2 * Kernel-based Virtual Machine driver for Linux
6 * Copyright (C) 2006 Qumranet, Inc.
7 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
10 * Yaniv Kamay <yaniv@qumranet.com>
11 * Avi Kivity <avi@qumranet.com>
13 * This work is licensed under the terms of the GNU GPL, version 2. See
14 * the COPYING file in the top-level directory.
18 #define pr_fmt(fmt) "SVM: " fmt
20 #include <linux/kvm_host.h>
24 #include "kvm_cache_regs.h"
29 #include <linux/module.h>
30 #include <linux/mod_devicetable.h>
31 #include <linux/kernel.h>
32 #include <linux/vmalloc.h>
33 #include <linux/highmem.h>
34 #include <linux/sched.h>
35 #include <linux/trace_events.h>
36 #include <linux/slab.h>
37 #include <linux/amd-iommu.h>
38 #include <linux/hashtable.h>
39 #include <linux/frame.h>
40 #include <linux/psp-sev.h>
41 #include <linux/file.h>
42 #include <linux/pagemap.h>
43 #include <linux/swap.h>
46 #include <asm/perf_event.h>
47 #include <asm/tlbflush.h>
49 #include <asm/debugreg.h>
50 #include <asm/kvm_para.h>
51 #include <asm/irq_remapping.h>
52 #include <asm/spec-ctrl.h>
54 #include <asm/virtext.h>
57 #define __ex(x) __kvm_handle_fault_on_reboot(x)
59 MODULE_AUTHOR("Qumranet");
60 MODULE_LICENSE("GPL");
62 static const struct x86_cpu_id svm_cpu_id[] = {
63 X86_FEATURE_MATCH(X86_FEATURE_SVM),
66 MODULE_DEVICE_TABLE(x86cpu, svm_cpu_id);
68 #define IOPM_ALLOC_ORDER 2
69 #define MSRPM_ALLOC_ORDER 1
71 #define SEG_TYPE_LDT 2
72 #define SEG_TYPE_BUSY_TSS16 3
74 #define SVM_FEATURE_NPT (1 << 0)
75 #define SVM_FEATURE_LBRV (1 << 1)
76 #define SVM_FEATURE_SVML (1 << 2)
77 #define SVM_FEATURE_NRIP (1 << 3)
78 #define SVM_FEATURE_TSC_RATE (1 << 4)
79 #define SVM_FEATURE_VMCB_CLEAN (1 << 5)
80 #define SVM_FEATURE_FLUSH_ASID (1 << 6)
81 #define SVM_FEATURE_DECODE_ASSIST (1 << 7)
82 #define SVM_FEATURE_PAUSE_FILTER (1 << 10)
84 #define SVM_AVIC_DOORBELL 0xc001011b
86 #define NESTED_EXIT_HOST 0 /* Exit handled on host level */
87 #define NESTED_EXIT_DONE 1 /* Exit caused nested vmexit */
88 #define NESTED_EXIT_CONTINUE 2 /* Further checks needed */
90 #define DEBUGCTL_RESERVED_BITS (~(0x3fULL))
92 #define TSC_RATIO_RSVD 0xffffff0000000000ULL
93 #define TSC_RATIO_MIN 0x0000000000000001ULL
94 #define TSC_RATIO_MAX 0x000000ffffffffffULL
96 #define AVIC_HPA_MASK ~((0xFFFULL << 52) | 0xFFF)
99 * 0xff is broadcast, so the max index allowed for physical APIC ID
100 * table is 0xfe. APIC IDs above 0xff are reserved.
102 #define AVIC_MAX_PHYSICAL_ID_COUNT 255
104 #define AVIC_UNACCEL_ACCESS_WRITE_MASK 1
105 #define AVIC_UNACCEL_ACCESS_OFFSET_MASK 0xFF0
106 #define AVIC_UNACCEL_ACCESS_VECTOR_MASK 0xFFFFFFFF
108 /* AVIC GATAG is encoded using VM and VCPU IDs */
109 #define AVIC_VCPU_ID_BITS 8
110 #define AVIC_VCPU_ID_MASK ((1 << AVIC_VCPU_ID_BITS) - 1)
112 #define AVIC_VM_ID_BITS 24
113 #define AVIC_VM_ID_NR (1 << AVIC_VM_ID_BITS)
114 #define AVIC_VM_ID_MASK ((1 << AVIC_VM_ID_BITS) - 1)
116 #define AVIC_GATAG(x, y) (((x & AVIC_VM_ID_MASK) << AVIC_VCPU_ID_BITS) | \
117 (y & AVIC_VCPU_ID_MASK))
118 #define AVIC_GATAG_TO_VMID(x) ((x >> AVIC_VCPU_ID_BITS) & AVIC_VM_ID_MASK)
119 #define AVIC_GATAG_TO_VCPUID(x) (x & AVIC_VCPU_ID_MASK)
121 static bool erratum_383_found __read_mostly;
123 static const u32 host_save_user_msrs[] = {
125 MSR_STAR, MSR_LSTAR, MSR_CSTAR, MSR_SYSCALL_MASK, MSR_KERNEL_GS_BASE,
128 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
132 #define NR_HOST_SAVE_USER_MSRS ARRAY_SIZE(host_save_user_msrs)
134 struct kvm_sev_info {
135 bool active; /* SEV enabled guest */
136 unsigned int asid; /* ASID used for this guest */
137 unsigned int handle; /* SEV firmware handle */
138 int fd; /* SEV device fd */
139 unsigned long pages_locked; /* Number of pages locked */
140 struct list_head regions_list; /* List of registered regions */
146 /* Struct members for AVIC */
148 struct page *avic_logical_id_table_page;
149 struct page *avic_physical_id_table_page;
150 struct hlist_node hnode;
152 struct kvm_sev_info sev_info;
157 struct nested_state {
163 /* These are the merged vectors */
166 /* gpa pointers to the real vectors */
170 /* A VMEXIT is required but not yet emulated */
173 /* cache for intercepts of the guest */
176 u32 intercept_exceptions;
179 /* Nested Paging related state */
183 #define MSRPM_OFFSETS 16
184 static u32 msrpm_offsets[MSRPM_OFFSETS] __read_mostly;
187 * Set osvw_len to higher value when updated Revision Guides
188 * are published and we know what the new status bits are
190 static uint64_t osvw_len = 4, osvw_status;
193 struct kvm_vcpu vcpu;
195 unsigned long vmcb_pa;
196 struct svm_cpu_data *svm_data;
197 uint64_t asid_generation;
198 uint64_t sysenter_esp;
199 uint64_t sysenter_eip;
206 u64 host_user_msrs[NR_HOST_SAVE_USER_MSRS];
216 * Contains guest-controlled bits of VIRT_SPEC_CTRL, which will be
217 * translated into the appropriate L2_CFG bits on the host to
218 * perform speculative control.
226 struct nested_state nested;
229 u64 nmi_singlestep_guest_rflags;
231 unsigned int3_injected;
232 unsigned long int3_rip;
234 /* cached guest cpuid flags for faster access */
235 bool nrips_enabled : 1;
239 struct page *avic_backing_page;
240 u64 *avic_physical_id_cache;
241 bool avic_is_running;
244 * Per-vcpu list of struct amd_svm_iommu_ir:
245 * This is used mainly to store interrupt remapping information used
246 * when update the vcpu affinity. This avoids the need to scan for
247 * IRTE and try to match ga_tag in the IOMMU driver.
249 struct list_head ir_list;
250 spinlock_t ir_list_lock;
252 /* which host CPU was used for running this vcpu */
253 unsigned int last_cpu;
257 * This is a wrapper of struct amd_iommu_ir_data.
259 struct amd_svm_iommu_ir {
260 struct list_head node; /* Used by SVM for per-vcpu ir_list */
261 void *data; /* Storing pointer to struct amd_ir_data */
264 #define AVIC_LOGICAL_ID_ENTRY_GUEST_PHYSICAL_ID_MASK (0xFF)
265 #define AVIC_LOGICAL_ID_ENTRY_VALID_BIT 31
266 #define AVIC_LOGICAL_ID_ENTRY_VALID_MASK (1 << 31)
268 #define AVIC_PHYSICAL_ID_ENTRY_HOST_PHYSICAL_ID_MASK (0xFFULL)
269 #define AVIC_PHYSICAL_ID_ENTRY_BACKING_PAGE_MASK (0xFFFFFFFFFFULL << 12)
270 #define AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK (1ULL << 62)
271 #define AVIC_PHYSICAL_ID_ENTRY_VALID_MASK (1ULL << 63)
273 static DEFINE_PER_CPU(u64, current_tsc_ratio);
274 #define TSC_RATIO_DEFAULT 0x0100000000ULL
276 #define MSR_INVALID 0xffffffffU
278 static const struct svm_direct_access_msrs {
279 u32 index; /* Index of the MSR */
280 bool always; /* True if intercept is always on */
281 } direct_access_msrs[] = {
282 { .index = MSR_STAR, .always = true },
283 { .index = MSR_IA32_SYSENTER_CS, .always = true },
285 { .index = MSR_GS_BASE, .always = true },
286 { .index = MSR_FS_BASE, .always = true },
287 { .index = MSR_KERNEL_GS_BASE, .always = true },
288 { .index = MSR_LSTAR, .always = true },
289 { .index = MSR_CSTAR, .always = true },
290 { .index = MSR_SYSCALL_MASK, .always = true },
292 { .index = MSR_IA32_SPEC_CTRL, .always = false },
293 { .index = MSR_IA32_PRED_CMD, .always = false },
294 { .index = MSR_IA32_LASTBRANCHFROMIP, .always = false },
295 { .index = MSR_IA32_LASTBRANCHTOIP, .always = false },
296 { .index = MSR_IA32_LASTINTFROMIP, .always = false },
297 { .index = MSR_IA32_LASTINTTOIP, .always = false },
298 { .index = MSR_INVALID, .always = false },
301 /* enable NPT for AMD64 and X86 with PAE */
302 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
303 static bool npt_enabled = true;
305 static bool npt_enabled;
309 * These 2 parameters are used to config the controls for Pause-Loop Exiting:
310 * pause_filter_count: On processors that support Pause filtering(indicated
311 * by CPUID Fn8000_000A_EDX), the VMCB provides a 16 bit pause filter
312 * count value. On VMRUN this value is loaded into an internal counter.
313 * Each time a pause instruction is executed, this counter is decremented
314 * until it reaches zero at which time a #VMEXIT is generated if pause
315 * intercept is enabled. Refer to AMD APM Vol 2 Section 15.14.4 Pause
316 * Intercept Filtering for more details.
317 * This also indicate if ple logic enabled.
319 * pause_filter_thresh: In addition, some processor families support advanced
320 * pause filtering (indicated by CPUID Fn8000_000A_EDX) upper bound on
321 * the amount of time a guest is allowed to execute in a pause loop.
322 * In this mode, a 16-bit pause filter threshold field is added in the
323 * VMCB. The threshold value is a cycle count that is used to reset the
324 * pause counter. As with simple pause filtering, VMRUN loads the pause
325 * count value from VMCB into an internal counter. Then, on each pause
326 * instruction the hardware checks the elapsed number of cycles since
327 * the most recent pause instruction against the pause filter threshold.
328 * If the elapsed cycle count is greater than the pause filter threshold,
329 * then the internal pause count is reloaded from the VMCB and execution
330 * continues. If the elapsed cycle count is less than the pause filter
331 * threshold, then the internal pause count is decremented. If the count
332 * value is less than zero and PAUSE intercept is enabled, a #VMEXIT is
333 * triggered. If advanced pause filtering is supported and pause filter
334 * threshold field is set to zero, the filter will operate in the simpler,
338 static unsigned short pause_filter_thresh = KVM_DEFAULT_PLE_GAP;
339 module_param(pause_filter_thresh, ushort, 0444);
341 static unsigned short pause_filter_count = KVM_SVM_DEFAULT_PLE_WINDOW;
342 module_param(pause_filter_count, ushort, 0444);
344 /* Default doubles per-vcpu window every exit. */
345 static unsigned short pause_filter_count_grow = KVM_DEFAULT_PLE_WINDOW_GROW;
346 module_param(pause_filter_count_grow, ushort, 0444);
348 /* Default resets per-vcpu window every exit to pause_filter_count. */
349 static unsigned short pause_filter_count_shrink = KVM_DEFAULT_PLE_WINDOW_SHRINK;
350 module_param(pause_filter_count_shrink, ushort, 0444);
352 /* Default is to compute the maximum so we can never overflow. */
353 static unsigned short pause_filter_count_max = KVM_SVM_DEFAULT_PLE_WINDOW_MAX;
354 module_param(pause_filter_count_max, ushort, 0444);
356 /* allow nested paging (virtualized MMU) for all guests */
357 static int npt = true;
358 module_param(npt, int, S_IRUGO);
360 /* allow nested virtualization in KVM/SVM */
361 static int nested = true;
362 module_param(nested, int, S_IRUGO);
364 /* enable / disable AVIC */
366 #ifdef CONFIG_X86_LOCAL_APIC
367 module_param(avic, int, S_IRUGO);
370 /* enable/disable Virtual VMLOAD VMSAVE */
371 static int vls = true;
372 module_param(vls, int, 0444);
374 /* enable/disable Virtual GIF */
375 static int vgif = true;
376 module_param(vgif, int, 0444);
378 /* enable/disable SEV support */
379 static int sev = IS_ENABLED(CONFIG_AMD_MEM_ENCRYPT_ACTIVE_BY_DEFAULT);
380 module_param(sev, int, 0444);
382 static bool __read_mostly dump_invalid_vmcb = 0;
383 module_param(dump_invalid_vmcb, bool, 0644);
385 static u8 rsm_ins_bytes[] = "\x0f\xaa";
387 static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0);
388 static void svm_flush_tlb(struct kvm_vcpu *vcpu, bool invalidate_gpa);
389 static void svm_complete_interrupts(struct vcpu_svm *svm);
391 static int nested_svm_exit_handled(struct vcpu_svm *svm);
392 static int nested_svm_intercept(struct vcpu_svm *svm);
393 static int nested_svm_vmexit(struct vcpu_svm *svm);
394 static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
395 bool has_error_code, u32 error_code);
398 VMCB_INTERCEPTS, /* Intercept vectors, TSC offset,
399 pause filter count */
400 VMCB_PERM_MAP, /* IOPM Base and MSRPM Base */
401 VMCB_ASID, /* ASID */
402 VMCB_INTR, /* int_ctl, int_vector */
403 VMCB_NPT, /* npt_en, nCR3, gPAT */
404 VMCB_CR, /* CR0, CR3, CR4, EFER */
405 VMCB_DR, /* DR6, DR7 */
406 VMCB_DT, /* GDT, IDT */
407 VMCB_SEG, /* CS, DS, SS, ES, CPL */
408 VMCB_CR2, /* CR2 only */
409 VMCB_LBR, /* DBGCTL, BR_FROM, BR_TO, LAST_EX_FROM, LAST_EX_TO */
410 VMCB_AVIC, /* AVIC APIC_BAR, AVIC APIC_BACKING_PAGE,
411 * AVIC PHYSICAL_TABLE pointer,
412 * AVIC LOGICAL_TABLE pointer
417 /* TPR and CR2 are always written before VMRUN */
418 #define VMCB_ALWAYS_DIRTY_MASK ((1U << VMCB_INTR) | (1U << VMCB_CR2))
420 #define VMCB_AVIC_APIC_BAR_MASK 0xFFFFFFFFFF000ULL
422 static unsigned int max_sev_asid;
423 static unsigned int min_sev_asid;
424 static unsigned long *sev_asid_bitmap;
425 #define __sme_page_pa(x) __sme_set(page_to_pfn(x) << PAGE_SHIFT)
428 struct list_head list;
429 unsigned long npages;
436 static inline struct kvm_svm *to_kvm_svm(struct kvm *kvm)
438 return container_of(kvm, struct kvm_svm, kvm);
441 static inline bool svm_sev_enabled(void)
443 return IS_ENABLED(CONFIG_KVM_AMD_SEV) ? max_sev_asid : 0;
446 static inline bool sev_guest(struct kvm *kvm)
448 #ifdef CONFIG_KVM_AMD_SEV
449 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
457 static inline int sev_get_asid(struct kvm *kvm)
459 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
464 static inline void mark_all_dirty(struct vmcb *vmcb)
466 vmcb->control.clean = 0;
469 static inline void mark_all_clean(struct vmcb *vmcb)
471 vmcb->control.clean = ((1 << VMCB_DIRTY_MAX) - 1)
472 & ~VMCB_ALWAYS_DIRTY_MASK;
475 static inline void mark_dirty(struct vmcb *vmcb, int bit)
477 vmcb->control.clean &= ~(1 << bit);
480 static inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu)
482 return container_of(vcpu, struct vcpu_svm, vcpu);
485 static inline void avic_update_vapic_bar(struct vcpu_svm *svm, u64 data)
487 svm->vmcb->control.avic_vapic_bar = data & VMCB_AVIC_APIC_BAR_MASK;
488 mark_dirty(svm->vmcb, VMCB_AVIC);
491 static inline bool avic_vcpu_is_running(struct kvm_vcpu *vcpu)
493 struct vcpu_svm *svm = to_svm(vcpu);
494 u64 *entry = svm->avic_physical_id_cache;
499 return (READ_ONCE(*entry) & AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK);
502 static void recalc_intercepts(struct vcpu_svm *svm)
504 struct vmcb_control_area *c, *h;
505 struct nested_state *g;
507 mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
509 if (!is_guest_mode(&svm->vcpu))
512 c = &svm->vmcb->control;
513 h = &svm->nested.hsave->control;
516 c->intercept_cr = h->intercept_cr | g->intercept_cr;
517 c->intercept_dr = h->intercept_dr | g->intercept_dr;
518 c->intercept_exceptions = h->intercept_exceptions | g->intercept_exceptions;
519 c->intercept = h->intercept | g->intercept;
522 static inline struct vmcb *get_host_vmcb(struct vcpu_svm *svm)
524 if (is_guest_mode(&svm->vcpu))
525 return svm->nested.hsave;
530 static inline void set_cr_intercept(struct vcpu_svm *svm, int bit)
532 struct vmcb *vmcb = get_host_vmcb(svm);
534 vmcb->control.intercept_cr |= (1U << bit);
536 recalc_intercepts(svm);
539 static inline void clr_cr_intercept(struct vcpu_svm *svm, int bit)
541 struct vmcb *vmcb = get_host_vmcb(svm);
543 vmcb->control.intercept_cr &= ~(1U << bit);
545 recalc_intercepts(svm);
548 static inline bool is_cr_intercept(struct vcpu_svm *svm, int bit)
550 struct vmcb *vmcb = get_host_vmcb(svm);
552 return vmcb->control.intercept_cr & (1U << bit);
555 static inline void set_dr_intercepts(struct vcpu_svm *svm)
557 struct vmcb *vmcb = get_host_vmcb(svm);
559 vmcb->control.intercept_dr = (1 << INTERCEPT_DR0_READ)
560 | (1 << INTERCEPT_DR1_READ)
561 | (1 << INTERCEPT_DR2_READ)
562 | (1 << INTERCEPT_DR3_READ)
563 | (1 << INTERCEPT_DR4_READ)
564 | (1 << INTERCEPT_DR5_READ)
565 | (1 << INTERCEPT_DR6_READ)
566 | (1 << INTERCEPT_DR7_READ)
567 | (1 << INTERCEPT_DR0_WRITE)
568 | (1 << INTERCEPT_DR1_WRITE)
569 | (1 << INTERCEPT_DR2_WRITE)
570 | (1 << INTERCEPT_DR3_WRITE)
571 | (1 << INTERCEPT_DR4_WRITE)
572 | (1 << INTERCEPT_DR5_WRITE)
573 | (1 << INTERCEPT_DR6_WRITE)
574 | (1 << INTERCEPT_DR7_WRITE);
576 recalc_intercepts(svm);
579 static inline void clr_dr_intercepts(struct vcpu_svm *svm)
581 struct vmcb *vmcb = get_host_vmcb(svm);
583 vmcb->control.intercept_dr = 0;
585 recalc_intercepts(svm);
588 static inline void set_exception_intercept(struct vcpu_svm *svm, int bit)
590 struct vmcb *vmcb = get_host_vmcb(svm);
592 vmcb->control.intercept_exceptions |= (1U << bit);
594 recalc_intercepts(svm);
597 static inline void clr_exception_intercept(struct vcpu_svm *svm, int bit)
599 struct vmcb *vmcb = get_host_vmcb(svm);
601 vmcb->control.intercept_exceptions &= ~(1U << bit);
603 recalc_intercepts(svm);
606 static inline void set_intercept(struct vcpu_svm *svm, int bit)
608 struct vmcb *vmcb = get_host_vmcb(svm);
610 vmcb->control.intercept |= (1ULL << bit);
612 recalc_intercepts(svm);
615 static inline void clr_intercept(struct vcpu_svm *svm, int bit)
617 struct vmcb *vmcb = get_host_vmcb(svm);
619 vmcb->control.intercept &= ~(1ULL << bit);
621 recalc_intercepts(svm);
624 static inline bool vgif_enabled(struct vcpu_svm *svm)
626 return !!(svm->vmcb->control.int_ctl & V_GIF_ENABLE_MASK);
629 static inline void enable_gif(struct vcpu_svm *svm)
631 if (vgif_enabled(svm))
632 svm->vmcb->control.int_ctl |= V_GIF_MASK;
634 svm->vcpu.arch.hflags |= HF_GIF_MASK;
637 static inline void disable_gif(struct vcpu_svm *svm)
639 if (vgif_enabled(svm))
640 svm->vmcb->control.int_ctl &= ~V_GIF_MASK;
642 svm->vcpu.arch.hflags &= ~HF_GIF_MASK;
645 static inline bool gif_set(struct vcpu_svm *svm)
647 if (vgif_enabled(svm))
648 return !!(svm->vmcb->control.int_ctl & V_GIF_MASK);
650 return !!(svm->vcpu.arch.hflags & HF_GIF_MASK);
653 static unsigned long iopm_base;
655 struct kvm_ldttss_desc {
658 unsigned base1:8, type:5, dpl:2, p:1;
659 unsigned limit1:4, zero0:3, g:1, base2:8;
662 } __attribute__((packed));
664 struct svm_cpu_data {
671 struct kvm_ldttss_desc *tss_desc;
673 struct page *save_area;
674 struct vmcb *current_vmcb;
676 /* index = sev_asid, value = vmcb pointer */
677 struct vmcb **sev_vmcbs;
680 static DEFINE_PER_CPU(struct svm_cpu_data *, svm_data);
682 static const u32 msrpm_ranges[] = {0, 0xc0000000, 0xc0010000};
684 #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
685 #define MSRS_RANGE_SIZE 2048
686 #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
688 static u32 svm_msrpm_offset(u32 msr)
693 for (i = 0; i < NUM_MSR_MAPS; i++) {
694 if (msr < msrpm_ranges[i] ||
695 msr >= msrpm_ranges[i] + MSRS_IN_RANGE)
698 offset = (msr - msrpm_ranges[i]) / 4; /* 4 msrs per u8 */
699 offset += (i * MSRS_RANGE_SIZE); /* add range offset */
701 /* Now we have the u8 offset - but need the u32 offset */
705 /* MSR not in any range */
709 #define MAX_INST_SIZE 15
711 static inline void clgi(void)
713 asm volatile (__ex("clgi"));
716 static inline void stgi(void)
718 asm volatile (__ex("stgi"));
721 static inline void invlpga(unsigned long addr, u32 asid)
723 asm volatile (__ex("invlpga %1, %0") : : "c"(asid), "a"(addr));
726 static int get_npt_level(struct kvm_vcpu *vcpu)
729 return PT64_ROOT_4LEVEL;
731 return PT32E_ROOT_LEVEL;
735 static void svm_set_efer(struct kvm_vcpu *vcpu, u64 efer)
737 vcpu->arch.efer = efer;
738 if (!npt_enabled && !(efer & EFER_LMA))
741 to_svm(vcpu)->vmcb->save.efer = efer | EFER_SVME;
742 mark_dirty(to_svm(vcpu)->vmcb, VMCB_CR);
745 static int is_external_interrupt(u32 info)
747 info &= SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
748 return info == (SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR);
751 static u32 svm_get_interrupt_shadow(struct kvm_vcpu *vcpu)
753 struct vcpu_svm *svm = to_svm(vcpu);
756 if (svm->vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK)
757 ret = KVM_X86_SHADOW_INT_STI | KVM_X86_SHADOW_INT_MOV_SS;
761 static void svm_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
763 struct vcpu_svm *svm = to_svm(vcpu);
766 svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK;
768 svm->vmcb->control.int_state |= SVM_INTERRUPT_SHADOW_MASK;
772 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
774 struct vcpu_svm *svm = to_svm(vcpu);
776 if (svm->vmcb->control.next_rip != 0) {
777 WARN_ON_ONCE(!static_cpu_has(X86_FEATURE_NRIPS));
778 svm->next_rip = svm->vmcb->control.next_rip;
781 if (!svm->next_rip) {
782 if (kvm_emulate_instruction(vcpu, EMULTYPE_SKIP) !=
784 printk(KERN_DEBUG "%s: NOP\n", __func__);
787 if (svm->next_rip - kvm_rip_read(vcpu) > MAX_INST_SIZE)
788 printk(KERN_ERR "%s: ip 0x%lx next 0x%llx\n",
789 __func__, kvm_rip_read(vcpu), svm->next_rip);
791 kvm_rip_write(vcpu, svm->next_rip);
792 svm_set_interrupt_shadow(vcpu, 0);
795 static void svm_queue_exception(struct kvm_vcpu *vcpu)
797 struct vcpu_svm *svm = to_svm(vcpu);
798 unsigned nr = vcpu->arch.exception.nr;
799 bool has_error_code = vcpu->arch.exception.has_error_code;
800 bool reinject = vcpu->arch.exception.injected;
801 u32 error_code = vcpu->arch.exception.error_code;
804 * If we are within a nested VM we'd better #VMEXIT and let the guest
805 * handle the exception
808 nested_svm_check_exception(svm, nr, has_error_code, error_code))
811 kvm_deliver_exception_payload(&svm->vcpu);
813 if (nr == BP_VECTOR && !static_cpu_has(X86_FEATURE_NRIPS)) {
814 unsigned long rip, old_rip = kvm_rip_read(&svm->vcpu);
817 * For guest debugging where we have to reinject #BP if some
818 * INT3 is guest-owned:
819 * Emulate nRIP by moving RIP forward. Will fail if injection
820 * raises a fault that is not intercepted. Still better than
821 * failing in all cases.
823 skip_emulated_instruction(&svm->vcpu);
824 rip = kvm_rip_read(&svm->vcpu);
825 svm->int3_rip = rip + svm->vmcb->save.cs.base;
826 svm->int3_injected = rip - old_rip;
829 svm->vmcb->control.event_inj = nr
831 | (has_error_code ? SVM_EVTINJ_VALID_ERR : 0)
832 | SVM_EVTINJ_TYPE_EXEPT;
833 svm->vmcb->control.event_inj_err = error_code;
836 static void svm_init_erratum_383(void)
842 if (!static_cpu_has_bug(X86_BUG_AMD_TLB_MMATCH))
845 /* Use _safe variants to not break nested virtualization */
846 val = native_read_msr_safe(MSR_AMD64_DC_CFG, &err);
852 low = lower_32_bits(val);
853 high = upper_32_bits(val);
855 native_write_msr_safe(MSR_AMD64_DC_CFG, low, high);
857 erratum_383_found = true;
860 static void svm_init_osvw(struct kvm_vcpu *vcpu)
863 * Guests should see errata 400 and 415 as fixed (assuming that
864 * HLT and IO instructions are intercepted).
866 vcpu->arch.osvw.length = (osvw_len >= 3) ? (osvw_len) : 3;
867 vcpu->arch.osvw.status = osvw_status & ~(6ULL);
870 * By increasing VCPU's osvw.length to 3 we are telling the guest that
871 * all osvw.status bits inside that length, including bit 0 (which is
872 * reserved for erratum 298), are valid. However, if host processor's
873 * osvw_len is 0 then osvw_status[0] carries no information. We need to
874 * be conservative here and therefore we tell the guest that erratum 298
875 * is present (because we really don't know).
877 if (osvw_len == 0 && boot_cpu_data.x86 == 0x10)
878 vcpu->arch.osvw.status |= 1;
881 static int has_svm(void)
885 if (!cpu_has_svm(&msg)) {
886 printk(KERN_INFO "has_svm: %s\n", msg);
893 static void svm_hardware_disable(void)
895 /* Make sure we clean up behind us */
896 if (static_cpu_has(X86_FEATURE_TSCRATEMSR))
897 wrmsrl(MSR_AMD64_TSC_RATIO, TSC_RATIO_DEFAULT);
901 amd_pmu_disable_virt();
904 static int svm_hardware_enable(void)
907 struct svm_cpu_data *sd;
909 struct desc_struct *gdt;
910 int me = raw_smp_processor_id();
912 rdmsrl(MSR_EFER, efer);
913 if (efer & EFER_SVME)
917 pr_err("%s: err EOPNOTSUPP on %d\n", __func__, me);
920 sd = per_cpu(svm_data, me);
922 pr_err("%s: svm_data is NULL on %d\n", __func__, me);
926 sd->asid_generation = 1;
927 sd->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1;
928 sd->next_asid = sd->max_asid + 1;
929 sd->min_asid = max_sev_asid + 1;
931 gdt = get_current_gdt_rw();
932 sd->tss_desc = (struct kvm_ldttss_desc *)(gdt + GDT_ENTRY_TSS);
934 wrmsrl(MSR_EFER, efer | EFER_SVME);
936 wrmsrl(MSR_VM_HSAVE_PA, page_to_pfn(sd->save_area) << PAGE_SHIFT);
938 if (static_cpu_has(X86_FEATURE_TSCRATEMSR)) {
939 wrmsrl(MSR_AMD64_TSC_RATIO, TSC_RATIO_DEFAULT);
940 __this_cpu_write(current_tsc_ratio, TSC_RATIO_DEFAULT);
947 * Note that it is possible to have a system with mixed processor
948 * revisions and therefore different OSVW bits. If bits are not the same
949 * on different processors then choose the worst case (i.e. if erratum
950 * is present on one processor and not on another then assume that the
951 * erratum is present everywhere).
