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 */
149 struct page *avic_logical_id_table_page;
150 struct page *avic_physical_id_table_page;
151 struct hlist_node hnode;
153 struct kvm_sev_info sev_info;
158 struct nested_state {
164 /* These are the merged vectors */
167 /* gpa pointers to the real vectors */
171 /* A VMEXIT is required but not yet emulated */
174 /* cache for intercepts of the guest */
177 u32 intercept_exceptions;
180 /* Nested Paging related state */
184 #define MSRPM_OFFSETS 16
185 static u32 msrpm_offsets[MSRPM_OFFSETS] __read_mostly;
188 * Set osvw_len to higher value when updated Revision Guides
189 * are published and we know what the new status bits are
191 static uint64_t osvw_len = 4, osvw_status;
194 struct kvm_vcpu vcpu;
196 unsigned long vmcb_pa;
197 struct svm_cpu_data *svm_data;
198 uint64_t asid_generation;
199 uint64_t sysenter_esp;
200 uint64_t sysenter_eip;
207 u64 host_user_msrs[NR_HOST_SAVE_USER_MSRS];
217 * Contains guest-controlled bits of VIRT_SPEC_CTRL, which will be
218 * translated into the appropriate L2_CFG bits on the host to
219 * perform speculative control.
227 struct nested_state nested;
230 u64 nmi_singlestep_guest_rflags;
232 unsigned int3_injected;
233 unsigned long int3_rip;
235 /* cached guest cpuid flags for faster access */
236 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_MASK (1 << 31)
267 #define AVIC_PHYSICAL_ID_ENTRY_HOST_PHYSICAL_ID_MASK (0xFFULL)
268 #define AVIC_PHYSICAL_ID_ENTRY_BACKING_PAGE_MASK (0xFFFFFFFFFFULL << 12)
269 #define AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK (1ULL << 62)
270 #define AVIC_PHYSICAL_ID_ENTRY_VALID_MASK (1ULL << 63)
272 static DEFINE_PER_CPU(u64, current_tsc_ratio);
273 #define TSC_RATIO_DEFAULT 0x0100000000ULL
275 #define MSR_INVALID 0xffffffffU
277 static const struct svm_direct_access_msrs {
278 u32 index; /* Index of the MSR */
279 bool always; /* True if intercept is always on */
280 } direct_access_msrs[] = {
281 { .index = MSR_STAR, .always = true },
282 { .index = MSR_IA32_SYSENTER_CS, .always = true },
284 { .index = MSR_GS_BASE, .always = true },
285 { .index = MSR_FS_BASE, .always = true },
286 { .index = MSR_KERNEL_GS_BASE, .always = true },
287 { .index = MSR_LSTAR, .always = true },
288 { .index = MSR_CSTAR, .always = true },
289 { .index = MSR_SYSCALL_MASK, .always = true },
291 { .index = MSR_IA32_SPEC_CTRL, .always = false },
292 { .index = MSR_IA32_PRED_CMD, .always = false },
293 { .index = MSR_IA32_LASTBRANCHFROMIP, .always = false },
294 { .index = MSR_IA32_LASTBRANCHTOIP, .always = false },
295 { .index = MSR_IA32_LASTINTFROMIP, .always = false },
296 { .index = MSR_IA32_LASTINTTOIP, .always = false },
297 { .index = MSR_INVALID, .always = false },
300 /* enable NPT for AMD64 and X86 with PAE */
301 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
302 static bool npt_enabled = true;
304 static bool npt_enabled;
308 * These 2 parameters are used to config the controls for Pause-Loop Exiting:
309 * pause_filter_count: On processors that support Pause filtering(indicated
310 * by CPUID Fn8000_000A_EDX), the VMCB provides a 16 bit pause filter
311 * count value. On VMRUN this value is loaded into an internal counter.
312 * Each time a pause instruction is executed, this counter is decremented
313 * until it reaches zero at which time a #VMEXIT is generated if pause
314 * intercept is enabled. Refer to AMD APM Vol 2 Section 15.14.4 Pause
315 * Intercept Filtering for more details.
316 * This also indicate if ple logic enabled.
318 * pause_filter_thresh: In addition, some processor families support advanced
319 * pause filtering (indicated by CPUID Fn8000_000A_EDX) upper bound on
320 * the amount of time a guest is allowed to execute in a pause loop.
321 * In this mode, a 16-bit pause filter threshold field is added in the
322 * VMCB. The threshold value is a cycle count that is used to reset the
323 * pause counter. As with simple pause filtering, VMRUN loads the pause
324 * count value from VMCB into an internal counter. Then, on each pause
325 * instruction the hardware checks the elapsed number of cycles since
326 * the most recent pause instruction against the pause filter threshold.
327 * If the elapsed cycle count is greater than the pause filter threshold,
328 * then the internal pause count is reloaded from the VMCB and execution
329 * continues. If the elapsed cycle count is less than the pause filter
330 * threshold, then the internal pause count is decremented. If the count
331 * value is less than zero and PAUSE intercept is enabled, a #VMEXIT is
332 * triggered. If advanced pause filtering is supported and pause filter
333 * threshold field is set to zero, the filter will operate in the simpler,
337 static unsigned short pause_filter_thresh = KVM_DEFAULT_PLE_GAP;
338 module_param(pause_filter_thresh, ushort, 0444);
340 static unsigned short pause_filter_count = KVM_SVM_DEFAULT_PLE_WINDOW;
341 module_param(pause_filter_count, ushort, 0444);
343 /* Default doubles per-vcpu window every exit. */
344 static unsigned short pause_filter_count_grow = KVM_DEFAULT_PLE_WINDOW_GROW;
345 module_param(pause_filter_count_grow, ushort, 0444);
347 /* Default resets per-vcpu window every exit to pause_filter_count. */
348 static unsigned short pause_filter_count_shrink = KVM_DEFAULT_PLE_WINDOW_SHRINK;
349 module_param(pause_filter_count_shrink, ushort, 0444);
351 /* Default is to compute the maximum so we can never overflow. */
352 static unsigned short pause_filter_count_max = KVM_SVM_DEFAULT_PLE_WINDOW_MAX;
353 module_param(pause_filter_count_max, ushort, 0444);
355 /* allow nested paging (virtualized MMU) for all guests */
356 static int npt = true;
357 module_param(npt, int, S_IRUGO);
359 /* allow nested virtualization in KVM/SVM */
360 static int nested = true;
361 module_param(nested, int, S_IRUGO);
363 /* enable / disable AVIC */
365 #ifdef CONFIG_X86_LOCAL_APIC
366 module_param(avic, int, S_IRUGO);
369 /* enable/disable Virtual VMLOAD VMSAVE */
370 static int vls = true;
371 module_param(vls, int, 0444);
373 /* enable/disable Virtual GIF */
374 static int vgif = true;
375 module_param(vgif, int, 0444);
377 /* enable/disable SEV support */
378 static int sev = IS_ENABLED(CONFIG_AMD_MEM_ENCRYPT_ACTIVE_BY_DEFAULT);
379 module_param(sev, int, 0444);
381 static u8 rsm_ins_bytes[] = "\x0f\xaa";
383 static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0);
384 static void svm_flush_tlb(struct kvm_vcpu *vcpu, bool invalidate_gpa);
385 static void svm_complete_interrupts(struct vcpu_svm *svm);
387 static int nested_svm_exit_handled(struct vcpu_svm *svm);
388 static int nested_svm_intercept(struct vcpu_svm *svm);
389 static int nested_svm_vmexit(struct vcpu_svm *svm);
390 static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
391 bool has_error_code, u32 error_code);
394 VMCB_INTERCEPTS, /* Intercept vectors, TSC offset,
395 pause filter count */
396 VMCB_PERM_MAP, /* IOPM Base and MSRPM Base */
397 VMCB_ASID, /* ASID */
398 VMCB_INTR, /* int_ctl, int_vector */
399 VMCB_NPT, /* npt_en, nCR3, gPAT */
400 VMCB_CR, /* CR0, CR3, CR4, EFER */
401 VMCB_DR, /* DR6, DR7 */
402 VMCB_DT, /* GDT, IDT */
403 VMCB_SEG, /* CS, DS, SS, ES, CPL */
404 VMCB_CR2, /* CR2 only */
405 VMCB_LBR, /* DBGCTL, BR_FROM, BR_TO, LAST_EX_FROM, LAST_EX_TO */
406 VMCB_AVIC, /* AVIC APIC_BAR, AVIC APIC_BACKING_PAGE,
407 * AVIC PHYSICAL_TABLE pointer,
408 * AVIC LOGICAL_TABLE pointer
413 /* TPR and CR2 are always written before VMRUN */
414 #define VMCB_ALWAYS_DIRTY_MASK ((1U << VMCB_INTR) | (1U << VMCB_CR2))
416 #define VMCB_AVIC_APIC_BAR_MASK 0xFFFFFFFFFF000ULL
418 static unsigned int max_sev_asid;
419 static unsigned int min_sev_asid;
420 static unsigned long *sev_asid_bitmap;
421 #define __sme_page_pa(x) __sme_set(page_to_pfn(x) << PAGE_SHIFT)
424 struct list_head list;
425 unsigned long npages;
432 static inline struct kvm_svm *to_kvm_svm(struct kvm *kvm)
434 return container_of(kvm, struct kvm_svm, kvm);
437 static inline bool svm_sev_enabled(void)
439 return IS_ENABLED(CONFIG_KVM_AMD_SEV) ? max_sev_asid : 0;
442 static inline bool sev_guest(struct kvm *kvm)
444 #ifdef CONFIG_KVM_AMD_SEV
445 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
453 static inline int sev_get_asid(struct kvm *kvm)
455 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
460 static inline void mark_all_dirty(struct vmcb *vmcb)
462 vmcb->control.clean = 0;
465 static inline void mark_all_clean(struct vmcb *vmcb)
467 vmcb->control.clean = ((1 << VMCB_DIRTY_MAX) - 1)
468 & ~VMCB_ALWAYS_DIRTY_MASK;
471 static inline void mark_dirty(struct vmcb *vmcb, int bit)
473 vmcb->control.clean &= ~(1 << bit);
476 static inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu)
478 return container_of(vcpu, struct vcpu_svm, vcpu);
481 static inline void avic_update_vapic_bar(struct vcpu_svm *svm, u64 data)
483 svm->vmcb->control.avic_vapic_bar = data & VMCB_AVIC_APIC_BAR_MASK;
484 mark_dirty(svm->vmcb, VMCB_AVIC);
487 static inline bool avic_vcpu_is_running(struct kvm_vcpu *vcpu)
489 struct vcpu_svm *svm = to_svm(vcpu);
490 u64 *entry = svm->avic_physical_id_cache;
495 return (READ_ONCE(*entry) & AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK);
498 static void recalc_intercepts(struct vcpu_svm *svm)
500 struct vmcb_control_area *c, *h;
501 struct nested_state *g;
503 mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
505 if (!is_guest_mode(&svm->vcpu))
508 c = &svm->vmcb->control;
509 h = &svm->nested.hsave->control;
512 c->intercept_cr = h->intercept_cr | g->intercept_cr;
513 c->intercept_dr = h->intercept_dr | g->intercept_dr;
514 c->intercept_exceptions = h->intercept_exceptions | g->intercept_exceptions;
515 c->intercept = h->intercept | g->intercept;
518 static inline struct vmcb *get_host_vmcb(struct vcpu_svm *svm)
520 if (is_guest_mode(&svm->vcpu))
521 return svm->nested.hsave;
526 static inline void set_cr_intercept(struct vcpu_svm *svm, int bit)
528 struct vmcb *vmcb = get_host_vmcb(svm);
530 vmcb->control.intercept_cr |= (1U << bit);
532 recalc_intercepts(svm);
535 static inline void clr_cr_intercept(struct vcpu_svm *svm, int bit)
537 struct vmcb *vmcb = get_host_vmcb(svm);
539 vmcb->control.intercept_cr &= ~(1U << bit);
541 recalc_intercepts(svm);
544 static inline bool is_cr_intercept(struct vcpu_svm *svm, int bit)
546 struct vmcb *vmcb = get_host_vmcb(svm);
548 return vmcb->control.intercept_cr & (1U << bit);
551 static inline void set_dr_intercepts(struct vcpu_svm *svm)
553 struct vmcb *vmcb = get_host_vmcb(svm);
555 vmcb->control.intercept_dr = (1 << INTERCEPT_DR0_READ)
556 | (1 << INTERCEPT_DR1_READ)
557 | (1 << INTERCEPT_DR2_READ)
558 | (1 << INTERCEPT_DR3_READ)
559 | (1 << INTERCEPT_DR4_READ)
560 | (1 << INTERCEPT_DR5_READ)
561 | (1 << INTERCEPT_DR6_READ)
562 | (1 << INTERCEPT_DR7_READ)
563 | (1 << INTERCEPT_DR0_WRITE)
564 | (1 << INTERCEPT_DR1_WRITE)
565 | (1 << INTERCEPT_DR2_WRITE)
566 | (1 << INTERCEPT_DR3_WRITE)
567 | (1 << INTERCEPT_DR4_WRITE)
568 | (1 << INTERCEPT_DR5_WRITE)
569 | (1 << INTERCEPT_DR6_WRITE)
570 | (1 << INTERCEPT_DR7_WRITE);
572 recalc_intercepts(svm);
575 static inline void clr_dr_intercepts(struct vcpu_svm *svm)
577 struct vmcb *vmcb = get_host_vmcb(svm);
579 vmcb->control.intercept_dr = 0;
581 recalc_intercepts(svm);
584 static inline void set_exception_intercept(struct vcpu_svm *svm, int bit)
586 struct vmcb *vmcb = get_host_vmcb(svm);
588 vmcb->control.intercept_exceptions |= (1U << bit);
590 recalc_intercepts(svm);
593 static inline void clr_exception_intercept(struct vcpu_svm *svm, int bit)
595 struct vmcb *vmcb = get_host_vmcb(svm);
597 vmcb->control.intercept_exceptions &= ~(1U << bit);
599 recalc_intercepts(svm);
602 static inline void set_intercept(struct vcpu_svm *svm, int bit)
604 struct vmcb *vmcb = get_host_vmcb(svm);
606 vmcb->control.intercept |= (1ULL << bit);
608 recalc_intercepts(svm);
611 static inline void clr_intercept(struct vcpu_svm *svm, int bit)
613 struct vmcb *vmcb = get_host_vmcb(svm);
615 vmcb->control.intercept &= ~(1ULL << bit);
617 recalc_intercepts(svm);
620 static inline bool vgif_enabled(struct vcpu_svm *svm)
622 return !!(svm->vmcb->control.int_ctl & V_GIF_ENABLE_MASK);
625 static inline void enable_gif(struct vcpu_svm *svm)
627 if (vgif_enabled(svm))
628 svm->vmcb->control.int_ctl |= V_GIF_MASK;
630 svm->vcpu.arch.hflags |= HF_GIF_MASK;
633 static inline void disable_gif(struct vcpu_svm *svm)
635 if (vgif_enabled(svm))
636 svm->vmcb->control.int_ctl &= ~V_GIF_MASK;
638 svm->vcpu.arch.hflags &= ~HF_GIF_MASK;
641 static inline bool gif_set(struct vcpu_svm *svm)
643 if (vgif_enabled(svm))
644 return !!(svm->vmcb->control.int_ctl & V_GIF_MASK);
646 return !!(svm->vcpu.arch.hflags & HF_GIF_MASK);
649 static unsigned long iopm_base;
651 struct kvm_ldttss_desc {
654 unsigned base1:8, type:5, dpl:2, p:1;
655 unsigned limit1:4, zero0:3, g:1, base2:8;
658 } __attribute__((packed));
660 struct svm_cpu_data {
667 struct kvm_ldttss_desc *tss_desc;
669 struct page *save_area;
670 struct vmcb *current_vmcb;
672 /* index = sev_asid, value = vmcb pointer */
673 struct vmcb **sev_vmcbs;
676 static DEFINE_PER_CPU(struct svm_cpu_data *, svm_data);
678 struct svm_init_data {
683 static const u32 msrpm_ranges[] = {0, 0xc0000000, 0xc0010000};
685 #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
686 #define MSRS_RANGE_SIZE 2048
687 #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
689 static u32 svm_msrpm_offset(u32 msr)
694 for (i = 0; i < NUM_MSR_MAPS; i++) {
695 if (msr < msrpm_ranges[i] ||
696 msr >= msrpm_ranges[i] + MSRS_IN_RANGE)
699 offset = (msr - msrpm_ranges[i]) / 4; /* 4 msrs per u8 */
700 offset += (i * MSRS_RANGE_SIZE); /* add range offset */
702 /* Now we have the u8 offset - but need the u32 offset */
706 /* MSR not in any range */
710 #define MAX_INST_SIZE 15
712 static inline void clgi(void)
714 asm volatile (__ex(SVM_CLGI));
717 static inline void stgi(void)
719 asm volatile (__ex(SVM_STGI));
722 static inline void invlpga(unsigned long addr, u32 asid)
724 asm volatile (__ex(SVM_INVLPGA) : : "a"(addr), "c"(asid));
727 static int get_npt_level(struct kvm_vcpu *vcpu)
730 return PT64_ROOT_4LEVEL;
732 return PT32E_ROOT_LEVEL;
736 static void svm_set_efer(struct kvm_vcpu *vcpu, u64 efer)
738 vcpu->arch.efer = efer;
739 if (!npt_enabled && !(efer & EFER_LMA))
742 to_svm(vcpu)->vmcb->save.efer = efer | EFER_SVME;
743 mark_dirty(to_svm(vcpu)->vmcb, VMCB_CR);
746 static int is_external_interrupt(u32 info)
748 info &= SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
749 return info == (SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR);
752 static u32 svm_get_interrupt_shadow(struct kvm_vcpu *vcpu)
754 struct vcpu_svm *svm = to_svm(vcpu);
757 if (svm->vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK)
758 ret = KVM_X86_SHADOW_INT_STI | KVM_X86_SHADOW_INT_MOV_SS;
762 static void svm_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
764 struct vcpu_svm *svm = to_svm(vcpu);
767 svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK;
769 svm->vmcb->control.int_state |= SVM_INTERRUPT_SHADOW_MASK;
773 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
775 struct vcpu_svm *svm = to_svm(vcpu);
777 if (svm->vmcb->control.next_rip != 0) {
778 WARN_ON_ONCE(!static_cpu_has(X86_FEATURE_NRIPS));
779 svm->next_rip = svm->vmcb->control.next_rip;
782 if (!svm->next_rip) {
783 if (kvm_emulate_instruction(vcpu, EMULTYPE_SKIP) !=
785 printk(KERN_DEBUG "%s: NOP\n", __func__);
788 if (svm->next_rip - kvm_rip_read(vcpu) > MAX_INST_SIZE)
789 printk(KERN_ERR "%s: ip 0x%lx next 0x%llx\n",
790 __func__, kvm_rip_read(vcpu), svm->next_rip);
792 kvm_rip_write(vcpu, svm->next_rip);
793 svm_set_interrupt_shadow(vcpu, 0);
796 static void svm_queue_exception(struct kvm_vcpu *vcpu)
798 struct vcpu_svm *svm = to_svm(vcpu);
799 unsigned nr = vcpu->arch.exception.nr;
800 bool has_error_code = vcpu->arch.exception.has_error_code;
801 bool reinject = vcpu->arch.exception.injected;
802 u32 error_code = vcpu->arch.exception.error_code;
805 * If we are within a nested VM we'd better #VMEXIT and let the guest
806 * handle the exception
809 nested_svm_check_exception(svm, nr, has_error_code, error_code))
812 kvm_deliver_exception_payload(&svm->vcpu);
814 if (nr == BP_VECTOR && !static_cpu_has(X86_FEATURE_NRIPS)) {
815 unsigned long rip, old_rip = kvm_rip_read(&svm->vcpu);
818 * For guest debugging where we have to reinject #BP if some
819 * INT3 is guest-owned:
820 * Emulate nRIP by moving RIP forward. Will fail if injection
821 * raises a fault that is not intercepted. Still better than
822 * failing in all cases.
824 skip_emulated_instruction(&svm->vcpu);
825 rip = kvm_rip_read(&svm->vcpu);
826 svm->int3_rip = rip + svm->vmcb->save.cs.base;
827 svm->int3_injected = rip - old_rip;
830 svm->vmcb->control.event_inj = nr
832 | (has_error_code ? SVM_EVTINJ_VALID_ERR : 0)
833 | SVM_EVTINJ_TYPE_EXEPT;
834 svm->vmcb->control.event_inj_err = error_code;
837 static void svm_init_erratum_383(void)
843 if (!static_cpu_has_bug(X86_BUG_AMD_TLB_MMATCH))
846 /* Use _safe variants to not break nested virtualization */
847 val = native_read_msr_safe(MSR_AMD64_DC_CFG, &err);
853 low = lower_32_bits(val);
854 high = upper_32_bits(val);
856 native_write_msr_safe(MSR_AMD64_DC_CFG, low, high);
858 erratum_383_found = true;
861 static void svm_init_osvw(struct kvm_vcpu *vcpu)
864 * Guests should see errata 400 and 415 as fixed (assuming that
865 * HLT and IO instructions are intercepted).
867 vcpu->arch.osvw.length = (osvw_len >= 3) ? (osvw_len) : 3;
868 vcpu->arch.osvw.status = osvw_status & ~(6ULL);
871 * By increasing VCPU's osvw.length to 3 we are telling the guest that
872 * all osvw.status bits inside that length, including bit 0 (which is
873 * reserved for erratum 298), are valid. However, if host processor's
874 * osvw_len is 0 then osvw_status[0] carries no information. We need to
875 * be conservative here and therefore we tell the guest that erratum 298
876 * is present (because we really don't know).
878 if (osvw_len == 0 && boot_cpu_data.x86 == 0x10)
879 vcpu->arch.osvw.status |= 1;
882 static int has_svm(void)
886 if (!cpu_has_svm(&msg)) {
887 printk(KERN_INFO "has_svm: %s\n", msg);
894 static void svm_hardware_disable(void)
896 /* Make sure we clean up behind us */
897 if (static_cpu_has(X86_FEATURE_TSCRATEMSR))
898 wrmsrl(MSR_AMD64_TSC_RATIO, TSC_RATIO_DEFAULT);
902 amd_pmu_disable_virt();
905 static int svm_hardware_enable(void)
908 struct svm_cpu_data *sd;
910 struct desc_struct *gdt;
911 int me = raw_smp_processor_id();
913 rdmsrl(MSR_EFER, efer);
914 if (efer & EFER_SVME)
918 pr_err("%s: err EOPNOTSUPP on %d\n", __func__, me);
921 sd = per_cpu(svm_data, me);
923 pr_err("%s: svm_data is NULL on %d\n", __func__, me);
927 sd->asid_generation = 1;
928 sd->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1;
929 sd->next_asid = sd->max_asid + 1;
930 sd->min_asid = max_sev_asid + 1;
932 gdt = get_current_gdt_rw();
933 sd->tss_desc = (struct kvm_ldttss_desc *)(gdt + GDT_ENTRY_TSS);
935 wrmsrl(MSR_EFER, efer | EFER_SVME);
937 wrmsrl(MSR_VM_HSAVE_PA, page_to_pfn(sd->save_area) << PAGE_SHIFT);
939 if (static_cpu_has(X86_FEATURE_TSCRATEMSR)) {
940 wrmsrl(MSR_AMD64_TSC_RATIO, TSC_RATIO_DEFAULT);
941 __this_cpu_write(current_tsc_ratio, TSC_RATIO_DEFAULT);
948 * Note that it is possible to have a system with mixed processor
949 * revisions and therefore different OSVW bits. If bits are not the same
950 * on different processors then choose the worst case (i.e. if erratum
951 * is present on one processor and not on another then assume that the
952 * erratum is present everywhere).
954 if (cpu_has(&boot_cpu_data, X86_FEATURE_OSVW)) {
955 uint64_t len, status = 0;
958 len = native_read_msr_safe(MSR_AMD64_OSVW_ID_LENGTH, &err);
960 status = native_read_msr_safe(MSR_AMD64_OSVW_STATUS,
964 osvw_status = osvw_len = 0;
968 osvw_status |= status;
969 osvw_status &= (1ULL << osvw_len) - 1;
972 osvw_status = osvw_len = 0;
974 svm_init_erratum_383();
976 amd_pmu_enable_virt();
981 static void svm_cpu_uninit(int cpu)
983 struct svm_cpu_data *sd = per_cpu(svm_data, raw_smp_processor_id());
988 per_cpu(svm_data, raw_smp_processor_id()) = NULL;
989 kfree(sd->sev_vmcbs);
990 __free_page(sd->save_area);
994 static int svm_cpu_init(int cpu)
996 struct svm_cpu_data *sd;
999 sd = kzalloc(sizeof(struct svm_cpu_data), GFP_KERNEL);
1004 sd->save_area = alloc_page(GFP_KERNEL);
1008 if (svm_sev_enabled()) {
1010 sd->sev_vmcbs = kmalloc_array(max_sev_asid + 1,
1017 per_cpu(svm_data, cpu) = sd;
1027 static bool valid_msr_intercept(u32 index)
1031 for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++)
1032 if (direct_access_msrs[i].index == index)
1038 static bool msr_write_intercepted(struct kvm_vcpu *vcpu, unsigned msr)
1045 msrpm = is_guest_mode(vcpu) ? to_svm(vcpu)->nested.msrpm:
1046 to_svm(vcpu)->msrpm;
1048 offset = svm_msrpm_offset(msr);
1049 bit_write = 2 * (msr & 0x0f) + 1;
1050 tmp = msrpm[offset];
1052 BUG_ON(offset == MSR_INVALID);
1054 return !!test_bit(bit_write, &tmp);
1057 static void set_msr_interception(u32 *msrpm, unsigned msr,
1058 int read, int write)
1060 u8 bit_read, bit_write;
1065 * If this warning triggers extend the direct_access_msrs list at the
1066 * beginning of the file
1068 WARN_ON(!valid_msr_intercept(msr));
1070 offset = svm_msrpm_offset(msr);
1071 bit_read = 2 * (msr & 0x0f);
1072 bit_write = 2 * (msr & 0x0f) + 1;
1073 tmp = msrpm[offset];
1075 BUG_ON(offset == MSR_INVALID);
1077 read ? clear_bit(bit_read, &tmp) : set_bit(bit_read, &tmp);
1078 write ? clear_bit(bit_write, &tmp) : set_bit(bit_write, &tmp);
1080 msrpm[offset] = tmp;
1083 static void svm_vcpu_init_msrpm(u32 *msrpm)
1087 memset(msrpm, 0xff, PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER));
1089 for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
1090 if (!direct_access_msrs[i].always)
1093 set_msr_interception(msrpm, direct_access_msrs[i].index, 1, 1);
1097 static void add_msr_offset(u32 offset)
1101 for (i = 0; i < MSRPM_OFFSETS; ++i) {
1103 /* Offset already in list? */
1104 if (msrpm_offsets[i] == offset)
1107 /* Slot used by another offset? */
1108 if (msrpm_offsets[i] != MSR_INVALID)
1111 /* Add offset to list */
1112 msrpm_offsets[i] = offset;
1118 * If this BUG triggers the msrpm_offsets table has an overflow. Just
1119 * increase MSRPM_OFFSETS in this case.
