KVM: Remove pointless desc_ptr #ifdef
[linux-2.6/verdex.git] / arch / x86 / kvm / svm.c
blob51741f96e7fb06d80721542eff09cf2b00ed54d7
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.
16 #include <linux/kvm_host.h>
18 #include "kvm_svm.h"
19 #include "irq.h"
20 #include "mmu.h"
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>
28 #include <asm/desc.h>
30 MODULE_AUTHOR("Qumranet");
31 MODULE_LICENSE("GPL");
33 #define IOPM_ALLOC_ORDER 2
34 #define MSRPM_ALLOC_ORDER 1
36 #define DB_VECTOR 1
37 #define UD_VECTOR 6
38 #define GP_VECTOR 13
40 #define DR7_GD_MASK (1 << 13)
41 #define DR6_BD_MASK (1 << 13)
43 #define SEG_TYPE_LDT 2
44 #define SEG_TYPE_BUSY_TSS16 3
46 #define SVM_FEATURE_NPT (1 << 0)
47 #define SVM_FEATURE_LBRV (1 << 1)
48 #define SVM_DEATURE_SVML (1 << 2)
50 #define DEBUGCTL_RESERVED_BITS (~(0x3fULL))
52 /* enable NPT for AMD64 and X86 with PAE */
53 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
54 static bool npt_enabled = true;
55 #else
56 static bool npt_enabled = false;
57 #endif
58 static int npt = 1;
60 module_param(npt, int, S_IRUGO);
62 static void kvm_reput_irq(struct vcpu_svm *svm);
64 static inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu)
66 return container_of(vcpu, struct vcpu_svm, vcpu);
69 static unsigned long iopm_base;
71 struct kvm_ldttss_desc {
72 u16 limit0;
73 u16 base0;
74 unsigned base1 : 8, type : 5, dpl : 2, p : 1;
75 unsigned limit1 : 4, zero0 : 3, g : 1, base2 : 8;
76 u32 base3;
77 u32 zero1;
78 } __attribute__((packed));
80 struct svm_cpu_data {
81 int cpu;
83 u64 asid_generation;
84 u32 max_asid;
85 u32 next_asid;
86 struct kvm_ldttss_desc *tss_desc;
88 struct page *save_area;
91 static DEFINE_PER_CPU(struct svm_cpu_data *, svm_data);
92 static uint32_t svm_features;
94 struct svm_init_data {
95 int cpu;
96 int r;
99 static u32 msrpm_ranges[] = {0, 0xc0000000, 0xc0010000};
101 #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
102 #define MSRS_RANGE_SIZE 2048
103 #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
105 #define MAX_INST_SIZE 15
107 static inline u32 svm_has(u32 feat)
109 return svm_features & feat;
112 static inline u8 pop_irq(struct kvm_vcpu *vcpu)
114 int word_index = __ffs(vcpu->arch.irq_summary);
115 int bit_index = __ffs(vcpu->arch.irq_pending[word_index]);
116 int irq = word_index * BITS_PER_LONG + bit_index;
118 clear_bit(bit_index, &vcpu->arch.irq_pending[word_index]);
119 if (!vcpu->arch.irq_pending[word_index])
120 clear_bit(word_index, &vcpu->arch.irq_summary);
121 return irq;
124 static inline void push_irq(struct kvm_vcpu *vcpu, u8 irq)
126 set_bit(irq, vcpu->arch.irq_pending);
127 set_bit(irq / BITS_PER_LONG, &vcpu->arch.irq_summary);
130 static inline void clgi(void)
132 asm volatile (SVM_CLGI);
135 static inline void stgi(void)
137 asm volatile (SVM_STGI);
140 static inline void invlpga(unsigned long addr, u32 asid)
142 asm volatile (SVM_INVLPGA :: "a"(addr), "c"(asid));
145 static inline unsigned long kvm_read_cr2(void)
147 unsigned long cr2;
149 asm volatile ("mov %%cr2, %0" : "=r" (cr2));
150 return cr2;
153 static inline void kvm_write_cr2(unsigned long val)
155 asm volatile ("mov %0, %%cr2" :: "r" (val));
158 static inline unsigned long read_dr6(void)
160 unsigned long dr6;
162 asm volatile ("mov %%dr6, %0" : "=r" (dr6));
163 return dr6;
166 static inline void write_dr6(unsigned long val)
168 asm volatile ("mov %0, %%dr6" :: "r" (val));
171 static inline unsigned long read_dr7(void)
173 unsigned long dr7;
175 asm volatile ("mov %%dr7, %0" : "=r" (dr7));
176 return dr7;
179 static inline void write_dr7(unsigned long val)
181 asm volatile ("mov %0, %%dr7" :: "r" (val));
184 static inline void force_new_asid(struct kvm_vcpu *vcpu)
186 to_svm(vcpu)->asid_generation--;
189 static inline void flush_guest_tlb(struct kvm_vcpu *vcpu)
191 force_new_asid(vcpu);
194 static void svm_set_efer(struct kvm_vcpu *vcpu, u64 efer)
196 if (!npt_enabled && !(efer & EFER_LMA))
197 efer &= ~EFER_LME;
199 to_svm(vcpu)->vmcb->save.efer = efer | MSR_EFER_SVME_MASK;
200 vcpu->arch.shadow_efer = efer;
203 static void svm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
204 bool has_error_code, u32 error_code)
206 struct vcpu_svm *svm = to_svm(vcpu);
208 svm->vmcb->control.event_inj = nr
209 | SVM_EVTINJ_VALID
210 | (has_error_code ? SVM_EVTINJ_VALID_ERR : 0)
211 | SVM_EVTINJ_TYPE_EXEPT;
212 svm->vmcb->control.event_inj_err = error_code;
215 static bool svm_exception_injected(struct kvm_vcpu *vcpu)
217 struct vcpu_svm *svm = to_svm(vcpu);
219 return !(svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_VALID);
222 static int is_external_interrupt(u32 info)
224 info &= SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
225 return info == (SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR);
228 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
230 struct vcpu_svm *svm = to_svm(vcpu);
232 if (!svm->next_rip) {
233 printk(KERN_DEBUG "%s: NOP\n", __func__);
234 return;
236 if (svm->next_rip - svm->vmcb->save.rip > MAX_INST_SIZE)
237 printk(KERN_ERR "%s: ip 0x%llx next 0x%llx\n",
238 __func__,
239 svm->vmcb->save.rip,
240 svm->next_rip);
242 vcpu->arch.rip = svm->vmcb->save.rip = svm->next_rip;
243 svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK;
245 vcpu->arch.interrupt_window_open = 1;
248 static int has_svm(void)
250 uint32_t eax, ebx, ecx, edx;
252 if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD) {
253 printk(KERN_INFO "has_svm: not amd\n");
254 return 0;
257 cpuid(0x80000000, &eax, &ebx, &ecx, &edx);
258 if (eax < SVM_CPUID_FUNC) {
259 printk(KERN_INFO "has_svm: can't execute cpuid_8000000a\n");
260 return 0;
263 cpuid(0x80000001, &eax, &ebx, &ecx, &edx);
264 if (!(ecx & (1 << SVM_CPUID_FEATURE_SHIFT))) {
265 printk(KERN_DEBUG "has_svm: svm not available\n");
266 return 0;
268 return 1;
271 static void svm_hardware_disable(void *garbage)
273 struct svm_cpu_data *svm_data
274 = per_cpu(svm_data, raw_smp_processor_id());
276 if (svm_data) {
277 uint64_t efer;
279 wrmsrl(MSR_VM_HSAVE_PA, 0);
280 rdmsrl(MSR_EFER, efer);
