projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge branch 'for-2.6.30' of git://git.kernel.org/pub/scm/linux/kernel/git/broonie...
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / arch / x86 / kvm / vmx.c
1 /*
2 * Kernel-based Virtual Machine driver for Linux
3 *
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
6 *
7 * Copyright (C) 2006 Qumranet, Inc.
8 *
9 * Authors:
10 * Avi Kivity <avi@qumranet.com>
11 * Yaniv Kamay <yaniv@qumranet.com>
12 *
13 * This work is licensed under the terms of the GNU GPL, version 2. See
14 * the COPYING file in the top-level directory.
15 *
16 */
17
18 #include "irq.h"
19 #include "mmu.h"
20
21 #include <linux/kvm_host.h>
22 #include <linux/module.h>
23 #include <linux/kernel.h>
24 #include <linux/mm.h>
25 #include <linux/highmem.h>
26 #include <linux/sched.h>
27 #include <linux/moduleparam.h>
28 #include "kvm_cache_regs.h"
29 #include "x86.h"
30
31 #include <asm/io.h>
32 #include <asm/desc.h>
33 #include <asm/vmx.h>
34 #include <asm/virtext.h>
35
36 #define __ex(x) __kvm_handle_fault_on_reboot(x)
37
38 MODULE_AUTHOR("Qumranet");
39 MODULE_LICENSE("GPL");
40
41 static int bypass_guest_pf = 1;
42 module_param(bypass_guest_pf, bool, 0);
43
44 static int enable_vpid = 1;
45 module_param(enable_vpid, bool, 0);
46
47 static int flexpriority_enabled = 1;
48 module_param(flexpriority_enabled, bool, 0);
49
50 static int enable_ept = 1;
51 module_param(enable_ept, bool, 0);
52
53 static int emulate_invalid_guest_state = 0;
54 module_param(emulate_invalid_guest_state, bool, 0);
55
56 struct vmcs {
57 u32 revision_id;
58 u32 abort;
59 char data[0];
60 };
61
62 struct vcpu_vmx {
63 struct kvm_vcpu vcpu;
64 struct list_head local_vcpus_link;
65 unsigned long host_rsp;
66 int launched;
67 u8 fail;
68 u32 idt_vectoring_info;
69 struct kvm_msr_entry *guest_msrs;
70 struct kvm_msr_entry *host_msrs;
71 int nmsrs;
72 int save_nmsrs;
73 int msr_offset_efer;
74 #ifdef CONFIG_X86_64
75 int msr_offset_kernel_gs_base;
76 #endif
77 struct vmcs *vmcs;
78 struct {
79 int loaded;
80 u16 fs_sel, gs_sel, ldt_sel;
81 int gs_ldt_reload_needed;
82 int fs_reload_needed;
83 int guest_efer_loaded;
84 } host_state;
85 struct {
86 struct {
87 bool pending;
88 u8 vector;
89 unsigned rip;
90 } irq;
91 } rmode;
92 int vpid;
93 bool emulation_required;
94 enum emulation_result invalid_state_emulation_result;
95
96 /* Support for vnmi-less CPUs */
97 int soft_vnmi_blocked;
98 ktime_t entry_time;
99 s64 vnmi_blocked_time;
100 };
101
102 static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
103 {
104 return container_of(vcpu, struct vcpu_vmx, vcpu);
105 }
106
107 static int init_rmode(struct kvm *kvm);
108 static u64 construct_eptp(unsigned long root_hpa);
109
110 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
111 static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
112 static DEFINE_PER_CPU(struct list_head, vcpus_on_cpu);
113
114 static struct page *vmx_io_bitmap_a;
115 static struct page *vmx_io_bitmap_b;
116 static struct page *vmx_msr_bitmap;
117
118 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
119 static DEFINE_SPINLOCK(vmx_vpid_lock);
120
121 static struct vmcs_config {
122 int size;
123 int order;
124 u32 revision_id;
125 u32 pin_based_exec_ctrl;
126 u32 cpu_based_exec_ctrl;
127 u32 cpu_based_2nd_exec_ctrl;
128 u32 vmexit_ctrl;
129 u32 vmentry_ctrl;
130 } vmcs_config;
131
132 static struct vmx_capability {
133 u32 ept;
134 u32 vpid;
135 } vmx_capability;
136
137 #define VMX_SEGMENT_FIELD(seg) \
138 [VCPU_SREG_##seg] = { \
139 .selector = GUEST_##seg##_SELECTOR, \
140 .base = GUEST_##seg##_BASE, \
141 .limit = GUEST_##seg##_LIMIT, \
142 .ar_bytes = GUEST_##seg##_AR_BYTES, \
143 }
144
145 static struct kvm_vmx_segment_field {
146 unsigned selector;
147 unsigned base;
148 unsigned limit;
149 unsigned ar_bytes;
150 } kvm_vmx_segment_fields[] = {
151 VMX_SEGMENT_FIELD(CS),
152 VMX_SEGMENT_FIELD(DS),
153 VMX_SEGMENT_FIELD(ES),
154 VMX_SEGMENT_FIELD(FS),
155 VMX_SEGMENT_FIELD(GS),
156 VMX_SEGMENT_FIELD(SS),
157 VMX_SEGMENT_FIELD(TR),
158 VMX_SEGMENT_FIELD(LDTR),
159 };
160
161 /*
162 * Keep MSR_K6_STAR at the end, as setup_msrs() will try to optimize it
163 * away by decrementing the array size.
164 */
165 static const u32 vmx_msr_index[] = {
166 #ifdef CONFIG_X86_64
167 MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR, MSR_KERNEL_GS_BASE,
168 #endif
169 MSR_EFER, MSR_K6_STAR,
170 };
171 #define NR_VMX_MSR ARRAY_SIZE(vmx_msr_index)
172
173 static void load_msrs(struct kvm_msr_entry *e, int n)
174 {
175 int i;
176
177 for (i = 0; i < n; ++i)
178 wrmsrl(e[i].index, e[i].data);
179 }
180
181 static void save_msrs(struct kvm_msr_entry *e, int n)
182 {
183 int i;
184
185 for (i = 0; i < n; ++i)
186 rdmsrl(e[i].index, e[i].data);
187 }
188
189 static inline int is_page_fault(u32 intr_info)
190 {
191 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
192 INTR_INFO_VALID_MASK)) ==
193 (INTR_TYPE_HARD_EXCEPTION | PF_VECTOR | INTR_INFO_VALID_MASK);
194 }
195
196 static inline int is_no_device(u32 intr_info)
197 {
198 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
199 INTR_INFO_VALID_MASK)) ==
200 (INTR_TYPE_HARD_EXCEPTION | NM_VECTOR | INTR_INFO_VALID_MASK);
201 }
202
203 static inline int is_invalid_opcode(u32 intr_info)
204 {
205 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
206 INTR_INFO_VALID_MASK)) ==
207 (INTR_TYPE_HARD_EXCEPTION | UD_VECTOR | INTR_INFO_VALID_MASK);
208 }
209
210 static inline int is_external_interrupt(u32 intr_info)
211 {
212 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
213 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
214 }
215
216 static inline int cpu_has_vmx_msr_bitmap(void)
217 {
218 return (vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS);
219 }
220
221 static inline int cpu_has_vmx_tpr_shadow(void)
222 {
223 return (vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW);
224 }
225
226 static inline int vm_need_tpr_shadow(struct kvm *kvm)
227 {
228 return ((cpu_has_vmx_tpr_shadow()) && (irqchip_in_kernel(kvm)));
229 }
230
231 static inline int cpu_has_secondary_exec_ctrls(void)
232 {
233 return (vmcs_config.cpu_based_exec_ctrl &
234 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS);
235 }
236
237 static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
238 {
239 return flexpriority_enabled
240 && (vmcs_config.cpu_based_2nd_exec_ctrl &
241 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
242 }
243
244 static inline int cpu_has_vmx_invept_individual_addr(void)
245 {
246 return (!!(vmx_capability.ept & VMX_EPT_EXTENT_INDIVIDUAL_BIT));
247 }
248
249 static inline int cpu_has_vmx_invept_context(void)
250 {
251 return (!!(vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT));
252 }
253
254 static inline int cpu_has_vmx_invept_global(void)
255 {
256 return (!!(vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT));
257 }
258
259 static inline int cpu_has_vmx_ept(void)
260 {
261 return (vmcs_config.cpu_based_2nd_exec_ctrl &
262 SECONDARY_EXEC_ENABLE_EPT);
263 }
264
265 static inline int vm_need_ept(void)
266 {
267 return (cpu_has_vmx_ept() && enable_ept);
268 }
269
270 static inline int vm_need_virtualize_apic_accesses(struct kvm *kvm)
271 {
272 return ((cpu_has_vmx_virtualize_apic_accesses()) &&
273 (irqchip_in_kernel(kvm)));
274 }
275
276 static inline int cpu_has_vmx_vpid(void)
277 {
278 return (vmcs_config.cpu_based_2nd_exec_ctrl &
279 SECONDARY_EXEC_ENABLE_VPID);
280 }
281
282 static inline int cpu_has_virtual_nmis(void)
283 {
284 return vmcs_config.pin_based_exec_ctrl & PIN_BASED_VIRTUAL_NMIS;
285 }
286
287 static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
288 {
289 int i;
290
291 for (i = 0; i < vmx->nmsrs; ++i)
292 if (vmx->guest_msrs[i].index == msr)
293 return i;
294 return -1;
295 }
296
297 static inline void __invvpid(int ext, u16 vpid, gva_t gva)
298 {
299 struct {
300 u64 vpid : 16;
301 u64 rsvd : 48;
302 u64 gva;
303 } operand = { vpid, 0, gva };
304
305 asm volatile (__ex(ASM_VMX_INVVPID)
306 /* CF==1 or ZF==1 --> rc = -1 */
307 "; ja 1f ; ud2 ; 1:"
308 : : "a"(&operand), "c"(ext) : "cc", "memory");
309 }
310
311 static inline void __invept(int ext, u64 eptp, gpa_t gpa)
312 {
313 struct {
314 u64 eptp, gpa;
315 } operand = {eptp, gpa};
316
317 asm volatile (__ex(ASM_VMX_INVEPT)
318 /* CF==1 or ZF==1 --> rc = -1 */
319 "; ja 1f ; ud2 ; 1:\n"
320 : : "a" (&operand), "c" (ext) : "cc", "memory");
321 }
322
323 static struct kvm_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
324 {
325 int i;
326
327 i = __find_msr_index(vmx, msr);
328 if (i >= 0)
329 return &vmx->guest_msrs[i];
330 return NULL;
331 }
332
333 static void vmcs_clear(struct vmcs *vmcs)
334 {
335 u64 phys_addr = __pa(vmcs);
336 u8 error;
337
338 asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0"
339 : "=g"(error) : "a"(&phys_addr), "m"(phys_addr)
340 : "cc", "memory");
341 if (error)
342 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
343 vmcs, phys_addr);
344 }
345
346 static void __vcpu_clear(void *arg)
347 {
348 struct vcpu_vmx *vmx = arg;
349 int cpu = raw_smp_processor_id();
350
351 if (vmx->vcpu.cpu == cpu)
352 vmcs_clear(vmx->vmcs);
353 if (per_cpu(current_vmcs, cpu) == vmx->vmcs)
354 per_cpu(current_vmcs, cpu) = NULL;
355 rdtscll(vmx->vcpu.arch.host_tsc);
356 list_del(&vmx->local_vcpus_link);
357 vmx->vcpu.cpu = -1;
358 vmx->launched = 0;
359 }
360
361 static void vcpu_clear(struct vcpu_vmx *vmx)
362 {
363 if (vmx->vcpu.cpu == -1)
364 return;
365 smp_call_function_single(vmx->vcpu.cpu, __vcpu_clear, vmx, 1);
366 }
367
368 static inline void vpid_sync_vcpu_all(struct vcpu_vmx *vmx)
369 {
370 if (vmx->vpid == 0)
371 return;
372
373 __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vmx->vpid, 0);
374 }
375
376 static inline void ept_sync_global(void)
377 {
378 if (cpu_has_vmx_invept_global())
379 __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0);
380 }
381
382 static inline void ept_sync_context(u64 eptp)
383 {
384 if (vm_need_ept()) {
385 if (cpu_has_vmx_invept_context())
386 __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0);
387 else
388 ept_sync_global();
389 }
390 }
391
392 static inline void ept_sync_individual_addr(u64 eptp, gpa_t gpa)
393 {
394 if (vm_need_ept()) {
395 if (cpu_has_vmx_invept_individual_addr())
396 __invept(VMX_EPT_EXTENT_INDIVIDUAL_ADDR,
397 eptp, gpa);
398 else
399 ept_sync_context(eptp);
400 }
401 }
402
403 static unsigned long vmcs_readl(unsigned long field)
404 {
405 unsigned long value;
406
407 asm volatile (__ex(ASM_VMX_VMREAD_RDX_RAX)
408 : "=a"(value) : "d"(field) : "cc");
409 return value;
410 }
411
412 static u16 vmcs_read16(unsigned long field)
413 {
414 return vmcs_readl(field);
415 }
416
417 static u32 vmcs_read32(unsigned long field)
418 {
419 return vmcs_readl(field);
420 }
421
422 static u64 vmcs_read64(unsigned long field)
423 {
424 #ifdef CONFIG_X86_64
425 return vmcs_readl(field);
426 #else
427 return vmcs_readl(field) | ((u64)vmcs_readl(field+1) << 32);
428 #endif
429 }
430
431 static noinline void vmwrite_error(unsigned long field, unsigned long value)
432 {
433 printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
434 field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
435 dump_stack();
436 }
437
438 static void vmcs_writel(unsigned long field, unsigned long value)
439 {
440 u8 error;
441
442 asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0"
443 : "=q"(error) : "a"(value), "d"(field) : "cc");
444 if (unlikely(error))
445 vmwrite_error(field, value);
446 }
447
448 static void vmcs_write16(unsigned long field, u16 value)
449 {
450 vmcs_writel(field, value);
451 }
452
453 static void vmcs_write32(unsigned long field, u32 value)
454 {
455 vmcs_writel(field, value);
456 }
457
458 static void vmcs_write64(unsigned long field, u64 value)
459 {
460 vmcs_writel(field, value);
461 #ifndef CONFIG_X86_64
462 asm volatile ("");
463 vmcs_writel(field+1, value >> 32);
464 #endif
465 }
466
467 static void vmcs_clear_bits(unsigned long field, u32 mask)
468 {
469 vmcs_writel(field, vmcs_readl(field) & ~mask);
470 }
471
472 static void vmcs_set_bits(unsigned long field, u32 mask)
473 {
474 vmcs_writel(field, vmcs_readl(field) | mask);
475 }
476
477 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
478 {
479 u32 eb;
480
481 eb = (1u << PF_VECTOR) | (1u << UD_VECTOR);
482 if (!vcpu->fpu_active)
483 eb |= 1u << NM_VECTOR;
484 if (vcpu->guest_debug & KVM_GUESTDBG_ENABLE) {
485 if (vcpu->guest_debug &
486 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
487 eb |= 1u << DB_VECTOR;
488 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
489 eb |= 1u << BP_VECTOR;
490 }
491 if (vcpu->arch.rmode.active)
492 eb = ~0;
493 if (vm_need_ept())
494 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
495 vmcs_write32(EXCEPTION_BITMAP, eb);
496 }
497
498 static void reload_tss(void)
499 {
500 /*
501 * VT restores TR but not its size. Useless.
502 */
503 struct descriptor_table gdt;
504 struct desc_struct *descs;
505
506 kvm_get_gdt(&gdt);
507 descs = (void *)gdt.base;
508 descs[GDT_ENTRY_TSS].type = 9; /* available TSS */
509 load_TR_desc();
510 }
511
512 static void load_transition_efer(struct vcpu_vmx *vmx)
513 {
514 int efer_offset = vmx->msr_offset_efer;
515 u64 host_efer = vmx->host_msrs[efer_offset].data;
516 u64 guest_efer = vmx->guest_msrs[efer_offset].data;
517 u64 ignore_bits;
518
519 if (efer_offset < 0)
520 return;
521 /*
522 * NX is emulated; LMA and LME handled by hardware; SCE meaninless
523 * outside long mode
524 */
525 ignore_bits = EFER_NX | EFER_SCE;
526 #ifdef CONFIG_X86_64
527 ignore_bits |= EFER_LMA | EFER_LME;
528 /* SCE is meaningful only in long mode on Intel */
529 if (guest_efer & EFER_LMA)
530 ignore_bits &= ~(u64)EFER_SCE;
531 #endif
532 if ((guest_efer & ~ignore_bits) == (host_efer & ~ignore_bits))
533 return;
534
535 vmx->host_state.guest_efer_loaded = 1;
536 guest_efer &= ~ignore_bits;
537 guest_efer |= host_efer & ignore_bits;
538 wrmsrl(MSR_EFER, guest_efer);
539 vmx->vcpu.stat.efer_reload++;
540 }
541
542 static void reload_host_efer(struct vcpu_vmx *vmx)
543 {
544 if (vmx->host_state.guest_efer_loaded) {
545 vmx->host_state.guest_efer_loaded = 0;
546 load_msrs(vmx->host_msrs + vmx->msr_offset_efer, 1);
547 }
548 }
549
550 static void vmx_save_host_state(struct kvm_vcpu *vcpu)
551 {
552 struct vcpu_vmx *vmx = to_vmx(vcpu);
553
554 if (vmx->host_state.loaded)
555 return;
556
557 vmx->host_state.loaded = 1;
558 /*
559 * Set host fs and gs selectors. Unfortunately, 22.2.3 does not
560 * allow segment selectors with cpl > 0 or ti == 1.