953 if (cpu_has(&boot_cpu_data, X86_FEATURE_OSVW)) {
954 uint64_t len, status = 0;
957 len = native_read_msr_safe(MSR_AMD64_OSVW_ID_LENGTH, &err);
959 status = native_read_msr_safe(MSR_AMD64_OSVW_STATUS,
963 osvw_status = osvw_len = 0;
967 osvw_status |= status;
968 osvw_status &= (1ULL << osvw_len) - 1;
971 osvw_status = osvw_len = 0;
973 svm_init_erratum_383();
975 amd_pmu_enable_virt();
980 static void svm_cpu_uninit(int cpu)
982 struct svm_cpu_data *sd = per_cpu(svm_data, raw_smp_processor_id());
987 per_cpu(svm_data, raw_smp_processor_id()) = NULL;
988 kfree(sd->sev_vmcbs);
989 __free_page(sd->save_area);
993 static int svm_cpu_init(int cpu)
995 struct svm_cpu_data *sd;
998 sd = kzalloc(sizeof(struct svm_cpu_data), GFP_KERNEL);
1003 sd->save_area = alloc_page(GFP_KERNEL);
1007 if (svm_sev_enabled()) {
1009 sd->sev_vmcbs = kmalloc_array(max_sev_asid + 1,
1016 per_cpu(svm_data, cpu) = sd;
1026 static bool valid_msr_intercept(u32 index)
1030 for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++)
1031 if (direct_access_msrs[i].index == index)
1037 static bool msr_write_intercepted(struct kvm_vcpu *vcpu, unsigned msr)
1044 msrpm = is_guest_mode(vcpu) ? to_svm(vcpu)->nested.msrpm:
1045 to_svm(vcpu)->msrpm;
1047 offset = svm_msrpm_offset(msr);
1048 bit_write = 2 * (msr & 0x0f) + 1;
1049 tmp = msrpm[offset];
1051 BUG_ON(offset == MSR_INVALID);
1053 return !!test_bit(bit_write, &tmp);
1056 static void set_msr_interception(u32 *msrpm, unsigned msr,
1057 int read, int write)
1059 u8 bit_read, bit_write;
1064 * If this warning triggers extend the direct_access_msrs list at the
1065 * beginning of the file
1067 WARN_ON(!valid_msr_intercept(msr));
1069 offset = svm_msrpm_offset(msr);
1070 bit_read = 2 * (msr & 0x0f);
1071 bit_write = 2 * (msr & 0x0f) + 1;
1072 tmp = msrpm[offset];
1074 BUG_ON(offset == MSR_INVALID);
1076 read ? clear_bit(bit_read, &tmp) : set_bit(bit_read, &tmp);
1077 write ? clear_bit(bit_write, &tmp) : set_bit(bit_write, &tmp);
1079 msrpm[offset] = tmp;
1082 static void svm_vcpu_init_msrpm(u32 *msrpm)
1086 memset(msrpm, 0xff, PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER));
1088 for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
1089 if (!direct_access_msrs[i].always)
1092 set_msr_interception(msrpm, direct_access_msrs[i].index, 1, 1);
1096 static void add_msr_offset(u32 offset)
1100 for (i = 0; i < MSRPM_OFFSETS; ++i) {
1102 /* Offset already in list? */
1103 if (msrpm_offsets[i] == offset)
1106 /* Slot used by another offset? */
1107 if (msrpm_offsets[i] != MSR_INVALID)
1110 /* Add offset to list */
1111 msrpm_offsets[i] = offset;
1117 * If this BUG triggers the msrpm_offsets table has an overflow. Just
1118 * increase MSRPM_OFFSETS in this case.
1123 static void init_msrpm_offsets(void)
1127 memset(msrpm_offsets, 0xff, sizeof(msrpm_offsets));
1129 for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
1132 offset = svm_msrpm_offset(direct_access_msrs[i].index);
1133 BUG_ON(offset == MSR_INVALID);
1135 add_msr_offset(offset);
1139 static void svm_enable_lbrv(struct vcpu_svm *svm)
1141 u32 *msrpm = svm->msrpm;
1143 svm->vmcb->control.virt_ext |= LBR_CTL_ENABLE_MASK;
1144 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 1, 1);
1145 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 1, 1);
1146 set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 1, 1);
1147 set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 1, 1);
1150 static void svm_disable_lbrv(struct vcpu_svm *svm)
1152 u32 *msrpm = svm->msrpm;
1154 svm->vmcb->control.virt_ext &= ~LBR_CTL_ENABLE_MASK;
1155 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 0, 0);
1156 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 0, 0);
1157 set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 0, 0);
1158 set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 0, 0);
1161 static void disable_nmi_singlestep(struct vcpu_svm *svm)
1163 svm->nmi_singlestep = false;
1165 if (!(svm->vcpu.guest_debug & KVM_GUESTDBG_SINGLESTEP)) {
1166 /* Clear our flags if they were not set by the guest */
1167 if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_TF))
1168 svm->vmcb->save.rflags &= ~X86_EFLAGS_TF;
1169 if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_RF))
1170 svm->vmcb->save.rflags &= ~X86_EFLAGS_RF;
1175 * This hash table is used to map VM_ID to a struct kvm_svm,
1176 * when handling AMD IOMMU GALOG notification to schedule in
1177 * a particular vCPU.
1179 #define SVM_VM_DATA_HASH_BITS 8
1180 static DEFINE_HASHTABLE(svm_vm_data_hash, SVM_VM_DATA_HASH_BITS);
1181 static u32 next_vm_id = 0;
1182 static bool next_vm_id_wrapped = 0;
1183 static DEFINE_SPINLOCK(svm_vm_data_hash_lock);
1186 * This function is called from IOMMU driver to notify
1187 * SVM to schedule in a particular vCPU of a particular VM.
1189 static int avic_ga_log_notifier(u32 ga_tag)
1191 unsigned long flags;
1192 struct kvm_svm *kvm_svm;
1193 struct kvm_vcpu *vcpu = NULL;
1194 u32 vm_id = AVIC_GATAG_TO_VMID(ga_tag);
1195 u32 vcpu_id = AVIC_GATAG_TO_VCPUID(ga_tag);
1197 pr_debug("SVM: %s: vm_id=%#x, vcpu_id=%#x\n", __func__, vm_id, vcpu_id);
1199 spin_lock_irqsave(&svm_vm_data_hash_lock, flags);
1200 hash_for_each_possible(svm_vm_data_hash, kvm_svm, hnode, vm_id) {
1201 if (kvm_svm->avic_vm_id != vm_id)
1203 vcpu = kvm_get_vcpu_by_id(&kvm_svm->kvm, vcpu_id);
1206 spin_unlock_irqrestore(&svm_vm_data_hash_lock, flags);
1209 * At this point, the IOMMU should have already set the pending
1210 * bit in the vAPIC backing page. So, we just need to schedule
1214 kvm_vcpu_wake_up(vcpu);
1219 static __init int sev_hardware_setup(void)
1221 struct sev_user_data_status *status;
1224 /* Maximum number of encrypted guests supported simultaneously */
1225 max_sev_asid = cpuid_ecx(0x8000001F);
1230 /* Minimum ASID value that should be used for SEV guest */
1231 min_sev_asid = cpuid_edx(0x8000001F);
1233 /* Initialize SEV ASID bitmap */
1234 sev_asid_bitmap = bitmap_zalloc(max_sev_asid, GFP_KERNEL);
1235 if (!sev_asid_bitmap)
1238 status = kmalloc(sizeof(*status), GFP_KERNEL);
1243 * Check SEV platform status.
1245 * PLATFORM_STATUS can be called in any state, if we failed to query
1246 * the PLATFORM status then either PSP firmware does not support SEV
1247 * feature or SEV firmware is dead.
1249 rc = sev_platform_status(status, NULL);
1253 pr_info("SEV supported\n");
1260 static void grow_ple_window(struct kvm_vcpu *vcpu)
1262 struct vcpu_svm *svm = to_svm(vcpu);
1263 struct vmcb_control_area *control = &svm->vmcb->control;
1264 int old = control->pause_filter_count;
1266 control->pause_filter_count = __grow_ple_window(old,
1268 pause_filter_count_grow,
1269 pause_filter_count_max);
1271 if (control->pause_filter_count != old)
1272 mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
1274 trace_kvm_ple_window_grow(vcpu->vcpu_id,
1275 control->pause_filter_count, old);
1278 static void shrink_ple_window(struct kvm_vcpu *vcpu)
1280 struct vcpu_svm *svm = to_svm(vcpu);
1281 struct vmcb_control_area *control = &svm->vmcb->control;
1282 int old = control->pause_filter_count;
1284 control->pause_filter_count =
1285 __shrink_ple_window(old,
1287 pause_filter_count_shrink,
1288 pause_filter_count);
1289 if (control->pause_filter_count != old)
1290 mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
1292 trace_kvm_ple_window_shrink(vcpu->vcpu_id,
1293 control->pause_filter_count, old);
1296 static __init int svm_hardware_setup(void)
1299 struct page *iopm_pages;
1303 iopm_pages = alloc_pages(GFP_KERNEL, IOPM_ALLOC_ORDER);
1308 iopm_va = page_address(iopm_pages);
1309 memset(iopm_va, 0xff, PAGE_SIZE * (1 << IOPM_ALLOC_ORDER));
1310 iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT;
1312 init_msrpm_offsets();
1314 if (boot_cpu_has(X86_FEATURE_NX))
1315 kvm_enable_efer_bits(EFER_NX);
1317 if (boot_cpu_has(X86_FEATURE_FXSR_OPT))
1318 kvm_enable_efer_bits(EFER_FFXSR);
1320 if (boot_cpu_has(X86_FEATURE_TSCRATEMSR)) {
1321 kvm_has_tsc_control = true;
1322 kvm_max_tsc_scaling_ratio = TSC_RATIO_MAX;
1323 kvm_tsc_scaling_ratio_frac_bits = 32;
1326 /* Check for pause filtering support */
1327 if (!boot_cpu_has(X86_FEATURE_PAUSEFILTER)) {
1328 pause_filter_count = 0;
1329 pause_filter_thresh = 0;
1330 } else if (!boot_cpu_has(X86_FEATURE_PFTHRESHOLD)) {
1331 pause_filter_thresh = 0;
1335 printk(KERN_INFO "kvm: Nested Virtualization enabled\n");
1336 kvm_enable_efer_bits(EFER_SVME | EFER_LMSLE);
1340 if (boot_cpu_has(X86_FEATURE_SEV) &&
1341 IS_ENABLED(CONFIG_KVM_AMD_SEV)) {
1342 r = sev_hardware_setup();
1350 for_each_possible_cpu(cpu) {
1351 r = svm_cpu_init(cpu);
1356 if (!boot_cpu_has(X86_FEATURE_NPT))
1357 npt_enabled = false;
1359 if (npt_enabled && !npt) {
1360 printk(KERN_INFO "kvm: Nested Paging disabled\n");
1361 npt_enabled = false;
1365 printk(KERN_INFO "kvm: Nested Paging enabled\n");
1372 !boot_cpu_has(X86_FEATURE_AVIC) ||
1373 !IS_ENABLED(CONFIG_X86_LOCAL_APIC)) {
1376 pr_info("AVIC enabled\n");
1378 amd_iommu_register_ga_log_notifier(&avic_ga_log_notifier);
1384 !boot_cpu_has(X86_FEATURE_V_VMSAVE_VMLOAD) ||
1385 !IS_ENABLED(CONFIG_X86_64)) {
1388 pr_info("Virtual VMLOAD VMSAVE supported\n");
1393 if (!boot_cpu_has(X86_FEATURE_VGIF))
1396 pr_info("Virtual GIF supported\n");
1402 __free_pages(iopm_pages, IOPM_ALLOC_ORDER);
1407 static __exit void svm_hardware_unsetup(void)
1411 if (svm_sev_enabled())
1412 bitmap_free(sev_asid_bitmap);
1414 for_each_possible_cpu(cpu)
1415 svm_cpu_uninit(cpu);
1417 __free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT), IOPM_ALLOC_ORDER);
1421 static void init_seg(struct vmcb_seg *seg)
1424 seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK |
1425 SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */
1426 seg->limit = 0xffff;
1430 static void init_sys_seg(struct vmcb_seg *seg, uint32_t type)
1433 seg->attrib = SVM_SELECTOR_P_MASK | type;
1434 seg->limit = 0xffff;
1438 static u64 svm_read_l1_tsc_offset(struct kvm_vcpu *vcpu)
1440 struct vcpu_svm *svm = to_svm(vcpu);
1442 if (is_guest_mode(vcpu))
1443 return svm->nested.hsave->control.tsc_offset;
1445 return vcpu->arch.tsc_offset;
1448 static u64 svm_write_l1_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1450 struct vcpu_svm *svm = to_svm(vcpu);
1451 u64 g_tsc_offset = 0;
1453 if (is_guest_mode(vcpu)) {
1454 /* Write L1's TSC offset. */
1455 g_tsc_offset = svm->vmcb->control.tsc_offset -
1456 svm->nested.hsave->control.tsc_offset;
1457 svm->nested.hsave->control.tsc_offset = offset;
1460 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
1461 svm->vmcb->control.tsc_offset - g_tsc_offset,
1464 svm->vmcb->control.tsc_offset = offset + g_tsc_offset;
1466 mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
1467 return svm->vmcb->control.tsc_offset;
1470 static void avic_init_vmcb(struct vcpu_svm *svm)
1472 struct vmcb *vmcb = svm->vmcb;
1473 struct kvm_svm *kvm_svm = to_kvm_svm(svm->vcpu.kvm);
1474 phys_addr_t bpa = __sme_set(page_to_phys(svm->avic_backing_page));
1475 phys_addr_t lpa = __sme_set(page_to_phys(kvm_svm->avic_logical_id_table_page));
1476 phys_addr_t ppa = __sme_set(page_to_phys(kvm_svm->avic_physical_id_table_page));
1478 vmcb->control.avic_backing_page = bpa & AVIC_HPA_MASK;
1479 vmcb->control.avic_logical_id = lpa & AVIC_HPA_MASK;
1480 vmcb->control.avic_physical_id = ppa & AVIC_HPA_MASK;
1481 vmcb->control.avic_physical_id |= AVIC_MAX_PHYSICAL_ID_COUNT;
1482 vmcb->control.int_ctl |= AVIC_ENABLE_MASK;
1485 static void init_vmcb(struct vcpu_svm *svm)
1487 struct vmcb_control_area *control = &svm->vmcb->control;
1488 struct vmcb_save_area *save = &svm->vmcb->save;
1490 svm->vcpu.arch.hflags = 0;
1492 set_cr_intercept(svm, INTERCEPT_CR0_READ);
1493 set_cr_intercept(svm, INTERCEPT_CR3_READ);
1494 set_cr_intercept(svm, INTERCEPT_CR4_READ);
1495 set_cr_intercept(svm, INTERCEPT_CR0_WRITE);
1496 set_cr_intercept(svm, INTERCEPT_CR3_WRITE);
1497 set_cr_intercept(svm, INTERCEPT_CR4_WRITE);
1498 if (!kvm_vcpu_apicv_active(&svm->vcpu))
1499 set_cr_intercept(svm, INTERCEPT_CR8_WRITE);
1501 set_dr_intercepts(svm);
1503 set_exception_intercept(svm, PF_VECTOR);
1504 set_exception_intercept(svm, UD_VECTOR);
1505 set_exception_intercept(svm, MC_VECTOR);
1506 set_exception_intercept(svm, AC_VECTOR);
1507 set_exception_intercept(svm, DB_VECTOR);
1509 * Guest access to VMware backdoor ports could legitimately
1510 * trigger #GP because of TSS I/O permission bitmap.
1511 * We intercept those #GP and allow access to them anyway
1514 if (enable_vmware_backdoor)
1515 set_exception_intercept(svm, GP_VECTOR);
1517 set_intercept(svm, INTERCEPT_INTR);
1518 set_intercept(svm, INTERCEPT_NMI);
1519 set_intercept(svm, INTERCEPT_SMI);
1520 set_intercept(svm, INTERCEPT_SELECTIVE_CR0);
1521 set_intercept(svm, INTERCEPT_RDPMC);
1522 set_intercept(svm, INTERCEPT_CPUID);
1523 set_intercept(svm, INTERCEPT_INVD);
1524 set_intercept(svm, INTERCEPT_INVLPG);
1525 set_intercept(svm, INTERCEPT_INVLPGA);
1526 set_intercept(svm, INTERCEPT_IOIO_PROT);
1527 set_intercept(svm, INTERCEPT_MSR_PROT);
1528 set_intercept(svm, INTERCEPT_TASK_SWITCH);
1529 set_intercept(svm, INTERCEPT_SHUTDOWN);
1530 set_intercept(svm, INTERCEPT_VMRUN);
1531 set_intercept(svm, INTERCEPT_VMMCALL);
1532 set_intercept(svm, INTERCEPT_VMLOAD);
1533 set_intercept(svm, INTERCEPT_VMSAVE);
1534 set_intercept(svm, INTERCEPT_STGI);
1535 set_intercept(svm, INTERCEPT_CLGI);
1536 set_intercept(svm, INTERCEPT_SKINIT);
1537 set_intercept(svm, INTERCEPT_WBINVD);
1538 set_intercept(svm, INTERCEPT_XSETBV);
1539 set_intercept(svm, INTERCEPT_RSM);
1541 if (!kvm_mwait_in_guest(svm->vcpu.kvm)) {
1542 set_intercept(svm, INTERCEPT_MONITOR);
1543 set_intercept(svm, INTERCEPT_MWAIT);
1546 if (!kvm_hlt_in_guest(svm->vcpu.kvm))
1547 set_intercept(svm, INTERCEPT_HLT);
1549 control->iopm_base_pa = __sme_set(iopm_base);
1550 control->msrpm_base_pa = __sme_set(__pa(svm->msrpm));
1551 control->int_ctl = V_INTR_MASKING_MASK;
1553 init_seg(&save->es);
1554 init_seg(&save->ss);
1555 init_seg(&save->ds);
1556 init_seg(&save->fs);
1557 init_seg(&save->gs);
1559 save->cs.selector = 0xf000;
1560 save->cs.base = 0xffff0000;
1561 /* Executable/Readable Code Segment */
1562 save->cs.attrib = SVM_SELECTOR_READ_MASK | SVM_SELECTOR_P_MASK |
1563 SVM_SELECTOR_S_MASK | SVM_SELECTOR_CODE_MASK;
1564 save->cs.limit = 0xffff;
1566 save->gdtr.limit = 0xffff;
1567 save->idtr.limit = 0xffff;
1569 init_sys_seg(&save->ldtr, SEG_TYPE_LDT);
1570 init_sys_seg(&save->tr, SEG_TYPE_BUSY_TSS16);
1572 svm_set_efer(&svm->vcpu, 0);
1573 save->dr6 = 0xffff0ff0;
1574 kvm_set_rflags(&svm->vcpu, 2);
1575 save->rip = 0x0000fff0;
1576 svm->vcpu.arch.regs[VCPU_REGS_RIP] = save->rip;
1579 * svm_set_cr0() sets PG and WP and clears NW and CD on save->cr0.
1580 * It also updates the guest-visible cr0 value.
1582 svm_set_cr0(&svm->vcpu, X86_CR0_NW | X86_CR0_CD | X86_CR0_ET);
1583 kvm_mmu_reset_context(&svm->vcpu);
1585 save->cr4 = X86_CR4_PAE;
1589 /* Setup VMCB for Nested Paging */
1590 control->nested_ctl |= SVM_NESTED_CTL_NP_ENABLE;
1591 clr_intercept(svm, INTERCEPT_INVLPG);
1592 clr_exception_intercept(svm, PF_VECTOR);
1593 clr_cr_intercept(svm, INTERCEPT_CR3_READ);
1594 clr_cr_intercept(svm, INTERCEPT_CR3_WRITE);
1595 save->g_pat = svm->vcpu.arch.pat;
1599 svm->asid_generation = 0;
1601 svm->nested.vmcb = 0;
1602 svm->vcpu.arch.hflags = 0;
1604 if (pause_filter_count) {
1605 control->pause_filter_count = pause_filter_count;
1606 if (pause_filter_thresh)
1607 control->pause_filter_thresh = pause_filter_thresh;
1608 set_intercept(svm, INTERCEPT_PAUSE);
1610 clr_intercept(svm, INTERCEPT_PAUSE);
1613 if (kvm_vcpu_apicv_active(&svm->vcpu))
1614 avic_init_vmcb(svm);
1617 * If hardware supports Virtual VMLOAD VMSAVE then enable it
1618 * in VMCB and clear intercepts to avoid #VMEXIT.
1621 clr_intercept(svm, INTERCEPT_VMLOAD);
1622 clr_intercept(svm, INTERCEPT_VMSAVE);
1623 svm->vmcb->control.virt_ext |= VIRTUAL_VMLOAD_VMSAVE_ENABLE_MASK;
1627 clr_intercept(svm, INTERCEPT_STGI);
1628 clr_intercept(svm, INTERCEPT_CLGI);
1629 svm->vmcb->control.int_ctl |= V_GIF_ENABLE_MASK;
1632 if (sev_guest(svm->vcpu.kvm)) {
1633 svm->vmcb->control.nested_ctl |= SVM_NESTED_CTL_SEV_ENABLE;
1634 clr_exception_intercept(svm, UD_VECTOR);
1637 mark_all_dirty(svm->vmcb);
1643 static u64 *avic_get_physical_id_entry(struct kvm_vcpu *vcpu,
1646 u64 *avic_physical_id_table;
1647 struct kvm_svm *kvm_svm = to_kvm_svm(vcpu->kvm);
1649 if (index >= AVIC_MAX_PHYSICAL_ID_COUNT)
1652 avic_physical_id_table = page_address(kvm_svm->avic_physical_id_table_page);
1654 return &avic_physical_id_table[index];
1659 * AVIC hardware walks the nested page table to check permissions,
1660 * but does not use the SPA address specified in the leaf page
1661 * table entry since it uses address in the AVIC_BACKING_PAGE pointer
1662 * field of the VMCB. Therefore, we set up the
1663 * APIC_ACCESS_PAGE_PRIVATE_MEMSLOT (4KB) here.
1665 static int avic_init_access_page(struct kvm_vcpu *vcpu)
1667 struct kvm *kvm = vcpu->kvm;
1670 mutex_lock(&kvm->slots_lock);
1671 if (kvm->arch.apic_access_page_done)
1674 ret = __x86_set_memory_region(kvm,
1675 APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
1676 APIC_DEFAULT_PHYS_BASE,
1681 kvm->arch.apic_access_page_done = true;
1683 mutex_unlock(&kvm->slots_lock);
1687 static int avic_init_backing_page(struct kvm_vcpu *vcpu)
1690 u64 *entry, new_entry;
1691 int id = vcpu->vcpu_id;
1692 struct vcpu_svm *svm = to_svm(vcpu);
1694 ret = avic_init_access_page(vcpu);
1698 if (id >= AVIC_MAX_PHYSICAL_ID_COUNT)
1701 if (!svm->vcpu.arch.apic->regs)
1704 svm->avic_backing_page = virt_to_page(svm->vcpu.arch.apic->regs);
1706 /* Setting AVIC backing page address in the phy APIC ID table */
1707 entry = avic_get_physical_id_entry(vcpu, id);
1711 new_entry = READ_ONCE(*entry);
1712 new_entry = __sme_set((page_to_phys(svm->avic_backing_page) &
1713 AVIC_PHYSICAL_ID_ENTRY_BACKING_PAGE_MASK) |
1714 AVIC_PHYSICAL_ID_ENTRY_VALID_MASK);
1715 WRITE_ONCE(*entry, new_entry);
1717 svm->avic_physical_id_cache = entry;
1722 static void __sev_asid_free(int asid)
1724 struct svm_cpu_data *sd;
1728 clear_bit(pos, sev_asid_bitmap);
1730 for_each_possible_cpu(cpu) {
1731 sd = per_cpu(svm_data, cpu);
1732 sd->sev_vmcbs[pos] = NULL;
1736 static void sev_asid_free(struct kvm *kvm)
1738 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1740 __sev_asid_free(sev->asid);
1743 static void sev_unbind_asid(struct kvm *kvm, unsigned int handle)
1745 struct sev_data_decommission *decommission;
1746 struct sev_data_deactivate *data;
1751 data = kzalloc(sizeof(*data), GFP_KERNEL);
1755 /* deactivate handle */
1756 data->handle = handle;
1757 sev_guest_deactivate(data, NULL);
1759 wbinvd_on_all_cpus();
1760 sev_guest_df_flush(NULL);
1763 decommission = kzalloc(sizeof(*decommission), GFP_KERNEL);
1767 /* decommission handle */
1768 decommission->handle = handle;
1769 sev_guest_decommission(decommission, NULL);
1771 kfree(decommission);
1774 static struct page **sev_pin_memory(struct kvm *kvm, unsigned long uaddr,
1775 unsigned long ulen, unsigned long *n,
1778 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1779 unsigned long npages, npinned, size;
1780 unsigned long locked, lock_limit;
1781 struct page **pages;
1782 unsigned long first, last;
1784 if (ulen == 0 || uaddr + ulen < uaddr)
1787 /* Calculate number of pages. */
1788 first = (uaddr & PAGE_MASK) >> PAGE_SHIFT;
1789 last = ((uaddr + ulen - 1) & PAGE_MASK) >> PAGE_SHIFT;
1790 npages = (last - first + 1);
1792 locked = sev->pages_locked + npages;
1793 lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1794 if (locked > lock_limit && !capable(CAP_IPC_LOCK)) {
1795 pr_err("SEV: %lu locked pages exceed the lock limit of %lu.\n", locked, lock_limit);
1799 /* Avoid using vmalloc for smaller buffers. */
1800 size = npages * sizeof(struct page *);
1801 if (size > PAGE_SIZE)
1802 pages = __vmalloc(size, GFP_KERNEL_ACCOUNT | __GFP_ZERO,
1805 pages = kmalloc(size, GFP_KERNEL_ACCOUNT);
1810 /* Pin the user virtual address. */
1811 npinned = get_user_pages_fast(uaddr, npages, FOLL_WRITE, pages);
1812 if (npinned != npages) {
1813 pr_err("SEV: Failure locking %lu pages.\n", npages);
1818 sev->pages_locked = locked;
1824 release_pages(pages, npinned);
1830 static void sev_unpin_memory(struct kvm *kvm, struct page **pages,
1831 unsigned long npages)
1833 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1835 release_pages(pages, npages);
1837 sev->pages_locked -= npages;
1840 static void sev_clflush_pages(struct page *pages[], unsigned long npages)
1842 uint8_t *page_virtual;
1845 if (npages == 0 || pages == NULL)
1848 for (i = 0; i < npages; i++) {
1849 page_virtual = kmap_atomic(pages[i]);
1850 clflush_cache_range(page_virtual, PAGE_SIZE);
1851 kunmap_atomic(page_virtual);
1855 static void __unregister_enc_region_locked(struct kvm *kvm,
1856 struct enc_region *region)
1859 * The guest may change the memory encryption attribute from C=0 -> C=1
1860 * or vice versa for this memory range. Lets make sure caches are
1861 * flushed to ensure that guest data gets written into memory with
1864 sev_clflush_pages(region->pages, region->npages);
1866 sev_unpin_memory(kvm, region->pages, region->npages);
1867 list_del(®ion->list);
1871 static struct kvm *svm_vm_alloc(void)
1873 struct kvm_svm *kvm_svm = __vmalloc(sizeof(struct kvm_svm),
1874 GFP_KERNEL_ACCOUNT | __GFP_ZERO,
1876 return &kvm_svm->kvm;
1879 static void svm_vm_free(struct kvm *kvm)
1881 vfree(to_kvm_svm(kvm));
1884 static void sev_vm_destroy(struct kvm *kvm)
1886 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1887 struct list_head *head = &sev->regions_list;
1888 struct list_head *pos, *q;
1890 if (!sev_guest(kvm))
1893 mutex_lock(&kvm->lock);
1896 * if userspace was terminated before unregistering the memory regions
1897 * then lets unpin all the registered memory.
