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 static const u32 msrpm_ranges[] = {0, 0xc0000000, 0xc0010000};
680 #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
681 #define MSRS_RANGE_SIZE 2048
682 #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
684 static u32 svm_msrpm_offset(u32 msr)
689 for (i = 0; i < NUM_MSR_MAPS; i++) {
690 if (msr < msrpm_ranges[i] ||
691 msr >= msrpm_ranges[i] + MSRS_IN_RANGE)
694 offset = (msr - msrpm_ranges[i]) / 4; /* 4 msrs per u8 */
695 offset += (i * MSRS_RANGE_SIZE); /* add range offset */
697 /* Now we have the u8 offset - but need the u32 offset */
701 /* MSR not in any range */
705 #define MAX_INST_SIZE 15
707 static inline void clgi(void)
709 asm volatile (__ex("clgi"));
712 static inline void stgi(void)
714 asm volatile (__ex("stgi"));
717 static inline void invlpga(unsigned long addr, u32 asid)
719 asm volatile (__ex("invlpga %1, %0") : : "c"(asid), "a"(addr));
722 static int get_npt_level(struct kvm_vcpu *vcpu)
725 return PT64_ROOT_4LEVEL;
727 return PT32E_ROOT_LEVEL;
731 static void svm_set_efer(struct kvm_vcpu *vcpu, u64 efer)
733 vcpu->arch.efer = efer;
734 if (!npt_enabled && !(efer & EFER_LMA))
737 to_svm(vcpu)->vmcb->save.efer = efer | EFER_SVME;
738 mark_dirty(to_svm(vcpu)->vmcb, VMCB_CR);
741 static int is_external_interrupt(u32 info)
743 info &= SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
744 return info == (SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR);
747 static u32 svm_get_interrupt_shadow(struct kvm_vcpu *vcpu)
749 struct vcpu_svm *svm = to_svm(vcpu);
752 if (svm->vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK)
753 ret = KVM_X86_SHADOW_INT_STI | KVM_X86_SHADOW_INT_MOV_SS;
757 static void svm_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
759 struct vcpu_svm *svm = to_svm(vcpu);
762 svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK;
764 svm->vmcb->control.int_state |= SVM_INTERRUPT_SHADOW_MASK;
768 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
770 struct vcpu_svm *svm = to_svm(vcpu);
772 if (svm->vmcb->control.next_rip != 0) {
773 WARN_ON_ONCE(!static_cpu_has(X86_FEATURE_NRIPS));
774 svm->next_rip = svm->vmcb->control.next_rip;
777 if (!svm->next_rip) {
778 if (kvm_emulate_instruction(vcpu, EMULTYPE_SKIP) !=
780 printk(KERN_DEBUG "%s: NOP\n", __func__);
783 if (svm->next_rip - kvm_rip_read(vcpu) > MAX_INST_SIZE)
784 printk(KERN_ERR "%s: ip 0x%lx next 0x%llx\n",
785 __func__, kvm_rip_read(vcpu), svm->next_rip);
787 kvm_rip_write(vcpu, svm->next_rip);
788 svm_set_interrupt_shadow(vcpu, 0);
791 static void svm_queue_exception(struct kvm_vcpu *vcpu)
793 struct vcpu_svm *svm = to_svm(vcpu);
794 unsigned nr = vcpu->arch.exception.nr;
795 bool has_error_code = vcpu->arch.exception.has_error_code;
796 bool reinject = vcpu->arch.exception.injected;
797 u32 error_code = vcpu->arch.exception.error_code;
800 * If we are within a nested VM we'd better #VMEXIT and let the guest
801 * handle the exception
804 nested_svm_check_exception(svm, nr, has_error_code, error_code))
807 kvm_deliver_exception_payload(&svm->vcpu);
809 if (nr == BP_VECTOR && !static_cpu_has(X86_FEATURE_NRIPS)) {
810 unsigned long rip, old_rip = kvm_rip_read(&svm->vcpu);
813 * For guest debugging where we have to reinject #BP if some
814 * INT3 is guest-owned:
815 * Emulate nRIP by moving RIP forward. Will fail if injection
816 * raises a fault that is not intercepted. Still better than
817 * failing in all cases.
819 skip_emulated_instruction(&svm->vcpu);
820 rip = kvm_rip_read(&svm->vcpu);
821 svm->int3_rip = rip + svm->vmcb->save.cs.base;
822 svm->int3_injected = rip - old_rip;
825 svm->vmcb->control.event_inj = nr
827 | (has_error_code ? SVM_EVTINJ_VALID_ERR : 0)
828 | SVM_EVTINJ_TYPE_EXEPT;
829 svm->vmcb->control.event_inj_err = error_code;
832 static void svm_init_erratum_383(void)
838 if (!static_cpu_has_bug(X86_BUG_AMD_TLB_MMATCH))
841 /* Use _safe variants to not break nested virtualization */
842 val = native_read_msr_safe(MSR_AMD64_DC_CFG, &err);
848 low = lower_32_bits(val);
849 high = upper_32_bits(val);
851 native_write_msr_safe(MSR_AMD64_DC_CFG, low, high);
853 erratum_383_found = true;
856 static void svm_init_osvw(struct kvm_vcpu *vcpu)
859 * Guests should see errata 400 and 415 as fixed (assuming that
860 * HLT and IO instructions are intercepted).
862 vcpu->arch.osvw.length = (osvw_len >= 3) ? (osvw_len) : 3;
863 vcpu->arch.osvw.status = osvw_status & ~(6ULL);
866 * By increasing VCPU's osvw.length to 3 we are telling the guest that
867 * all osvw.status bits inside that length, including bit 0 (which is
868 * reserved for erratum 298), are valid. However, if host processor's
869 * osvw_len is 0 then osvw_status[0] carries no information. We need to
870 * be conservative here and therefore we tell the guest that erratum 298
871 * is present (because we really don't know).
873 if (osvw_len == 0 && boot_cpu_data.x86 == 0x10)
874 vcpu->arch.osvw.status |= 1;
877 static int has_svm(void)
881 if (!cpu_has_svm(&msg)) {
882 printk(KERN_INFO "has_svm: %s\n", msg);
889 static void svm_hardware_disable(void)
891 /* Make sure we clean up behind us */
892 if (static_cpu_has(X86_FEATURE_TSCRATEMSR))
893 wrmsrl(MSR_AMD64_TSC_RATIO, TSC_RATIO_DEFAULT);
897 amd_pmu_disable_virt();
900 static int svm_hardware_enable(void)
903 struct svm_cpu_data *sd;
905 struct desc_struct *gdt;
906 int me = raw_smp_processor_id();
908 rdmsrl(MSR_EFER, efer);
909 if (efer & EFER_SVME)
913 pr_err("%s: err EOPNOTSUPP on %d\n", __func__, me);
916 sd = per_cpu(svm_data, me);
918 pr_err("%s: svm_data is NULL on %d\n", __func__, me);
922 sd->asid_generation = 1;
923 sd->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1;
924 sd->next_asid = sd->max_asid + 1;
925 sd->min_asid = max_sev_asid + 1;
927 gdt = get_current_gdt_rw();
928 sd->tss_desc = (struct kvm_ldttss_desc *)(gdt + GDT_ENTRY_TSS);
930 wrmsrl(MSR_EFER, efer | EFER_SVME);
932 wrmsrl(MSR_VM_HSAVE_PA, page_to_pfn(sd->save_area) << PAGE_SHIFT);
934 if (static_cpu_has(X86_FEATURE_TSCRATEMSR)) {
935 wrmsrl(MSR_AMD64_TSC_RATIO, TSC_RATIO_DEFAULT);
936 __this_cpu_write(current_tsc_ratio, TSC_RATIO_DEFAULT);
943 * Note that it is possible to have a system with mixed processor
944 * revisions and therefore different OSVW bits. If bits are not the same
945 * on different processors then choose the worst case (i.e. if erratum
946 * is present on one processor and not on another then assume that the
947 * erratum is present everywhere).
949 if (cpu_has(&boot_cpu_data, X86_FEATURE_OSVW)) {
950 uint64_t len, status = 0;
953 len = native_read_msr_safe(MSR_AMD64_OSVW_ID_LENGTH, &err);
955 status = native_read_msr_safe(MSR_AMD64_OSVW_STATUS,
959 osvw_status = osvw_len = 0;
963 osvw_status |= status;
964 osvw_status &= (1ULL << osvw_len) - 1;
967 osvw_status = osvw_len = 0;
969 svm_init_erratum_383();
971 amd_pmu_enable_virt();
976 static void svm_cpu_uninit(int cpu)
978 struct svm_cpu_data *sd = per_cpu(svm_data, raw_smp_processor_id());
983 per_cpu(svm_data, raw_smp_processor_id()) = NULL;
984 kfree(sd->sev_vmcbs);
985 __free_page(sd->save_area);
989 static int svm_cpu_init(int cpu)
991 struct svm_cpu_data *sd;
994 sd = kzalloc(sizeof(struct svm_cpu_data), GFP_KERNEL);
999 sd->save_area = alloc_page(GFP_KERNEL);
1003 if (svm_sev_enabled()) {
1005 sd->sev_vmcbs = kmalloc_array(max_sev_asid + 1,
1012 per_cpu(svm_data, cpu) = sd;
1022 static bool valid_msr_intercept(u32 index)
1026 for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++)
1027 if (direct_access_msrs[i].index == index)
1033 static bool msr_write_intercepted(struct kvm_vcpu *vcpu, unsigned msr)
1040 msrpm = is_guest_mode(vcpu) ? to_svm(vcpu)->nested.msrpm:
1041 to_svm(vcpu)->msrpm;
1043 offset = svm_msrpm_offset(msr);
1044 bit_write = 2 * (msr & 0x0f) + 1;
1045 tmp = msrpm[offset];
1047 BUG_ON(offset == MSR_INVALID);
1049 return !!test_bit(bit_write, &tmp);
1052 static void set_msr_interception(u32 *msrpm, unsigned msr,
1053 int read, int write)
1055 u8 bit_read, bit_write;
1060 * If this warning triggers extend the direct_access_msrs list at the
1061 * beginning of the file
1063 WARN_ON(!valid_msr_intercept(msr));
1065 offset = svm_msrpm_offset(msr);
1066 bit_read = 2 * (msr & 0x0f);
1067 bit_write = 2 * (msr & 0x0f) + 1;
1068 tmp = msrpm[offset];
1070 BUG_ON(offset == MSR_INVALID);
1072 read ? clear_bit(bit_read, &tmp) : set_bit(bit_read, &tmp);
1073 write ? clear_bit(bit_write, &tmp) : set_bit(bit_write, &tmp);
1075 msrpm[offset] = tmp;
1078 static void svm_vcpu_init_msrpm(u32 *msrpm)
1082 memset(msrpm, 0xff, PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER));
1084 for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
1085 if (!direct_access_msrs[i].always)
1088 set_msr_interception(msrpm, direct_access_msrs[i].index, 1, 1);
1092 static void add_msr_offset(u32 offset)
1096 for (i = 0; i < MSRPM_OFFSETS; ++i) {
1098 /* Offset already in list? */
1099 if (msrpm_offsets[i] == offset)
1102 /* Slot used by another offset? */
1103 if (msrpm_offsets[i] != MSR_INVALID)
1106 /* Add offset to list */
1107 msrpm_offsets[i] = offset;
1113 * If this BUG triggers the msrpm_offsets table has an overflow. Just
1114 * increase MSRPM_OFFSETS in this case.
1119 static void init_msrpm_offsets(void)
1123 memset(msrpm_offsets, 0xff, sizeof(msrpm_offsets));
1125 for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
1128 offset = svm_msrpm_offset(direct_access_msrs[i].index);
1129 BUG_ON(offset == MSR_INVALID);
1131 add_msr_offset(offset);
1135 static void svm_enable_lbrv(struct vcpu_svm *svm)
1137 u32 *msrpm = svm->msrpm;
1139 svm->vmcb->control.virt_ext |= LBR_CTL_ENABLE_MASK;
1140 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 1, 1);
1141 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 1, 1);
1142 set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 1, 1);
1143 set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 1, 1);
1146 static void svm_disable_lbrv(struct vcpu_svm *svm)
1148 u32 *msrpm = svm->msrpm;
1150 svm->vmcb->control.virt_ext &= ~LBR_CTL_ENABLE_MASK;
1151 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 0, 0);
1152 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 0, 0);
1153 set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 0, 0);
1154 set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 0, 0);
1157 static void disable_nmi_singlestep(struct vcpu_svm *svm)
1159 svm->nmi_singlestep = false;
1161 if (!(svm->vcpu.guest_debug & KVM_GUESTDBG_SINGLESTEP)) {
1162 /* Clear our flags if they were not set by the guest */
1163 if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_TF))
1164 svm->vmcb->save.rflags &= ~X86_EFLAGS_TF;
1165 if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_RF))
1166 svm->vmcb->save.rflags &= ~X86_EFLAGS_RF;
1171 * This hash table is used to map VM_ID to a struct kvm_svm,
1172 * when handling AMD IOMMU GALOG notification to schedule in
1173 * a particular vCPU.
1175 #define SVM_VM_DATA_HASH_BITS 8
1176 static DEFINE_HASHTABLE(svm_vm_data_hash, SVM_VM_DATA_HASH_BITS);
1177 static u32 next_vm_id = 0;
1178 static bool next_vm_id_wrapped = 0;
1179 static DEFINE_SPINLOCK(svm_vm_data_hash_lock);
1182 * This function is called from IOMMU driver to notify
1183 * SVM to schedule in a particular vCPU of a particular VM.
1185 static int avic_ga_log_notifier(u32 ga_tag)
1187 unsigned long flags;
1188 struct kvm_svm *kvm_svm;
1189 struct kvm_vcpu *vcpu = NULL;
1190 u32 vm_id = AVIC_GATAG_TO_VMID(ga_tag);
1191 u32 vcpu_id = AVIC_GATAG_TO_VCPUID(ga_tag);
1193 pr_debug("SVM: %s: vm_id=%#x, vcpu_id=%#x\n", __func__, vm_id, vcpu_id);
1195 spin_lock_irqsave(&svm_vm_data_hash_lock, flags);
1196 hash_for_each_possible(svm_vm_data_hash, kvm_svm, hnode, vm_id) {
1197 if (kvm_svm->avic_vm_id != vm_id)
1199 vcpu = kvm_get_vcpu_by_id(&kvm_svm->kvm, vcpu_id);
1202 spin_unlock_irqrestore(&svm_vm_data_hash_lock, flags);
1205 * At this point, the IOMMU should have already set the pending
1206 * bit in the vAPIC backing page. So, we just need to schedule
1210 kvm_vcpu_wake_up(vcpu);
1215 static __init int sev_hardware_setup(void)
1217 struct sev_user_data_status *status;
1220 /* Maximum number of encrypted guests supported simultaneously */
1221 max_sev_asid = cpuid_ecx(0x8000001F);
1226 /* Minimum ASID value that should be used for SEV guest */
1227 min_sev_asid = cpuid_edx(0x8000001F);
1229 /* Initialize SEV ASID bitmap */
1230 sev_asid_bitmap = bitmap_zalloc(max_sev_asid, GFP_KERNEL);
1231 if (!sev_asid_bitmap)
1234 status = kmalloc(sizeof(*status), GFP_KERNEL);
1239 * Check SEV platform status.
1241 * PLATFORM_STATUS can be called in any state, if we failed to query
1242 * the PLATFORM status then either PSP firmware does not support SEV
1243 * feature or SEV firmware is dead.
1245 rc = sev_platform_status(status, NULL);
1249 pr_info("SEV supported\n");
1256 static void grow_ple_window(struct kvm_vcpu *vcpu)
1258 struct vcpu_svm *svm = to_svm(vcpu);
1259 struct vmcb_control_area *control = &svm->vmcb->control;
1260 int old = control->pause_filter_count;
1262 control->pause_filter_count = __grow_ple_window(old,
1264 pause_filter_count_grow,
1265 pause_filter_count_max);
1267 if (control->pause_filter_count != old)
1268 mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
1270 trace_kvm_ple_window_grow(vcpu->vcpu_id,
1271 control->pause_filter_count, old);
1274 static void shrink_ple_window(struct kvm_vcpu *vcpu)
1276 struct vcpu_svm *svm = to_svm(vcpu);
1277 struct vmcb_control_area *control = &svm->vmcb->control;
1278 int old = control->pause_filter_count;
1280 control->pause_filter_count =
1281 __shrink_ple_window(old,
1283 pause_filter_count_shrink,
1284 pause_filter_count);
1285 if (control->pause_filter_count != old)
1286 mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
1288 trace_kvm_ple_window_shrink(vcpu->vcpu_id,
1289 control->pause_filter_count, old);
1292 static __init int svm_hardware_setup(void)
1295 struct page *iopm_pages;
1299 iopm_pages = alloc_pages(GFP_KERNEL, IOPM_ALLOC_ORDER);
1304 iopm_va = page_address(iopm_pages);
1305 memset(iopm_va, 0xff, PAGE_SIZE * (1 << IOPM_ALLOC_ORDER));
1306 iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT;
1308 init_msrpm_offsets();
1310 if (boot_cpu_has(X86_FEATURE_NX))
1311 kvm_enable_efer_bits(EFER_NX);
1313 if (boot_cpu_has(X86_FEATURE_FXSR_OPT))
1314 kvm_enable_efer_bits(EFER_FFXSR);
1316 if (boot_cpu_has(X86_FEATURE_TSCRATEMSR)) {
1317 kvm_has_tsc_control = true;
1318 kvm_max_tsc_scaling_ratio = TSC_RATIO_MAX;
1319 kvm_tsc_scaling_ratio_frac_bits = 32;
1322 /* Check for pause filtering support */
1323 if (!boot_cpu_has(X86_FEATURE_PAUSEFILTER)) {
1324 pause_filter_count = 0;
1325 pause_filter_thresh = 0;
1326 } else if (!boot_cpu_has(X86_FEATURE_PFTHRESHOLD)) {
1327 pause_filter_thresh = 0;
1331 printk(KERN_INFO "kvm: Nested Virtualization enabled\n");
1332 kvm_enable_efer_bits(EFER_SVME | EFER_LMSLE);
1336 if (boot_cpu_has(X86_FEATURE_SEV) &&
1337 IS_ENABLED(CONFIG_KVM_AMD_SEV)) {
1338 r = sev_hardware_setup();
1346 for_each_possible_cpu(cpu) {
1347 r = svm_cpu_init(cpu);
1352 if (!boot_cpu_has(X86_FEATURE_NPT))
1353 npt_enabled = false;
1355 if (npt_enabled && !npt) {
1356 printk(KERN_INFO "kvm: Nested Paging disabled\n");
1357 npt_enabled = false;
1361 printk(KERN_INFO "kvm: Nested Paging enabled\n");
1368 !boot_cpu_has(X86_FEATURE_AVIC) ||
1369 !IS_ENABLED(CONFIG_X86_LOCAL_APIC)) {
1372 pr_info("AVIC enabled\n");
1374 amd_iommu_register_ga_log_notifier(&avic_ga_log_notifier);
1380 !boot_cpu_has(X86_FEATURE_V_VMSAVE_VMLOAD) ||
1381 !IS_ENABLED(CONFIG_X86_64)) {
1384 pr_info("Virtual VMLOAD VMSAVE supported\n");
1389 if (!boot_cpu_has(X86_FEATURE_VGIF))
1392 pr_info("Virtual GIF supported\n");
1398 __free_pages(iopm_pages, IOPM_ALLOC_ORDER);
1403 static __exit void svm_hardware_unsetup(void)
1407 if (svm_sev_enabled())
1408 bitmap_free(sev_asid_bitmap);
1410 for_each_possible_cpu(cpu)
1411 svm_cpu_uninit(cpu);
1413 __free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT), IOPM_ALLOC_ORDER);
1417 static void init_seg(struct vmcb_seg *seg)
1420 seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK |
1421 SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */
1422 seg->limit = 0xffff;
1426 static void init_sys_seg(struct vmcb_seg *seg, uint32_t type)
1429 seg->attrib = SVM_SELECTOR_P_MASK | type;
1430 seg->limit = 0xffff;
1434 static u64 svm_read_l1_tsc_offset(struct kvm_vcpu *vcpu)
1436 struct vcpu_svm *svm = to_svm(vcpu);
1438 if (is_guest_mode(vcpu))
1439 return svm->nested.hsave->control.tsc_offset;
1441 return vcpu->arch.tsc_offset;
1444 static u64 svm_write_l1_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1446 struct vcpu_svm *svm = to_svm(vcpu);
1447 u64 g_tsc_offset = 0;
1449 if (is_guest_mode(vcpu)) {
1450 /* Write L1's TSC offset. */
1451 g_tsc_offset = svm->vmcb->control.tsc_offset -
1452 svm->nested.hsave->control.tsc_offset;
1453 svm->nested.hsave->control.tsc_offset = offset;
1456 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
1457 svm->vmcb->control.tsc_offset - g_tsc_offset,
1460 svm->vmcb->control.tsc_offset = offset + g_tsc_offset;
1462 mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
1463 return svm->vmcb->control.tsc_offset;
1466 static void avic_init_vmcb(struct vcpu_svm *svm)
1468 struct vmcb *vmcb = svm->vmcb;
1469 struct kvm_svm *kvm_svm = to_kvm_svm(svm->vcpu.kvm);
1470 phys_addr_t bpa = __sme_set(page_to_phys(svm->avic_backing_page));
1471 phys_addr_t lpa = __sme_set(page_to_phys(kvm_svm->avic_logical_id_table_page));
1472 phys_addr_t ppa = __sme_set(page_to_phys(kvm_svm->avic_physical_id_table_page));
1474 vmcb->control.avic_backing_page = bpa & AVIC_HPA_MASK;
1475 vmcb->control.avic_logical_id = lpa & AVIC_HPA_MASK;
1476 vmcb->control.avic_physical_id = ppa & AVIC_HPA_MASK;
1477 vmcb->control.avic_physical_id |= AVIC_MAX_PHYSICAL_ID_COUNT;
1478 vmcb->control.int_ctl |= AVIC_ENABLE_MASK;
1481 static void init_vmcb(struct vcpu_svm *svm)
1483 struct vmcb_control_area *control = &svm->vmcb->control;
1484 struct vmcb_save_area *save = &svm->vmcb->save;
1486 svm->vcpu.arch.hflags = 0;
1488 set_cr_intercept(svm, INTERCEPT_CR0_READ);
1489 set_cr_intercept(svm, INTERCEPT_CR3_READ);
1490 set_cr_intercept(svm, INTERCEPT_CR4_READ);
1491 set_cr_intercept(svm, INTERCEPT_CR0_WRITE);
1492 set_cr_intercept(svm, INTERCEPT_CR3_WRITE);
1493 set_cr_intercept(svm, INTERCEPT_CR4_WRITE);
1494 if (!kvm_vcpu_apicv_active(&svm->vcpu))
1495 set_cr_intercept(svm, INTERCEPT_CR8_WRITE);
1497 set_dr_intercepts(svm);
1499 set_exception_intercept(svm, PF_VECTOR);
1500 set_exception_intercept(svm, UD_VECTOR);
1501 set_exception_intercept(svm, MC_VECTOR);
1502 set_exception_intercept(svm, AC_VECTOR);
1503 set_exception_intercept(svm, DB_VECTOR);
1505 * Guest access to VMware backdoor ports could legitimately
1506 * trigger #GP because of TSS I/O permission bitmap.
1507 * We intercept those #GP and allow access to them anyway
1510 if (enable_vmware_backdoor)
1511 set_exception_intercept(svm, GP_VECTOR);
1513 set_intercept(svm, INTERCEPT_INTR);
1514 set_intercept(svm, INTERCEPT_NMI);
1515 set_intercept(svm, INTERCEPT_SMI);
1516 set_intercept(svm, INTERCEPT_SELECTIVE_CR0);
1517 set_intercept(svm, INTERCEPT_RDPMC);
1518 set_intercept(svm, INTERCEPT_CPUID);
1519 set_intercept(svm, INTERCEPT_INVD);
1520 set_intercept(svm, INTERCEPT_INVLPG);
1521 set_intercept(svm, INTERCEPT_INVLPGA);
1522 set_intercept(svm, INTERCEPT_IOIO_PROT);
1523 set_intercept(svm, INTERCEPT_MSR_PROT);
1524 set_intercept(svm, INTERCEPT_TASK_SWITCH);
1525 set_intercept(svm, INTERCEPT_SHUTDOWN);
1526 set_intercept(svm, INTERCEPT_VMRUN);
1527 set_intercept(svm, INTERCEPT_VMMCALL);
1528 set_intercept(svm, INTERCEPT_VMLOAD);
1529 set_intercept(svm, INTERCEPT_VMSAVE);
1530 set_intercept(svm, INTERCEPT_STGI);
1531 set_intercept(svm, INTERCEPT_CLGI);
1532 set_intercept(svm, INTERCEPT_SKINIT);
1533 set_intercept(svm, INTERCEPT_WBINVD);
1534 set_intercept(svm, INTERCEPT_XSETBV);
1535 set_intercept(svm, INTERCEPT_RSM);
1537 if (!kvm_mwait_in_guest(svm->vcpu.kvm)) {
1538 set_intercept(svm, INTERCEPT_MONITOR);
1539 set_intercept(svm, INTERCEPT_MWAIT);
1542 if (!kvm_hlt_in_guest(svm->vcpu.kvm))
1543 set_intercept(svm, INTERCEPT_HLT);
1545 control->iopm_base_pa = __sme_set(iopm_base);
1546 control->msrpm_base_pa = __sme_set(__pa(svm->msrpm));
1547 control->int_ctl = V_INTR_MASKING_MASK;
1549 init_seg(&save->es);
1550 init_seg(&save->ss);
1551 init_seg(&save->ds);
1552 init_seg(&save->fs);
1553 init_seg(&save->gs);
1555 save->cs.selector = 0xf000;
1556 save->cs.base = 0xffff0000;
1557 /* Executable/Readable Code Segment */
1558 save->cs.attrib = SVM_SELECTOR_READ_MASK | SVM_SELECTOR_P_MASK |
1559 SVM_SELECTOR_S_MASK | SVM_SELECTOR_CODE_MASK;
1560 save->cs.limit = 0xffff;
1562 save->gdtr.limit = 0xffff;
1563 save->idtr.limit = 0xffff;
1565 init_sys_seg(&save->ldtr, SEG_TYPE_LDT);
1566 init_sys_seg(&save->tr, SEG_TYPE_BUSY_TSS16);
1568 svm_set_efer(&svm->vcpu, 0);
1569 save->dr6 = 0xffff0ff0;
1570 kvm_set_rflags(&svm->vcpu, 2);
1571 save->rip = 0x0000fff0;
1572 svm->vcpu.arch.regs[VCPU_REGS_RIP] = save->rip;
1575 * svm_set_cr0() sets PG and WP and clears NW and CD on save->cr0.
