
2414 data = ~0;
2415 r = kvm_write_guest_page(kvm, fn, &data,
2416 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
2417 sizeof(u8));
2418 if (r < 0)
2419 goto out;
2420
2421 ret = 1;
2422 out:
2423 return ret;
2424 }
2425
2426 static int init_rmode_identity_map(struct kvm *kvm)
2427 {
2428 int i, r, ret;
2429 pfn_t identity_map_pfn;
2430 u32 tmp;
2431
2432 if (!enable_ept)
2433 return 1;
2434 if (unlikely(!kvm->arch.ept_identity_pagetable)) {
2435 printk(KERN_ERR "EPT: identity-mapping pagetable "
2436 "haven't been allocated!\n");
2437 return 0;
2438 }
2439 if (likely(kvm->arch.ept_identity_pagetable_done))
2440 return 1;
2441 ret = 0;
2442 identity_map_pfn = kvm->arch.ept_identity_map_addr >> PAGE_SHIFT;
2443 r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
2444 if (r < 0)
2445 goto out;
2446 /* Set up identity-mapping pagetable for EPT in real mode */
2447 for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
2448 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
2449 _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
2450 r = kvm_write_guest_page(kvm, identity_map_pfn,
2451 &tmp, i * sizeof(tmp), sizeof(tmp));
2452 if (r < 0)
2453 goto out;
2454 }
2455 kvm->arch.ept_identity_pagetable_done = true;
2456 ret = 1;
2457 out:
2458 return ret;
2459 }
2460
2461 static void seg_setup(int seg)
2462 {
2463 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2464 unsigned int ar;
2465
2466 vmcs_write16(sf->selector, 0);
2467 vmcs_writel(sf->base, 0);
2468 vmcs_write32(sf->limit, 0xffff);
2469 if (enable_unrestricted_guest) {
2470 ar = 0x93;
2471 if (seg == VCPU_SREG_CS)
2472 ar |= 0x08; /* code segment */
2473 } else
2474 ar = 0xf3;
2475
2476 vmcs_write32(sf->ar_bytes, ar);
2477 }
2478
2479 static int alloc_apic_access_page(struct kvm *kvm)
2480 {
2481 struct kvm_userspace_memory_region kvm_userspace_mem;
2482 int r = 0;
2483
2484 mutex_lock(&kvm->slots_lock);
2485 if (kvm->arch.apic_access_page)
2486 goto out;
2487 kvm_userspace_mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
2488 kvm_userspace_mem.flags = 0;
2489 kvm_userspace_mem.guest_phys_addr = 0xfee00000ULL;
2490 kvm_userspace_mem.memory_size = PAGE_SIZE;
2491 r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
2492 if (r)
2493 goto out;
2494
2495 kvm->arch.apic_access_page = gfn_to_page(kvm, 0xfee00);
2496 out:
2497 mutex_unlock(&kvm->slots_lock);
2498 return r;
2499 }
2500
2501 static int alloc_identity_pagetable(struct kvm *kvm)
2502 {
2503 struct kvm_userspace_memory_region kvm_userspace_mem;
2504 int r = 0;
2505
2506 mutex_lock(&kvm->slots_lock);
2507 if (kvm->arch.ept_identity_pagetable)
2508 goto out;
2509 kvm_userspace_mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
2510 kvm_userspace_mem.flags = 0;
2511 kvm_userspace_mem.guest_phys_addr =
2512 kvm->arch.ept_identity_map_addr;
2513 kvm_userspace_mem.memory_size = PAGE_SIZE;
2514 r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
2515 if (r)
2516 goto out;
2517
2518 kvm->arch.ept_identity_pagetable = gfn_to_page(kvm,
2519 kvm->arch.ept_identity_map_addr >> PAGE_SHIFT);
2520 out:
2521 mutex_unlock(&kvm->slots_lock);
2522 return r;
2523 }
2524
2525 static void allocate_vpid(struct vcpu_vmx *vmx)
2526 {
2527 int vpid;
2528
2529 vmx->vpid = 0;
2530 if (!enable_vpid)
2531 return;
2532 spin_lock(&vmx_vpid_lock);
2533 vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
2534 if (vpid < VMX_NR_VPIDS) {
2535 vmx->vpid = vpid;
2536 __set_bit(vpid, vmx_vpid_bitmap);
2537 }
2538 spin_unlock(&vmx_vpid_lock);
2539 }
2540
2541 static void free_vpid(struct vcpu_vmx *vmx)
2542 {
2543 if (!enable_vpid)
2544 return;
2545 spin_lock(&vmx_vpid_lock);
2546 if (vmx->vpid != 0)
2547 __clear_bit(vmx->vpid, vmx_vpid_bitmap);
2548 spin_unlock(&vmx_vpid_lock);
2549 }
2550
2551 static void __vmx_disable_intercept_for_msr(unsigned long *msr_bitmap, u32 msr)
2552 {
2553 int f = sizeof(unsigned long);
2554
2555 if (!cpu_has_vmx_msr_bitmap())
2556 return;
2557
2558 /*
2559 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
2560 * have the write-low and read-high bitmap offsets the wrong way round.
2561 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
2562 */
2563 if (msr <= 0x1fff) {
2564 __clear_bit(msr, msr_bitmap + 0x000 / f); /* read-low */
2565 __clear_bit(msr, msr_bitmap + 0x800 / f); /* write-low */
2566 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
2567 msr &= 0x1fff;
2568 __clear_bit(msr, msr_bitmap + 0x400 / f); /* read-high */
2569 __clear_bit(msr, msr_bitmap + 0xc00 / f); /* write-high */
2570 }
2571 }
2572
2573 static void vmx_disable_intercept_for_msr(u32 msr, bool longmode_only)
2574 {
2575 if (!longmode_only)
2576 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy, msr);
2577 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode, msr);
2578 }
2579
2580 /*
2581 * Sets up the vmcs for emulated real mode.
2582 */
2583 static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
2584 {
2585 u32 host_sysenter_cs, msr_low, msr_high;
2586 u32 junk;
2587 u64 host_pat;
2588 unsigned long a;
2589 struct desc_ptr dt;
2590 int i;
2591 unsigned long kvm_vmx_return;
2592 u32 exec_control;
2593
2594 /* I/O */
2595 vmcs_write64(IO_BITMAP_A, __pa(vmx_io_bitmap_a));
2596 vmcs_write64(IO_BITMAP_B, __pa(vmx_io_bitmap_b));
2597
2598 if (cpu_has_vmx_msr_bitmap())
2599 vmcs_write64(MSR_BITMAP, __pa(vmx_msr_bitmap_legacy));
2600
2601 vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
2602
2603 /* Control */
2604 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL,
2605 vmcs_config.pin_based_exec_ctrl);
2606
2607 exec_control = vmcs_config.cpu_based_exec_ctrl;
2608 if (!vm_need_tpr_shadow(vmx->vcpu.kvm)) {
2609 exec_control &= ~CPU_BASED_TPR_SHADOW;
2610 #ifdef CONFIG_X86_64
2611 exec_control |= CPU_BASED_CR8_STORE_EXITING |
2612 CPU_BASED_CR8_LOAD_EXITING;
2613 #endif
2614 }
2615 if (!enable_ept)
2616 exec_control |= CPU_BASED_CR3_STORE_EXITING |
2617 CPU_BASED_CR3_LOAD_EXITING |
2618 CPU_BASED_INVLPG_EXITING;
2619 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
2620
2621 if (cpu_has_secondary_exec_ctrls()) {
2622 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
2623 if (!vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
2624 exec_control &=
2625 ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
2626 if (vmx->vpid == 0)
2627 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
2628 if (!enable_ept) {
2629 exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
2630 enable_unrestricted_guest = 0;
2631 }
2632 if (!enable_unrestricted_guest)
2633 exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
2634 if (!ple_gap)
2635 exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
2636 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
2637 }
2638
2639 if (ple_gap) {
2640 vmcs_write32(PLE_GAP, ple_gap);
2641 vmcs_write32(PLE_WINDOW, ple_window);
2642 }
2643
2644 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, !!bypass_guest_pf);
2645 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, !!bypass_guest_pf);
2646 vmcs_write32(CR3_TARGET_COUNT, 0); /* 22.2.1 */
2647
2648 vmcs_writel(HOST_CR0, read_cr0() | X86_CR0_TS); /* 22.2.3 */
2649 vmcs_writel(HOST_CR4, read_cr4()); /* 22.2.3, 22.2.5 */
2650 vmcs_writel(HOST_CR3, read_cr3()); /* 22.2.3 FIXME: shadow tables */
2651
2652 vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */
2653 vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
2654 vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS); /* 22.2.4 */
2655 vmcs_write16(HOST_FS_SELECTOR, 0); /* 22.2.4 */
2656 vmcs_write16(HOST_GS_SELECTOR, 0); /* 22.2.4 */
2657 vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
2658 #ifdef CONFIG_X86_64
2659 rdmsrl(MSR_FS_BASE, a);
2660 vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
2661 rdmsrl(MSR_GS_BASE, a);
2662 vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
2663 #else
2664 vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
2665 vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
2666 #endif
2667
2668 vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8); /* 22.2.4 */
2669
2670 native_store_idt(&dt);
2671 vmcs_writel(HOST_IDTR_BASE, dt.address); /* 22.2.4 */
2672
2673 asm("mov $.Lkvm_vmx_return, %0" : "=r"(kvm_vmx_return));
2674 vmcs_writel(HOST_RIP, kvm_vmx_return); /* 22.2.5 */