1124 static void init_msrpm_offsets(void)
1128 memset(msrpm_offsets, 0xff, sizeof(msrpm_offsets));
1130 for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
1133 offset = svm_msrpm_offset(direct_access_msrs[i].index);
1134 BUG_ON(offset == MSR_INVALID);
1136 add_msr_offset(offset);
1140 static void svm_enable_lbrv(struct vcpu_svm *svm)
1142 u32 *msrpm = svm->msrpm;
1144 svm->vmcb->control.virt_ext |= LBR_CTL_ENABLE_MASK;
1145 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 1, 1);
1146 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 1, 1);
1147 set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 1, 1);
1148 set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 1, 1);
1151 static void svm_disable_lbrv(struct vcpu_svm *svm)
1153 u32 *msrpm = svm->msrpm;
1155 svm->vmcb->control.virt_ext &= ~LBR_CTL_ENABLE_MASK;
1156 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 0, 0);
1157 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 0, 0);
1158 set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 0, 0);
1159 set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 0, 0);
1162 static void disable_nmi_singlestep(struct vcpu_svm *svm)
1164 svm->nmi_singlestep = false;
1166 if (!(svm->vcpu.guest_debug & KVM_GUESTDBG_SINGLESTEP)) {
1167 /* Clear our flags if they were not set by the guest */
1168 if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_TF))
1169 svm->vmcb->save.rflags &= ~X86_EFLAGS_TF;
1170 if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_RF))
1171 svm->vmcb->save.rflags &= ~X86_EFLAGS_RF;
1176 * This hash table is used to map VM_ID to a struct kvm_svm,
1177 * when handling AMD IOMMU GALOG notification to schedule in
1178 * a particular vCPU.
1180 #define SVM_VM_DATA_HASH_BITS 8
1181 static DEFINE_HASHTABLE(svm_vm_data_hash, SVM_VM_DATA_HASH_BITS);
1182 static u32 next_vm_id = 0;
1183 static bool next_vm_id_wrapped = 0;
1184 static DEFINE_SPINLOCK(svm_vm_data_hash_lock);
1187 * This function is called from IOMMU driver to notify
1188 * SVM to schedule in a particular vCPU of a particular VM.
1190 static int avic_ga_log_notifier(u32 ga_tag)
1192 unsigned long flags;
1193 struct kvm_svm *kvm_svm;
1194 struct kvm_vcpu *vcpu = NULL;
1195 u32 vm_id = AVIC_GATAG_TO_VMID(ga_tag);
1196 u32 vcpu_id = AVIC_GATAG_TO_VCPUID(ga_tag);
1198 pr_debug("SVM: %s: vm_id=%#x, vcpu_id=%#x\n", __func__, vm_id, vcpu_id);
1200 spin_lock_irqsave(&svm_vm_data_hash_lock, flags);
1201 hash_for_each_possible(svm_vm_data_hash, kvm_svm, hnode, vm_id) {
1202 if (kvm_svm->avic_vm_id != vm_id)
1204 vcpu = kvm_get_vcpu_by_id(&kvm_svm->kvm, vcpu_id);
1207 spin_unlock_irqrestore(&svm_vm_data_hash_lock, flags);
1210 * At this point, the IOMMU should have already set the pending
1211 * bit in the vAPIC backing page. So, we just need to schedule
1215 kvm_vcpu_wake_up(vcpu);
1220 static __init int sev_hardware_setup(void)
1222 struct sev_user_data_status *status;
1225 /* Maximum number of encrypted guests supported simultaneously */
1226 max_sev_asid = cpuid_ecx(0x8000001F);
1231 /* Minimum ASID value that should be used for SEV guest */
1232 min_sev_asid = cpuid_edx(0x8000001F);
1234 /* Initialize SEV ASID bitmap */
1235 sev_asid_bitmap = bitmap_zalloc(max_sev_asid, GFP_KERNEL);
1236 if (!sev_asid_bitmap)
1239 status = kmalloc(sizeof(*status), GFP_KERNEL);
1244 * Check SEV platform status.
1246 * PLATFORM_STATUS can be called in any state, if we failed to query
1247 * the PLATFORM status then either PSP firmware does not support SEV
1248 * feature or SEV firmware is dead.
1250 rc = sev_platform_status(status, NULL);
1254 pr_info("SEV supported\n");
1261 static void grow_ple_window(struct kvm_vcpu *vcpu)
1263 struct vcpu_svm *svm = to_svm(vcpu);
1264 struct vmcb_control_area *control = &svm->vmcb->control;
1265 int old = control->pause_filter_count;
1267 control->pause_filter_count = __grow_ple_window(old,
1269 pause_filter_count_grow,
1270 pause_filter_count_max);
1272 if (control->pause_filter_count != old)
1273 mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
1275 trace_kvm_ple_window_grow(vcpu->vcpu_id,
1276 control->pause_filter_count, old);
1279 static void shrink_ple_window(struct kvm_vcpu *vcpu)
1281 struct vcpu_svm *svm = to_svm(vcpu);
1282 struct vmcb_control_area *control = &svm->vmcb->control;
1283 int old = control->pause_filter_count;
1285 control->pause_filter_count =
1286 __shrink_ple_window(old,
1288 pause_filter_count_shrink,
1289 pause_filter_count);
1290 if (control->pause_filter_count != old)
1291 mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
1293 trace_kvm_ple_window_shrink(vcpu->vcpu_id,
1294 control->pause_filter_count, old);
1297 static __init int svm_hardware_setup(void)
1300 struct page *iopm_pages;
1304 iopm_pages = alloc_pages(GFP_KERNEL, IOPM_ALLOC_ORDER);
1309 iopm_va = page_address(iopm_pages);
1310 memset(iopm_va, 0xff, PAGE_SIZE * (1 << IOPM_ALLOC_ORDER));
1311 iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT;
1313 init_msrpm_offsets();
1315 if (boot_cpu_has(X86_FEATURE_NX))
1316 kvm_enable_efer_bits(EFER_NX);
1318 if (boot_cpu_has(X86_FEATURE_FXSR_OPT))
1319 kvm_enable_efer_bits(EFER_FFXSR);
1321 if (boot_cpu_has(X86_FEATURE_TSCRATEMSR)) {
1322 kvm_has_tsc_control = true;
1323 kvm_max_tsc_scaling_ratio = TSC_RATIO_MAX;
1324 kvm_tsc_scaling_ratio_frac_bits = 32;
1327 /* Check for pause filtering support */
1328 if (!boot_cpu_has(X86_FEATURE_PAUSEFILTER)) {
1329 pause_filter_count = 0;
1330 pause_filter_thresh = 0;
1331 } else if (!boot_cpu_has(X86_FEATURE_PFTHRESHOLD)) {
1332 pause_filter_thresh = 0;
1336 printk(KERN_INFO "kvm: Nested Virtualization enabled\n");
1337 kvm_enable_efer_bits(EFER_SVME | EFER_LMSLE);
1341 if (boot_cpu_has(X86_FEATURE_SEV) &&
1342 IS_ENABLED(CONFIG_KVM_AMD_SEV)) {
1343 r = sev_hardware_setup();
1351 for_each_possible_cpu(cpu) {
1352 r = svm_cpu_init(cpu);
1357 if (!boot_cpu_has(X86_FEATURE_NPT))
1358 npt_enabled = false;
1360 if (npt_enabled && !npt) {
1361 printk(KERN_INFO "kvm: Nested Paging disabled\n");
1362 npt_enabled = false;
1366 printk(KERN_INFO "kvm: Nested Paging enabled\n");
1373 !boot_cpu_has(X86_FEATURE_AVIC) ||
1374 !IS_ENABLED(CONFIG_X86_LOCAL_APIC)) {
1377 pr_info("AVIC enabled\n");
1379 amd_iommu_register_ga_log_notifier(&avic_ga_log_notifier);
1385 !boot_cpu_has(X86_FEATURE_V_VMSAVE_VMLOAD) ||
1386 !IS_ENABLED(CONFIG_X86_64)) {
1389 pr_info("Virtual VMLOAD VMSAVE supported\n");
1394 if (!boot_cpu_has(X86_FEATURE_VGIF))
1397 pr_info("Virtual GIF supported\n");
1403 __free_pages(iopm_pages, IOPM_ALLOC_ORDER);
1408 static __exit void svm_hardware_unsetup(void)
1412 if (svm_sev_enabled())
1413 bitmap_free(sev_asid_bitmap);
1415 for_each_possible_cpu(cpu)
1416 svm_cpu_uninit(cpu);
1418 __free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT), IOPM_ALLOC_ORDER);
1422 static void init_seg(struct vmcb_seg *seg)
1425 seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK |
1426 SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */
1427 seg->limit = 0xffff;
1431 static void init_sys_seg(struct vmcb_seg *seg, uint32_t type)
1434 seg->attrib = SVM_SELECTOR_P_MASK | type;
1435 seg->limit = 0xffff;
1439 static u64 svm_read_l1_tsc_offset(struct kvm_vcpu *vcpu)
1441 struct vcpu_svm *svm = to_svm(vcpu);
1443 if (is_guest_mode(vcpu))
1444 return svm->nested.hsave->control.tsc_offset;
1446 return vcpu->arch.tsc_offset;
1449 static u64 svm_write_l1_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1451 struct vcpu_svm *svm = to_svm(vcpu);
1452 u64 g_tsc_offset = 0;
1454 if (is_guest_mode(vcpu)) {
1455 /* Write L1's TSC offset. */
1456 g_tsc_offset = svm->vmcb->control.tsc_offset -
1457 svm->nested.hsave->control.tsc_offset;
1458 svm->nested.hsave->control.tsc_offset = offset;
1461 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
1462 svm->vmcb->control.tsc_offset - g_tsc_offset,
1465 svm->vmcb->control.tsc_offset = offset + g_tsc_offset;
1467 mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
1468 return svm->vmcb->control.tsc_offset;
1471 static void avic_init_vmcb(struct vcpu_svm *svm)
1473 struct vmcb *vmcb = svm->vmcb;
1474 struct kvm_svm *kvm_svm = to_kvm_svm(svm->vcpu.kvm);
1475 phys_addr_t bpa = __sme_set(page_to_phys(svm->avic_backing_page));
1476 phys_addr_t lpa = __sme_set(page_to_phys(kvm_svm->avic_logical_id_table_page));
1477 phys_addr_t ppa = __sme_set(page_to_phys(kvm_svm->avic_physical_id_table_page));
1479 vmcb->control.avic_backing_page = bpa & AVIC_HPA_MASK;
1480 vmcb->control.avic_logical_id = lpa & AVIC_HPA_MASK;
1481 vmcb->control.avic_physical_id = ppa & AVIC_HPA_MASK;
1482 vmcb->control.avic_physical_id |= AVIC_MAX_PHYSICAL_ID_COUNT;
1483 vmcb->control.int_ctl |= AVIC_ENABLE_MASK;
1486 static void init_vmcb(struct vcpu_svm *svm)
1488 struct vmcb_control_area *control = &svm->vmcb->control;
1489 struct vmcb_save_area *save = &svm->vmcb->save;
1491 svm->vcpu.arch.hflags = 0;
1493 set_cr_intercept(svm, INTERCEPT_CR0_READ);
1494 set_cr_intercept(svm, INTERCEPT_CR3_READ);
1495 set_cr_intercept(svm, INTERCEPT_CR4_READ);
1496 set_cr_intercept(svm, INTERCEPT_CR0_WRITE);
1497 set_cr_intercept(svm, INTERCEPT_CR3_WRITE);
1498 set_cr_intercept(svm, INTERCEPT_CR4_WRITE);
1499 if (!kvm_vcpu_apicv_active(&svm->vcpu))
1500 set_cr_intercept(svm, INTERCEPT_CR8_WRITE);
1502 set_dr_intercepts(svm);
1504 set_exception_intercept(svm, PF_VECTOR);
1505 set_exception_intercept(svm, UD_VECTOR);
1506 set_exception_intercept(svm, MC_VECTOR);
1507 set_exception_intercept(svm, AC_VECTOR);
1508 set_exception_intercept(svm, DB_VECTOR);
1510 * Guest access to VMware backdoor ports could legitimately
1511 * trigger #GP because of TSS I/O permission bitmap.
1512 * We intercept those #GP and allow access to them anyway
1515 if (enable_vmware_backdoor)
1516 set_exception_intercept(svm, GP_VECTOR);
1518 set_intercept(svm, INTERCEPT_INTR);
1519 set_intercept(svm, INTERCEPT_NMI);
1520 set_intercept(svm, INTERCEPT_SMI);
1521 set_intercept(svm, INTERCEPT_SELECTIVE_CR0);
1522 set_intercept(svm, INTERCEPT_RDPMC);
1523 set_intercept(svm, INTERCEPT_CPUID);
1524 set_intercept(svm, INTERCEPT_INVD);
1525 set_intercept(svm, INTERCEPT_INVLPG);
1526 set_intercept(svm, INTERCEPT_INVLPGA);
1527 set_intercept(svm, INTERCEPT_IOIO_PROT);
1528 set_intercept(svm, INTERCEPT_MSR_PROT);
1529 set_intercept(svm, INTERCEPT_TASK_SWITCH);
1530 set_intercept(svm, INTERCEPT_SHUTDOWN);
1531 set_intercept(svm, INTERCEPT_VMRUN);
1532 set_intercept(svm, INTERCEPT_VMMCALL);
1533 set_intercept(svm, INTERCEPT_VMLOAD);
1534 set_intercept(svm, INTERCEPT_VMSAVE);
1535 set_intercept(svm, INTERCEPT_STGI);
1536 set_intercept(svm, INTERCEPT_CLGI);
1537 set_intercept(svm, INTERCEPT_SKINIT);
1538 set_intercept(svm, INTERCEPT_WBINVD);
1539 set_intercept(svm, INTERCEPT_XSETBV);
1540 set_intercept(svm, INTERCEPT_RSM);
1542 if (!kvm_mwait_in_guest(svm->vcpu.kvm)) {
1543 set_intercept(svm, INTERCEPT_MONITOR);
1544 set_intercept(svm, INTERCEPT_MWAIT);
1547 if (!kvm_hlt_in_guest(svm->vcpu.kvm))
1548 set_intercept(svm, INTERCEPT_HLT);
1550 control->iopm_base_pa = __sme_set(iopm_base);
1551 control->msrpm_base_pa = __sme_set(__pa(svm->msrpm));
1552 control->int_ctl = V_INTR_MASKING_MASK;
1554 init_seg(&save->es);
1555 init_seg(&save->ss);
1556 init_seg(&save->ds);
1557 init_seg(&save->fs);
1558 init_seg(&save->gs);
1560 save->cs.selector = 0xf000;
1561 save->cs.base = 0xffff0000;
1562 /* Executable/Readable Code Segment */
1563 save->cs.attrib = SVM_SELECTOR_READ_MASK | SVM_SELECTOR_P_MASK |
1564 SVM_SELECTOR_S_MASK | SVM_SELECTOR_CODE_MASK;
1565 save->cs.limit = 0xffff;
1567 save->gdtr.limit = 0xffff;
1568 save->idtr.limit = 0xffff;
1570 init_sys_seg(&save->ldtr, SEG_TYPE_LDT);
1571 init_sys_seg(&save->tr, SEG_TYPE_BUSY_TSS16);
1573 svm_set_efer(&svm->vcpu, 0);
1574 save->dr6 = 0xffff0ff0;
1575 kvm_set_rflags(&svm->vcpu, 2);
1576 save->rip = 0x0000fff0;
1577 svm->vcpu.arch.regs[VCPU_REGS_RIP] = save->rip;
1580 * svm_set_cr0() sets PG and WP and clears NW and CD on save->cr0.
1581 * It also updates the guest-visible cr0 value.
1583 svm_set_cr0(&svm->vcpu, X86_CR0_NW | X86_CR0_CD | X86_CR0_ET);
1584 kvm_mmu_reset_context(&svm->vcpu);
1586 save->cr4 = X86_CR4_PAE;
1590 /* Setup VMCB for Nested Paging */
1591 control->nested_ctl |= SVM_NESTED_CTL_NP_ENABLE;
1592 clr_intercept(svm, INTERCEPT_INVLPG);
1593 clr_exception_intercept(svm, PF_VECTOR);
1594 clr_cr_intercept(svm, INTERCEPT_CR3_READ);
1595 clr_cr_intercept(svm, INTERCEPT_CR3_WRITE);
1596 save->g_pat = svm->vcpu.arch.pat;
1600 svm->asid_generation = 0;
1602 svm->nested.vmcb = 0;
1603 svm->vcpu.arch.hflags = 0;
1605 if (pause_filter_count) {
1606 control->pause_filter_count = pause_filter_count;
1607 if (pause_filter_thresh)
1608 control->pause_filter_thresh = pause_filter_thresh;
1609 set_intercept(svm, INTERCEPT_PAUSE);
1611 clr_intercept(svm, INTERCEPT_PAUSE);
1614 if (kvm_vcpu_apicv_active(&svm->vcpu))
1615 avic_init_vmcb(svm);
1618 * If hardware supports Virtual VMLOAD VMSAVE then enable it
1619 * in VMCB and clear intercepts to avoid #VMEXIT.
1622 clr_intercept(svm, INTERCEPT_VMLOAD);
1623 clr_intercept(svm, INTERCEPT_VMSAVE);
1624 svm->vmcb->control.virt_ext |= VIRTUAL_VMLOAD_VMSAVE_ENABLE_MASK;
1628 clr_intercept(svm, INTERCEPT_STGI);
1629 clr_intercept(svm, INTERCEPT_CLGI);
1630 svm->vmcb->control.int_ctl |= V_GIF_ENABLE_MASK;
1633 if (sev_guest(svm->vcpu.kvm)) {
1634 svm->vmcb->control.nested_ctl |= SVM_NESTED_CTL_SEV_ENABLE;
1635 clr_exception_intercept(svm, UD_VECTOR);
1638 mark_all_dirty(svm->vmcb);
1644 static u64 *avic_get_physical_id_entry(struct kvm_vcpu *vcpu,
1647 u64 *avic_physical_id_table;
1648 struct kvm_svm *kvm_svm = to_kvm_svm(vcpu->kvm);
1650 if (index >= AVIC_MAX_PHYSICAL_ID_COUNT)
1653 avic_physical_id_table = page_address(kvm_svm->avic_physical_id_table_page);
1655 return &avic_physical_id_table[index];
1660 * AVIC hardware walks the nested page table to check permissions,
1661 * but does not use the SPA address specified in the leaf page
1662 * table entry since it uses address in the AVIC_BACKING_PAGE pointer
1663 * field of the VMCB. Therefore, we set up the
1664 * APIC_ACCESS_PAGE_PRIVATE_MEMSLOT (4KB) here.
1666 static int avic_init_access_page(struct kvm_vcpu *vcpu)
1668 struct kvm *kvm = vcpu->kvm;
1671 mutex_lock(&kvm->slots_lock);
1672 if (kvm->arch.apic_access_page_done)
1675 ret = __x86_set_memory_region(kvm,
1676 APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
1677 APIC_DEFAULT_PHYS_BASE,
1682 kvm->arch.apic_access_page_done = true;
1684 mutex_unlock(&kvm->slots_lock);
1688 static int avic_init_backing_page(struct kvm_vcpu *vcpu)
1691 u64 *entry, new_entry;
1692 int id = vcpu->vcpu_id;
1693 struct vcpu_svm *svm = to_svm(vcpu);
1695 ret = avic_init_access_page(vcpu);
1699 if (id >= AVIC_MAX_PHYSICAL_ID_COUNT)
1702 if (!svm->vcpu.arch.apic->regs)
1705 svm->avic_backing_page = virt_to_page(svm->vcpu.arch.apic->regs);
1707 /* Setting AVIC backing page address in the phy APIC ID table */
1708 entry = avic_get_physical_id_entry(vcpu, id);
1712 new_entry = READ_ONCE(*entry);
1713 new_entry = __sme_set((page_to_phys(svm->avic_backing_page) &
1714 AVIC_PHYSICAL_ID_ENTRY_BACKING_PAGE_MASK) |
1715 AVIC_PHYSICAL_ID_ENTRY_VALID_MASK);
1716 WRITE_ONCE(*entry, new_entry);
1718 svm->avic_physical_id_cache = entry;
1723 static void __sev_asid_free(int asid)
1725 struct svm_cpu_data *sd;
1729 clear_bit(pos, sev_asid_bitmap);
1731 for_each_possible_cpu(cpu) {
1732 sd = per_cpu(svm_data, cpu);
1733 sd->sev_vmcbs[pos] = NULL;
1737 static void sev_asid_free(struct kvm *kvm)
1739 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1741 __sev_asid_free(sev->asid);
1744 static void sev_unbind_asid(struct kvm *kvm, unsigned int handle)
1746 struct sev_data_decommission *decommission;
1747 struct sev_data_deactivate *data;
1752 data = kzalloc(sizeof(*data), GFP_KERNEL);
1756 /* deactivate handle */
1757 data->handle = handle;
1758 sev_guest_deactivate(data, NULL);
1760 wbinvd_on_all_cpus();
1761 sev_guest_df_flush(NULL);
1764 decommission = kzalloc(sizeof(*decommission), GFP_KERNEL);
1768 /* decommission handle */
1769 decommission->handle = handle;
1770 sev_guest_decommission(decommission, NULL);
1772 kfree(decommission);
1775 static struct page **sev_pin_memory(struct kvm *kvm, unsigned long uaddr,
1776 unsigned long ulen, unsigned long *n,
1779 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1780 unsigned long npages, npinned, size;
1781 unsigned long locked, lock_limit;
1782 struct page **pages;
1783 unsigned long first, last;
1785 if (ulen == 0 || uaddr + ulen < uaddr)
1788 /* Calculate number of pages. */
1789 first = (uaddr & PAGE_MASK) >> PAGE_SHIFT;
1790 last = ((uaddr + ulen - 1) & PAGE_MASK) >> PAGE_SHIFT;
1791 npages = (last - first + 1);
1793 locked = sev->pages_locked + npages;
1794 lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1795 if (locked > lock_limit && !capable(CAP_IPC_LOCK)) {
1796 pr_err("SEV: %lu locked pages exceed the lock limit of %lu.\n", locked, lock_limit);
1800 /* Avoid using vmalloc for smaller buffers. */
1801 size = npages * sizeof(struct page *);
1802 if (size > PAGE_SIZE)
1803 pages = vmalloc(size);
1805 pages = kmalloc(size, GFP_KERNEL);
1810 /* Pin the user virtual address. */
1811 npinned = get_user_pages_fast(uaddr, npages, write ? FOLL_WRITE : 0, 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 = vzalloc(sizeof(struct kvm_svm));
1874 return &kvm_svm->kvm;
1877 static void svm_vm_free(struct kvm *kvm)
1879 vfree(to_kvm_svm(kvm));
1882 static void sev_vm_destroy(struct kvm *kvm)
1884 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1885 struct list_head *head = &sev->regions_list;
1886 struct list_head *pos, *q;
1888 if (!sev_guest(kvm))
1891 mutex_lock(&kvm->lock);
1894 * if userspace was terminated before unregistering the memory regions
1895 * then lets unpin all the registered memory.
1897 if (!list_empty(head)) {
1898 list_for_each_safe(pos, q, head) {
1899 __unregister_enc_region_locked(kvm,
1900 list_entry(pos, struct enc_region, list));
1904 mutex_unlock(&kvm->lock);
1906 sev_unbind_asid(kvm, sev->handle);
1910 static void avic_vm_destroy(struct kvm *kvm)
1912 unsigned long flags;
1913 struct kvm_svm *kvm_svm = to_kvm_svm(kvm);
1918 if (kvm_svm->avic_logical_id_table_page)
1919 __free_page(kvm_svm->avic_logical_id_table_page);
1920 if (kvm_svm->avic_physical_id_table_page)
1921 __free_page(kvm_svm->avic_physical_id_table_page);
1923 spin_lock_irqsave(&svm_vm_data_hash_lock, flags);
1924 hash_del(&kvm_svm->hnode);
1925 spin_unlock_irqrestore(&svm_vm_data_hash_lock, flags);
1928 static void svm_vm_destroy(struct kvm *kvm)
1930 avic_vm_destroy(kvm);
1931 sev_vm_destroy(kvm);
1934 static int avic_vm_init(struct kvm *kvm)
1936 unsigned long flags;
1938 struct kvm_svm *kvm_svm = to_kvm_svm(kvm);
1940 struct page *p_page;
1941 struct page *l_page;
1947 /* Allocating physical APIC ID table (4KB) */
1948 p_page = alloc_page(GFP_KERNEL);
1952 kvm_svm->avic_physical_id_table_page = p_page;
1953 clear_page(page_address(p_page));
1955 /* Allocating logical APIC ID table (4KB) */
1956 l_page = alloc_page(GFP_KERNEL);
1960 kvm_svm->avic_logical_id_table_page = l_page;
1961 clear_page(page_address(l_page));
1963 spin_lock_irqsave(&svm_vm_data_hash_lock, flags);
1965 vm_id = next_vm_id = (next_vm_id + 1) & AVIC_VM_ID_MASK;
1966 if (vm_id == 0) { /* id is 1-based, zero is not okay */
1967 next_vm_id_wrapped = 1;
1970 /* Is it still in use? Only possible if wrapped at least once */
1971 if (next_vm_id_wrapped) {
1972 hash_for_each_possible(svm_vm_data_hash, k2, hnode, vm_id) {
1973 if (k2->avic_vm_id == vm_id)
1977 kvm_svm->avic_vm_id = vm_id;
1978 hash_add(svm_vm_data_hash, &kvm_svm->hnode, kvm_svm->avic_vm_id);
1979 spin_unlock_irqrestore(&svm_vm_data_hash_lock, flags);
1984 avic_vm_destroy(kvm);
1989 avic_update_iommu_vcpu_affinity(struct kvm_vcpu *vcpu, int cpu, bool r)
1992 unsigned long flags;
1993 struct amd_svm_iommu_ir *ir;
1994 struct vcpu_svm *svm = to_svm(vcpu);
1996 if (!kvm_arch_has_assigned_device(vcpu->kvm))
2000 * Here, we go through the per-vcpu ir_list to update all existing
2001 * interrupt remapping table entry targeting this vcpu.
2003 spin_lock_irqsave(&svm->ir_list_lock, flags);
2005 if (list_empty(&svm->ir_list))
2008 list_for_each_entry(ir, &svm->ir_list, node) {
2009 ret = amd_iommu_update_ga(cpu, r, ir->data);
2014 spin_unlock_irqrestore(&svm->ir_list_lock, flags);
2018 static void avic_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2021 /* ID = 0xff (broadcast), ID > 0xff (reserved) */
2022 int h_physical_id = kvm_cpu_get_apicid(cpu);
2023 struct vcpu_svm *svm = to_svm(vcpu);
2025 if (!kvm_vcpu_apicv_active(vcpu))
2028 if (WARN_ON(h_physical_id >= AVIC_MAX_PHYSICAL_ID_COUNT))
2031 entry = READ_ONCE(*(svm->avic_physical_id_cache));
2032 WARN_ON(entry & AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK);
2034 entry &= ~AVIC_PHYSICAL_ID_ENTRY_HOST_PHYSICAL_ID_MASK;
2035 entry |= (h_physical_id & AVIC_PHYSICAL_ID_ENTRY_HOST_PHYSICAL_ID_MASK);
2037 entry &= ~AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK;
2038 if (svm->avic_is_running)
2039 entry |= AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK;
2041 WRITE_ONCE(*(svm->avic_physical_id_cache), entry);
2042 avic_update_iommu_vcpu_affinity(vcpu, h_physical_id,
2043 svm->avic_is_running);
2046 static void avic_vcpu_put(struct kvm_vcpu *vcpu)
2049 struct vcpu_svm *svm = to_svm(vcpu);
2051 if (!kvm_vcpu_apicv_active(vcpu))
2054 entry = READ_ONCE(*(svm->avic_physical_id_cache));
2055 if (entry & AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK)
2056 avic_update_iommu_vcpu_affinity(vcpu, -1, 0);
2058 entry &= ~AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK;
2059 WRITE_ONCE(*(svm->avic_physical_id_cache), entry);
2063 * This function is called during VCPU halt/unhalt.