561 */
562 vmx->host_state.ldt_sel = kvm_read_ldt();
563 vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
564 vmx->host_state.fs_sel = kvm_read_fs();
565 if (!(vmx->host_state.fs_sel & 7)) {
566 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
567 vmx->host_state.fs_reload_needed = 0;
568 } else {
569 vmcs_write16(HOST_FS_SELECTOR, 0);
570 vmx->host_state.fs_reload_needed = 1;
571 }
572 vmx->host_state.gs_sel = kvm_read_gs();
573 if (!(vmx->host_state.gs_sel & 7))
574 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
575 else {
576 vmcs_write16(HOST_GS_SELECTOR, 0);
577 vmx->host_state.gs_ldt_reload_needed = 1;
578 }
579
580 #ifdef CONFIG_X86_64
581 vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
582 vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
583 #else
584 vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
585 vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
586 #endif
587
588 #ifdef CONFIG_X86_64
589 if (is_long_mode(&vmx->vcpu))
590 save_msrs(vmx->host_msrs +
591 vmx->msr_offset_kernel_gs_base, 1);
592
593 #endif
594 load_msrs(vmx->guest_msrs, vmx->save_nmsrs);
595 load_transition_efer(vmx);
596 }
597
598 static void __vmx_load_host_state(struct vcpu_vmx *vmx)
599 {
600 unsigned long flags;
601
602 if (!vmx->host_state.loaded)
603 return;
604
605 ++vmx->vcpu.stat.host_state_reload;
606 vmx->host_state.loaded = 0;
607 if (vmx->host_state.fs_reload_needed)
608 kvm_load_fs(vmx->host_state.fs_sel);
609 if (vmx->host_state.gs_ldt_reload_needed) {
610 kvm_load_ldt(vmx->host_state.ldt_sel);
611 /*
612 * If we have to reload gs, we must take care to
613 * preserve our gs base.
614 */
615 local_irq_save(flags);
616 kvm_load_gs(vmx->host_state.gs_sel);
617 #ifdef CONFIG_X86_64
618 wrmsrl(MSR_GS_BASE, vmcs_readl(HOST_GS_BASE));
619 #endif
620 local_irq_restore(flags);
621 }
622 reload_tss();
623 save_msrs(vmx->guest_msrs, vmx->save_nmsrs);
624 load_msrs(vmx->host_msrs, vmx->save_nmsrs);
625 reload_host_efer(vmx);
626 }
627
628 static void vmx_load_host_state(struct vcpu_vmx *vmx)
629 {
630 preempt_disable();
631 __vmx_load_host_state(vmx);
632 preempt_enable();
633 }
634
635 /*
636 * Switches to specified vcpu, until a matching vcpu_put(), but assumes
637 * vcpu mutex is already taken.
638 */
639 static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
640 {
641 struct vcpu_vmx *vmx = to_vmx(vcpu);
642 u64 phys_addr = __pa(vmx->vmcs);
643 u64 tsc_this, delta, new_offset;
644
645 if (vcpu->cpu != cpu) {
646 vcpu_clear(vmx);
647 kvm_migrate_timers(vcpu);
648 vpid_sync_vcpu_all(vmx);
649 local_irq_disable();
650 list_add(&vmx->local_vcpus_link,
651 &per_cpu(vcpus_on_cpu, cpu));
652 local_irq_enable();
653 }
654
655 if (per_cpu(current_vmcs, cpu) != vmx->vmcs) {
656 u8 error;
657
658 per_cpu(current_vmcs, cpu) = vmx->vmcs;
659 asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0"
660 : "=g"(error) : "a"(&phys_addr), "m"(phys_addr)
661 : "cc");
662 if (error)
663 printk(KERN_ERR "kvm: vmptrld %p/%llx fail\n",
664 vmx->vmcs, phys_addr);
665 }
666
667 if (vcpu->cpu != cpu) {
668 struct descriptor_table dt;
669 unsigned long sysenter_esp;
670
671 vcpu->cpu = cpu;
672 /*
673 * Linux uses per-cpu TSS and GDT, so set these when switching
674 * processors.
675 */
676 vmcs_writel(HOST_TR_BASE, kvm_read_tr_base()); /* 22.2.4 */
677 kvm_get_gdt(&dt);
678 vmcs_writel(HOST_GDTR_BASE, dt.base); /* 22.2.4 */
679
680 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
681 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
682
683 /*
684 * Make sure the time stamp counter is monotonous.
685 */
686 rdtscll(tsc_this);
687 if (tsc_this < vcpu->arch.host_tsc) {
688 delta = vcpu->arch.host_tsc - tsc_this;
689 new_offset = vmcs_read64(TSC_OFFSET) + delta;
690 vmcs_write64(TSC_OFFSET, new_offset);
691 }
692 }
693 }
694
695 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
696 {
697 __vmx_load_host_state(to_vmx(vcpu));
698 }
699
700 static void vmx_fpu_activate(struct kvm_vcpu *vcpu)
701 {
702 if (vcpu->fpu_active)
703 return;
704 vcpu->fpu_active = 1;
705 vmcs_clear_bits(GUEST_CR0, X86_CR0_TS);
706 if (vcpu->arch.cr0 & X86_CR0_TS)
707 vmcs_set_bits(GUEST_CR0, X86_CR0_TS);
708 update_exception_bitmap(vcpu);
709 }
710
711 static void vmx_fpu_deactivate(struct kvm_vcpu *vcpu)
712 {
713 if (!vcpu->fpu_active)
714 return;
715 vcpu->fpu_active = 0;
716 vmcs_set_bits(GUEST_CR0, X86_CR0_TS);
717 update_exception_bitmap(vcpu);
718 }
719
720 static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
721 {
722 return vmcs_readl(GUEST_RFLAGS);
723 }
724
725 static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
726 {
727 if (vcpu->arch.rmode.active)
728 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
729 vmcs_writel(GUEST_RFLAGS, rflags);
730 }
731
732 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
733 {
734 unsigned long rip;
735 u32 interruptibility;
736
737 rip = kvm_rip_read(vcpu);
738 rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
739 kvm_rip_write(vcpu, rip);
740
741 /*
742 * We emulated an instruction, so temporary interrupt blocking
743 * should be removed, if set.
744 */
745 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
746 if (interruptibility & 3)
747 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
748 interruptibility & ~3);
749 vcpu->arch.interrupt_window_open = 1;
750 }
751
752 static void vmx_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
753 bool has_error_code, u32 error_code)
754 {
755 struct vcpu_vmx *vmx = to_vmx(vcpu);
756 u32 intr_info = nr | INTR_INFO_VALID_MASK;
757
758 if (has_error_code) {
759 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
760 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
761 }
762
763 if (vcpu->arch.rmode.active) {
764 vmx->rmode.irq.pending = true;
765 vmx->rmode.irq.vector = nr;
766 vmx->rmode.irq.rip = kvm_rip_read(vcpu);
767 if (nr == BP_VECTOR || nr == OF_VECTOR)
768 vmx->rmode.irq.rip++;
769 intr_info |= INTR_TYPE_SOFT_INTR;
770 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
771 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, 1);
772 kvm_rip_write(vcpu, vmx->rmode.irq.rip - 1);
773 return;
774 }
775
776 if (nr == BP_VECTOR || nr == OF_VECTOR) {
777 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, 1);
778 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
779 } else
780 intr_info |= INTR_TYPE_HARD_EXCEPTION;
781
782 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
783 }
784
785 static bool vmx_exception_injected(struct kvm_vcpu *vcpu)
786 {
787 return false;
788 }
789
790 /*
791 * Swap MSR entry in host/guest MSR entry array.
792 */
793 #ifdef CONFIG_X86_64
794 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
795 {
796 struct kvm_msr_entry tmp;
797
798 tmp = vmx->guest_msrs[to];
799 vmx->guest_msrs[to] = vmx->guest_msrs[from];
800 vmx->guest_msrs[from] = tmp;
801 tmp = vmx->host_msrs[to];
802 vmx->host_msrs[to] = vmx->host_msrs[from];
803 vmx->host_msrs[from] = tmp;
804 }
805 #endif
806
807 /*
808 * Set up the vmcs to automatically save and restore system
809 * msrs. Don't touch the 64-bit msrs if the guest is in legacy
810 * mode, as fiddling with msrs is very expensive.
811 */
812 static void setup_msrs(struct vcpu_vmx *vmx)
813 {
814 int save_nmsrs;
815
816 vmx_load_host_state(vmx);
817 save_nmsrs = 0;
818 #ifdef CONFIG_X86_64
819 if (is_long_mode(&vmx->vcpu)) {
820 int index;
821
822 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
823 if (index >= 0)
824 move_msr_up(vmx, index, save_nmsrs++);
825 index = __find_msr_index(vmx, MSR_LSTAR);
826 if (index >= 0)
827 move_msr_up(vmx, index, save_nmsrs++);
828 index = __find_msr_index(vmx, MSR_CSTAR);
829 if (index >= 0)
830 move_msr_up(vmx, index, save_nmsrs++);
831 index = __find_msr_index(vmx, MSR_KERNEL_GS_BASE);
832 if (index >= 0)
833 move_msr_up(vmx, index, save_nmsrs++);
834 /*
835 * MSR_K6_STAR is only needed on long mode guests, and only
836 * if efer.sce is enabled.
837 */
838 index = __find_msr_index(vmx, MSR_K6_STAR);
839 if ((index >= 0) && (vmx->vcpu.arch.shadow_efer & EFER_SCE))
840 move_msr_up(vmx, index, save_nmsrs++);
841 }
842 #endif
843 vmx->save_nmsrs = save_nmsrs;
844
845 #ifdef CONFIG_X86_64
846 vmx->msr_offset_kernel_gs_base =
847 __find_msr_index(vmx, MSR_KERNEL_GS_BASE);
848 #endif
849 vmx->msr_offset_efer = __find_msr_index(vmx, MSR_EFER);
850 }
851
852 /*
853 * reads and returns guest's timestamp counter "register"
854 * guest_tsc = host_tsc + tsc_offset -- 21.3
855 */
856 static u64 guest_read_tsc(void)
857 {
858 u64 host_tsc, tsc_offset;
859
860 rdtscll(host_tsc);
861 tsc_offset = vmcs_read64(TSC_OFFSET);
862 return host_tsc + tsc_offset;
863 }
864
865 /*
866 * writes 'guest_tsc' into guest's timestamp counter "register"
867 * guest_tsc = host_tsc + tsc_offset ==> tsc_offset = guest_tsc - host_tsc
868 */
869 static void guest_write_tsc(u64 guest_tsc, u64 host_tsc)
870 {
871 vmcs_write64(TSC_OFFSET, guest_tsc - host_tsc);
872 }
873
874 /*
875 * Reads an msr value (of 'msr_index') into 'pdata'.
876 * Returns 0 on success, non-0 otherwise.
877 * Assumes vcpu_load() was already called.
878 */
879 static int vmx_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
880 {
881 u64 data;
882 struct kvm_msr_entry *msr;
883
884 if (!pdata) {
885 printk(KERN_ERR "BUG: get_msr called with NULL pdata\n");
886 return -EINVAL;
887 }
888
889 switch (msr_index) {
890 #ifdef CONFIG_X86_64
891 case MSR_FS_BASE:
892 data = vmcs_readl(GUEST_FS_BASE);
893 break;
894 case MSR_GS_BASE:
895 data = vmcs_readl(GUEST_GS_BASE);
896 break;
897 case MSR_EFER:
898 return kvm_get_msr_common(vcpu, msr_index, pdata);
899 #endif
900 case MSR_IA32_TIME_STAMP_COUNTER:
901 data = guest_read_tsc();
902 break;
903 case MSR_IA32_SYSENTER_CS:
904 data = vmcs_read32(GUEST_SYSENTER_CS);
905 break;
906 case MSR_IA32_SYSENTER_EIP:
907 data = vmcs_readl(GUEST_SYSENTER_EIP);
908 break;
909 case MSR_IA32_SYSENTER_ESP:
910 data = vmcs_readl(GUEST_SYSENTER_ESP);
911 break;
912 default:
913 vmx_load_host_state(to_vmx(vcpu));
914 msr = find_msr_entry(to_vmx(vcpu), msr_index);
915 if (msr) {
916 data = msr->data;
917 break;
918 }
919 return kvm_get_msr_common(vcpu, msr_index, pdata);
920 }
921
922 *pdata = data;
923 return 0;
924 }
925
926 /*
927 * Writes msr value into into the appropriate "register".
928 * Returns 0 on success, non-0 otherwise.
929 * Assumes vcpu_load() was already called.
930 */
931 static int vmx_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
932 {
933 struct vcpu_vmx *vmx = to_vmx(vcpu);
934 struct kvm_msr_entry *msr;
935 u64 host_tsc;
936 int ret = 0;
937
938 switch (msr_index) {
939 case MSR_EFER:
940 vmx_load_host_state(vmx);
941 ret = kvm_set_msr_common(vcpu, msr_index, data);
942 break;
943 #ifdef CONFIG_X86_64
944 case MSR_FS_BASE:
945 vmcs_writel(GUEST_FS_BASE, data);
946 break;
947 case MSR_GS_BASE:
948 vmcs_writel(GUEST_GS_BASE, data);
949 break;
950 #endif
951 case MSR_IA32_SYSENTER_CS:
952 vmcs_write32(GUEST_SYSENTER_CS, data);
953 break;
954 case MSR_IA32_SYSENTER_EIP:
955 vmcs_writel(GUEST_SYSENTER_EIP, data);
956 break;
957 case MSR_IA32_SYSENTER_ESP:
958 vmcs_writel(GUEST_SYSENTER_ESP, data);
959 break;
960 case MSR_IA32_TIME_STAMP_COUNTER:
961 rdtscll(host_tsc);
962 guest_write_tsc(data, host_tsc);
963 break;
964 case MSR_P6_PERFCTR0:
965 case MSR_P6_PERFCTR1:
966 case MSR_P6_EVNTSEL0:
967 case MSR_P6_EVNTSEL1:
968 /*
969 * Just discard all writes to the performance counters; this
970 * should keep both older linux and windows 64-bit guests
971 * happy
972 */
973 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: 0x%x data 0x%llx\n", msr_index, data);
974
975 break;
976 case MSR_IA32_CR_PAT:
977 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
978 vmcs_write64(GUEST_IA32_PAT, data);
979 vcpu->arch.pat = data;
980 break;
981 }
982 /* Otherwise falls through to kvm_set_msr_common */
983 default:
984 vmx_load_host_state(vmx);
985 msr = find_msr_entry(vmx, msr_index);
986 if (msr) {
987 msr->data = data;
988 break;
989 }
990 ret = kvm_set_msr_common(vcpu, msr_index, data);
991 }
992
993 return ret;
994 }
995
996 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
997 {
998 __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
999 switch (reg) {
1000 case VCPU_REGS_RSP:
1001 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
1002 break;
1003 case VCPU_REGS_RIP:
1004 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
1005 break;
1006 default:
1007 break;
1008 }
1009 }
1010
1011 static int set_guest_debug(struct kvm_vcpu *vcpu, struct kvm_guest_debug *dbg)
1012 {
1013 int old_debug = vcpu->guest_debug;
1014 unsigned long flags;
1015
1016 vcpu->guest_debug = dbg->control;
1017 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
1018 vcpu->guest_debug = 0;
1019
1020 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1021 vmcs_writel(GUEST_DR7, dbg->arch.debugreg[7]);
1022 else
1023 vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
1024
1025 flags = vmcs_readl(GUEST_RFLAGS);
1026 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
1027 flags |= X86_EFLAGS_TF | X86_EFLAGS_RF;
1028 else if (old_debug & KVM_GUESTDBG_SINGLESTEP)
1029 flags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1030 vmcs_writel(GUEST_RFLAGS, flags);
1031
1032 update_exception_bitmap(vcpu);
1033
1034 return 0;
1035 }
1036
1037 static int vmx_get_irq(struct kvm_vcpu *vcpu)
1038 {
1039 if (!vcpu->arch.interrupt.pending)
1040 return -1;
1041 return vcpu->arch.interrupt.nr;
1042 }
1043
1044 static __init int cpu_has_kvm_support(void)
1045 {
1046 return cpu_has_vmx();
1047 }
1048
1049 static __init int vmx_disabled_by_bios(void)
1050 {
1051 u64 msr;
1052
1053 rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
1054 return (msr & (FEATURE_CONTROL_LOCKED |
1055 FEATURE_CONTROL_VMXON_ENABLED))
1056 == FEATURE_CONTROL_LOCKED;
1057 /* locked but not enabled */
1058 }
1059
1060 static void hardware_enable(void *garbage)
1061 {
1062 int cpu = raw_smp_processor_id();
1063 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
1064 u64 old;
1065
1066 INIT_LIST_HEAD(&per_cpu(vcpus_on_cpu, cpu));
1067 rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
1068 if ((old & (FEATURE_CONTROL_LOCKED |
1069 FEATURE_CONTROL_VMXON_ENABLED))
1070 != (FEATURE_CONTROL_LOCKED |
1071 FEATURE_CONTROL_VMXON_ENABLED))
1072 /* enable and lock */
1073 wrmsrl(MSR_IA32_FEATURE_CONTROL, old |
1074 FEATURE_CONTROL_LOCKED |
1075 FEATURE_CONTROL_VMXON_ENABLED);
1076 write_cr4(read_cr4() | X86_CR4_VMXE); /* FIXME: not cpu hotplug safe */
1077 asm volatile (ASM_VMX_VMXON_RAX
1078 : : "a"(&phys_addr), "m"(phys_addr)
1079 : "memory", "cc");
1080 }
1081
1082 static void vmclear_local_vcpus(void)
1083 {
1084 int cpu = raw_smp_processor_id();
1085 struct vcpu_vmx *vmx, *n;
1086
1087 list_for_each_entry_safe(vmx, n, &per_cpu(vcpus_on_cpu, cpu),
1088 local_vcpus_link)
1089 __vcpu_clear(vmx);
1090 }
1091
1092
1093 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
1094 * tricks.
