KVM: Add and use pr_unimpl for standard formatting of unimplemented features
[linux-2.6/mini2440.git] / drivers / kvm / svm.c
blob3ec30d7c9b936ee74389167d1a084b136a6c674c
1 /*
2 * Kernel-based Virtual Machine driver for Linux
4 * AMD SVM support
6 * Copyright (C) 2006 Qumranet, Inc.
8 * Authors:
9 * Yaniv Kamay <yaniv@qumranet.com>
10 * Avi Kivity <avi@qumranet.com>
12 * This work is licensed under the terms of the GNU GPL, version 2. See
13 * the COPYING file in the top-level directory.
17 #include "kvm_svm.h"
18 #include "x86_emulate.h"
20 #include <linux/module.h>
21 #include <linux/kernel.h>
22 #include <linux/vmalloc.h>
23 #include <linux/highmem.h>
24 #include <linux/profile.h>
25 #include <linux/sched.h>
27 #include <asm/desc.h>
29 MODULE_AUTHOR("Qumranet");
30 MODULE_LICENSE("GPL");
32 #define IOPM_ALLOC_ORDER 2
33 #define MSRPM_ALLOC_ORDER 1
35 #define DB_VECTOR 1
36 #define UD_VECTOR 6
37 #define GP_VECTOR 13
39 #define DR7_GD_MASK (1 << 13)
40 #define DR6_BD_MASK (1 << 13)
42 #define SEG_TYPE_LDT 2
43 #define SEG_TYPE_BUSY_TSS16 3
45 #define KVM_EFER_LMA (1 << 10)
46 #define KVM_EFER_LME (1 << 8)
48 #define SVM_FEATURE_NPT (1 << 0)
49 #define SVM_FEATURE_LBRV (1 << 1)
50 #define SVM_DEATURE_SVML (1 << 2)
52 static inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu)
54 return container_of(vcpu, struct vcpu_svm, vcpu);
57 unsigned long iopm_base;
58 unsigned long msrpm_base;
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));
69 struct svm_cpu_data {
70 int cpu;
72 u64 asid_generation;
73 u32 max_asid;
74 u32 next_asid;
75 struct kvm_ldttss_desc *tss_desc;
77 struct page *save_area;
80 static DEFINE_PER_CPU(struct svm_cpu_data *, svm_data);
81 static uint32_t svm_features;
83 struct svm_init_data {
84 int cpu;
85 int r;
88 static u32 msrpm_ranges[] = {0, 0xc0000000, 0xc0010000};
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)
94 #define MAX_INST_SIZE 15
96 static inline u32 svm_has(u32 feat)
98 return svm_features & feat;
101 static unsigned get_addr_size(struct vcpu_svm *svm)
103 struct vmcb_save_area *sa = &svm->vmcb->save;
104 u16 cs_attrib;
106 if (!(sa->cr0 & X86_CR0_PE) || (sa->rflags & X86_EFLAGS_VM))
107 return 2;
109 cs_attrib = sa->cs.attrib;
111 return (cs_attrib & SVM_SELECTOR_L_MASK) ? 8 :
112 (cs_attrib & SVM_SELECTOR_DB_MASK) ? 4 : 2;
115 static inline u8 pop_irq(struct kvm_vcpu *vcpu)
117 int word_index = __ffs(vcpu->irq_summary);
118 int bit_index = __ffs(vcpu->irq_pending[word_index]);
119 int irq = word_index * BITS_PER_LONG + bit_index;
121 clear_bit(bit_index, &vcpu->irq_pending[word_index]);
122 if (!vcpu->irq_pending[word_index])
123 clear_bit(word_index, &vcpu->irq_summary);
124 return irq;
127 static inline void push_irq(struct kvm_vcpu *vcpu, u8 irq)
129 set_bit(irq, vcpu->irq_pending);
130 set_bit(irq / BITS_PER_LONG, &vcpu->irq_summary);
133 static inline void clgi(void)
135 asm volatile (SVM_CLGI);
138 static inline void stgi(void)
140 asm volatile (SVM_STGI);
143 static inline void invlpga(unsigned long addr, u32 asid)
145 asm volatile (SVM_INVLPGA :: "a"(addr), "c"(asid));
148 static inline unsigned long kvm_read_cr2(void)
150 unsigned long cr2;
152 asm volatile ("mov %%cr2, %0" : "=r" (cr2));
153 return cr2;
156 static inline void kvm_write_cr2(unsigned long val)
158 asm volatile ("mov %0, %%cr2" :: "r" (val));
161 static inline unsigned long read_dr6(void)
163 unsigned long dr6;
165 asm volatile ("mov %%dr6, %0" : "=r" (dr6));
166 return dr6;
169 static inline void write_dr6(unsigned long val)
171 asm volatile ("mov %0, %%dr6" :: "r" (val));
174 static inline unsigned long read_dr7(void)
176 unsigned long dr7;
178 asm volatile ("mov %%dr7, %0" : "=r" (dr7));
179 return dr7;
182 static inline void write_dr7(unsigned long val)
184 asm volatile ("mov %0, %%dr7" :: "r" (val));
187 static inline void force_new_asid(struct kvm_vcpu *vcpu)
189 to_svm(vcpu)->asid_generation--;
192 static inline void flush_guest_tlb(struct kvm_vcpu *vcpu)
194 force_new_asid(vcpu);
197 static void svm_set_efer(struct kvm_vcpu *vcpu, u64 efer)
199 if (!(efer & KVM_EFER_LMA))
200 efer &= ~KVM_EFER_LME;
202 to_svm(vcpu)->vmcb->save.efer = efer | MSR_EFER_SVME_MASK;
203 vcpu->shadow_efer = efer;
206 static void svm_inject_gp(struct kvm_vcpu *vcpu, unsigned error_code)
208 struct vcpu_svm *svm = to_svm(vcpu);
210 svm->vmcb->control.event_inj = SVM_EVTINJ_VALID |
211 SVM_EVTINJ_VALID_ERR |
212 SVM_EVTINJ_TYPE_EXEPT |
213 GP_VECTOR;
214 svm->vmcb->control.event_inj_err = error_code;
217 static void inject_ud(struct kvm_vcpu *vcpu)
219 to_svm(vcpu)->vmcb->control.event_inj = SVM_EVTINJ_VALID |
220 SVM_EVTINJ_TYPE_EXEPT |
221 UD_VECTOR;
224 static int is_page_fault(uint32_t info)
226 info &= SVM_EVTINJ_VEC_MASK | SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
227 return info == (PF_VECTOR | SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_EXEPT);
230 static int is_external_interrupt(u32 info)
232 info &= SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
233 return info == (SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR);
236 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
238 struct vcpu_svm *svm = to_svm(vcpu);
240 if (!svm->next_rip) {
241 printk(KERN_DEBUG "%s: NOP\n", __FUNCTION__);
242 return;
244 if (svm->next_rip - svm->vmcb->save.rip > MAX_INST_SIZE) {
245 printk(KERN_ERR "%s: ip 0x%llx next 0x%llx\n",
246 __FUNCTION__,
247 svm->vmcb->save.rip,
248 svm->next_rip);
251 vcpu->rip = svm->vmcb->save.rip = svm->next_rip;
252 svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK;
254 vcpu->interrupt_window_open = 1;
257 static int has_svm(void)
259 uint32_t eax, ebx, ecx, edx;
261 if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD) {
262 printk(KERN_INFO "has_svm: not amd\n");
263 return 0;
266 cpuid(0x80000000, &eax, &ebx, &ecx, &edx);
267 if (eax < SVM_CPUID_FUNC) {
268 printk(KERN_INFO "has_svm: can't execute cpuid_8000000a\n");
