?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Linux 2.6.25.3
[linux/fpc-iii.git] / arch / x86 / kvm / svm.c
blob1a582f1090e895aaa19634aa21d02ccf6c1584f0
1 /*
2 * Kernel-based Virtual Machine driver for Linux
3 *
4 * AMD SVM support
5 *
6 * Copyright (C) 2006 Qumranet, Inc.
7 *
8 * Authors:
9 * Yaniv Kamay <yaniv@qumranet.com>
10 * Avi Kivity <avi@qumranet.com>
11 *
12 * This work is licensed under the terms of the GNU GPL, version 2. See
13 * the COPYING file in the top-level directory.
14 *
15 */
16 #include <linux/kvm_host.h>
17
18 #include "kvm_svm.h"
19 #include "irq.h"
20 #include "mmu.h"
21
22 #include <linux/module.h>
23 #include <linux/kernel.h>
24 #include <linux/vmalloc.h>
25 #include <linux/highmem.h>
26 #include <linux/sched.h>
27
28 #include <asm/desc.h>
29
30 MODULE_AUTHOR("Qumranet");
31 MODULE_LICENSE("GPL");
32
33 #define IOPM_ALLOC_ORDER 2
34 #define MSRPM_ALLOC_ORDER 1
35
36 #define DB_VECTOR 1
37 #define UD_VECTOR 6
38 #define GP_VECTOR 13
39
40 #define DR7_GD_MASK (1 << 13)
41 #define DR6_BD_MASK (1 << 13)
42
43 #define SEG_TYPE_LDT 2
44 #define SEG_TYPE_BUSY_TSS16 3
45
46 #define SVM_FEATURE_NPT (1 << 0)
47 #define SVM_FEATURE_LBRV (1 << 1)
48 #define SVM_DEATURE_SVML (1 << 2)
49
50 static void kvm_reput_irq(struct vcpu_svm *svm);
51
52 static inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu)
53 {
54 return container_of(vcpu, struct vcpu_svm, vcpu);
55 }
56
57 unsigned long iopm_base;
58 unsigned long msrpm_base;
59
60 struct kvm_ldttss_desc {
61 u16 limit0;
62 u16 base0;
63 unsigned base1 : 8, type : 5, dpl : 2, p : 1;
64 unsigned limit1 : 4, zero0 : 3, g : 1, base2 : 8;
65 u32 base3;
66 u32 zero1;
67 } __attribute__((packed));
68
69 struct svm_cpu_data {
70 int cpu;
71
72 u64 asid_generation;
73 u32 max_asid;
74 u32 next_asid;
75 struct kvm_ldttss_desc *tss_desc;
76
77 struct page *save_area;
78 };
79
80 static DEFINE_PER_CPU(struct svm_cpu_data *, svm_data);
81 static uint32_t svm_features;
82
83 struct svm_init_data {
84 int cpu;
85 int r;
86 };
87
88 static u32 msrpm_ranges[] = {0, 0xc0000000, 0xc0010000};
89
90 #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
91 #define MSRS_RANGE_SIZE 2048
92 #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
93
94 #define MAX_INST_SIZE 15
95
96 static inline u32 svm_has(u32 feat)
97 {
98 return svm_features & feat;
99 }
100
101 static inline u8 pop_irq(struct kvm_vcpu *vcpu)
102 {
103 int word_index = __ffs(vcpu->arch.irq_summary);
104 int bit_index = __ffs(vcpu->arch.irq_pending[word_index]);
105 int irq = word_index * BITS_PER_LONG + bit_index;
106
107 clear_bit(bit_index, &vcpu->arch.irq_pending[word_index]);
108 if (!vcpu->arch.irq_pending[word_index])
109 clear_bit(word_index, &vcpu->arch.irq_summary);
110 return irq;
111 }
112
113 static inline void push_irq(struct kvm_vcpu *vcpu, u8 irq)
114 {
115 set_bit(irq, vcpu->arch.irq_pending);
116 set_bit(irq / BITS_PER_LONG, &vcpu->arch.irq_summary);
117 }
118
119 static inline void clgi(void)
120 {
121 asm volatile (SVM_CLGI);
122 }
123
124 static inline void stgi(void)
125 {
126 asm volatile (SVM_STGI);
127 }
128
129 static inline void invlpga(unsigned long addr, u32 asid)
130 {
131 asm volatile (SVM_INVLPGA :: "a"(addr), "c"(asid));
132 }
133
134 static inline unsigned long kvm_read_cr2(void)
135 {
136 unsigned long cr2;
137
138 asm volatile ("mov %%cr2, %0" : "=r" (cr2));
139 return cr2;
140 }
141
142 static inline void kvm_write_cr2(unsigned long val)
143 {
144 asm volatile ("mov %0, %%cr2" :: "r" (val));
145 }
146
147 static inline unsigned long read_dr6(void)
148 {
149 unsigned long dr6;
150
151 asm volatile ("mov %%dr6, %0" : "=r" (dr6));
152 return dr6;
153 }
154
155 static inline void write_dr6(unsigned long val)
156 {
157 asm volatile ("mov %0, %%dr6" :: "r" (val));
158 }
159
160 static inline unsigned long read_dr7(void)
161 {
162 unsigned long dr7;
163
164 asm volatile ("mov %%dr7, %0" : "=r" (dr7));
165 return dr7;
166 }
167
168 static inline void write_dr7(unsigned long val)
169 {
170 asm volatile ("mov %0, %%dr7" :: "r" (val));
171 }
172
173 static inline void force_new_asid(struct kvm_vcpu *vcpu)
174 {
175 to_svm(vcpu)->asid_generation--;
176 }
177
178 static inline void flush_guest_tlb(struct kvm_vcpu *vcpu)
179 {
180 force_new_asid(vcpu);
181 }
182
183 static void svm_set_efer(struct kvm_vcpu *vcpu, u64 efer)
184 {
185 if (!(efer & EFER_LMA))
186 efer &= ~EFER_LME;
187
188 to_svm(vcpu)->vmcb->save.efer = efer | MSR_EFER_SVME_MASK;
189 vcpu->arch.shadow_efer = efer;
190 }
191
192 static void svm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
193 bool has_error_code, u32 error_code)
194 {
195 struct vcpu_svm *svm = to_svm(vcpu);
196
197 svm->vmcb->control.event_inj = nr
198 | SVM_EVTINJ_VALID
199 | (has_error_code ? SVM_EVTINJ_VALID_ERR : 0)
200 | SVM_EVTINJ_TYPE_EXEPT;
201 svm->vmcb->control.event_inj_err = error_code;
202 }
203
204 static bool svm_exception_injected(struct kvm_vcpu *vcpu)
205 {
206 struct vcpu_svm *svm = to_svm(vcpu);
207
208 return !(svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_VALID);
209 }
210
211 static int is_external_interrupt(u32 info)
212 {
213 info &= SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
214 return info == (SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR);
215 }
216
217 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
218 {
219 struct vcpu_svm *svm = to_svm(vcpu);
220
221 if (!svm->next_rip) {
222 printk(KERN_DEBUG "%s: NOP\n", __FUNCTION__);
223 return;
224 }
225 if (svm->next_rip - svm->vmcb->save.rip > MAX_INST_SIZE)
226 printk(KERN_ERR "%s: ip 0x%llx next 0x%llx\n",
227 __FUNCTION__,
228 svm->vmcb->save.rip,
229 svm->next_rip);
230
231 vcpu->arch.rip = svm->vmcb->save.rip = svm->next_rip;
232 svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK;
233
234 vcpu->arch.interrupt_window_open = 1;
235 }
236
237 static int has_svm(void)
238 {
239 uint32_t eax, ebx, ecx, edx;
240
241 if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD) {
242 printk(KERN_INFO "has_svm: not amd\n");
243 return 0;
244 }
245
246 cpuid(0x80000000, &eax, &ebx, &ecx, &edx);
247 if (eax < SVM_CPUID_FUNC) {
248 printk(KERN_INFO "has_svm: can't execute cpuid_8000000a\n");
249 return 0;
