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.
17 #include <linux/kvm_host.h>
21 #include "kvm_cache_regs.h"
25 #include <linux/module.h>
26 #include <linux/mod_devicetable.h>
27 #include <linux/kernel.h>
28 #include <linux/vmalloc.h>
29 #include <linux/highmem.h>
30 #include <linux/sched.h>
31 #include <linux/ftrace_event.h>
32 #include <linux/slab.h>
34 #include <asm/perf_event.h>
35 #include <asm/tlbflush.h>
37 #include <asm/kvm_para.h>
39 #include <asm/virtext.h>
42 #define __ex(x) __kvm_handle_fault_on_reboot(x)
44 MODULE_AUTHOR("Qumranet");
45 MODULE_LICENSE("GPL");
47 static const struct x86_cpu_id svm_cpu_id[] = {
48 X86_FEATURE_MATCH(X86_FEATURE_SVM),
51 MODULE_DEVICE_TABLE(x86cpu, svm_cpu_id);
53 #define IOPM_ALLOC_ORDER 2
54 #define MSRPM_ALLOC_ORDER 1
56 #define SEG_TYPE_LDT 2
57 #define SEG_TYPE_BUSY_TSS16 3
59 #define SVM_FEATURE_NPT (1 << 0)
60 #define SVM_FEATURE_LBRV (1 << 1)
61 #define SVM_FEATURE_SVML (1 << 2)
62 #define SVM_FEATURE_NRIP (1 << 3)
63 #define SVM_FEATURE_TSC_RATE (1 << 4)
64 #define SVM_FEATURE_VMCB_CLEAN (1 << 5)
65 #define SVM_FEATURE_FLUSH_ASID (1 << 6)
66 #define SVM_FEATURE_DECODE_ASSIST (1 << 7)
67 #define SVM_FEATURE_PAUSE_FILTER (1 << 10)
69 #define NESTED_EXIT_HOST 0 /* Exit handled on host level */
70 #define NESTED_EXIT_DONE 1 /* Exit caused nested vmexit */
71 #define NESTED_EXIT_CONTINUE 2 /* Further checks needed */
73 #define DEBUGCTL_RESERVED_BITS (~(0x3fULL))
75 #define TSC_RATIO_RSVD 0xffffff0000000000ULL
76 #define TSC_RATIO_MIN 0x0000000000000001ULL
77 #define TSC_RATIO_MAX 0x000000ffffffffffULL
79 static bool erratum_383_found __read_mostly;
81 static const u32 host_save_user_msrs[] = {
83 MSR_STAR, MSR_LSTAR, MSR_CSTAR, MSR_SYSCALL_MASK, MSR_KERNEL_GS_BASE,
86 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
89 #define NR_HOST_SAVE_USER_MSRS ARRAY_SIZE(host_save_user_msrs)
99 /* These are the merged vectors */
102 /* gpa pointers to the real vectors */
106 /* A VMEXIT is required but not yet emulated */
109 /* cache for intercepts of the guest */
112 u32 intercept_exceptions;
115 /* Nested Paging related state */
119 #define MSRPM_OFFSETS 16
120 static u32 msrpm_offsets[MSRPM_OFFSETS] __read_mostly;
123 * Set osvw_len to higher value when updated Revision Guides
124 * are published and we know what the new status bits are
126 static uint64_t osvw_len = 4, osvw_status;
129 struct kvm_vcpu vcpu;
131 unsigned long vmcb_pa;
132 struct svm_cpu_data *svm_data;
133 uint64_t asid_generation;
134 uint64_t sysenter_esp;
135 uint64_t sysenter_eip;
139 u64 host_user_msrs[NR_HOST_SAVE_USER_MSRS];
151 struct nested_state nested;
155 unsigned int3_injected;
156 unsigned long int3_rip;
162 static DEFINE_PER_CPU(u64, current_tsc_ratio);
163 #define TSC_RATIO_DEFAULT 0x0100000000ULL
165 #define MSR_INVALID 0xffffffffU
167 static const struct svm_direct_access_msrs {
168 u32 index; /* Index of the MSR */
169 bool always; /* True if intercept is always on */
170 } direct_access_msrs[] = {
171 { .index = MSR_STAR, .always = true },
172 { .index = MSR_IA32_SYSENTER_CS, .always = true },
174 { .index = MSR_GS_BASE, .always = true },
175 { .index = MSR_FS_BASE, .always = true },
176 { .index = MSR_KERNEL_GS_BASE, .always = true },
177 { .index = MSR_LSTAR, .always = true },
178 { .index = MSR_CSTAR, .always = true },
179 { .index = MSR_SYSCALL_MASK, .always = true },
181 { .index = MSR_IA32_LASTBRANCHFROMIP, .always = false },
182 { .index = MSR_IA32_LASTBRANCHTOIP, .always = false },
183 { .index = MSR_IA32_LASTINTFROMIP, .always = false },
184 { .index = MSR_IA32_LASTINTTOIP, .always = false },
185 { .index = MSR_INVALID, .always = false },
188 /* enable NPT for AMD64 and X86 with PAE */
189 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
190 static bool npt_enabled = true;
192 static bool npt_enabled;
195 /* allow nested paging (virtualized MMU) for all guests */
196 static int npt = true;
197 module_param(npt, int, S_IRUGO);
199 /* allow nested virtualization in KVM/SVM */
200 static int nested = true;
201 module_param(nested, int, S_IRUGO);
203 static void svm_flush_tlb(struct kvm_vcpu *vcpu);
204 static void svm_complete_interrupts(struct vcpu_svm *svm);
206 static int nested_svm_exit_handled(struct vcpu_svm *svm);
207 static int nested_svm_intercept(struct vcpu_svm *svm);
208 static int nested_svm_vmexit(struct vcpu_svm *svm);
209 static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
210 bool has_error_code, u32 error_code);
211 static u64 __scale_tsc(u64 ratio, u64 tsc);
214 VMCB_INTERCEPTS, /* Intercept vectors, TSC offset,
215 pause filter count */
216 VMCB_PERM_MAP, /* IOPM Base and MSRPM Base */
217 VMCB_ASID, /* ASID */
218 VMCB_INTR, /* int_ctl, int_vector */
219 VMCB_NPT, /* npt_en, nCR3, gPAT */
220 VMCB_CR, /* CR0, CR3, CR4, EFER */
221 VMCB_DR, /* DR6, DR7 */
222 VMCB_DT, /* GDT, IDT */
223 VMCB_SEG, /* CS, DS, SS, ES, CPL */
224 VMCB_CR2, /* CR2 only */
225 VMCB_LBR, /* DBGCTL, BR_FROM, BR_TO, LAST_EX_FROM, LAST_EX_TO */
229 /* TPR and CR2 are always written before VMRUN */
230 #define VMCB_ALWAYS_DIRTY_MASK ((1U << VMCB_INTR) | (1U << VMCB_CR2))
232 static inline void mark_all_dirty(struct vmcb *vmcb)
234 vmcb->control.clean = 0;
237 static inline void mark_all_clean(struct vmcb *vmcb)
239 vmcb->control.clean = ((1 << VMCB_DIRTY_MAX) - 1)
240 & ~VMCB_ALWAYS_DIRTY_MASK;
243 static inline void mark_dirty(struct vmcb *vmcb, int bit)
245 vmcb->control.clean &= ~(1 << bit);
248 static inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu)
250 return container_of(vcpu, struct vcpu_svm, vcpu);
253 static void recalc_intercepts(struct vcpu_svm *svm)
255 struct vmcb_control_area *c, *h;
256 struct nested_state *g;
258 mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
260 if (!is_guest_mode(&svm->vcpu))
263 c = &svm->vmcb->control;
264 h = &svm->nested.hsave->control;
267 c->intercept_cr = h->intercept_cr | g->intercept_cr;
268 c->intercept_dr = h->intercept_dr | g->intercept_dr;
269 c->intercept_exceptions = h->intercept_exceptions | g->intercept_exceptions;
270 c->intercept = h->intercept | g->intercept;
273 static inline struct vmcb *get_host_vmcb(struct vcpu_svm *svm)
275 if (is_guest_mode(&svm->vcpu))
276 return svm->nested.hsave;
281 static inline void set_cr_intercept(struct vcpu_svm *svm, int bit)
283 struct vmcb *vmcb = get_host_vmcb(svm);
285 vmcb->control.intercept_cr |= (1U << bit);
287 recalc_intercepts(svm);
290 static inline void clr_cr_intercept(struct vcpu_svm *svm, int bit)
292 struct vmcb *vmcb = get_host_vmcb(svm);
294 vmcb->control.intercept_cr &= ~(1U << bit);
296 recalc_intercepts(svm);
299 static inline bool is_cr_intercept(struct vcpu_svm *svm, int bit)
301 struct vmcb *vmcb = get_host_vmcb(svm);
303 return vmcb->control.intercept_cr & (1U << bit);
306 static inline void set_dr_intercept(struct vcpu_svm *svm, int bit)
308 struct vmcb *vmcb = get_host_vmcb(svm);
310 vmcb->control.intercept_dr |= (1U << bit);
312 recalc_intercepts(svm);
315 static inline void clr_dr_intercept(struct vcpu_svm *svm, int bit)
317 struct vmcb *vmcb = get_host_vmcb(svm);
319 vmcb->control.intercept_dr &= ~(1U << bit);
321 recalc_intercepts(svm);
324 static inline void set_exception_intercept(struct vcpu_svm *svm, int bit)
326 struct vmcb *vmcb = get_host_vmcb(svm);
328 vmcb->control.intercept_exceptions |= (1U << bit);
330 recalc_intercepts(svm);
333 static inline void clr_exception_intercept(struct vcpu_svm *svm, int bit)
335 struct vmcb *vmcb = get_host_vmcb(svm);
337 vmcb->control.intercept_exceptions &= ~(1U << bit);
339 recalc_intercepts(svm);
342 static inline void set_intercept(struct vcpu_svm *svm, int bit)
344 struct vmcb *vmcb = get_host_vmcb(svm);
346 vmcb->control.intercept |= (1ULL << bit);
348 recalc_intercepts(svm);
351 static inline void clr_intercept(struct vcpu_svm *svm, int bit)
353 struct vmcb *vmcb = get_host_vmcb(svm);
355 vmcb->control.intercept &= ~(1ULL << bit);
357 recalc_intercepts(svm);
360 static inline void enable_gif(struct vcpu_svm *svm)
362 svm->vcpu.arch.hflags |= HF_GIF_MASK;
365 static inline void disable_gif(struct vcpu_svm *svm)
367 svm->vcpu.arch.hflags &= ~HF_GIF_MASK;
370 static inline bool gif_set(struct vcpu_svm *svm)
372 return !!(svm->vcpu.arch.hflags & HF_GIF_MASK);
375 static unsigned long iopm_base;
377 struct kvm_ldttss_desc {
380 unsigned base1:8, type:5, dpl:2, p:1;
381 unsigned limit1:4, zero0:3, g:1, base2:8;
384 } __attribute__((packed));
386 struct svm_cpu_data {
392 struct kvm_ldttss_desc *tss_desc;
394 struct page *save_area;
397 static DEFINE_PER_CPU(struct svm_cpu_data *, svm_data);
399 struct svm_init_data {
404 static const u32 msrpm_ranges[] = {0, 0xc0000000, 0xc0010000};
406 #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
407 #define MSRS_RANGE_SIZE 2048
408 #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
410 static u32 svm_msrpm_offset(u32 msr)
415 for (i = 0; i < NUM_MSR_MAPS; i++) {
416 if (msr < msrpm_ranges[i] ||
417 msr >= msrpm_ranges[i] + MSRS_IN_RANGE)
420 offset = (msr - msrpm_ranges[i]) / 4; /* 4 msrs per u8 */
421 offset += (i * MSRS_RANGE_SIZE); /* add range offset */
423 /* Now we have the u8 offset - but need the u32 offset */
427 /* MSR not in any range */
431 #define MAX_INST_SIZE 15
433 static inline void clgi(void)
435 asm volatile (__ex(SVM_CLGI));
438 static inline void stgi(void)
440 asm volatile (__ex(SVM_STGI));
443 static inline void invlpga(unsigned long addr, u32 asid)
445 asm volatile (__ex(SVM_INVLPGA) : : "a"(addr), "c"(asid));
448 static int get_npt_level(void)
451 return PT64_ROOT_LEVEL;
453 return PT32E_ROOT_LEVEL;
457 static void svm_set_efer(struct kvm_vcpu *vcpu, u64 efer)
459 vcpu->arch.efer = efer;
460 if (!npt_enabled && !(efer & EFER_LMA))
463 to_svm(vcpu)->vmcb->save.efer = efer | EFER_SVME;
464 mark_dirty(to_svm(vcpu)->vmcb, VMCB_CR);
467 static int is_external_interrupt(u32 info)
469 info &= SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
470 return info == (SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR);
473 static u32 svm_get_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
475 struct vcpu_svm *svm = to_svm(vcpu);
478 if (svm->vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK)
479 ret |= KVM_X86_SHADOW_INT_STI | KVM_X86_SHADOW_INT_MOV_SS;
483 static void svm_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
485 struct vcpu_svm *svm = to_svm(vcpu);
488 svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK;
490 svm->vmcb->control.int_state |= SVM_INTERRUPT_SHADOW_MASK;
494 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
496 struct vcpu_svm *svm = to_svm(vcpu);
498 if (svm->vmcb->control.next_rip != 0) {
499 WARN_ON_ONCE(!static_cpu_has(X86_FEATURE_NRIPS));
500 svm->next_rip = svm->vmcb->control.next_rip;
503 if (!svm->next_rip) {
504 if (emulate_instruction(vcpu, EMULTYPE_SKIP) !=
506 printk(KERN_DEBUG "%s: NOP\n", __func__);
509 if (svm->next_rip - kvm_rip_read(vcpu) > MAX_INST_SIZE)
510 printk(KERN_ERR "%s: ip 0x%lx next 0x%llx\n",
511 __func__, kvm_rip_read(vcpu), svm->next_rip);
513 kvm_rip_write(vcpu, svm->next_rip);
514 svm_set_interrupt_shadow(vcpu, 0);
517 static void svm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
518 bool has_error_code, u32 error_code,
521 struct vcpu_svm *svm = to_svm(vcpu);
524 * If we are within a nested VM we'd better #VMEXIT and let the guest
525 * handle the exception
528 nested_svm_check_exception(svm, nr, has_error_code, error_code))
531 if (nr == BP_VECTOR && !static_cpu_has(X86_FEATURE_NRIPS)) {
532 unsigned long rip, old_rip = kvm_rip_read(&svm->vcpu);
535 * For guest debugging where we have to reinject #BP if some
536 * INT3 is guest-owned:
537 * Emulate nRIP by moving RIP forward. Will fail if injection
538 * raises a fault that is not intercepted. Still better than
539 * failing in all cases.
541 skip_emulated_instruction(&svm->vcpu);
542 rip = kvm_rip_read(&svm->vcpu);
543 svm->int3_rip = rip + svm->vmcb->save.cs.base;
544 svm->int3_injected = rip - old_rip;
547 svm->vmcb->control.event_inj = nr
549 | (has_error_code ? SVM_EVTINJ_VALID_ERR : 0)
550 | SVM_EVTINJ_TYPE_EXEPT;
551 svm->vmcb->control.event_inj_err = error_code;
554 static void svm_init_erratum_383(void)
560 if (!static_cpu_has_bug(X86_BUG_AMD_TLB_MMATCH))
563 /* Use _safe variants to not break nested virtualization */
564 val = native_read_msr_safe(MSR_AMD64_DC_CFG, &err);
570 low = lower_32_bits(val);
571 high = upper_32_bits(val);
573 native_write_msr_safe(MSR_AMD64_DC_CFG, low, high);
575 erratum_383_found = true;
578 static void svm_init_osvw(struct kvm_vcpu *vcpu)
581 * Guests should see errata 400 and 415 as fixed (assuming that
582 * HLT and IO instructions are intercepted).
584 vcpu->arch.osvw.length = (osvw_len >= 3) ? (osvw_len) : 3;
585 vcpu->arch.osvw.status = osvw_status & ~(6ULL);
588 * By increasing VCPU's osvw.length to 3 we are telling the guest that
589 * all osvw.status bits inside that length, including bit 0 (which is
590 * reserved for erratum 298), are valid. However, if host processor's
591 * osvw_len is 0 then osvw_status[0] carries no information. We need to
592 * be conservative here and therefore we tell the guest that erratum 298
593 * is present (because we really don't know).
595 if (osvw_len == 0 && boot_cpu_data.x86 == 0x10)
596 vcpu->arch.osvw.status |= 1;
599 static int has_svm(void)
603 if (!cpu_has_svm(&msg)) {
604 printk(KERN_INFO "has_svm: %s\n", msg);
611 static void svm_hardware_disable(void)
613 /* Make sure we clean up behind us */
614 if (static_cpu_has(X86_FEATURE_TSCRATEMSR))
615 wrmsrl(MSR_AMD64_TSC_RATIO, TSC_RATIO_DEFAULT);
619 amd_pmu_disable_virt();
622 static int svm_hardware_enable(void)
625 struct svm_cpu_data *sd;
627 struct desc_ptr gdt_descr;
628 struct desc_struct *gdt;
629 int me = raw_smp_processor_id();
631 rdmsrl(MSR_EFER, efer);
632 if (efer & EFER_SVME)
636 pr_err("%s: err EOPNOTSUPP on %d\n", __func__, me);
639 sd = per_cpu(svm_data, me);
641 pr_err("%s: svm_data is NULL on %d\n", __func__, me);
645 sd->asid_generation = 1;
646 sd->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1;
647 sd->next_asid = sd->max_asid + 1;
649 native_store_gdt(&gdt_descr);
650 gdt = (struct desc_struct *)gdt_descr.address;
651 sd->tss_desc = (struct kvm_ldttss_desc *)(gdt + GDT_ENTRY_TSS);
653 wrmsrl(MSR_EFER, efer | EFER_SVME);
655 wrmsrl(MSR_VM_HSAVE_PA, page_to_pfn(sd->save_area) << PAGE_SHIFT);
657 if (static_cpu_has(X86_FEATURE_TSCRATEMSR)) {
658 wrmsrl(MSR_AMD64_TSC_RATIO, TSC_RATIO_DEFAULT);
659 __get_cpu_var(current_tsc_ratio) = TSC_RATIO_DEFAULT;
666 * Note that it is possible to have a system with mixed processor
667 * revisions and therefore different OSVW bits. If bits are not the same
668 * on different processors then choose the worst case (i.e. if erratum
669 * is present on one processor and not on another then assume that the
670 * erratum is present everywhere).
