2 * Kernel-based Virtual Machine driver for Linux
6 * Copyright (C) 2006 Qumranet, Inc.
7 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
10 * Yaniv Kamay <yaniv@qumranet.com>
11 * Avi Kivity <avi@qumranet.com>
13 * This work is licensed under the terms of the GNU GPL, version 2. See
14 * the COPYING file in the top-level directory.
18 #define pr_fmt(fmt) "SVM: " fmt
20 #include <linux/kvm_host.h>
24 #include "kvm_cache_regs.h"
29 #include <linux/module.h>
30 #include <linux/mod_devicetable.h>
31 #include <linux/kernel.h>
32 #include <linux/vmalloc.h>
33 #include <linux/highmem.h>
34 #include <linux/sched.h>
35 #include <linux/trace_events.h>
36 #include <linux/slab.h>
38 #include <asm/perf_event.h>
39 #include <asm/tlbflush.h>
41 #include <asm/debugreg.h>
42 #include <asm/kvm_para.h>
44 #include <asm/virtext.h>
47 #define __ex(x) __kvm_handle_fault_on_reboot(x)
49 MODULE_AUTHOR("Qumranet");
50 MODULE_LICENSE("GPL");
52 static const struct x86_cpu_id svm_cpu_id[] = {
53 X86_FEATURE_MATCH(X86_FEATURE_SVM),
56 MODULE_DEVICE_TABLE(x86cpu, svm_cpu_id);
58 #define IOPM_ALLOC_ORDER 2
59 #define MSRPM_ALLOC_ORDER 1
61 #define SEG_TYPE_LDT 2
62 #define SEG_TYPE_BUSY_TSS16 3
64 #define SVM_FEATURE_NPT (1 << 0)
65 #define SVM_FEATURE_LBRV (1 << 1)
66 #define SVM_FEATURE_SVML (1 << 2)
67 #define SVM_FEATURE_NRIP (1 << 3)
68 #define SVM_FEATURE_TSC_RATE (1 << 4)
69 #define SVM_FEATURE_VMCB_CLEAN (1 << 5)
70 #define SVM_FEATURE_FLUSH_ASID (1 << 6)
71 #define SVM_FEATURE_DECODE_ASSIST (1 << 7)
72 #define SVM_FEATURE_PAUSE_FILTER (1 << 10)
74 #define SVM_AVIC_DOORBELL 0xc001011b
76 #define NESTED_EXIT_HOST 0 /* Exit handled on host level */
77 #define NESTED_EXIT_DONE 1 /* Exit caused nested vmexit */
78 #define NESTED_EXIT_CONTINUE 2 /* Further checks needed */
80 #define DEBUGCTL_RESERVED_BITS (~(0x3fULL))
82 #define TSC_RATIO_RSVD 0xffffff0000000000ULL
83 #define TSC_RATIO_MIN 0x0000000000000001ULL
84 #define TSC_RATIO_MAX 0x000000ffffffffffULL
86 #define AVIC_HPA_MASK ~((0xFFFULL << 52) || 0xFFF)
89 * 0xff is broadcast, so the max index allowed for physical APIC ID
90 * table is 0xfe. APIC IDs above 0xff are reserved.
92 #define AVIC_MAX_PHYSICAL_ID_COUNT 255
94 static bool erratum_383_found __read_mostly;
96 static const u32 host_save_user_msrs[] = {
98 MSR_STAR, MSR_LSTAR, MSR_CSTAR, MSR_SYSCALL_MASK, MSR_KERNEL_GS_BASE,
101 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
105 #define NR_HOST_SAVE_USER_MSRS ARRAY_SIZE(host_save_user_msrs)
109 struct nested_state {
115 /* These are the merged vectors */
118 /* gpa pointers to the real vectors */
122 /* A VMEXIT is required but not yet emulated */
125 /* cache for intercepts of the guest */
128 u32 intercept_exceptions;
131 /* Nested Paging related state */
135 #define MSRPM_OFFSETS 16
136 static u32 msrpm_offsets[MSRPM_OFFSETS] __read_mostly;
139 * Set osvw_len to higher value when updated Revision Guides
140 * are published and we know what the new status bits are
142 static uint64_t osvw_len = 4, osvw_status;
145 struct kvm_vcpu vcpu;
147 unsigned long vmcb_pa;
148 struct svm_cpu_data *svm_data;
149 uint64_t asid_generation;
150 uint64_t sysenter_esp;
151 uint64_t sysenter_eip;
156 u64 host_user_msrs[NR_HOST_SAVE_USER_MSRS];
168 struct nested_state nested;
172 unsigned int3_injected;
173 unsigned long int3_rip;
176 /* cached guest cpuid flags for faster access */
177 bool nrips_enabled : 1;
179 struct page *avic_backing_page;
180 u64 *avic_physical_id_cache;
183 #define AVIC_LOGICAL_ID_ENTRY_GUEST_PHYSICAL_ID_MASK (0xFF)
184 #define AVIC_LOGICAL_ID_ENTRY_VALID_MASK (1 << 31)
186 #define AVIC_PHYSICAL_ID_ENTRY_HOST_PHYSICAL_ID_MASK (0xFFULL)
187 #define AVIC_PHYSICAL_ID_ENTRY_BACKING_PAGE_MASK (0xFFFFFFFFFFULL << 12)
188 #define AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK (1ULL << 62)
189 #define AVIC_PHYSICAL_ID_ENTRY_VALID_MASK (1ULL << 63)
191 static DEFINE_PER_CPU(u64, current_tsc_ratio);
192 #define TSC_RATIO_DEFAULT 0x0100000000ULL
194 #define MSR_INVALID 0xffffffffU
196 static const struct svm_direct_access_msrs {
197 u32 index; /* Index of the MSR */
198 bool always; /* True if intercept is always on */
199 } direct_access_msrs[] = {
200 { .index = MSR_STAR, .always = true },
201 { .index = MSR_IA32_SYSENTER_CS, .always = true },
203 { .index = MSR_GS_BASE, .always = true },
204 { .index = MSR_FS_BASE, .always = true },
205 { .index = MSR_KERNEL_GS_BASE, .always = true },
206 { .index = MSR_LSTAR, .always = true },
207 { .index = MSR_CSTAR, .always = true },
208 { .index = MSR_SYSCALL_MASK, .always = true },
210 { .index = MSR_IA32_LASTBRANCHFROMIP, .always = false },
211 { .index = MSR_IA32_LASTBRANCHTOIP, .always = false },
212 { .index = MSR_IA32_LASTINTFROMIP, .always = false },
213 { .index = MSR_IA32_LASTINTTOIP, .always = false },
214 { .index = MSR_INVALID, .always = false },
217 /* enable NPT for AMD64 and X86 with PAE */
218 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
219 static bool npt_enabled = true;
221 static bool npt_enabled;
224 /* allow nested paging (virtualized MMU) for all guests */
225 static int npt = true;
226 module_param(npt, int, S_IRUGO);
228 /* allow nested virtualization in KVM/SVM */
229 static int nested = true;
230 module_param(nested, int, S_IRUGO);
232 /* enable / disable AVIC */
234 module_param(avic, int, S_IRUGO);
236 static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0);
237 static void svm_flush_tlb(struct kvm_vcpu *vcpu);
238 static void svm_complete_interrupts(struct vcpu_svm *svm);
240 static int nested_svm_exit_handled(struct vcpu_svm *svm);
241 static int nested_svm_intercept(struct vcpu_svm *svm);
242 static int nested_svm_vmexit(struct vcpu_svm *svm);
243 static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
244 bool has_error_code, u32 error_code);
247 VMCB_INTERCEPTS, /* Intercept vectors, TSC offset,
248 pause filter count */
249 VMCB_PERM_MAP, /* IOPM Base and MSRPM Base */
250 VMCB_ASID, /* ASID */
251 VMCB_INTR, /* int_ctl, int_vector */
252 VMCB_NPT, /* npt_en, nCR3, gPAT */
253 VMCB_CR, /* CR0, CR3, CR4, EFER */
254 VMCB_DR, /* DR6, DR7 */
255 VMCB_DT, /* GDT, IDT */
256 VMCB_SEG, /* CS, DS, SS, ES, CPL */
257 VMCB_CR2, /* CR2 only */
258 VMCB_LBR, /* DBGCTL, BR_FROM, BR_TO, LAST_EX_FROM, LAST_EX_TO */
259 VMCB_AVIC, /* AVIC APIC_BAR, AVIC APIC_BACKING_PAGE,
260 * AVIC PHYSICAL_TABLE pointer,
261 * AVIC LOGICAL_TABLE pointer
266 /* TPR and CR2 are always written before VMRUN */
267 #define VMCB_ALWAYS_DIRTY_MASK ((1U << VMCB_INTR) | (1U << VMCB_CR2))
269 #define VMCB_AVIC_APIC_BAR_MASK 0xFFFFFFFFFF000ULL
271 static inline void mark_all_dirty(struct vmcb *vmcb)
273 vmcb->control.clean = 0;
276 static inline void mark_all_clean(struct vmcb *vmcb)
278 vmcb->control.clean = ((1 << VMCB_DIRTY_MAX) - 1)
279 & ~VMCB_ALWAYS_DIRTY_MASK;
282 static inline void mark_dirty(struct vmcb *vmcb, int bit)
284 vmcb->control.clean &= ~(1 << bit);
287 static inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu)
289 return container_of(vcpu, struct vcpu_svm, vcpu);
292 static inline void avic_update_vapic_bar(struct vcpu_svm *svm, u64 data)
294 svm->vmcb->control.avic_vapic_bar = data & VMCB_AVIC_APIC_BAR_MASK;
295 mark_dirty(svm->vmcb, VMCB_AVIC);
298 static inline bool avic_vcpu_is_running(struct kvm_vcpu *vcpu)
300 struct vcpu_svm *svm = to_svm(vcpu);
301 u64 *entry = svm->avic_physical_id_cache;
306 return (READ_ONCE(*entry) & AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK);
309 static void recalc_intercepts(struct vcpu_svm *svm)
311 struct vmcb_control_area *c, *h;
312 struct nested_state *g;
314 mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
316 if (!is_guest_mode(&svm->vcpu))
319 c = &svm->vmcb->control;
320 h = &svm->nested.hsave->control;
323 c->intercept_cr = h->intercept_cr | g->intercept_cr;
324 c->intercept_dr = h->intercept_dr | g->intercept_dr;
325 c->intercept_exceptions = h->intercept_exceptions | g->intercept_exceptions;
326 c->intercept = h->intercept | g->intercept;
329 static inline struct vmcb *get_host_vmcb(struct vcpu_svm *svm)
331 if (is_guest_mode(&svm->vcpu))
332 return svm->nested.hsave;
337 static inline void set_cr_intercept(struct vcpu_svm *svm, int bit)
339 struct vmcb *vmcb = get_host_vmcb(svm);
341 vmcb->control.intercept_cr |= (1U << bit);
343 recalc_intercepts(svm);
346 static inline void clr_cr_intercept(struct vcpu_svm *svm, int bit)
348 struct vmcb *vmcb = get_host_vmcb(svm);
350 vmcb->control.intercept_cr &= ~(1U << bit);
352 recalc_intercepts(svm);
355 static inline bool is_cr_intercept(struct vcpu_svm *svm, int bit)
357 struct vmcb *vmcb = get_host_vmcb(svm);
359 return vmcb->control.intercept_cr & (1U << bit);
362 static inline void set_dr_intercepts(struct vcpu_svm *svm)
364 struct vmcb *vmcb = get_host_vmcb(svm);
366 vmcb->control.intercept_dr = (1 << INTERCEPT_DR0_READ)
367 | (1 << INTERCEPT_DR1_READ)
368 | (1 << INTERCEPT_DR2_READ)
369 | (1 << INTERCEPT_DR3_READ)
370 | (1 << INTERCEPT_DR4_READ)
371 | (1 << INTERCEPT_DR5_READ)
372 | (1 << INTERCEPT_DR6_READ)
373 | (1 << INTERCEPT_DR7_READ)
374 | (1 << INTERCEPT_DR0_WRITE)
375 | (1 << INTERCEPT_DR1_WRITE)
376 | (1 << INTERCEPT_DR2_WRITE)
377 | (1 << INTERCEPT_DR3_WRITE)
378 | (1 << INTERCEPT_DR4_WRITE)
379 | (1 << INTERCEPT_DR5_WRITE)
380 | (1 << INTERCEPT_DR6_WRITE)
381 | (1 << INTERCEPT_DR7_WRITE);
383 recalc_intercepts(svm);
386 static inline void clr_dr_intercepts(struct vcpu_svm *svm)
388 struct vmcb *vmcb = get_host_vmcb(svm);
390 vmcb->control.intercept_dr = 0;
392 recalc_intercepts(svm);
395 static inline void set_exception_intercept(struct vcpu_svm *svm, int bit)
397 struct vmcb *vmcb = get_host_vmcb(svm);
399 vmcb->control.intercept_exceptions |= (1U << bit);
401 recalc_intercepts(svm);
404 static inline void clr_exception_intercept(struct vcpu_svm *svm, int bit)
406 struct vmcb *vmcb = get_host_vmcb(svm);
408 vmcb->control.intercept_exceptions &= ~(1U << bit);
410 recalc_intercepts(svm);
413 static inline void set_intercept(struct vcpu_svm *svm, int bit)
415 struct vmcb *vmcb = get_host_vmcb(svm);
417 vmcb->control.intercept |= (1ULL << bit);
419 recalc_intercepts(svm);
422 static inline void clr_intercept(struct vcpu_svm *svm, int bit)
424 struct vmcb *vmcb = get_host_vmcb(svm);
426 vmcb->control.intercept &= ~(1ULL << bit);
428 recalc_intercepts(svm);
431 static inline void enable_gif(struct vcpu_svm *svm)
433 svm->vcpu.arch.hflags |= HF_GIF_MASK;
436 static inline void disable_gif(struct vcpu_svm *svm)
438 svm->vcpu.arch.hflags &= ~HF_GIF_MASK;
441 static inline bool gif_set(struct vcpu_svm *svm)
443 return !!(svm->vcpu.arch.hflags & HF_GIF_MASK);
446 static unsigned long iopm_base;
448 struct kvm_ldttss_desc {
451 unsigned base1:8, type:5, dpl:2, p:1;
452 unsigned limit1:4, zero0:3, g:1, base2:8;
455 } __attribute__((packed));
457 struct svm_cpu_data {
463 struct kvm_ldttss_desc *tss_desc;
465 struct page *save_area;
468 static DEFINE_PER_CPU(struct svm_cpu_data *, svm_data);
470 struct svm_init_data {
475 static const u32 msrpm_ranges[] = {0, 0xc0000000, 0xc0010000};
477 #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
478 #define MSRS_RANGE_SIZE 2048
479 #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
481 static u32 svm_msrpm_offset(u32 msr)
486 for (i = 0; i < NUM_MSR_MAPS; i++) {
487 if (msr < msrpm_ranges[i] ||
488 msr >= msrpm_ranges[i] + MSRS_IN_RANGE)
491 offset = (msr - msrpm_ranges[i]) / 4; /* 4 msrs per u8 */
492 offset += (i * MSRS_RANGE_SIZE); /* add range offset */
494 /* Now we have the u8 offset - but need the u32 offset */
498 /* MSR not in any range */
502 #define MAX_INST_SIZE 15
504 static inline void clgi(void)
506 asm volatile (__ex(SVM_CLGI));
509 static inline void stgi(void)
511 asm volatile (__ex(SVM_STGI));
514 static inline void invlpga(unsigned long addr, u32 asid)
516 asm volatile (__ex(SVM_INVLPGA) : : "a"(addr), "c"(asid));
519 static int get_npt_level(void)
522 return PT64_ROOT_LEVEL;
524 return PT32E_ROOT_LEVEL;
528 static void svm_set_efer(struct kvm_vcpu *vcpu, u64 efer)
530 vcpu->arch.efer = efer;
531 if (!npt_enabled && !(efer & EFER_LMA))
534 to_svm(vcpu)->vmcb->save.efer = efer | EFER_SVME;
535 mark_dirty(to_svm(vcpu)->vmcb, VMCB_CR);
538 static int is_external_interrupt(u32 info)
540 info &= SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
541 return info == (SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR);
544 static u32 svm_get_interrupt_shadow(struct kvm_vcpu *vcpu)
546 struct vcpu_svm *svm = to_svm(vcpu);
549 if (svm->vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK)
550 ret = KVM_X86_SHADOW_INT_STI | KVM_X86_SHADOW_INT_MOV_SS;
554 static void svm_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
556 struct vcpu_svm *svm = to_svm(vcpu);
559 svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK;
561 svm->vmcb->control.int_state |= SVM_INTERRUPT_SHADOW_MASK;
565 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
567 struct vcpu_svm *svm = to_svm(vcpu);
569 if (svm->vmcb->control.next_rip != 0) {
570 WARN_ON_ONCE(!static_cpu_has(X86_FEATURE_NRIPS));
571 svm->next_rip = svm->vmcb->control.next_rip;
574 if (!svm->next_rip) {
575 if (emulate_instruction(vcpu, EMULTYPE_SKIP) !=
577 printk(KERN_DEBUG "%s: NOP\n", __func__);
580 if (svm->next_rip - kvm_rip_read(vcpu) > MAX_INST_SIZE)
581 printk(KERN_ERR "%s: ip 0x%lx next 0x%llx\n",
582 __func__, kvm_rip_read(vcpu), svm->next_rip);
584 kvm_rip_write(vcpu, svm->next_rip);
585 svm_set_interrupt_shadow(vcpu, 0);
588 static void svm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
589 bool has_error_code, u32 error_code,
592 struct vcpu_svm *svm = to_svm(vcpu);
595 * If we are within a nested VM we'd better #VMEXIT and let the guest
596 * handle the exception
599 nested_svm_check_exception(svm, nr, has_error_code, error_code))
602 if (nr == BP_VECTOR && !static_cpu_has(X86_FEATURE_NRIPS)) {
603 unsigned long rip, old_rip = kvm_rip_read(&svm->vcpu);
606 * For guest debugging where we have to reinject #BP if some
607 * INT3 is guest-owned:
608 * Emulate nRIP by moving RIP forward. Will fail if injection
609 * raises a fault that is not intercepted. Still better than
610 * failing in all cases.
612 skip_emulated_instruction(&svm->vcpu);
613 rip = kvm_rip_read(&svm->vcpu);
614 svm->int3_rip = rip + svm->vmcb->save.cs.base;
615 svm->int3_injected = rip - old_rip;
618 svm->vmcb->control.event_inj = nr
620 | (has_error_code ? SVM_EVTINJ_VALID_ERR : 0)
621 | SVM_EVTINJ_TYPE_EXEPT;
622 svm->vmcb->control.event_inj_err = error_code;
625 static void svm_init_erratum_383(void)
631 if (!static_cpu_has_bug(X86_BUG_AMD_TLB_MMATCH))
634 /* Use _safe variants to not break nested virtualization */
635 val = native_read_msr_safe(MSR_AMD64_DC_CFG, &err);
641 low = lower_32_bits(val);
642 high = upper_32_bits(val);
644 native_write_msr_safe(MSR_AMD64_DC_CFG, low, high);
646 erratum_383_found = true;
649 static void svm_init_osvw(struct kvm_vcpu *vcpu)
652 * Guests should see errata 400 and 415 as fixed (assuming that
653 * HLT and IO instructions are intercepted).