2675 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
2676 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
2677 vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host));
2678 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
2679 vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest));
2680
2681 rdmsr(MSR_IA32_SYSENTER_CS, host_sysenter_cs, junk);
2682 vmcs_write32(HOST_IA32_SYSENTER_CS, host_sysenter_cs);
2683 rdmsrl(MSR_IA32_SYSENTER_ESP, a);
2684 vmcs_writel(HOST_IA32_SYSENTER_ESP, a); /* 22.2.3 */
2685 rdmsrl(MSR_IA32_SYSENTER_EIP, a);
2686 vmcs_writel(HOST_IA32_SYSENTER_EIP, a); /* 22.2.3 */
2687
2688 if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
2689 rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high);
2690 host_pat = msr_low | ((u64) msr_high << 32);
2691 vmcs_write64(HOST_IA32_PAT, host_pat);
2692 }
2693 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2694 rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high);
2695 host_pat = msr_low | ((u64) msr_high << 32);
2696 /* Write the default value follow host pat */
2697 vmcs_write64(GUEST_IA32_PAT, host_pat);
2698 /* Keep arch.pat sync with GUEST_IA32_PAT */
2699 vmx->vcpu.arch.pat = host_pat;
2700 }
2701
2702 for (i = 0; i < NR_VMX_MSR; ++i) {
2703 u32 index = vmx_msr_index[i];
2704 u32 data_low, data_high;
2705 int j = vmx->nmsrs;
2706
2707 if (rdmsr_safe(index, &data_low, &data_high) < 0)
2708 continue;
2709 if (wrmsr_safe(index, data_low, data_high) < 0)
2710 continue;
2711 vmx->guest_msrs[j].index = i;
2712 vmx->guest_msrs[j].data = 0;
2713 vmx->guest_msrs[j].mask = -1ull;
2714 ++vmx->nmsrs;
2715 }
2716
2717 vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
2718
2719 /* 22.2.1, 20.8.1 */
2720 vmcs_write32(VM_ENTRY_CONTROLS, vmcs_config.vmentry_ctrl);
2721
2722 vmcs_writel(CR0_GUEST_HOST_MASK, ~0UL);
2723 vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
2724 if (enable_ept)
2725 vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
2726 vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
2727
2728 kvm_write_tsc(&vmx->vcpu, 0);
2729
2730 return 0;
2731 }
2732
2733 static int init_rmode(struct kvm *kvm)
2734 {
2735 int idx, ret = 0;
2736
2737 idx = srcu_read_lock(&kvm->srcu);
2738 if (!init_rmode_tss(kvm))
2739 goto exit;
2740 if (!init_rmode_identity_map(kvm))
2741 goto exit;
2742
2743 ret = 1;
2744 exit:
2745 srcu_read_unlock(&kvm->srcu, idx);
2746 return ret;
2747 }
2748
2749 static int vmx_vcpu_reset(struct kvm_vcpu *vcpu)
2750 {
2751 struct vcpu_vmx *vmx = to_vmx(vcpu);
2752 u64 msr;
2753 int ret;
2754
2755 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP));
2756 if (!init_rmode(vmx->vcpu.kvm)) {
2757 ret = -ENOMEM;
2758 goto out;
2759 }
2760
2761 vmx->rmode.vm86_active = 0;
2762
2763 vmx->soft_vnmi_blocked = 0;
2764
2765 vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
2766 kvm_set_cr8(&vmx->vcpu, 0);
2767 msr = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
2768 if (kvm_vcpu_is_bsp(&vmx->vcpu))
2769 msr |= MSR_IA32_APICBASE_BSP;
2770 kvm_set_apic_base(&vmx->vcpu, msr);
2771
2772 ret = fx_init(&vmx->vcpu);
2773 if (ret != 0)
2774 goto out;
2775
2776 seg_setup(VCPU_SREG_CS);
2777 /*
2778 * GUEST_CS_BASE should really be 0xffff0000, but VT vm86 mode
2779 * insists on having GUEST_CS_BASE == GUEST_CS_SELECTOR << 4. Sigh.
2780 */
2781 if (kvm_vcpu_is_bsp(&vmx->vcpu)) {
2782 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
2783 vmcs_writel(GUEST_CS_BASE, 0x000f0000);
2784 } else {
2785 vmcs_write16(GUEST_CS_SELECTOR, vmx->vcpu.arch.sipi_vector << 8);
2786 vmcs_writel(GUEST_CS_BASE, vmx->vcpu.arch.sipi_vector << 12);
2787 }
2788
2789 seg_setup(VCPU_SREG_DS);
2790 seg_setup(VCPU_SREG_ES);
2791 seg_setup(VCPU_SREG_FS);
2792 seg_setup(VCPU_SREG_GS);
2793 seg_setup(VCPU_SREG_SS);
2794
2795 vmcs_write16(GUEST_TR_SELECTOR, 0);
2796 vmcs_writel(GUEST_TR_BASE, 0);
2797 vmcs_write32(GUEST_TR_LIMIT, 0xffff);
2798 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
2799
2800 vmcs_write16(GUEST_LDTR_SELECTOR, 0);
2801 vmcs_writel(GUEST_LDTR_BASE, 0);
2802 vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
2803 vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
2804
2805 vmcs_write32(GUEST_SYSENTER_CS, 0);
2806 vmcs_writel(GUEST_SYSENTER_ESP, 0);
2807 vmcs_writel(GUEST_SYSENTER_EIP, 0);
2808
2809 vmcs_writel(GUEST_RFLAGS, 0x02);
2810 if (kvm_vcpu_is_bsp(&vmx->vcpu))
2811 kvm_rip_write(vcpu, 0xfff0);
2812 else
2813 kvm_rip_write(vcpu, 0);
2814 kvm_register_write(vcpu, VCPU_REGS_RSP, 0);
2815
2816 vmcs_writel(GUEST_DR7, 0x400);
2817
2818 vmcs_writel(GUEST_GDTR_BASE, 0);
2819 vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
2820
2821 vmcs_writel(GUEST_IDTR_BASE, 0);
2822 vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
2823
2824 vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
2825 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
2826 vmcs_write32(GUEST_PENDING_DBG_EXCEPTIONS, 0);
2827
2828 /* Special registers */
2829 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
2830
2831 setup_msrs(vmx);
2832
2833 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */
2834
2835 if (cpu_has_vmx_tpr_shadow()) {
2836 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
2837 if (vm_need_tpr_shadow(vmx->vcpu.kvm))
2838 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
2839 page_to_phys(vmx->vcpu.arch.apic->regs_page));
2840 vmcs_write32(TPR_THRESHOLD, 0);
2841 }
2842
2843 if (vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
2844 vmcs_write64(APIC_ACCESS_ADDR,
2845 page_to_phys(vmx->vcpu.kvm->arch.apic_access_page));
2846
2847 if (vmx->vpid != 0)
2848 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
2849
2850 vmx->vcpu.arch.cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
2851 vmx_set_cr0(&vmx->vcpu, kvm_read_cr0(vcpu)); /* enter rmode */
2852 vmx_set_cr4(&vmx->vcpu, 0);
2853 vmx_set_efer(&vmx->vcpu, 0);
2854 vmx_fpu_activate(&vmx->vcpu);
2855 update_exception_bitmap(&vmx->vcpu);
2856
2857 vpid_sync_context(vmx);
2858
2859 ret = 0;
2860
2861 /* HACK: Don't enable emulation on guest boot/reset */
2862 vmx->emulation_required = 0;
2863
2864 out:
2865 return ret;
2866 }
2867
2868 static void enable_irq_window(struct kvm_vcpu *vcpu)
2869 {
2870 u32 cpu_based_vm_exec_control;
2871
2872 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2873 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
2874 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2875 }
2876
2877 static void enable_nmi_window(struct kvm_vcpu *vcpu)
2878 {
2879 u32 cpu_based_vm_exec_control;
2880
2881 if (!cpu_has_virtual_nmis()) {
2882 enable_irq_window(vcpu);
2883 return;
2884 }
2885
2886 if (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
2887 enable_irq_window(vcpu);
2888 return;
2889 }
2890 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2891 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_NMI_PENDING;
2892 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2893 }
2894
2895 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
2896 {
2897 struct vcpu_vmx *vmx = to_vmx(vcpu);
2898 uint32_t intr;
2899 int irq = vcpu->arch.interrupt.nr;
2900
2901 trace_kvm_inj_virq(irq);
2902
2903 ++vcpu->stat.irq_injections;
2904 if (vmx->rmode.vm86_active) {
2905 if (kvm_inject_realmode_interrupt(vcpu, irq) != EMULATE_DONE)
2906 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2907 return;
2908 }
2909 intr = irq | INTR_INFO_VALID_MASK;
2910 if (vcpu->arch.interrupt.soft) {
2911 intr |= INTR_TYPE_SOFT_INTR;
2912 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2913 vmx->vcpu.arch.event_exit_inst_len);
2914 } else
2915 intr |= INTR_TYPE_EXT_INTR;
2916 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
2917 vmx_clear_hlt(vcpu);
2918 }
2919
2920 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
2921 {
2922 struct vcpu_vmx *vmx = to_vmx(vcpu);
2923
2924 if (!cpu_has_virtual_nmis()) {
2925 /*
2926 * Tracking the NMI-blocked state in software is built upon
2927 * finding the next open IRQ window. This, in turn, depends on
2928 * well-behaving guests: They have to keep IRQs disabled at
2929 * least as long as the NMI handler runs. Otherwise we may
2930 * cause NMI nesting, maybe breaking the guest. But as this is
2931 * highly unlikely, we can live with the residual risk.