250 }
251
252 cpuid(0x80000001, &eax, &ebx, &ecx, &edx);
253 if (!(ecx & (1 << SVM_CPUID_FEATURE_SHIFT))) {
254 printk(KERN_DEBUG "has_svm: svm not available\n");
255 return 0;
256 }
257 return 1;
258 }
259
260 static void svm_hardware_disable(void *garbage)
261 {
262 struct svm_cpu_data *svm_data
263 = per_cpu(svm_data, raw_smp_processor_id());
264
265 if (svm_data) {
266 uint64_t efer;
267
268 wrmsrl(MSR_VM_HSAVE_PA, 0);
269 rdmsrl(MSR_EFER, efer);
270 wrmsrl(MSR_EFER, efer & ~MSR_EFER_SVME_MASK);
271 per_cpu(svm_data, raw_smp_processor_id()) = NULL;
272 __free_page(svm_data->save_area);
273 kfree(svm_data);
274 }
275 }
276
277 static void svm_hardware_enable(void *garbage)
278 {
279
280 struct svm_cpu_data *svm_data;
281 uint64_t efer;
282 #ifdef CONFIG_X86_64
283 struct desc_ptr gdt_descr;
284 #else
285 struct desc_ptr gdt_descr;
286 #endif
287 struct desc_struct *gdt;
288 int me = raw_smp_processor_id();
289
290 if (!has_svm()) {
291 printk(KERN_ERR "svm_cpu_init: err EOPNOTSUPP on %d\n", me);
292 return;
293 }
294 svm_data = per_cpu(svm_data, me);
295
296 if (!svm_data) {
297 printk(KERN_ERR "svm_cpu_init: svm_data is NULL on %d\n",
298 me);
299 return;
300 }
301
302 svm_data->asid_generation = 1;
303 svm_data->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1;
304 svm_data->next_asid = svm_data->max_asid + 1;
305 svm_features = cpuid_edx(SVM_CPUID_FUNC);
306
307 asm volatile ("sgdt %0" : "=m"(gdt_descr));
308 gdt = (struct desc_struct *)gdt_descr.address;
309 svm_data->tss_desc = (struct kvm_ldttss_desc *)(gdt + GDT_ENTRY_TSS);
310
311 rdmsrl(MSR_EFER, efer);
312 wrmsrl(MSR_EFER, efer | MSR_EFER_SVME_MASK);
313
314 wrmsrl(MSR_VM_HSAVE_PA,
315 page_to_pfn(svm_data->save_area) << PAGE_SHIFT);
316 }
317
318 static int svm_cpu_init(int cpu)
319 {
320 struct svm_cpu_data *svm_data;
321 int r;
322
323 svm_data = kzalloc(sizeof(struct svm_cpu_data), GFP_KERNEL);
324 if (!svm_data)
325 return -ENOMEM;
326 svm_data->cpu = cpu;
327 svm_data->save_area = alloc_page(GFP_KERNEL);
328 r = -ENOMEM;
329 if (!svm_data->save_area)
330 goto err_1;
331
332 per_cpu(svm_data, cpu) = svm_data;
333
334 return 0;
335
336 err_1:
337 kfree(svm_data);
338 return r;
339
340 }
341
342 static void set_msr_interception(u32 *msrpm, unsigned msr,
343 int read, int write)
344 {
345 int i;
346
347 for (i = 0; i < NUM_MSR_MAPS; i++) {
348 if (msr >= msrpm_ranges[i] &&
349 msr < msrpm_ranges[i] + MSRS_IN_RANGE) {
350 u32 msr_offset = (i * MSRS_IN_RANGE + msr -
351 msrpm_ranges[i]) * 2;
352
353 u32 *base = msrpm + (msr_offset / 32);
354 u32 msr_shift = msr_offset % 32;
355 u32 mask = ((write) ? 0 : 2) | ((read) ? 0 : 1);
356 *base = (*base & ~(0x3 << msr_shift)) |
357 (mask << msr_shift);
358 return;
359 }
360 }
361 BUG();
362 }
363
364 static __init int svm_hardware_setup(void)
365 {
366 int cpu;
367 struct page *iopm_pages;
368 struct page *msrpm_pages;
369 void *iopm_va, *msrpm_va;
370 int r;
371
372 iopm_pages = alloc_pages(GFP_KERNEL, IOPM_ALLOC_ORDER);
373
374 if (!iopm_pages)
375 return -ENOMEM;
376
377 iopm_va = page_address(iopm_pages);
378 memset(iopm_va, 0xff, PAGE_SIZE * (1 << IOPM_ALLOC_ORDER));
379 clear_bit(0x80, iopm_va); /* allow direct access to PC debug port */
380 iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT;
381
382
383 msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
384
385 r = -ENOMEM;
386 if (!msrpm_pages)
387 goto err_1;
388
389 msrpm_va = page_address(msrpm_pages);
390 memset(msrpm_va, 0xff, PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER));
391 msrpm_base = page_to_pfn(msrpm_pages) << PAGE_SHIFT;
392
393 #ifdef CONFIG_X86_64
394 set_msr_interception(msrpm_va, MSR_GS_BASE, 1, 1);
395 set_msr_interception(msrpm_va, MSR_FS_BASE, 1, 1);
396 set_msr_interception(msrpm_va, MSR_KERNEL_GS_BASE, 1, 1);
397 set_msr_interception(msrpm_va, MSR_LSTAR, 1, 1);
398 set_msr_interception(msrpm_va, MSR_CSTAR, 1, 1);
399 set_msr_interception(msrpm_va, MSR_SYSCALL_MASK, 1, 1);
400 #endif
401 set_msr_interception(msrpm_va, MSR_K6_STAR, 1, 1);
402 set_msr_interception(msrpm_va, MSR_IA32_SYSENTER_CS, 1, 1);
403 set_msr_interception(msrpm_va, MSR_IA32_SYSENTER_ESP, 1, 1);
404 set_msr_interception(msrpm_va, MSR_IA32_SYSENTER_EIP, 1, 1);
405
406 for_each_online_cpu(cpu) {
407 r = svm_cpu_init(cpu);
408 if (r)
409 goto err_2;
410 }
411 return 0;
412
413 err_2:
414 __free_pages(msrpm_pages, MSRPM_ALLOC_ORDER);
415 msrpm_base = 0;
416 err_1:
417 __free_pages(iopm_pages, IOPM_ALLOC_ORDER);
418 iopm_base = 0;
419 return r;
420 }
421
422 static __exit void svm_hardware_unsetup(void)
423 {
424 __free_pages(pfn_to_page(msrpm_base >> PAGE_SHIFT), MSRPM_ALLOC_ORDER);
425 __free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT), IOPM_ALLOC_ORDER);
426 iopm_base = msrpm_base = 0;
427 }
428
429 static void init_seg(struct vmcb_seg *seg)
430 {
431 seg->selector = 0;
432 seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK |
433 SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */
434 seg->limit = 0xffff;
435 seg->base = 0;
436 }
437
438 static void init_sys_seg(struct vmcb_seg *seg, uint32_t type)
439 {
440 seg->selector = 0;
441 seg->attrib = SVM_SELECTOR_P_MASK | type;
442 seg->limit = 0xffff;
443 seg->base = 0;
444 }
445
446 static void init_vmcb(struct vmcb *vmcb)
447 {
448 struct vmcb_control_area *control = &vmcb->control;
449 struct vmcb_save_area *save = &vmcb->save;
450
451 control->intercept_cr_read = INTERCEPT_CR0_MASK |
452 INTERCEPT_CR3_MASK |
453 INTERCEPT_CR4_MASK |
454 INTERCEPT_CR8_MASK;
455
456 control->intercept_cr_write = INTERCEPT_CR0_MASK |
457 INTERCEPT_CR3_MASK |
458 INTERCEPT_CR4_MASK |
459 INTERCEPT_CR8_MASK;
460
461 control->intercept_dr_read = INTERCEPT_DR0_MASK |
462 INTERCEPT_DR1_MASK |
463 INTERCEPT_DR2_MASK |
464 INTERCEPT_DR3_MASK;
465
466 control->intercept_dr_write = INTERCEPT_DR0_MASK |
467 INTERCEPT_DR1_MASK |
468 INTERCEPT_DR2_MASK |
469 INTERCEPT_DR3_MASK |
470 INTERCEPT_DR5_MASK |
471 INTERCEPT_DR7_MASK;
472
473 control->intercept_exceptions = (1 << PF_VECTOR) |
474 (1 << UD_VECTOR);
475
476
477 control->intercept = (1ULL << INTERCEPT_INTR) |
478 (1ULL << INTERCEPT_NMI) |
479 (1ULL << INTERCEPT_SMI) |
480 /*
481 * selective cr0 intercept bug?