1899 if (!list_empty(head)) {
1900 list_for_each_safe(pos, q, head) {
1901 __unregister_enc_region_locked(kvm,
1902 list_entry(pos, struct enc_region, list));
1906 mutex_unlock(&kvm->lock);
1908 sev_unbind_asid(kvm, sev->handle);
1912 static void avic_vm_destroy(struct kvm *kvm)
1914 unsigned long flags;
1915 struct kvm_svm *kvm_svm = to_kvm_svm(kvm);
1920 if (kvm_svm->avic_logical_id_table_page)
1921 __free_page(kvm_svm->avic_logical_id_table_page);
1922 if (kvm_svm->avic_physical_id_table_page)
1923 __free_page(kvm_svm->avic_physical_id_table_page);
1925 spin_lock_irqsave(&svm_vm_data_hash_lock, flags);
1926 hash_del(&kvm_svm->hnode);
1927 spin_unlock_irqrestore(&svm_vm_data_hash_lock, flags);
1930 static void svm_vm_destroy(struct kvm *kvm)
1932 avic_vm_destroy(kvm);
1933 sev_vm_destroy(kvm);
1936 static int avic_vm_init(struct kvm *kvm)
1938 unsigned long flags;
1940 struct kvm_svm *kvm_svm = to_kvm_svm(kvm);
1942 struct page *p_page;
1943 struct page *l_page;
1949 /* Allocating physical APIC ID table (4KB) */
1950 p_page = alloc_page(GFP_KERNEL_ACCOUNT);
1954 kvm_svm->avic_physical_id_table_page = p_page;
1955 clear_page(page_address(p_page));
1957 /* Allocating logical APIC ID table (4KB) */
1958 l_page = alloc_page(GFP_KERNEL_ACCOUNT);
1962 kvm_svm->avic_logical_id_table_page = l_page;
1963 clear_page(page_address(l_page));
1965 spin_lock_irqsave(&svm_vm_data_hash_lock, flags);
1967 vm_id = next_vm_id = (next_vm_id + 1) & AVIC_VM_ID_MASK;
1968 if (vm_id == 0) { /* id is 1-based, zero is not okay */
1969 next_vm_id_wrapped = 1;
1972 /* Is it still in use? Only possible if wrapped at least once */
1973 if (next_vm_id_wrapped) {
1974 hash_for_each_possible(svm_vm_data_hash, k2, hnode, vm_id) {
1975 if (k2->avic_vm_id == vm_id)
1979 kvm_svm->avic_vm_id = vm_id;
1980 hash_add(svm_vm_data_hash, &kvm_svm->hnode, kvm_svm->avic_vm_id);
1981 spin_unlock_irqrestore(&svm_vm_data_hash_lock, flags);
1986 avic_vm_destroy(kvm);
1991 avic_update_iommu_vcpu_affinity(struct kvm_vcpu *vcpu, int cpu, bool r)
1994 unsigned long flags;
1995 struct amd_svm_iommu_ir *ir;
1996 struct vcpu_svm *svm = to_svm(vcpu);
1998 if (!kvm_arch_has_assigned_device(vcpu->kvm))
2002 * Here, we go through the per-vcpu ir_list to update all existing
2003 * interrupt remapping table entry targeting this vcpu.
2005 spin_lock_irqsave(&svm->ir_list_lock, flags);
2007 if (list_empty(&svm->ir_list))
2010 list_for_each_entry(ir, &svm->ir_list, node) {
2011 ret = amd_iommu_update_ga(cpu, r, ir->data);
2016 spin_unlock_irqrestore(&svm->ir_list_lock, flags);
2020 static void avic_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2023 /* ID = 0xff (broadcast), ID > 0xff (reserved) */
2024 int h_physical_id = kvm_cpu_get_apicid(cpu);
2025 struct vcpu_svm *svm = to_svm(vcpu);
2027 if (!kvm_vcpu_apicv_active(vcpu))
2031 * Since the host physical APIC id is 8 bits,
2032 * we can support host APIC ID upto 255.
2034 if (WARN_ON(h_physical_id > AVIC_PHYSICAL_ID_ENTRY_HOST_PHYSICAL_ID_MASK))
2037 entry = READ_ONCE(*(svm->avic_physical_id_cache));
2038 WARN_ON(entry & AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK);
2040 entry &= ~AVIC_PHYSICAL_ID_ENTRY_HOST_PHYSICAL_ID_MASK;
2041 entry |= (h_physical_id & AVIC_PHYSICAL_ID_ENTRY_HOST_PHYSICAL_ID_MASK);
2043 entry &= ~AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK;
2044 if (svm->avic_is_running)
2045 entry |= AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK;
2047 WRITE_ONCE(*(svm->avic_physical_id_cache), entry);
2048 avic_update_iommu_vcpu_affinity(vcpu, h_physical_id,
2049 svm->avic_is_running);
2052 static void avic_vcpu_put(struct kvm_vcpu *vcpu)
2055 struct vcpu_svm *svm = to_svm(vcpu);
2057 if (!kvm_vcpu_apicv_active(vcpu))
2060 entry = READ_ONCE(*(svm->avic_physical_id_cache));
2061 if (entry & AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK)
2062 avic_update_iommu_vcpu_affinity(vcpu, -1, 0);
2064 entry &= ~AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK;
2065 WRITE_ONCE(*(svm->avic_physical_id_cache), entry);
2069 * This function is called during VCPU halt/unhalt.
2071 static void avic_set_running(struct kvm_vcpu *vcpu, bool is_run)
2073 struct vcpu_svm *svm = to_svm(vcpu);
2075 svm->avic_is_running = is_run;
2077 avic_vcpu_load(vcpu, vcpu->cpu);
2079 avic_vcpu_put(vcpu);
2082 static void svm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
2084 struct vcpu_svm *svm = to_svm(vcpu);
2088 vcpu->arch.microcode_version = 0x01000065;
2090 svm->virt_spec_ctrl = 0;
2093 svm->vcpu.arch.apic_base = APIC_DEFAULT_PHYS_BASE |
2094 MSR_IA32_APICBASE_ENABLE;
2095 if (kvm_vcpu_is_reset_bsp(&svm->vcpu))
2096 svm->vcpu.arch.apic_base |= MSR_IA32_APICBASE_BSP;
2100 kvm_cpuid(vcpu, &eax, &dummy, &dummy, &dummy, true);
2101 kvm_rdx_write(vcpu, eax);
2103 if (kvm_vcpu_apicv_active(vcpu) && !init_event)
2104 avic_update_vapic_bar(svm, APIC_DEFAULT_PHYS_BASE);
2107 static int avic_init_vcpu(struct vcpu_svm *svm)
2111 if (!kvm_vcpu_apicv_active(&svm->vcpu))
2114 ret = avic_init_backing_page(&svm->vcpu);
2118 INIT_LIST_HEAD(&svm->ir_list);
2119 spin_lock_init(&svm->ir_list_lock);
2120 svm->dfr_reg = APIC_DFR_FLAT;
2125 static struct kvm_vcpu *svm_create_vcpu(struct kvm *kvm, unsigned int id)
2127 struct vcpu_svm *svm;
2129 struct page *msrpm_pages;
2130 struct page *hsave_page;
2131 struct page *nested_msrpm_pages;
2134 svm = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL_ACCOUNT);
2140 svm->vcpu.arch.guest_fpu = kmem_cache_zalloc(x86_fpu_cache,
2141 GFP_KERNEL_ACCOUNT);
2142 if (!svm->vcpu.arch.guest_fpu) {
2143 printk(KERN_ERR "kvm: failed to allocate vcpu's fpu\n");
2145 goto free_partial_svm;
2148 err = kvm_vcpu_init(&svm->vcpu, kvm, id);
2153 page = alloc_page(GFP_KERNEL_ACCOUNT);
2157 msrpm_pages = alloc_pages(GFP_KERNEL_ACCOUNT, MSRPM_ALLOC_ORDER);
2161 nested_msrpm_pages = alloc_pages(GFP_KERNEL_ACCOUNT, MSRPM_ALLOC_ORDER);
2162 if (!nested_msrpm_pages)
2165 hsave_page = alloc_page(GFP_KERNEL_ACCOUNT);
2169 err = avic_init_vcpu(svm);
2173 /* We initialize this flag to true to make sure that the is_running
2174 * bit would be set the first time the vcpu is loaded.
2176 svm->avic_is_running = true;
2178 svm->nested.hsave = page_address(hsave_page);
2180 svm->msrpm = page_address(msrpm_pages);
2181 svm_vcpu_init_msrpm(svm->msrpm);
2183 svm->nested.msrpm = page_address(nested_msrpm_pages);
2184 svm_vcpu_init_msrpm(svm->nested.msrpm);
2186 svm->vmcb = page_address(page);
2187 clear_page(svm->vmcb);
2188 svm->vmcb_pa = __sme_set(page_to_pfn(page) << PAGE_SHIFT);
2189 svm->asid_generation = 0;
2192 svm_init_osvw(&svm->vcpu);
2197 __free_page(hsave_page);
2199 __free_pages(nested_msrpm_pages, MSRPM_ALLOC_ORDER);
2201 __free_pages(msrpm_pages, MSRPM_ALLOC_ORDER);
2205 kvm_vcpu_uninit(&svm->vcpu);
2207 kmem_cache_free(x86_fpu_cache, svm->vcpu.arch.guest_fpu);
2209 kmem_cache_free(kvm_vcpu_cache, svm);
2211 return ERR_PTR(err);
2214 static void svm_clear_current_vmcb(struct vmcb *vmcb)
2218 for_each_online_cpu(i)
2219 cmpxchg(&per_cpu(svm_data, i)->current_vmcb, vmcb, NULL);
2222 static void svm_free_vcpu(struct kvm_vcpu *vcpu)
2224 struct vcpu_svm *svm = to_svm(vcpu);
2227 * The vmcb page can be recycled, causing a false negative in
2228 * svm_vcpu_load(). So, ensure that no logical CPU has this
2229 * vmcb page recorded as its current vmcb.
2231 svm_clear_current_vmcb(svm->vmcb);
2233 __free_page(pfn_to_page(__sme_clr(svm->vmcb_pa) >> PAGE_SHIFT));
2234 __free_pages(virt_to_page(svm->msrpm), MSRPM_ALLOC_ORDER);
2235 __free_page(virt_to_page(svm->nested.hsave));
2236 __free_pages(virt_to_page(svm->nested.msrpm), MSRPM_ALLOC_ORDER);
2237 kvm_vcpu_uninit(vcpu);
2238 kmem_cache_free(x86_fpu_cache, svm->vcpu.arch.guest_fpu);
2239 kmem_cache_free(kvm_vcpu_cache, svm);
2242 static void svm_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2244 struct vcpu_svm *svm = to_svm(vcpu);
2245 struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
2248 if (unlikely(cpu != vcpu->cpu)) {
2249 svm->asid_generation = 0;
2250 mark_all_dirty(svm->vmcb);
2253 #ifdef CONFIG_X86_64
2254 rdmsrl(MSR_GS_BASE, to_svm(vcpu)->host.gs_base);
2256 savesegment(fs, svm->host.fs);
2257 savesegment(gs, svm->host.gs);
2258 svm->host.ldt = kvm_read_ldt();
2260 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
2261 rdmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
2263 if (static_cpu_has(X86_FEATURE_TSCRATEMSR)) {
2264 u64 tsc_ratio = vcpu->arch.tsc_scaling_ratio;
2265 if (tsc_ratio != __this_cpu_read(current_tsc_ratio)) {
2266 __this_cpu_write(current_tsc_ratio, tsc_ratio);
2267 wrmsrl(MSR_AMD64_TSC_RATIO, tsc_ratio);
2270 /* This assumes that the kernel never uses MSR_TSC_AUX */
2271 if (static_cpu_has(X86_FEATURE_RDTSCP))
2272 wrmsrl(MSR_TSC_AUX, svm->tsc_aux);
2274 if (sd->current_vmcb != svm->vmcb) {
2275 sd->current_vmcb = svm->vmcb;
2276 indirect_branch_prediction_barrier();
2278 avic_vcpu_load(vcpu, cpu);
2281 static void svm_vcpu_put(struct kvm_vcpu *vcpu)
2283 struct vcpu_svm *svm = to_svm(vcpu);
2286 avic_vcpu_put(vcpu);
2288 ++vcpu->stat.host_state_reload;
2289 kvm_load_ldt(svm->host.ldt);
2290 #ifdef CONFIG_X86_64
2291 loadsegment(fs, svm->host.fs);
2292 wrmsrl(MSR_KERNEL_GS_BASE, current->thread.gsbase);
2293 load_gs_index(svm->host.gs);
2295 #ifdef CONFIG_X86_32_LAZY_GS
2296 loadsegment(gs, svm->host.gs);
2299 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
2300 wrmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
2303 static void svm_vcpu_blocking(struct kvm_vcpu *vcpu)
2305 avic_set_running(vcpu, false);
2308 static void svm_vcpu_unblocking(struct kvm_vcpu *vcpu)
2310 avic_set_running(vcpu, true);
2313 static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu)
2315 struct vcpu_svm *svm = to_svm(vcpu);
2316 unsigned long rflags = svm->vmcb->save.rflags;
2318 if (svm->nmi_singlestep) {
2319 /* Hide our flags if they were not set by the guest */
2320 if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_TF))
2321 rflags &= ~X86_EFLAGS_TF;
2322 if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_RF))
2323 rflags &= ~X86_EFLAGS_RF;
2328 static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
2330 if (to_svm(vcpu)->nmi_singlestep)
2331 rflags |= (X86_EFLAGS_TF | X86_EFLAGS_RF);
2334 * Any change of EFLAGS.VM is accompanied by a reload of SS
2335 * (caused by either a task switch or an inter-privilege IRET),
2336 * so we do not need to update the CPL here.
2338 to_svm(vcpu)->vmcb->save.rflags = rflags;
2341 static void svm_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
2344 case VCPU_EXREG_PDPTR:
2345 BUG_ON(!npt_enabled);
2346 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
2353 static void svm_set_vintr(struct vcpu_svm *svm)
2355 set_intercept(svm, INTERCEPT_VINTR);
2358 static void svm_clear_vintr(struct vcpu_svm *svm)
2360 clr_intercept(svm, INTERCEPT_VINTR);
2363 static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg)
2365 struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
2368 case VCPU_SREG_CS: return &save->cs;
2369 case VCPU_SREG_DS: return &save->ds;
2370 case VCPU_SREG_ES: return &save->es;
2371 case VCPU_SREG_FS: return &save->fs;
2372 case VCPU_SREG_GS: return &save->gs;
2373 case VCPU_SREG_SS: return &save->ss;
2374 case VCPU_SREG_TR: return &save->tr;
2375 case VCPU_SREG_LDTR: return &save->ldtr;
2381 static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg)
2383 struct vmcb_seg *s = svm_seg(vcpu, seg);
2388 static void svm_get_segment(struct kvm_vcpu *vcpu,
2389 struct kvm_segment *var, int seg)
2391 struct vmcb_seg *s = svm_seg(vcpu, seg);
2393 var->base = s->base;
2394 var->limit = s->limit;
2395 var->selector = s->selector;
2396 var->type = s->attrib & SVM_SELECTOR_TYPE_MASK;
2397 var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1;
2398 var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
2399 var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1;
2400 var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1;
2401 var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
2402 var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
2405 * AMD CPUs circa 2014 track the G bit for all segments except CS.
2406 * However, the SVM spec states that the G bit is not observed by the
2407 * CPU, and some VMware virtual CPUs drop the G bit for all segments.
2408 * So let's synthesize a legal G bit for all segments, this helps
2409 * running KVM nested. It also helps cross-vendor migration, because
2410 * Intel's vmentry has a check on the 'G' bit.
2412 var->g = s->limit > 0xfffff;
2415 * AMD's VMCB does not have an explicit unusable field, so emulate it
2416 * for cross vendor migration purposes by "not present"
2418 var->unusable = !var->present;
2423 * Work around a bug where the busy flag in the tr selector
2433 * The accessed bit must always be set in the segment
2434 * descriptor cache, although it can be cleared in the
2435 * descriptor, the cached bit always remains at 1. Since
2436 * Intel has a check on this, set it here to support
2437 * cross-vendor migration.
2444 * On AMD CPUs sometimes the DB bit in the segment
2445 * descriptor is left as 1, although the whole segment has
2446 * been made unusable. Clear it here to pass an Intel VMX
2447 * entry check when cross vendor migrating.
2451 /* This is symmetric with svm_set_segment() */
2452 var->dpl = to_svm(vcpu)->vmcb->save.cpl;
2457 static int svm_get_cpl(struct kvm_vcpu *vcpu)
2459 struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
2464 static void svm_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
2466 struct vcpu_svm *svm = to_svm(vcpu);
2468 dt->size = svm->vmcb->save.idtr.limit;
2469 dt->address = svm->vmcb->save.idtr.base;
2472 static void svm_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
2474 struct vcpu_svm *svm = to_svm(vcpu);
2476 svm->vmcb->save.idtr.limit = dt->size;
2477 svm->vmcb->save.idtr.base = dt->address ;
2478 mark_dirty(svm->vmcb, VMCB_DT);
2481 static void svm_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
2483 struct vcpu_svm *svm = to_svm(vcpu);
2485 dt->size = svm->vmcb->save.gdtr.limit;
2486 dt->address = svm->vmcb->save.gdtr.base;
2489 static void svm_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
2491 struct vcpu_svm *svm = to_svm(vcpu);
2493 svm->vmcb->save.gdtr.limit = dt->size;
2494 svm->vmcb->save.gdtr.base = dt->address ;
2495 mark_dirty(svm->vmcb, VMCB_DT);
2498 static void svm_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
2502 static void svm_decache_cr3(struct kvm_vcpu *vcpu)
2506 static void svm_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
2510 static void update_cr0_intercept(struct vcpu_svm *svm)
2512 ulong gcr0 = svm->vcpu.arch.cr0;
2513 u64 *hcr0 = &svm->vmcb->save.cr0;
2515 *hcr0 = (*hcr0 & ~SVM_CR0_SELECTIVE_MASK)
2516 | (gcr0 & SVM_CR0_SELECTIVE_MASK);
2518 mark_dirty(svm->vmcb, VMCB_CR);
2520 if (gcr0 == *hcr0) {
2521 clr_cr_intercept(svm, INTERCEPT_CR0_READ);
2522 clr_cr_intercept(svm, INTERCEPT_CR0_WRITE);
2524 set_cr_intercept(svm, INTERCEPT_CR0_READ);
2525 set_cr_intercept(svm, INTERCEPT_CR0_WRITE);
2529 static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
2531 struct vcpu_svm *svm = to_svm(vcpu);
2533 #ifdef CONFIG_X86_64
2534 if (vcpu->arch.efer & EFER_LME) {
2535 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
2536 vcpu->arch.efer |= EFER_LMA;
2537 svm->vmcb->save.efer |= EFER_LMA | EFER_LME;
2540 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG)) {
2541 vcpu->arch.efer &= ~EFER_LMA;
2542 svm->vmcb->save.efer &= ~(EFER_LMA | EFER_LME);
2546 vcpu->arch.cr0 = cr0;
2549 cr0 |= X86_CR0_PG | X86_CR0_WP;
2552 * re-enable caching here because the QEMU bios
2553 * does not do it - this results in some delay at
2556 if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
2557 cr0 &= ~(X86_CR0_CD | X86_CR0_NW);
2558 svm->vmcb->save.cr0 = cr0;
2559 mark_dirty(svm->vmcb, VMCB_CR);
2560 update_cr0_intercept(svm);
2563 static int svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
2565 unsigned long host_cr4_mce = cr4_read_shadow() & X86_CR4_MCE;
2566 unsigned long old_cr4 = to_svm(vcpu)->vmcb->save.cr4;
2568 if (cr4 & X86_CR4_VMXE)
2571 if (npt_enabled && ((old_cr4 ^ cr4) & X86_CR4_PGE))
2572 svm_flush_tlb(vcpu, true);
2574 vcpu->arch.cr4 = cr4;
2577 cr4 |= host_cr4_mce;
2578 to_svm(vcpu)->vmcb->save.cr4 = cr4;
2579 mark_dirty(to_svm(vcpu)->vmcb, VMCB_CR);
2583 static void svm_set_segment(struct kvm_vcpu *vcpu,
2584 struct kvm_segment *var, int seg)
2586 struct vcpu_svm *svm = to_svm(vcpu);
2587 struct vmcb_seg *s = svm_seg(vcpu, seg);
2589 s->base = var->base;
2590 s->limit = var->limit;
2591 s->selector = var->selector;
2592 s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK);
2593 s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT;
2594 s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT;
2595 s->attrib |= ((var->present & 1) && !var->unusable) << SVM_SELECTOR_P_SHIFT;
2596 s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT;
2597 s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT;
2598 s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT;
2599 s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT;
2602 * This is always accurate, except if SYSRET returned to a segment
2603 * with SS.DPL != 3. Intel does not have this quirk, and always
2604 * forces SS.DPL to 3 on sysret, so we ignore that case; fixing it
2605 * would entail passing the CPL to userspace and back.
2607 if (seg == VCPU_SREG_SS)
2608 /* This is symmetric with svm_get_segment() */
2609 svm->vmcb->save.cpl = (var->dpl & 3);
2611 mark_dirty(svm->vmcb, VMCB_SEG);
2614 static void update_bp_intercept(struct kvm_vcpu *vcpu)
2616 struct vcpu_svm *svm = to_svm(vcpu);
2618 clr_exception_intercept(svm, BP_VECTOR);
2620 if (vcpu->guest_debug & KVM_GUESTDBG_ENABLE) {
2621 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
2622 set_exception_intercept(svm, BP_VECTOR);
2624 vcpu->guest_debug = 0;
2627 static void new_asid(struct vcpu_svm *svm, struct svm_cpu_data *sd)
2629 if (sd->next_asid > sd->max_asid) {
2630 ++sd->asid_generation;
2631 sd->next_asid = sd->min_asid;
2632 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID;
2635 svm->asid_generation = sd->asid_generation;
2636 svm->vmcb->control.asid = sd->next_asid++;
2638 mark_dirty(svm->vmcb, VMCB_ASID);
2641 static u64 svm_get_dr6(struct kvm_vcpu *vcpu)
2643 return to_svm(vcpu)->vmcb->save.dr6;
2646 static void svm_set_dr6(struct kvm_vcpu *vcpu, unsigned long value)
2648 struct vcpu_svm *svm = to_svm(vcpu);
2650 svm->vmcb->save.dr6 = value;
2651 mark_dirty(svm->vmcb, VMCB_DR);
2654 static void svm_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
2656 struct vcpu_svm *svm = to_svm(vcpu);
2658 get_debugreg(vcpu->arch.db[0], 0);
2659 get_debugreg(vcpu->arch.db[1], 1);
2660 get_debugreg(vcpu->arch.db[2], 2);
2661 get_debugreg(vcpu->arch.db[3], 3);
2662 vcpu->arch.dr6 = svm_get_dr6(vcpu);
2663 vcpu->arch.dr7 = svm->vmcb->save.dr7;
2665 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
2666 set_dr_intercepts(svm);
2669 static void svm_set_dr7(struct kvm_vcpu *vcpu, unsigned long value)
2671 struct vcpu_svm *svm = to_svm(vcpu);
2673 svm->vmcb->save.dr7 = value;
2674 mark_dirty(svm->vmcb, VMCB_DR);
2677 static int pf_interception(struct vcpu_svm *svm)
2679 u64 fault_address = __sme_clr(svm->vmcb->control.exit_info_2);
2680 u64 error_code = svm->vmcb->control.exit_info_1;
2682 return kvm_handle_page_fault(&svm->vcpu, error_code, fault_address,
2683 static_cpu_has(X86_FEATURE_DECODEASSISTS) ?
2684 svm->vmcb->control.insn_bytes : NULL,
2685 svm->vmcb->control.insn_len);
2688 static int npf_interception(struct vcpu_svm *svm)
2690 u64 fault_address = __sme_clr(svm->vmcb->control.exit_info_2);
2691 u64 error_code = svm->vmcb->control.exit_info_1;
2693 trace_kvm_page_fault(fault_address, error_code);
2694 return kvm_mmu_page_fault(&svm->vcpu, fault_address, error_code,
2695 static_cpu_has(X86_FEATURE_DECODEASSISTS) ?
2696 svm->vmcb->control.insn_bytes : NULL,
2697 svm->vmcb->control.insn_len);
2700 static int db_interception(struct vcpu_svm *svm)
2702 struct kvm_run *kvm_run = svm->vcpu.run;
2703 struct kvm_vcpu *vcpu = &svm->vcpu;
2705 if (!(svm->vcpu.guest_debug &
2706 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) &&
2707 !svm->nmi_singlestep) {
2708 kvm_queue_exception(&svm->vcpu, DB_VECTOR);
2712 if (svm->nmi_singlestep) {
2713 disable_nmi_singlestep(svm);
2714 /* Make sure we check for pending NMIs upon entry */
2715 kvm_make_request(KVM_REQ_EVENT, vcpu);
2718 if (svm->vcpu.guest_debug &
2719 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) {
2720 kvm_run->exit_reason = KVM_EXIT_DEBUG;
2721 kvm_run->debug.arch.pc =
2722 svm->vmcb->save.cs.base + svm->vmcb->save.rip;
2723 kvm_run->debug.arch.exception = DB_VECTOR;
2730 static int bp_interception(struct vcpu_svm *svm)
2732 struct kvm_run *kvm_run = svm->vcpu.run;
2734 kvm_run->exit_reason = KVM_EXIT_DEBUG;
2735 kvm_run->debug.arch.pc = svm->vmcb->save.cs.base + svm->vmcb->save.rip;
2736 kvm_run->debug.arch.exception = BP_VECTOR;
2740 static int ud_interception(struct vcpu_svm *svm)
2742 return handle_ud(&svm->vcpu);
2745 static int ac_interception(struct vcpu_svm *svm)
2747 kvm_queue_exception_e(&svm->vcpu, AC_VECTOR, 0);
2751 static int gp_interception(struct vcpu_svm *svm)
2753 struct kvm_vcpu *vcpu = &svm->vcpu;
2754 u32 error_code = svm->vmcb->control.exit_info_1;
2757 WARN_ON_ONCE(!enable_vmware_backdoor);
2759 er = kvm_emulate_instruction(vcpu,
2760 EMULTYPE_VMWARE | EMULTYPE_NO_UD_ON_FAIL);
2761 if (er == EMULATE_USER_EXIT)
2763 else if (er != EMULATE_DONE)
2764 kvm_queue_exception_e(vcpu, GP_VECTOR, error_code);
2768 static bool is_erratum_383(void)
2773 if (!erratum_383_found)
2776 value = native_read_msr_safe(MSR_IA32_MC0_STATUS, &err);
2780 /* Bit 62 may or may not be set for this mce */
2781 value &= ~(1ULL << 62);
2783 if (value != 0xb600000000010015ULL)
2786 /* Clear MCi_STATUS registers */
2787 for (i = 0; i < 6; ++i)
2788 native_write_msr_safe(MSR_IA32_MCx_STATUS(i), 0, 0);
2790 value = native_read_msr_safe(MSR_IA32_MCG_STATUS, &err);
2794 value &= ~(1ULL << 2);
2795 low = lower_32_bits(value);
2796 high = upper_32_bits(value);
2798 native_write_msr_safe(MSR_IA32_MCG_STATUS, low, high);
2801 /* Flush tlb to evict multi-match entries */
2807 static void svm_handle_mce(struct vcpu_svm *svm)
2809 if (is_erratum_383()) {
2811 * Erratum 383 triggered. Guest state is corrupt so kill the
2814 pr_err("KVM: Guest triggered AMD Erratum 383\n");
2816 kvm_make_request(KVM_REQ_TRIPLE_FAULT, &svm->vcpu);
2822 * On an #MC intercept the MCE handler is not called automatically in
2823 * the host. So do it by hand here.
2827 /* not sure if we ever come back to this point */
2832 static int mc_interception(struct vcpu_svm *svm)
2837 static int shutdown_interception(struct vcpu_svm *svm)
2839 struct kvm_run *kvm_run = svm->vcpu.run;
2842 * VMCB is undefined after a SHUTDOWN intercept
2843 * so reinitialize it.