655 vcpu->arch.osvw.length = (osvw_len >= 3) ? (osvw_len) : 3;
656 vcpu->arch.osvw.status = osvw_status & ~(6ULL);
659 * By increasing VCPU's osvw.length to 3 we are telling the guest that
660 * all osvw.status bits inside that length, including bit 0 (which is
661 * reserved for erratum 298), are valid. However, if host processor's
662 * osvw_len is 0 then osvw_status[0] carries no information. We need to
663 * be conservative here and therefore we tell the guest that erratum 298
664 * is present (because we really don't know).
666 if (osvw_len == 0 && boot_cpu_data.x86 == 0x10)
667 vcpu->arch.osvw.status |= 1;
670 static int has_svm(void)
674 if (!cpu_has_svm(&msg)) {
675 printk(KERN_INFO "has_svm: %s\n", msg);
682 static void svm_hardware_disable(void)
684 /* Make sure we clean up behind us */
685 if (static_cpu_has(X86_FEATURE_TSCRATEMSR))
686 wrmsrl(MSR_AMD64_TSC_RATIO, TSC_RATIO_DEFAULT);
690 amd_pmu_disable_virt();
693 static int svm_hardware_enable(void)
696 struct svm_cpu_data *sd;
698 struct desc_ptr gdt_descr;
699 struct desc_struct *gdt;
700 int me = raw_smp_processor_id();
702 rdmsrl(MSR_EFER, efer);
703 if (efer & EFER_SVME)
707 pr_err("%s: err EOPNOTSUPP on %d\n", __func__, me);
710 sd = per_cpu(svm_data, me);
712 pr_err("%s: svm_data is NULL on %d\n", __func__, me);
716 sd->asid_generation = 1;
717 sd->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1;
718 sd->next_asid = sd->max_asid + 1;
720 native_store_gdt(&gdt_descr);
721 gdt = (struct desc_struct *)gdt_descr.address;
722 sd->tss_desc = (struct kvm_ldttss_desc *)(gdt + GDT_ENTRY_TSS);
724 wrmsrl(MSR_EFER, efer | EFER_SVME);
726 wrmsrl(MSR_VM_HSAVE_PA, page_to_pfn(sd->save_area) << PAGE_SHIFT);
728 if (static_cpu_has(X86_FEATURE_TSCRATEMSR)) {
729 wrmsrl(MSR_AMD64_TSC_RATIO, TSC_RATIO_DEFAULT);
730 __this_cpu_write(current_tsc_ratio, TSC_RATIO_DEFAULT);
737 * Note that it is possible to have a system with mixed processor
738 * revisions and therefore different OSVW bits. If bits are not the same
739 * on different processors then choose the worst case (i.e. if erratum
740 * is present on one processor and not on another then assume that the
741 * erratum is present everywhere).
743 if (cpu_has(&boot_cpu_data, X86_FEATURE_OSVW)) {
744 uint64_t len, status = 0;
747 len = native_read_msr_safe(MSR_AMD64_OSVW_ID_LENGTH, &err);
749 status = native_read_msr_safe(MSR_AMD64_OSVW_STATUS,
753 osvw_status = osvw_len = 0;
757 osvw_status |= status;
758 osvw_status &= (1ULL << osvw_len) - 1;
761 osvw_status = osvw_len = 0;
763 svm_init_erratum_383();
765 amd_pmu_enable_virt();
770 static void svm_cpu_uninit(int cpu)
772 struct svm_cpu_data *sd = per_cpu(svm_data, raw_smp_processor_id());
777 per_cpu(svm_data, raw_smp_processor_id()) = NULL;
778 __free_page(sd->save_area);
782 static int svm_cpu_init(int cpu)
784 struct svm_cpu_data *sd;
787 sd = kzalloc(sizeof(struct svm_cpu_data), GFP_KERNEL);
791 sd->save_area = alloc_page(GFP_KERNEL);
796 per_cpu(svm_data, cpu) = sd;
806 static bool valid_msr_intercept(u32 index)
810 for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++)
811 if (direct_access_msrs[i].index == index)
817 static void set_msr_interception(u32 *msrpm, unsigned msr,
820 u8 bit_read, bit_write;
825 * If this warning triggers extend the direct_access_msrs list at the
826 * beginning of the file
828 WARN_ON(!valid_msr_intercept(msr));
830 offset = svm_msrpm_offset(msr);
831 bit_read = 2 * (msr & 0x0f);
832 bit_write = 2 * (msr & 0x0f) + 1;
835 BUG_ON(offset == MSR_INVALID);
837 read ? clear_bit(bit_read, &tmp) : set_bit(bit_read, &tmp);
838 write ? clear_bit(bit_write, &tmp) : set_bit(bit_write, &tmp);
843 static void svm_vcpu_init_msrpm(u32 *msrpm)
847 memset(msrpm, 0xff, PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER));
849 for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
850 if (!direct_access_msrs[i].always)
853 set_msr_interception(msrpm, direct_access_msrs[i].index, 1, 1);
857 static void add_msr_offset(u32 offset)
861 for (i = 0; i < MSRPM_OFFSETS; ++i) {
863 /* Offset already in list? */
864 if (msrpm_offsets[i] == offset)
867 /* Slot used by another offset? */
868 if (msrpm_offsets[i] != MSR_INVALID)
871 /* Add offset to list */
872 msrpm_offsets[i] = offset;
878 * If this BUG triggers the msrpm_offsets table has an overflow. Just
879 * increase MSRPM_OFFSETS in this case.
884 static void init_msrpm_offsets(void)
888 memset(msrpm_offsets, 0xff, sizeof(msrpm_offsets));
890 for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
893 offset = svm_msrpm_offset(direct_access_msrs[i].index);
894 BUG_ON(offset == MSR_INVALID);
896 add_msr_offset(offset);
900 static void svm_enable_lbrv(struct vcpu_svm *svm)
902 u32 *msrpm = svm->msrpm;
904 svm->vmcb->control.lbr_ctl = 1;
905 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 1, 1);
906 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 1, 1);
907 set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 1, 1);
908 set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 1, 1);
911 static void svm_disable_lbrv(struct vcpu_svm *svm)
913 u32 *msrpm = svm->msrpm;
915 svm->vmcb->control.lbr_ctl = 0;
916 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 0, 0);
917 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 0, 0);
918 set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 0, 0);
919 set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 0, 0);
922 static __init int svm_hardware_setup(void)
925 struct page *iopm_pages;
929 iopm_pages = alloc_pages(GFP_KERNEL, IOPM_ALLOC_ORDER);
934 iopm_va = page_address(iopm_pages);
935 memset(iopm_va, 0xff, PAGE_SIZE * (1 << IOPM_ALLOC_ORDER));
936 iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT;
938 init_msrpm_offsets();
940 if (boot_cpu_has(X86_FEATURE_NX))
941 kvm_enable_efer_bits(EFER_NX);
943 if (boot_cpu_has(X86_FEATURE_FXSR_OPT))
944 kvm_enable_efer_bits(EFER_FFXSR);
946 if (boot_cpu_has(X86_FEATURE_TSCRATEMSR)) {
947 kvm_has_tsc_control = true;
948 kvm_max_tsc_scaling_ratio = TSC_RATIO_MAX;
949 kvm_tsc_scaling_ratio_frac_bits = 32;
953 printk(KERN_INFO "kvm: Nested Virtualization enabled\n");
954 kvm_enable_efer_bits(EFER_SVME | EFER_LMSLE);
957 for_each_possible_cpu(cpu) {
958 r = svm_cpu_init(cpu);
963 if (!boot_cpu_has(X86_FEATURE_NPT))
966 if (npt_enabled && !npt) {
967 printk(KERN_INFO "kvm: Nested Paging disabled\n");
972 printk(KERN_INFO "kvm: Nested Paging enabled\n");
977 if (avic && (!npt_enabled || !boot_cpu_has(X86_FEATURE_AVIC)))
981 pr_info("AVIC enabled\n");
986 __free_pages(iopm_pages, IOPM_ALLOC_ORDER);
991 static __exit void svm_hardware_unsetup(void)
995 for_each_possible_cpu(cpu)
998 __free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT), IOPM_ALLOC_ORDER);
1002 static void init_seg(struct vmcb_seg *seg)
1005 seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK |
1006 SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */
1007 seg->limit = 0xffff;
1011 static void init_sys_seg(struct vmcb_seg *seg, uint32_t type)
1014 seg->attrib = SVM_SELECTOR_P_MASK | type;
1015 seg->limit = 0xffff;
1019 static u64 svm_read_tsc_offset(struct kvm_vcpu *vcpu)
1021 struct vcpu_svm *svm = to_svm(vcpu);
1023 return svm->vmcb->control.tsc_offset;
1026 static void svm_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1028 struct vcpu_svm *svm = to_svm(vcpu);
1029 u64 g_tsc_offset = 0;
1031 if (is_guest_mode(vcpu)) {
1032 g_tsc_offset = svm->vmcb->control.tsc_offset -
1033 svm->nested.hsave->control.tsc_offset;
1034 svm->nested.hsave->control.tsc_offset = offset;
1036 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
1037 svm->vmcb->control.tsc_offset,
1040 svm->vmcb->control.tsc_offset = offset + g_tsc_offset;
1042 mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
1045 static void svm_adjust_tsc_offset_guest(struct kvm_vcpu *vcpu, s64 adjustment)
1047 struct vcpu_svm *svm = to_svm(vcpu);
1049 svm->vmcb->control.tsc_offset += adjustment;
1050 if (is_guest_mode(vcpu))
1051 svm->nested.hsave->control.tsc_offset += adjustment;
1053 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
1054 svm->vmcb->control.tsc_offset - adjustment,
1055 svm->vmcb->control.tsc_offset);
1057 mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
1060 static void avic_init_vmcb(struct vcpu_svm *svm)
1062 struct vmcb *vmcb = svm->vmcb;
1063 struct kvm_arch *vm_data = &svm->vcpu.kvm->arch;
1064 phys_addr_t bpa = page_to_phys(svm->avic_backing_page);
1065 phys_addr_t lpa = page_to_phys(vm_data->avic_logical_id_table_page);
1066 phys_addr_t ppa = page_to_phys(vm_data->avic_physical_id_table_page);
1068 vmcb->control.avic_backing_page = bpa & AVIC_HPA_MASK;
1069 vmcb->control.avic_logical_id = lpa & AVIC_HPA_MASK;
1070 vmcb->control.avic_physical_id = ppa & AVIC_HPA_MASK;
1071 vmcb->control.avic_physical_id |= AVIC_MAX_PHYSICAL_ID_COUNT;
1072 vmcb->control.int_ctl |= AVIC_ENABLE_MASK;
1073 svm->vcpu.arch.apicv_active = true;
1076 static void init_vmcb(struct vcpu_svm *svm)
1078 struct vmcb_control_area *control = &svm->vmcb->control;
1079 struct vmcb_save_area *save = &svm->vmcb->save;
1081 svm->vcpu.fpu_active = 1;
1082 svm->vcpu.arch.hflags = 0;
1084 set_cr_intercept(svm, INTERCEPT_CR0_READ);
1085 set_cr_intercept(svm, INTERCEPT_CR3_READ);
1086 set_cr_intercept(svm, INTERCEPT_CR4_READ);
1087 set_cr_intercept(svm, INTERCEPT_CR0_WRITE);
1088 set_cr_intercept(svm, INTERCEPT_CR3_WRITE);
1089 set_cr_intercept(svm, INTERCEPT_CR4_WRITE);
1090 set_cr_intercept(svm, INTERCEPT_CR8_WRITE);
1092 set_dr_intercepts(svm);
1094 set_exception_intercept(svm, PF_VECTOR);
1095 set_exception_intercept(svm, UD_VECTOR);
1096 set_exception_intercept(svm, MC_VECTOR);
1097 set_exception_intercept(svm, AC_VECTOR);
1098 set_exception_intercept(svm, DB_VECTOR);
1100 set_intercept(svm, INTERCEPT_INTR);
1101 set_intercept(svm, INTERCEPT_NMI);
1102 set_intercept(svm, INTERCEPT_SMI);
1103 set_intercept(svm, INTERCEPT_SELECTIVE_CR0);
1104 set_intercept(svm, INTERCEPT_RDPMC);
1105 set_intercept(svm, INTERCEPT_CPUID);
1106 set_intercept(svm, INTERCEPT_INVD);
1107 set_intercept(svm, INTERCEPT_HLT);
1108 set_intercept(svm, INTERCEPT_INVLPG);
1109 set_intercept(svm, INTERCEPT_INVLPGA);
1110 set_intercept(svm, INTERCEPT_IOIO_PROT);
1111 set_intercept(svm, INTERCEPT_MSR_PROT);
1112 set_intercept(svm, INTERCEPT_TASK_SWITCH);
1113 set_intercept(svm, INTERCEPT_SHUTDOWN);
1114 set_intercept(svm, INTERCEPT_VMRUN);
1115 set_intercept(svm, INTERCEPT_VMMCALL);
1116 set_intercept(svm, INTERCEPT_VMLOAD);
1117 set_intercept(svm, INTERCEPT_VMSAVE);
1118 set_intercept(svm, INTERCEPT_STGI);
1119 set_intercept(svm, INTERCEPT_CLGI);
1120 set_intercept(svm, INTERCEPT_SKINIT);
1121 set_intercept(svm, INTERCEPT_WBINVD);
1122 set_intercept(svm, INTERCEPT_MONITOR);
1123 set_intercept(svm, INTERCEPT_MWAIT);
1124 set_intercept(svm, INTERCEPT_XSETBV);
1126 control->iopm_base_pa = iopm_base;
1127 control->msrpm_base_pa = __pa(svm->msrpm);
1128 control->int_ctl = V_INTR_MASKING_MASK;
1130 init_seg(&save->es);
1131 init_seg(&save->ss);
1132 init_seg(&save->ds);
1133 init_seg(&save->fs);
1134 init_seg(&save->gs);
1136 save->cs.selector = 0xf000;
1137 save->cs.base = 0xffff0000;
1138 /* Executable/Readable Code Segment */
1139 save->cs.attrib = SVM_SELECTOR_READ_MASK | SVM_SELECTOR_P_MASK |
1140 SVM_SELECTOR_S_MASK | SVM_SELECTOR_CODE_MASK;
1141 save->cs.limit = 0xffff;
1143 save->gdtr.limit = 0xffff;
1144 save->idtr.limit = 0xffff;
1146 init_sys_seg(&save->ldtr, SEG_TYPE_LDT);
1147 init_sys_seg(&save->tr, SEG_TYPE_BUSY_TSS16);
1149 svm_set_efer(&svm->vcpu, 0);
1150 save->dr6 = 0xffff0ff0;
1151 kvm_set_rflags(&svm->vcpu, 2);
1152 save->rip = 0x0000fff0;
1153 svm->vcpu.arch.regs[VCPU_REGS_RIP] = save->rip;
1156 * svm_set_cr0() sets PG and WP and clears NW and CD on save->cr0.
1157 * It also updates the guest-visible cr0 value.
1159 svm_set_cr0(&svm->vcpu, X86_CR0_NW | X86_CR0_CD | X86_CR0_ET);
1160 kvm_mmu_reset_context(&svm->vcpu);
1162 save->cr4 = X86_CR4_PAE;
1166 /* Setup VMCB for Nested Paging */
1167 control->nested_ctl = 1;
1168 clr_intercept(svm, INTERCEPT_INVLPG);
1169 clr_exception_intercept(svm, PF_VECTOR);
1170 clr_cr_intercept(svm, INTERCEPT_CR3_READ);
1171 clr_cr_intercept(svm, INTERCEPT_CR3_WRITE);
1172 save->g_pat = svm->vcpu.arch.pat;
1176 svm->asid_generation = 0;
1178 svm->nested.vmcb = 0;
1179 svm->vcpu.arch.hflags = 0;
1181 if (boot_cpu_has(X86_FEATURE_PAUSEFILTER)) {
1182 control->pause_filter_count = 3000;
1183 set_intercept(svm, INTERCEPT_PAUSE);
1187 avic_init_vmcb(svm);
1189 mark_all_dirty(svm->vmcb);
1195 static u64 *avic_get_physical_id_entry(struct kvm_vcpu *vcpu, int index)
1197 u64 *avic_physical_id_table;
1198 struct kvm_arch *vm_data = &vcpu->kvm->arch;
1200 if (index >= AVIC_MAX_PHYSICAL_ID_COUNT)
1203 avic_physical_id_table = page_address(vm_data->avic_physical_id_table_page);
1205 return &avic_physical_id_table[index];
1210 * AVIC hardware walks the nested page table to check permissions,
1211 * but does not use the SPA address specified in the leaf page
1212 * table entry since it uses address in the AVIC_BACKING_PAGE pointer
1213 * field of the VMCB. Therefore, we set up the
1214 * APIC_ACCESS_PAGE_PRIVATE_MEMSLOT (4KB) here.