2932 */
2933 vmx->soft_vnmi_blocked = 1;
2934 vmx->vnmi_blocked_time = 0;
2935 }
2936
2937 ++vcpu->stat.nmi_injections;
2938 if (vmx->rmode.vm86_active) {
2939 if (kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR) != EMULATE_DONE)
2940 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2941 return;
2942 }
2943 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2944 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
2945 vmx_clear_hlt(vcpu);
2946 }
2947
2948 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
2949 {
2950 if (!cpu_has_virtual_nmis() && to_vmx(vcpu)->soft_vnmi_blocked)
2951 return 0;
2952
2953 return !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
2954 (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
2955 | GUEST_INTR_STATE_NMI));
2956 }
2957
2958 static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
2959 {
2960 if (!cpu_has_virtual_nmis())
2961 return to_vmx(vcpu)->soft_vnmi_blocked;
2962 return vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
2963 }
2964
2965 static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
2966 {
2967 struct vcpu_vmx *vmx = to_vmx(vcpu);
2968
2969 if (!cpu_has_virtual_nmis()) {
2970 if (vmx->soft_vnmi_blocked != masked) {
2971 vmx->soft_vnmi_blocked = masked;
2972 vmx->vnmi_blocked_time = 0;
2973 }
2974 } else {
2975 if (masked)
2976 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
2977 GUEST_INTR_STATE_NMI);
2978 else
2979 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
2980 GUEST_INTR_STATE_NMI);
2981 }
2982 }
2983
2984 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
2985 {
2986 return (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
2987 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
2988 (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
2989 }
2990
2991 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
2992 {
2993 int ret;
2994 struct kvm_userspace_memory_region tss_mem = {
2995 .slot = TSS_PRIVATE_MEMSLOT,
2996 .guest_phys_addr = addr,
2997 .memory_size = PAGE_SIZE * 3,
2998 .flags = 0,
2999 };
3000
3001 ret = kvm_set_memory_region(kvm, &tss_mem, 0);
3002 if (ret)
3003 return ret;
3004 kvm->arch.tss_addr = addr;
3005 return 0;
3006 }
3007
3008 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
3009 int vec, u32 err_code)
3010 {
3011 /*
3012 * Instruction with address size override prefix opcode 0x67
3013 * Cause the #SS fault with 0 error code in VM86 mode.
3014 */
3015 if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0)
3016 if (emulate_instruction(vcpu, 0) == EMULATE_DONE)
3017 return 1;
3018 /*
3019 * Forward all other exceptions that are valid in real mode.
3020 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
3021 * the required debugging infrastructure rework.
3022 */
3023 switch (vec) {
3024 case DB_VECTOR:
3025 if (vcpu->guest_debug &
3026 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
3027 return 0;
3028 kvm_queue_exception(vcpu, vec);
3029 return 1;
3030 case BP_VECTOR:
3031 /*
3032 * Update instruction length as we may reinject the exception
3033 * from user space while in guest debugging mode.
3034 */
3035 to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
3036 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
3037 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
3038 return 0;
3039 /* fall through */
3040 case DE_VECTOR:
3041 case OF_VECTOR:
3042 case BR_VECTOR:
3043 case UD_VECTOR:
3044 case DF_VECTOR:
3045 case SS_VECTOR:
3046 case GP_VECTOR:
3047 case MF_VECTOR:
3048 kvm_queue_exception(vcpu, vec);
3049 return 1;
3050 }
3051 return 0;
3052 }
3053
3054 /*
3055 * Trigger machine check on the host. We assume all the MSRs are already set up
3056 * by the CPU and that we still run on the same CPU as the MCE occurred on.
3057 * We pass a fake environment to the machine check handler because we want
3058 * the guest to be always treated like user space, no matter what context
3059 * it used internally.
3060 */
3061 static void kvm_machine_check(void)
3062 {
3063 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
3064 struct pt_regs regs = {
3065 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
3066 .flags = X86_EFLAGS_IF,
3067 };
3068
3069 do_machine_check(&regs, 0);
3070 #endif
3071 }
3072
3073 static int handle_machine_check(struct kvm_vcpu *vcpu)
3074 {
3075 /* already handled by vcpu_run */
3076 return 1;
3077 }
3078
3079 static int handle_exception(struct kvm_vcpu *vcpu)
3080 {
3081 struct vcpu_vmx *vmx = to_vmx(vcpu);
3082 struct kvm_run *kvm_run = vcpu->run;
3083 u32 intr_info, ex_no, error_code;
3084 unsigned long cr2, rip, dr6;
3085 u32 vect_info;
3086 enum emulation_result er;
3087
3088 vect_info = vmx->idt_vectoring_info;
3089 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
3090
3091 if (is_machine_check(intr_info))
3092 return handle_machine_check(vcpu);
3093
3094 if ((vect_info & VECTORING_INFO_VALID_MASK) &&
3095 !is_page_fault(intr_info)) {
3096 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3097 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
3098 vcpu->run->internal.ndata = 2;
3099 vcpu->run->internal.data[0] = vect_info;
3100 vcpu->run->internal.data[1] = intr_info;
3101 return 0;
3102 }
3103
3104 if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR)
3105 return 1; /* already handled by vmx_vcpu_run() */
3106
3107 if (is_no_device(intr_info)) {
3108 vmx_fpu_activate(vcpu);
3109 return 1;
3110 }
3111
3112 if (is_invalid_opcode(intr_info)) {
3113 er = emulate_instruction(vcpu, EMULTYPE_TRAP_UD);
3114 if (er != EMULATE_DONE)
3115 kvm_queue_exception(vcpu, UD_VECTOR);
3116 return 1;
3117 }
3118
3119 error_code = 0;
3120 rip = kvm_rip_read(vcpu);
3121 if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
3122 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
3123 if (is_page_fault(intr_info)) {
3124 /* EPT won't cause page fault directly */
3125 if (enable_ept)
3126 BUG();
3127 cr2 = vmcs_readl(EXIT_QUALIFICATION);
3128 trace_kvm_page_fault(cr2, error_code);
3129
3130 if (kvm_event_needs_reinjection(vcpu))
3131 kvm_mmu_unprotect_page_virt(vcpu, cr2);
3132 return kvm_mmu_page_fault(vcpu, cr2, error_code, NULL, 0);
3133 }
3134
3135 if (vmx->rmode.vm86_active &&
3136 handle_rmode_exception(vcpu, intr_info & INTR_INFO_VECTOR_MASK,
3137 error_code)) {
3138 if (vcpu->arch.halt_request) {
3139 vcpu->arch.halt_request = 0;
3140 return kvm_emulate_halt(vcpu);
3141 }
3142 return 1;
3143 }
3144
3145 ex_no = intr_info & INTR_INFO_VECTOR_MASK;
3146 switch (ex_no) {
3147 case DB_VECTOR:
3148 dr6 = vmcs_readl(EXIT_QUALIFICATION);
3149 if (!(vcpu->guest_debug &
3150 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
3151 vcpu->arch.dr6 = dr6 | DR6_FIXED_1;
3152 kvm_queue_exception(vcpu, DB_VECTOR);
3153 return 1;
3154 }
3155 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
3156 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
3157 /* fall through */
3158 case BP_VECTOR:
3159 /*
3160 * Update instruction length as we may reinject #BP from
3161 * user space while in guest debugging mode. Reading it for
3162 * #DB as well causes no harm, it is not used in that case.
3163 */
3164 vmx->vcpu.arch.event_exit_inst_len =
3165 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
3166 kvm_run->exit_reason = KVM_EXIT_DEBUG;
3167 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
3168 kvm_run->debug.arch.exception = ex_no;
3169 break;
3170 default:
3171 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
3172 kvm_run->ex.exception = ex_no;
3173 kvm_run->ex.error_code = error_code;
3174 break;
3175 }
3176 return 0;
3177 }
3178
3179 static int handle_external_interrupt(struct kvm_vcpu *vcpu)
3180 {
3181 ++vcpu->stat.irq_exits;
3182 return 1;
3183 }
3184
3185 static int handle_triple_fault(struct kvm_vcpu *vcpu)
3186 {
3187 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
3188 return 0;
3189 }
3190
3191 static int handle_io(struct kvm_vcpu *vcpu)
3192 {
3193 unsigned long exit_qualification;
3194 int size, in, string;
3195 unsigned port;
3196
3197 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3198 string = (exit_qualification & 16) != 0;
3199 in = (exit_qualification & 8) != 0;
3200
3201 ++vcpu->stat.io_exits;
3202
3203 if (string || in)
3204 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
3205
3206 port = exit_qualification >> 16;
3207 size = (exit_qualification & 7) + 1;
3208 skip_emulated_instruction(vcpu);
3209
3210 return kvm_fast_pio_out(vcpu, size, port);
3211 }
3212
3213 static void
3214 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
3215 {
3216 /*
3217 * Patch in the VMCALL instruction:
3218 */
3219 hypercall[0] = 0x0f;
3220 hypercall[1] = 0x01;
3221 hypercall[2] = 0xc1;
3222 }
3223
3224 static int handle_cr(struct kvm_vcpu *vcpu)
3225 {
3226 unsigned long exit_qualification, val;
3227 int cr;
3228 int reg;
3229 int err;
3230
3231 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3232 cr = exit_qualification & 15;
3233 reg = (exit_qualification >> 8) & 15;
3234 switch ((exit_qualification >> 4) & 3) {
3235 case 0: /* mov to cr */
3236 val = kvm_register_read(vcpu, reg);
3237 trace_kvm_cr_write(cr, val);
3238 switch (cr) {
3239 case 0:
3240 err = kvm_set_cr0(vcpu, val);
3241 kvm_complete_insn_gp(vcpu, err);
3242 return 1;
3243 case 3:
3244 err = kvm_set_cr3(vcpu, val);
3245 kvm_complete_insn_gp(vcpu, err);
3246 return 1;
3247 case 4:
3248 err = kvm_set_cr4(vcpu, val);
3249 kvm_complete_insn_gp(vcpu, err);
3250 return 1;
3251 case 8: {
3252 u8 cr8_prev = kvm_get_cr8(vcpu);
3253 u8 cr8 = kvm_register_read(vcpu, reg);
3254 err = kvm_set_cr8(vcpu, cr8);
3255 kvm_complete_insn_gp(vcpu, err);
3256 if (irqchip_in_kernel(vcpu->kvm))
3257 return 1;
3258 if (cr8_prev <= cr8)
3259 return 1;
3260 vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
3261 return 0;
3262 }
3263 };
3264 break;
3265 case 2: /* clts */
3266 vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
3267 trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
3268 skip_emulated_instruction(vcpu);
3269 vmx_fpu_activate(vcpu);
3270 return 1;
3271 case 1: /*mov from cr*/
3272 switch (cr) {
3273 case 3:
3274 val = kvm_read_cr3(vcpu);
3275 kvm_register_write(vcpu, reg, val);
3276 trace_kvm_cr_read(cr, val);
3277 skip_emulated_instruction(vcpu);
3278 return 1;
3279 case 8:
3280 val = kvm_get_cr8(vcpu);
3281 kvm_register_write(vcpu, reg, val);
3282 trace_kvm_cr_read(cr, val);
3283 skip_emulated_instruction(vcpu);
3284 return 1;
3285 }
3286 break;
3287 case 3: /* lmsw */
3288 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
3289 trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
3290 kvm_lmsw(vcpu, val);
3291
3292 skip_emulated_instruction(vcpu);
3293 return 1;
3294 default:
3295 break;
3296 }
3297 vcpu->run->exit_reason = 0;
3298 pr_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
3299 (int)(exit_qualification >> 4) & 3, cr);
3300 return 0;
3301 }
3302
3303 static int handle_dr(struct kvm_vcpu *vcpu)
3304 {
3305 unsigned long exit_qualification;
3306 int dr, reg;
3307
3308 /* Do not handle if the CPL > 0, will trigger GP on re-entry */
3309 if (!kvm_require_cpl(vcpu, 0))
3310 return 1;
3311 dr = vmcs_readl(GUEST_DR7);
3312 if (dr & DR7_GD) {
3313 /*
3314 * As the vm-exit takes precedence over the debug trap, we
3315 * need to emulate the latter, either for the host or the
3316 * guest debugging itself.
