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)
442 static inline bool sev_guest(struct kvm *kvm)
444 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
449 static inline int sev_get_asid(struct kvm *kvm)
451 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
456 static inline void mark_all_dirty(struct vmcb *vmcb)
458 vmcb->control.clean = 0;
461 static inline void mark_all_clean(struct vmcb *vmcb)
463 vmcb->control.clean = ((1 << VMCB_DIRTY_MAX) - 1)
464 & ~VMCB_ALWAYS_DIRTY_MASK;
467 static inline void mark_dirty(struct vmcb *vmcb, int bit)
469 vmcb->control.clean &= ~(1 << bit);
472 static inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu)
474 return container_of(vcpu, struct vcpu_svm, vcpu);
477 static inline void avic_update_vapic_bar(struct vcpu_svm *svm, u64 data)
479 svm->vmcb->control.avic_vapic_bar = data & VMCB_AVIC_APIC_BAR_MASK;
480 mark_dirty(svm->vmcb, VMCB_AVIC);
483 static inline bool avic_vcpu_is_running(struct kvm_vcpu *vcpu)
485 struct vcpu_svm *svm = to_svm(vcpu);
486 u64 *entry = svm->avic_physical_id_cache;
491 return (READ_ONCE(*entry) & AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK);
494 static void recalc_intercepts(struct vcpu_svm *svm)
496 struct vmcb_control_area *c, *h;
497 struct nested_state *g;
499 mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
501 if (!is_guest_mode(&svm->vcpu))
504 c = &svm->vmcb->control;
505 h = &svm->nested.hsave->control;
508 c->intercept_cr = h->intercept_cr | g->intercept_cr;
509 c->intercept_dr = h->intercept_dr | g->intercept_dr;
510 c->intercept_exceptions = h->intercept_exceptions | g->intercept_exceptions;
511 c->intercept = h->intercept | g->intercept;
514 static inline struct vmcb *get_host_vmcb(struct vcpu_svm *svm)
516 if (is_guest_mode(&svm->vcpu))
517 return svm->nested.hsave;
522 static inline void set_cr_intercept(struct vcpu_svm *svm, int bit)
524 struct vmcb *vmcb = get_host_vmcb(svm);
526 vmcb->control.intercept_cr |= (1U << bit);
528 recalc_intercepts(svm);
531 static inline void clr_cr_intercept(struct vcpu_svm *svm, int bit)
533 struct vmcb *vmcb = get_host_vmcb(svm);
535 vmcb->control.intercept_cr &= ~(1U << bit);
537 recalc_intercepts(svm);
540 static inline bool is_cr_intercept(struct vcpu_svm *svm, int bit)
542 struct vmcb *vmcb = get_host_vmcb(svm);
544 return vmcb->control.intercept_cr & (1U << bit);
547 static inline void set_dr_intercepts(struct vcpu_svm *svm)
549 struct vmcb *vmcb = get_host_vmcb(svm);
551 vmcb->control.intercept_dr = (1 << INTERCEPT_DR0_READ)
552 | (1 << INTERCEPT_DR1_READ)
553 | (1 << INTERCEPT_DR2_READ)
554 | (1 << INTERCEPT_DR3_READ)
555 | (1 << INTERCEPT_DR4_READ)
556 | (1 << INTERCEPT_DR5_READ)
557 | (1 << INTERCEPT_DR6_READ)
558 | (1 << INTERCEPT_DR7_READ)
559 | (1 << INTERCEPT_DR0_WRITE)
560 | (1 << INTERCEPT_DR1_WRITE)
561 | (1 << INTERCEPT_DR2_WRITE)
562 | (1 << INTERCEPT_DR3_WRITE)
563 | (1 << INTERCEPT_DR4_WRITE)
564 | (1 << INTERCEPT_DR5_WRITE)
565 | (1 << INTERCEPT_DR6_WRITE)
566 | (1 << INTERCEPT_DR7_WRITE);
568 recalc_intercepts(svm);
571 static inline void clr_dr_intercepts(struct vcpu_svm *svm)
573 struct vmcb *vmcb = get_host_vmcb(svm);
575 vmcb->control.intercept_dr = 0;
577 recalc_intercepts(svm);
580 static inline void set_exception_intercept(struct vcpu_svm *svm, int bit)
582 struct vmcb *vmcb = get_host_vmcb(svm);
584 vmcb->control.intercept_exceptions |= (1U << bit);
586 recalc_intercepts(svm);
589 static inline void clr_exception_intercept(struct vcpu_svm *svm, int bit)
591 struct vmcb *vmcb = get_host_vmcb(svm);
593 vmcb->control.intercept_exceptions &= ~(1U << bit);
595 recalc_intercepts(svm);
598 static inline void set_intercept(struct vcpu_svm *svm, int bit)
600 struct vmcb *vmcb = get_host_vmcb(svm);
602 vmcb->control.intercept |= (1ULL << bit);
604 recalc_intercepts(svm);
607 static inline void clr_intercept(struct vcpu_svm *svm, int bit)
609 struct vmcb *vmcb = get_host_vmcb(svm);
611 vmcb->control.intercept &= ~(1ULL << bit);
613 recalc_intercepts(svm);
616 static inline bool vgif_enabled(struct vcpu_svm *svm)
618 return !!(svm->vmcb->control.int_ctl & V_GIF_ENABLE_MASK);
621 static inline void enable_gif(struct vcpu_svm *svm)
623 if (vgif_enabled(svm))
624 svm->vmcb->control.int_ctl |= V_GIF_MASK;
626 svm->vcpu.arch.hflags |= HF_GIF_MASK;
629 static inline void disable_gif(struct vcpu_svm *svm)
631 if (vgif_enabled(svm))
632 svm->vmcb->control.int_ctl &= ~V_GIF_MASK;
634 svm->vcpu.arch.hflags &= ~HF_GIF_MASK;
637 static inline bool gif_set(struct vcpu_svm *svm)
639 if (vgif_enabled(svm))
640 return !!(svm->vmcb->control.int_ctl & V_GIF_MASK);
642 return !!(svm->vcpu.arch.hflags & HF_GIF_MASK);
645 static unsigned long iopm_base;
647 struct kvm_ldttss_desc {
650 unsigned base1:8, type:5, dpl:2, p:1;
651 unsigned limit1:4, zero0:3, g:1, base2:8;
654 } __attribute__((packed));
656 struct svm_cpu_data {
663 struct kvm_ldttss_desc *tss_desc;
665 struct page *save_area;
666 struct vmcb *current_vmcb;
668 /* index = sev_asid, value = vmcb pointer */
669 struct vmcb **sev_vmcbs;
672 static DEFINE_PER_CPU(struct svm_cpu_data *, svm_data);
674 struct svm_init_data {
679 static const u32 msrpm_ranges[] = {0, 0xc0000000, 0xc0010000};
681 #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
682 #define MSRS_RANGE_SIZE 2048
683 #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
685 static u32 svm_msrpm_offset(u32 msr)
690 for (i = 0; i < NUM_MSR_MAPS; i++) {
691 if (msr < msrpm_ranges[i] ||
692 msr >= msrpm_ranges[i] + MSRS_IN_RANGE)
695 offset = (msr - msrpm_ranges[i]) / 4; /* 4 msrs per u8 */
696 offset += (i * MSRS_RANGE_SIZE); /* add range offset */
698 /* Now we have the u8 offset - but need the u32 offset */
702 /* MSR not in any range */
706 #define MAX_INST_SIZE 15
708 static inline void clgi(void)
710 asm volatile (__ex(SVM_CLGI));
713 static inline void stgi(void)
715 asm volatile (__ex(SVM_STGI));
718 static inline void invlpga(unsigned long addr, u32 asid)
720 asm volatile (__ex(SVM_INVLPGA) : : "a"(addr), "c"(asid));
723 static int get_npt_level(struct kvm_vcpu *vcpu)
726 return PT64_ROOT_4LEVEL;
728 return PT32E_ROOT_LEVEL;
732 static void svm_set_efer(struct kvm_vcpu *vcpu, u64 efer)
734 vcpu->arch.efer = efer;
735 if (!npt_enabled && !(efer & EFER_LMA))
738 to_svm(vcpu)->vmcb->save.efer = efer | EFER_SVME;
739 mark_dirty(to_svm(vcpu)->vmcb, VMCB_CR);
742 static int is_external_interrupt(u32 info)
744 info &= SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
745 return info == (SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR);
748 static u32 svm_get_interrupt_shadow(struct kvm_vcpu *vcpu)
750 struct vcpu_svm *svm = to_svm(vcpu);
753 if (svm->vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK)
754 ret = KVM_X86_SHADOW_INT_STI | KVM_X86_SHADOW_INT_MOV_SS;
758 static void svm_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
760 struct vcpu_svm *svm = to_svm(vcpu);
763 svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK;
765 svm->vmcb->control.int_state |= SVM_INTERRUPT_SHADOW_MASK;
769 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
771 struct vcpu_svm *svm = to_svm(vcpu);
773 if (svm->vmcb->control.next_rip != 0) {
774 WARN_ON_ONCE(!static_cpu_has(X86_FEATURE_NRIPS));
775 svm->next_rip = svm->vmcb->control.next_rip;
778 if (!svm->next_rip) {
779 if (emulate_instruction(vcpu, EMULTYPE_SKIP) !=
781 printk(KERN_DEBUG "%s: NOP\n", __func__);
784 if (svm->next_rip - kvm_rip_read(vcpu) > MAX_INST_SIZE)
785 printk(KERN_ERR "%s: ip 0x%lx next 0x%llx\n",
786 __func__, kvm_rip_read(vcpu), svm->next_rip);
788 kvm_rip_write(vcpu, svm->next_rip);
789 svm_set_interrupt_shadow(vcpu, 0);
792 static void svm_queue_exception(struct kvm_vcpu *vcpu)
794 struct vcpu_svm *svm = to_svm(vcpu);
795 unsigned nr = vcpu->arch.exception.nr;
796 bool has_error_code = vcpu->arch.exception.has_error_code;
797 bool reinject = vcpu->arch.exception.injected;
798 u32 error_code = vcpu->arch.exception.error_code;
801 * If we are within a nested VM we'd better #VMEXIT and let the guest
802 * handle the exception
805 nested_svm_check_exception(svm, nr, has_error_code, error_code))
808 if (nr == BP_VECTOR && !static_cpu_has(X86_FEATURE_NRIPS)) {
809 unsigned long rip, old_rip = kvm_rip_read(&svm->vcpu);
812 * For guest debugging where we have to reinject #BP if some
813 * INT3 is guest-owned:
814 * Emulate nRIP by moving RIP forward. Will fail if injection
815 * raises a fault that is not intercepted. Still better than
816 * failing in all cases.
818 skip_emulated_instruction(&svm->vcpu);
819 rip = kvm_rip_read(&svm->vcpu);
820 svm->int3_rip = rip + svm->vmcb->save.cs.base;
821 svm->int3_injected = rip - old_rip;
824 svm->vmcb->control.event_inj = nr
826 | (has_error_code ? SVM_EVTINJ_VALID_ERR : 0)
827 | SVM_EVTINJ_TYPE_EXEPT;
828 svm->vmcb->control.event_inj_err = error_code;
831 static void svm_init_erratum_383(void)
837 if (!static_cpu_has_bug(X86_BUG_AMD_TLB_MMATCH))
840 /* Use _safe variants to not break nested virtualization */
841 val = native_read_msr_safe(MSR_AMD64_DC_CFG, &err);
847 low = lower_32_bits(val);
848 high = upper_32_bits(val);
850 native_write_msr_safe(MSR_AMD64_DC_CFG, low, high);
852 erratum_383_found = true;
855 static void svm_init_osvw(struct kvm_vcpu *vcpu)
858 * Guests should see errata 400 and 415 as fixed (assuming that
859 * HLT and IO instructions are intercepted).
861 vcpu->arch.osvw.length = (osvw_len >= 3) ? (osvw_len) : 3;
862 vcpu->arch.osvw.status = osvw_status & ~(6ULL);
865 * By increasing VCPU's osvw.length to 3 we are telling the guest that
866 * all osvw.status bits inside that length, including bit 0 (which is
867 * reserved for erratum 298), are valid. However, if host processor's
868 * osvw_len is 0 then osvw_status[0] carries no information. We need to
869 * be conservative here and therefore we tell the guest that erratum 298
870 * is present (because we really don't know).
872 if (osvw_len == 0 && boot_cpu_data.x86 == 0x10)
873 vcpu->arch.osvw.status |= 1;
876 static int has_svm(void)
880 if (!cpu_has_svm(&msg)) {
881 printk(KERN_INFO "has_svm: %s\n", msg);
888 static void svm_hardware_disable(void)
890 /* Make sure we clean up behind us */
891 if (static_cpu_has(X86_FEATURE_TSCRATEMSR))
892 wrmsrl(MSR_AMD64_TSC_RATIO, TSC_RATIO_DEFAULT);
896 amd_pmu_disable_virt();
899 static int svm_hardware_enable(void)
902 struct svm_cpu_data *sd;
904 struct desc_struct *gdt;
905 int me = raw_smp_processor_id();
907 rdmsrl(MSR_EFER, efer);
908 if (efer & EFER_SVME)
912 pr_err("%s: err EOPNOTSUPP on %d\n", __func__, me);
915 sd = per_cpu(svm_data, me);
917 pr_err("%s: svm_data is NULL on %d\n", __func__, me);
921 sd->asid_generation = 1;
922 sd->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1;
923 sd->next_asid = sd->max_asid + 1;
924 sd->min_asid = max_sev_asid + 1;
926 gdt = get_current_gdt_rw();
927 sd->tss_desc = (struct kvm_ldttss_desc *)(gdt + GDT_ENTRY_TSS);
929 wrmsrl(MSR_EFER, efer | EFER_SVME);
931 wrmsrl(MSR_VM_HSAVE_PA, page_to_pfn(sd->save_area) << PAGE_SHIFT);
933 if (static_cpu_has(X86_FEATURE_TSCRATEMSR)) {
934 wrmsrl(MSR_AMD64_TSC_RATIO, TSC_RATIO_DEFAULT);
935 __this_cpu_write(current_tsc_ratio, TSC_RATIO_DEFAULT);
942 * Note that it is possible to have a system with mixed processor
943 * revisions and therefore different OSVW bits. If bits are not the same
944 * on different processors then choose the worst case (i.e. if erratum
945 * is present on one processor and not on another then assume that the
946 * erratum is present everywhere).
948 if (cpu_has(&boot_cpu_data, X86_FEATURE_OSVW)) {
949 uint64_t len, status = 0;
952 len = native_read_msr_safe(MSR_AMD64_OSVW_ID_LENGTH, &err);
954 status = native_read_msr_safe(MSR_AMD64_OSVW_STATUS,
958 osvw_status = osvw_len = 0;
962 osvw_status |= status;
963 osvw_status &= (1ULL << osvw_len) - 1;
966 osvw_status = osvw_len = 0;
968 svm_init_erratum_383();
970 amd_pmu_enable_virt();
975 static void svm_cpu_uninit(int cpu)
977 struct svm_cpu_data *sd = per_cpu(svm_data, raw_smp_processor_id());
982 per_cpu(svm_data, raw_smp_processor_id()) = NULL;
983 kfree(sd->sev_vmcbs);
984 __free_page(sd->save_area);
988 static int svm_cpu_init(int cpu)
990 struct svm_cpu_data *sd;
993 sd = kzalloc(sizeof(struct svm_cpu_data), GFP_KERNEL);
998 sd->save_area = alloc_page(GFP_KERNEL);
1002 if (svm_sev_enabled()) {
1004 sd->sev_vmcbs = kmalloc((max_sev_asid + 1) * sizeof(void *), GFP_KERNEL);
1009 per_cpu(svm_data, cpu) = sd;
1019 static bool valid_msr_intercept(u32 index)
1023 for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++)
1024 if (direct_access_msrs[i].index == index)
1030 static bool msr_write_intercepted(struct kvm_vcpu *vcpu, unsigned msr)
1037 msrpm = is_guest_mode(vcpu) ? to_svm(vcpu)->nested.msrpm:
1038 to_svm(vcpu)->msrpm;
1040 offset = svm_msrpm_offset(msr);
1041 bit_write = 2 * (msr & 0x0f) + 1;
1042 tmp = msrpm[offset];
1044 BUG_ON(offset == MSR_INVALID);
1046 return !!test_bit(bit_write, &tmp);
1049 static void set_msr_interception(u32 *msrpm, unsigned msr,
1050 int read, int write)
1052 u8 bit_read, bit_write;
1057 * If this warning triggers extend the direct_access_msrs list at the
1058 * beginning of the file
1060 WARN_ON(!valid_msr_intercept(msr));
1062 offset = svm_msrpm_offset(msr);
1063 bit_read = 2 * (msr & 0x0f);
1064 bit_write = 2 * (msr & 0x0f) + 1;
1065 tmp = msrpm[offset];
1067 BUG_ON(offset == MSR_INVALID);
1069 read ? clear_bit(bit_read, &tmp) : set_bit(bit_read, &tmp);
1070 write ? clear_bit(bit_write, &tmp) : set_bit(bit_write, &tmp);
1072 msrpm[offset] = tmp;
1075 static void svm_vcpu_init_msrpm(u32 *msrpm)
1079 memset(msrpm, 0xff, PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER));
1081 for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
1082 if (!direct_access_msrs[i].always)
1085 set_msr_interception(msrpm, direct_access_msrs[i].index, 1, 1);
1089 static void add_msr_offset(u32 offset)
1093 for (i = 0; i < MSRPM_OFFSETS; ++i) {
1095 /* Offset already in list? */
1096 if (msrpm_offsets[i] == offset)
1099 /* Slot used by another offset? */
1100 if (msrpm_offsets[i] != MSR_INVALID)
1103 /* Add offset to list */
1104 msrpm_offsets[i] = offset;
1110 * If this BUG triggers the msrpm_offsets table has an overflow. Just
1111 * increase MSRPM_OFFSETS in this case.
1116 static void init_msrpm_offsets(void)
1120 memset(msrpm_offsets, 0xff, sizeof(msrpm_offsets));
1122 for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
1125 offset = svm_msrpm_offset(direct_access_msrs[i].index);
1126 BUG_ON(offset == MSR_INVALID);
1128 add_msr_offset(offset);
1132 static void svm_enable_lbrv(struct vcpu_svm *svm)
1134 u32 *msrpm = svm->msrpm;
1136 svm->vmcb->control.virt_ext |= LBR_CTL_ENABLE_MASK;
1137 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 1, 1);
1138 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 1, 1);
1139 set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 1, 1);
1140 set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 1, 1);
1143 static void svm_disable_lbrv(struct vcpu_svm *svm)
1145 u32 *msrpm = svm->msrpm;
1147 svm->vmcb->control.virt_ext &= ~LBR_CTL_ENABLE_MASK;
1148 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 0, 0);
1149 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 0, 0);
1150 set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 0, 0);
1151 set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 0, 0);
1154 static void disable_nmi_singlestep(struct vcpu_svm *svm)
1156 svm->nmi_singlestep = false;
1158 if (!(svm->vcpu.guest_debug & KVM_GUESTDBG_SINGLESTEP)) {
1159 /* Clear our flags if they were not set by the guest */
1160 if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_TF))
1161 svm->vmcb->save.rflags &= ~X86_EFLAGS_TF;
1162 if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_RF))
1163 svm->vmcb->save.rflags &= ~X86_EFLAGS_RF;
1168 * This hash table is used to map VM_ID to a struct kvm_svm,
1169 * when handling AMD IOMMU GALOG notification to schedule in
1170 * a particular vCPU.
1172 #define SVM_VM_DATA_HASH_BITS 8
1173 static DEFINE_HASHTABLE(svm_vm_data_hash, SVM_VM_DATA_HASH_BITS);
1174 static u32 next_vm_id = 0;
1175 static bool next_vm_id_wrapped = 0;
1176 static DEFINE_SPINLOCK(svm_vm_data_hash_lock);
1179 * This function is called from IOMMU driver to notify
1180 * SVM to schedule in a particular vCPU of a particular VM.
1182 static int avic_ga_log_notifier(u32 ga_tag)
1184 unsigned long flags;
1185 struct kvm_svm *kvm_svm;
1186 struct kvm_vcpu *vcpu = NULL;
1187 u32 vm_id = AVIC_GATAG_TO_VMID(ga_tag);
1188 u32 vcpu_id = AVIC_GATAG_TO_VCPUID(ga_tag);
1190 pr_debug("SVM: %s: vm_id=%#x, vcpu_id=%#x\n", __func__, vm_id, vcpu_id);
1192 spin_lock_irqsave(&svm_vm_data_hash_lock, flags);
1193 hash_for_each_possible(svm_vm_data_hash, kvm_svm, hnode, vm_id) {
1194 if (kvm_svm->avic_vm_id != vm_id)
1196 vcpu = kvm_get_vcpu_by_id(&kvm_svm->kvm, vcpu_id);
1199 spin_unlock_irqrestore(&svm_vm_data_hash_lock, flags);
1202 * At this point, the IOMMU should have already set the pending
1203 * bit in the vAPIC backing page. So, we just need to schedule
1207 kvm_vcpu_wake_up(vcpu);
1212 static __init int sev_hardware_setup(void)
1214 struct sev_user_data_status *status;
1217 /* Maximum number of encrypted guests supported simultaneously */
1218 max_sev_asid = cpuid_ecx(0x8000001F);
1223 /* Minimum ASID value that should be used for SEV guest */
1224 min_sev_asid = cpuid_edx(0x8000001F);
1226 /* Initialize SEV ASID bitmap */
1227 sev_asid_bitmap = kcalloc(BITS_TO_LONGS(max_sev_asid),
1228 sizeof(unsigned long), GFP_KERNEL);
1229 if (!sev_asid_bitmap)
1232 status = kmalloc(sizeof(*status), GFP_KERNEL);
1237 * Check SEV platform status.
1239 * PLATFORM_STATUS can be called in any state, if we failed to query
1240 * the PLATFORM status then either PSP firmware does not support SEV
1241 * feature or SEV firmware is dead.