2065 static void avic_set_running(struct kvm_vcpu *vcpu, bool is_run)
2067 struct vcpu_svm *svm = to_svm(vcpu);
2069 svm->avic_is_running = is_run;
2071 avic_vcpu_load(vcpu, vcpu->cpu);
2073 avic_vcpu_put(vcpu);
2076 static void svm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
2078 struct vcpu_svm *svm = to_svm(vcpu);
2082 vcpu->arch.microcode_version = 0x01000065;
2084 svm->virt_spec_ctrl = 0;
2087 svm->vcpu.arch.apic_base = APIC_DEFAULT_PHYS_BASE |
2088 MSR_IA32_APICBASE_ENABLE;
2089 if (kvm_vcpu_is_reset_bsp(&svm->vcpu))
2090 svm->vcpu.arch.apic_base |= MSR_IA32_APICBASE_BSP;
2094 kvm_cpuid(vcpu, &eax, &dummy, &dummy, &dummy, true);
2095 kvm_register_write(vcpu, VCPU_REGS_RDX, eax);
2097 if (kvm_vcpu_apicv_active(vcpu) && !init_event)
2098 avic_update_vapic_bar(svm, APIC_DEFAULT_PHYS_BASE);
2101 static int avic_init_vcpu(struct vcpu_svm *svm)
2105 if (!kvm_vcpu_apicv_active(&svm->vcpu))
2108 ret = avic_init_backing_page(&svm->vcpu);
2112 INIT_LIST_HEAD(&svm->ir_list);
2113 spin_lock_init(&svm->ir_list_lock);
2118 static struct kvm_vcpu *svm_create_vcpu(struct kvm *kvm, unsigned int id)
2120 struct vcpu_svm *svm;
2122 struct page *msrpm_pages;
2123 struct page *hsave_page;
2124 struct page *nested_msrpm_pages;
2127 svm = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
2133 err = kvm_vcpu_init(&svm->vcpu, kvm, id);
2138 page = alloc_page(GFP_KERNEL);
2142 msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
2146 nested_msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
2147 if (!nested_msrpm_pages)
2150 hsave_page = alloc_page(GFP_KERNEL);
2154 err = avic_init_vcpu(svm);
2158 /* We initialize this flag to true to make sure that the is_running
2159 * bit would be set the first time the vcpu is loaded.
2161 svm->avic_is_running = true;
2163 svm->nested.hsave = page_address(hsave_page);
2165 svm->msrpm = page_address(msrpm_pages);
2166 svm_vcpu_init_msrpm(svm->msrpm);
2168 svm->nested.msrpm = page_address(nested_msrpm_pages);
2169 svm_vcpu_init_msrpm(svm->nested.msrpm);
2171 svm->vmcb = page_address(page);
2172 clear_page(svm->vmcb);
2173 svm->vmcb_pa = __sme_set(page_to_pfn(page) << PAGE_SHIFT);
2174 svm->asid_generation = 0;
2177 svm_init_osvw(&svm->vcpu);
2182 __free_page(hsave_page);
2184 __free_pages(nested_msrpm_pages, MSRPM_ALLOC_ORDER);
2186 __free_pages(msrpm_pages, MSRPM_ALLOC_ORDER);
2190 kvm_vcpu_uninit(&svm->vcpu);
2192 kmem_cache_free(kvm_vcpu_cache, svm);
2194 return ERR_PTR(err);
2197 static void svm_clear_current_vmcb(struct vmcb *vmcb)
2201 for_each_online_cpu(i)
2202 cmpxchg(&per_cpu(svm_data, i)->current_vmcb, vmcb, NULL);
2205 static void svm_free_vcpu(struct kvm_vcpu *vcpu)
2207 struct vcpu_svm *svm = to_svm(vcpu);
2210 * The vmcb page can be recycled, causing a false negative in
2211 * svm_vcpu_load(). So, ensure that no logical CPU has this
2212 * vmcb page recorded as its current vmcb.
2214 svm_clear_current_vmcb(svm->vmcb);
2216 __free_page(pfn_to_page(__sme_clr(svm->vmcb_pa) >> PAGE_SHIFT));
2217 __free_pages(virt_to_page(svm->msrpm), MSRPM_ALLOC_ORDER);
2218 __free_page(virt_to_page(svm->nested.hsave));
2219 __free_pages(virt_to_page(svm->nested.msrpm), MSRPM_ALLOC_ORDER);
2220 kvm_vcpu_uninit(vcpu);
2221 kmem_cache_free(kvm_vcpu_cache, svm);
2224 static void svm_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2226 struct vcpu_svm *svm = to_svm(vcpu);
2227 struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
2230 if (unlikely(cpu != vcpu->cpu)) {
2231 svm->asid_generation = 0;
2232 mark_all_dirty(svm->vmcb);
2235 #ifdef CONFIG_X86_64
2236 rdmsrl(MSR_GS_BASE, to_svm(vcpu)->host.gs_base);
2238 savesegment(fs, svm->host.fs);
2239 savesegment(gs, svm->host.gs);
2240 svm->host.ldt = kvm_read_ldt();
2242 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
2243 rdmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
2245 if (static_cpu_has(X86_FEATURE_TSCRATEMSR)) {
2246 u64 tsc_ratio = vcpu->arch.tsc_scaling_ratio;
2247 if (tsc_ratio != __this_cpu_read(current_tsc_ratio)) {
2248 __this_cpu_write(current_tsc_ratio, tsc_ratio);
2249 wrmsrl(MSR_AMD64_TSC_RATIO, tsc_ratio);
2252 /* This assumes that the kernel never uses MSR_TSC_AUX */
2253 if (static_cpu_has(X86_FEATURE_RDTSCP))
2254 wrmsrl(MSR_TSC_AUX, svm->tsc_aux);
2256 if (sd->current_vmcb != svm->vmcb) {
2257 sd->current_vmcb = svm->vmcb;
2258 indirect_branch_prediction_barrier();
2260 avic_vcpu_load(vcpu, cpu);
2263 static void svm_vcpu_put(struct kvm_vcpu *vcpu)
2265 struct vcpu_svm *svm = to_svm(vcpu);
2268 avic_vcpu_put(vcpu);
2270 ++vcpu->stat.host_state_reload;
2271 kvm_load_ldt(svm->host.ldt);
2272 #ifdef CONFIG_X86_64
2273 loadsegment(fs, svm->host.fs);
2274 wrmsrl(MSR_KERNEL_GS_BASE, current->thread.gsbase);
2275 load_gs_index(svm->host.gs);
2277 #ifdef CONFIG_X86_32_LAZY_GS
2278 loadsegment(gs, svm->host.gs);
2281 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
2282 wrmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
2285 static void svm_vcpu_blocking(struct kvm_vcpu *vcpu)
2287 avic_set_running(vcpu, false);
2290 static void svm_vcpu_unblocking(struct kvm_vcpu *vcpu)
2292 avic_set_running(vcpu, true);
2295 static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu)
2297 struct vcpu_svm *svm = to_svm(vcpu);
2298 unsigned long rflags = svm->vmcb->save.rflags;
2300 if (svm->nmi_singlestep) {
2301 /* Hide our flags if they were not set by the guest */
2302 if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_TF))
2303 rflags &= ~X86_EFLAGS_TF;
2304 if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_RF))
2305 rflags &= ~X86_EFLAGS_RF;
2310 static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
2312 if (to_svm(vcpu)->nmi_singlestep)
2313 rflags |= (X86_EFLAGS_TF | X86_EFLAGS_RF);
2316 * Any change of EFLAGS.VM is accompanied by a reload of SS
2317 * (caused by either a task switch or an inter-privilege IRET),
2318 * so we do not need to update the CPL here.
2320 to_svm(vcpu)->vmcb->save.rflags = rflags;
2323 static void svm_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
2326 case VCPU_EXREG_PDPTR:
2327 BUG_ON(!npt_enabled);
2328 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
2335 static void svm_set_vintr(struct vcpu_svm *svm)
2337 set_intercept(svm, INTERCEPT_VINTR);
2340 static void svm_clear_vintr(struct vcpu_svm *svm)
2342 clr_intercept(svm, INTERCEPT_VINTR);
2345 static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg)
2347 struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
2350 case VCPU_SREG_CS: return &save->cs;
2351 case VCPU_SREG_DS: return &save->ds;
2352 case VCPU_SREG_ES: return &save->es;
2353 case VCPU_SREG_FS: return &save->fs;
2354 case VCPU_SREG_GS: return &save->gs;
2355 case VCPU_SREG_SS: return &save->ss;
2356 case VCPU_SREG_TR: return &save->tr;
2357 case VCPU_SREG_LDTR: return &save->ldtr;
2363 static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg)
2365 struct vmcb_seg *s = svm_seg(vcpu, seg);
2370 static void svm_get_segment(struct kvm_vcpu *vcpu,
2371 struct kvm_segment *var, int seg)
2373 struct vmcb_seg *s = svm_seg(vcpu, seg);
2375 var->base = s->base;
2376 var->limit = s->limit;
2377 var->selector = s->selector;
2378 var->type = s->attrib & SVM_SELECTOR_TYPE_MASK;
2379 var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1;
2380 var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
2381 var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1;
2382 var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1;
2383 var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
2384 var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
2387 * AMD CPUs circa 2014 track the G bit for all segments except CS.
2388 * However, the SVM spec states that the G bit is not observed by the
2389 * CPU, and some VMware virtual CPUs drop the G bit for all segments.
2390 * So let's synthesize a legal G bit for all segments, this helps
2391 * running KVM nested. It also helps cross-vendor migration, because
2392 * Intel's vmentry has a check on the 'G' bit.
2394 var->g = s->limit > 0xfffff;
2397 * AMD's VMCB does not have an explicit unusable field, so emulate it
2398 * for cross vendor migration purposes by "not present"
2400 var->unusable = !var->present;
2405 * Work around a bug where the busy flag in the tr selector
2415 * The accessed bit must always be set in the segment
2416 * descriptor cache, although it can be cleared in the
2417 * descriptor, the cached bit always remains at 1. Since
2418 * Intel has a check on this, set it here to support
2419 * cross-vendor migration.
2426 * On AMD CPUs sometimes the DB bit in the segment
2427 * descriptor is left as 1, although the whole segment has
2428 * been made unusable. Clear it here to pass an Intel VMX
2429 * entry check when cross vendor migrating.
2433 /* This is symmetric with svm_set_segment() */
2434 var->dpl = to_svm(vcpu)->vmcb->save.cpl;
2439 static int svm_get_cpl(struct kvm_vcpu *vcpu)
2441 struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
2446 static void svm_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
2448 struct vcpu_svm *svm = to_svm(vcpu);
2450 dt->size = svm->vmcb->save.idtr.limit;
2451 dt->address = svm->vmcb->save.idtr.base;
2454 static void svm_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
2456 struct vcpu_svm *svm = to_svm(vcpu);
2458 svm->vmcb->save.idtr.limit = dt->size;
2459 svm->vmcb->save.idtr.base = dt->address ;
2460 mark_dirty(svm->vmcb, VMCB_DT);
2463 static void svm_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
2465 struct vcpu_svm *svm = to_svm(vcpu);
2467 dt->size = svm->vmcb->save.gdtr.limit;
2468 dt->address = svm->vmcb->save.gdtr.base;
2471 static void svm_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
2473 struct vcpu_svm *svm = to_svm(vcpu);
2475 svm->vmcb->save.gdtr.limit = dt->size;
2476 svm->vmcb->save.gdtr.base = dt->address ;
2477 mark_dirty(svm->vmcb, VMCB_DT);
2480 static void svm_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
2484 static void svm_decache_cr3(struct kvm_vcpu *vcpu)
2488 static void svm_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
2492 static void update_cr0_intercept(struct vcpu_svm *svm)
2494 ulong gcr0 = svm->vcpu.arch.cr0;
2495 u64 *hcr0 = &svm->vmcb->save.cr0;
2497 *hcr0 = (*hcr0 & ~SVM_CR0_SELECTIVE_MASK)
2498 | (gcr0 & SVM_CR0_SELECTIVE_MASK);
2500 mark_dirty(svm->vmcb, VMCB_CR);
2502 if (gcr0 == *hcr0) {
2503 clr_cr_intercept(svm, INTERCEPT_CR0_READ);
2504 clr_cr_intercept(svm, INTERCEPT_CR0_WRITE);
2506 set_cr_intercept(svm, INTERCEPT_CR0_READ);
2507 set_cr_intercept(svm, INTERCEPT_CR0_WRITE);
2511 static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
2513 struct vcpu_svm *svm = to_svm(vcpu);
2515 #ifdef CONFIG_X86_64
2516 if (vcpu->arch.efer & EFER_LME) {
2517 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
2518 vcpu->arch.efer |= EFER_LMA;
2519 svm->vmcb->save.efer |= EFER_LMA | EFER_LME;
2522 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG)) {
2523 vcpu->arch.efer &= ~EFER_LMA;
2524 svm->vmcb->save.efer &= ~(EFER_LMA | EFER_LME);
2528 vcpu->arch.cr0 = cr0;
2531 cr0 |= X86_CR0_PG | X86_CR0_WP;
2534 * re-enable caching here because the QEMU bios
2535 * does not do it - this results in some delay at
2538 if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
2539 cr0 &= ~(X86_CR0_CD | X86_CR0_NW);
2540 svm->vmcb->save.cr0 = cr0;
2541 mark_dirty(svm->vmcb, VMCB_CR);
2542 update_cr0_intercept(svm);
2545 static int svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
2547 unsigned long host_cr4_mce = cr4_read_shadow() & X86_CR4_MCE;
2548 unsigned long old_cr4 = to_svm(vcpu)->vmcb->save.cr4;
2550 if (cr4 & X86_CR4_VMXE)
2553 if (npt_enabled && ((old_cr4 ^ cr4) & X86_CR4_PGE))
2554 svm_flush_tlb(vcpu, true);
2556 vcpu->arch.cr4 = cr4;
2559 cr4 |= host_cr4_mce;
2560 to_svm(vcpu)->vmcb->save.cr4 = cr4;
2561 mark_dirty(to_svm(vcpu)->vmcb, VMCB_CR);
2565 static void svm_set_segment(struct kvm_vcpu *vcpu,
2566 struct kvm_segment *var, int seg)
2568 struct vcpu_svm *svm = to_svm(vcpu);
2569 struct vmcb_seg *s = svm_seg(vcpu, seg);
2571 s->base = var->base;
2572 s->limit = var->limit;
2573 s->selector = var->selector;
2574 s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK);
2575 s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT;
2576 s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT;
2577 s->attrib |= ((var->present & 1) && !var->unusable) << SVM_SELECTOR_P_SHIFT;
2578 s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT;
2579 s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT;
2580 s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT;
2581 s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT;
2584 * This is always accurate, except if SYSRET returned to a segment
2585 * with SS.DPL != 3. Intel does not have this quirk, and always
2586 * forces SS.DPL to 3 on sysret, so we ignore that case; fixing it
2587 * would entail passing the CPL to userspace and back.
2589 if (seg == VCPU_SREG_SS)
2590 /* This is symmetric with svm_get_segment() */
2591 svm->vmcb->save.cpl = (var->dpl & 3);
2593 mark_dirty(svm->vmcb, VMCB_SEG);
2596 static void update_bp_intercept(struct kvm_vcpu *vcpu)
2598 struct vcpu_svm *svm = to_svm(vcpu);
2600 clr_exception_intercept(svm, BP_VECTOR);
2602 if (vcpu->guest_debug & KVM_GUESTDBG_ENABLE) {
2603 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
2604 set_exception_intercept(svm, BP_VECTOR);
2606 vcpu->guest_debug = 0;
2609 static void new_asid(struct vcpu_svm *svm, struct svm_cpu_data *sd)
2611 if (sd->next_asid > sd->max_asid) {
2612 ++sd->asid_generation;
2613 sd->next_asid = sd->min_asid;
2614 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID;
2617 svm->asid_generation = sd->asid_generation;
2618 svm->vmcb->control.asid = sd->next_asid++;
2620 mark_dirty(svm->vmcb, VMCB_ASID);
2623 static u64 svm_get_dr6(struct kvm_vcpu *vcpu)
2625 return to_svm(vcpu)->vmcb->save.dr6;
2628 static void svm_set_dr6(struct kvm_vcpu *vcpu, unsigned long value)
2630 struct vcpu_svm *svm = to_svm(vcpu);
2632 svm->vmcb->save.dr6 = value;
2633 mark_dirty(svm->vmcb, VMCB_DR);
2636 static void svm_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
2638 struct vcpu_svm *svm = to_svm(vcpu);
2640 get_debugreg(vcpu->arch.db[0], 0);
2641 get_debugreg(vcpu->arch.db[1], 1);
2642 get_debugreg(vcpu->arch.db[2], 2);
2643 get_debugreg(vcpu->arch.db[3], 3);
2644 vcpu->arch.dr6 = svm_get_dr6(vcpu);
2645 vcpu->arch.dr7 = svm->vmcb->save.dr7;
2647 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
2648 set_dr_intercepts(svm);
2651 static void svm_set_dr7(struct kvm_vcpu *vcpu, unsigned long value)
2653 struct vcpu_svm *svm = to_svm(vcpu);
2655 svm->vmcb->save.dr7 = value;
2656 mark_dirty(svm->vmcb, VMCB_DR);
2659 static int pf_interception(struct vcpu_svm *svm)
2661 u64 fault_address = __sme_clr(svm->vmcb->control.exit_info_2);
2662 u64 error_code = svm->vmcb->control.exit_info_1;
2664 return kvm_handle_page_fault(&svm->vcpu, error_code, fault_address,
2665 static_cpu_has(X86_FEATURE_DECODEASSISTS) ?
2666 svm->vmcb->control.insn_bytes : NULL,
2667 svm->vmcb->control.insn_len);
2670 static int npf_interception(struct vcpu_svm *svm)
2672 u64 fault_address = __sme_clr(svm->vmcb->control.exit_info_2);
2673 u64 error_code = svm->vmcb->control.exit_info_1;
2675 trace_kvm_page_fault(fault_address, error_code);
2676 return kvm_mmu_page_fault(&svm->vcpu, fault_address, error_code,
2677 static_cpu_has(X86_FEATURE_DECODEASSISTS) ?
2678 svm->vmcb->control.insn_bytes : NULL,
2679 svm->vmcb->control.insn_len);
2682 static int db_interception(struct vcpu_svm *svm)
2684 struct kvm_run *kvm_run = svm->vcpu.run;
2686 if (!(svm->vcpu.guest_debug &
2687 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) &&
2688 !svm->nmi_singlestep) {
2689 kvm_queue_exception(&svm->vcpu, DB_VECTOR);
2693 if (svm->nmi_singlestep) {
2694 disable_nmi_singlestep(svm);
2697 if (svm->vcpu.guest_debug &
2698 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) {
2699 kvm_run->exit_reason = KVM_EXIT_DEBUG;
2700 kvm_run->debug.arch.pc =
2701 svm->vmcb->save.cs.base + svm->vmcb->save.rip;
2702 kvm_run->debug.arch.exception = DB_VECTOR;
2709 static int bp_interception(struct vcpu_svm *svm)
2711 struct kvm_run *kvm_run = svm->vcpu.run;
2713 kvm_run->exit_reason = KVM_EXIT_DEBUG;
2714 kvm_run->debug.arch.pc = svm->vmcb->save.cs.base + svm->vmcb->save.rip;
2715 kvm_run->debug.arch.exception = BP_VECTOR;
2719 static int ud_interception(struct vcpu_svm *svm)
2721 return handle_ud(&svm->vcpu);
2724 static int ac_interception(struct vcpu_svm *svm)
2726 kvm_queue_exception_e(&svm->vcpu, AC_VECTOR, 0);
2730 static int gp_interception(struct vcpu_svm *svm)
2732 struct kvm_vcpu *vcpu = &svm->vcpu;
2733 u32 error_code = svm->vmcb->control.exit_info_1;
2736 WARN_ON_ONCE(!enable_vmware_backdoor);
2738 er = kvm_emulate_instruction(vcpu,
2739 EMULTYPE_VMWARE | EMULTYPE_NO_UD_ON_FAIL);
2740 if (er == EMULATE_USER_EXIT)
2742 else if (er != EMULATE_DONE)
2743 kvm_queue_exception_e(vcpu, GP_VECTOR, error_code);
2747 static bool is_erratum_383(void)
2752 if (!erratum_383_found)
2755 value = native_read_msr_safe(MSR_IA32_MC0_STATUS, &err);
2759 /* Bit 62 may or may not be set for this mce */
2760 value &= ~(1ULL << 62);
2762 if (value != 0xb600000000010015ULL)
2765 /* Clear MCi_STATUS registers */
2766 for (i = 0; i < 6; ++i)
2767 native_write_msr_safe(MSR_IA32_MCx_STATUS(i), 0, 0);
2769 value = native_read_msr_safe(MSR_IA32_MCG_STATUS, &err);
2773 value &= ~(1ULL << 2);
2774 low = lower_32_bits(value);
2775 high = upper_32_bits(value);
2777 native_write_msr_safe(MSR_IA32_MCG_STATUS, low, high);
2780 /* Flush tlb to evict multi-match entries */
2786 static void svm_handle_mce(struct vcpu_svm *svm)
2788 if (is_erratum_383()) {
2790 * Erratum 383 triggered. Guest state is corrupt so kill the
2793 pr_err("KVM: Guest triggered AMD Erratum 383\n");
2795 kvm_make_request(KVM_REQ_TRIPLE_FAULT, &svm->vcpu);
2801 * On an #MC intercept the MCE handler is not called automatically in
2802 * the host. So do it by hand here.
2806 /* not sure if we ever come back to this point */
2811 static int mc_interception(struct vcpu_svm *svm)
2816 static int shutdown_interception(struct vcpu_svm *svm)
2818 struct kvm_run *kvm_run = svm->vcpu.run;
2821 * VMCB is undefined after a SHUTDOWN intercept
2822 * so reinitialize it.
2824 clear_page(svm->vmcb);
2827 kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
2831 static int io_interception(struct vcpu_svm *svm)
2833 struct kvm_vcpu *vcpu = &svm->vcpu;
2834 u32 io_info = svm->vmcb->control.exit_info_1; /* address size bug? */
2835 int size, in, string;
2838 ++svm->vcpu.stat.io_exits;
2839 string = (io_info & SVM_IOIO_STR_MASK) != 0;
2840 in = (io_info & SVM_IOIO_TYPE_MASK) != 0;
2842 return kvm_emulate_instruction(vcpu, 0) == EMULATE_DONE;
2844 port = io_info >> 16;
2845 size = (io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT;
2846 svm->next_rip = svm->vmcb->control.exit_info_2;
2848 return kvm_fast_pio(&svm->vcpu, size, port, in);
2851 static int nmi_interception(struct vcpu_svm *svm)
2856 static int intr_interception(struct vcpu_svm *svm)
2858 ++svm->vcpu.stat.irq_exits;
2862 static int nop_on_interception(struct vcpu_svm *svm)
2867 static int halt_interception(struct vcpu_svm *svm)
2869 svm->next_rip = kvm_rip_read(&svm->vcpu) + 1;
2870 return kvm_emulate_halt(&svm->vcpu);
2873 static int vmmcall_interception(struct vcpu_svm *svm)
2875 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2876 return kvm_emulate_hypercall(&svm->vcpu);
2879 static unsigned long nested_svm_get_tdp_cr3(struct kvm_vcpu *vcpu)
2881 struct vcpu_svm *svm = to_svm(vcpu);
2883 return svm->nested.nested_cr3;
2886 static u64 nested_svm_get_tdp_pdptr(struct kvm_vcpu *vcpu, int index)
2888 struct vcpu_svm *svm = to_svm(vcpu);
2889 u64 cr3 = svm->nested.nested_cr3;
2893 ret = kvm_vcpu_read_guest_page(vcpu, gpa_to_gfn(__sme_clr(cr3)), &pdpte,
2894 offset_in_page(cr3) + index * 8, 8);
2900 static void nested_svm_set_tdp_cr3(struct kvm_vcpu *vcpu,
2903 struct vcpu_svm *svm = to_svm(vcpu);
2905 svm->vmcb->control.nested_cr3 = __sme_set(root);
2906 mark_dirty(svm->vmcb, VMCB_NPT);
2909 static void nested_svm_inject_npf_exit(struct kvm_vcpu *vcpu,
2910 struct x86_exception *fault)
2912 struct vcpu_svm *svm = to_svm(vcpu);
2914 if (svm->vmcb->control.exit_code != SVM_EXIT_NPF) {
2916 * TODO: track the cause of the nested page fault, and
2917 * correctly fill in the high bits of exit_info_1.