1216 static int avic_init_access_page(struct kvm_vcpu *vcpu)
1218 struct kvm *kvm = vcpu->kvm;
1221 if (kvm->arch.apic_access_page_done)
1224 ret = x86_set_memory_region(kvm,
1225 APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
1226 APIC_DEFAULT_PHYS_BASE,
1231 kvm->arch.apic_access_page_done = true;
1235 static int avic_init_backing_page(struct kvm_vcpu *vcpu)
1238 u64 *entry, new_entry;
1239 int id = vcpu->vcpu_id;
1240 struct vcpu_svm *svm = to_svm(vcpu);
1242 ret = avic_init_access_page(vcpu);
1246 if (id >= AVIC_MAX_PHYSICAL_ID_COUNT)
1249 if (!svm->vcpu.arch.apic->regs)
1252 svm->avic_backing_page = virt_to_page(svm->vcpu.arch.apic->regs);
1254 /* Setting AVIC backing page address in the phy APIC ID table */
1255 entry = avic_get_physical_id_entry(vcpu, id);
1259 new_entry = READ_ONCE(*entry);
1260 new_entry = (page_to_phys(svm->avic_backing_page) &
1261 AVIC_PHYSICAL_ID_ENTRY_BACKING_PAGE_MASK) |
1262 AVIC_PHYSICAL_ID_ENTRY_VALID_MASK;
1263 WRITE_ONCE(*entry, new_entry);
1265 svm->avic_physical_id_cache = entry;
1270 static void avic_vm_destroy(struct kvm *kvm)
1272 struct kvm_arch *vm_data = &kvm->arch;
1274 if (vm_data->avic_logical_id_table_page)
1275 __free_page(vm_data->avic_logical_id_table_page);
1276 if (vm_data->avic_physical_id_table_page)
1277 __free_page(vm_data->avic_physical_id_table_page);
1280 static int avic_vm_init(struct kvm *kvm)
1283 struct kvm_arch *vm_data = &kvm->arch;
1284 struct page *p_page;
1285 struct page *l_page;
1290 /* Allocating physical APIC ID table (4KB) */
1291 p_page = alloc_page(GFP_KERNEL);
1295 vm_data->avic_physical_id_table_page = p_page;
1296 clear_page(page_address(p_page));
1298 /* Allocating logical APIC ID table (4KB) */
1299 l_page = alloc_page(GFP_KERNEL);
1303 vm_data->avic_logical_id_table_page = l_page;
1304 clear_page(page_address(l_page));
1309 avic_vm_destroy(kvm);
1313 static void svm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
1315 struct vcpu_svm *svm = to_svm(vcpu);
1320 svm->vcpu.arch.apic_base = APIC_DEFAULT_PHYS_BASE |
1321 MSR_IA32_APICBASE_ENABLE;
1322 if (kvm_vcpu_is_reset_bsp(&svm->vcpu))
1323 svm->vcpu.arch.apic_base |= MSR_IA32_APICBASE_BSP;
1327 kvm_cpuid(vcpu, &eax, &dummy, &dummy, &dummy);
1328 kvm_register_write(vcpu, VCPU_REGS_RDX, eax);
1330 if (kvm_vcpu_apicv_active(vcpu) && !init_event)
1331 avic_update_vapic_bar(svm, APIC_DEFAULT_PHYS_BASE);
1334 static struct kvm_vcpu *svm_create_vcpu(struct kvm *kvm, unsigned int id)
1336 struct vcpu_svm *svm;
1338 struct page *msrpm_pages;
1339 struct page *hsave_page;
1340 struct page *nested_msrpm_pages;
1343 svm = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1349 err = kvm_vcpu_init(&svm->vcpu, kvm, id);
1354 page = alloc_page(GFP_KERNEL);
1358 msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
1362 nested_msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
1363 if (!nested_msrpm_pages)
1366 hsave_page = alloc_page(GFP_KERNEL);
1371 err = avic_init_backing_page(&svm->vcpu);
1376 svm->nested.hsave = page_address(hsave_page);
1378 svm->msrpm = page_address(msrpm_pages);
1379 svm_vcpu_init_msrpm(svm->msrpm);
1381 svm->nested.msrpm = page_address(nested_msrpm_pages);
1382 svm_vcpu_init_msrpm(svm->nested.msrpm);
1384 svm->vmcb = page_address(page);
1385 clear_page(svm->vmcb);
1386 svm->vmcb_pa = page_to_pfn(page) << PAGE_SHIFT;
1387 svm->asid_generation = 0;
1390 svm_init_osvw(&svm->vcpu);
1395 __free_page(hsave_page);
1397 __free_pages(nested_msrpm_pages, MSRPM_ALLOC_ORDER);
1399 __free_pages(msrpm_pages, MSRPM_ALLOC_ORDER);
1403 kvm_vcpu_uninit(&svm->vcpu);
1405 kmem_cache_free(kvm_vcpu_cache, svm);
1407 return ERR_PTR(err);
1410 static void svm_free_vcpu(struct kvm_vcpu *vcpu)
1412 struct vcpu_svm *svm = to_svm(vcpu);
1414 __free_page(pfn_to_page(svm->vmcb_pa >> PAGE_SHIFT));
1415 __free_pages(virt_to_page(svm->msrpm), MSRPM_ALLOC_ORDER);
1416 __free_page(virt_to_page(svm->nested.hsave));
1417 __free_pages(virt_to_page(svm->nested.msrpm), MSRPM_ALLOC_ORDER);
1418 kvm_vcpu_uninit(vcpu);
1419 kmem_cache_free(kvm_vcpu_cache, svm);
1422 static void svm_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1424 struct vcpu_svm *svm = to_svm(vcpu);
1427 if (unlikely(cpu != vcpu->cpu)) {
1428 svm->asid_generation = 0;
1429 mark_all_dirty(svm->vmcb);
1432 #ifdef CONFIG_X86_64
1433 rdmsrl(MSR_GS_BASE, to_svm(vcpu)->host.gs_base);
1435 savesegment(fs, svm->host.fs);
1436 savesegment(gs, svm->host.gs);
1437 svm->host.ldt = kvm_read_ldt();
1439 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
1440 rdmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
1442 if (static_cpu_has(X86_FEATURE_TSCRATEMSR)) {
1443 u64 tsc_ratio = vcpu->arch.tsc_scaling_ratio;
1444 if (tsc_ratio != __this_cpu_read(current_tsc_ratio)) {
1445 __this_cpu_write(current_tsc_ratio, tsc_ratio);
1446 wrmsrl(MSR_AMD64_TSC_RATIO, tsc_ratio);
1449 /* This assumes that the kernel never uses MSR_TSC_AUX */
1450 if (static_cpu_has(X86_FEATURE_RDTSCP))
1451 wrmsrl(MSR_TSC_AUX, svm->tsc_aux);
1454 static void svm_vcpu_put(struct kvm_vcpu *vcpu)
1456 struct vcpu_svm *svm = to_svm(vcpu);
1459 ++vcpu->stat.host_state_reload;
1460 kvm_load_ldt(svm->host.ldt);
1461 #ifdef CONFIG_X86_64
1462 loadsegment(fs, svm->host.fs);
1463 wrmsrl(MSR_KERNEL_GS_BASE, current->thread.gs);
1464 load_gs_index(svm->host.gs);
1466 #ifdef CONFIG_X86_32_LAZY_GS
1467 loadsegment(gs, svm->host.gs);
1470 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
1471 wrmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
1474 static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu)
1476 return to_svm(vcpu)->vmcb->save.rflags;
1479 static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1482 * Any change of EFLAGS.VM is accompained by a reload of SS
1483 * (caused by either a task switch or an inter-privilege IRET),
1484 * so we do not need to update the CPL here.
1486 to_svm(vcpu)->vmcb->save.rflags = rflags;
1489 static u32 svm_get_pkru(struct kvm_vcpu *vcpu)
1494 static void svm_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
1497 case VCPU_EXREG_PDPTR:
1498 BUG_ON(!npt_enabled);
1499 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
1506 static void svm_set_vintr(struct vcpu_svm *svm)
1508 set_intercept(svm, INTERCEPT_VINTR);
1511 static void svm_clear_vintr(struct vcpu_svm *svm)
1513 clr_intercept(svm, INTERCEPT_VINTR);
1516 static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg)
1518 struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
1521 case VCPU_SREG_CS: return &save->cs;
1522 case VCPU_SREG_DS: return &save->ds;
1523 case VCPU_SREG_ES: return &save->es;
1524 case VCPU_SREG_FS: return &save->fs;
1525 case VCPU_SREG_GS: return &save->gs;
1526 case VCPU_SREG_SS: return &save->ss;
1527 case VCPU_SREG_TR: return &save->tr;
1528 case VCPU_SREG_LDTR: return &save->ldtr;
1534 static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg)
1536 struct vmcb_seg *s = svm_seg(vcpu, seg);
1541 static void svm_get_segment(struct kvm_vcpu *vcpu,
1542 struct kvm_segment *var, int seg)
1544 struct vmcb_seg *s = svm_seg(vcpu, seg);
1546 var->base = s->base;
1547 var->limit = s->limit;
1548 var->selector = s->selector;
1549 var->type = s->attrib & SVM_SELECTOR_TYPE_MASK;
1550 var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1;
1551 var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
1552 var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1;
1553 var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1;
1554 var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
1555 var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
1558 * AMD CPUs circa 2014 track the G bit for all segments except CS.
1559 * However, the SVM spec states that the G bit is not observed by the
1560 * CPU, and some VMware virtual CPUs drop the G bit for all segments.
1561 * So let's synthesize a legal G bit for all segments, this helps
1562 * running KVM nested. It also helps cross-vendor migration, because
1563 * Intel's vmentry has a check on the 'G' bit.
1565 var->g = s->limit > 0xfffff;
1568 * AMD's VMCB does not have an explicit unusable field, so emulate it
1569 * for cross vendor migration purposes by "not present"
1571 var->unusable = !var->present || (var->type == 0);
1576 * Work around a bug where the busy flag in the tr selector
1586 * The accessed bit must always be set in the segment
1587 * descriptor cache, although it can be cleared in the
1588 * descriptor, the cached bit always remains at 1. Since
1589 * Intel has a check on this, set it here to support
1590 * cross-vendor migration.
1597 * On AMD CPUs sometimes the DB bit in the segment
1598 * descriptor is left as 1, although the whole segment has
1599 * been made unusable. Clear it here to pass an Intel VMX
1600 * entry check when cross vendor migrating.
1604 var->dpl = to_svm(vcpu)->vmcb->save.cpl;
1609 static int svm_get_cpl(struct kvm_vcpu *vcpu)
1611 struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
1616 static void svm_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1618 struct vcpu_svm *svm = to_svm(vcpu);
1620 dt->size = svm->vmcb->save.idtr.limit;
1621 dt->address = svm->vmcb->save.idtr.base;
1624 static void svm_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1626 struct vcpu_svm *svm = to_svm(vcpu);
1628 svm->vmcb->save.idtr.limit = dt->size;
1629 svm->vmcb->save.idtr.base = dt->address ;
1630 mark_dirty(svm->vmcb, VMCB_DT);
1633 static void svm_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1635 struct vcpu_svm *svm = to_svm(vcpu);
1637 dt->size = svm->vmcb->save.gdtr.limit;
1638 dt->address = svm->vmcb->save.gdtr.base;
1641 static void svm_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1643 struct vcpu_svm *svm = to_svm(vcpu);
1645 svm->vmcb->save.gdtr.limit = dt->size;
1646 svm->vmcb->save.gdtr.base = dt->address ;
1647 mark_dirty(svm->vmcb, VMCB_DT);
1650 static void svm_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
1654 static void svm_decache_cr3(struct kvm_vcpu *vcpu)
1658 static void svm_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
1662 static void update_cr0_intercept(struct vcpu_svm *svm)
1664 ulong gcr0 = svm->vcpu.arch.cr0;
1665 u64 *hcr0 = &svm->vmcb->save.cr0;
1667 if (!svm->vcpu.fpu_active)
1668 *hcr0 |= SVM_CR0_SELECTIVE_MASK;
1670 *hcr0 = (*hcr0 & ~SVM_CR0_SELECTIVE_MASK)
1671 | (gcr0 & SVM_CR0_SELECTIVE_MASK);
1673 mark_dirty(svm->vmcb, VMCB_CR);
1675 if (gcr0 == *hcr0 && svm->vcpu.fpu_active) {
1676 clr_cr_intercept(svm, INTERCEPT_CR0_READ);
1677 clr_cr_intercept(svm, INTERCEPT_CR0_WRITE);
1679 set_cr_intercept(svm, INTERCEPT_CR0_READ);
1680 set_cr_intercept(svm, INTERCEPT_CR0_WRITE);
1684 static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
1686 struct vcpu_svm *svm = to_svm(vcpu);
1688 #ifdef CONFIG_X86_64
1689 if (vcpu->arch.efer & EFER_LME) {
1690 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
1691 vcpu->arch.efer |= EFER_LMA;
1692 svm->vmcb->save.efer |= EFER_LMA | EFER_LME;
1695 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG)) {
1696 vcpu->arch.efer &= ~EFER_LMA;
1697 svm->vmcb->save.efer &= ~(EFER_LMA | EFER_LME);
1701 vcpu->arch.cr0 = cr0;
1704 cr0 |= X86_CR0_PG | X86_CR0_WP;
1706 if (!vcpu->fpu_active)
1709 * re-enable caching here because the QEMU bios
1710 * does not do it - this results in some delay at
1713 if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
1714 cr0 &= ~(X86_CR0_CD | X86_CR0_NW);
1715 svm->vmcb->save.cr0 = cr0;
1716 mark_dirty(svm->vmcb, VMCB_CR);
1717 update_cr0_intercept(svm);
1720 static int svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1722 unsigned long host_cr4_mce = cr4_read_shadow() & X86_CR4_MCE;
1723 unsigned long old_cr4 = to_svm(vcpu)->vmcb->save.cr4;
1725 if (cr4 & X86_CR4_VMXE)
1728 if (npt_enabled && ((old_cr4 ^ cr4) & X86_CR4_PGE))
1729 svm_flush_tlb(vcpu);
1731 vcpu->arch.cr4 = cr4;
1734 cr4 |= host_cr4_mce;
1735 to_svm(vcpu)->vmcb->save.cr4 = cr4;
1736 mark_dirty(to_svm(vcpu)->vmcb, VMCB_CR);
1740 static void svm_set_segment(struct kvm_vcpu *vcpu,
1741 struct kvm_segment *var, int seg)
1743 struct vcpu_svm *svm = to_svm(vcpu);
1744 struct vmcb_seg *s = svm_seg(vcpu, seg);
1746 s->base = var->base;
1747 s->limit = var->limit;
1748 s->selector = var->selector;
1752 s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK);
1753 s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT;
1754 s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT;
1755 s->attrib |= (var->present & 1) << SVM_SELECTOR_P_SHIFT;
1756 s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT;
1757 s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT;
1758 s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT;
1759 s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT;
1763 * This is always accurate, except if SYSRET returned to a segment
1764 * with SS.DPL != 3. Intel does not have this quirk, and always
1765 * forces SS.DPL to 3 on sysret, so we ignore that case; fixing it
1766 * would entail passing the CPL to userspace and back.
1768 if (seg == VCPU_SREG_SS)
1769 svm->vmcb->save.cpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
1771 mark_dirty(svm->vmcb, VMCB_SEG);
1774 static void update_bp_intercept(struct kvm_vcpu *vcpu)
1776 struct vcpu_svm *svm = to_svm(vcpu);
1778 clr_exception_intercept(svm, BP_VECTOR);
1780 if (vcpu->guest_debug & KVM_GUESTDBG_ENABLE) {
1781 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
1782 set_exception_intercept(svm, BP_VECTOR);
1784 vcpu->guest_debug = 0;
1787 static void new_asid(struct vcpu_svm *svm, struct svm_cpu_data *sd)
1789 if (sd->next_asid > sd->max_asid) {
1790 ++sd->asid_generation;
1792 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID;
1795 svm->asid_generation = sd->asid_generation;
1796 svm->vmcb->control.asid = sd->next_asid++;
1798 mark_dirty(svm->vmcb, VMCB_ASID);
1801 static u64 svm_get_dr6(struct kvm_vcpu *vcpu)
1803 return to_svm(vcpu)->vmcb->save.dr6;
1806 static void svm_set_dr6(struct kvm_vcpu *vcpu, unsigned long value)
1808 struct vcpu_svm *svm = to_svm(vcpu);
1810 svm->vmcb->save.dr6 = value;
1811 mark_dirty(svm->vmcb, VMCB_DR);
1814 static void svm_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
1816 struct vcpu_svm *svm = to_svm(vcpu);
1818 get_debugreg(vcpu->arch.db[0], 0);
1819 get_debugreg(vcpu->arch.db[1], 1);
1820 get_debugreg(vcpu->arch.db[2], 2);
1821 get_debugreg(vcpu->arch.db[3], 3);
1822 vcpu->arch.dr6 = svm_get_dr6(vcpu);
1823 vcpu->arch.dr7 = svm->vmcb->save.dr7;
1825 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
1826 set_dr_intercepts(svm);
1829 static void svm_set_dr7(struct kvm_vcpu *vcpu, unsigned long value)
1831 struct vcpu_svm *svm = to_svm(vcpu);
1833 svm->vmcb->save.dr7 = value;
1834 mark_dirty(svm->vmcb, VMCB_DR);
1837 static int pf_interception(struct vcpu_svm *svm)
1839 u64 fault_address = svm->vmcb->control.exit_info_2;
1843 switch (svm->apf_reason) {
1845 error_code = svm->vmcb->control.exit_info_1;
1847 trace_kvm_page_fault(fault_address, error_code);
1848 if (!npt_enabled && kvm_event_needs_reinjection(&svm->vcpu))
1849 kvm_mmu_unprotect_page_virt(&svm->vcpu, fault_address);
1850 r = kvm_mmu_page_fault(&svm->vcpu, fault_address, error_code,
1851 svm->vmcb->control.insn_bytes,
1852 svm->vmcb->control.insn_len);
1854 case KVM_PV_REASON_PAGE_NOT_PRESENT:
1855 svm->apf_reason = 0;
1856 local_irq_disable();
1857 kvm_async_pf_task_wait(fault_address);
1860 case KVM_PV_REASON_PAGE_READY:
1861 svm->apf_reason = 0;
1862 local_irq_disable();
1863 kvm_async_pf_task_wake(fault_address);
1870 static int db_interception(struct vcpu_svm *svm)
1872 struct kvm_run *kvm_run = svm->vcpu.run;
1874 if (!(svm->vcpu.guest_debug &
1875 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) &&
1876 !svm->nmi_singlestep) {
1877 kvm_queue_exception(&svm->vcpu, DB_VECTOR);
1881 if (svm->nmi_singlestep) {
1882 svm->nmi_singlestep = false;
1883 if (!(svm->vcpu.guest_debug & KVM_GUESTDBG_SINGLESTEP))
1884 svm->vmcb->save.rflags &=
1885 ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1888 if (svm->vcpu.guest_debug &
1889 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) {
1890 kvm_run->exit_reason = KVM_EXIT_DEBUG;
1891 kvm_run->debug.arch.pc =
1892 svm->vmcb->save.cs.base + svm->vmcb->save.rip;
1893 kvm_run->debug.arch.exception = DB_VECTOR;
1900 static int bp_interception(struct vcpu_svm *svm)
1902 struct kvm_run *kvm_run = svm->vcpu.run;
1904 kvm_run->exit_reason = KVM_EXIT_DEBUG;
1905 kvm_run->debug.arch.pc = svm->vmcb->save.cs.base + svm->vmcb->save.rip;
1906 kvm_run->debug.arch.exception = BP_VECTOR;
1910 static int ud_interception(struct vcpu_svm *svm)
1914 er = emulate_instruction(&svm->vcpu, EMULTYPE_TRAP_UD);
1915 if (er != EMULATE_DONE)
1916 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
1920 static int ac_interception(struct vcpu_svm *svm)
1922 kvm_queue_exception_e(&svm->vcpu, AC_VECTOR, 0);
1926 static void svm_fpu_activate(struct kvm_vcpu *vcpu)
1928 struct vcpu_svm *svm = to_svm(vcpu);
1930 clr_exception_intercept(svm, NM_VECTOR);
1932 svm->vcpu.fpu_active = 1;
1933 update_cr0_intercept(svm);
1936 static int nm_interception(struct vcpu_svm *svm)
1938 svm_fpu_activate(&svm->vcpu);
1942 static bool is_erratum_383(void)
1947 if (!erratum_383_found)
1950 value = native_read_msr_safe(MSR_IA32_MC0_STATUS, &err);
1954 /* Bit 62 may or may not be set for this mce */
1955 value &= ~(1ULL << 62);
1957 if (value != 0xb600000000010015ULL)
1960 /* Clear MCi_STATUS registers */
1961 for (i = 0; i < 6; ++i)
1962 native_write_msr_safe(MSR_IA32_MCx_STATUS(i), 0, 0);
1964 value = native_read_msr_safe(MSR_IA32_MCG_STATUS, &err);
1968 value &= ~(1ULL << 2);
1969 low = lower_32_bits(value);
1970 high = upper_32_bits(value);
1972 native_write_msr_safe(MSR_IA32_MCG_STATUS, low, high);
1975 /* Flush tlb to evict multi-match entries */
1981 static void svm_handle_mce(struct vcpu_svm *svm)
1983 if (is_erratum_383()) {
1985 * Erratum 383 triggered. Guest state is corrupt so kill the
1988 pr_err("KVM: Guest triggered AMD Erratum 383\n");
1990 kvm_make_request(KVM_REQ_TRIPLE_FAULT, &svm->vcpu);
1996 * On an #MC intercept the MCE handler is not called automatically in
1997 * the host. So do it by hand here.