1243 rc = sev_platform_status(status, NULL);
1247 pr_info("SEV supported\n");
1254 static void grow_ple_window(struct kvm_vcpu *vcpu)
1256 struct vcpu_svm *svm = to_svm(vcpu);
1257 struct vmcb_control_area *control = &svm->vmcb->control;
1258 int old = control->pause_filter_count;
1260 control->pause_filter_count = __grow_ple_window(old,
1262 pause_filter_count_grow,
1263 pause_filter_count_max);
1265 if (control->pause_filter_count != old)
1266 mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
1268 trace_kvm_ple_window_grow(vcpu->vcpu_id,
1269 control->pause_filter_count, old);
1272 static void shrink_ple_window(struct kvm_vcpu *vcpu)
1274 struct vcpu_svm *svm = to_svm(vcpu);
1275 struct vmcb_control_area *control = &svm->vmcb->control;
1276 int old = control->pause_filter_count;
1278 control->pause_filter_count =
1279 __shrink_ple_window(old,
1281 pause_filter_count_shrink,
1282 pause_filter_count);
1283 if (control->pause_filter_count != old)
1284 mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
1286 trace_kvm_ple_window_shrink(vcpu->vcpu_id,
1287 control->pause_filter_count, old);
1290 static __init int svm_hardware_setup(void)
1293 struct page *iopm_pages;
1297 iopm_pages = alloc_pages(GFP_KERNEL, IOPM_ALLOC_ORDER);
1302 iopm_va = page_address(iopm_pages);
1303 memset(iopm_va, 0xff, PAGE_SIZE * (1 << IOPM_ALLOC_ORDER));
1304 iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT;
1306 init_msrpm_offsets();
1308 if (boot_cpu_has(X86_FEATURE_NX))
1309 kvm_enable_efer_bits(EFER_NX);
1311 if (boot_cpu_has(X86_FEATURE_FXSR_OPT))
1312 kvm_enable_efer_bits(EFER_FFXSR);
1314 if (boot_cpu_has(X86_FEATURE_TSCRATEMSR)) {
1315 kvm_has_tsc_control = true;
1316 kvm_max_tsc_scaling_ratio = TSC_RATIO_MAX;
1317 kvm_tsc_scaling_ratio_frac_bits = 32;
1320 /* Check for pause filtering support */
1321 if (!boot_cpu_has(X86_FEATURE_PAUSEFILTER)) {
1322 pause_filter_count = 0;
1323 pause_filter_thresh = 0;
1324 } else if (!boot_cpu_has(X86_FEATURE_PFTHRESHOLD)) {
1325 pause_filter_thresh = 0;
1329 printk(KERN_INFO "kvm: Nested Virtualization enabled\n");
1330 kvm_enable_efer_bits(EFER_SVME | EFER_LMSLE);
1334 if (boot_cpu_has(X86_FEATURE_SEV) &&
1335 IS_ENABLED(CONFIG_KVM_AMD_SEV)) {
1336 r = sev_hardware_setup();
1344 for_each_possible_cpu(cpu) {
1345 r = svm_cpu_init(cpu);
1350 if (!boot_cpu_has(X86_FEATURE_NPT))
1351 npt_enabled = false;
1353 if (npt_enabled && !npt) {
1354 printk(KERN_INFO "kvm: Nested Paging disabled\n");
1355 npt_enabled = false;
1359 printk(KERN_INFO "kvm: Nested Paging enabled\n");
1366 !boot_cpu_has(X86_FEATURE_AVIC) ||
1367 !IS_ENABLED(CONFIG_X86_LOCAL_APIC)) {
1370 pr_info("AVIC enabled\n");
1372 amd_iommu_register_ga_log_notifier(&avic_ga_log_notifier);
1378 !boot_cpu_has(X86_FEATURE_V_VMSAVE_VMLOAD) ||
1379 !IS_ENABLED(CONFIG_X86_64)) {
1382 pr_info("Virtual VMLOAD VMSAVE supported\n");
1387 if (!boot_cpu_has(X86_FEATURE_VGIF))
1390 pr_info("Virtual GIF supported\n");
1396 __free_pages(iopm_pages, IOPM_ALLOC_ORDER);
1401 static __exit void svm_hardware_unsetup(void)
1405 if (svm_sev_enabled())
1406 kfree(sev_asid_bitmap);
1408 for_each_possible_cpu(cpu)
1409 svm_cpu_uninit(cpu);
1411 __free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT), IOPM_ALLOC_ORDER);
1415 static void init_seg(struct vmcb_seg *seg)
1418 seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK |
1419 SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */
1420 seg->limit = 0xffff;
1424 static void init_sys_seg(struct vmcb_seg *seg, uint32_t type)
1427 seg->attrib = SVM_SELECTOR_P_MASK | type;
1428 seg->limit = 0xffff;
1432 static u64 svm_read_l1_tsc_offset(struct kvm_vcpu *vcpu)
1434 struct vcpu_svm *svm = to_svm(vcpu);
1436 if (is_guest_mode(vcpu))
1437 return svm->nested.hsave->control.tsc_offset;
1439 return vcpu->arch.tsc_offset;
1442 static void svm_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1444 struct vcpu_svm *svm = to_svm(vcpu);
1445 u64 g_tsc_offset = 0;
1447 if (is_guest_mode(vcpu)) {
1448 /* Write L1's TSC offset. */
1449 g_tsc_offset = svm->vmcb->control.tsc_offset -
1450 svm->nested.hsave->control.tsc_offset;
1451 svm->nested.hsave->control.tsc_offset = offset;
1453 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
1454 svm->vmcb->control.tsc_offset,
1457 svm->vmcb->control.tsc_offset = offset + g_tsc_offset;
1459 mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
1462 static void avic_init_vmcb(struct vcpu_svm *svm)
1464 struct vmcb *vmcb = svm->vmcb;
1465 struct kvm_svm *kvm_svm = to_kvm_svm(svm->vcpu.kvm);
1466 phys_addr_t bpa = __sme_set(page_to_phys(svm->avic_backing_page));
1467 phys_addr_t lpa = __sme_set(page_to_phys(kvm_svm->avic_logical_id_table_page));
1468 phys_addr_t ppa = __sme_set(page_to_phys(kvm_svm->avic_physical_id_table_page));
1470 vmcb->control.avic_backing_page = bpa & AVIC_HPA_MASK;
1471 vmcb->control.avic_logical_id = lpa & AVIC_HPA_MASK;
1472 vmcb->control.avic_physical_id = ppa & AVIC_HPA_MASK;
1473 vmcb->control.avic_physical_id |= AVIC_MAX_PHYSICAL_ID_COUNT;
1474 vmcb->control.int_ctl |= AVIC_ENABLE_MASK;
1477 static void init_vmcb(struct vcpu_svm *svm)
1479 struct vmcb_control_area *control = &svm->vmcb->control;
1480 struct vmcb_save_area *save = &svm->vmcb->save;
1482 svm->vcpu.arch.hflags = 0;
1484 set_cr_intercept(svm, INTERCEPT_CR0_READ);
1485 set_cr_intercept(svm, INTERCEPT_CR3_READ);
1486 set_cr_intercept(svm, INTERCEPT_CR4_READ);
1487 set_cr_intercept(svm, INTERCEPT_CR0_WRITE);
1488 set_cr_intercept(svm, INTERCEPT_CR3_WRITE);
1489 set_cr_intercept(svm, INTERCEPT_CR4_WRITE);
1490 if (!kvm_vcpu_apicv_active(&svm->vcpu))
1491 set_cr_intercept(svm, INTERCEPT_CR8_WRITE);
1493 set_dr_intercepts(svm);
1495 set_exception_intercept(svm, PF_VECTOR);
1496 set_exception_intercept(svm, UD_VECTOR);
1497 set_exception_intercept(svm, MC_VECTOR);
1498 set_exception_intercept(svm, AC_VECTOR);
1499 set_exception_intercept(svm, DB_VECTOR);
1501 * Guest access to VMware backdoor ports could legitimately
1502 * trigger #GP because of TSS I/O permission bitmap.
1503 * We intercept those #GP and allow access to them anyway
1506 if (enable_vmware_backdoor)
1507 set_exception_intercept(svm, GP_VECTOR);
1509 set_intercept(svm, INTERCEPT_INTR);
1510 set_intercept(svm, INTERCEPT_NMI);
1511 set_intercept(svm, INTERCEPT_SMI);
1512 set_intercept(svm, INTERCEPT_SELECTIVE_CR0);
1513 set_intercept(svm, INTERCEPT_RDPMC);
1514 set_intercept(svm, INTERCEPT_CPUID);
1515 set_intercept(svm, INTERCEPT_INVD);
1516 set_intercept(svm, INTERCEPT_INVLPG);
1517 set_intercept(svm, INTERCEPT_INVLPGA);
1518 set_intercept(svm, INTERCEPT_IOIO_PROT);
1519 set_intercept(svm, INTERCEPT_MSR_PROT);
1520 set_intercept(svm, INTERCEPT_TASK_SWITCH);
1521 set_intercept(svm, INTERCEPT_SHUTDOWN);
1522 set_intercept(svm, INTERCEPT_VMRUN);
1523 set_intercept(svm, INTERCEPT_VMMCALL);
1524 set_intercept(svm, INTERCEPT_VMLOAD);
1525 set_intercept(svm, INTERCEPT_VMSAVE);
1526 set_intercept(svm, INTERCEPT_STGI);
1527 set_intercept(svm, INTERCEPT_CLGI);
1528 set_intercept(svm, INTERCEPT_SKINIT);
1529 set_intercept(svm, INTERCEPT_WBINVD);
1530 set_intercept(svm, INTERCEPT_XSETBV);
1531 set_intercept(svm, INTERCEPT_RSM);
1533 if (!kvm_mwait_in_guest(svm->vcpu.kvm)) {
1534 set_intercept(svm, INTERCEPT_MONITOR);
1535 set_intercept(svm, INTERCEPT_MWAIT);
1538 if (!kvm_hlt_in_guest(svm->vcpu.kvm))
1539 set_intercept(svm, INTERCEPT_HLT);
1541 control->iopm_base_pa = __sme_set(iopm_base);
1542 control->msrpm_base_pa = __sme_set(__pa(svm->msrpm));
1543 control->int_ctl = V_INTR_MASKING_MASK;
1545 init_seg(&save->es);
1546 init_seg(&save->ss);
1547 init_seg(&save->ds);
1548 init_seg(&save->fs);
1549 init_seg(&save->gs);
1551 save->cs.selector = 0xf000;
1552 save->cs.base = 0xffff0000;
1553 /* Executable/Readable Code Segment */
1554 save->cs.attrib = SVM_SELECTOR_READ_MASK | SVM_SELECTOR_P_MASK |
1555 SVM_SELECTOR_S_MASK | SVM_SELECTOR_CODE_MASK;
1556 save->cs.limit = 0xffff;
1558 save->gdtr.limit = 0xffff;
1559 save->idtr.limit = 0xffff;
1561 init_sys_seg(&save->ldtr, SEG_TYPE_LDT);
1562 init_sys_seg(&save->tr, SEG_TYPE_BUSY_TSS16);
1564 svm_set_efer(&svm->vcpu, 0);
1565 save->dr6 = 0xffff0ff0;
1566 kvm_set_rflags(&svm->vcpu, 2);
1567 save->rip = 0x0000fff0;
1568 svm->vcpu.arch.regs[VCPU_REGS_RIP] = save->rip;
1571 * svm_set_cr0() sets PG and WP and clears NW and CD on save->cr0.
1572 * It also updates the guest-visible cr0 value.
1574 svm_set_cr0(&svm->vcpu, X86_CR0_NW | X86_CR0_CD | X86_CR0_ET);
1575 kvm_mmu_reset_context(&svm->vcpu);
1577 save->cr4 = X86_CR4_PAE;
1581 /* Setup VMCB for Nested Paging */
1582 control->nested_ctl |= SVM_NESTED_CTL_NP_ENABLE;
1583 clr_intercept(svm, INTERCEPT_INVLPG);
1584 clr_exception_intercept(svm, PF_VECTOR);
1585 clr_cr_intercept(svm, INTERCEPT_CR3_READ);
1586 clr_cr_intercept(svm, INTERCEPT_CR3_WRITE);
1587 save->g_pat = svm->vcpu.arch.pat;
1591 svm->asid_generation = 0;
1593 svm->nested.vmcb = 0;
1594 svm->vcpu.arch.hflags = 0;
1596 if (pause_filter_count) {
1597 control->pause_filter_count = pause_filter_count;
1598 if (pause_filter_thresh)
1599 control->pause_filter_thresh = pause_filter_thresh;
1600 set_intercept(svm, INTERCEPT_PAUSE);
1602 clr_intercept(svm, INTERCEPT_PAUSE);
1605 if (kvm_vcpu_apicv_active(&svm->vcpu))
1606 avic_init_vmcb(svm);
1609 * If hardware supports Virtual VMLOAD VMSAVE then enable it
1610 * in VMCB and clear intercepts to avoid #VMEXIT.
1613 clr_intercept(svm, INTERCEPT_VMLOAD);
1614 clr_intercept(svm, INTERCEPT_VMSAVE);
1615 svm->vmcb->control.virt_ext |= VIRTUAL_VMLOAD_VMSAVE_ENABLE_MASK;
1619 clr_intercept(svm, INTERCEPT_STGI);
1620 clr_intercept(svm, INTERCEPT_CLGI);
1621 svm->vmcb->control.int_ctl |= V_GIF_ENABLE_MASK;
1624 if (sev_guest(svm->vcpu.kvm)) {
1625 svm->vmcb->control.nested_ctl |= SVM_NESTED_CTL_SEV_ENABLE;
1626 clr_exception_intercept(svm, UD_VECTOR);
1629 mark_all_dirty(svm->vmcb);
1635 static u64 *avic_get_physical_id_entry(struct kvm_vcpu *vcpu,
1638 u64 *avic_physical_id_table;
1639 struct kvm_svm *kvm_svm = to_kvm_svm(vcpu->kvm);
1641 if (index >= AVIC_MAX_PHYSICAL_ID_COUNT)
1644 avic_physical_id_table = page_address(kvm_svm->avic_physical_id_table_page);
1646 return &avic_physical_id_table[index];
1651 * AVIC hardware walks the nested page table to check permissions,
1652 * but does not use the SPA address specified in the leaf page
1653 * table entry since it uses address in the AVIC_BACKING_PAGE pointer
1654 * field of the VMCB. Therefore, we set up the
1655 * APIC_ACCESS_PAGE_PRIVATE_MEMSLOT (4KB) here.
1657 static int avic_init_access_page(struct kvm_vcpu *vcpu)
1659 struct kvm *kvm = vcpu->kvm;
1662 if (kvm->arch.apic_access_page_done)
1665 ret = x86_set_memory_region(kvm,
1666 APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
1667 APIC_DEFAULT_PHYS_BASE,
1672 kvm->arch.apic_access_page_done = true;
1676 static int avic_init_backing_page(struct kvm_vcpu *vcpu)
1679 u64 *entry, new_entry;
1680 int id = vcpu->vcpu_id;
1681 struct vcpu_svm *svm = to_svm(vcpu);
1683 ret = avic_init_access_page(vcpu);
1687 if (id >= AVIC_MAX_PHYSICAL_ID_COUNT)
1690 if (!svm->vcpu.arch.apic->regs)
1693 svm->avic_backing_page = virt_to_page(svm->vcpu.arch.apic->regs);
1695 /* Setting AVIC backing page address in the phy APIC ID table */
1696 entry = avic_get_physical_id_entry(vcpu, id);
1700 new_entry = READ_ONCE(*entry);
1701 new_entry = __sme_set((page_to_phys(svm->avic_backing_page) &
1702 AVIC_PHYSICAL_ID_ENTRY_BACKING_PAGE_MASK) |
1703 AVIC_PHYSICAL_ID_ENTRY_VALID_MASK);
1704 WRITE_ONCE(*entry, new_entry);
1706 svm->avic_physical_id_cache = entry;
1711 static void __sev_asid_free(int asid)
1713 struct svm_cpu_data *sd;
1717 clear_bit(pos, sev_asid_bitmap);
1719 for_each_possible_cpu(cpu) {
1720 sd = per_cpu(svm_data, cpu);
1721 sd->sev_vmcbs[pos] = NULL;
1725 static void sev_asid_free(struct kvm *kvm)
1727 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1729 __sev_asid_free(sev->asid);
1732 static void sev_unbind_asid(struct kvm *kvm, unsigned int handle)
1734 struct sev_data_decommission *decommission;
1735 struct sev_data_deactivate *data;
1740 data = kzalloc(sizeof(*data), GFP_KERNEL);
1744 /* deactivate handle */
1745 data->handle = handle;
1746 sev_guest_deactivate(data, NULL);
1748 wbinvd_on_all_cpus();
1749 sev_guest_df_flush(NULL);
1752 decommission = kzalloc(sizeof(*decommission), GFP_KERNEL);
1756 /* decommission handle */
1757 decommission->handle = handle;
1758 sev_guest_decommission(decommission, NULL);
1760 kfree(decommission);
1763 static struct page **sev_pin_memory(struct kvm *kvm, unsigned long uaddr,
1764 unsigned long ulen, unsigned long *n,
1767 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1768 unsigned long npages, npinned, size;
1769 unsigned long locked, lock_limit;
1770 struct page **pages;
1773 /* Calculate number of pages. */
1774 first = (uaddr & PAGE_MASK) >> PAGE_SHIFT;
1775 last = ((uaddr + ulen - 1) & PAGE_MASK) >> PAGE_SHIFT;
1776 npages = (last - first + 1);
1778 locked = sev->pages_locked + npages;
1779 lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1780 if (locked > lock_limit && !capable(CAP_IPC_LOCK)) {
1781 pr_err("SEV: %lu locked pages exceed the lock limit of %lu.\n", locked, lock_limit);
1785 /* Avoid using vmalloc for smaller buffers. */
1786 size = npages * sizeof(struct page *);
1787 if (size > PAGE_SIZE)
1788 pages = vmalloc(size);
1790 pages = kmalloc(size, GFP_KERNEL);
1795 /* Pin the user virtual address. */
1796 npinned = get_user_pages_fast(uaddr, npages, write ? FOLL_WRITE : 0, pages);
1797 if (npinned != npages) {
1798 pr_err("SEV: Failure locking %lu pages.\n", npages);
1803 sev->pages_locked = locked;
1809 release_pages(pages, npinned);
1815 static void sev_unpin_memory(struct kvm *kvm, struct page **pages,
1816 unsigned long npages)
1818 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1820 release_pages(pages, npages);
1822 sev->pages_locked -= npages;
1825 static void sev_clflush_pages(struct page *pages[], unsigned long npages)
1827 uint8_t *page_virtual;
1830 if (npages == 0 || pages == NULL)
1833 for (i = 0; i < npages; i++) {
1834 page_virtual = kmap_atomic(pages[i]);
1835 clflush_cache_range(page_virtual, PAGE_SIZE);
1836 kunmap_atomic(page_virtual);
1840 static void __unregister_enc_region_locked(struct kvm *kvm,
1841 struct enc_region *region)
1844 * The guest may change the memory encryption attribute from C=0 -> C=1
1845 * or vice versa for this memory range. Lets make sure caches are
1846 * flushed to ensure that guest data gets written into memory with
1849 sev_clflush_pages(region->pages, region->npages);
1851 sev_unpin_memory(kvm, region->pages, region->npages);
1852 list_del(®ion->list);
1856 static struct kvm *svm_vm_alloc(void)
1858 struct kvm_svm *kvm_svm = kzalloc(sizeof(struct kvm_svm), GFP_KERNEL);
1859 return &kvm_svm->kvm;
1862 static void svm_vm_free(struct kvm *kvm)
1864 kfree(to_kvm_svm(kvm));
1867 static void sev_vm_destroy(struct kvm *kvm)
1869 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1870 struct list_head *head = &sev->regions_list;
1871 struct list_head *pos, *q;
1873 if (!sev_guest(kvm))
1876 mutex_lock(&kvm->lock);
1879 * if userspace was terminated before unregistering the memory regions
1880 * then lets unpin all the registered memory.
1882 if (!list_empty(head)) {
1883 list_for_each_safe(pos, q, head) {
1884 __unregister_enc_region_locked(kvm,
1885 list_entry(pos, struct enc_region, list));
1889 mutex_unlock(&kvm->lock);
1891 sev_unbind_asid(kvm, sev->handle);
1895 static void avic_vm_destroy(struct kvm *kvm)
1897 unsigned long flags;
1898 struct kvm_svm *kvm_svm = to_kvm_svm(kvm);
1903 if (kvm_svm->avic_logical_id_table_page)
1904 __free_page(kvm_svm->avic_logical_id_table_page);
1905 if (kvm_svm->avic_physical_id_table_page)
1906 __free_page(kvm_svm->avic_physical_id_table_page);
1908 spin_lock_irqsave(&svm_vm_data_hash_lock, flags);
1909 hash_del(&kvm_svm->hnode);
1910 spin_unlock_irqrestore(&svm_vm_data_hash_lock, flags);
1913 static void svm_vm_destroy(struct kvm *kvm)
1915 avic_vm_destroy(kvm);
1916 sev_vm_destroy(kvm);
1919 static int avic_vm_init(struct kvm *kvm)
1921 unsigned long flags;
1923 struct kvm_svm *kvm_svm = to_kvm_svm(kvm);
1925 struct page *p_page;
1926 struct page *l_page;
1932 /* Allocating physical APIC ID table (4KB) */
1933 p_page = alloc_page(GFP_KERNEL);
1937 kvm_svm->avic_physical_id_table_page = p_page;
1938 clear_page(page_address(p_page));
1940 /* Allocating logical APIC ID table (4KB) */
1941 l_page = alloc_page(GFP_KERNEL);
1945 kvm_svm->avic_logical_id_table_page = l_page;
1946 clear_page(page_address(l_page));
1948 spin_lock_irqsave(&svm_vm_data_hash_lock, flags);
1950 vm_id = next_vm_id = (next_vm_id + 1) & AVIC_VM_ID_MASK;
1951 if (vm_id == 0) { /* id is 1-based, zero is not okay */
1952 next_vm_id_wrapped = 1;
1955 /* Is it still in use? Only possible if wrapped at least once */
1956 if (next_vm_id_wrapped) {
1957 hash_for_each_possible(svm_vm_data_hash, k2, hnode, vm_id) {
1958 if (k2->avic_vm_id == vm_id)
1962 kvm_svm->avic_vm_id = vm_id;
1963 hash_add(svm_vm_data_hash, &kvm_svm->hnode, kvm_svm->avic_vm_id);
1964 spin_unlock_irqrestore(&svm_vm_data_hash_lock, flags);
1969 avic_vm_destroy(kvm);
1974 avic_update_iommu_vcpu_affinity(struct kvm_vcpu *vcpu, int cpu, bool r)
1977 unsigned long flags;
1978 struct amd_svm_iommu_ir *ir;
1979 struct vcpu_svm *svm = to_svm(vcpu);
1981 if (!kvm_arch_has_assigned_device(vcpu->kvm))
1985 * Here, we go through the per-vcpu ir_list to update all existing
1986 * interrupt remapping table entry targeting this vcpu.
1988 spin_lock_irqsave(&svm->ir_list_lock, flags);
1990 if (list_empty(&svm->ir_list))
1993 list_for_each_entry(ir, &svm->ir_list, node) {
1994 ret = amd_iommu_update_ga(cpu, r, ir->data);
1999 spin_unlock_irqrestore(&svm->ir_list_lock, flags);
2003 static void avic_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2006 /* ID = 0xff (broadcast), ID > 0xff (reserved) */
2007 int h_physical_id = kvm_cpu_get_apicid(cpu);
2008 struct vcpu_svm *svm = to_svm(vcpu);
2010 if (!kvm_vcpu_apicv_active(vcpu))
2013 if (WARN_ON(h_physical_id >= AVIC_MAX_PHYSICAL_ID_COUNT))
2016 entry = READ_ONCE(*(svm->avic_physical_id_cache));
2017 WARN_ON(entry & AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK);
2019 entry &= ~AVIC_PHYSICAL_ID_ENTRY_HOST_PHYSICAL_ID_MASK;
2020 entry |= (h_physical_id & AVIC_PHYSICAL_ID_ENTRY_HOST_PHYSICAL_ID_MASK);
2022 entry &= ~AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK;
2023 if (svm->avic_is_running)
2024 entry |= AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK;
2026 WRITE_ONCE(*(svm->avic_physical_id_cache), entry);
2027 avic_update_iommu_vcpu_affinity(vcpu, h_physical_id,
2028 svm->avic_is_running);
2031 static void avic_vcpu_put(struct kvm_vcpu *vcpu)
2034 struct vcpu_svm *svm = to_svm(vcpu);
2036 if (!kvm_vcpu_apicv_active(vcpu))
2039 entry = READ_ONCE(*(svm->avic_physical_id_cache));
2040 if (entry & AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK)
2041 avic_update_iommu_vcpu_affinity(vcpu, -1, 0);
2043 entry &= ~AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK;
2044 WRITE_ONCE(*(svm->avic_physical_id_cache), entry);
2048 * This function is called during VCPU halt/unhalt.
2050 static void avic_set_running(struct kvm_vcpu *vcpu, bool is_run)
2052 struct vcpu_svm *svm = to_svm(vcpu);
2054 svm->avic_is_running = is_run;
2056 avic_vcpu_load(vcpu, vcpu->cpu);
2058 avic_vcpu_put(vcpu);
2061 static void svm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
2063 struct vcpu_svm *svm = to_svm(vcpu);
2067 vcpu->arch.microcode_version = 0x01000065;
2069 svm->virt_spec_ctrl = 0;
2072 svm->vcpu.arch.apic_base = APIC_DEFAULT_PHYS_BASE |
2073 MSR_IA32_APICBASE_ENABLE;
2074 if (kvm_vcpu_is_reset_bsp(&svm->vcpu))
2075 svm->vcpu.arch.apic_base |= MSR_IA32_APICBASE_BSP;
2079 kvm_cpuid(vcpu, &eax, &dummy, &dummy, &dummy, true);
2080 kvm_register_write(vcpu, VCPU_REGS_RDX, eax);
2082 if (kvm_vcpu_apicv_active(vcpu) && !init_event)
2083 avic_update_vapic_bar(svm, APIC_DEFAULT_PHYS_BASE);
2086 static int avic_init_vcpu(struct vcpu_svm *svm)
2090 if (!kvm_vcpu_apicv_active(&svm->vcpu))
2093 ret = avic_init_backing_page(&svm->vcpu);
2097 INIT_LIST_HEAD(&svm->ir_list);
2098 spin_lock_init(&svm->ir_list_lock);
2103 static struct kvm_vcpu *svm_create_vcpu(struct kvm *kvm, unsigned int id)
2105 struct vcpu_svm *svm;
2107 struct page *msrpm_pages;
2108 struct page *hsave_page;
2109 struct page *nested_msrpm_pages;
2112 svm = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
2118 err = kvm_vcpu_init(&svm->vcpu, kvm, id);
2123 page = alloc_page(GFP_KERNEL);
2127 msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
2131 nested_msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
2132 if (!nested_msrpm_pages)
2135 hsave_page = alloc_page(GFP_KERNEL);
2139 err = avic_init_vcpu(svm);
2143 /* We initialize this flag to true to make sure that the is_running
2144 * bit would be set the first time the vcpu is loaded.