2845 clear_page(svm->vmcb);
2848 kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
2852 static int io_interception(struct vcpu_svm *svm)
2854 struct kvm_vcpu *vcpu = &svm->vcpu;
2855 u32 io_info = svm->vmcb->control.exit_info_1; /* address size bug? */
2856 int size, in, string;
2859 ++svm->vcpu.stat.io_exits;
2860 string = (io_info & SVM_IOIO_STR_MASK) != 0;
2861 in = (io_info & SVM_IOIO_TYPE_MASK) != 0;
2863 return kvm_emulate_instruction(vcpu, 0) == EMULATE_DONE;
2865 port = io_info >> 16;
2866 size = (io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT;
2867 svm->next_rip = svm->vmcb->control.exit_info_2;
2869 return kvm_fast_pio(&svm->vcpu, size, port, in);
2872 static int nmi_interception(struct vcpu_svm *svm)
2877 static int intr_interception(struct vcpu_svm *svm)
2879 ++svm->vcpu.stat.irq_exits;
2883 static int nop_on_interception(struct vcpu_svm *svm)
2888 static int halt_interception(struct vcpu_svm *svm)
2890 svm->next_rip = kvm_rip_read(&svm->vcpu) + 1;
2891 return kvm_emulate_halt(&svm->vcpu);
2894 static int vmmcall_interception(struct vcpu_svm *svm)
2896 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2897 return kvm_emulate_hypercall(&svm->vcpu);
2900 static unsigned long nested_svm_get_tdp_cr3(struct kvm_vcpu *vcpu)
2902 struct vcpu_svm *svm = to_svm(vcpu);
2904 return svm->nested.nested_cr3;
2907 static u64 nested_svm_get_tdp_pdptr(struct kvm_vcpu *vcpu, int index)
2909 struct vcpu_svm *svm = to_svm(vcpu);
2910 u64 cr3 = svm->nested.nested_cr3;
2914 ret = kvm_vcpu_read_guest_page(vcpu, gpa_to_gfn(__sme_clr(cr3)), &pdpte,
2915 offset_in_page(cr3) + index * 8, 8);
2921 static void nested_svm_set_tdp_cr3(struct kvm_vcpu *vcpu,
2924 struct vcpu_svm *svm = to_svm(vcpu);
2926 svm->vmcb->control.nested_cr3 = __sme_set(root);
2927 mark_dirty(svm->vmcb, VMCB_NPT);
2930 static void nested_svm_inject_npf_exit(struct kvm_vcpu *vcpu,
2931 struct x86_exception *fault)
2933 struct vcpu_svm *svm = to_svm(vcpu);
2935 if (svm->vmcb->control.exit_code != SVM_EXIT_NPF) {
2937 * TODO: track the cause of the nested page fault, and
2938 * correctly fill in the high bits of exit_info_1.
2940 svm->vmcb->control.exit_code = SVM_EXIT_NPF;
2941 svm->vmcb->control.exit_code_hi = 0;
2942 svm->vmcb->control.exit_info_1 = (1ULL << 32);
2943 svm->vmcb->control.exit_info_2 = fault->address;
2946 svm->vmcb->control.exit_info_1 &= ~0xffffffffULL;
2947 svm->vmcb->control.exit_info_1 |= fault->error_code;
2950 * The present bit is always zero for page structure faults on real
2953 if (svm->vmcb->control.exit_info_1 & (2ULL << 32))
2954 svm->vmcb->control.exit_info_1 &= ~1;
2956 nested_svm_vmexit(svm);
2959 static void nested_svm_init_mmu_context(struct kvm_vcpu *vcpu)
2961 WARN_ON(mmu_is_nested(vcpu));
2963 vcpu->arch.mmu = &vcpu->arch.guest_mmu;
2964 kvm_init_shadow_mmu(vcpu);
2965 vcpu->arch.mmu->set_cr3 = nested_svm_set_tdp_cr3;
2966 vcpu->arch.mmu->get_cr3 = nested_svm_get_tdp_cr3;
2967 vcpu->arch.mmu->get_pdptr = nested_svm_get_tdp_pdptr;
2968 vcpu->arch.mmu->inject_page_fault = nested_svm_inject_npf_exit;
2969 vcpu->arch.mmu->shadow_root_level = get_npt_level(vcpu);
2970 reset_shadow_zero_bits_mask(vcpu, vcpu->arch.mmu);
2971 vcpu->arch.walk_mmu = &vcpu->arch.nested_mmu;
2974 static void nested_svm_uninit_mmu_context(struct kvm_vcpu *vcpu)
2976 vcpu->arch.mmu = &vcpu->arch.root_mmu;
2977 vcpu->arch.walk_mmu = &vcpu->arch.root_mmu;
2980 static int nested_svm_check_permissions(struct vcpu_svm *svm)
2982 if (!(svm->vcpu.arch.efer & EFER_SVME) ||
2983 !is_paging(&svm->vcpu)) {
2984 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2988 if (svm->vmcb->save.cpl) {
2989 kvm_inject_gp(&svm->vcpu, 0);
2996 static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
2997 bool has_error_code, u32 error_code)
3001 if (!is_guest_mode(&svm->vcpu))
3004 vmexit = nested_svm_intercept(svm);
3005 if (vmexit != NESTED_EXIT_DONE)
3008 svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + nr;
3009 svm->vmcb->control.exit_code_hi = 0;
3010 svm->vmcb->control.exit_info_1 = error_code;
3013 * EXITINFO2 is undefined for all exception intercepts other
3016 if (svm->vcpu.arch.exception.nested_apf)
3017 svm->vmcb->control.exit_info_2 = svm->vcpu.arch.apf.nested_apf_token;
3018 else if (svm->vcpu.arch.exception.has_payload)
3019 svm->vmcb->control.exit_info_2 = svm->vcpu.arch.exception.payload;
3021 svm->vmcb->control.exit_info_2 = svm->vcpu.arch.cr2;
3023 svm->nested.exit_required = true;
3027 /* This function returns true if it is save to enable the irq window */
3028 static inline bool nested_svm_intr(struct vcpu_svm *svm)
3030 if (!is_guest_mode(&svm->vcpu))
3033 if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
3036 if (!(svm->vcpu.arch.hflags & HF_HIF_MASK))
3040 * if vmexit was already requested (by intercepted exception
3041 * for instance) do not overwrite it with "external interrupt"
3044 if (svm->nested.exit_required)
3047 svm->vmcb->control.exit_code = SVM_EXIT_INTR;
3048 svm->vmcb->control.exit_info_1 = 0;
3049 svm->vmcb->control.exit_info_2 = 0;
3051 if (svm->nested.intercept & 1ULL) {
3053 * The #vmexit can't be emulated here directly because this
3054 * code path runs with irqs and preemption disabled. A
3055 * #vmexit emulation might sleep. Only signal request for
3058 svm->nested.exit_required = true;
3059 trace_kvm_nested_intr_vmexit(svm->vmcb->save.rip);
3066 /* This function returns true if it is save to enable the nmi window */
3067 static inline bool nested_svm_nmi(struct vcpu_svm *svm)
3069 if (!is_guest_mode(&svm->vcpu))
3072 if (!(svm->nested.intercept & (1ULL << INTERCEPT_NMI)))
3075 svm->vmcb->control.exit_code = SVM_EXIT_NMI;
3076 svm->nested.exit_required = true;
3081 static int nested_svm_intercept_ioio(struct vcpu_svm *svm)
3083 unsigned port, size, iopm_len;
3088 if (!(svm->nested.intercept & (1ULL << INTERCEPT_IOIO_PROT)))
3089 return NESTED_EXIT_HOST;
3091 port = svm->vmcb->control.exit_info_1 >> 16;
3092 size = (svm->vmcb->control.exit_info_1 & SVM_IOIO_SIZE_MASK) >>
3093 SVM_IOIO_SIZE_SHIFT;
3094 gpa = svm->nested.vmcb_iopm + (port / 8);
3095 start_bit = port % 8;
3096 iopm_len = (start_bit + size > 8) ? 2 : 1;
3097 mask = (0xf >> (4 - size)) << start_bit;
3100 if (kvm_vcpu_read_guest(&svm->vcpu, gpa, &val, iopm_len))
3101 return NESTED_EXIT_DONE;
3103 return (val & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
3106 static int nested_svm_exit_handled_msr(struct vcpu_svm *svm)
3108 u32 offset, msr, value;
3111 if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT)))
3112 return NESTED_EXIT_HOST;
3114 msr = svm->vcpu.arch.regs[VCPU_REGS_RCX];
3115 offset = svm_msrpm_offset(msr);
3116 write = svm->vmcb->control.exit_info_1 & 1;
3117 mask = 1 << ((2 * (msr & 0xf)) + write);
3119 if (offset == MSR_INVALID)
3120 return NESTED_EXIT_DONE;
3122 /* Offset is in 32 bit units but need in 8 bit units */
3125 if (kvm_vcpu_read_guest(&svm->vcpu, svm->nested.vmcb_msrpm + offset, &value, 4))
3126 return NESTED_EXIT_DONE;
3128 return (value & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
3131 /* DB exceptions for our internal use must not cause vmexit */
3132 static int nested_svm_intercept_db(struct vcpu_svm *svm)
3136 /* if we're not singlestepping, it's not ours */
3137 if (!svm->nmi_singlestep)
3138 return NESTED_EXIT_DONE;
3140 /* if it's not a singlestep exception, it's not ours */
3141 if (kvm_get_dr(&svm->vcpu, 6, &dr6))
3142 return NESTED_EXIT_DONE;
3143 if (!(dr6 & DR6_BS))
3144 return NESTED_EXIT_DONE;
3146 /* if the guest is singlestepping, it should get the vmexit */
3147 if (svm->nmi_singlestep_guest_rflags & X86_EFLAGS_TF) {
3148 disable_nmi_singlestep(svm);
3149 return NESTED_EXIT_DONE;
3152 /* it's ours, the nested hypervisor must not see this one */
3153 return NESTED_EXIT_HOST;
3156 static int nested_svm_exit_special(struct vcpu_svm *svm)
3158 u32 exit_code = svm->vmcb->control.exit_code;
3160 switch (exit_code) {
3163 case SVM_EXIT_EXCP_BASE + MC_VECTOR:
3164 return NESTED_EXIT_HOST;
3166 /* For now we are always handling NPFs when using them */
3168 return NESTED_EXIT_HOST;
3170 case SVM_EXIT_EXCP_BASE + PF_VECTOR:
3171 /* When we're shadowing, trap PFs, but not async PF */
3172 if (!npt_enabled && svm->vcpu.arch.apf.host_apf_reason == 0)
3173 return NESTED_EXIT_HOST;
3179 return NESTED_EXIT_CONTINUE;
3183 * If this function returns true, this #vmexit was already handled
3185 static int nested_svm_intercept(struct vcpu_svm *svm)
3187 u32 exit_code = svm->vmcb->control.exit_code;
3188 int vmexit = NESTED_EXIT_HOST;
3190 switch (exit_code) {
3192 vmexit = nested_svm_exit_handled_msr(svm);
3195 vmexit = nested_svm_intercept_ioio(svm);
3197 case SVM_EXIT_READ_CR0 ... SVM_EXIT_WRITE_CR8: {
3198 u32 bit = 1U << (exit_code - SVM_EXIT_READ_CR0);
3199 if (svm->nested.intercept_cr & bit)
3200 vmexit = NESTED_EXIT_DONE;
3203 case SVM_EXIT_READ_DR0 ... SVM_EXIT_WRITE_DR7: {
3204 u32 bit = 1U << (exit_code - SVM_EXIT_READ_DR0);
3205 if (svm->nested.intercept_dr & bit)
3206 vmexit = NESTED_EXIT_DONE;
3209 case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
3210 u32 excp_bits = 1 << (exit_code - SVM_EXIT_EXCP_BASE);
3211 if (svm->nested.intercept_exceptions & excp_bits) {
3212 if (exit_code == SVM_EXIT_EXCP_BASE + DB_VECTOR)
3213 vmexit = nested_svm_intercept_db(svm);
3215 vmexit = NESTED_EXIT_DONE;
3217 /* async page fault always cause vmexit */
3218 else if ((exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR) &&
3219 svm->vcpu.arch.exception.nested_apf != 0)
3220 vmexit = NESTED_EXIT_DONE;
3223 case SVM_EXIT_ERR: {
3224 vmexit = NESTED_EXIT_DONE;
3228 u64 exit_bits = 1ULL << (exit_code - SVM_EXIT_INTR);
3229 if (svm->nested.intercept & exit_bits)
3230 vmexit = NESTED_EXIT_DONE;
3237 static int nested_svm_exit_handled(struct vcpu_svm *svm)
3241 vmexit = nested_svm_intercept(svm);
3243 if (vmexit == NESTED_EXIT_DONE)
3244 nested_svm_vmexit(svm);
3249 static inline void copy_vmcb_control_area(struct vmcb *dst_vmcb, struct vmcb *from_vmcb)
3251 struct vmcb_control_area *dst = &dst_vmcb->control;
3252 struct vmcb_control_area *from = &from_vmcb->control;
3254 dst->intercept_cr = from->intercept_cr;
3255 dst->intercept_dr = from->intercept_dr;
3256 dst->intercept_exceptions = from->intercept_exceptions;
3257 dst->intercept = from->intercept;
3258 dst->iopm_base_pa = from->iopm_base_pa;
3259 dst->msrpm_base_pa = from->msrpm_base_pa;
3260 dst->tsc_offset = from->tsc_offset;
3261 dst->asid = from->asid;
3262 dst->tlb_ctl = from->tlb_ctl;
3263 dst->int_ctl = from->int_ctl;
3264 dst->int_vector = from->int_vector;
3265 dst->int_state = from->int_state;
3266 dst->exit_code = from->exit_code;
3267 dst->exit_code_hi = from->exit_code_hi;
3268 dst->exit_info_1 = from->exit_info_1;
3269 dst->exit_info_2 = from->exit_info_2;
3270 dst->exit_int_info = from->exit_int_info;
3271 dst->exit_int_info_err = from->exit_int_info_err;
3272 dst->nested_ctl = from->nested_ctl;
3273 dst->event_inj = from->event_inj;
3274 dst->event_inj_err = from->event_inj_err;
3275 dst->nested_cr3 = from->nested_cr3;
3276 dst->virt_ext = from->virt_ext;
3277 dst->pause_filter_count = from->pause_filter_count;
3278 dst->pause_filter_thresh = from->pause_filter_thresh;
3281 static int nested_svm_vmexit(struct vcpu_svm *svm)
3284 struct vmcb *nested_vmcb;
3285 struct vmcb *hsave = svm->nested.hsave;
3286 struct vmcb *vmcb = svm->vmcb;
3287 struct kvm_host_map map;
3289 trace_kvm_nested_vmexit_inject(vmcb->control.exit_code,
3290 vmcb->control.exit_info_1,
3291 vmcb->control.exit_info_2,
3292 vmcb->control.exit_int_info,
3293 vmcb->control.exit_int_info_err,
3296 rc = kvm_vcpu_map(&svm->vcpu, gfn_to_gpa(svm->nested.vmcb), &map);
3299 kvm_inject_gp(&svm->vcpu, 0);
3303 nested_vmcb = map.hva;
3305 /* Exit Guest-Mode */
3306 leave_guest_mode(&svm->vcpu);
3307 svm->nested.vmcb = 0;
3309 /* Give the current vmcb to the guest */
3312 nested_vmcb->save.es = vmcb->save.es;
3313 nested_vmcb->save.cs = vmcb->save.cs;
3314 nested_vmcb->save.ss = vmcb->save.ss;
3315 nested_vmcb->save.ds = vmcb->save.ds;
3316 nested_vmcb->save.gdtr = vmcb->save.gdtr;
3317 nested_vmcb->save.idtr = vmcb->save.idtr;
3318 nested_vmcb->save.efer = svm->vcpu.arch.efer;
3319 nested_vmcb->save.cr0 = kvm_read_cr0(&svm->vcpu);
3320 nested_vmcb->save.cr3 = kvm_read_cr3(&svm->vcpu);
3321 nested_vmcb->save.cr2 = vmcb->save.cr2;
3322 nested_vmcb->save.cr4 = svm->vcpu.arch.cr4;
3323 nested_vmcb->save.rflags = kvm_get_rflags(&svm->vcpu);
3324 nested_vmcb->save.rip = vmcb->save.rip;
3325 nested_vmcb->save.rsp = vmcb->save.rsp;
3326 nested_vmcb->save.rax = vmcb->save.rax;
3327 nested_vmcb->save.dr7 = vmcb->save.dr7;
3328 nested_vmcb->save.dr6 = vmcb->save.dr6;
3329 nested_vmcb->save.cpl = vmcb->save.cpl;
3331 nested_vmcb->control.int_ctl = vmcb->control.int_ctl;
3332 nested_vmcb->control.int_vector = vmcb->control.int_vector;
3333 nested_vmcb->control.int_state = vmcb->control.int_state;
3334 nested_vmcb->control.exit_code = vmcb->control.exit_code;
3335 nested_vmcb->control.exit_code_hi = vmcb->control.exit_code_hi;
3336 nested_vmcb->control.exit_info_1 = vmcb->control.exit_info_1;
3337 nested_vmcb->control.exit_info_2 = vmcb->control.exit_info_2;
3338 nested_vmcb->control.exit_int_info = vmcb->control.exit_int_info;
3339 nested_vmcb->control.exit_int_info_err = vmcb->control.exit_int_info_err;
3341 if (svm->nrips_enabled)
3342 nested_vmcb->control.next_rip = vmcb->control.next_rip;
3345 * If we emulate a VMRUN/#VMEXIT in the same host #vmexit cycle we have
3346 * to make sure that we do not lose injected events. So check event_inj
3347 * here and copy it to exit_int_info if it is valid.
3348 * Exit_int_info and event_inj can't be both valid because the case
3349 * below only happens on a VMRUN instruction intercept which has
3350 * no valid exit_int_info set.
3352 if (vmcb->control.event_inj & SVM_EVTINJ_VALID) {
3353 struct vmcb_control_area *nc = &nested_vmcb->control;
3355 nc->exit_int_info = vmcb->control.event_inj;
3356 nc->exit_int_info_err = vmcb->control.event_inj_err;
3359 nested_vmcb->control.tlb_ctl = 0;
3360 nested_vmcb->control.event_inj = 0;
3361 nested_vmcb->control.event_inj_err = 0;
3363 nested_vmcb->control.pause_filter_count =
3364 svm->vmcb->control.pause_filter_count;
3365 nested_vmcb->control.pause_filter_thresh =
3366 svm->vmcb->control.pause_filter_thresh;
3368 /* We always set V_INTR_MASKING and remember the old value in hflags */
3369 if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
3370 nested_vmcb->control.int_ctl &= ~V_INTR_MASKING_MASK;
3372 /* Restore the original control entries */
3373 copy_vmcb_control_area(vmcb, hsave);
3375 svm->vcpu.arch.tsc_offset = svm->vmcb->control.tsc_offset;
3376 kvm_clear_exception_queue(&svm->vcpu);
3377 kvm_clear_interrupt_queue(&svm->vcpu);
3379 svm->nested.nested_cr3 = 0;
3381 /* Restore selected save entries */
3382 svm->vmcb->save.es = hsave->save.es;
3383 svm->vmcb->save.cs = hsave->save.cs;
3384 svm->vmcb->save.ss = hsave->save.ss;
3385 svm->vmcb->save.ds = hsave->save.ds;
3386 svm->vmcb->save.gdtr = hsave->save.gdtr;
3387 svm->vmcb->save.idtr = hsave->save.idtr;
3388 kvm_set_rflags(&svm->vcpu, hsave->save.rflags);
3389 svm_set_efer(&svm->vcpu, hsave->save.efer);
3390 svm_set_cr0(&svm->vcpu, hsave->save.cr0 | X86_CR0_PE);
3391 svm_set_cr4(&svm->vcpu, hsave->save.cr4);
3393 svm->vmcb->save.cr3 = hsave->save.cr3;
3394 svm->vcpu.arch.cr3 = hsave->save.cr3;
3396 (void)kvm_set_cr3(&svm->vcpu, hsave->save.cr3);
3398 kvm_rax_write(&svm->vcpu, hsave->save.rax);
3399 kvm_rsp_write(&svm->vcpu, hsave->save.rsp);
3400 kvm_rip_write(&svm->vcpu, hsave->save.rip);
3401 svm->vmcb->save.dr7 = 0;
3402 svm->vmcb->save.cpl = 0;
3403 svm->vmcb->control.exit_int_info = 0;
3405 mark_all_dirty(svm->vmcb);
3407 kvm_vcpu_unmap(&svm->vcpu, &map, true);
3409 nested_svm_uninit_mmu_context(&svm->vcpu);
3410 kvm_mmu_reset_context(&svm->vcpu);
3411 kvm_mmu_load(&svm->vcpu);
3414 * Drop what we picked up for L2 via svm_complete_interrupts() so it
3415 * doesn't end up in L1.
3417 svm->vcpu.arch.nmi_injected = false;
3418 kvm_clear_exception_queue(&svm->vcpu);
3419 kvm_clear_interrupt_queue(&svm->vcpu);
3424 static bool nested_svm_vmrun_msrpm(struct vcpu_svm *svm)
3427 * This function merges the msr permission bitmaps of kvm and the
3428 * nested vmcb. It is optimized in that it only merges the parts where
3429 * the kvm msr permission bitmap may contain zero bits
3433 if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT)))
3436 for (i = 0; i < MSRPM_OFFSETS; i++) {
3440 if (msrpm_offsets[i] == 0xffffffff)
3443 p = msrpm_offsets[i];
3444 offset = svm->nested.vmcb_msrpm + (p * 4);
3446 if (kvm_vcpu_read_guest(&svm->vcpu, offset, &value, 4))
3449 svm->nested.msrpm[p] = svm->msrpm[p] | value;
3452 svm->vmcb->control.msrpm_base_pa = __sme_set(__pa(svm->nested.msrpm));
3457 static bool nested_vmcb_checks(struct vmcb *vmcb)
3459 if ((vmcb->control.intercept & (1ULL << INTERCEPT_VMRUN)) == 0)
3462 if (vmcb->control.asid == 0)
3465 if ((vmcb->control.nested_ctl & SVM_NESTED_CTL_NP_ENABLE) &&
3472 static void enter_svm_guest_mode(struct vcpu_svm *svm, u64 vmcb_gpa,
3473 struct vmcb *nested_vmcb, struct kvm_host_map *map)
3475 if (kvm_get_rflags(&svm->vcpu) & X86_EFLAGS_IF)
3476 svm->vcpu.arch.hflags |= HF_HIF_MASK;
3478 svm->vcpu.arch.hflags &= ~HF_HIF_MASK;
3480 if (nested_vmcb->control.nested_ctl & SVM_NESTED_CTL_NP_ENABLE) {
3481 svm->nested.nested_cr3 = nested_vmcb->control.nested_cr3;
3482 nested_svm_init_mmu_context(&svm->vcpu);
3485 /* Load the nested guest state */
3486 svm->vmcb->save.es = nested_vmcb->save.es;
3487 svm->vmcb->save.cs = nested_vmcb->save.cs;
3488 svm->vmcb->save.ss = nested_vmcb->save.ss;
3489 svm->vmcb->save.ds = nested_vmcb->save.ds;
3490 svm->vmcb->save.gdtr = nested_vmcb->save.gdtr;
3491 svm->vmcb->save.idtr = nested_vmcb->save.idtr;
3492 kvm_set_rflags(&svm->vcpu, nested_vmcb->save.rflags);
3493 svm_set_efer(&svm->vcpu, nested_vmcb->save.efer);
3494 svm_set_cr0(&svm->vcpu, nested_vmcb->save.cr0);
3495 svm_set_cr4(&svm->vcpu, nested_vmcb->save.cr4);
3497 svm->vmcb->save.cr3 = nested_vmcb->save.cr3;
3498 svm->vcpu.arch.cr3 = nested_vmcb->save.cr3;
3500 (void)kvm_set_cr3(&svm->vcpu, nested_vmcb->save.cr3);
3502 /* Guest paging mode is active - reset mmu */
3503 kvm_mmu_reset_context(&svm->vcpu);
3505 svm->vmcb->save.cr2 = svm->vcpu.arch.cr2 = nested_vmcb->save.cr2;
3506 kvm_rax_write(&svm->vcpu, nested_vmcb->save.rax);
3507 kvm_rsp_write(&svm->vcpu, nested_vmcb->save.rsp);
3508 kvm_rip_write(&svm->vcpu, nested_vmcb->save.rip);
3510 /* In case we don't even reach vcpu_run, the fields are not updated */
3511 svm->vmcb->save.rax = nested_vmcb->save.rax;
3512 svm->vmcb->save.rsp = nested_vmcb->save.rsp;
3513 svm->vmcb->save.rip = nested_vmcb->save.rip;
3514 svm->vmcb->save.dr7 = nested_vmcb->save.dr7;
3515 svm->vmcb->save.dr6 = nested_vmcb->save.dr6;
3516 svm->vmcb->save.cpl = nested_vmcb->save.cpl;
3518 svm->nested.vmcb_msrpm = nested_vmcb->control.msrpm_base_pa & ~0x0fffULL;
3519 svm->nested.vmcb_iopm = nested_vmcb->control.iopm_base_pa & ~0x0fffULL;
3521 /* cache intercepts */
3522 svm->nested.intercept_cr = nested_vmcb->control.intercept_cr;
3523 svm->nested.intercept_dr = nested_vmcb->control.intercept_dr;
3524 svm->nested.intercept_exceptions = nested_vmcb->control.intercept_exceptions;
3525 svm->nested.intercept = nested_vmcb->control.intercept;
3527 svm_flush_tlb(&svm->vcpu, true);
3528 svm->vmcb->control.int_ctl = nested_vmcb->control.int_ctl | V_INTR_MASKING_MASK;
3529 if (nested_vmcb->control.int_ctl & V_INTR_MASKING_MASK)
3530 svm->vcpu.arch.hflags |= HF_VINTR_MASK;
3532 svm->vcpu.arch.hflags &= ~HF_VINTR_MASK;
3534 if (svm->vcpu.arch.hflags & HF_VINTR_MASK) {
3535 /* We only want the cr8 intercept bits of the guest */
3536 clr_cr_intercept(svm, INTERCEPT_CR8_READ);
3537 clr_cr_intercept(svm, INTERCEPT_CR8_WRITE);
3540 /* We don't want to see VMMCALLs from a nested guest */
3541 clr_intercept(svm, INTERCEPT_VMMCALL);
3543 svm->vcpu.arch.tsc_offset += nested_vmcb->control.tsc_offset;
3544 svm->vmcb->control.tsc_offset = svm->vcpu.arch.tsc_offset;
3546 svm->vmcb->control.virt_ext = nested_vmcb->control.virt_ext;
3547 svm->vmcb->control.int_vector = nested_vmcb->control.int_vector;
3548 svm->vmcb->control.int_state = nested_vmcb->control.int_state;
3549 svm->vmcb->control.event_inj = nested_vmcb->control.event_inj;
3550 svm->vmcb->control.event_inj_err = nested_vmcb->control.event_inj_err;
3552 svm->vmcb->control.pause_filter_count =
3553 nested_vmcb->control.pause_filter_count;
3554 svm->vmcb->control.pause_filter_thresh =
3555 nested_vmcb->control.pause_filter_thresh;
3557 kvm_vcpu_unmap(&svm->vcpu, map, true);
3559 /* Enter Guest-Mode */
3560 enter_guest_mode(&svm->vcpu);
3563 * Merge guest and host intercepts - must be called with vcpu in
3564 * guest-mode to take affect here
3566 recalc_intercepts(svm);
3568 svm->nested.vmcb = vmcb_gpa;
3572 mark_all_dirty(svm->vmcb);
3575 static bool nested_svm_vmrun(struct vcpu_svm *svm)
3578 struct vmcb *nested_vmcb;
3579 struct vmcb *hsave = svm->nested.hsave;
3580 struct vmcb *vmcb = svm->vmcb;
3581 struct kvm_host_map map;
3584 vmcb_gpa = svm->vmcb->save.rax;
3586 rc = kvm_vcpu_map(&svm->vcpu, gfn_to_gpa(vmcb_gpa), &map);
3589 kvm_inject_gp(&svm->vcpu, 0);
3593 nested_vmcb = map.hva;
3595 if (!nested_vmcb_checks(nested_vmcb)) {
3596 nested_vmcb->control.exit_code = SVM_EXIT_ERR;
3597 nested_vmcb->control.exit_code_hi = 0;
3598 nested_vmcb->control.exit_info_1 = 0;
3599 nested_vmcb->control.exit_info_2 = 0;
3601 kvm_vcpu_unmap(&svm->vcpu, &map, true);
3606 trace_kvm_nested_vmrun(svm->vmcb->save.rip, vmcb_gpa,
3607 nested_vmcb->save.rip,
3608 nested_vmcb->control.int_ctl,
3609 nested_vmcb->control.event_inj,
3610 nested_vmcb->control.nested_ctl);
3612 trace_kvm_nested_intercepts(nested_vmcb->control.intercept_cr & 0xffff,
3613 nested_vmcb->control.intercept_cr >> 16,
3614 nested_vmcb->control.intercept_exceptions,
3615 nested_vmcb->control.intercept);
3617 /* Clear internal status */
3618 kvm_clear_exception_queue(&svm->vcpu);
3619 kvm_clear_interrupt_queue(&svm->vcpu);
3622 * Save the old vmcb, so we don't need to pick what we save, but can
3623 * restore everything when a VMEXIT occurs
3625 hsave->save.es = vmcb->save.es;
3626 hsave->save.cs = vmcb->save.cs;
3627 hsave->save.ss = vmcb->save.ss;
3628 hsave->save.ds = vmcb->save.ds;
3629 hsave->save.gdtr = vmcb->save.gdtr;
3630 hsave->save.idtr = vmcb->save.idtr;
3631 hsave->save.efer = svm->vcpu.arch.efer;
3632 hsave->save.cr0 = kvm_read_cr0(&svm->vcpu);
3633 hsave->save.cr4 = svm->vcpu.arch.cr4;
3634 hsave->save.rflags = kvm_get_rflags(&svm->vcpu);
3635 hsave->save.rip = kvm_rip_read(&svm->vcpu);
3636 hsave->save.rsp = vmcb->save.rsp;
3637 hsave->save.rax = vmcb->save.rax;
3639 hsave->save.cr3 = vmcb->save.cr3;
3641 hsave->save.cr3 = kvm_read_cr3(&svm->vcpu);
3643 copy_vmcb_control_area(hsave, vmcb);
3645 enter_svm_guest_mode(svm, vmcb_gpa, nested_vmcb, &map);
3650 static void nested_svm_vmloadsave(struct vmcb *from_vmcb, struct vmcb *to_vmcb)
3652 to_vmcb->save.fs = from_vmcb->save.fs;
3653 to_vmcb->save.gs = from_vmcb->save.gs;
3654 to_vmcb->save.tr = from_vmcb->save.tr;
3655 to_vmcb->save.ldtr = from_vmcb->save.ldtr;
3656 to_vmcb->save.kernel_gs_base = from_vmcb->save.kernel_gs_base;
3657 to_vmcb->save.star = from_vmcb->save.star;
3658 to_vmcb->save.lstar = from_vmcb->save.lstar;
3659 to_vmcb->save.cstar = from_vmcb->save.cstar;
3660 to_vmcb->save.sfmask = from_vmcb->save.sfmask;
3661 to_vmcb->save.sysenter_cs = from_vmcb->save.sysenter_cs;
3662 to_vmcb->save.sysenter_esp = from_vmcb->save.sysenter_esp;
3663 to_vmcb->save.sysenter_eip = from_vmcb->save.sysenter_eip;
3666 static int vmload_interception(struct vcpu_svm *svm)
3668 struct vmcb *nested_vmcb;
3669 struct kvm_host_map map;
3672 if (nested_svm_check_permissions(svm))
3675 ret = kvm_vcpu_map(&svm->vcpu, gpa_to_gfn(svm->vmcb->save.rax), &map);
3678 kvm_inject_gp(&svm->vcpu, 0);
3682 nested_vmcb = map.hva;
3684 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
3685 ret = kvm_skip_emulated_instruction(&svm->vcpu);
3687 nested_svm_vmloadsave(nested_vmcb, svm->vmcb);
3688 kvm_vcpu_unmap(&svm->vcpu, &map, true);
3693 static int vmsave_interception(struct vcpu_svm *svm)
3695 struct vmcb *nested_vmcb;
3696 struct kvm_host_map map;
3699 if (nested_svm_check_permissions(svm))
3702 ret = kvm_vcpu_map(&svm->vcpu, gpa_to_gfn(svm->vmcb->save.rax), &map);
3705 kvm_inject_gp(&svm->vcpu, 0);
3709 nested_vmcb = map.hva;
3711 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
3712 ret = kvm_skip_emulated_instruction(&svm->vcpu);
3714 nested_svm_vmloadsave(svm->vmcb, nested_vmcb);
3715 kvm_vcpu_unmap(&svm->vcpu, &map, true);
3720 static int vmrun_interception(struct vcpu_svm *svm)
3722 if (nested_svm_check_permissions(svm))
3725 /* Save rip after vmrun instruction */
3726 kvm_rip_write(&svm->vcpu, kvm_rip_read(&svm->vcpu) + 3);
3728 if (!nested_svm_vmrun(svm))
3731 if (!nested_svm_vmrun_msrpm(svm))
3738 svm->vmcb->control.exit_code = SVM_EXIT_ERR;
3739 svm->vmcb->control.exit_code_hi = 0;
3740 svm->vmcb->control.exit_info_1 = 0;
3741 svm->vmcb->control.exit_info_2 = 0;
3743 nested_svm_vmexit(svm);
3748 static int stgi_interception(struct vcpu_svm *svm)
3752 if (nested_svm_check_permissions(svm))
3756 * If VGIF is enabled, the STGI intercept is only added to
3757 * detect the opening of the SMI/NMI window; remove it now.