2001 /* not sure if we ever come back to this point */
2006 static int mc_interception(struct vcpu_svm *svm)
2011 static int shutdown_interception(struct vcpu_svm *svm)
2013 struct kvm_run *kvm_run = svm->vcpu.run;
2016 * VMCB is undefined after a SHUTDOWN intercept
2017 * so reinitialize it.
2019 clear_page(svm->vmcb);
2022 kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
2026 static int io_interception(struct vcpu_svm *svm)
2028 struct kvm_vcpu *vcpu = &svm->vcpu;
2029 u32 io_info = svm->vmcb->control.exit_info_1; /* address size bug? */
2030 int size, in, string;
2033 ++svm->vcpu.stat.io_exits;
2034 string = (io_info & SVM_IOIO_STR_MASK) != 0;
2035 in = (io_info & SVM_IOIO_TYPE_MASK) != 0;
2037 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
2039 port = io_info >> 16;
2040 size = (io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT;
2041 svm->next_rip = svm->vmcb->control.exit_info_2;
2042 skip_emulated_instruction(&svm->vcpu);
2044 return kvm_fast_pio_out(vcpu, size, port);
2047 static int nmi_interception(struct vcpu_svm *svm)
2052 static int intr_interception(struct vcpu_svm *svm)
2054 ++svm->vcpu.stat.irq_exits;
2058 static int nop_on_interception(struct vcpu_svm *svm)
2063 static int halt_interception(struct vcpu_svm *svm)
2065 svm->next_rip = kvm_rip_read(&svm->vcpu) + 1;
2066 return kvm_emulate_halt(&svm->vcpu);
2069 static int vmmcall_interception(struct vcpu_svm *svm)
2071 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2072 return kvm_emulate_hypercall(&svm->vcpu);
2075 static unsigned long nested_svm_get_tdp_cr3(struct kvm_vcpu *vcpu)
2077 struct vcpu_svm *svm = to_svm(vcpu);
2079 return svm->nested.nested_cr3;
2082 static u64 nested_svm_get_tdp_pdptr(struct kvm_vcpu *vcpu, int index)
2084 struct vcpu_svm *svm = to_svm(vcpu);
2085 u64 cr3 = svm->nested.nested_cr3;
2089 ret = kvm_vcpu_read_guest_page(vcpu, gpa_to_gfn(cr3), &pdpte,
2090 offset_in_page(cr3) + index * 8, 8);
2096 static void nested_svm_set_tdp_cr3(struct kvm_vcpu *vcpu,
2099 struct vcpu_svm *svm = to_svm(vcpu);
2101 svm->vmcb->control.nested_cr3 = root;
2102 mark_dirty(svm->vmcb, VMCB_NPT);
2103 svm_flush_tlb(vcpu);
2106 static void nested_svm_inject_npf_exit(struct kvm_vcpu *vcpu,
2107 struct x86_exception *fault)
2109 struct vcpu_svm *svm = to_svm(vcpu);
2111 if (svm->vmcb->control.exit_code != SVM_EXIT_NPF) {
2113 * TODO: track the cause of the nested page fault, and
2114 * correctly fill in the high bits of exit_info_1.
2116 svm->vmcb->control.exit_code = SVM_EXIT_NPF;
2117 svm->vmcb->control.exit_code_hi = 0;
2118 svm->vmcb->control.exit_info_1 = (1ULL << 32);
2119 svm->vmcb->control.exit_info_2 = fault->address;
2122 svm->vmcb->control.exit_info_1 &= ~0xffffffffULL;
2123 svm->vmcb->control.exit_info_1 |= fault->error_code;
2126 * The present bit is always zero for page structure faults on real
2129 if (svm->vmcb->control.exit_info_1 & (2ULL << 32))
2130 svm->vmcb->control.exit_info_1 &= ~1;
2132 nested_svm_vmexit(svm);
2135 static void nested_svm_init_mmu_context(struct kvm_vcpu *vcpu)
2137 WARN_ON(mmu_is_nested(vcpu));
2138 kvm_init_shadow_mmu(vcpu);
2139 vcpu->arch.mmu.set_cr3 = nested_svm_set_tdp_cr3;
2140 vcpu->arch.mmu.get_cr3 = nested_svm_get_tdp_cr3;
2141 vcpu->arch.mmu.get_pdptr = nested_svm_get_tdp_pdptr;
2142 vcpu->arch.mmu.inject_page_fault = nested_svm_inject_npf_exit;
2143 vcpu->arch.mmu.shadow_root_level = get_npt_level();
2144 reset_shadow_zero_bits_mask(vcpu, &vcpu->arch.mmu);
2145 vcpu->arch.walk_mmu = &vcpu->arch.nested_mmu;
2148 static void nested_svm_uninit_mmu_context(struct kvm_vcpu *vcpu)
2150 vcpu->arch.walk_mmu = &vcpu->arch.mmu;
2153 static int nested_svm_check_permissions(struct vcpu_svm *svm)
2155 if (!(svm->vcpu.arch.efer & EFER_SVME)
2156 || !is_paging(&svm->vcpu)) {
2157 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2161 if (svm->vmcb->save.cpl) {
2162 kvm_inject_gp(&svm->vcpu, 0);
2169 static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
2170 bool has_error_code, u32 error_code)
2174 if (!is_guest_mode(&svm->vcpu))
2177 svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + nr;
2178 svm->vmcb->control.exit_code_hi = 0;
2179 svm->vmcb->control.exit_info_1 = error_code;
2180 svm->vmcb->control.exit_info_2 = svm->vcpu.arch.cr2;
2182 vmexit = nested_svm_intercept(svm);
2183 if (vmexit == NESTED_EXIT_DONE)
2184 svm->nested.exit_required = true;
2189 /* This function returns true if it is save to enable the irq window */
2190 static inline bool nested_svm_intr(struct vcpu_svm *svm)
2192 if (!is_guest_mode(&svm->vcpu))
2195 if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
2198 if (!(svm->vcpu.arch.hflags & HF_HIF_MASK))
2202 * if vmexit was already requested (by intercepted exception
2203 * for instance) do not overwrite it with "external interrupt"
2206 if (svm->nested.exit_required)
2209 svm->vmcb->control.exit_code = SVM_EXIT_INTR;
2210 svm->vmcb->control.exit_info_1 = 0;
2211 svm->vmcb->control.exit_info_2 = 0;
2213 if (svm->nested.intercept & 1ULL) {
2215 * The #vmexit can't be emulated here directly because this
2216 * code path runs with irqs and preemption disabled. A
2217 * #vmexit emulation might sleep. Only signal request for
2220 svm->nested.exit_required = true;
2221 trace_kvm_nested_intr_vmexit(svm->vmcb->save.rip);
2228 /* This function returns true if it is save to enable the nmi window */
2229 static inline bool nested_svm_nmi(struct vcpu_svm *svm)
2231 if (!is_guest_mode(&svm->vcpu))
2234 if (!(svm->nested.intercept & (1ULL << INTERCEPT_NMI)))
2237 svm->vmcb->control.exit_code = SVM_EXIT_NMI;
2238 svm->nested.exit_required = true;
2243 static void *nested_svm_map(struct vcpu_svm *svm, u64 gpa, struct page **_page)
2249 page = kvm_vcpu_gfn_to_page(&svm->vcpu, gpa >> PAGE_SHIFT);
2250 if (is_error_page(page))
2258 kvm_inject_gp(&svm->vcpu, 0);
2263 static void nested_svm_unmap(struct page *page)
2266 kvm_release_page_dirty(page);
2269 static int nested_svm_intercept_ioio(struct vcpu_svm *svm)
2271 unsigned port, size, iopm_len;
2276 if (!(svm->nested.intercept & (1ULL << INTERCEPT_IOIO_PROT)))
2277 return NESTED_EXIT_HOST;
2279 port = svm->vmcb->control.exit_info_1 >> 16;
2280 size = (svm->vmcb->control.exit_info_1 & SVM_IOIO_SIZE_MASK) >>
2281 SVM_IOIO_SIZE_SHIFT;
2282 gpa = svm->nested.vmcb_iopm + (port / 8);
2283 start_bit = port % 8;
2284 iopm_len = (start_bit + size > 8) ? 2 : 1;
2285 mask = (0xf >> (4 - size)) << start_bit;
2288 if (kvm_vcpu_read_guest(&svm->vcpu, gpa, &val, iopm_len))
2289 return NESTED_EXIT_DONE;
2291 return (val & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
2294 static int nested_svm_exit_handled_msr(struct vcpu_svm *svm)
2296 u32 offset, msr, value;
2299 if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT)))
2300 return NESTED_EXIT_HOST;
2302 msr = svm->vcpu.arch.regs[VCPU_REGS_RCX];
2303 offset = svm_msrpm_offset(msr);
2304 write = svm->vmcb->control.exit_info_1 & 1;
2305 mask = 1 << ((2 * (msr & 0xf)) + write);
2307 if (offset == MSR_INVALID)
2308 return NESTED_EXIT_DONE;
2310 /* Offset is in 32 bit units but need in 8 bit units */
2313 if (kvm_vcpu_read_guest(&svm->vcpu, svm->nested.vmcb_msrpm + offset, &value, 4))
2314 return NESTED_EXIT_DONE;
2316 return (value & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
2319 static int nested_svm_exit_special(struct vcpu_svm *svm)
2321 u32 exit_code = svm->vmcb->control.exit_code;
2323 switch (exit_code) {
2326 case SVM_EXIT_EXCP_BASE + MC_VECTOR:
2327 return NESTED_EXIT_HOST;
2329 /* For now we are always handling NPFs when using them */
2331 return NESTED_EXIT_HOST;
2333 case SVM_EXIT_EXCP_BASE + PF_VECTOR:
2334 /* When we're shadowing, trap PFs, but not async PF */
2335 if (!npt_enabled && svm->apf_reason == 0)
2336 return NESTED_EXIT_HOST;
2338 case SVM_EXIT_EXCP_BASE + NM_VECTOR:
2339 nm_interception(svm);
2345 return NESTED_EXIT_CONTINUE;
2349 * If this function returns true, this #vmexit was already handled
2351 static int nested_svm_intercept(struct vcpu_svm *svm)
2353 u32 exit_code = svm->vmcb->control.exit_code;
2354 int vmexit = NESTED_EXIT_HOST;
2356 switch (exit_code) {
2358 vmexit = nested_svm_exit_handled_msr(svm);
2361 vmexit = nested_svm_intercept_ioio(svm);
2363 case SVM_EXIT_READ_CR0 ... SVM_EXIT_WRITE_CR8: {
2364 u32 bit = 1U << (exit_code - SVM_EXIT_READ_CR0);
2365 if (svm->nested.intercept_cr & bit)
2366 vmexit = NESTED_EXIT_DONE;
2369 case SVM_EXIT_READ_DR0 ... SVM_EXIT_WRITE_DR7: {
2370 u32 bit = 1U << (exit_code - SVM_EXIT_READ_DR0);
2371 if (svm->nested.intercept_dr & bit)
2372 vmexit = NESTED_EXIT_DONE;
2375 case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
2376 u32 excp_bits = 1 << (exit_code - SVM_EXIT_EXCP_BASE);
2377 if (svm->nested.intercept_exceptions & excp_bits)
2378 vmexit = NESTED_EXIT_DONE;
2379 /* async page fault always cause vmexit */
2380 else if ((exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR) &&
2381 svm->apf_reason != 0)
2382 vmexit = NESTED_EXIT_DONE;
2385 case SVM_EXIT_ERR: {
2386 vmexit = NESTED_EXIT_DONE;
2390 u64 exit_bits = 1ULL << (exit_code - SVM_EXIT_INTR);
2391 if (svm->nested.intercept & exit_bits)
2392 vmexit = NESTED_EXIT_DONE;
2399 static int nested_svm_exit_handled(struct vcpu_svm *svm)
2403 vmexit = nested_svm_intercept(svm);
2405 if (vmexit == NESTED_EXIT_DONE)
2406 nested_svm_vmexit(svm);
2411 static inline void copy_vmcb_control_area(struct vmcb *dst_vmcb, struct vmcb *from_vmcb)
2413 struct vmcb_control_area *dst = &dst_vmcb->control;
2414 struct vmcb_control_area *from = &from_vmcb->control;
2416 dst->intercept_cr = from->intercept_cr;
2417 dst->intercept_dr = from->intercept_dr;
2418 dst->intercept_exceptions = from->intercept_exceptions;
2419 dst->intercept = from->intercept;
2420 dst->iopm_base_pa = from->iopm_base_pa;
2421 dst->msrpm_base_pa = from->msrpm_base_pa;
2422 dst->tsc_offset = from->tsc_offset;
2423 dst->asid = from->asid;
2424 dst->tlb_ctl = from->tlb_ctl;
2425 dst->int_ctl = from->int_ctl;
2426 dst->int_vector = from->int_vector;
2427 dst->int_state = from->int_state;
2428 dst->exit_code = from->exit_code;
2429 dst->exit_code_hi = from->exit_code_hi;
2430 dst->exit_info_1 = from->exit_info_1;
2431 dst->exit_info_2 = from->exit_info_2;
2432 dst->exit_int_info = from->exit_int_info;
2433 dst->exit_int_info_err = from->exit_int_info_err;
2434 dst->nested_ctl = from->nested_ctl;
2435 dst->event_inj = from->event_inj;
2436 dst->event_inj_err = from->event_inj_err;
2437 dst->nested_cr3 = from->nested_cr3;
2438 dst->lbr_ctl = from->lbr_ctl;
2441 static int nested_svm_vmexit(struct vcpu_svm *svm)
2443 struct vmcb *nested_vmcb;
2444 struct vmcb *hsave = svm->nested.hsave;
2445 struct vmcb *vmcb = svm->vmcb;
2448 trace_kvm_nested_vmexit_inject(vmcb->control.exit_code,
2449 vmcb->control.exit_info_1,
2450 vmcb->control.exit_info_2,
2451 vmcb->control.exit_int_info,
2452 vmcb->control.exit_int_info_err,
2455 nested_vmcb = nested_svm_map(svm, svm->nested.vmcb, &page);
2459 /* Exit Guest-Mode */
2460 leave_guest_mode(&svm->vcpu);
2461 svm->nested.vmcb = 0;
2463 /* Give the current vmcb to the guest */
2466 nested_vmcb->save.es = vmcb->save.es;
2467 nested_vmcb->save.cs = vmcb->save.cs;
2468 nested_vmcb->save.ss = vmcb->save.ss;
2469 nested_vmcb->save.ds = vmcb->save.ds;
2470 nested_vmcb->save.gdtr = vmcb->save.gdtr;
2471 nested_vmcb->save.idtr = vmcb->save.idtr;
2472 nested_vmcb->save.efer = svm->vcpu.arch.efer;
2473 nested_vmcb->save.cr0 = kvm_read_cr0(&svm->vcpu);
2474 nested_vmcb->save.cr3 = kvm_read_cr3(&svm->vcpu);
2475 nested_vmcb->save.cr2 = vmcb->save.cr2;
2476 nested_vmcb->save.cr4 = svm->vcpu.arch.cr4;
2477 nested_vmcb->save.rflags = kvm_get_rflags(&svm->vcpu);
2478 nested_vmcb->save.rip = vmcb->save.rip;
2479 nested_vmcb->save.rsp = vmcb->save.rsp;
2480 nested_vmcb->save.rax = vmcb->save.rax;
2481 nested_vmcb->save.dr7 = vmcb->save.dr7;
2482 nested_vmcb->save.dr6 = vmcb->save.dr6;
2483 nested_vmcb->save.cpl = vmcb->save.cpl;
2485 nested_vmcb->control.int_ctl = vmcb->control.int_ctl;
2486 nested_vmcb->control.int_vector = vmcb->control.int_vector;
2487 nested_vmcb->control.int_state = vmcb->control.int_state;
2488 nested_vmcb->control.exit_code = vmcb->control.exit_code;
2489 nested_vmcb->control.exit_code_hi = vmcb->control.exit_code_hi;
2490 nested_vmcb->control.exit_info_1 = vmcb->control.exit_info_1;
2491 nested_vmcb->control.exit_info_2 = vmcb->control.exit_info_2;
2492 nested_vmcb->control.exit_int_info = vmcb->control.exit_int_info;
2493 nested_vmcb->control.exit_int_info_err = vmcb->control.exit_int_info_err;
2495 if (svm->nrips_enabled)
2496 nested_vmcb->control.next_rip = vmcb->control.next_rip;
2499 * If we emulate a VMRUN/#VMEXIT in the same host #vmexit cycle we have
2500 * to make sure that we do not lose injected events. So check event_inj
2501 * here and copy it to exit_int_info if it is valid.
2502 * Exit_int_info and event_inj can't be both valid because the case
2503 * below only happens on a VMRUN instruction intercept which has
2504 * no valid exit_int_info set.