482 * 0: 0f 22 d8 mov %eax,%cr3
483 * 3: 0f 20 c0 mov %cr0,%eax
484 * 6: 0d 00 00 00 80 or $0x80000000,%eax
485 * b: 0f 22 c0 mov %eax,%cr0
486 * set cr3 ->interception
487 * get cr0 ->interception
488 * set cr0 -> no interception
489 */
490 /* (1ULL << INTERCEPT_SELECTIVE_CR0) | */
491 (1ULL << INTERCEPT_CPUID) |
492 (1ULL << INTERCEPT_INVD) |
493 (1ULL << INTERCEPT_HLT) |
494 (1ULL << INTERCEPT_INVLPGA) |
495 (1ULL << INTERCEPT_IOIO_PROT) |
496 (1ULL << INTERCEPT_MSR_PROT) |
497 (1ULL << INTERCEPT_TASK_SWITCH) |
498 (1ULL << INTERCEPT_SHUTDOWN) |
499 (1ULL << INTERCEPT_VMRUN) |
500 (1ULL << INTERCEPT_VMMCALL) |
501 (1ULL << INTERCEPT_VMLOAD) |
502 (1ULL << INTERCEPT_VMSAVE) |
503 (1ULL << INTERCEPT_STGI) |
504 (1ULL << INTERCEPT_CLGI) |
505 (1ULL << INTERCEPT_SKINIT) |
506 (1ULL << INTERCEPT_WBINVD) |
507 (1ULL << INTERCEPT_MONITOR) |
508 (1ULL << INTERCEPT_MWAIT);
509
510 control->iopm_base_pa = iopm_base;
511 control->msrpm_base_pa = msrpm_base;
512 control->tsc_offset = 0;
513 control->int_ctl = V_INTR_MASKING_MASK;
514
515 init_seg(&save->es);
516 init_seg(&save->ss);
517 init_seg(&save->ds);
518 init_seg(&save->fs);
519 init_seg(&save->gs);
520
521 save->cs.selector = 0xf000;
522 /* Executable/Readable Code Segment */
523 save->cs.attrib = SVM_SELECTOR_READ_MASK | SVM_SELECTOR_P_MASK |
524 SVM_SELECTOR_S_MASK | SVM_SELECTOR_CODE_MASK;
525 save->cs.limit = 0xffff;
526 /*
527 * cs.base should really be 0xffff0000, but vmx can't handle that, so
528 * be consistent with it.
529 *
530 * Replace when we have real mode working for vmx.
531 */
532 save->cs.base = 0xf0000;
533
534 save->gdtr.limit = 0xffff;
535 save->idtr.limit = 0xffff;
536
537 init_sys_seg(&save->ldtr, SEG_TYPE_LDT);
538 init_sys_seg(&save->tr, SEG_TYPE_BUSY_TSS16);
539
540 save->efer = MSR_EFER_SVME_MASK;
541 save->dr6 = 0xffff0ff0;
542 save->dr7 = 0x400;
543 save->rflags = 2;
544 save->rip = 0x0000fff0;
545
546 /*
547 * cr0 val on cpu init should be 0x60000010, we enable cpu
548 * cache by default. the orderly way is to enable cache in bios.
549 */
550 save->cr0 = 0x00000010 | X86_CR0_PG | X86_CR0_WP;
551 save->cr4 = X86_CR4_PAE;
552 /* rdx = ?? */
553 }
554
555 static int svm_vcpu_reset(struct kvm_vcpu *vcpu)
556 {
557 struct vcpu_svm *svm = to_svm(vcpu);
558
559 init_vmcb(svm->vmcb);
560
561 if (vcpu->vcpu_id != 0) {
562 svm->vmcb->save.rip = 0;
563 svm->vmcb->save.cs.base = svm->vcpu.arch.sipi_vector << 12;
564 svm->vmcb->save.cs.selector = svm->vcpu.arch.sipi_vector << 8;
565 }
566
567 return 0;
568 }
569
570 static struct kvm_vcpu *svm_create_vcpu(struct kvm *kvm, unsigned int id)
571 {
572 struct vcpu_svm *svm;
573 struct page *page;
574 int err;
575
576 svm = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
577 if (!svm) {
578 err = -ENOMEM;
579 goto out;
580 }
581
582 err = kvm_vcpu_init(&svm->vcpu, kvm, id);
583 if (err)
584 goto free_svm;
585
586 page = alloc_page(GFP_KERNEL);
587 if (!page) {
588 err = -ENOMEM;
589 goto uninit;
590 }
591
592 svm->vmcb = page_address(page);
593 clear_page(svm->vmcb);
594 svm->vmcb_pa = page_to_pfn(page) << PAGE_SHIFT;
595 svm->asid_generation = 0;
596 memset(svm->db_regs, 0, sizeof(svm->db_regs));
597 init_vmcb(svm->vmcb);
598
599 fx_init(&svm->vcpu);
600 svm->vcpu.fpu_active = 1;
601 svm->vcpu.arch.apic_base = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
602 if (svm->vcpu.vcpu_id == 0)
603 svm->vcpu.arch.apic_base |= MSR_IA32_APICBASE_BSP;
604
605 return &svm->vcpu;
606
607 uninit:
608 kvm_vcpu_uninit(&svm->vcpu);
609 free_svm:
610 kmem_cache_free(kvm_vcpu_cache, svm);
611 out:
612 return ERR_PTR(err);
613 }
614
615 static void svm_free_vcpu(struct kvm_vcpu *vcpu)
616 {
617 struct vcpu_svm *svm = to_svm(vcpu);
618
619 __free_page(pfn_to_page(svm->vmcb_pa >> PAGE_SHIFT));
620 kvm_vcpu_uninit(vcpu);
621 kmem_cache_free(kvm_vcpu_cache, svm);
622 }
623
624 static void svm_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
625 {
626 struct vcpu_svm *svm = to_svm(vcpu);
627 int i;
628
629 if (unlikely(cpu != vcpu->cpu)) {
630 u64 tsc_this, delta;
631
632 /*
633 * Make sure that the guest sees a monotonically
634 * increasing TSC.