1095 */
1096 static void kvm_cpu_vmxoff(void)
1097 {
1098 asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
1099 write_cr4(read_cr4() & ~X86_CR4_VMXE);
1100 }
1101
1102 static void hardware_disable(void *garbage)
1103 {
1104 vmclear_local_vcpus();
1105 kvm_cpu_vmxoff();
1106 }
1107
1108 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
1109 u32 msr, u32 *result)
1110 {
1111 u32 vmx_msr_low, vmx_msr_high;
1112 u32 ctl = ctl_min | ctl_opt;
1113
1114 rdmsr(msr, vmx_msr_low, vmx_msr_high);
1115
1116 ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
1117 ctl |= vmx_msr_low; /* bit == 1 in low word ==> must be one */
1118
1119 /* Ensure minimum (required) set of control bits are supported. */
1120 if (ctl_min & ~ctl)
1121 return -EIO;
1122
1123 *result = ctl;
1124 return 0;
1125 }
1126
1127 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
1128 {
1129 u32 vmx_msr_low, vmx_msr_high;
1130 u32 min, opt, min2, opt2;
1131 u32 _pin_based_exec_control = 0;
1132 u32 _cpu_based_exec_control = 0;
1133 u32 _cpu_based_2nd_exec_control = 0;
1134 u32 _vmexit_control = 0;
1135 u32 _vmentry_control = 0;
1136
1137 min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
1138 opt = PIN_BASED_VIRTUAL_NMIS;
1139 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
1140 &_pin_based_exec_control) < 0)
1141 return -EIO;
1142
1143 min = CPU_BASED_HLT_EXITING |
1144 #ifdef CONFIG_X86_64
1145 CPU_BASED_CR8_LOAD_EXITING |
1146 CPU_BASED_CR8_STORE_EXITING |
1147 #endif
1148 CPU_BASED_CR3_LOAD_EXITING |
1149 CPU_BASED_CR3_STORE_EXITING |
1150 CPU_BASED_USE_IO_BITMAPS |
1151 CPU_BASED_MOV_DR_EXITING |
1152 CPU_BASED_USE_TSC_OFFSETING |
1153 CPU_BASED_INVLPG_EXITING;
1154 opt = CPU_BASED_TPR_SHADOW |
1155 CPU_BASED_USE_MSR_BITMAPS |
1156 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
1157 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
1158 &_cpu_based_exec_control) < 0)
1159 return -EIO;
1160 #ifdef CONFIG_X86_64
1161 if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
1162 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
1163 ~CPU_BASED_CR8_STORE_EXITING;
1164 #endif
1165 if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
1166 min2 = 0;
1167 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
1168 SECONDARY_EXEC_WBINVD_EXITING |
1169 SECONDARY_EXEC_ENABLE_VPID |
1170 SECONDARY_EXEC_ENABLE_EPT;
1171 if (adjust_vmx_controls(min2, opt2,
1172 MSR_IA32_VMX_PROCBASED_CTLS2,
1173 &_cpu_based_2nd_exec_control) < 0)
1174 return -EIO;
1175 }
1176 #ifndef CONFIG_X86_64
1177 if (!(_cpu_based_2nd_exec_control &
1178 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
1179 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
1180 #endif
1181 if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
1182 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
1183 enabled */
1184 min &= ~(CPU_BASED_CR3_LOAD_EXITING |
1185 CPU_BASED_CR3_STORE_EXITING |
1186 CPU_BASED_INVLPG_EXITING);
1187 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
1188 &_cpu_based_exec_control) < 0)
1189 return -EIO;
1190 rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
1191 vmx_capability.ept, vmx_capability.vpid);
1192 }
1193
1194 min = 0;
1195 #ifdef CONFIG_X86_64
1196 min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
1197 #endif
1198 opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT;
1199 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
1200 &_vmexit_control) < 0)
1201 return -EIO;
1202
1203 min = 0;
1204 opt = VM_ENTRY_LOAD_IA32_PAT;
1205 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
1206 &_vmentry_control) < 0)
1207 return -EIO;
1208
1209 rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
1210
1211 /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
1212 if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
1213 return -EIO;
1214
1215 #ifdef CONFIG_X86_64
1216 /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
1217 if (vmx_msr_high & (1u<<16))
1218 return -EIO;
1219 #endif
1220
1221 /* Require Write-Back (WB) memory type for VMCS accesses. */
1222 if (((vmx_msr_high >> 18) & 15) != 6)
1223 return -EIO;
1224
1225 vmcs_conf->size = vmx_msr_high & 0x1fff;
1226 vmcs_conf->order = get_order(vmcs_config.size);
1227 vmcs_conf->revision_id = vmx_msr_low;
1228
1229 vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
1230 vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
1231 vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
1232 vmcs_conf->vmexit_ctrl = _vmexit_control;
1233 vmcs_conf->vmentry_ctrl = _vmentry_control;
1234
1235 return 0;
1236 }
1237
1238 static struct vmcs *alloc_vmcs_cpu(int cpu)
1239 {
1240 int node = cpu_to_node(cpu);
1241 struct page *pages;
1242 struct vmcs *vmcs;
1243
1244 pages = alloc_pages_node(node, GFP_KERNEL, vmcs_config.order);
1245 if (!pages)
1246 return NULL;
1247 vmcs = page_address(pages);
1248 memset(vmcs, 0, vmcs_config.size);
1249 vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
1250 return vmcs;
1251 }
1252
1253 static struct vmcs *alloc_vmcs(void)
1254 {
1255 return alloc_vmcs_cpu(raw_smp_processor_id());
1256 }
1257
1258 static void free_vmcs(struct vmcs *vmcs)
1259 {
1260 free_pages((unsigned long)vmcs, vmcs_config.order);
1261 }
1262
1263 static void free_kvm_area(void)
1264 {
1265 int cpu;
1266
1267 for_each_online_cpu(cpu)
1268 free_vmcs(per_cpu(vmxarea, cpu));
1269 }
1270
1271 static __init int alloc_kvm_area(void)
1272 {
1273 int cpu;
1274
1275 for_each_online_cpu(cpu) {
1276 struct vmcs *vmcs;
1277
1278 vmcs = alloc_vmcs_cpu(cpu);
1279 if (!vmcs) {
1280 free_kvm_area();
1281 return -ENOMEM;
1282 }
1283
1284 per_cpu(vmxarea, cpu) = vmcs;
1285 }
1286 return 0;
1287 }
1288
1289 static __init int hardware_setup(void)
1290 {
1291 if (setup_vmcs_config(&vmcs_config) < 0)
1292 return -EIO;
1293
1294 if (boot_cpu_has(X86_FEATURE_NX))
1295 kvm_enable_efer_bits(EFER_NX);
1296
1297 return alloc_kvm_area();
1298 }
1299
1300 static __exit void hardware_unsetup(void)
1301 {
1302 free_kvm_area();
1303 }
1304
1305 static void fix_pmode_dataseg(int seg, struct kvm_save_segment *save)
1306 {
1307 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1308
1309 if (vmcs_readl(sf->base) == save->base && (save->base & AR_S_MASK)) {
1310 vmcs_write16(sf->selector, save->selector);
1311 vmcs_writel(sf->base, save->base);
1312 vmcs_write32(sf->limit, save->limit);
1313 vmcs_write32(sf->ar_bytes, save->ar);
1314 } else {
1315 u32 dpl = (vmcs_read16(sf->selector) & SELECTOR_RPL_MASK)
1316 << AR_DPL_SHIFT;
1317 vmcs_write32(sf->ar_bytes, 0x93 | dpl);
1318 }
1319 }
1320
1321 static void enter_pmode(struct kvm_vcpu *vcpu)
1322 {
1323 unsigned long flags;
1324 struct vcpu_vmx *vmx = to_vmx(vcpu);
1325
1326 vmx->emulation_required = 1;
1327 vcpu->arch.rmode.active = 0;
1328
1329 vmcs_writel(GUEST_TR_BASE, vcpu->arch.rmode.tr.base);
1330 vmcs_write32(GUEST_TR_LIMIT, vcpu->arch.rmode.tr.limit);
1331 vmcs_write32(GUEST_TR_AR_BYTES, vcpu->arch.rmode.tr.ar);
1332
1333 flags = vmcs_readl(GUEST_RFLAGS);
1334 flags &= ~(X86_EFLAGS_IOPL | X86_EFLAGS_VM);
1335 flags |= (vcpu->arch.rmode.save_iopl << IOPL_SHIFT);
1336 vmcs_writel(GUEST_RFLAGS, flags);
1337
1338 vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
1339 (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
1340
1341 update_exception_bitmap(vcpu);
1342
1343 if (emulate_invalid_guest_state)
1344 return;
1345
1346 fix_pmode_dataseg(VCPU_SREG_ES, &vcpu->arch.rmode.es);
1347 fix_pmode_dataseg(VCPU_SREG_DS, &vcpu->arch.rmode.ds);
1348 fix_pmode_dataseg(VCPU_SREG_GS, &vcpu->arch.rmode.gs);
1349 fix_pmode_dataseg(VCPU_SREG_FS, &vcpu->arch.rmode.fs);
1350
1351 vmcs_write16(GUEST_SS_SELECTOR, 0);
1352 vmcs_write32(GUEST_SS_AR_BYTES, 0x93);
1353
1354 vmcs_write16(GUEST_CS_SELECTOR,
1355 vmcs_read16(GUEST_CS_SELECTOR) & ~SELECTOR_RPL_MASK);
1356 vmcs_write32(GUEST_CS_AR_BYTES, 0x9b);
1357 }
1358
1359 static gva_t rmode_tss_base(struct kvm *kvm)
1360 {
1361 if (!kvm->arch.tss_addr) {
1362 gfn_t base_gfn = kvm->memslots[0].base_gfn +
1363 kvm->memslots[0].npages - 3;
1364 return base_gfn << PAGE_SHIFT;
1365 }
1366 return kvm->arch.tss_addr;
1367 }
1368
1369 static void fix_rmode_seg(int seg, struct kvm_save_segment *save)
1370 {
1371 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1372
1373 save->selector = vmcs_read16(sf->selector);
1374 save->base = vmcs_readl(sf->base);
1375 save->limit = vmcs_read32(sf->limit);
1376 save->ar = vmcs_read32(sf->ar_bytes);
1377 vmcs_write16(sf->selector, save->base >> 4);
1378 vmcs_write32(sf->base, save->base & 0xfffff);
1379 vmcs_write32(sf->limit, 0xffff);
1380 vmcs_write32(sf->ar_bytes, 0xf3);
1381 }
1382
1383 static void enter_rmode(struct kvm_vcpu *vcpu)
1384 {
1385 unsigned long flags;
1386 struct vcpu_vmx *vmx = to_vmx(vcpu);
1387
1388 vmx->emulation_required = 1;
1389 vcpu->arch.rmode.active = 1;
1390
1391 vcpu->arch.rmode.tr.base = vmcs_readl(GUEST_TR_BASE);
1392 vmcs_writel(GUEST_TR_BASE, rmode_tss_base(vcpu->kvm));
1393
1394 vcpu->arch.rmode.tr.limit = vmcs_read32(GUEST_TR_LIMIT);
1395 vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
1396
1397 vcpu->arch.rmode.tr.ar = vmcs_read32(GUEST_TR_AR_BYTES);
1398 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
1399
1400 flags = vmcs_readl(GUEST_RFLAGS);
1401 vcpu->arch.rmode.save_iopl
1402 = (flags & X86_EFLAGS_IOPL) >> IOPL_SHIFT;
1403
1404 flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1405
1406 vmcs_writel(GUEST_RFLAGS, flags);
1407 vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
1408 update_exception_bitmap(vcpu);
1409
1410 if (emulate_invalid_guest_state)
1411 goto continue_rmode;
1412
1413 vmcs_write16(GUEST_SS_SELECTOR, vmcs_readl(GUEST_SS_BASE) >> 4);
1414 vmcs_write32(GUEST_SS_LIMIT, 0xffff);
1415 vmcs_write32(GUEST_SS_AR_BYTES, 0xf3);
1416
1417 vmcs_write32(GUEST_CS_AR_BYTES, 0xf3);
1418 vmcs_write32(GUEST_CS_LIMIT, 0xffff);
1419 if (vmcs_readl(GUEST_CS_BASE) == 0xffff0000)
1420 vmcs_writel(GUEST_CS_BASE, 0xf0000);
1421 vmcs_write16(GUEST_CS_SELECTOR, vmcs_readl(GUEST_CS_BASE) >> 4);
1422
1423 fix_rmode_seg(VCPU_SREG_ES, &vcpu->arch.rmode.es);
1424 fix_rmode_seg(VCPU_SREG_DS, &vcpu->arch.rmode.ds);
1425 fix_rmode_seg(VCPU_SREG_GS, &vcpu->arch.rmode.gs);
1426 fix_rmode_seg(VCPU_SREG_FS, &vcpu->arch.rmode.fs);
1427
1428 continue_rmode:
1429 kvm_mmu_reset_context(vcpu);
1430 init_rmode(vcpu->kvm);
1431 }
1432
1433 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
1434 {
1435 struct vcpu_vmx *vmx = to_vmx(vcpu);
1436 struct kvm_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
1437
1438 vcpu->arch.shadow_efer = efer;
1439 if (!msr)
1440 return;
1441 if (efer & EFER_LMA) {
1442 vmcs_write32(VM_ENTRY_CONTROLS,
1443 vmcs_read32(VM_ENTRY_CONTROLS) |
1444 VM_ENTRY_IA32E_MODE);
1445 msr->data = efer;
1446 } else {
1447 vmcs_write32(VM_ENTRY_CONTROLS,
1448 vmcs_read32(VM_ENTRY_CONTROLS) &
1449 ~VM_ENTRY_IA32E_MODE);
1450
1451 msr->data = efer & ~EFER_LME;
1452 }
1453 setup_msrs(vmx);
1454 }
1455
1456 #ifdef CONFIG_X86_64
1457
1458 static void enter_lmode(struct kvm_vcpu *vcpu)
1459 {
1460 u32 guest_tr_ar;
1461
1462 guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
1463 if ((guest_tr_ar & AR_TYPE_MASK) != AR_TYPE_BUSY_64_TSS) {
1464 printk(KERN_DEBUG "%s: tss fixup for long mode. \n",
1465 __func__);
1466 vmcs_write32(GUEST_TR_AR_BYTES,
1467 (guest_tr_ar & ~AR_TYPE_MASK)
1468 | AR_TYPE_BUSY_64_TSS);
1469 }
1470 vcpu->arch.shadow_efer |= EFER_LMA;
1471 vmx_set_efer(vcpu, vcpu->arch.shadow_efer);
1472 }
1473
1474 static void exit_lmode(struct kvm_vcpu *vcpu)
1475 {