1576 * It also updates the guest-visible cr0 value.
1578 svm_set_cr0(&svm->vcpu, X86_CR0_NW | X86_CR0_CD | X86_CR0_ET);
1579 kvm_mmu_reset_context(&svm->vcpu);
1581 save->cr4 = X86_CR4_PAE;
1585 /* Setup VMCB for Nested Paging */
1586 control->nested_ctl |= SVM_NESTED_CTL_NP_ENABLE;
1587 clr_intercept(svm, INTERCEPT_INVLPG);
1588 clr_exception_intercept(svm, PF_VECTOR);
1589 clr_cr_intercept(svm, INTERCEPT_CR3_READ);
1590 clr_cr_intercept(svm, INTERCEPT_CR3_WRITE);
1591 save->g_pat = svm->vcpu.arch.pat;
1595 svm->asid_generation = 0;
1597 svm->nested.vmcb = 0;
1598 svm->vcpu.arch.hflags = 0;
1600 if (pause_filter_count) {
1601 control->pause_filter_count = pause_filter_count;
1602 if (pause_filter_thresh)
1603 control->pause_filter_thresh = pause_filter_thresh;
1604 set_intercept(svm, INTERCEPT_PAUSE);
1606 clr_intercept(svm, INTERCEPT_PAUSE);
1609 if (kvm_vcpu_apicv_active(&svm->vcpu))
1610 avic_init_vmcb(svm);
1613 * If hardware supports Virtual VMLOAD VMSAVE then enable it
1614 * in VMCB and clear intercepts to avoid #VMEXIT.
1617 clr_intercept(svm, INTERCEPT_VMLOAD);
1618 clr_intercept(svm, INTERCEPT_VMSAVE);
1619 svm->vmcb->control.virt_ext |= VIRTUAL_VMLOAD_VMSAVE_ENABLE_MASK;
1623 clr_intercept(svm, INTERCEPT_STGI);
1624 clr_intercept(svm, INTERCEPT_CLGI);
1625 svm->vmcb->control.int_ctl |= V_GIF_ENABLE_MASK;
1628 if (sev_guest(svm->vcpu.kvm)) {
1629 svm->vmcb->control.nested_ctl |= SVM_NESTED_CTL_SEV_ENABLE;
1630 clr_exception_intercept(svm, UD_VECTOR);
1633 mark_all_dirty(svm->vmcb);
1639 static u64 *avic_get_physical_id_entry(struct kvm_vcpu *vcpu,
1642 u64 *avic_physical_id_table;
1643 struct kvm_svm *kvm_svm = to_kvm_svm(vcpu->kvm);
1645 if (index >= AVIC_MAX_PHYSICAL_ID_COUNT)
1648 avic_physical_id_table = page_address(kvm_svm->avic_physical_id_table_page);
1650 return &avic_physical_id_table[index];
1655 * AVIC hardware walks the nested page table to check permissions,
1656 * but does not use the SPA address specified in the leaf page
1657 * table entry since it uses address in the AVIC_BACKING_PAGE pointer
1658 * field of the VMCB. Therefore, we set up the
1659 * APIC_ACCESS_PAGE_PRIVATE_MEMSLOT (4KB) here.
1661 static int avic_init_access_page(struct kvm_vcpu *vcpu)
1663 struct kvm *kvm = vcpu->kvm;
1666 mutex_lock(&kvm->slots_lock);
1667 if (kvm->arch.apic_access_page_done)
1670 ret = __x86_set_memory_region(kvm,
1671 APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
1672 APIC_DEFAULT_PHYS_BASE,
1677 kvm->arch.apic_access_page_done = true;
1679 mutex_unlock(&kvm->slots_lock);
1683 static int avic_init_backing_page(struct kvm_vcpu *vcpu)
1686 u64 *entry, new_entry;
1687 int id = vcpu->vcpu_id;
1688 struct vcpu_svm *svm = to_svm(vcpu);
1690 ret = avic_init_access_page(vcpu);
1694 if (id >= AVIC_MAX_PHYSICAL_ID_COUNT)
1697 if (!svm->vcpu.arch.apic->regs)
1700 svm->avic_backing_page = virt_to_page(svm->vcpu.arch.apic->regs);
1702 /* Setting AVIC backing page address in the phy APIC ID table */
1703 entry = avic_get_physical_id_entry(vcpu, id);
1707 new_entry = READ_ONCE(*entry);
1708 new_entry = __sme_set((page_to_phys(svm->avic_backing_page) &
1709 AVIC_PHYSICAL_ID_ENTRY_BACKING_PAGE_MASK) |
1710 AVIC_PHYSICAL_ID_ENTRY_VALID_MASK);
1711 WRITE_ONCE(*entry, new_entry);
1713 svm->avic_physical_id_cache = entry;
1718 static void __sev_asid_free(int asid)
1720 struct svm_cpu_data *sd;
1724 clear_bit(pos, sev_asid_bitmap);
1726 for_each_possible_cpu(cpu) {
1727 sd = per_cpu(svm_data, cpu);
1728 sd->sev_vmcbs[pos] = NULL;
1732 static void sev_asid_free(struct kvm *kvm)
1734 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1736 __sev_asid_free(sev->asid);
1739 static void sev_unbind_asid(struct kvm *kvm, unsigned int handle)
1741 struct sev_data_decommission *decommission;
1742 struct sev_data_deactivate *data;
1747 data = kzalloc(sizeof(*data), GFP_KERNEL);
1751 /* deactivate handle */
1752 data->handle = handle;
1753 sev_guest_deactivate(data, NULL);
1755 wbinvd_on_all_cpus();
1756 sev_guest_df_flush(NULL);
1759 decommission = kzalloc(sizeof(*decommission), GFP_KERNEL);
1763 /* decommission handle */
1764 decommission->handle = handle;
1765 sev_guest_decommission(decommission, NULL);
1767 kfree(decommission);
1770 static struct page **sev_pin_memory(struct kvm *kvm, unsigned long uaddr,
1771 unsigned long ulen, unsigned long *n,
1774 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1775 unsigned long npages, npinned, size;
1776 unsigned long locked, lock_limit;
1777 struct page **pages;
1778 unsigned long first, last;
1780 if (ulen == 0 || uaddr + ulen < uaddr)
1783 /* Calculate number of pages. */
1784 first = (uaddr & PAGE_MASK) >> PAGE_SHIFT;
1785 last = ((uaddr + ulen - 1) & PAGE_MASK) >> PAGE_SHIFT;
1786 npages = (last - first + 1);
1788 locked = sev->pages_locked + npages;
1789 lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1790 if (locked > lock_limit && !capable(CAP_IPC_LOCK)) {
1791 pr_err("SEV: %lu locked pages exceed the lock limit of %lu.\n", locked, lock_limit);
1795 /* Avoid using vmalloc for smaller buffers. */
1796 size = npages * sizeof(struct page *);
1797 if (size > PAGE_SIZE)
1798 pages = vmalloc(size);
1800 pages = kmalloc(size, GFP_KERNEL);
1805 /* Pin the user virtual address. */
1806 npinned = get_user_pages_fast(uaddr, npages, write ? FOLL_WRITE : 0, pages);
1807 if (npinned != npages) {
1808 pr_err("SEV: Failure locking %lu pages.\n", npages);
1813 sev->pages_locked = locked;
1819 release_pages(pages, npinned);
1825 static void sev_unpin_memory(struct kvm *kvm, struct page **pages,
1826 unsigned long npages)
1828 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1830 release_pages(pages, npages);
1832 sev->pages_locked -= npages;
1835 static void sev_clflush_pages(struct page *pages[], unsigned long npages)
1837 uint8_t *page_virtual;
1840 if (npages == 0 || pages == NULL)
1843 for (i = 0; i < npages; i++) {
1844 page_virtual = kmap_atomic(pages[i]);
1845 clflush_cache_range(page_virtual, PAGE_SIZE);
1846 kunmap_atomic(page_virtual);
1850 static void __unregister_enc_region_locked(struct kvm *kvm,
1851 struct enc_region *region)
1854 * The guest may change the memory encryption attribute from C=0 -> C=1
1855 * or vice versa for this memory range. Lets make sure caches are
1856 * flushed to ensure that guest data gets written into memory with
1859 sev_clflush_pages(region->pages, region->npages);
1861 sev_unpin_memory(kvm, region->pages, region->npages);
1862 list_del(®ion->list);
1866 static struct kvm *svm_vm_alloc(void)
1868 struct kvm_svm *kvm_svm = vzalloc(sizeof(struct kvm_svm));
1869 return &kvm_svm->kvm;
1872 static void svm_vm_free(struct kvm *kvm)
1874 vfree(to_kvm_svm(kvm));
1877 static void sev_vm_destroy(struct kvm *kvm)
1879 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1880 struct list_head *head = &sev->regions_list;
1881 struct list_head *pos, *q;
1883 if (!sev_guest(kvm))
1886 mutex_lock(&kvm->lock);
1889 * if userspace was terminated before unregistering the memory regions
1890 * then lets unpin all the registered memory.
1892 if (!list_empty(head)) {
1893 list_for_each_safe(pos, q, head) {
1894 __unregister_enc_region_locked(kvm,
1895 list_entry(pos, struct enc_region, list));
1899 mutex_unlock(&kvm->lock);
1901 sev_unbind_asid(kvm, sev->handle);
1905 static void avic_vm_destroy(struct kvm *kvm)
1907 unsigned long flags;
1908 struct kvm_svm *kvm_svm = to_kvm_svm(kvm);
1913 if (kvm_svm->avic_logical_id_table_page)
1914 __free_page(kvm_svm->avic_logical_id_table_page);
1915 if (kvm_svm->avic_physical_id_table_page)
1916 __free_page(kvm_svm->avic_physical_id_table_page);
1918 spin_lock_irqsave(&svm_vm_data_hash_lock, flags);
1919 hash_del(&kvm_svm->hnode);
1920 spin_unlock_irqrestore(&svm_vm_data_hash_lock, flags);
1923 static void svm_vm_destroy(struct kvm *kvm)
1925 avic_vm_destroy(kvm);
1926 sev_vm_destroy(kvm);
1929 static int avic_vm_init(struct kvm *kvm)
1931 unsigned long flags;
1933 struct kvm_svm *kvm_svm = to_kvm_svm(kvm);
1935 struct page *p_page;
1936 struct page *l_page;
1942 /* Allocating physical APIC ID table (4KB) */
1943 p_page = alloc_page(GFP_KERNEL);
1947 kvm_svm->avic_physical_id_table_page = p_page;
1948 clear_page(page_address(p_page));
1950 /* Allocating logical APIC ID table (4KB) */
1951 l_page = alloc_page(GFP_KERNEL);
1955 kvm_svm->avic_logical_id_table_page = l_page;
1956 clear_page(page_address(l_page));
1958 spin_lock_irqsave(&svm_vm_data_hash_lock, flags);
1960 vm_id = next_vm_id = (next_vm_id + 1) & AVIC_VM_ID_MASK;
1961 if (vm_id == 0) { /* id is 1-based, zero is not okay */
1962 next_vm_id_wrapped = 1;
1965 /* Is it still in use? Only possible if wrapped at least once */
1966 if (next_vm_id_wrapped) {
1967 hash_for_each_possible(svm_vm_data_hash, k2, hnode, vm_id) {
1968 if (k2->avic_vm_id == vm_id)
1972 kvm_svm->avic_vm_id = vm_id;
1973 hash_add(svm_vm_data_hash, &kvm_svm->hnode, kvm_svm->avic_vm_id);
1974 spin_unlock_irqrestore(&svm_vm_data_hash_lock, flags);
1979 avic_vm_destroy(kvm);
1984 avic_update_iommu_vcpu_affinity(struct kvm_vcpu *vcpu, int cpu, bool r)
1987 unsigned long flags;
1988 struct amd_svm_iommu_ir *ir;
1989 struct vcpu_svm *svm = to_svm(vcpu);
1991 if (!kvm_arch_has_assigned_device(vcpu->kvm))
1995 * Here, we go through the per-vcpu ir_list to update all existing
1996 * interrupt remapping table entry targeting this vcpu.
1998 spin_lock_irqsave(&svm->ir_list_lock, flags);
2000 if (list_empty(&svm->ir_list))
2003 list_for_each_entry(ir, &svm->ir_list, node) {
2004 ret = amd_iommu_update_ga(cpu, r, ir->data);
2009 spin_unlock_irqrestore(&svm->ir_list_lock, flags);
2013 static void avic_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2016 /* ID = 0xff (broadcast), ID > 0xff (reserved) */
2017 int h_physical_id = kvm_cpu_get_apicid(cpu);
2018 struct vcpu_svm *svm = to_svm(vcpu);
2020 if (!kvm_vcpu_apicv_active(vcpu))
2023 if (WARN_ON(h_physical_id >= AVIC_MAX_PHYSICAL_ID_COUNT))
2026 entry = READ_ONCE(*(svm->avic_physical_id_cache));
2027 WARN_ON(entry & AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK);
2029 entry &= ~AVIC_PHYSICAL_ID_ENTRY_HOST_PHYSICAL_ID_MASK;
2030 entry |= (h_physical_id & AVIC_PHYSICAL_ID_ENTRY_HOST_PHYSICAL_ID_MASK);
2032 entry &= ~AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK;
2033 if (svm->avic_is_running)
2034 entry |= AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK;
2036 WRITE_ONCE(*(svm->avic_physical_id_cache), entry);
2037 avic_update_iommu_vcpu_affinity(vcpu, h_physical_id,
2038 svm->avic_is_running);
2041 static void avic_vcpu_put(struct kvm_vcpu *vcpu)
2044 struct vcpu_svm *svm = to_svm(vcpu);
2046 if (!kvm_vcpu_apicv_active(vcpu))
2049 entry = READ_ONCE(*(svm->avic_physical_id_cache));
2050 if (entry & AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK)
2051 avic_update_iommu_vcpu_affinity(vcpu, -1, 0);
2053 entry &= ~AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK;
2054 WRITE_ONCE(*(svm->avic_physical_id_cache), entry);
2058 * This function is called during VCPU halt/unhalt.
2060 static void avic_set_running(struct kvm_vcpu *vcpu, bool is_run)
2062 struct vcpu_svm *svm = to_svm(vcpu);
2064 svm->avic_is_running = is_run;
2066 avic_vcpu_load(vcpu, vcpu->cpu);
2068 avic_vcpu_put(vcpu);
2071 static void svm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
2073 struct vcpu_svm *svm = to_svm(vcpu);
2077 vcpu->arch.microcode_version = 0x01000065;
2079 svm->virt_spec_ctrl = 0;
2082 svm->vcpu.arch.apic_base = APIC_DEFAULT_PHYS_BASE |
2083 MSR_IA32_APICBASE_ENABLE;
2084 if (kvm_vcpu_is_reset_bsp(&svm->vcpu))
2085 svm->vcpu.arch.apic_base |= MSR_IA32_APICBASE_BSP;
2089 kvm_cpuid(vcpu, &eax, &dummy, &dummy, &dummy, true);
2090 kvm_register_write(vcpu, VCPU_REGS_RDX, eax);
2092 if (kvm_vcpu_apicv_active(vcpu) && !init_event)
2093 avic_update_vapic_bar(svm, APIC_DEFAULT_PHYS_BASE);
2096 static int avic_init_vcpu(struct vcpu_svm *svm)
2100 if (!kvm_vcpu_apicv_active(&svm->vcpu))
2103 ret = avic_init_backing_page(&svm->vcpu);
2107 INIT_LIST_HEAD(&svm->ir_list);
2108 spin_lock_init(&svm->ir_list_lock);
2113 static struct kvm_vcpu *svm_create_vcpu(struct kvm *kvm, unsigned int id)
2115 struct vcpu_svm *svm;
2117 struct page *msrpm_pages;
2118 struct page *hsave_page;
2119 struct page *nested_msrpm_pages;
2122 svm = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
2128 svm->vcpu.arch.guest_fpu = kmem_cache_zalloc(x86_fpu_cache, GFP_KERNEL);
2129 if (!svm->vcpu.arch.guest_fpu) {
2130 printk(KERN_ERR "kvm: failed to allocate vcpu's fpu\n");
2132 goto free_partial_svm;
2135 err = kvm_vcpu_init(&svm->vcpu, kvm, id);
2140 page = alloc_page(GFP_KERNEL);
2144 msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
2148 nested_msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
2149 if (!nested_msrpm_pages)
2152 hsave_page = alloc_page(GFP_KERNEL);
2156 err = avic_init_vcpu(svm);
2160 /* We initialize this flag to true to make sure that the is_running
2161 * bit would be set the first time the vcpu is loaded.
2163 svm->avic_is_running = true;
2165 svm->nested.hsave = page_address(hsave_page);
2167 svm->msrpm = page_address(msrpm_pages);
2168 svm_vcpu_init_msrpm(svm->msrpm);
2170 svm->nested.msrpm = page_address(nested_msrpm_pages);
2171 svm_vcpu_init_msrpm(svm->nested.msrpm);
2173 svm->vmcb = page_address(page);
2174 clear_page(svm->vmcb);
2175 svm->vmcb_pa = __sme_set(page_to_pfn(page) << PAGE_SHIFT);
2176 svm->asid_generation = 0;
2179 svm_init_osvw(&svm->vcpu);
2184 __free_page(hsave_page);
2186 __free_pages(nested_msrpm_pages, MSRPM_ALLOC_ORDER);
2188 __free_pages(msrpm_pages, MSRPM_ALLOC_ORDER);
2192 kvm_vcpu_uninit(&svm->vcpu);
2194 kmem_cache_free(x86_fpu_cache, svm->vcpu.arch.guest_fpu);
2196 kmem_cache_free(kvm_vcpu_cache, svm);
2198 return ERR_PTR(err);
2201 static void svm_clear_current_vmcb(struct vmcb *vmcb)
2205 for_each_online_cpu(i)
2206 cmpxchg(&per_cpu(svm_data, i)->current_vmcb, vmcb, NULL);
2209 static void svm_free_vcpu(struct kvm_vcpu *vcpu)
2211 struct vcpu_svm *svm = to_svm(vcpu);
2214 * The vmcb page can be recycled, causing a false negative in
2215 * svm_vcpu_load(). So, ensure that no logical CPU has this
2216 * vmcb page recorded as its current vmcb.
2218 svm_clear_current_vmcb(svm->vmcb);
2220 __free_page(pfn_to_page(__sme_clr(svm->vmcb_pa) >> PAGE_SHIFT));
2221 __free_pages(virt_to_page(svm->msrpm), MSRPM_ALLOC_ORDER);
2222 __free_page(virt_to_page(svm->nested.hsave));
2223 __free_pages(virt_to_page(svm->nested.msrpm), MSRPM_ALLOC_ORDER);
2224 kvm_vcpu_uninit(vcpu);
2225 kmem_cache_free(x86_fpu_cache, svm->vcpu.arch.guest_fpu);
2226 kmem_cache_free(kvm_vcpu_cache, svm);
2229 static void svm_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2231 struct vcpu_svm *svm = to_svm(vcpu);
2232 struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
2235 if (unlikely(cpu != vcpu->cpu)) {
2236 svm->asid_generation = 0;
2237 mark_all_dirty(svm->vmcb);
2240 #ifdef CONFIG_X86_64
2241 rdmsrl(MSR_GS_BASE, to_svm(vcpu)->host.gs_base);
2243 savesegment(fs, svm->host.fs);
2244 savesegment(gs, svm->host.gs);
2245 svm->host.ldt = kvm_read_ldt();
2247 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
2248 rdmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
2250 if (static_cpu_has(X86_FEATURE_TSCRATEMSR)) {
2251 u64 tsc_ratio = vcpu->arch.tsc_scaling_ratio;
2252 if (tsc_ratio != __this_cpu_read(current_tsc_ratio)) {
2253 __this_cpu_write(current_tsc_ratio, tsc_ratio);
2254 wrmsrl(MSR_AMD64_TSC_RATIO, tsc_ratio);
2257 /* This assumes that the kernel never uses MSR_TSC_AUX */
2258 if (static_cpu_has(X86_FEATURE_RDTSCP))
2259 wrmsrl(MSR_TSC_AUX, svm->tsc_aux);
2261 if (sd->current_vmcb != svm->vmcb) {
2262 sd->current_vmcb = svm->vmcb;
2263 indirect_branch_prediction_barrier();
2265 avic_vcpu_load(vcpu, cpu);
2268 static void svm_vcpu_put(struct kvm_vcpu *vcpu)
2270 struct vcpu_svm *svm = to_svm(vcpu);
2273 avic_vcpu_put(vcpu);
2275 ++vcpu->stat.host_state_reload;
2276 kvm_load_ldt(svm->host.ldt);
2277 #ifdef CONFIG_X86_64
2278 loadsegment(fs, svm->host.fs);
2279 wrmsrl(MSR_KERNEL_GS_BASE, current->thread.gsbase);
2280 load_gs_index(svm->host.gs);
2282 #ifdef CONFIG_X86_32_LAZY_GS
2283 loadsegment(gs, svm->host.gs);
2286 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
2287 wrmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
2290 static void svm_vcpu_blocking(struct kvm_vcpu *vcpu)
2292 avic_set_running(vcpu, false);
2295 static void svm_vcpu_unblocking(struct kvm_vcpu *vcpu)
2297 avic_set_running(vcpu, true);
2300 static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu)
2302 struct vcpu_svm *svm = to_svm(vcpu);
2303 unsigned long rflags = svm->vmcb->save.rflags;
2305 if (svm->nmi_singlestep) {
2306 /* Hide our flags if they were not set by the guest */
2307 if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_TF))
2308 rflags &= ~X86_EFLAGS_TF;
2309 if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_RF))
2310 rflags &= ~X86_EFLAGS_RF;
2315 static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
2317 if (to_svm(vcpu)->nmi_singlestep)
2318 rflags |= (X86_EFLAGS_TF | X86_EFLAGS_RF);
2321 * Any change of EFLAGS.VM is accompanied by a reload of SS
2322 * (caused by either a task switch or an inter-privilege IRET),
2323 * so we do not need to update the CPL here.
2325 to_svm(vcpu)->vmcb->save.rflags = rflags;
2328 static void svm_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
2331 case VCPU_EXREG_PDPTR:
2332 BUG_ON(!npt_enabled);
2333 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
2340 static void svm_set_vintr(struct vcpu_svm *svm)
2342 set_intercept(svm, INTERCEPT_VINTR);
2345 static void svm_clear_vintr(struct vcpu_svm *svm)
2347 clr_intercept(svm, INTERCEPT_VINTR);
2350 static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg)
2352 struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
2355 case VCPU_SREG_CS: return &save->cs;
2356 case VCPU_SREG_DS: return &save->ds;
2357 case VCPU_SREG_ES: return &save->es;
2358 case VCPU_SREG_FS: return &save->fs;
2359 case VCPU_SREG_GS: return &save->gs;
2360 case VCPU_SREG_SS: return &save->ss;
2361 case VCPU_SREG_TR: return &save->tr;
2362 case VCPU_SREG_LDTR: return &save->ldtr;
2368 static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg)
2370 struct vmcb_seg *s = svm_seg(vcpu, seg);
2375 static void svm_get_segment(struct kvm_vcpu *vcpu,
2376 struct kvm_segment *var, int seg)
2378 struct vmcb_seg *s = svm_seg(vcpu, seg);
2380 var->base = s->base;
2381 var->limit = s->limit;
2382 var->selector = s->selector;
2383 var->type = s->attrib & SVM_SELECTOR_TYPE_MASK;
2384 var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1;
2385 var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
2386 var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1;
2387 var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1;
2388 var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
2389 var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
2392 * AMD CPUs circa 2014 track the G bit for all segments except CS.
2393 * However, the SVM spec states that the G bit is not observed by the
2394 * CPU, and some VMware virtual CPUs drop the G bit for all segments.
2395 * So let's synthesize a legal G bit for all segments, this helps
2396 * running KVM nested. It also helps cross-vendor migration, because
2397 * Intel's vmentry has a check on the 'G' bit.
2399 var->g = s->limit > 0xfffff;
2402 * AMD's VMCB does not have an explicit unusable field, so emulate it
2403 * for cross vendor migration purposes by "not present"
2405 var->unusable = !var->present;
2410 * Work around a bug where the busy flag in the tr selector
2420 * The accessed bit must always be set in the segment
2421 * descriptor cache, although it can be cleared in the
2422 * descriptor, the cached bit always remains at 1. Since
2423 * Intel has a check on this, set it here to support
2424 * cross-vendor migration.
2431 * On AMD CPUs sometimes the DB bit in the segment
2432 * descriptor is left as 1, although the whole segment has
2433 * been made unusable. Clear it here to pass an Intel VMX
2434 * entry check when cross vendor migrating.