635 */
636 rdtscll(tsc_this);
637 delta = vcpu->arch.host_tsc - tsc_this;
638 svm->vmcb->control.tsc_offset += delta;
639 vcpu->cpu = cpu;
640 kvm_migrate_apic_timer(vcpu);
641 }
642
643 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
644 rdmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
645 }
646
647 static void svm_vcpu_put(struct kvm_vcpu *vcpu)
648 {
649 struct vcpu_svm *svm = to_svm(vcpu);
650 int i;
651
652 ++vcpu->stat.host_state_reload;
653 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
654 wrmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
655
656 rdtscll(vcpu->arch.host_tsc);
657 }
658
659 static void svm_vcpu_decache(struct kvm_vcpu *vcpu)
660 {
661 }
662
663 static void svm_cache_regs(struct kvm_vcpu *vcpu)
664 {
665 struct vcpu_svm *svm = to_svm(vcpu);
666
667 vcpu->arch.regs[VCPU_REGS_RAX] = svm->vmcb->save.rax;
668 vcpu->arch.regs[VCPU_REGS_RSP] = svm->vmcb->save.rsp;
669 vcpu->arch.rip = svm->vmcb->save.rip;
670 }
671
672 static void svm_decache_regs(struct kvm_vcpu *vcpu)
673 {
674 struct vcpu_svm *svm = to_svm(vcpu);
675 svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX];
676 svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP];
677 svm->vmcb->save.rip = vcpu->arch.rip;
678 }
679
680 static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu)
681 {
682 return to_svm(vcpu)->vmcb->save.rflags;
683 }
684
685 static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
686 {
687 to_svm(vcpu)->vmcb->save.rflags = rflags;
688 }
689
690 static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg)
691 {
692 struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
693
694 switch (seg) {
695 case VCPU_SREG_CS: return &save->cs;
696 case VCPU_SREG_DS: return &save->ds;
697 case VCPU_SREG_ES: return &save->es;
698 case VCPU_SREG_FS: return &save->fs;
699 case VCPU_SREG_GS: return &save->gs;
700 case VCPU_SREG_SS: return &save->ss;
701 case VCPU_SREG_TR: return &save->tr;
702 case VCPU_SREG_LDTR: return &save->ldtr;
703 }
704 BUG();
705 return NULL;
706 }
707
708 static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg)
709 {
710 struct vmcb_seg *s = svm_seg(vcpu, seg);
711
712 return s->base;
713 }
714
715 static void svm_get_segment(struct kvm_vcpu *vcpu,
716 struct kvm_segment *var, int seg)
717 {
718 struct vmcb_seg *s = svm_seg(vcpu, seg);
719
720 var->base = s->base;
721 var->limit = s->limit;
722 var->selector = s->selector;
723 var->type = s->attrib & SVM_SELECTOR_TYPE_MASK;
724 var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1;
725 var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
726 var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1;
727 var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1;
728 var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
729 var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
730 var->g = (s->attrib >> SVM_SELECTOR_G_SHIFT) & 1;
731 var->unusable = !var->present;
732 }
733
734 static void svm_get_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
735 {
736 struct vcpu_svm *svm = to_svm(vcpu);
737
738 dt->limit = svm->vmcb->save.idtr.limit;
739 dt->base = svm->vmcb->save.idtr.base;
740 }
741
742 static void svm_set_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
743 {
744 struct vcpu_svm *svm = to_svm(vcpu);
745
746 svm->vmcb->save.idtr.limit = dt->limit;
747 svm->vmcb->save.idtr.base = dt->base ;
748 }
749
750 static void svm_get_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
751 {
752 struct vcpu_svm *svm = to_svm(vcpu);
753
754 dt->limit = svm->vmcb->save.gdtr.limit;
755 dt->base = svm->vmcb->save.gdtr.base;
756 }
757
758 static void svm_set_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
759 {
760 struct vcpu_svm *svm = to_svm(vcpu);
761
762 svm->vmcb->save.gdtr.limit = dt->limit;
763 svm->vmcb->save.gdtr.base = dt->base ;
764 }
765
766 static void svm_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
767 {
768 }
769
770 static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
771 {
772 struct vcpu_svm *svm = to_svm(vcpu);
773
774 #ifdef CONFIG_X86_64
775 if (vcpu->arch.shadow_efer & EFER_LME) {
776 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
777 vcpu->arch.shadow_efer |= EFER_LMA;
778 svm->vmcb->save.efer |= EFER_LMA | EFER_LME;
779 }
780
781 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG)) {
782 vcpu->arch.shadow_efer &= ~EFER_LMA;
783 svm->vmcb->save.efer &= ~(EFER_LMA | EFER_LME);
784 }
785 }
786 #endif
787 if ((vcpu->arch.cr0 & X86_CR0_TS) && !(cr0 & X86_CR0_TS)) {
788 svm->vmcb->control.intercept_exceptions &= ~(1 << NM_VECTOR);
789 vcpu->fpu_active = 1;
790 }
791
792 vcpu->arch.cr0 = cr0;
793 cr0 |= X86_CR0_PG | X86_CR0_WP;
794 cr0 &= ~(X86_CR0_CD | X86_CR0_NW);
795 if (!vcpu->fpu_active) {
796 svm->vmcb->control.intercept_exceptions |= (1 << NM_VECTOR);
797 cr0 |= X86_CR0_TS;
798 }
799 svm->vmcb->save.cr0 = cr0;
800 }
801
802 static void svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
803 {
804 vcpu->arch.cr4 = cr4;
805 to_svm(vcpu)->vmcb->save.cr4 = cr4 | X86_CR4_PAE;
806 }
807
808 static void svm_set_segment(struct kvm_vcpu *vcpu,
809 struct kvm_segment *var, int seg)
810 {
811 struct vcpu_svm *svm = to_svm(vcpu);
812 struct vmcb_seg *s = svm_seg(vcpu, seg);
813
814 s->base = var->base;
815 s->limit = var->limit;
816 s->selector = var->selector;
817 if (var->unusable)
818 s->attrib = 0;
819 else {
820 s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK);
821 s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT;
822 s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT;
823 s->attrib |= (var->present & 1) << SVM_SELECTOR_P_SHIFT;
824 s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT;
825 s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT;
826 s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT;
827 s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT;
828 }
829 if (seg == VCPU_SREG_CS)
830 svm->vmcb->save.cpl
831 = (svm->vmcb->save.cs.attrib
832 >> SVM_SELECTOR_DPL_SHIFT) & 3;
833
834 }
835
836 /* FIXME:
837
838 svm(vcpu)->vmcb->control.int_ctl &= ~V_TPR_MASK;
839 svm(vcpu)->vmcb->control.int_ctl |= (sregs->cr8 & V_TPR_MASK);
840
841 */
842
843 static int svm_guest_debug(struct kvm_vcpu *vcpu, struct kvm_debug_guest *dbg)
844 {
845 return -EOPNOTSUPP;
846 }
847
848 static int svm_get_irq(struct kvm_vcpu *vcpu)
849 {
850 struct vcpu_svm *svm = to_svm(vcpu);
851 u32 exit_int_info = svm->vmcb->control.exit_int_info;
852
853 if (is_external_interrupt(exit_int_info))
854 return exit_int_info & SVM_EVTINJ_VEC_MASK;
855 return -1;
856 }
857
858 static void load_host_msrs(struct kvm_vcpu *vcpu)
859 {
860 #ifdef CONFIG_X86_64
861 wrmsrl(MSR_GS_BASE, to_svm(vcpu)->host_gs_base);
862 #endif
863 }
864
865 static void save_host_msrs(struct kvm_vcpu *vcpu)
866 {
867 #ifdef CONFIG_X86_64
868 rdmsrl(MSR_GS_BASE, to_svm(vcpu)->host_gs_base);
869 #endif
870 }
871
872 static void new_asid(struct vcpu_svm *svm, struct svm_cpu_data *svm_data)
873 {
874 if (svm_data->next_asid > svm_data->max_asid) {
875 ++svm_data->asid_generation;
876 svm_data->next_asid = 1;
877 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID;
878 }
879
880 svm->vcpu.cpu = svm_data->cpu;
881 svm->asid_generation = svm_data->asid_generation;
882 svm->vmcb->control.asid = svm_data->next_asid++;
883 }
884
885 static unsigned long svm_get_dr(struct kvm_vcpu *vcpu, int dr)
886 {
887 return to_svm(vcpu)->db_regs[dr];
888 }
889
890 static void svm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long value,
891 int *exception)
892 {
893 struct vcpu_svm *svm = to_svm(vcpu);
894
895 *exception = 0;
896
897 if (svm->vmcb->save.dr7 & DR7_GD_MASK) {
898 svm->vmcb->save.dr7 &= ~DR7_GD_MASK;
899 svm->vmcb->save.dr6 |= DR6_BD_MASK;
900 *exception = DB_VECTOR;
901 return;
902 }
903
904 switch (dr) {
905 case 0 ... 3:
906 svm->db_regs[dr] = value;
907 return;
908 case 4 ... 5:
909 if (vcpu->arch.cr4 & X86_CR4_DE) {
910 *exception = UD_VECTOR;
911 return;
912 }
913 case 7: {
914 if (value & ~((1ULL << 32) - 1)) {
915 *exception = GP_VECTOR;
916 return;
917 }
918 svm->vmcb->save.dr7 = value;
919 return;
920 }
921 default:
922 printk(KERN_DEBUG "%s: unexpected dr %u\n",
923 __FUNCTION__, dr);
924 *exception = UD_VECTOR;
925 return;
926 }
927 }
928
929 static int pf_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
930 {
931 u32 exit_int_info = svm->vmcb->control.exit_int_info;
932 struct kvm *kvm = svm->vcpu.kvm;
933 u64 fault_address;
934 u32 error_code;
935
936 if (!irqchip_in_kernel(kvm) &&
937 is_external_interrupt(exit_int_info))
938 push_irq(&svm->vcpu, exit_int_info & SVM_EVTINJ_VEC_MASK);
939
940 fault_address = svm->vmcb->control.exit_info_2;
941 error_code = svm->vmcb->control.exit_info_1;
942 return kvm_mmu_page_fault(&svm->vcpu, fault_address, error_code);
943 }
944
945 static int ud_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
946 {
947 int er;
948
949 er = emulate_instruction(&svm->vcpu, kvm_run, 0, 0, EMULTYPE_TRAP_UD);
950 if (er != EMULATE_DONE)
951 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
952 return 1;
953 }
954
955 static int nm_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
956 {
957 svm->vmcb->control.intercept_exceptions &= ~(1 << NM_VECTOR);
958 if (!(svm->vcpu.arch.cr0 & X86_CR0_TS))
959 svm->vmcb->save.cr0 &= ~X86_CR0_TS;
960 svm->vcpu.fpu_active = 1;
961
962 return 1;
963 }
964
965 static int shutdown_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
966 {
967 /*
968 * VMCB is undefined after a SHUTDOWN intercept
969 * so reinitialize it.