269 return 0;
272 cpuid(0x80000001, &eax, &ebx, &ecx, &edx);
273 if (!(ecx & (1 << SVM_CPUID_FEATURE_SHIFT))) {
274 printk(KERN_DEBUG "has_svm: svm not available\n");
275 return 0;
277 return 1;
280 static void svm_hardware_disable(void *garbage)
282 struct svm_cpu_data *svm_data
283 = per_cpu(svm_data, raw_smp_processor_id());
285 if (svm_data) {
286 uint64_t efer;
288 wrmsrl(MSR_VM_HSAVE_PA, 0);
289 rdmsrl(MSR_EFER, efer);
290 wrmsrl(MSR_EFER, efer & ~MSR_EFER_SVME_MASK);
291 per_cpu(svm_data, raw_smp_processor_id()) = NULL;
292 __free_page(svm_data->save_area);
293 kfree(svm_data);
297 static void svm_hardware_enable(void *garbage)
300 struct svm_cpu_data *svm_data;
301 uint64_t efer;
302 #ifdef CONFIG_X86_64
303 struct desc_ptr gdt_descr;
304 #else
305 struct Xgt_desc_struct gdt_descr;
306 #endif
307 struct desc_struct *gdt;
308 int me = raw_smp_processor_id();
310 if (!has_svm()) {
311 printk(KERN_ERR "svm_cpu_init: err EOPNOTSUPP on %d\n", me);
312 return;
314 svm_data = per_cpu(svm_data, me);
316 if (!svm_data) {
317 printk(KERN_ERR "svm_cpu_init: svm_data is NULL on %d\n",
318 me);
319 return;
322 svm_data->asid_generation = 1;
323 svm_data->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1;
324 svm_data->next_asid = svm_data->max_asid + 1;
325 svm_features = cpuid_edx(SVM_CPUID_FUNC);
327 asm volatile ( "sgdt %0" : "=m"(gdt_descr) );
328 gdt = (struct desc_struct *)gdt_descr.address;
329 svm_data->tss_desc = (struct kvm_ldttss_desc *)(gdt + GDT_ENTRY_TSS);
331 rdmsrl(MSR_EFER, efer);
332 wrmsrl(MSR_EFER, efer | MSR_EFER_SVME_MASK);
334 wrmsrl(MSR_VM_HSAVE_PA,
335 page_to_pfn(svm_data->save_area) << PAGE_SHIFT);
338 static int svm_cpu_init(int cpu)
340 struct svm_cpu_data *svm_data;
341 int r;
343 svm_data = kzalloc(sizeof(struct svm_cpu_data), GFP_KERNEL);
344 if (!svm_data)
345 return -ENOMEM;
346 svm_data->cpu = cpu;
347 svm_data->save_area = alloc_page(GFP_KERNEL);
348 r = -ENOMEM;
349 if (!svm_data->save_area)
350 goto err_1;
352 per_cpu(svm_data, cpu) = svm_data;
354 return 0;
356 err_1:
357 kfree(svm_data);
358 return r;
362 static void set_msr_interception(u32 *msrpm, unsigned msr,
363 int read, int write)
365 int i;
367 for (i = 0; i < NUM_MSR_MAPS; i++) {
368 if (msr >= msrpm_ranges[i] &&
369 msr < msrpm_ranges[i] + MSRS_IN_RANGE) {
370 u32 msr_offset = (i * MSRS_IN_RANGE + msr -
371 msrpm_ranges[i]) * 2;
373 u32 *base = msrpm + (msr_offset / 32);
374 u32 msr_shift = msr_offset % 32;
375 u32 mask = ((write) ? 0 : 2) | ((read) ? 0 : 1);
376 *base = (*base & ~(0x3 << msr_shift)) |
377 (mask << msr_shift);
378 return;
381 BUG();
384 static __init int svm_hardware_setup(void)
386 int cpu;
387 struct page *iopm_pages;
388 struct page *msrpm_pages;
389 void *iopm_va, *msrpm_va;
390 int r;
392 kvm_emulator_want_group7_invlpg();
394 iopm_pages = alloc_pages(GFP_KERNEL, IOPM_ALLOC_ORDER);
396 if (!iopm_pages)
397 return -ENOMEM;
399 iopm_va = page_address(iopm_pages);
400 memset(iopm_va, 0xff, PAGE_SIZE * (1 << IOPM_ALLOC_ORDER));
401 clear_bit(0x80, iopm_va); /* allow direct access to PC debug port */
402 iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT;
405 msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
407 r = -ENOMEM;
408 if (!msrpm_pages)
409 goto err_1;
411 msrpm_va = page_address(msrpm_pages);
412 memset(msrpm_va, 0xff, PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER));
413 msrpm_base = page_to_pfn(msrpm_pages) << PAGE_SHIFT;
415 #ifdef CONFIG_X86_64
416 set_msr_interception(msrpm_va, MSR_GS_BASE, 1, 1);
417 set_msr_interception(msrpm_va, MSR_FS_BASE, 1, 1);
418 set_msr_interception(msrpm_va, MSR_KERNEL_GS_BASE, 1, 1);
419 set_msr_interception(msrpm_va, MSR_LSTAR, 1, 1);
420 set_msr_interception(msrpm_va, MSR_CSTAR, 1, 1);
421 set_msr_interception(msrpm_va, MSR_SYSCALL_MASK, 1, 1);
422 #endif
423 set_msr_interception(msrpm_va, MSR_K6_STAR, 1, 1);
424 set_msr_interception(msrpm_va, MSR_IA32_SYSENTER_CS, 1, 1);
425 set_msr_interception(msrpm_va, MSR_IA32_SYSENTER_ESP, 1, 1);
426 set_msr_interception(msrpm_va, MSR_IA32_SYSENTER_EIP, 1, 1);
428 for_each_online_cpu(cpu) {
429 r = svm_cpu_init(cpu);
430 if (r)
431 goto err_2;
433 return 0;
435 err_2:
436 __free_pages(msrpm_pages, MSRPM_ALLOC_ORDER);
437 msrpm_base = 0;
438 err_1:
439 __free_pages(iopm_pages, IOPM_ALLOC_ORDER);
440 iopm_base = 0;
441 return r;
444 static __exit void svm_hardware_unsetup(void)
446 __free_pages(pfn_to_page(msrpm_base >> PAGE_SHIFT), MSRPM_ALLOC_ORDER);
447 __free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT), IOPM_ALLOC_ORDER);
448 iopm_base = msrpm_base = 0;
451 static void init_seg(struct vmcb_seg *seg)
453 seg->selector = 0;
454 seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK |
455 SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */
456 seg->limit = 0xffff;
457 seg->base = 0;
460 static void init_sys_seg(struct vmcb_seg *seg, uint32_t type)
462 seg->selector = 0;
463 seg->attrib = SVM_SELECTOR_P_MASK | type;
464 seg->limit = 0xffff;
465 seg->base = 0;
468 static void init_vmcb(struct vmcb *vmcb)
470 struct vmcb_control_area *control = &vmcb->control;
471 struct vmcb_save_area *save = &vmcb->save;
473 control->intercept_cr_read = INTERCEPT_CR0_MASK |
474 INTERCEPT_CR3_MASK |
475 INTERCEPT_CR4_MASK;
477 control->intercept_cr_write = INTERCEPT_CR0_MASK |
478 INTERCEPT_CR3_MASK |
479 INTERCEPT_CR4_MASK;
481 control->intercept_dr_read = INTERCEPT_DR0_MASK |
482 INTERCEPT_DR1_MASK |
483 INTERCEPT_DR2_MASK |
484 INTERCEPT_DR3_MASK;
486 control->intercept_dr_write = INTERCEPT_DR0_MASK |
487 INTERCEPT_DR1_MASK |
488 INTERCEPT_DR2_MASK |
489 INTERCEPT_DR3_MASK |
490 INTERCEPT_DR5_MASK |
491 INTERCEPT_DR7_MASK;
493 control->intercept_exceptions = 1 << PF_VECTOR;
496 control->intercept = (1ULL << INTERCEPT_INTR) |
497 (1ULL << INTERCEPT_NMI) |
498 (1ULL << INTERCEPT_SMI) |
500 * selective cr0 intercept bug?