3317 */
3318 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
3319 vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
3320 vcpu->run->debug.arch.dr7 = dr;
3321 vcpu->run->debug.arch.pc =
3322 vmcs_readl(GUEST_CS_BASE) +
3323 vmcs_readl(GUEST_RIP);
3324 vcpu->run->debug.arch.exception = DB_VECTOR;
3325 vcpu->run->exit_reason = KVM_EXIT_DEBUG;
3326 return 0;
3327 } else {
3328 vcpu->arch.dr7 &= ~DR7_GD;
3329 vcpu->arch.dr6 |= DR6_BD;
3330 vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
3331 kvm_queue_exception(vcpu, DB_VECTOR);
3332 return 1;
3333 }
3334 }
3335
3336 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3337 dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
3338 reg = DEBUG_REG_ACCESS_REG(exit_qualification);
3339 if (exit_qualification & TYPE_MOV_FROM_DR) {
3340 unsigned long val;
3341 if (!kvm_get_dr(vcpu, dr, &val))
3342 kvm_register_write(vcpu, reg, val);
3343 } else
3344 kvm_set_dr(vcpu, dr, vcpu->arch.regs[reg]);
3345 skip_emulated_instruction(vcpu);
3346 return 1;
3347 }
3348
3349 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
3350 {
3351 vmcs_writel(GUEST_DR7, val);
3352 }
3353
3354 static int handle_cpuid(struct kvm_vcpu *vcpu)
3355 {
3356 kvm_emulate_cpuid(vcpu);
3357 return 1;
3358 }
3359
3360 static int handle_rdmsr(struct kvm_vcpu *vcpu)
3361 {
3362 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
3363 u64 data;
3364
3365 if (vmx_get_msr(vcpu, ecx, &data)) {
3366 trace_kvm_msr_read_ex(ecx);
3367 kvm_inject_gp(vcpu, 0);
3368 return 1;
3369 }
3370
3371 trace_kvm_msr_read(ecx, data);
3372
3373 /* FIXME: handling of bits 32:63 of rax, rdx */
3374 vcpu->arch.regs[VCPU_REGS_RAX] = data & -1u;
3375 vcpu->arch.regs[VCPU_REGS_RDX] = (data >> 32) & -1u;
3376 skip_emulated_instruction(vcpu);
3377 return 1;
3378 }
3379
3380 static int handle_wrmsr(struct kvm_vcpu *vcpu)
3381 {
3382 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
3383 u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
3384 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
3385
3386 if (vmx_set_msr(vcpu, ecx, data) != 0) {
3387 trace_kvm_msr_write_ex(ecx, data);
3388 kvm_inject_gp(vcpu, 0);
3389 return 1;
3390 }
3391
3392 trace_kvm_msr_write(ecx, data);
3393 skip_emulated_instruction(vcpu);
3394 return 1;
3395 }
3396
3397 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
3398 {
3399 kvm_make_request(KVM_REQ_EVENT, vcpu);
3400 return 1;
3401 }
3402
3403 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
3404 {
3405 u32 cpu_based_vm_exec_control;
3406
3407 /* clear pending irq */
3408 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
3409 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
3410 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
3411
3412 kvm_make_request(KVM_REQ_EVENT, vcpu);
3413
3414 ++vcpu->stat.irq_window_exits;
3415
3416 /*
3417 * If the user space waits to inject interrupts, exit as soon as
3418 * possible
3419 */
3420 if (!irqchip_in_kernel(vcpu->kvm) &&
3421 vcpu->run->request_interrupt_window &&
3422 !kvm_cpu_has_interrupt(vcpu)) {
3423 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
3424 return 0;
3425 }
3426 return 1;
3427 }
3428
3429 static int handle_halt(struct kvm_vcpu *vcpu)
3430 {
3431 skip_emulated_instruction(vcpu);
3432 return kvm_emulate_halt(vcpu);
3433 }
3434
3435 static int handle_vmcall(struct kvm_vcpu *vcpu)
3436 {
3437 skip_emulated_instruction(vcpu);
3438 kvm_emulate_hypercall(vcpu);
3439 return 1;
3440 }
3441
3442 static int handle_vmx_insn(struct kvm_vcpu *vcpu)
3443 {
3444 kvm_queue_exception(vcpu, UD_VECTOR);
3445 return 1;
3446 }
3447
3448 static int handle_invd(struct kvm_vcpu *vcpu)
3449 {
3450 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
3451 }
3452
3453 static int handle_invlpg(struct kvm_vcpu *vcpu)
3454 {
3455 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3456
3457 kvm_mmu_invlpg(vcpu, exit_qualification);
3458 skip_emulated_instruction(vcpu);
3459 return 1;
3460 }
3461
3462 static int handle_wbinvd(struct kvm_vcpu *vcpu)
3463 {
3464 skip_emulated_instruction(vcpu);
3465 kvm_emulate_wbinvd(vcpu);
3466 return 1;
3467 }
3468
3469 static int handle_xsetbv(struct kvm_vcpu *vcpu)
3470 {
3471 u64 new_bv = kvm_read_edx_eax(vcpu);
3472 u32 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
3473
3474 if (kvm_set_xcr(vcpu, index, new_bv) == 0)
3475 skip_emulated_instruction(vcpu);
3476 return 1;
3477 }
3478
3479 static int handle_apic_access(struct kvm_vcpu *vcpu)
3480 {
3481 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
3482 }
3483
3484 static int handle_task_switch(struct kvm_vcpu *vcpu)
3485 {
3486 struct vcpu_vmx *vmx = to_vmx(vcpu);
3487 unsigned long exit_qualification;
3488 bool has_error_code = false;
3489 u32 error_code = 0;
3490 u16 tss_selector;
3491 int reason, type, idt_v;
3492
3493 idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
3494 type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
3495
3496 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3497
3498 reason = (u32)exit_qualification >> 30;
3499 if (reason == TASK_SWITCH_GATE && idt_v) {
3500 switch (type) {
3501 case INTR_TYPE_NMI_INTR:
3502 vcpu->arch.nmi_injected = false;
3503 if (cpu_has_virtual_nmis())
3504 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
3505 GUEST_INTR_STATE_NMI);
3506 break;
3507 case INTR_TYPE_EXT_INTR:
3508 case INTR_TYPE_SOFT_INTR:
3509 kvm_clear_interrupt_queue(vcpu);
3510 break;
3511 case INTR_TYPE_HARD_EXCEPTION:
3512 if (vmx->idt_vectoring_info &
3513 VECTORING_INFO_DELIVER_CODE_MASK) {
3514 has_error_code = true;
3515 error_code =
3516 vmcs_read32(IDT_VECTORING_ERROR_CODE);
3517 }
3518 /* fall through */
3519 case INTR_TYPE_SOFT_EXCEPTION:
3520 kvm_clear_exception_queue(vcpu);
3521 break;
3522 default:
3523 break;
3524 }
3525 }
3526 tss_selector = exit_qualification;
3527
3528 if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
3529 type != INTR_TYPE_EXT_INTR &&
3530 type != INTR_TYPE_NMI_INTR))
3531 skip_emulated_instruction(vcpu);
3532
3533 if (kvm_task_switch(vcpu, tss_selector, reason,
3534 has_error_code, error_code) == EMULATE_FAIL) {
3535 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3536 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
3537 vcpu->run->internal.ndata = 0;
3538 return 0;
3539 }
3540
3541 /* clear all local breakpoint enable flags */
3542 vmcs_writel(GUEST_DR7, vmcs_readl(GUEST_DR7) & ~55);
3543
3544 /*
3545 * TODO: What about debug traps on tss switch?