2919 svm->vmcb->control.exit_code = SVM_EXIT_NPF;
2920 svm->vmcb->control.exit_code_hi = 0;
2921 svm->vmcb->control.exit_info_1 = (1ULL << 32);
2922 svm->vmcb->control.exit_info_2 = fault->address;
2925 svm->vmcb->control.exit_info_1 &= ~0xffffffffULL;
2926 svm->vmcb->control.exit_info_1 |= fault->error_code;
2929 * The present bit is always zero for page structure faults on real
2932 if (svm->vmcb->control.exit_info_1 & (2ULL << 32))
2933 svm->vmcb->control.exit_info_1 &= ~1;
2935 nested_svm_vmexit(svm);
2938 static void nested_svm_init_mmu_context(struct kvm_vcpu *vcpu)
2940 WARN_ON(mmu_is_nested(vcpu));
2941 kvm_init_shadow_mmu(vcpu);
2942 vcpu->arch.mmu->set_cr3 = nested_svm_set_tdp_cr3;
2943 vcpu->arch.mmu->get_cr3 = nested_svm_get_tdp_cr3;
2944 vcpu->arch.mmu->get_pdptr = nested_svm_get_tdp_pdptr;
2945 vcpu->arch.mmu->inject_page_fault = nested_svm_inject_npf_exit;
2946 vcpu->arch.mmu->shadow_root_level = get_npt_level(vcpu);
2947 reset_shadow_zero_bits_mask(vcpu, vcpu->arch.mmu);
2948 vcpu->arch.walk_mmu = &vcpu->arch.nested_mmu;
2951 static void nested_svm_uninit_mmu_context(struct kvm_vcpu *vcpu)
2953 vcpu->arch.walk_mmu = &vcpu->arch.root_mmu;
2956 static int nested_svm_check_permissions(struct vcpu_svm *svm)
2958 if (!(svm->vcpu.arch.efer & EFER_SVME) ||
2959 !is_paging(&svm->vcpu)) {
2960 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2964 if (svm->vmcb->save.cpl) {
2965 kvm_inject_gp(&svm->vcpu, 0);
2972 static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
2973 bool has_error_code, u32 error_code)
2977 if (!is_guest_mode(&svm->vcpu))
2980 vmexit = nested_svm_intercept(svm);
2981 if (vmexit != NESTED_EXIT_DONE)
2984 svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + nr;
2985 svm->vmcb->control.exit_code_hi = 0;
2986 svm->vmcb->control.exit_info_1 = error_code;
2989 * EXITINFO2 is undefined for all exception intercepts other
2992 if (svm->vcpu.arch.exception.nested_apf)
2993 svm->vmcb->control.exit_info_2 = svm->vcpu.arch.apf.nested_apf_token;
2994 else if (svm->vcpu.arch.exception.has_payload)
2995 svm->vmcb->control.exit_info_2 = svm->vcpu.arch.exception.payload;
2997 svm->vmcb->control.exit_info_2 = svm->vcpu.arch.cr2;
2999 svm->nested.exit_required = true;
3003 /* This function returns true if it is save to enable the irq window */
3004 static inline bool nested_svm_intr(struct vcpu_svm *svm)
3006 if (!is_guest_mode(&svm->vcpu))
3009 if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
3012 if (!(svm->vcpu.arch.hflags & HF_HIF_MASK))
3016 * if vmexit was already requested (by intercepted exception
3017 * for instance) do not overwrite it with "external interrupt"
3020 if (svm->nested.exit_required)
3023 svm->vmcb->control.exit_code = SVM_EXIT_INTR;
3024 svm->vmcb->control.exit_info_1 = 0;
3025 svm->vmcb->control.exit_info_2 = 0;
3027 if (svm->nested.intercept & 1ULL) {
3029 * The #vmexit can't be emulated here directly because this
3030 * code path runs with irqs and preemption disabled. A
3031 * #vmexit emulation might sleep. Only signal request for
3034 svm->nested.exit_required = true;
3035 trace_kvm_nested_intr_vmexit(svm->vmcb->save.rip);
3042 /* This function returns true if it is save to enable the nmi window */
3043 static inline bool nested_svm_nmi(struct vcpu_svm *svm)
3045 if (!is_guest_mode(&svm->vcpu))
3048 if (!(svm->nested.intercept & (1ULL << INTERCEPT_NMI)))
3051 svm->vmcb->control.exit_code = SVM_EXIT_NMI;
3052 svm->nested.exit_required = true;
3057 static void *nested_svm_map(struct vcpu_svm *svm, u64 gpa, struct page **_page)
3063 page = kvm_vcpu_gfn_to_page(&svm->vcpu, gpa >> PAGE_SHIFT);
3064 if (is_error_page(page))
3072 kvm_inject_gp(&svm->vcpu, 0);
3077 static void nested_svm_unmap(struct page *page)
3080 kvm_release_page_dirty(page);
3083 static int nested_svm_intercept_ioio(struct vcpu_svm *svm)
3085 unsigned port, size, iopm_len;
3090 if (!(svm->nested.intercept & (1ULL << INTERCEPT_IOIO_PROT)))
3091 return NESTED_EXIT_HOST;
3093 port = svm->vmcb->control.exit_info_1 >> 16;
3094 size = (svm->vmcb->control.exit_info_1 & SVM_IOIO_SIZE_MASK) >>
3095 SVM_IOIO_SIZE_SHIFT;
3096 gpa = svm->nested.vmcb_iopm + (port / 8);
3097 start_bit = port % 8;
3098 iopm_len = (start_bit + size > 8) ? 2 : 1;
3099 mask = (0xf >> (4 - size)) << start_bit;
3102 if (kvm_vcpu_read_guest(&svm->vcpu, gpa, &val, iopm_len))
3103 return NESTED_EXIT_DONE;
3105 return (val & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
3108 static int nested_svm_exit_handled_msr(struct vcpu_svm *svm)
3110 u32 offset, msr, value;
3113 if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT)))
3114 return NESTED_EXIT_HOST;
3116 msr = svm->vcpu.arch.regs[VCPU_REGS_RCX];
3117 offset = svm_msrpm_offset(msr);
3118 write = svm->vmcb->control.exit_info_1 & 1;
3119 mask = 1 << ((2 * (msr & 0xf)) + write);
3121 if (offset == MSR_INVALID)
3122 return NESTED_EXIT_DONE;
3124 /* Offset is in 32 bit units but need in 8 bit units */
3127 if (kvm_vcpu_read_guest(&svm->vcpu, svm->nested.vmcb_msrpm + offset, &value, 4))
3128 return NESTED_EXIT_DONE;
3130 return (value & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
3133 /* DB exceptions for our internal use must not cause vmexit */
3134 static int nested_svm_intercept_db(struct vcpu_svm *svm)
3138 /* if we're not singlestepping, it's not ours */
3139 if (!svm->nmi_singlestep)
3140 return NESTED_EXIT_DONE;
3142 /* if it's not a singlestep exception, it's not ours */
3143 if (kvm_get_dr(&svm->vcpu, 6, &dr6))
3144 return NESTED_EXIT_DONE;
3145 if (!(dr6 & DR6_BS))
3146 return NESTED_EXIT_DONE;
3148 /* if the guest is singlestepping, it should get the vmexit */
3149 if (svm->nmi_singlestep_guest_rflags & X86_EFLAGS_TF) {
3150 disable_nmi_singlestep(svm);
3151 return NESTED_EXIT_DONE;
3154 /* it's ours, the nested hypervisor must not see this one */
3155 return NESTED_EXIT_HOST;
3158 static int nested_svm_exit_special(struct vcpu_svm *svm)
3160 u32 exit_code = svm->vmcb->control.exit_code;
3162 switch (exit_code) {
3165 case SVM_EXIT_EXCP_BASE + MC_VECTOR:
3166 return NESTED_EXIT_HOST;
3168 /* For now we are always handling NPFs when using them */
3170 return NESTED_EXIT_HOST;
3172 case SVM_EXIT_EXCP_BASE + PF_VECTOR:
3173 /* When we're shadowing, trap PFs, but not async PF */
3174 if (!npt_enabled && svm->vcpu.arch.apf.host_apf_reason == 0)
3175 return NESTED_EXIT_HOST;
3181 return NESTED_EXIT_CONTINUE;
3185 * If this function returns true, this #vmexit was already handled
3187 static int nested_svm_intercept(struct vcpu_svm *svm)
3189 u32 exit_code = svm->vmcb->control.exit_code;
3190 int vmexit = NESTED_EXIT_HOST;
3192 switch (exit_code) {
3194 vmexit = nested_svm_exit_handled_msr(svm);
3197 vmexit = nested_svm_intercept_ioio(svm);
3199 case SVM_EXIT_READ_CR0 ... SVM_EXIT_WRITE_CR8: {
3200 u32 bit = 1U << (exit_code - SVM_EXIT_READ_CR0);
3201 if (svm->nested.intercept_cr & bit)
3202 vmexit = NESTED_EXIT_DONE;
3205 case SVM_EXIT_READ_DR0 ... SVM_EXIT_WRITE_DR7: {
3206 u32 bit = 1U << (exit_code - SVM_EXIT_READ_DR0);
3207 if (svm->nested.intercept_dr & bit)
3208 vmexit = NESTED_EXIT_DONE;
3211 case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
3212 u32 excp_bits = 1 << (exit_code - SVM_EXIT_EXCP_BASE);
3213 if (svm->nested.intercept_exceptions & excp_bits) {
3214 if (exit_code == SVM_EXIT_EXCP_BASE + DB_VECTOR)
3215 vmexit = nested_svm_intercept_db(svm);
3217 vmexit = NESTED_EXIT_DONE;
3219 /* async page fault always cause vmexit */
3220 else if ((exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR) &&
3221 svm->vcpu.arch.exception.nested_apf != 0)
3222 vmexit = NESTED_EXIT_DONE;
3225 case SVM_EXIT_ERR: {
3226 vmexit = NESTED_EXIT_DONE;
3230 u64 exit_bits = 1ULL << (exit_code - SVM_EXIT_INTR);
3231 if (svm->nested.intercept & exit_bits)
3232 vmexit = NESTED_EXIT_DONE;
3239 static int nested_svm_exit_handled(struct vcpu_svm *svm)
3243 vmexit = nested_svm_intercept(svm);
3245 if (vmexit == NESTED_EXIT_DONE)
3246 nested_svm_vmexit(svm);
3251 static inline void copy_vmcb_control_area(struct vmcb *dst_vmcb, struct vmcb *from_vmcb)
3253 struct vmcb_control_area *dst = &dst_vmcb->control;
3254 struct vmcb_control_area *from = &from_vmcb->control;
3256 dst->intercept_cr = from->intercept_cr;
3257 dst->intercept_dr = from->intercept_dr;
3258 dst->intercept_exceptions = from->intercept_exceptions;
3259 dst->intercept = from->intercept;
3260 dst->iopm_base_pa = from->iopm_base_pa;
3261 dst->msrpm_base_pa = from->msrpm_base_pa;
3262 dst->tsc_offset = from->tsc_offset;
3263 dst->asid = from->asid;
3264 dst->tlb_ctl = from->tlb_ctl;
3265 dst->int_ctl = from->int_ctl;
3266 dst->int_vector = from->int_vector;
3267 dst->int_state = from->int_state;
3268 dst->exit_code = from->exit_code;
3269 dst->exit_code_hi = from->exit_code_hi;
3270 dst->exit_info_1 = from->exit_info_1;
3271 dst->exit_info_2 = from->exit_info_2;
3272 dst->exit_int_info = from->exit_int_info;
3273 dst->exit_int_info_err = from->exit_int_info_err;
3274 dst->nested_ctl = from->nested_ctl;
3275 dst->event_inj = from->event_inj;
3276 dst->event_inj_err = from->event_inj_err;
3277 dst->nested_cr3 = from->nested_cr3;
3278 dst->virt_ext = from->virt_ext;
3281 static int nested_svm_vmexit(struct vcpu_svm *svm)
3283 struct vmcb *nested_vmcb;
3284 struct vmcb *hsave = svm->nested.hsave;
3285 struct vmcb *vmcb = svm->vmcb;
3288 trace_kvm_nested_vmexit_inject(vmcb->control.exit_code,
3289 vmcb->control.exit_info_1,
3290 vmcb->control.exit_info_2,
3291 vmcb->control.exit_int_info,
3292 vmcb->control.exit_int_info_err,
3295 nested_vmcb = nested_svm_map(svm, svm->nested.vmcb, &page);
3299 /* Exit Guest-Mode */
3300 leave_guest_mode(&svm->vcpu);
3301 svm->nested.vmcb = 0;
3303 /* Give the current vmcb to the guest */
3306 nested_vmcb->save.es = vmcb->save.es;
3307 nested_vmcb->save.cs = vmcb->save.cs;
3308 nested_vmcb->save.ss = vmcb->save.ss;
3309 nested_vmcb->save.ds = vmcb->save.ds;
3310 nested_vmcb->save.gdtr = vmcb->save.gdtr;
3311 nested_vmcb->save.idtr = vmcb->save.idtr;
3312 nested_vmcb->save.efer = svm->vcpu.arch.efer;
3313 nested_vmcb->save.cr0 = kvm_read_cr0(&svm->vcpu);
3314 nested_vmcb->save.cr3 = kvm_read_cr3(&svm->vcpu);
3315 nested_vmcb->save.cr2 = vmcb->save.cr2;
3316 nested_vmcb->save.cr4 = svm->vcpu.arch.cr4;
3317 nested_vmcb->save.rflags = kvm_get_rflags(&svm->vcpu);
3318 nested_vmcb->save.rip = vmcb->save.rip;
3319 nested_vmcb->save.rsp = vmcb->save.rsp;
3320 nested_vmcb->save.rax = vmcb->save.rax;
3321 nested_vmcb->save.dr7 = vmcb->save.dr7;
3322 nested_vmcb->save.dr6 = vmcb->save.dr6;
3323 nested_vmcb->save.cpl = vmcb->save.cpl;
3325 nested_vmcb->control.int_ctl = vmcb->control.int_ctl;
3326 nested_vmcb->control.int_vector = vmcb->control.int_vector;
3327 nested_vmcb->control.int_state = vmcb->control.int_state;
3328 nested_vmcb->control.exit_code = vmcb->control.exit_code;
3329 nested_vmcb->control.exit_code_hi = vmcb->control.exit_code_hi;
3330 nested_vmcb->control.exit_info_1 = vmcb->control.exit_info_1;
3331 nested_vmcb->control.exit_info_2 = vmcb->control.exit_info_2;
3332 nested_vmcb->control.exit_int_info = vmcb->control.exit_int_info;
3333 nested_vmcb->control.exit_int_info_err = vmcb->control.exit_int_info_err;
3335 if (svm->nrips_enabled)
3336 nested_vmcb->control.next_rip = vmcb->control.next_rip;
3339 * If we emulate a VMRUN/#VMEXIT in the same host #vmexit cycle we have
3340 * to make sure that we do not lose injected events. So check event_inj
3341 * here and copy it to exit_int_info if it is valid.
3342 * Exit_int_info and event_inj can't be both valid because the case
3343 * below only happens on a VMRUN instruction intercept which has
3344 * no valid exit_int_info set.
3346 if (vmcb->control.event_inj & SVM_EVTINJ_VALID) {
3347 struct vmcb_control_area *nc = &nested_vmcb->control;
3349 nc->exit_int_info = vmcb->control.event_inj;
3350 nc->exit_int_info_err = vmcb->control.event_inj_err;
3353 nested_vmcb->control.tlb_ctl = 0;
3354 nested_vmcb->control.event_inj = 0;
3355 nested_vmcb->control.event_inj_err = 0;
3357 /* We always set V_INTR_MASKING and remember the old value in hflags */
3358 if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
3359 nested_vmcb->control.int_ctl &= ~V_INTR_MASKING_MASK;
3361 /* Restore the original control entries */
3362 copy_vmcb_control_area(vmcb, hsave);
3364 svm->vcpu.arch.tsc_offset = svm->vmcb->control.tsc_offset;
3365 kvm_clear_exception_queue(&svm->vcpu);
3366 kvm_clear_interrupt_queue(&svm->vcpu);
3368 svm->nested.nested_cr3 = 0;
3370 /* Restore selected save entries */
3371 svm->vmcb->save.es = hsave->save.es;
3372 svm->vmcb->save.cs = hsave->save.cs;
3373 svm->vmcb->save.ss = hsave->save.ss;
3374 svm->vmcb->save.ds = hsave->save.ds;
3375 svm->vmcb->save.gdtr = hsave->save.gdtr;
3376 svm->vmcb->save.idtr = hsave->save.idtr;
3377 kvm_set_rflags(&svm->vcpu, hsave->save.rflags);
3378 svm_set_efer(&svm->vcpu, hsave->save.efer);
3379 svm_set_cr0(&svm->vcpu, hsave->save.cr0 | X86_CR0_PE);
3380 svm_set_cr4(&svm->vcpu, hsave->save.cr4);
3382 svm->vmcb->save.cr3 = hsave->save.cr3;
3383 svm->vcpu.arch.cr3 = hsave->save.cr3;
3385 (void)kvm_set_cr3(&svm->vcpu, hsave->save.cr3);
3387 kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, hsave->save.rax);
3388 kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, hsave->save.rsp);
3389 kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, hsave->save.rip);
3390 svm->vmcb->save.dr7 = 0;
3391 svm->vmcb->save.cpl = 0;
3392 svm->vmcb->control.exit_int_info = 0;
3394 mark_all_dirty(svm->vmcb);
3396 nested_svm_unmap(page);
3398 nested_svm_uninit_mmu_context(&svm->vcpu);
3399 kvm_mmu_reset_context(&svm->vcpu);
3400 kvm_mmu_load(&svm->vcpu);
3405 static bool nested_svm_vmrun_msrpm(struct vcpu_svm *svm)
3408 * This function merges the msr permission bitmaps of kvm and the
3409 * nested vmcb. It is optimized in that it only merges the parts where
3410 * the kvm msr permission bitmap may contain zero bits
3414 if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT)))
3417 for (i = 0; i < MSRPM_OFFSETS; i++) {
3421 if (msrpm_offsets[i] == 0xffffffff)
3424 p = msrpm_offsets[i];
3425 offset = svm->nested.vmcb_msrpm + (p * 4);
3427 if (kvm_vcpu_read_guest(&svm->vcpu, offset, &value, 4))
3430 svm->nested.msrpm[p] = svm->msrpm[p] | value;
3433 svm->vmcb->control.msrpm_base_pa = __sme_set(__pa(svm->nested.msrpm));
3438 static bool nested_vmcb_checks(struct vmcb *vmcb)
3440 if ((vmcb->control.intercept & (1ULL << INTERCEPT_VMRUN)) == 0)
3443 if (vmcb->control.asid == 0)
3446 if ((vmcb->control.nested_ctl & SVM_NESTED_CTL_NP_ENABLE) &&
3453 static void enter_svm_guest_mode(struct vcpu_svm *svm, u64 vmcb_gpa,
3454 struct vmcb *nested_vmcb, struct page *page)
3456 if (kvm_get_rflags(&svm->vcpu) & X86_EFLAGS_IF)
3457 svm->vcpu.arch.hflags |= HF_HIF_MASK;
3459 svm->vcpu.arch.hflags &= ~HF_HIF_MASK;
3461 if (nested_vmcb->control.nested_ctl & SVM_NESTED_CTL_NP_ENABLE) {
3462 kvm_mmu_unload(&svm->vcpu);
3463 svm->nested.nested_cr3 = nested_vmcb->control.nested_cr3;
3464 nested_svm_init_mmu_context(&svm->vcpu);
3467 /* Load the nested guest state */
3468 svm->vmcb->save.es = nested_vmcb->save.es;
3469 svm->vmcb->save.cs = nested_vmcb->save.cs;
3470 svm->vmcb->save.ss = nested_vmcb->save.ss;
3471 svm->vmcb->save.ds = nested_vmcb->save.ds;
3472 svm->vmcb->save.gdtr = nested_vmcb->save.gdtr;
3473 svm->vmcb->save.idtr = nested_vmcb->save.idtr;
3474 kvm_set_rflags(&svm->vcpu, nested_vmcb->save.rflags);
3475 svm_set_efer(&svm->vcpu, nested_vmcb->save.efer);
3476 svm_set_cr0(&svm->vcpu, nested_vmcb->save.cr0);
3477 svm_set_cr4(&svm->vcpu, nested_vmcb->save.cr4);
3479 svm->vmcb->save.cr3 = nested_vmcb->save.cr3;
3480 svm->vcpu.arch.cr3 = nested_vmcb->save.cr3;
3482 (void)kvm_set_cr3(&svm->vcpu, nested_vmcb->save.cr3);
3484 /* Guest paging mode is active - reset mmu */
3485 kvm_mmu_reset_context(&svm->vcpu);
3487 svm->vmcb->save.cr2 = svm->vcpu.arch.cr2 = nested_vmcb->save.cr2;
3488 kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, nested_vmcb->save.rax);
3489 kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, nested_vmcb->save.rsp);
3490 kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, nested_vmcb->save.rip);
3492 /* In case we don't even reach vcpu_run, the fields are not updated */
3493 svm->vmcb->save.rax = nested_vmcb->save.rax;
3494 svm->vmcb->save.rsp = nested_vmcb->save.rsp;
3495 svm->vmcb->save.rip = nested_vmcb->save.rip;
3496 svm->vmcb->save.dr7 = nested_vmcb->save.dr7;
3497 svm->vmcb->save.dr6 = nested_vmcb->save.dr6;
3498 svm->vmcb->save.cpl = nested_vmcb->save.cpl;
3500 svm->nested.vmcb_msrpm = nested_vmcb->control.msrpm_base_pa & ~0x0fffULL;
3501 svm->nested.vmcb_iopm = nested_vmcb->control.iopm_base_pa & ~0x0fffULL;
3503 /* cache intercepts */
3504 svm->nested.intercept_cr = nested_vmcb->control.intercept_cr;
3505 svm->nested.intercept_dr = nested_vmcb->control.intercept_dr;
3506 svm->nested.intercept_exceptions = nested_vmcb->control.intercept_exceptions;
3507 svm->nested.intercept = nested_vmcb->control.intercept;
3509 svm_flush_tlb(&svm->vcpu, true);
3510 svm->vmcb->control.int_ctl = nested_vmcb->control.int_ctl | V_INTR_MASKING_MASK;
3511 if (nested_vmcb->control.int_ctl & V_INTR_MASKING_MASK)
3512 svm->vcpu.arch.hflags |= HF_VINTR_MASK;
3514 svm->vcpu.arch.hflags &= ~HF_VINTR_MASK;
3516 if (svm->vcpu.arch.hflags & HF_VINTR_MASK) {
3517 /* We only want the cr8 intercept bits of the guest */
3518 clr_cr_intercept(svm, INTERCEPT_CR8_READ);
3519 clr_cr_intercept(svm, INTERCEPT_CR8_WRITE);
3522 /* We don't want to see VMMCALLs from a nested guest */
3523 clr_intercept(svm, INTERCEPT_VMMCALL);
3525 svm->vcpu.arch.tsc_offset += nested_vmcb->control.tsc_offset;
3526 svm->vmcb->control.tsc_offset = svm->vcpu.arch.tsc_offset;
3528 svm->vmcb->control.virt_ext = nested_vmcb->control.virt_ext;
3529 svm->vmcb->control.int_vector = nested_vmcb->control.int_vector;
3530 svm->vmcb->control.int_state = nested_vmcb->control.int_state;
3531 svm->vmcb->control.event_inj = nested_vmcb->control.event_inj;
3532 svm->vmcb->control.event_inj_err = nested_vmcb->control.event_inj_err;
3534 nested_svm_unmap(page);
3536 /* Enter Guest-Mode */
3537 enter_guest_mode(&svm->vcpu);
3540 * Merge guest and host intercepts - must be called with vcpu in
3541 * guest-mode to take affect here
3543 recalc_intercepts(svm);
3545 svm->nested.vmcb = vmcb_gpa;
3549 mark_all_dirty(svm->vmcb);
3552 static bool nested_svm_vmrun(struct vcpu_svm *svm)
3554 struct vmcb *nested_vmcb;
3555 struct vmcb *hsave = svm->nested.hsave;
3556 struct vmcb *vmcb = svm->vmcb;
3560 vmcb_gpa = svm->vmcb->save.rax;
3562 nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
3566 if (!nested_vmcb_checks(nested_vmcb)) {
3567 nested_vmcb->control.exit_code = SVM_EXIT_ERR;
3568 nested_vmcb->control.exit_code_hi = 0;
3569 nested_vmcb->control.exit_info_1 = 0;
3570 nested_vmcb->control.exit_info_2 = 0;
3572 nested_svm_unmap(page);
3577 trace_kvm_nested_vmrun(svm->vmcb->save.rip, vmcb_gpa,
3578 nested_vmcb->save.rip,
3579 nested_vmcb->control.int_ctl,
3580 nested_vmcb->control.event_inj,
3581 nested_vmcb->control.nested_ctl);
3583 trace_kvm_nested_intercepts(nested_vmcb->control.intercept_cr & 0xffff,
3584 nested_vmcb->control.intercept_cr >> 16,
3585 nested_vmcb->control.intercept_exceptions,
3586 nested_vmcb->control.intercept);
3588 /* Clear internal status */
3589 kvm_clear_exception_queue(&svm->vcpu);
3590 kvm_clear_interrupt_queue(&svm->vcpu);
3593 * Save the old vmcb, so we don't need to pick what we save, but can
3594 * restore everything when a VMEXIT occurs
3596 hsave->save.es = vmcb->save.es;
3597 hsave->save.cs = vmcb->save.cs;
3598 hsave->save.ss = vmcb->save.ss;
3599 hsave->save.ds = vmcb->save.ds;
3600 hsave->save.gdtr = vmcb->save.gdtr;
3601 hsave->save.idtr = vmcb->save.idtr;
3602 hsave->save.efer = svm->vcpu.arch.efer;
3603 hsave->save.cr0 = kvm_read_cr0(&svm->vcpu);
3604 hsave->save.cr4 = svm->vcpu.arch.cr4;
3605 hsave->save.rflags = kvm_get_rflags(&svm->vcpu);
3606 hsave->save.rip = kvm_rip_read(&svm->vcpu);
3607 hsave->save.rsp = vmcb->save.rsp;
3608 hsave->save.rax = vmcb->save.rax;
3610 hsave->save.cr3 = vmcb->save.cr3;
3612 hsave->save.cr3 = kvm_read_cr3(&svm->vcpu);
3614 copy_vmcb_control_area(hsave, vmcb);
3616 enter_svm_guest_mode(svm, vmcb_gpa, nested_vmcb, page);
3621 static void nested_svm_vmloadsave(struct vmcb *from_vmcb, struct vmcb *to_vmcb)
3623 to_vmcb->save.fs = from_vmcb->save.fs;
3624 to_vmcb->save.gs = from_vmcb->save.gs;
3625 to_vmcb->save.tr = from_vmcb->save.tr;
3626 to_vmcb->save.ldtr = from_vmcb->save.ldtr;
3627 to_vmcb->save.kernel_gs_base = from_vmcb->save.kernel_gs_base;
3628 to_vmcb->save.star = from_vmcb->save.star;
3629 to_vmcb->save.lstar = from_vmcb->save.lstar;
3630 to_vmcb->save.cstar = from_vmcb->save.cstar;
3631 to_vmcb->save.sfmask = from_vmcb->save.sfmask;
3632 to_vmcb->save.sysenter_cs = from_vmcb->save.sysenter_cs;
3633 to_vmcb->save.sysenter_esp = from_vmcb->save.sysenter_esp;
3634 to_vmcb->save.sysenter_eip = from_vmcb->save.sysenter_eip;
3637 static int vmload_interception(struct vcpu_svm *svm)
3639 struct vmcb *nested_vmcb;
3643 if (nested_svm_check_permissions(svm))
3646 nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
3650 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
3651 ret = kvm_skip_emulated_instruction(&svm->vcpu);
3653 nested_svm_vmloadsave(nested_vmcb, svm->vmcb);
3654 nested_svm_unmap(page);
3659 static int vmsave_interception(struct vcpu_svm *svm)
3661 struct vmcb *nested_vmcb;
3665 if (nested_svm_check_permissions(svm))
3668 nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
3672 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
3673 ret = kvm_skip_emulated_instruction(&svm->vcpu);
3675 nested_svm_vmloadsave(svm->vmcb, nested_vmcb);
3676 nested_svm_unmap(page);
3681 static int vmrun_interception(struct vcpu_svm *svm)
3683 if (nested_svm_check_permissions(svm))
3686 /* Save rip after vmrun instruction */
3687 kvm_rip_write(&svm->vcpu, kvm_rip_read(&svm->vcpu) + 3);
3689 if (!nested_svm_vmrun(svm))
3692 if (!nested_svm_vmrun_msrpm(svm))
3699 svm->vmcb->control.exit_code = SVM_EXIT_ERR;
3700 svm->vmcb->control.exit_code_hi = 0;
3701 svm->vmcb->control.exit_info_1 = 0;
3702 svm->vmcb->control.exit_info_2 = 0;
3704 nested_svm_vmexit(svm);
3709 static int stgi_interception(struct vcpu_svm *svm)
3713 if (nested_svm_check_permissions(svm))
3717 * If VGIF is enabled, the STGI intercept is only added to
3718 * detect the opening of the SMI/NMI window; remove it now.