501 * 0: 0f 22 d8 mov %eax,%cr3
502 * 3: 0f 20 c0 mov %cr0,%eax
503 * 6: 0d 00 00 00 80 or $0x80000000,%eax
504 * b: 0f 22 c0 mov %eax,%cr0
505 * set cr3 ->interception
506 * get cr0 ->interception
507 * set cr0 -> no interception
509 /* (1ULL << INTERCEPT_SELECTIVE_CR0) | */
510 (1ULL << INTERCEPT_CPUID) |
511 (1ULL << INTERCEPT_HLT) |
512 (1ULL << INTERCEPT_INVLPGA) |
513 (1ULL << INTERCEPT_IOIO_PROT) |
514 (1ULL << INTERCEPT_MSR_PROT) |
515 (1ULL << INTERCEPT_TASK_SWITCH) |
516 (1ULL << INTERCEPT_SHUTDOWN) |
517 (1ULL << INTERCEPT_VMRUN) |
518 (1ULL << INTERCEPT_VMMCALL) |
519 (1ULL << INTERCEPT_VMLOAD) |
520 (1ULL << INTERCEPT_VMSAVE) |
521 (1ULL << INTERCEPT_STGI) |
522 (1ULL << INTERCEPT_CLGI) |
523 (1ULL << INTERCEPT_SKINIT) |
524 (1ULL << INTERCEPT_MONITOR) |
525 (1ULL << INTERCEPT_MWAIT);
527 control->iopm_base_pa = iopm_base;
528 control->msrpm_base_pa = msrpm_base;
529 control->tsc_offset = 0;
530 control->int_ctl = V_INTR_MASKING_MASK;
532 init_seg(&save->es);
533 init_seg(&save->ss);
534 init_seg(&save->ds);
535 init_seg(&save->fs);
536 init_seg(&save->gs);
538 save->cs.selector = 0xf000;
539 /* Executable/Readable Code Segment */
540 save->cs.attrib = SVM_SELECTOR_READ_MASK | SVM_SELECTOR_P_MASK |
541 SVM_SELECTOR_S_MASK | SVM_SELECTOR_CODE_MASK;
542 save->cs.limit = 0xffff;
544 * cs.base should really be 0xffff0000, but vmx can't handle that, so
545 * be consistent with it.
547 * Replace when we have real mode working for vmx.
549 save->cs.base = 0xf0000;
551 save->gdtr.limit = 0xffff;
552 save->idtr.limit = 0xffff;
554 init_sys_seg(&save->ldtr, SEG_TYPE_LDT);
555 init_sys_seg(&save->tr, SEG_TYPE_BUSY_TSS16);
557 save->efer = MSR_EFER_SVME_MASK;
559 save->dr6 = 0xffff0ff0;
560 save->dr7 = 0x400;
561 save->rflags = 2;
562 save->rip = 0x0000fff0;
565 * cr0 val on cpu init should be 0x60000010, we enable cpu
566 * cache by default. the orderly way is to enable cache in bios.
568 save->cr0 = 0x00000010 | X86_CR0_PG | X86_CR0_WP;
569 save->cr4 = X86_CR4_PAE;
570 /* rdx = ?? */
573 static struct kvm_vcpu *svm_create_vcpu(struct kvm *kvm, unsigned int id)
575 struct vcpu_svm *svm;
576 struct page *page;
577 int err;
579 svm = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
580 if (!svm) {
581 err = -ENOMEM;
582 goto out;
585 err = kvm_vcpu_init(&svm->vcpu, kvm, id);
586 if (err)
587 goto free_svm;
589 page = alloc_page(GFP_KERNEL);
590 if (!page) {
591 err = -ENOMEM;
592 goto uninit;
595 svm->vmcb = page_address(page);
596 clear_page(svm->vmcb);
597 svm->vmcb_pa = page_to_pfn(page) << PAGE_SHIFT;
598 svm->asid_generation = 0;
599 memset(svm->db_regs, 0, sizeof(svm->db_regs));
600 init_vmcb(svm->vmcb);
602 fx_init(&svm->vcpu);
603 svm->vcpu.fpu_active = 1;
604 svm->vcpu.apic_base = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
605 if (svm->vcpu.vcpu_id == 0)
606 svm->vcpu.apic_base |= MSR_IA32_APICBASE_BSP;
608 return &svm->vcpu;
610 uninit:
611 kvm_vcpu_uninit(&svm->vcpu);
612 free_svm:
613 kfree(svm);
614 out:
615 return ERR_PTR(err);
618 static void svm_free_vcpu(struct kvm_vcpu *vcpu)
620 struct vcpu_svm *svm = to_svm(vcpu);
622 __free_page(pfn_to_page(svm->vmcb_pa >> PAGE_SHIFT));
623 kvm_vcpu_uninit(vcpu);
624 kfree(svm);
627 static void svm_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
629 struct vcpu_svm *svm = to_svm(vcpu);
630 int i;
632 if (unlikely(cpu != vcpu->cpu)) {
633 u64 tsc_this, delta;
636 * Make sure that the guest sees a monotonically
637 * increasing TSC.
639 rdtscll(tsc_this);
640 delta = vcpu->host_tsc - tsc_this;
641 svm->vmcb->control.tsc_offset += delta;
642 vcpu->cpu = cpu;
645 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
646 rdmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
649 static void svm_vcpu_put(struct kvm_vcpu *vcpu)
651 struct vcpu_svm *svm = to_svm(vcpu);
652 int i;
654 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
655 wrmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
657 rdtscll(vcpu->host_tsc);
660 static void svm_vcpu_decache(struct kvm_vcpu *vcpu)
664 static void svm_cache_regs(struct kvm_vcpu *vcpu)
666 struct vcpu_svm *svm = to_svm(vcpu);
668 vcpu->regs[VCPU_REGS_RAX] = svm->vmcb->save.rax;
669 vcpu->regs[VCPU_REGS_RSP] = svm->vmcb->save.rsp;
670 vcpu->rip = svm->vmcb->save.rip;
673 static void svm_decache_regs(struct kvm_vcpu *vcpu)
675 struct vcpu_svm *svm = to_svm(vcpu);
676 svm->vmcb->save.rax = vcpu->regs[VCPU_REGS_RAX];
677 svm->vmcb->save.rsp = vcpu->regs[VCPU_REGS_RSP];
678 svm->vmcb->save.rip = vcpu->rip;
681 static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu)
683 return to_svm(vcpu)->vmcb->save.rflags;
686 static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
688 to_svm(vcpu)->vmcb->save.rflags = rflags;
691 static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg)
693 struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
695 switch (seg) {
696 case VCPU_SREG_CS: return &save->cs;
697 case VCPU_SREG_DS: return &save->ds;
698 case VCPU_SREG_ES: return &save->es;
699 case VCPU_SREG_FS: return &save->fs;
700 case VCPU_SREG_GS: return &save->gs;
701 case VCPU_SREG_SS: return &save->ss;
702 case VCPU_SREG_TR: return &save->tr;
703 case VCPU_SREG_LDTR: return &save->ldtr;
705 BUG();
706 return NULL;
709 static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg)
711 struct vmcb_seg *s = svm_seg(vcpu, seg);
713 return s->base;
716 static void svm_get_segment(struct kvm_vcpu *vcpu,
717 struct kvm_segment *var, int seg)
719 struct vmcb_seg *s = svm_seg(vcpu, seg);
721 var->base = s->base;
722 var->limit = s->limit;
723 var->selector = s->selector;
724 var->type = s->attrib & SVM_SELECTOR_TYPE_MASK;
725 var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1;
726 var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
727 var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1;
728 var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1;
729 var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
730 var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
731 var->g = (s->attrib >> SVM_SELECTOR_G_SHIFT) & 1;
732 var->unusable = !var->present;
735 static void svm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