1476 vcpu->arch.shadow_efer &= ~EFER_LMA;
1477
1478 vmcs_write32(VM_ENTRY_CONTROLS,
1479 vmcs_read32(VM_ENTRY_CONTROLS)
1480 & ~VM_ENTRY_IA32E_MODE);
1481 }
1482
1483 #endif
1484
1485 static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
1486 {
1487 vpid_sync_vcpu_all(to_vmx(vcpu));
1488 if (vm_need_ept())
1489 ept_sync_context(construct_eptp(vcpu->arch.mmu.root_hpa));
1490 }
1491
1492 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
1493 {
1494 vcpu->arch.cr4 &= KVM_GUEST_CR4_MASK;
1495 vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & ~KVM_GUEST_CR4_MASK;
1496 }
1497
1498 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
1499 {
1500 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
1501 if (!load_pdptrs(vcpu, vcpu->arch.cr3)) {
1502 printk(KERN_ERR "EPT: Fail to load pdptrs!\n");
1503 return;
1504 }
1505 vmcs_write64(GUEST_PDPTR0, vcpu->arch.pdptrs[0]);
1506 vmcs_write64(GUEST_PDPTR1, vcpu->arch.pdptrs[1]);
1507 vmcs_write64(GUEST_PDPTR2, vcpu->arch.pdptrs[2]);
1508 vmcs_write64(GUEST_PDPTR3, vcpu->arch.pdptrs[3]);
1509 }
1510 }
1511
1512 static void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
1513
1514 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
1515 unsigned long cr0,
1516 struct kvm_vcpu *vcpu)
1517 {
1518 if (!(cr0 & X86_CR0_PG)) {
1519 /* From paging/starting to nonpaging */
1520 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
1521 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
1522 (CPU_BASED_CR3_LOAD_EXITING |
1523 CPU_BASED_CR3_STORE_EXITING));
1524 vcpu->arch.cr0 = cr0;
1525 vmx_set_cr4(vcpu, vcpu->arch.cr4);
1526 *hw_cr0 |= X86_CR0_PE | X86_CR0_PG;
1527 *hw_cr0 &= ~X86_CR0_WP;
1528 } else if (!is_paging(vcpu)) {
1529 /* From nonpaging to paging */
1530 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
1531 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
1532 ~(CPU_BASED_CR3_LOAD_EXITING |
1533 CPU_BASED_CR3_STORE_EXITING));
1534 vcpu->arch.cr0 = cr0;
1535 vmx_set_cr4(vcpu, vcpu->arch.cr4);
1536 if (!(vcpu->arch.cr0 & X86_CR0_WP))
1537 *hw_cr0 &= ~X86_CR0_WP;
1538 }
1539 }
1540
1541 static void ept_update_paging_mode_cr4(unsigned long *hw_cr4,
1542 struct kvm_vcpu *vcpu)
1543 {
1544 if (!is_paging(vcpu)) {
1545 *hw_cr4 &= ~X86_CR4_PAE;
1546 *hw_cr4 |= X86_CR4_PSE;
1547 } else if (!(vcpu->arch.cr4 & X86_CR4_PAE))
1548 *hw_cr4 &= ~X86_CR4_PAE;
1549 }
1550
1551 static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
1552 {
1553 unsigned long hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK) |
1554 KVM_VM_CR0_ALWAYS_ON;
1555
1556 vmx_fpu_deactivate(vcpu);
1557
1558 if (vcpu->arch.rmode.active && (cr0 & X86_CR0_PE))
1559 enter_pmode(vcpu);
1560
1561 if (!vcpu->arch.rmode.active && !(cr0 & X86_CR0_PE))
1562 enter_rmode(vcpu);
1563
1564 #ifdef CONFIG_X86_64
1565 if (vcpu->arch.shadow_efer & EFER_LME) {
1566 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
1567 enter_lmode(vcpu);
1568 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
1569 exit_lmode(vcpu);
1570 }
1571 #endif
1572
1573 if (vm_need_ept())
1574 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
1575
1576 vmcs_writel(CR0_READ_SHADOW, cr0);
1577 vmcs_writel(GUEST_CR0, hw_cr0);
1578 vcpu->arch.cr0 = cr0;
1579
1580 if (!(cr0 & X86_CR0_TS) || !(cr0 & X86_CR0_PE))
1581 vmx_fpu_activate(vcpu);
1582 }
1583
1584 static u64 construct_eptp(unsigned long root_hpa)
1585 {
1586 u64 eptp;
1587
1588 /* TODO write the value reading from MSR */
1589 eptp = VMX_EPT_DEFAULT_MT |
1590 VMX_EPT_DEFAULT_GAW << VMX_EPT_GAW_EPTP_SHIFT;
1591 eptp |= (root_hpa & PAGE_MASK);
1592
1593 return eptp;
1594 }
1595
1596 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1597 {
1598 unsigned long guest_cr3;
1599 u64 eptp;
1600
1601 guest_cr3 = cr3;
1602 if (vm_need_ept()) {
1603 eptp = construct_eptp(cr3);
1604 vmcs_write64(EPT_POINTER, eptp);
1605 ept_sync_context(eptp);
1606 ept_load_pdptrs(vcpu);
1607 guest_cr3 = is_paging(vcpu) ? vcpu->arch.cr3 :
1608 VMX_EPT_IDENTITY_PAGETABLE_ADDR;
1609 }
1610
1611 vmx_flush_tlb(vcpu);
1612 vmcs_writel(GUEST_CR3, guest_cr3);
1613 if (vcpu->arch.cr0 & X86_CR0_PE)
1614 vmx_fpu_deactivate(vcpu);
1615 }
1616
1617 static void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1618 {
1619 unsigned long hw_cr4 = cr4 | (vcpu->arch.rmode.active ?
1620 KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON);
1621
1622 vcpu->arch.cr4 = cr4;
1623 if (vm_need_ept())
1624 ept_update_paging_mode_cr4(&hw_cr4, vcpu);
1625
1626 vmcs_writel(CR4_READ_SHADOW, cr4);
1627 vmcs_writel(GUEST_CR4, hw_cr4);
1628 }
1629
1630 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
1631 {
1632 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1633
1634 return vmcs_readl(sf->base);
1635 }
1636
1637 static void vmx_get_segment(struct kvm_vcpu *vcpu,
1638 struct kvm_segment *var, int seg)
1639 {
1640 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1641 u32 ar;
1642
1643 var->base = vmcs_readl(sf->base);
1644 var->limit = vmcs_read32(sf->limit);
1645 var->selector = vmcs_read16(sf->selector);
1646 ar = vmcs_read32(sf->ar_bytes);
1647 if ((ar & AR_UNUSABLE_MASK) && !emulate_invalid_guest_state)
1648 ar = 0;
1649 var->type = ar & 15;
1650 var->s = (ar >> 4) & 1;
1651 var->dpl = (ar >> 5) & 3;
1652 var->present = (ar >> 7) & 1;
1653 var->avl = (ar >> 12) & 1;
1654 var->l = (ar >> 13) & 1;
1655 var->db = (ar >> 14) & 1;
1656 var->g = (ar >> 15) & 1;
1657 var->unusable = (ar >> 16) & 1;
1658 }
1659
1660 static int vmx_get_cpl(struct kvm_vcpu *vcpu)
1661 {
1662 struct kvm_segment kvm_seg;
1663
1664 if (!(vcpu->arch.cr0 & X86_CR0_PE)) /* if real mode */
1665 return 0;
1666
1667 if (vmx_get_rflags(vcpu) & X86_EFLAGS_VM) /* if virtual 8086 */
1668 return 3;
1669
1670 vmx_get_segment(vcpu, &kvm_seg, VCPU_SREG_CS);
1671 return kvm_seg.selector & 3;
1672 }
1673
1674 static u32 vmx_segment_access_rights(struct kvm_segment *var)
1675 {
1676 u32 ar;
1677
1678 if (var->unusable)
1679 ar = 1 << 16;
1680 else {
1681 ar = var->type & 15;
1682 ar |= (var->s & 1) << 4;
1683 ar |= (var->dpl & 3) << 5;
1684 ar |= (var->present & 1) << 7;
1685 ar |= (var->avl & 1) << 12;
1686 ar |= (var->l & 1) << 13;
1687 ar |= (var->db & 1) << 14;
1688 ar |= (var->g & 1) << 15;
1689 }
1690 if (ar == 0) /* a 0 value means unusable */
1691 ar = AR_UNUSABLE_MASK;
1692
1693 return ar;
1694 }
1695
1696 static void vmx_set_segment(struct kvm_vcpu *vcpu,
1697 struct kvm_segment *var, int seg)
1698 {
1699 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1700 u32 ar;
1701
1702 if (vcpu->arch.rmode.active && seg == VCPU_SREG_TR) {
1703 vcpu->arch.rmode.tr.selector = var->selector;
1704 vcpu->arch.rmode.tr.base = var->base;
1705 vcpu->arch.rmode.tr.limit = var->limit;
1706 vcpu->arch.rmode.tr.ar = vmx_segment_access_rights(var);
1707 return;
1708 }
1709 vmcs_writel(sf->base, var->base);
1710 vmcs_write32(sf->limit, var->limit);
1711 vmcs_write16(sf->selector, var->selector);
1712 if (vcpu->arch.rmode.active && var->s) {
1713 /*
1714 * Hack real-mode segments into vm86 compatibility.
1715 */
1716 if (var->base == 0xffff0000 && var->selector == 0xf000)
1717 vmcs_writel(sf->base, 0xf0000);
1718 ar = 0xf3;
1719 } else
1720 ar = vmx_segment_access_rights(var);
1721 vmcs_write32(sf->ar_bytes, ar);
1722 }
1723
1724 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
1725 {
1726 u32 ar = vmcs_read32(GUEST_CS_AR_BYTES);
1727
1728 *db = (ar >> 14) & 1;
1729 *l = (ar >> 13) & 1;
1730 }
1731
1732 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1733 {
1734 dt->limit = vmcs_read32(GUEST_IDTR_LIMIT);
1735 dt->base = vmcs_readl(GUEST_IDTR_BASE);
1736 }
1737
1738 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1739 {
1740 vmcs_write32(GUEST_IDTR_LIMIT, dt->limit);
1741 vmcs_writel(GUEST_IDTR_BASE, dt->base);
1742 }
1743
1744 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1745 {
1746 dt->limit = vmcs_read32(GUEST_GDTR_LIMIT);
1747 dt->base = vmcs_readl(GUEST_GDTR_BASE);
1748 }
1749
1750 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1751 {
1752 vmcs_write32(GUEST_GDTR_LIMIT, dt->limit);
1753 vmcs_writel(GUEST_GDTR_BASE, dt->base);
1754 }
1755
1756 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
1757 {
1758 struct kvm_segment var;
1759 u32 ar;
1760
1761 vmx_get_segment(vcpu, &var, seg);
1762 ar = vmx_segment_access_rights(&var);
1763
1764 if (var.base != (var.selector << 4))
1765 return false;
1766 if (var.limit != 0xffff)
1767 return false;
1768 if (ar != 0xf3)
1769 return false;
1770
1771 return true;
1772 }
1773
1774 static bool code_segment_valid(struct kvm_vcpu *vcpu)
1775 {
1776 struct kvm_segment cs;
1777 unsigned int cs_rpl;
1778
1779 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
1780 cs_rpl = cs.selector & SELECTOR_RPL_MASK;
1781
1782 if (cs.unusable)
1783 return false;
1784 if (~cs.type & (AR_TYPE_CODE_MASK|AR_TYPE_ACCESSES_MASK))
1785 return false;
1786 if (!cs.s)
1787 return false;
1788 if (cs.type & AR_TYPE_WRITEABLE_MASK) {
1789 if (cs.dpl > cs_rpl)
1790 return false;
1791 } else {
1792 if (cs.dpl != cs_rpl)
1793 return false;
1794 }
1795 if (!cs.present)
1796 return false;
1797
1798 /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
1799 return true;
1800 }
1801
1802 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
1803 {
1804 struct kvm_segment ss;
1805 unsigned int ss_rpl;
1806
1807 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
1808 ss_rpl = ss.selector & SELECTOR_RPL_MASK;
1809
1810 if (ss.unusable)
1811 return true;
1812 if (ss.type != 3 && ss.type != 7)
1813 return false;
1814 if (!ss.s)
1815 return false;
1816 if (ss.dpl != ss_rpl) /* DPL != RPL */
1817 return false;
1818 if (!ss.present)
1819 return false;
1820
1821 return true;
1822 }
1823
1824 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
1825 {
1826 struct kvm_segment var;
1827 unsigned int rpl;
1828
1829 vmx_get_segment(vcpu, &var, seg);
1830 rpl = var.selector & SELECTOR_RPL_MASK;
1831
1832 if (var.unusable)
1833 return true;
1834 if (!var.s)
1835 return false;
1836 if (!var.present)
1837 return false;
1838 if (~var.type & (AR_TYPE_CODE_MASK|AR_TYPE_WRITEABLE_MASK)) {
1839 if (var.dpl < rpl) /* DPL < RPL */
1840 return false;
1841 }
1842
1843 /* TODO: Add other members to kvm_segment_field to allow checking for other access
1844 * rights flags
1845 */
1846 return true;
1847 }
1848
1849 static bool tr_valid(struct kvm_vcpu *vcpu)
1850 {
1851 struct kvm_segment tr;
1852
1853 vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
1854
1855 if (tr.unusable)
1856 return false;
1857 if (tr.selector & SELECTOR_TI_MASK) /* TI = 1 */
1858 return false;
1859 if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
1860 return false;
1861 if (!tr.present)
1862 return false;
1863
1864 return true;
1865 }
1866
1867 static bool ldtr_valid(struct kvm_vcpu *vcpu)
1868 {
1869 struct kvm_segment ldtr;
1870
1871 vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
1872
1873 if (ldtr.unusable)
1874 return true;
1875 if (ldtr.selector & SELECTOR_TI_MASK) /* TI = 1 */
1876 return false;
1877 if (ldtr.type != 2)
1878 return false;
1879 if (!ldtr.present)
1880 return false;
1881
1882 return true;
1883 }
1884
1885 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
1886 {
1887 struct kvm_segment cs, ss;
1888
1889 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
1890 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
1891
1892 return ((cs.selector & SELECTOR_RPL_MASK) ==
1893 (ss.selector & SELECTOR_RPL_MASK));
1894 }
1895
1896 /*
1897 * Check if guest state is valid. Returns true if valid, false if
1898 * not.