672 if (cpu_has(&boot_cpu_data, X86_FEATURE_OSVW)) {
673 uint64_t len, status = 0;
676 len = native_read_msr_safe(MSR_AMD64_OSVW_ID_LENGTH, &err);
678 status = native_read_msr_safe(MSR_AMD64_OSVW_STATUS,
682 osvw_status = osvw_len = 0;
686 osvw_status |= status;
687 osvw_status &= (1ULL << osvw_len) - 1;
690 osvw_status = osvw_len = 0;
692 svm_init_erratum_383();
694 amd_pmu_enable_virt();
699 static void svm_cpu_uninit(int cpu)
701 struct svm_cpu_data *sd = per_cpu(svm_data, raw_smp_processor_id());
706 per_cpu(svm_data, raw_smp_processor_id()) = NULL;
707 __free_page(sd->save_area);
711 static int svm_cpu_init(int cpu)
713 struct svm_cpu_data *sd;
716 sd = kzalloc(sizeof(struct svm_cpu_data), GFP_KERNEL);
720 sd->save_area = alloc_page(GFP_KERNEL);
725 per_cpu(svm_data, cpu) = sd;
735 static bool valid_msr_intercept(u32 index)
739 for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++)
740 if (direct_access_msrs[i].index == index)
746 static void set_msr_interception(u32 *msrpm, unsigned msr,
749 u8 bit_read, bit_write;
754 * If this warning triggers extend the direct_access_msrs list at the
755 * beginning of the file
757 WARN_ON(!valid_msr_intercept(msr));
759 offset = svm_msrpm_offset(msr);
760 bit_read = 2 * (msr & 0x0f);
761 bit_write = 2 * (msr & 0x0f) + 1;
764 BUG_ON(offset == MSR_INVALID);
766 read ? clear_bit(bit_read, &tmp) : set_bit(bit_read, &tmp);
767 write ? clear_bit(bit_write, &tmp) : set_bit(bit_write, &tmp);
772 static void svm_vcpu_init_msrpm(u32 *msrpm)
776 memset(msrpm, 0xff, PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER));
778 for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
779 if (!direct_access_msrs[i].always)
782 set_msr_interception(msrpm, direct_access_msrs[i].index, 1, 1);
786 static void add_msr_offset(u32 offset)
790 for (i = 0; i < MSRPM_OFFSETS; ++i) {
792 /* Offset already in list? */
793 if (msrpm_offsets[i] == offset)
796 /* Slot used by another offset? */
797 if (msrpm_offsets[i] != MSR_INVALID)
800 /* Add offset to list */
801 msrpm_offsets[i] = offset;
807 * If this BUG triggers the msrpm_offsets table has an overflow. Just
808 * increase MSRPM_OFFSETS in this case.
813 static void init_msrpm_offsets(void)
817 memset(msrpm_offsets, 0xff, sizeof(msrpm_offsets));
819 for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
822 offset = svm_msrpm_offset(direct_access_msrs[i].index);
823 BUG_ON(offset == MSR_INVALID);
825 add_msr_offset(offset);
829 static void svm_enable_lbrv(struct vcpu_svm *svm)
831 u32 *msrpm = svm->msrpm;
833 svm->vmcb->control.lbr_ctl = 1;
834 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 1, 1);
835 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 1, 1);
836 set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 1, 1);
837 set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 1, 1);
840 static void svm_disable_lbrv(struct vcpu_svm *svm)
842 u32 *msrpm = svm->msrpm;
844 svm->vmcb->control.lbr_ctl = 0;
845 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 0, 0);
846 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 0, 0);
847 set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 0, 0);
848 set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 0, 0);
851 static __init int svm_hardware_setup(void)
854 struct page *iopm_pages;
858 iopm_pages = alloc_pages(GFP_KERNEL, IOPM_ALLOC_ORDER);
863 iopm_va = page_address(iopm_pages);
864 memset(iopm_va, 0xff, PAGE_SIZE * (1 << IOPM_ALLOC_ORDER));
865 iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT;
867 init_msrpm_offsets();
869 if (boot_cpu_has(X86_FEATURE_NX))
870 kvm_enable_efer_bits(EFER_NX);
872 if (boot_cpu_has(X86_FEATURE_FXSR_OPT))
873 kvm_enable_efer_bits(EFER_FFXSR);
875 if (boot_cpu_has(X86_FEATURE_TSCRATEMSR)) {
878 kvm_has_tsc_control = true;
881 * Make sure the user can only configure tsc_khz values that
882 * fit into a signed integer.
883 * A min value is not calculated needed because it will always
884 * be 1 on all machines and a value of 0 is used to disable
885 * tsc-scaling for the vcpu.
887 max = min(0x7fffffffULL, __scale_tsc(tsc_khz, TSC_RATIO_MAX));
889 kvm_max_guest_tsc_khz = max;
893 printk(KERN_INFO "kvm: Nested Virtualization enabled\n");
894 kvm_enable_efer_bits(EFER_SVME | EFER_LMSLE);
897 for_each_possible_cpu(cpu) {
898 r = svm_cpu_init(cpu);
903 if (!boot_cpu_has(X86_FEATURE_NPT))
906 if (npt_enabled && !npt) {
907 printk(KERN_INFO "kvm: Nested Paging disabled\n");
912 printk(KERN_INFO "kvm: Nested Paging enabled\n");
920 __free_pages(iopm_pages, IOPM_ALLOC_ORDER);
925 static __exit void svm_hardware_unsetup(void)
929 for_each_possible_cpu(cpu)
932 __free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT), IOPM_ALLOC_ORDER);
936 static void init_seg(struct vmcb_seg *seg)
939 seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK |
940 SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */
945 static void init_sys_seg(struct vmcb_seg *seg, uint32_t type)
948 seg->attrib = SVM_SELECTOR_P_MASK | type;
953 static u64 __scale_tsc(u64 ratio, u64 tsc)
955 u64 mult, frac, _tsc;
958 frac = ratio & ((1ULL << 32) - 1);
962 _tsc += (tsc >> 32) * frac;
963 _tsc += ((tsc & ((1ULL << 32) - 1)) * frac) >> 32;
968 static u64 svm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
970 struct vcpu_svm *svm = to_svm(vcpu);
973 if (svm->tsc_ratio != TSC_RATIO_DEFAULT)
974 _tsc = __scale_tsc(svm->tsc_ratio, tsc);
979 static void svm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
981 struct vcpu_svm *svm = to_svm(vcpu);
985 /* Guest TSC same frequency as host TSC? */
987 svm->tsc_ratio = TSC_RATIO_DEFAULT;
991 /* TSC scaling supported? */
992 if (!boot_cpu_has(X86_FEATURE_TSCRATEMSR)) {
993 if (user_tsc_khz > tsc_khz) {
994 vcpu->arch.tsc_catchup = 1;
995 vcpu->arch.tsc_always_catchup = 1;
997 WARN(1, "user requested TSC rate below hardware speed\n");
1003 /* TSC scaling required - calculate ratio */
1005 do_div(ratio, tsc_khz);
1007 if (ratio == 0 || ratio & TSC_RATIO_RSVD) {
1008 WARN_ONCE(1, "Invalid TSC ratio - virtual-tsc-khz=%u\n",
1012 svm->tsc_ratio = ratio;
1015 static u64 svm_read_tsc_offset(struct kvm_vcpu *vcpu)
1017 struct vcpu_svm *svm = to_svm(vcpu);
1019 return svm->vmcb->control.tsc_offset;
1022 static void svm_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1024 struct vcpu_svm *svm = to_svm(vcpu);
1025 u64 g_tsc_offset = 0;
1027 if (is_guest_mode(vcpu)) {
1028 g_tsc_offset = svm->vmcb->control.tsc_offset -
1029 svm->nested.hsave->control.tsc_offset;
1030 svm->nested.hsave->control.tsc_offset = offset;
1033 svm->vmcb->control.tsc_offset = offset + g_tsc_offset;
1035 mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
1038 static void svm_adjust_tsc_offset(struct kvm_vcpu *vcpu, s64 adjustment, bool host)
1040 struct vcpu_svm *svm = to_svm(vcpu);
1042 WARN_ON(adjustment < 0);
1044 adjustment = svm_scale_tsc(vcpu, adjustment);
1046 svm->vmcb->control.tsc_offset += adjustment;
1047 if (is_guest_mode(vcpu))
1048 svm->nested.hsave->control.tsc_offset += adjustment;
1049 mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
1052 static u64 svm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1056 tsc = svm_scale_tsc(vcpu, native_read_tsc());
1058 return target_tsc - tsc;
1061 static void init_vmcb(struct vcpu_svm *svm)
1063 struct vmcb_control_area *control = &svm->vmcb->control;
1064 struct vmcb_save_area *save = &svm->vmcb->save;
1066 svm->vcpu.fpu_active = 1;
1067 svm->vcpu.arch.hflags = 0;
1069 set_cr_intercept(svm, INTERCEPT_CR0_READ);
1070 set_cr_intercept(svm, INTERCEPT_CR3_READ);
1071 set_cr_intercept(svm, INTERCEPT_CR4_READ);
1072 set_cr_intercept(svm, INTERCEPT_CR0_WRITE);
1073 set_cr_intercept(svm, INTERCEPT_CR3_WRITE);
1074 set_cr_intercept(svm, INTERCEPT_CR4_WRITE);
1075 set_cr_intercept(svm, INTERCEPT_CR8_WRITE);
1077 set_dr_intercept(svm, INTERCEPT_DR0_READ);
1078 set_dr_intercept(svm, INTERCEPT_DR1_READ);
1079 set_dr_intercept(svm, INTERCEPT_DR2_READ);
1080 set_dr_intercept(svm, INTERCEPT_DR3_READ);
1081 set_dr_intercept(svm, INTERCEPT_DR4_READ);
1082 set_dr_intercept(svm, INTERCEPT_DR5_READ);
1083 set_dr_intercept(svm, INTERCEPT_DR6_READ);
1084 set_dr_intercept(svm, INTERCEPT_DR7_READ);
1086 set_dr_intercept(svm, INTERCEPT_DR0_WRITE);
1087 set_dr_intercept(svm, INTERCEPT_DR1_WRITE);
1088 set_dr_intercept(svm, INTERCEPT_DR2_WRITE);
1089 set_dr_intercept(svm, INTERCEPT_DR3_WRITE);
1090 set_dr_intercept(svm, INTERCEPT_DR4_WRITE);
1091 set_dr_intercept(svm, INTERCEPT_DR5_WRITE);
1092 set_dr_intercept(svm, INTERCEPT_DR6_WRITE);
1093 set_dr_intercept(svm, INTERCEPT_DR7_WRITE);
1095 set_exception_intercept(svm, PF_VECTOR);
1096 set_exception_intercept(svm, UD_VECTOR);
1097 set_exception_intercept(svm, MC_VECTOR);
1099 set_intercept(svm, INTERCEPT_INTR);
1100 set_intercept(svm, INTERCEPT_NMI);
1101 set_intercept(svm, INTERCEPT_SMI);
1102 set_intercept(svm, INTERCEPT_SELECTIVE_CR0);
1103 set_intercept(svm, INTERCEPT_RDPMC);
1104 set_intercept(svm, INTERCEPT_CPUID);
1105 set_intercept(svm, INTERCEPT_INVD);
1106 set_intercept(svm, INTERCEPT_HLT);
1107 set_intercept(svm, INTERCEPT_INVLPG);
1108 set_intercept(svm, INTERCEPT_INVLPGA);
1109 set_intercept(svm, INTERCEPT_IOIO_PROT);
1110 set_intercept(svm, INTERCEPT_MSR_PROT);
1111 set_intercept(svm, INTERCEPT_TASK_SWITCH);
1112 set_intercept(svm, INTERCEPT_SHUTDOWN);
1113 set_intercept(svm, INTERCEPT_VMRUN);
1114 set_intercept(svm, INTERCEPT_VMMCALL);
1115 set_intercept(svm, INTERCEPT_VMLOAD);
1116 set_intercept(svm, INTERCEPT_VMSAVE);
1117 set_intercept(svm, INTERCEPT_STGI);
1118 set_intercept(svm, INTERCEPT_CLGI);
1119 set_intercept(svm, INTERCEPT_SKINIT);
1120 set_intercept(svm, INTERCEPT_WBINVD);
1121 set_intercept(svm, INTERCEPT_MONITOR);
1122 set_intercept(svm, INTERCEPT_MWAIT);
1123 set_intercept(svm, INTERCEPT_XSETBV);
1125 control->iopm_base_pa = iopm_base;
1126 control->msrpm_base_pa = __pa(svm->msrpm);
1127 control->int_ctl = V_INTR_MASKING_MASK;
1129 init_seg(&save->es);
1130 init_seg(&save->ss);
1131 init_seg(&save->ds);
1132 init_seg(&save->fs);
1133 init_seg(&save->gs);
1135 save->cs.selector = 0xf000;
1136 save->cs.base = 0xffff0000;
1137 /* Executable/Readable Code Segment */
1138 save->cs.attrib = SVM_SELECTOR_READ_MASK | SVM_SELECTOR_P_MASK |
1139 SVM_SELECTOR_S_MASK | SVM_SELECTOR_CODE_MASK;
1140 save->cs.limit = 0xffff;
1142 save->gdtr.limit = 0xffff;
1143 save->idtr.limit = 0xffff;
1145 init_sys_seg(&save->ldtr, SEG_TYPE_LDT);
1146 init_sys_seg(&save->tr, SEG_TYPE_BUSY_TSS16);
1148 svm_set_efer(&svm->vcpu, 0);
1149 save->dr6 = 0xffff0ff0;
1150 kvm_set_rflags(&svm->vcpu, 2);
1151 save->rip = 0x0000fff0;
1152 svm->vcpu.arch.regs[VCPU_REGS_RIP] = save->rip;
1155 * This is the guest-visible cr0 value.
1156 * svm_set_cr0() sets PG and WP and clears NW and CD on save->cr0.