2146 svm->avic_is_running = true;
2148 svm->nested.hsave = page_address(hsave_page);
2150 svm->msrpm = page_address(msrpm_pages);
2151 svm_vcpu_init_msrpm(svm->msrpm);
2153 svm->nested.msrpm = page_address(nested_msrpm_pages);
2154 svm_vcpu_init_msrpm(svm->nested.msrpm);
2156 svm->vmcb = page_address(page);
2157 clear_page(svm->vmcb);
2158 svm->vmcb_pa = __sme_set(page_to_pfn(page) << PAGE_SHIFT);
2159 svm->asid_generation = 0;
2162 svm_init_osvw(&svm->vcpu);
2167 __free_page(hsave_page);
2169 __free_pages(nested_msrpm_pages, MSRPM_ALLOC_ORDER);
2171 __free_pages(msrpm_pages, MSRPM_ALLOC_ORDER);
2175 kvm_vcpu_uninit(&svm->vcpu);
2177 kmem_cache_free(kvm_vcpu_cache, svm);
2179 return ERR_PTR(err);
2182 static void svm_free_vcpu(struct kvm_vcpu *vcpu)
2184 struct vcpu_svm *svm = to_svm(vcpu);
2186 __free_page(pfn_to_page(__sme_clr(svm->vmcb_pa) >> PAGE_SHIFT));
2187 __free_pages(virt_to_page(svm->msrpm), MSRPM_ALLOC_ORDER);
2188 __free_page(virt_to_page(svm->nested.hsave));
2189 __free_pages(virt_to_page(svm->nested.msrpm), MSRPM_ALLOC_ORDER);
2190 kvm_vcpu_uninit(vcpu);
2191 kmem_cache_free(kvm_vcpu_cache, svm);
2193 * The vmcb page can be recycled, causing a false negative in
2194 * svm_vcpu_load(). So do a full IBPB now.
2196 indirect_branch_prediction_barrier();
2199 static void svm_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2201 struct vcpu_svm *svm = to_svm(vcpu);
2202 struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
2205 if (unlikely(cpu != vcpu->cpu)) {
2206 svm->asid_generation = 0;
2207 mark_all_dirty(svm->vmcb);
2210 #ifdef CONFIG_X86_64
2211 rdmsrl(MSR_GS_BASE, to_svm(vcpu)->host.gs_base);
2213 savesegment(fs, svm->host.fs);
2214 savesegment(gs, svm->host.gs);
2215 svm->host.ldt = kvm_read_ldt();
2217 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
2218 rdmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
2220 if (static_cpu_has(X86_FEATURE_TSCRATEMSR)) {
2221 u64 tsc_ratio = vcpu->arch.tsc_scaling_ratio;
2222 if (tsc_ratio != __this_cpu_read(current_tsc_ratio)) {
2223 __this_cpu_write(current_tsc_ratio, tsc_ratio);
2224 wrmsrl(MSR_AMD64_TSC_RATIO, tsc_ratio);
2227 /* This assumes that the kernel never uses MSR_TSC_AUX */
2228 if (static_cpu_has(X86_FEATURE_RDTSCP))
2229 wrmsrl(MSR_TSC_AUX, svm->tsc_aux);
2231 if (sd->current_vmcb != svm->vmcb) {
2232 sd->current_vmcb = svm->vmcb;
2233 indirect_branch_prediction_barrier();
2235 avic_vcpu_load(vcpu, cpu);
2238 static void svm_vcpu_put(struct kvm_vcpu *vcpu)
2240 struct vcpu_svm *svm = to_svm(vcpu);
2243 avic_vcpu_put(vcpu);
2245 ++vcpu->stat.host_state_reload;
2246 kvm_load_ldt(svm->host.ldt);
2247 #ifdef CONFIG_X86_64
2248 loadsegment(fs, svm->host.fs);
2249 wrmsrl(MSR_KERNEL_GS_BASE, current->thread.gsbase);
2250 load_gs_index(svm->host.gs);
2252 #ifdef CONFIG_X86_32_LAZY_GS
2253 loadsegment(gs, svm->host.gs);
2256 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
2257 wrmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
2260 static void svm_vcpu_blocking(struct kvm_vcpu *vcpu)
2262 avic_set_running(vcpu, false);
2265 static void svm_vcpu_unblocking(struct kvm_vcpu *vcpu)
2267 avic_set_running(vcpu, true);
2270 static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu)
2272 struct vcpu_svm *svm = to_svm(vcpu);
2273 unsigned long rflags = svm->vmcb->save.rflags;
2275 if (svm->nmi_singlestep) {
2276 /* Hide our flags if they were not set by the guest */
2277 if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_TF))
2278 rflags &= ~X86_EFLAGS_TF;
2279 if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_RF))
2280 rflags &= ~X86_EFLAGS_RF;
2285 static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
2287 if (to_svm(vcpu)->nmi_singlestep)
2288 rflags |= (X86_EFLAGS_TF | X86_EFLAGS_RF);
2291 * Any change of EFLAGS.VM is accompanied by a reload of SS
2292 * (caused by either a task switch or an inter-privilege IRET),
2293 * so we do not need to update the CPL here.
2295 to_svm(vcpu)->vmcb->save.rflags = rflags;
2298 static void svm_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
2301 case VCPU_EXREG_PDPTR:
2302 BUG_ON(!npt_enabled);
2303 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
2310 static void svm_set_vintr(struct vcpu_svm *svm)
2312 set_intercept(svm, INTERCEPT_VINTR);
2315 static void svm_clear_vintr(struct vcpu_svm *svm)
2317 clr_intercept(svm, INTERCEPT_VINTR);
2320 static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg)
2322 struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
2325 case VCPU_SREG_CS: return &save->cs;
2326 case VCPU_SREG_DS: return &save->ds;
2327 case VCPU_SREG_ES: return &save->es;
2328 case VCPU_SREG_FS: return &save->fs;
2329 case VCPU_SREG_GS: return &save->gs;
2330 case VCPU_SREG_SS: return &save->ss;
2331 case VCPU_SREG_TR: return &save->tr;
2332 case VCPU_SREG_LDTR: return &save->ldtr;
2338 static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg)
2340 struct vmcb_seg *s = svm_seg(vcpu, seg);
2345 static void svm_get_segment(struct kvm_vcpu *vcpu,
2346 struct kvm_segment *var, int seg)
2348 struct vmcb_seg *s = svm_seg(vcpu, seg);
2350 var->base = s->base;
2351 var->limit = s->limit;
2352 var->selector = s->selector;
2353 var->type = s->attrib & SVM_SELECTOR_TYPE_MASK;
2354 var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1;
2355 var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
2356 var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1;
2357 var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1;
2358 var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
2359 var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
2362 * AMD CPUs circa 2014 track the G bit for all segments except CS.
2363 * However, the SVM spec states that the G bit is not observed by the
2364 * CPU, and some VMware virtual CPUs drop the G bit for all segments.
2365 * So let's synthesize a legal G bit for all segments, this helps
2366 * running KVM nested. It also helps cross-vendor migration, because
2367 * Intel's vmentry has a check on the 'G' bit.
2369 var->g = s->limit > 0xfffff;
2372 * AMD's VMCB does not have an explicit unusable field, so emulate it
2373 * for cross vendor migration purposes by "not present"
2375 var->unusable = !var->present;
2380 * Work around a bug where the busy flag in the tr selector
2390 * The accessed bit must always be set in the segment
2391 * descriptor cache, although it can be cleared in the
2392 * descriptor, the cached bit always remains at 1. Since
2393 * Intel has a check on this, set it here to support
2394 * cross-vendor migration.
2401 * On AMD CPUs sometimes the DB bit in the segment
2402 * descriptor is left as 1, although the whole segment has
2403 * been made unusable. Clear it here to pass an Intel VMX
2404 * entry check when cross vendor migrating.
2408 /* This is symmetric with svm_set_segment() */
2409 var->dpl = to_svm(vcpu)->vmcb->save.cpl;
2414 static int svm_get_cpl(struct kvm_vcpu *vcpu)
2416 struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
2421 static void svm_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
2423 struct vcpu_svm *svm = to_svm(vcpu);
2425 dt->size = svm->vmcb->save.idtr.limit;
2426 dt->address = svm->vmcb->save.idtr.base;
2429 static void svm_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
2431 struct vcpu_svm *svm = to_svm(vcpu);
2433 svm->vmcb->save.idtr.limit = dt->size;
2434 svm->vmcb->save.idtr.base = dt->address ;
2435 mark_dirty(svm->vmcb, VMCB_DT);
2438 static void svm_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
2440 struct vcpu_svm *svm = to_svm(vcpu);
2442 dt->size = svm->vmcb->save.gdtr.limit;
2443 dt->address = svm->vmcb->save.gdtr.base;
2446 static void svm_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
2448 struct vcpu_svm *svm = to_svm(vcpu);
2450 svm->vmcb->save.gdtr.limit = dt->size;
2451 svm->vmcb->save.gdtr.base = dt->address ;
2452 mark_dirty(svm->vmcb, VMCB_DT);
2455 static void svm_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
2459 static void svm_decache_cr3(struct kvm_vcpu *vcpu)
2463 static void svm_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
2467 static void update_cr0_intercept(struct vcpu_svm *svm)
2469 ulong gcr0 = svm->vcpu.arch.cr0;
2470 u64 *hcr0 = &svm->vmcb->save.cr0;
2472 *hcr0 = (*hcr0 & ~SVM_CR0_SELECTIVE_MASK)
2473 | (gcr0 & SVM_CR0_SELECTIVE_MASK);
2475 mark_dirty(svm->vmcb, VMCB_CR);
2477 if (gcr0 == *hcr0) {
2478 clr_cr_intercept(svm, INTERCEPT_CR0_READ);
2479 clr_cr_intercept(svm, INTERCEPT_CR0_WRITE);
2481 set_cr_intercept(svm, INTERCEPT_CR0_READ);
2482 set_cr_intercept(svm, INTERCEPT_CR0_WRITE);
2486 static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
2488 struct vcpu_svm *svm = to_svm(vcpu);
2490 #ifdef CONFIG_X86_64
2491 if (vcpu->arch.efer & EFER_LME) {
2492 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
2493 vcpu->arch.efer |= EFER_LMA;
2494 svm->vmcb->save.efer |= EFER_LMA | EFER_LME;
2497 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG)) {
2498 vcpu->arch.efer &= ~EFER_LMA;
2499 svm->vmcb->save.efer &= ~(EFER_LMA | EFER_LME);
2503 vcpu->arch.cr0 = cr0;
2506 cr0 |= X86_CR0_PG | X86_CR0_WP;
2509 * re-enable caching here because the QEMU bios
2510 * does not do it - this results in some delay at
2513 if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
2514 cr0 &= ~(X86_CR0_CD | X86_CR0_NW);
2515 svm->vmcb->save.cr0 = cr0;
2516 mark_dirty(svm->vmcb, VMCB_CR);
2517 update_cr0_intercept(svm);
2520 static int svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
2522 unsigned long host_cr4_mce = cr4_read_shadow() & X86_CR4_MCE;
2523 unsigned long old_cr4 = to_svm(vcpu)->vmcb->save.cr4;
2525 if (cr4 & X86_CR4_VMXE)
2528 if (npt_enabled && ((old_cr4 ^ cr4) & X86_CR4_PGE))
2529 svm_flush_tlb(vcpu, true);
2531 vcpu->arch.cr4 = cr4;
2534 cr4 |= host_cr4_mce;
2535 to_svm(vcpu)->vmcb->save.cr4 = cr4;
2536 mark_dirty(to_svm(vcpu)->vmcb, VMCB_CR);
2540 static void svm_set_segment(struct kvm_vcpu *vcpu,
2541 struct kvm_segment *var, int seg)
2543 struct vcpu_svm *svm = to_svm(vcpu);
2544 struct vmcb_seg *s = svm_seg(vcpu, seg);
2546 s->base = var->base;
2547 s->limit = var->limit;
2548 s->selector = var->selector;
2549 s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK);
2550 s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT;
2551 s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT;
2552 s->attrib |= ((var->present & 1) && !var->unusable) << SVM_SELECTOR_P_SHIFT;
2553 s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT;
2554 s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT;
2555 s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT;
2556 s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT;
2559 * This is always accurate, except if SYSRET returned to a segment
2560 * with SS.DPL != 3. Intel does not have this quirk, and always
2561 * forces SS.DPL to 3 on sysret, so we ignore that case; fixing it
2562 * would entail passing the CPL to userspace and back.
2564 if (seg == VCPU_SREG_SS)
2565 /* This is symmetric with svm_get_segment() */
2566 svm->vmcb->save.cpl = (var->dpl & 3);
2568 mark_dirty(svm->vmcb, VMCB_SEG);
2571 static void update_bp_intercept(struct kvm_vcpu *vcpu)
2573 struct vcpu_svm *svm = to_svm(vcpu);
2575 clr_exception_intercept(svm, BP_VECTOR);
2577 if (vcpu->guest_debug & KVM_GUESTDBG_ENABLE) {
2578 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
2579 set_exception_intercept(svm, BP_VECTOR);
2581 vcpu->guest_debug = 0;
2584 static void new_asid(struct vcpu_svm *svm, struct svm_cpu_data *sd)
2586 if (sd->next_asid > sd->max_asid) {
2587 ++sd->asid_generation;
2588 sd->next_asid = sd->min_asid;
2589 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID;
2592 svm->asid_generation = sd->asid_generation;
2593 svm->vmcb->control.asid = sd->next_asid++;
2595 mark_dirty(svm->vmcb, VMCB_ASID);
2598 static u64 svm_get_dr6(struct kvm_vcpu *vcpu)
2600 return to_svm(vcpu)->vmcb->save.dr6;
2603 static void svm_set_dr6(struct kvm_vcpu *vcpu, unsigned long value)
2605 struct vcpu_svm *svm = to_svm(vcpu);
2607 svm->vmcb->save.dr6 = value;
2608 mark_dirty(svm->vmcb, VMCB_DR);
2611 static void svm_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
2613 struct vcpu_svm *svm = to_svm(vcpu);
2615 get_debugreg(vcpu->arch.db[0], 0);
2616 get_debugreg(vcpu->arch.db[1], 1);
2617 get_debugreg(vcpu->arch.db[2], 2);
2618 get_debugreg(vcpu->arch.db[3], 3);
2619 vcpu->arch.dr6 = svm_get_dr6(vcpu);
2620 vcpu->arch.dr7 = svm->vmcb->save.dr7;
2622 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
2623 set_dr_intercepts(svm);
2626 static void svm_set_dr7(struct kvm_vcpu *vcpu, unsigned long value)
2628 struct vcpu_svm *svm = to_svm(vcpu);
2630 svm->vmcb->save.dr7 = value;
2631 mark_dirty(svm->vmcb, VMCB_DR);
2634 static int pf_interception(struct vcpu_svm *svm)
2636 u64 fault_address = __sme_clr(svm->vmcb->control.exit_info_2);
2637 u64 error_code = svm->vmcb->control.exit_info_1;
2639 return kvm_handle_page_fault(&svm->vcpu, error_code, fault_address,
2640 static_cpu_has(X86_FEATURE_DECODEASSISTS) ?
2641 svm->vmcb->control.insn_bytes : NULL,
2642 svm->vmcb->control.insn_len);
2645 static int npf_interception(struct vcpu_svm *svm)
2647 u64 fault_address = __sme_clr(svm->vmcb->control.exit_info_2);
2648 u64 error_code = svm->vmcb->control.exit_info_1;
2650 trace_kvm_page_fault(fault_address, error_code);
2651 return kvm_mmu_page_fault(&svm->vcpu, fault_address, error_code,
2652 static_cpu_has(X86_FEATURE_DECODEASSISTS) ?
2653 svm->vmcb->control.insn_bytes : NULL,
2654 svm->vmcb->control.insn_len);
2657 static int db_interception(struct vcpu_svm *svm)
2659 struct kvm_run *kvm_run = svm->vcpu.run;
2661 if (!(svm->vcpu.guest_debug &
2662 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) &&
2663 !svm->nmi_singlestep) {
2664 kvm_queue_exception(&svm->vcpu, DB_VECTOR);
2668 if (svm->nmi_singlestep) {
2669 disable_nmi_singlestep(svm);
2672 if (svm->vcpu.guest_debug &
2673 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) {
2674 kvm_run->exit_reason = KVM_EXIT_DEBUG;
2675 kvm_run->debug.arch.pc =
2676 svm->vmcb->save.cs.base + svm->vmcb->save.rip;
2677 kvm_run->debug.arch.exception = DB_VECTOR;
2684 static int bp_interception(struct vcpu_svm *svm)
2686 struct kvm_run *kvm_run = svm->vcpu.run;
2688 kvm_run->exit_reason = KVM_EXIT_DEBUG;
2689 kvm_run->debug.arch.pc = svm->vmcb->save.cs.base + svm->vmcb->save.rip;
2690 kvm_run->debug.arch.exception = BP_VECTOR;
2694 static int ud_interception(struct vcpu_svm *svm)
2696 return handle_ud(&svm->vcpu);
2699 static int ac_interception(struct vcpu_svm *svm)
2701 kvm_queue_exception_e(&svm->vcpu, AC_VECTOR, 0);
2705 static int gp_interception(struct vcpu_svm *svm)
2707 struct kvm_vcpu *vcpu = &svm->vcpu;
2708 u32 error_code = svm->vmcb->control.exit_info_1;
2711 WARN_ON_ONCE(!enable_vmware_backdoor);
2713 er = emulate_instruction(vcpu,
2714 EMULTYPE_VMWARE | EMULTYPE_NO_UD_ON_FAIL);
2715 if (er == EMULATE_USER_EXIT)
2717 else if (er != EMULATE_DONE)
2718 kvm_queue_exception_e(vcpu, GP_VECTOR, error_code);
2722 static bool is_erratum_383(void)
2727 if (!erratum_383_found)
2730 value = native_read_msr_safe(MSR_IA32_MC0_STATUS, &err);
2734 /* Bit 62 may or may not be set for this mce */
2735 value &= ~(1ULL << 62);
2737 if (value != 0xb600000000010015ULL)
2740 /* Clear MCi_STATUS registers */
2741 for (i = 0; i < 6; ++i)
2742 native_write_msr_safe(MSR_IA32_MCx_STATUS(i), 0, 0);
2744 value = native_read_msr_safe(MSR_IA32_MCG_STATUS, &err);
2748 value &= ~(1ULL << 2);
2749 low = lower_32_bits(value);
2750 high = upper_32_bits(value);
2752 native_write_msr_safe(MSR_IA32_MCG_STATUS, low, high);
2755 /* Flush tlb to evict multi-match entries */
2761 static void svm_handle_mce(struct vcpu_svm *svm)
2763 if (is_erratum_383()) {
2765 * Erratum 383 triggered. Guest state is corrupt so kill the
2768 pr_err("KVM: Guest triggered AMD Erratum 383\n");
2770 kvm_make_request(KVM_REQ_TRIPLE_FAULT, &svm->vcpu);
2776 * On an #MC intercept the MCE handler is not called automatically in
2777 * the host. So do it by hand here.
2781 /* not sure if we ever come back to this point */
2786 static int mc_interception(struct vcpu_svm *svm)
2791 static int shutdown_interception(struct vcpu_svm *svm)
2793 struct kvm_run *kvm_run = svm->vcpu.run;
2796 * VMCB is undefined after a SHUTDOWN intercept
2797 * so reinitialize it.
2799 clear_page(svm->vmcb);
2802 kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
2806 static int io_interception(struct vcpu_svm *svm)
2808 struct kvm_vcpu *vcpu = &svm->vcpu;
2809 u32 io_info = svm->vmcb->control.exit_info_1; /* address size bug? */
2810 int size, in, string;
2813 ++svm->vcpu.stat.io_exits;
2814 string = (io_info & SVM_IOIO_STR_MASK) != 0;
2815 in = (io_info & SVM_IOIO_TYPE_MASK) != 0;
2817 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
2819 port = io_info >> 16;
2820 size = (io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT;
2821 svm->next_rip = svm->vmcb->control.exit_info_2;
2823 return kvm_fast_pio(&svm->vcpu, size, port, in);
2826 static int nmi_interception(struct vcpu_svm *svm)
2831 static int intr_interception(struct vcpu_svm *svm)
2833 ++svm->vcpu.stat.irq_exits;
2837 static int nop_on_interception(struct vcpu_svm *svm)
2842 static int halt_interception(struct vcpu_svm *svm)
2844 svm->next_rip = kvm_rip_read(&svm->vcpu) + 1;
2845 return kvm_emulate_halt(&svm->vcpu);
2848 static int vmmcall_interception(struct vcpu_svm *svm)
2850 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2851 return kvm_emulate_hypercall(&svm->vcpu);
2854 static unsigned long nested_svm_get_tdp_cr3(struct kvm_vcpu *vcpu)
2856 struct vcpu_svm *svm = to_svm(vcpu);
2858 return svm->nested.nested_cr3;
2861 static u64 nested_svm_get_tdp_pdptr(struct kvm_vcpu *vcpu, int index)
2863 struct vcpu_svm *svm = to_svm(vcpu);
2864 u64 cr3 = svm->nested.nested_cr3;
2868 ret = kvm_vcpu_read_guest_page(vcpu, gpa_to_gfn(__sme_clr(cr3)), &pdpte,
2869 offset_in_page(cr3) + index * 8, 8);
2875 static void nested_svm_set_tdp_cr3(struct kvm_vcpu *vcpu,
2878 struct vcpu_svm *svm = to_svm(vcpu);
2880 svm->vmcb->control.nested_cr3 = __sme_set(root);
2881 mark_dirty(svm->vmcb, VMCB_NPT);
2882 svm_flush_tlb(vcpu, true);
2885 static void nested_svm_inject_npf_exit(struct kvm_vcpu *vcpu,
2886 struct x86_exception *fault)
2888 struct vcpu_svm *svm = to_svm(vcpu);
2890 if (svm->vmcb->control.exit_code != SVM_EXIT_NPF) {
2892 * TODO: track the cause of the nested page fault, and
2893 * correctly fill in the high bits of exit_info_1.
2895 svm->vmcb->control.exit_code = SVM_EXIT_NPF;
2896 svm->vmcb->control.exit_code_hi = 0;
2897 svm->vmcb->control.exit_info_1 = (1ULL << 32);
2898 svm->vmcb->control.exit_info_2 = fault->address;
2901 svm->vmcb->control.exit_info_1 &= ~0xffffffffULL;
2902 svm->vmcb->control.exit_info_1 |= fault->error_code;
2905 * The present bit is always zero for page structure faults on real
2908 if (svm->vmcb->control.exit_info_1 & (2ULL << 32))
2909 svm->vmcb->control.exit_info_1 &= ~1;
2911 nested_svm_vmexit(svm);
2914 static void nested_svm_init_mmu_context(struct kvm_vcpu *vcpu)
2916 WARN_ON(mmu_is_nested(vcpu));
2917 kvm_init_shadow_mmu(vcpu);
2918 vcpu->arch.mmu.set_cr3 = nested_svm_set_tdp_cr3;
2919 vcpu->arch.mmu.get_cr3 = nested_svm_get_tdp_cr3;
2920 vcpu->arch.mmu.get_pdptr = nested_svm_get_tdp_pdptr;
2921 vcpu->arch.mmu.inject_page_fault = nested_svm_inject_npf_exit;
2922 vcpu->arch.mmu.shadow_root_level = get_npt_level(vcpu);
2923 reset_shadow_zero_bits_mask(vcpu, &vcpu->arch.mmu);
2924 vcpu->arch.walk_mmu = &vcpu->arch.nested_mmu;
2927 static void nested_svm_uninit_mmu_context(struct kvm_vcpu *vcpu)
2929 vcpu->arch.walk_mmu = &vcpu->arch.mmu;
2932 static int nested_svm_check_permissions(struct vcpu_svm *svm)
2934 if (!(svm->vcpu.arch.efer & EFER_SVME) ||
2935 !is_paging(&svm->vcpu)) {
2936 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2940 if (svm->vmcb->save.cpl) {
2941 kvm_inject_gp(&svm->vcpu, 0);
2948 static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
2949 bool has_error_code, u32 error_code)
2953 if (!is_guest_mode(&svm->vcpu))
2956 vmexit = nested_svm_intercept(svm);
2957 if (vmexit != NESTED_EXIT_DONE)
2960 svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + nr;
2961 svm->vmcb->control.exit_code_hi = 0;
2962 svm->vmcb->control.exit_info_1 = error_code;
2965 * FIXME: we should not write CR2 when L1 intercepts an L2 #PF exception.
2966 * The fix is to add the ancillary datum (CR2 or DR6) to structs
2967 * kvm_queued_exception and kvm_vcpu_events, so that CR2 and DR6 can be
2968 * written only when inject_pending_event runs (DR6 would written here
2969 * too). This should be conditional on a new capability---if the
2970 * capability is disabled, kvm_multiple_exception would write the
2971 * ancillary information to CR2 or DR6, for backwards ABI-compatibility.