281 wrmsrl(MSR_EFER, efer & ~MSR_EFER_SVME_MASK);
282 per_cpu(svm_data, raw_smp_processor_id()) = NULL;
283 __free_page(svm_data->save_area);
284 kfree(svm_data);
288 static void svm_hardware_enable(void *garbage)
291 struct svm_cpu_data *svm_data;
292 uint64_t efer;
293 struct desc_ptr gdt_descr;
294 struct desc_struct *gdt;
295 int me = raw_smp_processor_id();
297 if (!has_svm()) {
298 printk(KERN_ERR "svm_cpu_init: err EOPNOTSUPP on %d\n", me);
299 return;
301 svm_data = per_cpu(svm_data, me);
303 if (!svm_data) {
304 printk(KERN_ERR "svm_cpu_init: svm_data is NULL on %d\n",
305 me);
306 return;
309 svm_data->asid_generation = 1;
310 svm_data->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1;
311 svm_data->next_asid = svm_data->max_asid + 1;
313 asm volatile ("sgdt %0" : "=m"(gdt_descr));
314 gdt = (struct desc_struct *)gdt_descr.address;
315 svm_data->tss_desc = (struct kvm_ldttss_desc *)(gdt + GDT_ENTRY_TSS);
317 rdmsrl(MSR_EFER, efer);
318 wrmsrl(MSR_EFER, efer | MSR_EFER_SVME_MASK);
320 wrmsrl(MSR_VM_HSAVE_PA,
321 page_to_pfn(svm_data->save_area) << PAGE_SHIFT);
324 static int svm_cpu_init(int cpu)
326 struct svm_cpu_data *svm_data;
327 int r;
329 svm_data = kzalloc(sizeof(struct svm_cpu_data), GFP_KERNEL);
330 if (!svm_data)
331 return -ENOMEM;
332 svm_data->cpu = cpu;
333 svm_data->save_area = alloc_page(GFP_KERNEL);
334 r = -ENOMEM;
335 if (!svm_data->save_area)
336 goto err_1;
338 per_cpu(svm_data, cpu) = svm_data;
340 return 0;
342 err_1:
343 kfree(svm_data);
344 return r;
348 static void set_msr_interception(u32 *msrpm, unsigned msr,
349 int read, int write)
351 int i;
353 for (i = 0; i < NUM_MSR_MAPS; i++) {
354 if (msr >= msrpm_ranges[i] &&
355 msr < msrpm_ranges[i] + MSRS_IN_RANGE) {
356 u32 msr_offset = (i * MSRS_IN_RANGE + msr -
357 msrpm_ranges[i]) * 2;
359 u32 *base = msrpm + (msr_offset / 32);
360 u32 msr_shift = msr_offset % 32;
361 u32 mask = ((write) ? 0 : 2) | ((read) ? 0 : 1);
362 *base = (*base & ~(0x3 << msr_shift)) |
363 (mask << msr_shift);
364 return;
367 BUG();
370 static void svm_vcpu_init_msrpm(u32 *msrpm)
372 memset(msrpm, 0xff, PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER));
374 #ifdef CONFIG_X86_64
375 set_msr_interception(msrpm, MSR_GS_BASE, 1, 1);
376 set_msr_interception(msrpm, MSR_FS_BASE, 1, 1);
377 set_msr_interception(msrpm, MSR_KERNEL_GS_BASE, 1, 1);
378 set_msr_interception(msrpm, MSR_LSTAR, 1, 1);
379 set_msr_interception(msrpm, MSR_CSTAR, 1, 1);
380 set_msr_interception(msrpm, MSR_SYSCALL_MASK, 1, 1);
381 #endif
382 set_msr_interception(msrpm, MSR_K6_STAR, 1, 1);
383 set_msr_interception(msrpm, MSR_IA32_SYSENTER_CS, 1, 1);
384 set_msr_interception(msrpm, MSR_IA32_SYSENTER_ESP, 1, 1);
385 set_msr_interception(msrpm, MSR_IA32_SYSENTER_EIP, 1, 1);
388 static void svm_enable_lbrv(struct vcpu_svm *svm)
390 u32 *msrpm = svm->msrpm;
392 svm->vmcb->control.lbr_ctl = 1;
393 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 1, 1);
394 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 1, 1);
395 set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 1, 1);
396 set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 1, 1);
399 static void svm_disable_lbrv(struct vcpu_svm *svm)
401 u32 *msrpm = svm->msrpm;
403 svm->vmcb->control.lbr_ctl = 0;
404 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 0, 0);
405 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 0, 0);
406 set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 0, 0);
407 set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 0, 0);
410 static __init int svm_hardware_setup(void)
412 int cpu;
413 struct page *iopm_pages;
414 void *iopm_va;
415 int r;
417 iopm_pages = alloc_pages(GFP_KERNEL, IOPM_ALLOC_ORDER);
419 if (!iopm_pages)
420 return -ENOMEM;
422 iopm_va = page_address(iopm_pages);
423 memset(iopm_va, 0xff, PAGE_SIZE * (1 << IOPM_ALLOC_ORDER));
424 clear_bit(0x80, iopm_va); /* allow direct access to PC debug port */
425 iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT;
427 if (boot_cpu_has(X86_FEATURE_NX))
428 kvm_enable_efer_bits(EFER_NX);
430 for_each_online_cpu(cpu) {
431 r = svm_cpu_init(cpu);
432 if (r)
433 goto err;
436 svm_features = cpuid_edx(SVM_CPUID_FUNC);
438 if (!svm_has(SVM_FEATURE_NPT))
439 npt_enabled = false;
441 if (npt_enabled && !npt) {
442 printk(KERN_INFO "kvm: Nested Paging disabled\n");
443 npt_enabled = false;
446 if (npt_enabled) {
447 printk(KERN_INFO "kvm: Nested Paging enabled\n");
448 kvm_enable_tdp();
451 return 0;
453 err:
454 __free_pages(iopm_pages, IOPM_ALLOC_ORDER);
455 iopm_base = 0;
456 return r;
459 static __exit void svm_hardware_unsetup(void)
461 __free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT), IOPM_ALLOC_ORDER);
462 iopm_base = 0;
465 static void init_seg(struct vmcb_seg *seg)
467 seg->selector = 0;
468 seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK |
469 SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */
470 seg->limit = 0xffff;
471 seg->base = 0;
474 static void init_sys_seg(struct vmcb_seg *seg, uint32_t type)
476 seg->selector = 0;
477 seg->attrib = SVM_SELECTOR_P_MASK | type;
478 seg->limit = 0xffff;
479 seg->base = 0;
482 static void init_vmcb(struct vcpu_svm *svm)
484 struct vmcb_control_area *control = &svm->vmcb->control;
485 struct vmcb_save_area *save = &svm->vmcb->save;
487 control->intercept_cr_read = INTERCEPT_CR0_MASK |
488 INTERCEPT_CR3_MASK |
489 INTERCEPT_CR4_MASK |
490 INTERCEPT_CR8_MASK;
492 control->intercept_cr_write = INTERCEPT_CR0_MASK |
493 INTERCEPT_CR3_MASK |
494 INTERCEPT_CR4_MASK |
495 INTERCEPT_CR8_MASK;
497 control->intercept_dr_read = INTERCEPT_DR0_MASK |
498 INTERCEPT_DR1_MASK |
499 INTERCEPT_DR2_MASK |
500 INTERCEPT_DR3_MASK;
502 control->intercept_dr_write = INTERCEPT_DR0_MASK |
503 INTERCEPT_DR1_MASK |
504 INTERCEPT_DR2_MASK |
505 INTERCEPT_DR3_MASK |
506 INTERCEPT_DR5_MASK |
507 INTERCEPT_DR7_MASK;
509 control->intercept_exceptions = (1 << PF_VECTOR) |
510 (1 << UD_VECTOR);
513 control->intercept = (1ULL << INTERCEPT_INTR) |
514 (1ULL << INTERCEPT_NMI) |
515 (1ULL << INTERCEPT_SMI) |
517 * selective cr0 intercept bug?