3546 * Are we supposed to inject them and update dr6?
3547 */
3548
3549 return 1;
3550 }
3551
3552 static int handle_ept_violation(struct kvm_vcpu *vcpu)
3553 {
3554 unsigned long exit_qualification;
3555 gpa_t gpa;
3556 int gla_validity;
3557
3558 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3559
3560 if (exit_qualification & (1 << 6)) {
3561 printk(KERN_ERR "EPT: GPA exceeds GAW!\n");
3562 return -EINVAL;
3563 }
3564
3565 gla_validity = (exit_qualification >> 7) & 0x3;
3566 if (gla_validity != 0x3 && gla_validity != 0x1 && gla_validity != 0) {
3567 printk(KERN_ERR "EPT: Handling EPT violation failed!\n");
3568 printk(KERN_ERR "EPT: GPA: 0x%lx, GVA: 0x%lx\n",
3569 (long unsigned int)vmcs_read64(GUEST_PHYSICAL_ADDRESS),
3570 vmcs_readl(GUEST_LINEAR_ADDRESS));
3571 printk(KERN_ERR "EPT: Exit qualification is 0x%lx\n",
3572 (long unsigned int)exit_qualification);
3573 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
3574 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_VIOLATION;
3575 return 0;
3576 }
3577
3578 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
3579 trace_kvm_page_fault(gpa, exit_qualification);
3580 return kvm_mmu_page_fault(vcpu, gpa, exit_qualification & 0x3, NULL, 0);
3581 }
3582
3583 static u64 ept_rsvd_mask(u64 spte, int level)
3584 {
3585 int i;
3586 u64 mask = 0;
3587
3588 for (i = 51; i > boot_cpu_data.x86_phys_bits; i--)
3589 mask |= (1ULL << i);
3590
3591 if (level > 2)
3592 /* bits 7:3 reserved */
3593 mask |= 0xf8;
3594 else if (level == 2) {
3595 if (spte & (1ULL << 7))
3596 /* 2MB ref, bits 20:12 reserved */
3597 mask |= 0x1ff000;
3598 else
3599 /* bits 6:3 reserved */
3600 mask |= 0x78;
3601 }
3602
3603 return mask;
3604 }
3605
3606 static void ept_misconfig_inspect_spte(struct kvm_vcpu *vcpu, u64 spte,
3607 int level)
3608 {
3609 printk(KERN_ERR "%s: spte 0x%llx level %d\n", __func__, spte, level);
3610
3611 /* 010b (write-only) */
3612 WARN_ON((spte & 0x7) == 0x2);
3613
3614 /* 110b (write/execute) */
3615 WARN_ON((spte & 0x7) == 0x6);
3616
3617 /* 100b (execute-only) and value not supported by logical processor */
3618 if (!cpu_has_vmx_ept_execute_only())
3619 WARN_ON((spte & 0x7) == 0x4);
3620
3621 /* not 000b */
3622 if ((spte & 0x7)) {
3623 u64 rsvd_bits = spte & ept_rsvd_mask(spte, level);
3624
3625 if (rsvd_bits != 0) {
3626 printk(KERN_ERR "%s: rsvd_bits = 0x%llx\n",
3627 __func__, rsvd_bits);
3628 WARN_ON(1);
3629 }
3630
3631 if (level == 1 || (level == 2 && (spte & (1ULL << 7)))) {
3632 u64 ept_mem_type = (spte & 0x38) >> 3;
3633
3634 if (ept_mem_type == 2 || ept_mem_type == 3 ||
3635 ept_mem_type == 7) {
3636 printk(KERN_ERR "%s: ept_mem_type=0x%llx\n",
3637 __func__, ept_mem_type);
3638 WARN_ON(1);
3639 }
3640 }
3641 }
3642 }
3643
3644 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
3645 {
3646 u64 sptes[4];
3647 int nr_sptes, i;
3648 gpa_t gpa;
3649
3650 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
3651
3652 printk(KERN_ERR "EPT: Misconfiguration.\n");
3653 printk(KERN_ERR "EPT: GPA: 0x%llx\n", gpa);
3654
3655 nr_sptes = kvm_mmu_get_spte_hierarchy(vcpu, gpa, sptes);
3656
3657 for (i = PT64_ROOT_LEVEL; i > PT64_ROOT_LEVEL - nr_sptes; --i)
3658 ept_misconfig_inspect_spte(vcpu, sptes[i-1], i);
3659
3660 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
3661 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_MISCONFIG;
3662
3663 return 0;
3664 }
3665
3666 static int handle_nmi_window(struct kvm_vcpu *vcpu)
3667 {
3668 u32 cpu_based_vm_exec_control;
3669
3670 /* clear pending NMI */
3671 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
3672 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
3673 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
3674 ++vcpu->stat.nmi_window_exits;
3675 kvm_make_request(KVM_REQ_EVENT, vcpu);
3676
3677 return 1;
3678 }
3679
3680 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
3681 {
3682 struct vcpu_vmx *vmx = to_vmx(vcpu);
3683 enum emulation_result err = EMULATE_DONE;
3684 int ret = 1;
3685 u32 cpu_exec_ctrl;
3686 bool intr_window_requested;
3687
3688 cpu_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
3689 intr_window_requested = cpu_exec_ctrl & CPU_BASED_VIRTUAL_INTR_PENDING;
3690
3691 while (!guest_state_valid(vcpu)) {
3692 if (intr_window_requested
3693 && (kvm_get_rflags(&vmx->vcpu) & X86_EFLAGS_IF))
3694 return handle_interrupt_window(&vmx->vcpu);
3695
3696 err = emulate_instruction(vcpu, 0);
3697
3698 if (err == EMULATE_DO_MMIO) {
3699 ret = 0;
3700 goto out;
3701 }
3702
3703 if (err != EMULATE_DONE)
3704 return 0;
3705
3706 if (signal_pending(current))
3707 goto out;
3708 if (need_resched())
3709 schedule();
3710 }
3711
3712 vmx->emulation_required = 0;
3713 out:
3714 return ret;
3715 }
3716
3717 /*
3718 * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
3719 * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
3720 */
3721 static int handle_pause(struct kvm_vcpu *vcpu)
3722 {
3723 skip_emulated_instruction(vcpu);
3724 kvm_vcpu_on_spin(vcpu);
3725
3726 return 1;
3727 }
3728
3729 static int handle_invalid_op(struct kvm_vcpu *vcpu)
3730 {
3731 kvm_queue_exception(vcpu, UD_VECTOR);
3732 return 1;
3733 }
3734
3735 /*
3736 * The exit handlers return 1 if the exit was handled fully and guest execution
3737 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
3738 * to be done to userspace and return 0.
3739 */
3740 static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
3741 [EXIT_REASON_EXCEPTION_NMI] = handle_exception,
3742 [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt,
3743 [EXIT_REASON_TRIPLE_FAULT] = handle_triple_fault,
3744 [EXIT_REASON_NMI_WINDOW] = handle_nmi_window,
3745 [EXIT_REASON_IO_INSTRUCTION] = handle_io,
3746 [EXIT_REASON_CR_ACCESS] = handle_cr,
3747 [EXIT_REASON_DR_ACCESS] = handle_dr,
3748 [EXIT_REASON_CPUID] = handle_cpuid,
3749 [EXIT_REASON_MSR_READ] = handle_rdmsr,
3750 [EXIT_REASON_MSR_WRITE] = handle_wrmsr,
3751 [EXIT_REASON_PENDING_INTERRUPT] = handle_interrupt_window,
3752 [EXIT_REASON_HLT] = handle_halt,
3753 [EXIT_REASON_INVD] = handle_invd,
3754 [EXIT_REASON_INVLPG] = handle_invlpg,
3755 [EXIT_REASON_VMCALL] = handle_vmcall,
3756 [EXIT_REASON_VMCLEAR] = handle_vmx_insn,
3757 [EXIT_REASON_VMLAUNCH] = handle_vmx_insn,
3758 [EXIT_REASON_VMPTRLD] = handle_vmx_insn,
3759 [EXIT_REASON_VMPTRST] = handle_vmx_insn,
3760 [EXIT_REASON_VMREAD] = handle_vmx_insn,
3761 [EXIT_REASON_VMRESUME] = handle_vmx_insn,
3762 [EXIT_REASON_VMWRITE] = handle_vmx_insn,
3763 [EXIT_REASON_VMOFF] = handle_vmx_insn,
3764 [EXIT_REASON_VMON] = handle_vmx_insn,
3765 [EXIT_REASON_TPR_BELOW_THRESHOLD] = handle_tpr_below_threshold,
3766 [EXIT_REASON_APIC_ACCESS] = handle_apic_access,
3767 [EXIT_REASON_WBINVD] = handle_wbinvd,
3768 [EXIT_REASON_XSETBV] = handle_xsetbv,
3769 [EXIT_REASON_TASK_SWITCH] = handle_task_switch,
3770 [EXIT_REASON_MCE_DURING_VMENTRY] = handle_machine_check,
3771 [EXIT_REASON_EPT_VIOLATION] = handle_ept_violation,
3772 [EXIT_REASON_EPT_MISCONFIG] = handle_ept_misconfig,
3773 [EXIT_REASON_PAUSE_INSTRUCTION] = handle_pause,
3774 [EXIT_REASON_MWAIT_INSTRUCTION] = handle_invalid_op,
3775 [EXIT_REASON_MONITOR_INSTRUCTION] = handle_invalid_op,
3776 };
3777
3778 static const int kvm_vmx_max_exit_handlers =
3779 ARRAY_SIZE(kvm_vmx_exit_handlers);
3780
3781 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
3782 {
3783 *info1 = vmcs_readl(EXIT_QUALIFICATION);
3784 *info2 = vmcs_read32(VM_EXIT_INTR_INFO);
3785 }
3786
3787 /*
3788 * The guest has exited. See if we can fix it or if we need userspace
3789 * assistance.