2438 /* This is symmetric with svm_set_segment() */
2439 var->dpl = to_svm(vcpu)->vmcb->save.cpl;
2444 static int svm_get_cpl(struct kvm_vcpu *vcpu)
2446 struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
2451 static void svm_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
2453 struct vcpu_svm *svm = to_svm(vcpu);
2455 dt->size = svm->vmcb->save.idtr.limit;
2456 dt->address = svm->vmcb->save.idtr.base;
2459 static void svm_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
2461 struct vcpu_svm *svm = to_svm(vcpu);
2463 svm->vmcb->save.idtr.limit = dt->size;
2464 svm->vmcb->save.idtr.base = dt->address ;
2465 mark_dirty(svm->vmcb, VMCB_DT);
2468 static void svm_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
2470 struct vcpu_svm *svm = to_svm(vcpu);
2472 dt->size = svm->vmcb->save.gdtr.limit;
2473 dt->address = svm->vmcb->save.gdtr.base;
2476 static void svm_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
2478 struct vcpu_svm *svm = to_svm(vcpu);
2480 svm->vmcb->save.gdtr.limit = dt->size;
2481 svm->vmcb->save.gdtr.base = dt->address ;
2482 mark_dirty(svm->vmcb, VMCB_DT);
2485 static void svm_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
2489 static void svm_decache_cr3(struct kvm_vcpu *vcpu)
2493 static void svm_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
2497 static void update_cr0_intercept(struct vcpu_svm *svm)
2499 ulong gcr0 = svm->vcpu.arch.cr0;
2500 u64 *hcr0 = &svm->vmcb->save.cr0;
2502 *hcr0 = (*hcr0 & ~SVM_CR0_SELECTIVE_MASK)
2503 | (gcr0 & SVM_CR0_SELECTIVE_MASK);
2505 mark_dirty(svm->vmcb, VMCB_CR);
2507 if (gcr0 == *hcr0) {
2508 clr_cr_intercept(svm, INTERCEPT_CR0_READ);
2509 clr_cr_intercept(svm, INTERCEPT_CR0_WRITE);
2511 set_cr_intercept(svm, INTERCEPT_CR0_READ);
2512 set_cr_intercept(svm, INTERCEPT_CR0_WRITE);
2516 static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
2518 struct vcpu_svm *svm = to_svm(vcpu);
2520 #ifdef CONFIG_X86_64
2521 if (vcpu->arch.efer & 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;
2527 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG)) {
2528 vcpu->arch.efer &= ~EFER_LMA;
2529 svm->vmcb->save.efer &= ~(EFER_LMA | EFER_LME);
2533 vcpu->arch.cr0 = cr0;
2536 cr0 |= X86_CR0_PG | X86_CR0_WP;
2539 * re-enable caching here because the QEMU bios
2540 * does not do it - this results in some delay at
2543 if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
2544 cr0 &= ~(X86_CR0_CD | X86_CR0_NW);
2545 svm->vmcb->save.cr0 = cr0;
2546 mark_dirty(svm->vmcb, VMCB_CR);
2547 update_cr0_intercept(svm);
2550 static int svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
2552 unsigned long host_cr4_mce = cr4_read_shadow() & X86_CR4_MCE;
2553 unsigned long old_cr4 = to_svm(vcpu)->vmcb->save.cr4;
2555 if (cr4 & X86_CR4_VMXE)
2558 if (npt_enabled && ((old_cr4 ^ cr4) & X86_CR4_PGE))
2559 svm_flush_tlb(vcpu, true);
2561 vcpu->arch.cr4 = cr4;
2564 cr4 |= host_cr4_mce;
2565 to_svm(vcpu)->vmcb->save.cr4 = cr4;
2566 mark_dirty(to_svm(vcpu)->vmcb, VMCB_CR);
2570 static void svm_set_segment(struct kvm_vcpu *vcpu,
2571 struct kvm_segment *var, int seg)
2573 struct vcpu_svm *svm = to_svm(vcpu);
2574 struct vmcb_seg *s = svm_seg(vcpu, seg);
2576 s->base = var->base;
2577 s->limit = var->limit;
2578 s->selector = var->selector;
2579 s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK);
2580 s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT;
2581 s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT;
2582 s->attrib |= ((var->present & 1) && !var->unusable) << SVM_SELECTOR_P_SHIFT;
2583 s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT;
2584 s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT;
2585 s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT;
2586 s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT;
2589 * This is always accurate, except if SYSRET returned to a segment
2590 * with SS.DPL != 3. Intel does not have this quirk, and always
2591 * forces SS.DPL to 3 on sysret, so we ignore that case; fixing it
2592 * would entail passing the CPL to userspace and back.
2594 if (seg == VCPU_SREG_SS)
2595 /* This is symmetric with svm_get_segment() */
2596 svm->vmcb->save.cpl = (var->dpl & 3);
2598 mark_dirty(svm->vmcb, VMCB_SEG);
2601 static void update_bp_intercept(struct kvm_vcpu *vcpu)
2603 struct vcpu_svm *svm = to_svm(vcpu);
2605 clr_exception_intercept(svm, BP_VECTOR);
2607 if (vcpu->guest_debug & KVM_GUESTDBG_ENABLE) {
2608 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
2609 set_exception_intercept(svm, BP_VECTOR);
2611 vcpu->guest_debug = 0;
2614 static void new_asid(struct vcpu_svm *svm, struct svm_cpu_data *sd)
2616 if (sd->next_asid > sd->max_asid) {
2617 ++sd->asid_generation;
2618 sd->next_asid = sd->min_asid;
2619 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID;
2622 svm->asid_generation = sd->asid_generation;
2623 svm->vmcb->control.asid = sd->next_asid++;
2625 mark_dirty(svm->vmcb, VMCB_ASID);
2628 static u64 svm_get_dr6(struct kvm_vcpu *vcpu)
2630 return to_svm(vcpu)->vmcb->save.dr6;
2633 static void svm_set_dr6(struct kvm_vcpu *vcpu, unsigned long value)
2635 struct vcpu_svm *svm = to_svm(vcpu);
2637 svm->vmcb->save.dr6 = value;
2638 mark_dirty(svm->vmcb, VMCB_DR);
2641 static void svm_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
2643 struct vcpu_svm *svm = to_svm(vcpu);
2645 get_debugreg(vcpu->arch.db[0], 0);
2646 get_debugreg(vcpu->arch.db[1], 1);
2647 get_debugreg(vcpu->arch.db[2], 2);
2648 get_debugreg(vcpu->arch.db[3], 3);
2649 vcpu->arch.dr6 = svm_get_dr6(vcpu);
2650 vcpu->arch.dr7 = svm->vmcb->save.dr7;
2652 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
2653 set_dr_intercepts(svm);
2656 static void svm_set_dr7(struct kvm_vcpu *vcpu, unsigned long value)
2658 struct vcpu_svm *svm = to_svm(vcpu);
2660 svm->vmcb->save.dr7 = value;
2661 mark_dirty(svm->vmcb, VMCB_DR);
2664 static int pf_interception(struct vcpu_svm *svm)
2666 u64 fault_address = __sme_clr(svm->vmcb->control.exit_info_2);
2667 u64 error_code = svm->vmcb->control.exit_info_1;
2669 return kvm_handle_page_fault(&svm->vcpu, error_code, fault_address,
2670 static_cpu_has(X86_FEATURE_DECODEASSISTS) ?
2671 svm->vmcb->control.insn_bytes : NULL,
2672 svm->vmcb->control.insn_len);
2675 static int npf_interception(struct vcpu_svm *svm)
2677 u64 fault_address = __sme_clr(svm->vmcb->control.exit_info_2);
2678 u64 error_code = svm->vmcb->control.exit_info_1;
2680 trace_kvm_page_fault(fault_address, error_code);
2681 return kvm_mmu_page_fault(&svm->vcpu, fault_address, error_code,
2682 static_cpu_has(X86_FEATURE_DECODEASSISTS) ?
2683 svm->vmcb->control.insn_bytes : NULL,
2684 svm->vmcb->control.insn_len);
2687 static int db_interception(struct vcpu_svm *svm)
2689 struct kvm_run *kvm_run = svm->vcpu.run;
2691 if (!(svm->vcpu.guest_debug &
2692 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) &&
2693 !svm->nmi_singlestep) {
2694 kvm_queue_exception(&svm->vcpu, DB_VECTOR);
2698 if (svm->nmi_singlestep) {
2699 disable_nmi_singlestep(svm);
2702 if (svm->vcpu.guest_debug &
2703 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) {
2704 kvm_run->exit_reason = KVM_EXIT_DEBUG;
2705 kvm_run->debug.arch.pc =
2706 svm->vmcb->save.cs.base + svm->vmcb->save.rip;
2707 kvm_run->debug.arch.exception = DB_VECTOR;
2714 static int bp_interception(struct vcpu_svm *svm)
2716 struct kvm_run *kvm_run = svm->vcpu.run;
2718 kvm_run->exit_reason = KVM_EXIT_DEBUG;
2719 kvm_run->debug.arch.pc = svm->vmcb->save.cs.base + svm->vmcb->save.rip;
2720 kvm_run->debug.arch.exception = BP_VECTOR;
2724 static int ud_interception(struct vcpu_svm *svm)
2726 return handle_ud(&svm->vcpu);
2729 static int ac_interception(struct vcpu_svm *svm)
2731 kvm_queue_exception_e(&svm->vcpu, AC_VECTOR, 0);
2735 static int gp_interception(struct vcpu_svm *svm)
2737 struct kvm_vcpu *vcpu = &svm->vcpu;
2738 u32 error_code = svm->vmcb->control.exit_info_1;
2741 WARN_ON_ONCE(!enable_vmware_backdoor);
2743 er = kvm_emulate_instruction(vcpu,
2744 EMULTYPE_VMWARE | EMULTYPE_NO_UD_ON_FAIL);
2745 if (er == EMULATE_USER_EXIT)
2747 else if (er != EMULATE_DONE)
2748 kvm_queue_exception_e(vcpu, GP_VECTOR, error_code);
2752 static bool is_erratum_383(void)
2757 if (!erratum_383_found)
2760 value = native_read_msr_safe(MSR_IA32_MC0_STATUS, &err);
2764 /* Bit 62 may or may not be set for this mce */
2765 value &= ~(1ULL << 62);
2767 if (value != 0xb600000000010015ULL)
2770 /* Clear MCi_STATUS registers */
2771 for (i = 0; i < 6; ++i)
2772 native_write_msr_safe(MSR_IA32_MCx_STATUS(i), 0, 0);
2774 value = native_read_msr_safe(MSR_IA32_MCG_STATUS, &err);
2778 value &= ~(1ULL << 2);
2779 low = lower_32_bits(value);
2780 high = upper_32_bits(value);
2782 native_write_msr_safe(MSR_IA32_MCG_STATUS, low, high);
2785 /* Flush tlb to evict multi-match entries */
2791 static void svm_handle_mce(struct vcpu_svm *svm)
2793 if (is_erratum_383()) {
2795 * Erratum 383 triggered. Guest state is corrupt so kill the
2798 pr_err("KVM: Guest triggered AMD Erratum 383\n");
2800 kvm_make_request(KVM_REQ_TRIPLE_FAULT, &svm->vcpu);
2806 * On an #MC intercept the MCE handler is not called automatically in
2807 * the host. So do it by hand here.
2811 /* not sure if we ever come back to this point */
2816 static int mc_interception(struct vcpu_svm *svm)
2821 static int shutdown_interception(struct vcpu_svm *svm)
2823 struct kvm_run *kvm_run = svm->vcpu.run;
2826 * VMCB is undefined after a SHUTDOWN intercept
2827 * so reinitialize it.
2829 clear_page(svm->vmcb);
2832 kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
2836 static int io_interception(struct vcpu_svm *svm)
2838 struct kvm_vcpu *vcpu = &svm->vcpu;
2839 u32 io_info = svm->vmcb->control.exit_info_1; /* address size bug? */
2840 int size, in, string;
2843 ++svm->vcpu.stat.io_exits;
2844 string = (io_info & SVM_IOIO_STR_MASK) != 0;
2845 in = (io_info & SVM_IOIO_TYPE_MASK) != 0;
2847 return kvm_emulate_instruction(vcpu, 0) == EMULATE_DONE;
2849 port = io_info >> 16;
2850 size = (io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT;
2851 svm->next_rip = svm->vmcb->control.exit_info_2;
2853 return kvm_fast_pio(&svm->vcpu, size, port, in);
2856 static int nmi_interception(struct vcpu_svm *svm)
2861 static int intr_interception(struct vcpu_svm *svm)
2863 ++svm->vcpu.stat.irq_exits;
2867 static int nop_on_interception(struct vcpu_svm *svm)
2872 static int halt_interception(struct vcpu_svm *svm)
2874 svm->next_rip = kvm_rip_read(&svm->vcpu) + 1;
2875 return kvm_emulate_halt(&svm->vcpu);
2878 static int vmmcall_interception(struct vcpu_svm *svm)
2880 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2881 return kvm_emulate_hypercall(&svm->vcpu);
2884 static unsigned long nested_svm_get_tdp_cr3(struct kvm_vcpu *vcpu)
2886 struct vcpu_svm *svm = to_svm(vcpu);
2888 return svm->nested.nested_cr3;
2891 static u64 nested_svm_get_tdp_pdptr(struct kvm_vcpu *vcpu, int index)
2893 struct vcpu_svm *svm = to_svm(vcpu);
2894 u64 cr3 = svm->nested.nested_cr3;
2898 ret = kvm_vcpu_read_guest_page(vcpu, gpa_to_gfn(__sme_clr(cr3)), &pdpte,
2899 offset_in_page(cr3) + index * 8, 8);
2905 static void nested_svm_set_tdp_cr3(struct kvm_vcpu *vcpu,
2908 struct vcpu_svm *svm = to_svm(vcpu);
2910 svm->vmcb->control.nested_cr3 = __sme_set(root);
2911 mark_dirty(svm->vmcb, VMCB_NPT);
2914 static void nested_svm_inject_npf_exit(struct kvm_vcpu *vcpu,
2915 struct x86_exception *fault)
2917 struct vcpu_svm *svm = to_svm(vcpu);
2919 if (svm->vmcb->control.exit_code != SVM_EXIT_NPF) {
2921 * TODO: track the cause of the nested page fault, and
2922 * correctly fill in the high bits of exit_info_1.
2924 svm->vmcb->control.exit_code = SVM_EXIT_NPF;
2925 svm->vmcb->control.exit_code_hi = 0;
2926 svm->vmcb->control.exit_info_1 = (1ULL << 32);
2927 svm->vmcb->control.exit_info_2 = fault->address;
2930 svm->vmcb->control.exit_info_1 &= ~0xffffffffULL;
2931 svm->vmcb->control.exit_info_1 |= fault->error_code;
2934 * The present bit is always zero for page structure faults on real
2937 if (svm->vmcb->control.exit_info_1 & (2ULL << 32))
2938 svm->vmcb->control.exit_info_1 &= ~1;
2940 nested_svm_vmexit(svm);
2943 static void nested_svm_init_mmu_context(struct kvm_vcpu *vcpu)
2945 WARN_ON(mmu_is_nested(vcpu));
2947 vcpu->arch.mmu = &vcpu->arch.guest_mmu;
2948 kvm_init_shadow_mmu(vcpu);
2949 vcpu->arch.mmu->set_cr3 = nested_svm_set_tdp_cr3;
2950 vcpu->arch.mmu->get_cr3 = nested_svm_get_tdp_cr3;
2951 vcpu->arch.mmu->get_pdptr = nested_svm_get_tdp_pdptr;
2952 vcpu->arch.mmu->inject_page_fault = nested_svm_inject_npf_exit;
2953 vcpu->arch.mmu->shadow_root_level = get_npt_level(vcpu);
2954 reset_shadow_zero_bits_mask(vcpu, vcpu->arch.mmu);
2955 vcpu->arch.walk_mmu = &vcpu->arch.nested_mmu;
2958 static void nested_svm_uninit_mmu_context(struct kvm_vcpu *vcpu)
2960 vcpu->arch.mmu = &vcpu->arch.root_mmu;
2961 vcpu->arch.walk_mmu = &vcpu->arch.root_mmu;
2964 static int nested_svm_check_permissions(struct vcpu_svm *svm)
2966 if (!(svm->vcpu.arch.efer & EFER_SVME) ||
2967 !is_paging(&svm->vcpu)) {
2968 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2972 if (svm->vmcb->save.cpl) {
2973 kvm_inject_gp(&svm->vcpu, 0);
2980 static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
2981 bool has_error_code, u32 error_code)
2985 if (!is_guest_mode(&svm->vcpu))
2988 vmexit = nested_svm_intercept(svm);
2989 if (vmexit != NESTED_EXIT_DONE)
2992 svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + nr;
2993 svm->vmcb->control.exit_code_hi = 0;
2994 svm->vmcb->control.exit_info_1 = error_code;
2997 * EXITINFO2 is undefined for all exception intercepts other
3000 if (svm->vcpu.arch.exception.nested_apf)
3001 svm->vmcb->control.exit_info_2 = svm->vcpu.arch.apf.nested_apf_token;
3002 else if (svm->vcpu.arch.exception.has_payload)
3003 svm->vmcb->control.exit_info_2 = svm->vcpu.arch.exception.payload;
3005 svm->vmcb->control.exit_info_2 = svm->vcpu.arch.cr2;
3007 svm->nested.exit_required = true;
3011 /* This function returns true if it is save to enable the irq window */
3012 static inline bool nested_svm_intr(struct vcpu_svm *svm)
3014 if (!is_guest_mode(&svm->vcpu))
3017 if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
3020 if (!(svm->vcpu.arch.hflags & HF_HIF_MASK))
3024 * if vmexit was already requested (by intercepted exception
3025 * for instance) do not overwrite it with "external interrupt"
3028 if (svm->nested.exit_required)
3031 svm->vmcb->control.exit_code = SVM_EXIT_INTR;
3032 svm->vmcb->control.exit_info_1 = 0;
3033 svm->vmcb->control.exit_info_2 = 0;
3035 if (svm->nested.intercept & 1ULL) {
3037 * The #vmexit can't be emulated here directly because this
3038 * code path runs with irqs and preemption disabled. A
3039 * #vmexit emulation might sleep. Only signal request for
3042 svm->nested.exit_required = true;
3043 trace_kvm_nested_intr_vmexit(svm->vmcb->save.rip);
3050 /* This function returns true if it is save to enable the nmi window */
3051 static inline bool nested_svm_nmi(struct vcpu_svm *svm)
3053 if (!is_guest_mode(&svm->vcpu))
3056 if (!(svm->nested.intercept & (1ULL << INTERCEPT_NMI)))
3059 svm->vmcb->control.exit_code = SVM_EXIT_NMI;
3060 svm->nested.exit_required = true;
3065 static void *nested_svm_map(struct vcpu_svm *svm, u64 gpa, struct page **_page)
3071 page = kvm_vcpu_gfn_to_page(&svm->vcpu, gpa >> PAGE_SHIFT);
3072 if (is_error_page(page))
3080 kvm_inject_gp(&svm->vcpu, 0);
3085 static void nested_svm_unmap(struct page *page)
3088 kvm_release_page_dirty(page);
3091 static int nested_svm_intercept_ioio(struct vcpu_svm *svm)
3093 unsigned port, size, iopm_len;
3098 if (!(svm->nested.intercept & (1ULL << INTERCEPT_IOIO_PROT)))
3099 return NESTED_EXIT_HOST;
3101 port = svm->vmcb->control.exit_info_1 >> 16;
3102 size = (svm->vmcb->control.exit_info_1 & SVM_IOIO_SIZE_MASK) >>
3103 SVM_IOIO_SIZE_SHIFT;
3104 gpa = svm->nested.vmcb_iopm + (port / 8);
3105 start_bit = port % 8;
3106 iopm_len = (start_bit + size > 8) ? 2 : 1;
3107 mask = (0xf >> (4 - size)) << start_bit;
3110 if (kvm_vcpu_read_guest(&svm->vcpu, gpa, &val, iopm_len))
3111 return NESTED_EXIT_DONE;
3113 return (val & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
3116 static int nested_svm_exit_handled_msr(struct vcpu_svm *svm)
3118 u32 offset, msr, value;
3121 if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT)))
3122 return NESTED_EXIT_HOST;
3124 msr = svm->vcpu.arch.regs[VCPU_REGS_RCX];
3125 offset = svm_msrpm_offset(msr);
3126 write = svm->vmcb->control.exit_info_1 & 1;
3127 mask = 1 << ((2 * (msr & 0xf)) + write);
3129 if (offset == MSR_INVALID)
3130 return NESTED_EXIT_DONE;
3132 /* Offset is in 32 bit units but need in 8 bit units */
3135 if (kvm_vcpu_read_guest(&svm->vcpu, svm->nested.vmcb_msrpm + offset, &value, 4))
3136 return NESTED_EXIT_DONE;
3138 return (value & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
3141 /* DB exceptions for our internal use must not cause vmexit */
3142 static int nested_svm_intercept_db(struct vcpu_svm *svm)
3146 /* if we're not singlestepping, it's not ours */
3147 if (!svm->nmi_singlestep)
3148 return NESTED_EXIT_DONE;
3150 /* if it's not a singlestep exception, it's not ours */
3151 if (kvm_get_dr(&svm->vcpu, 6, &dr6))
3152 return NESTED_EXIT_DONE;
3153 if (!(dr6 & DR6_BS))
3154 return NESTED_EXIT_DONE;
3156 /* if the guest is singlestepping, it should get the vmexit */
3157 if (svm->nmi_singlestep_guest_rflags & X86_EFLAGS_TF) {
3158 disable_nmi_singlestep(svm);
3159 return NESTED_EXIT_DONE;
3162 /* it's ours, the nested hypervisor must not see this one */
3163 return NESTED_EXIT_HOST;
3166 static int nested_svm_exit_special(struct vcpu_svm *svm)
3168 u32 exit_code = svm->vmcb->control.exit_code;
3170 switch (exit_code) {
3173 case SVM_EXIT_EXCP_BASE + MC_VECTOR:
3174 return NESTED_EXIT_HOST;
3176 /* For now we are always handling NPFs when using them */
3178 return NESTED_EXIT_HOST;
3180 case SVM_EXIT_EXCP_BASE + PF_VECTOR:
3181 /* When we're shadowing, trap PFs, but not async PF */
3182 if (!npt_enabled && svm->vcpu.arch.apf.host_apf_reason == 0)
3183 return NESTED_EXIT_HOST;
3189 return NESTED_EXIT_CONTINUE;
3193 * If this function returns true, this #vmexit was already handled
3195 static int nested_svm_intercept(struct vcpu_svm *svm)
3197 u32 exit_code = svm->vmcb->control.exit_code;
3198 int vmexit = NESTED_EXIT_HOST;
3200 switch (exit_code) {
3202 vmexit = nested_svm_exit_handled_msr(svm);
3205 vmexit = nested_svm_intercept_ioio(svm);
3207 case SVM_EXIT_READ_CR0 ... SVM_EXIT_WRITE_CR8: {
3208 u32 bit = 1U << (exit_code - SVM_EXIT_READ_CR0);
3209 if (svm->nested.intercept_cr & bit)
3210 vmexit = NESTED_EXIT_DONE;
3213 case SVM_EXIT_READ_DR0 ... SVM_EXIT_WRITE_DR7: {
3214 u32 bit = 1U << (exit_code - SVM_EXIT_READ_DR0);
3215 if (svm->nested.intercept_dr & bit)
3216 vmexit = NESTED_EXIT_DONE;
3219 case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
3220 u32 excp_bits = 1 << (exit_code - SVM_EXIT_EXCP_BASE);
3221 if (svm->nested.intercept_exceptions & excp_bits) {
3222 if (exit_code == SVM_EXIT_EXCP_BASE + DB_VECTOR)
3223 vmexit = nested_svm_intercept_db(svm);
3225 vmexit = NESTED_EXIT_DONE;
3227 /* async page fault always cause vmexit */
3228 else if ((exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR) &&
3229 svm->vcpu.arch.exception.nested_apf != 0)
3230 vmexit = NESTED_EXIT_DONE;
3233 case SVM_EXIT_ERR: {
3234 vmexit = NESTED_EXIT_DONE;
3238 u64 exit_bits = 1ULL << (exit_code - SVM_EXIT_INTR);
3239 if (svm->nested.intercept & exit_bits)
3240 vmexit = NESTED_EXIT_DONE;
3247 static int nested_svm_exit_handled(struct vcpu_svm *svm)
3251 vmexit = nested_svm_intercept(svm);
3253 if (vmexit == NESTED_EXIT_DONE)
3254 nested_svm_vmexit(svm);
3259 static inline void copy_vmcb_control_area(struct vmcb *dst_vmcb, struct vmcb *from_vmcb)
3261 struct vmcb_control_area *dst = &dst_vmcb->control;
3262 struct vmcb_control_area *from = &from_vmcb->control;
3264 dst->intercept_cr = from->intercept_cr;
3265 dst->intercept_dr = from->intercept_dr;
3266 dst->intercept_exceptions = from->intercept_exceptions;
3267 dst->intercept = from->intercept;
3268 dst->iopm_base_pa = from->iopm_base_pa;
3269 dst->msrpm_base_pa = from->msrpm_base_pa;
3270 dst->tsc_offset = from->tsc_offset;
3271 dst->asid = from->asid;
3272 dst->tlb_ctl = from->tlb_ctl;
3273 dst->int_ctl = from->int_ctl;
3274 dst->int_vector = from->int_vector;
3275 dst->int_state = from->int_state;
3276 dst->exit_code = from->exit_code;
3277 dst->exit_code_hi = from->exit_code_hi;
3278 dst->exit_info_1 = from->exit_info_1;
3279 dst->exit_info_2 = from->exit_info_2;
3280 dst->exit_int_info = from->exit_int_info;
3281 dst->exit_int_info_err = from->exit_int_info_err;
3282 dst->nested_ctl = from->nested_ctl;
3283 dst->event_inj = from->event_inj;
3284 dst->event_inj_err = from->event_inj_err;
3285 dst->nested_cr3 = from->nested_cr3;
3286 dst->virt_ext = from->virt_ext;
3287 dst->pause_filter_count = from->pause_filter_count;
3288 dst->pause_filter_thresh = from->pause_filter_thresh;
3291 static int nested_svm_vmexit(struct vcpu_svm *svm)
3293 struct vmcb *nested_vmcb;
3294 struct vmcb *hsave = svm->nested.hsave;
3295 struct vmcb *vmcb = svm->vmcb;
3298 trace_kvm_nested_vmexit_inject(vmcb->control.exit_code,
3299 vmcb->control.exit_info_1,
3300 vmcb->control.exit_info_2,
3301 vmcb->control.exit_int_info,
3302 vmcb->control.exit_int_info_err,
3305 nested_vmcb = nested_svm_map(svm, svm->nested.vmcb, &page);
3309 /* Exit Guest-Mode */
3310 leave_guest_mode(&svm->vcpu);
3311 svm->nested.vmcb = 0;
3313 /* Give the current vmcb to the guest */
3316 nested_vmcb->save.es = vmcb->save.es;
3317 nested_vmcb->save.cs = vmcb->save.cs;
3318 nested_vmcb->save.ss = vmcb->save.ss;
3319 nested_vmcb->save.ds = vmcb->save.ds;
3320 nested_vmcb->save.gdtr = vmcb->save.gdtr;
3321 nested_vmcb->save.idtr = vmcb->save.idtr;
3322 nested_vmcb->save.efer = svm->vcpu.arch.efer;
3323 nested_vmcb->save.cr0 = kvm_read_cr0(&svm->vcpu);
3324 nested_vmcb->save.cr3 = kvm_read_cr3(&svm->vcpu);
3325 nested_vmcb->save.cr2 = vmcb->save.cr2;
3326 nested_vmcb->save.cr4 = svm->vcpu.arch.cr4;
3327 nested_vmcb->save.rflags = kvm_get_rflags(&svm->vcpu);
3328 nested_vmcb->save.rip = vmcb->save.rip;
3329 nested_vmcb->save.rsp = vmcb->save.rsp;
3330 nested_vmcb->save.rax = vmcb->save.rax;
3331 nested_vmcb->save.dr7 = vmcb->save.dr7;
3332 nested_vmcb->save.dr6 = vmcb->save.dr6;
3333 nested_vmcb->save.cpl = vmcb->save.cpl;
3335 nested_vmcb->control.int_ctl = vmcb->control.int_ctl;
3336 nested_vmcb->control.int_vector = vmcb->control.int_vector;
3337 nested_vmcb->control.int_state = vmcb->control.int_state;
3338 nested_vmcb->control.exit_code = vmcb->control.exit_code;
3339 nested_vmcb->control.exit_code_hi = vmcb->control.exit_code_hi;
3340 nested_vmcb->control.exit_info_1 = vmcb->control.exit_info_1;
3341 nested_vmcb->control.exit_info_2 = vmcb->control.exit_info_2;
3342 nested_vmcb->control.exit_int_info = vmcb->control.exit_int_info;
3343 nested_vmcb->control.exit_int_info_err = vmcb->control.exit_int_info_err;
3345 if (svm->nrips_enabled)
3346 nested_vmcb->control.next_rip = vmcb->control.next_rip;
3349 * If we emulate a VMRUN/#VMEXIT in the same host #vmexit cycle we have
3350 * to make sure that we do not lose injected events. So check event_inj
3351 * here and copy it to exit_int_info if it is valid.