970 */
971 clear_page(svm->vmcb);
972 init_vmcb(svm->vmcb);
973
974 kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
975 return 0;
976 }
977
978 static int io_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
979 {
980 u32 io_info = svm->vmcb->control.exit_info_1; /* address size bug? */
981 int size, down, in, string, rep;
982 unsigned port;
983
984 ++svm->vcpu.stat.io_exits;
985
986 svm->next_rip = svm->vmcb->control.exit_info_2;
987
988 string = (io_info & SVM_IOIO_STR_MASK) != 0;
989
990 if (string) {
991 if (emulate_instruction(&svm->vcpu,
992 kvm_run, 0, 0, 0) == EMULATE_DO_MMIO)
993 return 0;
994 return 1;
995 }
996
997 in = (io_info & SVM_IOIO_TYPE_MASK) != 0;
998 port = io_info >> 16;
999 size = (io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT;
1000 rep = (io_info & SVM_IOIO_REP_MASK) != 0;
1001 down = (svm->vmcb->save.rflags & X86_EFLAGS_DF) != 0;
1002
1003 return kvm_emulate_pio(&svm->vcpu, kvm_run, in, size, port);
1004 }
1005
1006 static int nop_on_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1007 {
1008 return 1;
1009 }
1010
1011 static int halt_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1012 {
1013 svm->next_rip = svm->vmcb->save.rip + 1;
1014 skip_emulated_instruction(&svm->vcpu);
1015 return kvm_emulate_halt(&svm->vcpu);
1016 }
1017
1018 static int vmmcall_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1019 {
1020 svm->next_rip = svm->vmcb->save.rip + 3;
1021 skip_emulated_instruction(&svm->vcpu);
1022 kvm_emulate_hypercall(&svm->vcpu);
1023 return 1;
1024 }
1025
1026 static int invalid_op_interception(struct vcpu_svm *svm,
1027 struct kvm_run *kvm_run)
1028 {
1029 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
1030 return 1;
1031 }
1032
1033 static int task_switch_interception(struct vcpu_svm *svm,
1034 struct kvm_run *kvm_run)
1035 {
1036 pr_unimpl(&svm->vcpu, "%s: task switch is unsupported\n", __FUNCTION__);
1037 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
1038 return 0;
1039 }
1040
1041 static int cpuid_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1042 {
1043 svm->next_rip = svm->vmcb->save.rip + 2;
1044 kvm_emulate_cpuid(&svm->vcpu);
1045 return 1;
1046 }
1047
1048 static int emulate_on_interception(struct vcpu_svm *svm,
1049 struct kvm_run *kvm_run)
1050 {
1051 if (emulate_instruction(&svm->vcpu, NULL, 0, 0, 0) != EMULATE_DONE)
1052 pr_unimpl(&svm->vcpu, "%s: failed\n", __FUNCTION__);
1053 return 1;
1054 }
1055
1056 static int cr8_write_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1057 {
1058 emulate_instruction(&svm->vcpu, NULL, 0, 0, 0);
1059 if (irqchip_in_kernel(svm->vcpu.kvm))
1060 return 1;
1061 kvm_run->exit_reason = KVM_EXIT_SET_TPR;
1062 return 0;
1063 }
1064
1065 static int svm_get_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 *data)
1066 {
1067 struct vcpu_svm *svm = to_svm(vcpu);
1068
1069 switch (ecx) {
1070 case MSR_IA32_TIME_STAMP_COUNTER: {
1071 u64 tsc;
1072
1073 rdtscll(tsc);
1074 *data = svm->vmcb->control.tsc_offset + tsc;
1075 break;
1076 }
1077 case MSR_K6_STAR:
1078 *data = svm->vmcb->save.star;
1079 break;
1080 #ifdef CONFIG_X86_64
1081 case MSR_LSTAR:
1082 *data = svm->vmcb->save.lstar;
1083 break;
1084 case MSR_CSTAR:
1085 *data = svm->vmcb->save.cstar;
1086 break;
1087 case MSR_KERNEL_GS_BASE:
1088 *data = svm->vmcb->save.kernel_gs_base;
1089 break;
1090 case MSR_SYSCALL_MASK:
1091 *data = svm->vmcb->save.sfmask;
1092 break;
1093 #endif
1094 case MSR_IA32_SYSENTER_CS:
1095 *data = svm->vmcb->save.sysenter_cs;
1096 break;
1097 case MSR_IA32_SYSENTER_EIP:
1098 *data = svm->vmcb->save.sysenter_eip;
1099 break;
1100 case MSR_IA32_SYSENTER_ESP:
1101 *data = svm->vmcb->save.sysenter_esp;
1102 break;
1103 /* Nobody will change the following 5 values in the VMCB so
1104 we can safely return them on rdmsr. They will always be 0
1105 until LBRV is implemented. */
1106 case MSR_IA32_DEBUGCTLMSR:
1107 *data = svm->vmcb->save.dbgctl;
1108 break;
1109 case MSR_IA32_LASTBRANCHFROMIP:
1110 *data = svm->vmcb->save.br_from;
1111 break;
1112 case MSR_IA32_LASTBRANCHTOIP:
1113 *data = svm->vmcb->save.br_to;
1114 break;
1115 case MSR_IA32_LASTINTFROMIP:
1116 *data = svm->vmcb->save.last_excp_from;
1117 break;
1118 case MSR_IA32_LASTINTTOIP:
1119 *data = svm->vmcb->save.last_excp_to;
1120 break;
1121 default:
1122 return kvm_get_msr_common(vcpu, ecx, data);
1123 }
1124 return 0;
1125 }
1126
1127 static int rdmsr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1128 {
1129 u32 ecx = svm->vcpu.arch.regs[VCPU_REGS_RCX];
1130 u64 data;
1131
1132 if (svm_get_msr(&svm->vcpu, ecx, &data))
1133 kvm_inject_gp(&svm->vcpu, 0);
1134 else {
1135 svm->vmcb->save.rax = data & 0xffffffff;
1136 svm->vcpu.arch.regs[VCPU_REGS_RDX] = data >> 32;
1137 svm->next_rip = svm->vmcb->save.rip + 2;
1138 skip_emulated_instruction(&svm->vcpu);
1139 }
1140 return 1;
1141 }
1142
1143 static int svm_set_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 data)
1144 {
1145 struct vcpu_svm *svm = to_svm(vcpu);
1146
1147 switch (ecx) {
1148 case MSR_IA32_TIME_STAMP_COUNTER: {
1149 u64 tsc;
1150
1151 rdtscll(tsc);
1152 svm->vmcb->control.tsc_offset = data - tsc;
1153 break;
1154 }
1155 case MSR_K6_STAR:
1156 svm->vmcb->save.star = data;
1157 break;
1158 #ifdef CONFIG_X86_64
1159 case MSR_LSTAR:
1160 svm->vmcb->save.lstar = data;
1161 break;
1162 case MSR_CSTAR:
1163 svm->vmcb->save.cstar = data;
1164 break;
1165 case MSR_KERNEL_GS_BASE:
1166 svm->vmcb->save.kernel_gs_base = data;
1167 break;
1168 case MSR_SYSCALL_MASK:
1169 svm->vmcb->save.sfmask = data;
1170 break;
1171 #endif
1172 case MSR_IA32_SYSENTER_CS:
1173 svm->vmcb->save.sysenter_cs = data;
1174 break;
1175 case MSR_IA32_SYSENTER_EIP:
1176 svm->vmcb->save.sysenter_eip = data;
1177 break;
1178 case MSR_IA32_SYSENTER_ESP:
1179 svm->vmcb->save.sysenter_esp = data;
1180 break;
1181 case MSR_IA32_DEBUGCTLMSR:
1182 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1183 __FUNCTION__, data);
1184 break;
1185 case MSR_K7_EVNTSEL0:
1186 case MSR_K7_EVNTSEL1:
1187 case MSR_K7_EVNTSEL2:
1188 case MSR_K7_EVNTSEL3:
1189 /*
1190 * only support writing 0 to the performance counters for now
1191 * to make Windows happy. Should be replaced by a real
1192 * performance counter emulation later.