3759 if (vgif_enabled(svm))
3760 clr_intercept(svm, INTERCEPT_STGI);
3762 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
3763 ret = kvm_skip_emulated_instruction(&svm->vcpu);
3764 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
3771 static int clgi_interception(struct vcpu_svm *svm)
3775 if (nested_svm_check_permissions(svm))
3778 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
3779 ret = kvm_skip_emulated_instruction(&svm->vcpu);
3783 /* After a CLGI no interrupts should come */
3784 if (!kvm_vcpu_apicv_active(&svm->vcpu)) {
3785 svm_clear_vintr(svm);
3786 svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
3787 mark_dirty(svm->vmcb, VMCB_INTR);
3793 static int invlpga_interception(struct vcpu_svm *svm)
3795 struct kvm_vcpu *vcpu = &svm->vcpu;
3797 trace_kvm_invlpga(svm->vmcb->save.rip, kvm_rcx_read(&svm->vcpu),
3798 kvm_rax_read(&svm->vcpu));
3800 /* Let's treat INVLPGA the same as INVLPG (can be optimized!) */
3801 kvm_mmu_invlpg(vcpu, kvm_rax_read(&svm->vcpu));
3803 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
3804 return kvm_skip_emulated_instruction(&svm->vcpu);
3807 static int skinit_interception(struct vcpu_svm *svm)
3809 trace_kvm_skinit(svm->vmcb->save.rip, kvm_rax_read(&svm->vcpu));
3811 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
3815 static int wbinvd_interception(struct vcpu_svm *svm)
3817 return kvm_emulate_wbinvd(&svm->vcpu);
3820 static int xsetbv_interception(struct vcpu_svm *svm)
3822 u64 new_bv = kvm_read_edx_eax(&svm->vcpu);
3823 u32 index = kvm_rcx_read(&svm->vcpu);
3825 if (kvm_set_xcr(&svm->vcpu, index, new_bv) == 0) {
3826 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
3827 return kvm_skip_emulated_instruction(&svm->vcpu);
3833 static int task_switch_interception(struct vcpu_svm *svm)
3837 int int_type = svm->vmcb->control.exit_int_info &
3838 SVM_EXITINTINFO_TYPE_MASK;
3839 int int_vec = svm->vmcb->control.exit_int_info & SVM_EVTINJ_VEC_MASK;
3841 svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_TYPE_MASK;
3843 svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_VALID;
3844 bool has_error_code = false;
3847 tss_selector = (u16)svm->vmcb->control.exit_info_1;
3849 if (svm->vmcb->control.exit_info_2 &
3850 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_IRET))
3851 reason = TASK_SWITCH_IRET;
3852 else if (svm->vmcb->control.exit_info_2 &
3853 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_JMP))
3854 reason = TASK_SWITCH_JMP;
3856 reason = TASK_SWITCH_GATE;
3858 reason = TASK_SWITCH_CALL;
3860 if (reason == TASK_SWITCH_GATE) {
3862 case SVM_EXITINTINFO_TYPE_NMI:
3863 svm->vcpu.arch.nmi_injected = false;
3865 case SVM_EXITINTINFO_TYPE_EXEPT:
3866 if (svm->vmcb->control.exit_info_2 &
3867 (1ULL << SVM_EXITINFOSHIFT_TS_HAS_ERROR_CODE)) {
3868 has_error_code = true;
3870 (u32)svm->vmcb->control.exit_info_2;
3872 kvm_clear_exception_queue(&svm->vcpu);
3874 case SVM_EXITINTINFO_TYPE_INTR:
3875 kvm_clear_interrupt_queue(&svm->vcpu);
3882 if (reason != TASK_SWITCH_GATE ||
3883 int_type == SVM_EXITINTINFO_TYPE_SOFT ||
3884 (int_type == SVM_EXITINTINFO_TYPE_EXEPT &&
3885 (int_vec == OF_VECTOR || int_vec == BP_VECTOR)))
3886 skip_emulated_instruction(&svm->vcpu);
3888 if (int_type != SVM_EXITINTINFO_TYPE_SOFT)
3891 if (kvm_task_switch(&svm->vcpu, tss_selector, int_vec, reason,
3892 has_error_code, error_code) == EMULATE_FAIL) {
3893 svm->vcpu.run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3894 svm->vcpu.run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
3895 svm->vcpu.run->internal.ndata = 0;
3901 static int cpuid_interception(struct vcpu_svm *svm)
3903 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
3904 return kvm_emulate_cpuid(&svm->vcpu);
3907 static int iret_interception(struct vcpu_svm *svm)
3909 ++svm->vcpu.stat.nmi_window_exits;
3910 clr_intercept(svm, INTERCEPT_IRET);
3911 svm->vcpu.arch.hflags |= HF_IRET_MASK;
3912 svm->nmi_iret_rip = kvm_rip_read(&svm->vcpu);
3913 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
3917 static int invlpg_interception(struct vcpu_svm *svm)
3919 if (!static_cpu_has(X86_FEATURE_DECODEASSISTS))
3920 return kvm_emulate_instruction(&svm->vcpu, 0) == EMULATE_DONE;
3922 kvm_mmu_invlpg(&svm->vcpu, svm->vmcb->control.exit_info_1);
3923 return kvm_skip_emulated_instruction(&svm->vcpu);
3926 static int emulate_on_interception(struct vcpu_svm *svm)
3928 return kvm_emulate_instruction(&svm->vcpu, 0) == EMULATE_DONE;
3931 static int rsm_interception(struct vcpu_svm *svm)
3933 return kvm_emulate_instruction_from_buffer(&svm->vcpu,
3934 rsm_ins_bytes, 2) == EMULATE_DONE;
3937 static int rdpmc_interception(struct vcpu_svm *svm)
3941 if (!static_cpu_has(X86_FEATURE_NRIPS))
3942 return emulate_on_interception(svm);
3944 err = kvm_rdpmc(&svm->vcpu);
3945 return kvm_complete_insn_gp(&svm->vcpu, err);
3948 static bool check_selective_cr0_intercepted(struct vcpu_svm *svm,
3951 unsigned long cr0 = svm->vcpu.arch.cr0;
3955 intercept = svm->nested.intercept;
3957 if (!is_guest_mode(&svm->vcpu) ||
3958 (!(intercept & (1ULL << INTERCEPT_SELECTIVE_CR0))))
3961 cr0 &= ~SVM_CR0_SELECTIVE_MASK;
3962 val &= ~SVM_CR0_SELECTIVE_MASK;
3965 svm->vmcb->control.exit_code = SVM_EXIT_CR0_SEL_WRITE;
3966 ret = (nested_svm_exit_handled(svm) == NESTED_EXIT_DONE);
3972 #define CR_VALID (1ULL << 63)
3974 static int cr_interception(struct vcpu_svm *svm)
3980 if (!static_cpu_has(X86_FEATURE_DECODEASSISTS))
3981 return emulate_on_interception(svm);
3983 if (unlikely((svm->vmcb->control.exit_info_1 & CR_VALID) == 0))
3984 return emulate_on_interception(svm);
3986 reg = svm->vmcb->control.exit_info_1 & SVM_EXITINFO_REG_MASK;
3987 if (svm->vmcb->control.exit_code == SVM_EXIT_CR0_SEL_WRITE)
3988 cr = SVM_EXIT_WRITE_CR0 - SVM_EXIT_READ_CR0;
3990 cr = svm->vmcb->control.exit_code - SVM_EXIT_READ_CR0;
3993 if (cr >= 16) { /* mov to cr */
3995 val = kvm_register_read(&svm->vcpu, reg);
3998 if (!check_selective_cr0_intercepted(svm, val))
3999 err = kvm_set_cr0(&svm->vcpu, val);
4005 err = kvm_set_cr3(&svm->vcpu, val);
4008 err = kvm_set_cr4(&svm->vcpu, val);
4011 err = kvm_set_cr8(&svm->vcpu, val);
4014 WARN(1, "unhandled write to CR%d", cr);
4015 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
4018 } else { /* mov from cr */
4021 val = kvm_read_cr0(&svm->vcpu);
4024 val = svm->vcpu.arch.cr2;
4027 val = kvm_read_cr3(&svm->vcpu);
4030 val = kvm_read_cr4(&svm->vcpu);
4033 val = kvm_get_cr8(&svm->vcpu);
4036 WARN(1, "unhandled read from CR%d", cr);
4037 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
4040 kvm_register_write(&svm->vcpu, reg, val);
4042 return kvm_complete_insn_gp(&svm->vcpu, err);
4045 static int dr_interception(struct vcpu_svm *svm)
4050 if (svm->vcpu.guest_debug == 0) {
4052 * No more DR vmexits; force a reload of the debug registers
4053 * and reenter on this instruction. The next vmexit will
4054 * retrieve the full state of the debug registers.
4056 clr_dr_intercepts(svm);
4057 svm->vcpu.arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
4061 if (!boot_cpu_has(X86_FEATURE_DECODEASSISTS))
4062 return emulate_on_interception(svm);
4064 reg = svm->vmcb->control.exit_info_1 & SVM_EXITINFO_REG_MASK;
4065 dr = svm->vmcb->control.exit_code - SVM_EXIT_READ_DR0;
4067 if (dr >= 16) { /* mov to DRn */
4068 if (!kvm_require_dr(&svm->vcpu, dr - 16))
4070 val = kvm_register_read(&svm->vcpu, reg);
4071 kvm_set_dr(&svm->vcpu, dr - 16, val);
4073 if (!kvm_require_dr(&svm->vcpu, dr))
4075 kvm_get_dr(&svm->vcpu, dr, &val);
4076 kvm_register_write(&svm->vcpu, reg, val);
4079 return kvm_skip_emulated_instruction(&svm->vcpu);
4082 static int cr8_write_interception(struct vcpu_svm *svm)
4084 struct kvm_run *kvm_run = svm->vcpu.run;
4087 u8 cr8_prev = kvm_get_cr8(&svm->vcpu);
4088 /* instruction emulation calls kvm_set_cr8() */
4089 r = cr_interception(svm);
4090 if (lapic_in_kernel(&svm->vcpu))
4092 if (cr8_prev <= kvm_get_cr8(&svm->vcpu))
4094 kvm_run->exit_reason = KVM_EXIT_SET_TPR;
4098 static int svm_get_msr_feature(struct kvm_msr_entry *msr)
4102 switch (msr->index) {
4103 case MSR_F10H_DECFG:
4104 if (boot_cpu_has(X86_FEATURE_LFENCE_RDTSC))
4105 msr->data |= MSR_F10H_DECFG_LFENCE_SERIALIZE;
4114 static int svm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
4116 struct vcpu_svm *svm = to_svm(vcpu);
4118 switch (msr_info->index) {
4120 msr_info->data = svm->vmcb->save.star;
4122 #ifdef CONFIG_X86_64
4124 msr_info->data = svm->vmcb->save.lstar;
4127 msr_info->data = svm->vmcb->save.cstar;
4129 case MSR_KERNEL_GS_BASE:
4130 msr_info->data = svm->vmcb->save.kernel_gs_base;
4132 case MSR_SYSCALL_MASK:
4133 msr_info->data = svm->vmcb->save.sfmask;
4136 case MSR_IA32_SYSENTER_CS:
4137 msr_info->data = svm->vmcb->save.sysenter_cs;
4139 case MSR_IA32_SYSENTER_EIP:
4140 msr_info->data = svm->sysenter_eip;
4142 case MSR_IA32_SYSENTER_ESP:
4143 msr_info->data = svm->sysenter_esp;
4146 if (!boot_cpu_has(X86_FEATURE_RDTSCP))
4148 msr_info->data = svm->tsc_aux;
4151 * Nobody will change the following 5 values in the VMCB so we can
4152 * safely return them on rdmsr. They will always be 0 until LBRV is
4155 case MSR_IA32_DEBUGCTLMSR:
4156 msr_info->data = svm->vmcb->save.dbgctl;
4158 case MSR_IA32_LASTBRANCHFROMIP:
4159 msr_info->data = svm->vmcb->save.br_from;
4161 case MSR_IA32_LASTBRANCHTOIP:
4162 msr_info->data = svm->vmcb->save.br_to;
4164 case MSR_IA32_LASTINTFROMIP:
4165 msr_info->data = svm->vmcb->save.last_excp_from;
4167 case MSR_IA32_LASTINTTOIP:
4168 msr_info->data = svm->vmcb->save.last_excp_to;
4170 case MSR_VM_HSAVE_PA:
4171 msr_info->data = svm->nested.hsave_msr;
4174 msr_info->data = svm->nested.vm_cr_msr;
4176 case MSR_IA32_SPEC_CTRL:
4177 if (!msr_info->host_initiated &&
4178 !guest_cpuid_has(vcpu, X86_FEATURE_AMD_IBRS) &&
4179 !guest_cpuid_has(vcpu, X86_FEATURE_AMD_SSBD))
4182 msr_info->data = svm->spec_ctrl;
4184 case MSR_AMD64_VIRT_SPEC_CTRL:
4185 if (!msr_info->host_initiated &&
4186 !guest_cpuid_has(vcpu, X86_FEATURE_VIRT_SSBD))
4189 msr_info->data = svm->virt_spec_ctrl;
4191 case MSR_F15H_IC_CFG: {
4195 family = guest_cpuid_family(vcpu);
4196 model = guest_cpuid_model(vcpu);
4198 if (family < 0 || model < 0)
4199 return kvm_get_msr_common(vcpu, msr_info);
4203 if (family == 0x15 &&
4204 (model >= 0x2 && model < 0x20))
4205 msr_info->data = 0x1E;
4208 case MSR_F10H_DECFG:
4209 msr_info->data = svm->msr_decfg;
4212 return kvm_get_msr_common(vcpu, msr_info);
4217 static int rdmsr_interception(struct vcpu_svm *svm)
4219 u32 ecx = kvm_rcx_read(&svm->vcpu);
4220 struct msr_data msr_info;
4222 msr_info.index = ecx;
4223 msr_info.host_initiated = false;
4224 if (svm_get_msr(&svm->vcpu, &msr_info)) {
4225 trace_kvm_msr_read_ex(ecx);
4226 kvm_inject_gp(&svm->vcpu, 0);
4229 trace_kvm_msr_read(ecx, msr_info.data);
4231 kvm_rax_write(&svm->vcpu, msr_info.data & 0xffffffff);
4232 kvm_rdx_write(&svm->vcpu, msr_info.data >> 32);
4233 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
4234 return kvm_skip_emulated_instruction(&svm->vcpu);
4238 static int svm_set_vm_cr(struct kvm_vcpu *vcpu, u64 data)
4240 struct vcpu_svm *svm = to_svm(vcpu);
4241 int svm_dis, chg_mask;
4243 if (data & ~SVM_VM_CR_VALID_MASK)
4246 chg_mask = SVM_VM_CR_VALID_MASK;
4248 if (svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK)
4249 chg_mask &= ~(SVM_VM_CR_SVM_LOCK_MASK | SVM_VM_CR_SVM_DIS_MASK);
4251 svm->nested.vm_cr_msr &= ~chg_mask;
4252 svm->nested.vm_cr_msr |= (data & chg_mask);
4254 svm_dis = svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK;
4256 /* check for svm_disable while efer.svme is set */
4257 if (svm_dis && (vcpu->arch.efer & EFER_SVME))
4263 static int svm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
4265 struct vcpu_svm *svm = to_svm(vcpu);
4267 u32 ecx = msr->index;
4268 u64 data = msr->data;
4270 case MSR_IA32_CR_PAT:
4271 if (!kvm_mtrr_valid(vcpu, MSR_IA32_CR_PAT, data))
4273 vcpu->arch.pat = data;
4274 svm->vmcb->save.g_pat = data;
4275 mark_dirty(svm->vmcb, VMCB_NPT);
4277 case MSR_IA32_SPEC_CTRL:
4278 if (!msr->host_initiated &&
4279 !guest_cpuid_has(vcpu, X86_FEATURE_AMD_IBRS) &&
4280 !guest_cpuid_has(vcpu, X86_FEATURE_AMD_SSBD))
4283 /* The STIBP bit doesn't fault even if it's not advertised */
4284 if (data & ~(SPEC_CTRL_IBRS | SPEC_CTRL_STIBP | SPEC_CTRL_SSBD))
4287 svm->spec_ctrl = data;
4294 * When it's written (to non-zero) for the first time, pass
4298 * The handling of the MSR bitmap for L2 guests is done in
4299 * nested_svm_vmrun_msrpm.
4300 * We update the L1 MSR bit as well since it will end up
4301 * touching the MSR anyway now.
4303 set_msr_interception(svm->msrpm, MSR_IA32_SPEC_CTRL, 1, 1);
4305 case MSR_IA32_PRED_CMD:
4306 if (!msr->host_initiated &&
4307 !guest_cpuid_has(vcpu, X86_FEATURE_AMD_IBPB))
4310 if (data & ~PRED_CMD_IBPB)
4316 wrmsrl(MSR_IA32_PRED_CMD, PRED_CMD_IBPB);
4317 if (is_guest_mode(vcpu))
4319 set_msr_interception(svm->msrpm, MSR_IA32_PRED_CMD, 0, 1);
4321 case MSR_AMD64_VIRT_SPEC_CTRL:
4322 if (!msr->host_initiated &&
4323 !guest_cpuid_has(vcpu, X86_FEATURE_VIRT_SSBD))
4326 if (data & ~SPEC_CTRL_SSBD)
4329 svm->virt_spec_ctrl = data;
4332 svm->vmcb->save.star = data;
4334 #ifdef CONFIG_X86_64
4336 svm->vmcb->save.lstar = data;
4339 svm->vmcb->save.cstar = data;
4341 case MSR_KERNEL_GS_BASE:
4342 svm->vmcb->save.kernel_gs_base = data;
4344 case MSR_SYSCALL_MASK:
4345 svm->vmcb->save.sfmask = data;
4348 case MSR_IA32_SYSENTER_CS:
4349 svm->vmcb->save.sysenter_cs = data;
4351 case MSR_IA32_SYSENTER_EIP:
4352 svm->sysenter_eip = data;
4353 svm->vmcb->save.sysenter_eip = data;
4355 case MSR_IA32_SYSENTER_ESP:
4356 svm->sysenter_esp = data;
4357 svm->vmcb->save.sysenter_esp = data;
4360 if (!boot_cpu_has(X86_FEATURE_RDTSCP))
4364 * This is rare, so we update the MSR here instead of using
4365 * direct_access_msrs. Doing that would require a rdmsr in
4368 svm->tsc_aux = data;
4369 wrmsrl(MSR_TSC_AUX, svm->tsc_aux);
4371 case MSR_IA32_DEBUGCTLMSR:
4372 if (!boot_cpu_has(X86_FEATURE_LBRV)) {
4373 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTL 0x%llx, nop\n",
4377 if (data & DEBUGCTL_RESERVED_BITS)
4380 svm->vmcb->save.dbgctl = data;
4381 mark_dirty(svm->vmcb, VMCB_LBR);
4382 if (data & (1ULL<<0))
4383 svm_enable_lbrv(svm);
4385 svm_disable_lbrv(svm);
4387 case MSR_VM_HSAVE_PA:
4388 svm->nested.hsave_msr = data;
4391 return svm_set_vm_cr(vcpu, data);
4393 vcpu_unimpl(vcpu, "unimplemented wrmsr: 0x%x data 0x%llx\n", ecx, data);
4395 case MSR_F10H_DECFG: {
4396 struct kvm_msr_entry msr_entry;
4398 msr_entry.index = msr->index;
4399 if (svm_get_msr_feature(&msr_entry))
4402 /* Check the supported bits */
4403 if (data & ~msr_entry.data)
4406 /* Don't allow the guest to change a bit, #GP */
4407 if (!msr->host_initiated && (data ^ msr_entry.data))
4410 svm->msr_decfg = data;
4413 case MSR_IA32_APICBASE:
4414 if (kvm_vcpu_apicv_active(vcpu))
4415 avic_update_vapic_bar(to_svm(vcpu), data);
4418 return kvm_set_msr_common(vcpu, msr);
4423 static int wrmsr_interception(struct vcpu_svm *svm)
4425 struct msr_data msr;
4426 u32 ecx = kvm_rcx_read(&svm->vcpu);
4427 u64 data = kvm_read_edx_eax(&svm->vcpu);
4431 msr.host_initiated = false;
4433 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
4434 if (kvm_set_msr(&svm->vcpu, &msr)) {
4435 trace_kvm_msr_write_ex(ecx, data);
4436 kvm_inject_gp(&svm->vcpu, 0);
4439 trace_kvm_msr_write(ecx, data);
4440 return kvm_skip_emulated_instruction(&svm->vcpu);
4444 static int msr_interception(struct vcpu_svm *svm)
4446 if (svm->vmcb->control.exit_info_1)
4447 return wrmsr_interception(svm);
4449 return rdmsr_interception(svm);
4452 static int interrupt_window_interception(struct vcpu_svm *svm)
4454 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
4455 svm_clear_vintr(svm);
4456 svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
4457 mark_dirty(svm->vmcb, VMCB_INTR);
4458 ++svm->vcpu.stat.irq_window_exits;
4462 static int pause_interception(struct vcpu_svm *svm)
4464 struct kvm_vcpu *vcpu = &svm->vcpu;
4465 bool in_kernel = (svm_get_cpl(vcpu) == 0);
4467 if (pause_filter_thresh)
4468 grow_ple_window(vcpu);
4470 kvm_vcpu_on_spin(vcpu, in_kernel);
4474 static int nop_interception(struct vcpu_svm *svm)
4476 return kvm_skip_emulated_instruction(&(svm->vcpu));
4479 static int monitor_interception(struct vcpu_svm *svm)
4481 printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
4482 return nop_interception(svm);
4485 static int mwait_interception(struct vcpu_svm *svm)
4487 printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
4488 return nop_interception(svm);
4491 enum avic_ipi_failure_cause {
4492 AVIC_IPI_FAILURE_INVALID_INT_TYPE,
4493 AVIC_IPI_FAILURE_TARGET_NOT_RUNNING,
4494 AVIC_IPI_FAILURE_INVALID_TARGET,
4495 AVIC_IPI_FAILURE_INVALID_BACKING_PAGE,
4498 static int avic_incomplete_ipi_interception(struct vcpu_svm *svm)
4500 u32 icrh = svm->vmcb->control.exit_info_1 >> 32;
4501 u32 icrl = svm->vmcb->control.exit_info_1;
4502 u32 id = svm->vmcb->control.exit_info_2 >> 32;
4503 u32 index = svm->vmcb->control.exit_info_2 & 0xFF;
4504 struct kvm_lapic *apic = svm->vcpu.arch.apic;
4506 trace_kvm_avic_incomplete_ipi(svm->vcpu.vcpu_id, icrh, icrl, id, index);
4509 case AVIC_IPI_FAILURE_INVALID_INT_TYPE:
4511 * AVIC hardware handles the generation of
4512 * IPIs when the specified Message Type is Fixed
4513 * (also known as fixed delivery mode) and
4514 * the Trigger Mode is edge-triggered. The hardware
4515 * also supports self and broadcast delivery modes
4516 * specified via the Destination Shorthand(DSH)
4517 * field of the ICRL. Logical and physical APIC ID
4518 * formats are supported. All other IPI types cause
4519 * a #VMEXIT, which needs to emulated.
4521 kvm_lapic_reg_write(apic, APIC_ICR2, icrh);
4522 kvm_lapic_reg_write(apic, APIC_ICR, icrl);
4524 case AVIC_IPI_FAILURE_TARGET_NOT_RUNNING: {
4526 struct kvm_vcpu *vcpu;
4527 struct kvm *kvm = svm->vcpu.kvm;
4528 struct kvm_lapic *apic = svm->vcpu.arch.apic;
4531 * At this point, we expect that the AVIC HW has already
4532 * set the appropriate IRR bits on the valid target
4533 * vcpus. So, we just need to kick the appropriate vcpu.
4535 kvm_for_each_vcpu(i, vcpu, kvm) {
4536 bool m = kvm_apic_match_dest(vcpu, apic,
4537 icrl & KVM_APIC_SHORT_MASK,
4538 GET_APIC_DEST_FIELD(icrh),
4539 icrl & KVM_APIC_DEST_MASK);
4541 if (m && !avic_vcpu_is_running(vcpu))
4542 kvm_vcpu_wake_up(vcpu);
4546 case AVIC_IPI_FAILURE_INVALID_TARGET:
4547 WARN_ONCE(1, "Invalid IPI target: index=%u, vcpu=%d, icr=%#0x:%#0x\n",
4548 index, svm->vcpu.vcpu_id, icrh, icrl);
4550 case AVIC_IPI_FAILURE_INVALID_BACKING_PAGE:
4551 WARN_ONCE(1, "Invalid backing page\n");
4554 pr_err("Unknown IPI interception\n");
4560 static u32 *avic_get_logical_id_entry(struct kvm_vcpu *vcpu, u32 ldr, bool flat)
4562 struct kvm_svm *kvm_svm = to_kvm_svm(vcpu->kvm);
4564 u32 *logical_apic_id_table;
4565 int dlid = GET_APIC_LOGICAL_ID(ldr);
4570 if (flat) { /* flat */
4571 index = ffs(dlid) - 1;
4574 } else { /* cluster */
4575 int cluster = (dlid & 0xf0) >> 4;
4576 int apic = ffs(dlid & 0x0f) - 1;
4578 if ((apic < 0) || (apic > 7) ||
4581 index = (cluster << 2) + apic;
4584 logical_apic_id_table = (u32 *) page_address(kvm_svm->avic_logical_id_table_page);
4586 return &logical_apic_id_table[index];
4589 static int avic_ldr_write(struct kvm_vcpu *vcpu, u8 g_physical_id, u32 ldr)
4592 u32 *entry, new_entry;
4594 flat = kvm_lapic_get_reg(vcpu->arch.apic, APIC_DFR) == APIC_DFR_FLAT;
4595 entry = avic_get_logical_id_entry(vcpu, ldr, flat);
4599 new_entry = READ_ONCE(*entry);
4600 new_entry &= ~AVIC_LOGICAL_ID_ENTRY_GUEST_PHYSICAL_ID_MASK;
4601 new_entry |= (g_physical_id & AVIC_LOGICAL_ID_ENTRY_GUEST_PHYSICAL_ID_MASK);
4602 new_entry |= AVIC_LOGICAL_ID_ENTRY_VALID_MASK;
4603 WRITE_ONCE(*entry, new_entry);
4608 static void avic_invalidate_logical_id_entry(struct kvm_vcpu *vcpu)
4610 struct vcpu_svm *svm = to_svm(vcpu);
4611 bool flat = svm->dfr_reg == APIC_DFR_FLAT;
4612 u32 *entry = avic_get_logical_id_entry(vcpu, svm->ldr_reg, flat);
4615 clear_bit(AVIC_LOGICAL_ID_ENTRY_VALID_BIT, (unsigned long *)entry);
4618 static int avic_handle_ldr_update(struct kvm_vcpu *vcpu)
4621 struct vcpu_svm *svm = to_svm(vcpu);
4622 u32 ldr = kvm_lapic_get_reg(vcpu->arch.apic, APIC_LDR);
4624 if (ldr == svm->ldr_reg)
4627 avic_invalidate_logical_id_entry(vcpu);
4630 ret = avic_ldr_write(vcpu, vcpu->vcpu_id, ldr);
4638 static int avic_handle_apic_id_update(struct kvm_vcpu *vcpu)
4641 struct vcpu_svm *svm = to_svm(vcpu);
4642 u32 apic_id_reg = kvm_lapic_get_reg(vcpu->arch.apic, APIC_ID);
4643 u32 id = (apic_id_reg >> 24) & 0xff;
4645 if (vcpu->vcpu_id == id)
4648 old = avic_get_physical_id_entry(vcpu, vcpu->vcpu_id);
4649 new = avic_get_physical_id_entry(vcpu, id);
4653 /* We need to move physical_id_entry to new offset */
4656 to_svm(vcpu)->avic_physical_id_cache = new;
4659 * Also update the guest physical APIC ID in the logical
4660 * APIC ID table entry if already setup the LDR.