3352 * Exit_int_info and event_inj can't be both valid because the case
3353 * below only happens on a VMRUN instruction intercept which has
3354 * no valid exit_int_info set.
3356 if (vmcb->control.event_inj & SVM_EVTINJ_VALID) {
3357 struct vmcb_control_area *nc = &nested_vmcb->control;
3359 nc->exit_int_info = vmcb->control.event_inj;
3360 nc->exit_int_info_err = vmcb->control.event_inj_err;
3363 nested_vmcb->control.tlb_ctl = 0;
3364 nested_vmcb->control.event_inj = 0;
3365 nested_vmcb->control.event_inj_err = 0;
3367 nested_vmcb->control.pause_filter_count =
3368 svm->vmcb->control.pause_filter_count;
3369 nested_vmcb->control.pause_filter_thresh =
3370 svm->vmcb->control.pause_filter_thresh;
3372 /* We always set V_INTR_MASKING and remember the old value in hflags */
3373 if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
3374 nested_vmcb->control.int_ctl &= ~V_INTR_MASKING_MASK;
3376 /* Restore the original control entries */
3377 copy_vmcb_control_area(vmcb, hsave);
3379 svm->vcpu.arch.tsc_offset = svm->vmcb->control.tsc_offset;
3380 kvm_clear_exception_queue(&svm->vcpu);
3381 kvm_clear_interrupt_queue(&svm->vcpu);
3383 svm->nested.nested_cr3 = 0;
3385 /* Restore selected save entries */
3386 svm->vmcb->save.es = hsave->save.es;
3387 svm->vmcb->save.cs = hsave->save.cs;
3388 svm->vmcb->save.ss = hsave->save.ss;
3389 svm->vmcb->save.ds = hsave->save.ds;
3390 svm->vmcb->save.gdtr = hsave->save.gdtr;
3391 svm->vmcb->save.idtr = hsave->save.idtr;
3392 kvm_set_rflags(&svm->vcpu, hsave->save.rflags);
3393 svm_set_efer(&svm->vcpu, hsave->save.efer);
3394 svm_set_cr0(&svm->vcpu, hsave->save.cr0 | X86_CR0_PE);
3395 svm_set_cr4(&svm->vcpu, hsave->save.cr4);
3397 svm->vmcb->save.cr3 = hsave->save.cr3;
3398 svm->vcpu.arch.cr3 = hsave->save.cr3;
3400 (void)kvm_set_cr3(&svm->vcpu, hsave->save.cr3);
3402 kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, hsave->save.rax);
3403 kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, hsave->save.rsp);
3404 kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, hsave->save.rip);
3405 svm->vmcb->save.dr7 = 0;
3406 svm->vmcb->save.cpl = 0;
3407 svm->vmcb->control.exit_int_info = 0;
3409 mark_all_dirty(svm->vmcb);
3411 nested_svm_unmap(page);
3413 nested_svm_uninit_mmu_context(&svm->vcpu);
3414 kvm_mmu_reset_context(&svm->vcpu);
3415 kvm_mmu_load(&svm->vcpu);
3418 * Drop what we picked up for L2 via svm_complete_interrupts() so it
3419 * doesn't end up in L1.
3421 svm->vcpu.arch.nmi_injected = false;
3422 kvm_clear_exception_queue(&svm->vcpu);
3423 kvm_clear_interrupt_queue(&svm->vcpu);
3428 static bool nested_svm_vmrun_msrpm(struct vcpu_svm *svm)
3431 * This function merges the msr permission bitmaps of kvm and the
3432 * nested vmcb. It is optimized in that it only merges the parts where
3433 * the kvm msr permission bitmap may contain zero bits
3437 if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT)))
3440 for (i = 0; i < MSRPM_OFFSETS; i++) {
3444 if (msrpm_offsets[i] == 0xffffffff)
3447 p = msrpm_offsets[i];
3448 offset = svm->nested.vmcb_msrpm + (p * 4);
3450 if (kvm_vcpu_read_guest(&svm->vcpu, offset, &value, 4))
3453 svm->nested.msrpm[p] = svm->msrpm[p] | value;
3456 svm->vmcb->control.msrpm_base_pa = __sme_set(__pa(svm->nested.msrpm));
3461 static bool nested_vmcb_checks(struct vmcb *vmcb)
3463 if ((vmcb->control.intercept & (1ULL << INTERCEPT_VMRUN)) == 0)
3466 if (vmcb->control.asid == 0)
3469 if ((vmcb->control.nested_ctl & SVM_NESTED_CTL_NP_ENABLE) &&
3476 static void enter_svm_guest_mode(struct vcpu_svm *svm, u64 vmcb_gpa,
3477 struct vmcb *nested_vmcb, struct page *page)
3479 if (kvm_get_rflags(&svm->vcpu) & X86_EFLAGS_IF)
3480 svm->vcpu.arch.hflags |= HF_HIF_MASK;
3482 svm->vcpu.arch.hflags &= ~HF_HIF_MASK;
3484 if (nested_vmcb->control.nested_ctl & SVM_NESTED_CTL_NP_ENABLE) {
3485 svm->nested.nested_cr3 = nested_vmcb->control.nested_cr3;
3486 nested_svm_init_mmu_context(&svm->vcpu);
3489 /* Load the nested guest state */
3490 svm->vmcb->save.es = nested_vmcb->save.es;
3491 svm->vmcb->save.cs = nested_vmcb->save.cs;
3492 svm->vmcb->save.ss = nested_vmcb->save.ss;
3493 svm->vmcb->save.ds = nested_vmcb->save.ds;
3494 svm->vmcb->save.gdtr = nested_vmcb->save.gdtr;
3495 svm->vmcb->save.idtr = nested_vmcb->save.idtr;
3496 kvm_set_rflags(&svm->vcpu, nested_vmcb->save.rflags);
3497 svm_set_efer(&svm->vcpu, nested_vmcb->save.efer);
3498 svm_set_cr0(&svm->vcpu, nested_vmcb->save.cr0);
3499 svm_set_cr4(&svm->vcpu, nested_vmcb->save.cr4);
3501 svm->vmcb->save.cr3 = nested_vmcb->save.cr3;
3502 svm->vcpu.arch.cr3 = nested_vmcb->save.cr3;
3504 (void)kvm_set_cr3(&svm->vcpu, nested_vmcb->save.cr3);
3506 /* Guest paging mode is active - reset mmu */
3507 kvm_mmu_reset_context(&svm->vcpu);
3509 svm->vmcb->save.cr2 = svm->vcpu.arch.cr2 = nested_vmcb->save.cr2;
3510 kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, nested_vmcb->save.rax);
3511 kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, nested_vmcb->save.rsp);
3512 kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, nested_vmcb->save.rip);
3514 /* In case we don't even reach vcpu_run, the fields are not updated */
3515 svm->vmcb->save.rax = nested_vmcb->save.rax;
3516 svm->vmcb->save.rsp = nested_vmcb->save.rsp;
3517 svm->vmcb->save.rip = nested_vmcb->save.rip;
3518 svm->vmcb->save.dr7 = nested_vmcb->save.dr7;
3519 svm->vmcb->save.dr6 = nested_vmcb->save.dr6;
3520 svm->vmcb->save.cpl = nested_vmcb->save.cpl;
3522 svm->nested.vmcb_msrpm = nested_vmcb->control.msrpm_base_pa & ~0x0fffULL;
3523 svm->nested.vmcb_iopm = nested_vmcb->control.iopm_base_pa & ~0x0fffULL;
3525 /* cache intercepts */
3526 svm->nested.intercept_cr = nested_vmcb->control.intercept_cr;
3527 svm->nested.intercept_dr = nested_vmcb->control.intercept_dr;
3528 svm->nested.intercept_exceptions = nested_vmcb->control.intercept_exceptions;
3529 svm->nested.intercept = nested_vmcb->control.intercept;
3531 svm_flush_tlb(&svm->vcpu, true);
3532 svm->vmcb->control.int_ctl = nested_vmcb->control.int_ctl | V_INTR_MASKING_MASK;
3533 if (nested_vmcb->control.int_ctl & V_INTR_MASKING_MASK)
3534 svm->vcpu.arch.hflags |= HF_VINTR_MASK;
3536 svm->vcpu.arch.hflags &= ~HF_VINTR_MASK;
3538 if (svm->vcpu.arch.hflags & HF_VINTR_MASK) {
3539 /* We only want the cr8 intercept bits of the guest */
3540 clr_cr_intercept(svm, INTERCEPT_CR8_READ);
3541 clr_cr_intercept(svm, INTERCEPT_CR8_WRITE);
3544 /* We don't want to see VMMCALLs from a nested guest */
3545 clr_intercept(svm, INTERCEPT_VMMCALL);
3547 svm->vcpu.arch.tsc_offset += nested_vmcb->control.tsc_offset;
3548 svm->vmcb->control.tsc_offset = svm->vcpu.arch.tsc_offset;
3550 svm->vmcb->control.virt_ext = nested_vmcb->control.virt_ext;
3551 svm->vmcb->control.int_vector = nested_vmcb->control.int_vector;
3552 svm->vmcb->control.int_state = nested_vmcb->control.int_state;
3553 svm->vmcb->control.event_inj = nested_vmcb->control.event_inj;
3554 svm->vmcb->control.event_inj_err = nested_vmcb->control.event_inj_err;
3556 svm->vmcb->control.pause_filter_count =
3557 nested_vmcb->control.pause_filter_count;
3558 svm->vmcb->control.pause_filter_thresh =
3559 nested_vmcb->control.pause_filter_thresh;
3561 nested_svm_unmap(page);
3563 /* Enter Guest-Mode */
3564 enter_guest_mode(&svm->vcpu);
3567 * Merge guest and host intercepts - must be called with vcpu in
3568 * guest-mode to take affect here
3570 recalc_intercepts(svm);
3572 svm->nested.vmcb = vmcb_gpa;
3576 mark_all_dirty(svm->vmcb);
3579 static bool nested_svm_vmrun(struct vcpu_svm *svm)
3581 struct vmcb *nested_vmcb;
3582 struct vmcb *hsave = svm->nested.hsave;
3583 struct vmcb *vmcb = svm->vmcb;
3587 vmcb_gpa = svm->vmcb->save.rax;
3589 nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
3593 if (!nested_vmcb_checks(nested_vmcb)) {
3594 nested_vmcb->control.exit_code = SVM_EXIT_ERR;
3595 nested_vmcb->control.exit_code_hi = 0;
3596 nested_vmcb->control.exit_info_1 = 0;
3597 nested_vmcb->control.exit_info_2 = 0;
3599 nested_svm_unmap(page);
3604 trace_kvm_nested_vmrun(svm->vmcb->save.rip, vmcb_gpa,
3605 nested_vmcb->save.rip,
3606 nested_vmcb->control.int_ctl,
3607 nested_vmcb->control.event_inj,
3608 nested_vmcb->control.nested_ctl);
3610 trace_kvm_nested_intercepts(nested_vmcb->control.intercept_cr & 0xffff,
3611 nested_vmcb->control.intercept_cr >> 16,
3612 nested_vmcb->control.intercept_exceptions,
3613 nested_vmcb->control.intercept);
3615 /* Clear internal status */
3616 kvm_clear_exception_queue(&svm->vcpu);
3617 kvm_clear_interrupt_queue(&svm->vcpu);
3620 * Save the old vmcb, so we don't need to pick what we save, but can
3621 * restore everything when a VMEXIT occurs
3623 hsave->save.es = vmcb->save.es;
3624 hsave->save.cs = vmcb->save.cs;
3625 hsave->save.ss = vmcb->save.ss;
3626 hsave->save.ds = vmcb->save.ds;
3627 hsave->save.gdtr = vmcb->save.gdtr;
3628 hsave->save.idtr = vmcb->save.idtr;
3629 hsave->save.efer = svm->vcpu.arch.efer;
3630 hsave->save.cr0 = kvm_read_cr0(&svm->vcpu);
3631 hsave->save.cr4 = svm->vcpu.arch.cr4;
3632 hsave->save.rflags = kvm_get_rflags(&svm->vcpu);
3633 hsave->save.rip = kvm_rip_read(&svm->vcpu);
3634 hsave->save.rsp = vmcb->save.rsp;
3635 hsave->save.rax = vmcb->save.rax;
3637 hsave->save.cr3 = vmcb->save.cr3;
3639 hsave->save.cr3 = kvm_read_cr3(&svm->vcpu);
3641 copy_vmcb_control_area(hsave, vmcb);
3643 enter_svm_guest_mode(svm, vmcb_gpa, nested_vmcb, page);
3648 static void nested_svm_vmloadsave(struct vmcb *from_vmcb, struct vmcb *to_vmcb)
3650 to_vmcb->save.fs = from_vmcb->save.fs;
3651 to_vmcb->save.gs = from_vmcb->save.gs;
3652 to_vmcb->save.tr = from_vmcb->save.tr;
3653 to_vmcb->save.ldtr = from_vmcb->save.ldtr;
3654 to_vmcb->save.kernel_gs_base = from_vmcb->save.kernel_gs_base;
3655 to_vmcb->save.star = from_vmcb->save.star;
3656 to_vmcb->save.lstar = from_vmcb->save.lstar;
3657 to_vmcb->save.cstar = from_vmcb->save.cstar;
3658 to_vmcb->save.sfmask = from_vmcb->save.sfmask;
3659 to_vmcb->save.sysenter_cs = from_vmcb->save.sysenter_cs;
3660 to_vmcb->save.sysenter_esp = from_vmcb->save.sysenter_esp;
3661 to_vmcb->save.sysenter_eip = from_vmcb->save.sysenter_eip;
3664 static int vmload_interception(struct vcpu_svm *svm)
3666 struct vmcb *nested_vmcb;
3670 if (nested_svm_check_permissions(svm))
3673 nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
3677 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
3678 ret = kvm_skip_emulated_instruction(&svm->vcpu);
3680 nested_svm_vmloadsave(nested_vmcb, svm->vmcb);
3681 nested_svm_unmap(page);
3686 static int vmsave_interception(struct vcpu_svm *svm)
3688 struct vmcb *nested_vmcb;
3692 if (nested_svm_check_permissions(svm))
3695 nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
3699 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
3700 ret = kvm_skip_emulated_instruction(&svm->vcpu);
3702 nested_svm_vmloadsave(svm->vmcb, nested_vmcb);
3703 nested_svm_unmap(page);
3708 static int vmrun_interception(struct vcpu_svm *svm)
3710 if (nested_svm_check_permissions(svm))
3713 /* Save rip after vmrun instruction */
3714 kvm_rip_write(&svm->vcpu, kvm_rip_read(&svm->vcpu) + 3);
3716 if (!nested_svm_vmrun(svm))
3719 if (!nested_svm_vmrun_msrpm(svm))
3726 svm->vmcb->control.exit_code = SVM_EXIT_ERR;
3727 svm->vmcb->control.exit_code_hi = 0;
3728 svm->vmcb->control.exit_info_1 = 0;
3729 svm->vmcb->control.exit_info_2 = 0;
3731 nested_svm_vmexit(svm);
3736 static int stgi_interception(struct vcpu_svm *svm)
3740 if (nested_svm_check_permissions(svm))
3744 * If VGIF is enabled, the STGI intercept is only added to
3745 * detect the opening of the SMI/NMI window; remove it now.
3747 if (vgif_enabled(svm))
3748 clr_intercept(svm, INTERCEPT_STGI);
3750 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
3751 ret = kvm_skip_emulated_instruction(&svm->vcpu);
3752 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
3759 static int clgi_interception(struct vcpu_svm *svm)
3763 if (nested_svm_check_permissions(svm))
3766 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
3767 ret = kvm_skip_emulated_instruction(&svm->vcpu);
3771 /* After a CLGI no interrupts should come */
3772 if (!kvm_vcpu_apicv_active(&svm->vcpu)) {
3773 svm_clear_vintr(svm);
3774 svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
3775 mark_dirty(svm->vmcb, VMCB_INTR);
3781 static int invlpga_interception(struct vcpu_svm *svm)
3783 struct kvm_vcpu *vcpu = &svm->vcpu;
3785 trace_kvm_invlpga(svm->vmcb->save.rip, kvm_register_read(&svm->vcpu, VCPU_REGS_RCX),
3786 kvm_register_read(&svm->vcpu, VCPU_REGS_RAX));
3788 /* Let's treat INVLPGA the same as INVLPG (can be optimized!) */
3789 kvm_mmu_invlpg(vcpu, kvm_register_read(&svm->vcpu, VCPU_REGS_RAX));
3791 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
3792 return kvm_skip_emulated_instruction(&svm->vcpu);
3795 static int skinit_interception(struct vcpu_svm *svm)
3797 trace_kvm_skinit(svm->vmcb->save.rip, kvm_register_read(&svm->vcpu, VCPU_REGS_RAX));
3799 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
3803 static int wbinvd_interception(struct vcpu_svm *svm)
3805 return kvm_emulate_wbinvd(&svm->vcpu);
3808 static int xsetbv_interception(struct vcpu_svm *svm)
3810 u64 new_bv = kvm_read_edx_eax(&svm->vcpu);
3811 u32 index = kvm_register_read(&svm->vcpu, VCPU_REGS_RCX);
3813 if (kvm_set_xcr(&svm->vcpu, index, new_bv) == 0) {
3814 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
3815 return kvm_skip_emulated_instruction(&svm->vcpu);
3821 static int task_switch_interception(struct vcpu_svm *svm)
3825 int int_type = svm->vmcb->control.exit_int_info &
3826 SVM_EXITINTINFO_TYPE_MASK;
3827 int int_vec = svm->vmcb->control.exit_int_info & SVM_EVTINJ_VEC_MASK;
3829 svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_TYPE_MASK;
3831 svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_VALID;
3832 bool has_error_code = false;
3835 tss_selector = (u16)svm->vmcb->control.exit_info_1;
3837 if (svm->vmcb->control.exit_info_2 &
3838 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_IRET))
3839 reason = TASK_SWITCH_IRET;
3840 else if (svm->vmcb->control.exit_info_2 &
3841 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_JMP))
3842 reason = TASK_SWITCH_JMP;
3844 reason = TASK_SWITCH_GATE;
3846 reason = TASK_SWITCH_CALL;
3848 if (reason == TASK_SWITCH_GATE) {
3850 case SVM_EXITINTINFO_TYPE_NMI:
3851 svm->vcpu.arch.nmi_injected = false;
3853 case SVM_EXITINTINFO_TYPE_EXEPT:
3854 if (svm->vmcb->control.exit_info_2 &
3855 (1ULL << SVM_EXITINFOSHIFT_TS_HAS_ERROR_CODE)) {
3856 has_error_code = true;
3858 (u32)svm->vmcb->control.exit_info_2;
3860 kvm_clear_exception_queue(&svm->vcpu);
3862 case SVM_EXITINTINFO_TYPE_INTR:
3863 kvm_clear_interrupt_queue(&svm->vcpu);
3870 if (reason != TASK_SWITCH_GATE ||
3871 int_type == SVM_EXITINTINFO_TYPE_SOFT ||
3872 (int_type == SVM_EXITINTINFO_TYPE_EXEPT &&
3873 (int_vec == OF_VECTOR || int_vec == BP_VECTOR)))
3874 skip_emulated_instruction(&svm->vcpu);
3876 if (int_type != SVM_EXITINTINFO_TYPE_SOFT)
3879 if (kvm_task_switch(&svm->vcpu, tss_selector, int_vec, reason,
3880 has_error_code, error_code) == EMULATE_FAIL) {
3881 svm->vcpu.run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3882 svm->vcpu.run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
3883 svm->vcpu.run->internal.ndata = 0;
3889 static int cpuid_interception(struct vcpu_svm *svm)
3891 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
3892 return kvm_emulate_cpuid(&svm->vcpu);
3895 static int iret_interception(struct vcpu_svm *svm)
3897 ++svm->vcpu.stat.nmi_window_exits;
3898 clr_intercept(svm, INTERCEPT_IRET);
3899 svm->vcpu.arch.hflags |= HF_IRET_MASK;
3900 svm->nmi_iret_rip = kvm_rip_read(&svm->vcpu);
3901 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
3905 static int invlpg_interception(struct vcpu_svm *svm)
3907 if (!static_cpu_has(X86_FEATURE_DECODEASSISTS))
3908 return kvm_emulate_instruction(&svm->vcpu, 0) == EMULATE_DONE;
3910 kvm_mmu_invlpg(&svm->vcpu, svm->vmcb->control.exit_info_1);
3911 return kvm_skip_emulated_instruction(&svm->vcpu);
3914 static int emulate_on_interception(struct vcpu_svm *svm)
3916 return kvm_emulate_instruction(&svm->vcpu, 0) == EMULATE_DONE;
3919 static int rsm_interception(struct vcpu_svm *svm)
3921 return kvm_emulate_instruction_from_buffer(&svm->vcpu,
3922 rsm_ins_bytes, 2) == EMULATE_DONE;
3925 static int rdpmc_interception(struct vcpu_svm *svm)
3929 if (!static_cpu_has(X86_FEATURE_NRIPS))
3930 return emulate_on_interception(svm);
3932 err = kvm_rdpmc(&svm->vcpu);
3933 return kvm_complete_insn_gp(&svm->vcpu, err);
3936 static bool check_selective_cr0_intercepted(struct vcpu_svm *svm,
3939 unsigned long cr0 = svm->vcpu.arch.cr0;
3943 intercept = svm->nested.intercept;
3945 if (!is_guest_mode(&svm->vcpu) ||
3946 (!(intercept & (1ULL << INTERCEPT_SELECTIVE_CR0))))
3949 cr0 &= ~SVM_CR0_SELECTIVE_MASK;
3950 val &= ~SVM_CR0_SELECTIVE_MASK;
3953 svm->vmcb->control.exit_code = SVM_EXIT_CR0_SEL_WRITE;
3954 ret = (nested_svm_exit_handled(svm) == NESTED_EXIT_DONE);
3960 #define CR_VALID (1ULL << 63)
3962 static int cr_interception(struct vcpu_svm *svm)
3968 if (!static_cpu_has(X86_FEATURE_DECODEASSISTS))
3969 return emulate_on_interception(svm);
3971 if (unlikely((svm->vmcb->control.exit_info_1 & CR_VALID) == 0))
3972 return emulate_on_interception(svm);
3974 reg = svm->vmcb->control.exit_info_1 & SVM_EXITINFO_REG_MASK;
3975 if (svm->vmcb->control.exit_code == SVM_EXIT_CR0_SEL_WRITE)
3976 cr = SVM_EXIT_WRITE_CR0 - SVM_EXIT_READ_CR0;
3978 cr = svm->vmcb->control.exit_code - SVM_EXIT_READ_CR0;
3981 if (cr >= 16) { /* mov to cr */
3983 val = kvm_register_read(&svm->vcpu, reg);
3986 if (!check_selective_cr0_intercepted(svm, val))
3987 err = kvm_set_cr0(&svm->vcpu, val);
3993 err = kvm_set_cr3(&svm->vcpu, val);
3996 err = kvm_set_cr4(&svm->vcpu, val);
3999 err = kvm_set_cr8(&svm->vcpu, val);
4002 WARN(1, "unhandled write to CR%d", cr);
4003 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
4006 } else { /* mov from cr */
4009 val = kvm_read_cr0(&svm->vcpu);
4012 val = svm->vcpu.arch.cr2;
4015 val = kvm_read_cr3(&svm->vcpu);
4018 val = kvm_read_cr4(&svm->vcpu);
4021 val = kvm_get_cr8(&svm->vcpu);
4024 WARN(1, "unhandled read from CR%d", cr);
4025 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
4028 kvm_register_write(&svm->vcpu, reg, val);
4030 return kvm_complete_insn_gp(&svm->vcpu, err);
4033 static int dr_interception(struct vcpu_svm *svm)
4038 if (svm->vcpu.guest_debug == 0) {
4040 * No more DR vmexits; force a reload of the debug registers
4041 * and reenter on this instruction. The next vmexit will
4042 * retrieve the full state of the debug registers.