3720 if (vgif_enabled(svm))
3721 clr_intercept(svm, INTERCEPT_STGI);
3723 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
3724 ret = kvm_skip_emulated_instruction(&svm->vcpu);
3725 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
3732 static int clgi_interception(struct vcpu_svm *svm)
3736 if (nested_svm_check_permissions(svm))
3739 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
3740 ret = kvm_skip_emulated_instruction(&svm->vcpu);
3744 /* After a CLGI no interrupts should come */
3745 if (!kvm_vcpu_apicv_active(&svm->vcpu)) {
3746 svm_clear_vintr(svm);
3747 svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
3748 mark_dirty(svm->vmcb, VMCB_INTR);
3754 static int invlpga_interception(struct vcpu_svm *svm)
3756 struct kvm_vcpu *vcpu = &svm->vcpu;
3758 trace_kvm_invlpga(svm->vmcb->save.rip, kvm_register_read(&svm->vcpu, VCPU_REGS_RCX),
3759 kvm_register_read(&svm->vcpu, VCPU_REGS_RAX));
3761 /* Let's treat INVLPGA the same as INVLPG (can be optimized!) */
3762 kvm_mmu_invlpg(vcpu, kvm_register_read(&svm->vcpu, VCPU_REGS_RAX));
3764 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
3765 return kvm_skip_emulated_instruction(&svm->vcpu);
3768 static int skinit_interception(struct vcpu_svm *svm)
3770 trace_kvm_skinit(svm->vmcb->save.rip, kvm_register_read(&svm->vcpu, VCPU_REGS_RAX));
3772 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
3776 static int wbinvd_interception(struct vcpu_svm *svm)
3778 return kvm_emulate_wbinvd(&svm->vcpu);
3781 static int xsetbv_interception(struct vcpu_svm *svm)
3783 u64 new_bv = kvm_read_edx_eax(&svm->vcpu);
3784 u32 index = kvm_register_read(&svm->vcpu, VCPU_REGS_RCX);
3786 if (kvm_set_xcr(&svm->vcpu, index, new_bv) == 0) {
3787 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
3788 return kvm_skip_emulated_instruction(&svm->vcpu);
3794 static int task_switch_interception(struct vcpu_svm *svm)
3798 int int_type = svm->vmcb->control.exit_int_info &
3799 SVM_EXITINTINFO_TYPE_MASK;
3800 int int_vec = svm->vmcb->control.exit_int_info & SVM_EVTINJ_VEC_MASK;
3802 svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_TYPE_MASK;
3804 svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_VALID;
3805 bool has_error_code = false;
3808 tss_selector = (u16)svm->vmcb->control.exit_info_1;
3810 if (svm->vmcb->control.exit_info_2 &
3811 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_IRET))
3812 reason = TASK_SWITCH_IRET;
3813 else if (svm->vmcb->control.exit_info_2 &
3814 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_JMP))
3815 reason = TASK_SWITCH_JMP;
3817 reason = TASK_SWITCH_GATE;
3819 reason = TASK_SWITCH_CALL;
3821 if (reason == TASK_SWITCH_GATE) {
3823 case SVM_EXITINTINFO_TYPE_NMI:
3824 svm->vcpu.arch.nmi_injected = false;
3826 case SVM_EXITINTINFO_TYPE_EXEPT:
3827 if (svm->vmcb->control.exit_info_2 &
3828 (1ULL << SVM_EXITINFOSHIFT_TS_HAS_ERROR_CODE)) {
3829 has_error_code = true;
3831 (u32)svm->vmcb->control.exit_info_2;
3833 kvm_clear_exception_queue(&svm->vcpu);
3835 case SVM_EXITINTINFO_TYPE_INTR:
3836 kvm_clear_interrupt_queue(&svm->vcpu);
3843 if (reason != TASK_SWITCH_GATE ||
3844 int_type == SVM_EXITINTINFO_TYPE_SOFT ||
3845 (int_type == SVM_EXITINTINFO_TYPE_EXEPT &&
3846 (int_vec == OF_VECTOR || int_vec == BP_VECTOR)))
3847 skip_emulated_instruction(&svm->vcpu);
3849 if (int_type != SVM_EXITINTINFO_TYPE_SOFT)
3852 if (kvm_task_switch(&svm->vcpu, tss_selector, int_vec, reason,
3853 has_error_code, error_code) == EMULATE_FAIL) {
3854 svm->vcpu.run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3855 svm->vcpu.run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
3856 svm->vcpu.run->internal.ndata = 0;
3862 static int cpuid_interception(struct vcpu_svm *svm)
3864 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
3865 return kvm_emulate_cpuid(&svm->vcpu);
3868 static int iret_interception(struct vcpu_svm *svm)
3870 ++svm->vcpu.stat.nmi_window_exits;
3871 clr_intercept(svm, INTERCEPT_IRET);
3872 svm->vcpu.arch.hflags |= HF_IRET_MASK;
3873 svm->nmi_iret_rip = kvm_rip_read(&svm->vcpu);
3874 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
3878 static int invlpg_interception(struct vcpu_svm *svm)
3880 if (!static_cpu_has(X86_FEATURE_DECODEASSISTS))
3881 return kvm_emulate_instruction(&svm->vcpu, 0) == EMULATE_DONE;
3883 kvm_mmu_invlpg(&svm->vcpu, svm->vmcb->control.exit_info_1);
3884 return kvm_skip_emulated_instruction(&svm->vcpu);
3887 static int emulate_on_interception(struct vcpu_svm *svm)
3889 return kvm_emulate_instruction(&svm->vcpu, 0) == EMULATE_DONE;
3892 static int rsm_interception(struct vcpu_svm *svm)
3894 return kvm_emulate_instruction_from_buffer(&svm->vcpu,
3895 rsm_ins_bytes, 2) == EMULATE_DONE;
3898 static int rdpmc_interception(struct vcpu_svm *svm)
3902 if (!static_cpu_has(X86_FEATURE_NRIPS))
3903 return emulate_on_interception(svm);
3905 err = kvm_rdpmc(&svm->vcpu);
3906 return kvm_complete_insn_gp(&svm->vcpu, err);
3909 static bool check_selective_cr0_intercepted(struct vcpu_svm *svm,
3912 unsigned long cr0 = svm->vcpu.arch.cr0;
3916 intercept = svm->nested.intercept;
3918 if (!is_guest_mode(&svm->vcpu) ||
3919 (!(intercept & (1ULL << INTERCEPT_SELECTIVE_CR0))))
3922 cr0 &= ~SVM_CR0_SELECTIVE_MASK;
3923 val &= ~SVM_CR0_SELECTIVE_MASK;
3926 svm->vmcb->control.exit_code = SVM_EXIT_CR0_SEL_WRITE;
3927 ret = (nested_svm_exit_handled(svm) == NESTED_EXIT_DONE);
3933 #define CR_VALID (1ULL << 63)
3935 static int cr_interception(struct vcpu_svm *svm)
3941 if (!static_cpu_has(X86_FEATURE_DECODEASSISTS))
3942 return emulate_on_interception(svm);
3944 if (unlikely((svm->vmcb->control.exit_info_1 & CR_VALID) == 0))
3945 return emulate_on_interception(svm);
3947 reg = svm->vmcb->control.exit_info_1 & SVM_EXITINFO_REG_MASK;
3948 if (svm->vmcb->control.exit_code == SVM_EXIT_CR0_SEL_WRITE)
3949 cr = SVM_EXIT_WRITE_CR0 - SVM_EXIT_READ_CR0;
3951 cr = svm->vmcb->control.exit_code - SVM_EXIT_READ_CR0;
3954 if (cr >= 16) { /* mov to cr */
3956 val = kvm_register_read(&svm->vcpu, reg);
3959 if (!check_selective_cr0_intercepted(svm, val))
3960 err = kvm_set_cr0(&svm->vcpu, val);
3966 err = kvm_set_cr3(&svm->vcpu, val);
3969 err = kvm_set_cr4(&svm->vcpu, val);
3972 err = kvm_set_cr8(&svm->vcpu, val);
3975 WARN(1, "unhandled write to CR%d", cr);
3976 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
3979 } else { /* mov from cr */
3982 val = kvm_read_cr0(&svm->vcpu);
3985 val = svm->vcpu.arch.cr2;
3988 val = kvm_read_cr3(&svm->vcpu);
3991 val = kvm_read_cr4(&svm->vcpu);
3994 val = kvm_get_cr8(&svm->vcpu);
3997 WARN(1, "unhandled read from CR%d", cr);
3998 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
4001 kvm_register_write(&svm->vcpu, reg, val);
4003 return kvm_complete_insn_gp(&svm->vcpu, err);
4006 static int dr_interception(struct vcpu_svm *svm)
4011 if (svm->vcpu.guest_debug == 0) {
4013 * No more DR vmexits; force a reload of the debug registers
4014 * and reenter on this instruction. The next vmexit will
4015 * retrieve the full state of the debug registers.
4017 clr_dr_intercepts(svm);
4018 svm->vcpu.arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
4022 if (!boot_cpu_has(X86_FEATURE_DECODEASSISTS))
4023 return emulate_on_interception(svm);
4025 reg = svm->vmcb->control.exit_info_1 & SVM_EXITINFO_REG_MASK;
4026 dr = svm->vmcb->control.exit_code - SVM_EXIT_READ_DR0;
4028 if (dr >= 16) { /* mov to DRn */
4029 if (!kvm_require_dr(&svm->vcpu, dr - 16))
4031 val = kvm_register_read(&svm->vcpu, reg);
4032 kvm_set_dr(&svm->vcpu, dr - 16, val);
4034 if (!kvm_require_dr(&svm->vcpu, dr))
4036 kvm_get_dr(&svm->vcpu, dr, &val);
4037 kvm_register_write(&svm->vcpu, reg, val);
4040 return kvm_skip_emulated_instruction(&svm->vcpu);
4043 static int cr8_write_interception(struct vcpu_svm *svm)
4045 struct kvm_run *kvm_run = svm->vcpu.run;
4048 u8 cr8_prev = kvm_get_cr8(&svm->vcpu);
4049 /* instruction emulation calls kvm_set_cr8() */
4050 r = cr_interception(svm);
4051 if (lapic_in_kernel(&svm->vcpu))
4053 if (cr8_prev <= kvm_get_cr8(&svm->vcpu))
4055 kvm_run->exit_reason = KVM_EXIT_SET_TPR;
4059 static int svm_get_msr_feature(struct kvm_msr_entry *msr)
4063 switch (msr->index) {
4064 case MSR_F10H_DECFG:
4065 if (boot_cpu_has(X86_FEATURE_LFENCE_RDTSC))
4066 msr->data |= MSR_F10H_DECFG_LFENCE_SERIALIZE;
4075 static int svm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
4077 struct vcpu_svm *svm = to_svm(vcpu);
4079 switch (msr_info->index) {
4081 msr_info->data = svm->vmcb->save.star;
4083 #ifdef CONFIG_X86_64
4085 msr_info->data = svm->vmcb->save.lstar;
4088 msr_info->data = svm->vmcb->save.cstar;
4090 case MSR_KERNEL_GS_BASE:
4091 msr_info->data = svm->vmcb->save.kernel_gs_base;
4093 case MSR_SYSCALL_MASK:
4094 msr_info->data = svm->vmcb->save.sfmask;
4097 case MSR_IA32_SYSENTER_CS:
4098 msr_info->data = svm->vmcb->save.sysenter_cs;
4100 case MSR_IA32_SYSENTER_EIP:
4101 msr_info->data = svm->sysenter_eip;
4103 case MSR_IA32_SYSENTER_ESP:
4104 msr_info->data = svm->sysenter_esp;
4107 if (!boot_cpu_has(X86_FEATURE_RDTSCP))
4109 msr_info->data = svm->tsc_aux;
4112 * Nobody will change the following 5 values in the VMCB so we can
4113 * safely return them on rdmsr. They will always be 0 until LBRV is
4116 case MSR_IA32_DEBUGCTLMSR:
4117 msr_info->data = svm->vmcb->save.dbgctl;
4119 case MSR_IA32_LASTBRANCHFROMIP:
4120 msr_info->data = svm->vmcb->save.br_from;
4122 case MSR_IA32_LASTBRANCHTOIP:
4123 msr_info->data = svm->vmcb->save.br_to;
4125 case MSR_IA32_LASTINTFROMIP:
4126 msr_info->data = svm->vmcb->save.last_excp_from;
4128 case MSR_IA32_LASTINTTOIP:
4129 msr_info->data = svm->vmcb->save.last_excp_to;
4131 case MSR_VM_HSAVE_PA:
4132 msr_info->data = svm->nested.hsave_msr;
4135 msr_info->data = svm->nested.vm_cr_msr;
4137 case MSR_IA32_SPEC_CTRL:
4138 if (!msr_info->host_initiated &&
4139 !guest_cpuid_has(vcpu, X86_FEATURE_AMD_IBRS) &&
4140 !guest_cpuid_has(vcpu, X86_FEATURE_AMD_SSBD))
4143 msr_info->data = svm->spec_ctrl;
4145 case MSR_AMD64_VIRT_SPEC_CTRL:
4146 if (!msr_info->host_initiated &&
4147 !guest_cpuid_has(vcpu, X86_FEATURE_VIRT_SSBD))
4150 msr_info->data = svm->virt_spec_ctrl;
4152 case MSR_F15H_IC_CFG: {
4156 family = guest_cpuid_family(vcpu);
4157 model = guest_cpuid_model(vcpu);
4159 if (family < 0 || model < 0)
4160 return kvm_get_msr_common(vcpu, msr_info);
4164 if (family == 0x15 &&
4165 (model >= 0x2 && model < 0x20))
4166 msr_info->data = 0x1E;
4169 case MSR_F10H_DECFG:
4170 msr_info->data = svm->msr_decfg;
4173 return kvm_get_msr_common(vcpu, msr_info);
4178 static int rdmsr_interception(struct vcpu_svm *svm)
4180 u32 ecx = kvm_register_read(&svm->vcpu, VCPU_REGS_RCX);
4181 struct msr_data msr_info;
4183 msr_info.index = ecx;
4184 msr_info.host_initiated = false;
4185 if (svm_get_msr(&svm->vcpu, &msr_info)) {
4186 trace_kvm_msr_read_ex(ecx);
4187 kvm_inject_gp(&svm->vcpu, 0);
4190 trace_kvm_msr_read(ecx, msr_info.data);
4192 kvm_register_write(&svm->vcpu, VCPU_REGS_RAX,
4193 msr_info.data & 0xffffffff);
4194 kvm_register_write(&svm->vcpu, VCPU_REGS_RDX,
4195 msr_info.data >> 32);
4196 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
4197 return kvm_skip_emulated_instruction(&svm->vcpu);
4201 static int svm_set_vm_cr(struct kvm_vcpu *vcpu, u64 data)
4203 struct vcpu_svm *svm = to_svm(vcpu);
4204 int svm_dis, chg_mask;
4206 if (data & ~SVM_VM_CR_VALID_MASK)
4209 chg_mask = SVM_VM_CR_VALID_MASK;
4211 if (svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK)
4212 chg_mask &= ~(SVM_VM_CR_SVM_LOCK_MASK | SVM_VM_CR_SVM_DIS_MASK);
4214 svm->nested.vm_cr_msr &= ~chg_mask;
4215 svm->nested.vm_cr_msr |= (data & chg_mask);
4217 svm_dis = svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK;
4219 /* check for svm_disable while efer.svme is set */
4220 if (svm_dis && (vcpu->arch.efer & EFER_SVME))
4226 static int svm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
4228 struct vcpu_svm *svm = to_svm(vcpu);
4230 u32 ecx = msr->index;
4231 u64 data = msr->data;
4233 case MSR_IA32_CR_PAT:
4234 if (!kvm_mtrr_valid(vcpu, MSR_IA32_CR_PAT, data))
4236 vcpu->arch.pat = data;
4237 svm->vmcb->save.g_pat = data;
4238 mark_dirty(svm->vmcb, VMCB_NPT);
4240 case MSR_IA32_SPEC_CTRL:
4241 if (!msr->host_initiated &&
4242 !guest_cpuid_has(vcpu, X86_FEATURE_AMD_IBRS) &&
4243 !guest_cpuid_has(vcpu, X86_FEATURE_AMD_SSBD))
4246 /* The STIBP bit doesn't fault even if it's not advertised */
4247 if (data & ~(SPEC_CTRL_IBRS | SPEC_CTRL_STIBP | SPEC_CTRL_SSBD))
4250 svm->spec_ctrl = data;
4257 * When it's written (to non-zero) for the first time, pass
4261 * The handling of the MSR bitmap for L2 guests is done in
4262 * nested_svm_vmrun_msrpm.
4263 * We update the L1 MSR bit as well since it will end up
4264 * touching the MSR anyway now.
4266 set_msr_interception(svm->msrpm, MSR_IA32_SPEC_CTRL, 1, 1);
4268 case MSR_IA32_PRED_CMD:
4269 if (!msr->host_initiated &&
4270 !guest_cpuid_has(vcpu, X86_FEATURE_AMD_IBPB))
4273 if (data & ~PRED_CMD_IBPB)
4279 wrmsrl(MSR_IA32_PRED_CMD, PRED_CMD_IBPB);
4280 if (is_guest_mode(vcpu))
4282 set_msr_interception(svm->msrpm, MSR_IA32_PRED_CMD, 0, 1);
4284 case MSR_AMD64_VIRT_SPEC_CTRL:
4285 if (!msr->host_initiated &&
4286 !guest_cpuid_has(vcpu, X86_FEATURE_VIRT_SSBD))
4289 if (data & ~SPEC_CTRL_SSBD)
4292 svm->virt_spec_ctrl = data;
4295 svm->vmcb->save.star = data;
4297 #ifdef CONFIG_X86_64
4299 svm->vmcb->save.lstar = data;
4302 svm->vmcb->save.cstar = data;
4304 case MSR_KERNEL_GS_BASE:
4305 svm->vmcb->save.kernel_gs_base = data;
4307 case MSR_SYSCALL_MASK:
4308 svm->vmcb->save.sfmask = data;
4311 case MSR_IA32_SYSENTER_CS:
4312 svm->vmcb->save.sysenter_cs = data;
4314 case MSR_IA32_SYSENTER_EIP:
4315 svm->sysenter_eip = data;
4316 svm->vmcb->save.sysenter_eip = data;
4318 case MSR_IA32_SYSENTER_ESP:
4319 svm->sysenter_esp = data;
4320 svm->vmcb->save.sysenter_esp = data;
4323 if (!boot_cpu_has(X86_FEATURE_RDTSCP))
4327 * This is rare, so we update the MSR here instead of using
4328 * direct_access_msrs. Doing that would require a rdmsr in
4331 svm->tsc_aux = data;
4332 wrmsrl(MSR_TSC_AUX, svm->tsc_aux);
4334 case MSR_IA32_DEBUGCTLMSR:
4335 if (!boot_cpu_has(X86_FEATURE_LBRV)) {
4336 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTL 0x%llx, nop\n",
4340 if (data & DEBUGCTL_RESERVED_BITS)
4343 svm->vmcb->save.dbgctl = data;
4344 mark_dirty(svm->vmcb, VMCB_LBR);
4345 if (data & (1ULL<<0))
4346 svm_enable_lbrv(svm);
4348 svm_disable_lbrv(svm);
4350 case MSR_VM_HSAVE_PA:
4351 svm->nested.hsave_msr = data;
4354 return svm_set_vm_cr(vcpu, data);
4356 vcpu_unimpl(vcpu, "unimplemented wrmsr: 0x%x data 0x%llx\n", ecx, data);
4358 case MSR_F10H_DECFG: {
4359 struct kvm_msr_entry msr_entry;
4361 msr_entry.index = msr->index;
4362 if (svm_get_msr_feature(&msr_entry))
4365 /* Check the supported bits */
4366 if (data & ~msr_entry.data)
4369 /* Don't allow the guest to change a bit, #GP */
4370 if (!msr->host_initiated && (data ^ msr_entry.data))
4373 svm->msr_decfg = data;
4376 case MSR_IA32_APICBASE:
4377 if (kvm_vcpu_apicv_active(vcpu))
4378 avic_update_vapic_bar(to_svm(vcpu), data);
4379 /* Follow through */
4381 return kvm_set_msr_common(vcpu, msr);
4386 static int wrmsr_interception(struct vcpu_svm *svm)
4388 struct msr_data msr;
4389 u32 ecx = kvm_register_read(&svm->vcpu, VCPU_REGS_RCX);
4390 u64 data = kvm_read_edx_eax(&svm->vcpu);
4394 msr.host_initiated = false;
4396 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
4397 if (kvm_set_msr(&svm->vcpu, &msr)) {
4398 trace_kvm_msr_write_ex(ecx, data);
4399 kvm_inject_gp(&svm->vcpu, 0);
4402 trace_kvm_msr_write(ecx, data);
4403 return kvm_skip_emulated_instruction(&svm->vcpu);
4407 static int msr_interception(struct vcpu_svm *svm)
4409 if (svm->vmcb->control.exit_info_1)
4410 return wrmsr_interception(svm);
4412 return rdmsr_interception(svm);
4415 static int interrupt_window_interception(struct vcpu_svm *svm)
4417 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
4418 svm_clear_vintr(svm);
4419 svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
4420 mark_dirty(svm->vmcb, VMCB_INTR);
4421 ++svm->vcpu.stat.irq_window_exits;
4425 static int pause_interception(struct vcpu_svm *svm)
4427 struct kvm_vcpu *vcpu = &svm->vcpu;
4428 bool in_kernel = (svm_get_cpl(vcpu) == 0);
4430 if (pause_filter_thresh)
4431 grow_ple_window(vcpu);
4433 kvm_vcpu_on_spin(vcpu, in_kernel);
4437 static int nop_interception(struct vcpu_svm *svm)
4439 return kvm_skip_emulated_instruction(&(svm->vcpu));
4442 static int monitor_interception(struct vcpu_svm *svm)
4444 printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
4445 return nop_interception(svm);
4448 static int mwait_interception(struct vcpu_svm *svm)
4450 printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
4451 return nop_interception(svm);
4454 enum avic_ipi_failure_cause {
4455 AVIC_IPI_FAILURE_INVALID_INT_TYPE,
4456 AVIC_IPI_FAILURE_TARGET_NOT_RUNNING,
4457 AVIC_IPI_FAILURE_INVALID_TARGET,
4458 AVIC_IPI_FAILURE_INVALID_BACKING_PAGE,
4461 static int avic_incomplete_ipi_interception(struct vcpu_svm *svm)
4463 u32 icrh = svm->vmcb->control.exit_info_1 >> 32;
4464 u32 icrl = svm->vmcb->control.exit_info_1;
4465 u32 id = svm->vmcb->control.exit_info_2 >> 32;
4466 u32 index = svm->vmcb->control.exit_info_2 & 0xFF;
4467 struct kvm_lapic *apic = svm->vcpu.arch.apic;
4469 trace_kvm_avic_incomplete_ipi(svm->vcpu.vcpu_id, icrh, icrl, id, index);
4472 case AVIC_IPI_FAILURE_INVALID_INT_TYPE:
4474 * AVIC hardware handles the generation of
4475 * IPIs when the specified Message Type is Fixed
4476 * (also known as fixed delivery mode) and
4477 * the Trigger Mode is edge-triggered. The hardware
4478 * also supports self and broadcast delivery modes
4479 * specified via the Destination Shorthand(DSH)
4480 * field of the ICRL. Logical and physical APIC ID
4481 * formats are supported. All other IPI types cause
4482 * a #VMEXIT, which needs to emulated.
4484 kvm_lapic_reg_write(apic, APIC_ICR2, icrh);
4485 kvm_lapic_reg_write(apic, APIC_ICR, icrl);
4487 case AVIC_IPI_FAILURE_TARGET_NOT_RUNNING: {
4489 struct kvm_vcpu *vcpu;
4490 struct kvm *kvm = svm->vcpu.kvm;
4491 struct kvm_lapic *apic = svm->vcpu.arch.apic;
4494 * At this point, we expect that the AVIC HW has already
4495 * set the appropriate IRR bits on the valid target
4496 * vcpus. So, we just need to kick the appropriate vcpu.
4498 kvm_for_each_vcpu(i, vcpu, kvm) {
4499 bool m = kvm_apic_match_dest(vcpu, apic,
4500 icrl & KVM_APIC_SHORT_MASK,
4501 GET_APIC_DEST_FIELD(icrh),
4502 icrl & KVM_APIC_DEST_MASK);
4504 if (m && !avic_vcpu_is_running(vcpu))
4505 kvm_vcpu_wake_up(vcpu);
4509 case AVIC_IPI_FAILURE_INVALID_TARGET:
4511 case AVIC_IPI_FAILURE_INVALID_BACKING_PAGE:
4512 WARN_ONCE(1, "Invalid backing page\n");
4515 pr_err("Unknown IPI interception\n");
4521 static u32 *avic_get_logical_id_entry(struct kvm_vcpu *vcpu, u32 ldr, bool flat)
4523 struct kvm_svm *kvm_svm = to_kvm_svm(vcpu->kvm);
4525 u32 *logical_apic_id_table;
4526 int dlid = GET_APIC_LOGICAL_ID(ldr);
4531 if (flat) { /* flat */
4532 index = ffs(dlid) - 1;
4535 } else { /* cluster */
4536 int cluster = (dlid & 0xf0) >> 4;
4537 int apic = ffs(dlid & 0x0f) - 1;
4539 if ((apic < 0) || (apic > 7) ||
4542 index = (cluster << 2) + apic;
4545 logical_apic_id_table = (u32 *) page_address(kvm_svm->avic_logical_id_table_page);
4547 return &logical_apic_id_table[index];
4550 static int avic_ldr_write(struct kvm_vcpu *vcpu, u8 g_physical_id, u32 ldr,
4554 u32 *entry, new_entry;
4556 flat = kvm_lapic_get_reg(vcpu->arch.apic, APIC_DFR) == APIC_DFR_FLAT;
4557 entry = avic_get_logical_id_entry(vcpu, ldr, flat);
4561 new_entry = READ_ONCE(*entry);
4562 new_entry &= ~AVIC_LOGICAL_ID_ENTRY_GUEST_PHYSICAL_ID_MASK;
4563 new_entry |= (g_physical_id & AVIC_LOGICAL_ID_ENTRY_GUEST_PHYSICAL_ID_MASK);
4565 new_entry |= AVIC_LOGICAL_ID_ENTRY_VALID_MASK;
4567 new_entry &= ~AVIC_LOGICAL_ID_ENTRY_VALID_MASK;
4568 WRITE_ONCE(*entry, new_entry);
4573 static int avic_handle_ldr_update(struct kvm_vcpu *vcpu)
4576 struct vcpu_svm *svm = to_svm(vcpu);
4577 u32 ldr = kvm_lapic_get_reg(vcpu->arch.apic, APIC_LDR);
4582 ret = avic_ldr_write(vcpu, vcpu->vcpu_id, ldr, true);
4583 if (ret && svm->ldr_reg) {
4584 avic_ldr_write(vcpu, 0, svm->ldr_reg, false);
4592 static int avic_handle_apic_id_update(struct kvm_vcpu *vcpu)
4595 struct vcpu_svm *svm = to_svm(vcpu);
4596 u32 apic_id_reg = kvm_lapic_get_reg(vcpu->arch.apic, APIC_ID);
4597 u32 id = (apic_id_reg >> 24) & 0xff;
4599 if (vcpu->vcpu_id == id)
4602 old = avic_get_physical_id_entry(vcpu, vcpu->vcpu_id);
4603 new = avic_get_physical_id_entry(vcpu, id);
4607 /* We need to move physical_id_entry to new offset */
4610 to_svm(vcpu)->avic_physical_id_cache = new;
4613 * Also update the guest physical APIC ID in the logical
4614 * APIC ID table entry if already setup the LDR.
4617 avic_handle_ldr_update(vcpu);
4622 static int avic_handle_dfr_update(struct kvm_vcpu *vcpu)
4624 struct vcpu_svm *svm = to_svm(vcpu);
4625 struct kvm_svm *kvm_svm = to_kvm_svm(vcpu->kvm);
4626 u32 dfr = kvm_lapic_get_reg(vcpu->arch.apic, APIC_DFR);
4627 u32 mod = (dfr >> 28) & 0xf;
4630 * We assume that all local APICs are using the same type.