1158 svm->vcpu.arch.cr0 = 0;
1159 (void)kvm_set_cr0(&svm->vcpu, X86_CR0_NW | X86_CR0_CD | X86_CR0_ET);
1161 save->cr4 = X86_CR4_PAE;
1165 /* Setup VMCB for Nested Paging */
1166 control->nested_ctl = 1;
1167 clr_intercept(svm, INTERCEPT_INVLPG);
1168 clr_exception_intercept(svm, PF_VECTOR);
1169 clr_cr_intercept(svm, INTERCEPT_CR3_READ);
1170 clr_cr_intercept(svm, INTERCEPT_CR3_WRITE);
1171 save->g_pat = 0x0007040600070406ULL;
1175 svm->asid_generation = 0;
1177 svm->nested.vmcb = 0;
1178 svm->vcpu.arch.hflags = 0;
1180 if (boot_cpu_has(X86_FEATURE_PAUSEFILTER)) {
1181 control->pause_filter_count = 3000;
1182 set_intercept(svm, INTERCEPT_PAUSE);
1185 mark_all_dirty(svm->vmcb);
1190 static void svm_vcpu_reset(struct kvm_vcpu *vcpu)
1192 struct vcpu_svm *svm = to_svm(vcpu);
1198 kvm_cpuid(vcpu, &eax, &dummy, &dummy, &dummy);
1199 kvm_register_write(vcpu, VCPU_REGS_RDX, eax);
1202 static struct kvm_vcpu *svm_create_vcpu(struct kvm *kvm, unsigned int id)
1204 struct vcpu_svm *svm;
1206 struct page *msrpm_pages;
1207 struct page *hsave_page;
1208 struct page *nested_msrpm_pages;
1211 svm = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1217 svm->tsc_ratio = TSC_RATIO_DEFAULT;
1219 err = kvm_vcpu_init(&svm->vcpu, kvm, id);
1224 page = alloc_page(GFP_KERNEL);
1228 msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
1232 nested_msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
1233 if (!nested_msrpm_pages)
1236 hsave_page = alloc_page(GFP_KERNEL);
1240 svm->nested.hsave = page_address(hsave_page);
1242 svm->msrpm = page_address(msrpm_pages);
1243 svm_vcpu_init_msrpm(svm->msrpm);
1245 svm->nested.msrpm = page_address(nested_msrpm_pages);
1246 svm_vcpu_init_msrpm(svm->nested.msrpm);
1248 svm->vmcb = page_address(page);
1249 clear_page(svm->vmcb);
1250 svm->vmcb_pa = page_to_pfn(page) << PAGE_SHIFT;
1251 svm->asid_generation = 0;
1254 svm->vcpu.arch.apic_base = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
1255 if (kvm_vcpu_is_bsp(&svm->vcpu))
1256 svm->vcpu.arch.apic_base |= MSR_IA32_APICBASE_BSP;
1258 svm_init_osvw(&svm->vcpu);
1263 __free_pages(nested_msrpm_pages, MSRPM_ALLOC_ORDER);
1265 __free_pages(msrpm_pages, MSRPM_ALLOC_ORDER);
1269 kvm_vcpu_uninit(&svm->vcpu);
1271 kmem_cache_free(kvm_vcpu_cache, svm);
1273 return ERR_PTR(err);
1276 static void svm_free_vcpu(struct kvm_vcpu *vcpu)
1278 struct vcpu_svm *svm = to_svm(vcpu);
1280 __free_page(pfn_to_page(svm->vmcb_pa >> PAGE_SHIFT));
1281 __free_pages(virt_to_page(svm->msrpm), MSRPM_ALLOC_ORDER);
1282 __free_page(virt_to_page(svm->nested.hsave));
1283 __free_pages(virt_to_page(svm->nested.msrpm), MSRPM_ALLOC_ORDER);
1284 kvm_vcpu_uninit(vcpu);
1285 kmem_cache_free(kvm_vcpu_cache, svm);
1288 static void svm_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1290 struct vcpu_svm *svm = to_svm(vcpu);
1293 if (unlikely(cpu != vcpu->cpu)) {
1294 svm->asid_generation = 0;
1295 mark_all_dirty(svm->vmcb);
1298 #ifdef CONFIG_X86_64
1299 rdmsrl(MSR_GS_BASE, to_svm(vcpu)->host.gs_base);
1301 savesegment(fs, svm->host.fs);
1302 savesegment(gs, svm->host.gs);
1303 svm->host.ldt = kvm_read_ldt();
1305 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
1306 rdmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
1308 if (static_cpu_has(X86_FEATURE_TSCRATEMSR) &&
1309 svm->tsc_ratio != __get_cpu_var(current_tsc_ratio)) {
1310 __get_cpu_var(current_tsc_ratio) = svm->tsc_ratio;
1311 wrmsrl(MSR_AMD64_TSC_RATIO, svm->tsc_ratio);
1315 static void svm_vcpu_put(struct kvm_vcpu *vcpu)
1317 struct vcpu_svm *svm = to_svm(vcpu);
1320 ++vcpu->stat.host_state_reload;
1321 kvm_load_ldt(svm->host.ldt);
1322 #ifdef CONFIG_X86_64
1323 loadsegment(fs, svm->host.fs);
1324 wrmsrl(MSR_KERNEL_GS_BASE, current->thread.gs);
1325 load_gs_index(svm->host.gs);
1327 #ifdef CONFIG_X86_32_LAZY_GS
1328 loadsegment(gs, svm->host.gs);
1331 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
1332 wrmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
1335 static void svm_update_cpl(struct kvm_vcpu *vcpu)
1337 struct vcpu_svm *svm = to_svm(vcpu);
1340 if (!is_protmode(vcpu))
1342 else if (svm->vmcb->save.rflags & X86_EFLAGS_VM)
1345 cpl = svm->vmcb->save.cs.selector & 0x3;
1347 svm->vmcb->save.cpl = cpl;
1350 static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu)
1352 return to_svm(vcpu)->vmcb->save.rflags;
1355 static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1357 unsigned long old_rflags = to_svm(vcpu)->vmcb->save.rflags;
1359 to_svm(vcpu)->vmcb->save.rflags = rflags;
1360 if ((old_rflags ^ rflags) & X86_EFLAGS_VM)
1361 svm_update_cpl(vcpu);
1364 static void svm_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
1367 case VCPU_EXREG_PDPTR:
1368 BUG_ON(!npt_enabled);
1369 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
1376 static void svm_set_vintr(struct vcpu_svm *svm)
1378 set_intercept(svm, INTERCEPT_VINTR);
1381 static void svm_clear_vintr(struct vcpu_svm *svm)
1383 clr_intercept(svm, INTERCEPT_VINTR);
1386 static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg)
1388 struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
1391 case VCPU_SREG_CS: return &save->cs;
1392 case VCPU_SREG_DS: return &save->ds;
1393 case VCPU_SREG_ES: return &save->es;
1394 case VCPU_SREG_FS: return &save->fs;
1395 case VCPU_SREG_GS: return &save->gs;
1396 case VCPU_SREG_SS: return &save->ss;
1397 case VCPU_SREG_TR: return &save->tr;
1398 case VCPU_SREG_LDTR: return &save->ldtr;
1404 static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg)
1406 struct vmcb_seg *s = svm_seg(vcpu, seg);
1411 static void svm_get_segment(struct kvm_vcpu *vcpu,
1412 struct kvm_segment *var, int seg)
1414 struct vmcb_seg *s = svm_seg(vcpu, seg);
1416 var->base = s->base;
1417 var->limit = s->limit;
1418 var->selector = s->selector;
1419 var->type = s->attrib & SVM_SELECTOR_TYPE_MASK;
1420 var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1;
1421 var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
1422 var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1;
1423 var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1;
1424 var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
1425 var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
1426 var->g = (s->attrib >> SVM_SELECTOR_G_SHIFT) & 1;
1429 * AMD's VMCB does not have an explicit unusable field, so emulate it
1430 * for cross vendor migration purposes by "not present"
1432 var->unusable = !var->present || (var->type == 0);
1437 * SVM always stores 0 for the 'G' bit in the CS selector in
1438 * the VMCB on a VMEXIT. This hurts cross-vendor migration:
1439 * Intel's VMENTRY has a check on the 'G' bit.
1441 var->g = s->limit > 0xfffff;
1445 * Work around a bug where the busy flag in the tr selector
1455 * The accessed bit must always be set in the segment
1456 * descriptor cache, although it can be cleared in the
1457 * descriptor, the cached bit always remains at 1. Since
1458 * Intel has a check on this, set it here to support
1459 * cross-vendor migration.
1466 * On AMD CPUs sometimes the DB bit in the segment
1467 * descriptor is left as 1, although the whole segment has
1468 * been made unusable. Clear it here to pass an Intel VMX
1469 * entry check when cross vendor migrating.
1477 static int svm_get_cpl(struct kvm_vcpu *vcpu)
1479 struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
1484 static void svm_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1486 struct vcpu_svm *svm = to_svm(vcpu);
1488 dt->size = svm->vmcb->save.idtr.limit;
1489 dt->address = svm->vmcb->save.idtr.base;
1492 static void svm_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1494 struct vcpu_svm *svm = to_svm(vcpu);
1496 svm->vmcb->save.idtr.limit = dt->size;
1497 svm->vmcb->save.idtr.base = dt->address ;
1498 mark_dirty(svm->vmcb, VMCB_DT);
1501 static void svm_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1503 struct vcpu_svm *svm = to_svm(vcpu);
1505 dt->size = svm->vmcb->save.gdtr.limit;
1506 dt->address = svm->vmcb->save.gdtr.base;
1509 static void svm_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1511 struct vcpu_svm *svm = to_svm(vcpu);
1513 svm->vmcb->save.gdtr.limit = dt->size;
1514 svm->vmcb->save.gdtr.base = dt->address ;
1515 mark_dirty(svm->vmcb, VMCB_DT);
1518 static void svm_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
1522 static void svm_decache_cr3(struct kvm_vcpu *vcpu)
1526 static void svm_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
1530 static void update_cr0_intercept(struct vcpu_svm *svm)
1532 ulong gcr0 = svm->vcpu.arch.cr0;
1533 u64 *hcr0 = &svm->vmcb->save.cr0;
1535 if (!svm->vcpu.fpu_active)
1536 *hcr0 |= SVM_CR0_SELECTIVE_MASK;
1538 *hcr0 = (*hcr0 & ~SVM_CR0_SELECTIVE_MASK)
1539 | (gcr0 & SVM_CR0_SELECTIVE_MASK);
1541 mark_dirty(svm->vmcb, VMCB_CR);
1543 if (gcr0 == *hcr0 && svm->vcpu.fpu_active) {
1544 clr_cr_intercept(svm, INTERCEPT_CR0_READ);
1545 clr_cr_intercept(svm, INTERCEPT_CR0_WRITE);
1547 set_cr_intercept(svm, INTERCEPT_CR0_READ);
1548 set_cr_intercept(svm, INTERCEPT_CR0_WRITE);
1552 static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
1554 struct vcpu_svm *svm = to_svm(vcpu);
1556 #ifdef CONFIG_X86_64
1557 if (vcpu->arch.efer & EFER_LME) {
1558 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
1559 vcpu->arch.efer |= EFER_LMA;
1560 svm->vmcb->save.efer |= EFER_LMA | EFER_LME;
1563 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG)) {
1564 vcpu->arch.efer &= ~EFER_LMA;
1565 svm->vmcb->save.efer &= ~(EFER_LMA | EFER_LME);
1569 vcpu->arch.cr0 = cr0;
1572 cr0 |= X86_CR0_PG | X86_CR0_WP;
1574 if (!vcpu->fpu_active)
1577 * re-enable caching here because the QEMU bios
1578 * does not do it - this results in some delay at
1581 cr0 &= ~(X86_CR0_CD | X86_CR0_NW);
1582 svm->vmcb->save.cr0 = cr0;
1583 mark_dirty(svm->vmcb, VMCB_CR);
1584 update_cr0_intercept(svm);
1587 static int svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1589 unsigned long host_cr4_mce = read_cr4() & X86_CR4_MCE;
1590 unsigned long old_cr4 = to_svm(vcpu)->vmcb->save.cr4;
1592 if (cr4 & X86_CR4_VMXE)
1595 if (npt_enabled && ((old_cr4 ^ cr4) & X86_CR4_PGE))
1596 svm_flush_tlb(vcpu);
1598 vcpu->arch.cr4 = cr4;
1601 cr4 |= host_cr4_mce;
1602 to_svm(vcpu)->vmcb->save.cr4 = cr4;
1603 mark_dirty(to_svm(vcpu)->vmcb, VMCB_CR);
1607 static void svm_set_segment(struct kvm_vcpu *vcpu,
1608 struct kvm_segment *var, int seg)
1610 struct vcpu_svm *svm = to_svm(vcpu);
1611 struct vmcb_seg *s = svm_seg(vcpu, seg);
1613 s->base = var->base;
1614 s->limit = var->limit;
1615 s->selector = var->selector;
1619 s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK);
1620 s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT;
1621 s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT;
1622 s->attrib |= (var->present & 1) << SVM_SELECTOR_P_SHIFT;
1623 s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT;
1624 s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT;
1625 s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT;
1626 s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT;
1628 if (seg == VCPU_SREG_CS)
1629 svm_update_cpl(vcpu);
1631 mark_dirty(svm->vmcb, VMCB_SEG);
1634 static void update_db_bp_intercept(struct kvm_vcpu *vcpu)
1636 struct vcpu_svm *svm = to_svm(vcpu);
1638 clr_exception_intercept(svm, DB_VECTOR);
1639 clr_exception_intercept(svm, BP_VECTOR);
1641 if (svm->nmi_singlestep)
1642 set_exception_intercept(svm, DB_VECTOR);
1644 if (vcpu->guest_debug & KVM_GUESTDBG_ENABLE) {
1645 if (vcpu->guest_debug &
1646 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
1647 set_exception_intercept(svm, DB_VECTOR);
1648 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
1649 set_exception_intercept(svm, BP_VECTOR);
1651 vcpu->guest_debug = 0;
1654 static void new_asid(struct vcpu_svm *svm, struct svm_cpu_data *sd)
1656 if (sd->next_asid > sd->max_asid) {
1657 ++sd->asid_generation;
1659 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID;
1662 svm->asid_generation = sd->asid_generation;
1663 svm->vmcb->control.asid = sd->next_asid++;
1665 mark_dirty(svm->vmcb, VMCB_ASID);
1668 static void svm_set_dr7(struct kvm_vcpu *vcpu, unsigned long value)
1670 struct vcpu_svm *svm = to_svm(vcpu);
1672 svm->vmcb->save.dr7 = value;
1673 mark_dirty(svm->vmcb, VMCB_DR);
1676 static int pf_interception(struct vcpu_svm *svm)
1678 u64 fault_address = svm->vmcb->control.exit_info_2;
1682 switch (svm->apf_reason) {
1684 error_code = svm->vmcb->control.exit_info_1;
1686 trace_kvm_page_fault(fault_address, error_code);
1687 if (!npt_enabled && kvm_event_needs_reinjection(&svm->vcpu))
1688 kvm_mmu_unprotect_page_virt(&svm->vcpu, fault_address);
1689 r = kvm_mmu_page_fault(&svm->vcpu, fault_address, error_code,
1690 svm->vmcb->control.insn_bytes,
1691 svm->vmcb->control.insn_len);
1693 case KVM_PV_REASON_PAGE_NOT_PRESENT:
1694 svm->apf_reason = 0;
1695 local_irq_disable();
1696 kvm_async_pf_task_wait(fault_address);
1699 case KVM_PV_REASON_PAGE_READY:
1700 svm->apf_reason = 0;
1701 local_irq_disable();
1702 kvm_async_pf_task_wake(fault_address);
1709 static int db_interception(struct vcpu_svm *svm)
1711 struct kvm_run *kvm_run = svm->vcpu.run;
1713 if (!(svm->vcpu.guest_debug &
1714 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) &&
1715 !svm->nmi_singlestep) {
1716 kvm_queue_exception(&svm->vcpu, DB_VECTOR);
1720 if (svm->nmi_singlestep) {
1721 svm->nmi_singlestep = false;
1722 if (!(svm->vcpu.guest_debug & KVM_GUESTDBG_SINGLESTEP))
1723 svm->vmcb->save.rflags &=
1724 ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1725 update_db_bp_intercept(&svm->vcpu);
1728 if (svm->vcpu.guest_debug &
1729 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) {
1730 kvm_run->exit_reason = KVM_EXIT_DEBUG;
1731 kvm_run->debug.arch.pc =
1732 svm->vmcb->save.cs.base + svm->vmcb->save.rip;
1733 kvm_run->debug.arch.exception = DB_VECTOR;
1740 static int bp_interception(struct vcpu_svm *svm)
1742 struct kvm_run *kvm_run = svm->vcpu.run;
1744 kvm_run->exit_reason = KVM_EXIT_DEBUG;
1745 kvm_run->debug.arch.pc = svm->vmcb->save.cs.base + svm->vmcb->save.rip;
1746 kvm_run->debug.arch.exception = BP_VECTOR;
1750 static int ud_interception(struct vcpu_svm *svm)
1754 er = emulate_instruction(&svm->vcpu, EMULTYPE_TRAP_UD);
1755 if (er != EMULATE_DONE)
1756 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
1760 static void svm_fpu_activate(struct kvm_vcpu *vcpu)
1762 struct vcpu_svm *svm = to_svm(vcpu);
1764 clr_exception_intercept(svm, NM_VECTOR);
1766 svm->vcpu.fpu_active = 1;
1767 update_cr0_intercept(svm);
1770 static int nm_interception(struct vcpu_svm *svm)
1772 svm_fpu_activate(&svm->vcpu);
1776 static bool is_erratum_383(void)
1781 if (!erratum_383_found)
1784 value = native_read_msr_safe(MSR_IA32_MC0_STATUS, &err);
1788 /* Bit 62 may or may not be set for this mce */
1789 value &= ~(1ULL << 62);
1791 if (value != 0xb600000000010015ULL)
1794 /* Clear MCi_STATUS registers */
1795 for (i = 0; i < 6; ++i)
1796 native_write_msr_safe(MSR_IA32_MCx_STATUS(i), 0, 0);
1798 value = native_read_msr_safe(MSR_IA32_MCG_STATUS, &err);
1802 value &= ~(1ULL << 2);
1803 low = lower_32_bits(value);
1804 high = upper_32_bits(value);
1806 native_write_msr_safe(MSR_IA32_MCG_STATUS, low, high);
1809 /* Flush tlb to evict multi-match entries */
1815 static void svm_handle_mce(struct vcpu_svm *svm)
1817 if (is_erratum_383()) {
1819 * Erratum 383 triggered. Guest state is corrupt so kill the
1822 pr_err("KVM: Guest triggered AMD Erratum 383\n");
1824 kvm_make_request(KVM_REQ_TRIPLE_FAULT, &svm->vcpu);
1830 * On an #MC intercept the MCE handler is not called automatically in
1831 * the host. So do it by hand here.
1835 /* not sure if we ever come back to this point */
1840 static int mc_interception(struct vcpu_svm *svm)
1845 static int shutdown_interception(struct vcpu_svm *svm)
1847 struct kvm_run *kvm_run = svm->vcpu.run;
1850 * VMCB is undefined after a SHUTDOWN intercept
1851 * so reinitialize it.