2973 if (svm->vcpu.arch.exception.nested_apf)
2974 svm->vmcb->control.exit_info_2 = svm->vcpu.arch.apf.nested_apf_token;
2976 svm->vmcb->control.exit_info_2 = svm->vcpu.arch.cr2;
2978 svm->nested.exit_required = true;
2982 /* This function returns true if it is save to enable the irq window */
2983 static inline bool nested_svm_intr(struct vcpu_svm *svm)
2985 if (!is_guest_mode(&svm->vcpu))
2988 if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
2991 if (!(svm->vcpu.arch.hflags & HF_HIF_MASK))
2995 * if vmexit was already requested (by intercepted exception
2996 * for instance) do not overwrite it with "external interrupt"
2999 if (svm->nested.exit_required)
3002 svm->vmcb->control.exit_code = SVM_EXIT_INTR;
3003 svm->vmcb->control.exit_info_1 = 0;
3004 svm->vmcb->control.exit_info_2 = 0;
3006 if (svm->nested.intercept & 1ULL) {
3008 * The #vmexit can't be emulated here directly because this
3009 * code path runs with irqs and preemption disabled. A
3010 * #vmexit emulation might sleep. Only signal request for
3013 svm->nested.exit_required = true;
3014 trace_kvm_nested_intr_vmexit(svm->vmcb->save.rip);
3021 /* This function returns true if it is save to enable the nmi window */
3022 static inline bool nested_svm_nmi(struct vcpu_svm *svm)
3024 if (!is_guest_mode(&svm->vcpu))
3027 if (!(svm->nested.intercept & (1ULL << INTERCEPT_NMI)))
3030 svm->vmcb->control.exit_code = SVM_EXIT_NMI;
3031 svm->nested.exit_required = true;
3036 static void *nested_svm_map(struct vcpu_svm *svm, u64 gpa, struct page **_page)
3042 page = kvm_vcpu_gfn_to_page(&svm->vcpu, gpa >> PAGE_SHIFT);
3043 if (is_error_page(page))
3051 kvm_inject_gp(&svm->vcpu, 0);
3056 static void nested_svm_unmap(struct page *page)
3059 kvm_release_page_dirty(page);
3062 static int nested_svm_intercept_ioio(struct vcpu_svm *svm)
3064 unsigned port, size, iopm_len;
3069 if (!(svm->nested.intercept & (1ULL << INTERCEPT_IOIO_PROT)))
3070 return NESTED_EXIT_HOST;
3072 port = svm->vmcb->control.exit_info_1 >> 16;
3073 size = (svm->vmcb->control.exit_info_1 & SVM_IOIO_SIZE_MASK) >>
3074 SVM_IOIO_SIZE_SHIFT;
3075 gpa = svm->nested.vmcb_iopm + (port / 8);
3076 start_bit = port % 8;
3077 iopm_len = (start_bit + size > 8) ? 2 : 1;
3078 mask = (0xf >> (4 - size)) << start_bit;
3081 if (kvm_vcpu_read_guest(&svm->vcpu, gpa, &val, iopm_len))
3082 return NESTED_EXIT_DONE;
3084 return (val & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
3087 static int nested_svm_exit_handled_msr(struct vcpu_svm *svm)
3089 u32 offset, msr, value;
3092 if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT)))
3093 return NESTED_EXIT_HOST;
3095 msr = svm->vcpu.arch.regs[VCPU_REGS_RCX];
3096 offset = svm_msrpm_offset(msr);
3097 write = svm->vmcb->control.exit_info_1 & 1;
3098 mask = 1 << ((2 * (msr & 0xf)) + write);
3100 if (offset == MSR_INVALID)
3101 return NESTED_EXIT_DONE;
3103 /* Offset is in 32 bit units but need in 8 bit units */
3106 if (kvm_vcpu_read_guest(&svm->vcpu, svm->nested.vmcb_msrpm + offset, &value, 4))
3107 return NESTED_EXIT_DONE;
3109 return (value & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
3112 /* DB exceptions for our internal use must not cause vmexit */
3113 static int nested_svm_intercept_db(struct vcpu_svm *svm)
3117 /* if we're not singlestepping, it's not ours */
3118 if (!svm->nmi_singlestep)
3119 return NESTED_EXIT_DONE;
3121 /* if it's not a singlestep exception, it's not ours */
3122 if (kvm_get_dr(&svm->vcpu, 6, &dr6))
3123 return NESTED_EXIT_DONE;
3124 if (!(dr6 & DR6_BS))
3125 return NESTED_EXIT_DONE;
3127 /* if the guest is singlestepping, it should get the vmexit */
3128 if (svm->nmi_singlestep_guest_rflags & X86_EFLAGS_TF) {
3129 disable_nmi_singlestep(svm);
3130 return NESTED_EXIT_DONE;
3133 /* it's ours, the nested hypervisor must not see this one */
3134 return NESTED_EXIT_HOST;
3137 static int nested_svm_exit_special(struct vcpu_svm *svm)
3139 u32 exit_code = svm->vmcb->control.exit_code;
3141 switch (exit_code) {
3144 case SVM_EXIT_EXCP_BASE + MC_VECTOR:
3145 return NESTED_EXIT_HOST;
3147 /* For now we are always handling NPFs when using them */
3149 return NESTED_EXIT_HOST;
3151 case SVM_EXIT_EXCP_BASE + PF_VECTOR:
3152 /* When we're shadowing, trap PFs, but not async PF */
3153 if (!npt_enabled && svm->vcpu.arch.apf.host_apf_reason == 0)
3154 return NESTED_EXIT_HOST;
3160 return NESTED_EXIT_CONTINUE;
3164 * If this function returns true, this #vmexit was already handled
3166 static int nested_svm_intercept(struct vcpu_svm *svm)
3168 u32 exit_code = svm->vmcb->control.exit_code;
3169 int vmexit = NESTED_EXIT_HOST;
3171 switch (exit_code) {
3173 vmexit = nested_svm_exit_handled_msr(svm);
3176 vmexit = nested_svm_intercept_ioio(svm);
3178 case SVM_EXIT_READ_CR0 ... SVM_EXIT_WRITE_CR8: {
3179 u32 bit = 1U << (exit_code - SVM_EXIT_READ_CR0);
3180 if (svm->nested.intercept_cr & bit)
3181 vmexit = NESTED_EXIT_DONE;
3184 case SVM_EXIT_READ_DR0 ... SVM_EXIT_WRITE_DR7: {
3185 u32 bit = 1U << (exit_code - SVM_EXIT_READ_DR0);
3186 if (svm->nested.intercept_dr & bit)
3187 vmexit = NESTED_EXIT_DONE;
3190 case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
3191 u32 excp_bits = 1 << (exit_code - SVM_EXIT_EXCP_BASE);
3192 if (svm->nested.intercept_exceptions & excp_bits) {
3193 if (exit_code == SVM_EXIT_EXCP_BASE + DB_VECTOR)
3194 vmexit = nested_svm_intercept_db(svm);
3196 vmexit = NESTED_EXIT_DONE;
3198 /* async page fault always cause vmexit */
3199 else if ((exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR) &&
3200 svm->vcpu.arch.exception.nested_apf != 0)
3201 vmexit = NESTED_EXIT_DONE;
3204 case SVM_EXIT_ERR: {
3205 vmexit = NESTED_EXIT_DONE;
3209 u64 exit_bits = 1ULL << (exit_code - SVM_EXIT_INTR);
3210 if (svm->nested.intercept & exit_bits)
3211 vmexit = NESTED_EXIT_DONE;
3218 static int nested_svm_exit_handled(struct vcpu_svm *svm)
3222 vmexit = nested_svm_intercept(svm);
3224 if (vmexit == NESTED_EXIT_DONE)
3225 nested_svm_vmexit(svm);
3230 static inline void copy_vmcb_control_area(struct vmcb *dst_vmcb, struct vmcb *from_vmcb)
3232 struct vmcb_control_area *dst = &dst_vmcb->control;
3233 struct vmcb_control_area *from = &from_vmcb->control;
3235 dst->intercept_cr = from->intercept_cr;
3236 dst->intercept_dr = from->intercept_dr;
3237 dst->intercept_exceptions = from->intercept_exceptions;
3238 dst->intercept = from->intercept;
3239 dst->iopm_base_pa = from->iopm_base_pa;
3240 dst->msrpm_base_pa = from->msrpm_base_pa;
3241 dst->tsc_offset = from->tsc_offset;
3242 dst->asid = from->asid;
3243 dst->tlb_ctl = from->tlb_ctl;
3244 dst->int_ctl = from->int_ctl;
3245 dst->int_vector = from->int_vector;
3246 dst->int_state = from->int_state;
3247 dst->exit_code = from->exit_code;
3248 dst->exit_code_hi = from->exit_code_hi;
3249 dst->exit_info_1 = from->exit_info_1;
3250 dst->exit_info_2 = from->exit_info_2;
3251 dst->exit_int_info = from->exit_int_info;
3252 dst->exit_int_info_err = from->exit_int_info_err;
3253 dst->nested_ctl = from->nested_ctl;
3254 dst->event_inj = from->event_inj;
3255 dst->event_inj_err = from->event_inj_err;
3256 dst->nested_cr3 = from->nested_cr3;
3257 dst->virt_ext = from->virt_ext;
3260 static int nested_svm_vmexit(struct vcpu_svm *svm)
3262 struct vmcb *nested_vmcb;
3263 struct vmcb *hsave = svm->nested.hsave;
3264 struct vmcb *vmcb = svm->vmcb;
3267 trace_kvm_nested_vmexit_inject(vmcb->control.exit_code,
3268 vmcb->control.exit_info_1,
3269 vmcb->control.exit_info_2,
3270 vmcb->control.exit_int_info,
3271 vmcb->control.exit_int_info_err,
3274 nested_vmcb = nested_svm_map(svm, svm->nested.vmcb, &page);
3278 /* Exit Guest-Mode */
3279 leave_guest_mode(&svm->vcpu);
3280 svm->nested.vmcb = 0;
3282 /* Give the current vmcb to the guest */
3285 nested_vmcb->save.es = vmcb->save.es;
3286 nested_vmcb->save.cs = vmcb->save.cs;
3287 nested_vmcb->save.ss = vmcb->save.ss;
3288 nested_vmcb->save.ds = vmcb->save.ds;
3289 nested_vmcb->save.gdtr = vmcb->save.gdtr;
3290 nested_vmcb->save.idtr = vmcb->save.idtr;
3291 nested_vmcb->save.efer = svm->vcpu.arch.efer;
3292 nested_vmcb->save.cr0 = kvm_read_cr0(&svm->vcpu);
3293 nested_vmcb->save.cr3 = kvm_read_cr3(&svm->vcpu);
3294 nested_vmcb->save.cr2 = vmcb->save.cr2;
3295 nested_vmcb->save.cr4 = svm->vcpu.arch.cr4;
3296 nested_vmcb->save.rflags = kvm_get_rflags(&svm->vcpu);
3297 nested_vmcb->save.rip = vmcb->save.rip;
3298 nested_vmcb->save.rsp = vmcb->save.rsp;
3299 nested_vmcb->save.rax = vmcb->save.rax;
3300 nested_vmcb->save.dr7 = vmcb->save.dr7;
3301 nested_vmcb->save.dr6 = vmcb->save.dr6;
3302 nested_vmcb->save.cpl = vmcb->save.cpl;
3304 nested_vmcb->control.int_ctl = vmcb->control.int_ctl;
3305 nested_vmcb->control.int_vector = vmcb->control.int_vector;
3306 nested_vmcb->control.int_state = vmcb->control.int_state;
3307 nested_vmcb->control.exit_code = vmcb->control.exit_code;
3308 nested_vmcb->control.exit_code_hi = vmcb->control.exit_code_hi;
3309 nested_vmcb->control.exit_info_1 = vmcb->control.exit_info_1;
3310 nested_vmcb->control.exit_info_2 = vmcb->control.exit_info_2;
3311 nested_vmcb->control.exit_int_info = vmcb->control.exit_int_info;
3312 nested_vmcb->control.exit_int_info_err = vmcb->control.exit_int_info_err;
3314 if (svm->nrips_enabled)
3315 nested_vmcb->control.next_rip = vmcb->control.next_rip;
3318 * If we emulate a VMRUN/#VMEXIT in the same host #vmexit cycle we have
3319 * to make sure that we do not lose injected events. So check event_inj
3320 * here and copy it to exit_int_info if it is valid.
3321 * Exit_int_info and event_inj can't be both valid because the case
3322 * below only happens on a VMRUN instruction intercept which has
3323 * no valid exit_int_info set.
3325 if (vmcb->control.event_inj & SVM_EVTINJ_VALID) {
3326 struct vmcb_control_area *nc = &nested_vmcb->control;
3328 nc->exit_int_info = vmcb->control.event_inj;
3329 nc->exit_int_info_err = vmcb->control.event_inj_err;
3332 nested_vmcb->control.tlb_ctl = 0;
3333 nested_vmcb->control.event_inj = 0;
3334 nested_vmcb->control.event_inj_err = 0;
3336 /* We always set V_INTR_MASKING and remember the old value in hflags */
3337 if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
3338 nested_vmcb->control.int_ctl &= ~V_INTR_MASKING_MASK;
3340 /* Restore the original control entries */
3341 copy_vmcb_control_area(vmcb, hsave);
3343 svm->vcpu.arch.tsc_offset = svm->vmcb->control.tsc_offset;
3344 kvm_clear_exception_queue(&svm->vcpu);
3345 kvm_clear_interrupt_queue(&svm->vcpu);
3347 svm->nested.nested_cr3 = 0;
3349 /* Restore selected save entries */
3350 svm->vmcb->save.es = hsave->save.es;
3351 svm->vmcb->save.cs = hsave->save.cs;
3352 svm->vmcb->save.ss = hsave->save.ss;
3353 svm->vmcb->save.ds = hsave->save.ds;
3354 svm->vmcb->save.gdtr = hsave->save.gdtr;
3355 svm->vmcb->save.idtr = hsave->save.idtr;
3356 kvm_set_rflags(&svm->vcpu, hsave->save.rflags);
3357 svm_set_efer(&svm->vcpu, hsave->save.efer);
3358 svm_set_cr0(&svm->vcpu, hsave->save.cr0 | X86_CR0_PE);
3359 svm_set_cr4(&svm->vcpu, hsave->save.cr4);
3361 svm->vmcb->save.cr3 = hsave->save.cr3;
3362 svm->vcpu.arch.cr3 = hsave->save.cr3;
3364 (void)kvm_set_cr3(&svm->vcpu, hsave->save.cr3);
3366 kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, hsave->save.rax);
3367 kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, hsave->save.rsp);
3368 kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, hsave->save.rip);
3369 svm->vmcb->save.dr7 = 0;
3370 svm->vmcb->save.cpl = 0;
3371 svm->vmcb->control.exit_int_info = 0;
3373 mark_all_dirty(svm->vmcb);
3375 nested_svm_unmap(page);
3377 nested_svm_uninit_mmu_context(&svm->vcpu);
3378 kvm_mmu_reset_context(&svm->vcpu);
3379 kvm_mmu_load(&svm->vcpu);
3384 static bool nested_svm_vmrun_msrpm(struct vcpu_svm *svm)
3387 * This function merges the msr permission bitmaps of kvm and the
3388 * nested vmcb. It is optimized in that it only merges the parts where
3389 * the kvm msr permission bitmap may contain zero bits
3393 if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT)))
3396 for (i = 0; i < MSRPM_OFFSETS; i++) {
3400 if (msrpm_offsets[i] == 0xffffffff)
3403 p = msrpm_offsets[i];
3404 offset = svm->nested.vmcb_msrpm + (p * 4);
3406 if (kvm_vcpu_read_guest(&svm->vcpu, offset, &value, 4))
3409 svm->nested.msrpm[p] = svm->msrpm[p] | value;
3412 svm->vmcb->control.msrpm_base_pa = __sme_set(__pa(svm->nested.msrpm));
3417 static bool nested_vmcb_checks(struct vmcb *vmcb)
3419 if ((vmcb->control.intercept & (1ULL << INTERCEPT_VMRUN)) == 0)
3422 if (vmcb->control.asid == 0)
3425 if ((vmcb->control.nested_ctl & SVM_NESTED_CTL_NP_ENABLE) &&
3432 static void enter_svm_guest_mode(struct vcpu_svm *svm, u64 vmcb_gpa,
3433 struct vmcb *nested_vmcb, struct page *page)
3435 if (kvm_get_rflags(&svm->vcpu) & X86_EFLAGS_IF)
3436 svm->vcpu.arch.hflags |= HF_HIF_MASK;
3438 svm->vcpu.arch.hflags &= ~HF_HIF_MASK;
3440 if (nested_vmcb->control.nested_ctl & SVM_NESTED_CTL_NP_ENABLE) {
3441 kvm_mmu_unload(&svm->vcpu);
3442 svm->nested.nested_cr3 = nested_vmcb->control.nested_cr3;
3443 nested_svm_init_mmu_context(&svm->vcpu);
3446 /* Load the nested guest state */
3447 svm->vmcb->save.es = nested_vmcb->save.es;
3448 svm->vmcb->save.cs = nested_vmcb->save.cs;
3449 svm->vmcb->save.ss = nested_vmcb->save.ss;
3450 svm->vmcb->save.ds = nested_vmcb->save.ds;
3451 svm->vmcb->save.gdtr = nested_vmcb->save.gdtr;
3452 svm->vmcb->save.idtr = nested_vmcb->save.idtr;
3453 kvm_set_rflags(&svm->vcpu, nested_vmcb->save.rflags);
3454 svm_set_efer(&svm->vcpu, nested_vmcb->save.efer);
3455 svm_set_cr0(&svm->vcpu, nested_vmcb->save.cr0);
3456 svm_set_cr4(&svm->vcpu, nested_vmcb->save.cr4);
3458 svm->vmcb->save.cr3 = nested_vmcb->save.cr3;
3459 svm->vcpu.arch.cr3 = nested_vmcb->save.cr3;
3461 (void)kvm_set_cr3(&svm->vcpu, nested_vmcb->save.cr3);
3463 /* Guest paging mode is active - reset mmu */
3464 kvm_mmu_reset_context(&svm->vcpu);
3466 svm->vmcb->save.cr2 = svm->vcpu.arch.cr2 = nested_vmcb->save.cr2;
3467 kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, nested_vmcb->save.rax);
3468 kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, nested_vmcb->save.rsp);
3469 kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, nested_vmcb->save.rip);
3471 /* In case we don't even reach vcpu_run, the fields are not updated */
3472 svm->vmcb->save.rax = nested_vmcb->save.rax;
3473 svm->vmcb->save.rsp = nested_vmcb->save.rsp;
3474 svm->vmcb->save.rip = nested_vmcb->save.rip;
3475 svm->vmcb->save.dr7 = nested_vmcb->save.dr7;
3476 svm->vmcb->save.dr6 = nested_vmcb->save.dr6;
3477 svm->vmcb->save.cpl = nested_vmcb->save.cpl;
3479 svm->nested.vmcb_msrpm = nested_vmcb->control.msrpm_base_pa & ~0x0fffULL;
3480 svm->nested.vmcb_iopm = nested_vmcb->control.iopm_base_pa & ~0x0fffULL;
3482 /* cache intercepts */
3483 svm->nested.intercept_cr = nested_vmcb->control.intercept_cr;
3484 svm->nested.intercept_dr = nested_vmcb->control.intercept_dr;
3485 svm->nested.intercept_exceptions = nested_vmcb->control.intercept_exceptions;
3486 svm->nested.intercept = nested_vmcb->control.intercept;
3488 svm_flush_tlb(&svm->vcpu, true);
3489 svm->vmcb->control.int_ctl = nested_vmcb->control.int_ctl | V_INTR_MASKING_MASK;
3490 if (nested_vmcb->control.int_ctl & V_INTR_MASKING_MASK)
3491 svm->vcpu.arch.hflags |= HF_VINTR_MASK;
3493 svm->vcpu.arch.hflags &= ~HF_VINTR_MASK;
3495 if (svm->vcpu.arch.hflags & HF_VINTR_MASK) {
3496 /* We only want the cr8 intercept bits of the guest */
3497 clr_cr_intercept(svm, INTERCEPT_CR8_READ);
3498 clr_cr_intercept(svm, INTERCEPT_CR8_WRITE);
3501 /* We don't want to see VMMCALLs from a nested guest */
3502 clr_intercept(svm, INTERCEPT_VMMCALL);
3504 svm->vcpu.arch.tsc_offset += nested_vmcb->control.tsc_offset;
3505 svm->vmcb->control.tsc_offset = svm->vcpu.arch.tsc_offset;
3507 svm->vmcb->control.virt_ext = nested_vmcb->control.virt_ext;
3508 svm->vmcb->control.int_vector = nested_vmcb->control.int_vector;
3509 svm->vmcb->control.int_state = nested_vmcb->control.int_state;
3510 svm->vmcb->control.event_inj = nested_vmcb->control.event_inj;
3511 svm->vmcb->control.event_inj_err = nested_vmcb->control.event_inj_err;
3513 nested_svm_unmap(page);
3515 /* Enter Guest-Mode */
3516 enter_guest_mode(&svm->vcpu);
3519 * Merge guest and host intercepts - must be called with vcpu in
3520 * guest-mode to take affect here
3522 recalc_intercepts(svm);
3524 svm->nested.vmcb = vmcb_gpa;
3528 mark_all_dirty(svm->vmcb);
3531 static bool nested_svm_vmrun(struct vcpu_svm *svm)
3533 struct vmcb *nested_vmcb;
3534 struct vmcb *hsave = svm->nested.hsave;
3535 struct vmcb *vmcb = svm->vmcb;
3539 vmcb_gpa = svm->vmcb->save.rax;
3541 nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
3545 if (!nested_vmcb_checks(nested_vmcb)) {
3546 nested_vmcb->control.exit_code = SVM_EXIT_ERR;
3547 nested_vmcb->control.exit_code_hi = 0;
3548 nested_vmcb->control.exit_info_1 = 0;
3549 nested_vmcb->control.exit_info_2 = 0;
3551 nested_svm_unmap(page);
3556 trace_kvm_nested_vmrun(svm->vmcb->save.rip, vmcb_gpa,
3557 nested_vmcb->save.rip,
3558 nested_vmcb->control.int_ctl,
3559 nested_vmcb->control.event_inj,
3560 nested_vmcb->control.nested_ctl);
3562 trace_kvm_nested_intercepts(nested_vmcb->control.intercept_cr & 0xffff,
3563 nested_vmcb->control.intercept_cr >> 16,
3564 nested_vmcb->control.intercept_exceptions,
3565 nested_vmcb->control.intercept);
3567 /* Clear internal status */
3568 kvm_clear_exception_queue(&svm->vcpu);
3569 kvm_clear_interrupt_queue(&svm->vcpu);
3572 * Save the old vmcb, so we don't need to pick what we save, but can
3573 * restore everything when a VMEXIT occurs
3575 hsave->save.es = vmcb->save.es;
3576 hsave->save.cs = vmcb->save.cs;
3577 hsave->save.ss = vmcb->save.ss;
3578 hsave->save.ds = vmcb->save.ds;
3579 hsave->save.gdtr = vmcb->save.gdtr;
3580 hsave->save.idtr = vmcb->save.idtr;
3581 hsave->save.efer = svm->vcpu.arch.efer;
3582 hsave->save.cr0 = kvm_read_cr0(&svm->vcpu);
3583 hsave->save.cr4 = svm->vcpu.arch.cr4;
3584 hsave->save.rflags = kvm_get_rflags(&svm->vcpu);
3585 hsave->save.rip = kvm_rip_read(&svm->vcpu);
3586 hsave->save.rsp = vmcb->save.rsp;
3587 hsave->save.rax = vmcb->save.rax;
3589 hsave->save.cr3 = vmcb->save.cr3;
3591 hsave->save.cr3 = kvm_read_cr3(&svm->vcpu);
3593 copy_vmcb_control_area(hsave, vmcb);
3595 enter_svm_guest_mode(svm, vmcb_gpa, nested_vmcb, page);
3600 static void nested_svm_vmloadsave(struct vmcb *from_vmcb, struct vmcb *to_vmcb)
3602 to_vmcb->save.fs = from_vmcb->save.fs;
3603 to_vmcb->save.gs = from_vmcb->save.gs;
3604 to_vmcb->save.tr = from_vmcb->save.tr;
3605 to_vmcb->save.ldtr = from_vmcb->save.ldtr;
3606 to_vmcb->save.kernel_gs_base = from_vmcb->save.kernel_gs_base;
3607 to_vmcb->save.star = from_vmcb->save.star;
3608 to_vmcb->save.lstar = from_vmcb->save.lstar;
3609 to_vmcb->save.cstar = from_vmcb->save.cstar;
3610 to_vmcb->save.sfmask = from_vmcb->save.sfmask;
3611 to_vmcb->save.sysenter_cs = from_vmcb->save.sysenter_cs;
3612 to_vmcb->save.sysenter_esp = from_vmcb->save.sysenter_esp;
3613 to_vmcb->save.sysenter_eip = from_vmcb->save.sysenter_eip;
3616 static int vmload_interception(struct vcpu_svm *svm)
3618 struct vmcb *nested_vmcb;
3622 if (nested_svm_check_permissions(svm))
3625 nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
3629 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
3630 ret = kvm_skip_emulated_instruction(&svm->vcpu);
3632 nested_svm_vmloadsave(nested_vmcb, svm->vmcb);
3633 nested_svm_unmap(page);
3638 static int vmsave_interception(struct vcpu_svm *svm)
3640 struct vmcb *nested_vmcb;
3644 if (nested_svm_check_permissions(svm))
3647 nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
3651 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
3652 ret = kvm_skip_emulated_instruction(&svm->vcpu);
3654 nested_svm_vmloadsave(svm->vmcb, nested_vmcb);
3655 nested_svm_unmap(page);
3660 static int vmrun_interception(struct vcpu_svm *svm)
3662 if (nested_svm_check_permissions(svm))
3665 /* Save rip after vmrun instruction */
3666 kvm_rip_write(&svm->vcpu, kvm_rip_read(&svm->vcpu) + 3);
3668 if (!nested_svm_vmrun(svm))
3671 if (!nested_svm_vmrun_msrpm(svm))
3678 svm->vmcb->control.exit_code = SVM_EXIT_ERR;
3679 svm->vmcb->control.exit_code_hi = 0;
3680 svm->vmcb->control.exit_info_1 = 0;
3681 svm->vmcb->control.exit_info_2 = 0;
3683 nested_svm_vmexit(svm);
3688 static int stgi_interception(struct vcpu_svm *svm)
3692 if (nested_svm_check_permissions(svm))
3696 * If VGIF is enabled, the STGI intercept is only added to
3697 * detect the opening of the SMI/NMI window; remove it now.