4631 * If this changes, we need to flush the AVIC logical
4634 if (kvm_svm->ldr_mode == mod)
4637 clear_page(page_address(kvm_svm->avic_logical_id_table_page));
4638 kvm_svm->ldr_mode = mod;
4641 avic_handle_ldr_update(vcpu);
4645 static int avic_unaccel_trap_write(struct vcpu_svm *svm)
4647 struct kvm_lapic *apic = svm->vcpu.arch.apic;
4648 u32 offset = svm->vmcb->control.exit_info_1 &
4649 AVIC_UNACCEL_ACCESS_OFFSET_MASK;
4653 if (avic_handle_apic_id_update(&svm->vcpu))
4657 if (avic_handle_ldr_update(&svm->vcpu))
4661 avic_handle_dfr_update(&svm->vcpu);
4667 kvm_lapic_reg_write(apic, offset, kvm_lapic_get_reg(apic, offset));
4672 static bool is_avic_unaccelerated_access_trap(u32 offset)
4701 static int avic_unaccelerated_access_interception(struct vcpu_svm *svm)
4704 u32 offset = svm->vmcb->control.exit_info_1 &
4705 AVIC_UNACCEL_ACCESS_OFFSET_MASK;
4706 u32 vector = svm->vmcb->control.exit_info_2 &
4707 AVIC_UNACCEL_ACCESS_VECTOR_MASK;
4708 bool write = (svm->vmcb->control.exit_info_1 >> 32) &
4709 AVIC_UNACCEL_ACCESS_WRITE_MASK;
4710 bool trap = is_avic_unaccelerated_access_trap(offset);
4712 trace_kvm_avic_unaccelerated_access(svm->vcpu.vcpu_id, offset,
4713 trap, write, vector);
4716 WARN_ONCE(!write, "svm: Handling trap read.\n");
4717 ret = avic_unaccel_trap_write(svm);
4719 /* Handling Fault */
4720 ret = (kvm_emulate_instruction(&svm->vcpu, 0) == EMULATE_DONE);
4726 static int (*const svm_exit_handlers[])(struct vcpu_svm *svm) = {
4727 [SVM_EXIT_READ_CR0] = cr_interception,
4728 [SVM_EXIT_READ_CR3] = cr_interception,
4729 [SVM_EXIT_READ_CR4] = cr_interception,
4730 [SVM_EXIT_READ_CR8] = cr_interception,
4731 [SVM_EXIT_CR0_SEL_WRITE] = cr_interception,
4732 [SVM_EXIT_WRITE_CR0] = cr_interception,
4733 [SVM_EXIT_WRITE_CR3] = cr_interception,
4734 [SVM_EXIT_WRITE_CR4] = cr_interception,
4735 [SVM_EXIT_WRITE_CR8] = cr8_write_interception,
4736 [SVM_EXIT_READ_DR0] = dr_interception,
4737 [SVM_EXIT_READ_DR1] = dr_interception,
4738 [SVM_EXIT_READ_DR2] = dr_interception,
4739 [SVM_EXIT_READ_DR3] = dr_interception,
4740 [SVM_EXIT_READ_DR4] = dr_interception,
4741 [SVM_EXIT_READ_DR5] = dr_interception,
4742 [SVM_EXIT_READ_DR6] = dr_interception,
4743 [SVM_EXIT_READ_DR7] = dr_interception,
4744 [SVM_EXIT_WRITE_DR0] = dr_interception,
4745 [SVM_EXIT_WRITE_DR1] = dr_interception,
4746 [SVM_EXIT_WRITE_DR2] = dr_interception,
4747 [SVM_EXIT_WRITE_DR3] = dr_interception,
4748 [SVM_EXIT_WRITE_DR4] = dr_interception,
4749 [SVM_EXIT_WRITE_DR5] = dr_interception,
4750 [SVM_EXIT_WRITE_DR6] = dr_interception,
4751 [SVM_EXIT_WRITE_DR7] = dr_interception,
4752 [SVM_EXIT_EXCP_BASE + DB_VECTOR] = db_interception,
4753 [SVM_EXIT_EXCP_BASE + BP_VECTOR] = bp_interception,
4754 [SVM_EXIT_EXCP_BASE + UD_VECTOR] = ud_interception,
4755 [SVM_EXIT_EXCP_BASE + PF_VECTOR] = pf_interception,
4756 [SVM_EXIT_EXCP_BASE + MC_VECTOR] = mc_interception,
4757 [SVM_EXIT_EXCP_BASE + AC_VECTOR] = ac_interception,
4758 [SVM_EXIT_EXCP_BASE + GP_VECTOR] = gp_interception,
4759 [SVM_EXIT_INTR] = intr_interception,
4760 [SVM_EXIT_NMI] = nmi_interception,
4761 [SVM_EXIT_SMI] = nop_on_interception,
4762 [SVM_EXIT_INIT] = nop_on_interception,
4763 [SVM_EXIT_VINTR] = interrupt_window_interception,
4764 [SVM_EXIT_RDPMC] = rdpmc_interception,
4765 [SVM_EXIT_CPUID] = cpuid_interception,
4766 [SVM_EXIT_IRET] = iret_interception,
4767 [SVM_EXIT_INVD] = emulate_on_interception,
4768 [SVM_EXIT_PAUSE] = pause_interception,
4769 [SVM_EXIT_HLT] = halt_interception,
4770 [SVM_EXIT_INVLPG] = invlpg_interception,
4771 [SVM_EXIT_INVLPGA] = invlpga_interception,
4772 [SVM_EXIT_IOIO] = io_interception,
4773 [SVM_EXIT_MSR] = msr_interception,
4774 [SVM_EXIT_TASK_SWITCH] = task_switch_interception,
4775 [SVM_EXIT_SHUTDOWN] = shutdown_interception,
4776 [SVM_EXIT_VMRUN] = vmrun_interception,
4777 [SVM_EXIT_VMMCALL] = vmmcall_interception,
4778 [SVM_EXIT_VMLOAD] = vmload_interception,
4779 [SVM_EXIT_VMSAVE] = vmsave_interception,
4780 [SVM_EXIT_STGI] = stgi_interception,
4781 [SVM_EXIT_CLGI] = clgi_interception,
4782 [SVM_EXIT_SKINIT] = skinit_interception,
4783 [SVM_EXIT_WBINVD] = wbinvd_interception,
4784 [SVM_EXIT_MONITOR] = monitor_interception,
4785 [SVM_EXIT_MWAIT] = mwait_interception,
4786 [SVM_EXIT_XSETBV] = xsetbv_interception,
4787 [SVM_EXIT_NPF] = npf_interception,
4788 [SVM_EXIT_RSM] = rsm_interception,
4789 [SVM_EXIT_AVIC_INCOMPLETE_IPI] = avic_incomplete_ipi_interception,
4790 [SVM_EXIT_AVIC_UNACCELERATED_ACCESS] = avic_unaccelerated_access_interception,
4793 static void dump_vmcb(struct kvm_vcpu *vcpu)
4795 struct vcpu_svm *svm = to_svm(vcpu);
4796 struct vmcb_control_area *control = &svm->vmcb->control;
4797 struct vmcb_save_area *save = &svm->vmcb->save;
4799 pr_err("VMCB Control Area:\n");
4800 pr_err("%-20s%04x\n", "cr_read:", control->intercept_cr & 0xffff);
4801 pr_err("%-20s%04x\n", "cr_write:", control->intercept_cr >> 16);
4802 pr_err("%-20s%04x\n", "dr_read:", control->intercept_dr & 0xffff);
4803 pr_err("%-20s%04x\n", "dr_write:", control->intercept_dr >> 16);
4804 pr_err("%-20s%08x\n", "exceptions:", control->intercept_exceptions);
4805 pr_err("%-20s%016llx\n", "intercepts:", control->intercept);
4806 pr_err("%-20s%d\n", "pause filter count:", control->pause_filter_count);
4807 pr_err("%-20s%d\n", "pause filter threshold:",
4808 control->pause_filter_thresh);
4809 pr_err("%-20s%016llx\n", "iopm_base_pa:", control->iopm_base_pa);
4810 pr_err("%-20s%016llx\n", "msrpm_base_pa:", control->msrpm_base_pa);
4811 pr_err("%-20s%016llx\n", "tsc_offset:", control->tsc_offset);
4812 pr_err("%-20s%d\n", "asid:", control->asid);
4813 pr_err("%-20s%d\n", "tlb_ctl:", control->tlb_ctl);
4814 pr_err("%-20s%08x\n", "int_ctl:", control->int_ctl);
4815 pr_err("%-20s%08x\n", "int_vector:", control->int_vector);
4816 pr_err("%-20s%08x\n", "int_state:", control->int_state);
4817 pr_err("%-20s%08x\n", "exit_code:", control->exit_code);
4818 pr_err("%-20s%016llx\n", "exit_info1:", control->exit_info_1);
4819 pr_err("%-20s%016llx\n", "exit_info2:", control->exit_info_2);
4820 pr_err("%-20s%08x\n", "exit_int_info:", control->exit_int_info);
4821 pr_err("%-20s%08x\n", "exit_int_info_err:", control->exit_int_info_err);
4822 pr_err("%-20s%lld\n", "nested_ctl:", control->nested_ctl);
4823 pr_err("%-20s%016llx\n", "nested_cr3:", control->nested_cr3);
4824 pr_err("%-20s%016llx\n", "avic_vapic_bar:", control->avic_vapic_bar);
4825 pr_err("%-20s%08x\n", "event_inj:", control->event_inj);
4826 pr_err("%-20s%08x\n", "event_inj_err:", control->event_inj_err);
4827 pr_err("%-20s%lld\n", "virt_ext:", control->virt_ext);
4828 pr_err("%-20s%016llx\n", "next_rip:", control->next_rip);
4829 pr_err("%-20s%016llx\n", "avic_backing_page:", control->avic_backing_page);
4830 pr_err("%-20s%016llx\n", "avic_logical_id:", control->avic_logical_id);
4831 pr_err("%-20s%016llx\n", "avic_physical_id:", control->avic_physical_id);
4832 pr_err("VMCB State Save Area:\n");
4833 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4835 save->es.selector, save->es.attrib,
4836 save->es.limit, save->es.base);
4837 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4839 save->cs.selector, save->cs.attrib,
4840 save->cs.limit, save->cs.base);
4841 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4843 save->ss.selector, save->ss.attrib,
4844 save->ss.limit, save->ss.base);
4845 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4847 save->ds.selector, save->ds.attrib,
4848 save->ds.limit, save->ds.base);
4849 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4851 save->fs.selector, save->fs.attrib,
4852 save->fs.limit, save->fs.base);
4853 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4855 save->gs.selector, save->gs.attrib,
4856 save->gs.limit, save->gs.base);
4857 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4859 save->gdtr.selector, save->gdtr.attrib,
4860 save->gdtr.limit, save->gdtr.base);
4861 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4863 save->ldtr.selector, save->ldtr.attrib,
4864 save->ldtr.limit, save->ldtr.base);
4865 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4867 save->idtr.selector, save->idtr.attrib,
4868 save->idtr.limit, save->idtr.base);
4869 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4871 save->tr.selector, save->tr.attrib,
4872 save->tr.limit, save->tr.base);
4873 pr_err("cpl: %d efer: %016llx\n",
4874 save->cpl, save->efer);
4875 pr_err("%-15s %016llx %-13s %016llx\n",
4876 "cr0:", save->cr0, "cr2:", save->cr2);
4877 pr_err("%-15s %016llx %-13s %016llx\n",
4878 "cr3:", save->cr3, "cr4:", save->cr4);
4879 pr_err("%-15s %016llx %-13s %016llx\n",
4880 "dr6:", save->dr6, "dr7:", save->dr7);
4881 pr_err("%-15s %016llx %-13s %016llx\n",
4882 "rip:", save->rip, "rflags:", save->rflags);
4883 pr_err("%-15s %016llx %-13s %016llx\n",
4884 "rsp:", save->rsp, "rax:", save->rax);
4885 pr_err("%-15s %016llx %-13s %016llx\n",
4886 "star:", save->star, "lstar:", save->lstar);
4887 pr_err("%-15s %016llx %-13s %016llx\n",
4888 "cstar:", save->cstar, "sfmask:", save->sfmask);
4889 pr_err("%-15s %016llx %-13s %016llx\n",
4890 "kernel_gs_base:", save->kernel_gs_base,
4891 "sysenter_cs:", save->sysenter_cs);
4892 pr_err("%-15s %016llx %-13s %016llx\n",
4893 "sysenter_esp:", save->sysenter_esp,
4894 "sysenter_eip:", save->sysenter_eip);
4895 pr_err("%-15s %016llx %-13s %016llx\n",
4896 "gpat:", save->g_pat, "dbgctl:", save->dbgctl);
4897 pr_err("%-15s %016llx %-13s %016llx\n",
4898 "br_from:", save->br_from, "br_to:", save->br_to);
4899 pr_err("%-15s %016llx %-13s %016llx\n",
4900 "excp_from:", save->last_excp_from,
4901 "excp_to:", save->last_excp_to);
4904 static void svm_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
4906 struct vmcb_control_area *control = &to_svm(vcpu)->vmcb->control;
4908 *info1 = control->exit_info_1;
4909 *info2 = control->exit_info_2;
4912 static int handle_exit(struct kvm_vcpu *vcpu)
4914 struct vcpu_svm *svm = to_svm(vcpu);
4915 struct kvm_run *kvm_run = vcpu->run;
4916 u32 exit_code = svm->vmcb->control.exit_code;
4918 trace_kvm_exit(exit_code, vcpu, KVM_ISA_SVM);
4920 if (!is_cr_intercept(svm, INTERCEPT_CR0_WRITE))
4921 vcpu->arch.cr0 = svm->vmcb->save.cr0;
4923 vcpu->arch.cr3 = svm->vmcb->save.cr3;
4925 if (unlikely(svm->nested.exit_required)) {
4926 nested_svm_vmexit(svm);
4927 svm->nested.exit_required = false;
4932 if (is_guest_mode(vcpu)) {
4935 trace_kvm_nested_vmexit(svm->vmcb->save.rip, exit_code,
4936 svm->vmcb->control.exit_info_1,
4937 svm->vmcb->control.exit_info_2,
4938 svm->vmcb->control.exit_int_info,
4939 svm->vmcb->control.exit_int_info_err,
4942 vmexit = nested_svm_exit_special(svm);
4944 if (vmexit == NESTED_EXIT_CONTINUE)
4945 vmexit = nested_svm_exit_handled(svm);
4947 if (vmexit == NESTED_EXIT_DONE)
4951 svm_complete_interrupts(svm);
4953 if (svm->vmcb->control.exit_code == SVM_EXIT_ERR) {
4954 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4955 kvm_run->fail_entry.hardware_entry_failure_reason
4956 = svm->vmcb->control.exit_code;
4957 pr_err("KVM: FAILED VMRUN WITH VMCB:\n");
4962 if (is_external_interrupt(svm->vmcb->control.exit_int_info) &&
4963 exit_code != SVM_EXIT_EXCP_BASE + PF_VECTOR &&
4964 exit_code != SVM_EXIT_NPF && exit_code != SVM_EXIT_TASK_SWITCH &&
4965 exit_code != SVM_EXIT_INTR && exit_code != SVM_EXIT_NMI)
4966 printk(KERN_ERR "%s: unexpected exit_int_info 0x%x "
4968 __func__, svm->vmcb->control.exit_int_info,
4971 if (exit_code >= ARRAY_SIZE(svm_exit_handlers)
4972 || !svm_exit_handlers[exit_code]) {
4973 WARN_ONCE(1, "svm: unexpected exit reason 0x%x\n", exit_code);
4974 kvm_queue_exception(vcpu, UD_VECTOR);
4978 return svm_exit_handlers[exit_code](svm);
4981 static void reload_tss(struct kvm_vcpu *vcpu)
4983 int cpu = raw_smp_processor_id();
4985 struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
4986 sd->tss_desc->type = 9; /* available 32/64-bit TSS */
4990 static void pre_sev_run(struct vcpu_svm *svm, int cpu)
4992 struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
4993 int asid = sev_get_asid(svm->vcpu.kvm);
4995 /* Assign the asid allocated with this SEV guest */
4996 svm->vmcb->control.asid = asid;
5001 * 1) when different VMCB for the same ASID is to be run on the same host CPU.
5002 * 2) or this VMCB was executed on different host CPU in previous VMRUNs.
5004 if (sd->sev_vmcbs[asid] == svm->vmcb &&
5005 svm->last_cpu == cpu)
5008 svm->last_cpu = cpu;
5009 sd->sev_vmcbs[asid] = svm->vmcb;
5010 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ASID;
5011 mark_dirty(svm->vmcb, VMCB_ASID);
5014 static void pre_svm_run(struct vcpu_svm *svm)
5016 int cpu = raw_smp_processor_id();
5018 struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
5020 if (sev_guest(svm->vcpu.kvm))
5021 return pre_sev_run(svm, cpu);
5023 /* FIXME: handle wraparound of asid_generation */
5024 if (svm->asid_generation != sd->asid_generation)
5028 static void svm_inject_nmi(struct kvm_vcpu *vcpu)
5030 struct vcpu_svm *svm = to_svm(vcpu);
5032 svm->vmcb->control.event_inj = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI;
5033 vcpu->arch.hflags |= HF_NMI_MASK;
5034 set_intercept(svm, INTERCEPT_IRET);
5035 ++vcpu->stat.nmi_injections;
5038 static inline void svm_inject_irq(struct vcpu_svm *svm, int irq)
5040 struct vmcb_control_area *control;
5042 /* The following fields are ignored when AVIC is enabled */
5043 control = &svm->vmcb->control;
5044 control->int_vector = irq;
5045 control->int_ctl &= ~V_INTR_PRIO_MASK;
5046 control->int_ctl |= V_IRQ_MASK |
5047 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT);
5048 mark_dirty(svm->vmcb, VMCB_INTR);
5051 static void svm_set_irq(struct kvm_vcpu *vcpu)
5053 struct vcpu_svm *svm = to_svm(vcpu);
5055 BUG_ON(!(gif_set(svm)));
5057 trace_kvm_inj_virq(vcpu->arch.interrupt.nr);
5058 ++vcpu->stat.irq_injections;
5060 svm->vmcb->control.event_inj = vcpu->arch.interrupt.nr |
5061 SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR;
5064 static inline bool svm_nested_virtualize_tpr(struct kvm_vcpu *vcpu)
5066 return is_guest_mode(vcpu) && (vcpu->arch.hflags & HF_VINTR_MASK);
5069 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
5071 struct vcpu_svm *svm = to_svm(vcpu);
5073 if (svm_nested_virtualize_tpr(vcpu) ||
5074 kvm_vcpu_apicv_active(vcpu))
5077 clr_cr_intercept(svm, INTERCEPT_CR8_WRITE);
5083 set_cr_intercept(svm, INTERCEPT_CR8_WRITE);
5086 static void svm_set_virtual_apic_mode(struct kvm_vcpu *vcpu)
5091 static bool svm_get_enable_apicv(struct kvm_vcpu *vcpu)
5093 return avic && irqchip_split(vcpu->kvm);
5096 static void svm_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
5100 static void svm_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr)
5104 /* Note: Currently only used by Hyper-V. */
5105 static void svm_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu)
5107 struct vcpu_svm *svm = to_svm(vcpu);
5108 struct vmcb *vmcb = svm->vmcb;
5110 if (!kvm_vcpu_apicv_active(&svm->vcpu))
5113 vmcb->control.int_ctl &= ~AVIC_ENABLE_MASK;
5114 mark_dirty(vmcb, VMCB_INTR);
5117 static void svm_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
5122 static void svm_deliver_avic_intr(struct kvm_vcpu *vcpu, int vec)
5124 kvm_lapic_set_irr(vec, vcpu->arch.apic);
5125 smp_mb__after_atomic();
5127 if (avic_vcpu_is_running(vcpu))
5128 wrmsrl(SVM_AVIC_DOORBELL,
5129 kvm_cpu_get_apicid(vcpu->cpu));
5131 kvm_vcpu_wake_up(vcpu);
5134 static void svm_ir_list_del(struct vcpu_svm *svm, struct amd_iommu_pi_data *pi)
5136 unsigned long flags;
5137 struct amd_svm_iommu_ir *cur;
5139 spin_lock_irqsave(&svm->ir_list_lock, flags);
5140 list_for_each_entry(cur, &svm->ir_list, node) {
5141 if (cur->data != pi->ir_data)
5143 list_del(&cur->node);
5147 spin_unlock_irqrestore(&svm->ir_list_lock, flags);
5150 static int svm_ir_list_add(struct vcpu_svm *svm, struct amd_iommu_pi_data *pi)
5153 unsigned long flags;
5154 struct amd_svm_iommu_ir *ir;
5157 * In some cases, the existing irte is updaed and re-set,
5158 * so we need to check here if it's already been * added
5161 if (pi->ir_data && (pi->prev_ga_tag != 0)) {
5162 struct kvm *kvm = svm->vcpu.kvm;
5163 u32 vcpu_id = AVIC_GATAG_TO_VCPUID(pi->prev_ga_tag);
5164 struct kvm_vcpu *prev_vcpu = kvm_get_vcpu_by_id(kvm, vcpu_id);
5165 struct vcpu_svm *prev_svm;
5172 prev_svm = to_svm(prev_vcpu);
5173 svm_ir_list_del(prev_svm, pi);
5177 * Allocating new amd_iommu_pi_data, which will get
5178 * add to the per-vcpu ir_list.
5180 ir = kzalloc(sizeof(struct amd_svm_iommu_ir), GFP_KERNEL);
5185 ir->data = pi->ir_data;
5187 spin_lock_irqsave(&svm->ir_list_lock, flags);
5188 list_add(&ir->node, &svm->ir_list);
5189 spin_unlock_irqrestore(&svm->ir_list_lock, flags);
5196 * The HW cannot support posting multicast/broadcast
5197 * interrupts to a vCPU. So, we still use legacy interrupt
5198 * remapping for these kind of interrupts.
5200 * For lowest-priority interrupts, we only support
5201 * those with single CPU as the destination, e.g. user
5202 * configures the interrupts via /proc/irq or uses
5203 * irqbalance to make the interrupts single-CPU.
5206 get_pi_vcpu_info(struct kvm *kvm, struct kvm_kernel_irq_routing_entry *e,
5207 struct vcpu_data *vcpu_info, struct vcpu_svm **svm)
5209 struct kvm_lapic_irq irq;
5210 struct kvm_vcpu *vcpu = NULL;
5212 kvm_set_msi_irq(kvm, e, &irq);
5214 if (!kvm_intr_is_single_vcpu(kvm, &irq, &vcpu)) {
5215 pr_debug("SVM: %s: use legacy intr remap mode for irq %u\n",
5216 __func__, irq.vector);
5220 pr_debug("SVM: %s: use GA mode for irq %u\n", __func__,
5222 *svm = to_svm(vcpu);
5223 vcpu_info->pi_desc_addr = __sme_set(page_to_phys((*svm)->avic_backing_page));
5224 vcpu_info->vector = irq.vector;
5230 * svm_update_pi_irte - set IRTE for Posted-Interrupts
5233 * @host_irq: host irq of the interrupt
5234 * @guest_irq: gsi of the interrupt
5235 * @set: set or unset PI
5236 * returns 0 on success, < 0 on failure
5238 static int svm_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
5239 uint32_t guest_irq, bool set)
5241 struct kvm_kernel_irq_routing_entry *e;
5242 struct kvm_irq_routing_table *irq_rt;
5243 int idx, ret = -EINVAL;
5245 if (!kvm_arch_has_assigned_device(kvm) ||
5246 !irq_remapping_cap(IRQ_POSTING_CAP))
5249 pr_debug("SVM: %s: host_irq=%#x, guest_irq=%#x, set=%#x\n",
5250 __func__, host_irq, guest_irq, set);
5252 idx = srcu_read_lock(&kvm->irq_srcu);
5253 irq_rt = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu);
5254 WARN_ON(guest_irq >= irq_rt->nr_rt_entries);
5256 hlist_for_each_entry(e, &irq_rt->map[guest_irq], link) {
5257 struct vcpu_data vcpu_info;
5258 struct vcpu_svm *svm = NULL;
5260 if (e->type != KVM_IRQ_ROUTING_MSI)
5264 * Here, we setup with legacy mode in the following cases:
5265 * 1. When cannot target interrupt to a specific vcpu.
5266 * 2. Unsetting posted interrupt.
5267 * 3. APIC virtialization is disabled for the vcpu.
5269 if (!get_pi_vcpu_info(kvm, e, &vcpu_info, &svm) && set &&
5270 kvm_vcpu_apicv_active(&svm->vcpu)) {
5271 struct amd_iommu_pi_data pi;
5273 /* Try to enable guest_mode in IRTE */
5274 pi.base = __sme_set(page_to_phys(svm->avic_backing_page) &
5276 pi.ga_tag = AVIC_GATAG(to_kvm_svm(kvm)->avic_vm_id,
5278 pi.is_guest_mode = true;
5279 pi.vcpu_data = &vcpu_info;
5280 ret = irq_set_vcpu_affinity(host_irq, &pi);
5283 * Here, we successfully setting up vcpu affinity in
5284 * IOMMU guest mode. Now, we need to store the posted
5285 * interrupt information in a per-vcpu ir_list so that
5286 * we can reference to them directly when we update vcpu
5287 * scheduling information in IOMMU irte.
5289 if (!ret && pi.is_guest_mode)
5290 svm_ir_list_add(svm, &pi);
5292 /* Use legacy mode in IRTE */
5293 struct amd_iommu_pi_data pi;
5296 * Here, pi is used to:
5297 * - Tell IOMMU to use legacy mode for this interrupt.
5298 * - Retrieve ga_tag of prior interrupt remapping data.
5300 pi.is_guest_mode = false;
5301 ret = irq_set_vcpu_affinity(host_irq, &pi);
5304 * Check if the posted interrupt was previously
5305 * setup with the guest_mode by checking if the ga_tag
5306 * was cached. If so, we need to clean up the per-vcpu
5309 if (!ret && pi.prev_ga_tag) {
5310 int id = AVIC_GATAG_TO_VCPUID(pi.prev_ga_tag);
5311 struct kvm_vcpu *vcpu;
5313 vcpu = kvm_get_vcpu_by_id(kvm, id);
5315 svm_ir_list_del(to_svm(vcpu), &pi);
5320 trace_kvm_pi_irte_update(host_irq, svm->vcpu.vcpu_id,
5321 e->gsi, vcpu_info.vector,
5322 vcpu_info.pi_desc_addr, set);
5326 pr_err("%s: failed to update PI IRTE\n", __func__);
5333 srcu_read_unlock(&kvm->irq_srcu, idx);
5337 static int svm_nmi_allowed(struct kvm_vcpu *vcpu)
5339 struct vcpu_svm *svm = to_svm(vcpu);
5340 struct vmcb *vmcb = svm->vmcb;
5342 ret = !(vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK) &&
5343 !(svm->vcpu.arch.hflags & HF_NMI_MASK);
5344 ret = ret && gif_set(svm) && nested_svm_nmi(svm);
5349 static bool svm_get_nmi_mask(struct kvm_vcpu *vcpu)
5351 struct vcpu_svm *svm = to_svm(vcpu);
5353 return !!(svm->vcpu.arch.hflags & HF_NMI_MASK);
5356 static void svm_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
5358 struct vcpu_svm *svm = to_svm(vcpu);
5361 svm->vcpu.arch.hflags |= HF_NMI_MASK;
5362 set_intercept(svm, INTERCEPT_IRET);
5364 svm->vcpu.arch.hflags &= ~HF_NMI_MASK;
5365 clr_intercept(svm, INTERCEPT_IRET);
5369 static int svm_interrupt_allowed(struct kvm_vcpu *vcpu)
5371 struct vcpu_svm *svm = to_svm(vcpu);
5372 struct vmcb *vmcb = svm->vmcb;
5375 if (!gif_set(svm) ||
5376 (vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK))
5379 ret = !!(kvm_get_rflags(vcpu) & X86_EFLAGS_IF);
5381 if (is_guest_mode(vcpu))
5382 return ret && !(svm->vcpu.arch.hflags & HF_VINTR_MASK);
5387 static void enable_irq_window(struct kvm_vcpu *vcpu)
5389 struct vcpu_svm *svm = to_svm(vcpu);
5391 if (kvm_vcpu_apicv_active(vcpu))
5395 * In case GIF=0 we can't rely on the CPU to tell us when GIF becomes
5396 * 1, because that's a separate STGI/VMRUN intercept. The next time we
5397 * get that intercept, this function will be called again though and
5398 * we'll get the vintr intercept. However, if the vGIF feature is
5399 * enabled, the STGI interception will not occur. Enable the irq
5400 * window under the assumption that the hardware will set the GIF.