1853 clear_page(svm->vmcb);
1856 kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
1860 static int io_interception(struct vcpu_svm *svm)
1862 struct kvm_vcpu *vcpu = &svm->vcpu;
1863 u32 io_info = svm->vmcb->control.exit_info_1; /* address size bug? */
1864 int size, in, string;
1867 ++svm->vcpu.stat.io_exits;
1868 string = (io_info & SVM_IOIO_STR_MASK) != 0;
1869 in = (io_info & SVM_IOIO_TYPE_MASK) != 0;
1871 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
1873 port = io_info >> 16;
1874 size = (io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT;
1875 svm->next_rip = svm->vmcb->control.exit_info_2;
1876 skip_emulated_instruction(&svm->vcpu);
1878 return kvm_fast_pio_out(vcpu, size, port);
1881 static int nmi_interception(struct vcpu_svm *svm)
1886 static int intr_interception(struct vcpu_svm *svm)
1888 ++svm->vcpu.stat.irq_exits;
1892 static int nop_on_interception(struct vcpu_svm *svm)
1897 static int halt_interception(struct vcpu_svm *svm)
1899 svm->next_rip = kvm_rip_read(&svm->vcpu) + 1;
1900 skip_emulated_instruction(&svm->vcpu);
1901 return kvm_emulate_halt(&svm->vcpu);
1904 static int vmmcall_interception(struct vcpu_svm *svm)
1906 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
1907 skip_emulated_instruction(&svm->vcpu);
1908 kvm_emulate_hypercall(&svm->vcpu);
1912 static unsigned long nested_svm_get_tdp_cr3(struct kvm_vcpu *vcpu)
1914 struct vcpu_svm *svm = to_svm(vcpu);
1916 return svm->nested.nested_cr3;
1919 static u64 nested_svm_get_tdp_pdptr(struct kvm_vcpu *vcpu, int index)
1921 struct vcpu_svm *svm = to_svm(vcpu);
1922 u64 cr3 = svm->nested.nested_cr3;
1926 ret = kvm_read_guest_page(vcpu->kvm, gpa_to_gfn(cr3), &pdpte,
1927 offset_in_page(cr3) + index * 8, 8);
1933 static void nested_svm_set_tdp_cr3(struct kvm_vcpu *vcpu,
1936 struct vcpu_svm *svm = to_svm(vcpu);
1938 svm->vmcb->control.nested_cr3 = root;
1939 mark_dirty(svm->vmcb, VMCB_NPT);
1940 svm_flush_tlb(vcpu);
1943 static void nested_svm_inject_npf_exit(struct kvm_vcpu *vcpu,
1944 struct x86_exception *fault)
1946 struct vcpu_svm *svm = to_svm(vcpu);
1948 svm->vmcb->control.exit_code = SVM_EXIT_NPF;
1949 svm->vmcb->control.exit_code_hi = 0;
1950 svm->vmcb->control.exit_info_1 = fault->error_code;
1951 svm->vmcb->control.exit_info_2 = fault->address;
1953 nested_svm_vmexit(svm);
1956 static int nested_svm_init_mmu_context(struct kvm_vcpu *vcpu)
1960 r = kvm_init_shadow_mmu(vcpu, &vcpu->arch.mmu);
1962 vcpu->arch.mmu.set_cr3 = nested_svm_set_tdp_cr3;
1963 vcpu->arch.mmu.get_cr3 = nested_svm_get_tdp_cr3;
1964 vcpu->arch.mmu.get_pdptr = nested_svm_get_tdp_pdptr;
1965 vcpu->arch.mmu.inject_page_fault = nested_svm_inject_npf_exit;
1966 vcpu->arch.mmu.shadow_root_level = get_npt_level();
1967 vcpu->arch.walk_mmu = &vcpu->arch.nested_mmu;
1972 static void nested_svm_uninit_mmu_context(struct kvm_vcpu *vcpu)
1974 vcpu->arch.walk_mmu = &vcpu->arch.mmu;
1977 static int nested_svm_check_permissions(struct vcpu_svm *svm)
1979 if (!(svm->vcpu.arch.efer & EFER_SVME)
1980 || !is_paging(&svm->vcpu)) {
1981 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
1985 if (svm->vmcb->save.cpl) {
1986 kvm_inject_gp(&svm->vcpu, 0);
1993 static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
1994 bool has_error_code, u32 error_code)
1998 if (!is_guest_mode(&svm->vcpu))
2001 svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + nr;
2002 svm->vmcb->control.exit_code_hi = 0;
2003 svm->vmcb->control.exit_info_1 = error_code;
2004 svm->vmcb->control.exit_info_2 = svm->vcpu.arch.cr2;
2006 vmexit = nested_svm_intercept(svm);
2007 if (vmexit == NESTED_EXIT_DONE)
2008 svm->nested.exit_required = true;
2013 /* This function returns true if it is save to enable the irq window */
2014 static inline bool nested_svm_intr(struct vcpu_svm *svm)
2016 if (!is_guest_mode(&svm->vcpu))
2019 if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
2022 if (!(svm->vcpu.arch.hflags & HF_HIF_MASK))
2026 * if vmexit was already requested (by intercepted exception
2027 * for instance) do not overwrite it with "external interrupt"
2030 if (svm->nested.exit_required)
2033 svm->vmcb->control.exit_code = SVM_EXIT_INTR;
2034 svm->vmcb->control.exit_info_1 = 0;
2035 svm->vmcb->control.exit_info_2 = 0;
2037 if (svm->nested.intercept & 1ULL) {
2039 * The #vmexit can't be emulated here directly because this
2040 * code path runs with irqs and preemption disabled. A
2041 * #vmexit emulation might sleep. Only signal request for
2044 svm->nested.exit_required = true;
2045 trace_kvm_nested_intr_vmexit(svm->vmcb->save.rip);
2052 /* This function returns true if it is save to enable the nmi window */
2053 static inline bool nested_svm_nmi(struct vcpu_svm *svm)
2055 if (!is_guest_mode(&svm->vcpu))
2058 if (!(svm->nested.intercept & (1ULL << INTERCEPT_NMI)))
2061 svm->vmcb->control.exit_code = SVM_EXIT_NMI;
2062 svm->nested.exit_required = true;
2067 static void *nested_svm_map(struct vcpu_svm *svm, u64 gpa, struct page **_page)
2073 page = gfn_to_page(svm->vcpu.kvm, gpa >> PAGE_SHIFT);
2074 if (is_error_page(page))
2082 kvm_inject_gp(&svm->vcpu, 0);
2087 static void nested_svm_unmap(struct page *page)
2090 kvm_release_page_dirty(page);
2093 static int nested_svm_intercept_ioio(struct vcpu_svm *svm)
2099 if (!(svm->nested.intercept & (1ULL << INTERCEPT_IOIO_PROT)))
2100 return NESTED_EXIT_HOST;
2102 port = svm->vmcb->control.exit_info_1 >> 16;
2103 gpa = svm->nested.vmcb_iopm + (port / 8);
2107 if (kvm_read_guest(svm->vcpu.kvm, gpa, &val, 1))
2110 return val ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
2113 static int nested_svm_exit_handled_msr(struct vcpu_svm *svm)
2115 u32 offset, msr, value;
2118 if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT)))
2119 return NESTED_EXIT_HOST;
2121 msr = svm->vcpu.arch.regs[VCPU_REGS_RCX];
2122 offset = svm_msrpm_offset(msr);
2123 write = svm->vmcb->control.exit_info_1 & 1;
2124 mask = 1 << ((2 * (msr & 0xf)) + write);
2126 if (offset == MSR_INVALID)
2127 return NESTED_EXIT_DONE;
2129 /* Offset is in 32 bit units but need in 8 bit units */
2132 if (kvm_read_guest(svm->vcpu.kvm, svm->nested.vmcb_msrpm + offset, &value, 4))
2133 return NESTED_EXIT_DONE;
2135 return (value & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
2138 static int nested_svm_exit_special(struct vcpu_svm *svm)
2140 u32 exit_code = svm->vmcb->control.exit_code;
2142 switch (exit_code) {
2145 case SVM_EXIT_EXCP_BASE + MC_VECTOR:
2146 return NESTED_EXIT_HOST;
2148 /* For now we are always handling NPFs when using them */
2150 return NESTED_EXIT_HOST;
2152 case SVM_EXIT_EXCP_BASE + PF_VECTOR:
2153 /* When we're shadowing, trap PFs, but not async PF */
2154 if (!npt_enabled && svm->apf_reason == 0)
2155 return NESTED_EXIT_HOST;
2157 case SVM_EXIT_EXCP_BASE + NM_VECTOR:
2158 nm_interception(svm);
2164 return NESTED_EXIT_CONTINUE;
2168 * If this function returns true, this #vmexit was already handled
2170 static int nested_svm_intercept(struct vcpu_svm *svm)
2172 u32 exit_code = svm->vmcb->control.exit_code;
2173 int vmexit = NESTED_EXIT_HOST;
2175 switch (exit_code) {
2177 vmexit = nested_svm_exit_handled_msr(svm);
2180 vmexit = nested_svm_intercept_ioio(svm);
2182 case SVM_EXIT_READ_CR0 ... SVM_EXIT_WRITE_CR8: {
2183 u32 bit = 1U << (exit_code - SVM_EXIT_READ_CR0);
2184 if (svm->nested.intercept_cr & bit)
2185 vmexit = NESTED_EXIT_DONE;
2188 case SVM_EXIT_READ_DR0 ... SVM_EXIT_WRITE_DR7: {
2189 u32 bit = 1U << (exit_code - SVM_EXIT_READ_DR0);
2190 if (svm->nested.intercept_dr & bit)
2191 vmexit = NESTED_EXIT_DONE;
2194 case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
2195 u32 excp_bits = 1 << (exit_code - SVM_EXIT_EXCP_BASE);
2196 if (svm->nested.intercept_exceptions & excp_bits)
2197 vmexit = NESTED_EXIT_DONE;
2198 /* async page fault always cause vmexit */
2199 else if ((exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR) &&
2200 svm->apf_reason != 0)
2201 vmexit = NESTED_EXIT_DONE;
2204 case SVM_EXIT_ERR: {
2205 vmexit = NESTED_EXIT_DONE;
2209 u64 exit_bits = 1ULL << (exit_code - SVM_EXIT_INTR);
2210 if (svm->nested.intercept & exit_bits)
2211 vmexit = NESTED_EXIT_DONE;
2218 static int nested_svm_exit_handled(struct vcpu_svm *svm)
2222 vmexit = nested_svm_intercept(svm);
2224 if (vmexit == NESTED_EXIT_DONE)
2225 nested_svm_vmexit(svm);
2230 static inline void copy_vmcb_control_area(struct vmcb *dst_vmcb, struct vmcb *from_vmcb)
2232 struct vmcb_control_area *dst = &dst_vmcb->control;
2233 struct vmcb_control_area *from = &from_vmcb->control;
2235 dst->intercept_cr = from->intercept_cr;
2236 dst->intercept_dr = from->intercept_dr;
2237 dst->intercept_exceptions = from->intercept_exceptions;
2238 dst->intercept = from->intercept;
2239 dst->iopm_base_pa = from->iopm_base_pa;
2240 dst->msrpm_base_pa = from->msrpm_base_pa;
2241 dst->tsc_offset = from->tsc_offset;
2242 dst->asid = from->asid;
2243 dst->tlb_ctl = from->tlb_ctl;
2244 dst->int_ctl = from->int_ctl;
2245 dst->int_vector = from->int_vector;
2246 dst->int_state = from->int_state;
2247 dst->exit_code = from->exit_code;
2248 dst->exit_code_hi = from->exit_code_hi;
2249 dst->exit_info_1 = from->exit_info_1;
2250 dst->exit_info_2 = from->exit_info_2;
2251 dst->exit_int_info = from->exit_int_info;
2252 dst->exit_int_info_err = from->exit_int_info_err;
2253 dst->nested_ctl = from->nested_ctl;
2254 dst->event_inj = from->event_inj;
2255 dst->event_inj_err = from->event_inj_err;
2256 dst->nested_cr3 = from->nested_cr3;
2257 dst->lbr_ctl = from->lbr_ctl;
2260 static int nested_svm_vmexit(struct vcpu_svm *svm)
2262 struct vmcb *nested_vmcb;
2263 struct vmcb *hsave = svm->nested.hsave;
2264 struct vmcb *vmcb = svm->vmcb;
2267 trace_kvm_nested_vmexit_inject(vmcb->control.exit_code,
2268 vmcb->control.exit_info_1,
2269 vmcb->control.exit_info_2,
2270 vmcb->control.exit_int_info,
2271 vmcb->control.exit_int_info_err,
2274 nested_vmcb = nested_svm_map(svm, svm->nested.vmcb, &page);
2278 /* Exit Guest-Mode */
2279 leave_guest_mode(&svm->vcpu);
2280 svm->nested.vmcb = 0;
2282 /* Give the current vmcb to the guest */
2285 nested_vmcb->save.es = vmcb->save.es;
2286 nested_vmcb->save.cs = vmcb->save.cs;
2287 nested_vmcb->save.ss = vmcb->save.ss;
2288 nested_vmcb->save.ds = vmcb->save.ds;
2289 nested_vmcb->save.gdtr = vmcb->save.gdtr;
2290 nested_vmcb->save.idtr = vmcb->save.idtr;
2291 nested_vmcb->save.efer = svm->vcpu.arch.efer;
2292 nested_vmcb->save.cr0 = kvm_read_cr0(&svm->vcpu);
2293 nested_vmcb->save.cr3 = kvm_read_cr3(&svm->vcpu);
2294 nested_vmcb->save.cr2 = vmcb->save.cr2;
2295 nested_vmcb->save.cr4 = svm->vcpu.arch.cr4;
2296 nested_vmcb->save.rflags = kvm_get_rflags(&svm->vcpu);
2297 nested_vmcb->save.rip = vmcb->save.rip;
2298 nested_vmcb->save.rsp = vmcb->save.rsp;
2299 nested_vmcb->save.rax = vmcb->save.rax;
2300 nested_vmcb->save.dr7 = vmcb->save.dr7;
2301 nested_vmcb->save.dr6 = vmcb->save.dr6;
2302 nested_vmcb->save.cpl = vmcb->save.cpl;
2304 nested_vmcb->control.int_ctl = vmcb->control.int_ctl;
2305 nested_vmcb->control.int_vector = vmcb->control.int_vector;
2306 nested_vmcb->control.int_state = vmcb->control.int_state;
2307 nested_vmcb->control.exit_code = vmcb->control.exit_code;
2308 nested_vmcb->control.exit_code_hi = vmcb->control.exit_code_hi;
2309 nested_vmcb->control.exit_info_1 = vmcb->control.exit_info_1;
2310 nested_vmcb->control.exit_info_2 = vmcb->control.exit_info_2;
2311 nested_vmcb->control.exit_int_info = vmcb->control.exit_int_info;
2312 nested_vmcb->control.exit_int_info_err = vmcb->control.exit_int_info_err;
2313 nested_vmcb->control.next_rip = vmcb->control.next_rip;
2316 * If we emulate a VMRUN/#VMEXIT in the same host #vmexit cycle we have
2317 * to make sure that we do not lose injected events. So check event_inj
2318 * here and copy it to exit_int_info if it is valid.
2319 * Exit_int_info and event_inj can't be both valid because the case
2320 * below only happens on a VMRUN instruction intercept which has
2321 * no valid exit_int_info set.