3790 */
3791 static int vmx_handle_exit(struct kvm_vcpu *vcpu)
3792 {
3793 struct vcpu_vmx *vmx = to_vmx(vcpu);
3794 u32 exit_reason = vmx->exit_reason;
3795 u32 vectoring_info = vmx->idt_vectoring_info;
3796
3797 trace_kvm_exit(exit_reason, vcpu, KVM_ISA_VMX);
3798
3799 /* If guest state is invalid, start emulating */
3800 if (vmx->emulation_required && emulate_invalid_guest_state)
3801 return handle_invalid_guest_state(vcpu);
3802
3803 if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
3804 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3805 vcpu->run->fail_entry.hardware_entry_failure_reason
3806 = exit_reason;
3807 return 0;
3808 }
3809
3810 if (unlikely(vmx->fail)) {
3811 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3812 vcpu->run->fail_entry.hardware_entry_failure_reason
3813 = vmcs_read32(VM_INSTRUCTION_ERROR);
3814 return 0;
3815 }
3816
3817 if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
3818 (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
3819 exit_reason != EXIT_REASON_EPT_VIOLATION &&
3820 exit_reason != EXIT_REASON_TASK_SWITCH))
3821 printk(KERN_WARNING "%s: unexpected, valid vectoring info "
3822 "(0x%x) and exit reason is 0x%x\n",
3823 __func__, vectoring_info, exit_reason);
3824
3825 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked)) {
3826 if (vmx_interrupt_allowed(vcpu)) {
3827 vmx->soft_vnmi_blocked = 0;
3828 } else if (vmx->vnmi_blocked_time > 1000000000LL &&
3829 vcpu->arch.nmi_pending) {
3830 /*
3831 * This CPU don't support us in finding the end of an
3832 * NMI-blocked window if the guest runs with IRQs
3833 * disabled. So we pull the trigger after 1 s of
3834 * futile waiting, but inform the user about this.
3835 */
3836 printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
3837 "state on VCPU %d after 1 s timeout\n",
3838 __func__, vcpu->vcpu_id);
3839 vmx->soft_vnmi_blocked = 0;
3840 }
3841 }
3842
3843 if (exit_reason < kvm_vmx_max_exit_handlers
3844 && kvm_vmx_exit_handlers[exit_reason])
3845 return kvm_vmx_exit_handlers[exit_reason](vcpu);
3846 else {
3847 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
3848 vcpu->run->hw.hardware_exit_reason = exit_reason;
3849 }
3850 return 0;
3851 }
3852
3853 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
3854 {
3855 if (irr == -1 || tpr < irr) {
3856 vmcs_write32(TPR_THRESHOLD, 0);
3857 return;
3858 }
3859
3860 vmcs_write32(TPR_THRESHOLD, irr);
3861 }
3862
3863 static void vmx_complete_atomic_exit(struct vcpu_vmx *vmx)
3864 {
3865 u32 exit_intr_info = vmx->exit_intr_info;
3866
3867 /* Handle machine checks before interrupts are enabled */
3868 if ((vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY)
3869 || (vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI
3870 && is_machine_check(exit_intr_info)))
3871 kvm_machine_check();
3872
3873 /* We need to handle NMIs before interrupts are enabled */
3874 if ((exit_intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR &&
3875 (exit_intr_info & INTR_INFO_VALID_MASK)) {
3876 kvm_before_handle_nmi(&vmx->vcpu);
3877 asm("int $2");
3878 kvm_after_handle_nmi(&vmx->vcpu);
3879 }
3880 }
3881
3882 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
3883 {
3884 u32 exit_intr_info = vmx->exit_intr_info;
3885 bool unblock_nmi;
3886 u8 vector;
3887 bool idtv_info_valid;
3888
3889 idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
3890
3891 if (cpu_has_virtual_nmis()) {
3892 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
3893 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
3894 /*
3895 * SDM 3: 27.7.1.2 (September 2008)
3896 * Re-set bit "block by NMI" before VM entry if vmexit caused by
3897 * a guest IRET fault.
3898 * SDM 3: 23.2.2 (September 2008)
3899 * Bit 12 is undefined in any of the following cases:
3900 * If the VM exit sets the valid bit in the IDT-vectoring
3901 * information field.
3902 * If the VM exit is due to a double fault.
3903 */
3904 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
3905 vector != DF_VECTOR && !idtv_info_valid)
3906 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
3907 GUEST_INTR_STATE_NMI);
3908 } else if (unlikely(vmx->soft_vnmi_blocked))
3909 vmx->vnmi_blocked_time +=
3910 ktime_to_ns(ktime_sub(ktime_get(), vmx->entry_time));
3911 }
3912
3913 static void __vmx_complete_interrupts(struct vcpu_vmx *vmx,
3914 u32 idt_vectoring_info,
3915 int instr_len_field,
3916 int error_code_field)
3917 {
3918 u8 vector;
3919 int type;
3920 bool idtv_info_valid;
3921
3922 idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
3923
3924 vmx->vcpu.arch.nmi_injected = false;
3925 kvm_clear_exception_queue(&vmx->vcpu);
3926 kvm_clear_interrupt_queue(&vmx->vcpu);
3927
3928 if (!idtv_info_valid)
3929 return;
3930
3931 kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
3932
3933 vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
3934 type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
3935
3936 switch (type) {
3937 case INTR_TYPE_NMI_INTR:
3938 vmx->vcpu.arch.nmi_injected = true;
3939 /*
3940 * SDM 3: 27.7.1.2 (September 2008)
3941 * Clear bit "block by NMI" before VM entry if a NMI
3942 * delivery faulted.
3943 */
3944 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
3945 GUEST_INTR_STATE_NMI);
3946 break;
3947 case INTR_TYPE_SOFT_EXCEPTION:
3948 vmx->vcpu.arch.event_exit_inst_len =
3949 vmcs_read32(instr_len_field);
3950 /* fall through */
3951 case INTR_TYPE_HARD_EXCEPTION:
3952 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
3953 u32 err = vmcs_read32(error_code_field);
3954 kvm_queue_exception_e(&vmx->vcpu, vector, err);
3955 } else
3956 kvm_queue_exception(&vmx->vcpu, vector);
3957 break;
3958 case INTR_TYPE_SOFT_INTR:
3959 vmx->vcpu.arch.event_exit_inst_len =
3960 vmcs_read32(instr_len_field);
3961 /* fall through */
3962 case INTR_TYPE_EXT_INTR:
3963 kvm_queue_interrupt(&vmx->vcpu, vector,
3964 type == INTR_TYPE_SOFT_INTR);
3965 break;
3966 default:
3967 break;
3968 }
3969 }
3970
3971 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
3972 {
3973 __vmx_complete_interrupts(vmx, vmx->idt_vectoring_info,
3974 VM_EXIT_INSTRUCTION_LEN,
3975 IDT_VECTORING_ERROR_CODE);
3976 }
3977
3978 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
3979 {
3980 __vmx_complete_interrupts(to_vmx(vcpu),
3981 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
3982 VM_ENTRY_INSTRUCTION_LEN,
3983 VM_ENTRY_EXCEPTION_ERROR_CODE);
3984
3985 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
3986 }
3987
3988 #ifdef CONFIG_X86_64
3989 #define R "r"
3990 #define Q "q"
3991 #else
3992 #define R "e"
3993 #define Q "l"
3994 #endif
3995
3996 static void vmx_vcpu_run(struct kvm_vcpu *vcpu)
3997 {
3998 struct vcpu_vmx *vmx = to_vmx(vcpu);
3999
4000 /* Record the guest's net vcpu time for enforced NMI injections. */
4001 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked))
4002 vmx->entry_time = ktime_get();
4003
4004 /* Don't enter VMX if guest state is invalid, let the exit handler
4005 start emulation until we arrive back to a valid state */
4006 if (vmx->emulation_required && emulate_invalid_guest_state)
4007 return;
4008
4009 if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
4010 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
4011 if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
4012 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
4013
4014 /* When single-stepping over STI and MOV SS, we must clear the
4015 * corresponding interruptibility bits in the guest state. Otherwise
4016 * vmentry fails as it then expects bit 14 (BS) in pending debug