3699 if (vgif_enabled(svm))
3700 clr_intercept(svm, INTERCEPT_STGI);
3702 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
3703 ret = kvm_skip_emulated_instruction(&svm->vcpu);
3704 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
3711 static int clgi_interception(struct vcpu_svm *svm)
3715 if (nested_svm_check_permissions(svm))
3718 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
3719 ret = kvm_skip_emulated_instruction(&svm->vcpu);
3723 /* After a CLGI no interrupts should come */
3724 if (!kvm_vcpu_apicv_active(&svm->vcpu)) {
3725 svm_clear_vintr(svm);
3726 svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
3727 mark_dirty(svm->vmcb, VMCB_INTR);
3733 static int invlpga_interception(struct vcpu_svm *svm)
3735 struct kvm_vcpu *vcpu = &svm->vcpu;
3737 trace_kvm_invlpga(svm->vmcb->save.rip, kvm_register_read(&svm->vcpu, VCPU_REGS_RCX),
3738 kvm_register_read(&svm->vcpu, VCPU_REGS_RAX));
3740 /* Let's treat INVLPGA the same as INVLPG (can be optimized!) */
3741 kvm_mmu_invlpg(vcpu, kvm_register_read(&svm->vcpu, VCPU_REGS_RAX));
3743 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
3744 return kvm_skip_emulated_instruction(&svm->vcpu);
3747 static int skinit_interception(struct vcpu_svm *svm)
3749 trace_kvm_skinit(svm->vmcb->save.rip, kvm_register_read(&svm->vcpu, VCPU_REGS_RAX));
3751 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
3755 static int wbinvd_interception(struct vcpu_svm *svm)
3757 return kvm_emulate_wbinvd(&svm->vcpu);
3760 static int xsetbv_interception(struct vcpu_svm *svm)
3762 u64 new_bv = kvm_read_edx_eax(&svm->vcpu);
3763 u32 index = kvm_register_read(&svm->vcpu, VCPU_REGS_RCX);
3765 if (kvm_set_xcr(&svm->vcpu, index, new_bv) == 0) {
3766 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
3767 return kvm_skip_emulated_instruction(&svm->vcpu);
3773 static int task_switch_interception(struct vcpu_svm *svm)
3777 int int_type = svm->vmcb->control.exit_int_info &
3778 SVM_EXITINTINFO_TYPE_MASK;
3779 int int_vec = svm->vmcb->control.exit_int_info & SVM_EVTINJ_VEC_MASK;
3781 svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_TYPE_MASK;
3783 svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_VALID;
3784 bool has_error_code = false;
3787 tss_selector = (u16)svm->vmcb->control.exit_info_1;
3789 if (svm->vmcb->control.exit_info_2 &
3790 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_IRET))
3791 reason = TASK_SWITCH_IRET;
3792 else if (svm->vmcb->control.exit_info_2 &
3793 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_JMP))
3794 reason = TASK_SWITCH_JMP;
3796 reason = TASK_SWITCH_GATE;
3798 reason = TASK_SWITCH_CALL;
3800 if (reason == TASK_SWITCH_GATE) {
3802 case SVM_EXITINTINFO_TYPE_NMI:
3803 svm->vcpu.arch.nmi_injected = false;
3805 case SVM_EXITINTINFO_TYPE_EXEPT:
3806 if (svm->vmcb->control.exit_info_2 &
3807 (1ULL << SVM_EXITINFOSHIFT_TS_HAS_ERROR_CODE)) {
3808 has_error_code = true;
3810 (u32)svm->vmcb->control.exit_info_2;
3812 kvm_clear_exception_queue(&svm->vcpu);
3814 case SVM_EXITINTINFO_TYPE_INTR:
3815 kvm_clear_interrupt_queue(&svm->vcpu);
3822 if (reason != TASK_SWITCH_GATE ||
3823 int_type == SVM_EXITINTINFO_TYPE_SOFT ||
3824 (int_type == SVM_EXITINTINFO_TYPE_EXEPT &&
3825 (int_vec == OF_VECTOR || int_vec == BP_VECTOR)))
3826 skip_emulated_instruction(&svm->vcpu);
3828 if (int_type != SVM_EXITINTINFO_TYPE_SOFT)
3831 if (kvm_task_switch(&svm->vcpu, tss_selector, int_vec, reason,
3832 has_error_code, error_code) == EMULATE_FAIL) {
3833 svm->vcpu.run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3834 svm->vcpu.run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
3835 svm->vcpu.run->internal.ndata = 0;
3841 static int cpuid_interception(struct vcpu_svm *svm)
3843 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
3844 return kvm_emulate_cpuid(&svm->vcpu);
3847 static int iret_interception(struct vcpu_svm *svm)
3849 ++svm->vcpu.stat.nmi_window_exits;
3850 clr_intercept(svm, INTERCEPT_IRET);
3851 svm->vcpu.arch.hflags |= HF_IRET_MASK;
3852 svm->nmi_iret_rip = kvm_rip_read(&svm->vcpu);
3853 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
3857 static int invlpg_interception(struct vcpu_svm *svm)
3859 if (!static_cpu_has(X86_FEATURE_DECODEASSISTS))
3860 return emulate_instruction(&svm->vcpu, 0) == EMULATE_DONE;
3862 kvm_mmu_invlpg(&svm->vcpu, svm->vmcb->control.exit_info_1);
3863 return kvm_skip_emulated_instruction(&svm->vcpu);
3866 static int emulate_on_interception(struct vcpu_svm *svm)
3868 return emulate_instruction(&svm->vcpu, 0) == EMULATE_DONE;
3871 static int rsm_interception(struct vcpu_svm *svm)
3873 return x86_emulate_instruction(&svm->vcpu, 0, 0,
3874 rsm_ins_bytes, 2) == EMULATE_DONE;
3877 static int rdpmc_interception(struct vcpu_svm *svm)
3881 if (!static_cpu_has(X86_FEATURE_NRIPS))
3882 return emulate_on_interception(svm);
3884 err = kvm_rdpmc(&svm->vcpu);
3885 return kvm_complete_insn_gp(&svm->vcpu, err);
3888 static bool check_selective_cr0_intercepted(struct vcpu_svm *svm,
3891 unsigned long cr0 = svm->vcpu.arch.cr0;
3895 intercept = svm->nested.intercept;
3897 if (!is_guest_mode(&svm->vcpu) ||
3898 (!(intercept & (1ULL << INTERCEPT_SELECTIVE_CR0))))
3901 cr0 &= ~SVM_CR0_SELECTIVE_MASK;
3902 val &= ~SVM_CR0_SELECTIVE_MASK;
3905 svm->vmcb->control.exit_code = SVM_EXIT_CR0_SEL_WRITE;
3906 ret = (nested_svm_exit_handled(svm) == NESTED_EXIT_DONE);
3912 #define CR_VALID (1ULL << 63)
3914 static int cr_interception(struct vcpu_svm *svm)
3920 if (!static_cpu_has(X86_FEATURE_DECODEASSISTS))
3921 return emulate_on_interception(svm);
3923 if (unlikely((svm->vmcb->control.exit_info_1 & CR_VALID) == 0))
3924 return emulate_on_interception(svm);
3926 reg = svm->vmcb->control.exit_info_1 & SVM_EXITINFO_REG_MASK;
3927 if (svm->vmcb->control.exit_code == SVM_EXIT_CR0_SEL_WRITE)
3928 cr = SVM_EXIT_WRITE_CR0 - SVM_EXIT_READ_CR0;
3930 cr = svm->vmcb->control.exit_code - SVM_EXIT_READ_CR0;
3933 if (cr >= 16) { /* mov to cr */
3935 val = kvm_register_read(&svm->vcpu, reg);
3938 if (!check_selective_cr0_intercepted(svm, val))
3939 err = kvm_set_cr0(&svm->vcpu, val);
3945 err = kvm_set_cr3(&svm->vcpu, val);
3948 err = kvm_set_cr4(&svm->vcpu, val);
3951 err = kvm_set_cr8(&svm->vcpu, val);
3954 WARN(1, "unhandled write to CR%d", cr);
3955 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
3958 } else { /* mov from cr */
3961 val = kvm_read_cr0(&svm->vcpu);
3964 val = svm->vcpu.arch.cr2;
3967 val = kvm_read_cr3(&svm->vcpu);
3970 val = kvm_read_cr4(&svm->vcpu);
3973 val = kvm_get_cr8(&svm->vcpu);
3976 WARN(1, "unhandled read from CR%d", cr);
3977 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
3980 kvm_register_write(&svm->vcpu, reg, val);
3982 return kvm_complete_insn_gp(&svm->vcpu, err);
3985 static int dr_interception(struct vcpu_svm *svm)
3990 if (svm->vcpu.guest_debug == 0) {
3992 * No more DR vmexits; force a reload of the debug registers
3993 * and reenter on this instruction. The next vmexit will
3994 * retrieve the full state of the debug registers.
3996 clr_dr_intercepts(svm);
3997 svm->vcpu.arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
4001 if (!boot_cpu_has(X86_FEATURE_DECODEASSISTS))
4002 return emulate_on_interception(svm);
4004 reg = svm->vmcb->control.exit_info_1 & SVM_EXITINFO_REG_MASK;
4005 dr = svm->vmcb->control.exit_code - SVM_EXIT_READ_DR0;
4007 if (dr >= 16) { /* mov to DRn */
4008 if (!kvm_require_dr(&svm->vcpu, dr - 16))
4010 val = kvm_register_read(&svm->vcpu, reg);
4011 kvm_set_dr(&svm->vcpu, dr - 16, val);
4013 if (!kvm_require_dr(&svm->vcpu, dr))
4015 kvm_get_dr(&svm->vcpu, dr, &val);
4016 kvm_register_write(&svm->vcpu, reg, val);
4019 return kvm_skip_emulated_instruction(&svm->vcpu);
4022 static int cr8_write_interception(struct vcpu_svm *svm)
4024 struct kvm_run *kvm_run = svm->vcpu.run;
4027 u8 cr8_prev = kvm_get_cr8(&svm->vcpu);
4028 /* instruction emulation calls kvm_set_cr8() */
4029 r = cr_interception(svm);
4030 if (lapic_in_kernel(&svm->vcpu))
4032 if (cr8_prev <= kvm_get_cr8(&svm->vcpu))
4034 kvm_run->exit_reason = KVM_EXIT_SET_TPR;
4038 static int svm_get_msr_feature(struct kvm_msr_entry *msr)
4042 switch (msr->index) {
4043 case MSR_F10H_DECFG:
4044 if (boot_cpu_has(X86_FEATURE_LFENCE_RDTSC))
4045 msr->data |= MSR_F10H_DECFG_LFENCE_SERIALIZE;
4054 static int svm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
4056 struct vcpu_svm *svm = to_svm(vcpu);
4058 switch (msr_info->index) {
4060 msr_info->data = svm->vmcb->save.star;
4062 #ifdef CONFIG_X86_64
4064 msr_info->data = svm->vmcb->save.lstar;
4067 msr_info->data = svm->vmcb->save.cstar;
4069 case MSR_KERNEL_GS_BASE:
4070 msr_info->data = svm->vmcb->save.kernel_gs_base;
4072 case MSR_SYSCALL_MASK:
4073 msr_info->data = svm->vmcb->save.sfmask;
4076 case MSR_IA32_SYSENTER_CS:
4077 msr_info->data = svm->vmcb->save.sysenter_cs;
4079 case MSR_IA32_SYSENTER_EIP:
4080 msr_info->data = svm->sysenter_eip;
4082 case MSR_IA32_SYSENTER_ESP:
4083 msr_info->data = svm->sysenter_esp;
4086 if (!boot_cpu_has(X86_FEATURE_RDTSCP))
4088 msr_info->data = svm->tsc_aux;
4091 * Nobody will change the following 5 values in the VMCB so we can
4092 * safely return them on rdmsr. They will always be 0 until LBRV is
4095 case MSR_IA32_DEBUGCTLMSR:
4096 msr_info->data = svm->vmcb->save.dbgctl;
4098 case MSR_IA32_LASTBRANCHFROMIP:
4099 msr_info->data = svm->vmcb->save.br_from;
4101 case MSR_IA32_LASTBRANCHTOIP:
4102 msr_info->data = svm->vmcb->save.br_to;
4104 case MSR_IA32_LASTINTFROMIP:
4105 msr_info->data = svm->vmcb->save.last_excp_from;
4107 case MSR_IA32_LASTINTTOIP:
4108 msr_info->data = svm->vmcb->save.last_excp_to;
4110 case MSR_VM_HSAVE_PA:
4111 msr_info->data = svm->nested.hsave_msr;
4114 msr_info->data = svm->nested.vm_cr_msr;
4116 case MSR_IA32_SPEC_CTRL:
4117 if (!msr_info->host_initiated &&
4118 !guest_cpuid_has(vcpu, X86_FEATURE_AMD_IBRS))
4121 msr_info->data = svm->spec_ctrl;
4123 case MSR_AMD64_VIRT_SPEC_CTRL:
4124 if (!msr_info->host_initiated &&
4125 !guest_cpuid_has(vcpu, X86_FEATURE_VIRT_SSBD))
4128 msr_info->data = svm->virt_spec_ctrl;
4130 case MSR_F15H_IC_CFG: {
4134 family = guest_cpuid_family(vcpu);
4135 model = guest_cpuid_model(vcpu);
4137 if (family < 0 || model < 0)
4138 return kvm_get_msr_common(vcpu, msr_info);
4142 if (family == 0x15 &&
4143 (model >= 0x2 && model < 0x20))
4144 msr_info->data = 0x1E;
4147 case MSR_F10H_DECFG:
4148 msr_info->data = svm->msr_decfg;
4151 return kvm_get_msr_common(vcpu, msr_info);
4156 static int rdmsr_interception(struct vcpu_svm *svm)
4158 u32 ecx = kvm_register_read(&svm->vcpu, VCPU_REGS_RCX);
4159 struct msr_data msr_info;
4161 msr_info.index = ecx;
4162 msr_info.host_initiated = false;
4163 if (svm_get_msr(&svm->vcpu, &msr_info)) {
4164 trace_kvm_msr_read_ex(ecx);
4165 kvm_inject_gp(&svm->vcpu, 0);
4168 trace_kvm_msr_read(ecx, msr_info.data);
4170 kvm_register_write(&svm->vcpu, VCPU_REGS_RAX,
4171 msr_info.data & 0xffffffff);
4172 kvm_register_write(&svm->vcpu, VCPU_REGS_RDX,
4173 msr_info.data >> 32);
4174 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
4175 return kvm_skip_emulated_instruction(&svm->vcpu);
4179 static int svm_set_vm_cr(struct kvm_vcpu *vcpu, u64 data)
4181 struct vcpu_svm *svm = to_svm(vcpu);
4182 int svm_dis, chg_mask;
4184 if (data & ~SVM_VM_CR_VALID_MASK)
4187 chg_mask = SVM_VM_CR_VALID_MASK;
4189 if (svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK)
4190 chg_mask &= ~(SVM_VM_CR_SVM_LOCK_MASK | SVM_VM_CR_SVM_DIS_MASK);
4192 svm->nested.vm_cr_msr &= ~chg_mask;
4193 svm->nested.vm_cr_msr |= (data & chg_mask);
4195 svm_dis = svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK;
4197 /* check for svm_disable while efer.svme is set */
4198 if (svm_dis && (vcpu->arch.efer & EFER_SVME))
4204 static int svm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
4206 struct vcpu_svm *svm = to_svm(vcpu);
4208 u32 ecx = msr->index;
4209 u64 data = msr->data;
4211 case MSR_IA32_CR_PAT:
4212 if (!kvm_mtrr_valid(vcpu, MSR_IA32_CR_PAT, data))
4214 vcpu->arch.pat = data;
4215 svm->vmcb->save.g_pat = data;
4216 mark_dirty(svm->vmcb, VMCB_NPT);
4218 case MSR_IA32_SPEC_CTRL:
4219 if (!msr->host_initiated &&
4220 !guest_cpuid_has(vcpu, X86_FEATURE_AMD_IBRS))
4223 /* The STIBP bit doesn't fault even if it's not advertised */
4224 if (data & ~(SPEC_CTRL_IBRS | SPEC_CTRL_STIBP))
4227 svm->spec_ctrl = data;
4234 * When it's written (to non-zero) for the first time, pass
4238 * The handling of the MSR bitmap for L2 guests is done in
4239 * nested_svm_vmrun_msrpm.
4240 * We update the L1 MSR bit as well since it will end up
4241 * touching the MSR anyway now.
4243 set_msr_interception(svm->msrpm, MSR_IA32_SPEC_CTRL, 1, 1);
4245 case MSR_IA32_PRED_CMD:
4246 if (!msr->host_initiated &&
4247 !guest_cpuid_has(vcpu, X86_FEATURE_AMD_IBPB))
4250 if (data & ~PRED_CMD_IBPB)
4256 wrmsrl(MSR_IA32_PRED_CMD, PRED_CMD_IBPB);
4257 if (is_guest_mode(vcpu))
4259 set_msr_interception(svm->msrpm, MSR_IA32_PRED_CMD, 0, 1);
4261 case MSR_AMD64_VIRT_SPEC_CTRL:
4262 if (!msr->host_initiated &&
4263 !guest_cpuid_has(vcpu, X86_FEATURE_VIRT_SSBD))
4266 if (data & ~SPEC_CTRL_SSBD)
4269 svm->virt_spec_ctrl = data;
4272 svm->vmcb->save.star = data;
4274 #ifdef CONFIG_X86_64
4276 svm->vmcb->save.lstar = data;
4279 svm->vmcb->save.cstar = data;
4281 case MSR_KERNEL_GS_BASE:
4282 svm->vmcb->save.kernel_gs_base = data;
4284 case MSR_SYSCALL_MASK:
4285 svm->vmcb->save.sfmask = data;
4288 case MSR_IA32_SYSENTER_CS:
4289 svm->vmcb->save.sysenter_cs = data;
4291 case MSR_IA32_SYSENTER_EIP:
4292 svm->sysenter_eip = data;
4293 svm->vmcb->save.sysenter_eip = data;
4295 case MSR_IA32_SYSENTER_ESP:
4296 svm->sysenter_esp = data;
4297 svm->vmcb->save.sysenter_esp = data;
4300 if (!boot_cpu_has(X86_FEATURE_RDTSCP))
4304 * This is rare, so we update the MSR here instead of using
4305 * direct_access_msrs. Doing that would require a rdmsr in
4308 svm->tsc_aux = data;
4309 wrmsrl(MSR_TSC_AUX, svm->tsc_aux);
4311 case MSR_IA32_DEBUGCTLMSR:
4312 if (!boot_cpu_has(X86_FEATURE_LBRV)) {
4313 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTL 0x%llx, nop\n",
4317 if (data & DEBUGCTL_RESERVED_BITS)
4320 svm->vmcb->save.dbgctl = data;
4321 mark_dirty(svm->vmcb, VMCB_LBR);
4322 if (data & (1ULL<<0))
4323 svm_enable_lbrv(svm);
4325 svm_disable_lbrv(svm);
4327 case MSR_VM_HSAVE_PA:
4328 svm->nested.hsave_msr = data;
4331 return svm_set_vm_cr(vcpu, data);
4333 vcpu_unimpl(vcpu, "unimplemented wrmsr: 0x%x data 0x%llx\n", ecx, data);
4335 case MSR_F10H_DECFG: {
4336 struct kvm_msr_entry msr_entry;
4338 msr_entry.index = msr->index;
4339 if (svm_get_msr_feature(&msr_entry))
4342 /* Check the supported bits */
4343 if (data & ~msr_entry.data)
4346 /* Don't allow the guest to change a bit, #GP */
4347 if (!msr->host_initiated && (data ^ msr_entry.data))
4350 svm->msr_decfg = data;
4353 case MSR_IA32_APICBASE:
4354 if (kvm_vcpu_apicv_active(vcpu))
4355 avic_update_vapic_bar(to_svm(vcpu), data);
4356 /* Follow through */
4358 return kvm_set_msr_common(vcpu, msr);
4363 static int wrmsr_interception(struct vcpu_svm *svm)
4365 struct msr_data msr;
4366 u32 ecx = kvm_register_read(&svm->vcpu, VCPU_REGS_RCX);
4367 u64 data = kvm_read_edx_eax(&svm->vcpu);
4371 msr.host_initiated = false;
4373 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
4374 if (kvm_set_msr(&svm->vcpu, &msr)) {
4375 trace_kvm_msr_write_ex(ecx, data);
4376 kvm_inject_gp(&svm->vcpu, 0);
4379 trace_kvm_msr_write(ecx, data);
4380 return kvm_skip_emulated_instruction(&svm->vcpu);
4384 static int msr_interception(struct vcpu_svm *svm)
4386 if (svm->vmcb->control.exit_info_1)
4387 return wrmsr_interception(svm);
4389 return rdmsr_interception(svm);
4392 static int interrupt_window_interception(struct vcpu_svm *svm)
4394 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
4395 svm_clear_vintr(svm);
4396 svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
4397 mark_dirty(svm->vmcb, VMCB_INTR);
4398 ++svm->vcpu.stat.irq_window_exits;
4402 static int pause_interception(struct vcpu_svm *svm)
4404 struct kvm_vcpu *vcpu = &svm->vcpu;
4405 bool in_kernel = (svm_get_cpl(vcpu) == 0);
4407 if (pause_filter_thresh)
4408 grow_ple_window(vcpu);
4410 kvm_vcpu_on_spin(vcpu, in_kernel);
4414 static int nop_interception(struct vcpu_svm *svm)
4416 return kvm_skip_emulated_instruction(&(svm->vcpu));
4419 static int monitor_interception(struct vcpu_svm *svm)
4421 printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
4422 return nop_interception(svm);
4425 static int mwait_interception(struct vcpu_svm *svm)
4427 printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
4428 return nop_interception(svm);
4431 enum avic_ipi_failure_cause {
4432 AVIC_IPI_FAILURE_INVALID_INT_TYPE,
4433 AVIC_IPI_FAILURE_TARGET_NOT_RUNNING,
4434 AVIC_IPI_FAILURE_INVALID_TARGET,
4435 AVIC_IPI_FAILURE_INVALID_BACKING_PAGE,
4438 static int avic_incomplete_ipi_interception(struct vcpu_svm *svm)
4440 u32 icrh = svm->vmcb->control.exit_info_1 >> 32;
4441 u32 icrl = svm->vmcb->control.exit_info_1;
4442 u32 id = svm->vmcb->control.exit_info_2 >> 32;
4443 u32 index = svm->vmcb->control.exit_info_2 & 0xFF;
4444 struct kvm_lapic *apic = svm->vcpu.arch.apic;
4446 trace_kvm_avic_incomplete_ipi(svm->vcpu.vcpu_id, icrh, icrl, id, index);
4449 case AVIC_IPI_FAILURE_INVALID_INT_TYPE:
4451 * AVIC hardware handles the generation of
4452 * IPIs when the specified Message Type is Fixed
4453 * (also known as fixed delivery mode) and
4454 * the Trigger Mode is edge-triggered. The hardware
4455 * also supports self and broadcast delivery modes
4456 * specified via the Destination Shorthand(DSH)
4457 * field of the ICRL. Logical and physical APIC ID
4458 * formats are supported. All other IPI types cause
4459 * a #VMEXIT, which needs to emulated.
4461 kvm_lapic_reg_write(apic, APIC_ICR2, icrh);
4462 kvm_lapic_reg_write(apic, APIC_ICR, icrl);
4464 case AVIC_IPI_FAILURE_TARGET_NOT_RUNNING: {
4466 struct kvm_vcpu *vcpu;
4467 struct kvm *kvm = svm->vcpu.kvm;
4468 struct kvm_lapic *apic = svm->vcpu.arch.apic;
4471 * At this point, we expect that the AVIC HW has already
4472 * set the appropriate IRR bits on the valid target
4473 * vcpus. So, we just need to kick the appropriate vcpu.