2323 if (vmcb->control.event_inj & SVM_EVTINJ_VALID) {
2324 struct vmcb_control_area *nc = &nested_vmcb->control;
2326 nc->exit_int_info = vmcb->control.event_inj;
2327 nc->exit_int_info_err = vmcb->control.event_inj_err;
2330 nested_vmcb->control.tlb_ctl = 0;
2331 nested_vmcb->control.event_inj = 0;
2332 nested_vmcb->control.event_inj_err = 0;
2334 /* We always set V_INTR_MASKING and remember the old value in hflags */
2335 if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
2336 nested_vmcb->control.int_ctl &= ~V_INTR_MASKING_MASK;
2338 /* Restore the original control entries */
2339 copy_vmcb_control_area(vmcb, hsave);
2341 kvm_clear_exception_queue(&svm->vcpu);
2342 kvm_clear_interrupt_queue(&svm->vcpu);
2344 svm->nested.nested_cr3 = 0;
2346 /* Restore selected save entries */
2347 svm->vmcb->save.es = hsave->save.es;
2348 svm->vmcb->save.cs = hsave->save.cs;
2349 svm->vmcb->save.ss = hsave->save.ss;
2350 svm->vmcb->save.ds = hsave->save.ds;
2351 svm->vmcb->save.gdtr = hsave->save.gdtr;
2352 svm->vmcb->save.idtr = hsave->save.idtr;
2353 kvm_set_rflags(&svm->vcpu, hsave->save.rflags);
2354 svm_set_efer(&svm->vcpu, hsave->save.efer);
2355 svm_set_cr0(&svm->vcpu, hsave->save.cr0 | X86_CR0_PE);
2356 svm_set_cr4(&svm->vcpu, hsave->save.cr4);
2358 svm->vmcb->save.cr3 = hsave->save.cr3;
2359 svm->vcpu.arch.cr3 = hsave->save.cr3;
2361 (void)kvm_set_cr3(&svm->vcpu, hsave->save.cr3);
2363 kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, hsave->save.rax);
2364 kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, hsave->save.rsp);
2365 kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, hsave->save.rip);
2366 svm->vmcb->save.dr7 = 0;
2367 svm->vmcb->save.cpl = 0;
2368 svm->vmcb->control.exit_int_info = 0;
2370 mark_all_dirty(svm->vmcb);
2372 nested_svm_unmap(page);
2374 nested_svm_uninit_mmu_context(&svm->vcpu);
2375 kvm_mmu_reset_context(&svm->vcpu);
2376 kvm_mmu_load(&svm->vcpu);
2381 static bool nested_svm_vmrun_msrpm(struct vcpu_svm *svm)
2384 * This function merges the msr permission bitmaps of kvm and the
2385 * nested vmcb. It is optimized in that it only merges the parts where
2386 * the kvm msr permission bitmap may contain zero bits
2390 if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT)))
2393 for (i = 0; i < MSRPM_OFFSETS; i++) {
2397 if (msrpm_offsets[i] == 0xffffffff)
2400 p = msrpm_offsets[i];
2401 offset = svm->nested.vmcb_msrpm + (p * 4);
2403 if (kvm_read_guest(svm->vcpu.kvm, offset, &value, 4))
2406 svm->nested.msrpm[p] = svm->msrpm[p] | value;
2409 svm->vmcb->control.msrpm_base_pa = __pa(svm->nested.msrpm);
2414 static bool nested_vmcb_checks(struct vmcb *vmcb)
2416 if ((vmcb->control.intercept & (1ULL << INTERCEPT_VMRUN)) == 0)
2419 if (vmcb->control.asid == 0)
2422 if (vmcb->control.nested_ctl && !npt_enabled)
2428 static bool nested_svm_vmrun(struct vcpu_svm *svm)
2430 struct vmcb *nested_vmcb;
2431 struct vmcb *hsave = svm->nested.hsave;
2432 struct vmcb *vmcb = svm->vmcb;
2436 vmcb_gpa = svm->vmcb->save.rax;
2438 nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
2442 if (!nested_vmcb_checks(nested_vmcb)) {
2443 nested_vmcb->control.exit_code = SVM_EXIT_ERR;
2444 nested_vmcb->control.exit_code_hi = 0;
2445 nested_vmcb->control.exit_info_1 = 0;
2446 nested_vmcb->control.exit_info_2 = 0;
2448 nested_svm_unmap(page);
2453 trace_kvm_nested_vmrun(svm->vmcb->save.rip, vmcb_gpa,
2454 nested_vmcb->save.rip,
2455 nested_vmcb->control.int_ctl,
2456 nested_vmcb->control.event_inj,
2457 nested_vmcb->control.nested_ctl);
2459 trace_kvm_nested_intercepts(nested_vmcb->control.intercept_cr & 0xffff,
2460 nested_vmcb->control.intercept_cr >> 16,
2461 nested_vmcb->control.intercept_exceptions,
2462 nested_vmcb->control.intercept);
2464 /* Clear internal status */
2465 kvm_clear_exception_queue(&svm->vcpu);
2466 kvm_clear_interrupt_queue(&svm->vcpu);
2469 * Save the old vmcb, so we don't need to pick what we save, but can
2470 * restore everything when a VMEXIT occurs
2472 hsave->save.es = vmcb->save.es;
2473 hsave->save.cs = vmcb->save.cs;
2474 hsave->save.ss = vmcb->save.ss;
2475 hsave->save.ds = vmcb->save.ds;
2476 hsave->save.gdtr = vmcb->save.gdtr;
2477 hsave->save.idtr = vmcb->save.idtr;
2478 hsave->save.efer = svm->vcpu.arch.efer;
2479 hsave->save.cr0 = kvm_read_cr0(&svm->vcpu);
2480 hsave->save.cr4 = svm->vcpu.arch.cr4;
2481 hsave->save.rflags = kvm_get_rflags(&svm->vcpu);
2482 hsave->save.rip = kvm_rip_read(&svm->vcpu);
2483 hsave->save.rsp = vmcb->save.rsp;
2484 hsave->save.rax = vmcb->save.rax;
2486 hsave->save.cr3 = vmcb->save.cr3;
2488 hsave->save.cr3 = kvm_read_cr3(&svm->vcpu);
2490 copy_vmcb_control_area(hsave, vmcb);
2492 if (kvm_get_rflags(&svm->vcpu) & X86_EFLAGS_IF)
2493 svm->vcpu.arch.hflags |= HF_HIF_MASK;
2495 svm->vcpu.arch.hflags &= ~HF_HIF_MASK;
2497 if (nested_vmcb->control.nested_ctl) {
2498 kvm_mmu_unload(&svm->vcpu);
2499 svm->nested.nested_cr3 = nested_vmcb->control.nested_cr3;
2500 nested_svm_init_mmu_context(&svm->vcpu);
2503 /* Load the nested guest state */
2504 svm->vmcb->save.es = nested_vmcb->save.es;
2505 svm->vmcb->save.cs = nested_vmcb->save.cs;
2506 svm->vmcb->save.ss = nested_vmcb->save.ss;
2507 svm->vmcb->save.ds = nested_vmcb->save.ds;
2508 svm->vmcb->save.gdtr = nested_vmcb->save.gdtr;
2509 svm->vmcb->save.idtr = nested_vmcb->save.idtr;
2510 kvm_set_rflags(&svm->vcpu, nested_vmcb->save.rflags);
2511 svm_set_efer(&svm->vcpu, nested_vmcb->save.efer);
2512 svm_set_cr0(&svm->vcpu, nested_vmcb->save.cr0);
2513 svm_set_cr4(&svm->vcpu, nested_vmcb->save.cr4);
2515 svm->vmcb->save.cr3 = nested_vmcb->save.cr3;
2516 svm->vcpu.arch.cr3 = nested_vmcb->save.cr3;
2518 (void)kvm_set_cr3(&svm->vcpu, nested_vmcb->save.cr3);
2520 /* Guest paging mode is active - reset mmu */
2521 kvm_mmu_reset_context(&svm->vcpu);
2523 svm->vmcb->save.cr2 = svm->vcpu.arch.cr2 = nested_vmcb->save.cr2;
2524 kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, nested_vmcb->save.rax);
2525 kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, nested_vmcb->save.rsp);
2526 kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, nested_vmcb->save.rip);
2528 /* In case we don't even reach vcpu_run, the fields are not updated */
2529 svm->vmcb->save.rax = nested_vmcb->save.rax;
2530 svm->vmcb->save.rsp = nested_vmcb->save.rsp;
2531 svm->vmcb->save.rip = nested_vmcb->save.rip;
2532 svm->vmcb->save.dr7 = nested_vmcb->save.dr7;
2533 svm->vmcb->save.dr6 = nested_vmcb->save.dr6;
2534 svm->vmcb->save.cpl = nested_vmcb->save.cpl;
2536 svm->nested.vmcb_msrpm = nested_vmcb->control.msrpm_base_pa & ~0x0fffULL;
2537 svm->nested.vmcb_iopm = nested_vmcb->control.iopm_base_pa & ~0x0fffULL;
2539 /* cache intercepts */
2540 svm->nested.intercept_cr = nested_vmcb->control.intercept_cr;
2541 svm->nested.intercept_dr = nested_vmcb->control.intercept_dr;
2542 svm->nested.intercept_exceptions = nested_vmcb->control.intercept_exceptions;
2543 svm->nested.intercept = nested_vmcb->control.intercept;
2545 svm_flush_tlb(&svm->vcpu);
2546 svm->vmcb->control.int_ctl = nested_vmcb->control.int_ctl | V_INTR_MASKING_MASK;
2547 if (nested_vmcb->control.int_ctl & V_INTR_MASKING_MASK)
2548 svm->vcpu.arch.hflags |= HF_VINTR_MASK;
2550 svm->vcpu.arch.hflags &= ~HF_VINTR_MASK;
2552 if (svm->vcpu.arch.hflags & HF_VINTR_MASK) {
2553 /* We only want the cr8 intercept bits of the guest */
2554 clr_cr_intercept(svm, INTERCEPT_CR8_READ);
2555 clr_cr_intercept(svm, INTERCEPT_CR8_WRITE);
2558 /* We don't want to see VMMCALLs from a nested guest */
2559 clr_intercept(svm, INTERCEPT_VMMCALL);
2561 svm->vmcb->control.lbr_ctl = nested_vmcb->control.lbr_ctl;
2562 svm->vmcb->control.int_vector = nested_vmcb->control.int_vector;
2563 svm->vmcb->control.int_state = nested_vmcb->control.int_state;
2564 svm->vmcb->control.tsc_offset += nested_vmcb->control.tsc_offset;
2565 svm->vmcb->control.event_inj = nested_vmcb->control.event_inj;
2566 svm->vmcb->control.event_inj_err = nested_vmcb->control.event_inj_err;
2568 nested_svm_unmap(page);
2570 /* Enter Guest-Mode */
2571 enter_guest_mode(&svm->vcpu);
2574 * Merge guest and host intercepts - must be called with vcpu in
2575 * guest-mode to take affect here
2577 recalc_intercepts(svm);
2579 svm->nested.vmcb = vmcb_gpa;
2583 mark_all_dirty(svm->vmcb);
2588 static void nested_svm_vmloadsave(struct vmcb *from_vmcb, struct vmcb *to_vmcb)
2590 to_vmcb->save.fs = from_vmcb->save.fs;
2591 to_vmcb->save.gs = from_vmcb->save.gs;
2592 to_vmcb->save.tr = from_vmcb->save.tr;
2593 to_vmcb->save.ldtr = from_vmcb->save.ldtr;
2594 to_vmcb->save.kernel_gs_base = from_vmcb->save.kernel_gs_base;
2595 to_vmcb->save.star = from_vmcb->save.star;
2596 to_vmcb->save.lstar = from_vmcb->save.lstar;
2597 to_vmcb->save.cstar = from_vmcb->save.cstar;
2598 to_vmcb->save.sfmask = from_vmcb->save.sfmask;
2599 to_vmcb->save.sysenter_cs = from_vmcb->save.sysenter_cs;
2600 to_vmcb->save.sysenter_esp = from_vmcb->save.sysenter_esp;
2601 to_vmcb->save.sysenter_eip = from_vmcb->save.sysenter_eip;
2604 static int vmload_interception(struct vcpu_svm *svm)
2606 struct vmcb *nested_vmcb;
2609 if (nested_svm_check_permissions(svm))
2612 nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
2616 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2617 skip_emulated_instruction(&svm->vcpu);
2619 nested_svm_vmloadsave(nested_vmcb, svm->vmcb);
2620 nested_svm_unmap(page);
2625 static int vmsave_interception(struct vcpu_svm *svm)
2627 struct vmcb *nested_vmcb;
2630 if (nested_svm_check_permissions(svm))
2633 nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
2637 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2638 skip_emulated_instruction(&svm->vcpu);
2640 nested_svm_vmloadsave(svm->vmcb, nested_vmcb);
2641 nested_svm_unmap(page);
2646 static int vmrun_interception(struct vcpu_svm *svm)
2648 if (nested_svm_check_permissions(svm))
2651 /* Save rip after vmrun instruction */
2652 kvm_rip_write(&svm->vcpu, kvm_rip_read(&svm->vcpu) + 3);
2654 if (!nested_svm_vmrun(svm))
2657 if (!nested_svm_vmrun_msrpm(svm))
2664 svm->vmcb->control.exit_code = SVM_EXIT_ERR;
2665 svm->vmcb->control.exit_code_hi = 0;
2666 svm->vmcb->control.exit_info_1 = 0;
2667 svm->vmcb->control.exit_info_2 = 0;
2669 nested_svm_vmexit(svm);
2674 static int stgi_interception(struct vcpu_svm *svm)
2676 if (nested_svm_check_permissions(svm))
2679 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2680 skip_emulated_instruction(&svm->vcpu);
2681 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
2688 static int clgi_interception(struct vcpu_svm *svm)
2690 if (nested_svm_check_permissions(svm))
2693 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2694 skip_emulated_instruction(&svm->vcpu);
2698 /* After a CLGI no interrupts should come */
2699 svm_clear_vintr(svm);
2700 svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
2702 mark_dirty(svm->vmcb, VMCB_INTR);
2707 static int invlpga_interception(struct vcpu_svm *svm)
2709 struct kvm_vcpu *vcpu = &svm->vcpu;
2711 trace_kvm_invlpga(svm->vmcb->save.rip, vcpu->arch.regs[VCPU_REGS_RCX],
2712 vcpu->arch.regs[VCPU_REGS_RAX]);
2714 /* Let's treat INVLPGA the same as INVLPG (can be optimized!) */
2715 kvm_mmu_invlpg(vcpu, vcpu->arch.regs[VCPU_REGS_RAX]);
2717 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2718 skip_emulated_instruction(&svm->vcpu);
2722 static int skinit_interception(struct vcpu_svm *svm)
2724 trace_kvm_skinit(svm->vmcb->save.rip, svm->vcpu.arch.regs[VCPU_REGS_RAX]);
2726 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2730 static int xsetbv_interception(struct vcpu_svm *svm)
2732 u64 new_bv = kvm_read_edx_eax(&svm->vcpu);
2733 u32 index = kvm_register_read(&svm->vcpu, VCPU_REGS_RCX);
2735 if (kvm_set_xcr(&svm->vcpu, index, new_bv) == 0) {
2736 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2737 skip_emulated_instruction(&svm->vcpu);
2743 static int invalid_op_interception(struct vcpu_svm *svm)
2745 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2749 static int task_switch_interception(struct vcpu_svm *svm)
2753 int int_type = svm->vmcb->control.exit_int_info &
2754 SVM_EXITINTINFO_TYPE_MASK;
2755 int int_vec = svm->vmcb->control.exit_int_info & SVM_EVTINJ_VEC_MASK;
2757 svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_TYPE_MASK;
2759 svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_VALID;
2760 bool has_error_code = false;
2763 tss_selector = (u16)svm->vmcb->control.exit_info_1;
2765 if (svm->vmcb->control.exit_info_2 &
2766 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_IRET))
2767 reason = TASK_SWITCH_IRET;
2768 else if (svm->vmcb->control.exit_info_2 &
2769 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_JMP))
2770 reason = TASK_SWITCH_JMP;
2772 reason = TASK_SWITCH_GATE;
2774 reason = TASK_SWITCH_CALL;
2776 if (reason == TASK_SWITCH_GATE) {
2778 case SVM_EXITINTINFO_TYPE_NMI:
2779 svm->vcpu.arch.nmi_injected = false;
2781 case SVM_EXITINTINFO_TYPE_EXEPT:
2782 if (svm->vmcb->control.exit_info_2 &
2783 (1ULL << SVM_EXITINFOSHIFT_TS_HAS_ERROR_CODE)) {
2784 has_error_code = true;
2786 (u32)svm->vmcb->control.exit_info_2;
2788 kvm_clear_exception_queue(&svm->vcpu);
2790 case SVM_EXITINTINFO_TYPE_INTR:
2791 kvm_clear_interrupt_queue(&svm->vcpu);
2798 if (reason != TASK_SWITCH_GATE ||
2799 int_type == SVM_EXITINTINFO_TYPE_SOFT ||
2800 (int_type == SVM_EXITINTINFO_TYPE_EXEPT &&
2801 (int_vec == OF_VECTOR || int_vec == BP_VECTOR)))
2802 skip_emulated_instruction(&svm->vcpu);
2804 if (int_type != SVM_EXITINTINFO_TYPE_SOFT)
2807 if (kvm_task_switch(&svm->vcpu, tss_selector, int_vec, reason,
2808 has_error_code, error_code) == EMULATE_FAIL) {
2809 svm->vcpu.run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
2810 svm->vcpu.run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
2811 svm->vcpu.run->internal.ndata = 0;
2817 static int cpuid_interception(struct vcpu_svm *svm)
2819 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