518 * 0: 0f 22 d8 mov %eax,%cr3
519 * 3: 0f 20 c0 mov %cr0,%eax
520 * 6: 0d 00 00 00 80 or $0x80000000,%eax
521 * b: 0f 22 c0 mov %eax,%cr0
522 * set cr3 ->interception
523 * get cr0 ->interception
524 * set cr0 -> no interception
526 /* (1ULL << INTERCEPT_SELECTIVE_CR0) | */
527 (1ULL << INTERCEPT_CPUID) |
528 (1ULL << INTERCEPT_INVD) |
529 (1ULL << INTERCEPT_HLT) |
530 (1ULL << INTERCEPT_INVLPGA) |
531 (1ULL << INTERCEPT_IOIO_PROT) |
532 (1ULL << INTERCEPT_MSR_PROT) |
533 (1ULL << INTERCEPT_TASK_SWITCH) |
534 (1ULL << INTERCEPT_SHUTDOWN) |
535 (1ULL << INTERCEPT_VMRUN) |
536 (1ULL << INTERCEPT_VMMCALL) |
537 (1ULL << INTERCEPT_VMLOAD) |
538 (1ULL << INTERCEPT_VMSAVE) |
539 (1ULL << INTERCEPT_STGI) |
540 (1ULL << INTERCEPT_CLGI) |
541 (1ULL << INTERCEPT_SKINIT) |
542 (1ULL << INTERCEPT_WBINVD) |
543 (1ULL << INTERCEPT_MONITOR) |
544 (1ULL << INTERCEPT_MWAIT);
546 control->iopm_base_pa = iopm_base;
547 control->msrpm_base_pa = __pa(svm->msrpm);
548 control->tsc_offset = 0;
549 control->int_ctl = V_INTR_MASKING_MASK;
551 init_seg(&save->es);
552 init_seg(&save->ss);
553 init_seg(&save->ds);
554 init_seg(&save->fs);
555 init_seg(&save->gs);
557 save->cs.selector = 0xf000;
558 /* Executable/Readable Code Segment */
559 save->cs.attrib = SVM_SELECTOR_READ_MASK | SVM_SELECTOR_P_MASK |
560 SVM_SELECTOR_S_MASK | SVM_SELECTOR_CODE_MASK;
561 save->cs.limit = 0xffff;
563 * cs.base should really be 0xffff0000, but vmx can't handle that, so
564 * be consistent with it.
566 * Replace when we have real mode working for vmx.
568 save->cs.base = 0xf0000;
570 save->gdtr.limit = 0xffff;
571 save->idtr.limit = 0xffff;
573 init_sys_seg(&save->ldtr, SEG_TYPE_LDT);
574 init_sys_seg(&save->tr, SEG_TYPE_BUSY_TSS16);
576 save->efer = MSR_EFER_SVME_MASK;
577 save->dr6 = 0xffff0ff0;
578 save->dr7 = 0x400;
579 save->rflags = 2;
580 save->rip = 0x0000fff0;
583 * cr0 val on cpu init should be 0x60000010, we enable cpu
584 * cache by default. the orderly way is to enable cache in bios.
586 save->cr0 = 0x00000010 | X86_CR0_PG | X86_CR0_WP;
587 save->cr4 = X86_CR4_PAE;
588 /* rdx = ?? */
590 if (npt_enabled) {
591 /* Setup VMCB for Nested Paging */
592 control->nested_ctl = 1;
593 control->intercept_exceptions &= ~(1 << PF_VECTOR);
594 control->intercept_cr_read &= ~(INTERCEPT_CR0_MASK|
595 INTERCEPT_CR3_MASK);
596 control->intercept_cr_write &= ~(INTERCEPT_CR0_MASK|
597 INTERCEPT_CR3_MASK);
598 save->g_pat = 0x0007040600070406ULL;
599 /* enable caching because the QEMU Bios doesn't enable it */
600 save->cr0 = X86_CR0_ET;
601 save->cr3 = 0;
602 save->cr4 = 0;
607 static int svm_vcpu_reset(struct kvm_vcpu *vcpu)
609 struct vcpu_svm *svm = to_svm(vcpu);
611 init_vmcb(svm);
613 if (vcpu->vcpu_id != 0) {
614 svm->vmcb->save.rip = 0;
615 svm->vmcb->save.cs.base = svm->vcpu.arch.sipi_vector << 12;
616 svm->vmcb->save.cs.selector = svm->vcpu.arch.sipi_vector << 8;
619 return 0;
622 static struct kvm_vcpu *svm_create_vcpu(struct kvm *kvm, unsigned int id)
624 struct vcpu_svm *svm;
625 struct page *page;
626 struct page *msrpm_pages;
627 int err;
629 svm = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
630 if (!svm) {
631 err = -ENOMEM;
632 goto out;
635 err = kvm_vcpu_init(&svm->vcpu, kvm, id);
636 if (err)
637 goto free_svm;
639 page = alloc_page(GFP_KERNEL);
640 if (!page) {
641 err = -ENOMEM;
642 goto uninit;
645 err = -ENOMEM;
646 msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
647 if (!msrpm_pages)
648 goto uninit;
649 svm->msrpm = page_address(msrpm_pages);
650 svm_vcpu_init_msrpm(svm->msrpm);
652 svm->vmcb = page_address(page);
653 clear_page(svm->vmcb);
654 svm->vmcb_pa = page_to_pfn(page) << PAGE_SHIFT;
655 svm->asid_generation = 0;
656 memset(svm->db_regs, 0, sizeof(svm->db_regs));
657 init_vmcb(svm);
659 fx_init(&svm->vcpu);
660 svm->vcpu.fpu_active = 1;
661 svm->vcpu.arch.apic_base = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
662 if (svm->vcpu.vcpu_id == 0)
663 svm->vcpu.arch.apic_base |= MSR_IA32_APICBASE_BSP;
665 return &svm->vcpu;
667 uninit:
668 kvm_vcpu_uninit(&svm->vcpu);
669 free_svm:
670 kmem_cache_free(kvm_vcpu_cache, svm);
671 out:
672 return ERR_PTR(err);
675 static void svm_free_vcpu(struct kvm_vcpu *vcpu)
677 struct vcpu_svm *svm = to_svm(vcpu);
679 __free_page(pfn_to_page(svm->vmcb_pa >> PAGE_SHIFT));
680 __free_pages(virt_to_page(svm->msrpm), MSRPM_ALLOC_ORDER);
681 kvm_vcpu_uninit(vcpu);
682 kmem_cache_free(kvm_vcpu_cache, svm);
685 static void svm_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
687 struct vcpu_svm *svm = to_svm(vcpu);
688 int i;
690 if (unlikely(cpu != vcpu->cpu)) {
691 u64 tsc_this, delta;
694 * Make sure that the guest sees a monotonically
695 * increasing TSC.