4663 avic_handle_ldr_update(vcpu);
4668 static void avic_handle_dfr_update(struct kvm_vcpu *vcpu)
4670 struct vcpu_svm *svm = to_svm(vcpu);
4671 u32 dfr = kvm_lapic_get_reg(vcpu->arch.apic, APIC_DFR);
4673 if (svm->dfr_reg == dfr)
4676 avic_invalidate_logical_id_entry(vcpu);
4680 static int avic_unaccel_trap_write(struct vcpu_svm *svm)
4682 struct kvm_lapic *apic = svm->vcpu.arch.apic;
4683 u32 offset = svm->vmcb->control.exit_info_1 &
4684 AVIC_UNACCEL_ACCESS_OFFSET_MASK;
4688 if (avic_handle_apic_id_update(&svm->vcpu))
4692 if (avic_handle_ldr_update(&svm->vcpu))
4696 avic_handle_dfr_update(&svm->vcpu);
4702 kvm_lapic_reg_write(apic, offset, kvm_lapic_get_reg(apic, offset));
4707 static bool is_avic_unaccelerated_access_trap(u32 offset)
4736 static int avic_unaccelerated_access_interception(struct vcpu_svm *svm)
4739 u32 offset = svm->vmcb->control.exit_info_1 &
4740 AVIC_UNACCEL_ACCESS_OFFSET_MASK;
4741 u32 vector = svm->vmcb->control.exit_info_2 &
4742 AVIC_UNACCEL_ACCESS_VECTOR_MASK;
4743 bool write = (svm->vmcb->control.exit_info_1 >> 32) &
4744 AVIC_UNACCEL_ACCESS_WRITE_MASK;
4745 bool trap = is_avic_unaccelerated_access_trap(offset);
4747 trace_kvm_avic_unaccelerated_access(svm->vcpu.vcpu_id, offset,
4748 trap, write, vector);
4751 WARN_ONCE(!write, "svm: Handling trap read.\n");
4752 ret = avic_unaccel_trap_write(svm);
4754 /* Handling Fault */
4755 ret = (kvm_emulate_instruction(&svm->vcpu, 0) == EMULATE_DONE);
4761 static int (*const svm_exit_handlers[])(struct vcpu_svm *svm) = {
4762 [SVM_EXIT_READ_CR0] = cr_interception,
4763 [SVM_EXIT_READ_CR3] = cr_interception,
4764 [SVM_EXIT_READ_CR4] = cr_interception,
4765 [SVM_EXIT_READ_CR8] = cr_interception,
4766 [SVM_EXIT_CR0_SEL_WRITE] = cr_interception,
4767 [SVM_EXIT_WRITE_CR0] = cr_interception,
4768 [SVM_EXIT_WRITE_CR3] = cr_interception,
4769 [SVM_EXIT_WRITE_CR4] = cr_interception,
4770 [SVM_EXIT_WRITE_CR8] = cr8_write_interception,
4771 [SVM_EXIT_READ_DR0] = dr_interception,
4772 [SVM_EXIT_READ_DR1] = dr_interception,
4773 [SVM_EXIT_READ_DR2] = dr_interception,
4774 [SVM_EXIT_READ_DR3] = dr_interception,
4775 [SVM_EXIT_READ_DR4] = dr_interception,
4776 [SVM_EXIT_READ_DR5] = dr_interception,
4777 [SVM_EXIT_READ_DR6] = dr_interception,
4778 [SVM_EXIT_READ_DR7] = dr_interception,
4779 [SVM_EXIT_WRITE_DR0] = dr_interception,
4780 [SVM_EXIT_WRITE_DR1] = dr_interception,
4781 [SVM_EXIT_WRITE_DR2] = dr_interception,
4782 [SVM_EXIT_WRITE_DR3] = dr_interception,
4783 [SVM_EXIT_WRITE_DR4] = dr_interception,
4784 [SVM_EXIT_WRITE_DR5] = dr_interception,
4785 [SVM_EXIT_WRITE_DR6] = dr_interception,
4786 [SVM_EXIT_WRITE_DR7] = dr_interception,
4787 [SVM_EXIT_EXCP_BASE + DB_VECTOR] = db_interception,
4788 [SVM_EXIT_EXCP_BASE + BP_VECTOR] = bp_interception,
4789 [SVM_EXIT_EXCP_BASE + UD_VECTOR] = ud_interception,
4790 [SVM_EXIT_EXCP_BASE + PF_VECTOR] = pf_interception,
4791 [SVM_EXIT_EXCP_BASE + MC_VECTOR] = mc_interception,
4792 [SVM_EXIT_EXCP_BASE + AC_VECTOR] = ac_interception,
4793 [SVM_EXIT_EXCP_BASE + GP_VECTOR] = gp_interception,
4794 [SVM_EXIT_INTR] = intr_interception,
4795 [SVM_EXIT_NMI] = nmi_interception,
4796 [SVM_EXIT_SMI] = nop_on_interception,
4797 [SVM_EXIT_INIT] = nop_on_interception,
4798 [SVM_EXIT_VINTR] = interrupt_window_interception,
4799 [SVM_EXIT_RDPMC] = rdpmc_interception,
4800 [SVM_EXIT_CPUID] = cpuid_interception,
4801 [SVM_EXIT_IRET] = iret_interception,
4802 [SVM_EXIT_INVD] = emulate_on_interception,
4803 [SVM_EXIT_PAUSE] = pause_interception,
4804 [SVM_EXIT_HLT] = halt_interception,
4805 [SVM_EXIT_INVLPG] = invlpg_interception,
4806 [SVM_EXIT_INVLPGA] = invlpga_interception,
4807 [SVM_EXIT_IOIO] = io_interception,
4808 [SVM_EXIT_MSR] = msr_interception,
4809 [SVM_EXIT_TASK_SWITCH] = task_switch_interception,
4810 [SVM_EXIT_SHUTDOWN] = shutdown_interception,
4811 [SVM_EXIT_VMRUN] = vmrun_interception,
4812 [SVM_EXIT_VMMCALL] = vmmcall_interception,
4813 [SVM_EXIT_VMLOAD] = vmload_interception,
4814 [SVM_EXIT_VMSAVE] = vmsave_interception,
4815 [SVM_EXIT_STGI] = stgi_interception,
4816 [SVM_EXIT_CLGI] = clgi_interception,
4817 [SVM_EXIT_SKINIT] = skinit_interception,
4818 [SVM_EXIT_WBINVD] = wbinvd_interception,
4819 [SVM_EXIT_MONITOR] = monitor_interception,
4820 [SVM_EXIT_MWAIT] = mwait_interception,
4821 [SVM_EXIT_XSETBV] = xsetbv_interception,
4822 [SVM_EXIT_NPF] = npf_interception,
4823 [SVM_EXIT_RSM] = rsm_interception,
4824 [SVM_EXIT_AVIC_INCOMPLETE_IPI] = avic_incomplete_ipi_interception,
4825 [SVM_EXIT_AVIC_UNACCELERATED_ACCESS] = avic_unaccelerated_access_interception,
4828 static void dump_vmcb(struct kvm_vcpu *vcpu)
4830 struct vcpu_svm *svm = to_svm(vcpu);
4831 struct vmcb_control_area *control = &svm->vmcb->control;
4832 struct vmcb_save_area *save = &svm->vmcb->save;
4834 if (!dump_invalid_vmcb) {
4835 pr_warn_ratelimited("set kvm_amd.dump_invalid_vmcb=1 to dump internal KVM state.\n");
4839 pr_err("VMCB Control Area:\n");
4840 pr_err("%-20s%04x\n", "cr_read:", control->intercept_cr & 0xffff);
4841 pr_err("%-20s%04x\n", "cr_write:", control->intercept_cr >> 16);
4842 pr_err("%-20s%04x\n", "dr_read:", control->intercept_dr & 0xffff);
4843 pr_err("%-20s%04x\n", "dr_write:", control->intercept_dr >> 16);
4844 pr_err("%-20s%08x\n", "exceptions:", control->intercept_exceptions);
4845 pr_err("%-20s%016llx\n", "intercepts:", control->intercept);
4846 pr_err("%-20s%d\n", "pause filter count:", control->pause_filter_count);
4847 pr_err("%-20s%d\n", "pause filter threshold:",
4848 control->pause_filter_thresh);
4849 pr_err("%-20s%016llx\n", "iopm_base_pa:", control->iopm_base_pa);
4850 pr_err("%-20s%016llx\n", "msrpm_base_pa:", control->msrpm_base_pa);
4851 pr_err("%-20s%016llx\n", "tsc_offset:", control->tsc_offset);
4852 pr_err("%-20s%d\n", "asid:", control->asid);
4853 pr_err("%-20s%d\n", "tlb_ctl:", control->tlb_ctl);
4854 pr_err("%-20s%08x\n", "int_ctl:", control->int_ctl);
4855 pr_err("%-20s%08x\n", "int_vector:", control->int_vector);
4856 pr_err("%-20s%08x\n", "int_state:", control->int_state);
4857 pr_err("%-20s%08x\n", "exit_code:", control->exit_code);
4858 pr_err("%-20s%016llx\n", "exit_info1:", control->exit_info_1);
4859 pr_err("%-20s%016llx\n", "exit_info2:", control->exit_info_2);
4860 pr_err("%-20s%08x\n", "exit_int_info:", control->exit_int_info);
4861 pr_err("%-20s%08x\n", "exit_int_info_err:", control->exit_int_info_err);
4862 pr_err("%-20s%lld\n", "nested_ctl:", control->nested_ctl);
4863 pr_err("%-20s%016llx\n", "nested_cr3:", control->nested_cr3);
4864 pr_err("%-20s%016llx\n", "avic_vapic_bar:", control->avic_vapic_bar);
4865 pr_err("%-20s%08x\n", "event_inj:", control->event_inj);
4866 pr_err("%-20s%08x\n", "event_inj_err:", control->event_inj_err);
4867 pr_err("%-20s%lld\n", "virt_ext:", control->virt_ext);
4868 pr_err("%-20s%016llx\n", "next_rip:", control->next_rip);
4869 pr_err("%-20s%016llx\n", "avic_backing_page:", control->avic_backing_page);
4870 pr_err("%-20s%016llx\n", "avic_logical_id:", control->avic_logical_id);
4871 pr_err("%-20s%016llx\n", "avic_physical_id:", control->avic_physical_id);
4872 pr_err("VMCB State Save Area:\n");
4873 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4875 save->es.selector, save->es.attrib,
4876 save->es.limit, save->es.base);
4877 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4879 save->cs.selector, save->cs.attrib,
4880 save->cs.limit, save->cs.base);
4881 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4883 save->ss.selector, save->ss.attrib,
4884 save->ss.limit, save->ss.base);
4885 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4887 save->ds.selector, save->ds.attrib,
4888 save->ds.limit, save->ds.base);
4889 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4891 save->fs.selector, save->fs.attrib,
4892 save->fs.limit, save->fs.base);
4893 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4895 save->gs.selector, save->gs.attrib,
4896 save->gs.limit, save->gs.base);
4897 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4899 save->gdtr.selector, save->gdtr.attrib,
4900 save->gdtr.limit, save->gdtr.base);
4901 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4903 save->ldtr.selector, save->ldtr.attrib,
4904 save->ldtr.limit, save->ldtr.base);
4905 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4907 save->idtr.selector, save->idtr.attrib,
4908 save->idtr.limit, save->idtr.base);
4909 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4911 save->tr.selector, save->tr.attrib,
4912 save->tr.limit, save->tr.base);
4913 pr_err("cpl: %d efer: %016llx\n",
4914 save->cpl, save->efer);
4915 pr_err("%-15s %016llx %-13s %016llx\n",
4916 "cr0:", save->cr0, "cr2:", save->cr2);
4917 pr_err("%-15s %016llx %-13s %016llx\n",
4918 "cr3:", save->cr3, "cr4:", save->cr4);
4919 pr_err("%-15s %016llx %-13s %016llx\n",
4920 "dr6:", save->dr6, "dr7:", save->dr7);
4921 pr_err("%-15s %016llx %-13s %016llx\n",
4922 "rip:", save->rip, "rflags:", save->rflags);
4923 pr_err("%-15s %016llx %-13s %016llx\n",
4924 "rsp:", save->rsp, "rax:", save->rax);
4925 pr_err("%-15s %016llx %-13s %016llx\n",
4926 "star:", save->star, "lstar:", save->lstar);
4927 pr_err("%-15s %016llx %-13s %016llx\n",
4928 "cstar:", save->cstar, "sfmask:", save->sfmask);
4929 pr_err("%-15s %016llx %-13s %016llx\n",
4930 "kernel_gs_base:", save->kernel_gs_base,
4931 "sysenter_cs:", save->sysenter_cs);
4932 pr_err("%-15s %016llx %-13s %016llx\n",
4933 "sysenter_esp:", save->sysenter_esp,
4934 "sysenter_eip:", save->sysenter_eip);
4935 pr_err("%-15s %016llx %-13s %016llx\n",
4936 "gpat:", save->g_pat, "dbgctl:", save->dbgctl);
4937 pr_err("%-15s %016llx %-13s %016llx\n",
4938 "br_from:", save->br_from, "br_to:", save->br_to);
4939 pr_err("%-15s %016llx %-13s %016llx\n",
4940 "excp_from:", save->last_excp_from,
4941 "excp_to:", save->last_excp_to);
4944 static void svm_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
4946 struct vmcb_control_area *control = &to_svm(vcpu)->vmcb->control;
4948 *info1 = control->exit_info_1;
4949 *info2 = control->exit_info_2;
4952 static int handle_exit(struct kvm_vcpu *vcpu)
4954 struct vcpu_svm *svm = to_svm(vcpu);
4955 struct kvm_run *kvm_run = vcpu->run;
4956 u32 exit_code = svm->vmcb->control.exit_code;
4958 trace_kvm_exit(exit_code, vcpu, KVM_ISA_SVM);
4960 if (!is_cr_intercept(svm, INTERCEPT_CR0_WRITE))
4961 vcpu->arch.cr0 = svm->vmcb->save.cr0;
4963 vcpu->arch.cr3 = svm->vmcb->save.cr3;
4965 if (unlikely(svm->nested.exit_required)) {
4966 nested_svm_vmexit(svm);
4967 svm->nested.exit_required = false;
4972 if (is_guest_mode(vcpu)) {
4975 trace_kvm_nested_vmexit(svm->vmcb->save.rip, exit_code,
4976 svm->vmcb->control.exit_info_1,
4977 svm->vmcb->control.exit_info_2,
4978 svm->vmcb->control.exit_int_info,
4979 svm->vmcb->control.exit_int_info_err,
4982 vmexit = nested_svm_exit_special(svm);
4984 if (vmexit == NESTED_EXIT_CONTINUE)
4985 vmexit = nested_svm_exit_handled(svm);
4987 if (vmexit == NESTED_EXIT_DONE)
4991 svm_complete_interrupts(svm);
4993 if (svm->vmcb->control.exit_code == SVM_EXIT_ERR) {
4994 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4995 kvm_run->fail_entry.hardware_entry_failure_reason
4996 = svm->vmcb->control.exit_code;
5001 if (is_external_interrupt(svm->vmcb->control.exit_int_info) &&
5002 exit_code != SVM_EXIT_EXCP_BASE + PF_VECTOR &&
5003 exit_code != SVM_EXIT_NPF && exit_code != SVM_EXIT_TASK_SWITCH &&
5004 exit_code != SVM_EXIT_INTR && exit_code != SVM_EXIT_NMI)
5005 printk(KERN_ERR "%s: unexpected exit_int_info 0x%x "
5007 __func__, svm->vmcb->control.exit_int_info,
5010 if (exit_code >= ARRAY_SIZE(svm_exit_handlers)
5011 || !svm_exit_handlers[exit_code]) {
5012 WARN_ONCE(1, "svm: unexpected exit reason 0x%x\n", exit_code);
5013 kvm_queue_exception(vcpu, UD_VECTOR);
5017 return svm_exit_handlers[exit_code](svm);
5020 static void reload_tss(struct kvm_vcpu *vcpu)
5022 int cpu = raw_smp_processor_id();
5024 struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
5025 sd->tss_desc->type = 9; /* available 32/64-bit TSS */
5029 static void pre_sev_run(struct vcpu_svm *svm, int cpu)
5031 struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
5032 int asid = sev_get_asid(svm->vcpu.kvm);
5034 /* Assign the asid allocated with this SEV guest */
5035 svm->vmcb->control.asid = asid;
5040 * 1) when different VMCB for the same ASID is to be run on the same host CPU.
5041 * 2) or this VMCB was executed on different host CPU in previous VMRUNs.
5043 if (sd->sev_vmcbs[asid] == svm->vmcb &&
5044 svm->last_cpu == cpu)
5047 svm->last_cpu = cpu;
5048 sd->sev_vmcbs[asid] = svm->vmcb;
5049 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ASID;
5050 mark_dirty(svm->vmcb, VMCB_ASID);
5053 static void pre_svm_run(struct vcpu_svm *svm)
5055 int cpu = raw_smp_processor_id();
5057 struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
5059 if (sev_guest(svm->vcpu.kvm))
5060 return pre_sev_run(svm, cpu);
5062 /* FIXME: handle wraparound of asid_generation */
5063 if (svm->asid_generation != sd->asid_generation)
5067 static void svm_inject_nmi(struct kvm_vcpu *vcpu)
5069 struct vcpu_svm *svm = to_svm(vcpu);
5071 svm->vmcb->control.event_inj = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI;
5072 vcpu->arch.hflags |= HF_NMI_MASK;
5073 set_intercept(svm, INTERCEPT_IRET);
5074 ++vcpu->stat.nmi_injections;
5077 static inline void svm_inject_irq(struct vcpu_svm *svm, int irq)
5079 struct vmcb_control_area *control;
5081 /* The following fields are ignored when AVIC is enabled */
5082 control = &svm->vmcb->control;
5083 control->int_vector = irq;
5084 control->int_ctl &= ~V_INTR_PRIO_MASK;
5085 control->int_ctl |= V_IRQ_MASK |
5086 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT);
5087 mark_dirty(svm->vmcb, VMCB_INTR);
5090 static void svm_set_irq(struct kvm_vcpu *vcpu)
5092 struct vcpu_svm *svm = to_svm(vcpu);
5094 BUG_ON(!(gif_set(svm)));
5096 trace_kvm_inj_virq(vcpu->arch.interrupt.nr);
5097 ++vcpu->stat.irq_injections;
5099 svm->vmcb->control.event_inj = vcpu->arch.interrupt.nr |
5100 SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR;
5103 static inline bool svm_nested_virtualize_tpr(struct kvm_vcpu *vcpu)
5105 return is_guest_mode(vcpu) && (vcpu->arch.hflags & HF_VINTR_MASK);
5108 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
5110 struct vcpu_svm *svm = to_svm(vcpu);
5112 if (svm_nested_virtualize_tpr(vcpu) ||
5113 kvm_vcpu_apicv_active(vcpu))
5116 clr_cr_intercept(svm, INTERCEPT_CR8_WRITE);
5122 set_cr_intercept(svm, INTERCEPT_CR8_WRITE);
5125 static void svm_set_virtual_apic_mode(struct kvm_vcpu *vcpu)
5130 static bool svm_get_enable_apicv(struct kvm_vcpu *vcpu)
5132 return avic && irqchip_split(vcpu->kvm);
5135 static void svm_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
5139 static void svm_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr)
5143 /* Note: Currently only used by Hyper-V. */
5144 static void svm_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu)
5146 struct vcpu_svm *svm = to_svm(vcpu);
5147 struct vmcb *vmcb = svm->vmcb;
5149 if (kvm_vcpu_apicv_active(vcpu))
5150 vmcb->control.int_ctl |= AVIC_ENABLE_MASK;
5152 vmcb->control.int_ctl &= ~AVIC_ENABLE_MASK;
5153 mark_dirty(vmcb, VMCB_AVIC);
5156 static void svm_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
5161 static void svm_deliver_avic_intr(struct kvm_vcpu *vcpu, int vec)
5163 kvm_lapic_set_irr(vec, vcpu->arch.apic);
5164 smp_mb__after_atomic();
5166 if (avic_vcpu_is_running(vcpu))
5167 wrmsrl(SVM_AVIC_DOORBELL,
5168 kvm_cpu_get_apicid(vcpu->cpu));
5170 kvm_vcpu_wake_up(vcpu);
5173 static void svm_ir_list_del(struct vcpu_svm *svm, struct amd_iommu_pi_data *pi)
5175 unsigned long flags;
5176 struct amd_svm_iommu_ir *cur;
5178 spin_lock_irqsave(&svm->ir_list_lock, flags);
5179 list_for_each_entry(cur, &svm->ir_list, node) {
5180 if (cur->data != pi->ir_data)
5182 list_del(&cur->node);
5186 spin_unlock_irqrestore(&svm->ir_list_lock, flags);
5189 static int svm_ir_list_add(struct vcpu_svm *svm, struct amd_iommu_pi_data *pi)
5192 unsigned long flags;
5193 struct amd_svm_iommu_ir *ir;
5196 * In some cases, the existing irte is updaed and re-set,
5197 * so we need to check here if it's already been * added
5200 if (pi->ir_data && (pi->prev_ga_tag != 0)) {
5201 struct kvm *kvm = svm->vcpu.kvm;
5202 u32 vcpu_id = AVIC_GATAG_TO_VCPUID(pi->prev_ga_tag);
5203 struct kvm_vcpu *prev_vcpu = kvm_get_vcpu_by_id(kvm, vcpu_id);
5204 struct vcpu_svm *prev_svm;
5211 prev_svm = to_svm(prev_vcpu);
5212 svm_ir_list_del(prev_svm, pi);
5216 * Allocating new amd_iommu_pi_data, which will get
5217 * add to the per-vcpu ir_list.
5219 ir = kzalloc(sizeof(struct amd_svm_iommu_ir), GFP_KERNEL_ACCOUNT);
5224 ir->data = pi->ir_data;
5226 spin_lock_irqsave(&svm->ir_list_lock, flags);
5227 list_add(&ir->node, &svm->ir_list);
5228 spin_unlock_irqrestore(&svm->ir_list_lock, flags);
5235 * The HW cannot support posting multicast/broadcast
5236 * interrupts to a vCPU. So, we still use legacy interrupt
5237 * remapping for these kind of interrupts.
5239 * For lowest-priority interrupts, we only support
5240 * those with single CPU as the destination, e.g. user
5241 * configures the interrupts via /proc/irq or uses
5242 * irqbalance to make the interrupts single-CPU.
5245 get_pi_vcpu_info(struct kvm *kvm, struct kvm_kernel_irq_routing_entry *e,
5246 struct vcpu_data *vcpu_info, struct vcpu_svm **svm)
5248 struct kvm_lapic_irq irq;
5249 struct kvm_vcpu *vcpu = NULL;
5251 kvm_set_msi_irq(kvm, e, &irq);
5253 if (!kvm_intr_is_single_vcpu(kvm, &irq, &vcpu)) {
5254 pr_debug("SVM: %s: use legacy intr remap mode for irq %u\n",
5255 __func__, irq.vector);
5259 pr_debug("SVM: %s: use GA mode for irq %u\n", __func__,
5261 *svm = to_svm(vcpu);
5262 vcpu_info->pi_desc_addr = __sme_set(page_to_phys((*svm)->avic_backing_page));
5263 vcpu_info->vector = irq.vector;
5269 * svm_update_pi_irte - set IRTE for Posted-Interrupts
5272 * @host_irq: host irq of the interrupt
5273 * @guest_irq: gsi of the interrupt
5274 * @set: set or unset PI
5275 * returns 0 on success, < 0 on failure
5277 static int svm_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
5278 uint32_t guest_irq, bool set)
5280 struct kvm_kernel_irq_routing_entry *e;
5281 struct kvm_irq_routing_table *irq_rt;
5282 int idx, ret = -EINVAL;
5284 if (!kvm_arch_has_assigned_device(kvm) ||
5285 !irq_remapping_cap(IRQ_POSTING_CAP))
5288 pr_debug("SVM: %s: host_irq=%#x, guest_irq=%#x, set=%#x\n",
5289 __func__, host_irq, guest_irq, set);
5291 idx = srcu_read_lock(&kvm->irq_srcu);
5292 irq_rt = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu);
5293 WARN_ON(guest_irq >= irq_rt->nr_rt_entries);
5295 hlist_for_each_entry(e, &irq_rt->map[guest_irq], link) {
5296 struct vcpu_data vcpu_info;
5297 struct vcpu_svm *svm = NULL;
5299 if (e->type != KVM_IRQ_ROUTING_MSI)
5303 * Here, we setup with legacy mode in the following cases:
5304 * 1. When cannot target interrupt to a specific vcpu.
5305 * 2. Unsetting posted interrupt.
5306 * 3. APIC virtialization is disabled for the vcpu.
5308 if (!get_pi_vcpu_info(kvm, e, &vcpu_info, &svm) && set &&
5309 kvm_vcpu_apicv_active(&svm->vcpu)) {
5310 struct amd_iommu_pi_data pi;
5312 /* Try to enable guest_mode in IRTE */
5313 pi.base = __sme_set(page_to_phys(svm->avic_backing_page) &
5315 pi.ga_tag = AVIC_GATAG(to_kvm_svm(kvm)->avic_vm_id,
5317 pi.is_guest_mode = true;
5318 pi.vcpu_data = &vcpu_info;
5319 ret = irq_set_vcpu_affinity(host_irq, &pi);
5322 * Here, we successfully setting up vcpu affinity in
5323 * IOMMU guest mode. Now, we need to store the posted
5324 * interrupt information in a per-vcpu ir_list so that
5325 * we can reference to them directly when we update vcpu
5326 * scheduling information in IOMMU irte.
5328 if (!ret && pi.is_guest_mode)
5329 svm_ir_list_add(svm, &pi);
5331 /* Use legacy mode in IRTE */
5332 struct amd_iommu_pi_data pi;
5335 * Here, pi is used to:
5336 * - Tell IOMMU to use legacy mode for this interrupt.
5337 * - Retrieve ga_tag of prior interrupt remapping data.