2820 kvm_emulate_cpuid(&svm->vcpu);
2824 static int iret_interception(struct vcpu_svm *svm)
2826 ++svm->vcpu.stat.nmi_window_exits;
2827 clr_intercept(svm, INTERCEPT_IRET);
2828 svm->vcpu.arch.hflags |= HF_IRET_MASK;
2829 svm->nmi_iret_rip = kvm_rip_read(&svm->vcpu);
2833 static int invlpg_interception(struct vcpu_svm *svm)
2835 if (!static_cpu_has(X86_FEATURE_DECODEASSISTS))
2836 return emulate_instruction(&svm->vcpu, 0) == EMULATE_DONE;
2838 kvm_mmu_invlpg(&svm->vcpu, svm->vmcb->control.exit_info_1);
2839 skip_emulated_instruction(&svm->vcpu);
2843 static int emulate_on_interception(struct vcpu_svm *svm)
2845 return emulate_instruction(&svm->vcpu, 0) == EMULATE_DONE;
2848 static int rdpmc_interception(struct vcpu_svm *svm)
2852 if (!static_cpu_has(X86_FEATURE_NRIPS))
2853 return emulate_on_interception(svm);
2855 err = kvm_rdpmc(&svm->vcpu);
2856 kvm_complete_insn_gp(&svm->vcpu, err);
2861 bool check_selective_cr0_intercepted(struct vcpu_svm *svm, unsigned long val)
2863 unsigned long cr0 = svm->vcpu.arch.cr0;
2867 intercept = svm->nested.intercept;
2869 if (!is_guest_mode(&svm->vcpu) ||
2870 (!(intercept & (1ULL << INTERCEPT_SELECTIVE_CR0))))
2873 cr0 &= ~SVM_CR0_SELECTIVE_MASK;
2874 val &= ~SVM_CR0_SELECTIVE_MASK;
2877 svm->vmcb->control.exit_code = SVM_EXIT_CR0_SEL_WRITE;
2878 ret = (nested_svm_exit_handled(svm) == NESTED_EXIT_DONE);
2884 #define CR_VALID (1ULL << 63)
2886 static int cr_interception(struct vcpu_svm *svm)
2892 if (!static_cpu_has(X86_FEATURE_DECODEASSISTS))
2893 return emulate_on_interception(svm);
2895 if (unlikely((svm->vmcb->control.exit_info_1 & CR_VALID) == 0))
2896 return emulate_on_interception(svm);
2898 reg = svm->vmcb->control.exit_info_1 & SVM_EXITINFO_REG_MASK;
2899 cr = svm->vmcb->control.exit_code - SVM_EXIT_READ_CR0;
2902 if (cr >= 16) { /* mov to cr */
2904 val = kvm_register_read(&svm->vcpu, reg);
2907 if (!check_selective_cr0_intercepted(svm, val))
2908 err = kvm_set_cr0(&svm->vcpu, val);
2914 err = kvm_set_cr3(&svm->vcpu, val);
2917 err = kvm_set_cr4(&svm->vcpu, val);
2920 err = kvm_set_cr8(&svm->vcpu, val);
2923 WARN(1, "unhandled write to CR%d", cr);
2924 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2927 } else { /* mov from cr */
2930 val = kvm_read_cr0(&svm->vcpu);
2933 val = svm->vcpu.arch.cr2;
2936 val = kvm_read_cr3(&svm->vcpu);
2939 val = kvm_read_cr4(&svm->vcpu);
2942 val = kvm_get_cr8(&svm->vcpu);
2945 WARN(1, "unhandled read from CR%d", cr);
2946 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2949 kvm_register_write(&svm->vcpu, reg, val);
2951 kvm_complete_insn_gp(&svm->vcpu, err);
2956 static int dr_interception(struct vcpu_svm *svm)
2962 if (!boot_cpu_has(X86_FEATURE_DECODEASSISTS))
2963 return emulate_on_interception(svm);
2965 reg = svm->vmcb->control.exit_info_1 & SVM_EXITINFO_REG_MASK;
2966 dr = svm->vmcb->control.exit_code - SVM_EXIT_READ_DR0;
2968 if (dr >= 16) { /* mov to DRn */
2969 val = kvm_register_read(&svm->vcpu, reg);
2970 kvm_set_dr(&svm->vcpu, dr - 16, val);
2972 err = kvm_get_dr(&svm->vcpu, dr, &val);
2974 kvm_register_write(&svm->vcpu, reg, val);
2977 skip_emulated_instruction(&svm->vcpu);
2982 static int cr8_write_interception(struct vcpu_svm *svm)
2984 struct kvm_run *kvm_run = svm->vcpu.run;
2987 u8 cr8_prev = kvm_get_cr8(&svm->vcpu);
2988 /* instruction emulation calls kvm_set_cr8() */
2989 r = cr_interception(svm);
2990 if (irqchip_in_kernel(svm->vcpu.kvm))
2992 if (cr8_prev <= kvm_get_cr8(&svm->vcpu))
2994 kvm_run->exit_reason = KVM_EXIT_SET_TPR;
2998 u64 svm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
3000 struct vmcb *vmcb = get_host_vmcb(to_svm(vcpu));
3001 return vmcb->control.tsc_offset +
3002 svm_scale_tsc(vcpu, host_tsc);
3005 static int svm_get_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 *data)
3007 struct vcpu_svm *svm = to_svm(vcpu);
3010 case MSR_IA32_TSC: {
3011 *data = svm->vmcb->control.tsc_offset +
3012 svm_scale_tsc(vcpu, native_read_tsc());
3017 *data = svm->vmcb->save.star;
3019 #ifdef CONFIG_X86_64
3021 *data = svm->vmcb->save.lstar;
3024 *data = svm->vmcb->save.cstar;
3026 case MSR_KERNEL_GS_BASE:
3027 *data = svm->vmcb->save.kernel_gs_base;
3029 case MSR_SYSCALL_MASK:
3030 *data = svm->vmcb->save.sfmask;
3033 case MSR_IA32_SYSENTER_CS:
3034 *data = svm->vmcb->save.sysenter_cs;
3036 case MSR_IA32_SYSENTER_EIP:
3037 *data = svm->sysenter_eip;
3039 case MSR_IA32_SYSENTER_ESP:
3040 *data = svm->sysenter_esp;
3043 * Nobody will change the following 5 values in the VMCB so we can
3044 * safely return them on rdmsr. They will always be 0 until LBRV is
3047 case MSR_IA32_DEBUGCTLMSR:
3048 *data = svm->vmcb->save.dbgctl;
3050 case MSR_IA32_LASTBRANCHFROMIP:
3051 *data = svm->vmcb->save.br_from;
3053 case MSR_IA32_LASTBRANCHTOIP:
3054 *data = svm->vmcb->save.br_to;
3056 case MSR_IA32_LASTINTFROMIP:
3057 *data = svm->vmcb->save.last_excp_from;
3059 case MSR_IA32_LASTINTTOIP:
3060 *data = svm->vmcb->save.last_excp_to;
3062 case MSR_VM_HSAVE_PA:
3063 *data = svm->nested.hsave_msr;
3066 *data = svm->nested.vm_cr_msr;
3068 case MSR_IA32_UCODE_REV:
3072 return kvm_get_msr_common(vcpu, ecx, data);
3077 static int rdmsr_interception(struct vcpu_svm *svm)
3079 u32 ecx = svm->vcpu.arch.regs[VCPU_REGS_RCX];
3082 if (svm_get_msr(&svm->vcpu, ecx, &data)) {
3083 trace_kvm_msr_read_ex(ecx);
3084 kvm_inject_gp(&svm->vcpu, 0);
3086 trace_kvm_msr_read(ecx, data);
3088 svm->vcpu.arch.regs[VCPU_REGS_RAX] = data & 0xffffffff;
3089 svm->vcpu.arch.regs[VCPU_REGS_RDX] = data >> 32;
3090 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
3091 skip_emulated_instruction(&svm->vcpu);
3096 static int svm_set_vm_cr(struct kvm_vcpu *vcpu, u64 data)
3098 struct vcpu_svm *svm = to_svm(vcpu);
3099 int svm_dis, chg_mask;
3101 if (data & ~SVM_VM_CR_VALID_MASK)
3104 chg_mask = SVM_VM_CR_VALID_MASK;
3106 if (svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK)
3107 chg_mask &= ~(SVM_VM_CR_SVM_LOCK_MASK | SVM_VM_CR_SVM_DIS_MASK);
3109 svm->nested.vm_cr_msr &= ~chg_mask;
3110 svm->nested.vm_cr_msr |= (data & chg_mask);
3112 svm_dis = svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK;
3114 /* check for svm_disable while efer.svme is set */
3115 if (svm_dis && (vcpu->arch.efer & EFER_SVME))
3121 static int svm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
3123 struct vcpu_svm *svm = to_svm(vcpu);
3125 u32 ecx = msr->index;
3126 u64 data = msr->data;
3129 kvm_write_tsc(vcpu, msr);
3132 svm->vmcb->save.star = data;
3134 #ifdef CONFIG_X86_64
3136 svm->vmcb->save.lstar = data;
3139 svm->vmcb->save.cstar = data;
3141 case MSR_KERNEL_GS_BASE:
3142 svm->vmcb->save.kernel_gs_base = data;
3144 case MSR_SYSCALL_MASK:
3145 svm->vmcb->save.sfmask = data;
3148 case MSR_IA32_SYSENTER_CS:
3149 svm->vmcb->save.sysenter_cs = data;
3151 case MSR_IA32_SYSENTER_EIP:
3152 svm->sysenter_eip = data;
3153 svm->vmcb->save.sysenter_eip = data;
3155 case MSR_IA32_SYSENTER_ESP:
3156 svm->sysenter_esp = data;
3157 svm->vmcb->save.sysenter_esp = data;
3159 case MSR_IA32_DEBUGCTLMSR:
3160 if (!boot_cpu_has(X86_FEATURE_LBRV)) {
3161 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTL 0x%llx, nop\n",
3165 if (data & DEBUGCTL_RESERVED_BITS)
3168 svm->vmcb->save.dbgctl = data;
3169 mark_dirty(svm->vmcb, VMCB_LBR);
3170 if (data & (1ULL<<0))
3171 svm_enable_lbrv(svm);
3173 svm_disable_lbrv(svm);
3175 case MSR_VM_HSAVE_PA:
3176 svm->nested.hsave_msr = data;
3179 return svm_set_vm_cr(vcpu, data);
3181 vcpu_unimpl(vcpu, "unimplemented wrmsr: 0x%x data 0x%llx\n", ecx, data);
3184 return kvm_set_msr_common(vcpu, msr);
3189 static int wrmsr_interception(struct vcpu_svm *svm)
3191 struct msr_data msr;
3192 u32 ecx = svm->vcpu.arch.regs[VCPU_REGS_RCX];
3193 u64 data = (svm->vcpu.arch.regs[VCPU_REGS_RAX] & -1u)
3194 | ((u64)(svm->vcpu.arch.regs[VCPU_REGS_RDX] & -1u) << 32);
3198 msr.host_initiated = false;
3200 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
3201 if (kvm_set_msr(&svm->vcpu, &msr)) {
3202 trace_kvm_msr_write_ex(ecx, data);
3203 kvm_inject_gp(&svm->vcpu, 0);
3205 trace_kvm_msr_write(ecx, data);
3206 skip_emulated_instruction(&svm->vcpu);
3211 static int msr_interception(struct vcpu_svm *svm)
3213 if (svm->vmcb->control.exit_info_1)
3214 return wrmsr_interception(svm);
3216 return rdmsr_interception(svm);
3219 static int interrupt_window_interception(struct vcpu_svm *svm)
3221 struct kvm_run *kvm_run = svm->vcpu.run;
3223 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
3224 svm_clear_vintr(svm);
3225 svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
3226 mark_dirty(svm->vmcb, VMCB_INTR);
3227 ++svm->vcpu.stat.irq_window_exits;
3229 * If the user space waits to inject interrupts, exit as soon as
3232 if (!irqchip_in_kernel(svm->vcpu.kvm) &&
3233 kvm_run->request_interrupt_window &&
3234 !kvm_cpu_has_interrupt(&svm->vcpu)) {
3235 kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
3242 static int pause_interception(struct vcpu_svm *svm)
3244 kvm_vcpu_on_spin(&(svm->vcpu));
3248 static int (*const svm_exit_handlers[])(struct vcpu_svm *svm) = {
3249 [SVM_EXIT_READ_CR0] = cr_interception,
3250 [SVM_EXIT_READ_CR3] = cr_interception,
3251 [SVM_EXIT_READ_CR4] = cr_interception,
3252 [SVM_EXIT_READ_CR8] = cr_interception,
3253 [SVM_EXIT_CR0_SEL_WRITE] = emulate_on_interception,
3254 [SVM_EXIT_WRITE_CR0] = cr_interception,
3255 [SVM_EXIT_WRITE_CR3] = cr_interception,
3256 [SVM_EXIT_WRITE_CR4] = cr_interception,
3257 [SVM_EXIT_WRITE_CR8] = cr8_write_interception,
3258 [SVM_EXIT_READ_DR0] = dr_interception,
3259 [SVM_EXIT_READ_DR1] = dr_interception,
3260 [SVM_EXIT_READ_DR2] = dr_interception,
3261 [SVM_EXIT_READ_DR3] = dr_interception,
3262 [SVM_EXIT_READ_DR4] = dr_interception,
3263 [SVM_EXIT_READ_DR5] = dr_interception,
3264 [SVM_EXIT_READ_DR6] = dr_interception,
3265 [SVM_EXIT_READ_DR7] = dr_interception,
3266 [SVM_EXIT_WRITE_DR0] = dr_interception,
3267 [SVM_EXIT_WRITE_DR1] = dr_interception,
3268 [SVM_EXIT_WRITE_DR2] = dr_interception,
3269 [SVM_EXIT_WRITE_DR3] = dr_interception,
3270 [SVM_EXIT_WRITE_DR4] = dr_interception,
3271 [SVM_EXIT_WRITE_DR5] = dr_interception,
3272 [SVM_EXIT_WRITE_DR6] = dr_interception,
3273 [SVM_EXIT_WRITE_DR7] = dr_interception,
3274 [SVM_EXIT_EXCP_BASE + DB_VECTOR] = db_interception,
3275 [SVM_EXIT_EXCP_BASE + BP_VECTOR] = bp_interception,
3276 [SVM_EXIT_EXCP_BASE + UD_VECTOR] = ud_interception,
3277 [SVM_EXIT_EXCP_BASE + PF_VECTOR] = pf_interception,
3278 [SVM_EXIT_EXCP_BASE + NM_VECTOR] = nm_interception,
3279 [SVM_EXIT_EXCP_BASE + MC_VECTOR] = mc_interception,
3280 [SVM_EXIT_INTR] = intr_interception,
3281 [SVM_EXIT_NMI] = nmi_interception,
3282 [SVM_EXIT_SMI] = nop_on_interception,
3283 [SVM_EXIT_INIT] = nop_on_interception,
3284 [SVM_EXIT_VINTR] = interrupt_window_interception,
3285 [SVM_EXIT_RDPMC] = rdpmc_interception,
3286 [SVM_EXIT_CPUID] = cpuid_interception,
3287 [SVM_EXIT_IRET] = iret_interception,
3288 [SVM_EXIT_INVD] = emulate_on_interception,
3289 [SVM_EXIT_PAUSE] = pause_interception,
3290 [SVM_EXIT_HLT] = halt_interception,
3291 [SVM_EXIT_INVLPG] = invlpg_interception,
3292 [SVM_EXIT_INVLPGA] = invlpga_interception,
3293 [SVM_EXIT_IOIO] = io_interception,
3294 [SVM_EXIT_MSR] = msr_interception,
3295 [SVM_EXIT_TASK_SWITCH] = task_switch_interception,
3296 [SVM_EXIT_SHUTDOWN] = shutdown_interception,
3297 [SVM_EXIT_VMRUN] = vmrun_interception,
3298 [SVM_EXIT_VMMCALL] = vmmcall_interception,
3299 [SVM_EXIT_VMLOAD] = vmload_interception,
3300 [SVM_EXIT_VMSAVE] = vmsave_interception,
3301 [SVM_EXIT_STGI] = stgi_interception,
3302 [SVM_EXIT_CLGI] = clgi_interception,
3303 [SVM_EXIT_SKINIT] = skinit_interception,
3304 [SVM_EXIT_WBINVD] = emulate_on_interception,
3305 [SVM_EXIT_MONITOR] = invalid_op_interception,
3306 [SVM_EXIT_MWAIT] = invalid_op_interception,
3307 [SVM_EXIT_XSETBV] = xsetbv_interception,
3308 [SVM_EXIT_NPF] = pf_interception,
3311 static void dump_vmcb(struct kvm_vcpu *vcpu)
3313 struct vcpu_svm *svm = to_svm(vcpu);
3314 struct vmcb_control_area *control = &svm->vmcb->control;
3315 struct vmcb_save_area *save = &svm->vmcb->save;
3317 pr_err("VMCB Control Area:\n");
3318 pr_err("%-20s%04x\n", "cr_read:", control->intercept_cr & 0xffff);
3319 pr_err("%-20s%04x\n", "cr_write:", control->intercept_cr >> 16);
3320 pr_err("%-20s%04x\n", "dr_read:", control->intercept_dr & 0xffff);
3321 pr_err("%-20s%04x\n", "dr_write:", control->intercept_dr >> 16);
3322 pr_err("%-20s%08x\n", "exceptions:", control->intercept_exceptions);
3323 pr_err("%-20s%016llx\n", "intercepts:", control->intercept);
3324 pr_err("%-20s%d\n", "pause filter count:", control->pause_filter_count);
3325 pr_err("%-20s%016llx\n", "iopm_base_pa:", control->iopm_base_pa);
3326 pr_err("%-20s%016llx\n", "msrpm_base_pa:", control->msrpm_base_pa);
3327 pr_err("%-20s%016llx\n", "tsc_offset:", control->tsc_offset);
3328 pr_err("%-20s%d\n", "asid:", control->asid);
3329 pr_err("%-20s%d\n", "tlb_ctl:", control->tlb_ctl);
3330 pr_err("%-20s%08x\n", "int_ctl:", control->int_ctl);
3331 pr_err("%-20s%08x\n", "int_vector:", control->int_vector);
3332 pr_err("%-20s%08x\n", "int_state:", control->int_state);
3333 pr_err("%-20s%08x\n", "exit_code:", control->exit_code);
3334 pr_err("%-20s%016llx\n", "exit_info1:", control->exit_info_1);
3335 pr_err("%-20s%016llx\n", "exit_info2:", control->exit_info_2);
3336 pr_err("%-20s%08x\n", "exit_int_info:", control->exit_int_info);
3337 pr_err("%-20s%08x\n", "exit_int_info_err:", control->exit_int_info_err);
3338 pr_err("%-20s%lld\n", "nested_ctl:", control->nested_ctl);
3339 pr_err("%-20s%016llx\n", "nested_cr3:", control->nested_cr3);
3340 pr_err("%-20s%08x\n", "event_inj:", control->event_inj);
3341 pr_err("%-20s%08x\n", "event_inj_err:", control->event_inj_err);
3342 pr_err("%-20s%lld\n", "lbr_ctl:", control->lbr_ctl);
3343 pr_err("%-20s%016llx\n", "next_rip:", control->next_rip);
3344 pr_err("VMCB State Save Area:\n");