4475 kvm_for_each_vcpu(i, vcpu, kvm) {
4476 bool m = kvm_apic_match_dest(vcpu, apic,
4477 icrl & KVM_APIC_SHORT_MASK,
4478 GET_APIC_DEST_FIELD(icrh),
4479 icrl & KVM_APIC_DEST_MASK);
4481 if (m && !avic_vcpu_is_running(vcpu))
4482 kvm_vcpu_wake_up(vcpu);
4486 case AVIC_IPI_FAILURE_INVALID_TARGET:
4488 case AVIC_IPI_FAILURE_INVALID_BACKING_PAGE:
4489 WARN_ONCE(1, "Invalid backing page\n");
4492 pr_err("Unknown IPI interception\n");
4498 static u32 *avic_get_logical_id_entry(struct kvm_vcpu *vcpu, u32 ldr, bool flat)
4500 struct kvm_svm *kvm_svm = to_kvm_svm(vcpu->kvm);
4502 u32 *logical_apic_id_table;
4503 int dlid = GET_APIC_LOGICAL_ID(ldr);
4508 if (flat) { /* flat */
4509 index = ffs(dlid) - 1;
4512 } else { /* cluster */
4513 int cluster = (dlid & 0xf0) >> 4;
4514 int apic = ffs(dlid & 0x0f) - 1;
4516 if ((apic < 0) || (apic > 7) ||
4519 index = (cluster << 2) + apic;
4522 logical_apic_id_table = (u32 *) page_address(kvm_svm->avic_logical_id_table_page);
4524 return &logical_apic_id_table[index];
4527 static int avic_ldr_write(struct kvm_vcpu *vcpu, u8 g_physical_id, u32 ldr,
4531 u32 *entry, new_entry;
4533 flat = kvm_lapic_get_reg(vcpu->arch.apic, APIC_DFR) == APIC_DFR_FLAT;
4534 entry = avic_get_logical_id_entry(vcpu, ldr, flat);
4538 new_entry = READ_ONCE(*entry);
4539 new_entry &= ~AVIC_LOGICAL_ID_ENTRY_GUEST_PHYSICAL_ID_MASK;
4540 new_entry |= (g_physical_id & AVIC_LOGICAL_ID_ENTRY_GUEST_PHYSICAL_ID_MASK);
4542 new_entry |= AVIC_LOGICAL_ID_ENTRY_VALID_MASK;
4544 new_entry &= ~AVIC_LOGICAL_ID_ENTRY_VALID_MASK;
4545 WRITE_ONCE(*entry, new_entry);
4550 static int avic_handle_ldr_update(struct kvm_vcpu *vcpu)
4553 struct vcpu_svm *svm = to_svm(vcpu);
4554 u32 ldr = kvm_lapic_get_reg(vcpu->arch.apic, APIC_LDR);
4559 ret = avic_ldr_write(vcpu, vcpu->vcpu_id, ldr, true);
4560 if (ret && svm->ldr_reg) {
4561 avic_ldr_write(vcpu, 0, svm->ldr_reg, false);
4569 static int avic_handle_apic_id_update(struct kvm_vcpu *vcpu)
4572 struct vcpu_svm *svm = to_svm(vcpu);
4573 u32 apic_id_reg = kvm_lapic_get_reg(vcpu->arch.apic, APIC_ID);
4574 u32 id = (apic_id_reg >> 24) & 0xff;
4576 if (vcpu->vcpu_id == id)
4579 old = avic_get_physical_id_entry(vcpu, vcpu->vcpu_id);
4580 new = avic_get_physical_id_entry(vcpu, id);
4584 /* We need to move physical_id_entry to new offset */
4587 to_svm(vcpu)->avic_physical_id_cache = new;
4590 * Also update the guest physical APIC ID in the logical
4591 * APIC ID table entry if already setup the LDR.
4594 avic_handle_ldr_update(vcpu);
4599 static int avic_handle_dfr_update(struct kvm_vcpu *vcpu)
4601 struct vcpu_svm *svm = to_svm(vcpu);
4602 struct kvm_svm *kvm_svm = to_kvm_svm(vcpu->kvm);
4603 u32 dfr = kvm_lapic_get_reg(vcpu->arch.apic, APIC_DFR);
4604 u32 mod = (dfr >> 28) & 0xf;
4607 * We assume that all local APICs are using the same type.
4608 * If this changes, we need to flush the AVIC logical
4611 if (kvm_svm->ldr_mode == mod)
4614 clear_page(page_address(kvm_svm->avic_logical_id_table_page));
4615 kvm_svm->ldr_mode = mod;
4618 avic_handle_ldr_update(vcpu);
4622 static int avic_unaccel_trap_write(struct vcpu_svm *svm)
4624 struct kvm_lapic *apic = svm->vcpu.arch.apic;
4625 u32 offset = svm->vmcb->control.exit_info_1 &
4626 AVIC_UNACCEL_ACCESS_OFFSET_MASK;
4630 if (avic_handle_apic_id_update(&svm->vcpu))
4634 if (avic_handle_ldr_update(&svm->vcpu))
4638 avic_handle_dfr_update(&svm->vcpu);
4644 kvm_lapic_reg_write(apic, offset, kvm_lapic_get_reg(apic, offset));
4649 static bool is_avic_unaccelerated_access_trap(u32 offset)
4678 static int avic_unaccelerated_access_interception(struct vcpu_svm *svm)
4681 u32 offset = svm->vmcb->control.exit_info_1 &
4682 AVIC_UNACCEL_ACCESS_OFFSET_MASK;
4683 u32 vector = svm->vmcb->control.exit_info_2 &
4684 AVIC_UNACCEL_ACCESS_VECTOR_MASK;
4685 bool write = (svm->vmcb->control.exit_info_1 >> 32) &
4686 AVIC_UNACCEL_ACCESS_WRITE_MASK;
4687 bool trap = is_avic_unaccelerated_access_trap(offset);
4689 trace_kvm_avic_unaccelerated_access(svm->vcpu.vcpu_id, offset,
4690 trap, write, vector);
4693 WARN_ONCE(!write, "svm: Handling trap read.\n");
4694 ret = avic_unaccel_trap_write(svm);
4696 /* Handling Fault */
4697 ret = (emulate_instruction(&svm->vcpu, 0) == EMULATE_DONE);
4703 static int (*const svm_exit_handlers[])(struct vcpu_svm *svm) = {
4704 [SVM_EXIT_READ_CR0] = cr_interception,
4705 [SVM_EXIT_READ_CR3] = cr_interception,
4706 [SVM_EXIT_READ_CR4] = cr_interception,
4707 [SVM_EXIT_READ_CR8] = cr_interception,
4708 [SVM_EXIT_CR0_SEL_WRITE] = cr_interception,
4709 [SVM_EXIT_WRITE_CR0] = cr_interception,
4710 [SVM_EXIT_WRITE_CR3] = cr_interception,
4711 [SVM_EXIT_WRITE_CR4] = cr_interception,
4712 [SVM_EXIT_WRITE_CR8] = cr8_write_interception,
4713 [SVM_EXIT_READ_DR0] = dr_interception,
4714 [SVM_EXIT_READ_DR1] = dr_interception,
4715 [SVM_EXIT_READ_DR2] = dr_interception,
4716 [SVM_EXIT_READ_DR3] = dr_interception,
4717 [SVM_EXIT_READ_DR4] = dr_interception,
4718 [SVM_EXIT_READ_DR5] = dr_interception,
4719 [SVM_EXIT_READ_DR6] = dr_interception,
4720 [SVM_EXIT_READ_DR7] = dr_interception,
4721 [SVM_EXIT_WRITE_DR0] = dr_interception,
4722 [SVM_EXIT_WRITE_DR1] = dr_interception,
4723 [SVM_EXIT_WRITE_DR2] = dr_interception,
4724 [SVM_EXIT_WRITE_DR3] = dr_interception,
4725 [SVM_EXIT_WRITE_DR4] = dr_interception,
4726 [SVM_EXIT_WRITE_DR5] = dr_interception,
4727 [SVM_EXIT_WRITE_DR6] = dr_interception,
4728 [SVM_EXIT_WRITE_DR7] = dr_interception,
4729 [SVM_EXIT_EXCP_BASE + DB_VECTOR] = db_interception,
4730 [SVM_EXIT_EXCP_BASE + BP_VECTOR] = bp_interception,
4731 [SVM_EXIT_EXCP_BASE + UD_VECTOR] = ud_interception,
4732 [SVM_EXIT_EXCP_BASE + PF_VECTOR] = pf_interception,
4733 [SVM_EXIT_EXCP_BASE + MC_VECTOR] = mc_interception,
4734 [SVM_EXIT_EXCP_BASE + AC_VECTOR] = ac_interception,
4735 [SVM_EXIT_EXCP_BASE + GP_VECTOR] = gp_interception,
4736 [SVM_EXIT_INTR] = intr_interception,
4737 [SVM_EXIT_NMI] = nmi_interception,
4738 [SVM_EXIT_SMI] = nop_on_interception,
4739 [SVM_EXIT_INIT] = nop_on_interception,
4740 [SVM_EXIT_VINTR] = interrupt_window_interception,
4741 [SVM_EXIT_RDPMC] = rdpmc_interception,
4742 [SVM_EXIT_CPUID] = cpuid_interception,
4743 [SVM_EXIT_IRET] = iret_interception,
4744 [SVM_EXIT_INVD] = emulate_on_interception,
4745 [SVM_EXIT_PAUSE] = pause_interception,
4746 [SVM_EXIT_HLT] = halt_interception,
4747 [SVM_EXIT_INVLPG] = invlpg_interception,
4748 [SVM_EXIT_INVLPGA] = invlpga_interception,
4749 [SVM_EXIT_IOIO] = io_interception,
4750 [SVM_EXIT_MSR] = msr_interception,
4751 [SVM_EXIT_TASK_SWITCH] = task_switch_interception,
4752 [SVM_EXIT_SHUTDOWN] = shutdown_interception,
4753 [SVM_EXIT_VMRUN] = vmrun_interception,
4754 [SVM_EXIT_VMMCALL] = vmmcall_interception,
4755 [SVM_EXIT_VMLOAD] = vmload_interception,
4756 [SVM_EXIT_VMSAVE] = vmsave_interception,
4757 [SVM_EXIT_STGI] = stgi_interception,
4758 [SVM_EXIT_CLGI] = clgi_interception,
4759 [SVM_EXIT_SKINIT] = skinit_interception,
4760 [SVM_EXIT_WBINVD] = wbinvd_interception,
4761 [SVM_EXIT_MONITOR] = monitor_interception,
4762 [SVM_EXIT_MWAIT] = mwait_interception,
4763 [SVM_EXIT_XSETBV] = xsetbv_interception,
4764 [SVM_EXIT_NPF] = npf_interception,
4765 [SVM_EXIT_RSM] = rsm_interception,
4766 [SVM_EXIT_AVIC_INCOMPLETE_IPI] = avic_incomplete_ipi_interception,
4767 [SVM_EXIT_AVIC_UNACCELERATED_ACCESS] = avic_unaccelerated_access_interception,
4770 static void dump_vmcb(struct kvm_vcpu *vcpu)
4772 struct vcpu_svm *svm = to_svm(vcpu);
4773 struct vmcb_control_area *control = &svm->vmcb->control;
4774 struct vmcb_save_area *save = &svm->vmcb->save;
4776 pr_err("VMCB Control Area:\n");
4777 pr_err("%-20s%04x\n", "cr_read:", control->intercept_cr & 0xffff);
4778 pr_err("%-20s%04x\n", "cr_write:", control->intercept_cr >> 16);
4779 pr_err("%-20s%04x\n", "dr_read:", control->intercept_dr & 0xffff);
4780 pr_err("%-20s%04x\n", "dr_write:", control->intercept_dr >> 16);
4781 pr_err("%-20s%08x\n", "exceptions:", control->intercept_exceptions);
4782 pr_err("%-20s%016llx\n", "intercepts:", control->intercept);
4783 pr_err("%-20s%d\n", "pause filter count:", control->pause_filter_count);
4784 pr_err("%-20s%d\n", "pause filter threshold:",
4785 control->pause_filter_thresh);
4786 pr_err("%-20s%016llx\n", "iopm_base_pa:", control->iopm_base_pa);
4787 pr_err("%-20s%016llx\n", "msrpm_base_pa:", control->msrpm_base_pa);
4788 pr_err("%-20s%016llx\n", "tsc_offset:", control->tsc_offset);
4789 pr_err("%-20s%d\n", "asid:", control->asid);
4790 pr_err("%-20s%d\n", "tlb_ctl:", control->tlb_ctl);
4791 pr_err("%-20s%08x\n", "int_ctl:", control->int_ctl);
4792 pr_err("%-20s%08x\n", "int_vector:", control->int_vector);
4793 pr_err("%-20s%08x\n", "int_state:", control->int_state);
4794 pr_err("%-20s%08x\n", "exit_code:", control->exit_code);
4795 pr_err("%-20s%016llx\n", "exit_info1:", control->exit_info_1);
4796 pr_err("%-20s%016llx\n", "exit_info2:", control->exit_info_2);
4797 pr_err("%-20s%08x\n", "exit_int_info:", control->exit_int_info);
4798 pr_err("%-20s%08x\n", "exit_int_info_err:", control->exit_int_info_err);
4799 pr_err("%-20s%lld\n", "nested_ctl:", control->nested_ctl);
4800 pr_err("%-20s%016llx\n", "nested_cr3:", control->nested_cr3);
4801 pr_err("%-20s%016llx\n", "avic_vapic_bar:", control->avic_vapic_bar);
4802 pr_err("%-20s%08x\n", "event_inj:", control->event_inj);
4803 pr_err("%-20s%08x\n", "event_inj_err:", control->event_inj_err);
4804 pr_err("%-20s%lld\n", "virt_ext:", control->virt_ext);
4805 pr_err("%-20s%016llx\n", "next_rip:", control->next_rip);
4806 pr_err("%-20s%016llx\n", "avic_backing_page:", control->avic_backing_page);
4807 pr_err("%-20s%016llx\n", "avic_logical_id:", control->avic_logical_id);
4808 pr_err("%-20s%016llx\n", "avic_physical_id:", control->avic_physical_id);
4809 pr_err("VMCB State Save Area:\n");
4810 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4812 save->es.selector, save->es.attrib,
4813 save->es.limit, save->es.base);
4814 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4816 save->cs.selector, save->cs.attrib,
4817 save->cs.limit, save->cs.base);
4818 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4820 save->ss.selector, save->ss.attrib,
4821 save->ss.limit, save->ss.base);
4822 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4824 save->ds.selector, save->ds.attrib,
4825 save->ds.limit, save->ds.base);
4826 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4828 save->fs.selector, save->fs.attrib,
4829 save->fs.limit, save->fs.base);
4830 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4832 save->gs.selector, save->gs.attrib,
4833 save->gs.limit, save->gs.base);
4834 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4836 save->gdtr.selector, save->gdtr.attrib,
4837 save->gdtr.limit, save->gdtr.base);
4838 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4840 save->ldtr.selector, save->ldtr.attrib,
4841 save->ldtr.limit, save->ldtr.base);
4842 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4844 save->idtr.selector, save->idtr.attrib,
4845 save->idtr.limit, save->idtr.base);
4846 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4848 save->tr.selector, save->tr.attrib,
4849 save->tr.limit, save->tr.base);
4850 pr_err("cpl: %d efer: %016llx\n",
4851 save->cpl, save->efer);
4852 pr_err("%-15s %016llx %-13s %016llx\n",
4853 "cr0:", save->cr0, "cr2:", save->cr2);
4854 pr_err("%-15s %016llx %-13s %016llx\n",
4855 "cr3:", save->cr3, "cr4:", save->cr4);
4856 pr_err("%-15s %016llx %-13s %016llx\n",
4857 "dr6:", save->dr6, "dr7:", save->dr7);
4858 pr_err("%-15s %016llx %-13s %016llx\n",
4859 "rip:", save->rip, "rflags:", save->rflags);
4860 pr_err("%-15s %016llx %-13s %016llx\n",
4861 "rsp:", save->rsp, "rax:", save->rax);
4862 pr_err("%-15s %016llx %-13s %016llx\n",
4863 "star:", save->star, "lstar:", save->lstar);
4864 pr_err("%-15s %016llx %-13s %016llx\n",
4865 "cstar:", save->cstar, "sfmask:", save->sfmask);
4866 pr_err("%-15s %016llx %-13s %016llx\n",
4867 "kernel_gs_base:", save->kernel_gs_base,
4868 "sysenter_cs:", save->sysenter_cs);
4869 pr_err("%-15s %016llx %-13s %016llx\n",
4870 "sysenter_esp:", save->sysenter_esp,
4871 "sysenter_eip:", save->sysenter_eip);
4872 pr_err("%-15s %016llx %-13s %016llx\n",
4873 "gpat:", save->g_pat, "dbgctl:", save->dbgctl);
4874 pr_err("%-15s %016llx %-13s %016llx\n",
4875 "br_from:", save->br_from, "br_to:", save->br_to);
4876 pr_err("%-15s %016llx %-13s %016llx\n",
4877 "excp_from:", save->last_excp_from,
4878 "excp_to:", save->last_excp_to);
4881 static void svm_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
4883 struct vmcb_control_area *control = &to_svm(vcpu)->vmcb->control;
4885 *info1 = control->exit_info_1;
4886 *info2 = control->exit_info_2;
4889 static int handle_exit(struct kvm_vcpu *vcpu)
4891 struct vcpu_svm *svm = to_svm(vcpu);
4892 struct kvm_run *kvm_run = vcpu->run;
4893 u32 exit_code = svm->vmcb->control.exit_code;
4895 trace_kvm_exit(exit_code, vcpu, KVM_ISA_SVM);
4897 if (!is_cr_intercept(svm, INTERCEPT_CR0_WRITE))
4898 vcpu->arch.cr0 = svm->vmcb->save.cr0;
4900 vcpu->arch.cr3 = svm->vmcb->save.cr3;
4902 if (unlikely(svm->nested.exit_required)) {
4903 nested_svm_vmexit(svm);
4904 svm->nested.exit_required = false;
4909 if (is_guest_mode(vcpu)) {
4912 trace_kvm_nested_vmexit(svm->vmcb->save.rip, exit_code,
4913 svm->vmcb->control.exit_info_1,
4914 svm->vmcb->control.exit_info_2,
4915 svm->vmcb->control.exit_int_info,
4916 svm->vmcb->control.exit_int_info_err,
4919 vmexit = nested_svm_exit_special(svm);
4921 if (vmexit == NESTED_EXIT_CONTINUE)
4922 vmexit = nested_svm_exit_handled(svm);
4924 if (vmexit == NESTED_EXIT_DONE)
4928 svm_complete_interrupts(svm);
4930 if (svm->vmcb->control.exit_code == SVM_EXIT_ERR) {
4931 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4932 kvm_run->fail_entry.hardware_entry_failure_reason
4933 = svm->vmcb->control.exit_code;
4934 pr_err("KVM: FAILED VMRUN WITH VMCB:\n");
4939 if (is_external_interrupt(svm->vmcb->control.exit_int_info) &&
4940 exit_code != SVM_EXIT_EXCP_BASE + PF_VECTOR &&
4941 exit_code != SVM_EXIT_NPF && exit_code != SVM_EXIT_TASK_SWITCH &&
4942 exit_code != SVM_EXIT_INTR && exit_code != SVM_EXIT_NMI)
4943 printk(KERN_ERR "%s: unexpected exit_int_info 0x%x "
4945 __func__, svm->vmcb->control.exit_int_info,
4948 if (exit_code >= ARRAY_SIZE(svm_exit_handlers)
4949 || !svm_exit_handlers[exit_code]) {
4950 WARN_ONCE(1, "svm: unexpected exit reason 0x%x\n", exit_code);
4951 kvm_queue_exception(vcpu, UD_VECTOR);
4955 return svm_exit_handlers[exit_code](svm);
4958 static void reload_tss(struct kvm_vcpu *vcpu)
4960 int cpu = raw_smp_processor_id();
4962 struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
4963 sd->tss_desc->type = 9; /* available 32/64-bit TSS */
4967 static void pre_sev_run(struct vcpu_svm *svm, int cpu)
4969 struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
4970 int asid = sev_get_asid(svm->vcpu.kvm);
4972 /* Assign the asid allocated with this SEV guest */
4973 svm->vmcb->control.asid = asid;
4978 * 1) when different VMCB for the same ASID is to be run on the same host CPU.
4979 * 2) or this VMCB was executed on different host CPU in previous VMRUNs.
4981 if (sd->sev_vmcbs[asid] == svm->vmcb &&
4982 svm->last_cpu == cpu)
4985 svm->last_cpu = cpu;
4986 sd->sev_vmcbs[asid] = svm->vmcb;
4987 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ASID;
4988 mark_dirty(svm->vmcb, VMCB_ASID);
4991 static void pre_svm_run(struct vcpu_svm *svm)
4993 int cpu = raw_smp_processor_id();
4995 struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
4997 if (sev_guest(svm->vcpu.kvm))
4998 return pre_sev_run(svm, cpu);
5000 /* FIXME: handle wraparound of asid_generation */
5001 if (svm->asid_generation != sd->asid_generation)
5005 static void svm_inject_nmi(struct kvm_vcpu *vcpu)
5007 struct vcpu_svm *svm = to_svm(vcpu);
5009 svm->vmcb->control.event_inj = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI;
5010 vcpu->arch.hflags |= HF_NMI_MASK;
5011 set_intercept(svm, INTERCEPT_IRET);
5012 ++vcpu->stat.nmi_injections;
5015 static inline void svm_inject_irq(struct vcpu_svm *svm, int irq)
5017 struct vmcb_control_area *control;
5019 /* The following fields are ignored when AVIC is enabled */
5020 control = &svm->vmcb->control;
5021 control->int_vector = irq;
5022 control->int_ctl &= ~V_INTR_PRIO_MASK;
5023 control->int_ctl |= V_IRQ_MASK |
5024 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT);
5025 mark_dirty(svm->vmcb, VMCB_INTR);
5028 static void svm_set_irq(struct kvm_vcpu *vcpu)
5030 struct vcpu_svm *svm = to_svm(vcpu);
5032 BUG_ON(!(gif_set(svm)));
5034 trace_kvm_inj_virq(vcpu->arch.interrupt.nr);
5035 ++vcpu->stat.irq_injections;
5037 svm->vmcb->control.event_inj = vcpu->arch.interrupt.nr |
5038 SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR;
5041 static inline bool svm_nested_virtualize_tpr(struct kvm_vcpu *vcpu)
5043 return is_guest_mode(vcpu) && (vcpu->arch.hflags & HF_VINTR_MASK);
5046 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
5048 struct vcpu_svm *svm = to_svm(vcpu);
5050 if (svm_nested_virtualize_tpr(vcpu) ||
5051 kvm_vcpu_apicv_active(vcpu))
5054 clr_cr_intercept(svm, INTERCEPT_CR8_WRITE);
5060 set_cr_intercept(svm, INTERCEPT_CR8_WRITE);
5063 static void svm_set_virtual_x2apic_mode(struct kvm_vcpu *vcpu, bool set)
5068 static bool svm_get_enable_apicv(struct kvm_vcpu *vcpu)
5070 return avic && irqchip_split(vcpu->kvm);
5073 static void svm_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
5077 static void svm_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr)
5081 /* Note: Currently only used by Hyper-V. */
5082 static void svm_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu)
5084 struct vcpu_svm *svm = to_svm(vcpu);
5085 struct vmcb *vmcb = svm->vmcb;
5087 if (!kvm_vcpu_apicv_active(&svm->vcpu))
5090 vmcb->control.int_ctl &= ~AVIC_ENABLE_MASK;
5091 mark_dirty(vmcb, VMCB_INTR);
5094 static void svm_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
5099 static void svm_deliver_avic_intr(struct kvm_vcpu *vcpu, int vec)
5101 kvm_lapic_set_irr(vec, vcpu->arch.apic);
5102 smp_mb__after_atomic();
5104 if (avic_vcpu_is_running(vcpu))
5105 wrmsrl(SVM_AVIC_DOORBELL,
5106 kvm_cpu_get_apicid(vcpu->cpu));
5108 kvm_vcpu_wake_up(vcpu);
5111 static void svm_ir_list_del(struct vcpu_svm *svm, struct amd_iommu_pi_data *pi)
5113 unsigned long flags;
5114 struct amd_svm_iommu_ir *cur;
5116 spin_lock_irqsave(&svm->ir_list_lock, flags);
5117 list_for_each_entry(cur, &svm->ir_list, node) {
5118 if (cur->data != pi->ir_data)
5120 list_del(&cur->node);
5124 spin_unlock_irqrestore(&svm->ir_list_lock, flags);
5127 static int svm_ir_list_add(struct vcpu_svm *svm, struct amd_iommu_pi_data *pi)
5130 unsigned long flags;
5131 struct amd_svm_iommu_ir *ir;
5134 * In some cases, the existing irte is updaed and re-set,
5135 * so we need to check here if it's already been * added
5138 if (pi->ir_data && (pi->prev_ga_tag != 0)) {
5139 struct kvm *kvm = svm->vcpu.kvm;
5140 u32 vcpu_id = AVIC_GATAG_TO_VCPUID(pi->prev_ga_tag);
5141 struct kvm_vcpu *prev_vcpu = kvm_get_vcpu_by_id(kvm, vcpu_id);
5142 struct vcpu_svm *prev_svm;
5149 prev_svm = to_svm(prev_vcpu);
5150 svm_ir_list_del(prev_svm, pi);
5154 * Allocating new amd_iommu_pi_data, which will get
5155 * add to the per-vcpu ir_list.