737 struct vmcb_seg *s = svm_seg(vcpu, VCPU_SREG_CS);
739 *db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
740 *l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
743 static void svm_get_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
745 struct vcpu_svm *svm = to_svm(vcpu);
747 dt->limit = svm->vmcb->save.idtr.limit;
748 dt->base = svm->vmcb->save.idtr.base;
751 static void svm_set_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
753 struct vcpu_svm *svm = to_svm(vcpu);
755 svm->vmcb->save.idtr.limit = dt->limit;
756 svm->vmcb->save.idtr.base = dt->base ;
759 static void svm_get_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
761 struct vcpu_svm *svm = to_svm(vcpu);
763 dt->limit = svm->vmcb->save.gdtr.limit;
764 dt->base = svm->vmcb->save.gdtr.base;
767 static void svm_set_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
769 struct vcpu_svm *svm = to_svm(vcpu);
771 svm->vmcb->save.gdtr.limit = dt->limit;
772 svm->vmcb->save.gdtr.base = dt->base ;
775 static void svm_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
779 static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
781 struct vcpu_svm *svm = to_svm(vcpu);
783 #ifdef CONFIG_X86_64
784 if (vcpu->shadow_efer & KVM_EFER_LME) {
785 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
786 vcpu->shadow_efer |= KVM_EFER_LMA;
787 svm->vmcb->save.efer |= KVM_EFER_LMA | KVM_EFER_LME;
790 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG) ) {
791 vcpu->shadow_efer &= ~KVM_EFER_LMA;
792 svm->vmcb->save.efer &= ~(KVM_EFER_LMA | KVM_EFER_LME);
795 #endif
796 if ((vcpu->cr0 & X86_CR0_TS) && !(cr0 & X86_CR0_TS)) {
797 svm->vmcb->control.intercept_exceptions &= ~(1 << NM_VECTOR);
798 vcpu->fpu_active = 1;
801 vcpu->cr0 = cr0;
802 cr0 |= X86_CR0_PG | X86_CR0_WP;
803 cr0 &= ~(X86_CR0_CD | X86_CR0_NW);
804 svm->vmcb->save.cr0 = cr0;
807 static void svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
809 vcpu->cr4 = cr4;
810 to_svm(vcpu)->vmcb->save.cr4 = cr4 | X86_CR4_PAE;
813 static void svm_set_segment(struct kvm_vcpu *vcpu,
814 struct kvm_segment *var, int seg)
816 struct vcpu_svm *svm = to_svm(vcpu);
817 struct vmcb_seg *s = svm_seg(vcpu, seg);
819 s->base = var->base;
820 s->limit = var->limit;
821 s->selector = var->selector;
822 if (var->unusable)
823 s->attrib = 0;
824 else {
825 s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK);
826 s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT;
827 s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT;
828 s->attrib |= (var->present & 1) << SVM_SELECTOR_P_SHIFT;
829 s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT;
830 s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT;
831 s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT;
832 s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT;
834 if (seg == VCPU_SREG_CS)
835 svm->vmcb->save.cpl
836 = (svm->vmcb->save.cs.attrib
837 >> SVM_SELECTOR_DPL_SHIFT) & 3;
841 /* FIXME:
843 svm(vcpu)->vmcb->control.int_ctl &= ~V_TPR_MASK;
844 svm(vcpu)->vmcb->control.int_ctl |= (sregs->cr8 & V_TPR_MASK);
848 static int svm_guest_debug(struct kvm_vcpu *vcpu, struct kvm_debug_guest *dbg)
850 return -EOPNOTSUPP;
853 static void load_host_msrs(struct kvm_vcpu *vcpu)
855 #ifdef CONFIG_X86_64
856 wrmsrl(MSR_GS_BASE, to_svm(vcpu)->host_gs_base);
857 #endif
860 static void save_host_msrs(struct kvm_vcpu *vcpu)
862 #ifdef CONFIG_X86_64
863 rdmsrl(MSR_GS_BASE, to_svm(vcpu)->host_gs_base);
864 #endif
867 static void new_asid(struct vcpu_svm *svm, struct svm_cpu_data *svm_data)
869 if (svm_data->next_asid > svm_data->max_asid) {
870 ++svm_data->asid_generation;
871 svm_data->next_asid = 1;
872 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID;
875 svm->vcpu.cpu = svm_data->cpu;
876 svm->asid_generation = svm_data->asid_generation;
877 svm->vmcb->control.asid = svm_data->next_asid++;
880 static void svm_invlpg(struct kvm_vcpu *vcpu, gva_t address)
882 invlpga(address, to_svm(vcpu)->vmcb->control.asid); // is needed?
885 static unsigned long svm_get_dr(struct kvm_vcpu *vcpu, int dr)
887 return to_svm(vcpu)->db_regs[dr];
890 static void svm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long value,
891 int *exception)
893 struct vcpu_svm *svm = to_svm(vcpu);
895 *exception = 0;
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;
904 switch (dr) {
905 case 0 ... 3:
906 svm->db_regs[dr] = value;
907 return;
908 case 4 ... 5:
909 if (vcpu->cr4 & X86_CR4_DE) {
910 *exception = UD_VECTOR;
911 return;
913 case 7: {
914 if (value & ~((1ULL << 32) - 1)) {
915 *exception = GP_VECTOR;
916 return;
918 svm->vmcb->save.dr7 = value;
919 return;
921 default:
922 printk(KERN_DEBUG "%s: unexpected dr %u\n",
923 __FUNCTION__, dr);
924 *exception = UD_VECTOR;
925 return;
929 static int pf_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
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 enum emulation_result er;
936 int r;
938 if (is_external_interrupt(exit_int_info))
939 push_irq(&svm->vcpu, exit_int_info & SVM_EVTINJ_VEC_MASK);
941 mutex_lock(&kvm->lock);
943 fault_address = svm->vmcb->control.exit_info_2;
944 error_code = svm->vmcb->control.exit_info_1;
945 r = kvm_mmu_page_fault(&svm->vcpu, fault_address, error_code);
946 if (r < 0) {
947 mutex_unlock(&kvm->lock);
948 return r;
950 if (!r) {
951 mutex_unlock(&kvm->lock);
952 return 1;
954 er = emulate_instruction(&svm->vcpu, kvm_run, fault_address,
955 error_code);
956 mutex_unlock(&kvm->lock);
958 switch (er) {
959 case EMULATE_DONE:
960 return 1;
961 case EMULATE_DO_MMIO:
962 ++svm->vcpu.stat.mmio_exits;
963 return 0;
964 case EMULATE_FAIL:
965 vcpu_printf(&svm->vcpu, "%s: emulate fail\n", __FUNCTION__);
966 break;
967 default:
968 BUG();
971 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
972 return 0;
975 static int nm_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
977 svm->vmcb->control.intercept_exceptions &= ~(1 << NM_VECTOR);
978 if (!(svm->vcpu.cr0 & X86_CR0_TS))
979 svm->vmcb->save.cr0 &= ~X86_CR0_TS;
980 svm->vcpu.fpu_active = 1;
982 return 1;
985 static int shutdown_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
988 * VMCB is undefined after a SHUTDOWN intercept
989 * so reinitialize it.