697 rdtscll(tsc_this);
698 delta = vcpu->arch.host_tsc - tsc_this;
699 svm->vmcb->control.tsc_offset += delta;
700 vcpu->cpu = cpu;
701 kvm_migrate_apic_timer(vcpu);
704 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
705 rdmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
708 static void svm_vcpu_put(struct kvm_vcpu *vcpu)
710 struct vcpu_svm *svm = to_svm(vcpu);
711 int i;
713 ++vcpu->stat.host_state_reload;
714 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
715 wrmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
717 rdtscll(vcpu->arch.host_tsc);
720 static void svm_vcpu_decache(struct kvm_vcpu *vcpu)
724 static void svm_cache_regs(struct kvm_vcpu *vcpu)
726 struct vcpu_svm *svm = to_svm(vcpu);
728 vcpu->arch.regs[VCPU_REGS_RAX] = svm->vmcb->save.rax;
729 vcpu->arch.regs[VCPU_REGS_RSP] = svm->vmcb->save.rsp;
730 vcpu->arch.rip = svm->vmcb->save.rip;
733 static void svm_decache_regs(struct kvm_vcpu *vcpu)
735 struct vcpu_svm *svm = to_svm(vcpu);
736 svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX];
737 svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP];
738 svm->vmcb->save.rip = vcpu->arch.rip;
741 static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu)
743 return to_svm(vcpu)->vmcb->save.rflags;
746 static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
748 to_svm(vcpu)->vmcb->save.rflags = rflags;
751 static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg)
753 struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
755 switch (seg) {
756 case VCPU_SREG_CS: return &save->cs;
757 case VCPU_SREG_DS: return &save->ds;
758 case VCPU_SREG_ES: return &save->es;
759 case VCPU_SREG_FS: return &save->fs;
760 case VCPU_SREG_GS: return &save->gs;
761 case VCPU_SREG_SS: return &save->ss;
762 case VCPU_SREG_TR: return &save->tr;
763 case VCPU_SREG_LDTR: return &save->ldtr;
765 BUG();
766 return NULL;
769 static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg)
771 struct vmcb_seg *s = svm_seg(vcpu, seg);
773 return s->base;
776 static void svm_get_segment(struct kvm_vcpu *vcpu,
777 struct kvm_segment *var, int seg)
779 struct vmcb_seg *s = svm_seg(vcpu, seg);
781 var->base = s->base;
782 var->limit = s->limit;
783 var->selector = s->selector;
784 var->type = s->attrib & SVM_SELECTOR_TYPE_MASK;
785 var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1;
786 var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
787 var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1;
788 var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1;
789 var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
790 var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
791 var->g = (s->attrib >> SVM_SELECTOR_G_SHIFT) & 1;
792 var->unusable = !var->present;
795 static void svm_get_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
797 struct vcpu_svm *svm = to_svm(vcpu);
799 dt->limit = svm->vmcb->save.idtr.limit;
800 dt->base = svm->vmcb->save.idtr.base;
803 static void svm_set_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
805 struct vcpu_svm *svm = to_svm(vcpu);
807 svm->vmcb->save.idtr.limit = dt->limit;
808 svm->vmcb->save.idtr.base = dt->base ;
811 static void svm_get_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
813 struct vcpu_svm *svm = to_svm(vcpu);
815 dt->limit = svm->vmcb->save.gdtr.limit;
816 dt->base = svm->vmcb->save.gdtr.base;
819 static void svm_set_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
821 struct vcpu_svm *svm = to_svm(vcpu);
823 svm->vmcb->save.gdtr.limit = dt->limit;
824 svm->vmcb->save.gdtr.base = dt->base ;
827 static void svm_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
831 static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
833 struct vcpu_svm *svm = to_svm(vcpu);
835 #ifdef CONFIG_X86_64
836 if (vcpu->arch.shadow_efer & EFER_LME) {
837 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
838 vcpu->arch.shadow_efer |= EFER_LMA;
839 svm->vmcb->save.efer |= EFER_LMA | EFER_LME;
842 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG)) {
843 vcpu->arch.shadow_efer &= ~EFER_LMA;
844 svm->vmcb->save.efer &= ~(EFER_LMA | EFER_LME);
847 #endif
848 if (npt_enabled)
849 goto set;
851 if ((vcpu->arch.cr0 & X86_CR0_TS) && !(cr0 & X86_CR0_TS)) {
852 svm->vmcb->control.intercept_exceptions &= ~(1 << NM_VECTOR);
853 vcpu->fpu_active = 1;
856 vcpu->arch.cr0 = cr0;
857 cr0 |= X86_CR0_PG | X86_CR0_WP;
858 if (!vcpu->fpu_active) {
859 svm->vmcb->control.intercept_exceptions |= (1 << NM_VECTOR);
860 cr0 |= X86_CR0_TS;
862 set:
864 * re-enable caching here because the QEMU bios
865 * does not do it - this results in some delay at
866 * reboot
868 cr0 &= ~(X86_CR0_CD | X86_CR0_NW);
869 svm->vmcb->save.cr0 = cr0;
872 static void svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
874 vcpu->arch.cr4 = cr4;
875 if (!npt_enabled)
876 cr4 |= X86_CR4_PAE;
877 to_svm(vcpu)->vmcb->save.cr4 = cr4;
880 static void svm_set_segment(struct kvm_vcpu *vcpu,
881 struct kvm_segment *var, int seg)
883 struct vcpu_svm *svm = to_svm(vcpu);
884 struct vmcb_seg *s = svm_seg(vcpu, seg);
886 s->base = var->base;
887 s->limit = var->limit;
888 s->selector = var->selector;
889 if (var->unusable)
890 s->attrib = 0;
891 else {
892 s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK);
893 s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT;
894 s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT;
895 s->attrib |= (var->present & 1) << SVM_SELECTOR_P_SHIFT;
896 s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT;
897 s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT;
898 s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT;
899 s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT;
901 if (seg == VCPU_SREG_CS)
902 svm->vmcb->save.cpl
903 = (svm->vmcb->save.cs.attrib
904 >> SVM_SELECTOR_DPL_SHIFT) & 3;
908 /* FIXME:
910 svm(vcpu)->vmcb->control.int_ctl &= ~V_TPR_MASK;
911 svm(vcpu)->vmcb->control.int_ctl |= (sregs->cr8 & V_TPR_MASK);
915 static int svm_guest_debug(struct kvm_vcpu *vcpu, struct kvm_debug_guest *dbg)
917 return -EOPNOTSUPP;
920 static int svm_get_irq(struct kvm_vcpu *vcpu)
922 struct vcpu_svm *svm = to_svm(vcpu);
923 u32 exit_int_info = svm->vmcb->control.exit_int_info;
925 if (is_external_interrupt(exit_int_info))
926 return exit_int_info & SVM_EVTINJ_VEC_MASK;
927 return -1;
930 static void load_host_msrs(struct kvm_vcpu *vcpu)
932 #ifdef CONFIG_X86_64
933 wrmsrl(MSR_GS_BASE, to_svm(vcpu)->host_gs_base);
934 #endif
937 static void save_host_msrs(struct kvm_vcpu *vcpu)
939 #ifdef CONFIG_X86_64
940 rdmsrl(MSR_GS_BASE, to_svm(vcpu)->host_gs_base);
941 #endif
944 static void new_asid(struct vcpu_svm *svm, struct svm_cpu_data *svm_data)
946 if (svm_data->next_asid > svm_data->max_asid) {
947 ++svm_data->asid_generation;
948 svm_data->next_asid = 1;
949 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID;
952 svm->vcpu.cpu = svm_data->cpu;
953 svm->asid_generation = svm_data->asid_generation;
954 svm->vmcb->control.asid = svm_data->next_asid++;
957 static unsigned long svm_get_dr(struct kvm_vcpu *vcpu, int dr)
959 return to_svm(vcpu)->db_regs[dr];
962 static void svm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long value,
963 int *exception)
965 struct vcpu_svm *svm = to_svm(vcpu);
967 *exception = 0;
969 if (svm->vmcb->save.dr7 & DR7_GD_MASK) {
970 svm->vmcb->save.dr7 &= ~DR7_GD_MASK;
971 svm->vmcb->save.dr6 |= DR6_BD_MASK;
972 *exception = DB_VECTOR;
973 return;
976 switch (dr) {
977 case 0 ... 3:
978 svm->db_regs[dr] = value;
979 return;
980 case 4 ... 5:
981 if (vcpu->arch.cr4 & X86_CR4_DE) {
982 *exception = UD_VECTOR;
983 return;
985 case 7: {
986 if (value & ~((1ULL << 32) - 1)) {
987 *exception = GP_VECTOR;
988 return;
990 svm->vmcb->save.dr7 = value;
991 return;
993 default:
994 printk(KERN_DEBUG "%s: unexpected dr %u\n",
995 __func__, dr);
996 *exception = UD_VECTOR;
997 return;
1001 static int pf_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1003 u32 exit_int_info = svm->vmcb->control.exit_int_info;
1004 struct kvm *kvm = svm->vcpu.kvm;
1005 u64 fault_address;
1006 u32 error_code;
1008 if (!irqchip_in_kernel(kvm) &&
1009 is_external_interrupt(exit_int_info))
1010 push_irq(&svm->vcpu, exit_int_info & SVM_EVTINJ_VEC_MASK);
1012 fault_address = svm->vmcb->control.exit_info_2;
1013 error_code = svm->vmcb->control.exit_info_1;
1014 return kvm_mmu_page_fault(&svm->vcpu, fault_address, error_code);
1017 static int ud_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1019 int er;
1021 er = emulate_instruction(&svm->vcpu, kvm_run, 0, 0, EMULTYPE_TRAP_UD);
1022 if (er != EMULATE_DONE)
1023 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
1024 return 1;
1027 static int nm_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1029 svm->vmcb->control.intercept_exceptions &= ~(1 << NM_VECTOR);
1030 if (!(svm->vcpu.arch.cr0 & X86_CR0_TS))
1031 svm->vmcb->save.cr0 &= ~X86_CR0_TS;
1032 svm->vcpu.fpu_active = 1;
1034 return 1;
1037 static int shutdown_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1040 * VMCB is undefined after a SHUTDOWN intercept
1041 * so reinitialize it.