2506 if (vmcb->control.event_inj & SVM_EVTINJ_VALID) {
2507 struct vmcb_control_area *nc = &nested_vmcb->control;
2509 nc->exit_int_info = vmcb->control.event_inj;
2510 nc->exit_int_info_err = vmcb->control.event_inj_err;
2513 nested_vmcb->control.tlb_ctl = 0;
2514 nested_vmcb->control.event_inj = 0;
2515 nested_vmcb->control.event_inj_err = 0;
2517 /* We always set V_INTR_MASKING and remember the old value in hflags */
2518 if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
2519 nested_vmcb->control.int_ctl &= ~V_INTR_MASKING_MASK;
2521 /* Restore the original control entries */
2522 copy_vmcb_control_area(vmcb, hsave);
2524 kvm_clear_exception_queue(&svm->vcpu);
2525 kvm_clear_interrupt_queue(&svm->vcpu);
2527 svm->nested.nested_cr3 = 0;
2529 /* Restore selected save entries */
2530 svm->vmcb->save.es = hsave->save.es;
2531 svm->vmcb->save.cs = hsave->save.cs;
2532 svm->vmcb->save.ss = hsave->save.ss;
2533 svm->vmcb->save.ds = hsave->save.ds;
2534 svm->vmcb->save.gdtr = hsave->save.gdtr;
2535 svm->vmcb->save.idtr = hsave->save.idtr;
2536 kvm_set_rflags(&svm->vcpu, hsave->save.rflags);
2537 svm_set_efer(&svm->vcpu, hsave->save.efer);
2538 svm_set_cr0(&svm->vcpu, hsave->save.cr0 | X86_CR0_PE);
2539 svm_set_cr4(&svm->vcpu, hsave->save.cr4);
2541 svm->vmcb->save.cr3 = hsave->save.cr3;
2542 svm->vcpu.arch.cr3 = hsave->save.cr3;
2544 (void)kvm_set_cr3(&svm->vcpu, hsave->save.cr3);
2546 kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, hsave->save.rax);
2547 kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, hsave->save.rsp);
2548 kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, hsave->save.rip);
2549 svm->vmcb->save.dr7 = 0;
2550 svm->vmcb->save.cpl = 0;
2551 svm->vmcb->control.exit_int_info = 0;
2553 mark_all_dirty(svm->vmcb);
2555 nested_svm_unmap(page);
2557 nested_svm_uninit_mmu_context(&svm->vcpu);
2558 kvm_mmu_reset_context(&svm->vcpu);
2559 kvm_mmu_load(&svm->vcpu);
2564 static bool nested_svm_vmrun_msrpm(struct vcpu_svm *svm)
2567 * This function merges the msr permission bitmaps of kvm and the
2568 * nested vmcb. It is optimized in that it only merges the parts where
2569 * the kvm msr permission bitmap may contain zero bits
2573 if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT)))
2576 for (i = 0; i < MSRPM_OFFSETS; i++) {
2580 if (msrpm_offsets[i] == 0xffffffff)
2583 p = msrpm_offsets[i];
2584 offset = svm->nested.vmcb_msrpm + (p * 4);
2586 if (kvm_vcpu_read_guest(&svm->vcpu, offset, &value, 4))
2589 svm->nested.msrpm[p] = svm->msrpm[p] | value;
2592 svm->vmcb->control.msrpm_base_pa = __pa(svm->nested.msrpm);
2597 static bool nested_vmcb_checks(struct vmcb *vmcb)
2599 if ((vmcb->control.intercept & (1ULL << INTERCEPT_VMRUN)) == 0)
2602 if (vmcb->control.asid == 0)
2605 if (vmcb->control.nested_ctl && !npt_enabled)
2611 static bool nested_svm_vmrun(struct vcpu_svm *svm)
2613 struct vmcb *nested_vmcb;
2614 struct vmcb *hsave = svm->nested.hsave;
2615 struct vmcb *vmcb = svm->vmcb;
2619 vmcb_gpa = svm->vmcb->save.rax;
2621 nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
2625 if (!nested_vmcb_checks(nested_vmcb)) {
2626 nested_vmcb->control.exit_code = SVM_EXIT_ERR;
2627 nested_vmcb->control.exit_code_hi = 0;
2628 nested_vmcb->control.exit_info_1 = 0;
2629 nested_vmcb->control.exit_info_2 = 0;
2631 nested_svm_unmap(page);
2636 trace_kvm_nested_vmrun(svm->vmcb->save.rip, vmcb_gpa,
2637 nested_vmcb->save.rip,
2638 nested_vmcb->control.int_ctl,
2639 nested_vmcb->control.event_inj,
2640 nested_vmcb->control.nested_ctl);
2642 trace_kvm_nested_intercepts(nested_vmcb->control.intercept_cr & 0xffff,
2643 nested_vmcb->control.intercept_cr >> 16,
2644 nested_vmcb->control.intercept_exceptions,
2645 nested_vmcb->control.intercept);
2647 /* Clear internal status */
2648 kvm_clear_exception_queue(&svm->vcpu);
2649 kvm_clear_interrupt_queue(&svm->vcpu);
2652 * Save the old vmcb, so we don't need to pick what we save, but can
2653 * restore everything when a VMEXIT occurs
2655 hsave->save.es = vmcb->save.es;
2656 hsave->save.cs = vmcb->save.cs;
2657 hsave->save.ss = vmcb->save.ss;
2658 hsave->save.ds = vmcb->save.ds;
2659 hsave->save.gdtr = vmcb->save.gdtr;
2660 hsave->save.idtr = vmcb->save.idtr;
2661 hsave->save.efer = svm->vcpu.arch.efer;
2662 hsave->save.cr0 = kvm_read_cr0(&svm->vcpu);
2663 hsave->save.cr4 = svm->vcpu.arch.cr4;
2664 hsave->save.rflags = kvm_get_rflags(&svm->vcpu);
2665 hsave->save.rip = kvm_rip_read(&svm->vcpu);
2666 hsave->save.rsp = vmcb->save.rsp;
2667 hsave->save.rax = vmcb->save.rax;
2669 hsave->save.cr3 = vmcb->save.cr3;
2671 hsave->save.cr3 = kvm_read_cr3(&svm->vcpu);
2673 copy_vmcb_control_area(hsave, vmcb);
2675 if (kvm_get_rflags(&svm->vcpu) & X86_EFLAGS_IF)
2676 svm->vcpu.arch.hflags |= HF_HIF_MASK;
2678 svm->vcpu.arch.hflags &= ~HF_HIF_MASK;
2680 if (nested_vmcb->control.nested_ctl) {
2681 kvm_mmu_unload(&svm->vcpu);
2682 svm->nested.nested_cr3 = nested_vmcb->control.nested_cr3;
2683 nested_svm_init_mmu_context(&svm->vcpu);
2686 /* Load the nested guest state */
2687 svm->vmcb->save.es = nested_vmcb->save.es;
2688 svm->vmcb->save.cs = nested_vmcb->save.cs;
2689 svm->vmcb->save.ss = nested_vmcb->save.ss;
2690 svm->vmcb->save.ds = nested_vmcb->save.ds;
2691 svm->vmcb->save.gdtr = nested_vmcb->save.gdtr;
2692 svm->vmcb->save.idtr = nested_vmcb->save.idtr;
2693 kvm_set_rflags(&svm->vcpu, nested_vmcb->save.rflags);
2694 svm_set_efer(&svm->vcpu, nested_vmcb->save.efer);
2695 svm_set_cr0(&svm->vcpu, nested_vmcb->save.cr0);
2696 svm_set_cr4(&svm->vcpu, nested_vmcb->save.cr4);
2698 svm->vmcb->save.cr3 = nested_vmcb->save.cr3;
2699 svm->vcpu.arch.cr3 = nested_vmcb->save.cr3;
2701 (void)kvm_set_cr3(&svm->vcpu, nested_vmcb->save.cr3);
2703 /* Guest paging mode is active - reset mmu */
2704 kvm_mmu_reset_context(&svm->vcpu);
2706 svm->vmcb->save.cr2 = svm->vcpu.arch.cr2 = nested_vmcb->save.cr2;
2707 kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, nested_vmcb->save.rax);
2708 kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, nested_vmcb->save.rsp);
2709 kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, nested_vmcb->save.rip);
2711 /* In case we don't even reach vcpu_run, the fields are not updated */
2712 svm->vmcb->save.rax = nested_vmcb->save.rax;
2713 svm->vmcb->save.rsp = nested_vmcb->save.rsp;
2714 svm->vmcb->save.rip = nested_vmcb->save.rip;
2715 svm->vmcb->save.dr7 = nested_vmcb->save.dr7;
2716 svm->vmcb->save.dr6 = nested_vmcb->save.dr6;
2717 svm->vmcb->save.cpl = nested_vmcb->save.cpl;
2719 svm->nested.vmcb_msrpm = nested_vmcb->control.msrpm_base_pa & ~0x0fffULL;
2720 svm->nested.vmcb_iopm = nested_vmcb->control.iopm_base_pa & ~0x0fffULL;
2722 /* cache intercepts */
2723 svm->nested.intercept_cr = nested_vmcb->control.intercept_cr;
2724 svm->nested.intercept_dr = nested_vmcb->control.intercept_dr;
2725 svm->nested.intercept_exceptions = nested_vmcb->control.intercept_exceptions;
2726 svm->nested.intercept = nested_vmcb->control.intercept;
2728 svm_flush_tlb(&svm->vcpu);
2729 svm->vmcb->control.int_ctl = nested_vmcb->control.int_ctl | V_INTR_MASKING_MASK;
2730 if (nested_vmcb->control.int_ctl & V_INTR_MASKING_MASK)
2731 svm->vcpu.arch.hflags |= HF_VINTR_MASK;
2733 svm->vcpu.arch.hflags &= ~HF_VINTR_MASK;
2735 if (svm->vcpu.arch.hflags & HF_VINTR_MASK) {
2736 /* We only want the cr8 intercept bits of the guest */
2737 clr_cr_intercept(svm, INTERCEPT_CR8_READ);
2738 clr_cr_intercept(svm, INTERCEPT_CR8_WRITE);
2741 /* We don't want to see VMMCALLs from a nested guest */
2742 clr_intercept(svm, INTERCEPT_VMMCALL);
2744 svm->vmcb->control.lbr_ctl = nested_vmcb->control.lbr_ctl;
2745 svm->vmcb->control.int_vector = nested_vmcb->control.int_vector;
2746 svm->vmcb->control.int_state = nested_vmcb->control.int_state;
2747 svm->vmcb->control.tsc_offset += nested_vmcb->control.tsc_offset;
2748 svm->vmcb->control.event_inj = nested_vmcb->control.event_inj;
2749 svm->vmcb->control.event_inj_err = nested_vmcb->control.event_inj_err;
2751 nested_svm_unmap(page);
2753 /* Enter Guest-Mode */
2754 enter_guest_mode(&svm->vcpu);
2757 * Merge guest and host intercepts - must be called with vcpu in
2758 * guest-mode to take affect here
2760 recalc_intercepts(svm);
2762 svm->nested.vmcb = vmcb_gpa;
2766 mark_all_dirty(svm->vmcb);
2771 static void nested_svm_vmloadsave(struct vmcb *from_vmcb, struct vmcb *to_vmcb)
2773 to_vmcb->save.fs = from_vmcb->save.fs;
2774 to_vmcb->save.gs = from_vmcb->save.gs;
2775 to_vmcb->save.tr = from_vmcb->save.tr;
2776 to_vmcb->save.ldtr = from_vmcb->save.ldtr;
2777 to_vmcb->save.kernel_gs_base = from_vmcb->save.kernel_gs_base;
2778 to_vmcb->save.star = from_vmcb->save.star;
2779 to_vmcb->save.lstar = from_vmcb->save.lstar;
2780 to_vmcb->save.cstar = from_vmcb->save.cstar;
2781 to_vmcb->save.sfmask = from_vmcb->save.sfmask;
2782 to_vmcb->save.sysenter_cs = from_vmcb->save.sysenter_cs;
2783 to_vmcb->save.sysenter_esp = from_vmcb->save.sysenter_esp;
2784 to_vmcb->save.sysenter_eip = from_vmcb->save.sysenter_eip;
2787 static int vmload_interception(struct vcpu_svm *svm)
2789 struct vmcb *nested_vmcb;
2792 if (nested_svm_check_permissions(svm))
2795 nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
2799 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2800 skip_emulated_instruction(&svm->vcpu);
2802 nested_svm_vmloadsave(nested_vmcb, svm->vmcb);
2803 nested_svm_unmap(page);
2808 static int vmsave_interception(struct vcpu_svm *svm)
2810 struct vmcb *nested_vmcb;
2813 if (nested_svm_check_permissions(svm))
2816 nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
2820 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2821 skip_emulated_instruction(&svm->vcpu);
2823 nested_svm_vmloadsave(svm->vmcb, nested_vmcb);
2824 nested_svm_unmap(page);
2829 static int vmrun_interception(struct vcpu_svm *svm)
2831 if (nested_svm_check_permissions(svm))
2834 /* Save rip after vmrun instruction */
2835 kvm_rip_write(&svm->vcpu, kvm_rip_read(&svm->vcpu) + 3);
2837 if (!nested_svm_vmrun(svm))
2840 if (!nested_svm_vmrun_msrpm(svm))
2847 svm->vmcb->control.exit_code = SVM_EXIT_ERR;
2848 svm->vmcb->control.exit_code_hi = 0;
2849 svm->vmcb->control.exit_info_1 = 0;
2850 svm->vmcb->control.exit_info_2 = 0;
2852 nested_svm_vmexit(svm);
2857 static int stgi_interception(struct vcpu_svm *svm)
2859 if (nested_svm_check_permissions(svm))
2862 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2863 skip_emulated_instruction(&svm->vcpu);
2864 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
2871 static int clgi_interception(struct vcpu_svm *svm)
2873 if (nested_svm_check_permissions(svm))
2876 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2877 skip_emulated_instruction(&svm->vcpu);
2881 /* After a CLGI no interrupts should come */
2882 if (!kvm_vcpu_apicv_active(&svm->vcpu)) {
2883 svm_clear_vintr(svm);
2884 svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
2885 mark_dirty(svm->vmcb, VMCB_INTR);
2891 static int invlpga_interception(struct vcpu_svm *svm)
2893 struct kvm_vcpu *vcpu = &svm->vcpu;
2895 trace_kvm_invlpga(svm->vmcb->save.rip, kvm_register_read(&svm->vcpu, VCPU_REGS_RCX),
2896 kvm_register_read(&svm->vcpu, VCPU_REGS_RAX));
2898 /* Let's treat INVLPGA the same as INVLPG (can be optimized!) */
2899 kvm_mmu_invlpg(vcpu, kvm_register_read(&svm->vcpu, VCPU_REGS_RAX));
2901 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2902 skip_emulated_instruction(&svm->vcpu);
2906 static int skinit_interception(struct vcpu_svm *svm)
2908 trace_kvm_skinit(svm->vmcb->save.rip, kvm_register_read(&svm->vcpu, VCPU_REGS_RAX));
2910 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2914 static int wbinvd_interception(struct vcpu_svm *svm)
2916 kvm_emulate_wbinvd(&svm->vcpu);
2920 static int xsetbv_interception(struct vcpu_svm *svm)
2922 u64 new_bv = kvm_read_edx_eax(&svm->vcpu);
2923 u32 index = kvm_register_read(&svm->vcpu, VCPU_REGS_RCX);
2925 if (kvm_set_xcr(&svm->vcpu, index, new_bv) == 0) {
2926 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2927 skip_emulated_instruction(&svm->vcpu);
2933 static int task_switch_interception(struct vcpu_svm *svm)
2937 int int_type = svm->vmcb->control.exit_int_info &
2938 SVM_EXITINTINFO_TYPE_MASK;
2939 int int_vec = svm->vmcb->control.exit_int_info & SVM_EVTINJ_VEC_MASK;
2941 svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_TYPE_MASK;
2943 svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_VALID;
2944 bool has_error_code = false;
2947 tss_selector = (u16)svm->vmcb->control.exit_info_1;
2949 if (svm->vmcb->control.exit_info_2 &
2950 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_IRET))
2951 reason = TASK_SWITCH_IRET;
2952 else if (svm->vmcb->control.exit_info_2 &
2953 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_JMP))
2954 reason = TASK_SWITCH_JMP;
2956 reason = TASK_SWITCH_GATE;
2958 reason = TASK_SWITCH_CALL;
2960 if (reason == TASK_SWITCH_GATE) {
2962 case SVM_EXITINTINFO_TYPE_NMI:
2963 svm->vcpu.arch.nmi_injected = false;
2965 case SVM_EXITINTINFO_TYPE_EXEPT:
2966 if (svm->vmcb->control.exit_info_2 &
2967 (1ULL << SVM_EXITINFOSHIFT_TS_HAS_ERROR_CODE)) {
2968 has_error_code = true;
2970 (u32)svm->vmcb->control.exit_info_2;
2972 kvm_clear_exception_queue(&svm->vcpu);
2974 case SVM_EXITINTINFO_TYPE_INTR:
2975 kvm_clear_interrupt_queue(&svm->vcpu);
2982 if (reason != TASK_SWITCH_GATE ||
2983 int_type == SVM_EXITINTINFO_TYPE_SOFT ||
2984 (int_type == SVM_EXITINTINFO_TYPE_EXEPT &&
2985 (int_vec == OF_VECTOR || int_vec == BP_VECTOR)))
2986 skip_emulated_instruction(&svm->vcpu);
2988 if (int_type != SVM_EXITINTINFO_TYPE_SOFT)
2991 if (kvm_task_switch(&svm->vcpu, tss_selector, int_vec, reason,
2992 has_error_code, error_code) == EMULATE_FAIL) {
2993 svm->vcpu.run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
2994 svm->vcpu.run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
2995 svm->vcpu.run->internal.ndata = 0;
3001 static int cpuid_interception(struct vcpu_svm *svm)
3003 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
3004 kvm_emulate_cpuid(&svm->vcpu);
3008 static int iret_interception(struct vcpu_svm *svm)
3010 ++svm->vcpu.stat.nmi_window_exits;
3011 clr_intercept(svm, INTERCEPT_IRET);
3012 svm->vcpu.arch.hflags |= HF_IRET_MASK;
3013 svm->nmi_iret_rip = kvm_rip_read(&svm->vcpu);
3014 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
3018 static int invlpg_interception(struct vcpu_svm *svm)
3020 if (!static_cpu_has(X86_FEATURE_DECODEASSISTS))
3021 return emulate_instruction(&svm->vcpu, 0) == EMULATE_DONE;
3023 kvm_mmu_invlpg(&svm->vcpu, svm->vmcb->control.exit_info_1);
3024 skip_emulated_instruction(&svm->vcpu);
3028 static int emulate_on_interception(struct vcpu_svm *svm)
3030 return emulate_instruction(&svm->vcpu, 0) == EMULATE_DONE;
3033 static int rdpmc_interception(struct vcpu_svm *svm)
3037 if (!static_cpu_has(X86_FEATURE_NRIPS))
3038 return emulate_on_interception(svm);
3040 err = kvm_rdpmc(&svm->vcpu);
3041 kvm_complete_insn_gp(&svm->vcpu, err);
3046 static bool check_selective_cr0_intercepted(struct vcpu_svm *svm,
3049 unsigned long cr0 = svm->vcpu.arch.cr0;
3053 intercept = svm->nested.intercept;
3055 if (!is_guest_mode(&svm->vcpu) ||
3056 (!(intercept & (1ULL << INTERCEPT_SELECTIVE_CR0))))
3059 cr0 &= ~SVM_CR0_SELECTIVE_MASK;
3060 val &= ~SVM_CR0_SELECTIVE_MASK;
3063 svm->vmcb->control.exit_code = SVM_EXIT_CR0_SEL_WRITE;
3064 ret = (nested_svm_exit_handled(svm) == NESTED_EXIT_DONE);
3070 #define CR_VALID (1ULL << 63)
3072 static int cr_interception(struct vcpu_svm *svm)
3078 if (!static_cpu_has(X86_FEATURE_DECODEASSISTS))
3079 return emulate_on_interception(svm);
3081 if (unlikely((svm->vmcb->control.exit_info_1 & CR_VALID) == 0))
3082 return emulate_on_interception(svm);
3084 reg = svm->vmcb->control.exit_info_1 & SVM_EXITINFO_REG_MASK;
3085 if (svm->vmcb->control.exit_code == SVM_EXIT_CR0_SEL_WRITE)
3086 cr = SVM_EXIT_WRITE_CR0 - SVM_EXIT_READ_CR0;
3088 cr = svm->vmcb->control.exit_code - SVM_EXIT_READ_CR0;
3091 if (cr >= 16) { /* mov to cr */
3093 val = kvm_register_read(&svm->vcpu, reg);
3096 if (!check_selective_cr0_intercepted(svm, val))
3097 err = kvm_set_cr0(&svm->vcpu, val);
3103 err = kvm_set_cr3(&svm->vcpu, val);
3106 err = kvm_set_cr4(&svm->vcpu, val);
3109 err = kvm_set_cr8(&svm->vcpu, val);
3112 WARN(1, "unhandled write to CR%d", cr);
3113 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
3116 } else { /* mov from cr */
3119 val = kvm_read_cr0(&svm->vcpu);
3122 val = svm->vcpu.arch.cr2;
3125 val = kvm_read_cr3(&svm->vcpu);
3128 val = kvm_read_cr4(&svm->vcpu);
3131 val = kvm_get_cr8(&svm->vcpu);
3134 WARN(1, "unhandled read from CR%d", cr);
3135 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
3138 kvm_register_write(&svm->vcpu, reg, val);
3140 kvm_complete_insn_gp(&svm->vcpu, err);
3145 static int dr_interception(struct vcpu_svm *svm)
3150 if (svm->vcpu.guest_debug == 0) {
3152 * No more DR vmexits; force a reload of the debug registers
3153 * and reenter on this instruction. The next vmexit will
3154 * retrieve the full state of the debug registers.