1193 */
1194 if (data != 0)
1195 goto unhandled;
1196 break;
1197 default:
1198 unhandled:
1199 return kvm_set_msr_common(vcpu, ecx, data);
1200 }
1201 return 0;
1202 }
1203
1204 static int wrmsr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1205 {
1206 u32 ecx = svm->vcpu.arch.regs[VCPU_REGS_RCX];
1207 u64 data = (svm->vmcb->save.rax & -1u)
1208 | ((u64)(svm->vcpu.arch.regs[VCPU_REGS_RDX] & -1u) << 32);
1209 svm->next_rip = svm->vmcb->save.rip + 2;
1210 if (svm_set_msr(&svm->vcpu, ecx, data))
1211 kvm_inject_gp(&svm->vcpu, 0);
1212 else
1213 skip_emulated_instruction(&svm->vcpu);
1214 return 1;
1215 }
1216
1217 static int msr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1218 {
1219 if (svm->vmcb->control.exit_info_1)
1220 return wrmsr_interception(svm, kvm_run);
1221 else
1222 return rdmsr_interception(svm, kvm_run);
1223 }
1224
1225 static int interrupt_window_interception(struct vcpu_svm *svm,
1226 struct kvm_run *kvm_run)
1227 {
1228 svm->vmcb->control.intercept &= ~(1ULL << INTERCEPT_VINTR);
1229 svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
1230 /*
1231 * If the user space waits to inject interrupts, exit as soon as
1232 * possible
1233 */
1234 if (kvm_run->request_interrupt_window &&
1235 !svm->vcpu.arch.irq_summary) {
1236 ++svm->vcpu.stat.irq_window_exits;
1237 kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
1238 return 0;
1239 }
1240
1241 return 1;
1242 }
1243
1244 static int (*svm_exit_handlers[])(struct vcpu_svm *svm,
1245 struct kvm_run *kvm_run) = {
1246 [SVM_EXIT_READ_CR0] = emulate_on_interception,
1247 [SVM_EXIT_READ_CR3] = emulate_on_interception,
1248 [SVM_EXIT_READ_CR4] = emulate_on_interception,
1249 [SVM_EXIT_READ_CR8] = emulate_on_interception,
1250 /* for now: */
1251 [SVM_EXIT_WRITE_CR0] = emulate_on_interception,
1252 [SVM_EXIT_WRITE_CR3] = emulate_on_interception,
1253 [SVM_EXIT_WRITE_CR4] = emulate_on_interception,
1254 [SVM_EXIT_WRITE_CR8] = cr8_write_interception,
1255 [SVM_EXIT_READ_DR0] = emulate_on_interception,
1256 [SVM_EXIT_READ_DR1] = emulate_on_interception,
1257 [SVM_EXIT_READ_DR2] = emulate_on_interception,
1258 [SVM_EXIT_READ_DR3] = emulate_on_interception,
1259 [SVM_EXIT_WRITE_DR0] = emulate_on_interception,
1260 [SVM_EXIT_WRITE_DR1] = emulate_on_interception,
1261 [SVM_EXIT_WRITE_DR2] = emulate_on_interception,
1262 [SVM_EXIT_WRITE_DR3] = emulate_on_interception,
1263 [SVM_EXIT_WRITE_DR5] = emulate_on_interception,
1264 [SVM_EXIT_WRITE_DR7] = emulate_on_interception,
1265 [SVM_EXIT_EXCP_BASE + UD_VECTOR] = ud_interception,
1266 [SVM_EXIT_EXCP_BASE + PF_VECTOR] = pf_interception,
1267 [SVM_EXIT_EXCP_BASE + NM_VECTOR] = nm_interception,
1268 [SVM_EXIT_INTR] = nop_on_interception,
1269 [SVM_EXIT_NMI] = nop_on_interception,
1270 [SVM_EXIT_SMI] = nop_on_interception,
1271 [SVM_EXIT_INIT] = nop_on_interception,
1272 [SVM_EXIT_VINTR] = interrupt_window_interception,
1273 /* [SVM_EXIT_CR0_SEL_WRITE] = emulate_on_interception, */
1274 [SVM_EXIT_CPUID] = cpuid_interception,
1275 [SVM_EXIT_INVD] = emulate_on_interception,
1276 [SVM_EXIT_HLT] = halt_interception,
1277 [SVM_EXIT_INVLPG] = emulate_on_interception,
1278 [SVM_EXIT_INVLPGA] = invalid_op_interception,
1279 [SVM_EXIT_IOIO] = io_interception,
1280 [SVM_EXIT_MSR] = msr_interception,
1281 [SVM_EXIT_TASK_SWITCH] = task_switch_interception,
1282 [SVM_EXIT_SHUTDOWN] = shutdown_interception,
1283 [SVM_EXIT_VMRUN] = invalid_op_interception,
1284 [SVM_EXIT_VMMCALL] = vmmcall_interception,
1285 [SVM_EXIT_VMLOAD] = invalid_op_interception,
1286 [SVM_EXIT_VMSAVE] = invalid_op_interception,
1287 [SVM_EXIT_STGI] = invalid_op_interception,
1288 [SVM_EXIT_CLGI] = invalid_op_interception,
1289 [SVM_EXIT_SKINIT] = invalid_op_interception,
1290 [SVM_EXIT_WBINVD] = emulate_on_interception,
1291 [SVM_EXIT_MONITOR] = invalid_op_interception,
1292 [SVM_EXIT_MWAIT] = invalid_op_interception,
1293 };
1294
1295
1296 static int handle_exit(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1297 {
1298 struct vcpu_svm *svm = to_svm(vcpu);
1299 u32 exit_code = svm->vmcb->control.exit_code;
1300
1301 kvm_reput_irq(svm);
1302
1303 if (svm->vmcb->control.exit_code == SVM_EXIT_ERR) {
1304 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
1305 kvm_run->fail_entry.hardware_entry_failure_reason
1306 = svm->vmcb->control.exit_code;
1307 return 0;
1308 }
1309
1310 if (is_external_interrupt(svm->vmcb->control.exit_int_info) &&
1311 exit_code != SVM_EXIT_EXCP_BASE + PF_VECTOR)
1312 printk(KERN_ERR "%s: unexpected exit_ini_info 0x%x "
1313 "exit_code 0x%x\n",
1314 __FUNCTION__, svm->vmcb->control.exit_int_info,
1315 exit_code);
1316
1317 if (exit_code >= ARRAY_SIZE(svm_exit_handlers)
1318 || !svm_exit_handlers[exit_code]) {
1319 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
1320 kvm_run->hw.hardware_exit_reason = exit_code;
1321 return 0;
1322 }
1323
1324 return svm_exit_handlers[exit_code](svm, kvm_run);
1325 }
1326
1327 static void reload_tss(struct kvm_vcpu *vcpu)
1328 {
1329 int cpu = raw_smp_processor_id();
1330
1331 struct svm_cpu_data *svm_data = per_cpu(svm_data, cpu);
1332 svm_data->tss_desc->type = 9; /* available 32/64-bit TSS */
1333 load_TR_desc();
1334 }
1335
1336 static void pre_svm_run(struct vcpu_svm *svm)
1337 {
1338 int cpu = raw_smp_processor_id();
1339
1340 struct svm_cpu_data *svm_data = per_cpu(svm_data, cpu);
1341
1342 svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
1343 if (svm->vcpu.cpu != cpu ||
1344 svm->asid_generation != svm_data->asid_generation)
1345 new_asid(svm, svm_data);
1346 }
1347
1348
1349 static inline void svm_inject_irq(struct vcpu_svm *svm, int irq)
1350 {
1351 struct vmcb_control_area *control;
1352
1353 control = &svm->vmcb->control;
1354 control->int_vector = irq;
1355 control->int_ctl &= ~V_INTR_PRIO_MASK;
1356 control->int_ctl |= V_IRQ_MASK |
1357 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT);
1358 }
1359
1360 static void svm_set_irq(struct kvm_vcpu *vcpu, int irq)
1361 {