1043 clear_page(svm->vmcb);
1044 init_vmcb(svm);
1046 kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
1047 return 0;
1050 static int io_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1052 u32 io_info = svm->vmcb->control.exit_info_1; /* address size bug? */
1053 int size, down, in, string, rep;
1054 unsigned port;
1056 ++svm->vcpu.stat.io_exits;
1058 svm->next_rip = svm->vmcb->control.exit_info_2;
1060 string = (io_info & SVM_IOIO_STR_MASK) != 0;
1062 if (string) {
1063 if (emulate_instruction(&svm->vcpu,
1064 kvm_run, 0, 0, 0) == EMULATE_DO_MMIO)
1065 return 0;
1066 return 1;
1069 in = (io_info & SVM_IOIO_TYPE_MASK) != 0;
1070 port = io_info >> 16;
1071 size = (io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT;
1072 rep = (io_info & SVM_IOIO_REP_MASK) != 0;
1073 down = (svm->vmcb->save.rflags & X86_EFLAGS_DF) != 0;
1075 return kvm_emulate_pio(&svm->vcpu, kvm_run, in, size, port);
1078 static int nop_on_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1080 return 1;
1083 static int halt_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1085 svm->next_rip = svm->vmcb->save.rip + 1;
1086 skip_emulated_instruction(&svm->vcpu);
1087 return kvm_emulate_halt(&svm->vcpu);
1090 static int vmmcall_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1092 svm->next_rip = svm->vmcb->save.rip + 3;
1093 skip_emulated_instruction(&svm->vcpu);
1094 kvm_emulate_hypercall(&svm->vcpu);
1095 return 1;
1098 static int invalid_op_interception(struct vcpu_svm *svm,
1099 struct kvm_run *kvm_run)
1101 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
1102 return 1;
1105 static int task_switch_interception(struct vcpu_svm *svm,
1106 struct kvm_run *kvm_run)
1108 pr_unimpl(&svm->vcpu, "%s: task switch is unsupported\n", __func__);
1109 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
1110 return 0;
1113 static int cpuid_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1115 svm->next_rip = svm->vmcb->save.rip + 2;
1116 kvm_emulate_cpuid(&svm->vcpu);
1117 return 1;
1120 static int emulate_on_interception(struct vcpu_svm *svm,
1121 struct kvm_run *kvm_run)
1123 if (emulate_instruction(&svm->vcpu, NULL, 0, 0, 0) != EMULATE_DONE)
1124 pr_unimpl(&svm->vcpu, "%s: failed\n", __func__);
1125 return 1;
1128 static int cr8_write_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1130 emulate_instruction(&svm->vcpu, NULL, 0, 0, 0);
1131 if (irqchip_in_kernel(svm->vcpu.kvm))
1132 return 1;
1133 kvm_run->exit_reason = KVM_EXIT_SET_TPR;
1134 return 0;
1137 static int svm_get_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 *data)
1139 struct vcpu_svm *svm = to_svm(vcpu);
1141 switch (ecx) {
1142 case MSR_IA32_TIME_STAMP_COUNTER: {
1143 u64 tsc;
1145 rdtscll(tsc);
1146 *data = svm->vmcb->control.tsc_offset + tsc;
1147 break;
1149 case MSR_K6_STAR:
1150 *data = svm->vmcb->save.star;
1151 break;
1152 #ifdef CONFIG_X86_64
1153 case MSR_LSTAR:
1154 *data = svm->vmcb->save.lstar;
1155 break;
1156 case MSR_CSTAR:
1157 *data = svm->vmcb->save.cstar;
1158 break;
1159 case MSR_KERNEL_GS_BASE:
1160 *data = svm->vmcb->save.kernel_gs_base;
1161 break;
1162 case MSR_SYSCALL_MASK:
1163 *data = svm->vmcb->save.sfmask;
1164 break;
1165 #endif
1166 case MSR_IA32_SYSENTER_CS:
1167 *data = svm->vmcb->save.sysenter_cs;
1168 break;
1169 case MSR_IA32_SYSENTER_EIP:
1170 *data = svm->vmcb->save.sysenter_eip;
1171 break;
1172 case MSR_IA32_SYSENTER_ESP:
1173 *data = svm->vmcb->save.sysenter_esp;
1174 break;
1175 /* Nobody will change the following 5 values in the VMCB so
1176 we can safely return them on rdmsr. They will always be 0
1177 until LBRV is implemented. */
1178 case MSR_IA32_DEBUGCTLMSR:
1179 *data = svm->vmcb->save.dbgctl;
1180 break;
1181 case MSR_IA32_LASTBRANCHFROMIP:
1182 *data = svm->vmcb->save.br_from;
1183 break;
1184 case MSR_IA32_LASTBRANCHTOIP:
1185 *data = svm->vmcb->save.br_to;
1186 break;
1187 case MSR_IA32_LASTINTFROMIP:
1188 *data = svm->vmcb->save.last_excp_from;
1189 break;
1190 case MSR_IA32_LASTINTTOIP:
1191 *data = svm->vmcb->save.last_excp_to;
1192 break;
1193 default:
1194 return kvm_get_msr_common(vcpu, ecx, data);
1196 return 0;
1199 static int rdmsr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1201 u32 ecx = svm->vcpu.arch.regs[VCPU_REGS_RCX];
1202 u64 data;
1204 if (svm_get_msr(&svm->vcpu, ecx, &data))
1205 kvm_inject_gp(&svm->vcpu, 0);
1206 else {
1207 svm->vmcb->save.rax = data & 0xffffffff;
1208 svm->vcpu.arch.regs[VCPU_REGS_RDX] = data >> 32;
1209 svm->next_rip = svm->vmcb->save.rip + 2;
1210 skip_emulated_instruction(&svm->vcpu);
1212 return 1;
1215 static int svm_set_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 data)
1217 struct vcpu_svm *svm = to_svm(vcpu);
1219 switch (ecx) {
1220 case MSR_IA32_TIME_STAMP_COUNTER: {
1221 u64 tsc;
1223 rdtscll(tsc);
1224 svm->vmcb->control.tsc_offset = data - tsc;
1225 break;
1227 case MSR_K6_STAR:
1228 svm->vmcb->save.star = data;
1229 break;
1230 #ifdef CONFIG_X86_64
1231 case MSR_LSTAR:
1232 svm->vmcb->save.lstar = data;
1233 break;
1234 case MSR_CSTAR:
1235 svm->vmcb->save.cstar = data;
1236 break;
1237 case MSR_KERNEL_GS_BASE:
1238 svm->vmcb->save.kernel_gs_base = data;
1239 break;
1240 case MSR_SYSCALL_MASK:
1241 svm->vmcb->save.sfmask = data;
1242 break;
1243 #endif
1244 case MSR_IA32_SYSENTER_CS:
1245 svm->vmcb->save.sysenter_cs = data;
1246 break;
1247 case MSR_IA32_SYSENTER_EIP:
1248 svm->vmcb->save.sysenter_eip = data;
1249 break;
1250 case MSR_IA32_SYSENTER_ESP:
1251 svm->vmcb->save.sysenter_esp = data;
1252 break;
1253 case MSR_IA32_DEBUGCTLMSR:
1254 if (!svm_has(SVM_FEATURE_LBRV)) {
1255 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTL 0x%llx, nop\n",
1256 __func__, data);
1257 break;
1259 if (data & DEBUGCTL_RESERVED_BITS)
1260 return 1;
1262 svm->vmcb->save.dbgctl = data;
1263 if (data & (1ULL<<0))
1264 svm_enable_lbrv(svm);
1265 else
1266 svm_disable_lbrv(svm);
1267 break;
1268 case MSR_K7_EVNTSEL0:
1269 case MSR_K7_EVNTSEL1:
1270 case MSR_K7_EVNTSEL2:
1271 case MSR_K7_EVNTSEL3:
1273 * only support writing 0 to the performance counters for now
1274 * to make Windows happy. Should be replaced by a real
1275 * performance counter emulation later.