4017 * exceptions being set, but that's not correct for the guest debugging
4018 * case. */
4019 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
4020 vmx_set_interrupt_shadow(vcpu, 0);
4021
4022 asm(
4023 /* Store host registers */
4024 "push %%"R"dx; push %%"R"bp;"
4025 "push %%"R"cx \n\t"
4026 "cmp %%"R"sp, %c[host_rsp](%0) \n\t"
4027 "je 1f \n\t"
4028 "mov %%"R"sp, %c[host_rsp](%0) \n\t"
4029 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
4030 "1: \n\t"
4031 /* Reload cr2 if changed */
4032 "mov %c[cr2](%0), %%"R"ax \n\t"
4033 "mov %%cr2, %%"R"dx \n\t"
4034 "cmp %%"R"ax, %%"R"dx \n\t"
4035 "je 2f \n\t"
4036 "mov %%"R"ax, %%cr2 \n\t"
4037 "2: \n\t"
4038 /* Check if vmlaunch of vmresume is needed */
4039 "cmpl $0, %c[launched](%0) \n\t"
4040 /* Load guest registers. Don't clobber flags. */
4041 "mov %c[rax](%0), %%"R"ax \n\t"
4042 "mov %c[rbx](%0), %%"R"bx \n\t"
4043 "mov %c[rdx](%0), %%"R"dx \n\t"
4044 "mov %c[rsi](%0), %%"R"si \n\t"
4045 "mov %c[rdi](%0), %%"R"di \n\t"
4046 "mov %c[rbp](%0), %%"R"bp \n\t"
4047 #ifdef CONFIG_X86_64
4048 "mov %c[r8](%0), %%r8 \n\t"
4049 "mov %c[r9](%0), %%r9 \n\t"
4050 "mov %c[r10](%0), %%r10 \n\t"
4051 "mov %c[r11](%0), %%r11 \n\t"
4052 "mov %c[r12](%0), %%r12 \n\t"
4053 "mov %c[r13](%0), %%r13 \n\t"
4054 "mov %c[r14](%0), %%r14 \n\t"
4055 "mov %c[r15](%0), %%r15 \n\t"
4056 #endif
4057 "mov %c[rcx](%0), %%"R"cx \n\t" /* kills %0 (ecx) */
4058
4059 /* Enter guest mode */
4060 "jne .Llaunched \n\t"
4061 __ex(ASM_VMX_VMLAUNCH) "\n\t"
4062 "jmp .Lkvm_vmx_return \n\t"
4063 ".Llaunched: " __ex(ASM_VMX_VMRESUME) "\n\t"
4064 ".Lkvm_vmx_return: "
4065 /* Save guest registers, load host registers, keep flags */
4066 "xchg %0, (%%"R"sp) \n\t"
4067 "mov %%"R"ax, %c[rax](%0) \n\t"
4068 "mov %%"R"bx, %c[rbx](%0) \n\t"
4069 "pop"Q" %c[rcx](%0) \n\t"
4070 "mov %%"R"dx, %c[rdx](%0) \n\t"
4071 "mov %%"R"si, %c[rsi](%0) \n\t"
4072 "mov %%"R"di, %c[rdi](%0) \n\t"
4073 "mov %%"R"bp, %c[rbp](%0) \n\t"
4074 #ifdef CONFIG_X86_64
4075 "mov %%r8, %c[r8](%0) \n\t"
4076 "mov %%r9, %c[r9](%0) \n\t"
4077 "mov %%r10, %c[r10](%0) \n\t"
4078 "mov %%r11, %c[r11](%0) \n\t"
4079 "mov %%r12, %c[r12](%0) \n\t"
4080 "mov %%r13, %c[r13](%0) \n\t"
4081 "mov %%r14, %c[r14](%0) \n\t"
4082 "mov %%r15, %c[r15](%0) \n\t"
4083 #endif
4084 "mov %%cr2, %%"R"ax \n\t"
4085 "mov %%"R"ax, %c[cr2](%0) \n\t"
4086
4087 "pop %%"R"bp; pop %%"R"dx \n\t"
4088 "setbe %c[fail](%0) \n\t"
4089 : : "c"(vmx), "d"((unsigned long)HOST_RSP),
4090 [launched]"i"(offsetof(struct vcpu_vmx, launched)),
4091 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
4092 [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
4093 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
4094 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
4095 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
4096 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
4097 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
4098 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
4099 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
4100 #ifdef CONFIG_X86_64
4101 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
4102 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
4103 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
4104 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
4105 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
4106 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
4107 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
4108 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
4109 #endif
4110 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2))
4111 : "cc", "memory"
4112 , R"ax", R"bx", R"di", R"si"
4113 #ifdef CONFIG_X86_64
4114 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
4115 #endif
4116 );
4117
4118 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
4119 | (1 << VCPU_EXREG_PDPTR)
4120 | (1 << VCPU_EXREG_CR3));
4121 vcpu->arch.regs_dirty = 0;
4122
4123 vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
4124
4125 asm("mov %0, %%ds; mov %0, %%es" : : "r"(__USER_DS));
4126 vmx->launched = 1;
4127
4128 vmx->exit_reason = vmcs_read32(VM_EXIT_REASON);
4129 vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
4130
4131 vmx_complete_atomic_exit(vmx);
4132 vmx_recover_nmi_blocking(vmx);
4133 vmx_complete_interrupts(vmx);
4134 }
4135
4136 #undef R
4137 #undef Q
4138
4139 static void vmx_free_vmcs(struct kvm_vcpu *vcpu)
4140 {
4141 struct vcpu_vmx *vmx = to_vmx(vcpu);
4142
4143 if (vmx->vmcs) {
4144 vcpu_clear(vmx);
4145 free_vmcs(vmx->vmcs);
4146 vmx->vmcs = NULL;
4147 }
4148 }
4149
4150 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
4151 {
4152 struct vcpu_vmx *vmx = to_vmx(vcpu);
4153
4154 free_vpid(vmx);
4155 vmx_free_vmcs(vcpu);
4156 kfree(vmx->guest_msrs);
4157 kvm_vcpu_uninit(vcpu);
4158 kmem_cache_free(kvm_vcpu_cache, vmx);
4159 }
4160
4161 static inline void vmcs_init(struct vmcs *vmcs)
4162 {
4163 u64 phys_addr = __pa(per_cpu(vmxarea, raw_smp_processor_id()));
4164
4165 if (!vmm_exclusive)
4166 kvm_cpu_vmxon(phys_addr);
4167
4168 vmcs_clear(vmcs);
4169
4170 if (!vmm_exclusive)
4171 kvm_cpu_vmxoff();
4172 }
4173
4174 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
4175 {
4176 int err;
4177 struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
4178 int cpu;
4179
4180 if (!vmx)
4181 return ERR_PTR(-ENOMEM);
4182
4183 allocate_vpid(vmx);
4184
4185 err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
4186 if (err)
4187 goto free_vcpu;
4188
4189 vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
4190 if (!vmx->guest_msrs) {
4191 err = -ENOMEM;
4192 goto uninit_vcpu;
4193 }
4194
4195 vmx->vmcs = alloc_vmcs();
4196 if (!vmx->vmcs)
4197 goto free_msrs;
4198
4199 vmcs_init(vmx->vmcs);
4200
4201 cpu = get_cpu();
4202 vmx_vcpu_load(&vmx->vcpu, cpu);
4203 vmx->vcpu.cpu = cpu;
4204 err = vmx_vcpu_setup(vmx);
4205 vmx_vcpu_put(&vmx->vcpu);
4206 put_cpu();
4207 if (err)
4208 goto free_vmcs;
4209 if (vm_need_virtualize_apic_accesses(kvm))
4210 if (alloc_apic_access_page(kvm) != 0)
4211 goto free_vmcs;
4212
4213 if (enable_ept) {
4214 if (!kvm->arch.ept_identity_map_addr)
4215 kvm->arch.ept_identity_map_addr =
4216 VMX_EPT_IDENTITY_PAGETABLE_ADDR;
4217 if (alloc_identity_pagetable(kvm) != 0)
4218 goto free_vmcs;
4219 }
4220
4221 return &vmx->vcpu;
4222
4223 free_vmcs:
4224 free_vmcs(vmx->vmcs);
4225 free_msrs:
4226 kfree(vmx->guest_msrs);
4227 uninit_vcpu:
4228 kvm_vcpu_uninit(&vmx->vcpu);
4229 free_vcpu:
4230 free_vpid(vmx);
4231 kmem_cache_free(kvm_vcpu_cache, vmx);
4232 return ERR_PTR(err);
4233 }
4234
4235 static void __init vmx_check_processor_compat(void *rtn)
4236 {
4237 struct vmcs_config vmcs_conf;
4238
4239 *(int *)rtn = 0;
4240 if (setup_vmcs_config(&vmcs_conf) < 0)
4241 *(int *)rtn = -EIO;
4242 if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
4243 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
4244 smp_processor_id());
4245 *(int *)rtn = -EIO;
4246 }
4247 }
4248
4249 static int get_ept_level(void)
4250 {
4251 return VMX_EPT_DEFAULT_GAW + 1;
4252 }
4253
4254 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
4255 {
4256 u64 ret;
4257
4258 /* For VT-d and EPT combination
4259 * 1. MMIO: always map as UC
4260 * 2. EPT with VT-d:
4261 * a. VT-d without snooping control feature: can't guarantee the
4262 * result, try to trust guest.
4263 * b. VT-d with snooping control feature: snooping control feature of
4264 * VT-d engine can guarantee the cache correctness. Just set it
4265 * to WB to keep consistent with host. So the same as item 3.