5339 pi.is_guest_mode = false;
5340 ret = irq_set_vcpu_affinity(host_irq, &pi);
5343 * Check if the posted interrupt was previously
5344 * setup with the guest_mode by checking if the ga_tag
5345 * was cached. If so, we need to clean up the per-vcpu
5348 if (!ret && pi.prev_ga_tag) {
5349 int id = AVIC_GATAG_TO_VCPUID(pi.prev_ga_tag);
5350 struct kvm_vcpu *vcpu;
5352 vcpu = kvm_get_vcpu_by_id(kvm, id);
5354 svm_ir_list_del(to_svm(vcpu), &pi);
5359 trace_kvm_pi_irte_update(host_irq, svm->vcpu.vcpu_id,
5360 e->gsi, vcpu_info.vector,
5361 vcpu_info.pi_desc_addr, set);
5365 pr_err("%s: failed to update PI IRTE\n", __func__);
5372 srcu_read_unlock(&kvm->irq_srcu, idx);
5376 static int svm_nmi_allowed(struct kvm_vcpu *vcpu)
5378 struct vcpu_svm *svm = to_svm(vcpu);
5379 struct vmcb *vmcb = svm->vmcb;
5381 ret = !(vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK) &&
5382 !(svm->vcpu.arch.hflags & HF_NMI_MASK);
5383 ret = ret && gif_set(svm) && nested_svm_nmi(svm);
5388 static bool svm_get_nmi_mask(struct kvm_vcpu *vcpu)
5390 struct vcpu_svm *svm = to_svm(vcpu);
5392 return !!(svm->vcpu.arch.hflags & HF_NMI_MASK);
5395 static void svm_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
5397 struct vcpu_svm *svm = to_svm(vcpu);
5400 svm->vcpu.arch.hflags |= HF_NMI_MASK;
5401 set_intercept(svm, INTERCEPT_IRET);
5403 svm->vcpu.arch.hflags &= ~HF_NMI_MASK;
5404 clr_intercept(svm, INTERCEPT_IRET);
5408 static int svm_interrupt_allowed(struct kvm_vcpu *vcpu)
5410 struct vcpu_svm *svm = to_svm(vcpu);
5411 struct vmcb *vmcb = svm->vmcb;
5414 if (!gif_set(svm) ||
5415 (vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK))
5418 ret = !!(kvm_get_rflags(vcpu) & X86_EFLAGS_IF);
5420 if (is_guest_mode(vcpu))
5421 return ret && !(svm->vcpu.arch.hflags & HF_VINTR_MASK);
5426 static void enable_irq_window(struct kvm_vcpu *vcpu)
5428 struct vcpu_svm *svm = to_svm(vcpu);
5430 if (kvm_vcpu_apicv_active(vcpu))
5434 * In case GIF=0 we can't rely on the CPU to tell us when GIF becomes
5435 * 1, because that's a separate STGI/VMRUN intercept. The next time we
5436 * get that intercept, this function will be called again though and
5437 * we'll get the vintr intercept. However, if the vGIF feature is
5438 * enabled, the STGI interception will not occur. Enable the irq
5439 * window under the assumption that the hardware will set the GIF.
5441 if ((vgif_enabled(svm) || gif_set(svm)) && nested_svm_intr(svm)) {
5443 svm_inject_irq(svm, 0x0);
5447 static void enable_nmi_window(struct kvm_vcpu *vcpu)
5449 struct vcpu_svm *svm = to_svm(vcpu);
5451 if ((svm->vcpu.arch.hflags & (HF_NMI_MASK | HF_IRET_MASK))
5453 return; /* IRET will cause a vm exit */
5455 if (!gif_set(svm)) {
5456 if (vgif_enabled(svm))
5457 set_intercept(svm, INTERCEPT_STGI);
5458 return; /* STGI will cause a vm exit */
5461 if (svm->nested.exit_required)
5462 return; /* we're not going to run the guest yet */
5465 * Something prevents NMI from been injected. Single step over possible
5466 * problem (IRET or exception injection or interrupt shadow)
5468 svm->nmi_singlestep_guest_rflags = svm_get_rflags(vcpu);
5469 svm->nmi_singlestep = true;
5470 svm->vmcb->save.rflags |= (X86_EFLAGS_TF | X86_EFLAGS_RF);
5473 static int svm_set_tss_addr(struct kvm *kvm, unsigned int addr)
5478 static int svm_set_identity_map_addr(struct kvm *kvm, u64 ident_addr)
5483 static void svm_flush_tlb(struct kvm_vcpu *vcpu, bool invalidate_gpa)
5485 struct vcpu_svm *svm = to_svm(vcpu);
5487 if (static_cpu_has(X86_FEATURE_FLUSHBYASID))
5488 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ASID;
5490 svm->asid_generation--;
5493 static void svm_flush_tlb_gva(struct kvm_vcpu *vcpu, gva_t gva)
5495 struct vcpu_svm *svm = to_svm(vcpu);
5497 invlpga(gva, svm->vmcb->control.asid);
5500 static void svm_prepare_guest_switch(struct kvm_vcpu *vcpu)
5504 static inline void sync_cr8_to_lapic(struct kvm_vcpu *vcpu)
5506 struct vcpu_svm *svm = to_svm(vcpu);
5508 if (svm_nested_virtualize_tpr(vcpu))
5511 if (!is_cr_intercept(svm, INTERCEPT_CR8_WRITE)) {
5512 int cr8 = svm->vmcb->control.int_ctl & V_TPR_MASK;
5513 kvm_set_cr8(vcpu, cr8);
5517 static inline void sync_lapic_to_cr8(struct kvm_vcpu *vcpu)
5519 struct vcpu_svm *svm = to_svm(vcpu);
5522 if (svm_nested_virtualize_tpr(vcpu) ||
5523 kvm_vcpu_apicv_active(vcpu))
5526 cr8 = kvm_get_cr8(vcpu);
5527 svm->vmcb->control.int_ctl &= ~V_TPR_MASK;
5528 svm->vmcb->control.int_ctl |= cr8 & V_TPR_MASK;
5531 static void svm_complete_interrupts(struct vcpu_svm *svm)
5535 u32 exitintinfo = svm->vmcb->control.exit_int_info;
5536 unsigned int3_injected = svm->int3_injected;
5538 svm->int3_injected = 0;
5541 * If we've made progress since setting HF_IRET_MASK, we've
5542 * executed an IRET and can allow NMI injection.
5544 if ((svm->vcpu.arch.hflags & HF_IRET_MASK)
5545 && kvm_rip_read(&svm->vcpu) != svm->nmi_iret_rip) {
5546 svm->vcpu.arch.hflags &= ~(HF_NMI_MASK | HF_IRET_MASK);
5547 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
5550 svm->vcpu.arch.nmi_injected = false;
5551 kvm_clear_exception_queue(&svm->vcpu);
5552 kvm_clear_interrupt_queue(&svm->vcpu);
5554 if (!(exitintinfo & SVM_EXITINTINFO_VALID))
5557 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
5559 vector = exitintinfo & SVM_EXITINTINFO_VEC_MASK;
5560 type = exitintinfo & SVM_EXITINTINFO_TYPE_MASK;
5563 case SVM_EXITINTINFO_TYPE_NMI:
5564 svm->vcpu.arch.nmi_injected = true;
5566 case SVM_EXITINTINFO_TYPE_EXEPT:
5568 * In case of software exceptions, do not reinject the vector,
5569 * but re-execute the instruction instead. Rewind RIP first
5570 * if we emulated INT3 before.
5572 if (kvm_exception_is_soft(vector)) {
5573 if (vector == BP_VECTOR && int3_injected &&
5574 kvm_is_linear_rip(&svm->vcpu, svm->int3_rip))
5575 kvm_rip_write(&svm->vcpu,
5576 kvm_rip_read(&svm->vcpu) -
5580 if (exitintinfo & SVM_EXITINTINFO_VALID_ERR) {
5581 u32 err = svm->vmcb->control.exit_int_info_err;
5582 kvm_requeue_exception_e(&svm->vcpu, vector, err);
5585 kvm_requeue_exception(&svm->vcpu, vector);
5587 case SVM_EXITINTINFO_TYPE_INTR:
5588 kvm_queue_interrupt(&svm->vcpu, vector, false);
5595 static void svm_cancel_injection(struct kvm_vcpu *vcpu)
5597 struct vcpu_svm *svm = to_svm(vcpu);
5598 struct vmcb_control_area *control = &svm->vmcb->control;
5600 control->exit_int_info = control->event_inj;
5601 control->exit_int_info_err = control->event_inj_err;
5602 control->event_inj = 0;
5603 svm_complete_interrupts(svm);
5606 static void svm_vcpu_run(struct kvm_vcpu *vcpu)
5608 struct vcpu_svm *svm = to_svm(vcpu);
5610 svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX];
5611 svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP];
5612 svm->vmcb->save.rip = vcpu->arch.regs[VCPU_REGS_RIP];
5615 * A vmexit emulation is required before the vcpu can be executed
5618 if (unlikely(svm->nested.exit_required))
5622 * Disable singlestep if we're injecting an interrupt/exception.
5623 * We don't want our modified rflags to be pushed on the stack where
5624 * we might not be able to easily reset them if we disabled NMI
5627 if (svm->nmi_singlestep && svm->vmcb->control.event_inj) {
5629 * Event injection happens before external interrupts cause a
5630 * vmexit and interrupts are disabled here, so smp_send_reschedule
5631 * is enough to force an immediate vmexit.
5633 disable_nmi_singlestep(svm);
5634 smp_send_reschedule(vcpu->cpu);
5639 sync_lapic_to_cr8(vcpu);
5641 svm->vmcb->save.cr2 = vcpu->arch.cr2;
5644 kvm_load_guest_xcr0(vcpu);
5647 * If this vCPU has touched SPEC_CTRL, restore the guest's value if
5648 * it's non-zero. Since vmentry is serialising on affected CPUs, there
5649 * is no need to worry about the conditional branch over the wrmsr
5650 * being speculatively taken.
5652 x86_spec_ctrl_set_guest(svm->spec_ctrl, svm->virt_spec_ctrl);
5657 "push %%" _ASM_BP "; \n\t"
5658 "mov %c[rbx](%[svm]), %%" _ASM_BX " \n\t"
5659 "mov %c[rcx](%[svm]), %%" _ASM_CX " \n\t"
5660 "mov %c[rdx](%[svm]), %%" _ASM_DX " \n\t"
5661 "mov %c[rsi](%[svm]), %%" _ASM_SI " \n\t"
5662 "mov %c[rdi](%[svm]), %%" _ASM_DI " \n\t"
5663 "mov %c[rbp](%[svm]), %%" _ASM_BP " \n\t"
5664 #ifdef CONFIG_X86_64
5665 "mov %c[r8](%[svm]), %%r8 \n\t"
5666 "mov %c[r9](%[svm]), %%r9 \n\t"
5667 "mov %c[r10](%[svm]), %%r10 \n\t"
5668 "mov %c[r11](%[svm]), %%r11 \n\t"
5669 "mov %c[r12](%[svm]), %%r12 \n\t"
5670 "mov %c[r13](%[svm]), %%r13 \n\t"
5671 "mov %c[r14](%[svm]), %%r14 \n\t"
5672 "mov %c[r15](%[svm]), %%r15 \n\t"
5675 /* Enter guest mode */
5676 "push %%" _ASM_AX " \n\t"
5677 "mov %c[vmcb](%[svm]), %%" _ASM_AX " \n\t"
5678 __ex("vmload %%" _ASM_AX) "\n\t"
5679 __ex("vmrun %%" _ASM_AX) "\n\t"
5680 __ex("vmsave %%" _ASM_AX) "\n\t"
5681 "pop %%" _ASM_AX " \n\t"
5683 /* Save guest registers, load host registers */
5684 "mov %%" _ASM_BX ", %c[rbx](%[svm]) \n\t"
5685 "mov %%" _ASM_CX ", %c[rcx](%[svm]) \n\t"
5686 "mov %%" _ASM_DX ", %c[rdx](%[svm]) \n\t"
5687 "mov %%" _ASM_SI ", %c[rsi](%[svm]) \n\t"
5688 "mov %%" _ASM_DI ", %c[rdi](%[svm]) \n\t"
5689 "mov %%" _ASM_BP ", %c[rbp](%[svm]) \n\t"
5690 #ifdef CONFIG_X86_64
5691 "mov %%r8, %c[r8](%[svm]) \n\t"
5692 "mov %%r9, %c[r9](%[svm]) \n\t"
5693 "mov %%r10, %c[r10](%[svm]) \n\t"
5694 "mov %%r11, %c[r11](%[svm]) \n\t"
5695 "mov %%r12, %c[r12](%[svm]) \n\t"
5696 "mov %%r13, %c[r13](%[svm]) \n\t"
5697 "mov %%r14, %c[r14](%[svm]) \n\t"
5698 "mov %%r15, %c[r15](%[svm]) \n\t"
5700 * Clear host registers marked as clobbered to prevent
5703 "xor %%r8d, %%r8d \n\t"
5704 "xor %%r9d, %%r9d \n\t"
5705 "xor %%r10d, %%r10d \n\t"
5706 "xor %%r11d, %%r11d \n\t"
5707 "xor %%r12d, %%r12d \n\t"
5708 "xor %%r13d, %%r13d \n\t"
5709 "xor %%r14d, %%r14d \n\t"
5710 "xor %%r15d, %%r15d \n\t"
5712 "xor %%ebx, %%ebx \n\t"
5713 "xor %%ecx, %%ecx \n\t"
5714 "xor %%edx, %%edx \n\t"
5715 "xor %%esi, %%esi \n\t"
5716 "xor %%edi, %%edi \n\t"
5720 [vmcb]"i"(offsetof(struct vcpu_svm, vmcb_pa)),
5721 [rbx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBX])),
5722 [rcx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RCX])),
5723 [rdx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDX])),
5724 [rsi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RSI])),
5725 [rdi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDI])),
5726 [rbp]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBP]))
5727 #ifdef CONFIG_X86_64
5728 , [r8]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R8])),
5729 [r9]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R9])),
5730 [r10]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R10])),
5731 [r11]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R11])),
5732 [r12]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R12])),
5733 [r13]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R13])),
5734 [r14]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R14])),
5735 [r15]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R15]))
5738 #ifdef CONFIG_X86_64
5739 , "rbx", "rcx", "rdx", "rsi", "rdi"
5740 , "r8", "r9", "r10", "r11" , "r12", "r13", "r14", "r15"
5742 , "ebx", "ecx", "edx", "esi", "edi"
5746 /* Eliminate branch target predictions from guest mode */
5749 #ifdef CONFIG_X86_64
5750 wrmsrl(MSR_GS_BASE, svm->host.gs_base);
5752 loadsegment(fs, svm->host.fs);
5753 #ifndef CONFIG_X86_32_LAZY_GS
5754 loadsegment(gs, svm->host.gs);
5759 * We do not use IBRS in the kernel. If this vCPU has used the
5760 * SPEC_CTRL MSR it may have left it on; save the value and
5761 * turn it off. This is much more efficient than blindly adding
5762 * it to the atomic save/restore list. Especially as the former
5763 * (Saving guest MSRs on vmexit) doesn't even exist in KVM.
5765 * For non-nested case:
5766 * If the L01 MSR bitmap does not intercept the MSR, then we need to
5770 * If the L02 MSR bitmap does not intercept the MSR, then we need to
5773 if (unlikely(!msr_write_intercepted(vcpu, MSR_IA32_SPEC_CTRL)))
5774 svm->spec_ctrl = native_read_msr(MSR_IA32_SPEC_CTRL);
5778 local_irq_disable();
5780 x86_spec_ctrl_restore_host(svm->spec_ctrl, svm->virt_spec_ctrl);
5782 vcpu->arch.cr2 = svm->vmcb->save.cr2;
5783 vcpu->arch.regs[VCPU_REGS_RAX] = svm->vmcb->save.rax;
5784 vcpu->arch.regs[VCPU_REGS_RSP] = svm->vmcb->save.rsp;
5785 vcpu->arch.regs[VCPU_REGS_RIP] = svm->vmcb->save.rip;
5787 if (unlikely(svm->vmcb->control.exit_code == SVM_EXIT_NMI))
5788 kvm_before_interrupt(&svm->vcpu);
5790 kvm_put_guest_xcr0(vcpu);
5793 /* Any pending NMI will happen here */
5795 if (unlikely(svm->vmcb->control.exit_code == SVM_EXIT_NMI))
5796 kvm_after_interrupt(&svm->vcpu);
5798 sync_cr8_to_lapic(vcpu);
5802 svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
5804 /* if exit due to PF check for async PF */
5805 if (svm->vmcb->control.exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR)
5806 svm->vcpu.arch.apf.host_apf_reason = kvm_read_and_reset_pf_reason();
5809 vcpu->arch.regs_avail &= ~(1 << VCPU_EXREG_PDPTR);
5810 vcpu->arch.regs_dirty &= ~(1 << VCPU_EXREG_PDPTR);
5814 * We need to handle MC intercepts here before the vcpu has a chance to
5815 * change the physical cpu
5817 if (unlikely(svm->vmcb->control.exit_code ==
5818 SVM_EXIT_EXCP_BASE + MC_VECTOR))
5819 svm_handle_mce(svm);
5821 mark_all_clean(svm->vmcb);
5823 STACK_FRAME_NON_STANDARD(svm_vcpu_run);
5825 static void svm_set_cr3(struct kvm_vcpu *vcpu, unsigned long root)
5827 struct vcpu_svm *svm = to_svm(vcpu);
5829 svm->vmcb->save.cr3 = __sme_set(root);
5830 mark_dirty(svm->vmcb, VMCB_CR);
5833 static void set_tdp_cr3(struct kvm_vcpu *vcpu, unsigned long root)
5835 struct vcpu_svm *svm = to_svm(vcpu);
5837 svm->vmcb->control.nested_cr3 = __sme_set(root);
5838 mark_dirty(svm->vmcb, VMCB_NPT);
5840 /* Also sync guest cr3 here in case we live migrate */
5841 svm->vmcb->save.cr3 = kvm_read_cr3(vcpu);
5842 mark_dirty(svm->vmcb, VMCB_CR);
5845 static int is_disabled(void)
5849 rdmsrl(MSR_VM_CR, vm_cr);
5850 if (vm_cr & (1 << SVM_VM_CR_SVM_DISABLE))
5857 svm_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
5860 * Patch in the VMMCALL instruction:
5862 hypercall[0] = 0x0f;
5863 hypercall[1] = 0x01;
5864 hypercall[2] = 0xd9;
5867 static void svm_check_processor_compat(void *rtn)
5872 static bool svm_cpu_has_accelerated_tpr(void)
5877 static bool svm_has_emulated_msr(int index)
5880 case MSR_IA32_MCG_EXT_CTL:
5889 static u64 svm_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
5894 static void svm_cpuid_update(struct kvm_vcpu *vcpu)
5896 struct vcpu_svm *svm = to_svm(vcpu);
5898 /* Update nrips enabled cache */
5899 svm->nrips_enabled = !!guest_cpuid_has(&svm->vcpu, X86_FEATURE_NRIPS);
5901 if (!kvm_vcpu_apicv_active(vcpu))
5904 guest_cpuid_clear(vcpu, X86_FEATURE_X2APIC);
5907 static void svm_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
5912 entry->ecx &= ~bit(X86_FEATURE_X2APIC);
5916 entry->ecx |= (1 << 2); /* Set SVM bit */
5919 entry->eax = 1; /* SVM revision 1 */
5920 entry->ebx = 8; /* Lets support 8 ASIDs in case we add proper
5921 ASID emulation to nested SVM */
5922 entry->ecx = 0; /* Reserved */
5923 entry->edx = 0; /* Per default do not support any
5924 additional features */
5926 /* Support next_rip if host supports it */
5927 if (boot_cpu_has(X86_FEATURE_NRIPS))
5928 entry->edx |= SVM_FEATURE_NRIP;
5930 /* Support NPT for the guest if enabled */
5932 entry->edx |= SVM_FEATURE_NPT;
5936 /* Support memory encryption cpuid if host supports it */
5937 if (boot_cpu_has(X86_FEATURE_SEV))
5938 cpuid(0x8000001f, &entry->eax, &entry->ebx,
5939 &entry->ecx, &entry->edx);
5944 static int svm_get_lpage_level(void)
5946 return PT_PDPE_LEVEL;
5949 static bool svm_rdtscp_supported(void)
5951 return boot_cpu_has(X86_FEATURE_RDTSCP);
5954 static bool svm_invpcid_supported(void)
5959 static bool svm_mpx_supported(void)
5964 static bool svm_xsaves_supported(void)
5969 static bool svm_umip_emulated(void)
5974 static bool svm_pt_supported(void)
5979 static bool svm_has_wbinvd_exit(void)
5984 #define PRE_EX(exit) { .exit_code = (exit), \
5985 .stage = X86_ICPT_PRE_EXCEPT, }
5986 #define POST_EX(exit) { .exit_code = (exit), \
5987 .stage = X86_ICPT_POST_EXCEPT, }
5988 #define POST_MEM(exit) { .exit_code = (exit), \
5989 .stage = X86_ICPT_POST_MEMACCESS, }
5991 static const struct __x86_intercept {
5993 enum x86_intercept_stage stage;
5994 } x86_intercept_map[] = {
5995 [x86_intercept_cr_read] = POST_EX(SVM_EXIT_READ_CR0),
5996 [x86_intercept_cr_write] = POST_EX(SVM_EXIT_WRITE_CR0),
5997 [x86_intercept_clts] = POST_EX(SVM_EXIT_WRITE_CR0),
5998 [x86_intercept_lmsw] = POST_EX(SVM_EXIT_WRITE_CR0),
5999 [x86_intercept_smsw] = POST_EX(SVM_EXIT_READ_CR0),
6000 [x86_intercept_dr_read] = POST_EX(SVM_EXIT_READ_DR0),
6001 [x86_intercept_dr_write] = POST_EX(SVM_EXIT_WRITE_DR0),
6002 [x86_intercept_sldt] = POST_EX(SVM_EXIT_LDTR_READ),
6003 [x86_intercept_str] = POST_EX(SVM_EXIT_TR_READ),
6004 [x86_intercept_lldt] = POST_EX(SVM_EXIT_LDTR_WRITE),
6005 [x86_intercept_ltr] = POST_EX(SVM_EXIT_TR_WRITE),
6006 [x86_intercept_sgdt] = POST_EX(SVM_EXIT_GDTR_READ),
6007 [x86_intercept_sidt] = POST_EX(SVM_EXIT_IDTR_READ),
6008 [x86_intercept_lgdt] = POST_EX(SVM_EXIT_GDTR_WRITE),
6009 [x86_intercept_lidt] = POST_EX(SVM_EXIT_IDTR_WRITE),
6010 [x86_intercept_vmrun] = POST_EX(SVM_EXIT_VMRUN),
6011 [x86_intercept_vmmcall] = POST_EX(SVM_EXIT_VMMCALL),
6012 [x86_intercept_vmload] = POST_EX(SVM_EXIT_VMLOAD),
6013 [x86_intercept_vmsave] = POST_EX(SVM_EXIT_VMSAVE),
6014 [x86_intercept_stgi] = POST_EX(SVM_EXIT_STGI),
6015 [x86_intercept_clgi] = POST_EX(SVM_EXIT_CLGI),
6016 [x86_intercept_skinit] = POST_EX(SVM_EXIT_SKINIT),
6017 [x86_intercept_invlpga] = POST_EX(SVM_EXIT_INVLPGA),
6018 [x86_intercept_rdtscp] = POST_EX(SVM_EXIT_RDTSCP),
6019 [x86_intercept_monitor] = POST_MEM(SVM_EXIT_MONITOR),
6020 [x86_intercept_mwait] = POST_EX(SVM_EXIT_MWAIT),
6021 [x86_intercept_invlpg] = POST_EX(SVM_EXIT_INVLPG),
6022 [x86_intercept_invd] = POST_EX(SVM_EXIT_INVD),
6023 [x86_intercept_wbinvd] = POST_EX(SVM_EXIT_WBINVD),
6024 [x86_intercept_wrmsr] = POST_EX(SVM_EXIT_MSR),
6025 [x86_intercept_rdtsc] = POST_EX(SVM_EXIT_RDTSC),
6026 [x86_intercept_rdmsr] = POST_EX(SVM_EXIT_MSR),
6027 [x86_intercept_rdpmc] = POST_EX(SVM_EXIT_RDPMC),
6028 [x86_intercept_cpuid] = PRE_EX(SVM_EXIT_CPUID),
6029 [x86_intercept_rsm] = PRE_EX(SVM_EXIT_RSM),
6030 [x86_intercept_pause] = PRE_EX(SVM_EXIT_PAUSE),
6031 [x86_intercept_pushf] = PRE_EX(SVM_EXIT_PUSHF),
6032 [x86_intercept_popf] = PRE_EX(SVM_EXIT_POPF),
6033 [x86_intercept_intn] = PRE_EX(SVM_EXIT_SWINT),
6034 [x86_intercept_iret] = PRE_EX(SVM_EXIT_IRET),
6035 [x86_intercept_icebp] = PRE_EX(SVM_EXIT_ICEBP),
6036 [x86_intercept_hlt] = POST_EX(SVM_EXIT_HLT),
6037 [x86_intercept_in] = POST_EX(SVM_EXIT_IOIO),
6038 [x86_intercept_ins] = POST_EX(SVM_EXIT_IOIO),
6039 [x86_intercept_out] = POST_EX(SVM_EXIT_IOIO),
6040 [x86_intercept_outs] = POST_EX(SVM_EXIT_IOIO),
6047 static int svm_check_intercept(struct kvm_vcpu *vcpu,
6048 struct x86_instruction_info *info,
6049 enum x86_intercept_stage stage)
6051 struct vcpu_svm *svm = to_svm(vcpu);
6052 int vmexit, ret = X86EMUL_CONTINUE;
6053 struct __x86_intercept icpt_info;
6054 struct vmcb *vmcb = svm->vmcb;
6056 if (info->intercept >= ARRAY_SIZE(x86_intercept_map))
6059 icpt_info = x86_intercept_map[info->intercept];
6061 if (stage != icpt_info.stage)
6064 switch (icpt_info.exit_code) {
6065 case SVM_EXIT_READ_CR0:
6066 if (info->intercept == x86_intercept_cr_read)
6067 icpt_info.exit_code += info->modrm_reg;
6069 case SVM_EXIT_WRITE_CR0: {
6070 unsigned long cr0, val;
6073 if (info->intercept == x86_intercept_cr_write)
6074 icpt_info.exit_code += info->modrm_reg;
6076 if (icpt_info.exit_code != SVM_EXIT_WRITE_CR0 ||
6077 info->intercept == x86_intercept_clts)
6080 intercept = svm->nested.intercept;
6082 if (!(intercept & (1ULL << INTERCEPT_SELECTIVE_CR0)))
6085 cr0 = vcpu->arch.cr0 & ~SVM_CR0_SELECTIVE_MASK;
6086 val = info->src_val & ~SVM_CR0_SELECTIVE_MASK;
6088 if (info->intercept == x86_intercept_lmsw) {
6091 /* lmsw can't clear PE - catch this here */
6092 if (cr0 & X86_CR0_PE)
6097 icpt_info.exit_code = SVM_EXIT_CR0_SEL_WRITE;
6101 case SVM_EXIT_READ_DR0:
6102 case SVM_EXIT_WRITE_DR0:
6103 icpt_info.exit_code += info->modrm_reg;
6106 if (info->intercept == x86_intercept_wrmsr)
6107 vmcb->control.exit_info_1 = 1;
6109 vmcb->control.exit_info_1 = 0;
6111 case SVM_EXIT_PAUSE:
6113 * We get this for NOP only, but pause
6114 * is rep not, check this here
6116 if (info->rep_prefix != REPE_PREFIX)
6119 case SVM_EXIT_IOIO: {
6123 if (info->intercept == x86_intercept_in ||
6124 info->intercept == x86_intercept_ins) {
6125 exit_info = ((info->src_val & 0xffff) << 16) |
6127 bytes = info->dst_bytes;
6129 exit_info = (info->dst_val & 0xffff) << 16;
6130 bytes = info->src_bytes;
6133 if (info->intercept == x86_intercept_outs ||
6134 info->intercept == x86_intercept_ins)
6135 exit_info |= SVM_IOIO_STR_MASK;
6137 if (info->rep_prefix)
6138 exit_info |= SVM_IOIO_REP_MASK;
6140 bytes = min(bytes, 4u);
6142 exit_info |= bytes << SVM_IOIO_SIZE_SHIFT;
6144 exit_info |= (u32)info->ad_bytes << (SVM_IOIO_ASIZE_SHIFT - 1);
6146 vmcb->control.exit_info_1 = exit_info;
6147 vmcb->control.exit_info_2 = info->next_rip;
6155 /* TODO: Advertise NRIPS to guest hypervisor unconditionally */
6156 if (static_cpu_has(X86_FEATURE_NRIPS))
6157 vmcb->control.next_rip = info->next_rip;
6158 vmcb->control.exit_code = icpt_info.exit_code;
6159 vmexit = nested_svm_exit_handled(svm);
6161 ret = (vmexit == NESTED_EXIT_DONE) ? X86EMUL_INTERCEPTED
6168 static void svm_handle_external_intr(struct kvm_vcpu *vcpu)
6172 * We must have an instruction with interrupts enabled, so
6173 * the timer interrupt isn't delayed by the interrupt shadow.