991 clear_page(svm->vmcb);
992 init_vmcb(svm->vmcb);
994 kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
995 return 0;
998 static int io_get_override(struct vcpu_svm *svm,
999 struct vmcb_seg **seg,
1000 int *addr_override)
1002 u8 inst[MAX_INST_SIZE];
1003 unsigned ins_length;
1004 gva_t rip;
1005 int i;
1007 rip = svm->vmcb->save.rip;
1008 ins_length = svm->next_rip - rip;
1009 rip += svm->vmcb->save.cs.base;
1011 if (ins_length > MAX_INST_SIZE)
1012 printk(KERN_DEBUG
1013 "%s: inst length err, cs base 0x%llx rip 0x%llx "
1014 "next rip 0x%llx ins_length %u\n",
1015 __FUNCTION__,
1016 svm->vmcb->save.cs.base,
1017 svm->vmcb->save.rip,
1018 svm->vmcb->control.exit_info_2,
1019 ins_length);
1021 if (emulator_read_std(rip, inst, ins_length, &svm->vcpu)
1022 != X86EMUL_CONTINUE)
1023 /* #PF */
1024 return 0;
1026 *addr_override = 0;
1027 *seg = NULL;
1028 for (i = 0; i < ins_length; i++)
1029 switch (inst[i]) {
1030 case 0xf0:
1031 case 0xf2:
1032 case 0xf3:
1033 case 0x66:
1034 continue;
1035 case 0x67:
1036 *addr_override = 1;
1037 continue;
1038 case 0x2e:
1039 *seg = &svm->vmcb->save.cs;
1040 continue;
1041 case 0x36:
1042 *seg = &svm->vmcb->save.ss;
1043 continue;
1044 case 0x3e:
1045 *seg = &svm->vmcb->save.ds;
1046 continue;
1047 case 0x26:
1048 *seg = &svm->vmcb->save.es;
1049 continue;
1050 case 0x64:
1051 *seg = &svm->vmcb->save.fs;
1052 continue;
1053 case 0x65:
1054 *seg = &svm->vmcb->save.gs;
1055 continue;
1056 default:
1057 return 1;
1059 printk(KERN_DEBUG "%s: unexpected\n", __FUNCTION__);
1060 return 0;
1063 static unsigned long io_address(struct vcpu_svm *svm, int ins, gva_t *address)
1065 unsigned long addr_mask;
1066 unsigned long *reg;
1067 struct vmcb_seg *seg;
1068 int addr_override;
1069 struct vmcb_save_area *save_area = &svm->vmcb->save;
1070 u16 cs_attrib = save_area->cs.attrib;
1071 unsigned addr_size = get_addr_size(svm);
1073 if (!io_get_override(svm, &seg, &addr_override))
1074 return 0;
1076 if (addr_override)
1077 addr_size = (addr_size == 2) ? 4: (addr_size >> 1);
1079 if (ins) {
1080 reg = &svm->vcpu.regs[VCPU_REGS_RDI];
1081 seg = &svm->vmcb->save.es;
1082 } else {
1083 reg = &svm->vcpu.regs[VCPU_REGS_RSI];
1084 seg = (seg) ? seg : &svm->vmcb->save.ds;
1087 addr_mask = ~0ULL >> (64 - (addr_size * 8));
1089 if ((cs_attrib & SVM_SELECTOR_L_MASK) &&
1090 !(svm->vmcb->save.rflags & X86_EFLAGS_VM)) {
1091 *address = (*reg & addr_mask);
1092 return addr_mask;
1095 if (!(seg->attrib & SVM_SELECTOR_P_SHIFT)) {
1096 svm_inject_gp(&svm->vcpu, 0);
1097 return 0;
1100 *address = (*reg & addr_mask) + seg->base;
1101 return addr_mask;
1104 static int io_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1106 u32 io_info = svm->vmcb->control.exit_info_1; //address size bug?
1107 int size, down, in, string, rep;
1108 unsigned port;
1109 unsigned long count;
1110 gva_t address = 0;
1112 ++svm->vcpu.stat.io_exits;
1114 svm->next_rip = svm->vmcb->control.exit_info_2;
1116 in = (io_info & SVM_IOIO_TYPE_MASK) != 0;
1117 port = io_info >> 16;
1118 size = (io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT;
1119 string = (io_info & SVM_IOIO_STR_MASK) != 0;
1120 rep = (io_info & SVM_IOIO_REP_MASK) != 0;
1121 count = 1;
1122 down = (svm->vmcb->save.rflags & X86_EFLAGS_DF) != 0;
1124 if (string) {
1125 unsigned addr_mask;
1127 addr_mask = io_address(svm, in, &address);
1128 if (!addr_mask) {
1129 printk(KERN_DEBUG "%s: get io address failed\n",
1130 __FUNCTION__);
1131 return 1;
1134 if (rep)
1135 count = svm->vcpu.regs[VCPU_REGS_RCX] & addr_mask;
1137 return kvm_setup_pio(&svm->vcpu, kvm_run, in, size, count, string,
1138 down, address, rep, port);
1141 static int nop_on_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1143 return 1;
1146 static int halt_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1148 svm->next_rip = svm->vmcb->save.rip + 1;
1149 skip_emulated_instruction(&svm->vcpu);
1150 return kvm_emulate_halt(&svm->vcpu);
1153 static int vmmcall_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1155 svm->next_rip = svm->vmcb->save.rip + 3;
1156 skip_emulated_instruction(&svm->vcpu);
1157 return kvm_hypercall(&svm->vcpu, kvm_run);
1160 static int invalid_op_interception(struct vcpu_svm *svm,
1161 struct kvm_run *kvm_run)
1163 inject_ud(&svm->vcpu);
1164 return 1;
1167 static int task_switch_interception(struct vcpu_svm *svm,
1168 struct kvm_run *kvm_run)
1170 pr_unimpl(&svm->vcpu, "%s: task switch is unsupported\n", __FUNCTION__);
1171 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
1172 return 0;
1175 static int cpuid_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1177 svm->next_rip = svm->vmcb->save.rip + 2;
1178 kvm_emulate_cpuid(&svm->vcpu);
1179 return 1;
1182 static int emulate_on_interception(struct vcpu_svm *svm,
1183 struct kvm_run *kvm_run)
1185 if (emulate_instruction(&svm->vcpu, NULL, 0, 0) != EMULATE_DONE)
1186 pr_unimpl(&svm->vcpu, "%s: failed\n", __FUNCTION__);
1187 return 1;
1190 static int svm_get_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 *data)
1192 struct vcpu_svm *svm = to_svm(vcpu);
1194 switch (ecx) {
1195 case MSR_IA32_TIME_STAMP_COUNTER: {
1196 u64 tsc;
1198 rdtscll(tsc);
1199 *data = svm->vmcb->control.tsc_offset + tsc;
1200 break;
1202 case MSR_K6_STAR:
1203 *data = svm->vmcb->save.star;
1204 break;
1205 #ifdef CONFIG_X86_64
1206 case MSR_LSTAR:
1207 *data = svm->vmcb->save.lstar;
1208 break;
1209 case MSR_CSTAR:
1210 *data = svm->vmcb->save.cstar;
1211 break;
1212 case MSR_KERNEL_GS_BASE:
1213 *data = svm->vmcb->save.kernel_gs_base;
1214 break;
1215 case MSR_SYSCALL_MASK:
1216 *data = svm->vmcb->save.sfmask;
1217 break;
1218 #endif
1219 case MSR_IA32_SYSENTER_CS:
1220 *data = svm->vmcb->save.sysenter_cs;
1221 break;
1222 case MSR_IA32_SYSENTER_EIP:
1223 *data = svm->vmcb->save.sysenter_eip;
1224 break;
1225 case MSR_IA32_SYSENTER_ESP:
1226 *data = svm->vmcb->save.sysenter_esp;
1227 break;
1228 default:
1229 return kvm_get_msr_common(vcpu, ecx, data);
1231 return 0;
1234 static int rdmsr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1236 u32 ecx = svm->vcpu.regs[VCPU_REGS_RCX];
1237 u64 data;
1239 if (svm_get_msr(&svm->vcpu, ecx, &data))
1240 svm_inject_gp(&svm->vcpu, 0);
1241 else {
1242 svm->vmcb->save.rax = data & 0xffffffff;
1243 svm->vcpu.regs[VCPU_REGS_RDX] = data >> 32;
1244 svm->next_rip = svm->vmcb->save.rip + 2;
1245 skip_emulated_instruction(&svm->vcpu);
1247 return 1;
1250 static int svm_set_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 data)
1252 struct vcpu_svm *svm = to_svm(vcpu);