4266 * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
4267 * consistent with host MTRR
4268 */
4269 if (is_mmio)
4270 ret = MTRR_TYPE_UNCACHABLE << VMX_EPT_MT_EPTE_SHIFT;
4271 else if (vcpu->kvm->arch.iommu_domain &&
4272 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY))
4273 ret = kvm_get_guest_memory_type(vcpu, gfn) <<
4274 VMX_EPT_MT_EPTE_SHIFT;
4275 else
4276 ret = (MTRR_TYPE_WRBACK << VMX_EPT_MT_EPTE_SHIFT)
4277 | VMX_EPT_IPAT_BIT;
4278
4279 return ret;
4280 }
4281
4282 #define _ER(x) { EXIT_REASON_##x, #x }
4283
4284 static const struct trace_print_flags vmx_exit_reasons_str[] = {
4285 _ER(EXCEPTION_NMI),
4286 _ER(EXTERNAL_INTERRUPT),
4287 _ER(TRIPLE_FAULT),
4288 _ER(PENDING_INTERRUPT),
4289 _ER(NMI_WINDOW),
4290 _ER(TASK_SWITCH),
4291 _ER(CPUID),
4292 _ER(HLT),
4293 _ER(INVLPG),
4294 _ER(RDPMC),
4295 _ER(RDTSC),
4296 _ER(VMCALL),
4297 _ER(VMCLEAR),
4298 _ER(VMLAUNCH),
4299 _ER(VMPTRLD),
4300 _ER(VMPTRST),
4301 _ER(VMREAD),
4302 _ER(VMRESUME),
4303 _ER(VMWRITE),
4304 _ER(VMOFF),
4305 _ER(VMON),
4306 _ER(CR_ACCESS),
4307 _ER(DR_ACCESS),
4308 _ER(IO_INSTRUCTION),
4309 _ER(MSR_READ),
4310 _ER(MSR_WRITE),
4311 _ER(MWAIT_INSTRUCTION),
4312 _ER(MONITOR_INSTRUCTION),
4313 _ER(PAUSE_INSTRUCTION),
4314 _ER(MCE_DURING_VMENTRY),
4315 _ER(TPR_BELOW_THRESHOLD),
4316 _ER(APIC_ACCESS),
4317 _ER(EPT_VIOLATION),
4318 _ER(EPT_MISCONFIG),
4319 _ER(WBINVD),
4320 { -1, NULL }
4321 };
4322
4323 #undef _ER
4324
4325 static int vmx_get_lpage_level(void)
4326 {
4327 if (enable_ept && !cpu_has_vmx_ept_1g_page())
4328 return PT_DIRECTORY_LEVEL;
4329 else
4330 /* For shadow and EPT supported 1GB page */
4331 return PT_PDPE_LEVEL;
4332 }
4333
4334 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
4335 {
4336 struct kvm_cpuid_entry2 *best;
4337 struct vcpu_vmx *vmx = to_vmx(vcpu);
4338 u32 exec_control;
4339
4340 vmx->rdtscp_enabled = false;
4341 if (vmx_rdtscp_supported()) {
4342 exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
4343 if (exec_control & SECONDARY_EXEC_RDTSCP) {
4344 best = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
4345 if (best && (best->edx & bit(X86_FEATURE_RDTSCP)))
4346 vmx->rdtscp_enabled = true;
4347 else {
4348 exec_control &= ~SECONDARY_EXEC_RDTSCP;
4349 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
4350 exec_control);
4351 }
4352 }
4353 }
4354 }
4355
4356 static void vmx_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
4357 {
4358 }
4359
4360 static struct kvm_x86_ops vmx_x86_ops = {
4361 .cpu_has_kvm_support = cpu_has_kvm_support,
4362 .disabled_by_bios = vmx_disabled_by_bios,
4363 .hardware_setup = hardware_setup,
4364 .hardware_unsetup = hardware_unsetup,
4365 .check_processor_compatibility = vmx_check_processor_compat,
4366 .hardware_enable = hardware_enable,
4367 .hardware_disable = hardware_disable,
4368 .cpu_has_accelerated_tpr = report_flexpriority,
4369
4370 .vcpu_create = vmx_create_vcpu,
4371 .vcpu_free = vmx_free_vcpu,
4372 .vcpu_reset = vmx_vcpu_reset,
4373
4374 .prepare_guest_switch = vmx_save_host_state,
4375 .vcpu_load = vmx_vcpu_load,
4376 .vcpu_put = vmx_vcpu_put,
4377
4378 .set_guest_debug = set_guest_debug,
4379 .get_msr = vmx_get_msr,
4380 .set_msr = vmx_set_msr,
4381 .get_segment_base = vmx_get_segment_base,
4382 .get_segment = vmx_get_segment,
4383 .set_segment = vmx_set_segment,
4384 .get_cpl = vmx_get_cpl,
4385 .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
4386 .decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
4387 .decache_cr3 = vmx_decache_cr3,
4388 .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
4389 .set_cr0 = vmx_set_cr0,
4390 .set_cr3 = vmx_set_cr3,
4391 .set_cr4 = vmx_set_cr4,
4392 .set_efer = vmx_set_efer,
4393 .get_idt = vmx_get_idt,
4394 .set_idt = vmx_set_idt,
4395 .get_gdt = vmx_get_gdt,
4396 .set_gdt = vmx_set_gdt,
4397 .set_dr7 = vmx_set_dr7,
4398 .cache_reg = vmx_cache_reg,
4399 .get_rflags = vmx_get_rflags,
4400 .set_rflags = vmx_set_rflags,
4401 .fpu_activate = vmx_fpu_activate,
4402 .fpu_deactivate = vmx_fpu_deactivate,
4403
4404 .tlb_flush = vmx_flush_tlb,
4405
4406 .run = vmx_vcpu_run,
4407 .handle_exit = vmx_handle_exit,
4408 .skip_emulated_instruction = skip_emulated_instruction,
4409 .set_interrupt_shadow = vmx_set_interrupt_shadow,
4410 .get_interrupt_shadow = vmx_get_interrupt_shadow,
4411 .patch_hypercall = vmx_patch_hypercall,
4412 .set_irq = vmx_inject_irq,
4413 .set_nmi = vmx_inject_nmi,
4414 .queue_exception = vmx_queue_exception,
4415 .cancel_injection = vmx_cancel_injection,
4416 .interrupt_allowed = vmx_interrupt_allowed,
4417 .nmi_allowed = vmx_nmi_allowed,
4418 .get_nmi_mask = vmx_get_nmi_mask,
4419 .set_nmi_mask = vmx_set_nmi_mask,
4420 .enable_nmi_window = enable_nmi_window,
4421 .enable_irq_window = enable_irq_window,
4422 .update_cr8_intercept = update_cr8_intercept,
4423
4424 .set_tss_addr = vmx_set_tss_addr,
4425 .get_tdp_level = get_ept_level,
4426 .get_mt_mask = vmx_get_mt_mask,
4427
4428 .get_exit_info = vmx_get_exit_info,
4429 .exit_reasons_str = vmx_exit_reasons_str,
4430
4431 .get_lpage_level = vmx_get_lpage_level,
4432
4433 .cpuid_update = vmx_cpuid_update,
4434
4435 .rdtscp_supported = vmx_rdtscp_supported,
4436
4437 .set_supported_cpuid = vmx_set_supported_cpuid,
4438
4439 .has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
4440
4441 .write_tsc_offset = vmx_write_tsc_offset,
4442 .adjust_tsc_offset = vmx_adjust_tsc_offset,
4443
4444 .set_tdp_cr3 = vmx_set_cr3,
4445 };
4446
4447 static int __init vmx_init(void)
4448 {
4449 int r, i;
4450
4451 rdmsrl_safe(MSR_EFER, &host_efer);
4452
4453 for (i = 0; i < NR_VMX_MSR; ++i)
4454 kvm_define_shared_msr(i, vmx_msr_index[i]);
4455
4456 vmx_io_bitmap_a = (unsigned long *)__get_free_page(GFP_KERNEL);
4457 if (!vmx_io_bitmap_a)
4458 return -ENOMEM;
4459
4460 vmx_io_bitmap_b = (unsigned long *)__get_free_page(GFP_KERNEL);
4461 if (!vmx_io_bitmap_b) {
4462 r = -ENOMEM;
4463 goto out;
4464 }
4465
4466 vmx_msr_bitmap_legacy = (unsigned long *)__get_free_page(GFP_KERNEL);
4467 if (!vmx_msr_bitmap_legacy) {
4468 r = -ENOMEM;
4469 goto out1;
4470 }
4471
4472 vmx_msr_bitmap_longmode = (unsigned long *)__get_free_page(GFP_KERNEL);
4473 if (!vmx_msr_bitmap_longmode) {
4474 r = -ENOMEM;
4475 goto out2;
4476 }
4477
4478 /*
4479 * Allow direct access to the PC debug port (it is often used for I/O
4480 * delays, but the vmexits simply slow things down).
4481 */
4482 memset(vmx_io_bitmap_a, 0xff, PAGE_SIZE);
4483 clear_bit(0x80, vmx_io_bitmap_a);
4484
4485 memset(vmx_io_bitmap_b, 0xff, PAGE_SIZE);
4486
4487 memset(vmx_msr_bitmap_legacy, 0xff, PAGE_SIZE);
4488 memset(vmx_msr_bitmap_longmode, 0xff, PAGE_SIZE);
4489
4490 set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
4491
4492 r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx),
4493 __alignof__(struct vcpu_vmx), THIS_MODULE);
4494 if (r)
4495 goto out3;
4496
4497 vmx_disable_intercept_for_msr(MSR_FS_BASE, false);
4498 vmx_disable_intercept_for_msr(MSR_GS_BASE, false);
4499 vmx_disable_intercept_for_msr(MSR_KERNEL_GS_BASE, true);
4500 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_CS, false);
4501 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_ESP, false);
4502 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_EIP, false);
4503
4504 if (enable_ept) {
4505 bypass_guest_pf = 0;
4506 kvm_mmu_set_mask_ptes(0ull, 0ull, 0ull, 0ull,
4507 VMX_EPT_EXECUTABLE_MASK);
4508 kvm_enable_tdp();
4509 } else
4510 kvm_disable_tdp();
4511
4512 if (bypass_guest_pf)
4513 kvm_mmu_set_nonpresent_ptes(~0xffeull, 0ull);
4514
4515 return 0;
4516
4517 out3:
4518 free_page((unsigned long)vmx_msr_bitmap_longmode);
4519 out2:
4520 free_page((unsigned long)vmx_msr_bitmap_legacy);
4521 out1:
4522 free_page((unsigned long)vmx_io_bitmap_b);
4523 out:
4524 free_page((unsigned long)vmx_io_bitmap_a);
4525 return r;
4526 }
4527
4528 static void __exit vmx_exit(void)
4529 {
4530 free_page((unsigned long)vmx_msr_bitmap_legacy);
4531 free_page((unsigned long)vmx_msr_bitmap_longmode);
4532 free_page((unsigned long)vmx_io_bitmap_b);
4533 free_page((unsigned long)vmx_io_bitmap_a);
4534
4535 kvm_exit();
4536 }
4537
4538 module_init(vmx_init)
4539 module_exit(vmx_exit)
Linux 2.6 ibm-acpi/thinkpad-acpi driver git tree