4044 clr_dr_intercepts(svm);
4045 svm->vcpu.arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
4049 if (!boot_cpu_has(X86_FEATURE_DECODEASSISTS))
4050 return emulate_on_interception(svm);
4052 reg = svm->vmcb->control.exit_info_1 & SVM_EXITINFO_REG_MASK;
4053 dr = svm->vmcb->control.exit_code - SVM_EXIT_READ_DR0;
4055 if (dr >= 16) { /* mov to DRn */
4056 if (!kvm_require_dr(&svm->vcpu, dr - 16))
4058 val = kvm_register_read(&svm->vcpu, reg);
4059 kvm_set_dr(&svm->vcpu, dr - 16, val);
4061 if (!kvm_require_dr(&svm->vcpu, dr))
4063 kvm_get_dr(&svm->vcpu, dr, &val);
4064 kvm_register_write(&svm->vcpu, reg, val);
4067 return kvm_skip_emulated_instruction(&svm->vcpu);
4070 static int cr8_write_interception(struct vcpu_svm *svm)
4072 struct kvm_run *kvm_run = svm->vcpu.run;
4075 u8 cr8_prev = kvm_get_cr8(&svm->vcpu);
4076 /* instruction emulation calls kvm_set_cr8() */
4077 r = cr_interception(svm);
4078 if (lapic_in_kernel(&svm->vcpu))
4080 if (cr8_prev <= kvm_get_cr8(&svm->vcpu))
4082 kvm_run->exit_reason = KVM_EXIT_SET_TPR;
4086 static int svm_get_msr_feature(struct kvm_msr_entry *msr)
4090 switch (msr->index) {
4091 case MSR_F10H_DECFG:
4092 if (boot_cpu_has(X86_FEATURE_LFENCE_RDTSC))
4093 msr->data |= MSR_F10H_DECFG_LFENCE_SERIALIZE;
4102 static int svm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
4104 struct vcpu_svm *svm = to_svm(vcpu);
4106 switch (msr_info->index) {
4108 msr_info->data = svm->vmcb->save.star;
4110 #ifdef CONFIG_X86_64
4112 msr_info->data = svm->vmcb->save.lstar;
4115 msr_info->data = svm->vmcb->save.cstar;
4117 case MSR_KERNEL_GS_BASE:
4118 msr_info->data = svm->vmcb->save.kernel_gs_base;
4120 case MSR_SYSCALL_MASK:
4121 msr_info->data = svm->vmcb->save.sfmask;
4124 case MSR_IA32_SYSENTER_CS:
4125 msr_info->data = svm->vmcb->save.sysenter_cs;
4127 case MSR_IA32_SYSENTER_EIP:
4128 msr_info->data = svm->sysenter_eip;
4130 case MSR_IA32_SYSENTER_ESP:
4131 msr_info->data = svm->sysenter_esp;
4134 if (!boot_cpu_has(X86_FEATURE_RDTSCP))
4136 msr_info->data = svm->tsc_aux;
4139 * Nobody will change the following 5 values in the VMCB so we can
4140 * safely return them on rdmsr. They will always be 0 until LBRV is
4143 case MSR_IA32_DEBUGCTLMSR:
4144 msr_info->data = svm->vmcb->save.dbgctl;
4146 case MSR_IA32_LASTBRANCHFROMIP:
4147 msr_info->data = svm->vmcb->save.br_from;
4149 case MSR_IA32_LASTBRANCHTOIP:
4150 msr_info->data = svm->vmcb->save.br_to;
4152 case MSR_IA32_LASTINTFROMIP:
4153 msr_info->data = svm->vmcb->save.last_excp_from;
4155 case MSR_IA32_LASTINTTOIP:
4156 msr_info->data = svm->vmcb->save.last_excp_to;
4158 case MSR_VM_HSAVE_PA:
4159 msr_info->data = svm->nested.hsave_msr;
4162 msr_info->data = svm->nested.vm_cr_msr;
4164 case MSR_IA32_SPEC_CTRL:
4165 if (!msr_info->host_initiated &&
4166 !guest_cpuid_has(vcpu, X86_FEATURE_AMD_IBRS) &&
4167 !guest_cpuid_has(vcpu, X86_FEATURE_AMD_SSBD))
4170 msr_info->data = svm->spec_ctrl;
4172 case MSR_AMD64_VIRT_SPEC_CTRL:
4173 if (!msr_info->host_initiated &&
4174 !guest_cpuid_has(vcpu, X86_FEATURE_VIRT_SSBD))
4177 msr_info->data = svm->virt_spec_ctrl;
4179 case MSR_F15H_IC_CFG: {
4183 family = guest_cpuid_family(vcpu);
4184 model = guest_cpuid_model(vcpu);
4186 if (family < 0 || model < 0)
4187 return kvm_get_msr_common(vcpu, msr_info);
4191 if (family == 0x15 &&
4192 (model >= 0x2 && model < 0x20))
4193 msr_info->data = 0x1E;
4196 case MSR_F10H_DECFG:
4197 msr_info->data = svm->msr_decfg;
4200 return kvm_get_msr_common(vcpu, msr_info);
4205 static int rdmsr_interception(struct vcpu_svm *svm)
4207 u32 ecx = kvm_register_read(&svm->vcpu, VCPU_REGS_RCX);
4208 struct msr_data msr_info;
4210 msr_info.index = ecx;
4211 msr_info.host_initiated = false;
4212 if (svm_get_msr(&svm->vcpu, &msr_info)) {
4213 trace_kvm_msr_read_ex(ecx);
4214 kvm_inject_gp(&svm->vcpu, 0);
4217 trace_kvm_msr_read(ecx, msr_info.data);
4219 kvm_register_write(&svm->vcpu, VCPU_REGS_RAX,
4220 msr_info.data & 0xffffffff);
4221 kvm_register_write(&svm->vcpu, VCPU_REGS_RDX,
4222 msr_info.data >> 32);
4223 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
4224 return kvm_skip_emulated_instruction(&svm->vcpu);
4228 static int svm_set_vm_cr(struct kvm_vcpu *vcpu, u64 data)
4230 struct vcpu_svm *svm = to_svm(vcpu);
4231 int svm_dis, chg_mask;
4233 if (data & ~SVM_VM_CR_VALID_MASK)
4236 chg_mask = SVM_VM_CR_VALID_MASK;
4238 if (svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK)
4239 chg_mask &= ~(SVM_VM_CR_SVM_LOCK_MASK | SVM_VM_CR_SVM_DIS_MASK);
4241 svm->nested.vm_cr_msr &= ~chg_mask;
4242 svm->nested.vm_cr_msr |= (data & chg_mask);
4244 svm_dis = svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK;
4246 /* check for svm_disable while efer.svme is set */
4247 if (svm_dis && (vcpu->arch.efer & EFER_SVME))
4253 static int svm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
4255 struct vcpu_svm *svm = to_svm(vcpu);
4257 u32 ecx = msr->index;
4258 u64 data = msr->data;
4260 case MSR_IA32_CR_PAT:
4261 if (!kvm_mtrr_valid(vcpu, MSR_IA32_CR_PAT, data))
4263 vcpu->arch.pat = data;
4264 svm->vmcb->save.g_pat = data;
4265 mark_dirty(svm->vmcb, VMCB_NPT);
4267 case MSR_IA32_SPEC_CTRL:
4268 if (!msr->host_initiated &&
4269 !guest_cpuid_has(vcpu, X86_FEATURE_AMD_IBRS) &&
4270 !guest_cpuid_has(vcpu, X86_FEATURE_AMD_SSBD))
4273 /* The STIBP bit doesn't fault even if it's not advertised */
4274 if (data & ~(SPEC_CTRL_IBRS | SPEC_CTRL_STIBP | SPEC_CTRL_SSBD))
4277 svm->spec_ctrl = data;
4284 * When it's written (to non-zero) for the first time, pass
4288 * The handling of the MSR bitmap for L2 guests is done in
4289 * nested_svm_vmrun_msrpm.
4290 * We update the L1 MSR bit as well since it will end up
4291 * touching the MSR anyway now.
4293 set_msr_interception(svm->msrpm, MSR_IA32_SPEC_CTRL, 1, 1);
4295 case MSR_IA32_PRED_CMD:
4296 if (!msr->host_initiated &&
4297 !guest_cpuid_has(vcpu, X86_FEATURE_AMD_IBPB))
4300 if (data & ~PRED_CMD_IBPB)
4306 wrmsrl(MSR_IA32_PRED_CMD, PRED_CMD_IBPB);
4307 if (is_guest_mode(vcpu))
4309 set_msr_interception(svm->msrpm, MSR_IA32_PRED_CMD, 0, 1);
4311 case MSR_AMD64_VIRT_SPEC_CTRL:
4312 if (!msr->host_initiated &&
4313 !guest_cpuid_has(vcpu, X86_FEATURE_VIRT_SSBD))
4316 if (data & ~SPEC_CTRL_SSBD)
4319 svm->virt_spec_ctrl = data;
4322 svm->vmcb->save.star = data;
4324 #ifdef CONFIG_X86_64
4326 svm->vmcb->save.lstar = data;
4329 svm->vmcb->save.cstar = data;
4331 case MSR_KERNEL_GS_BASE:
4332 svm->vmcb->save.kernel_gs_base = data;
4334 case MSR_SYSCALL_MASK:
4335 svm->vmcb->save.sfmask = data;
4338 case MSR_IA32_SYSENTER_CS:
4339 svm->vmcb->save.sysenter_cs = data;
4341 case MSR_IA32_SYSENTER_EIP:
4342 svm->sysenter_eip = data;
4343 svm->vmcb->save.sysenter_eip = data;
4345 case MSR_IA32_SYSENTER_ESP:
4346 svm->sysenter_esp = data;
4347 svm->vmcb->save.sysenter_esp = data;
4350 if (!boot_cpu_has(X86_FEATURE_RDTSCP))
4354 * This is rare, so we update the MSR here instead of using
4355 * direct_access_msrs. Doing that would require a rdmsr in
4358 svm->tsc_aux = data;
4359 wrmsrl(MSR_TSC_AUX, svm->tsc_aux);
4361 case MSR_IA32_DEBUGCTLMSR:
4362 if (!boot_cpu_has(X86_FEATURE_LBRV)) {
4363 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTL 0x%llx, nop\n",
4367 if (data & DEBUGCTL_RESERVED_BITS)
4370 svm->vmcb->save.dbgctl = data;
4371 mark_dirty(svm->vmcb, VMCB_LBR);
4372 if (data & (1ULL<<0))
4373 svm_enable_lbrv(svm);
4375 svm_disable_lbrv(svm);
4377 case MSR_VM_HSAVE_PA:
4378 svm->nested.hsave_msr = data;
4381 return svm_set_vm_cr(vcpu, data);
4383 vcpu_unimpl(vcpu, "unimplemented wrmsr: 0x%x data 0x%llx\n", ecx, data);
4385 case MSR_F10H_DECFG: {
4386 struct kvm_msr_entry msr_entry;
4388 msr_entry.index = msr->index;
4389 if (svm_get_msr_feature(&msr_entry))
4392 /* Check the supported bits */
4393 if (data & ~msr_entry.data)
4396 /* Don't allow the guest to change a bit, #GP */
4397 if (!msr->host_initiated && (data ^ msr_entry.data))
4400 svm->msr_decfg = data;
4403 case MSR_IA32_APICBASE:
4404 if (kvm_vcpu_apicv_active(vcpu))
4405 avic_update_vapic_bar(to_svm(vcpu), data);
4408 return kvm_set_msr_common(vcpu, msr);
4413 static int wrmsr_interception(struct vcpu_svm *svm)
4415 struct msr_data msr;
4416 u32 ecx = kvm_register_read(&svm->vcpu, VCPU_REGS_RCX);
4417 u64 data = kvm_read_edx_eax(&svm->vcpu);
4421 msr.host_initiated = false;
4423 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
4424 if (kvm_set_msr(&svm->vcpu, &msr)) {
4425 trace_kvm_msr_write_ex(ecx, data);
4426 kvm_inject_gp(&svm->vcpu, 0);
4429 trace_kvm_msr_write(ecx, data);
4430 return kvm_skip_emulated_instruction(&svm->vcpu);
4434 static int msr_interception(struct vcpu_svm *svm)
4436 if (svm->vmcb->control.exit_info_1)
4437 return wrmsr_interception(svm);
4439 return rdmsr_interception(svm);
4442 static int interrupt_window_interception(struct vcpu_svm *svm)
4444 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
4445 svm_clear_vintr(svm);
4446 svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
4447 mark_dirty(svm->vmcb, VMCB_INTR);
4448 ++svm->vcpu.stat.irq_window_exits;
4452 static int pause_interception(struct vcpu_svm *svm)
4454 struct kvm_vcpu *vcpu = &svm->vcpu;
4455 bool in_kernel = (svm_get_cpl(vcpu) == 0);
4457 if (pause_filter_thresh)
4458 grow_ple_window(vcpu);
4460 kvm_vcpu_on_spin(vcpu, in_kernel);
4464 static int nop_interception(struct vcpu_svm *svm)
4466 return kvm_skip_emulated_instruction(&(svm->vcpu));
4469 static int monitor_interception(struct vcpu_svm *svm)
4471 printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
4472 return nop_interception(svm);
4475 static int mwait_interception(struct vcpu_svm *svm)
4477 printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
4478 return nop_interception(svm);
4481 enum avic_ipi_failure_cause {
4482 AVIC_IPI_FAILURE_INVALID_INT_TYPE,
4483 AVIC_IPI_FAILURE_TARGET_NOT_RUNNING,
4484 AVIC_IPI_FAILURE_INVALID_TARGET,
4485 AVIC_IPI_FAILURE_INVALID_BACKING_PAGE,
4488 static int avic_incomplete_ipi_interception(struct vcpu_svm *svm)
4490 u32 icrh = svm->vmcb->control.exit_info_1 >> 32;
4491 u32 icrl = svm->vmcb->control.exit_info_1;
4492 u32 id = svm->vmcb->control.exit_info_2 >> 32;
4493 u32 index = svm->vmcb->control.exit_info_2 & 0xFF;
4494 struct kvm_lapic *apic = svm->vcpu.arch.apic;
4496 trace_kvm_avic_incomplete_ipi(svm->vcpu.vcpu_id, icrh, icrl, id, index);
4499 case AVIC_IPI_FAILURE_INVALID_INT_TYPE:
4501 * AVIC hardware handles the generation of
4502 * IPIs when the specified Message Type is Fixed
4503 * (also known as fixed delivery mode) and
4504 * the Trigger Mode is edge-triggered. The hardware
4505 * also supports self and broadcast delivery modes
4506 * specified via the Destination Shorthand(DSH)
4507 * field of the ICRL. Logical and physical APIC ID
4508 * formats are supported. All other IPI types cause
4509 * a #VMEXIT, which needs to emulated.
4511 kvm_lapic_reg_write(apic, APIC_ICR2, icrh);
4512 kvm_lapic_reg_write(apic, APIC_ICR, icrl);
4514 case AVIC_IPI_FAILURE_TARGET_NOT_RUNNING: {
4515 struct kvm_lapic *apic = svm->vcpu.arch.apic;
4518 * Update ICR high and low, then emulate sending IPI,
4519 * which is handled when writing APIC_ICR.
4521 kvm_lapic_reg_write(apic, APIC_ICR2, icrh);
4522 kvm_lapic_reg_write(apic, APIC_ICR, icrl);
4525 case AVIC_IPI_FAILURE_INVALID_TARGET:
4526 WARN_ONCE(1, "Invalid IPI target: index=%u, vcpu=%d, icr=%#0x:%#0x\n",
4527 index, svm->vcpu.vcpu_id, icrh, icrl);
4529 case AVIC_IPI_FAILURE_INVALID_BACKING_PAGE:
4530 WARN_ONCE(1, "Invalid backing page\n");
4533 pr_err("Unknown IPI interception\n");
4539 static u32 *avic_get_logical_id_entry(struct kvm_vcpu *vcpu, u32 ldr, bool flat)
4541 struct kvm_svm *kvm_svm = to_kvm_svm(vcpu->kvm);
4543 u32 *logical_apic_id_table;
4544 int dlid = GET_APIC_LOGICAL_ID(ldr);
4549 if (flat) { /* flat */
4550 index = ffs(dlid) - 1;
4553 } else { /* cluster */
4554 int cluster = (dlid & 0xf0) >> 4;
4555 int apic = ffs(dlid & 0x0f) - 1;
4557 if ((apic < 0) || (apic > 7) ||
4560 index = (cluster << 2) + apic;
4563 logical_apic_id_table = (u32 *) page_address(kvm_svm->avic_logical_id_table_page);
4565 return &logical_apic_id_table[index];
4568 static int avic_ldr_write(struct kvm_vcpu *vcpu, u8 g_physical_id, u32 ldr,
4572 u32 *entry, new_entry;
4574 flat = kvm_lapic_get_reg(vcpu->arch.apic, APIC_DFR) == APIC_DFR_FLAT;
4575 entry = avic_get_logical_id_entry(vcpu, ldr, flat);
4579 new_entry = READ_ONCE(*entry);
4580 new_entry &= ~AVIC_LOGICAL_ID_ENTRY_GUEST_PHYSICAL_ID_MASK;
4581 new_entry |= (g_physical_id & AVIC_LOGICAL_ID_ENTRY_GUEST_PHYSICAL_ID_MASK);
4583 new_entry |= AVIC_LOGICAL_ID_ENTRY_VALID_MASK;
4585 new_entry &= ~AVIC_LOGICAL_ID_ENTRY_VALID_MASK;
4586 WRITE_ONCE(*entry, new_entry);
4591 static int avic_handle_ldr_update(struct kvm_vcpu *vcpu)
4594 struct vcpu_svm *svm = to_svm(vcpu);
4595 u32 ldr = kvm_lapic_get_reg(vcpu->arch.apic, APIC_LDR);
4600 ret = avic_ldr_write(vcpu, vcpu->vcpu_id, ldr, true);
4601 if (ret && svm->ldr_reg) {
4602 avic_ldr_write(vcpu, 0, svm->ldr_reg, false);
4610 static int avic_handle_apic_id_update(struct kvm_vcpu *vcpu)
4613 struct vcpu_svm *svm = to_svm(vcpu);
4614 u32 apic_id_reg = kvm_lapic_get_reg(vcpu->arch.apic, APIC_ID);
4615 u32 id = (apic_id_reg >> 24) & 0xff;
4617 if (vcpu->vcpu_id == id)
4620 old = avic_get_physical_id_entry(vcpu, vcpu->vcpu_id);
4621 new = avic_get_physical_id_entry(vcpu, id);
4625 /* We need to move physical_id_entry to new offset */
4628 to_svm(vcpu)->avic_physical_id_cache = new;
4631 * Also update the guest physical APIC ID in the logical
4632 * APIC ID table entry if already setup the LDR.
4635 avic_handle_ldr_update(vcpu);
4640 static int avic_handle_dfr_update(struct kvm_vcpu *vcpu)
4642 struct vcpu_svm *svm = to_svm(vcpu);
4643 struct kvm_svm *kvm_svm = to_kvm_svm(vcpu->kvm);
4644 u32 dfr = kvm_lapic_get_reg(vcpu->arch.apic, APIC_DFR);
4645 u32 mod = (dfr >> 28) & 0xf;
4648 * We assume that all local APICs are using the same type.