1254 switch (ecx) {
1255 case MSR_IA32_TIME_STAMP_COUNTER: {
1256 u64 tsc;
1258 rdtscll(tsc);
1259 svm->vmcb->control.tsc_offset = data - tsc;
1260 break;
1262 case MSR_K6_STAR:
1263 svm->vmcb->save.star = data;
1264 break;
1265 #ifdef CONFIG_X86_64
1266 case MSR_LSTAR:
1267 svm->vmcb->save.lstar = data;
1268 break;
1269 case MSR_CSTAR:
1270 svm->vmcb->save.cstar = data;
1271 break;
1272 case MSR_KERNEL_GS_BASE:
1273 svm->vmcb->save.kernel_gs_base = data;
1274 break;
1275 case MSR_SYSCALL_MASK:
1276 svm->vmcb->save.sfmask = data;
1277 break;
1278 #endif
1279 case MSR_IA32_SYSENTER_CS:
1280 svm->vmcb->save.sysenter_cs = data;
1281 break;
1282 case MSR_IA32_SYSENTER_EIP:
1283 svm->vmcb->save.sysenter_eip = data;
1284 break;
1285 case MSR_IA32_SYSENTER_ESP:
1286 svm->vmcb->save.sysenter_esp = data;
1287 break;
1288 default:
1289 return kvm_set_msr_common(vcpu, ecx, data);
1291 return 0;
1294 static int wrmsr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1296 u32 ecx = svm->vcpu.regs[VCPU_REGS_RCX];
1297 u64 data = (svm->vmcb->save.rax & -1u)
1298 | ((u64)(svm->vcpu.regs[VCPU_REGS_RDX] & -1u) << 32);
1299 svm->next_rip = svm->vmcb->save.rip + 2;
1300 if (svm_set_msr(&svm->vcpu, ecx, data))
1301 svm_inject_gp(&svm->vcpu, 0);
1302 else
1303 skip_emulated_instruction(&svm->vcpu);
1304 return 1;
1307 static int msr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1309 if (svm->vmcb->control.exit_info_1)
1310 return wrmsr_interception(svm, kvm_run);
1311 else
1312 return rdmsr_interception(svm, kvm_run);
1315 static int interrupt_window_interception(struct vcpu_svm *svm,
1316 struct kvm_run *kvm_run)
1319 * If the user space waits to inject interrupts, exit as soon as
1320 * possible
1322 if (kvm_run->request_interrupt_window &&
1323 !svm->vcpu.irq_summary) {
1324 ++svm->vcpu.stat.irq_window_exits;
1325 kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
1326 return 0;
1329 return 1;
1332 static int (*svm_exit_handlers[])(struct vcpu_svm *svm,
1333 struct kvm_run *kvm_run) = {
1334 [SVM_EXIT_READ_CR0] = emulate_on_interception,
1335 [SVM_EXIT_READ_CR3] = emulate_on_interception,
1336 [SVM_EXIT_READ_CR4] = emulate_on_interception,
1337 /* for now: */
1338 [SVM_EXIT_WRITE_CR0] = emulate_on_interception,
1339 [SVM_EXIT_WRITE_CR3] = emulate_on_interception,
1340 [SVM_EXIT_WRITE_CR4] = emulate_on_interception,
1341 [SVM_EXIT_READ_DR0] = emulate_on_interception,
1342 [SVM_EXIT_READ_DR1] = emulate_on_interception,
1343 [SVM_EXIT_READ_DR2] = emulate_on_interception,
1344 [SVM_EXIT_READ_DR3] = emulate_on_interception,
1345 [SVM_EXIT_WRITE_DR0] = emulate_on_interception,
1346 [SVM_EXIT_WRITE_DR1] = emulate_on_interception,
1347 [SVM_EXIT_WRITE_DR2] = emulate_on_interception,
1348 [SVM_EXIT_WRITE_DR3] = emulate_on_interception,
1349 [SVM_EXIT_WRITE_DR5] = emulate_on_interception,
1350 [SVM_EXIT_WRITE_DR7] = emulate_on_interception,
1351 [SVM_EXIT_EXCP_BASE + PF_VECTOR] = pf_interception,
1352 [SVM_EXIT_EXCP_BASE + NM_VECTOR] = nm_interception,
1353 [SVM_EXIT_INTR] = nop_on_interception,
1354 [SVM_EXIT_NMI] = nop_on_interception,
1355 [SVM_EXIT_SMI] = nop_on_interception,
1356 [SVM_EXIT_INIT] = nop_on_interception,
1357 [SVM_EXIT_VINTR] = interrupt_window_interception,
1358 /* [SVM_EXIT_CR0_SEL_WRITE] = emulate_on_interception, */
1359 [SVM_EXIT_CPUID] = cpuid_interception,
1360 [SVM_EXIT_HLT] = halt_interception,
1361 [SVM_EXIT_INVLPG] = emulate_on_interception,
1362 [SVM_EXIT_INVLPGA] = invalid_op_interception,
1363 [SVM_EXIT_IOIO] = io_interception,
1364 [SVM_EXIT_MSR] = msr_interception,
1365 [SVM_EXIT_TASK_SWITCH] = task_switch_interception,
1366 [SVM_EXIT_SHUTDOWN] = shutdown_interception,
1367 [SVM_EXIT_VMRUN] = invalid_op_interception,
1368 [SVM_EXIT_VMMCALL] = vmmcall_interception,
1369 [SVM_EXIT_VMLOAD] = invalid_op_interception,
1370 [SVM_EXIT_VMSAVE] = invalid_op_interception,
1371 [SVM_EXIT_STGI] = invalid_op_interception,
1372 [SVM_EXIT_CLGI] = invalid_op_interception,
1373 [SVM_EXIT_SKINIT] = invalid_op_interception,
1374 [SVM_EXIT_MONITOR] = invalid_op_interception,
1375 [SVM_EXIT_MWAIT] = invalid_op_interception,
1379 static int handle_exit(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1381 u32 exit_code = svm->vmcb->control.exit_code;
1383 if (is_external_interrupt(svm->vmcb->control.exit_int_info) &&
1384 exit_code != SVM_EXIT_EXCP_BASE + PF_VECTOR)
1385 printk(KERN_ERR "%s: unexpected exit_ini_info 0x%x "
1386 "exit_code 0x%x\n",
1387 __FUNCTION__, svm->vmcb->control.exit_int_info,
1388 exit_code);
1390 if (exit_code >= ARRAY_SIZE(svm_exit_handlers)
1391 || svm_exit_handlers[exit_code] == 0) {
1392 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
1393 kvm_run->hw.hardware_exit_reason = exit_code;
1394 return 0;
1397 return svm_exit_handlers[exit_code](svm, kvm_run);
1400 static void reload_tss(struct kvm_vcpu *vcpu)
1402 int cpu = raw_smp_processor_id();
1404 struct svm_cpu_data *svm_data = per_cpu(svm_data, cpu);
1405 svm_data->tss_desc->type = 9; //available 32/64-bit TSS
1406 load_TR_desc();
1409 static void pre_svm_run(struct vcpu_svm *svm)
1411 int cpu = raw_smp_processor_id();
1413 struct svm_cpu_data *svm_data = per_cpu(svm_data, cpu);
1415 svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
1416 if (svm->vcpu.cpu != cpu ||
1417 svm->asid_generation != svm_data->asid_generation)
1418 new_asid(svm, svm_data);
1422 static inline void inject_irq(struct vcpu_svm *svm)
1424 struct vmcb_control_area *control;
1426 control = &svm->vmcb->control;
1427 control->int_vector = pop_irq(&svm->vcpu);
1428 control->int_ctl &= ~V_INTR_PRIO_MASK;
1429 control->int_ctl |= V_IRQ_MASK |
1430 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT);
1433 static void reput_irq(struct vcpu_svm *svm)
1435 struct vmcb_control_area *control = &svm->vmcb->control;
1437 if (control->int_ctl & V_IRQ_MASK) {
1438 control->int_ctl &= ~V_IRQ_MASK;
1439 push_irq(&svm->vcpu, control->int_vector);
1442 svm->vcpu.interrupt_window_open =
1443 !(control->int_state & SVM_INTERRUPT_SHADOW_MASK);
1446 static void do_interrupt_requests(struct vcpu_svm *svm,
1447 struct kvm_run *kvm_run)
1449 struct vmcb_control_area *control = &svm->vmcb->control;
1451 svm->vcpu.interrupt_window_open =
1452 (!(control->int_state & SVM_INTERRUPT_SHADOW_MASK) &&
1453 (svm->vmcb->save.rflags & X86_EFLAGS_IF));
1455 if (svm->vcpu.interrupt_window_open && svm->vcpu.irq_summary)
1457 * If interrupts enabled, and not blocked by sti or mov ss. Good.
1459 inject_irq(svm);
1462 * Interrupts blocked. Wait for unblock.