3156 clr_dr_intercepts(svm);
3157 svm->vcpu.arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
3161 if (!boot_cpu_has(X86_FEATURE_DECODEASSISTS))
3162 return emulate_on_interception(svm);
3164 reg = svm->vmcb->control.exit_info_1 & SVM_EXITINFO_REG_MASK;
3165 dr = svm->vmcb->control.exit_code - SVM_EXIT_READ_DR0;
3167 if (dr >= 16) { /* mov to DRn */
3168 if (!kvm_require_dr(&svm->vcpu, dr - 16))
3170 val = kvm_register_read(&svm->vcpu, reg);
3171 kvm_set_dr(&svm->vcpu, dr - 16, val);
3173 if (!kvm_require_dr(&svm->vcpu, dr))
3175 kvm_get_dr(&svm->vcpu, dr, &val);
3176 kvm_register_write(&svm->vcpu, reg, val);
3179 skip_emulated_instruction(&svm->vcpu);
3184 static int cr8_write_interception(struct vcpu_svm *svm)
3186 struct kvm_run *kvm_run = svm->vcpu.run;
3189 u8 cr8_prev = kvm_get_cr8(&svm->vcpu);
3190 /* instruction emulation calls kvm_set_cr8() */
3191 r = cr_interception(svm);
3192 if (lapic_in_kernel(&svm->vcpu))
3194 if (cr8_prev <= kvm_get_cr8(&svm->vcpu))
3196 kvm_run->exit_reason = KVM_EXIT_SET_TPR;
3200 static u64 svm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
3202 struct vmcb *vmcb = get_host_vmcb(to_svm(vcpu));
3203 return vmcb->control.tsc_offset + host_tsc;
3206 static int svm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3208 struct vcpu_svm *svm = to_svm(vcpu);
3210 switch (msr_info->index) {
3211 case MSR_IA32_TSC: {
3212 msr_info->data = svm->vmcb->control.tsc_offset +
3213 kvm_scale_tsc(vcpu, rdtsc());
3218 msr_info->data = svm->vmcb->save.star;
3220 #ifdef CONFIG_X86_64
3222 msr_info->data = svm->vmcb->save.lstar;
3225 msr_info->data = svm->vmcb->save.cstar;
3227 case MSR_KERNEL_GS_BASE:
3228 msr_info->data = svm->vmcb->save.kernel_gs_base;
3230 case MSR_SYSCALL_MASK:
3231 msr_info->data = svm->vmcb->save.sfmask;
3234 case MSR_IA32_SYSENTER_CS:
3235 msr_info->data = svm->vmcb->save.sysenter_cs;
3237 case MSR_IA32_SYSENTER_EIP:
3238 msr_info->data = svm->sysenter_eip;
3240 case MSR_IA32_SYSENTER_ESP:
3241 msr_info->data = svm->sysenter_esp;
3244 if (!boot_cpu_has(X86_FEATURE_RDTSCP))
3246 msr_info->data = svm->tsc_aux;
3249 * Nobody will change the following 5 values in the VMCB so we can
3250 * safely return them on rdmsr. They will always be 0 until LBRV is
3253 case MSR_IA32_DEBUGCTLMSR:
3254 msr_info->data = svm->vmcb->save.dbgctl;
3256 case MSR_IA32_LASTBRANCHFROMIP:
3257 msr_info->data = svm->vmcb->save.br_from;
3259 case MSR_IA32_LASTBRANCHTOIP:
3260 msr_info->data = svm->vmcb->save.br_to;
3262 case MSR_IA32_LASTINTFROMIP:
3263 msr_info->data = svm->vmcb->save.last_excp_from;
3265 case MSR_IA32_LASTINTTOIP:
3266 msr_info->data = svm->vmcb->save.last_excp_to;
3268 case MSR_VM_HSAVE_PA:
3269 msr_info->data = svm->nested.hsave_msr;
3272 msr_info->data = svm->nested.vm_cr_msr;
3274 case MSR_IA32_UCODE_REV:
3275 msr_info->data = 0x01000065;
3277 case MSR_F15H_IC_CFG: {
3281 family = guest_cpuid_family(vcpu);
3282 model = guest_cpuid_model(vcpu);
3284 if (family < 0 || model < 0)
3285 return kvm_get_msr_common(vcpu, msr_info);
3289 if (family == 0x15 &&
3290 (model >= 0x2 && model < 0x20))
3291 msr_info->data = 0x1E;
3295 return kvm_get_msr_common(vcpu, msr_info);
3300 static int rdmsr_interception(struct vcpu_svm *svm)
3302 u32 ecx = kvm_register_read(&svm->vcpu, VCPU_REGS_RCX);
3303 struct msr_data msr_info;
3305 msr_info.index = ecx;
3306 msr_info.host_initiated = false;
3307 if (svm_get_msr(&svm->vcpu, &msr_info)) {
3308 trace_kvm_msr_read_ex(ecx);
3309 kvm_inject_gp(&svm->vcpu, 0);
3311 trace_kvm_msr_read(ecx, msr_info.data);
3313 kvm_register_write(&svm->vcpu, VCPU_REGS_RAX,
3314 msr_info.data & 0xffffffff);
3315 kvm_register_write(&svm->vcpu, VCPU_REGS_RDX,
3316 msr_info.data >> 32);
3317 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
3318 skip_emulated_instruction(&svm->vcpu);
3323 static int svm_set_vm_cr(struct kvm_vcpu *vcpu, u64 data)
3325 struct vcpu_svm *svm = to_svm(vcpu);
3326 int svm_dis, chg_mask;
3328 if (data & ~SVM_VM_CR_VALID_MASK)
3331 chg_mask = SVM_VM_CR_VALID_MASK;
3333 if (svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK)
3334 chg_mask &= ~(SVM_VM_CR_SVM_LOCK_MASK | SVM_VM_CR_SVM_DIS_MASK);
3336 svm->nested.vm_cr_msr &= ~chg_mask;
3337 svm->nested.vm_cr_msr |= (data & chg_mask);
3339 svm_dis = svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK;
3341 /* check for svm_disable while efer.svme is set */
3342 if (svm_dis && (vcpu->arch.efer & EFER_SVME))
3348 static int svm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
3350 struct vcpu_svm *svm = to_svm(vcpu);
3352 u32 ecx = msr->index;
3353 u64 data = msr->data;
3356 kvm_write_tsc(vcpu, msr);
3359 svm->vmcb->save.star = data;
3361 #ifdef CONFIG_X86_64
3363 svm->vmcb->save.lstar = data;
3366 svm->vmcb->save.cstar = data;
3368 case MSR_KERNEL_GS_BASE:
3369 svm->vmcb->save.kernel_gs_base = data;
3371 case MSR_SYSCALL_MASK:
3372 svm->vmcb->save.sfmask = data;
3375 case MSR_IA32_SYSENTER_CS:
3376 svm->vmcb->save.sysenter_cs = data;
3378 case MSR_IA32_SYSENTER_EIP:
3379 svm->sysenter_eip = data;
3380 svm->vmcb->save.sysenter_eip = data;
3382 case MSR_IA32_SYSENTER_ESP:
3383 svm->sysenter_esp = data;
3384 svm->vmcb->save.sysenter_esp = data;
3387 if (!boot_cpu_has(X86_FEATURE_RDTSCP))
3391 * This is rare, so we update the MSR here instead of using
3392 * direct_access_msrs. Doing that would require a rdmsr in
3395 svm->tsc_aux = data;
3396 wrmsrl(MSR_TSC_AUX, svm->tsc_aux);
3398 case MSR_IA32_DEBUGCTLMSR:
3399 if (!boot_cpu_has(X86_FEATURE_LBRV)) {
3400 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTL 0x%llx, nop\n",
3404 if (data & DEBUGCTL_RESERVED_BITS)
3407 svm->vmcb->save.dbgctl = data;
3408 mark_dirty(svm->vmcb, VMCB_LBR);
3409 if (data & (1ULL<<0))
3410 svm_enable_lbrv(svm);
3412 svm_disable_lbrv(svm);
3414 case MSR_VM_HSAVE_PA:
3415 svm->nested.hsave_msr = data;
3418 return svm_set_vm_cr(vcpu, data);
3420 vcpu_unimpl(vcpu, "unimplemented wrmsr: 0x%x data 0x%llx\n", ecx, data);
3422 case MSR_IA32_APICBASE:
3423 if (kvm_vcpu_apicv_active(vcpu))
3424 avic_update_vapic_bar(to_svm(vcpu), data);
3425 /* Follow through */
3427 return kvm_set_msr_common(vcpu, msr);
3432 static int wrmsr_interception(struct vcpu_svm *svm)
3434 struct msr_data msr;
3435 u32 ecx = kvm_register_read(&svm->vcpu, VCPU_REGS_RCX);
3436 u64 data = kvm_read_edx_eax(&svm->vcpu);
3440 msr.host_initiated = false;
3442 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
3443 if (kvm_set_msr(&svm->vcpu, &msr)) {
3444 trace_kvm_msr_write_ex(ecx, data);
3445 kvm_inject_gp(&svm->vcpu, 0);
3447 trace_kvm_msr_write(ecx, data);
3448 skip_emulated_instruction(&svm->vcpu);
3453 static int msr_interception(struct vcpu_svm *svm)
3455 if (svm->vmcb->control.exit_info_1)
3456 return wrmsr_interception(svm);
3458 return rdmsr_interception(svm);
3461 static int interrupt_window_interception(struct vcpu_svm *svm)
3463 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
3464 svm_clear_vintr(svm);
3465 svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
3466 mark_dirty(svm->vmcb, VMCB_INTR);
3467 ++svm->vcpu.stat.irq_window_exits;
3471 static int pause_interception(struct vcpu_svm *svm)
3473 kvm_vcpu_on_spin(&(svm->vcpu));
3477 static int nop_interception(struct vcpu_svm *svm)
3479 skip_emulated_instruction(&(svm->vcpu));
3483 static int monitor_interception(struct vcpu_svm *svm)
3485 printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
3486 return nop_interception(svm);
3489 static int mwait_interception(struct vcpu_svm *svm)
3491 printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
3492 return nop_interception(svm);
3495 static int (*const svm_exit_handlers[])(struct vcpu_svm *svm) = {
3496 [SVM_EXIT_READ_CR0] = cr_interception,
3497 [SVM_EXIT_READ_CR3] = cr_interception,
3498 [SVM_EXIT_READ_CR4] = cr_interception,
3499 [SVM_EXIT_READ_CR8] = cr_interception,
3500 [SVM_EXIT_CR0_SEL_WRITE] = cr_interception,
3501 [SVM_EXIT_WRITE_CR0] = cr_interception,
3502 [SVM_EXIT_WRITE_CR3] = cr_interception,
3503 [SVM_EXIT_WRITE_CR4] = cr_interception,
3504 [SVM_EXIT_WRITE_CR8] = cr8_write_interception,
3505 [SVM_EXIT_READ_DR0] = dr_interception,
3506 [SVM_EXIT_READ_DR1] = dr_interception,
3507 [SVM_EXIT_READ_DR2] = dr_interception,
3508 [SVM_EXIT_READ_DR3] = dr_interception,
3509 [SVM_EXIT_READ_DR4] = dr_interception,
3510 [SVM_EXIT_READ_DR5] = dr_interception,
3511 [SVM_EXIT_READ_DR6] = dr_interception,
3512 [SVM_EXIT_READ_DR7] = dr_interception,
3513 [SVM_EXIT_WRITE_DR0] = dr_interception,
3514 [SVM_EXIT_WRITE_DR1] = dr_interception,
3515 [SVM_EXIT_WRITE_DR2] = dr_interception,
3516 [SVM_EXIT_WRITE_DR3] = dr_interception,
3517 [SVM_EXIT_WRITE_DR4] = dr_interception,
3518 [SVM_EXIT_WRITE_DR5] = dr_interception,
3519 [SVM_EXIT_WRITE_DR6] = dr_interception,
3520 [SVM_EXIT_WRITE_DR7] = dr_interception,
3521 [SVM_EXIT_EXCP_BASE + DB_VECTOR] = db_interception,
3522 [SVM_EXIT_EXCP_BASE + BP_VECTOR] = bp_interception,
3523 [SVM_EXIT_EXCP_BASE + UD_VECTOR] = ud_interception,
3524 [SVM_EXIT_EXCP_BASE + PF_VECTOR] = pf_interception,
3525 [SVM_EXIT_EXCP_BASE + NM_VECTOR] = nm_interception,
3526 [SVM_EXIT_EXCP_BASE + MC_VECTOR] = mc_interception,
3527 [SVM_EXIT_EXCP_BASE + AC_VECTOR] = ac_interception,
3528 [SVM_EXIT_INTR] = intr_interception,
3529 [SVM_EXIT_NMI] = nmi_interception,
3530 [SVM_EXIT_SMI] = nop_on_interception,
3531 [SVM_EXIT_INIT] = nop_on_interception,
3532 [SVM_EXIT_VINTR] = interrupt_window_interception,
3533 [SVM_EXIT_RDPMC] = rdpmc_interception,
3534 [SVM_EXIT_CPUID] = cpuid_interception,
3535 [SVM_EXIT_IRET] = iret_interception,
3536 [SVM_EXIT_INVD] = emulate_on_interception,
3537 [SVM_EXIT_PAUSE] = pause_interception,
3538 [SVM_EXIT_HLT] = halt_interception,
3539 [SVM_EXIT_INVLPG] = invlpg_interception,
3540 [SVM_EXIT_INVLPGA] = invlpga_interception,
3541 [SVM_EXIT_IOIO] = io_interception,
3542 [SVM_EXIT_MSR] = msr_interception,
3543 [SVM_EXIT_TASK_SWITCH] = task_switch_interception,
3544 [SVM_EXIT_SHUTDOWN] = shutdown_interception,
3545 [SVM_EXIT_VMRUN] = vmrun_interception,
3546 [SVM_EXIT_VMMCALL] = vmmcall_interception,
3547 [SVM_EXIT_VMLOAD] = vmload_interception,
3548 [SVM_EXIT_VMSAVE] = vmsave_interception,
3549 [SVM_EXIT_STGI] = stgi_interception,
3550 [SVM_EXIT_CLGI] = clgi_interception,
3551 [SVM_EXIT_SKINIT] = skinit_interception,
3552 [SVM_EXIT_WBINVD] = wbinvd_interception,
3553 [SVM_EXIT_MONITOR] = monitor_interception,
3554 [SVM_EXIT_MWAIT] = mwait_interception,
3555 [SVM_EXIT_XSETBV] = xsetbv_interception,
3556 [SVM_EXIT_NPF] = pf_interception,
3557 [SVM_EXIT_RSM] = emulate_on_interception,
3560 static void dump_vmcb(struct kvm_vcpu *vcpu)
3562 struct vcpu_svm *svm = to_svm(vcpu);
3563 struct vmcb_control_area *control = &svm->vmcb->control;
3564 struct vmcb_save_area *save = &svm->vmcb->save;
3566 pr_err("VMCB Control Area:\n");
3567 pr_err("%-20s%04x\n", "cr_read:", control->intercept_cr & 0xffff);
3568 pr_err("%-20s%04x\n", "cr_write:", control->intercept_cr >> 16);
3569 pr_err("%-20s%04x\n", "dr_read:", control->intercept_dr & 0xffff);
3570 pr_err("%-20s%04x\n", "dr_write:", control->intercept_dr >> 16);
3571 pr_err("%-20s%08x\n", "exceptions:", control->intercept_exceptions);
3572 pr_err("%-20s%016llx\n", "intercepts:", control->intercept);
3573 pr_err("%-20s%d\n", "pause filter count:", control->pause_filter_count);
3574 pr_err("%-20s%016llx\n", "iopm_base_pa:", control->iopm_base_pa);
3575 pr_err("%-20s%016llx\n", "msrpm_base_pa:", control->msrpm_base_pa);
3576 pr_err("%-20s%016llx\n", "tsc_offset:", control->tsc_offset);
3577 pr_err("%-20s%d\n", "asid:", control->asid);
3578 pr_err("%-20s%d\n", "tlb_ctl:", control->tlb_ctl);
3579 pr_err("%-20s%08x\n", "int_ctl:", control->int_ctl);
3580 pr_err("%-20s%08x\n", "int_vector:", control->int_vector);
3581 pr_err("%-20s%08x\n", "int_state:", control->int_state);
3582 pr_err("%-20s%08x\n", "exit_code:", control->exit_code);
3583 pr_err("%-20s%016llx\n", "exit_info1:", control->exit_info_1);
3584 pr_err("%-20s%016llx\n", "exit_info2:", control->exit_info_2);
3585 pr_err("%-20s%08x\n", "exit_int_info:", control->exit_int_info);
3586 pr_err("%-20s%08x\n", "exit_int_info_err:", control->exit_int_info_err);
3587 pr_err("%-20s%lld\n", "nested_ctl:", control->nested_ctl);
3588 pr_err("%-20s%016llx\n", "nested_cr3:", control->nested_cr3);