3345 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3347 save->es.selector, save->es.attrib,
3348 save->es.limit, save->es.base);
3349 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3351 save->cs.selector, save->cs.attrib,
3352 save->cs.limit, save->cs.base);
3353 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3355 save->ss.selector, save->ss.attrib,
3356 save->ss.limit, save->ss.base);
3357 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3359 save->ds.selector, save->ds.attrib,
3360 save->ds.limit, save->ds.base);
3361 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3363 save->fs.selector, save->fs.attrib,
3364 save->fs.limit, save->fs.base);
3365 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3367 save->gs.selector, save->gs.attrib,
3368 save->gs.limit, save->gs.base);
3369 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3371 save->gdtr.selector, save->gdtr.attrib,
3372 save->gdtr.limit, save->gdtr.base);
3373 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3375 save->ldtr.selector, save->ldtr.attrib,
3376 save->ldtr.limit, save->ldtr.base);
3377 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3379 save->idtr.selector, save->idtr.attrib,
3380 save->idtr.limit, save->idtr.base);
3381 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3383 save->tr.selector, save->tr.attrib,
3384 save->tr.limit, save->tr.base);
3385 pr_err("cpl: %d efer: %016llx\n",
3386 save->cpl, save->efer);
3387 pr_err("%-15s %016llx %-13s %016llx\n",
3388 "cr0:", save->cr0, "cr2:", save->cr2);
3389 pr_err("%-15s %016llx %-13s %016llx\n",
3390 "cr3:", save->cr3, "cr4:", save->cr4);
3391 pr_err("%-15s %016llx %-13s %016llx\n",
3392 "dr6:", save->dr6, "dr7:", save->dr7);
3393 pr_err("%-15s %016llx %-13s %016llx\n",
3394 "rip:", save->rip, "rflags:", save->rflags);
3395 pr_err("%-15s %016llx %-13s %016llx\n",
3396 "rsp:", save->rsp, "rax:", save->rax);
3397 pr_err("%-15s %016llx %-13s %016llx\n",
3398 "star:", save->star, "lstar:", save->lstar);
3399 pr_err("%-15s %016llx %-13s %016llx\n",
3400 "cstar:", save->cstar, "sfmask:", save->sfmask);
3401 pr_err("%-15s %016llx %-13s %016llx\n",
3402 "kernel_gs_base:", save->kernel_gs_base,
3403 "sysenter_cs:", save->sysenter_cs);
3404 pr_err("%-15s %016llx %-13s %016llx\n",
3405 "sysenter_esp:", save->sysenter_esp,
3406 "sysenter_eip:", save->sysenter_eip);
3407 pr_err("%-15s %016llx %-13s %016llx\n",
3408 "gpat:", save->g_pat, "dbgctl:", save->dbgctl);
3409 pr_err("%-15s %016llx %-13s %016llx\n",
3410 "br_from:", save->br_from, "br_to:", save->br_to);
3411 pr_err("%-15s %016llx %-13s %016llx\n",
3412 "excp_from:", save->last_excp_from,
3413 "excp_to:", save->last_excp_to);
3416 static void svm_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
3418 struct vmcb_control_area *control = &to_svm(vcpu)->vmcb->control;
3420 *info1 = control->exit_info_1;
3421 *info2 = control->exit_info_2;
3424 static int handle_exit(struct kvm_vcpu *vcpu)
3426 struct vcpu_svm *svm = to_svm(vcpu);
3427 struct kvm_run *kvm_run = vcpu->run;
3428 u32 exit_code = svm->vmcb->control.exit_code;
3430 if (!is_cr_intercept(svm, INTERCEPT_CR0_WRITE))
3431 vcpu->arch.cr0 = svm->vmcb->save.cr0;
3433 vcpu->arch.cr3 = svm->vmcb->save.cr3;
3435 if (unlikely(svm->nested.exit_required)) {
3436 nested_svm_vmexit(svm);
3437 svm->nested.exit_required = false;
3442 if (is_guest_mode(vcpu)) {
3445 trace_kvm_nested_vmexit(svm->vmcb->save.rip, exit_code,
3446 svm->vmcb->control.exit_info_1,
3447 svm->vmcb->control.exit_info_2,
3448 svm->vmcb->control.exit_int_info,
3449 svm->vmcb->control.exit_int_info_err,
3452 vmexit = nested_svm_exit_special(svm);
3454 if (vmexit == NESTED_EXIT_CONTINUE)
3455 vmexit = nested_svm_exit_handled(svm);
3457 if (vmexit == NESTED_EXIT_DONE)
3461 svm_complete_interrupts(svm);
3463 if (svm->vmcb->control.exit_code == SVM_EXIT_ERR) {
3464 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3465 kvm_run->fail_entry.hardware_entry_failure_reason
3466 = svm->vmcb->control.exit_code;
3467 pr_err("KVM: FAILED VMRUN WITH VMCB:\n");
3472 if (is_external_interrupt(svm->vmcb->control.exit_int_info) &&
3473 exit_code != SVM_EXIT_EXCP_BASE + PF_VECTOR &&
3474 exit_code != SVM_EXIT_NPF && exit_code != SVM_EXIT_TASK_SWITCH &&
3475 exit_code != SVM_EXIT_INTR && exit_code != SVM_EXIT_NMI)
3476 printk(KERN_ERR "%s: unexpected exit_int_info 0x%x "
3478 __func__, svm->vmcb->control.exit_int_info,
3481 if (exit_code >= ARRAY_SIZE(svm_exit_handlers)
3482 || !svm_exit_handlers[exit_code]) {
3483 WARN_ONCE(1, "vmx: unexpected exit reason 0x%x\n", exit_code);
3484 kvm_queue_exception(vcpu, UD_VECTOR);
3488 return svm_exit_handlers[exit_code](svm);
3491 static void reload_tss(struct kvm_vcpu *vcpu)
3493 int cpu = raw_smp_processor_id();
3495 struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
3496 sd->tss_desc->type = 9; /* available 32/64-bit TSS */
3500 static void pre_svm_run(struct vcpu_svm *svm)
3502 int cpu = raw_smp_processor_id();
3504 struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
3506 /* FIXME: handle wraparound of asid_generation */
3507 if (svm->asid_generation != sd->asid_generation)
3511 static void svm_inject_nmi(struct kvm_vcpu *vcpu)
3513 struct vcpu_svm *svm = to_svm(vcpu);
3515 svm->vmcb->control.event_inj = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI;
3516 vcpu->arch.hflags |= HF_NMI_MASK;
3517 set_intercept(svm, INTERCEPT_IRET);
3518 ++vcpu->stat.nmi_injections;
3521 static inline void svm_inject_irq(struct vcpu_svm *svm, int irq)
3523 struct vmcb_control_area *control;
3525 control = &svm->vmcb->control;
3526 control->int_vector = irq;
3527 control->int_ctl &= ~V_INTR_PRIO_MASK;
3528 control->int_ctl |= V_IRQ_MASK |
3529 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT);
3530 mark_dirty(svm->vmcb, VMCB_INTR);
3533 static void svm_set_irq(struct kvm_vcpu *vcpu)
3535 struct vcpu_svm *svm = to_svm(vcpu);
3537 BUG_ON(!(gif_set(svm)));
3539 trace_kvm_inj_virq(vcpu->arch.interrupt.nr);
3540 ++vcpu->stat.irq_injections;
3542 svm->vmcb->control.event_inj = vcpu->arch.interrupt.nr |
3543 SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR;
3546 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
3548 struct vcpu_svm *svm = to_svm(vcpu);
3550 if (is_guest_mode(vcpu) && (vcpu->arch.hflags & HF_VINTR_MASK))
3553 clr_cr_intercept(svm, INTERCEPT_CR8_WRITE);
3559 set_cr_intercept(svm, INTERCEPT_CR8_WRITE);
3562 static void svm_set_virtual_x2apic_mode(struct kvm_vcpu *vcpu, bool set)
3567 static int svm_vm_has_apicv(struct kvm *kvm)
3572 static void svm_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
3577 static void svm_hwapic_isr_update(struct kvm *kvm, int isr)
3582 static void svm_sync_pir_to_irr(struct kvm_vcpu *vcpu)
3587 static int svm_nmi_allowed(struct kvm_vcpu *vcpu)
3589 struct vcpu_svm *svm = to_svm(vcpu);
3590 struct vmcb *vmcb = svm->vmcb;
3592 ret = !(vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK) &&
3593 !(svm->vcpu.arch.hflags & HF_NMI_MASK);
3594 ret = ret && gif_set(svm) && nested_svm_nmi(svm);
3599 static bool svm_get_nmi_mask(struct kvm_vcpu *vcpu)
3601 struct vcpu_svm *svm = to_svm(vcpu);
3603 return !!(svm->vcpu.arch.hflags & HF_NMI_MASK);
3606 static void svm_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
3608 struct vcpu_svm *svm = to_svm(vcpu);
3611 svm->vcpu.arch.hflags |= HF_NMI_MASK;
3612 set_intercept(svm, INTERCEPT_IRET);
3614 svm->vcpu.arch.hflags &= ~HF_NMI_MASK;
3615 clr_intercept(svm, INTERCEPT_IRET);
3619 static int svm_interrupt_allowed(struct kvm_vcpu *vcpu)
3621 struct vcpu_svm *svm = to_svm(vcpu);
3622 struct vmcb *vmcb = svm->vmcb;
3625 if (!gif_set(svm) ||
3626 (vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK))
3629 ret = !!(kvm_get_rflags(vcpu) & X86_EFLAGS_IF);
3631 if (is_guest_mode(vcpu))
3632 return ret && !(svm->vcpu.arch.hflags & HF_VINTR_MASK);
3637 static int enable_irq_window(struct kvm_vcpu *vcpu)
3639 struct vcpu_svm *svm = to_svm(vcpu);
3642 * In case GIF=0 we can't rely on the CPU to tell us when GIF becomes
3643 * 1, because that's a separate STGI/VMRUN intercept. The next time we
3644 * get that intercept, this function will be called again though and
3645 * we'll get the vintr intercept.
3647 if (gif_set(svm) && nested_svm_intr(svm)) {
3649 svm_inject_irq(svm, 0x0);
3654 static int enable_nmi_window(struct kvm_vcpu *vcpu)
3656 struct vcpu_svm *svm = to_svm(vcpu);
3658 if ((svm->vcpu.arch.hflags & (HF_NMI_MASK | HF_IRET_MASK))
3660 return 0; /* IRET will cause a vm exit */
3663 * Something prevents NMI from been injected. Single step over possible
3664 * problem (IRET or exception injection or interrupt shadow)
3666 svm->nmi_singlestep = true;
3667 svm->vmcb->save.rflags |= (X86_EFLAGS_TF | X86_EFLAGS_RF);
3668 update_db_bp_intercept(vcpu);
3672 static int svm_set_tss_addr(struct kvm *kvm, unsigned int addr)
3677 static void svm_flush_tlb(struct kvm_vcpu *vcpu)
3679 struct vcpu_svm *svm = to_svm(vcpu);
3681 if (static_cpu_has(X86_FEATURE_FLUSHBYASID))
3682 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ASID;
3684 svm->asid_generation--;
3687 static void svm_prepare_guest_switch(struct kvm_vcpu *vcpu)
3691 static inline void sync_cr8_to_lapic(struct kvm_vcpu *vcpu)
3693 struct vcpu_svm *svm = to_svm(vcpu);
3695 if (is_guest_mode(vcpu) && (vcpu->arch.hflags & HF_VINTR_MASK))
3698 if (!is_cr_intercept(svm, INTERCEPT_CR8_WRITE)) {
3699 int cr8 = svm->vmcb->control.int_ctl & V_TPR_MASK;
3700 kvm_set_cr8(vcpu, cr8);
3704 static inline void sync_lapic_to_cr8(struct kvm_vcpu *vcpu)
3706 struct vcpu_svm *svm = to_svm(vcpu);
3709 if (is_guest_mode(vcpu) && (vcpu->arch.hflags & HF_VINTR_MASK))
3712 cr8 = kvm_get_cr8(vcpu);
3713 svm->vmcb->control.int_ctl &= ~V_TPR_MASK;
3714 svm->vmcb->control.int_ctl |= cr8 & V_TPR_MASK;
3717 static void svm_complete_interrupts(struct vcpu_svm *svm)
3721 u32 exitintinfo = svm->vmcb->control.exit_int_info;
3722 unsigned int3_injected = svm->int3_injected;
3724 svm->int3_injected = 0;
3727 * If we've made progress since setting HF_IRET_MASK, we've
3728 * executed an IRET and can allow NMI injection.
3730 if ((svm->vcpu.arch.hflags & HF_IRET_MASK)
3731 && kvm_rip_read(&svm->vcpu) != svm->nmi_iret_rip) {
3732 svm->vcpu.arch.hflags &= ~(HF_NMI_MASK | HF_IRET_MASK);
3733 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
3736 svm->vcpu.arch.nmi_injected = false;
3737 kvm_clear_exception_queue(&svm->vcpu);
3738 kvm_clear_interrupt_queue(&svm->vcpu);
3740 if (!(exitintinfo & SVM_EXITINTINFO_VALID))
3743 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
3745 vector = exitintinfo & SVM_EXITINTINFO_VEC_MASK;
3746 type = exitintinfo & SVM_EXITINTINFO_TYPE_MASK;
3749 case SVM_EXITINTINFO_TYPE_NMI:
3750 svm->vcpu.arch.nmi_injected = true;
3752 case SVM_EXITINTINFO_TYPE_EXEPT:
3754 * In case of software exceptions, do not reinject the vector,
3755 * but re-execute the instruction instead. Rewind RIP first
3756 * if we emulated INT3 before.
3758 if (kvm_exception_is_soft(vector)) {
3759 if (vector == BP_VECTOR && int3_injected &&
3760 kvm_is_linear_rip(&svm->vcpu, svm->int3_rip))
3761 kvm_rip_write(&svm->vcpu,
3762 kvm_rip_read(&svm->vcpu) -
3766 if (exitintinfo & SVM_EXITINTINFO_VALID_ERR) {
3767 u32 err = svm->vmcb->control.exit_int_info_err;
3768 kvm_requeue_exception_e(&svm->vcpu, vector, err);
3771 kvm_requeue_exception(&svm->vcpu, vector);
3773 case SVM_EXITINTINFO_TYPE_INTR:
3774 kvm_queue_interrupt(&svm->vcpu, vector, false);
3781 static void svm_cancel_injection(struct kvm_vcpu *vcpu)
3783 struct vcpu_svm *svm = to_svm(vcpu);
3784 struct vmcb_control_area *control = &svm->vmcb->control;
3786 control->exit_int_info = control->event_inj;
3787 control->exit_int_info_err = control->event_inj_err;
3788 control->event_inj = 0;
3789 svm_complete_interrupts(svm);
3792 static void svm_vcpu_run(struct kvm_vcpu *vcpu)
3794 struct vcpu_svm *svm = to_svm(vcpu);
3796 svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX];
3797 svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP];
3798 svm->vmcb->save.rip = vcpu->arch.regs[VCPU_REGS_RIP];
3801 * A vmexit emulation is required before the vcpu can be executed
3804 if (unlikely(svm->nested.exit_required))
3809 sync_lapic_to_cr8(vcpu);
3811 svm->vmcb->save.cr2 = vcpu->arch.cr2;
3818 "push %%" _ASM_BP "; \n\t"
3819 "mov %c[rbx](%[svm]), %%" _ASM_BX " \n\t"
3820 "mov %c[rcx](%[svm]), %%" _ASM_CX " \n\t"
3821 "mov %c[rdx](%[svm]), %%" _ASM_DX " \n\t"
3822 "mov %c[rsi](%[svm]), %%" _ASM_SI " \n\t"
3823 "mov %c[rdi](%[svm]), %%" _ASM_DI " \n\t"
3824 "mov %c[rbp](%[svm]), %%" _ASM_BP " \n\t"
3825 #ifdef CONFIG_X86_64
3826 "mov %c[r8](%[svm]), %%r8 \n\t"
3827 "mov %c[r9](%[svm]), %%r9 \n\t"
3828 "mov %c[r10](%[svm]), %%r10 \n\t"
3829 "mov %c[r11](%[svm]), %%r11 \n\t"
3830 "mov %c[r12](%[svm]), %%r12 \n\t"
3831 "mov %c[r13](%[svm]), %%r13 \n\t"
3832 "mov %c[r14](%[svm]), %%r14 \n\t"
3833 "mov %c[r15](%[svm]), %%r15 \n\t"
3836 /* Enter guest mode */
3837 "push %%" _ASM_AX " \n\t"
3838 "mov %c[vmcb](%[svm]), %%" _ASM_AX " \n\t"
3839 __ex(SVM_VMLOAD) "\n\t"
3840 __ex(SVM_VMRUN) "\n\t"
3841 __ex(SVM_VMSAVE) "\n\t"
3842 "pop %%" _ASM_AX " \n\t"
3844 /* Save guest registers, load host registers */
3845 "mov %%" _ASM_BX ", %c[rbx](%[svm]) \n\t"
3846 "mov %%" _ASM_CX ", %c[rcx](%[svm]) \n\t"
3847 "mov %%" _ASM_DX ", %c[rdx](%[svm]) \n\t"
3848 "mov %%" _ASM_SI ", %c[rsi](%[svm]) \n\t"
3849 "mov %%" _ASM_DI ", %c[rdi](%[svm]) \n\t"
3850 "mov %%" _ASM_BP ", %c[rbp](%[svm]) \n\t"
3851 #ifdef CONFIG_X86_64
3852 "mov %%r8, %c[r8](%[svm]) \n\t"
3853 "mov %%r9, %c[r9](%[svm]) \n\t"
3854 "mov %%r10, %c[r10](%[svm]) \n\t"
3855 "mov %%r11, %c[r11](%[svm]) \n\t"
3856 "mov %%r12, %c[r12](%[svm]) \n\t"
3857 "mov %%r13, %c[r13](%[svm]) \n\t"
3858 "mov %%r14, %c[r14](%[svm]) \n\t"
3859 "mov %%r15, %c[r15](%[svm]) \n\t"
3864 [vmcb]"i"(offsetof(struct vcpu_svm, vmcb_pa)),