1277 if (data != 0)
1278 goto unhandled;
1279 break;
1280 default:
1281 unhandled:
1282 return kvm_set_msr_common(vcpu, ecx, data);
1284 return 0;
1287 static int wrmsr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1289 u32 ecx = svm->vcpu.arch.regs[VCPU_REGS_RCX];
1290 u64 data = (svm->vmcb->save.rax & -1u)
1291 | ((u64)(svm->vcpu.arch.regs[VCPU_REGS_RDX] & -1u) << 32);
1292 svm->next_rip = svm->vmcb->save.rip + 2;
1293 if (svm_set_msr(&svm->vcpu, ecx, data))
1294 kvm_inject_gp(&svm->vcpu, 0);
1295 else
1296 skip_emulated_instruction(&svm->vcpu);
1297 return 1;
1300 static int msr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1302 if (svm->vmcb->control.exit_info_1)
1303 return wrmsr_interception(svm, kvm_run);
1304 else
1305 return rdmsr_interception(svm, kvm_run);
1308 static int interrupt_window_interception(struct vcpu_svm *svm,
1309 struct kvm_run *kvm_run)
1311 svm->vmcb->control.intercept &= ~(1ULL << INTERCEPT_VINTR);
1312 svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
1314 * If the user space waits to inject interrupts, exit as soon as
1315 * possible
1317 if (kvm_run->request_interrupt_window &&
1318 !svm->vcpu.arch.irq_summary) {
1319 ++svm->vcpu.stat.irq_window_exits;
1320 kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
1321 return 0;
1324 return 1;
1327 static int (*svm_exit_handlers[])(struct vcpu_svm *svm,
1328 struct kvm_run *kvm_run) = {
1329 [SVM_EXIT_READ_CR0] = emulate_on_interception,
1330 [SVM_EXIT_READ_CR3] = emulate_on_interception,
1331 [SVM_EXIT_READ_CR4] = emulate_on_interception,
1332 [SVM_EXIT_READ_CR8] = emulate_on_interception,
1333 /* for now: */
1334 [SVM_EXIT_WRITE_CR0] = emulate_on_interception,
1335 [SVM_EXIT_WRITE_CR3] = emulate_on_interception,
1336 [SVM_EXIT_WRITE_CR4] = emulate_on_interception,
1337 [SVM_EXIT_WRITE_CR8] = cr8_write_interception,
1338 [SVM_EXIT_READ_DR0] = emulate_on_interception,
1339 [SVM_EXIT_READ_DR1] = emulate_on_interception,
1340 [SVM_EXIT_READ_DR2] = emulate_on_interception,
1341 [SVM_EXIT_READ_DR3] = emulate_on_interception,
1342 [SVM_EXIT_WRITE_DR0] = emulate_on_interception,
1343 [SVM_EXIT_WRITE_DR1] = emulate_on_interception,
1344 [SVM_EXIT_WRITE_DR2] = emulate_on_interception,
1345 [SVM_EXIT_WRITE_DR3] = emulate_on_interception,
1346 [SVM_EXIT_WRITE_DR5] = emulate_on_interception,
1347 [SVM_EXIT_WRITE_DR7] = emulate_on_interception,
1348 [SVM_EXIT_EXCP_BASE + UD_VECTOR] = ud_interception,
1349 [SVM_EXIT_EXCP_BASE + PF_VECTOR] = pf_interception,
1350 [SVM_EXIT_EXCP_BASE + NM_VECTOR] = nm_interception,
1351 [SVM_EXIT_INTR] = nop_on_interception,
1352 [SVM_EXIT_NMI] = nop_on_interception,
1353 [SVM_EXIT_SMI] = nop_on_interception,
1354 [SVM_EXIT_INIT] = nop_on_interception,
1355 [SVM_EXIT_VINTR] = interrupt_window_interception,
1356 /* [SVM_EXIT_CR0_SEL_WRITE] = emulate_on_interception, */
1357 [SVM_EXIT_CPUID] = cpuid_interception,
1358 [SVM_EXIT_INVD] = emulate_on_interception,
1359 [SVM_EXIT_HLT] = halt_interception,
1360 [SVM_EXIT_INVLPG] = emulate_on_interception,
1361 [SVM_EXIT_INVLPGA] = invalid_op_interception,
1362 [SVM_EXIT_IOIO] = io_interception,
1363 [SVM_EXIT_MSR] = msr_interception,
1364 [SVM_EXIT_TASK_SWITCH] = task_switch_interception,
1365 [SVM_EXIT_SHUTDOWN] = shutdown_interception,
1366 [SVM_EXIT_VMRUN] = invalid_op_interception,
1367 [SVM_EXIT_VMMCALL] = vmmcall_interception,
1368 [SVM_EXIT_VMLOAD] = invalid_op_interception,
1369 [SVM_EXIT_VMSAVE] = invalid_op_interception,
1370 [SVM_EXIT_STGI] = invalid_op_interception,
1371 [SVM_EXIT_CLGI] = invalid_op_interception,
1372 [SVM_EXIT_SKINIT] = invalid_op_interception,
1373 [SVM_EXIT_WBINVD] = emulate_on_interception,
1374 [SVM_EXIT_MONITOR] = invalid_op_interception,
1375 [SVM_EXIT_MWAIT] = invalid_op_interception,
1376 [SVM_EXIT_NPF] = pf_interception,
1379 static int handle_exit(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1381 struct vcpu_svm *svm = to_svm(vcpu);
1382 u32 exit_code = svm->vmcb->control.exit_code;
1384 if (npt_enabled) {
1385 int mmu_reload = 0;
1386 if ((vcpu->arch.cr0 ^ svm->vmcb->save.cr0) & X86_CR0_PG) {
1387 svm_set_cr0(vcpu, svm->vmcb->save.cr0);
1388 mmu_reload = 1;
1390 vcpu->arch.cr0 = svm->vmcb->save.cr0;
1391 vcpu->arch.cr3 = svm->vmcb->save.cr3;
1392 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
1393 if (!load_pdptrs(vcpu, vcpu->arch.cr3)) {
1394 kvm_inject_gp(vcpu, 0);
1395 return 1;
1398 if (mmu_reload) {
1399 kvm_mmu_reset_context(vcpu);
1400 kvm_mmu_load(vcpu);
1404 kvm_reput_irq(svm);
1406 if (svm->vmcb->control.exit_code == SVM_EXIT_ERR) {
1407 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
1408 kvm_run->fail_entry.hardware_entry_failure_reason
1409 = svm->vmcb->control.exit_code;
1410 return 0;
1413 if (is_external_interrupt(svm->vmcb->control.exit_int_info) &&
1414 exit_code != SVM_EXIT_EXCP_BASE + PF_VECTOR &&
1415 exit_code != SVM_EXIT_NPF)
1416 printk(KERN_ERR "%s: unexpected exit_ini_info 0x%x "
1417 "exit_code 0x%x\n",
1418 __func__, svm->vmcb->control.exit_int_info,
1419 exit_code);
1421 if (exit_code >= ARRAY_SIZE(svm_exit_handlers)
1422 || !svm_exit_handlers[exit_code]) {
1423 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
1424 kvm_run->hw.hardware_exit_reason = exit_code;
1425 return 0;
1428 return svm_exit_handlers[exit_code](svm, kvm_run);
1431 static void reload_tss(struct kvm_vcpu *vcpu)
1433 int cpu = raw_smp_processor_id();
1435 struct svm_cpu_data *svm_data = per_cpu(svm_data, cpu);
1436 svm_data->tss_desc->type = 9; /* available 32/64-bit TSS */
1437 load_TR_desc();
1440 static void pre_svm_run(struct vcpu_svm *svm)
1442 int cpu = raw_smp_processor_id();
1444 struct svm_cpu_data *svm_data = per_cpu(svm_data, cpu);
1446 svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
1447 if (svm->vcpu.cpu != cpu ||
1448 svm->asid_generation != svm_data->asid_generation)
1449 new_asid(svm, svm_data);
1453 static inline void svm_inject_irq(struct vcpu_svm *svm, int irq)