5157 ir = kzalloc(sizeof(struct amd_svm_iommu_ir), GFP_KERNEL);
5162 ir->data = pi->ir_data;
5164 spin_lock_irqsave(&svm->ir_list_lock, flags);
5165 list_add(&ir->node, &svm->ir_list);
5166 spin_unlock_irqrestore(&svm->ir_list_lock, flags);
5173 * The HW cannot support posting multicast/broadcast
5174 * interrupts to a vCPU. So, we still use legacy interrupt
5175 * remapping for these kind of interrupts.
5177 * For lowest-priority interrupts, we only support
5178 * those with single CPU as the destination, e.g. user
5179 * configures the interrupts via /proc/irq or uses
5180 * irqbalance to make the interrupts single-CPU.
5183 get_pi_vcpu_info(struct kvm *kvm, struct kvm_kernel_irq_routing_entry *e,
5184 struct vcpu_data *vcpu_info, struct vcpu_svm **svm)
5186 struct kvm_lapic_irq irq;
5187 struct kvm_vcpu *vcpu = NULL;
5189 kvm_set_msi_irq(kvm, e, &irq);
5191 if (!kvm_intr_is_single_vcpu(kvm, &irq, &vcpu)) {
5192 pr_debug("SVM: %s: use legacy intr remap mode for irq %u\n",
5193 __func__, irq.vector);
5197 pr_debug("SVM: %s: use GA mode for irq %u\n", __func__,
5199 *svm = to_svm(vcpu);
5200 vcpu_info->pi_desc_addr = __sme_set(page_to_phys((*svm)->avic_backing_page));
5201 vcpu_info->vector = irq.vector;
5207 * svm_update_pi_irte - set IRTE for Posted-Interrupts
5210 * @host_irq: host irq of the interrupt
5211 * @guest_irq: gsi of the interrupt
5212 * @set: set or unset PI
5213 * returns 0 on success, < 0 on failure
5215 static int svm_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
5216 uint32_t guest_irq, bool set)
5218 struct kvm_kernel_irq_routing_entry *e;
5219 struct kvm_irq_routing_table *irq_rt;
5220 int idx, ret = -EINVAL;
5222 if (!kvm_arch_has_assigned_device(kvm) ||
5223 !irq_remapping_cap(IRQ_POSTING_CAP))
5226 pr_debug("SVM: %s: host_irq=%#x, guest_irq=%#x, set=%#x\n",
5227 __func__, host_irq, guest_irq, set);
5229 idx = srcu_read_lock(&kvm->irq_srcu);
5230 irq_rt = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu);
5231 WARN_ON(guest_irq >= irq_rt->nr_rt_entries);
5233 hlist_for_each_entry(e, &irq_rt->map[guest_irq], link) {
5234 struct vcpu_data vcpu_info;
5235 struct vcpu_svm *svm = NULL;
5237 if (e->type != KVM_IRQ_ROUTING_MSI)
5241 * Here, we setup with legacy mode in the following cases:
5242 * 1. When cannot target interrupt to a specific vcpu.
5243 * 2. Unsetting posted interrupt.
5244 * 3. APIC virtialization is disabled for the vcpu.
5246 if (!get_pi_vcpu_info(kvm, e, &vcpu_info, &svm) && set &&
5247 kvm_vcpu_apicv_active(&svm->vcpu)) {
5248 struct amd_iommu_pi_data pi;
5250 /* Try to enable guest_mode in IRTE */
5251 pi.base = __sme_set(page_to_phys(svm->avic_backing_page) &
5253 pi.ga_tag = AVIC_GATAG(to_kvm_svm(kvm)->avic_vm_id,
5255 pi.is_guest_mode = true;
5256 pi.vcpu_data = &vcpu_info;
5257 ret = irq_set_vcpu_affinity(host_irq, &pi);
5260 * Here, we successfully setting up vcpu affinity in
5261 * IOMMU guest mode. Now, we need to store the posted
5262 * interrupt information in a per-vcpu ir_list so that
5263 * we can reference to them directly when we update vcpu
5264 * scheduling information in IOMMU irte.
5266 if (!ret && pi.is_guest_mode)
5267 svm_ir_list_add(svm, &pi);
5269 /* Use legacy mode in IRTE */
5270 struct amd_iommu_pi_data pi;
5273 * Here, pi is used to:
5274 * - Tell IOMMU to use legacy mode for this interrupt.
5275 * - Retrieve ga_tag of prior interrupt remapping data.
5277 pi.is_guest_mode = false;
5278 ret = irq_set_vcpu_affinity(host_irq, &pi);
5281 * Check if the posted interrupt was previously
5282 * setup with the guest_mode by checking if the ga_tag
5283 * was cached. If so, we need to clean up the per-vcpu
5286 if (!ret && pi.prev_ga_tag) {
5287 int id = AVIC_GATAG_TO_VCPUID(pi.prev_ga_tag);
5288 struct kvm_vcpu *vcpu;
5290 vcpu = kvm_get_vcpu_by_id(kvm, id);
5292 svm_ir_list_del(to_svm(vcpu), &pi);
5297 trace_kvm_pi_irte_update(host_irq, svm->vcpu.vcpu_id,
5298 e->gsi, vcpu_info.vector,
5299 vcpu_info.pi_desc_addr, set);
5303 pr_err("%s: failed to update PI IRTE\n", __func__);
5310 srcu_read_unlock(&kvm->irq_srcu, idx);
5314 static int svm_nmi_allowed(struct kvm_vcpu *vcpu)
5316 struct vcpu_svm *svm = to_svm(vcpu);
5317 struct vmcb *vmcb = svm->vmcb;
5319 ret = !(vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK) &&
5320 !(svm->vcpu.arch.hflags & HF_NMI_MASK);
5321 ret = ret && gif_set(svm) && nested_svm_nmi(svm);
5326 static bool svm_get_nmi_mask(struct kvm_vcpu *vcpu)
5328 struct vcpu_svm *svm = to_svm(vcpu);
5330 return !!(svm->vcpu.arch.hflags & HF_NMI_MASK);
5333 static void svm_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
5335 struct vcpu_svm *svm = to_svm(vcpu);
5338 svm->vcpu.arch.hflags |= HF_NMI_MASK;
5339 set_intercept(svm, INTERCEPT_IRET);
5341 svm->vcpu.arch.hflags &= ~HF_NMI_MASK;
5342 clr_intercept(svm, INTERCEPT_IRET);
5346 static int svm_interrupt_allowed(struct kvm_vcpu *vcpu)
5348 struct vcpu_svm *svm = to_svm(vcpu);
5349 struct vmcb *vmcb = svm->vmcb;
5352 if (!gif_set(svm) ||
5353 (vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK))
5356 ret = !!(kvm_get_rflags(vcpu) & X86_EFLAGS_IF);
5358 if (is_guest_mode(vcpu))
5359 return ret && !(svm->vcpu.arch.hflags & HF_VINTR_MASK);
5364 static void enable_irq_window(struct kvm_vcpu *vcpu)
5366 struct vcpu_svm *svm = to_svm(vcpu);
5368 if (kvm_vcpu_apicv_active(vcpu))
5372 * In case GIF=0 we can't rely on the CPU to tell us when GIF becomes
5373 * 1, because that's a separate STGI/VMRUN intercept. The next time we
5374 * get that intercept, this function will be called again though and
5375 * we'll get the vintr intercept. However, if the vGIF feature is
5376 * enabled, the STGI interception will not occur. Enable the irq
5377 * window under the assumption that the hardware will set the GIF.
5379 if ((vgif_enabled(svm) || gif_set(svm)) && nested_svm_intr(svm)) {
5381 svm_inject_irq(svm, 0x0);
5385 static void enable_nmi_window(struct kvm_vcpu *vcpu)
5387 struct vcpu_svm *svm = to_svm(vcpu);
5389 if ((svm->vcpu.arch.hflags & (HF_NMI_MASK | HF_IRET_MASK))
5391 return; /* IRET will cause a vm exit */
5393 if (!gif_set(svm)) {
5394 if (vgif_enabled(svm))
5395 set_intercept(svm, INTERCEPT_STGI);
5396 return; /* STGI will cause a vm exit */
5399 if (svm->nested.exit_required)
5400 return; /* we're not going to run the guest yet */
5403 * Something prevents NMI from been injected. Single step over possible
5404 * problem (IRET or exception injection or interrupt shadow)
5406 svm->nmi_singlestep_guest_rflags = svm_get_rflags(vcpu);
5407 svm->nmi_singlestep = true;
5408 svm->vmcb->save.rflags |= (X86_EFLAGS_TF | X86_EFLAGS_RF);
5411 static int svm_set_tss_addr(struct kvm *kvm, unsigned int addr)
5416 static int svm_set_identity_map_addr(struct kvm *kvm, u64 ident_addr)
5421 static void svm_flush_tlb(struct kvm_vcpu *vcpu, bool invalidate_gpa)
5423 struct vcpu_svm *svm = to_svm(vcpu);
5425 if (static_cpu_has(X86_FEATURE_FLUSHBYASID))
5426 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ASID;
5428 svm->asid_generation--;
5431 static void svm_prepare_guest_switch(struct kvm_vcpu *vcpu)
5435 static inline void sync_cr8_to_lapic(struct kvm_vcpu *vcpu)
5437 struct vcpu_svm *svm = to_svm(vcpu);
5439 if (svm_nested_virtualize_tpr(vcpu))
5442 if (!is_cr_intercept(svm, INTERCEPT_CR8_WRITE)) {
5443 int cr8 = svm->vmcb->control.int_ctl & V_TPR_MASK;
5444 kvm_set_cr8(vcpu, cr8);
5448 static inline void sync_lapic_to_cr8(struct kvm_vcpu *vcpu)
5450 struct vcpu_svm *svm = to_svm(vcpu);
5453 if (svm_nested_virtualize_tpr(vcpu) ||
5454 kvm_vcpu_apicv_active(vcpu))
5457 cr8 = kvm_get_cr8(vcpu);
5458 svm->vmcb->control.int_ctl &= ~V_TPR_MASK;
5459 svm->vmcb->control.int_ctl |= cr8 & V_TPR_MASK;
5462 static void svm_complete_interrupts(struct vcpu_svm *svm)
5466 u32 exitintinfo = svm->vmcb->control.exit_int_info;
5467 unsigned int3_injected = svm->int3_injected;
5469 svm->int3_injected = 0;
5472 * If we've made progress since setting HF_IRET_MASK, we've
5473 * executed an IRET and can allow NMI injection.
5475 if ((svm->vcpu.arch.hflags & HF_IRET_MASK)
5476 && kvm_rip_read(&svm->vcpu) != svm->nmi_iret_rip) {
5477 svm->vcpu.arch.hflags &= ~(HF_NMI_MASK | HF_IRET_MASK);
5478 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
5481 svm->vcpu.arch.nmi_injected = false;
5482 kvm_clear_exception_queue(&svm->vcpu);
5483 kvm_clear_interrupt_queue(&svm->vcpu);
5485 if (!(exitintinfo & SVM_EXITINTINFO_VALID))
5488 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
5490 vector = exitintinfo & SVM_EXITINTINFO_VEC_MASK;
5491 type = exitintinfo & SVM_EXITINTINFO_TYPE_MASK;
5494 case SVM_EXITINTINFO_TYPE_NMI:
5495 svm->vcpu.arch.nmi_injected = true;
5497 case SVM_EXITINTINFO_TYPE_EXEPT:
5499 * In case of software exceptions, do not reinject the vector,
5500 * but re-execute the instruction instead. Rewind RIP first
5501 * if we emulated INT3 before.
5503 if (kvm_exception_is_soft(vector)) {
5504 if (vector == BP_VECTOR && int3_injected &&
5505 kvm_is_linear_rip(&svm->vcpu, svm->int3_rip))
5506 kvm_rip_write(&svm->vcpu,
5507 kvm_rip_read(&svm->vcpu) -
5511 if (exitintinfo & SVM_EXITINTINFO_VALID_ERR) {
5512 u32 err = svm->vmcb->control.exit_int_info_err;
5513 kvm_requeue_exception_e(&svm->vcpu, vector, err);
5516 kvm_requeue_exception(&svm->vcpu, vector);
5518 case SVM_EXITINTINFO_TYPE_INTR:
5519 kvm_queue_interrupt(&svm->vcpu, vector, false);
5526 static void svm_cancel_injection(struct kvm_vcpu *vcpu)
5528 struct vcpu_svm *svm = to_svm(vcpu);
5529 struct vmcb_control_area *control = &svm->vmcb->control;
5531 control->exit_int_info = control->event_inj;
5532 control->exit_int_info_err = control->event_inj_err;
5533 control->event_inj = 0;
5534 svm_complete_interrupts(svm);
5537 static void svm_vcpu_run(struct kvm_vcpu *vcpu)
5539 struct vcpu_svm *svm = to_svm(vcpu);
5541 svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX];
5542 svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP];
5543 svm->vmcb->save.rip = vcpu->arch.regs[VCPU_REGS_RIP];
5546 * A vmexit emulation is required before the vcpu can be executed
5549 if (unlikely(svm->nested.exit_required))
5553 * Disable singlestep if we're injecting an interrupt/exception.
5554 * We don't want our modified rflags to be pushed on the stack where
5555 * we might not be able to easily reset them if we disabled NMI
5558 if (svm->nmi_singlestep && svm->vmcb->control.event_inj) {
5560 * Event injection happens before external interrupts cause a
5561 * vmexit and interrupts are disabled here, so smp_send_reschedule
5562 * is enough to force an immediate vmexit.
5564 disable_nmi_singlestep(svm);
5565 smp_send_reschedule(vcpu->cpu);
5570 sync_lapic_to_cr8(vcpu);
5572 svm->vmcb->save.cr2 = vcpu->arch.cr2;
5579 * If this vCPU has touched SPEC_CTRL, restore the guest's value if
5580 * it's non-zero. Since vmentry is serialising on affected CPUs, there
5581 * is no need to worry about the conditional branch over the wrmsr
5582 * being speculatively taken.
5584 x86_spec_ctrl_set_guest(svm->spec_ctrl, svm->virt_spec_ctrl);
5587 "push %%" _ASM_BP "; \n\t"
5588 "mov %c[rbx](%[svm]), %%" _ASM_BX " \n\t"
5589 "mov %c[rcx](%[svm]), %%" _ASM_CX " \n\t"
5590 "mov %c[rdx](%[svm]), %%" _ASM_DX " \n\t"
5591 "mov %c[rsi](%[svm]), %%" _ASM_SI " \n\t"
5592 "mov %c[rdi](%[svm]), %%" _ASM_DI " \n\t"
5593 "mov %c[rbp](%[svm]), %%" _ASM_BP " \n\t"
5594 #ifdef CONFIG_X86_64
5595 "mov %c[r8](%[svm]), %%r8 \n\t"
5596 "mov %c[r9](%[svm]), %%r9 \n\t"
5597 "mov %c[r10](%[svm]), %%r10 \n\t"
5598 "mov %c[r11](%[svm]), %%r11 \n\t"
5599 "mov %c[r12](%[svm]), %%r12 \n\t"
5600 "mov %c[r13](%[svm]), %%r13 \n\t"
5601 "mov %c[r14](%[svm]), %%r14 \n\t"
5602 "mov %c[r15](%[svm]), %%r15 \n\t"
5605 /* Enter guest mode */
5606 "push %%" _ASM_AX " \n\t"
5607 "mov %c[vmcb](%[svm]), %%" _ASM_AX " \n\t"
5608 __ex(SVM_VMLOAD) "\n\t"
5609 __ex(SVM_VMRUN) "\n\t"
5610 __ex(SVM_VMSAVE) "\n\t"
5611 "pop %%" _ASM_AX " \n\t"
5613 /* Save guest registers, load host registers */
5614 "mov %%" _ASM_BX ", %c[rbx](%[svm]) \n\t"
5615 "mov %%" _ASM_CX ", %c[rcx](%[svm]) \n\t"
5616 "mov %%" _ASM_DX ", %c[rdx](%[svm]) \n\t"
5617 "mov %%" _ASM_SI ", %c[rsi](%[svm]) \n\t"
5618 "mov %%" _ASM_DI ", %c[rdi](%[svm]) \n\t"
5619 "mov %%" _ASM_BP ", %c[rbp](%[svm]) \n\t"
5620 #ifdef CONFIG_X86_64
5621 "mov %%r8, %c[r8](%[svm]) \n\t"
5622 "mov %%r9, %c[r9](%[svm]) \n\t"
5623 "mov %%r10, %c[r10](%[svm]) \n\t"
5624 "mov %%r11, %c[r11](%[svm]) \n\t"
5625 "mov %%r12, %c[r12](%[svm]) \n\t"
5626 "mov %%r13, %c[r13](%[svm]) \n\t"
5627 "mov %%r14, %c[r14](%[svm]) \n\t"
5628 "mov %%r15, %c[r15](%[svm]) \n\t"
5631 * Clear host registers marked as clobbered to prevent
5634 "xor %%" _ASM_BX ", %%" _ASM_BX " \n\t"
5635 "xor %%" _ASM_CX ", %%" _ASM_CX " \n\t"
5636 "xor %%" _ASM_DX ", %%" _ASM_DX " \n\t"
5637 "xor %%" _ASM_SI ", %%" _ASM_SI " \n\t"
5638 "xor %%" _ASM_DI ", %%" _ASM_DI " \n\t"
5639 #ifdef CONFIG_X86_64
5640 "xor %%r8, %%r8 \n\t"
5641 "xor %%r9, %%r9 \n\t"
5642 "xor %%r10, %%r10 \n\t"
5643 "xor %%r11, %%r11 \n\t"
5644 "xor %%r12, %%r12 \n\t"
5645 "xor %%r13, %%r13 \n\t"
5646 "xor %%r14, %%r14 \n\t"
5647 "xor %%r15, %%r15 \n\t"
5652 [vmcb]"i"(offsetof(struct vcpu_svm, vmcb_pa)),
5653 [rbx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBX])),
5654 [rcx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RCX])),
5655 [rdx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDX])),
5656 [rsi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RSI])),
5657 [rdi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDI])),
5658 [rbp]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBP]))
5659 #ifdef CONFIG_X86_64
5660 , [r8]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R8])),
5661 [r9]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R9])),
5662 [r10]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R10])),
5663 [r11]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R11])),
5664 [r12]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R12])),
5665 [r13]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R13])),
5666 [r14]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R14])),
5667 [r15]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R15]))
5670 #ifdef CONFIG_X86_64
5671 , "rbx", "rcx", "rdx", "rsi", "rdi"
5672 , "r8", "r9", "r10", "r11" , "r12", "r13", "r14", "r15"
5674 , "ebx", "ecx", "edx", "esi", "edi"
5678 /* Eliminate branch target predictions from guest mode */
5681 #ifdef CONFIG_X86_64
5682 wrmsrl(MSR_GS_BASE, svm->host.gs_base);
5684 loadsegment(fs, svm->host.fs);
5685 #ifndef CONFIG_X86_32_LAZY_GS
5686 loadsegment(gs, svm->host.gs);
5691 * We do not use IBRS in the kernel. If this vCPU has used the
5692 * SPEC_CTRL MSR it may have left it on; save the value and
5693 * turn it off. This is much more efficient than blindly adding
5694 * it to the atomic save/restore list. Especially as the former
5695 * (Saving guest MSRs on vmexit) doesn't even exist in KVM.