4649 * If this changes, we need to flush the AVIC logical
4652 if (kvm_svm->ldr_mode == mod)
4655 clear_page(page_address(kvm_svm->avic_logical_id_table_page));
4656 kvm_svm->ldr_mode = mod;
4659 avic_handle_ldr_update(vcpu);
4663 static int avic_unaccel_trap_write(struct vcpu_svm *svm)
4665 struct kvm_lapic *apic = svm->vcpu.arch.apic;
4666 u32 offset = svm->vmcb->control.exit_info_1 &
4667 AVIC_UNACCEL_ACCESS_OFFSET_MASK;
4671 if (avic_handle_apic_id_update(&svm->vcpu))
4675 if (avic_handle_ldr_update(&svm->vcpu))
4679 avic_handle_dfr_update(&svm->vcpu);
4685 kvm_lapic_reg_write(apic, offset, kvm_lapic_get_reg(apic, offset));
4690 static bool is_avic_unaccelerated_access_trap(u32 offset)
4719 static int avic_unaccelerated_access_interception(struct vcpu_svm *svm)
4722 u32 offset = svm->vmcb->control.exit_info_1 &
4723 AVIC_UNACCEL_ACCESS_OFFSET_MASK;
4724 u32 vector = svm->vmcb->control.exit_info_2 &
4725 AVIC_UNACCEL_ACCESS_VECTOR_MASK;
4726 bool write = (svm->vmcb->control.exit_info_1 >> 32) &
4727 AVIC_UNACCEL_ACCESS_WRITE_MASK;
4728 bool trap = is_avic_unaccelerated_access_trap(offset);
4730 trace_kvm_avic_unaccelerated_access(svm->vcpu.vcpu_id, offset,
4731 trap, write, vector);
4734 WARN_ONCE(!write, "svm: Handling trap read.\n");
4735 ret = avic_unaccel_trap_write(svm);
4737 /* Handling Fault */
4738 ret = (kvm_emulate_instruction(&svm->vcpu, 0) == EMULATE_DONE);
4744 static int (*const svm_exit_handlers[])(struct vcpu_svm *svm) = {
4745 [SVM_EXIT_READ_CR0] = cr_interception,
4746 [SVM_EXIT_READ_CR3] = cr_interception,
4747 [SVM_EXIT_READ_CR4] = cr_interception,
4748 [SVM_EXIT_READ_CR8] = cr_interception,
4749 [SVM_EXIT_CR0_SEL_WRITE] = cr_interception,
4750 [SVM_EXIT_WRITE_CR0] = cr_interception,
4751 [SVM_EXIT_WRITE_CR3] = cr_interception,
4752 [SVM_EXIT_WRITE_CR4] = cr_interception,
4753 [SVM_EXIT_WRITE_CR8] = cr8_write_interception,
4754 [SVM_EXIT_READ_DR0] = dr_interception,
4755 [SVM_EXIT_READ_DR1] = dr_interception,
4756 [SVM_EXIT_READ_DR2] = dr_interception,
4757 [SVM_EXIT_READ_DR3] = dr_interception,
4758 [SVM_EXIT_READ_DR4] = dr_interception,
4759 [SVM_EXIT_READ_DR5] = dr_interception,
4760 [SVM_EXIT_READ_DR6] = dr_interception,
4761 [SVM_EXIT_READ_DR7] = dr_interception,
4762 [SVM_EXIT_WRITE_DR0] = dr_interception,
4763 [SVM_EXIT_WRITE_DR1] = dr_interception,
4764 [SVM_EXIT_WRITE_DR2] = dr_interception,
4765 [SVM_EXIT_WRITE_DR3] = dr_interception,
4766 [SVM_EXIT_WRITE_DR4] = dr_interception,
4767 [SVM_EXIT_WRITE_DR5] = dr_interception,
4768 [SVM_EXIT_WRITE_DR6] = dr_interception,
4769 [SVM_EXIT_WRITE_DR7] = dr_interception,
4770 [SVM_EXIT_EXCP_BASE + DB_VECTOR] = db_interception,
4771 [SVM_EXIT_EXCP_BASE + BP_VECTOR] = bp_interception,
4772 [SVM_EXIT_EXCP_BASE + UD_VECTOR] = ud_interception,
4773 [SVM_EXIT_EXCP_BASE + PF_VECTOR] = pf_interception,
4774 [SVM_EXIT_EXCP_BASE + MC_VECTOR] = mc_interception,
4775 [SVM_EXIT_EXCP_BASE + AC_VECTOR] = ac_interception,
4776 [SVM_EXIT_EXCP_BASE + GP_VECTOR] = gp_interception,
4777 [SVM_EXIT_INTR] = intr_interception,
4778 [SVM_EXIT_NMI] = nmi_interception,
4779 [SVM_EXIT_SMI] = nop_on_interception,
4780 [SVM_EXIT_INIT] = nop_on_interception,
4781 [SVM_EXIT_VINTR] = interrupt_window_interception,
4782 [SVM_EXIT_RDPMC] = rdpmc_interception,
4783 [SVM_EXIT_CPUID] = cpuid_interception,
4784 [SVM_EXIT_IRET] = iret_interception,
4785 [SVM_EXIT_INVD] = emulate_on_interception,
4786 [SVM_EXIT_PAUSE] = pause_interception,
4787 [SVM_EXIT_HLT] = halt_interception,
4788 [SVM_EXIT_INVLPG] = invlpg_interception,
4789 [SVM_EXIT_INVLPGA] = invlpga_interception,
4790 [SVM_EXIT_IOIO] = io_interception,
4791 [SVM_EXIT_MSR] = msr_interception,
4792 [SVM_EXIT_TASK_SWITCH] = task_switch_interception,
4793 [SVM_EXIT_SHUTDOWN] = shutdown_interception,
4794 [SVM_EXIT_VMRUN] = vmrun_interception,
4795 [SVM_EXIT_VMMCALL] = vmmcall_interception,
4796 [SVM_EXIT_VMLOAD] = vmload_interception,
4797 [SVM_EXIT_VMSAVE] = vmsave_interception,
4798 [SVM_EXIT_STGI] = stgi_interception,
4799 [SVM_EXIT_CLGI] = clgi_interception,
4800 [SVM_EXIT_SKINIT] = skinit_interception,
4801 [SVM_EXIT_WBINVD] = wbinvd_interception,
4802 [SVM_EXIT_MONITOR] = monitor_interception,
4803 [SVM_EXIT_MWAIT] = mwait_interception,
4804 [SVM_EXIT_XSETBV] = xsetbv_interception,
4805 [SVM_EXIT_NPF] = npf_interception,
4806 [SVM_EXIT_RSM] = rsm_interception,
4807 [SVM_EXIT_AVIC_INCOMPLETE_IPI] = avic_incomplete_ipi_interception,
4808 [SVM_EXIT_AVIC_UNACCELERATED_ACCESS] = avic_unaccelerated_access_interception,
4811 static void dump_vmcb(struct kvm_vcpu *vcpu)
4813 struct vcpu_svm *svm = to_svm(vcpu);
4814 struct vmcb_control_area *control = &svm->vmcb->control;
4815 struct vmcb_save_area *save = &svm->vmcb->save;
4817 pr_err("VMCB Control Area:\n");
4818 pr_err("%-20s%04x\n", "cr_read:", control->intercept_cr & 0xffff);
4819 pr_err("%-20s%04x\n", "cr_write:", control->intercept_cr >> 16);
4820 pr_err("%-20s%04x\n", "dr_read:", control->intercept_dr & 0xffff);
4821 pr_err("%-20s%04x\n", "dr_write:", control->intercept_dr >> 16);
4822 pr_err("%-20s%08x\n", "exceptions:", control->intercept_exceptions);
4823 pr_err("%-20s%016llx\n", "intercepts:", control->intercept);
4824 pr_err("%-20s%d\n", "pause filter count:", control->pause_filter_count);
4825 pr_err("%-20s%d\n", "pause filter threshold:",
4826 control->pause_filter_thresh);
4827 pr_err("%-20s%016llx\n", "iopm_base_pa:", control->iopm_base_pa);
4828 pr_err("%-20s%016llx\n", "msrpm_base_pa:", control->msrpm_base_pa);
4829 pr_err("%-20s%016llx\n", "tsc_offset:", control->tsc_offset);
4830 pr_err("%-20s%d\n", "asid:", control->asid);
4831 pr_err("%-20s%d\n", "tlb_ctl:", control->tlb_ctl);
4832 pr_err("%-20s%08x\n", "int_ctl:", control->int_ctl);
4833 pr_err("%-20s%08x\n", "int_vector:", control->int_vector);
4834 pr_err("%-20s%08x\n", "int_state:", control->int_state);
4835 pr_err("%-20s%08x\n", "exit_code:", control->exit_code);
4836 pr_err("%-20s%016llx\n", "exit_info1:", control->exit_info_1);
4837 pr_err("%-20s%016llx\n", "exit_info2:", control->exit_info_2);
4838 pr_err("%-20s%08x\n", "exit_int_info:", control->exit_int_info);
4839 pr_err("%-20s%08x\n", "exit_int_info_err:", control->exit_int_info_err);
4840 pr_err("%-20s%lld\n", "nested_ctl:", control->nested_ctl);
4841 pr_err("%-20s%016llx\n", "nested_cr3:", control->nested_cr3);
4842 pr_err("%-20s%016llx\n", "avic_vapic_bar:", control->avic_vapic_bar);
4843 pr_err("%-20s%08x\n", "event_inj:", control->event_inj);
4844 pr_err("%-20s%08x\n", "event_inj_err:", control->event_inj_err);
4845 pr_err("%-20s%lld\n", "virt_ext:", control->virt_ext);
4846 pr_err("%-20s%016llx\n", "next_rip:", control->next_rip);
4847 pr_err("%-20s%016llx\n", "avic_backing_page:", control->avic_backing_page);
4848 pr_err("%-20s%016llx\n", "avic_logical_id:", control->avic_logical_id);
4849 pr_err("%-20s%016llx\n", "avic_physical_id:", control->avic_physical_id);
4850 pr_err("VMCB State Save Area:\n");
4851 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4853 save->es.selector, save->es.attrib,
4854 save->es.limit, save->es.base);
4855 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4857 save->cs.selector, save->cs.attrib,
4858 save->cs.limit, save->cs.base);
4859 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4861 save->ss.selector, save->ss.attrib,
4862 save->ss.limit, save->ss.base);
4863 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4865 save->ds.selector, save->ds.attrib,
4866 save->ds.limit, save->ds.base);
4867 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4869 save->fs.selector, save->fs.attrib,
4870 save->fs.limit, save->fs.base);
4871 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4873 save->gs.selector, save->gs.attrib,
4874 save->gs.limit, save->gs.base);
4875 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4877 save->gdtr.selector, save->gdtr.attrib,
4878 save->gdtr.limit, save->gdtr.base);
4879 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4881 save->ldtr.selector, save->ldtr.attrib,
4882 save->ldtr.limit, save->ldtr.base);
4883 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4885 save->idtr.selector, save->idtr.attrib,
4886 save->idtr.limit, save->idtr.base);
4887 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4889 save->tr.selector, save->tr.attrib,
4890 save->tr.limit, save->tr.base);
4891 pr_err("cpl: %d efer: %016llx\n",
4892 save->cpl, save->efer);
4893 pr_err("%-15s %016llx %-13s %016llx\n",
4894 "cr0:", save->cr0, "cr2:", save->cr2);
4895 pr_err("%-15s %016llx %-13s %016llx\n",
4896 "cr3:", save->cr3, "cr4:", save->cr4);
4897 pr_err("%-15s %016llx %-13s %016llx\n",
4898 "dr6:", save->dr6, "dr7:", save->dr7);
4899 pr_err("%-15s %016llx %-13s %016llx\n",
4900 "rip:", save->rip, "rflags:", save->rflags);
4901 pr_err("%-15s %016llx %-13s %016llx\n",
4902 "rsp:", save->rsp, "rax:", save->rax);
4903 pr_err("%-15s %016llx %-13s %016llx\n",
4904 "star:", save->star, "lstar:", save->lstar);
4905 pr_err("%-15s %016llx %-13s %016llx\n",
4906 "cstar:", save->cstar, "sfmask:", save->sfmask);
4907 pr_err("%-15s %016llx %-13s %016llx\n",
4908 "kernel_gs_base:", save->kernel_gs_base,
4909 "sysenter_cs:", save->sysenter_cs);
4910 pr_err("%-15s %016llx %-13s %016llx\n",
4911 "sysenter_esp:", save->sysenter_esp,
4912 "sysenter_eip:", save->sysenter_eip);
4913 pr_err("%-15s %016llx %-13s %016llx\n",
4914 "gpat:", save->g_pat, "dbgctl:", save->dbgctl);
4915 pr_err("%-15s %016llx %-13s %016llx\n",
4916 "br_from:", save->br_from, "br_to:", save->br_to);
4917 pr_err("%-15s %016llx %-13s %016llx\n",
4918 "excp_from:", save->last_excp_from,
4919 "excp_to:", save->last_excp_to);
4922 static void svm_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
4924 struct vmcb_control_area *control = &to_svm(vcpu)->vmcb->control;
4926 *info1 = control->exit_info_1;
4927 *info2 = control->exit_info_2;
4930 static int handle_exit(struct kvm_vcpu *vcpu)
4932 struct vcpu_svm *svm = to_svm(vcpu);
4933 struct kvm_run *kvm_run = vcpu->run;
4934 u32 exit_code = svm->vmcb->control.exit_code;
4936 trace_kvm_exit(exit_code, vcpu, KVM_ISA_SVM);
4938 if (!is_cr_intercept(svm, INTERCEPT_CR0_WRITE))
4939 vcpu->arch.cr0 = svm->vmcb->save.cr0;
4941 vcpu->arch.cr3 = svm->vmcb->save.cr3;
4943 if (unlikely(svm->nested.exit_required)) {
4944 nested_svm_vmexit(svm);
4945 svm->nested.exit_required = false;
4950 if (is_guest_mode(vcpu)) {
4953 trace_kvm_nested_vmexit(svm->vmcb->save.rip, exit_code,
4954 svm->vmcb->control.exit_info_1,
4955 svm->vmcb->control.exit_info_2,
4956 svm->vmcb->control.exit_int_info,
4957 svm->vmcb->control.exit_int_info_err,
4960 vmexit = nested_svm_exit_special(svm);
4962 if (vmexit == NESTED_EXIT_CONTINUE)
4963 vmexit = nested_svm_exit_handled(svm);
4965 if (vmexit == NESTED_EXIT_DONE)
4969 svm_complete_interrupts(svm);
4971 if (svm->vmcb->control.exit_code == SVM_EXIT_ERR) {
4972 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4973 kvm_run->fail_entry.hardware_entry_failure_reason
4974 = svm->vmcb->control.exit_code;
4975 pr_err("KVM: FAILED VMRUN WITH VMCB:\n");
4980 if (is_external_interrupt(svm->vmcb->control.exit_int_info) &&
4981 exit_code != SVM_EXIT_EXCP_BASE + PF_VECTOR &&
4982 exit_code != SVM_EXIT_NPF && exit_code != SVM_EXIT_TASK_SWITCH &&
4983 exit_code != SVM_EXIT_INTR && exit_code != SVM_EXIT_NMI)
4984 printk(KERN_ERR "%s: unexpected exit_int_info 0x%x "
4986 __func__, svm->vmcb->control.exit_int_info,
4989 if (exit_code >= ARRAY_SIZE(svm_exit_handlers)
4990 || !svm_exit_handlers[exit_code]) {
4991 WARN_ONCE(1, "svm: unexpected exit reason 0x%x\n", exit_code);
4992 kvm_queue_exception(vcpu, UD_VECTOR);
4996 return svm_exit_handlers[exit_code](svm);
4999 static void reload_tss(struct kvm_vcpu *vcpu)
5001 int cpu = raw_smp_processor_id();
5003 struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
5004 sd->tss_desc->type = 9; /* available 32/64-bit TSS */
5008 static void pre_sev_run(struct vcpu_svm *svm, int cpu)
5010 struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
5011 int asid = sev_get_asid(svm->vcpu.kvm);
5013 /* Assign the asid allocated with this SEV guest */
5014 svm->vmcb->control.asid = asid;
5019 * 1) when different VMCB for the same ASID is to be run on the same host CPU.
5020 * 2) or this VMCB was executed on different host CPU in previous VMRUNs.
5022 if (sd->sev_vmcbs[asid] == svm->vmcb &&
5023 svm->last_cpu == cpu)
5026 svm->last_cpu = cpu;
5027 sd->sev_vmcbs[asid] = svm->vmcb;
5028 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ASID;
5029 mark_dirty(svm->vmcb, VMCB_ASID);
5032 static void pre_svm_run(struct vcpu_svm *svm)
5034 int cpu = raw_smp_processor_id();
5036 struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
5038 if (sev_guest(svm->vcpu.kvm))
5039 return pre_sev_run(svm, cpu);
5041 /* FIXME: handle wraparound of asid_generation */
5042 if (svm->asid_generation != sd->asid_generation)
5046 static void svm_inject_nmi(struct kvm_vcpu *vcpu)
5048 struct vcpu_svm *svm = to_svm(vcpu);
5050 svm->vmcb->control.event_inj = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI;
5051 vcpu->arch.hflags |= HF_NMI_MASK;
5052 set_intercept(svm, INTERCEPT_IRET);
5053 ++vcpu->stat.nmi_injections;
5056 static inline void svm_inject_irq(struct vcpu_svm *svm, int irq)
5058 struct vmcb_control_area *control;
5060 /* The following fields are ignored when AVIC is enabled */
5061 control = &svm->vmcb->control;
5062 control->int_vector = irq;
5063 control->int_ctl &= ~V_INTR_PRIO_MASK;
5064 control->int_ctl |= V_IRQ_MASK |
5065 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT);
5066 mark_dirty(svm->vmcb, VMCB_INTR);
5069 static void svm_set_irq(struct kvm_vcpu *vcpu)
5071 struct vcpu_svm *svm = to_svm(vcpu);
5073 BUG_ON(!(gif_set(svm)));
5075 trace_kvm_inj_virq(vcpu->arch.interrupt.nr);
5076 ++vcpu->stat.irq_injections;
5078 svm->vmcb->control.event_inj = vcpu->arch.interrupt.nr |
5079 SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR;
5082 static inline bool svm_nested_virtualize_tpr(struct kvm_vcpu *vcpu)
5084 return is_guest_mode(vcpu) && (vcpu->arch.hflags & HF_VINTR_MASK);
5087 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
5089 struct vcpu_svm *svm = to_svm(vcpu);
5091 if (svm_nested_virtualize_tpr(vcpu) ||
5092 kvm_vcpu_apicv_active(vcpu))
5095 clr_cr_intercept(svm, INTERCEPT_CR8_WRITE);
5101 set_cr_intercept(svm, INTERCEPT_CR8_WRITE);
5104 static void svm_set_virtual_apic_mode(struct kvm_vcpu *vcpu)
5109 static bool svm_get_enable_apicv(struct kvm_vcpu *vcpu)
5111 return avic && irqchip_split(vcpu->kvm);
5114 static void svm_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
5118 static void svm_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr)
5122 /* Note: Currently only used by Hyper-V. */
5123 static void svm_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu)
5125 struct vcpu_svm *svm = to_svm(vcpu);
5126 struct vmcb *vmcb = svm->vmcb;
5128 if (!kvm_vcpu_apicv_active(&svm->vcpu))
5131 vmcb->control.int_ctl &= ~AVIC_ENABLE_MASK;
5132 mark_dirty(vmcb, VMCB_INTR);
5135 static void svm_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
5140 static void svm_deliver_avic_intr(struct kvm_vcpu *vcpu, int vec)
5142 kvm_lapic_set_irr(vec, vcpu->arch.apic);
5143 smp_mb__after_atomic();
5145 if (avic_vcpu_is_running(vcpu))
5146 wrmsrl(SVM_AVIC_DOORBELL,
5147 kvm_cpu_get_apicid(vcpu->cpu));
5149 kvm_vcpu_wake_up(vcpu);
5152 static void svm_ir_list_del(struct vcpu_svm *svm, struct amd_iommu_pi_data *pi)
5154 unsigned long flags;
5155 struct amd_svm_iommu_ir *cur;
5157 spin_lock_irqsave(&svm->ir_list_lock, flags);
5158 list_for_each_entry(cur, &svm->ir_list, node) {
5159 if (cur->data != pi->ir_data)
5161 list_del(&cur->node);
5165 spin_unlock_irqrestore(&svm->ir_list_lock, flags);
5168 static int svm_ir_list_add(struct vcpu_svm *svm, struct amd_iommu_pi_data *pi)
5171 unsigned long flags;
5172 struct amd_svm_iommu_ir *ir;
5175 * In some cases, the existing irte is updaed and re-set,
5176 * so we need to check here if it's already been * added
5179 if (pi->ir_data && (pi->prev_ga_tag != 0)) {
5180 struct kvm *kvm = svm->vcpu.kvm;
5181 u32 vcpu_id = AVIC_GATAG_TO_VCPUID(pi->prev_ga_tag);
5182 struct kvm_vcpu *prev_vcpu = kvm_get_vcpu_by_id(kvm, vcpu_id);
5183 struct vcpu_svm *prev_svm;
5190 prev_svm = to_svm(prev_vcpu);
5191 svm_ir_list_del(prev_svm, pi);
5195 * Allocating new amd_iommu_pi_data, which will get
5196 * add to the per-vcpu ir_list.
5198 ir = kzalloc(sizeof(struct amd_svm_iommu_ir), GFP_KERNEL);
5203 ir->data = pi->ir_data;
5205 spin_lock_irqsave(&svm->ir_list_lock, flags);
5206 list_add(&ir->node, &svm->ir_list);
5207 spin_unlock_irqrestore(&svm->ir_list_lock, flags);
5214 * The HW cannot support posting multicast/broadcast
5215 * interrupts to a vCPU. So, we still use legacy interrupt
5216 * remapping for these kind of interrupts.
5218 * For lowest-priority interrupts, we only support
5219 * those with single CPU as the destination, e.g. user
5220 * configures the interrupts via /proc/irq or uses
5221 * irqbalance to make the interrupts single-CPU.
5224 get_pi_vcpu_info(struct kvm *kvm, struct kvm_kernel_irq_routing_entry *e,
5225 struct vcpu_data *vcpu_info, struct vcpu_svm **svm)
5227 struct kvm_lapic_irq irq;
5228 struct kvm_vcpu *vcpu = NULL;
5230 kvm_set_msi_irq(kvm, e, &irq);
5232 if (!kvm_intr_is_single_vcpu(kvm, &irq, &vcpu)) {
5233 pr_debug("SVM: %s: use legacy intr remap mode for irq %u\n",
5234 __func__, irq.vector);
5238 pr_debug("SVM: %s: use GA mode for irq %u\n", __func__,
5240 *svm = to_svm(vcpu);
5241 vcpu_info->pi_desc_addr = __sme_set(page_to_phys((*svm)->avic_backing_page));
5242 vcpu_info->vector = irq.vector;
5248 * svm_update_pi_irte - set IRTE for Posted-Interrupts
5251 * @host_irq: host irq of the interrupt
5252 * @guest_irq: gsi of the interrupt
5253 * @set: set or unset PI
5254 * returns 0 on success, < 0 on failure
5256 static int svm_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
5257 uint32_t guest_irq, bool set)
5259 struct kvm_kernel_irq_routing_entry *e;
5260 struct kvm_irq_routing_table *irq_rt;
5261 int idx, ret = -EINVAL;
5263 if (!kvm_arch_has_assigned_device(kvm) ||
5264 !irq_remapping_cap(IRQ_POSTING_CAP))
5267 pr_debug("SVM: %s: host_irq=%#x, guest_irq=%#x, set=%#x\n",
5268 __func__, host_irq, guest_irq, set);
5270 idx = srcu_read_lock(&kvm->irq_srcu);
5271 irq_rt = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu);
5272 WARN_ON(guest_irq >= irq_rt->nr_rt_entries);
5274 hlist_for_each_entry(e, &irq_rt->map[guest_irq], link) {
5275 struct vcpu_data vcpu_info;
5276 struct vcpu_svm *svm = NULL;
5278 if (e->type != KVM_IRQ_ROUTING_MSI)
5282 * Here, we setup with legacy mode in the following cases:
5283 * 1. When cannot target interrupt to a specific vcpu.
5284 * 2. Unsetting posted interrupt.
5285 * 3. APIC virtialization is disabled for the vcpu.
5287 if (!get_pi_vcpu_info(kvm, e, &vcpu_info, &svm) && set &&
5288 kvm_vcpu_apicv_active(&svm->vcpu)) {
5289 struct amd_iommu_pi_data pi;
5291 /* Try to enable guest_mode in IRTE */
5292 pi.base = __sme_set(page_to_phys(svm->avic_backing_page) &
5294 pi.ga_tag = AVIC_GATAG(to_kvm_svm(kvm)->avic_vm_id,
5296 pi.is_guest_mode = true;
5297 pi.vcpu_data = &vcpu_info;
5298 ret = irq_set_vcpu_affinity(host_irq, &pi);
5301 * Here, we successfully setting up vcpu affinity in
5302 * IOMMU guest mode. Now, we need to store the posted
5303 * interrupt information in a per-vcpu ir_list so that
5304 * we can reference to them directly when we update vcpu
5305 * scheduling information in IOMMU irte.
5307 if (!ret && pi.is_guest_mode)
5308 svm_ir_list_add(svm, &pi);
5310 /* Use legacy mode in IRTE */
5311 struct amd_iommu_pi_data pi;
5314 * Here, pi is used to:
5315 * - Tell IOMMU to use legacy mode for this interrupt.
5316 * - Retrieve ga_tag of prior interrupt remapping data.
5318 pi.is_guest_mode = false;
5319 ret = irq_set_vcpu_affinity(host_irq, &pi);
5322 * Check if the posted interrupt was previously
5323 * setup with the guest_mode by checking if the ga_tag
5324 * was cached. If so, we need to clean up the per-vcpu
5327 if (!ret && pi.prev_ga_tag) {
5328 int id = AVIC_GATAG_TO_VCPUID(pi.prev_ga_tag);
5329 struct kvm_vcpu *vcpu;
5331 vcpu = kvm_get_vcpu_by_id(kvm, id);
5333 svm_ir_list_del(to_svm(vcpu), &pi);
5338 trace_kvm_pi_irte_update(host_irq, svm->vcpu.vcpu_id,
5339 e->gsi, vcpu_info.vector,
5340 vcpu_info.pi_desc_addr, set);
5344 pr_err("%s: failed to update PI IRTE\n", __func__);
5351 srcu_read_unlock(&kvm->irq_srcu, idx);
5355 static int svm_nmi_allowed(struct kvm_vcpu *vcpu)
5357 struct vcpu_svm *svm = to_svm(vcpu);
5358 struct vmcb *vmcb = svm->vmcb;
5360 ret = !(vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK) &&
5361 !(svm->vcpu.arch.hflags & HF_NMI_MASK);
5362 ret = ret && gif_set(svm) && nested_svm_nmi(svm);
5367 static bool svm_get_nmi_mask(struct kvm_vcpu *vcpu)
5369 struct vcpu_svm *svm = to_svm(vcpu);
5371 return !!(svm->vcpu.arch.hflags & HF_NMI_MASK);
5374 static void svm_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
5376 struct vcpu_svm *svm = to_svm(vcpu);
5379 svm->vcpu.arch.hflags |= HF_NMI_MASK;
5380 set_intercept(svm, INTERCEPT_IRET);
5382 svm->vcpu.arch.hflags &= ~HF_NMI_MASK;
5383 clr_intercept(svm, INTERCEPT_IRET);
5387 static int svm_interrupt_allowed(struct kvm_vcpu *vcpu)
5389 struct vcpu_svm *svm = to_svm(vcpu);
5390 struct vmcb *vmcb = svm->vmcb;
5393 if (!gif_set(svm) ||
5394 (vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK))
5397 ret = !!(kvm_get_rflags(vcpu) & X86_EFLAGS_IF);
5399 if (is_guest_mode(vcpu))
5400 return ret && !(svm->vcpu.arch.hflags & HF_VINTR_MASK);
5405 static void enable_irq_window(struct kvm_vcpu *vcpu)
5407 struct vcpu_svm *svm = to_svm(vcpu);
5409 if (kvm_vcpu_apicv_active(vcpu))
5413 * In case GIF=0 we can't rely on the CPU to tell us when GIF becomes
5414 * 1, because that's a separate STGI/VMRUN intercept. The next time we
5415 * get that intercept, this function will be called again though and
5416 * we'll get the vintr intercept. However, if the vGIF feature is
5417 * enabled, the STGI interception will not occur. Enable the irq
5418 * window under the assumption that the hardware will set the GIF.
5420 if ((vgif_enabled(svm) || gif_set(svm)) && nested_svm_intr(svm)) {
5422 svm_inject_irq(svm, 0x0);
5426 static void enable_nmi_window(struct kvm_vcpu *vcpu)
5428 struct vcpu_svm *svm = to_svm(vcpu);
5430 if ((svm->vcpu.arch.hflags & (HF_NMI_MASK | HF_IRET_MASK))
5432 return; /* IRET will cause a vm exit */
5434 if (!gif_set(svm)) {
5435 if (vgif_enabled(svm))
5436 set_intercept(svm, INTERCEPT_STGI);
5437 return; /* STGI will cause a vm exit */
5440 if (svm->nested.exit_required)
5441 return; /* we're not going to run the guest yet */
5444 * Something prevents NMI from been injected. Single step over possible
5445 * problem (IRET or exception injection or interrupt shadow)
5447 svm->nmi_singlestep_guest_rflags = svm_get_rflags(vcpu);
5448 svm->nmi_singlestep = true;
5449 svm->vmcb->save.rflags |= (X86_EFLAGS_TF | X86_EFLAGS_RF);
5452 static int svm_set_tss_addr(struct kvm *kvm, unsigned int addr)
5457 static int svm_set_identity_map_addr(struct kvm *kvm, u64 ident_addr)
5462 static void svm_flush_tlb(struct kvm_vcpu *vcpu, bool invalidate_gpa)
5464 struct vcpu_svm *svm = to_svm(vcpu);
5466 if (static_cpu_has(X86_FEATURE_FLUSHBYASID))
5467 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ASID;
5469 svm->asid_generation--;
5472 static void svm_flush_tlb_gva(struct kvm_vcpu *vcpu, gva_t gva)
5474 struct vcpu_svm *svm = to_svm(vcpu);
5476 invlpga(gva, svm->vmcb->control.asid);
5479 static void svm_prepare_guest_switch(struct kvm_vcpu *vcpu)
5483 static inline void sync_cr8_to_lapic(struct kvm_vcpu *vcpu)
5485 struct vcpu_svm *svm = to_svm(vcpu);
5487 if (svm_nested_virtualize_tpr(vcpu))
5490 if (!is_cr_intercept(svm, INTERCEPT_CR8_WRITE)) {
5491 int cr8 = svm->vmcb->control.int_ctl & V_TPR_MASK;
5492 kvm_set_cr8(vcpu, cr8);
5496 static inline void sync_lapic_to_cr8(struct kvm_vcpu *vcpu)
5498 struct vcpu_svm *svm = to_svm(vcpu);
5501 if (svm_nested_virtualize_tpr(vcpu) ||
5502 kvm_vcpu_apicv_active(vcpu))
5505 cr8 = kvm_get_cr8(vcpu);
5506 svm->vmcb->control.int_ctl &= ~V_TPR_MASK;
5507 svm->vmcb->control.int_ctl |= cr8 & V_TPR_MASK;
5510 static void svm_complete_interrupts(struct vcpu_svm *svm)
5514 u32 exitintinfo = svm->vmcb->control.exit_int_info;
5515 unsigned int3_injected = svm->int3_injected;
5517 svm->int3_injected = 0;
5520 * If we've made progress since setting HF_IRET_MASK, we've
5521 * executed an IRET and can allow NMI injection.