1362 struct vcpu_svm *svm = to_svm(vcpu);
1363
1364 svm_inject_irq(svm, irq);
1365 }
1366
1367 static void svm_intr_assist(struct kvm_vcpu *vcpu)
1368 {
1369 struct vcpu_svm *svm = to_svm(vcpu);
1370 struct vmcb *vmcb = svm->vmcb;
1371 int intr_vector = -1;
1372
1373 if ((vmcb->control.exit_int_info & SVM_EVTINJ_VALID) &&
1374 ((vmcb->control.exit_int_info & SVM_EVTINJ_TYPE_MASK) == 0)) {
1375 intr_vector = vmcb->control.exit_int_info &
1376 SVM_EVTINJ_VEC_MASK;
1377 vmcb->control.exit_int_info = 0;
1378 svm_inject_irq(svm, intr_vector);
1379 return;
1380 }
1381
1382 if (vmcb->control.int_ctl & V_IRQ_MASK)
1383 return;
1384
1385 if (!kvm_cpu_has_interrupt(vcpu))
1386 return;
1387
1388 if (!(vmcb->save.rflags & X86_EFLAGS_IF) ||
1389 (vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK) ||
1390 (vmcb->control.event_inj & SVM_EVTINJ_VALID)) {
1391 /* unable to deliver irq, set pending irq */
1392 vmcb->control.intercept |= (1ULL << INTERCEPT_VINTR);
1393 svm_inject_irq(svm, 0x0);
1394 return;
1395 }
1396 /* Okay, we can deliver the interrupt: grab it and update PIC state. */
1397 intr_vector = kvm_cpu_get_interrupt(vcpu);
1398 svm_inject_irq(svm, intr_vector);
1399 kvm_timer_intr_post(vcpu, intr_vector);
1400 }
1401
1402 static void kvm_reput_irq(struct vcpu_svm *svm)
1403 {
1404 struct vmcb_control_area *control = &svm->vmcb->control;
1405
1406 if ((control->int_ctl & V_IRQ_MASK)
1407 && !irqchip_in_kernel(svm->vcpu.kvm)) {
1408 control->int_ctl &= ~V_IRQ_MASK;
1409 push_irq(&svm->vcpu, control->int_vector);
1410 }
1411
1412 svm->vcpu.arch.interrupt_window_open =
1413 !(control->int_state & SVM_INTERRUPT_SHADOW_MASK);
1414 }
1415
1416 static void svm_do_inject_vector(struct vcpu_svm *svm)
1417 {
1418 struct kvm_vcpu *vcpu = &svm->vcpu;
1419 int word_index = __ffs(vcpu->arch.irq_summary);
1420 int bit_index = __ffs(vcpu->arch.irq_pending[word_index]);
1421 int irq = word_index * BITS_PER_LONG + bit_index;
1422
1423 clear_bit(bit_index, &vcpu->arch.irq_pending[word_index]);
1424 if (!vcpu->arch.irq_pending[word_index])
1425 clear_bit(word_index, &vcpu->arch.irq_summary);
1426 svm_inject_irq(svm, irq);
1427 }
1428
1429 static void do_interrupt_requests(struct kvm_vcpu *vcpu,
1430 struct kvm_run *kvm_run)
1431 {
1432 struct vcpu_svm *svm = to_svm(vcpu);
1433 struct vmcb_control_area *control = &svm->vmcb->control;
1434
1435 svm->vcpu.arch.interrupt_window_open =
1436 (!(control->int_state & SVM_INTERRUPT_SHADOW_MASK) &&
1437 (svm->vmcb->save.rflags & X86_EFLAGS_IF));
1438
1439 if (svm->vcpu.arch.interrupt_window_open && svm->vcpu.arch.irq_summary)
1440 /*
1441 * If interrupts enabled, and not blocked by sti or mov ss. Good.
1442 */
1443 svm_do_inject_vector(svm);
1444
1445 /*
1446 * Interrupts blocked. Wait for unblock.
1447 */
1448 if (!svm->vcpu.arch.interrupt_window_open &&
1449 (svm->vcpu.arch.irq_summary || kvm_run->request_interrupt_window))
1450 control->intercept |= 1ULL << INTERCEPT_VINTR;
1451 else
1452 control->intercept &= ~(1ULL << INTERCEPT_VINTR);
1453 }
1454
1455 static int svm_set_tss_addr(struct kvm *kvm, unsigned int addr)
1456 {
1457 return 0;
1458 }
1459
1460 static void save_db_regs(unsigned long *db_regs)
1461 {
1462 asm volatile ("mov %%dr0, %0" : "=r"(db_regs[0]));
1463 asm volatile ("mov %%dr1, %0" : "=r"(db_regs[1]));
1464 asm volatile ("mov %%dr2, %0" : "=r"(db_regs[2]));
1465 asm volatile ("mov %%dr3, %0" : "=r"(db_regs[3]));
1466 }
1467
1468 static void load_db_regs(unsigned long *db_regs)
1469 {
1470 asm volatile ("mov %0, %%dr0" : : "r"(db_regs[0]));
1471 asm volatile ("mov %0, %%dr1" : : "r"(db_regs[1]));
1472 asm volatile ("mov %0, %%dr2" : : "r"(db_regs[2]));
1473 asm volatile ("mov %0, %%dr3" : : "r"(db_regs[3]));
1474 }
1475
1476 static void svm_flush_tlb(struct kvm_vcpu *vcpu)
1477 {
1478 force_new_asid(vcpu);
1479 }
1480
1481 static void svm_prepare_guest_switch(struct kvm_vcpu *vcpu)
1482 {
1483 }
1484
1485 static void svm_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1486 {
1487 struct vcpu_svm *svm = to_svm(vcpu);
1488 u16 fs_selector;
1489 u16 gs_selector;
1490 u16 ldt_selector;
1491
1492 pre_svm_run(svm);
1493
1494 save_host_msrs(vcpu);
1495 fs_selector = read_fs();
1496 gs_selector = read_gs();
1497 ldt_selector = read_ldt();
1498 svm->host_cr2 = kvm_read_cr2();
1499 svm->host_dr6 = read_dr6();
1500 svm->host_dr7 = read_dr7();
1501 svm->vmcb->save.cr2 = vcpu->arch.cr2;
1502
1503 if (svm->vmcb->save.dr7 & 0xff) {
1504 write_dr7(0);
1505 save_db_regs(svm->host_db_regs);
1506 load_db_regs(svm->db_regs);
1507 }
1508
1509 clgi();
1510
1511 local_irq_enable();
1512
1513 asm volatile (
1514 #ifdef CONFIG_X86_64
1515 "push %%rbp; \n\t"
1516 #else
1517 "push %%ebp; \n\t"
1518 #endif
1519
1520 #ifdef CONFIG_X86_64
1521 "mov %c[rbx](%[svm]), %%rbx \n\t"
1522 "mov %c[rcx](%[svm]), %%rcx \n\t"
1523 "mov %c[rdx](%[svm]), %%rdx \n\t"
1524 "mov %c[rsi](%[svm]), %%rsi \n\t"
1525 "mov %c[rdi](%[svm]), %%rdi \n\t"
1526 "mov %c[rbp](%[svm]), %%rbp \n\t"
1527 "mov %c[r8](%[svm]), %%r8 \n\t"
1528 "mov %c[r9](%[svm]), %%r9 \n\t"
1529 "mov %c[r10](%[svm]), %%r10 \n\t"
1530 "mov %c[r11](%[svm]), %%r11 \n\t"
1531 "mov %c[r12](%[svm]), %%r12 \n\t"
1532 "mov %c[r13](%[svm]), %%r13 \n\t"
1533 "mov %c[r14](%[svm]), %%r14 \n\t"
1534 "mov %c[r15](%[svm]), %%r15 \n\t"
1535 #else
1536 "mov %c[rbx](%[svm]), %%ebx \n\t"
1537 "mov %c[rcx](%[svm]), %%ecx \n\t"
1538 "mov %c[rdx](%[svm]), %%edx \n\t"
1539 "mov %c[rsi](%[svm]), %%esi \n\t"
1540 "mov %c[rdi](%[svm]), %%edi \n\t"
1541 "mov %c[rbp](%[svm]), %%ebp \n\t"
1542 #endif
1543
1544 #ifdef CONFIG_X86_64
1545 /* Enter guest mode */
1546 "push %%rax \n\t"
1547 "mov %c[vmcb](%[svm]), %%rax \n\t"
1548 SVM_VMLOAD "\n\t"
1549 SVM_VMRUN "\n\t"
1550 SVM_VMSAVE "\n\t"
1551 "pop %%rax \n\t"
1552 #else
1553 /* Enter guest mode */
1554 "push %%eax \n\t"
1555 "mov %c[vmcb](%[svm]), %%eax \n\t"
1556 SVM_VMLOAD "\n\t"
1557 SVM_VMRUN "\n\t"
1558 SVM_VMSAVE "\n\t"
1559 "pop %%eax \n\t"
1560 #endif
1561