3865 [rbx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBX])),
3866 [rcx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RCX])),
3867 [rdx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDX])),
3868 [rsi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RSI])),
3869 [rdi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDI])),
3870 [rbp]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBP]))
3871 #ifdef CONFIG_X86_64
3872 , [r8]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R8])),
3873 [r9]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R9])),
3874 [r10]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R10])),
3875 [r11]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R11])),
3876 [r12]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R12])),
3877 [r13]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R13])),
3878 [r14]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R14])),
3879 [r15]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R15]))
3882 #ifdef CONFIG_X86_64
3883 , "rbx", "rcx", "rdx", "rsi", "rdi"
3884 , "r8", "r9", "r10", "r11" , "r12", "r13", "r14", "r15"
3886 , "ebx", "ecx", "edx", "esi", "edi"
3890 #ifdef CONFIG_X86_64
3891 wrmsrl(MSR_GS_BASE, svm->host.gs_base);
3893 loadsegment(fs, svm->host.fs);
3894 #ifndef CONFIG_X86_32_LAZY_GS
3895 loadsegment(gs, svm->host.gs);
3901 local_irq_disable();
3903 vcpu->arch.cr2 = svm->vmcb->save.cr2;
3904 vcpu->arch.regs[VCPU_REGS_RAX] = svm->vmcb->save.rax;
3905 vcpu->arch.regs[VCPU_REGS_RSP] = svm->vmcb->save.rsp;
3906 vcpu->arch.regs[VCPU_REGS_RIP] = svm->vmcb->save.rip;
3908 trace_kvm_exit(svm->vmcb->control.exit_code, vcpu, KVM_ISA_SVM);
3910 if (unlikely(svm->vmcb->control.exit_code == SVM_EXIT_NMI))
3911 kvm_before_handle_nmi(&svm->vcpu);
3915 /* Any pending NMI will happen here */
3917 if (unlikely(svm->vmcb->control.exit_code == SVM_EXIT_NMI))
3918 kvm_after_handle_nmi(&svm->vcpu);
3920 sync_cr8_to_lapic(vcpu);
3924 svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
3926 /* if exit due to PF check for async PF */
3927 if (svm->vmcb->control.exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR)
3928 svm->apf_reason = kvm_read_and_reset_pf_reason();
3931 vcpu->arch.regs_avail &= ~(1 << VCPU_EXREG_PDPTR);
3932 vcpu->arch.regs_dirty &= ~(1 << VCPU_EXREG_PDPTR);
3936 * We need to handle MC intercepts here before the vcpu has a chance to
3937 * change the physical cpu
3939 if (unlikely(svm->vmcb->control.exit_code ==
3940 SVM_EXIT_EXCP_BASE + MC_VECTOR))
3941 svm_handle_mce(svm);
3943 mark_all_clean(svm->vmcb);
3946 static void svm_set_cr3(struct kvm_vcpu *vcpu, unsigned long root)
3948 struct vcpu_svm *svm = to_svm(vcpu);
3950 svm->vmcb->save.cr3 = root;
3951 mark_dirty(svm->vmcb, VMCB_CR);
3952 svm_flush_tlb(vcpu);
3955 static void set_tdp_cr3(struct kvm_vcpu *vcpu, unsigned long root)
3957 struct vcpu_svm *svm = to_svm(vcpu);
3959 svm->vmcb->control.nested_cr3 = root;
3960 mark_dirty(svm->vmcb, VMCB_NPT);
3962 /* Also sync guest cr3 here in case we live migrate */
3963 svm->vmcb->save.cr3 = kvm_read_cr3(vcpu);
3964 mark_dirty(svm->vmcb, VMCB_CR);
3966 svm_flush_tlb(vcpu);
3969 static int is_disabled(void)
3973 rdmsrl(MSR_VM_CR, vm_cr);
3974 if (vm_cr & (1 << SVM_VM_CR_SVM_DISABLE))
3981 svm_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
3984 * Patch in the VMMCALL instruction:
3986 hypercall[0] = 0x0f;
3987 hypercall[1] = 0x01;
3988 hypercall[2] = 0xd9;
3991 static void svm_check_processor_compat(void *rtn)
3996 static bool svm_cpu_has_accelerated_tpr(void)
4001 static u64 svm_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
4006 static void svm_cpuid_update(struct kvm_vcpu *vcpu)
4010 static void svm_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
4015 entry->ecx |= (1 << 2); /* Set SVM bit */
4018 entry->eax = 1; /* SVM revision 1 */
4019 entry->ebx = 8; /* Lets support 8 ASIDs in case we add proper
4020 ASID emulation to nested SVM */
4021 entry->ecx = 0; /* Reserved */
4022 entry->edx = 0; /* Per default do not support any
4023 additional features */
4025 /* Support next_rip if host supports it */
4026 if (boot_cpu_has(X86_FEATURE_NRIPS))
4027 entry->edx |= SVM_FEATURE_NRIP;
4029 /* Support NPT for the guest if enabled */
4031 entry->edx |= SVM_FEATURE_NPT;
4037 static int svm_get_lpage_level(void)
4039 return PT_PDPE_LEVEL;
4042 static bool svm_rdtscp_supported(void)
4047 static bool svm_invpcid_supported(void)
4052 static bool svm_has_wbinvd_exit(void)
4057 static void svm_fpu_deactivate(struct kvm_vcpu *vcpu)
4059 struct vcpu_svm *svm = to_svm(vcpu);
4061 set_exception_intercept(svm, NM_VECTOR);
4062 update_cr0_intercept(svm);
4065 #define PRE_EX(exit) { .exit_code = (exit), \
4066 .stage = X86_ICPT_PRE_EXCEPT, }
4067 #define POST_EX(exit) { .exit_code = (exit), \
4068 .stage = X86_ICPT_POST_EXCEPT, }
4069 #define POST_MEM(exit) { .exit_code = (exit), \
4070 .stage = X86_ICPT_POST_MEMACCESS, }
4072 static const struct __x86_intercept {
4074 enum x86_intercept_stage stage;
4075 } x86_intercept_map[] = {
4076 [x86_intercept_cr_read] = POST_EX(SVM_EXIT_READ_CR0),
4077 [x86_intercept_cr_write] = POST_EX(SVM_EXIT_WRITE_CR0),
4078 [x86_intercept_clts] = POST_EX(SVM_EXIT_WRITE_CR0),
4079 [x86_intercept_lmsw] = POST_EX(SVM_EXIT_WRITE_CR0),
4080 [x86_intercept_smsw] = POST_EX(SVM_EXIT_READ_CR0),
4081 [x86_intercept_dr_read] = POST_EX(SVM_EXIT_READ_DR0),
4082 [x86_intercept_dr_write] = POST_EX(SVM_EXIT_WRITE_DR0),
4083 [x86_intercept_sldt] = POST_EX(SVM_EXIT_LDTR_READ),
4084 [x86_intercept_str] = POST_EX(SVM_EXIT_TR_READ),
4085 [x86_intercept_lldt] = POST_EX(SVM_EXIT_LDTR_WRITE),
4086 [x86_intercept_ltr] = POST_EX(SVM_EXIT_TR_WRITE),
4087 [x86_intercept_sgdt] = POST_EX(SVM_EXIT_GDTR_READ),
4088 [x86_intercept_sidt] = POST_EX(SVM_EXIT_IDTR_READ),
4089 [x86_intercept_lgdt] = POST_EX(SVM_EXIT_GDTR_WRITE),
4090 [x86_intercept_lidt] = POST_EX(SVM_EXIT_IDTR_WRITE),
4091 [x86_intercept_vmrun] = POST_EX(SVM_EXIT_VMRUN),
4092 [x86_intercept_vmmcall] = POST_EX(SVM_EXIT_VMMCALL),
4093 [x86_intercept_vmload] = POST_EX(SVM_EXIT_VMLOAD),
4094 [x86_intercept_vmsave] = POST_EX(SVM_EXIT_VMSAVE),
4095 [x86_intercept_stgi] = POST_EX(SVM_EXIT_STGI),
4096 [x86_intercept_clgi] = POST_EX(SVM_EXIT_CLGI),
4097 [x86_intercept_skinit] = POST_EX(SVM_EXIT_SKINIT),
4098 [x86_intercept_invlpga] = POST_EX(SVM_EXIT_INVLPGA),
4099 [x86_intercept_rdtscp] = POST_EX(SVM_EXIT_RDTSCP),
4100 [x86_intercept_monitor] = POST_MEM(SVM_EXIT_MONITOR),
4101 [x86_intercept_mwait] = POST_EX(SVM_EXIT_MWAIT),
4102 [x86_intercept_invlpg] = POST_EX(SVM_EXIT_INVLPG),
4103 [x86_intercept_invd] = POST_EX(SVM_EXIT_INVD),
4104 [x86_intercept_wbinvd] = POST_EX(SVM_EXIT_WBINVD),
4105 [x86_intercept_wrmsr] = POST_EX(SVM_EXIT_MSR),
4106 [x86_intercept_rdtsc] = POST_EX(SVM_EXIT_RDTSC),
4107 [x86_intercept_rdmsr] = POST_EX(SVM_EXIT_MSR),
4108 [x86_intercept_rdpmc] = POST_EX(SVM_EXIT_RDPMC),
4109 [x86_intercept_cpuid] = PRE_EX(SVM_EXIT_CPUID),
4110 [x86_intercept_rsm] = PRE_EX(SVM_EXIT_RSM),
4111 [x86_intercept_pause] = PRE_EX(SVM_EXIT_PAUSE),
4112 [x86_intercept_pushf] = PRE_EX(SVM_EXIT_PUSHF),
4113 [x86_intercept_popf] = PRE_EX(SVM_EXIT_POPF),
4114 [x86_intercept_intn] = PRE_EX(SVM_EXIT_SWINT),
4115 [x86_intercept_iret] = PRE_EX(SVM_EXIT_IRET),
4116 [x86_intercept_icebp] = PRE_EX(SVM_EXIT_ICEBP),
4117 [x86_intercept_hlt] = POST_EX(SVM_EXIT_HLT),
4118 [x86_intercept_in] = POST_EX(SVM_EXIT_IOIO),
4119 [x86_intercept_ins] = POST_EX(SVM_EXIT_IOIO),
4120 [x86_intercept_out] = POST_EX(SVM_EXIT_IOIO),
4121 [x86_intercept_outs] = POST_EX(SVM_EXIT_IOIO),
4128 static int svm_check_intercept(struct kvm_vcpu *vcpu,
4129 struct x86_instruction_info *info,
4130 enum x86_intercept_stage stage)
4132 struct vcpu_svm *svm = to_svm(vcpu);
4133 int vmexit, ret = X86EMUL_CONTINUE;
4134 struct __x86_intercept icpt_info;
4135 struct vmcb *vmcb = svm->vmcb;
4137 if (info->intercept >= ARRAY_SIZE(x86_intercept_map))
4140 icpt_info = x86_intercept_map[info->intercept];
4142 if (stage != icpt_info.stage)
4145 switch (icpt_info.exit_code) {
4146 case SVM_EXIT_READ_CR0:
4147 if (info->intercept == x86_intercept_cr_read)
4148 icpt_info.exit_code += info->modrm_reg;
4150 case SVM_EXIT_WRITE_CR0: {
4151 unsigned long cr0, val;
4154 if (info->intercept == x86_intercept_cr_write)
4155 icpt_info.exit_code += info->modrm_reg;
4157 if (icpt_info.exit_code != SVM_EXIT_WRITE_CR0)
4160 intercept = svm->nested.intercept;
4162 if (!(intercept & (1ULL << INTERCEPT_SELECTIVE_CR0)))
4165 cr0 = vcpu->arch.cr0 & ~SVM_CR0_SELECTIVE_MASK;
4166 val = info->src_val & ~SVM_CR0_SELECTIVE_MASK;
4168 if (info->intercept == x86_intercept_lmsw) {
4171 /* lmsw can't clear PE - catch this here */
4172 if (cr0 & X86_CR0_PE)
4177 icpt_info.exit_code = SVM_EXIT_CR0_SEL_WRITE;
4181 case SVM_EXIT_READ_DR0:
4182 case SVM_EXIT_WRITE_DR0:
4183 icpt_info.exit_code += info->modrm_reg;
4186 if (info->intercept == x86_intercept_wrmsr)
4187 vmcb->control.exit_info_1 = 1;
4189 vmcb->control.exit_info_1 = 0;
4191 case SVM_EXIT_PAUSE:
4193 * We get this for NOP only, but pause
4194 * is rep not, check this here
4196 if (info->rep_prefix != REPE_PREFIX)
4198 case SVM_EXIT_IOIO: {
4202 exit_info = (vcpu->arch.regs[VCPU_REGS_RDX] & 0xffff) << 16;
4204 if (info->intercept == x86_intercept_in ||
4205 info->intercept == x86_intercept_ins) {
4206 exit_info |= SVM_IOIO_TYPE_MASK;
4207 bytes = info->src_bytes;
4209 bytes = info->dst_bytes;
4212 if (info->intercept == x86_intercept_outs ||
4213 info->intercept == x86_intercept_ins)
4214 exit_info |= SVM_IOIO_STR_MASK;
4216 if (info->rep_prefix)
4217 exit_info |= SVM_IOIO_REP_MASK;
4219 bytes = min(bytes, 4u);
4221 exit_info |= bytes << SVM_IOIO_SIZE_SHIFT;
4223 exit_info |= (u32)info->ad_bytes << (SVM_IOIO_ASIZE_SHIFT - 1);
4225 vmcb->control.exit_info_1 = exit_info;
4226 vmcb->control.exit_info_2 = info->next_rip;
4234 /* TODO: Advertise NRIPS to guest hypervisor unconditionally */
4235 if (static_cpu_has(X86_FEATURE_NRIPS))
4236 vmcb->control.next_rip = info->next_rip;
4237 vmcb->control.exit_code = icpt_info.exit_code;
4238 vmexit = nested_svm_exit_handled(svm);
4240 ret = (vmexit == NESTED_EXIT_DONE) ? X86EMUL_INTERCEPTED
4247 static void svm_handle_external_intr(struct kvm_vcpu *vcpu)
4252 static struct kvm_x86_ops svm_x86_ops = {
4253 .cpu_has_kvm_support = has_svm,
4254 .disabled_by_bios = is_disabled,
4255 .hardware_setup = svm_hardware_setup,
4256 .hardware_unsetup = svm_hardware_unsetup,
4257 .check_processor_compatibility = svm_check_processor_compat,
4258 .hardware_enable = svm_hardware_enable,
4259 .hardware_disable = svm_hardware_disable,
4260 .cpu_has_accelerated_tpr = svm_cpu_has_accelerated_tpr,
4262 .vcpu_create = svm_create_vcpu,
4263 .vcpu_free = svm_free_vcpu,
4264 .vcpu_reset = svm_vcpu_reset,
4266 .prepare_guest_switch = svm_prepare_guest_switch,
4267 .vcpu_load = svm_vcpu_load,
4268 .vcpu_put = svm_vcpu_put,
4270 .update_db_bp_intercept = update_db_bp_intercept,
4271 .get_msr = svm_get_msr,
4272 .set_msr = svm_set_msr,
4273 .get_segment_base = svm_get_segment_base,
4274 .get_segment = svm_get_segment,
4275 .set_segment = svm_set_segment,
4276 .get_cpl = svm_get_cpl,
4277 .get_cs_db_l_bits = kvm_get_cs_db_l_bits,
4278 .decache_cr0_guest_bits = svm_decache_cr0_guest_bits,
4279 .decache_cr3 = svm_decache_cr3,
4280 .decache_cr4_guest_bits = svm_decache_cr4_guest_bits,
4281 .set_cr0 = svm_set_cr0,
4282 .set_cr3 = svm_set_cr3,
4283 .set_cr4 = svm_set_cr4,
4284 .set_efer = svm_set_efer,
4285 .get_idt = svm_get_idt,
4286 .set_idt = svm_set_idt,
4287 .get_gdt = svm_get_gdt,
4288 .set_gdt = svm_set_gdt,
4289 .set_dr7 = svm_set_dr7,
4290 .cache_reg = svm_cache_reg,
4291 .get_rflags = svm_get_rflags,
4292 .set_rflags = svm_set_rflags,
4293 .fpu_activate = svm_fpu_activate,
4294 .fpu_deactivate = svm_fpu_deactivate,
4296 .tlb_flush = svm_flush_tlb,
4298 .run = svm_vcpu_run,
4299 .handle_exit = handle_exit,
4300 .skip_emulated_instruction = skip_emulated_instruction,
4301 .set_interrupt_shadow = svm_set_interrupt_shadow,
4302 .get_interrupt_shadow = svm_get_interrupt_shadow,
4303 .patch_hypercall = svm_patch_hypercall,
4304 .set_irq = svm_set_irq,
4305 .set_nmi = svm_inject_nmi,
4306 .queue_exception = svm_queue_exception,
4307 .cancel_injection = svm_cancel_injection,
4308 .interrupt_allowed = svm_interrupt_allowed,
4309 .nmi_allowed = svm_nmi_allowed,
4310 .get_nmi_mask = svm_get_nmi_mask,
4311 .set_nmi_mask = svm_set_nmi_mask,
4312 .enable_nmi_window = enable_nmi_window,
4313 .enable_irq_window = enable_irq_window,
4314 .update_cr8_intercept = update_cr8_intercept,
4315 .set_virtual_x2apic_mode = svm_set_virtual_x2apic_mode,
4316 .vm_has_apicv = svm_vm_has_apicv,
4317 .load_eoi_exitmap = svm_load_eoi_exitmap,
4318 .hwapic_isr_update = svm_hwapic_isr_update,
4319 .sync_pir_to_irr = svm_sync_pir_to_irr,
4321 .set_tss_addr = svm_set_tss_addr,
4322 .get_tdp_level = get_npt_level,
4323 .get_mt_mask = svm_get_mt_mask,
4325 .get_exit_info = svm_get_exit_info,
4327 .get_lpage_level = svm_get_lpage_level,
4329 .cpuid_update = svm_cpuid_update,
4331 .rdtscp_supported = svm_rdtscp_supported,
4332 .invpcid_supported = svm_invpcid_supported,
4334 .set_supported_cpuid = svm_set_supported_cpuid,
4336 .has_wbinvd_exit = svm_has_wbinvd_exit,
4338 .set_tsc_khz = svm_set_tsc_khz,
4339 .read_tsc_offset = svm_read_tsc_offset,
4340 .write_tsc_offset = svm_write_tsc_offset,
4341 .adjust_tsc_offset = svm_adjust_tsc_offset,
4342 .compute_tsc_offset = svm_compute_tsc_offset,
4343 .read_l1_tsc = svm_read_l1_tsc,
4345 .set_tdp_cr3 = set_tdp_cr3,
4347 .check_intercept = svm_check_intercept,
4348 .handle_external_intr = svm_handle_external_intr,
4351 static int __init svm_init(void)
4353 return kvm_init(&svm_x86_ops, sizeof(struct vcpu_svm),
4354 __alignof__(struct vcpu_svm), THIS_MODULE);
4357 static void __exit svm_exit(void)
4362 module_init(svm_init)
4363 module_exit(svm_exit)