1899 * We assume that registers are always usable
1900 */
1901 static bool guest_state_valid(struct kvm_vcpu *vcpu)
1902 {
1903 /* real mode guest state checks */
1904 if (!(vcpu->arch.cr0 & X86_CR0_PE)) {
1905 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
1906 return false;
1907 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
1908 return false;
1909 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
1910 return false;
1911 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
1912 return false;
1913 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
1914 return false;
1915 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
1916 return false;
1917 } else {
1918 /* protected mode guest state checks */
1919 if (!cs_ss_rpl_check(vcpu))
1920 return false;
1921 if (!code_segment_valid(vcpu))
1922 return false;
1923 if (!stack_segment_valid(vcpu))
1924 return false;
1925 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
1926 return false;
1927 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
1928 return false;
1929 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
1930 return false;
1931 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
1932 return false;
1933 if (!tr_valid(vcpu))
1934 return false;
1935 if (!ldtr_valid(vcpu))
1936 return false;
1937 }
1938 /* TODO:
1939 * - Add checks on RIP
1940 * - Add checks on RFLAGS
1941 */
1942
1943 return true;
1944 }
1945
1946 static int init_rmode_tss(struct kvm *kvm)
1947 {
1948 gfn_t fn = rmode_tss_base(kvm) >> PAGE_SHIFT;
1949 u16 data = 0;
1950 int ret = 0;
1951 int r;
1952
1953 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
1954 if (r < 0)
1955 goto out;
1956 data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
1957 r = kvm_write_guest_page(kvm, fn++, &data,
1958 TSS_IOPB_BASE_OFFSET, sizeof(u16));
1959 if (r < 0)
1960 goto out;
1961 r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
1962 if (r < 0)
1963 goto out;
1964 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
1965 if (r < 0)
1966 goto out;
1967 data = ~0;
1968 r = kvm_write_guest_page(kvm, fn, &data,
1969 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
1970 sizeof(u8));
1971 if (r < 0)
1972 goto out;
1973
1974 ret = 1;
1975 out:
1976 return ret;
1977 }
1978
1979 static int init_rmode_identity_map(struct kvm *kvm)
1980 {
1981 int i, r, ret;
1982 pfn_t identity_map_pfn;
1983 u32 tmp;
1984
1985 if (!vm_need_ept())
1986 return 1;
1987 if (unlikely(!kvm->arch.ept_identity_pagetable)) {
1988 printk(KERN_ERR "EPT: identity-mapping pagetable "
1989 "haven't been allocated!\n");
1990 return 0;
1991 }
1992 if (likely(kvm->arch.ept_identity_pagetable_done))
1993 return 1;
1994 ret = 0;
1995 identity_map_pfn = VMX_EPT_IDENTITY_PAGETABLE_ADDR >> PAGE_SHIFT;
1996 r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
1997 if (r < 0)
1998 goto out;
1999 /* Set up identity-mapping pagetable for EPT in real mode */
2000 for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
2001 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
2002 _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
2003 r = kvm_write_guest_page(kvm, identity_map_pfn,
2004 &tmp, i * sizeof(tmp), sizeof(tmp));
2005 if (r < 0)
2006 goto out;
2007 }
2008 kvm->arch.ept_identity_pagetable_done = true;
2009 ret = 1;
2010 out:
2011 return ret;
2012 }
2013
2014 static void seg_setup(int seg)
2015 {
2016 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2017
2018 vmcs_write16(sf->selector, 0);
2019 vmcs_writel(sf->base, 0);
2020 vmcs_write32(sf->limit, 0xffff);
2021 vmcs_write32(sf->ar_bytes, 0xf3);
2022 }
2023
2024 static int alloc_apic_access_page(struct kvm *kvm)
2025 {
2026 struct kvm_userspace_memory_region kvm_userspace_mem;
2027 int r = 0;
2028
2029 down_write(&kvm->slots_lock);
2030 if (kvm->arch.apic_access_page)
2031 goto out;
2032 kvm_userspace_mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
2033 kvm_userspace_mem.flags = 0;
2034 kvm_userspace_mem.guest_phys_addr = 0xfee00000ULL;
2035 kvm_userspace_mem.memory_size = PAGE_SIZE;
2036 r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
2037 if (r)
2038 goto out;
2039
2040 kvm->arch.apic_access_page = gfn_to_page(kvm, 0xfee00);
2041 out:
2042 up_write(&kvm->slots_lock);
2043 return r;
2044 }
2045
2046 static int alloc_identity_pagetable(struct kvm *kvm)
2047 {
2048 struct kvm_userspace_memory_region kvm_userspace_mem;
2049 int r = 0;
2050
2051 down_write(&kvm->slots_lock);
2052 if (kvm->arch.ept_identity_pagetable)
2053 goto out;
2054 kvm_userspace_mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
2055 kvm_userspace_mem.flags = 0;
2056 kvm_userspace_mem.guest_phys_addr = VMX_EPT_IDENTITY_PAGETABLE_ADDR;
2057 kvm_userspace_mem.memory_size = PAGE_SIZE;
2058 r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
2059 if (r)
2060 goto out;
2061
2062 kvm->arch.ept_identity_pagetable = gfn_to_page(kvm,
2063 VMX_EPT_IDENTITY_PAGETABLE_ADDR >> PAGE_SHIFT);
2064 out:
2065 up_write(&kvm->slots_lock);
2066 return r;
2067 }
2068
2069 static void allocate_vpid(struct vcpu_vmx *vmx)
2070 {
2071 int vpid;
2072
2073 vmx->vpid = 0;
2074 if (!enable_vpid || !cpu_has_vmx_vpid())
2075 return;
2076 spin_lock(&vmx_vpid_lock);
2077 vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
2078 if (vpid < VMX_NR_VPIDS) {
2079 vmx->vpid = vpid;
2080 __set_bit(vpid, vmx_vpid_bitmap);
2081 }
2082 spin_unlock(&vmx_vpid_lock);
2083 }
2084
2085 static void vmx_disable_intercept_for_msr(struct page *msr_bitmap, u32 msr)
2086 {
2087 void *va;
2088
2089 if (!cpu_has_vmx_msr_bitmap())
2090 return;
2091
2092 /*
2093 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
2094 * have the write-low and read-high bitmap offsets the wrong way round.
2095 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
2096 */
2097 va = kmap(msr_bitmap);
2098 if (msr <= 0x1fff) {
2099 __clear_bit(msr, va + 0x000); /* read-low */
2100 __clear_bit(msr, va + 0x800); /* write-low */
2101 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
2102 msr &= 0x1fff;
2103 __clear_bit(msr, va + 0x400); /* read-high */
2104 __clear_bit(msr, va + 0xc00); /* write-high */
2105 }
2106 kunmap(msr_bitmap);
2107 }
2108
2109 /*
2110 * Sets up the vmcs for emulated real mode.
2111 */
2112 static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
2113 {
2114 u32 host_sysenter_cs, msr_low, msr_high;
2115 u32 junk;
2116 u64 host_pat, tsc_this, tsc_base;
2117 unsigned long a;
2118 struct descriptor_table dt;
2119 int i;
2120 unsigned long kvm_vmx_return;
2121 u32 exec_control;
2122
2123 /* I/O */
2124 vmcs_write64(IO_BITMAP_A, page_to_phys(vmx_io_bitmap_a));
2125 vmcs_write64(IO_BITMAP_B, page_to_phys(vmx_io_bitmap_b));
2126
2127 if (cpu_has_vmx_msr_bitmap())
2128 vmcs_write64(MSR_BITMAP, page_to_phys(vmx_msr_bitmap));
2129
2130 vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
2131
2132 /* Control */
2133 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL,
2134 vmcs_config.pin_based_exec_ctrl);
2135
2136 exec_control = vmcs_config.cpu_based_exec_ctrl;
2137 if (!vm_need_tpr_shadow(vmx->vcpu.kvm)) {
2138 exec_control &= ~CPU_BASED_TPR_SHADOW;
2139 #ifdef CONFIG_X86_64
2140 exec_control |= CPU_BASED_CR8_STORE_EXITING |
2141 CPU_BASED_CR8_LOAD_EXITING;
2142 #endif
2143 }
2144 if (!vm_need_ept())
2145 exec_control |= CPU_BASED_CR3_STORE_EXITING |
2146 CPU_BASED_CR3_LOAD_EXITING |
2147 CPU_BASED_INVLPG_EXITING;
2148 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
2149
2150 if (cpu_has_secondary_exec_ctrls()) {
2151 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
2152 if (!vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
2153 exec_control &=
2154 ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
2155 if (vmx->vpid == 0)
2156 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
2157 if (!vm_need_ept())
2158 exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
2159 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
2160 }
2161
2162 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, !!bypass_guest_pf);
2163 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, !!bypass_guest_pf);
2164 vmcs_write32(CR3_TARGET_COUNT, 0); /* 22.2.1 */
2165
2166 vmcs_writel(HOST_CR0, read_cr0()); /* 22.2.3 */
2167 vmcs_writel(HOST_CR4, read_cr4()); /* 22.2.3, 22.2.5 */
2168 vmcs_writel(HOST_CR3, read_cr3()); /* 22.2.3 FIXME: shadow tables */
2169
2170 vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */
2171 vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
2172 vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS); /* 22.2.4 */
2173 vmcs_write16(HOST_FS_SELECTOR, kvm_read_fs()); /* 22.2.4 */
2174 vmcs_write16(HOST_GS_SELECTOR, kvm_read_gs()); /* 22.2.4 */
2175 vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
2176 #ifdef CONFIG_X86_64
2177 rdmsrl(MSR_FS_BASE, a);
2178 vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
2179 rdmsrl(MSR_GS_BASE, a);
2180 vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
2181 #else
2182 vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
2183 vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
2184 #endif
2185
2186 vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8); /* 22.2.4 */
2187
2188 kvm_get_idt(&dt);
2189 vmcs_writel(HOST_IDTR_BASE, dt.base); /* 22.2.4 */
2190
2191 asm("mov $.Lkvm_vmx_return, %0" : "=r"(kvm_vmx_return));
2192 vmcs_writel(HOST_RIP, kvm_vmx_return); /* 22.2.5 */
2193 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
2194 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
2195 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
2196
2197 rdmsr(MSR_IA32_SYSENTER_CS, host_sysenter_cs, junk);
2198 vmcs_write32(HOST_IA32_SYSENTER_CS, host_sysenter_cs);
2199 rdmsrl(MSR_IA32_SYSENTER_ESP, a);
2200 vmcs_writel(HOST_IA32_SYSENTER_ESP, a); /* 22.2.3 */
2201 rdmsrl(MSR_IA32_SYSENTER_EIP, a);
2202 vmcs_writel(HOST_IA32_SYSENTER_EIP, a); /* 22.2.3 */
2203
2204 if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
2205 rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high);
2206 host_pat = msr_low | ((u64) msr_high << 32);
2207 vmcs_write64(HOST_IA32_PAT, host_pat);
2208 }
2209 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2210 rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high);
2211 host_pat = msr_low | ((u64) msr_high << 32);
2212 /* Write the default value follow host pat */
2213 vmcs_write64(GUEST_IA32_PAT, host_pat);
2214 /* Keep arch.pat sync with GUEST_IA32_PAT */
2215 vmx->vcpu.arch.pat = host_pat;
2216 }
2217
2218 for (i = 0; i < NR_VMX_MSR; ++i) {
2219 u32 index = vmx_msr_index[i];
2220 u32 data_low, data_high;
2221 u64 data;
2222 int j = vmx->nmsrs;
2223
2224 if (rdmsr_safe(index, &data_low, &data_high) < 0)
2225 continue;
2226 if (wrmsr_safe(index, data_low, data_high) < 0)
2227 continue;
2228 data = data_low | ((u64)data_high << 32);
2229 vmx->host_msrs[j].index = index;
2230 vmx->host_msrs[j].reserved = 0;
2231 vmx->host_msrs[j].data = data;
2232 vmx->guest_msrs[j] = vmx->host_msrs[j];
2233 ++vmx->nmsrs;
2234 }
2235
2236 vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
2237
2238 /* 22.2.1, 20.8.1 */
2239 vmcs_write32(VM_ENTRY_CONTROLS, vmcs_config.vmentry_ctrl);
2240
2241 vmcs_writel(CR0_GUEST_HOST_MASK, ~0UL);
2242 vmcs_writel(CR4_GUEST_HOST_MASK, KVM_GUEST_CR4_MASK);
2243
2244 tsc_base = vmx->vcpu.kvm->arch.vm_init_tsc;
2245 rdtscll(tsc_this);
2246 if (tsc_this < vmx->vcpu.kvm->arch.vm_init_tsc)
2247 tsc_base = tsc_this;
2248
2249 guest_write_tsc(0, tsc_base);
2250
2251 return 0;
2252 }
2253
2254 static int init_rmode(struct kvm *kvm)
2255 {
2256 if (!init_rmode_tss(kvm))
2257 return 0;
2258 if (!init_rmode_identity_map(kvm))
2259 return 0;
2260 return 1;
2261 }
2262
2263 static int vmx_vcpu_reset(struct kvm_vcpu *vcpu)
2264 {
2265 struct vcpu_vmx *vmx = to_vmx(vcpu);
2266 u64 msr;
2267 int ret;
2268
2269 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP));
2270 down_read(&vcpu->kvm->slots_lock);
2271 if (!init_rmode(vmx->vcpu.kvm)) {
2272 ret = -ENOMEM;
2273 goto out;
2274 }
2275
2276 vmx->vcpu.arch.rmode.active = 0;
2277
2278 vmx->soft_vnmi_blocked = 0;
2279
2280 vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
2281 kvm_set_cr8(&vmx->vcpu, 0);
2282 msr = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
2283 if (vmx->vcpu.vcpu_id == 0)
2284 msr |= MSR_IA32_APICBASE_BSP;
2285 kvm_set_apic_base(&vmx->vcpu, msr);
2286
2287 fx_init(&vmx->vcpu);
2288
2289 seg_setup(VCPU_SREG_CS);
2290 /*
2291 * GUEST_CS_BASE should really be 0xffff0000, but VT vm86 mode
2292 * insists on having GUEST_CS_BASE == GUEST_CS_SELECTOR << 4. Sigh.
2293 */
2294 if (vmx->vcpu.vcpu_id == 0) {
2295 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
2296 vmcs_writel(GUEST_CS_BASE, 0x000f0000);
2297 } else {
2298 vmcs_write16(GUEST_CS_SELECTOR, vmx->vcpu.arch.sipi_vector << 8);
2299 vmcs_writel(GUEST_CS_BASE, vmx->vcpu.arch.sipi_vector << 12);
2300 }
2301
2302 seg_setup(VCPU_SREG_DS);
2303 seg_setup(VCPU_SREG_ES);
2304 seg_setup(VCPU_SREG_FS);
2305 seg_setup(VCPU_SREG_GS);
2306 seg_setup(VCPU_SREG_SS);
2307
2308 vmcs_write16(GUEST_TR_SELECTOR, 0);
2309 vmcs_writel(GUEST_TR_BASE, 0);
2310 vmcs_write32(GUEST_TR_LIMIT, 0xffff);
2311 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
2312
2313 vmcs_write16(GUEST_LDTR_SELECTOR, 0);
2314 vmcs_writel(GUEST_LDTR_BASE, 0);
2315 vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
2316 vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
2317
2318 vmcs_write32(GUEST_SYSENTER_CS, 0);
2319 vmcs_writel(GUEST_SYSENTER_ESP, 0);
2320 vmcs_writel(GUEST_SYSENTER_EIP, 0);
2321
2322 vmcs_writel(GUEST_RFLAGS, 0x02);
2323 if (vmx->vcpu.vcpu_id == 0)
2324 kvm_rip_write(vcpu, 0xfff0);
2325 else
2326 kvm_rip_write(vcpu, 0);
2327 kvm_register_write(vcpu, VCPU_REGS_RSP, 0);
2328
2329 vmcs_writel(GUEST_DR7, 0x400);
2330
2331 vmcs_writel(GUEST_GDTR_BASE, 0);
2332 vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
2333
2334 vmcs_writel(GUEST_IDTR_BASE, 0);
2335 vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
2336
2337 vmcs_write32(GUEST_ACTIVITY_STATE, 0);
2338 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
2339 vmcs_write32(GUEST_PENDING_DBG_EXCEPTIONS, 0);
2340
2341 /* Special registers */
2342 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
2343
2344 setup_msrs(vmx);
2345
2346 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */
2347
2348 if (cpu_has_vmx_tpr_shadow()) {
2349 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
2350 if (vm_need_tpr_shadow(vmx->vcpu.kvm))
2351 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
2352 page_to_phys(vmx->vcpu.arch.apic->regs_page));
2353 vmcs_write32(TPR_THRESHOLD, 0);
2354 }
2355
2356 if (vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
2357 vmcs_write64(APIC_ACCESS_ADDR,
2358 page_to_phys(vmx->vcpu.kvm->arch.apic_access_page));
2359
2360 if (vmx->vpid != 0)
2361 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
2362
2363 vmx->vcpu.arch.cr0 = 0x60000010;
2364 vmx_set_cr0(&vmx->vcpu, vmx->vcpu.arch.cr0); /* enter rmode */
2365 vmx_set_cr4(&vmx->vcpu, 0);
2366 vmx_set_efer(&vmx->vcpu, 0);
2367 vmx_fpu_activate(&vmx->vcpu);
2368 update_exception_bitmap(&vmx->vcpu);
2369
2370 vpid_sync_vcpu_all(vmx);
2371
2372 ret = 0;
2373
2374 /* HACK: Don't enable emulation on guest boot/reset */
2375 vmx->emulation_required = 0;
2376
2377 out:
2378 up_read(&vcpu->kvm->slots_lock);
2379 return ret;
2380 }
2381
2382 static void enable_irq_window(struct kvm_vcpu *vcpu)
2383 {
2384 u32 cpu_based_vm_exec_control;
2385
2386 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2387 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
2388 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2389 }
2390
2391 static void enable_nmi_window(struct kvm_vcpu *vcpu)
2392 {
2393 u32 cpu_based_vm_exec_control;
2394
2395 if (!cpu_has_virtual_nmis()) {
2396 enable_irq_window(vcpu);
2397 return;
2398 }
2399
2400 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2401 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_NMI_PENDING;
2402 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2403 }
2404
2405 static void vmx_inject_irq(struct kvm_vcpu *vcpu, int irq)
2406 {
2407 struct vcpu_vmx *vmx = to_vmx(vcpu);
2408
2409 KVMTRACE_1D(INJ_VIRQ, vcpu, (u32)irq, handler);
2410
2411 ++vcpu->stat.irq_injections;
2412 if (vcpu->arch.rmode.active) {
2413 vmx->rmode.irq.pending = true;
2414 vmx->rmode.irq.vector = irq;
2415 vmx->rmode.irq.rip = kvm_rip_read(vcpu);
2416 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2417 irq | INTR_TYPE_SOFT_INTR | INTR_INFO_VALID_MASK);
2418 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, 1);
2419 kvm_rip_write(vcpu, vmx->rmode.irq.rip - 1);
2420 return;
2421 }
2422 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2423 irq | INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
2424 }
2425
2426 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
2427 {
2428 struct vcpu_vmx *vmx = to_vmx(vcpu);
2429
2430 if (!cpu_has_virtual_nmis()) {
2431 /*
2432 * Tracking the NMI-blocked state in software is built upon
2433 * finding the next open IRQ window. This, in turn, depends on
2434 * well-behaving guests: They have to keep IRQs disabled at
2435 * least as long as the NMI handler runs. Otherwise we may
2436 * cause NMI nesting, maybe breaking the guest. But as this is
2437 * highly unlikely, we can live with the residual risk.