1464 if (!svm->vcpu.interrupt_window_open &&
1465 (svm->vcpu.irq_summary || kvm_run->request_interrupt_window)) {
1466 control->intercept |= 1ULL << INTERCEPT_VINTR;
1467 } else
1468 control->intercept &= ~(1ULL << INTERCEPT_VINTR);
1471 static void post_kvm_run_save(struct vcpu_svm *svm,
1472 struct kvm_run *kvm_run)
1474 kvm_run->ready_for_interrupt_injection
1475 = (svm->vcpu.interrupt_window_open &&
1476 svm->vcpu.irq_summary == 0);
1477 kvm_run->if_flag = (svm->vmcb->save.rflags & X86_EFLAGS_IF) != 0;
1478 kvm_run->cr8 = svm->vcpu.cr8;
1479 kvm_run->apic_base = svm->vcpu.apic_base;
1483 * Check if userspace requested an interrupt window, and that the
1484 * interrupt window is open.
1486 * No need to exit to userspace if we already have an interrupt queued.
1488 static int dm_request_for_irq_injection(struct vcpu_svm *svm,
1489 struct kvm_run *kvm_run)
1491 return (!svm->vcpu.irq_summary &&
1492 kvm_run->request_interrupt_window &&
1493 svm->vcpu.interrupt_window_open &&
1494 (svm->vmcb->save.rflags & X86_EFLAGS_IF));
1497 static void save_db_regs(unsigned long *db_regs)
1499 asm volatile ("mov %%dr0, %0" : "=r"(db_regs[0]));
1500 asm volatile ("mov %%dr1, %0" : "=r"(db_regs[1]));
1501 asm volatile ("mov %%dr2, %0" : "=r"(db_regs[2]));
1502 asm volatile ("mov %%dr3, %0" : "=r"(db_regs[3]));
1505 static void load_db_regs(unsigned long *db_regs)
1507 asm volatile ("mov %0, %%dr0" : : "r"(db_regs[0]));
1508 asm volatile ("mov %0, %%dr1" : : "r"(db_regs[1]));
1509 asm volatile ("mov %0, %%dr2" : : "r"(db_regs[2]));
1510 asm volatile ("mov %0, %%dr3" : : "r"(db_regs[3]));
1513 static void svm_flush_tlb(struct kvm_vcpu *vcpu)
1515 force_new_asid(vcpu);
1518 static int svm_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1520 struct vcpu_svm *svm = to_svm(vcpu);
1521 u16 fs_selector;
1522 u16 gs_selector;
1523 u16 ldt_selector;
1524 int r;
1526 again:
1527 r = kvm_mmu_reload(vcpu);
1528 if (unlikely(r))
1529 return r;
1531 if (!vcpu->mmio_read_completed)
1532 do_interrupt_requests(svm, kvm_run);
1534 clgi();
1536 vcpu->guest_mode = 1;
1537 if (vcpu->requests)
1538 if (test_and_clear_bit(KVM_TLB_FLUSH, &vcpu->requests))
1539 svm_flush_tlb(vcpu);
1541 pre_svm_run(svm);
1543 save_host_msrs(vcpu);
1544 fs_selector = read_fs();
1545 gs_selector = read_gs();
1546 ldt_selector = read_ldt();
1547 svm->host_cr2 = kvm_read_cr2();
1548 svm->host_dr6 = read_dr6();
1549 svm->host_dr7 = read_dr7();
1550 svm->vmcb->save.cr2 = vcpu->cr2;
1552 if (svm->vmcb->save.dr7 & 0xff) {
1553 write_dr7(0);
1554 save_db_regs(svm->host_db_regs);
1555 load_db_regs(svm->db_regs);
1558 if (vcpu->fpu_active) {
1559 fx_save(&vcpu->host_fx_image);
1560 fx_restore(&vcpu->guest_fx_image);
1563 asm volatile (
1564 #ifdef CONFIG_X86_64
1565 "push %%rbx; push %%rcx; push %%rdx;"
1566 "push %%rsi; push %%rdi; push %%rbp;"
1567 "push %%r8; push %%r9; push %%r10; push %%r11;"
1568 "push %%r12; push %%r13; push %%r14; push %%r15;"
1569 #else
1570 "push %%ebx; push %%ecx; push %%edx;"
1571 "push %%esi; push %%edi; push %%ebp;"
1572 #endif
1574 #ifdef CONFIG_X86_64
1575 "mov %c[rbx](%[svm]), %%rbx \n\t"
1576 "mov %c[rcx](%[svm]), %%rcx \n\t"
1577 "mov %c[rdx](%[svm]), %%rdx \n\t"
1578 "mov %c[rsi](%[svm]), %%rsi \n\t"
1579 "mov %c[rdi](%[svm]), %%rdi \n\t"
1580 "mov %c[rbp](%[svm]), %%rbp \n\t"
1581 "mov %c[r8](%[svm]), %%r8 \n\t"
1582 "mov %c[r9](%[svm]), %%r9 \n\t"
1583 "mov %c[r10](%[svm]), %%r10 \n\t"
1584 "mov %c[r11](%[svm]), %%r11 \n\t"
1585 "mov %c[r12](%[svm]), %%r12 \n\t"
1586 "mov %c[r13](%[svm]), %%r13 \n\t"
1587 "mov %c[r14](%[svm]), %%r14 \n\t"
1588 "mov %c[r15](%[svm]), %%r15 \n\t"
1589 #else
1590 "mov %c[rbx](%[svm]), %%ebx \n\t"
1591 "mov %c[rcx](%[svm]), %%ecx \n\t"
1592 "mov %c[rdx](%[svm]), %%edx \n\t"
1593 "mov %c[rsi](%[svm]), %%esi \n\t"
1594 "mov %c[rdi](%[svm]), %%edi \n\t"
1595 "mov %c[rbp](%[svm]), %%ebp \n\t"
1596 #endif
1598 #ifdef CONFIG_X86_64
1599 /* Enter guest mode */
1600 "push %%rax \n\t"
1601 "mov %c[vmcb](%[svm]), %%rax \n\t"
1602 SVM_VMLOAD "\n\t"
1603 SVM_VMRUN "\n\t"
1604 SVM_VMSAVE "\n\t"
1605 "pop %%rax \n\t"
1606 #else
1607 /* Enter guest mode */
1608 "push %%eax \n\t"
1609 "mov %c[vmcb](%[svm]), %%eax \n\t"
1610 SVM_VMLOAD "\n\t"
1611 SVM_VMRUN "\n\t"
1612 SVM_VMSAVE "\n\t"
1613 "pop %%eax \n\t"
1614 #endif
1616 /* Save guest registers, load host registers */
1617 #ifdef CONFIG_X86_64
1618 "mov %%rbx, %c[rbx](%[svm]) \n\t"
1619 "mov %%rcx, %c[rcx](%[svm]) \n\t"
1620 "mov %%rdx, %c[rdx](%[svm]) \n\t"
1621 "mov %%rsi, %c[rsi](%[svm]) \n\t"
1622 "mov %%rdi, %c[rdi](%[svm]) \n\t"
1623 "mov %%rbp, %c[rbp](%[svm]) \n\t"
1624 "mov %%r8, %c[r8](%[svm]) \n\t"
1625 "mov %%r9, %c[r9](%[svm]) \n\t"
1626 "mov %%r10, %c[r10](%[svm]) \n\t"
1627 "mov %%r11, %c[r11](%[svm]) \n\t"
1628 "mov %%r12, %c[r12](%[svm]) \n\t"
1629 "mov %%r13, %c[r13](%[svm]) \n\t"
1630 "mov %%r14, %c[r14](%[svm]) \n\t"
1631 "mov %%r15, %c[r15](%[svm]) \n\t"
1633 "pop %%r15; pop %%r14; pop %%r13; pop %%r12;"
1634 "pop %%r11; pop %%r10; pop %%r9; pop %%r8;"
1635 "pop %%rbp; pop %%rdi; pop %%rsi;"
1636 "pop %%rdx; pop %%rcx; pop %%rbx; \n\t"
1637 #else
1638 "mov %%ebx, %c[rbx](%[svm]) \n\t"
1639 "mov %%ecx, %c[rcx](%[svm]) \n\t"
1640 "mov %%edx, %c[rdx](%[svm]) \n\t"
1641 "mov %%esi, %c[rsi](%[svm]) \n\t"
1642 "mov %%edi, %c[rdi](%[svm]) \n\t"
1643 "mov %%ebp, %c[rbp](%[svm]) \n\t"
1645 "pop %%ebp; pop %%edi; pop %%esi;"
1646 "pop %%edx; pop %%ecx; pop %%ebx; \n\t"
1647 #endif
1649 : [svm]"a"(svm),
1650 [vmcb]"i"(offsetof(struct vcpu_svm, vmcb_pa)),
1651 [rbx]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_RBX])),
1652 [rcx]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_RCX])),