1562 /* Save guest registers, load host registers */
1563 #ifdef CONFIG_X86_64
1564 "mov %%rbx, %c[rbx](%[svm]) \n\t"
1565 "mov %%rcx, %c[rcx](%[svm]) \n\t"
1566 "mov %%rdx, %c[rdx](%[svm]) \n\t"
1567 "mov %%rsi, %c[rsi](%[svm]) \n\t"
1568 "mov %%rdi, %c[rdi](%[svm]) \n\t"
1569 "mov %%rbp, %c[rbp](%[svm]) \n\t"
1570 "mov %%r8, %c[r8](%[svm]) \n\t"
1571 "mov %%r9, %c[r9](%[svm]) \n\t"
1572 "mov %%r10, %c[r10](%[svm]) \n\t"
1573 "mov %%r11, %c[r11](%[svm]) \n\t"
1574 "mov %%r12, %c[r12](%[svm]) \n\t"
1575 "mov %%r13, %c[r13](%[svm]) \n\t"
1576 "mov %%r14, %c[r14](%[svm]) \n\t"
1577 "mov %%r15, %c[r15](%[svm]) \n\t"
1578
1579 "pop %%rbp; \n\t"
1580 #else
1581 "mov %%ebx, %c[rbx](%[svm]) \n\t"
1582 "mov %%ecx, %c[rcx](%[svm]) \n\t"
1583 "mov %%edx, %c[rdx](%[svm]) \n\t"
1584 "mov %%esi, %c[rsi](%[svm]) \n\t"
1585 "mov %%edi, %c[rdi](%[svm]) \n\t"
1586 "mov %%ebp, %c[rbp](%[svm]) \n\t"
1587
1588 "pop %%ebp; \n\t"
1589 #endif
1590 :
1591 : [svm]"a"(svm),
1592 [vmcb]"i"(offsetof(struct vcpu_svm, vmcb_pa)),
1593 [rbx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBX])),
1594 [rcx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RCX])),
1595 [rdx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDX])),
1596 [rsi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RSI])),
1597 [rdi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDI])),
1598 [rbp]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBP]))
1599 #ifdef CONFIG_X86_64
1600 , [r8]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R8])),
1601 [r9]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R9])),
1602 [r10]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R10])),
1603 [r11]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R11])),
1604 [r12]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R12])),
1605 [r13]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R13])),
1606 [r14]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R14])),
1607 [r15]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R15]))
1608 #endif
1609 : "cc", "memory"
1610 #ifdef CONFIG_X86_64
1611 , "rbx", "rcx", "rdx", "rsi", "rdi"
1612 , "r8", "r9", "r10", "r11" , "r12", "r13", "r14", "r15"
1613 #else
1614 , "ebx", "ecx", "edx" , "esi", "edi"
1615 #endif
1616 );
1617
1618 if ((svm->vmcb->save.dr7 & 0xff))
1619 load_db_regs(svm->host_db_regs);
1620
1621 vcpu->arch.cr2 = svm->vmcb->save.cr2;
1622
1623 write_dr6(svm->host_dr6);
1624 write_dr7(svm->host_dr7);
1625 kvm_write_cr2(svm->host_cr2);
1626
1627 load_fs(fs_selector);
1628 load_gs(gs_selector);
1629 load_ldt(ldt_selector);
1630 load_host_msrs(vcpu);
1631
1632 reload_tss(vcpu);
1633
1634 local_irq_disable();
1635
1636 stgi();
1637
1638 svm->next_rip = 0;
1639 }
1640
1641 static void svm_set_cr3(struct kvm_vcpu *vcpu, unsigned long root)
1642 {
1643 struct vcpu_svm *svm = to_svm(vcpu);
1644
1645 svm->vmcb->save.cr3 = root;
1646 force_new_asid(vcpu);
1647
1648 if (vcpu->fpu_active) {
1649 svm->vmcb->control.intercept_exceptions |= (1 << NM_VECTOR);
1650 svm->vmcb->save.cr0 |= X86_CR0_TS;
1651 vcpu->fpu_active = 0;
1652 }
1653 }
1654
1655 static int is_disabled(void)
1656 {
1657 u64 vm_cr;
1658
1659 rdmsrl(MSR_VM_CR, vm_cr);
1660 if (vm_cr & (1 << SVM_VM_CR_SVM_DISABLE))
1661 return 1;
1662
1663 return 0;
1664 }
1665
1666 static void
1667 svm_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
1668 {
1669 /*
1670 * Patch in the VMMCALL instruction:
1671 */
1672 hypercall[0] = 0x0f;
1673 hypercall[1] = 0x01;
1674 hypercall[2] = 0xd9;
1675 }
1676
1677 static void svm_check_processor_compat(void *rtn)
1678 {
1679 *(int *)rtn = 0;
1680 }
1681
1682 static bool svm_cpu_has_accelerated_tpr(void)
1683 {
1684 return false;
1685 }
1686
1687 static struct kvm_x86_ops svm_x86_ops = {
1688 .cpu_has_kvm_support = has_svm,
1689 .disabled_by_bios = is_disabled,
1690 .hardware_setup = svm_hardware_setup,
1691 .hardware_unsetup = svm_hardware_unsetup,
1692 .check_processor_compatibility = svm_check_processor_compat,
1693 .hardware_enable = svm_hardware_enable,
1694 .hardware_disable = svm_hardware_disable,
1695 .cpu_has_accelerated_tpr = svm_cpu_has_accelerated_tpr,
1696
1697 .vcpu_create = svm_create_vcpu,
1698 .vcpu_free = svm_free_vcpu,
1699 .vcpu_reset = svm_vcpu_reset,
1700
1701 .prepare_guest_switch = svm_prepare_guest_switch,
1702 .vcpu_load = svm_vcpu_load,
1703 .vcpu_put = svm_vcpu_put,
1704 .vcpu_decache = svm_vcpu_decache,
1705
1706 .set_guest_debug = svm_guest_debug,
1707 .get_msr = svm_get_msr,
1708 .set_msr = svm_set_msr,
1709 .get_segment_base = svm_get_segment_base,
1710 .get_segment = svm_get_segment,
1711 .set_segment = svm_set_segment,
1712 .get_cs_db_l_bits = kvm_get_cs_db_l_bits,
1713 .decache_cr4_guest_bits = svm_decache_cr4_guest_bits,
1714 .set_cr0 = svm_set_cr0,
1715 .set_cr3 = svm_set_cr3,
1716 .set_cr4 = svm_set_cr4,
1717 .set_efer = svm_set_efer,
1718 .get_idt = svm_get_idt,
1719 .set_idt = svm_set_idt,
1720 .get_gdt = svm_get_gdt,
1721 .set_gdt = svm_set_gdt,
1722 .get_dr = svm_get_dr,
1723 .set_dr = svm_set_dr,
1724 .cache_regs = svm_cache_regs,
1725 .decache_regs = svm_decache_regs,
1726 .get_rflags = svm_get_rflags,
1727 .set_rflags = svm_set_rflags,
1728
1729 .tlb_flush = svm_flush_tlb,
1730
1731 .run = svm_vcpu_run,
1732 .handle_exit = handle_exit,
1733 .skip_emulated_instruction = skip_emulated_instruction,
1734 .patch_hypercall = svm_patch_hypercall,
1735 .get_irq = svm_get_irq,
1736 .set_irq = svm_set_irq,
1737 .queue_exception = svm_queue_exception,
1738 .exception_injected = svm_exception_injected,
1739 .inject_pending_irq = svm_intr_assist,
1740 .inject_pending_vectors = do_interrupt_requests,
1741
1742 .set_tss_addr = svm_set_tss_addr,
1743 };
1744
1745 static int __init svm_init(void)
1746 {
1747 return kvm_init(&svm_x86_ops, sizeof(struct vcpu_svm),
1748 THIS_MODULE);
1749 }
1750
1751 static void __exit svm_exit(void)
1752 {
1753 kvm_exit();
1754 }
1755
1756 module_init(svm_init)
1757 module_exit(svm_exit)
Linux with added FPC-III support