2438 */
2439 vmx->soft_vnmi_blocked = 1;
2440 vmx->vnmi_blocked_time = 0;
2441 }
2442
2443 ++vcpu->stat.nmi_injections;
2444 if (vcpu->arch.rmode.active) {
2445 vmx->rmode.irq.pending = true;
2446 vmx->rmode.irq.vector = NMI_VECTOR;
2447 vmx->rmode.irq.rip = kvm_rip_read(vcpu);
2448 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2449 NMI_VECTOR | INTR_TYPE_SOFT_INTR |
2450 INTR_INFO_VALID_MASK);
2451 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, 1);
2452 kvm_rip_write(vcpu, vmx->rmode.irq.rip - 1);
2453 return;
2454 }
2455 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2456 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
2457 }
2458
2459 static void vmx_update_window_states(struct kvm_vcpu *vcpu)
2460 {
2461 u32 guest_intr = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
2462
2463 vcpu->arch.nmi_window_open =
2464 !(guest_intr & (GUEST_INTR_STATE_STI |
2465 GUEST_INTR_STATE_MOV_SS |
2466 GUEST_INTR_STATE_NMI));
2467 if (!cpu_has_virtual_nmis() && to_vmx(vcpu)->soft_vnmi_blocked)
2468 vcpu->arch.nmi_window_open = 0;
2469
2470 vcpu->arch.interrupt_window_open =
2471 ((vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
2472 !(guest_intr & (GUEST_INTR_STATE_STI |
2473 GUEST_INTR_STATE_MOV_SS)));
2474 }
2475
2476 static void kvm_do_inject_irq(struct kvm_vcpu *vcpu)
2477 {
2478 int word_index = __ffs(vcpu->arch.irq_summary);
2479 int bit_index = __ffs(vcpu->arch.irq_pending[word_index]);
2480 int irq = word_index * BITS_PER_LONG + bit_index;
2481
2482 clear_bit(bit_index, &vcpu->arch.irq_pending[word_index]);
2483 if (!vcpu->arch.irq_pending[word_index])
2484 clear_bit(word_index, &vcpu->arch.irq_summary);
2485 kvm_queue_interrupt(vcpu, irq);
2486 }
2487
2488 static void do_interrupt_requests(struct kvm_vcpu *vcpu,
2489 struct kvm_run *kvm_run)
2490 {
2491 vmx_update_window_states(vcpu);
2492
2493 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
2494 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
2495 GUEST_INTR_STATE_STI |
2496 GUEST_INTR_STATE_MOV_SS);
2497
2498 if (vcpu->arch.nmi_pending && !vcpu->arch.nmi_injected) {
2499 if (vcpu->arch.interrupt.pending) {
2500 enable_nmi_window(vcpu);
2501 } else if (vcpu->arch.nmi_window_open) {
2502 vcpu->arch.nmi_pending = false;
2503 vcpu->arch.nmi_injected = true;
2504 } else {
2505 enable_nmi_window(vcpu);
2506 return;
2507 }
2508 }
2509 if (vcpu->arch.nmi_injected) {
2510 vmx_inject_nmi(vcpu);
2511 if (vcpu->arch.nmi_pending)
2512 enable_nmi_window(vcpu);
2513 else if (vcpu->arch.irq_summary
2514 || kvm_run->request_interrupt_window)
2515 enable_irq_window(vcpu);
2516 return;
2517 }
2518
2519 if (vcpu->arch.interrupt_window_open) {
2520 if (vcpu->arch.irq_summary && !vcpu->arch.interrupt.pending)
2521 kvm_do_inject_irq(vcpu);
2522
2523 if (vcpu->arch.interrupt.pending)
2524 vmx_inject_irq(vcpu, vcpu->arch.interrupt.nr);
2525 }
2526 if (!vcpu->arch.interrupt_window_open &&
2527 (vcpu->arch.irq_summary || kvm_run->request_interrupt_window))
2528 enable_irq_window(vcpu);
2529 }
2530
2531 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
2532 {
2533 int ret;
2534 struct kvm_userspace_memory_region tss_mem = {
2535 .slot = TSS_PRIVATE_MEMSLOT,
2536 .guest_phys_addr = addr,
2537 .memory_size = PAGE_SIZE * 3,
2538 .flags = 0,
2539 };
2540
2541 ret = kvm_set_memory_region(kvm, &tss_mem, 0);
2542 if (ret)
2543 return ret;
2544 kvm->arch.tss_addr = addr;
2545 return 0;
2546 }
2547
2548 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
2549 int vec, u32 err_code)
2550 {
2551 /*
2552 * Instruction with address size override prefix opcode 0x67
2553 * Cause the #SS fault with 0 error code in VM86 mode.
2554 */
2555 if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0)
2556 if (emulate_instruction(vcpu, NULL, 0, 0, 0) == EMULATE_DONE)
2557 return 1;
2558 /*
2559 * Forward all other exceptions that are valid in real mode.
2560 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
2561 * the required debugging infrastructure rework.
2562 */
2563 switch (vec) {
2564 case DB_VECTOR:
2565 if (vcpu->guest_debug &
2566 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
2567 return 0;
2568 kvm_queue_exception(vcpu, vec);
2569 return 1;
2570 case BP_VECTOR:
2571 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
2572 return 0;
2573 /* fall through */
2574 case DE_VECTOR:
2575 case OF_VECTOR:
2576 case BR_VECTOR:
2577 case UD_VECTOR:
2578 case DF_VECTOR:
2579 case SS_VECTOR:
2580 case GP_VECTOR:
2581 case MF_VECTOR:
2582 kvm_queue_exception(vcpu, vec);
2583 return 1;
2584 }
2585 return 0;
2586 }
2587
2588 static int handle_exception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2589 {
2590 struct vcpu_vmx *vmx = to_vmx(vcpu);
2591 u32 intr_info, ex_no, error_code;
2592 unsigned long cr2, rip, dr6;
2593 u32 vect_info;
2594 enum emulation_result er;
2595
2596 vect_info = vmx->idt_vectoring_info;
2597 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
2598
2599 if ((vect_info & VECTORING_INFO_VALID_MASK) &&
2600 !is_page_fault(intr_info))
2601 printk(KERN_ERR "%s: unexpected, vectoring info 0x%x "
2602 "intr info 0x%x\n", __func__, vect_info, intr_info);
2603
2604 if (!irqchip_in_kernel(vcpu->kvm) && is_external_interrupt(vect_info)) {
2605 int irq = vect_info & VECTORING_INFO_VECTOR_MASK;
2606 set_bit(irq, vcpu->arch.irq_pending);
2607 set_bit(irq / BITS_PER_LONG, &vcpu->arch.irq_summary);
2608 }
2609
2610 if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR)
2611 return 1; /* already handled by vmx_vcpu_run() */
2612
2613 if (is_no_device(intr_info)) {
2614 vmx_fpu_activate(vcpu);
2615 return 1;
2616 }
2617
2618 if (is_invalid_opcode(intr_info)) {
2619 er = emulate_instruction(vcpu, kvm_run, 0, 0, EMULTYPE_TRAP_UD);
2620 if (er != EMULATE_DONE)
2621 kvm_queue_exception(vcpu, UD_VECTOR);
2622 return 1;
2623 }
2624
2625 error_code = 0;
2626 rip = kvm_rip_read(vcpu);
2627 if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
2628 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
2629 if (is_page_fault(intr_info)) {
2630 /* EPT won't cause page fault directly */
2631 if (vm_need_ept())
2632 BUG();
2633 cr2 = vmcs_readl(EXIT_QUALIFICATION);
2634 KVMTRACE_3D(PAGE_FAULT, vcpu, error_code, (u32)cr2,
2635 (u32)((u64)cr2 >> 32), handler);
2636 if (vcpu->arch.interrupt.pending || vcpu->arch.exception.pending)
2637 kvm_mmu_unprotect_page_virt(vcpu, cr2);
2638 return kvm_mmu_page_fault(vcpu, cr2, error_code);
2639 }
2640
2641 if (vcpu->arch.rmode.active &&
2642 handle_rmode_exception(vcpu, intr_info & INTR_INFO_VECTOR_MASK,
2643 error_code)) {
2644 if (vcpu->arch.halt_request) {
2645 vcpu->arch.halt_request = 0;
2646 return kvm_emulate_halt(vcpu);
2647 }
2648 return 1;
2649 }
2650
2651 ex_no = intr_info & INTR_INFO_VECTOR_MASK;
2652 switch (ex_no) {
2653 case DB_VECTOR:
2654 dr6 = vmcs_readl(EXIT_QUALIFICATION);
2655 if (!(vcpu->guest_debug &
2656 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
2657 vcpu->arch.dr6 = dr6 | DR6_FIXED_1;
2658 kvm_queue_exception(vcpu, DB_VECTOR);
2659 return 1;
2660 }
2661 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
2662 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
2663 /* fall through */
2664 case BP_VECTOR:
2665 kvm_run->exit_reason = KVM_EXIT_DEBUG;
2666 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
2667 kvm_run->debug.arch.exception = ex_no;
2668 break;
2669 default:
2670 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
2671 kvm_run->ex.exception = ex_no;
2672 kvm_run->ex.error_code = error_code;
2673 break;
2674 }
2675 return 0;
2676 }
2677
2678 static int handle_external_interrupt(struct kvm_vcpu *vcpu,
2679 struct kvm_run *kvm_run)
2680 {
2681 ++vcpu->stat.irq_exits;
2682 KVMTRACE_1D(INTR, vcpu, vmcs_read32(VM_EXIT_INTR_INFO), handler);
2683 return 1;
2684 }
2685
2686 static int handle_triple_fault(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2687 {
2688 kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
2689 return 0;
2690 }
2691
2692 static int handle_io(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2693 {
2694 unsigned long exit_qualification;
2695 int size, in, string;
2696 unsigned port;
2697
2698 ++vcpu->stat.io_exits;
2699 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
2700 string = (exit_qualification & 16) != 0;
2701
2702 if (string) {
2703 if (emulate_instruction(vcpu,
2704 kvm_run, 0, 0, 0) == EMULATE_DO_MMIO)
2705 return 0;
2706 return 1;
2707 }
2708
2709 size = (exit_qualification & 7) + 1;
2710 in = (exit_qualification & 8) != 0;
2711 port = exit_qualification >> 16;
2712
2713 skip_emulated_instruction(vcpu);
2714 return kvm_emulate_pio(vcpu, kvm_run, in, size, port);
2715 }
2716
2717 static void
2718 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
2719 {
2720 /*
2721 * Patch in the VMCALL instruction:
2722 */
2723 hypercall[0] = 0x0f;
2724 hypercall[1] = 0x01;
2725 hypercall[2] = 0xc1;
2726 }
2727
2728 static int handle_cr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2729 {
2730 unsigned long exit_qualification;
2731 int cr;
2732 int reg;
2733
2734 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
2735 cr = exit_qualification & 15;
2736 reg = (exit_qualification >> 8) & 15;
2737 switch ((exit_qualification >> 4) & 3) {
2738 case 0: /* mov to cr */
2739 KVMTRACE_3D(CR_WRITE, vcpu, (u32)cr,
2740 (u32)kvm_register_read(vcpu, reg),
2741 (u32)((u64)kvm_register_read(vcpu, reg) >> 32),
2742 handler);
2743 switch (cr) {
2744 case 0:
2745 kvm_set_cr0(vcpu, kvm_register_read(vcpu, reg));
2746 skip_emulated_instruction(vcpu);
2747 return 1;
2748 case 3:
2749 kvm_set_cr3(vcpu, kvm_register_read(vcpu, reg));
2750 skip_emulated_instruction(vcpu);
2751 return 1;
2752 case 4:
2753 kvm_set_cr4(vcpu, kvm_register_read(vcpu, reg));
2754 skip_emulated_instruction(vcpu);
2755 return 1;
2756 case 8:
2757 kvm_set_cr8(vcpu, kvm_register_read(vcpu, reg));
2758 skip_emulated_instruction(vcpu);
2759 if (irqchip_in_kernel(vcpu->kvm))
2760 return 1;
2761 kvm_run->exit_reason = KVM_EXIT_SET_TPR;
2762 return 0;
2763 };
2764 break;
2765 case 2: /* clts */
2766 vmx_fpu_deactivate(vcpu);
2767 vcpu->arch.cr0 &= ~X86_CR0_TS;
2768 vmcs_writel(CR0_READ_SHADOW, vcpu->arch.cr0);
2769 vmx_fpu_activate(vcpu);
2770 KVMTRACE_0D(CLTS, vcpu, handler);
2771 skip_emulated_instruction(vcpu);
2772 return 1;
2773 case 1: /*mov from cr*/
2774 switch (cr) {
2775 case 3:
2776 kvm_register_write(vcpu, reg, vcpu->arch.cr3);
2777 KVMTRACE_3D(CR_READ, vcpu, (u32)cr,
2778 (u32)kvm_register_read(vcpu, reg),
2779 (u32)((u64)kvm_register_read(vcpu, reg) >> 32),
2780 handler);
2781 skip_emulated_instruction(vcpu);
2782 return 1;
2783 case 8:
2784 kvm_register_write(vcpu, reg, kvm_get_cr8(vcpu));
2785 KVMTRACE_2D(CR_READ, vcpu, (u32)cr,
2786 (u32)kvm_register_read(vcpu, reg), handler);
2787 skip_emulated_instruction(vcpu);
2788 return 1;
2789 }
2790 break;
2791 case 3: /* lmsw */
2792 kvm_lmsw(vcpu, (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f);
2793
2794 skip_emulated_instruction(vcpu);
2795 return 1;
2796 default:
2797 break;
2798 }
2799 kvm_run->exit_reason = 0;
2800 pr_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
2801 (int)(exit_qualification >> 4) & 3, cr);
2802 return 0;
2803 }
2804
2805 static int handle_dr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2806 {
2807 unsigned long exit_qualification;
2808 unsigned long val;
2809 int dr, reg;
2810
2811 dr = vmcs_readl(GUEST_DR7);
2812 if (dr & DR7_GD) {
2813 /*
2814 * As the vm-exit takes precedence over the debug trap, we
2815 * need to emulate the latter, either for the host or the
2816 * guest debugging itself.