1653 [rdx]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_RDX])),
1654 [rsi]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_RSI])),
1655 [rdi]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_RDI])),
1656 [rbp]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_RBP]))
1657 #ifdef CONFIG_X86_64
1658 ,[r8 ]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_R8])),
1659 [r9 ]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_R9 ])),
1660 [r10]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_R10])),
1661 [r11]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_R11])),
1662 [r12]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_R12])),
1663 [r13]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_R13])),
1664 [r14]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_R14])),
1665 [r15]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_R15]))
1666 #endif
1667 : "cc", "memory" );
1669 vcpu->guest_mode = 0;
1671 if (vcpu->fpu_active) {
1672 fx_save(&vcpu->guest_fx_image);
1673 fx_restore(&vcpu->host_fx_image);
1676 if ((svm->vmcb->save.dr7 & 0xff))
1677 load_db_regs(svm->host_db_regs);
1679 vcpu->cr2 = svm->vmcb->save.cr2;
1681 write_dr6(svm->host_dr6);
1682 write_dr7(svm->host_dr7);
1683 kvm_write_cr2(svm->host_cr2);
1685 load_fs(fs_selector);
1686 load_gs(gs_selector);
1687 load_ldt(ldt_selector);
1688 load_host_msrs(vcpu);
1690 reload_tss(vcpu);
1693 * Profile KVM exit RIPs:
1695 if (unlikely(prof_on == KVM_PROFILING))
1696 profile_hit(KVM_PROFILING,
1697 (void *)(unsigned long)svm->vmcb->save.rip);
1699 stgi();
1701 reput_irq(svm);
1703 svm->next_rip = 0;
1705 if (svm->vmcb->control.exit_code == SVM_EXIT_ERR) {
1706 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
1707 kvm_run->fail_entry.hardware_entry_failure_reason
1708 = svm->vmcb->control.exit_code;
1709 post_kvm_run_save(svm, kvm_run);
1710 return 0;
1713 r = handle_exit(svm, kvm_run);
1714 if (r > 0) {
1715 if (signal_pending(current)) {
1716 ++vcpu->stat.signal_exits;
1717 post_kvm_run_save(svm, kvm_run);
1718 kvm_run->exit_reason = KVM_EXIT_INTR;
1719 return -EINTR;
1722 if (dm_request_for_irq_injection(svm, kvm_run)) {
1723 ++vcpu->stat.request_irq_exits;
1724 post_kvm_run_save(svm, kvm_run);
1725 kvm_run->exit_reason = KVM_EXIT_INTR;
1726 return -EINTR;
1728 kvm_resched(vcpu);
1729 goto again;
1731 post_kvm_run_save(svm, kvm_run);
1732 return r;
1735 static void svm_set_cr3(struct kvm_vcpu *vcpu, unsigned long root)
1737 struct vcpu_svm *svm = to_svm(vcpu);
1739 svm->vmcb->save.cr3 = root;
1740 force_new_asid(vcpu);
1742 if (vcpu->fpu_active) {
1743 svm->vmcb->control.intercept_exceptions |= (1 << NM_VECTOR);
1744 svm->vmcb->save.cr0 |= X86_CR0_TS;
1745 vcpu->fpu_active = 0;
1749 static void svm_inject_page_fault(struct kvm_vcpu *vcpu,
1750 unsigned long addr,
1751 uint32_t err_code)
1753 struct vcpu_svm *svm = to_svm(vcpu);
1754 uint32_t exit_int_info = svm->vmcb->control.exit_int_info;
1756 ++vcpu->stat.pf_guest;
1758 if (is_page_fault(exit_int_info)) {
1760 svm->vmcb->control.event_inj_err = 0;
1761 svm->vmcb->control.event_inj = SVM_EVTINJ_VALID |
1762 SVM_EVTINJ_VALID_ERR |
1763 SVM_EVTINJ_TYPE_EXEPT |
1764 DF_VECTOR;
1765 return;
1767 vcpu->cr2 = addr;
1768 svm->vmcb->save.cr2 = addr;
1769 svm->vmcb->control.event_inj = SVM_EVTINJ_VALID |
1770 SVM_EVTINJ_VALID_ERR |
1771 SVM_EVTINJ_TYPE_EXEPT |
1772 PF_VECTOR;
1773 svm->vmcb->control.event_inj_err = err_code;
1777 static int is_disabled(void)
1779 u64 vm_cr;
1781 rdmsrl(MSR_VM_CR, vm_cr);
1782 if (vm_cr & (1 << SVM_VM_CR_SVM_DISABLE))
1783 return 1;
1785 return 0;
1788 static void
1789 svm_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
1792 * Patch in the VMMCALL instruction:
1794 hypercall[0] = 0x0f;
1795 hypercall[1] = 0x01;
1796 hypercall[2] = 0xd9;
1797 hypercall[3] = 0xc3;
1800 static void svm_check_processor_compat(void *rtn)
1802 *(int *)rtn = 0;
1805 static struct kvm_arch_ops svm_arch_ops = {
1806 .cpu_has_kvm_support = has_svm,
1807 .disabled_by_bios = is_disabled,
1808 .hardware_setup = svm_hardware_setup,
1809 .hardware_unsetup = svm_hardware_unsetup,
1810 .check_processor_compatibility = svm_check_processor_compat,
1811 .hardware_enable = svm_hardware_enable,
1812 .hardware_disable = svm_hardware_disable,
1814 .vcpu_create = svm_create_vcpu,
1815 .vcpu_free = svm_free_vcpu,
1817 .vcpu_load = svm_vcpu_load,
1818 .vcpu_put = svm_vcpu_put,
1819 .vcpu_decache = svm_vcpu_decache,
1821 .set_guest_debug = svm_guest_debug,
1822 .get_msr = svm_get_msr,
1823 .set_msr = svm_set_msr,
1824 .get_segment_base = svm_get_segment_base,
1825 .get_segment = svm_get_segment,
1826 .set_segment = svm_set_segment,
1827 .get_cs_db_l_bits = svm_get_cs_db_l_bits,
1828 .decache_cr4_guest_bits = svm_decache_cr4_guest_bits,
1829 .set_cr0 = svm_set_cr0,
1830 .set_cr3 = svm_set_cr3,
1831 .set_cr4 = svm_set_cr4,
1832 .set_efer = svm_set_efer,
1833 .get_idt = svm_get_idt,
1834 .set_idt = svm_set_idt,
1835 .get_gdt = svm_get_gdt,
1836 .set_gdt = svm_set_gdt,
1837 .get_dr = svm_get_dr,
1838 .set_dr = svm_set_dr,
1839 .cache_regs = svm_cache_regs,
1840 .decache_regs = svm_decache_regs,
1841 .get_rflags = svm_get_rflags,
1842 .set_rflags = svm_set_rflags,
1844 .invlpg = svm_invlpg,
1845 .tlb_flush = svm_flush_tlb,
1846 .inject_page_fault = svm_inject_page_fault,
1848 .inject_gp = svm_inject_gp,
1850 .run = svm_vcpu_run,
1851 .skip_emulated_instruction = skip_emulated_instruction,
1852 .patch_hypercall = svm_patch_hypercall,
1855 static int __init svm_init(void)
1857 return kvm_init_arch(&svm_arch_ops, sizeof(struct vcpu_svm),
1858 THIS_MODULE);
1861 static void __exit svm_exit(void)
1863 kvm_exit_arch();
1866 module_init(svm_init)
1867 module_exit(svm_exit)