5523 if ((svm->vcpu.arch.hflags & HF_IRET_MASK)
5524 && kvm_rip_read(&svm->vcpu) != svm->nmi_iret_rip) {
5525 svm->vcpu.arch.hflags &= ~(HF_NMI_MASK | HF_IRET_MASK);
5526 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
5529 svm->vcpu.arch.nmi_injected = false;
5530 kvm_clear_exception_queue(&svm->vcpu);
5531 kvm_clear_interrupt_queue(&svm->vcpu);
5533 if (!(exitintinfo & SVM_EXITINTINFO_VALID))
5536 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
5538 vector = exitintinfo & SVM_EXITINTINFO_VEC_MASK;
5539 type = exitintinfo & SVM_EXITINTINFO_TYPE_MASK;
5542 case SVM_EXITINTINFO_TYPE_NMI:
5543 svm->vcpu.arch.nmi_injected = true;
5545 case SVM_EXITINTINFO_TYPE_EXEPT:
5547 * In case of software exceptions, do not reinject the vector,
5548 * but re-execute the instruction instead. Rewind RIP first
5549 * if we emulated INT3 before.
5551 if (kvm_exception_is_soft(vector)) {
5552 if (vector == BP_VECTOR && int3_injected &&
5553 kvm_is_linear_rip(&svm->vcpu, svm->int3_rip))
5554 kvm_rip_write(&svm->vcpu,
5555 kvm_rip_read(&svm->vcpu) -
5559 if (exitintinfo & SVM_EXITINTINFO_VALID_ERR) {
5560 u32 err = svm->vmcb->control.exit_int_info_err;
5561 kvm_requeue_exception_e(&svm->vcpu, vector, err);
5564 kvm_requeue_exception(&svm->vcpu, vector);
5566 case SVM_EXITINTINFO_TYPE_INTR:
5567 kvm_queue_interrupt(&svm->vcpu, vector, false);
5574 static void svm_cancel_injection(struct kvm_vcpu *vcpu)
5576 struct vcpu_svm *svm = to_svm(vcpu);
5577 struct vmcb_control_area *control = &svm->vmcb->control;
5579 control->exit_int_info = control->event_inj;
5580 control->exit_int_info_err = control->event_inj_err;
5581 control->event_inj = 0;
5582 svm_complete_interrupts(svm);
5585 static void svm_vcpu_run(struct kvm_vcpu *vcpu)
5587 struct vcpu_svm *svm = to_svm(vcpu);
5589 svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX];
5590 svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP];
5591 svm->vmcb->save.rip = vcpu->arch.regs[VCPU_REGS_RIP];
5594 * A vmexit emulation is required before the vcpu can be executed
5597 if (unlikely(svm->nested.exit_required))
5601 * Disable singlestep if we're injecting an interrupt/exception.
5602 * We don't want our modified rflags to be pushed on the stack where
5603 * we might not be able to easily reset them if we disabled NMI
5606 if (svm->nmi_singlestep && svm->vmcb->control.event_inj) {
5608 * Event injection happens before external interrupts cause a
5609 * vmexit and interrupts are disabled here, so smp_send_reschedule
5610 * is enough to force an immediate vmexit.
5612 disable_nmi_singlestep(svm);
5613 smp_send_reschedule(vcpu->cpu);
5618 sync_lapic_to_cr8(vcpu);
5620 svm->vmcb->save.cr2 = vcpu->arch.cr2;
5625 * If this vCPU has touched SPEC_CTRL, restore the guest's value if
5626 * it's non-zero. Since vmentry is serialising on affected CPUs, there
5627 * is no need to worry about the conditional branch over the wrmsr
5628 * being speculatively taken.
5630 x86_spec_ctrl_set_guest(svm->spec_ctrl, svm->virt_spec_ctrl);
5635 "push %%" _ASM_BP "; \n\t"
5636 "mov %c[rbx](%[svm]), %%" _ASM_BX " \n\t"
5637 "mov %c[rcx](%[svm]), %%" _ASM_CX " \n\t"
5638 "mov %c[rdx](%[svm]), %%" _ASM_DX " \n\t"
5639 "mov %c[rsi](%[svm]), %%" _ASM_SI " \n\t"
5640 "mov %c[rdi](%[svm]), %%" _ASM_DI " \n\t"
5641 "mov %c[rbp](%[svm]), %%" _ASM_BP " \n\t"
5642 #ifdef CONFIG_X86_64
5643 "mov %c[r8](%[svm]), %%r8 \n\t"
5644 "mov %c[r9](%[svm]), %%r9 \n\t"
5645 "mov %c[r10](%[svm]), %%r10 \n\t"
5646 "mov %c[r11](%[svm]), %%r11 \n\t"
5647 "mov %c[r12](%[svm]), %%r12 \n\t"
5648 "mov %c[r13](%[svm]), %%r13 \n\t"
5649 "mov %c[r14](%[svm]), %%r14 \n\t"
5650 "mov %c[r15](%[svm]), %%r15 \n\t"
5653 /* Enter guest mode */
5654 "push %%" _ASM_AX " \n\t"
5655 "mov %c[vmcb](%[svm]), %%" _ASM_AX " \n\t"
5656 __ex("vmload %%" _ASM_AX) "\n\t"
5657 __ex("vmrun %%" _ASM_AX) "\n\t"
5658 __ex("vmsave %%" _ASM_AX) "\n\t"
5659 "pop %%" _ASM_AX " \n\t"
5661 /* Save guest registers, load host registers */
5662 "mov %%" _ASM_BX ", %c[rbx](%[svm]) \n\t"
5663 "mov %%" _ASM_CX ", %c[rcx](%[svm]) \n\t"
5664 "mov %%" _ASM_DX ", %c[rdx](%[svm]) \n\t"
5665 "mov %%" _ASM_SI ", %c[rsi](%[svm]) \n\t"
5666 "mov %%" _ASM_DI ", %c[rdi](%[svm]) \n\t"
5667 "mov %%" _ASM_BP ", %c[rbp](%[svm]) \n\t"
5668 #ifdef CONFIG_X86_64
5669 "mov %%r8, %c[r8](%[svm]) \n\t"
5670 "mov %%r9, %c[r9](%[svm]) \n\t"
5671 "mov %%r10, %c[r10](%[svm]) \n\t"
5672 "mov %%r11, %c[r11](%[svm]) \n\t"
5673 "mov %%r12, %c[r12](%[svm]) \n\t"
5674 "mov %%r13, %c[r13](%[svm]) \n\t"
5675 "mov %%r14, %c[r14](%[svm]) \n\t"
5676 "mov %%r15, %c[r15](%[svm]) \n\t"
5678 * Clear host registers marked as clobbered to prevent
5681 "xor %%r8d, %%r8d \n\t"
5682 "xor %%r9d, %%r9d \n\t"
5683 "xor %%r10d, %%r10d \n\t"
5684 "xor %%r11d, %%r11d \n\t"
5685 "xor %%r12d, %%r12d \n\t"
5686 "xor %%r13d, %%r13d \n\t"
5687 "xor %%r14d, %%r14d \n\t"
5688 "xor %%r15d, %%r15d \n\t"
5690 "xor %%ebx, %%ebx \n\t"
5691 "xor %%ecx, %%ecx \n\t"
5692 "xor %%edx, %%edx \n\t"
5693 "xor %%esi, %%esi \n\t"
5694 "xor %%edi, %%edi \n\t"
5698 [vmcb]"i"(offsetof(struct vcpu_svm, vmcb_pa)),
5699 [rbx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBX])),
5700 [rcx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RCX])),
5701 [rdx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDX])),
5702 [rsi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RSI])),
5703 [rdi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDI])),
5704 [rbp]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBP]))
5705 #ifdef CONFIG_X86_64
5706 , [r8]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R8])),
5707 [r9]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R9])),
5708 [r10]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R10])),
5709 [r11]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R11])),
5710 [r12]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R12])),
5711 [r13]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R13])),
5712 [r14]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R14])),
5713 [r15]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R15]))
5716 #ifdef CONFIG_X86_64
5717 , "rbx", "rcx", "rdx", "rsi", "rdi"
5718 , "r8", "r9", "r10", "r11" , "r12", "r13", "r14", "r15"
5720 , "ebx", "ecx", "edx", "esi", "edi"
5724 /* Eliminate branch target predictions from guest mode */
5727 #ifdef CONFIG_X86_64
5728 wrmsrl(MSR_GS_BASE, svm->host.gs_base);
5730 loadsegment(fs, svm->host.fs);
5731 #ifndef CONFIG_X86_32_LAZY_GS
5732 loadsegment(gs, svm->host.gs);
5737 * We do not use IBRS in the kernel. If this vCPU has used the
5738 * SPEC_CTRL MSR it may have left it on; save the value and
5739 * turn it off. This is much more efficient than blindly adding
5740 * it to the atomic save/restore list. Especially as the former
5741 * (Saving guest MSRs on vmexit) doesn't even exist in KVM.
5743 * For non-nested case:
5744 * If the L01 MSR bitmap does not intercept the MSR, then we need to
5748 * If the L02 MSR bitmap does not intercept the MSR, then we need to
5751 if (unlikely(!msr_write_intercepted(vcpu, MSR_IA32_SPEC_CTRL)))
5752 svm->spec_ctrl = native_read_msr(MSR_IA32_SPEC_CTRL);
5756 local_irq_disable();
5758 x86_spec_ctrl_restore_host(svm->spec_ctrl, svm->virt_spec_ctrl);
5760 vcpu->arch.cr2 = svm->vmcb->save.cr2;
5761 vcpu->arch.regs[VCPU_REGS_RAX] = svm->vmcb->save.rax;
5762 vcpu->arch.regs[VCPU_REGS_RSP] = svm->vmcb->save.rsp;
5763 vcpu->arch.regs[VCPU_REGS_RIP] = svm->vmcb->save.rip;
5765 if (unlikely(svm->vmcb->control.exit_code == SVM_EXIT_NMI))
5766 kvm_before_interrupt(&svm->vcpu);
5770 /* Any pending NMI will happen here */
5772 if (unlikely(svm->vmcb->control.exit_code == SVM_EXIT_NMI))
5773 kvm_after_interrupt(&svm->vcpu);
5775 sync_cr8_to_lapic(vcpu);
5779 svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
5781 /* if exit due to PF check for async PF */
5782 if (svm->vmcb->control.exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR)
5783 svm->vcpu.arch.apf.host_apf_reason = kvm_read_and_reset_pf_reason();
5786 vcpu->arch.regs_avail &= ~(1 << VCPU_EXREG_PDPTR);
5787 vcpu->arch.regs_dirty &= ~(1 << VCPU_EXREG_PDPTR);
5791 * We need to handle MC intercepts here before the vcpu has a chance to
5792 * change the physical cpu
5794 if (unlikely(svm->vmcb->control.exit_code ==
5795 SVM_EXIT_EXCP_BASE + MC_VECTOR))
5796 svm_handle_mce(svm);
5798 mark_all_clean(svm->vmcb);
5800 STACK_FRAME_NON_STANDARD(svm_vcpu_run);
5802 static void svm_set_cr3(struct kvm_vcpu *vcpu, unsigned long root)
5804 struct vcpu_svm *svm = to_svm(vcpu);
5806 svm->vmcb->save.cr3 = __sme_set(root);
5807 mark_dirty(svm->vmcb, VMCB_CR);
5810 static void set_tdp_cr3(struct kvm_vcpu *vcpu, unsigned long root)
5812 struct vcpu_svm *svm = to_svm(vcpu);
5814 svm->vmcb->control.nested_cr3 = __sme_set(root);
5815 mark_dirty(svm->vmcb, VMCB_NPT);
5817 /* Also sync guest cr3 here in case we live migrate */
5818 svm->vmcb->save.cr3 = kvm_read_cr3(vcpu);
5819 mark_dirty(svm->vmcb, VMCB_CR);
5822 static int is_disabled(void)
5826 rdmsrl(MSR_VM_CR, vm_cr);
5827 if (vm_cr & (1 << SVM_VM_CR_SVM_DISABLE))
5834 svm_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
5837 * Patch in the VMMCALL instruction:
5839 hypercall[0] = 0x0f;
5840 hypercall[1] = 0x01;
5841 hypercall[2] = 0xd9;
5844 static void svm_check_processor_compat(void *rtn)
5849 static bool svm_cpu_has_accelerated_tpr(void)
5854 static bool svm_has_emulated_msr(int index)
5857 case MSR_IA32_MCG_EXT_CTL:
5866 static u64 svm_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
5871 static void svm_cpuid_update(struct kvm_vcpu *vcpu)
5873 struct vcpu_svm *svm = to_svm(vcpu);
5875 /* Update nrips enabled cache */
5876 svm->nrips_enabled = !!guest_cpuid_has(&svm->vcpu, X86_FEATURE_NRIPS);
5878 if (!kvm_vcpu_apicv_active(vcpu))
5881 guest_cpuid_clear(vcpu, X86_FEATURE_X2APIC);
5884 static void svm_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
5889 entry->ecx &= ~bit(X86_FEATURE_X2APIC);
5893 entry->ecx |= (1 << 2); /* Set SVM bit */
5896 entry->eax = 1; /* SVM revision 1 */
5897 entry->ebx = 8; /* Lets support 8 ASIDs in case we add proper
5898 ASID emulation to nested SVM */
5899 entry->ecx = 0; /* Reserved */
5900 entry->edx = 0; /* Per default do not support any
5901 additional features */
5903 /* Support next_rip if host supports it */
5904 if (boot_cpu_has(X86_FEATURE_NRIPS))
5905 entry->edx |= SVM_FEATURE_NRIP;
5907 /* Support NPT for the guest if enabled */
5909 entry->edx |= SVM_FEATURE_NPT;
5913 /* Support memory encryption cpuid if host supports it */
5914 if (boot_cpu_has(X86_FEATURE_SEV))
5915 cpuid(0x8000001f, &entry->eax, &entry->ebx,
5916 &entry->ecx, &entry->edx);
5921 static int svm_get_lpage_level(void)
5923 return PT_PDPE_LEVEL;
5926 static bool svm_rdtscp_supported(void)
5928 return boot_cpu_has(X86_FEATURE_RDTSCP);
5931 static bool svm_invpcid_supported(void)
5936 static bool svm_mpx_supported(void)
5941 static bool svm_xsaves_supported(void)
5946 static bool svm_umip_emulated(void)
5951 static bool svm_pt_supported(void)
5956 static bool svm_has_wbinvd_exit(void)
5961 #define PRE_EX(exit) { .exit_code = (exit), \
5962 .stage = X86_ICPT_PRE_EXCEPT, }
5963 #define POST_EX(exit) { .exit_code = (exit), \
5964 .stage = X86_ICPT_POST_EXCEPT, }
5965 #define POST_MEM(exit) { .exit_code = (exit), \
5966 .stage = X86_ICPT_POST_MEMACCESS, }
5968 static const struct __x86_intercept {
5970 enum x86_intercept_stage stage;
5971 } x86_intercept_map[] = {
5972 [x86_intercept_cr_read] = POST_EX(SVM_EXIT_READ_CR0),
5973 [x86_intercept_cr_write] = POST_EX(SVM_EXIT_WRITE_CR0),
5974 [x86_intercept_clts] = POST_EX(SVM_EXIT_WRITE_CR0),
5975 [x86_intercept_lmsw] = POST_EX(SVM_EXIT_WRITE_CR0),
5976 [x86_intercept_smsw] = POST_EX(SVM_EXIT_READ_CR0),
5977 [x86_intercept_dr_read] = POST_EX(SVM_EXIT_READ_DR0),
5978 [x86_intercept_dr_write] = POST_EX(SVM_EXIT_WRITE_DR0),
5979 [x86_intercept_sldt] = POST_EX(SVM_EXIT_LDTR_READ),
5980 [x86_intercept_str] = POST_EX(SVM_EXIT_TR_READ),
5981 [x86_intercept_lldt] = POST_EX(SVM_EXIT_LDTR_WRITE),
5982 [x86_intercept_ltr] = POST_EX(SVM_EXIT_TR_WRITE),
5983 [x86_intercept_sgdt] = POST_EX(SVM_EXIT_GDTR_READ),
5984 [x86_intercept_sidt] = POST_EX(SVM_EXIT_IDTR_READ),
5985 [x86_intercept_lgdt] = POST_EX(SVM_EXIT_GDTR_WRITE),
5986 [x86_intercept_lidt] = POST_EX(SVM_EXIT_IDTR_WRITE),
5987 [x86_intercept_vmrun] = POST_EX(SVM_EXIT_VMRUN),
5988 [x86_intercept_vmmcall] = POST_EX(SVM_EXIT_VMMCALL),
5989 [x86_intercept_vmload] = POST_EX(SVM_EXIT_VMLOAD),
5990 [x86_intercept_vmsave] = POST_EX(SVM_EXIT_VMSAVE),
5991 [x86_intercept_stgi] = POST_EX(SVM_EXIT_STGI),
5992 [x86_intercept_clgi] = POST_EX(SVM_EXIT_CLGI),
5993 [x86_intercept_skinit] = POST_EX(SVM_EXIT_SKINIT),
5994 [x86_intercept_invlpga] = POST_EX(SVM_EXIT_INVLPGA),
5995 [x86_intercept_rdtscp] = POST_EX(SVM_EXIT_RDTSCP),
5996 [x86_intercept_monitor] = POST_MEM(SVM_EXIT_MONITOR),
5997 [x86_intercept_mwait] = POST_EX(SVM_EXIT_MWAIT),
5998 [x86_intercept_invlpg] = POST_EX(SVM_EXIT_INVLPG),
5999 [x86_intercept_invd] = POST_EX(SVM_EXIT_INVD),
6000 [x86_intercept_wbinvd] = POST_EX(SVM_EXIT_WBINVD),
6001 [x86_intercept_wrmsr] = POST_EX(SVM_EXIT_MSR),
6002 [x86_intercept_rdtsc] = POST_EX(SVM_EXIT_RDTSC),
6003 [x86_intercept_rdmsr] = POST_EX(SVM_EXIT_MSR),
6004 [x86_intercept_rdpmc] = POST_EX(SVM_EXIT_RDPMC),
6005 [x86_intercept_cpuid] = PRE_EX(SVM_EXIT_CPUID),
6006 [x86_intercept_rsm] = PRE_EX(SVM_EXIT_RSM),
6007 [x86_intercept_pause] = PRE_EX(SVM_EXIT_PAUSE),
6008 [x86_intercept_pushf] = PRE_EX(SVM_EXIT_PUSHF),
6009 [x86_intercept_popf] = PRE_EX(SVM_EXIT_POPF),
6010 [x86_intercept_intn] = PRE_EX(SVM_EXIT_SWINT),
6011 [x86_intercept_iret] = PRE_EX(SVM_EXIT_IRET),
6012 [x86_intercept_icebp] = PRE_EX(SVM_EXIT_ICEBP),
6013 [x86_intercept_hlt] = POST_EX(SVM_EXIT_HLT),
6014 [x86_intercept_in] = POST_EX(SVM_EXIT_IOIO),
6015 [x86_intercept_ins] = POST_EX(SVM_EXIT_IOIO),
6016 [x86_intercept_out] = POST_EX(SVM_EXIT_IOIO),
6017 [x86_intercept_outs] = POST_EX(SVM_EXIT_IOIO),
6024 static int svm_check_intercept(struct kvm_vcpu *vcpu,
6025 struct x86_instruction_info *info,
6026 enum x86_intercept_stage stage)
6028 struct vcpu_svm *svm = to_svm(vcpu);
6029 int vmexit, ret = X86EMUL_CONTINUE;
6030 struct __x86_intercept icpt_info;
6031 struct vmcb *vmcb = svm->vmcb;
6033 if (info->intercept >= ARRAY_SIZE(x86_intercept_map))
6036 icpt_info = x86_intercept_map[info->intercept];
6038 if (stage != icpt_info.stage)
6041 switch (icpt_info.exit_code) {
6042 case SVM_EXIT_READ_CR0:
6043 if (info->intercept == x86_intercept_cr_read)
6044 icpt_info.exit_code += info->modrm_reg;
6046 case SVM_EXIT_WRITE_CR0: {
6047 unsigned long cr0, val;
6050 if (info->intercept == x86_intercept_cr_write)
6051 icpt_info.exit_code += info->modrm_reg;
6053 if (icpt_info.exit_code != SVM_EXIT_WRITE_CR0 ||
6054 info->intercept == x86_intercept_clts)
6057 intercept = svm->nested.intercept;
6059 if (!(intercept & (1ULL << INTERCEPT_SELECTIVE_CR0)))
6062 cr0 = vcpu->arch.cr0 & ~SVM_CR0_SELECTIVE_MASK;
6063 val = info->src_val & ~SVM_CR0_SELECTIVE_MASK;
6065 if (info->intercept == x86_intercept_lmsw) {
6068 /* lmsw can't clear PE - catch this here */
6069 if (cr0 & X86_CR0_PE)
6074 icpt_info.exit_code = SVM_EXIT_CR0_SEL_WRITE;
6078 case SVM_EXIT_READ_DR0:
6079 case SVM_EXIT_WRITE_DR0:
6080 icpt_info.exit_code += info->modrm_reg;
6083 if (info->intercept == x86_intercept_wrmsr)
6084 vmcb->control.exit_info_1 = 1;
6086 vmcb->control.exit_info_1 = 0;
6088 case SVM_EXIT_PAUSE:
6090 * We get this for NOP only, but pause
6091 * is rep not, check this here
6093 if (info->rep_prefix != REPE_PREFIX)
6096 case SVM_EXIT_IOIO: {
6100 if (info->intercept == x86_intercept_in ||
6101 info->intercept == x86_intercept_ins) {
6102 exit_info = ((info->src_val & 0xffff) << 16) |
6104 bytes = info->dst_bytes;
6106 exit_info = (info->dst_val & 0xffff) << 16;
6107 bytes = info->src_bytes;
6110 if (info->intercept == x86_intercept_outs ||
6111 info->intercept == x86_intercept_ins)
6112 exit_info |= SVM_IOIO_STR_MASK;
6114 if (info->rep_prefix)
6115 exit_info |= SVM_IOIO_REP_MASK;
6117 bytes = min(bytes, 4u);
6119 exit_info |= bytes << SVM_IOIO_SIZE_SHIFT;
6121 exit_info |= (u32)info->ad_bytes << (SVM_IOIO_ASIZE_SHIFT - 1);
6123 vmcb->control.exit_info_1 = exit_info;
6124 vmcb->control.exit_info_2 = info->next_rip;
6132 /* TODO: Advertise NRIPS to guest hypervisor unconditionally */
6133 if (static_cpu_has(X86_FEATURE_NRIPS))
6134 vmcb->control.next_rip = info->next_rip;
6135 vmcb->control.exit_code = icpt_info.exit_code;
6136 vmexit = nested_svm_exit_handled(svm);
6138 ret = (vmexit == NESTED_EXIT_DONE) ? X86EMUL_INTERCEPTED
6145 static void svm_handle_external_intr(struct kvm_vcpu *vcpu)
6149 * We must have an instruction with interrupts enabled, so
6150 * the timer interrupt isn't delayed by the interrupt shadow.
6153 local_irq_disable();
6156 static void svm_sched_in(struct kvm_vcpu *vcpu, int cpu)
6158 if (pause_filter_thresh)
6159 shrink_ple_window(vcpu);
6162 static inline void avic_post_state_restore(struct kvm_vcpu *vcpu)
6164 if (avic_handle_apic_id_update(vcpu) != 0)
6166 if (avic_handle_dfr_update(vcpu) != 0)
6168 avic_handle_ldr_update(vcpu);
6171 static void svm_setup_mce(struct kvm_vcpu *vcpu)
6173 /* [63:9] are reserved. */
6174 vcpu->arch.mcg_cap &= 0x1ff;
6177 static int svm_smi_allowed(struct kvm_vcpu *vcpu)
6179 struct vcpu_svm *svm = to_svm(vcpu);
6181 /* Per APM Vol.2 15.22.2 "Response to SMI" */
6185 if (is_guest_mode(&svm->vcpu) &&
6186 svm->nested.intercept & (1ULL << INTERCEPT_SMI)) {
6187 /* TODO: Might need to set exit_info_1 and exit_info_2 here */
6188 svm->vmcb->control.exit_code = SVM_EXIT_SMI;
6189 svm->nested.exit_required = true;
6196 static int svm_pre_enter_smm(struct kvm_vcpu *vcpu, char *smstate)
6198 struct vcpu_svm *svm = to_svm(vcpu);
6201 if (is_guest_mode(vcpu)) {
6202 /* FED8h - SVM Guest */
6203 put_smstate(u64, smstate, 0x7ed8, 1);
6204 /* FEE0h - SVM Guest VMCB Physical Address */
6205 put_smstate(u64, smstate, 0x7ee0, svm->nested.vmcb);
6207 svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX];
6208 svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP];
6209 svm->vmcb->save.rip = vcpu->arch.regs[VCPU_REGS_RIP];
6211 ret = nested_svm_vmexit(svm);
6218 static int svm_pre_leave_smm(struct kvm_vcpu *vcpu, u64 smbase)
6220 struct vcpu_svm *svm = to_svm(vcpu);
6221 struct vmcb *nested_vmcb;
6229 ret = kvm_vcpu_read_guest(vcpu, smbase + 0xfed8, &svm_state_save,
6230 sizeof(svm_state_save));
6234 if (svm_state_save.guest) {
6235 vcpu->arch.hflags &= ~HF_SMM_MASK;
6236 nested_vmcb = nested_svm_map(svm, svm_state_save.vmcb, &page);
6238 enter_svm_guest_mode(svm, svm_state_save.vmcb, nested_vmcb, page);
6241 vcpu->arch.hflags |= HF_SMM_MASK;
6246 static int enable_smi_window(struct kvm_vcpu *vcpu)
6248 struct vcpu_svm *svm = to_svm(vcpu);
6250 if (!gif_set(svm)) {
6251 if (vgif_enabled(svm))
6252 set_intercept(svm, INTERCEPT_STGI);
6253 /* STGI will cause a vm exit */
6259 static int sev_asid_new(void)
6264 * SEV-enabled guest must use asid from min_sev_asid to max_sev_asid.