2817 */
2818 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
2819 kvm_run->debug.arch.dr6 = vcpu->arch.dr6;
2820 kvm_run->debug.arch.dr7 = dr;
2821 kvm_run->debug.arch.pc =
2822 vmcs_readl(GUEST_CS_BASE) +
2823 vmcs_readl(GUEST_RIP);
2824 kvm_run->debug.arch.exception = DB_VECTOR;
2825 kvm_run->exit_reason = KVM_EXIT_DEBUG;
2826 return 0;
2827 } else {
2828 vcpu->arch.dr7 &= ~DR7_GD;
2829 vcpu->arch.dr6 |= DR6_BD;
2830 vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
2831 kvm_queue_exception(vcpu, DB_VECTOR);
2832 return 1;
2833 }
2834 }
2835
2836 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
2837 dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
2838 reg = DEBUG_REG_ACCESS_REG(exit_qualification);
2839 if (exit_qualification & TYPE_MOV_FROM_DR) {
2840 switch (dr) {
2841 case 0 ... 3:
2842 val = vcpu->arch.db[dr];
2843 break;
2844 case 6:
2845 val = vcpu->arch.dr6;
2846 break;
2847 case 7:
2848 val = vcpu->arch.dr7;
2849 break;
2850 default:
2851 val = 0;
2852 }
2853 kvm_register_write(vcpu, reg, val);
2854 KVMTRACE_2D(DR_READ, vcpu, (u32)dr, (u32)val, handler);
2855 } else {
2856 val = vcpu->arch.regs[reg];
2857 switch (dr) {
2858 case 0 ... 3:
2859 vcpu->arch.db[dr] = val;
2860 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
2861 vcpu->arch.eff_db[dr] = val;
2862 break;
2863 case 4 ... 5:
2864 if (vcpu->arch.cr4 & X86_CR4_DE)
2865 kvm_queue_exception(vcpu, UD_VECTOR);
2866 break;
2867 case 6:
2868 if (val & 0xffffffff00000000ULL) {
2869 kvm_queue_exception(vcpu, GP_VECTOR);
2870 break;
2871 }
2872 vcpu->arch.dr6 = (val & DR6_VOLATILE) | DR6_FIXED_1;
2873 break;
2874 case 7:
2875 if (val & 0xffffffff00000000ULL) {
2876 kvm_queue_exception(vcpu, GP_VECTOR);
2877 break;
2878 }
2879 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
2880 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
2881 vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
2882 vcpu->arch.switch_db_regs =
2883 (val & DR7_BP_EN_MASK);
2884 }
2885 break;
2886 }
2887 KVMTRACE_2D(DR_WRITE, vcpu, (u32)dr, (u32)val, handler);
2888 }
2889 skip_emulated_instruction(vcpu);
2890 return 1;
2891 }
2892
2893 static int handle_cpuid(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2894 {
2895 kvm_emulate_cpuid(vcpu);
2896 return 1;
2897 }
2898
2899 static int handle_rdmsr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2900 {
2901 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
2902 u64 data;
2903
2904 if (vmx_get_msr(vcpu, ecx, &data)) {
2905 kvm_inject_gp(vcpu, 0);
2906 return 1;
2907 }
2908
2909 KVMTRACE_3D(MSR_READ, vcpu, ecx, (u32)data, (u32)(data >> 32),
2910 handler);
2911
2912 /* FIXME: handling of bits 32:63 of rax, rdx */
2913 vcpu->arch.regs[VCPU_REGS_RAX] = data & -1u;
2914 vcpu->arch.regs[VCPU_REGS_RDX] = (data >> 32) & -1u;
2915 skip_emulated_instruction(vcpu);
2916 return 1;
2917 }
2918
2919 static int handle_wrmsr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2920 {
2921 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
2922 u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
2923 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
2924
2925 KVMTRACE_3D(MSR_WRITE, vcpu, ecx, (u32)data, (u32)(data >> 32),
2926 handler);
2927
2928 if (vmx_set_msr(vcpu, ecx, data) != 0) {
2929 kvm_inject_gp(vcpu, 0);
2930 return 1;
2931 }
2932
2933 skip_emulated_instruction(vcpu);
2934 return 1;
2935 }
2936
2937 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu,
2938 struct kvm_run *kvm_run)
2939 {
2940 return 1;
2941 }
2942
2943 static int handle_interrupt_window(struct kvm_vcpu *vcpu,
2944 struct kvm_run *kvm_run)
2945 {
2946 u32 cpu_based_vm_exec_control;
2947
2948 /* clear pending irq */
2949 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2950 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
2951 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2952
2953 KVMTRACE_0D(PEND_INTR, vcpu, handler);
2954 ++vcpu->stat.irq_window_exits;
2955
2956 /*
2957 * If the user space waits to inject interrupts, exit as soon as
2958 * possible
2959 */
2960 if (kvm_run->request_interrupt_window &&
2961 !vcpu->arch.irq_summary) {
2962 kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
2963 return 0;
2964 }
2965 return 1;
2966 }
2967
2968 static int handle_halt(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2969 {
2970 skip_emulated_instruction(vcpu);
2971 return kvm_emulate_halt(vcpu);
2972 }
2973
2974 static int handle_vmcall(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2975 {
2976 skip_emulated_instruction(vcpu);
2977 kvm_emulate_hypercall(vcpu);
2978 return 1;
2979 }
2980
2981 static int handle_invlpg(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2982 {
2983 u64 exit_qualification = vmcs_read64(EXIT_QUALIFICATION);
2984
2985 kvm_mmu_invlpg(vcpu, exit_qualification);
2986 skip_emulated_instruction(vcpu);
2987 return 1;
2988 }
2989
2990 static int handle_wbinvd(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2991 {
2992 skip_emulated_instruction(vcpu);
2993 /* TODO: Add support for VT-d/pass-through device */
2994 return 1;
2995 }
2996
2997 static int handle_apic_access(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2998 {
2999 u64 exit_qualification;
3000 enum emulation_result er;
3001 unsigned long offset;
3002
3003 exit_qualification = vmcs_read64(EXIT_QUALIFICATION);
3004 offset = exit_qualification & 0xffful;
3005
3006 er = emulate_instruction(vcpu, kvm_run, 0, 0, 0);
3007
3008 if (er != EMULATE_DONE) {
3009 printk(KERN_ERR
3010 "Fail to handle apic access vmexit! Offset is 0x%lx\n",
3011 offset);
3012 return -ENOTSUPP;
3013 }
3014 return 1;
3015 }
3016
3017 static int handle_task_switch(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3018 {
3019 struct vcpu_vmx *vmx = to_vmx(vcpu);
3020 unsigned long exit_qualification;
3021 u16 tss_selector;
3022 int reason;
3023
3024 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3025
3026 reason = (u32)exit_qualification >> 30;
3027 if (reason == TASK_SWITCH_GATE && vmx->vcpu.arch.nmi_injected &&
3028 (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
3029 (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK)
3030 == INTR_TYPE_NMI_INTR) {
3031 vcpu->arch.nmi_injected = false;
3032 if (cpu_has_virtual_nmis())
3033 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
3034 GUEST_INTR_STATE_NMI);
3035 }
3036 tss_selector = exit_qualification;
3037
3038 if (!kvm_task_switch(vcpu, tss_selector, reason))
3039 return 0;
3040
3041 /* clear all local breakpoint enable flags */
3042 vmcs_writel(GUEST_DR7, vmcs_readl(GUEST_DR7) & ~55);
3043
3044 /*
3045 * TODO: What about debug traps on tss switch?
3046 * Are we supposed to inject them and update dr6?
3047 */
3048
3049 return 1;
3050 }
3051
3052 static int handle_ept_violation(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3053 {
3054 u64 exit_qualification;
3055 gpa_t gpa;
3056 int gla_validity;
3057
3058 exit_qualification = vmcs_read64(EXIT_QUALIFICATION);
3059
3060 if (exit_qualification & (1 << 6)) {
3061 printk(KERN_ERR "EPT: GPA exceeds GAW!\n");
3062 return -ENOTSUPP;
3063 }
3064
3065 gla_validity = (exit_qualification >> 7) & 0x3;
3066 if (gla_validity != 0x3 && gla_validity != 0x1 && gla_validity != 0) {
3067 printk(KERN_ERR "EPT: Handling EPT violation failed!\n");
3068 printk(KERN_ERR "EPT: GPA: 0x%lx, GVA: 0x%lx\n",
3069 (long unsigned int)vmcs_read64(GUEST_PHYSICAL_ADDRESS),
3070 (long unsigned int)vmcs_read64(GUEST_LINEAR_ADDRESS));
3071 printk(KERN_ERR "EPT: Exit qualification is 0x%lx\n",
3072 (long unsigned int)exit_qualification);
3073 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
3074 kvm_run->hw.hardware_exit_reason = 0;
3075 return -ENOTSUPP;
3076 }
3077
3078 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
3079 return kvm_mmu_page_fault(vcpu, gpa & PAGE_MASK, 0);
3080 }
3081
3082 static int handle_nmi_window(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3083 {
3084 u32 cpu_based_vm_exec_control;
3085
3086 /* clear pending NMI */
3087 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
3088 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
3089 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
3090 ++vcpu->stat.nmi_window_exits;
3091
3092 return 1;
3093 }
3094
3095 static void handle_invalid_guest_state(struct kvm_vcpu *vcpu,
3096 struct kvm_run *kvm_run)
3097 {
3098 struct vcpu_vmx *vmx = to_vmx(vcpu);
3099 enum emulation_result err = EMULATE_DONE;
3100
3101 preempt_enable();
3102 local_irq_enable();
3103
3104 while (!guest_state_valid(vcpu)) {
3105 err = emulate_instruction(vcpu, kvm_run, 0, 0, 0);
3106
3107 if (err == EMULATE_DO_MMIO)
3108 break;
3109
3110 if (err != EMULATE_DONE) {
3111 kvm_report_emulation_failure(vcpu, "emulation failure");
3112 return;
3113 }
3114
3115 if (signal_pending(current))
3116 break;
3117 if (need_resched())
3118 schedule();
3119 }
3120
3121 local_irq_disable();
3122 preempt_disable();
3123
3124 vmx->invalid_state_emulation_result = err;
3125 }
3126
3127 /*
3128 * The exit handlers return 1 if the exit was handled fully and guest execution
3129 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
3130 * to be done to userspace and return 0.
3131 */
3132 static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu,
3133 struct kvm_run *kvm_run) = {
3134 [EXIT_REASON_EXCEPTION_NMI] = handle_exception,
3135 [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt,
3136 [EXIT_REASON_TRIPLE_FAULT] = handle_triple_fault,
3137 [EXIT_REASON_NMI_WINDOW] = handle_nmi_window,
3138 [EXIT_REASON_IO_INSTRUCTION] = handle_io,
3139 [EXIT_REASON_CR_ACCESS] = handle_cr,
3140 [EXIT_REASON_DR_ACCESS] = handle_dr,
3141 [EXIT_REASON_CPUID] = handle_cpuid,
3142 [EXIT_REASON_MSR_READ] = handle_rdmsr,
3143 [EXIT_REASON_MSR_WRITE] = handle_wrmsr,
3144 [EXIT_REASON_PENDING_INTERRUPT] = handle_interrupt_window,
3145 [EXIT_REASON_HLT] = handle_halt,
3146 [EXIT_REASON_INVLPG] = handle_invlpg,
3147 [EXIT_REASON_VMCALL] = handle_vmcall,
3148 [EXIT_REASON_TPR_BELOW_THRESHOLD] = handle_tpr_below_threshold,
3149 [EXIT_REASON_APIC_ACCESS] = handle_apic_access,
3150 [EXIT_REASON_WBINVD] = handle_wbinvd,
3151 [EXIT_REASON_TASK_SWITCH] = handle_task_switch,
3152 [EXIT_REASON_EPT_VIOLATION] = handle_ept_violation,
3153 };
3154
3155 static const int kvm_vmx_max_exit_handlers =
3156 ARRAY_SIZE(kvm_vmx_exit_handlers);
3157
3158 /*
3159 * The guest has exited. See if we can fix it or if we need userspace
3160 * assistance.
3161 */
3162 static int kvm_handle_exit(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
3163 {
3164 u32 exit_reason = vmcs_read32(VM_EXIT_REASON);
3165 struct vcpu_vmx *vmx = to_vmx(vcpu);
3166 u32 vectoring_info = vmx->idt_vectoring_info;
3167
3168 KVMTRACE_3D(VMEXIT, vcpu, exit_reason, (u32)kvm_rip_read(vcpu),
3169 (u32)((u64)kvm_rip_read(vcpu) >> 32), entryexit);
3170
3171 /* If we need to emulate an MMIO from handle_invalid_guest_state
3172 * we just return 0 */
3173 if (vmx->emulation_required && emulate_invalid_guest_state) {
3174 if (guest_state_valid(vcpu))
3175 vmx->emulation_required = 0;
3176 return vmx->invalid_state_emulation_result != EMULATE_DO_MMIO;
3177 }
3178
3179 /* Access CR3 don't cause VMExit in paging mode, so we need
3180 * to sync with guest real CR3. */
3181 if (vm_need_ept() && is_paging(vcpu)) {
3182 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
3183 ept_load_pdptrs(vcpu);
3184 }
3185
3186 if (unlikely(vmx->fail)) {
3187 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3188 kvm_run->fail_entry.hardware_entry_failure_reason
3189 = vmcs_read32(VM_INSTRUCTION_ERROR);
3190 return 0;
3191 }
3192
3193 if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
3194 (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
3195 exit_reason != EXIT_REASON_EPT_VIOLATION &&
3196 exit_reason != EXIT_REASON_TASK_SWITCH))
3197 printk(KERN_WARNING "%s: unexpected, valid vectoring info "
3198 "(0x%x) and exit reason is 0x%x\n",
3199 __func__, vectoring_info, exit_reason);
3200
3201 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked)) {
3202 if (vcpu->arch.interrupt_window_open) {
3203 vmx->soft_vnmi_blocked = 0;
3204 vcpu->arch.nmi_window_open = 1;
3205 } else if (vmx->vnmi_blocked_time > 1000000000LL &&
3206 vcpu->arch.nmi_pending) {
3207 /*
3208 * This CPU don't support us in finding the end of an
3209 * NMI-blocked window if the guest runs with IRQs
3210 * disabled. So we pull the trigger after 1 s of
3211 * futile waiting, but inform the user about this.
3212 */
3213 printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
3214 "state on VCPU %d after 1 s timeout\n",
3215 __func__, vcpu->vcpu_id);
3216 vmx->soft_vnmi_blocked = 0;
3217 vmx->vcpu.arch.nmi_window_open = 1;
3218 }
3219 }
3220
3221 if (exit_reason < kvm_vmx_max_exit_handlers
3222 && kvm_vmx_exit_handlers[exit_reason])
3223 return kvm_vmx_exit_handlers[exit_reason](vcpu, kvm_run);
3224 else {
3225 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
3226 kvm_run->hw.hardware_exit_reason = exit_reason;
3227 }
3228 return 0;
3229 }
3230
3231 static void update_tpr_threshold(struct kvm_vcpu *vcpu)
3232 {
3233 int max_irr, tpr;
3234
3235 if (!vm_need_tpr_shadow(vcpu->kvm))
3236 return;
3237
3238 if (!kvm_lapic_enabled(vcpu) ||
3239 ((max_irr = kvm_lapic_find_highest_irr(vcpu)) == -1)) {
3240 vmcs_write32(TPR_THRESHOLD, 0);
3241 return;
3242 }
3243
3244 tpr = (kvm_lapic_get_cr8(vcpu) & 0x0f) << 4;
3245 vmcs_write32(TPR_THRESHOLD, (max_irr > tpr) ? tpr >> 4 : max_irr >> 4);
3246 }
3247
3248 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
3249 {
3250 u32 exit_intr_info;
3251 u32 idt_vectoring_info;
3252 bool unblock_nmi;
3253 u8 vector;
3254 int type;
3255 bool idtv_info_valid;
3256 u32 error;
3257
3258 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
3259 if (cpu_has_virtual_nmis()) {
3260 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
3261 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
3262 /*
3263 * SDM 3: 25.7.1.2
3264 * Re-set bit "block by NMI" before VM entry if vmexit caused by
3265 * a guest IRET fault.
3266 */
3267 if (unblock_nmi && vector != DF_VECTOR)
3268 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
3269 GUEST_INTR_STATE_NMI);
3270 } else if (unlikely(vmx->soft_vnmi_blocked))
3271 vmx->vnmi_blocked_time +=
3272 ktime_to_ns(ktime_sub(ktime_get(), vmx->entry_time));
3273
3274 idt_vectoring_info = vmx->idt_vectoring_info;
3275 idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
3276 vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
3277 type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
3278 if (vmx->vcpu.arch.nmi_injected) {
3279 /*
3280 * SDM 3: 25.7.1.2
3281 * Clear bit "block by NMI" before VM entry if a NMI delivery
3282 * faulted.
3283 */
3284 if (idtv_info_valid && type == INTR_TYPE_NMI_INTR)
3285 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
3286 GUEST_INTR_STATE_NMI);
3287 else
3288 vmx->vcpu.arch.nmi_injected = false;
3289 }
3290 kvm_clear_exception_queue(&vmx->vcpu);
3291 if (idtv_info_valid && (type == INTR_TYPE_HARD_EXCEPTION ||
3292 type == INTR_TYPE_SOFT_EXCEPTION)) {
3293 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
3294 error = vmcs_read32(IDT_VECTORING_ERROR_CODE);
3295 kvm_queue_exception_e(&vmx->vcpu, vector, error);
3296 } else
3297 kvm_queue_exception(&vmx->vcpu, vector);
3298 vmx->idt_vectoring_info = 0;
3299 }
3300 kvm_clear_interrupt_queue(&vmx->vcpu);
3301 if (idtv_info_valid && type == INTR_TYPE_EXT_INTR) {
3302 kvm_queue_interrupt(&vmx->vcpu, vector);
3303 vmx->idt_vectoring_info = 0;
3304 }
3305 }
3306
3307 static void vmx_intr_assist(struct kvm_vcpu *vcpu)
3308 {
3309 update_tpr_threshold(vcpu);
3310
3311 vmx_update_window_states(vcpu);
3312
3313 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
3314 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
3315 GUEST_INTR_STATE_STI |
3316 GUEST_INTR_STATE_MOV_SS);
3317
3318 if (vcpu->arch.nmi_pending && !vcpu->arch.nmi_injected) {
3319 if (vcpu->arch.interrupt.pending) {
3320 enable_nmi_window(vcpu);
3321 } else if (vcpu->arch.nmi_window_open) {
3322 vcpu->arch.nmi_pending = false;
3323 vcpu->arch.nmi_injected = true;
3324 } else {
3325 enable_nmi_window(vcpu);
3326 return;
3327 }
3328 }
3329 if (vcpu->arch.nmi_injected) {
3330 vmx_inject_nmi(vcpu);
3331 if (vcpu->arch.nmi_pending)
3332 enable_nmi_window(vcpu);
3333 else if (kvm_cpu_has_interrupt(vcpu))
3334 enable_irq_window(vcpu);
3335 return;
3336 }
3337 if (!vcpu->arch.interrupt.pending && kvm_cpu_has_interrupt(vcpu)) {
3338 if (vcpu->arch.interrupt_window_open)
3339 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu));
3340 else
3341 enable_irq_window(vcpu);
3342 }
3343 if (vcpu->arch.interrupt.pending) {
3344 vmx_inject_irq(vcpu, vcpu->arch.interrupt.nr);
3345 if (kvm_cpu_has_interrupt(vcpu))
3346 enable_irq_window(vcpu);
3347 }
3348 }
3349
3350 /*
3351 * Failure to inject an interrupt should give us the information
3352 * in IDT_VECTORING_INFO_FIELD. However, if the failure occurs
3353 * when fetching the interrupt redirection bitmap in the real-mode
3354 * tss, this doesn't happen. So we do it ourselves.