3589 pr_err("%-20s%016llx\n", "avic_vapic_bar:", control->avic_vapic_bar);
3590 pr_err("%-20s%08x\n", "event_inj:", control->event_inj);
3591 pr_err("%-20s%08x\n", "event_inj_err:", control->event_inj_err);
3592 pr_err("%-20s%lld\n", "lbr_ctl:", control->lbr_ctl);
3593 pr_err("%-20s%016llx\n", "next_rip:", control->next_rip);
3594 pr_err("%-20s%016llx\n", "avic_backing_page:", control->avic_backing_page);
3595 pr_err("%-20s%016llx\n", "avic_logical_id:", control->avic_logical_id);
3596 pr_err("%-20s%016llx\n", "avic_physical_id:", control->avic_physical_id);
3597 pr_err("VMCB State Save Area:\n");
3598 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3600 save->es.selector, save->es.attrib,
3601 save->es.limit, save->es.base);
3602 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3604 save->cs.selector, save->cs.attrib,
3605 save->cs.limit, save->cs.base);
3606 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3608 save->ss.selector, save->ss.attrib,
3609 save->ss.limit, save->ss.base);
3610 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3612 save->ds.selector, save->ds.attrib,
3613 save->ds.limit, save->ds.base);
3614 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3616 save->fs.selector, save->fs.attrib,
3617 save->fs.limit, save->fs.base);
3618 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3620 save->gs.selector, save->gs.attrib,
3621 save->gs.limit, save->gs.base);
3622 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3624 save->gdtr.selector, save->gdtr.attrib,
3625 save->gdtr.limit, save->gdtr.base);
3626 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3628 save->ldtr.selector, save->ldtr.attrib,
3629 save->ldtr.limit, save->ldtr.base);
3630 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3632 save->idtr.selector, save->idtr.attrib,
3633 save->idtr.limit, save->idtr.base);
3634 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3636 save->tr.selector, save->tr.attrib,
3637 save->tr.limit, save->tr.base);
3638 pr_err("cpl: %d efer: %016llx\n",
3639 save->cpl, save->efer);
3640 pr_err("%-15s %016llx %-13s %016llx\n",
3641 "cr0:", save->cr0, "cr2:", save->cr2);
3642 pr_err("%-15s %016llx %-13s %016llx\n",
3643 "cr3:", save->cr3, "cr4:", save->cr4);
3644 pr_err("%-15s %016llx %-13s %016llx\n",
3645 "dr6:", save->dr6, "dr7:", save->dr7);
3646 pr_err("%-15s %016llx %-13s %016llx\n",
3647 "rip:", save->rip, "rflags:", save->rflags);
3648 pr_err("%-15s %016llx %-13s %016llx\n",
3649 "rsp:", save->rsp, "rax:", save->rax);
3650 pr_err("%-15s %016llx %-13s %016llx\n",
3651 "star:", save->star, "lstar:", save->lstar);
3652 pr_err("%-15s %016llx %-13s %016llx\n",
3653 "cstar:", save->cstar, "sfmask:", save->sfmask);
3654 pr_err("%-15s %016llx %-13s %016llx\n",
3655 "kernel_gs_base:", save->kernel_gs_base,
3656 "sysenter_cs:", save->sysenter_cs);
3657 pr_err("%-15s %016llx %-13s %016llx\n",
3658 "sysenter_esp:", save->sysenter_esp,
3659 "sysenter_eip:", save->sysenter_eip);
3660 pr_err("%-15s %016llx %-13s %016llx\n",
3661 "gpat:", save->g_pat, "dbgctl:", save->dbgctl);
3662 pr_err("%-15s %016llx %-13s %016llx\n",
3663 "br_from:", save->br_from, "br_to:", save->br_to);
3664 pr_err("%-15s %016llx %-13s %016llx\n",
3665 "excp_from:", save->last_excp_from,
3666 "excp_to:", save->last_excp_to);
3669 static void svm_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
3671 struct vmcb_control_area *control = &to_svm(vcpu)->vmcb->control;
3673 *info1 = control->exit_info_1;
3674 *info2 = control->exit_info_2;
3677 static int handle_exit(struct kvm_vcpu *vcpu)
3679 struct vcpu_svm *svm = to_svm(vcpu);
3680 struct kvm_run *kvm_run = vcpu->run;
3681 u32 exit_code = svm->vmcb->control.exit_code;
3683 trace_kvm_exit(exit_code, vcpu, KVM_ISA_SVM);
3685 if (!is_cr_intercept(svm, INTERCEPT_CR0_WRITE))
3686 vcpu->arch.cr0 = svm->vmcb->save.cr0;
3688 vcpu->arch.cr3 = svm->vmcb->save.cr3;
3690 if (unlikely(svm->nested.exit_required)) {
3691 nested_svm_vmexit(svm);
3692 svm->nested.exit_required = false;
3697 if (is_guest_mode(vcpu)) {
3700 trace_kvm_nested_vmexit(svm->vmcb->save.rip, exit_code,
3701 svm->vmcb->control.exit_info_1,
3702 svm->vmcb->control.exit_info_2,
3703 svm->vmcb->control.exit_int_info,
3704 svm->vmcb->control.exit_int_info_err,
3707 vmexit = nested_svm_exit_special(svm);
3709 if (vmexit == NESTED_EXIT_CONTINUE)
3710 vmexit = nested_svm_exit_handled(svm);
3712 if (vmexit == NESTED_EXIT_DONE)
3716 svm_complete_interrupts(svm);
3718 if (svm->vmcb->control.exit_code == SVM_EXIT_ERR) {
3719 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3720 kvm_run->fail_entry.hardware_entry_failure_reason
3721 = svm->vmcb->control.exit_code;
3722 pr_err("KVM: FAILED VMRUN WITH VMCB:\n");
3727 if (is_external_interrupt(svm->vmcb->control.exit_int_info) &&
3728 exit_code != SVM_EXIT_EXCP_BASE + PF_VECTOR &&
3729 exit_code != SVM_EXIT_NPF && exit_code != SVM_EXIT_TASK_SWITCH &&
3730 exit_code != SVM_EXIT_INTR && exit_code != SVM_EXIT_NMI)
3731 printk(KERN_ERR "%s: unexpected exit_int_info 0x%x "
3733 __func__, svm->vmcb->control.exit_int_info,
3736 if (exit_code >= ARRAY_SIZE(svm_exit_handlers)
3737 || !svm_exit_handlers[exit_code]) {
3738 WARN_ONCE(1, "svm: unexpected exit reason 0x%x\n", exit_code);
3739 kvm_queue_exception(vcpu, UD_VECTOR);
3743 return svm_exit_handlers[exit_code](svm);
3746 static void reload_tss(struct kvm_vcpu *vcpu)
3748 int cpu = raw_smp_processor_id();
3750 struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
3751 sd->tss_desc->type = 9; /* available 32/64-bit TSS */
3755 static void pre_svm_run(struct vcpu_svm *svm)
3757 int cpu = raw_smp_processor_id();
3759 struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
3761 /* FIXME: handle wraparound of asid_generation */
3762 if (svm->asid_generation != sd->asid_generation)
3766 static void svm_inject_nmi(struct kvm_vcpu *vcpu)
3768 struct vcpu_svm *svm = to_svm(vcpu);
3770 svm->vmcb->control.event_inj = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI;
3771 vcpu->arch.hflags |= HF_NMI_MASK;
3772 set_intercept(svm, INTERCEPT_IRET);
3773 ++vcpu->stat.nmi_injections;
3776 static inline void svm_inject_irq(struct vcpu_svm *svm, int irq)
3778 struct vmcb_control_area *control;
3780 /* The following fields are ignored when AVIC is enabled */
3781 control = &svm->vmcb->control;
3782 control->int_vector = irq;
3783 control->int_ctl &= ~V_INTR_PRIO_MASK;
3784 control->int_ctl |= V_IRQ_MASK |
3785 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT);
3786 mark_dirty(svm->vmcb, VMCB_INTR);
3789 static void svm_set_irq(struct kvm_vcpu *vcpu)
3791 struct vcpu_svm *svm = to_svm(vcpu);
3793 BUG_ON(!(gif_set(svm)));
3795 trace_kvm_inj_virq(vcpu->arch.interrupt.nr);
3796 ++vcpu->stat.irq_injections;
3798 svm->vmcb->control.event_inj = vcpu->arch.interrupt.nr |
3799 SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR;
3802 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
3804 struct vcpu_svm *svm = to_svm(vcpu);
3806 if (is_guest_mode(vcpu) && (vcpu->arch.hflags & HF_VINTR_MASK))
3809 clr_cr_intercept(svm, INTERCEPT_CR8_WRITE);
3815 set_cr_intercept(svm, INTERCEPT_CR8_WRITE);
3818 static void svm_set_virtual_x2apic_mode(struct kvm_vcpu *vcpu, bool set)
3823 static bool svm_get_enable_apicv(void)
3828 static void svm_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
3832 static void svm_hwapic_isr_update(struct kvm *kvm, int isr)
3836 /* Note: Currently only used by Hyper-V. */
3837 static void svm_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu)
3839 struct vcpu_svm *svm = to_svm(vcpu);
3840 struct vmcb *vmcb = svm->vmcb;
3845 vmcb->control.int_ctl &= ~AVIC_ENABLE_MASK;
3846 mark_dirty(vmcb, VMCB_INTR);
3849 static void svm_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
3854 static void svm_sync_pir_to_irr(struct kvm_vcpu *vcpu)
3859 static void svm_deliver_avic_intr(struct kvm_vcpu *vcpu, int vec)
3861 kvm_lapic_set_irr(vec, vcpu->arch.apic);
3862 smp_mb__after_atomic();
3864 if (avic_vcpu_is_running(vcpu))
3865 wrmsrl(SVM_AVIC_DOORBELL,
3866 __default_cpu_present_to_apicid(vcpu->cpu));
3868 kvm_vcpu_wake_up(vcpu);
3871 static int svm_nmi_allowed(struct kvm_vcpu *vcpu)
3873 struct vcpu_svm *svm = to_svm(vcpu);
3874 struct vmcb *vmcb = svm->vmcb;
3876 ret = !(vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK) &&
3877 !(svm->vcpu.arch.hflags & HF_NMI_MASK);
3878 ret = ret && gif_set(svm) && nested_svm_nmi(svm);
3883 static bool svm_get_nmi_mask(struct kvm_vcpu *vcpu)
3885 struct vcpu_svm *svm = to_svm(vcpu);
3887 return !!(svm->vcpu.arch.hflags & HF_NMI_MASK);
3890 static void svm_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
3892 struct vcpu_svm *svm = to_svm(vcpu);
3895 svm->vcpu.arch.hflags |= HF_NMI_MASK;
3896 set_intercept(svm, INTERCEPT_IRET);
3898 svm->vcpu.arch.hflags &= ~HF_NMI_MASK;
3899 clr_intercept(svm, INTERCEPT_IRET);
3903 static int svm_interrupt_allowed(struct kvm_vcpu *vcpu)
3905 struct vcpu_svm *svm = to_svm(vcpu);
3906 struct vmcb *vmcb = svm->vmcb;
3909 if (!gif_set(svm) ||
3910 (vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK))
3913 ret = !!(kvm_get_rflags(vcpu) & X86_EFLAGS_IF);
3915 if (is_guest_mode(vcpu))
3916 return ret && !(svm->vcpu.arch.hflags & HF_VINTR_MASK);
3921 static void enable_irq_window(struct kvm_vcpu *vcpu)
3923 struct vcpu_svm *svm = to_svm(vcpu);
3925 if (kvm_vcpu_apicv_active(vcpu))
3929 * In case GIF=0 we can't rely on the CPU to tell us when GIF becomes
3930 * 1, because that's a separate STGI/VMRUN intercept. The next time we
3931 * get that intercept, this function will be called again though and
3932 * we'll get the vintr intercept.
3934 if (gif_set(svm) && nested_svm_intr(svm)) {
3936 svm_inject_irq(svm, 0x0);
3940 static void enable_nmi_window(struct kvm_vcpu *vcpu)
3942 struct vcpu_svm *svm = to_svm(vcpu);
3944 if ((svm->vcpu.arch.hflags & (HF_NMI_MASK | HF_IRET_MASK))
3946 return; /* IRET will cause a vm exit */
3949 * Something prevents NMI from been injected. Single step over possible
3950 * problem (IRET or exception injection or interrupt shadow)
3952 svm->nmi_singlestep = true;
3953 svm->vmcb->save.rflags |= (X86_EFLAGS_TF | X86_EFLAGS_RF);
3956 static int svm_set_tss_addr(struct kvm *kvm, unsigned int addr)
3961 static void svm_flush_tlb(struct kvm_vcpu *vcpu)
3963 struct vcpu_svm *svm = to_svm(vcpu);
3965 if (static_cpu_has(X86_FEATURE_FLUSHBYASID))
3966 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ASID;
3968 svm->asid_generation--;
3971 static void svm_prepare_guest_switch(struct kvm_vcpu *vcpu)
3975 static inline void sync_cr8_to_lapic(struct kvm_vcpu *vcpu)
3977 struct vcpu_svm *svm = to_svm(vcpu);
3979 if (is_guest_mode(vcpu) && (vcpu->arch.hflags & HF_VINTR_MASK))
3982 if (!is_cr_intercept(svm, INTERCEPT_CR8_WRITE)) {
3983 int cr8 = svm->vmcb->control.int_ctl & V_TPR_MASK;
3984 kvm_set_cr8(vcpu, cr8);
3988 static inline void sync_lapic_to_cr8(struct kvm_vcpu *vcpu)
3990 struct vcpu_svm *svm = to_svm(vcpu);
3993 if (is_guest_mode(vcpu) && (vcpu->arch.hflags & HF_VINTR_MASK))
3996 cr8 = kvm_get_cr8(vcpu);
3997 svm->vmcb->control.int_ctl &= ~V_TPR_MASK;
3998 svm->vmcb->control.int_ctl |= cr8 & V_TPR_MASK;
4001 static void svm_complete_interrupts(struct vcpu_svm *svm)
4005 u32 exitintinfo = svm->vmcb->control.exit_int_info;
4006 unsigned int3_injected = svm->int3_injected;
4008 svm->int3_injected = 0;
4011 * If we've made progress since setting HF_IRET_MASK, we've
4012 * executed an IRET and can allow NMI injection.
4014 if ((svm->vcpu.arch.hflags & HF_IRET_MASK)
4015 && kvm_rip_read(&svm->vcpu) != svm->nmi_iret_rip) {
4016 svm->vcpu.arch.hflags &= ~(HF_NMI_MASK | HF_IRET_MASK);
4017 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
4020 svm->vcpu.arch.nmi_injected = false;
4021 kvm_clear_exception_queue(&svm->vcpu);
4022 kvm_clear_interrupt_queue(&svm->vcpu);
4024 if (!(exitintinfo & SVM_EXITINTINFO_VALID))
4027 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
4029 vector = exitintinfo & SVM_EXITINTINFO_VEC_MASK;
4030 type = exitintinfo & SVM_EXITINTINFO_TYPE_MASK;
4033 case SVM_EXITINTINFO_TYPE_NMI:
4034 svm->vcpu.arch.nmi_injected = true;
4036 case SVM_EXITINTINFO_TYPE_EXEPT:
4038 * In case of software exceptions, do not reinject the vector,
4039 * but re-execute the instruction instead. Rewind RIP first
4040 * if we emulated INT3 before.