6176 local_irq_disable();
6179 static void svm_sched_in(struct kvm_vcpu *vcpu, int cpu)
6181 if (pause_filter_thresh)
6182 shrink_ple_window(vcpu);
6185 static inline void avic_post_state_restore(struct kvm_vcpu *vcpu)
6187 if (avic_handle_apic_id_update(vcpu) != 0)
6189 avic_handle_dfr_update(vcpu);
6190 avic_handle_ldr_update(vcpu);
6193 static void svm_setup_mce(struct kvm_vcpu *vcpu)
6195 /* [63:9] are reserved. */
6196 vcpu->arch.mcg_cap &= 0x1ff;
6199 static int svm_smi_allowed(struct kvm_vcpu *vcpu)
6201 struct vcpu_svm *svm = to_svm(vcpu);
6203 /* Per APM Vol.2 15.22.2 "Response to SMI" */
6207 if (is_guest_mode(&svm->vcpu) &&
6208 svm->nested.intercept & (1ULL << INTERCEPT_SMI)) {
6209 /* TODO: Might need to set exit_info_1 and exit_info_2 here */
6210 svm->vmcb->control.exit_code = SVM_EXIT_SMI;
6211 svm->nested.exit_required = true;
6218 static int svm_pre_enter_smm(struct kvm_vcpu *vcpu, char *smstate)
6220 struct vcpu_svm *svm = to_svm(vcpu);
6223 if (is_guest_mode(vcpu)) {
6224 /* FED8h - SVM Guest */
6225 put_smstate(u64, smstate, 0x7ed8, 1);
6226 /* FEE0h - SVM Guest VMCB Physical Address */
6227 put_smstate(u64, smstate, 0x7ee0, svm->nested.vmcb);
6229 svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX];
6230 svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP];
6231 svm->vmcb->save.rip = vcpu->arch.regs[VCPU_REGS_RIP];
6233 ret = nested_svm_vmexit(svm);
6240 static int svm_pre_leave_smm(struct kvm_vcpu *vcpu, const char *smstate)
6242 struct vcpu_svm *svm = to_svm(vcpu);
6243 struct vmcb *nested_vmcb;
6244 struct kvm_host_map map;
6248 guest = GET_SMSTATE(u64, smstate, 0x7ed8);
6249 vmcb = GET_SMSTATE(u64, smstate, 0x7ee0);
6252 if (kvm_vcpu_map(&svm->vcpu, gpa_to_gfn(vmcb), &map) == -EINVAL)
6254 nested_vmcb = map.hva;
6255 enter_svm_guest_mode(svm, vmcb, nested_vmcb, &map);
6260 static int enable_smi_window(struct kvm_vcpu *vcpu)
6262 struct vcpu_svm *svm = to_svm(vcpu);
6264 if (!gif_set(svm)) {
6265 if (vgif_enabled(svm))
6266 set_intercept(svm, INTERCEPT_STGI);
6267 /* STGI will cause a vm exit */
6273 static int sev_asid_new(void)
6278 * SEV-enabled guest must use asid from min_sev_asid to max_sev_asid.
6280 pos = find_next_zero_bit(sev_asid_bitmap, max_sev_asid, min_sev_asid - 1);
6281 if (pos >= max_sev_asid)
6284 set_bit(pos, sev_asid_bitmap);
6288 static int sev_guest_init(struct kvm *kvm, struct kvm_sev_cmd *argp)
6290 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
6294 if (unlikely(sev->active))
6297 asid = sev_asid_new();
6301 ret = sev_platform_init(&argp->error);
6307 INIT_LIST_HEAD(&sev->regions_list);
6312 __sev_asid_free(asid);
6316 static int sev_bind_asid(struct kvm *kvm, unsigned int handle, int *error)
6318 struct sev_data_activate *data;
6319 int asid = sev_get_asid(kvm);
6322 wbinvd_on_all_cpus();
6324 ret = sev_guest_df_flush(error);
6328 data = kzalloc(sizeof(*data), GFP_KERNEL_ACCOUNT);
6332 /* activate ASID on the given handle */
6333 data->handle = handle;
6335 ret = sev_guest_activate(data, error);
6341 static int __sev_issue_cmd(int fd, int id, void *data, int *error)
6350 ret = sev_issue_cmd_external_user(f.file, id, data, error);
6356 static int sev_issue_cmd(struct kvm *kvm, int id, void *data, int *error)
6358 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
6360 return __sev_issue_cmd(sev->fd, id, data, error);
6363 static int sev_launch_start(struct kvm *kvm, struct kvm_sev_cmd *argp)
6365 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
6366 struct sev_data_launch_start *start;
6367 struct kvm_sev_launch_start params;
6368 void *dh_blob, *session_blob;
6369 int *error = &argp->error;
6372 if (!sev_guest(kvm))
6375 if (copy_from_user(¶ms, (void __user *)(uintptr_t)argp->data, sizeof(params)))
6378 start = kzalloc(sizeof(*start), GFP_KERNEL_ACCOUNT);
6383 if (params.dh_uaddr) {
6384 dh_blob = psp_copy_user_blob(params.dh_uaddr, params.dh_len);
6385 if (IS_ERR(dh_blob)) {
6386 ret = PTR_ERR(dh_blob);
6390 start->dh_cert_address = __sme_set(__pa(dh_blob));
6391 start->dh_cert_len = params.dh_len;
6394 session_blob = NULL;
6395 if (params.session_uaddr) {
6396 session_blob = psp_copy_user_blob(params.session_uaddr, params.session_len);
6397 if (IS_ERR(session_blob)) {
6398 ret = PTR_ERR(session_blob);
6402 start->session_address = __sme_set(__pa(session_blob));
6403 start->session_len = params.session_len;
6406 start->handle = params.handle;
6407 start->policy = params.policy;
6409 /* create memory encryption context */
6410 ret = __sev_issue_cmd(argp->sev_fd, SEV_CMD_LAUNCH_START, start, error);
6412 goto e_free_session;
6414 /* Bind ASID to this guest */
6415 ret = sev_bind_asid(kvm, start->handle, error);
6417 goto e_free_session;
6419 /* return handle to userspace */
6420 params.handle = start->handle;
6421 if (copy_to_user((void __user *)(uintptr_t)argp->data, ¶ms, sizeof(params))) {
6422 sev_unbind_asid(kvm, start->handle);
6424 goto e_free_session;
6427 sev->handle = start->handle;
6428 sev->fd = argp->sev_fd;
6431 kfree(session_blob);
6439 static unsigned long get_num_contig_pages(unsigned long idx,
6440 struct page **inpages, unsigned long npages)
6442 unsigned long paddr, next_paddr;
6443 unsigned long i = idx + 1, pages = 1;
6445 /* find the number of contiguous pages starting from idx */
6446 paddr = __sme_page_pa(inpages[idx]);
6447 while (i < npages) {
6448 next_paddr = __sme_page_pa(inpages[i++]);
6449 if ((paddr + PAGE_SIZE) == next_paddr) {
6460 static int sev_launch_update_data(struct kvm *kvm, struct kvm_sev_cmd *argp)
6462 unsigned long vaddr, vaddr_end, next_vaddr, npages, pages, size, i;
6463 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
6464 struct kvm_sev_launch_update_data params;
6465 struct sev_data_launch_update_data *data;
6466 struct page **inpages;
6469 if (!sev_guest(kvm))
6472 if (copy_from_user(¶ms, (void __user *)(uintptr_t)argp->data, sizeof(params)))
6475 data = kzalloc(sizeof(*data), GFP_KERNEL_ACCOUNT);
6479 vaddr = params.uaddr;
6481 vaddr_end = vaddr + size;
6483 /* Lock the user memory. */
6484 inpages = sev_pin_memory(kvm, vaddr, size, &npages, 1);
6491 * The LAUNCH_UPDATE command will perform in-place encryption of the
6492 * memory content (i.e it will write the same memory region with C=1).
6493 * It's possible that the cache may contain the data with C=0, i.e.,
6494 * unencrypted so invalidate it first.
6496 sev_clflush_pages(inpages, npages);
6498 for (i = 0; vaddr < vaddr_end; vaddr = next_vaddr, i += pages) {
6502 * If the user buffer is not page-aligned, calculate the offset
6505 offset = vaddr & (PAGE_SIZE - 1);
6507 /* Calculate the number of pages that can be encrypted in one go. */
6508 pages = get_num_contig_pages(i, inpages, npages);
6510 len = min_t(size_t, ((pages * PAGE_SIZE) - offset), size);
6512 data->handle = sev->handle;
6514 data->address = __sme_page_pa(inpages[i]) + offset;
6515 ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_UPDATE_DATA, data, &argp->error);
6520 next_vaddr = vaddr + len;
6524 /* content of memory is updated, mark pages dirty */
6525 for (i = 0; i < npages; i++) {
6526 set_page_dirty_lock(inpages[i]);
6527 mark_page_accessed(inpages[i]);
6529 /* unlock the user pages */
6530 sev_unpin_memory(kvm, inpages, npages);
6536 static int sev_launch_measure(struct kvm *kvm, struct kvm_sev_cmd *argp)
6538 void __user *measure = (void __user *)(uintptr_t)argp->data;
6539 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
6540 struct sev_data_launch_measure *data;
6541 struct kvm_sev_launch_measure params;
6542 void __user *p = NULL;
6546 if (!sev_guest(kvm))
6549 if (copy_from_user(¶ms, measure, sizeof(params)))
6552 data = kzalloc(sizeof(*data), GFP_KERNEL_ACCOUNT);
6556 /* User wants to query the blob length */
6560 p = (void __user *)(uintptr_t)params.uaddr;
6562 if (params.len > SEV_FW_BLOB_MAX_SIZE) {
6568 blob = kmalloc(params.len, GFP_KERNEL);
6572 data->address = __psp_pa(blob);
6573 data->len = params.len;
6577 data->handle = sev->handle;
6578 ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_MEASURE, data, &argp->error);
6581 * If we query the session length, FW responded with expected data.
6590 if (copy_to_user(p, blob, params.len))
6595 params.len = data->len;
6596 if (copy_to_user(measure, ¶ms, sizeof(params)))
6605 static int sev_launch_finish(struct kvm *kvm, struct kvm_sev_cmd *argp)
6607 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
6608 struct sev_data_launch_finish *data;
6611 if (!sev_guest(kvm))
6614 data = kzalloc(sizeof(*data), GFP_KERNEL_ACCOUNT);
6618 data->handle = sev->handle;
6619 ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_FINISH, data, &argp->error);
6625 static int sev_guest_status(struct kvm *kvm, struct kvm_sev_cmd *argp)
6627 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
6628 struct kvm_sev_guest_status params;
6629 struct sev_data_guest_status *data;
6632 if (!sev_guest(kvm))
6635 data = kzalloc(sizeof(*data), GFP_KERNEL_ACCOUNT);
6639 data->handle = sev->handle;
6640 ret = sev_issue_cmd(kvm, SEV_CMD_GUEST_STATUS, data, &argp->error);
6644 params.policy = data->policy;
6645 params.state = data->state;
6646 params.handle = data->handle;
6648 if (copy_to_user((void __user *)(uintptr_t)argp->data, ¶ms, sizeof(params)))
6655 static int __sev_issue_dbg_cmd(struct kvm *kvm, unsigned long src,
6656 unsigned long dst, int size,
6657 int *error, bool enc)
6659 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
6660 struct sev_data_dbg *data;
6663 data = kzalloc(sizeof(*data), GFP_KERNEL_ACCOUNT);
6667 data->handle = sev->handle;
6668 data->dst_addr = dst;
6669 data->src_addr = src;
6672 ret = sev_issue_cmd(kvm,
6673 enc ? SEV_CMD_DBG_ENCRYPT : SEV_CMD_DBG_DECRYPT,
6679 static int __sev_dbg_decrypt(struct kvm *kvm, unsigned long src_paddr,
6680 unsigned long dst_paddr, int sz, int *err)
6685 * Its safe to read more than we are asked, caller should ensure that
6686 * destination has enough space.
6688 src_paddr = round_down(src_paddr, 16);
6689 offset = src_paddr & 15;
6690 sz = round_up(sz + offset, 16);
6692 return __sev_issue_dbg_cmd(kvm, src_paddr, dst_paddr, sz, err, false);
6695 static int __sev_dbg_decrypt_user(struct kvm *kvm, unsigned long paddr,
6696 unsigned long __user dst_uaddr,
6697 unsigned long dst_paddr,
6700 struct page *tpage = NULL;
6703 /* if inputs are not 16-byte then use intermediate buffer */
6704 if (!IS_ALIGNED(dst_paddr, 16) ||
6705 !IS_ALIGNED(paddr, 16) ||
6706 !IS_ALIGNED(size, 16)) {
6707 tpage = (void *)alloc_page(GFP_KERNEL);
6711 dst_paddr = __sme_page_pa(tpage);
6714 ret = __sev_dbg_decrypt(kvm, paddr, dst_paddr, size, err);
6719 offset = paddr & 15;
6720 if (copy_to_user((void __user *)(uintptr_t)dst_uaddr,
6721 page_address(tpage) + offset, size))
6732 static int __sev_dbg_encrypt_user(struct kvm *kvm, unsigned long paddr,
6733 unsigned long __user vaddr,
6734 unsigned long dst_paddr,
6735 unsigned long __user dst_vaddr,
6736 int size, int *error)
6738 struct page *src_tpage = NULL;
6739 struct page *dst_tpage = NULL;
6740 int ret, len = size;
6742 /* If source buffer is not aligned then use an intermediate buffer */
6743 if (!IS_ALIGNED(vaddr, 16)) {
6744 src_tpage = alloc_page(GFP_KERNEL);
6748 if (copy_from_user(page_address(src_tpage),
6749 (void __user *)(uintptr_t)vaddr, size)) {
6750 __free_page(src_tpage);
6754 paddr = __sme_page_pa(src_tpage);
6758 * If destination buffer or length is not aligned then do read-modify-write:
6759 * - decrypt destination in an intermediate buffer
6760 * - copy the source buffer in an intermediate buffer
6761 * - use the intermediate buffer as source buffer
6763 if (!IS_ALIGNED(dst_vaddr, 16) || !IS_ALIGNED(size, 16)) {
6766 dst_tpage = alloc_page(GFP_KERNEL);
6772 ret = __sev_dbg_decrypt(kvm, dst_paddr,
6773 __sme_page_pa(dst_tpage), size, error);
6778 * If source is kernel buffer then use memcpy() otherwise
6781 dst_offset = dst_paddr & 15;
6784 memcpy(page_address(dst_tpage) + dst_offset,
6785 page_address(src_tpage), size);
6787 if (copy_from_user(page_address(dst_tpage) + dst_offset,
6788 (void __user *)(uintptr_t)vaddr, size)) {
6794 paddr = __sme_page_pa(dst_tpage);
6795 dst_paddr = round_down(dst_paddr, 16);
6796 len = round_up(size, 16);
6799 ret = __sev_issue_dbg_cmd(kvm, paddr, dst_paddr, len, error, true);
6803 __free_page(src_tpage);
6805 __free_page(dst_tpage);
6809 static int sev_dbg_crypt(struct kvm *kvm, struct kvm_sev_cmd *argp, bool dec)
6811 unsigned long vaddr, vaddr_end, next_vaddr;
6812 unsigned long dst_vaddr;
6813 struct page **src_p, **dst_p;
6814 struct kvm_sev_dbg debug;
6819 if (!sev_guest(kvm))
6822 if (copy_from_user(&debug, (void __user *)(uintptr_t)argp->data, sizeof(debug)))
6825 if (!debug.len || debug.src_uaddr + debug.len < debug.src_uaddr)
6827 if (!debug.dst_uaddr)
6830 vaddr = debug.src_uaddr;
6832 vaddr_end = vaddr + size;
6833 dst_vaddr = debug.dst_uaddr;
6835 for (; vaddr < vaddr_end; vaddr = next_vaddr) {
6836 int len, s_off, d_off;
6838 /* lock userspace source and destination page */
6839 src_p = sev_pin_memory(kvm, vaddr & PAGE_MASK, PAGE_SIZE, &n, 0);
6843 dst_p = sev_pin_memory(kvm, dst_vaddr & PAGE_MASK, PAGE_SIZE, &n, 1);
6845 sev_unpin_memory(kvm, src_p, n);
6850 * The DBG_{DE,EN}CRYPT commands will perform {dec,en}cryption of the
6851 * memory content (i.e it will write the same memory region with C=1).
6852 * It's possible that the cache may contain the data with C=0, i.e.,
6853 * unencrypted so invalidate it first.
6855 sev_clflush_pages(src_p, 1);
6856 sev_clflush_pages(dst_p, 1);
6859 * Since user buffer may not be page aligned, calculate the
6860 * offset within the page.
6862 s_off = vaddr & ~PAGE_MASK;
6863 d_off = dst_vaddr & ~PAGE_MASK;
6864 len = min_t(size_t, (PAGE_SIZE - s_off), size);
6867 ret = __sev_dbg_decrypt_user(kvm,
6868 __sme_page_pa(src_p[0]) + s_off,
6870 __sme_page_pa(dst_p[0]) + d_off,
6873 ret = __sev_dbg_encrypt_user(kvm,
6874 __sme_page_pa(src_p[0]) + s_off,
6876 __sme_page_pa(dst_p[0]) + d_off,
6880 sev_unpin_memory(kvm, src_p, n);
6881 sev_unpin_memory(kvm, dst_p, n);
6886 next_vaddr = vaddr + len;
6887 dst_vaddr = dst_vaddr + len;
6894 static int sev_launch_secret(struct kvm *kvm, struct kvm_sev_cmd *argp)
6896 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
6897 struct sev_data_launch_secret *data;
6898 struct kvm_sev_launch_secret params;
6899 struct page **pages;
6904 if (!sev_guest(kvm))
6907 if (copy_from_user(¶ms, (void __user *)(uintptr_t)argp->data, sizeof(params)))
6910 pages = sev_pin_memory(kvm, params.guest_uaddr, params.guest_len, &n, 1);
6915 * The secret must be copied into contiguous memory region, lets verify
6916 * that userspace memory pages are contiguous before we issue command.
6918 if (get_num_contig_pages(0, pages, n) != n) {
6920 goto e_unpin_memory;
6924 data = kzalloc(sizeof(*data), GFP_KERNEL_ACCOUNT);
6926 goto e_unpin_memory;
6928 offset = params.guest_uaddr & (PAGE_SIZE - 1);
6929 data->guest_address = __sme_page_pa(pages[0]) + offset;
6930 data->guest_len = params.guest_len;
6932 blob = psp_copy_user_blob(params.trans_uaddr, params.trans_len);
6934 ret = PTR_ERR(blob);
6938 data->trans_address = __psp_pa(blob);
6939 data->trans_len = params.trans_len;
6941 hdr = psp_copy_user_blob(params.hdr_uaddr, params.hdr_len);
6946 data->hdr_address = __psp_pa(hdr);
6947 data->hdr_len = params.hdr_len;
6949 data->handle = sev->handle;
6950 ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_UPDATE_SECRET, data, &argp->error);
6959 sev_unpin_memory(kvm, pages, n);
6963 static int svm_mem_enc_op(struct kvm *kvm, void __user *argp)
6965 struct kvm_sev_cmd sev_cmd;
6968 if (!svm_sev_enabled())
6971 if (copy_from_user(&sev_cmd, argp, sizeof(struct kvm_sev_cmd)))
6974 mutex_lock(&kvm->lock);
6976 switch (sev_cmd.id) {
6978 r = sev_guest_init(kvm, &sev_cmd);
6980 case KVM_SEV_LAUNCH_START:
6981 r = sev_launch_start(kvm, &sev_cmd);
6983 case KVM_SEV_LAUNCH_UPDATE_DATA:
6984 r = sev_launch_update_data(kvm, &sev_cmd);
6986 case KVM_SEV_LAUNCH_MEASURE:
6987 r = sev_launch_measure(kvm, &sev_cmd);
6989 case KVM_SEV_LAUNCH_FINISH:
6990 r = sev_launch_finish(kvm, &sev_cmd);
6992 case KVM_SEV_GUEST_STATUS:
6993 r = sev_guest_status(kvm, &sev_cmd);
6995 case KVM_SEV_DBG_DECRYPT:
6996 r = sev_dbg_crypt(kvm, &sev_cmd, true);
6998 case KVM_SEV_DBG_ENCRYPT:
6999 r = sev_dbg_crypt(kvm, &sev_cmd, false);
7001 case KVM_SEV_LAUNCH_SECRET:
7002 r = sev_launch_secret(kvm, &sev_cmd);
7009 if (copy_to_user(argp, &sev_cmd, sizeof(struct kvm_sev_cmd)))
7013 mutex_unlock(&kvm->lock);
7017 static int svm_register_enc_region(struct kvm *kvm,
7018 struct kvm_enc_region *range)
7020 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
7021 struct enc_region *region;
7024 if (!sev_guest(kvm))
7027 if (range->addr > ULONG_MAX || range->size > ULONG_MAX)
7030 region = kzalloc(sizeof(*region), GFP_KERNEL_ACCOUNT);
7034 region->pages = sev_pin_memory(kvm, range->addr, range->size, ®ion->npages, 1);
7035 if (!region->pages) {
7041 * The guest may change the memory encryption attribute from C=0 -> C=1
7042 * or vice versa for this memory range. Lets make sure caches are
7043 * flushed to ensure that guest data gets written into memory with
7046 sev_clflush_pages(region->pages, region->npages);
7048 region->uaddr = range->addr;
7049 region->size = range->size;
7051 mutex_lock(&kvm->lock);
7052 list_add_tail(®ion->list, &sev->regions_list);
7053 mutex_unlock(&kvm->lock);
7062 static struct enc_region *
7063 find_enc_region(struct kvm *kvm, struct kvm_enc_region *range)
7065 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
7066 struct list_head *head = &sev->regions_list;
7067 struct enc_region *i;
7069 list_for_each_entry(i, head, list) {
7070 if (i->uaddr == range->addr &&
7071 i->size == range->size)
7079 static int svm_unregister_enc_region(struct kvm *kvm,
7080 struct kvm_enc_region *range)
7082 struct enc_region *region;
7085 mutex_lock(&kvm->lock);
7087 if (!sev_guest(kvm)) {
7092 region = find_enc_region(kvm, range);
7098 __unregister_enc_region_locked(kvm, region);
7100 mutex_unlock(&kvm->lock);
7104 mutex_unlock(&kvm->lock);
7108 static uint16_t nested_get_evmcs_version(struct kvm_vcpu *vcpu)
7114 static int nested_enable_evmcs(struct kvm_vcpu *vcpu,
7115 uint16_t *vmcs_version)
7117 /* Intel-only feature */
7121 static bool svm_need_emulation_on_page_fault(struct kvm_vcpu *vcpu)
7125 is_user = svm_get_cpl(vcpu) == 3;
7126 smap = !kvm_read_cr4_bits(vcpu, X86_CR4_SMAP);
7129 * Detect and workaround Errata 1096 Fam_17h_00_0Fh
7131 * In non SEV guest, hypervisor will be able to read the guest
7132 * memory to decode the instruction pointer when insn_len is zero
7133 * so we return true to indicate that decoding is possible.
7135 * But in the SEV guest, the guest memory is encrypted with the
7136 * guest specific key and hypervisor will not be able to decode the
7137 * instruction pointer so we will not able to workaround it. Lets
7138 * print the error and request to kill the guest.
7140 if (is_user && smap) {
7141 if (!sev_guest(vcpu->kvm))
7144 pr_err_ratelimited("KVM: Guest triggered AMD Erratum 1096\n");
7145 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
7151 static struct kvm_x86_ops svm_x86_ops __ro_after_init = {
7152 .cpu_has_kvm_support = has_svm,
7153 .disabled_by_bios = is_disabled,
7154 .hardware_setup = svm_hardware_setup,
7155 .hardware_unsetup = svm_hardware_unsetup,
7156 .check_processor_compatibility = svm_check_processor_compat,
7157 .hardware_enable = svm_hardware_enable,
7158 .hardware_disable = svm_hardware_disable,
7159 .cpu_has_accelerated_tpr = svm_cpu_has_accelerated_tpr,
7160 .has_emulated_msr = svm_has_emulated_msr,
7162 .vcpu_create = svm_create_vcpu,
7163 .vcpu_free = svm_free_vcpu,
7164 .vcpu_reset = svm_vcpu_reset,
7166 .vm_alloc = svm_vm_alloc,
7167 .vm_free = svm_vm_free,
7168 .vm_init = avic_vm_init,
7169 .vm_destroy = svm_vm_destroy,
7171 .prepare_guest_switch = svm_prepare_guest_switch,
7172 .vcpu_load = svm_vcpu_load,
7173 .vcpu_put = svm_vcpu_put,
7174 .vcpu_blocking = svm_vcpu_blocking,
7175 .vcpu_unblocking = svm_vcpu_unblocking,
7177 .update_bp_intercept = update_bp_intercept,
7178 .get_msr_feature = svm_get_msr_feature,
7179 .get_msr = svm_get_msr,
7180 .set_msr = svm_set_msr,
7181 .get_segment_base = svm_get_segment_base,
7182 .get_segment = svm_get_segment,
7183 .set_segment = svm_set_segment,
7184 .get_cpl = svm_get_cpl,
7185 .get_cs_db_l_bits = kvm_get_cs_db_l_bits,
7186 .decache_cr0_guest_bits = svm_decache_cr0_guest_bits,
7187 .decache_cr3 = svm_decache_cr3,
7188 .decache_cr4_guest_bits = svm_decache_cr4_guest_bits,
7189 .set_cr0 = svm_set_cr0,
7190 .set_cr3 = svm_set_cr3,
7191 .set_cr4 = svm_set_cr4,
7192 .set_efer = svm_set_efer,
7193 .get_idt = svm_get_idt,
7194 .set_idt = svm_set_idt,
7195 .get_gdt = svm_get_gdt,
7196 .set_gdt = svm_set_gdt,
7197 .get_dr6 = svm_get_dr6,
7198 .set_dr6 = svm_set_dr6,
7199 .set_dr7 = svm_set_dr7,
7200 .sync_dirty_debug_regs = svm_sync_dirty_debug_regs,
7201 .cache_reg = svm_cache_reg,
7202 .get_rflags = svm_get_rflags,
7203 .set_rflags = svm_set_rflags,
7205 .tlb_flush = svm_flush_tlb,
7206 .tlb_flush_gva = svm_flush_tlb_gva,
7208 .run = svm_vcpu_run,
7209 .handle_exit = handle_exit,
7210 .skip_emulated_instruction = skip_emulated_instruction,
7211 .set_interrupt_shadow = svm_set_interrupt_shadow,
7212 .get_interrupt_shadow = svm_get_interrupt_shadow,
7213 .patch_hypercall = svm_patch_hypercall,
7214 .set_irq = svm_set_irq,
7215 .set_nmi = svm_inject_nmi,
7216 .queue_exception = svm_queue_exception,
7217 .cancel_injection = svm_cancel_injection,
7218 .interrupt_allowed = svm_interrupt_allowed,
7219 .nmi_allowed = svm_nmi_allowed,
7220 .get_nmi_mask = svm_get_nmi_mask,
7221 .set_nmi_mask = svm_set_nmi_mask,
7222 .enable_nmi_window = enable_nmi_window,
7223 .enable_irq_window = enable_irq_window,
7224 .update_cr8_intercept = update_cr8_intercept,
7225 .set_virtual_apic_mode = svm_set_virtual_apic_mode,
7226 .get_enable_apicv = svm_get_enable_apicv,
7227 .refresh_apicv_exec_ctrl = svm_refresh_apicv_exec_ctrl,
7228 .load_eoi_exitmap = svm_load_eoi_exitmap,
7229 .hwapic_irr_update = svm_hwapic_irr_update,
7230 .hwapic_isr_update = svm_hwapic_isr_update,
7231 .sync_pir_to_irr = kvm_lapic_find_highest_irr,
7232 .apicv_post_state_restore = avic_post_state_restore,
7234 .set_tss_addr = svm_set_tss_addr,
7235 .set_identity_map_addr = svm_set_identity_map_addr,
7236 .get_tdp_level = get_npt_level,
7237 .get_mt_mask = svm_get_mt_mask,
7239 .get_exit_info = svm_get_exit_info,
7241 .get_lpage_level = svm_get_lpage_level,
7243 .cpuid_update = svm_cpuid_update,
7245 .rdtscp_supported = svm_rdtscp_supported,
7246 .invpcid_supported = svm_invpcid_supported,
7247 .mpx_supported = svm_mpx_supported,
7248 .xsaves_supported = svm_xsaves_supported,
7249 .umip_emulated = svm_umip_emulated,
7250 .pt_supported = svm_pt_supported,
7252 .set_supported_cpuid = svm_set_supported_cpuid,
7254 .has_wbinvd_exit = svm_has_wbinvd_exit,
7256 .read_l1_tsc_offset = svm_read_l1_tsc_offset,
7257 .write_l1_tsc_offset = svm_write_l1_tsc_offset,
7259 .set_tdp_cr3 = set_tdp_cr3,
7261 .check_intercept = svm_check_intercept,
7262 .handle_external_intr = svm_handle_external_intr,
7264 .request_immediate_exit = __kvm_request_immediate_exit,
7266 .sched_in = svm_sched_in,
7268 .pmu_ops = &amd_pmu_ops,
7269 .deliver_posted_interrupt = svm_deliver_avic_intr,
7270 .update_pi_irte = svm_update_pi_irte,
7271 .setup_mce = svm_setup_mce,
7273 .smi_allowed = svm_smi_allowed,
7274 .pre_enter_smm = svm_pre_enter_smm,
7275 .pre_leave_smm = svm_pre_leave_smm,
7276 .enable_smi_window = enable_smi_window,
7278 .mem_enc_op = svm_mem_enc_op,
7279 .mem_enc_reg_region = svm_register_enc_region,
7280 .mem_enc_unreg_region = svm_unregister_enc_region,
7282 .nested_enable_evmcs = nested_enable_evmcs,
7283 .nested_get_evmcs_version = nested_get_evmcs_version,
7285 .need_emulation_on_page_fault = svm_need_emulation_on_page_fault,
7288 static int __init svm_init(void)
7290 return kvm_init(&svm_x86_ops, sizeof(struct vcpu_svm),
7291 __alignof__(struct vcpu_svm), THIS_MODULE);
7294 static void __exit svm_exit(void)
7299 module_init(svm_init)
7300 module_exit(svm_exit)