3355 */
3356 static void fixup_rmode_irq(struct vcpu_vmx *vmx)
3357 {
3358 vmx->rmode.irq.pending = 0;
3359 if (kvm_rip_read(&vmx->vcpu) + 1 != vmx->rmode.irq.rip)
3360 return;
3361 kvm_rip_write(&vmx->vcpu, vmx->rmode.irq.rip);
3362 if (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK) {
3363 vmx->idt_vectoring_info &= ~VECTORING_INFO_TYPE_MASK;
3364 vmx->idt_vectoring_info |= INTR_TYPE_EXT_INTR;
3365 return;
3366 }
3367 vmx->idt_vectoring_info =
3368 VECTORING_INFO_VALID_MASK
3369 | INTR_TYPE_EXT_INTR
3370 | vmx->rmode.irq.vector;
3371 }
3372
3373 #ifdef CONFIG_X86_64
3374 #define R "r"
3375 #define Q "q"
3376 #else
3377 #define R "e"
3378 #define Q "l"
3379 #endif
3380
3381 static void vmx_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3382 {
3383 struct vcpu_vmx *vmx = to_vmx(vcpu);
3384 u32 intr_info;
3385
3386 /* Record the guest's net vcpu time for enforced NMI injections. */
3387 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked))
3388 vmx->entry_time = ktime_get();
3389
3390 /* Handle invalid guest state instead of entering VMX */
3391 if (vmx->emulation_required && emulate_invalid_guest_state) {
3392 handle_invalid_guest_state(vcpu, kvm_run);
3393 return;
3394 }
3395
3396 if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
3397 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
3398 if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
3399 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
3400
3401 /*
3402 * Loading guest fpu may have cleared host cr0.ts
3403 */
3404 vmcs_writel(HOST_CR0, read_cr0());
3405
3406 set_debugreg(vcpu->arch.dr6, 6);
3407
3408 asm(
3409 /* Store host registers */
3410 "push %%"R"dx; push %%"R"bp;"
3411 "push %%"R"cx \n\t"
3412 "cmp %%"R"sp, %c[host_rsp](%0) \n\t"
3413 "je 1f \n\t"
3414 "mov %%"R"sp, %c[host_rsp](%0) \n\t"
3415 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
3416 "1: \n\t"
3417 /* Check if vmlaunch of vmresume is needed */
3418 "cmpl $0, %c[launched](%0) \n\t"
3419 /* Load guest registers. Don't clobber flags. */
3420 "mov %c[cr2](%0), %%"R"ax \n\t"
3421 "mov %%"R"ax, %%cr2 \n\t"
3422 "mov %c[rax](%0), %%"R"ax \n\t"
3423 "mov %c[rbx](%0), %%"R"bx \n\t"
3424 "mov %c[rdx](%0), %%"R"dx \n\t"
3425 "mov %c[rsi](%0), %%"R"si \n\t"
3426 "mov %c[rdi](%0), %%"R"di \n\t"
3427 "mov %c[rbp](%0), %%"R"bp \n\t"
3428 #ifdef CONFIG_X86_64
3429 "mov %c[r8](%0), %%r8 \n\t"
3430 "mov %c[r9](%0), %%r9 \n\t"
3431 "mov %c[r10](%0), %%r10 \n\t"
3432 "mov %c[r11](%0), %%r11 \n\t"
3433 "mov %c[r12](%0), %%r12 \n\t"
3434 "mov %c[r13](%0), %%r13 \n\t"
3435 "mov %c[r14](%0), %%r14 \n\t"
3436 "mov %c[r15](%0), %%r15 \n\t"
3437 #endif
3438 "mov %c[rcx](%0), %%"R"cx \n\t" /* kills %0 (ecx) */
3439
3440 /* Enter guest mode */
3441 "jne .Llaunched \n\t"
3442 __ex(ASM_VMX_VMLAUNCH) "\n\t"
3443 "jmp .Lkvm_vmx_return \n\t"
3444 ".Llaunched: " __ex(ASM_VMX_VMRESUME) "\n\t"
3445 ".Lkvm_vmx_return: "
3446 /* Save guest registers, load host registers, keep flags */
3447 "xchg %0, (%%"R"sp) \n\t"
3448 "mov %%"R"ax, %c[rax](%0) \n\t"
3449 "mov %%"R"bx, %c[rbx](%0) \n\t"
3450 "push"Q" (%%"R"sp); pop"Q" %c[rcx](%0) \n\t"
3451 "mov %%"R"dx, %c[rdx](%0) \n\t"
3452 "mov %%"R"si, %c[rsi](%0) \n\t"
3453 "mov %%"R"di, %c[rdi](%0) \n\t"
3454 "mov %%"R"bp, %c[rbp](%0) \n\t"
3455 #ifdef CONFIG_X86_64
3456 "mov %%r8, %c[r8](%0) \n\t"
3457 "mov %%r9, %c[r9](%0) \n\t"
3458 "mov %%r10, %c[r10](%0) \n\t"
3459 "mov %%r11, %c[r11](%0) \n\t"
3460 "mov %%r12, %c[r12](%0) \n\t"
3461 "mov %%r13, %c[r13](%0) \n\t"
3462 "mov %%r14, %c[r14](%0) \n\t"
3463 "mov %%r15, %c[r15](%0) \n\t"
3464 #endif
3465 "mov %%cr2, %%"R"ax \n\t"
3466 "mov %%"R"ax, %c[cr2](%0) \n\t"
3467
3468 "pop %%"R"bp; pop %%"R"bp; pop %%"R"dx \n\t"
3469 "setbe %c[fail](%0) \n\t"
3470 : : "c"(vmx), "d"((unsigned long)HOST_RSP),
3471 [launched]"i"(offsetof(struct vcpu_vmx, launched)),
3472 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
3473 [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
3474 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
3475 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
3476 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
3477 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
3478 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
3479 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
3480 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
3481 #ifdef CONFIG_X86_64
3482 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
3483 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
3484 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
3485 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
3486 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
3487 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
3488 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
3489 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
3490 #endif
3491 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2))
3492 : "cc", "memory"
3493 , R"bx", R"di", R"si"
3494 #ifdef CONFIG_X86_64
3495 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
3496 #endif
3497 );
3498
3499 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP));
3500 vcpu->arch.regs_dirty = 0;
3501
3502 get_debugreg(vcpu->arch.dr6, 6);
3503
3504 vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
3505 if (vmx->rmode.irq.pending)
3506 fixup_rmode_irq(vmx);
3507
3508 vmx_update_window_states(vcpu);
3509
3510 asm("mov %0, %%ds; mov %0, %%es" : : "r"(__USER_DS));
3511 vmx->launched = 1;
3512
3513 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
3514
3515 /* We need to handle NMIs before interrupts are enabled */
3516 if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR &&
3517 (intr_info & INTR_INFO_VALID_MASK)) {
3518 KVMTRACE_0D(NMI, vcpu, handler);
3519 asm("int $2");
3520 }
3521
3522 vmx_complete_interrupts(vmx);
3523 }
3524
3525 #undef R
3526 #undef Q
3527
3528 static void vmx_free_vmcs(struct kvm_vcpu *vcpu)
3529 {
3530 struct vcpu_vmx *vmx = to_vmx(vcpu);
3531
3532 if (vmx->vmcs) {
3533 vcpu_clear(vmx);
3534 free_vmcs(vmx->vmcs);
3535 vmx->vmcs = NULL;
3536 }
3537 }
3538
3539 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
3540 {
3541 struct vcpu_vmx *vmx = to_vmx(vcpu);
3542
3543 spin_lock(&vmx_vpid_lock);
3544 if (vmx->vpid != 0)
3545 __clear_bit(vmx->vpid, vmx_vpid_bitmap);
3546 spin_unlock(&vmx_vpid_lock);
3547 vmx_free_vmcs(vcpu);
3548 kfree(vmx->host_msrs);
3549 kfree(vmx->guest_msrs);
3550 kvm_vcpu_uninit(vcpu);
3551 kmem_cache_free(kvm_vcpu_cache, vmx);
3552 }
3553
3554 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
3555 {
3556 int err;
3557 struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
3558 int cpu;
3559
3560 if (!vmx)
3561 return ERR_PTR(-ENOMEM);
3562
3563 allocate_vpid(vmx);
3564
3565 err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
3566 if (err)
3567 goto free_vcpu;
3568
3569 vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
3570 if (!vmx->guest_msrs) {
3571 err = -ENOMEM;
3572 goto uninit_vcpu;
3573 }
3574
3575 vmx->host_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
3576 if (!vmx->host_msrs)
3577 goto free_guest_msrs;
3578
3579 vmx->vmcs = alloc_vmcs();
3580 if (!vmx->vmcs)
3581 goto free_msrs;
3582
3583 vmcs_clear(vmx->vmcs);
3584
3585 cpu = get_cpu();
3586 vmx_vcpu_load(&vmx->vcpu, cpu);
3587 err = vmx_vcpu_setup(vmx);
3588 vmx_vcpu_put(&vmx->vcpu);
3589 put_cpu();
3590 if (err)
3591 goto free_vmcs;
3592 if (vm_need_virtualize_apic_accesses(kvm))
3593 if (alloc_apic_access_page(kvm) != 0)
3594 goto free_vmcs;
3595
3596 if (vm_need_ept())
3597 if (alloc_identity_pagetable(kvm) != 0)
3598 goto free_vmcs;
3599
3600 return &vmx->vcpu;
3601
3602 free_vmcs:
3603 free_vmcs(vmx->vmcs);
3604 free_msrs:
3605 kfree(vmx->host_msrs);
3606 free_guest_msrs:
3607 kfree(vmx->guest_msrs);
3608 uninit_vcpu:
3609 kvm_vcpu_uninit(&vmx->vcpu);
3610 free_vcpu:
3611 kmem_cache_free(kvm_vcpu_cache, vmx);
3612 return ERR_PTR(err);
3613 }
3614
3615 static void __init vmx_check_processor_compat(void *rtn)
3616 {
3617 struct vmcs_config vmcs_conf;
3618
3619 *(int *)rtn = 0;
3620 if (setup_vmcs_config(&vmcs_conf) < 0)
3621 *(int *)rtn = -EIO;
3622 if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
3623 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
3624 smp_processor_id());
3625 *(int *)rtn = -EIO;
3626 }
3627 }
3628
3629 static int get_ept_level(void)
3630 {
3631 return VMX_EPT_DEFAULT_GAW + 1;
3632 }
3633
3634 static int vmx_get_mt_mask_shift(void)
3635 {
3636 return VMX_EPT_MT_EPTE_SHIFT;
3637 }
3638
3639 static struct kvm_x86_ops vmx_x86_ops = {
3640 .cpu_has_kvm_support = cpu_has_kvm_support,
3641 .disabled_by_bios = vmx_disabled_by_bios,
3642 .hardware_setup = hardware_setup,
3643 .hardware_unsetup = hardware_unsetup,
3644 .check_processor_compatibility = vmx_check_processor_compat,
3645 .hardware_enable = hardware_enable,
3646 .hardware_disable = hardware_disable,
3647 .cpu_has_accelerated_tpr = cpu_has_vmx_virtualize_apic_accesses,
3648
3649 .vcpu_create = vmx_create_vcpu,
3650 .vcpu_free = vmx_free_vcpu,
3651 .vcpu_reset = vmx_vcpu_reset,
3652
3653 .prepare_guest_switch = vmx_save_host_state,
3654 .vcpu_load = vmx_vcpu_load,
3655 .vcpu_put = vmx_vcpu_put,
3656
3657 .set_guest_debug = set_guest_debug,
3658 .get_msr = vmx_get_msr,
3659 .set_msr = vmx_set_msr,
3660 .get_segment_base = vmx_get_segment_base,
3661 .get_segment = vmx_get_segment,
3662 .set_segment = vmx_set_segment,
3663 .get_cpl = vmx_get_cpl,
3664 .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
3665 .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
3666 .set_cr0 = vmx_set_cr0,
3667 .set_cr3 = vmx_set_cr3,
3668 .set_cr4 = vmx_set_cr4,
3669 .set_efer = vmx_set_efer,
3670 .get_idt = vmx_get_idt,
3671 .set_idt = vmx_set_idt,
3672 .get_gdt = vmx_get_gdt,
3673 .set_gdt = vmx_set_gdt,
3674 .cache_reg = vmx_cache_reg,
3675 .get_rflags = vmx_get_rflags,
3676 .set_rflags = vmx_set_rflags,
3677
3678 .tlb_flush = vmx_flush_tlb,
3679
3680 .run = vmx_vcpu_run,
3681 .handle_exit = kvm_handle_exit,
3682 .skip_emulated_instruction = skip_emulated_instruction,
3683 .patch_hypercall = vmx_patch_hypercall,
3684 .get_irq = vmx_get_irq,
3685 .set_irq = vmx_inject_irq,
3686 .queue_exception = vmx_queue_exception,
3687 .exception_injected = vmx_exception_injected,
3688 .inject_pending_irq = vmx_intr_assist,
3689 .inject_pending_vectors = do_interrupt_requests,
3690
3691 .set_tss_addr = vmx_set_tss_addr,
3692 .get_tdp_level = get_ept_level,
3693 .get_mt_mask_shift = vmx_get_mt_mask_shift,
3694 };
3695
3696 static int __init vmx_init(void)
3697 {
3698 void *va;
3699 int r;
3700
3701 vmx_io_bitmap_a = alloc_page(GFP_KERNEL | __GFP_HIGHMEM);
3702 if (!vmx_io_bitmap_a)
3703 return -ENOMEM;
3704
3705 vmx_io_bitmap_b = alloc_page(GFP_KERNEL | __GFP_HIGHMEM);
3706 if (!vmx_io_bitmap_b) {
3707 r = -ENOMEM;
3708 goto out;
3709 }
3710
3711 vmx_msr_bitmap = alloc_page(GFP_KERNEL | __GFP_HIGHMEM);
3712 if (!vmx_msr_bitmap) {
3713 r = -ENOMEM;
3714 goto out1;
3715 }
3716
3717 /*
3718 * Allow direct access to the PC debug port (it is often used for I/O
3719 * delays, but the vmexits simply slow things down).
3720 */
3721 va = kmap(vmx_io_bitmap_a);
3722 memset(va, 0xff, PAGE_SIZE);
3723 clear_bit(0x80, va);
3724 kunmap(vmx_io_bitmap_a);
3725
3726 va = kmap(vmx_io_bitmap_b);
3727 memset(va, 0xff, PAGE_SIZE);
3728 kunmap(vmx_io_bitmap_b);
3729
3730 va = kmap(vmx_msr_bitmap);
3731 memset(va, 0xff, PAGE_SIZE);
3732 kunmap(vmx_msr_bitmap);
3733
3734 set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
3735
3736 r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx), THIS_MODULE);
3737 if (r)
3738 goto out2;
3739
3740 vmx_disable_intercept_for_msr(vmx_msr_bitmap, MSR_FS_BASE);
3741 vmx_disable_intercept_for_msr(vmx_msr_bitmap, MSR_GS_BASE);
3742 vmx_disable_intercept_for_msr(vmx_msr_bitmap, MSR_IA32_SYSENTER_CS);
3743 vmx_disable_intercept_for_msr(vmx_msr_bitmap, MSR_IA32_SYSENTER_ESP);
3744 vmx_disable_intercept_for_msr(vmx_msr_bitmap, MSR_IA32_SYSENTER_EIP);
3745
3746 if (vm_need_ept()) {
3747 bypass_guest_pf = 0;
3748 kvm_mmu_set_base_ptes(VMX_EPT_READABLE_MASK |
3749 VMX_EPT_WRITABLE_MASK);
3750 kvm_mmu_set_mask_ptes(0ull, 0ull, 0ull, 0ull,
3751 VMX_EPT_EXECUTABLE_MASK,
3752 VMX_EPT_DEFAULT_MT << VMX_EPT_MT_EPTE_SHIFT);
3753 kvm_enable_tdp();
3754 } else
3755 kvm_disable_tdp();
3756
3757 if (bypass_guest_pf)
3758 kvm_mmu_set_nonpresent_ptes(~0xffeull, 0ull);
3759
3760 ept_sync_global();
3761
3762 return 0;
3763
3764 out2:
3765 __free_page(vmx_msr_bitmap);
3766 out1:
3767 __free_page(vmx_io_bitmap_b);
3768 out:
3769 __free_page(vmx_io_bitmap_a);
3770 return r;
3771 }
3772
3773 static void __exit vmx_exit(void)
3774 {
3775 __free_page(vmx_msr_bitmap);
3776 __free_page(vmx_io_bitmap_b);
3777 __free_page(vmx_io_bitmap_a);
3778
3779 kvm_exit();
3780 }
3781
3782 module_init(vmx_init)
3783 module_exit(vmx_exit)
kernel source for telekom puls (alcatel/mediatek)