1455 struct vmcb_control_area *control;
1457 control = &svm->vmcb->control;
1458 control->int_vector = irq;
1459 control->int_ctl &= ~V_INTR_PRIO_MASK;
1460 control->int_ctl |= V_IRQ_MASK |
1461 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT);
1464 static void svm_set_irq(struct kvm_vcpu *vcpu, int irq)
1466 struct vcpu_svm *svm = to_svm(vcpu);
1468 svm_inject_irq(svm, irq);
1471 static void svm_intr_assist(struct kvm_vcpu *vcpu)
1473 struct vcpu_svm *svm = to_svm(vcpu);
1474 struct vmcb *vmcb = svm->vmcb;
1475 int intr_vector = -1;
1477 if ((vmcb->control.exit_int_info & SVM_EVTINJ_VALID) &&
1478 ((vmcb->control.exit_int_info & SVM_EVTINJ_TYPE_MASK) == 0)) {
1479 intr_vector = vmcb->control.exit_int_info &
1480 SVM_EVTINJ_VEC_MASK;
1481 vmcb->control.exit_int_info = 0;
1482 svm_inject_irq(svm, intr_vector);
1483 return;
1486 if (vmcb->control.int_ctl & V_IRQ_MASK)
1487 return;
1489 if (!kvm_cpu_has_interrupt(vcpu))
1490 return;
1492 if (!(vmcb->save.rflags & X86_EFLAGS_IF) ||
1493 (vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK) ||
1494 (vmcb->control.event_inj & SVM_EVTINJ_VALID)) {
1495 /* unable to deliver irq, set pending irq */
1496 vmcb->control.intercept |= (1ULL << INTERCEPT_VINTR);
1497 svm_inject_irq(svm, 0x0);
1498 return;
1500 /* Okay, we can deliver the interrupt: grab it and update PIC state. */
1501 intr_vector = kvm_cpu_get_interrupt(vcpu);
1502 svm_inject_irq(svm, intr_vector);
1503 kvm_timer_intr_post(vcpu, intr_vector);
1506 static void kvm_reput_irq(struct vcpu_svm *svm)
1508 struct vmcb_control_area *control = &svm->vmcb->control;
1510 if ((control->int_ctl & V_IRQ_MASK)
1511 && !irqchip_in_kernel(svm->vcpu.kvm)) {
1512 control->int_ctl &= ~V_IRQ_MASK;
1513 push_irq(&svm->vcpu, control->int_vector);
1516 svm->vcpu.arch.interrupt_window_open =
1517 !(control->int_state & SVM_INTERRUPT_SHADOW_MASK);
1520 static void svm_do_inject_vector(struct vcpu_svm *svm)
1522 struct kvm_vcpu *vcpu = &svm->vcpu;
1523 int word_index = __ffs(vcpu->arch.irq_summary);
1524 int bit_index = __ffs(vcpu->arch.irq_pending[word_index]);
1525 int irq = word_index * BITS_PER_LONG + bit_index;
1527 clear_bit(bit_index, &vcpu->arch.irq_pending[word_index]);
1528 if (!vcpu->arch.irq_pending[word_index])
1529 clear_bit(word_index, &vcpu->arch.irq_summary);
1530 svm_inject_irq(svm, irq);
1533 static void do_interrupt_requests(struct kvm_vcpu *vcpu,
1534 struct kvm_run *kvm_run)
1536 struct vcpu_svm *svm = to_svm(vcpu);
1537 struct vmcb_control_area *control = &svm->vmcb->control;
1539 svm->vcpu.arch.interrupt_window_open =
1540 (!(control->int_state & SVM_INTERRUPT_SHADOW_MASK) &&
1541 (svm->vmcb->save.rflags & X86_EFLAGS_IF));
1543 if (svm->vcpu.arch.interrupt_window_open && svm->vcpu.arch.irq_summary)
1545 * If interrupts enabled, and not blocked by sti or mov ss. Good.
1547 svm_do_inject_vector(svm);
1550 * Interrupts blocked. Wait for unblock.
1552 if (!svm->vcpu.arch.interrupt_window_open &&
1553 (svm->vcpu.arch.irq_summary || kvm_run->request_interrupt_window))
1554 control->intercept |= 1ULL << INTERCEPT_VINTR;
1555 else
1556 control->intercept &= ~(1ULL << INTERCEPT_VINTR);
1559 static int svm_set_tss_addr(struct kvm *kvm, unsigned int addr)
1561 return 0;
1564 static void save_db_regs(unsigned long *db_regs)
1566 asm volatile ("mov %%dr0, %0" : "=r"(db_regs[0]));
1567 asm volatile ("mov %%dr1, %0" : "=r"(db_regs[1]));
1568 asm volatile ("mov %%dr2, %0" : "=r"(db_regs[2]));
1569 asm volatile ("mov %%dr3, %0" : "=r"(db_regs[3]));
1572 static void load_db_regs(unsigned long *db_regs)
1574 asm volatile ("mov %0, %%dr0" : : "r"(db_regs[0]));
1575 asm volatile ("mov %0, %%dr1" : : "r"(db_regs[1]));
1576 asm volatile ("mov %0, %%dr2" : : "r"(db_regs[2]));
1577 asm volatile ("mov %0, %%dr3" : : "r"(db_regs[3]));
1580 static void svm_flush_tlb(struct kvm_vcpu *vcpu)
1582 force_new_asid(vcpu);
1585 static void svm_prepare_guest_switch(struct kvm_vcpu *vcpu)
1589 static void svm_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1591 struct vcpu_svm *svm = to_svm(vcpu);
1592 u16 fs_selector;
1593 u16 gs_selector;
1594 u16 ldt_selector;
1596 pre_svm_run(svm);
1598 save_host_msrs(vcpu);
1599 fs_selector = read_fs();
1600 gs_selector = read_gs();
1601 ldt_selector = read_ldt();
1602 svm->host_cr2 = kvm_read_cr2();
1603 svm->host_dr6 = read_dr6();
1604 svm->host_dr7 = read_dr7();
1605 svm->vmcb->save.cr2 = vcpu->arch.cr2;
1606 /* required for live migration with NPT */
1607 if (npt_enabled)
1608 svm->vmcb->save.cr3 = vcpu->arch.cr3;
1610 if (svm->vmcb->save.dr7 & 0xff) {
1611 write_dr7(0);
1612 save_db_regs(svm->host_db_regs);
1613 load_db_regs(svm->db_regs);
1616 clgi();
1618 local_irq_enable();
1620 asm volatile (
1621 #ifdef CONFIG_X86_64
1622 "push %%rbp; \n\t"
1623 #else
1624 "push %%ebp; \n\t"
1625 #endif
1627 #ifdef CONFIG_X86_64
1628 "mov %c[rbx](%[svm]), %%rbx \n\t"
1629 "mov %c[rcx](%[svm]), %%rcx \n\t"
1630 "mov %c[rdx](%[svm]), %%rdx \n\t"
1631 "mov %c[rsi](%[svm]), %%rsi \n\t"
1632 "mov %c[rdi](%[svm]), %%rdi \n\t"
1633 "mov %c[rbp](%[svm]), %%rbp \n\t"
1634 "mov %c[r8](%[svm]), %%r8 \n\t"
1635 "mov %c[r9](%[svm]), %%r9 \n\t"
1636 "mov %c[r10](%[svm]), %%r10 \n\t"
1637 "mov %c[r11](%[svm]), %%r11 \n\t"
1638 "mov %c[r12](%[svm]), %%r12 \n\t"
1639 "mov %c[r13](%[svm]), %%r13 \n\t"
1640 "mov %c[r14](%[svm]), %%r14 \n\t"
1641 "mov %c[r15](%[svm]), %%r15 \n\t"
1642 #else
1643 "mov %c[rbx](%[svm]), %%ebx \n\t"
1644 "mov %c[rcx](%[svm]), %%ecx \n\t"
1645 "mov %c[rdx](%[svm]), %%edx \n\t"
1646 "mov %c[rsi](%[svm]), %%esi \n\t"
1647 "mov %c[rdi](%[svm]), %%edi \n\t"
1648 "mov %c[rbp](%[svm]), %%ebp \n\t"
1649 #endif
1651 #ifdef CONFIG_X86_64
1652 /* Enter guest mode */
1653 "push %%rax \n\t"
1654 "mov %c[vmcb](%[svm]), %%rax \n\t"
1655 SVM_VMLOAD "\n\t"
1656 SVM_VMRUN "\n\t"
1657 SVM_VMSAVE "\n\t"
1658 "pop %%rax \n\t"
1659 #else
1660 /* Enter guest mode */
1661 "push %%eax \n\t"