5402 if ((vgif_enabled(svm) || gif_set(svm)) && nested_svm_intr(svm)) {
5404 svm_inject_irq(svm, 0x0);
5408 static void enable_nmi_window(struct kvm_vcpu *vcpu)
5410 struct vcpu_svm *svm = to_svm(vcpu);
5412 if ((svm->vcpu.arch.hflags & (HF_NMI_MASK | HF_IRET_MASK))
5414 return; /* IRET will cause a vm exit */
5416 if (!gif_set(svm)) {
5417 if (vgif_enabled(svm))
5418 set_intercept(svm, INTERCEPT_STGI);
5419 return; /* STGI will cause a vm exit */
5422 if (svm->nested.exit_required)
5423 return; /* we're not going to run the guest yet */
5426 * Something prevents NMI from been injected. Single step over possible
5427 * problem (IRET or exception injection or interrupt shadow)
5429 svm->nmi_singlestep_guest_rflags = svm_get_rflags(vcpu);
5430 svm->nmi_singlestep = true;
5431 svm->vmcb->save.rflags |= (X86_EFLAGS_TF | X86_EFLAGS_RF);
5434 static int svm_set_tss_addr(struct kvm *kvm, unsigned int addr)
5439 static int svm_set_identity_map_addr(struct kvm *kvm, u64 ident_addr)
5444 static void svm_flush_tlb(struct kvm_vcpu *vcpu, bool invalidate_gpa)
5446 struct vcpu_svm *svm = to_svm(vcpu);
5448 if (static_cpu_has(X86_FEATURE_FLUSHBYASID))
5449 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ASID;
5451 svm->asid_generation--;
5454 static void svm_flush_tlb_gva(struct kvm_vcpu *vcpu, gva_t gva)
5456 struct vcpu_svm *svm = to_svm(vcpu);
5458 invlpga(gva, svm->vmcb->control.asid);
5461 static void svm_prepare_guest_switch(struct kvm_vcpu *vcpu)
5465 static inline void sync_cr8_to_lapic(struct kvm_vcpu *vcpu)
5467 struct vcpu_svm *svm = to_svm(vcpu);
5469 if (svm_nested_virtualize_tpr(vcpu))
5472 if (!is_cr_intercept(svm, INTERCEPT_CR8_WRITE)) {
5473 int cr8 = svm->vmcb->control.int_ctl & V_TPR_MASK;
5474 kvm_set_cr8(vcpu, cr8);
5478 static inline void sync_lapic_to_cr8(struct kvm_vcpu *vcpu)
5480 struct vcpu_svm *svm = to_svm(vcpu);
5483 if (svm_nested_virtualize_tpr(vcpu) ||
5484 kvm_vcpu_apicv_active(vcpu))
5487 cr8 = kvm_get_cr8(vcpu);
5488 svm->vmcb->control.int_ctl &= ~V_TPR_MASK;
5489 svm->vmcb->control.int_ctl |= cr8 & V_TPR_MASK;
5492 static void svm_complete_interrupts(struct vcpu_svm *svm)
5496 u32 exitintinfo = svm->vmcb->control.exit_int_info;
5497 unsigned int3_injected = svm->int3_injected;
5499 svm->int3_injected = 0;
5502 * If we've made progress since setting HF_IRET_MASK, we've
5503 * executed an IRET and can allow NMI injection.
5505 if ((svm->vcpu.arch.hflags & HF_IRET_MASK)
5506 && kvm_rip_read(&svm->vcpu) != svm->nmi_iret_rip) {
5507 svm->vcpu.arch.hflags &= ~(HF_NMI_MASK | HF_IRET_MASK);
5508 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
5511 svm->vcpu.arch.nmi_injected = false;
5512 kvm_clear_exception_queue(&svm->vcpu);
5513 kvm_clear_interrupt_queue(&svm->vcpu);
5515 if (!(exitintinfo & SVM_EXITINTINFO_VALID))
5518 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
5520 vector = exitintinfo & SVM_EXITINTINFO_VEC_MASK;
5521 type = exitintinfo & SVM_EXITINTINFO_TYPE_MASK;
5524 case SVM_EXITINTINFO_TYPE_NMI:
5525 svm->vcpu.arch.nmi_injected = true;
5527 case SVM_EXITINTINFO_TYPE_EXEPT:
5529 * In case of software exceptions, do not reinject the vector,
5530 * but re-execute the instruction instead. Rewind RIP first
5531 * if we emulated INT3 before.
5533 if (kvm_exception_is_soft(vector)) {
5534 if (vector == BP_VECTOR && int3_injected &&
5535 kvm_is_linear_rip(&svm->vcpu, svm->int3_rip))
5536 kvm_rip_write(&svm->vcpu,
5537 kvm_rip_read(&svm->vcpu) -
5541 if (exitintinfo & SVM_EXITINTINFO_VALID_ERR) {
5542 u32 err = svm->vmcb->control.exit_int_info_err;
5543 kvm_requeue_exception_e(&svm->vcpu, vector, err);
5546 kvm_requeue_exception(&svm->vcpu, vector);
5548 case SVM_EXITINTINFO_TYPE_INTR:
5549 kvm_queue_interrupt(&svm->vcpu, vector, false);
5556 static void svm_cancel_injection(struct kvm_vcpu *vcpu)
5558 struct vcpu_svm *svm = to_svm(vcpu);
5559 struct vmcb_control_area *control = &svm->vmcb->control;
5561 control->exit_int_info = control->event_inj;
5562 control->exit_int_info_err = control->event_inj_err;
5563 control->event_inj = 0;
5564 svm_complete_interrupts(svm);
5567 static void svm_vcpu_run(struct kvm_vcpu *vcpu)
5569 struct vcpu_svm *svm = to_svm(vcpu);
5571 svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX];
5572 svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP];
5573 svm->vmcb->save.rip = vcpu->arch.regs[VCPU_REGS_RIP];
5576 * A vmexit emulation is required before the vcpu can be executed
5579 if (unlikely(svm->nested.exit_required))
5583 * Disable singlestep if we're injecting an interrupt/exception.
5584 * We don't want our modified rflags to be pushed on the stack where
5585 * we might not be able to easily reset them if we disabled NMI
5588 if (svm->nmi_singlestep && svm->vmcb->control.event_inj) {
5590 * Event injection happens before external interrupts cause a
5591 * vmexit and interrupts are disabled here, so smp_send_reschedule
5592 * is enough to force an immediate vmexit.
5594 disable_nmi_singlestep(svm);
5595 smp_send_reschedule(vcpu->cpu);
5600 sync_lapic_to_cr8(vcpu);
5602 svm->vmcb->save.cr2 = vcpu->arch.cr2;
5607 * If this vCPU has touched SPEC_CTRL, restore the guest's value if
5608 * it's non-zero. Since vmentry is serialising on affected CPUs, there
5609 * is no need to worry about the conditional branch over the wrmsr
5610 * being speculatively taken.
5612 x86_spec_ctrl_set_guest(svm->spec_ctrl, svm->virt_spec_ctrl);
5617 "push %%" _ASM_BP "; \n\t"
5618 "mov %c[rbx](%[svm]), %%" _ASM_BX " \n\t"
5619 "mov %c[rcx](%[svm]), %%" _ASM_CX " \n\t"
5620 "mov %c[rdx](%[svm]), %%" _ASM_DX " \n\t"
5621 "mov %c[rsi](%[svm]), %%" _ASM_SI " \n\t"
5622 "mov %c[rdi](%[svm]), %%" _ASM_DI " \n\t"
5623 "mov %c[rbp](%[svm]), %%" _ASM_BP " \n\t"
5624 #ifdef CONFIG_X86_64
5625 "mov %c[r8](%[svm]), %%r8 \n\t"
5626 "mov %c[r9](%[svm]), %%r9 \n\t"
5627 "mov %c[r10](%[svm]), %%r10 \n\t"
5628 "mov %c[r11](%[svm]), %%r11 \n\t"
5629 "mov %c[r12](%[svm]), %%r12 \n\t"
5630 "mov %c[r13](%[svm]), %%r13 \n\t"
5631 "mov %c[r14](%[svm]), %%r14 \n\t"
5632 "mov %c[r15](%[svm]), %%r15 \n\t"
5635 /* Enter guest mode */
5636 "push %%" _ASM_AX " \n\t"
5637 "mov %c[vmcb](%[svm]), %%" _ASM_AX " \n\t"
5638 __ex(SVM_VMLOAD) "\n\t"
5639 __ex(SVM_VMRUN) "\n\t"
5640 __ex(SVM_VMSAVE) "\n\t"
5641 "pop %%" _ASM_AX " \n\t"
5643 /* Save guest registers, load host registers */
5644 "mov %%" _ASM_BX ", %c[rbx](%[svm]) \n\t"
5645 "mov %%" _ASM_CX ", %c[rcx](%[svm]) \n\t"
5646 "mov %%" _ASM_DX ", %c[rdx](%[svm]) \n\t"
5647 "mov %%" _ASM_SI ", %c[rsi](%[svm]) \n\t"
5648 "mov %%" _ASM_DI ", %c[rdi](%[svm]) \n\t"
5649 "mov %%" _ASM_BP ", %c[rbp](%[svm]) \n\t"
5650 #ifdef CONFIG_X86_64
5651 "mov %%r8, %c[r8](%[svm]) \n\t"
5652 "mov %%r9, %c[r9](%[svm]) \n\t"
5653 "mov %%r10, %c[r10](%[svm]) \n\t"
5654 "mov %%r11, %c[r11](%[svm]) \n\t"
5655 "mov %%r12, %c[r12](%[svm]) \n\t"
5656 "mov %%r13, %c[r13](%[svm]) \n\t"
5657 "mov %%r14, %c[r14](%[svm]) \n\t"
5658 "mov %%r15, %c[r15](%[svm]) \n\t"
5660 * Clear host registers marked as clobbered to prevent
5663 "xor %%r8d, %%r8d \n\t"
5664 "xor %%r9d, %%r9d \n\t"
5665 "xor %%r10d, %%r10d \n\t"
5666 "xor %%r11d, %%r11d \n\t"
5667 "xor %%r12d, %%r12d \n\t"
5668 "xor %%r13d, %%r13d \n\t"
5669 "xor %%r14d, %%r14d \n\t"
5670 "xor %%r15d, %%r15d \n\t"
5672 "xor %%ebx, %%ebx \n\t"
5673 "xor %%ecx, %%ecx \n\t"
5674 "xor %%edx, %%edx \n\t"
5675 "xor %%esi, %%esi \n\t"
5676 "xor %%edi, %%edi \n\t"
5680 [vmcb]"i"(offsetof(struct vcpu_svm, vmcb_pa)),
5681 [rbx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBX])),
5682 [rcx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RCX])),
5683 [rdx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDX])),
5684 [rsi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RSI])),
5685 [rdi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDI])),
5686 [rbp]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBP]))
5687 #ifdef CONFIG_X86_64
5688 , [r8]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R8])),
5689 [r9]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R9])),
5690 [r10]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R10])),
5691 [r11]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R11])),
5692 [r12]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R12])),
5693 [r13]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R13])),
5694 [r14]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R14])),
5695 [r15]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R15]))
5698 #ifdef CONFIG_X86_64
5699 , "rbx", "rcx", "rdx", "rsi", "rdi"
5700 , "r8", "r9", "r10", "r11" , "r12", "r13", "r14", "r15"
5702 , "ebx", "ecx", "edx", "esi", "edi"
5706 /* Eliminate branch target predictions from guest mode */
5709 #ifdef CONFIG_X86_64
5710 wrmsrl(MSR_GS_BASE, svm->host.gs_base);
5712 loadsegment(fs, svm->host.fs);
5713 #ifndef CONFIG_X86_32_LAZY_GS
5714 loadsegment(gs, svm->host.gs);
5719 * We do not use IBRS in the kernel. If this vCPU has used the
5720 * SPEC_CTRL MSR it may have left it on; save the value and
5721 * turn it off. This is much more efficient than blindly adding
5722 * it to the atomic save/restore list. Especially as the former
5723 * (Saving guest MSRs on vmexit) doesn't even exist in KVM.
5725 * For non-nested case:
5726 * If the L01 MSR bitmap does not intercept the MSR, then we need to
5730 * If the L02 MSR bitmap does not intercept the MSR, then we need to
5733 if (unlikely(!msr_write_intercepted(vcpu, MSR_IA32_SPEC_CTRL)))
5734 svm->spec_ctrl = native_read_msr(MSR_IA32_SPEC_CTRL);
5738 local_irq_disable();
5740 x86_spec_ctrl_restore_host(svm->spec_ctrl, svm->virt_spec_ctrl);
5742 vcpu->arch.cr2 = svm->vmcb->save.cr2;
5743 vcpu->arch.regs[VCPU_REGS_RAX] = svm->vmcb->save.rax;
5744 vcpu->arch.regs[VCPU_REGS_RSP] = svm->vmcb->save.rsp;
5745 vcpu->arch.regs[VCPU_REGS_RIP] = svm->vmcb->save.rip;
5747 if (unlikely(svm->vmcb->control.exit_code == SVM_EXIT_NMI))
5748 kvm_before_interrupt(&svm->vcpu);
5752 /* Any pending NMI will happen here */
5754 if (unlikely(svm->vmcb->control.exit_code == SVM_EXIT_NMI))
5755 kvm_after_interrupt(&svm->vcpu);
5757 sync_cr8_to_lapic(vcpu);
5761 svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
5763 /* if exit due to PF check for async PF */
5764 if (svm->vmcb->control.exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR)
5765 svm->vcpu.arch.apf.host_apf_reason = kvm_read_and_reset_pf_reason();
5768 vcpu->arch.regs_avail &= ~(1 << VCPU_EXREG_PDPTR);
5769 vcpu->arch.regs_dirty &= ~(1 << VCPU_EXREG_PDPTR);
5773 * We need to handle MC intercepts here before the vcpu has a chance to
5774 * change the physical cpu
5776 if (unlikely(svm->vmcb->control.exit_code ==
5777 SVM_EXIT_EXCP_BASE + MC_VECTOR))
5778 svm_handle_mce(svm);
5780 mark_all_clean(svm->vmcb);
5782 STACK_FRAME_NON_STANDARD(svm_vcpu_run);
5784 static void svm_set_cr3(struct kvm_vcpu *vcpu, unsigned long root)
5786 struct vcpu_svm *svm = to_svm(vcpu);
5788 svm->vmcb->save.cr3 = __sme_set(root);
5789 mark_dirty(svm->vmcb, VMCB_CR);
5792 static void set_tdp_cr3(struct kvm_vcpu *vcpu, unsigned long root)
5794 struct vcpu_svm *svm = to_svm(vcpu);
5796 svm->vmcb->control.nested_cr3 = __sme_set(root);
5797 mark_dirty(svm->vmcb, VMCB_NPT);
5799 /* Also sync guest cr3 here in case we live migrate */
5800 svm->vmcb->save.cr3 = kvm_read_cr3(vcpu);
5801 mark_dirty(svm->vmcb, VMCB_CR);
5804 static int is_disabled(void)
5808 rdmsrl(MSR_VM_CR, vm_cr);
5809 if (vm_cr & (1 << SVM_VM_CR_SVM_DISABLE))
5816 svm_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
5819 * Patch in the VMMCALL instruction:
5821 hypercall[0] = 0x0f;
5822 hypercall[1] = 0x01;
5823 hypercall[2] = 0xd9;
5826 static void svm_check_processor_compat(void *rtn)
5831 static bool svm_cpu_has_accelerated_tpr(void)
5836 static bool svm_has_emulated_msr(int index)
5841 static u64 svm_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
5846 static void svm_cpuid_update(struct kvm_vcpu *vcpu)
5848 struct vcpu_svm *svm = to_svm(vcpu);
5850 /* Update nrips enabled cache */
5851 svm->nrips_enabled = !!guest_cpuid_has(&svm->vcpu, X86_FEATURE_NRIPS);
5853 if (!kvm_vcpu_apicv_active(vcpu))
5856 guest_cpuid_clear(vcpu, X86_FEATURE_X2APIC);
5859 static void svm_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
5864 entry->ecx &= ~bit(X86_FEATURE_X2APIC);
5868 entry->ecx |= (1 << 2); /* Set SVM bit */
5871 entry->eax = 1; /* SVM revision 1 */
5872 entry->ebx = 8; /* Lets support 8 ASIDs in case we add proper
5873 ASID emulation to nested SVM */
5874 entry->ecx = 0; /* Reserved */
5875 entry->edx = 0; /* Per default do not support any
5876 additional features */
5878 /* Support next_rip if host supports it */
5879 if (boot_cpu_has(X86_FEATURE_NRIPS))
5880 entry->edx |= SVM_FEATURE_NRIP;
5882 /* Support NPT for the guest if enabled */
5884 entry->edx |= SVM_FEATURE_NPT;
5888 /* Support memory encryption cpuid if host supports it */
5889 if (boot_cpu_has(X86_FEATURE_SEV))
5890 cpuid(0x8000001f, &entry->eax, &entry->ebx,
5891 &entry->ecx, &entry->edx);
5896 static int svm_get_lpage_level(void)
5898 return PT_PDPE_LEVEL;
5901 static bool svm_rdtscp_supported(void)
5903 return boot_cpu_has(X86_FEATURE_RDTSCP);
5906 static bool svm_invpcid_supported(void)
5911 static bool svm_mpx_supported(void)
5916 static bool svm_xsaves_supported(void)
5921 static bool svm_umip_emulated(void)
5926 static bool svm_has_wbinvd_exit(void)
5931 #define PRE_EX(exit) { .exit_code = (exit), \
5932 .stage = X86_ICPT_PRE_EXCEPT, }
5933 #define POST_EX(exit) { .exit_code = (exit), \
5934 .stage = X86_ICPT_POST_EXCEPT, }
5935 #define POST_MEM(exit) { .exit_code = (exit), \
5936 .stage = X86_ICPT_POST_MEMACCESS, }
5938 static const struct __x86_intercept {
5940 enum x86_intercept_stage stage;
5941 } x86_intercept_map[] = {
5942 [x86_intercept_cr_read] = POST_EX(SVM_EXIT_READ_CR0),
5943 [x86_intercept_cr_write] = POST_EX(SVM_EXIT_WRITE_CR0),
5944 [x86_intercept_clts] = POST_EX(SVM_EXIT_WRITE_CR0),
5945 [x86_intercept_lmsw] = POST_EX(SVM_EXIT_WRITE_CR0),
5946 [x86_intercept_smsw] = POST_EX(SVM_EXIT_READ_CR0),
5947 [x86_intercept_dr_read] = POST_EX(SVM_EXIT_READ_DR0),
5948 [x86_intercept_dr_write] = POST_EX(SVM_EXIT_WRITE_DR0),
5949 [x86_intercept_sldt] = POST_EX(SVM_EXIT_LDTR_READ),
5950 [x86_intercept_str] = POST_EX(SVM_EXIT_TR_READ),
5951 [x86_intercept_lldt] = POST_EX(SVM_EXIT_LDTR_WRITE),
5952 [x86_intercept_ltr] = POST_EX(SVM_EXIT_TR_WRITE),
5953 [x86_intercept_sgdt] = POST_EX(SVM_EXIT_GDTR_READ),
5954 [x86_intercept_sidt] = POST_EX(SVM_EXIT_IDTR_READ),
5955 [x86_intercept_lgdt] = POST_EX(SVM_EXIT_GDTR_WRITE),
5956 [x86_intercept_lidt] = POST_EX(SVM_EXIT_IDTR_WRITE),
5957 [x86_intercept_vmrun] = POST_EX(SVM_EXIT_VMRUN),
5958 [x86_intercept_vmmcall] = POST_EX(SVM_EXIT_VMMCALL),
5959 [x86_intercept_vmload] = POST_EX(SVM_EXIT_VMLOAD),
5960 [x86_intercept_vmsave] = POST_EX(SVM_EXIT_VMSAVE),
5961 [x86_intercept_stgi] = POST_EX(SVM_EXIT_STGI),
5962 [x86_intercept_clgi] = POST_EX(SVM_EXIT_CLGI),
5963 [x86_intercept_skinit] = POST_EX(SVM_EXIT_SKINIT),
5964 [x86_intercept_invlpga] = POST_EX(SVM_EXIT_INVLPGA),
5965 [x86_intercept_rdtscp] = POST_EX(SVM_EXIT_RDTSCP),
5966 [x86_intercept_monitor] = POST_MEM(SVM_EXIT_MONITOR),
5967 [x86_intercept_mwait] = POST_EX(SVM_EXIT_MWAIT),
5968 [x86_intercept_invlpg] = POST_EX(SVM_EXIT_INVLPG),
5969 [x86_intercept_invd] = POST_EX(SVM_EXIT_INVD),
5970 [x86_intercept_wbinvd] = POST_EX(SVM_EXIT_WBINVD),
5971 [x86_intercept_wrmsr] = POST_EX(SVM_EXIT_MSR),
5972 [x86_intercept_rdtsc] = POST_EX(SVM_EXIT_RDTSC),
5973 [x86_intercept_rdmsr] = POST_EX(SVM_EXIT_MSR),
5974 [x86_intercept_rdpmc] = POST_EX(SVM_EXIT_RDPMC),
5975 [x86_intercept_cpuid] = PRE_EX(SVM_EXIT_CPUID),
5976 [x86_intercept_rsm] = PRE_EX(SVM_EXIT_RSM),
5977 [x86_intercept_pause] = PRE_EX(SVM_EXIT_PAUSE),
5978 [x86_intercept_pushf] = PRE_EX(SVM_EXIT_PUSHF),
5979 [x86_intercept_popf] = PRE_EX(SVM_EXIT_POPF),
5980 [x86_intercept_intn] = PRE_EX(SVM_EXIT_SWINT),
5981 [x86_intercept_iret] = PRE_EX(SVM_EXIT_IRET),
5982 [x86_intercept_icebp] = PRE_EX(SVM_EXIT_ICEBP),
5983 [x86_intercept_hlt] = POST_EX(SVM_EXIT_HLT),
5984 [x86_intercept_in] = POST_EX(SVM_EXIT_IOIO),
5985 [x86_intercept_ins] = POST_EX(SVM_EXIT_IOIO),
5986 [x86_intercept_out] = POST_EX(SVM_EXIT_IOIO),
5987 [x86_intercept_outs] = POST_EX(SVM_EXIT_IOIO),
5994 static int svm_check_intercept(struct kvm_vcpu *vcpu,
5995 struct x86_instruction_info *info,
5996 enum x86_intercept_stage stage)
5998 struct vcpu_svm *svm = to_svm(vcpu);
5999 int vmexit, ret = X86EMUL_CONTINUE;
6000 struct __x86_intercept icpt_info;
6001 struct vmcb *vmcb = svm->vmcb;
6003 if (info->intercept >= ARRAY_SIZE(x86_intercept_map))
6006 icpt_info = x86_intercept_map[info->intercept];
6008 if (stage != icpt_info.stage)
6011 switch (icpt_info.exit_code) {
6012 case SVM_EXIT_READ_CR0:
6013 if (info->intercept == x86_intercept_cr_read)
6014 icpt_info.exit_code += info->modrm_reg;
6016 case SVM_EXIT_WRITE_CR0: {
6017 unsigned long cr0, val;
6020 if (info->intercept == x86_intercept_cr_write)
6021 icpt_info.exit_code += info->modrm_reg;
6023 if (icpt_info.exit_code != SVM_EXIT_WRITE_CR0 ||
6024 info->intercept == x86_intercept_clts)
6027 intercept = svm->nested.intercept;
6029 if (!(intercept & (1ULL << INTERCEPT_SELECTIVE_CR0)))
6032 cr0 = vcpu->arch.cr0 & ~SVM_CR0_SELECTIVE_MASK;
6033 val = info->src_val & ~SVM_CR0_SELECTIVE_MASK;
6035 if (info->intercept == x86_intercept_lmsw) {
6038 /* lmsw can't clear PE - catch this here */
6039 if (cr0 & X86_CR0_PE)
6044 icpt_info.exit_code = SVM_EXIT_CR0_SEL_WRITE;
6048 case SVM_EXIT_READ_DR0:
6049 case SVM_EXIT_WRITE_DR0:
6050 icpt_info.exit_code += info->modrm_reg;
6053 if (info->intercept == x86_intercept_wrmsr)
6054 vmcb->control.exit_info_1 = 1;
6056 vmcb->control.exit_info_1 = 0;
6058 case SVM_EXIT_PAUSE:
6060 * We get this for NOP only, but pause
6061 * is rep not, check this here
6063 if (info->rep_prefix != REPE_PREFIX)
6066 case SVM_EXIT_IOIO: {
6070 if (info->intercept == x86_intercept_in ||
6071 info->intercept == x86_intercept_ins) {
6072 exit_info = ((info->src_val & 0xffff) << 16) |
6074 bytes = info->dst_bytes;
6076 exit_info = (info->dst_val & 0xffff) << 16;
6077 bytes = info->src_bytes;
6080 if (info->intercept == x86_intercept_outs ||
6081 info->intercept == x86_intercept_ins)
6082 exit_info |= SVM_IOIO_STR_MASK;
6084 if (info->rep_prefix)
6085 exit_info |= SVM_IOIO_REP_MASK;
6087 bytes = min(bytes, 4u);
6089 exit_info |= bytes << SVM_IOIO_SIZE_SHIFT;
6091 exit_info |= (u32)info->ad_bytes << (SVM_IOIO_ASIZE_SHIFT - 1);
6093 vmcb->control.exit_info_1 = exit_info;
6094 vmcb->control.exit_info_2 = info->next_rip;
6102 /* TODO: Advertise NRIPS to guest hypervisor unconditionally */
6103 if (static_cpu_has(X86_FEATURE_NRIPS))
6104 vmcb->control.next_rip = info->next_rip;
6105 vmcb->control.exit_code = icpt_info.exit_code;
6106 vmexit = nested_svm_exit_handled(svm);
6108 ret = (vmexit == NESTED_EXIT_DONE) ? X86EMUL_INTERCEPTED
6115 static void svm_handle_external_intr(struct kvm_vcpu *vcpu)
6119 * We must have an instruction with interrupts enabled, so
6120 * the timer interrupt isn't delayed by the interrupt shadow.