4042 if (kvm_exception_is_soft(vector)) {
4043 if (vector == BP_VECTOR && int3_injected &&
4044 kvm_is_linear_rip(&svm->vcpu, svm->int3_rip))
4045 kvm_rip_write(&svm->vcpu,
4046 kvm_rip_read(&svm->vcpu) -
4050 if (exitintinfo & SVM_EXITINTINFO_VALID_ERR) {
4051 u32 err = svm->vmcb->control.exit_int_info_err;
4052 kvm_requeue_exception_e(&svm->vcpu, vector, err);
4055 kvm_requeue_exception(&svm->vcpu, vector);
4057 case SVM_EXITINTINFO_TYPE_INTR:
4058 kvm_queue_interrupt(&svm->vcpu, vector, false);
4065 static void svm_cancel_injection(struct kvm_vcpu *vcpu)
4067 struct vcpu_svm *svm = to_svm(vcpu);
4068 struct vmcb_control_area *control = &svm->vmcb->control;
4070 control->exit_int_info = control->event_inj;
4071 control->exit_int_info_err = control->event_inj_err;
4072 control->event_inj = 0;
4073 svm_complete_interrupts(svm);
4076 static void svm_vcpu_run(struct kvm_vcpu *vcpu)
4078 struct vcpu_svm *svm = to_svm(vcpu);
4080 svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX];
4081 svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP];
4082 svm->vmcb->save.rip = vcpu->arch.regs[VCPU_REGS_RIP];
4085 * A vmexit emulation is required before the vcpu can be executed
4088 if (unlikely(svm->nested.exit_required))
4093 sync_lapic_to_cr8(vcpu);
4095 svm->vmcb->save.cr2 = vcpu->arch.cr2;
4102 "push %%" _ASM_BP "; \n\t"
4103 "mov %c[rbx](%[svm]), %%" _ASM_BX " \n\t"
4104 "mov %c[rcx](%[svm]), %%" _ASM_CX " \n\t"
4105 "mov %c[rdx](%[svm]), %%" _ASM_DX " \n\t"
4106 "mov %c[rsi](%[svm]), %%" _ASM_SI " \n\t"
4107 "mov %c[rdi](%[svm]), %%" _ASM_DI " \n\t"
4108 "mov %c[rbp](%[svm]), %%" _ASM_BP " \n\t"
4109 #ifdef CONFIG_X86_64
4110 "mov %c[r8](%[svm]), %%r8 \n\t"
4111 "mov %c[r9](%[svm]), %%r9 \n\t"
4112 "mov %c[r10](%[svm]), %%r10 \n\t"
4113 "mov %c[r11](%[svm]), %%r11 \n\t"
4114 "mov %c[r12](%[svm]), %%r12 \n\t"
4115 "mov %c[r13](%[svm]), %%r13 \n\t"
4116 "mov %c[r14](%[svm]), %%r14 \n\t"
4117 "mov %c[r15](%[svm]), %%r15 \n\t"
4120 /* Enter guest mode */
4121 "push %%" _ASM_AX " \n\t"
4122 "mov %c[vmcb](%[svm]), %%" _ASM_AX " \n\t"
4123 __ex(SVM_VMLOAD) "\n\t"
4124 __ex(SVM_VMRUN) "\n\t"
4125 __ex(SVM_VMSAVE) "\n\t"
4126 "pop %%" _ASM_AX " \n\t"
4128 /* Save guest registers, load host registers */
4129 "mov %%" _ASM_BX ", %c[rbx](%[svm]) \n\t"
4130 "mov %%" _ASM_CX ", %c[rcx](%[svm]) \n\t"
4131 "mov %%" _ASM_DX ", %c[rdx](%[svm]) \n\t"
4132 "mov %%" _ASM_SI ", %c[rsi](%[svm]) \n\t"
4133 "mov %%" _ASM_DI ", %c[rdi](%[svm]) \n\t"
4134 "mov %%" _ASM_BP ", %c[rbp](%[svm]) \n\t"
4135 #ifdef CONFIG_X86_64
4136 "mov %%r8, %c[r8](%[svm]) \n\t"
4137 "mov %%r9, %c[r9](%[svm]) \n\t"
4138 "mov %%r10, %c[r10](%[svm]) \n\t"
4139 "mov %%r11, %c[r11](%[svm]) \n\t"
4140 "mov %%r12, %c[r12](%[svm]) \n\t"
4141 "mov %%r13, %c[r13](%[svm]) \n\t"
4142 "mov %%r14, %c[r14](%[svm]) \n\t"
4143 "mov %%r15, %c[r15](%[svm]) \n\t"
4148 [vmcb]"i"(offsetof(struct vcpu_svm, vmcb_pa)),
4149 [rbx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBX])),
4150 [rcx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RCX])),
4151 [rdx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDX])),
4152 [rsi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RSI])),
4153 [rdi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDI])),
4154 [rbp]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBP]))
4155 #ifdef CONFIG_X86_64
4156 , [r8]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R8])),
4157 [r9]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R9])),
4158 [r10]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R10])),
4159 [r11]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R11])),
4160 [r12]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R12])),
4161 [r13]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R13])),
4162 [r14]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R14])),
4163 [r15]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R15]))
4166 #ifdef CONFIG_X86_64
4167 , "rbx", "rcx", "rdx", "rsi", "rdi"
4168 , "r8", "r9", "r10", "r11" , "r12", "r13", "r14", "r15"
4170 , "ebx", "ecx", "edx", "esi", "edi"
4174 #ifdef CONFIG_X86_64
4175 wrmsrl(MSR_GS_BASE, svm->host.gs_base);
4177 loadsegment(fs, svm->host.fs);
4178 #ifndef CONFIG_X86_32_LAZY_GS
4179 loadsegment(gs, svm->host.gs);
4185 local_irq_disable();
4187 vcpu->arch.cr2 = svm->vmcb->save.cr2;
4188 vcpu->arch.regs[VCPU_REGS_RAX] = svm->vmcb->save.rax;
4189 vcpu->arch.regs[VCPU_REGS_RSP] = svm->vmcb->save.rsp;
4190 vcpu->arch.regs[VCPU_REGS_RIP] = svm->vmcb->save.rip;
4192 if (unlikely(svm->vmcb->control.exit_code == SVM_EXIT_NMI))
4193 kvm_before_handle_nmi(&svm->vcpu);
4197 /* Any pending NMI will happen here */
4199 if (unlikely(svm->vmcb->control.exit_code == SVM_EXIT_NMI))
4200 kvm_after_handle_nmi(&svm->vcpu);
4202 sync_cr8_to_lapic(vcpu);
4206 svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
4208 /* if exit due to PF check for async PF */
4209 if (svm->vmcb->control.exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR)
4210 svm->apf_reason = kvm_read_and_reset_pf_reason();
4213 vcpu->arch.regs_avail &= ~(1 << VCPU_EXREG_PDPTR);
4214 vcpu->arch.regs_dirty &= ~(1 << VCPU_EXREG_PDPTR);
4218 * We need to handle MC intercepts here before the vcpu has a chance to
4219 * change the physical cpu
4221 if (unlikely(svm->vmcb->control.exit_code ==
4222 SVM_EXIT_EXCP_BASE + MC_VECTOR))
4223 svm_handle_mce(svm);
4225 mark_all_clean(svm->vmcb);
4228 static void svm_set_cr3(struct kvm_vcpu *vcpu, unsigned long root)
4230 struct vcpu_svm *svm = to_svm(vcpu);
4232 svm->vmcb->save.cr3 = root;
4233 mark_dirty(svm->vmcb, VMCB_CR);
4234 svm_flush_tlb(vcpu);
4237 static void set_tdp_cr3(struct kvm_vcpu *vcpu, unsigned long root)
4239 struct vcpu_svm *svm = to_svm(vcpu);
4241 svm->vmcb->control.nested_cr3 = root;
4242 mark_dirty(svm->vmcb, VMCB_NPT);
4244 /* Also sync guest cr3 here in case we live migrate */
4245 svm->vmcb->save.cr3 = kvm_read_cr3(vcpu);
4246 mark_dirty(svm->vmcb, VMCB_CR);
4248 svm_flush_tlb(vcpu);
4251 static int is_disabled(void)
4255 rdmsrl(MSR_VM_CR, vm_cr);
4256 if (vm_cr & (1 << SVM_VM_CR_SVM_DISABLE))
4263 svm_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
4266 * Patch in the VMMCALL instruction:
4268 hypercall[0] = 0x0f;
4269 hypercall[1] = 0x01;
4270 hypercall[2] = 0xd9;
4273 static void svm_check_processor_compat(void *rtn)
4278 static bool svm_cpu_has_accelerated_tpr(void)
4283 static bool svm_has_high_real_mode_segbase(void)
4288 static u64 svm_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
4293 static void svm_cpuid_update(struct kvm_vcpu *vcpu)
4295 struct vcpu_svm *svm = to_svm(vcpu);
4297 /* Update nrips enabled cache */
4298 svm->nrips_enabled = !!guest_cpuid_has_nrips(&svm->vcpu);
4301 static void svm_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
4306 entry->ecx |= (1 << 2); /* Set SVM bit */
4309 entry->eax = 1; /* SVM revision 1 */
4310 entry->ebx = 8; /* Lets support 8 ASIDs in case we add proper
4311 ASID emulation to nested SVM */
4312 entry->ecx = 0; /* Reserved */
4313 entry->edx = 0; /* Per default do not support any
4314 additional features */
4316 /* Support next_rip if host supports it */
4317 if (boot_cpu_has(X86_FEATURE_NRIPS))
4318 entry->edx |= SVM_FEATURE_NRIP;
4320 /* Support NPT for the guest if enabled */
4322 entry->edx |= SVM_FEATURE_NPT;
4328 static int svm_get_lpage_level(void)
4330 return PT_PDPE_LEVEL;
4333 static bool svm_rdtscp_supported(void)
4335 return boot_cpu_has(X86_FEATURE_RDTSCP);
4338 static bool svm_invpcid_supported(void)
4343 static bool svm_mpx_supported(void)
4348 static bool svm_xsaves_supported(void)
4353 static bool svm_has_wbinvd_exit(void)
4358 static void svm_fpu_deactivate(struct kvm_vcpu *vcpu)
4360 struct vcpu_svm *svm = to_svm(vcpu);
4362 set_exception_intercept(svm, NM_VECTOR);
4363 update_cr0_intercept(svm);
4366 #define PRE_EX(exit) { .exit_code = (exit), \
4367 .stage = X86_ICPT_PRE_EXCEPT, }
4368 #define POST_EX(exit) { .exit_code = (exit), \
4369 .stage = X86_ICPT_POST_EXCEPT, }
4370 #define POST_MEM(exit) { .exit_code = (exit), \
4371 .stage = X86_ICPT_POST_MEMACCESS, }
4373 static const struct __x86_intercept {
4375 enum x86_intercept_stage stage;
4376 } x86_intercept_map[] = {
4377 [x86_intercept_cr_read] = POST_EX(SVM_EXIT_READ_CR0),
4378 [x86_intercept_cr_write] = POST_EX(SVM_EXIT_WRITE_CR0),
4379 [x86_intercept_clts] = POST_EX(SVM_EXIT_WRITE_CR0),
4380 [x86_intercept_lmsw] = POST_EX(SVM_EXIT_WRITE_CR0),
4381 [x86_intercept_smsw] = POST_EX(SVM_EXIT_READ_CR0),
4382 [x86_intercept_dr_read] = POST_EX(SVM_EXIT_READ_DR0),
4383 [x86_intercept_dr_write] = POST_EX(SVM_EXIT_WRITE_DR0),
4384 [x86_intercept_sldt] = POST_EX(SVM_EXIT_LDTR_READ),
4385 [x86_intercept_str] = POST_EX(SVM_EXIT_TR_READ),
4386 [x86_intercept_lldt] = POST_EX(SVM_EXIT_LDTR_WRITE),
4387 [x86_intercept_ltr] = POST_EX(SVM_EXIT_TR_WRITE),
4388 [x86_intercept_sgdt] = POST_EX(SVM_EXIT_GDTR_READ),
4389 [x86_intercept_sidt] = POST_EX(SVM_EXIT_IDTR_READ),
4390 [x86_intercept_lgdt] = POST_EX(SVM_EXIT_GDTR_WRITE),
4391 [x86_intercept_lidt] = POST_EX(SVM_EXIT_IDTR_WRITE),
4392 [x86_intercept_vmrun] = POST_EX(SVM_EXIT_VMRUN),
4393 [x86_intercept_vmmcall] = POST_EX(SVM_EXIT_VMMCALL),
4394 [x86_intercept_vmload] = POST_EX(SVM_EXIT_VMLOAD),
4395 [x86_intercept_vmsave] = POST_EX(SVM_EXIT_VMSAVE),
4396 [x86_intercept_stgi] = POST_EX(SVM_EXIT_STGI),
4397 [x86_intercept_clgi] = POST_EX(SVM_EXIT_CLGI),
4398 [x86_intercept_skinit] = POST_EX(SVM_EXIT_SKINIT),
4399 [x86_intercept_invlpga] = POST_EX(SVM_EXIT_INVLPGA),
4400 [x86_intercept_rdtscp] = POST_EX(SVM_EXIT_RDTSCP),
4401 [x86_intercept_monitor] = POST_MEM(SVM_EXIT_MONITOR),
4402 [x86_intercept_mwait] = POST_EX(SVM_EXIT_MWAIT),
4403 [x86_intercept_invlpg] = POST_EX(SVM_EXIT_INVLPG),
4404 [x86_intercept_invd] = POST_EX(SVM_EXIT_INVD),
4405 [x86_intercept_wbinvd] = POST_EX(SVM_EXIT_WBINVD),
4406 [x86_intercept_wrmsr] = POST_EX(SVM_EXIT_MSR),
4407 [x86_intercept_rdtsc] = POST_EX(SVM_EXIT_RDTSC),
4408 [x86_intercept_rdmsr] = POST_EX(SVM_EXIT_MSR),
4409 [x86_intercept_rdpmc] = POST_EX(SVM_EXIT_RDPMC),
4410 [x86_intercept_cpuid] = PRE_EX(SVM_EXIT_CPUID),
4411 [x86_intercept_rsm] = PRE_EX(SVM_EXIT_RSM),
4412 [x86_intercept_pause] = PRE_EX(SVM_EXIT_PAUSE),
4413 [x86_intercept_pushf] = PRE_EX(SVM_EXIT_PUSHF),
4414 [x86_intercept_popf] = PRE_EX(SVM_EXIT_POPF),
4415 [x86_intercept_intn] = PRE_EX(SVM_EXIT_SWINT),
4416 [x86_intercept_iret] = PRE_EX(SVM_EXIT_IRET),
4417 [x86_intercept_icebp] = PRE_EX(SVM_EXIT_ICEBP),
4418 [x86_intercept_hlt] = POST_EX(SVM_EXIT_HLT),
4419 [x86_intercept_in] = POST_EX(SVM_EXIT_IOIO),
4420 [x86_intercept_ins] = POST_EX(SVM_EXIT_IOIO),
4421 [x86_intercept_out] = POST_EX(SVM_EXIT_IOIO),
4422 [x86_intercept_outs] = POST_EX(SVM_EXIT_IOIO),
4429 static int svm_check_intercept(struct kvm_vcpu *vcpu,
4430 struct x86_instruction_info *info,
4431 enum x86_intercept_stage stage)
4433 struct vcpu_svm *svm = to_svm(vcpu);
4434 int vmexit, ret = X86EMUL_CONTINUE;
4435 struct __x86_intercept icpt_info;
4436 struct vmcb *vmcb = svm->vmcb;
4438 if (info->intercept >= ARRAY_SIZE(x86_intercept_map))
4441 icpt_info = x86_intercept_map[info->intercept];
4443 if (stage != icpt_info.stage)
4446 switch (icpt_info.exit_code) {
4447 case SVM_EXIT_READ_CR0:
4448 if (info->intercept == x86_intercept_cr_read)
4449 icpt_info.exit_code += info->modrm_reg;
4451 case SVM_EXIT_WRITE_CR0: {
4452 unsigned long cr0, val;
4455 if (info->intercept == x86_intercept_cr_write)
4456 icpt_info.exit_code += info->modrm_reg;
4458 if (icpt_info.exit_code != SVM_EXIT_WRITE_CR0 ||
4459 info->intercept == x86_intercept_clts)
4462 intercept = svm->nested.intercept;
4464 if (!(intercept & (1ULL << INTERCEPT_SELECTIVE_CR0)))
4467 cr0 = vcpu->arch.cr0 & ~SVM_CR0_SELECTIVE_MASK;
4468 val = info->src_val & ~SVM_CR0_SELECTIVE_MASK;
4470 if (info->intercept == x86_intercept_lmsw) {
4473 /* lmsw can't clear PE - catch this here */
4474 if (cr0 & X86_CR0_PE)
4479 icpt_info.exit_code = SVM_EXIT_CR0_SEL_WRITE;
4483 case SVM_EXIT_READ_DR0:
4484 case SVM_EXIT_WRITE_DR0:
4485 icpt_info.exit_code += info->modrm_reg;
4488 if (info->intercept == x86_intercept_wrmsr)
4489 vmcb->control.exit_info_1 = 1;
4491 vmcb->control.exit_info_1 = 0;
4493 case SVM_EXIT_PAUSE:
4495 * We get this for NOP only, but pause
4496 * is rep not, check this here
4498 if (info->rep_prefix != REPE_PREFIX)
4500 case SVM_EXIT_IOIO: {
4504 if (info->intercept == x86_intercept_in ||
4505 info->intercept == x86_intercept_ins) {
4506 exit_info = ((info->src_val & 0xffff) << 16) |
4508 bytes = info->dst_bytes;
4510 exit_info = (info->dst_val & 0xffff) << 16;
4511 bytes = info->src_bytes;
4514 if (info->intercept == x86_intercept_outs ||
4515 info->intercept == x86_intercept_ins)
4516 exit_info |= SVM_IOIO_STR_MASK;
4518 if (info->rep_prefix)
4519 exit_info |= SVM_IOIO_REP_MASK;
4521 bytes = min(bytes, 4u);
4523 exit_info |= bytes << SVM_IOIO_SIZE_SHIFT;
4525 exit_info |= (u32)info->ad_bytes << (SVM_IOIO_ASIZE_SHIFT - 1);
4527 vmcb->control.exit_info_1 = exit_info;
4528 vmcb->control.exit_info_2 = info->next_rip;
4536 /* TODO: Advertise NRIPS to guest hypervisor unconditionally */
4537 if (static_cpu_has(X86_FEATURE_NRIPS))
4538 vmcb->control.next_rip = info->next_rip;
4539 vmcb->control.exit_code = icpt_info.exit_code;
4540 vmexit = nested_svm_exit_handled(svm);
4542 ret = (vmexit == NESTED_EXIT_DONE) ? X86EMUL_INTERCEPTED
4549 static void svm_handle_external_intr(struct kvm_vcpu *vcpu)
4554 static void svm_sched_in(struct kvm_vcpu *vcpu, int cpu)
4558 static struct kvm_x86_ops svm_x86_ops = {
4559 .cpu_has_kvm_support = has_svm,
4560 .disabled_by_bios = is_disabled,
4561 .hardware_setup = svm_hardware_setup,
4562 .hardware_unsetup = svm_hardware_unsetup,
4563 .check_processor_compatibility = svm_check_processor_compat,
4564 .hardware_enable = svm_hardware_enable,
4565 .hardware_disable = svm_hardware_disable,
4566 .cpu_has_accelerated_tpr = svm_cpu_has_accelerated_tpr,
4567 .cpu_has_high_real_mode_segbase = svm_has_high_real_mode_segbase,
4569 .vcpu_create = svm_create_vcpu,
4570 .vcpu_free = svm_free_vcpu,
4571 .vcpu_reset = svm_vcpu_reset,
4573 .vm_init = avic_vm_init,
4574 .vm_destroy = avic_vm_destroy,
4576 .prepare_guest_switch = svm_prepare_guest_switch,
4577 .vcpu_load = svm_vcpu_load,
4578 .vcpu_put = svm_vcpu_put,
4580 .update_bp_intercept = update_bp_intercept,
4581 .get_msr = svm_get_msr,
4582 .set_msr = svm_set_msr,
4583 .get_segment_base = svm_get_segment_base,
4584 .get_segment = svm_get_segment,
4585 .set_segment = svm_set_segment,
4586 .get_cpl = svm_get_cpl,
4587 .get_cs_db_l_bits = kvm_get_cs_db_l_bits,
4588 .decache_cr0_guest_bits = svm_decache_cr0_guest_bits,
4589 .decache_cr3 = svm_decache_cr3,
4590 .decache_cr4_guest_bits = svm_decache_cr4_guest_bits,
4591 .set_cr0 = svm_set_cr0,
4592 .set_cr3 = svm_set_cr3,
4593 .set_cr4 = svm_set_cr4,
4594 .set_efer = svm_set_efer,
4595 .get_idt = svm_get_idt,
4596 .set_idt = svm_set_idt,
4597 .get_gdt = svm_get_gdt,
4598 .set_gdt = svm_set_gdt,
4599 .get_dr6 = svm_get_dr6,
4600 .set_dr6 = svm_set_dr6,
4601 .set_dr7 = svm_set_dr7,
4602 .sync_dirty_debug_regs = svm_sync_dirty_debug_regs,
4603 .cache_reg = svm_cache_reg,
4604 .get_rflags = svm_get_rflags,
4605 .set_rflags = svm_set_rflags,
4607 .get_pkru = svm_get_pkru,
4609 .fpu_activate = svm_fpu_activate,
4610 .fpu_deactivate = svm_fpu_deactivate,
4612 .tlb_flush = svm_flush_tlb,
4614 .run = svm_vcpu_run,
4615 .handle_exit = handle_exit,
4616 .skip_emulated_instruction = skip_emulated_instruction,
4617 .set_interrupt_shadow = svm_set_interrupt_shadow,
4618 .get_interrupt_shadow = svm_get_interrupt_shadow,
4619 .patch_hypercall = svm_patch_hypercall,
4620 .set_irq = svm_set_irq,
4621 .set_nmi = svm_inject_nmi,
4622 .queue_exception = svm_queue_exception,
4623 .cancel_injection = svm_cancel_injection,
4624 .interrupt_allowed = svm_interrupt_allowed,
4625 .nmi_allowed = svm_nmi_allowed,
4626 .get_nmi_mask = svm_get_nmi_mask,
4627 .set_nmi_mask = svm_set_nmi_mask,
4628 .enable_nmi_window = enable_nmi_window,
4629 .enable_irq_window = enable_irq_window,
4630 .update_cr8_intercept = update_cr8_intercept,
4631 .set_virtual_x2apic_mode = svm_set_virtual_x2apic_mode,
4632 .get_enable_apicv = svm_get_enable_apicv,
4633 .refresh_apicv_exec_ctrl = svm_refresh_apicv_exec_ctrl,
4634 .load_eoi_exitmap = svm_load_eoi_exitmap,
4635 .sync_pir_to_irr = svm_sync_pir_to_irr,
4636 .hwapic_irr_update = svm_hwapic_irr_update,
4637 .hwapic_isr_update = svm_hwapic_isr_update,
4639 .set_tss_addr = svm_set_tss_addr,
4640 .get_tdp_level = get_npt_level,
4641 .get_mt_mask = svm_get_mt_mask,
4643 .get_exit_info = svm_get_exit_info,
4645 .get_lpage_level = svm_get_lpage_level,
4647 .cpuid_update = svm_cpuid_update,
4649 .rdtscp_supported = svm_rdtscp_supported,
4650 .invpcid_supported = svm_invpcid_supported,
4651 .mpx_supported = svm_mpx_supported,
4652 .xsaves_supported = svm_xsaves_supported,
4654 .set_supported_cpuid = svm_set_supported_cpuid,
4656 .has_wbinvd_exit = svm_has_wbinvd_exit,
4658 .read_tsc_offset = svm_read_tsc_offset,
4659 .write_tsc_offset = svm_write_tsc_offset,
4660 .adjust_tsc_offset_guest = svm_adjust_tsc_offset_guest,
4661 .read_l1_tsc = svm_read_l1_tsc,
4663 .set_tdp_cr3 = set_tdp_cr3,
4665 .check_intercept = svm_check_intercept,
4666 .handle_external_intr = svm_handle_external_intr,
4668 .sched_in = svm_sched_in,
4670 .pmu_ops = &amd_pmu_ops,
4671 .deliver_posted_interrupt = svm_deliver_avic_intr,
4674 static int __init svm_init(void)
4676 return kvm_init(&svm_x86_ops, sizeof(struct vcpu_svm),
4677 __alignof__(struct vcpu_svm), THIS_MODULE);
4680 static void __exit svm_exit(void)
4685 module_init(svm_init)
4686 module_exit(svm_exit)