6266 pos = find_next_zero_bit(sev_asid_bitmap, max_sev_asid, min_sev_asid - 1);
6267 if (pos >= max_sev_asid)
6270 set_bit(pos, sev_asid_bitmap);
6274 static int sev_guest_init(struct kvm *kvm, struct kvm_sev_cmd *argp)
6276 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
6280 if (unlikely(sev->active))
6283 asid = sev_asid_new();
6287 ret = sev_platform_init(&argp->error);
6293 INIT_LIST_HEAD(&sev->regions_list);
6298 __sev_asid_free(asid);
6302 static int sev_bind_asid(struct kvm *kvm, unsigned int handle, int *error)
6304 struct sev_data_activate *data;
6305 int asid = sev_get_asid(kvm);
6308 wbinvd_on_all_cpus();
6310 ret = sev_guest_df_flush(error);
6314 data = kzalloc(sizeof(*data), GFP_KERNEL);
6318 /* activate ASID on the given handle */
6319 data->handle = handle;
6321 ret = sev_guest_activate(data, error);
6327 static int __sev_issue_cmd(int fd, int id, void *data, int *error)
6336 ret = sev_issue_cmd_external_user(f.file, id, data, error);
6342 static int sev_issue_cmd(struct kvm *kvm, int id, void *data, int *error)
6344 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
6346 return __sev_issue_cmd(sev->fd, id, data, error);
6349 static int sev_launch_start(struct kvm *kvm, struct kvm_sev_cmd *argp)
6351 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
6352 struct sev_data_launch_start *start;
6353 struct kvm_sev_launch_start params;
6354 void *dh_blob, *session_blob;
6355 int *error = &argp->error;
6358 if (!sev_guest(kvm))
6361 if (copy_from_user(¶ms, (void __user *)(uintptr_t)argp->data, sizeof(params)))
6364 start = kzalloc(sizeof(*start), GFP_KERNEL);
6369 if (params.dh_uaddr) {
6370 dh_blob = psp_copy_user_blob(params.dh_uaddr, params.dh_len);
6371 if (IS_ERR(dh_blob)) {
6372 ret = PTR_ERR(dh_blob);
6376 start->dh_cert_address = __sme_set(__pa(dh_blob));
6377 start->dh_cert_len = params.dh_len;
6380 session_blob = NULL;
6381 if (params.session_uaddr) {
6382 session_blob = psp_copy_user_blob(params.session_uaddr, params.session_len);
6383 if (IS_ERR(session_blob)) {
6384 ret = PTR_ERR(session_blob);
6388 start->session_address = __sme_set(__pa(session_blob));
6389 start->session_len = params.session_len;
6392 start->handle = params.handle;
6393 start->policy = params.policy;
6395 /* create memory encryption context */
6396 ret = __sev_issue_cmd(argp->sev_fd, SEV_CMD_LAUNCH_START, start, error);
6398 goto e_free_session;
6400 /* Bind ASID to this guest */
6401 ret = sev_bind_asid(kvm, start->handle, error);
6403 goto e_free_session;
6405 /* return handle to userspace */
6406 params.handle = start->handle;
6407 if (copy_to_user((void __user *)(uintptr_t)argp->data, ¶ms, sizeof(params))) {
6408 sev_unbind_asid(kvm, start->handle);
6410 goto e_free_session;
6413 sev->handle = start->handle;
6414 sev->fd = argp->sev_fd;
6417 kfree(session_blob);
6425 static int get_num_contig_pages(int idx, struct page **inpages,
6426 unsigned long npages)
6428 unsigned long paddr, next_paddr;
6429 int i = idx + 1, pages = 1;
6431 /* find the number of contiguous pages starting from idx */
6432 paddr = __sme_page_pa(inpages[idx]);
6433 while (i < npages) {
6434 next_paddr = __sme_page_pa(inpages[i++]);
6435 if ((paddr + PAGE_SIZE) == next_paddr) {
6446 static int sev_launch_update_data(struct kvm *kvm, struct kvm_sev_cmd *argp)
6448 unsigned long vaddr, vaddr_end, next_vaddr, npages, size;
6449 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
6450 struct kvm_sev_launch_update_data params;
6451 struct sev_data_launch_update_data *data;
6452 struct page **inpages;
6455 if (!sev_guest(kvm))
6458 if (copy_from_user(¶ms, (void __user *)(uintptr_t)argp->data, sizeof(params)))
6461 data = kzalloc(sizeof(*data), GFP_KERNEL);
6465 vaddr = params.uaddr;
6467 vaddr_end = vaddr + size;
6469 /* Lock the user memory. */
6470 inpages = sev_pin_memory(kvm, vaddr, size, &npages, 1);
6477 * The LAUNCH_UPDATE command will perform in-place encryption of the
6478 * memory content (i.e it will write the same memory region with C=1).
6479 * It's possible that the cache may contain the data with C=0, i.e.,
6480 * unencrypted so invalidate it first.
6482 sev_clflush_pages(inpages, npages);
6484 for (i = 0; vaddr < vaddr_end; vaddr = next_vaddr, i += pages) {
6488 * If the user buffer is not page-aligned, calculate the offset
6491 offset = vaddr & (PAGE_SIZE - 1);
6493 /* Calculate the number of pages that can be encrypted in one go. */
6494 pages = get_num_contig_pages(i, inpages, npages);
6496 len = min_t(size_t, ((pages * PAGE_SIZE) - offset), size);
6498 data->handle = sev->handle;
6500 data->address = __sme_page_pa(inpages[i]) + offset;
6501 ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_UPDATE_DATA, data, &argp->error);
6506 next_vaddr = vaddr + len;
6510 /* content of memory is updated, mark pages dirty */
6511 for (i = 0; i < npages; i++) {
6512 set_page_dirty_lock(inpages[i]);
6513 mark_page_accessed(inpages[i]);
6515 /* unlock the user pages */
6516 sev_unpin_memory(kvm, inpages, npages);
6522 static int sev_launch_measure(struct kvm *kvm, struct kvm_sev_cmd *argp)
6524 void __user *measure = (void __user *)(uintptr_t)argp->data;
6525 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
6526 struct sev_data_launch_measure *data;
6527 struct kvm_sev_launch_measure params;
6528 void __user *p = NULL;
6532 if (!sev_guest(kvm))
6535 if (copy_from_user(¶ms, measure, sizeof(params)))
6538 data = kzalloc(sizeof(*data), GFP_KERNEL);
6542 /* User wants to query the blob length */
6546 p = (void __user *)(uintptr_t)params.uaddr;
6548 if (params.len > SEV_FW_BLOB_MAX_SIZE) {
6554 blob = kmalloc(params.len, GFP_KERNEL);
6558 data->address = __psp_pa(blob);
6559 data->len = params.len;
6563 data->handle = sev->handle;
6564 ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_MEASURE, data, &argp->error);
6567 * If we query the session length, FW responded with expected data.
6576 if (copy_to_user(p, blob, params.len))
6581 params.len = data->len;
6582 if (copy_to_user(measure, ¶ms, sizeof(params)))
6591 static int sev_launch_finish(struct kvm *kvm, struct kvm_sev_cmd *argp)
6593 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
6594 struct sev_data_launch_finish *data;
6597 if (!sev_guest(kvm))
6600 data = kzalloc(sizeof(*data), GFP_KERNEL);
6604 data->handle = sev->handle;
6605 ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_FINISH, data, &argp->error);
6611 static int sev_guest_status(struct kvm *kvm, struct kvm_sev_cmd *argp)
6613 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
6614 struct kvm_sev_guest_status params;
6615 struct sev_data_guest_status *data;
6618 if (!sev_guest(kvm))
6621 data = kzalloc(sizeof(*data), GFP_KERNEL);
6625 data->handle = sev->handle;
6626 ret = sev_issue_cmd(kvm, SEV_CMD_GUEST_STATUS, data, &argp->error);
6630 params.policy = data->policy;
6631 params.state = data->state;
6632 params.handle = data->handle;
6634 if (copy_to_user((void __user *)(uintptr_t)argp->data, ¶ms, sizeof(params)))
6641 static int __sev_issue_dbg_cmd(struct kvm *kvm, unsigned long src,
6642 unsigned long dst, int size,
6643 int *error, bool enc)
6645 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
6646 struct sev_data_dbg *data;
6649 data = kzalloc(sizeof(*data), GFP_KERNEL);
6653 data->handle = sev->handle;
6654 data->dst_addr = dst;
6655 data->src_addr = src;
6658 ret = sev_issue_cmd(kvm,
6659 enc ? SEV_CMD_DBG_ENCRYPT : SEV_CMD_DBG_DECRYPT,
6665 static int __sev_dbg_decrypt(struct kvm *kvm, unsigned long src_paddr,
6666 unsigned long dst_paddr, int sz, int *err)
6671 * Its safe to read more than we are asked, caller should ensure that
6672 * destination has enough space.
6674 src_paddr = round_down(src_paddr, 16);
6675 offset = src_paddr & 15;
6676 sz = round_up(sz + offset, 16);
6678 return __sev_issue_dbg_cmd(kvm, src_paddr, dst_paddr, sz, err, false);
6681 static int __sev_dbg_decrypt_user(struct kvm *kvm, unsigned long paddr,
6682 unsigned long __user dst_uaddr,
6683 unsigned long dst_paddr,
6686 struct page *tpage = NULL;
6689 /* if inputs are not 16-byte then use intermediate buffer */
6690 if (!IS_ALIGNED(dst_paddr, 16) ||
6691 !IS_ALIGNED(paddr, 16) ||
6692 !IS_ALIGNED(size, 16)) {
6693 tpage = (void *)alloc_page(GFP_KERNEL);
6697 dst_paddr = __sme_page_pa(tpage);
6700 ret = __sev_dbg_decrypt(kvm, paddr, dst_paddr, size, err);
6705 offset = paddr & 15;
6706 if (copy_to_user((void __user *)(uintptr_t)dst_uaddr,
6707 page_address(tpage) + offset, size))
6718 static int __sev_dbg_encrypt_user(struct kvm *kvm, unsigned long paddr,
6719 unsigned long __user vaddr,
6720 unsigned long dst_paddr,
6721 unsigned long __user dst_vaddr,
6722 int size, int *error)
6724 struct page *src_tpage = NULL;
6725 struct page *dst_tpage = NULL;
6726 int ret, len = size;
6728 /* If source buffer is not aligned then use an intermediate buffer */
6729 if (!IS_ALIGNED(vaddr, 16)) {
6730 src_tpage = alloc_page(GFP_KERNEL);
6734 if (copy_from_user(page_address(src_tpage),
6735 (void __user *)(uintptr_t)vaddr, size)) {
6736 __free_page(src_tpage);
6740 paddr = __sme_page_pa(src_tpage);
6744 * If destination buffer or length is not aligned then do read-modify-write:
6745 * - decrypt destination in an intermediate buffer
6746 * - copy the source buffer in an intermediate buffer
6747 * - use the intermediate buffer as source buffer
6749 if (!IS_ALIGNED(dst_vaddr, 16) || !IS_ALIGNED(size, 16)) {
6752 dst_tpage = alloc_page(GFP_KERNEL);
6758 ret = __sev_dbg_decrypt(kvm, dst_paddr,
6759 __sme_page_pa(dst_tpage), size, error);
6764 * If source is kernel buffer then use memcpy() otherwise
6767 dst_offset = dst_paddr & 15;
6770 memcpy(page_address(dst_tpage) + dst_offset,
6771 page_address(src_tpage), size);
6773 if (copy_from_user(page_address(dst_tpage) + dst_offset,
6774 (void __user *)(uintptr_t)vaddr, size)) {
6780 paddr = __sme_page_pa(dst_tpage);
6781 dst_paddr = round_down(dst_paddr, 16);
6782 len = round_up(size, 16);
6785 ret = __sev_issue_dbg_cmd(kvm, paddr, dst_paddr, len, error, true);
6789 __free_page(src_tpage);
6791 __free_page(dst_tpage);
6795 static int sev_dbg_crypt(struct kvm *kvm, struct kvm_sev_cmd *argp, bool dec)
6797 unsigned long vaddr, vaddr_end, next_vaddr;
6798 unsigned long dst_vaddr;
6799 struct page **src_p, **dst_p;
6800 struct kvm_sev_dbg debug;
6804 if (!sev_guest(kvm))
6807 if (copy_from_user(&debug, (void __user *)(uintptr_t)argp->data, sizeof(debug)))
6810 vaddr = debug.src_uaddr;
6812 vaddr_end = vaddr + size;
6813 dst_vaddr = debug.dst_uaddr;
6815 for (; vaddr < vaddr_end; vaddr = next_vaddr) {
6816 int len, s_off, d_off;
6818 /* lock userspace source and destination page */
6819 src_p = sev_pin_memory(kvm, vaddr & PAGE_MASK, PAGE_SIZE, &n, 0);
6823 dst_p = sev_pin_memory(kvm, dst_vaddr & PAGE_MASK, PAGE_SIZE, &n, 1);
6825 sev_unpin_memory(kvm, src_p, n);
6830 * The DBG_{DE,EN}CRYPT commands will perform {dec,en}cryption of the
6831 * memory content (i.e it will write the same memory region with C=1).
6832 * It's possible that the cache may contain the data with C=0, i.e.,
6833 * unencrypted so invalidate it first.
6835 sev_clflush_pages(src_p, 1);
6836 sev_clflush_pages(dst_p, 1);
6839 * Since user buffer may not be page aligned, calculate the
6840 * offset within the page.
6842 s_off = vaddr & ~PAGE_MASK;
6843 d_off = dst_vaddr & ~PAGE_MASK;
6844 len = min_t(size_t, (PAGE_SIZE - s_off), size);
6847 ret = __sev_dbg_decrypt_user(kvm,
6848 __sme_page_pa(src_p[0]) + s_off,
6850 __sme_page_pa(dst_p[0]) + d_off,
6853 ret = __sev_dbg_encrypt_user(kvm,
6854 __sme_page_pa(src_p[0]) + s_off,
6856 __sme_page_pa(dst_p[0]) + d_off,
6860 sev_unpin_memory(kvm, src_p, 1);
6861 sev_unpin_memory(kvm, dst_p, 1);
6866 next_vaddr = vaddr + len;
6867 dst_vaddr = dst_vaddr + len;
6874 static int sev_launch_secret(struct kvm *kvm, struct kvm_sev_cmd *argp)
6876 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
6877 struct sev_data_launch_secret *data;
6878 struct kvm_sev_launch_secret params;
6879 struct page **pages;
6884 if (!sev_guest(kvm))
6887 if (copy_from_user(¶ms, (void __user *)(uintptr_t)argp->data, sizeof(params)))
6890 pages = sev_pin_memory(kvm, params.guest_uaddr, params.guest_len, &n, 1);
6895 * The secret must be copied into contiguous memory region, lets verify
6896 * that userspace memory pages are contiguous before we issue command.
6898 if (get_num_contig_pages(0, pages, n) != n) {
6900 goto e_unpin_memory;
6904 data = kzalloc(sizeof(*data), GFP_KERNEL);
6906 goto e_unpin_memory;
6908 offset = params.guest_uaddr & (PAGE_SIZE - 1);
6909 data->guest_address = __sme_page_pa(pages[0]) + offset;
6910 data->guest_len = params.guest_len;
6912 blob = psp_copy_user_blob(params.trans_uaddr, params.trans_len);
6914 ret = PTR_ERR(blob);
6918 data->trans_address = __psp_pa(blob);
6919 data->trans_len = params.trans_len;
6921 hdr = psp_copy_user_blob(params.hdr_uaddr, params.hdr_len);
6926 data->hdr_address = __psp_pa(hdr);
6927 data->hdr_len = params.hdr_len;
6929 data->handle = sev->handle;
6930 ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_UPDATE_SECRET, data, &argp->error);
6939 sev_unpin_memory(kvm, pages, n);
6943 static int svm_mem_enc_op(struct kvm *kvm, void __user *argp)
6945 struct kvm_sev_cmd sev_cmd;
6948 if (!svm_sev_enabled())
6951 if (copy_from_user(&sev_cmd, argp, sizeof(struct kvm_sev_cmd)))
6954 mutex_lock(&kvm->lock);
6956 switch (sev_cmd.id) {
6958 r = sev_guest_init(kvm, &sev_cmd);
6960 case KVM_SEV_LAUNCH_START:
6961 r = sev_launch_start(kvm, &sev_cmd);
6963 case KVM_SEV_LAUNCH_UPDATE_DATA:
6964 r = sev_launch_update_data(kvm, &sev_cmd);
6966 case KVM_SEV_LAUNCH_MEASURE:
6967 r = sev_launch_measure(kvm, &sev_cmd);
6969 case KVM_SEV_LAUNCH_FINISH:
6970 r = sev_launch_finish(kvm, &sev_cmd);
6972 case KVM_SEV_GUEST_STATUS:
6973 r = sev_guest_status(kvm, &sev_cmd);
6975 case KVM_SEV_DBG_DECRYPT:
6976 r = sev_dbg_crypt(kvm, &sev_cmd, true);
6978 case KVM_SEV_DBG_ENCRYPT:
6979 r = sev_dbg_crypt(kvm, &sev_cmd, false);
6981 case KVM_SEV_LAUNCH_SECRET:
6982 r = sev_launch_secret(kvm, &sev_cmd);
6989 if (copy_to_user(argp, &sev_cmd, sizeof(struct kvm_sev_cmd)))
6993 mutex_unlock(&kvm->lock);
6997 static int svm_register_enc_region(struct kvm *kvm,
6998 struct kvm_enc_region *range)
7000 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
7001 struct enc_region *region;
7004 if (!sev_guest(kvm))
7007 if (range->addr > ULONG_MAX || range->size > ULONG_MAX)
7010 region = kzalloc(sizeof(*region), GFP_KERNEL);
7014 region->pages = sev_pin_memory(kvm, range->addr, range->size, ®ion->npages, 1);
7015 if (!region->pages) {
7021 * The guest may change the memory encryption attribute from C=0 -> C=1
7022 * or vice versa for this memory range. Lets make sure caches are
7023 * flushed to ensure that guest data gets written into memory with
7026 sev_clflush_pages(region->pages, region->npages);
7028 region->uaddr = range->addr;
7029 region->size = range->size;
7031 mutex_lock(&kvm->lock);
7032 list_add_tail(®ion->list, &sev->regions_list);
7033 mutex_unlock(&kvm->lock);
7042 static struct enc_region *
7043 find_enc_region(struct kvm *kvm, struct kvm_enc_region *range)
7045 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
7046 struct list_head *head = &sev->regions_list;
7047 struct enc_region *i;
7049 list_for_each_entry(i, head, list) {
7050 if (i->uaddr == range->addr &&
7051 i->size == range->size)
7059 static int svm_unregister_enc_region(struct kvm *kvm,
7060 struct kvm_enc_region *range)
7062 struct enc_region *region;
7065 mutex_lock(&kvm->lock);
7067 if (!sev_guest(kvm)) {
7072 region = find_enc_region(kvm, range);
7078 __unregister_enc_region_locked(kvm, region);
7080 mutex_unlock(&kvm->lock);
7084 mutex_unlock(&kvm->lock);
7088 static uint16_t nested_get_evmcs_version(struct kvm_vcpu *vcpu)
7094 static int nested_enable_evmcs(struct kvm_vcpu *vcpu,
7095 uint16_t *vmcs_version)
7097 /* Intel-only feature */
7101 static struct kvm_x86_ops svm_x86_ops __ro_after_init = {
7102 .cpu_has_kvm_support = has_svm,
7103 .disabled_by_bios = is_disabled,
7104 .hardware_setup = svm_hardware_setup,
7105 .hardware_unsetup = svm_hardware_unsetup,
7106 .check_processor_compatibility = svm_check_processor_compat,
7107 .hardware_enable = svm_hardware_enable,
7108 .hardware_disable = svm_hardware_disable,
7109 .cpu_has_accelerated_tpr = svm_cpu_has_accelerated_tpr,
7110 .has_emulated_msr = svm_has_emulated_msr,
7112 .vcpu_create = svm_create_vcpu,
7113 .vcpu_free = svm_free_vcpu,
7114 .vcpu_reset = svm_vcpu_reset,
7116 .vm_alloc = svm_vm_alloc,
7117 .vm_free = svm_vm_free,
7118 .vm_init = avic_vm_init,
7119 .vm_destroy = svm_vm_destroy,
7121 .prepare_guest_switch = svm_prepare_guest_switch,
7122 .vcpu_load = svm_vcpu_load,
7123 .vcpu_put = svm_vcpu_put,
7124 .vcpu_blocking = svm_vcpu_blocking,
7125 .vcpu_unblocking = svm_vcpu_unblocking,
7127 .update_bp_intercept = update_bp_intercept,
7128 .get_msr_feature = svm_get_msr_feature,
7129 .get_msr = svm_get_msr,
7130 .set_msr = svm_set_msr,
7131 .get_segment_base = svm_get_segment_base,
7132 .get_segment = svm_get_segment,
7133 .set_segment = svm_set_segment,
7134 .get_cpl = svm_get_cpl,
7135 .get_cs_db_l_bits = kvm_get_cs_db_l_bits,
7136 .decache_cr0_guest_bits = svm_decache_cr0_guest_bits,
7137 .decache_cr3 = svm_decache_cr3,
7138 .decache_cr4_guest_bits = svm_decache_cr4_guest_bits,
7139 .set_cr0 = svm_set_cr0,
7140 .set_cr3 = svm_set_cr3,
7141 .set_cr4 = svm_set_cr4,
7142 .set_efer = svm_set_efer,
7143 .get_idt = svm_get_idt,
7144 .set_idt = svm_set_idt,
7145 .get_gdt = svm_get_gdt,
7146 .set_gdt = svm_set_gdt,
7147 .get_dr6 = svm_get_dr6,
7148 .set_dr6 = svm_set_dr6,
7149 .set_dr7 = svm_set_dr7,
7150 .sync_dirty_debug_regs = svm_sync_dirty_debug_regs,
7151 .cache_reg = svm_cache_reg,
7152 .get_rflags = svm_get_rflags,
7153 .set_rflags = svm_set_rflags,
7155 .tlb_flush = svm_flush_tlb,
7156 .tlb_flush_gva = svm_flush_tlb_gva,
7158 .run = svm_vcpu_run,
7159 .handle_exit = handle_exit,
7160 .skip_emulated_instruction = skip_emulated_instruction,
7161 .set_interrupt_shadow = svm_set_interrupt_shadow,
7162 .get_interrupt_shadow = svm_get_interrupt_shadow,
7163 .patch_hypercall = svm_patch_hypercall,
7164 .set_irq = svm_set_irq,
7165 .set_nmi = svm_inject_nmi,
7166 .queue_exception = svm_queue_exception,
7167 .cancel_injection = svm_cancel_injection,
7168 .interrupt_allowed = svm_interrupt_allowed,
7169 .nmi_allowed = svm_nmi_allowed,
7170 .get_nmi_mask = svm_get_nmi_mask,
7171 .set_nmi_mask = svm_set_nmi_mask,
7172 .enable_nmi_window = enable_nmi_window,
7173 .enable_irq_window = enable_irq_window,
7174 .update_cr8_intercept = update_cr8_intercept,
7175 .set_virtual_apic_mode = svm_set_virtual_apic_mode,
7176 .get_enable_apicv = svm_get_enable_apicv,
7177 .refresh_apicv_exec_ctrl = svm_refresh_apicv_exec_ctrl,
7178 .load_eoi_exitmap = svm_load_eoi_exitmap,
7179 .hwapic_irr_update = svm_hwapic_irr_update,
7180 .hwapic_isr_update = svm_hwapic_isr_update,
7181 .sync_pir_to_irr = kvm_lapic_find_highest_irr,
7182 .apicv_post_state_restore = avic_post_state_restore,
7184 .set_tss_addr = svm_set_tss_addr,
7185 .set_identity_map_addr = svm_set_identity_map_addr,
7186 .get_tdp_level = get_npt_level,
7187 .get_mt_mask = svm_get_mt_mask,
7189 .get_exit_info = svm_get_exit_info,
7191 .get_lpage_level = svm_get_lpage_level,
7193 .cpuid_update = svm_cpuid_update,
7195 .rdtscp_supported = svm_rdtscp_supported,
7196 .invpcid_supported = svm_invpcid_supported,
7197 .mpx_supported = svm_mpx_supported,
7198 .xsaves_supported = svm_xsaves_supported,
7199 .umip_emulated = svm_umip_emulated,
7200 .pt_supported = svm_pt_supported,
7202 .set_supported_cpuid = svm_set_supported_cpuid,
7204 .has_wbinvd_exit = svm_has_wbinvd_exit,
7206 .read_l1_tsc_offset = svm_read_l1_tsc_offset,
7207 .write_l1_tsc_offset = svm_write_l1_tsc_offset,
7209 .set_tdp_cr3 = set_tdp_cr3,
7211 .check_intercept = svm_check_intercept,
7212 .handle_external_intr = svm_handle_external_intr,
7214 .request_immediate_exit = __kvm_request_immediate_exit,
7216 .sched_in = svm_sched_in,
7218 .pmu_ops = &amd_pmu_ops,
7219 .deliver_posted_interrupt = svm_deliver_avic_intr,
7220 .update_pi_irte = svm_update_pi_irte,
7221 .setup_mce = svm_setup_mce,
7223 .smi_allowed = svm_smi_allowed,
7224 .pre_enter_smm = svm_pre_enter_smm,
7225 .pre_leave_smm = svm_pre_leave_smm,
7226 .enable_smi_window = enable_smi_window,
7228 .mem_enc_op = svm_mem_enc_op,
7229 .mem_enc_reg_region = svm_register_enc_region,
7230 .mem_enc_unreg_region = svm_unregister_enc_region,
7232 .nested_enable_evmcs = nested_enable_evmcs,
7233 .nested_get_evmcs_version = nested_get_evmcs_version,
7236 static int __init svm_init(void)
7238 return kvm_init(&svm_x86_ops, sizeof(struct vcpu_svm),
7239 __alignof__(struct vcpu_svm), THIS_MODULE);
7242 static void __exit svm_exit(void)
7247 module_init(svm_init)
7248 module_exit(svm_exit)