6123 local_irq_disable();
6126 static void svm_sched_in(struct kvm_vcpu *vcpu, int cpu)
6128 if (pause_filter_thresh)
6129 shrink_ple_window(vcpu);
6132 static inline void avic_post_state_restore(struct kvm_vcpu *vcpu)
6134 if (avic_handle_apic_id_update(vcpu) != 0)
6136 if (avic_handle_dfr_update(vcpu) != 0)
6138 avic_handle_ldr_update(vcpu);
6141 static void svm_setup_mce(struct kvm_vcpu *vcpu)
6143 /* [63:9] are reserved. */
6144 vcpu->arch.mcg_cap &= 0x1ff;
6147 static int svm_smi_allowed(struct kvm_vcpu *vcpu)
6149 struct vcpu_svm *svm = to_svm(vcpu);
6151 /* Per APM Vol.2 15.22.2 "Response to SMI" */
6155 if (is_guest_mode(&svm->vcpu) &&
6156 svm->nested.intercept & (1ULL << INTERCEPT_SMI)) {
6157 /* TODO: Might need to set exit_info_1 and exit_info_2 here */
6158 svm->vmcb->control.exit_code = SVM_EXIT_SMI;
6159 svm->nested.exit_required = true;
6166 static int svm_pre_enter_smm(struct kvm_vcpu *vcpu, char *smstate)
6168 struct vcpu_svm *svm = to_svm(vcpu);
6171 if (is_guest_mode(vcpu)) {
6172 /* FED8h - SVM Guest */
6173 put_smstate(u64, smstate, 0x7ed8, 1);
6174 /* FEE0h - SVM Guest VMCB Physical Address */
6175 put_smstate(u64, smstate, 0x7ee0, svm->nested.vmcb);
6177 svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX];
6178 svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP];
6179 svm->vmcb->save.rip = vcpu->arch.regs[VCPU_REGS_RIP];
6181 ret = nested_svm_vmexit(svm);
6188 static int svm_pre_leave_smm(struct kvm_vcpu *vcpu, u64 smbase)
6190 struct vcpu_svm *svm = to_svm(vcpu);
6191 struct vmcb *nested_vmcb;
6199 ret = kvm_vcpu_read_guest(vcpu, smbase + 0xfed8, &svm_state_save,
6200 sizeof(svm_state_save));
6204 if (svm_state_save.guest) {
6205 vcpu->arch.hflags &= ~HF_SMM_MASK;
6206 nested_vmcb = nested_svm_map(svm, svm_state_save.vmcb, &page);
6208 enter_svm_guest_mode(svm, svm_state_save.vmcb, nested_vmcb, page);
6211 vcpu->arch.hflags |= HF_SMM_MASK;
6216 static int enable_smi_window(struct kvm_vcpu *vcpu)
6218 struct vcpu_svm *svm = to_svm(vcpu);
6220 if (!gif_set(svm)) {
6221 if (vgif_enabled(svm))
6222 set_intercept(svm, INTERCEPT_STGI);
6223 /* STGI will cause a vm exit */
6229 static int sev_asid_new(void)
6234 * SEV-enabled guest must use asid from min_sev_asid to max_sev_asid.
6236 pos = find_next_zero_bit(sev_asid_bitmap, max_sev_asid, min_sev_asid - 1);
6237 if (pos >= max_sev_asid)
6240 set_bit(pos, sev_asid_bitmap);
6244 static int sev_guest_init(struct kvm *kvm, struct kvm_sev_cmd *argp)
6246 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
6250 asid = sev_asid_new();
6254 ret = sev_platform_init(&argp->error);
6260 INIT_LIST_HEAD(&sev->regions_list);
6265 __sev_asid_free(asid);
6269 static int sev_bind_asid(struct kvm *kvm, unsigned int handle, int *error)
6271 struct sev_data_activate *data;
6272 int asid = sev_get_asid(kvm);
6275 wbinvd_on_all_cpus();
6277 ret = sev_guest_df_flush(error);
6281 data = kzalloc(sizeof(*data), GFP_KERNEL);
6285 /* activate ASID on the given handle */
6286 data->handle = handle;
6288 ret = sev_guest_activate(data, error);
6294 static int __sev_issue_cmd(int fd, int id, void *data, int *error)
6303 ret = sev_issue_cmd_external_user(f.file, id, data, error);
6309 static int sev_issue_cmd(struct kvm *kvm, int id, void *data, int *error)
6311 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
6313 return __sev_issue_cmd(sev->fd, id, data, error);
6316 static int sev_launch_start(struct kvm *kvm, struct kvm_sev_cmd *argp)
6318 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
6319 struct sev_data_launch_start *start;
6320 struct kvm_sev_launch_start params;
6321 void *dh_blob, *session_blob;
6322 int *error = &argp->error;
6325 if (!sev_guest(kvm))
6328 if (copy_from_user(¶ms, (void __user *)(uintptr_t)argp->data, sizeof(params)))
6331 start = kzalloc(sizeof(*start), GFP_KERNEL);
6336 if (params.dh_uaddr) {
6337 dh_blob = psp_copy_user_blob(params.dh_uaddr, params.dh_len);
6338 if (IS_ERR(dh_blob)) {
6339 ret = PTR_ERR(dh_blob);
6343 start->dh_cert_address = __sme_set(__pa(dh_blob));
6344 start->dh_cert_len = params.dh_len;
6347 session_blob = NULL;
6348 if (params.session_uaddr) {
6349 session_blob = psp_copy_user_blob(params.session_uaddr, params.session_len);
6350 if (IS_ERR(session_blob)) {
6351 ret = PTR_ERR(session_blob);
6355 start->session_address = __sme_set(__pa(session_blob));
6356 start->session_len = params.session_len;
6359 start->handle = params.handle;
6360 start->policy = params.policy;
6362 /* create memory encryption context */
6363 ret = __sev_issue_cmd(argp->sev_fd, SEV_CMD_LAUNCH_START, start, error);
6365 goto e_free_session;
6367 /* Bind ASID to this guest */
6368 ret = sev_bind_asid(kvm, start->handle, error);
6370 goto e_free_session;
6372 /* return handle to userspace */
6373 params.handle = start->handle;
6374 if (copy_to_user((void __user *)(uintptr_t)argp->data, ¶ms, sizeof(params))) {
6375 sev_unbind_asid(kvm, start->handle);
6377 goto e_free_session;
6380 sev->handle = start->handle;
6381 sev->fd = argp->sev_fd;
6384 kfree(session_blob);
6392 static int get_num_contig_pages(int idx, struct page **inpages,
6393 unsigned long npages)
6395 unsigned long paddr, next_paddr;
6396 int i = idx + 1, pages = 1;
6398 /* find the number of contiguous pages starting from idx */
6399 paddr = __sme_page_pa(inpages[idx]);
6400 while (i < npages) {
6401 next_paddr = __sme_page_pa(inpages[i++]);
6402 if ((paddr + PAGE_SIZE) == next_paddr) {
6413 static int sev_launch_update_data(struct kvm *kvm, struct kvm_sev_cmd *argp)
6415 unsigned long vaddr, vaddr_end, next_vaddr, npages, size;
6416 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
6417 struct kvm_sev_launch_update_data params;
6418 struct sev_data_launch_update_data *data;
6419 struct page **inpages;
6422 if (!sev_guest(kvm))
6425 if (copy_from_user(¶ms, (void __user *)(uintptr_t)argp->data, sizeof(params)))
6428 data = kzalloc(sizeof(*data), GFP_KERNEL);
6432 vaddr = params.uaddr;
6434 vaddr_end = vaddr + size;
6436 /* Lock the user memory. */
6437 inpages = sev_pin_memory(kvm, vaddr, size, &npages, 1);
6444 * The LAUNCH_UPDATE command will perform in-place encryption of the
6445 * memory content (i.e it will write the same memory region with C=1).
6446 * It's possible that the cache may contain the data with C=0, i.e.,
6447 * unencrypted so invalidate it first.
6449 sev_clflush_pages(inpages, npages);
6451 for (i = 0; vaddr < vaddr_end; vaddr = next_vaddr, i += pages) {
6455 * If the user buffer is not page-aligned, calculate the offset
6458 offset = vaddr & (PAGE_SIZE - 1);
6460 /* Calculate the number of pages that can be encrypted in one go. */
6461 pages = get_num_contig_pages(i, inpages, npages);
6463 len = min_t(size_t, ((pages * PAGE_SIZE) - offset), size);
6465 data->handle = sev->handle;
6467 data->address = __sme_page_pa(inpages[i]) + offset;
6468 ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_UPDATE_DATA, data, &argp->error);
6473 next_vaddr = vaddr + len;
6477 /* content of memory is updated, mark pages dirty */
6478 for (i = 0; i < npages; i++) {
6479 set_page_dirty_lock(inpages[i]);
6480 mark_page_accessed(inpages[i]);
6482 /* unlock the user pages */
6483 sev_unpin_memory(kvm, inpages, npages);
6489 static int sev_launch_measure(struct kvm *kvm, struct kvm_sev_cmd *argp)
6491 void __user *measure = (void __user *)(uintptr_t)argp->data;
6492 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
6493 struct sev_data_launch_measure *data;
6494 struct kvm_sev_launch_measure params;
6495 void __user *p = NULL;
6499 if (!sev_guest(kvm))
6502 if (copy_from_user(¶ms, measure, sizeof(params)))
6505 data = kzalloc(sizeof(*data), GFP_KERNEL);
6509 /* User wants to query the blob length */
6513 p = (void __user *)(uintptr_t)params.uaddr;
6515 if (params.len > SEV_FW_BLOB_MAX_SIZE) {
6521 blob = kmalloc(params.len, GFP_KERNEL);
6525 data->address = __psp_pa(blob);
6526 data->len = params.len;
6530 data->handle = sev->handle;
6531 ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_MEASURE, data, &argp->error);
6534 * If we query the session length, FW responded with expected data.
6543 if (copy_to_user(p, blob, params.len))
6548 params.len = data->len;
6549 if (copy_to_user(measure, ¶ms, sizeof(params)))
6558 static int sev_launch_finish(struct kvm *kvm, struct kvm_sev_cmd *argp)
6560 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
6561 struct sev_data_launch_finish *data;
6564 if (!sev_guest(kvm))
6567 data = kzalloc(sizeof(*data), GFP_KERNEL);
6571 data->handle = sev->handle;
6572 ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_FINISH, data, &argp->error);
6578 static int sev_guest_status(struct kvm *kvm, struct kvm_sev_cmd *argp)
6580 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
6581 struct kvm_sev_guest_status params;
6582 struct sev_data_guest_status *data;
6585 if (!sev_guest(kvm))
6588 data = kzalloc(sizeof(*data), GFP_KERNEL);
6592 data->handle = sev->handle;
6593 ret = sev_issue_cmd(kvm, SEV_CMD_GUEST_STATUS, data, &argp->error);
6597 params.policy = data->policy;
6598 params.state = data->state;
6599 params.handle = data->handle;
6601 if (copy_to_user((void __user *)(uintptr_t)argp->data, ¶ms, sizeof(params)))
6608 static int __sev_issue_dbg_cmd(struct kvm *kvm, unsigned long src,
6609 unsigned long dst, int size,
6610 int *error, bool enc)
6612 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
6613 struct sev_data_dbg *data;
6616 data = kzalloc(sizeof(*data), GFP_KERNEL);
6620 data->handle = sev->handle;
6621 data->dst_addr = dst;
6622 data->src_addr = src;
6625 ret = sev_issue_cmd(kvm,
6626 enc ? SEV_CMD_DBG_ENCRYPT : SEV_CMD_DBG_DECRYPT,
6632 static int __sev_dbg_decrypt(struct kvm *kvm, unsigned long src_paddr,
6633 unsigned long dst_paddr, int sz, int *err)
6638 * Its safe to read more than we are asked, caller should ensure that
6639 * destination has enough space.
6641 src_paddr = round_down(src_paddr, 16);
6642 offset = src_paddr & 15;
6643 sz = round_up(sz + offset, 16);
6645 return __sev_issue_dbg_cmd(kvm, src_paddr, dst_paddr, sz, err, false);
6648 static int __sev_dbg_decrypt_user(struct kvm *kvm, unsigned long paddr,
6649 unsigned long __user dst_uaddr,
6650 unsigned long dst_paddr,
6653 struct page *tpage = NULL;
6656 /* if inputs are not 16-byte then use intermediate buffer */
6657 if (!IS_ALIGNED(dst_paddr, 16) ||
6658 !IS_ALIGNED(paddr, 16) ||
6659 !IS_ALIGNED(size, 16)) {
6660 tpage = (void *)alloc_page(GFP_KERNEL);
6664 dst_paddr = __sme_page_pa(tpage);
6667 ret = __sev_dbg_decrypt(kvm, paddr, dst_paddr, size, err);
6672 offset = paddr & 15;
6673 if (copy_to_user((void __user *)(uintptr_t)dst_uaddr,
6674 page_address(tpage) + offset, size))
6685 static int __sev_dbg_encrypt_user(struct kvm *kvm, unsigned long paddr,
6686 unsigned long __user vaddr,
6687 unsigned long dst_paddr,
6688 unsigned long __user dst_vaddr,
6689 int size, int *error)
6691 struct page *src_tpage = NULL;
6692 struct page *dst_tpage = NULL;
6693 int ret, len = size;
6695 /* If source buffer is not aligned then use an intermediate buffer */
6696 if (!IS_ALIGNED(vaddr, 16)) {
6697 src_tpage = alloc_page(GFP_KERNEL);
6701 if (copy_from_user(page_address(src_tpage),
6702 (void __user *)(uintptr_t)vaddr, size)) {
6703 __free_page(src_tpage);
6707 paddr = __sme_page_pa(src_tpage);
6711 * If destination buffer or length is not aligned then do read-modify-write:
6712 * - decrypt destination in an intermediate buffer
6713 * - copy the source buffer in an intermediate buffer
6714 * - use the intermediate buffer as source buffer
6716 if (!IS_ALIGNED(dst_vaddr, 16) || !IS_ALIGNED(size, 16)) {
6719 dst_tpage = alloc_page(GFP_KERNEL);
6725 ret = __sev_dbg_decrypt(kvm, dst_paddr,
6726 __sme_page_pa(dst_tpage), size, error);
6731 * If source is kernel buffer then use memcpy() otherwise
6734 dst_offset = dst_paddr & 15;
6737 memcpy(page_address(dst_tpage) + dst_offset,
6738 page_address(src_tpage), size);
6740 if (copy_from_user(page_address(dst_tpage) + dst_offset,
6741 (void __user *)(uintptr_t)vaddr, size)) {
6747 paddr = __sme_page_pa(dst_tpage);
6748 dst_paddr = round_down(dst_paddr, 16);
6749 len = round_up(size, 16);
6752 ret = __sev_issue_dbg_cmd(kvm, paddr, dst_paddr, len, error, true);
6756 __free_page(src_tpage);
6758 __free_page(dst_tpage);
6762 static int sev_dbg_crypt(struct kvm *kvm, struct kvm_sev_cmd *argp, bool dec)
6764 unsigned long vaddr, vaddr_end, next_vaddr;
6765 unsigned long dst_vaddr;
6766 struct page **src_p, **dst_p;
6767 struct kvm_sev_dbg debug;
6771 if (!sev_guest(kvm))
6774 if (copy_from_user(&debug, (void __user *)(uintptr_t)argp->data, sizeof(debug)))
6777 vaddr = debug.src_uaddr;
6779 vaddr_end = vaddr + size;
6780 dst_vaddr = debug.dst_uaddr;
6782 for (; vaddr < vaddr_end; vaddr = next_vaddr) {
6783 int len, s_off, d_off;
6785 /* lock userspace source and destination page */
6786 src_p = sev_pin_memory(kvm, vaddr & PAGE_MASK, PAGE_SIZE, &n, 0);
6790 dst_p = sev_pin_memory(kvm, dst_vaddr & PAGE_MASK, PAGE_SIZE, &n, 1);
6792 sev_unpin_memory(kvm, src_p, n);
6797 * The DBG_{DE,EN}CRYPT commands will perform {dec,en}cryption of the
6798 * memory content (i.e it will write the same memory region with C=1).
6799 * It's possible that the cache may contain the data with C=0, i.e.,
6800 * unencrypted so invalidate it first.
6802 sev_clflush_pages(src_p, 1);
6803 sev_clflush_pages(dst_p, 1);
6806 * Since user buffer may not be page aligned, calculate the
6807 * offset within the page.
6809 s_off = vaddr & ~PAGE_MASK;
6810 d_off = dst_vaddr & ~PAGE_MASK;
6811 len = min_t(size_t, (PAGE_SIZE - s_off), size);
6814 ret = __sev_dbg_decrypt_user(kvm,
6815 __sme_page_pa(src_p[0]) + s_off,
6817 __sme_page_pa(dst_p[0]) + d_off,
6820 ret = __sev_dbg_encrypt_user(kvm,
6821 __sme_page_pa(src_p[0]) + s_off,
6823 __sme_page_pa(dst_p[0]) + d_off,
6827 sev_unpin_memory(kvm, src_p, 1);
6828 sev_unpin_memory(kvm, dst_p, 1);
6833 next_vaddr = vaddr + len;
6834 dst_vaddr = dst_vaddr + len;
6841 static int sev_launch_secret(struct kvm *kvm, struct kvm_sev_cmd *argp)
6843 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
6844 struct sev_data_launch_secret *data;
6845 struct kvm_sev_launch_secret params;
6846 struct page **pages;
6851 if (!sev_guest(kvm))
6854 if (copy_from_user(¶ms, (void __user *)(uintptr_t)argp->data, sizeof(params)))
6857 pages = sev_pin_memory(kvm, params.guest_uaddr, params.guest_len, &n, 1);
6862 * The secret must be copied into contiguous memory region, lets verify
6863 * that userspace memory pages are contiguous before we issue command.
6865 if (get_num_contig_pages(0, pages, n) != n) {
6867 goto e_unpin_memory;
6871 data = kzalloc(sizeof(*data), GFP_KERNEL);
6873 goto e_unpin_memory;
6875 offset = params.guest_uaddr & (PAGE_SIZE - 1);
6876 data->guest_address = __sme_page_pa(pages[0]) + offset;
6877 data->guest_len = params.guest_len;
6879 blob = psp_copy_user_blob(params.trans_uaddr, params.trans_len);
6881 ret = PTR_ERR(blob);
6885 data->trans_address = __psp_pa(blob);
6886 data->trans_len = params.trans_len;
6888 hdr = psp_copy_user_blob(params.hdr_uaddr, params.hdr_len);
6893 data->hdr_address = __psp_pa(hdr);
6894 data->hdr_len = params.hdr_len;
6896 data->handle = sev->handle;
6897 ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_UPDATE_SECRET, data, &argp->error);
6906 sev_unpin_memory(kvm, pages, n);
6910 static int svm_mem_enc_op(struct kvm *kvm, void __user *argp)
6912 struct kvm_sev_cmd sev_cmd;
6915 if (!svm_sev_enabled())
6918 if (copy_from_user(&sev_cmd, argp, sizeof(struct kvm_sev_cmd)))
6921 mutex_lock(&kvm->lock);
6923 switch (sev_cmd.id) {
6925 r = sev_guest_init(kvm, &sev_cmd);
6927 case KVM_SEV_LAUNCH_START:
6928 r = sev_launch_start(kvm, &sev_cmd);
6930 case KVM_SEV_LAUNCH_UPDATE_DATA:
6931 r = sev_launch_update_data(kvm, &sev_cmd);
6933 case KVM_SEV_LAUNCH_MEASURE:
6934 r = sev_launch_measure(kvm, &sev_cmd);
6936 case KVM_SEV_LAUNCH_FINISH:
6937 r = sev_launch_finish(kvm, &sev_cmd);
6939 case KVM_SEV_GUEST_STATUS:
6940 r = sev_guest_status(kvm, &sev_cmd);
6942 case KVM_SEV_DBG_DECRYPT:
6943 r = sev_dbg_crypt(kvm, &sev_cmd, true);
6945 case KVM_SEV_DBG_ENCRYPT:
6946 r = sev_dbg_crypt(kvm, &sev_cmd, false);
6948 case KVM_SEV_LAUNCH_SECRET:
6949 r = sev_launch_secret(kvm, &sev_cmd);
6956 if (copy_to_user(argp, &sev_cmd, sizeof(struct kvm_sev_cmd)))
6960 mutex_unlock(&kvm->lock);
6964 static int svm_register_enc_region(struct kvm *kvm,
6965 struct kvm_enc_region *range)
6967 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
6968 struct enc_region *region;
6971 if (!sev_guest(kvm))
6974 if (range->addr > ULONG_MAX || range->size > ULONG_MAX)
6977 region = kzalloc(sizeof(*region), GFP_KERNEL);
6981 region->pages = sev_pin_memory(kvm, range->addr, range->size, ®ion->npages, 1);
6982 if (!region->pages) {
6988 * The guest may change the memory encryption attribute from C=0 -> C=1
6989 * or vice versa for this memory range. Lets make sure caches are
6990 * flushed to ensure that guest data gets written into memory with
6993 sev_clflush_pages(region->pages, region->npages);
6995 region->uaddr = range->addr;
6996 region->size = range->size;
6998 mutex_lock(&kvm->lock);
6999 list_add_tail(®ion->list, &sev->regions_list);
7000 mutex_unlock(&kvm->lock);
7009 static struct enc_region *
7010 find_enc_region(struct kvm *kvm, struct kvm_enc_region *range)
7012 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
7013 struct list_head *head = &sev->regions_list;
7014 struct enc_region *i;
7016 list_for_each_entry(i, head, list) {
7017 if (i->uaddr == range->addr &&
7018 i->size == range->size)
7026 static int svm_unregister_enc_region(struct kvm *kvm,
7027 struct kvm_enc_region *range)
7029 struct enc_region *region;
7032 mutex_lock(&kvm->lock);
7034 if (!sev_guest(kvm)) {
7039 region = find_enc_region(kvm, range);
7045 __unregister_enc_region_locked(kvm, region);
7047 mutex_unlock(&kvm->lock);
7051 mutex_unlock(&kvm->lock);
7055 static int nested_enable_evmcs(struct kvm_vcpu *vcpu,
7056 uint16_t *vmcs_version)
7058 /* Intel-only feature */
7062 static struct kvm_x86_ops svm_x86_ops __ro_after_init = {
7063 .cpu_has_kvm_support = has_svm,
7064 .disabled_by_bios = is_disabled,
7065 .hardware_setup = svm_hardware_setup,
7066 .hardware_unsetup = svm_hardware_unsetup,
7067 .check_processor_compatibility = svm_check_processor_compat,
7068 .hardware_enable = svm_hardware_enable,
7069 .hardware_disable = svm_hardware_disable,
7070 .cpu_has_accelerated_tpr = svm_cpu_has_accelerated_tpr,
7071 .has_emulated_msr = svm_has_emulated_msr,
7073 .vcpu_create = svm_create_vcpu,
7074 .vcpu_free = svm_free_vcpu,
7075 .vcpu_reset = svm_vcpu_reset,
7077 .vm_alloc = svm_vm_alloc,
7078 .vm_free = svm_vm_free,
7079 .vm_init = avic_vm_init,
7080 .vm_destroy = svm_vm_destroy,
7082 .prepare_guest_switch = svm_prepare_guest_switch,
7083 .vcpu_load = svm_vcpu_load,
7084 .vcpu_put = svm_vcpu_put,
7085 .vcpu_blocking = svm_vcpu_blocking,
7086 .vcpu_unblocking = svm_vcpu_unblocking,
7088 .update_bp_intercept = update_bp_intercept,
7089 .get_msr_feature = svm_get_msr_feature,
7090 .get_msr = svm_get_msr,
7091 .set_msr = svm_set_msr,
7092 .get_segment_base = svm_get_segment_base,
7093 .get_segment = svm_get_segment,
7094 .set_segment = svm_set_segment,
7095 .get_cpl = svm_get_cpl,
7096 .get_cs_db_l_bits = kvm_get_cs_db_l_bits,
7097 .decache_cr0_guest_bits = svm_decache_cr0_guest_bits,
7098 .decache_cr3 = svm_decache_cr3,
7099 .decache_cr4_guest_bits = svm_decache_cr4_guest_bits,
7100 .set_cr0 = svm_set_cr0,
7101 .set_cr3 = svm_set_cr3,
7102 .set_cr4 = svm_set_cr4,
7103 .set_efer = svm_set_efer,
7104 .get_idt = svm_get_idt,
7105 .set_idt = svm_set_idt,
7106 .get_gdt = svm_get_gdt,
7107 .set_gdt = svm_set_gdt,
7108 .get_dr6 = svm_get_dr6,
7109 .set_dr6 = svm_set_dr6,
7110 .set_dr7 = svm_set_dr7,
7111 .sync_dirty_debug_regs = svm_sync_dirty_debug_regs,
7112 .cache_reg = svm_cache_reg,
7113 .get_rflags = svm_get_rflags,
7114 .set_rflags = svm_set_rflags,
7116 .tlb_flush = svm_flush_tlb,
7117 .tlb_flush_gva = svm_flush_tlb_gva,
7119 .run = svm_vcpu_run,
7120 .handle_exit = handle_exit,
7121 .skip_emulated_instruction = skip_emulated_instruction,
7122 .set_interrupt_shadow = svm_set_interrupt_shadow,
7123 .get_interrupt_shadow = svm_get_interrupt_shadow,
7124 .patch_hypercall = svm_patch_hypercall,
7125 .set_irq = svm_set_irq,
7126 .set_nmi = svm_inject_nmi,
7127 .queue_exception = svm_queue_exception,
7128 .cancel_injection = svm_cancel_injection,
7129 .interrupt_allowed = svm_interrupt_allowed,
7130 .nmi_allowed = svm_nmi_allowed,
7131 .get_nmi_mask = svm_get_nmi_mask,
7132 .set_nmi_mask = svm_set_nmi_mask,
7133 .enable_nmi_window = enable_nmi_window,
7134 .enable_irq_window = enable_irq_window,
7135 .update_cr8_intercept = update_cr8_intercept,
7136 .set_virtual_apic_mode = svm_set_virtual_apic_mode,
7137 .get_enable_apicv = svm_get_enable_apicv,
7138 .refresh_apicv_exec_ctrl = svm_refresh_apicv_exec_ctrl,
7139 .load_eoi_exitmap = svm_load_eoi_exitmap,
7140 .hwapic_irr_update = svm_hwapic_irr_update,
7141 .hwapic_isr_update = svm_hwapic_isr_update,
7142 .sync_pir_to_irr = kvm_lapic_find_highest_irr,
7143 .apicv_post_state_restore = avic_post_state_restore,
7145 .set_tss_addr = svm_set_tss_addr,
7146 .set_identity_map_addr = svm_set_identity_map_addr,
7147 .get_tdp_level = get_npt_level,
7148 .get_mt_mask = svm_get_mt_mask,
7150 .get_exit_info = svm_get_exit_info,
7152 .get_lpage_level = svm_get_lpage_level,
7154 .cpuid_update = svm_cpuid_update,
7156 .rdtscp_supported = svm_rdtscp_supported,
7157 .invpcid_supported = svm_invpcid_supported,
7158 .mpx_supported = svm_mpx_supported,
7159 .xsaves_supported = svm_xsaves_supported,
7160 .umip_emulated = svm_umip_emulated,
7162 .set_supported_cpuid = svm_set_supported_cpuid,
7164 .has_wbinvd_exit = svm_has_wbinvd_exit,
7166 .read_l1_tsc_offset = svm_read_l1_tsc_offset,
7167 .write_l1_tsc_offset = svm_write_l1_tsc_offset,
7169 .set_tdp_cr3 = set_tdp_cr3,
7171 .check_intercept = svm_check_intercept,
7172 .handle_external_intr = svm_handle_external_intr,
7174 .request_immediate_exit = __kvm_request_immediate_exit,
7176 .sched_in = svm_sched_in,
7178 .pmu_ops = &amd_pmu_ops,
7179 .deliver_posted_interrupt = svm_deliver_avic_intr,
7180 .update_pi_irte = svm_update_pi_irte,
7181 .setup_mce = svm_setup_mce,
7183 .smi_allowed = svm_smi_allowed,
7184 .pre_enter_smm = svm_pre_enter_smm,
7185 .pre_leave_smm = svm_pre_leave_smm,
7186 .enable_smi_window = enable_smi_window,
7188 .mem_enc_op = svm_mem_enc_op,
7189 .mem_enc_reg_region = svm_register_enc_region,
7190 .mem_enc_unreg_region = svm_unregister_enc_region,
7192 .nested_enable_evmcs = nested_enable_evmcs,
7195 static int __init svm_init(void)
7197 return kvm_init(&svm_x86_ops, sizeof(struct vcpu_svm),
7198 __alignof__(struct vcpu_svm), THIS_MODULE);
7201 static void __exit svm_exit(void)
7206 module_init(svm_init)
7207 module_exit(svm_exit)