1662 "mov %c[vmcb](%[svm]), %%eax \n\t"
1663 SVM_VMLOAD "\n\t"
1664 SVM_VMRUN "\n\t"
1665 SVM_VMSAVE "\n\t"
1666 "pop %%eax \n\t"
1667 #endif
1669 /* Save guest registers, load host registers */
1670 #ifdef CONFIG_X86_64
1671 "mov %%rbx, %c[rbx](%[svm]) \n\t"
1672 "mov %%rcx, %c[rcx](%[svm]) \n\t"
1673 "mov %%rdx, %c[rdx](%[svm]) \n\t"
1674 "mov %%rsi, %c[rsi](%[svm]) \n\t"
1675 "mov %%rdi, %c[rdi](%[svm]) \n\t"
1676 "mov %%rbp, %c[rbp](%[svm]) \n\t"
1677 "mov %%r8, %c[r8](%[svm]) \n\t"
1678 "mov %%r9, %c[r9](%[svm]) \n\t"
1679 "mov %%r10, %c[r10](%[svm]) \n\t"
1680 "mov %%r11, %c[r11](%[svm]) \n\t"
1681 "mov %%r12, %c[r12](%[svm]) \n\t"
1682 "mov %%r13, %c[r13](%[svm]) \n\t"
1683 "mov %%r14, %c[r14](%[svm]) \n\t"
1684 "mov %%r15, %c[r15](%[svm]) \n\t"
1686 "pop %%rbp; \n\t"
1687 #else
1688 "mov %%ebx, %c[rbx](%[svm]) \n\t"
1689 "mov %%ecx, %c[rcx](%[svm]) \n\t"
1690 "mov %%edx, %c[rdx](%[svm]) \n\t"
1691 "mov %%esi, %c[rsi](%[svm]) \n\t"
1692 "mov %%edi, %c[rdi](%[svm]) \n\t"
1693 "mov %%ebp, %c[rbp](%[svm]) \n\t"
1695 "pop %%ebp; \n\t"
1696 #endif
1698 : [svm]"a"(svm),
1699 [vmcb]"i"(offsetof(struct vcpu_svm, vmcb_pa)),
1700 [rbx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBX])),
1701 [rcx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RCX])),
1702 [rdx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDX])),
1703 [rsi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RSI])),
1704 [rdi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDI])),
1705 [rbp]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBP]))
1706 #ifdef CONFIG_X86_64
1707 , [r8]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R8])),
1708 [r9]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R9])),
1709 [r10]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R10])),
1710 [r11]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R11])),
1711 [r12]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R12])),
1712 [r13]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R13])),
1713 [r14]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R14])),
1714 [r15]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R15]))
1715 #endif
1716 : "cc", "memory"
1717 #ifdef CONFIG_X86_64
1718 , "rbx", "rcx", "rdx", "rsi", "rdi"
1719 , "r8", "r9", "r10", "r11" , "r12", "r13", "r14", "r15"
1720 #else
1721 , "ebx", "ecx", "edx" , "esi", "edi"
1722 #endif
1725 if ((svm->vmcb->save.dr7 & 0xff))
1726 load_db_regs(svm->host_db_regs);
1728 vcpu->arch.cr2 = svm->vmcb->save.cr2;
1730 write_dr6(svm->host_dr6);
1731 write_dr7(svm->host_dr7);
1732 kvm_write_cr2(svm->host_cr2);
1734 load_fs(fs_selector);
1735 load_gs(gs_selector);
1736 load_ldt(ldt_selector);
1737 load_host_msrs(vcpu);
1739 reload_tss(vcpu);
1741 local_irq_disable();
1743 stgi();
1745 svm->next_rip = 0;
1748 static void svm_set_cr3(struct kvm_vcpu *vcpu, unsigned long root)
1750 struct vcpu_svm *svm = to_svm(vcpu);
1752 if (npt_enabled) {
1753 svm->vmcb->control.nested_cr3 = root;
1754 force_new_asid(vcpu);
1755 return;
1758 svm->vmcb->save.cr3 = root;
1759 force_new_asid(vcpu);
1761 if (vcpu->fpu_active) {
1762 svm->vmcb->control.intercept_exceptions |= (1 << NM_VECTOR);
1763 svm->vmcb->save.cr0 |= X86_CR0_TS;
1764 vcpu->fpu_active = 0;
1768 static int is_disabled(void)
1770 u64 vm_cr;
1772 rdmsrl(MSR_VM_CR, vm_cr);
1773 if (vm_cr & (1 << SVM_VM_CR_SVM_DISABLE))
1774 return 1;
1776 return 0;
1779 static void
1780 svm_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
1783 * Patch in the VMMCALL instruction:
1785 hypercall[0] = 0x0f;
1786 hypercall[1] = 0x01;
1787 hypercall[2] = 0xd9;
1790 static void svm_check_processor_compat(void *rtn)
1792 *(int *)rtn = 0;
1795 static bool svm_cpu_has_accelerated_tpr(void)
1797 return false;
1800 static struct kvm_x86_ops svm_x86_ops = {
1801 .cpu_has_kvm_support = has_svm,
1802 .disabled_by_bios = is_disabled,
1803 .hardware_setup = svm_hardware_setup,
1804 .hardware_unsetup = svm_hardware_unsetup,
1805 .check_processor_compatibility = svm_check_processor_compat,
1806 .hardware_enable = svm_hardware_enable,
1807 .hardware_disable = svm_hardware_disable,
1808 .cpu_has_accelerated_tpr = svm_cpu_has_accelerated_tpr,
1810 .vcpu_create = svm_create_vcpu,
1811 .vcpu_free = svm_free_vcpu,
1812 .vcpu_reset = svm_vcpu_reset,
1814 .prepare_guest_switch = svm_prepare_guest_switch,
1815 .vcpu_load = svm_vcpu_load,
1816 .vcpu_put = svm_vcpu_put,
1817 .vcpu_decache = svm_vcpu_decache,
1819 .set_guest_debug = svm_guest_debug,
1820 .get_msr = svm_get_msr,
1821 .set_msr = svm_set_msr,
1822 .get_segment_base = svm_get_segment_base,
1823 .get_segment = svm_get_segment,
1824 .set_segment = svm_set_segment,
1825 .get_cs_db_l_bits = kvm_get_cs_db_l_bits,
1826 .decache_cr4_guest_bits = svm_decache_cr4_guest_bits,
1827 .set_cr0 = svm_set_cr0,
1828 .set_cr3 = svm_set_cr3,
1829 .set_cr4 = svm_set_cr4,
1830 .set_efer = svm_set_efer,
1831 .get_idt = svm_get_idt,
1832 .set_idt = svm_set_idt,
1833 .get_gdt = svm_get_gdt,
1834 .set_gdt = svm_set_gdt,
1835 .get_dr = svm_get_dr,
1836 .set_dr = svm_set_dr,
1837 .cache_regs = svm_cache_regs,
1838 .decache_regs = svm_decache_regs,
1839 .get_rflags = svm_get_rflags,
1840 .set_rflags = svm_set_rflags,
1842 .tlb_flush = svm_flush_tlb,
1844 .run = svm_vcpu_run,
1845 .handle_exit = handle_exit,
1846 .skip_emulated_instruction = skip_emulated_instruction,
1847 .patch_hypercall = svm_patch_hypercall,
1848 .get_irq = svm_get_irq,
1849 .set_irq = svm_set_irq,
1850 .queue_exception = svm_queue_exception,
1851 .exception_injected = svm_exception_injected,
1852 .inject_pending_irq = svm_intr_assist,
1853 .inject_pending_vectors = do_interrupt_requests,
1855 .set_tss_addr = svm_set_tss_addr,
1858 static int __init svm_init(void)
1860 return kvm_init(&svm_x86_ops, sizeof(struct vcpu_svm),
1861 THIS_MODULE);
1864 static void __exit svm_exit(void)
1866 kvm_exit();
1869 module_init(svm_init)
1870 module_exit(svm_exit)