5697 * For non-nested case:
5698 * If the L01 MSR bitmap does not intercept the MSR, then we need to
5702 * If the L02 MSR bitmap does not intercept the MSR, then we need to
5705 if (unlikely(!msr_write_intercepted(vcpu, MSR_IA32_SPEC_CTRL)))
5706 svm->spec_ctrl = native_read_msr(MSR_IA32_SPEC_CTRL);
5708 x86_spec_ctrl_restore_host(svm->spec_ctrl, svm->virt_spec_ctrl);
5712 local_irq_disable();
5714 vcpu->arch.cr2 = svm->vmcb->save.cr2;
5715 vcpu->arch.regs[VCPU_REGS_RAX] = svm->vmcb->save.rax;
5716 vcpu->arch.regs[VCPU_REGS_RSP] = svm->vmcb->save.rsp;
5717 vcpu->arch.regs[VCPU_REGS_RIP] = svm->vmcb->save.rip;
5719 if (unlikely(svm->vmcb->control.exit_code == SVM_EXIT_NMI))
5720 kvm_before_interrupt(&svm->vcpu);
5724 /* Any pending NMI will happen here */
5726 if (unlikely(svm->vmcb->control.exit_code == SVM_EXIT_NMI))
5727 kvm_after_interrupt(&svm->vcpu);
5729 sync_cr8_to_lapic(vcpu);
5733 svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
5735 /* if exit due to PF check for async PF */
5736 if (svm->vmcb->control.exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR)
5737 svm->vcpu.arch.apf.host_apf_reason = kvm_read_and_reset_pf_reason();
5740 vcpu->arch.regs_avail &= ~(1 << VCPU_EXREG_PDPTR);
5741 vcpu->arch.regs_dirty &= ~(1 << VCPU_EXREG_PDPTR);
5745 * We need to handle MC intercepts here before the vcpu has a chance to
5746 * change the physical cpu
5748 if (unlikely(svm->vmcb->control.exit_code ==
5749 SVM_EXIT_EXCP_BASE + MC_VECTOR))
5750 svm_handle_mce(svm);
5752 mark_all_clean(svm->vmcb);
5754 STACK_FRAME_NON_STANDARD(svm_vcpu_run);
5756 static void svm_set_cr3(struct kvm_vcpu *vcpu, unsigned long root)
5758 struct vcpu_svm *svm = to_svm(vcpu);
5760 svm->vmcb->save.cr3 = __sme_set(root);
5761 mark_dirty(svm->vmcb, VMCB_CR);
5762 svm_flush_tlb(vcpu, true);
5765 static void set_tdp_cr3(struct kvm_vcpu *vcpu, unsigned long root)
5767 struct vcpu_svm *svm = to_svm(vcpu);
5769 svm->vmcb->control.nested_cr3 = __sme_set(root);
5770 mark_dirty(svm->vmcb, VMCB_NPT);
5772 /* Also sync guest cr3 here in case we live migrate */
5773 svm->vmcb->save.cr3 = kvm_read_cr3(vcpu);
5774 mark_dirty(svm->vmcb, VMCB_CR);
5776 svm_flush_tlb(vcpu, true);
5779 static int is_disabled(void)
5783 rdmsrl(MSR_VM_CR, vm_cr);
5784 if (vm_cr & (1 << SVM_VM_CR_SVM_DISABLE))
5791 svm_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
5794 * Patch in the VMMCALL instruction:
5796 hypercall[0] = 0x0f;
5797 hypercall[1] = 0x01;
5798 hypercall[2] = 0xd9;
5801 static void svm_check_processor_compat(void *rtn)
5806 static bool svm_cpu_has_accelerated_tpr(void)
5811 static bool svm_has_emulated_msr(int index)
5816 static u64 svm_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
5821 static void svm_cpuid_update(struct kvm_vcpu *vcpu)
5823 struct vcpu_svm *svm = to_svm(vcpu);
5825 /* Update nrips enabled cache */
5826 svm->nrips_enabled = !!guest_cpuid_has(&svm->vcpu, X86_FEATURE_NRIPS);
5828 if (!kvm_vcpu_apicv_active(vcpu))
5831 guest_cpuid_clear(vcpu, X86_FEATURE_X2APIC);
5834 static void svm_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
5839 entry->ecx &= ~bit(X86_FEATURE_X2APIC);
5843 entry->ecx |= (1 << 2); /* Set SVM bit */
5846 entry->eax = 1; /* SVM revision 1 */
5847 entry->ebx = 8; /* Lets support 8 ASIDs in case we add proper
5848 ASID emulation to nested SVM */
5849 entry->ecx = 0; /* Reserved */
5850 entry->edx = 0; /* Per default do not support any
5851 additional features */
5853 /* Support next_rip if host supports it */
5854 if (boot_cpu_has(X86_FEATURE_NRIPS))
5855 entry->edx |= SVM_FEATURE_NRIP;
5857 /* Support NPT for the guest if enabled */
5859 entry->edx |= SVM_FEATURE_NPT;
5863 /* Support memory encryption cpuid if host supports it */
5864 if (boot_cpu_has(X86_FEATURE_SEV))
5865 cpuid(0x8000001f, &entry->eax, &entry->ebx,
5866 &entry->ecx, &entry->edx);
5871 static int svm_get_lpage_level(void)
5873 return PT_PDPE_LEVEL;
5876 static bool svm_rdtscp_supported(void)
5878 return boot_cpu_has(X86_FEATURE_RDTSCP);
5881 static bool svm_invpcid_supported(void)
5886 static bool svm_mpx_supported(void)
5891 static bool svm_xsaves_supported(void)
5896 static bool svm_umip_emulated(void)
5901 static bool svm_has_wbinvd_exit(void)
5906 #define PRE_EX(exit) { .exit_code = (exit), \
5907 .stage = X86_ICPT_PRE_EXCEPT, }
5908 #define POST_EX(exit) { .exit_code = (exit), \
5909 .stage = X86_ICPT_POST_EXCEPT, }
5910 #define POST_MEM(exit) { .exit_code = (exit), \
5911 .stage = X86_ICPT_POST_MEMACCESS, }
5913 static const struct __x86_intercept {
5915 enum x86_intercept_stage stage;
5916 } x86_intercept_map[] = {
5917 [x86_intercept_cr_read] = POST_EX(SVM_EXIT_READ_CR0),
5918 [x86_intercept_cr_write] = POST_EX(SVM_EXIT_WRITE_CR0),
5919 [x86_intercept_clts] = POST_EX(SVM_EXIT_WRITE_CR0),
5920 [x86_intercept_lmsw] = POST_EX(SVM_EXIT_WRITE_CR0),
5921 [x86_intercept_smsw] = POST_EX(SVM_EXIT_READ_CR0),
5922 [x86_intercept_dr_read] = POST_EX(SVM_EXIT_READ_DR0),
5923 [x86_intercept_dr_write] = POST_EX(SVM_EXIT_WRITE_DR0),
5924 [x86_intercept_sldt] = POST_EX(SVM_EXIT_LDTR_READ),
5925 [x86_intercept_str] = POST_EX(SVM_EXIT_TR_READ),
5926 [x86_intercept_lldt] = POST_EX(SVM_EXIT_LDTR_WRITE),
5927 [x86_intercept_ltr] = POST_EX(SVM_EXIT_TR_WRITE),
5928 [x86_intercept_sgdt] = POST_EX(SVM_EXIT_GDTR_READ),
5929 [x86_intercept_sidt] = POST_EX(SVM_EXIT_IDTR_READ),
5930 [x86_intercept_lgdt] = POST_EX(SVM_EXIT_GDTR_WRITE),
5931 [x86_intercept_lidt] = POST_EX(SVM_EXIT_IDTR_WRITE),
5932 [x86_intercept_vmrun] = POST_EX(SVM_EXIT_VMRUN),
5933 [x86_intercept_vmmcall] = POST_EX(SVM_EXIT_VMMCALL),
5934 [x86_intercept_vmload] = POST_EX(SVM_EXIT_VMLOAD),
5935 [x86_intercept_vmsave] = POST_EX(SVM_EXIT_VMSAVE),
5936 [x86_intercept_stgi] = POST_EX(SVM_EXIT_STGI),
5937 [x86_intercept_clgi] = POST_EX(SVM_EXIT_CLGI),
5938 [x86_intercept_skinit] = POST_EX(SVM_EXIT_SKINIT),
5939 [x86_intercept_invlpga] = POST_EX(SVM_EXIT_INVLPGA),
5940 [x86_intercept_rdtscp] = POST_EX(SVM_EXIT_RDTSCP),
5941 [x86_intercept_monitor] = POST_MEM(SVM_EXIT_MONITOR),
5942 [x86_intercept_mwait] = POST_EX(SVM_EXIT_MWAIT),
5943 [x86_intercept_invlpg] = POST_EX(SVM_EXIT_INVLPG),
5944 [x86_intercept_invd] = POST_EX(SVM_EXIT_INVD),
5945 [x86_intercept_wbinvd] = POST_EX(SVM_EXIT_WBINVD),
5946 [x86_intercept_wrmsr] = POST_EX(SVM_EXIT_MSR),
5947 [x86_intercept_rdtsc] = POST_EX(SVM_EXIT_RDTSC),
5948 [x86_intercept_rdmsr] = POST_EX(SVM_EXIT_MSR),
5949 [x86_intercept_rdpmc] = POST_EX(SVM_EXIT_RDPMC),
5950 [x86_intercept_cpuid] = PRE_EX(SVM_EXIT_CPUID),
5951 [x86_intercept_rsm] = PRE_EX(SVM_EXIT_RSM),
5952 [x86_intercept_pause] = PRE_EX(SVM_EXIT_PAUSE),
5953 [x86_intercept_pushf] = PRE_EX(SVM_EXIT_PUSHF),
5954 [x86_intercept_popf] = PRE_EX(SVM_EXIT_POPF),
5955 [x86_intercept_intn] = PRE_EX(SVM_EXIT_SWINT),
5956 [x86_intercept_iret] = PRE_EX(SVM_EXIT_IRET),
5957 [x86_intercept_icebp] = PRE_EX(SVM_EXIT_ICEBP),
5958 [x86_intercept_hlt] = POST_EX(SVM_EXIT_HLT),
5959 [x86_intercept_in] = POST_EX(SVM_EXIT_IOIO),
5960 [x86_intercept_ins] = POST_EX(SVM_EXIT_IOIO),
5961 [x86_intercept_out] = POST_EX(SVM_EXIT_IOIO),
5962 [x86_intercept_outs] = POST_EX(SVM_EXIT_IOIO),
5969 static int svm_check_intercept(struct kvm_vcpu *vcpu,
5970 struct x86_instruction_info *info,
5971 enum x86_intercept_stage stage)
5973 struct vcpu_svm *svm = to_svm(vcpu);
5974 int vmexit, ret = X86EMUL_CONTINUE;
5975 struct __x86_intercept icpt_info;
5976 struct vmcb *vmcb = svm->vmcb;
5978 if (info->intercept >= ARRAY_SIZE(x86_intercept_map))
5981 icpt_info = x86_intercept_map[info->intercept];
5983 if (stage != icpt_info.stage)
5986 switch (icpt_info.exit_code) {
5987 case SVM_EXIT_READ_CR0:
5988 if (info->intercept == x86_intercept_cr_read)
5989 icpt_info.exit_code += info->modrm_reg;
5991 case SVM_EXIT_WRITE_CR0: {
5992 unsigned long cr0, val;
5995 if (info->intercept == x86_intercept_cr_write)
5996 icpt_info.exit_code += info->modrm_reg;
5998 if (icpt_info.exit_code != SVM_EXIT_WRITE_CR0 ||
5999 info->intercept == x86_intercept_clts)
6002 intercept = svm->nested.intercept;
6004 if (!(intercept & (1ULL << INTERCEPT_SELECTIVE_CR0)))
6007 cr0 = vcpu->arch.cr0 & ~SVM_CR0_SELECTIVE_MASK;
6008 val = info->src_val & ~SVM_CR0_SELECTIVE_MASK;
6010 if (info->intercept == x86_intercept_lmsw) {
6013 /* lmsw can't clear PE - catch this here */
6014 if (cr0 & X86_CR0_PE)
6019 icpt_info.exit_code = SVM_EXIT_CR0_SEL_WRITE;
6023 case SVM_EXIT_READ_DR0:
6024 case SVM_EXIT_WRITE_DR0:
6025 icpt_info.exit_code += info->modrm_reg;
6028 if (info->intercept == x86_intercept_wrmsr)
6029 vmcb->control.exit_info_1 = 1;
6031 vmcb->control.exit_info_1 = 0;
6033 case SVM_EXIT_PAUSE:
6035 * We get this for NOP only, but pause
6036 * is rep not, check this here
6038 if (info->rep_prefix != REPE_PREFIX)
6041 case SVM_EXIT_IOIO: {
6045 if (info->intercept == x86_intercept_in ||
6046 info->intercept == x86_intercept_ins) {
6047 exit_info = ((info->src_val & 0xffff) << 16) |
6049 bytes = info->dst_bytes;
6051 exit_info = (info->dst_val & 0xffff) << 16;
6052 bytes = info->src_bytes;
6055 if (info->intercept == x86_intercept_outs ||
6056 info->intercept == x86_intercept_ins)
6057 exit_info |= SVM_IOIO_STR_MASK;
6059 if (info->rep_prefix)
6060 exit_info |= SVM_IOIO_REP_MASK;
6062 bytes = min(bytes, 4u);
6064 exit_info |= bytes << SVM_IOIO_SIZE_SHIFT;
6066 exit_info |= (u32)info->ad_bytes << (SVM_IOIO_ASIZE_SHIFT - 1);
6068 vmcb->control.exit_info_1 = exit_info;
6069 vmcb->control.exit_info_2 = info->next_rip;
6077 /* TODO: Advertise NRIPS to guest hypervisor unconditionally */
6078 if (static_cpu_has(X86_FEATURE_NRIPS))
6079 vmcb->control.next_rip = info->next_rip;
6080 vmcb->control.exit_code = icpt_info.exit_code;
6081 vmexit = nested_svm_exit_handled(svm);
6083 ret = (vmexit == NESTED_EXIT_DONE) ? X86EMUL_INTERCEPTED
6090 static void svm_handle_external_intr(struct kvm_vcpu *vcpu)
6094 * We must have an instruction with interrupts enabled, so
6095 * the timer interrupt isn't delayed by the interrupt shadow.
6098 local_irq_disable();
6101 static void svm_sched_in(struct kvm_vcpu *vcpu, int cpu)
6103 if (pause_filter_thresh)
6104 shrink_ple_window(vcpu);
6107 static inline void avic_post_state_restore(struct kvm_vcpu *vcpu)
6109 if (avic_handle_apic_id_update(vcpu) != 0)
6111 if (avic_handle_dfr_update(vcpu) != 0)
6113 avic_handle_ldr_update(vcpu);
6116 static void svm_setup_mce(struct kvm_vcpu *vcpu)
6118 /* [63:9] are reserved. */
6119 vcpu->arch.mcg_cap &= 0x1ff;
6122 static int svm_smi_allowed(struct kvm_vcpu *vcpu)
6124 struct vcpu_svm *svm = to_svm(vcpu);
6126 /* Per APM Vol.2 15.22.2 "Response to SMI" */
6130 if (is_guest_mode(&svm->vcpu) &&
6131 svm->nested.intercept & (1ULL << INTERCEPT_SMI)) {
6132 /* TODO: Might need to set exit_info_1 and exit_info_2 here */
6133 svm->vmcb->control.exit_code = SVM_EXIT_SMI;
6134 svm->nested.exit_required = true;
6141 static int svm_pre_enter_smm(struct kvm_vcpu *vcpu, char *smstate)
6143 struct vcpu_svm *svm = to_svm(vcpu);
6146 if (is_guest_mode(vcpu)) {
6147 /* FED8h - SVM Guest */
6148 put_smstate(u64, smstate, 0x7ed8, 1);
6149 /* FEE0h - SVM Guest VMCB Physical Address */
6150 put_smstate(u64, smstate, 0x7ee0, svm->nested.vmcb);
6152 svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX];
6153 svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP];
6154 svm->vmcb->save.rip = vcpu->arch.regs[VCPU_REGS_RIP];
6156 ret = nested_svm_vmexit(svm);
6163 static int svm_pre_leave_smm(struct kvm_vcpu *vcpu, u64 smbase)
6165 struct vcpu_svm *svm = to_svm(vcpu);
6166 struct vmcb *nested_vmcb;
6174 ret = kvm_vcpu_read_guest(vcpu, smbase + 0xfed8, &svm_state_save,
6175 sizeof(svm_state_save));
6179 if (svm_state_save.guest) {
6180 vcpu->arch.hflags &= ~HF_SMM_MASK;
6181 nested_vmcb = nested_svm_map(svm, svm_state_save.vmcb, &page);
6183 enter_svm_guest_mode(svm, svm_state_save.vmcb, nested_vmcb, page);
6186 vcpu->arch.hflags |= HF_SMM_MASK;
6191 static int enable_smi_window(struct kvm_vcpu *vcpu)
6193 struct vcpu_svm *svm = to_svm(vcpu);
6195 if (!gif_set(svm)) {
6196 if (vgif_enabled(svm))
6197 set_intercept(svm, INTERCEPT_STGI);
6198 /* STGI will cause a vm exit */
6204 static int sev_asid_new(void)
6209 * SEV-enabled guest must use asid from min_sev_asid to max_sev_asid.
6211 pos = find_next_zero_bit(sev_asid_bitmap, max_sev_asid, min_sev_asid - 1);
6212 if (pos >= max_sev_asid)
6215 set_bit(pos, sev_asid_bitmap);
6219 static int sev_guest_init(struct kvm *kvm, struct kvm_sev_cmd *argp)
6221 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
6225 asid = sev_asid_new();
6229 ret = sev_platform_init(&argp->error);
6235 INIT_LIST_HEAD(&sev->regions_list);
6240 __sev_asid_free(asid);
6244 static int sev_bind_asid(struct kvm *kvm, unsigned int handle, int *error)
6246 struct sev_data_activate *data;
6247 int asid = sev_get_asid(kvm);
6250 wbinvd_on_all_cpus();
6252 ret = sev_guest_df_flush(error);
6256 data = kzalloc(sizeof(*data), GFP_KERNEL);
6260 /* activate ASID on the given handle */
6261 data->handle = handle;
6263 ret = sev_guest_activate(data, error);
6269 static int __sev_issue_cmd(int fd, int id, void *data, int *error)
6278 ret = sev_issue_cmd_external_user(f.file, id, data, error);
6284 static int sev_issue_cmd(struct kvm *kvm, int id, void *data, int *error)
6286 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
6288 return __sev_issue_cmd(sev->fd, id, data, error);
6291 static int sev_launch_start(struct kvm *kvm, struct kvm_sev_cmd *argp)
6293 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
6294 struct sev_data_launch_start *start;
6295 struct kvm_sev_launch_start params;
6296 void *dh_blob, *session_blob;
6297 int *error = &argp->error;
6300 if (!sev_guest(kvm))
6303 if (copy_from_user(¶ms, (void __user *)(uintptr_t)argp->data, sizeof(params)))
6306 start = kzalloc(sizeof(*start), GFP_KERNEL);
6311 if (params.dh_uaddr) {
6312 dh_blob = psp_copy_user_blob(params.dh_uaddr, params.dh_len);
6313 if (IS_ERR(dh_blob)) {
6314 ret = PTR_ERR(dh_blob);
6318 start->dh_cert_address = __sme_set(__pa(dh_blob));
6319 start->dh_cert_len = params.dh_len;
6322 session_blob = NULL;
6323 if (params.session_uaddr) {
6324 session_blob = psp_copy_user_blob(params.session_uaddr, params.session_len);
6325 if (IS_ERR(session_blob)) {
6326 ret = PTR_ERR(session_blob);
6330 start->session_address = __sme_set(__pa(session_blob));
6331 start->session_len = params.session_len;
6334 start->handle = params.handle;
6335 start->policy = params.policy;
6337 /* create memory encryption context */
6338 ret = __sev_issue_cmd(argp->sev_fd, SEV_CMD_LAUNCH_START, start, error);
6340 goto e_free_session;
6342 /* Bind ASID to this guest */
6343 ret = sev_bind_asid(kvm, start->handle, error);
6345 goto e_free_session;
6347 /* return handle to userspace */
6348 params.handle = start->handle;
6349 if (copy_to_user((void __user *)(uintptr_t)argp->data, ¶ms, sizeof(params))) {
6350 sev_unbind_asid(kvm, start->handle);
6352 goto e_free_session;
6355 sev->handle = start->handle;
6356 sev->fd = argp->sev_fd;
6359 kfree(session_blob);
6367 static int get_num_contig_pages(int idx, struct page **inpages,
6368 unsigned long npages)
6370 unsigned long paddr, next_paddr;
6371 int i = idx + 1, pages = 1;
6373 /* find the number of contiguous pages starting from idx */
6374 paddr = __sme_page_pa(inpages[idx]);
6375 while (i < npages) {
6376 next_paddr = __sme_page_pa(inpages[i++]);
6377 if ((paddr + PAGE_SIZE) == next_paddr) {
6388 static int sev_launch_update_data(struct kvm *kvm, struct kvm_sev_cmd *argp)
6390 unsigned long vaddr, vaddr_end, next_vaddr, npages, size;
6391 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
6392 struct kvm_sev_launch_update_data params;
6393 struct sev_data_launch_update_data *data;
6394 struct page **inpages;
6397 if (!sev_guest(kvm))
6400 if (copy_from_user(¶ms, (void __user *)(uintptr_t)argp->data, sizeof(params)))
6403 data = kzalloc(sizeof(*data), GFP_KERNEL);
6407 vaddr = params.uaddr;
6409 vaddr_end = vaddr + size;
6411 /* Lock the user memory. */
6412 inpages = sev_pin_memory(kvm, vaddr, size, &npages, 1);
6419 * The LAUNCH_UPDATE command will perform in-place encryption of the
6420 * memory content (i.e it will write the same memory region with C=1).
6421 * It's possible that the cache may contain the data with C=0, i.e.,
6422 * unencrypted so invalidate it first.
6424 sev_clflush_pages(inpages, npages);
6426 for (i = 0; vaddr < vaddr_end; vaddr = next_vaddr, i += pages) {
6430 * If the user buffer is not page-aligned, calculate the offset
6433 offset = vaddr & (PAGE_SIZE - 1);
6435 /* Calculate the number of pages that can be encrypted in one go. */
6436 pages = get_num_contig_pages(i, inpages, npages);
6438 len = min_t(size_t, ((pages * PAGE_SIZE) - offset), size);
6440 data->handle = sev->handle;
6442 data->address = __sme_page_pa(inpages[i]) + offset;
6443 ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_UPDATE_DATA, data, &argp->error);
6448 next_vaddr = vaddr + len;
6452 /* content of memory is updated, mark pages dirty */
6453 for (i = 0; i < npages; i++) {
6454 set_page_dirty_lock(inpages[i]);
6455 mark_page_accessed(inpages[i]);
6457 /* unlock the user pages */
6458 sev_unpin_memory(kvm, inpages, npages);
6464 static int sev_launch_measure(struct kvm *kvm, struct kvm_sev_cmd *argp)
6466 void __user *measure = (void __user *)(uintptr_t)argp->data;
6467 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
6468 struct sev_data_launch_measure *data;
6469 struct kvm_sev_launch_measure params;
6470 void __user *p = NULL;
6474 if (!sev_guest(kvm))
6477 if (copy_from_user(¶ms, measure, sizeof(params)))
6480 data = kzalloc(sizeof(*data), GFP_KERNEL);
6484 /* User wants to query the blob length */
6488 p = (void __user *)(uintptr_t)params.uaddr;
6490 if (params.len > SEV_FW_BLOB_MAX_SIZE) {
6496 blob = kmalloc(params.len, GFP_KERNEL);
6500 data->address = __psp_pa(blob);
6501 data->len = params.len;
6505 data->handle = sev->handle;
6506 ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_MEASURE, data, &argp->error);
6509 * If we query the session length, FW responded with expected data.
6518 if (copy_to_user(p, blob, params.len))
6523 params.len = data->len;
6524 if (copy_to_user(measure, ¶ms, sizeof(params)))
6533 static int sev_launch_finish(struct kvm *kvm, struct kvm_sev_cmd *argp)
6535 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
6536 struct sev_data_launch_finish *data;
6539 if (!sev_guest(kvm))
6542 data = kzalloc(sizeof(*data), GFP_KERNEL);
6546 data->handle = sev->handle;
6547 ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_FINISH, data, &argp->error);
6553 static int sev_guest_status(struct kvm *kvm, struct kvm_sev_cmd *argp)
6555 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
6556 struct kvm_sev_guest_status params;
6557 struct sev_data_guest_status *data;
6560 if (!sev_guest(kvm))
6563 data = kzalloc(sizeof(*data), GFP_KERNEL);
6567 data->handle = sev->handle;
6568 ret = sev_issue_cmd(kvm, SEV_CMD_GUEST_STATUS, data, &argp->error);
6572 params.policy = data->policy;
6573 params.state = data->state;
6574 params.handle = data->handle;
6576 if (copy_to_user((void __user *)(uintptr_t)argp->data, ¶ms, sizeof(params)))
6583 static int __sev_issue_dbg_cmd(struct kvm *kvm, unsigned long src,
6584 unsigned long dst, int size,
6585 int *error, bool enc)
6587 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
6588 struct sev_data_dbg *data;
6591 data = kzalloc(sizeof(*data), GFP_KERNEL);
6595 data->handle = sev->handle;
6596 data->dst_addr = dst;
6597 data->src_addr = src;
6600 ret = sev_issue_cmd(kvm,
6601 enc ? SEV_CMD_DBG_ENCRYPT : SEV_CMD_DBG_DECRYPT,
6607 static int __sev_dbg_decrypt(struct kvm *kvm, unsigned long src_paddr,
6608 unsigned long dst_paddr, int sz, int *err)
6613 * Its safe to read more than we are asked, caller should ensure that
6614 * destination has enough space.
6616 src_paddr = round_down(src_paddr, 16);
6617 offset = src_paddr & 15;
6618 sz = round_up(sz + offset, 16);
6620 return __sev_issue_dbg_cmd(kvm, src_paddr, dst_paddr, sz, err, false);
6623 static int __sev_dbg_decrypt_user(struct kvm *kvm, unsigned long paddr,
6624 unsigned long __user dst_uaddr,
6625 unsigned long dst_paddr,
6628 struct page *tpage = NULL;
6631 /* if inputs are not 16-byte then use intermediate buffer */
6632 if (!IS_ALIGNED(dst_paddr, 16) ||
6633 !IS_ALIGNED(paddr, 16) ||
6634 !IS_ALIGNED(size, 16)) {
6635 tpage = (void *)alloc_page(GFP_KERNEL);
6639 dst_paddr = __sme_page_pa(tpage);
6642 ret = __sev_dbg_decrypt(kvm, paddr, dst_paddr, size, err);
6647 offset = paddr & 15;
6648 if (copy_to_user((void __user *)(uintptr_t)dst_uaddr,
6649 page_address(tpage) + offset, size))
6660 static int __sev_dbg_encrypt_user(struct kvm *kvm, unsigned long paddr,
6661 unsigned long __user vaddr,
6662 unsigned long dst_paddr,
6663 unsigned long __user dst_vaddr,
6664 int size, int *error)
6666 struct page *src_tpage = NULL;
6667 struct page *dst_tpage = NULL;
6668 int ret, len = size;
6670 /* If source buffer is not aligned then use an intermediate buffer */
6671 if (!IS_ALIGNED(vaddr, 16)) {
6672 src_tpage = alloc_page(GFP_KERNEL);
6676 if (copy_from_user(page_address(src_tpage),
6677 (void __user *)(uintptr_t)vaddr, size)) {
6678 __free_page(src_tpage);
6682 paddr = __sme_page_pa(src_tpage);
6686 * If destination buffer or length is not aligned then do read-modify-write:
6687 * - decrypt destination in an intermediate buffer
6688 * - copy the source buffer in an intermediate buffer
6689 * - use the intermediate buffer as source buffer
6691 if (!IS_ALIGNED(dst_vaddr, 16) || !IS_ALIGNED(size, 16)) {
6694 dst_tpage = alloc_page(GFP_KERNEL);
6700 ret = __sev_dbg_decrypt(kvm, dst_paddr,
6701 __sme_page_pa(dst_tpage), size, error);
6706 * If source is kernel buffer then use memcpy() otherwise
6709 dst_offset = dst_paddr & 15;
6712 memcpy(page_address(dst_tpage) + dst_offset,
6713 page_address(src_tpage), size);
6715 if (copy_from_user(page_address(dst_tpage) + dst_offset,
6716 (void __user *)(uintptr_t)vaddr, size)) {
6722 paddr = __sme_page_pa(dst_tpage);
6723 dst_paddr = round_down(dst_paddr, 16);
6724 len = round_up(size, 16);
6727 ret = __sev_issue_dbg_cmd(kvm, paddr, dst_paddr, len, error, true);
6731 __free_page(src_tpage);
6733 __free_page(dst_tpage);
6737 static int sev_dbg_crypt(struct kvm *kvm, struct kvm_sev_cmd *argp, bool dec)
6739 unsigned long vaddr, vaddr_end, next_vaddr;
6740 unsigned long dst_vaddr, dst_vaddr_end;
6741 struct page **src_p, **dst_p;
6742 struct kvm_sev_dbg debug;
6746 if (!sev_guest(kvm))
6749 if (copy_from_user(&debug, (void __user *)(uintptr_t)argp->data, sizeof(debug)))
6752 vaddr = debug.src_uaddr;
6754 vaddr_end = vaddr + size;
6755 dst_vaddr = debug.dst_uaddr;
6756 dst_vaddr_end = dst_vaddr + size;
6758 for (; vaddr < vaddr_end; vaddr = next_vaddr) {
6759 int len, s_off, d_off;
6761 /* lock userspace source and destination page */
6762 src_p = sev_pin_memory(kvm, vaddr & PAGE_MASK, PAGE_SIZE, &n, 0);
6766 dst_p = sev_pin_memory(kvm, dst_vaddr & PAGE_MASK, PAGE_SIZE, &n, 1);
6768 sev_unpin_memory(kvm, src_p, n);
6773 * The DBG_{DE,EN}CRYPT commands will perform {dec,en}cryption of the
6774 * memory content (i.e it will write the same memory region with C=1).
6775 * It's possible that the cache may contain the data with C=0, i.e.,
6776 * unencrypted so invalidate it first.
6778 sev_clflush_pages(src_p, 1);
6779 sev_clflush_pages(dst_p, 1);
6782 * Since user buffer may not be page aligned, calculate the
6783 * offset within the page.
6785 s_off = vaddr & ~PAGE_MASK;
6786 d_off = dst_vaddr & ~PAGE_MASK;
6787 len = min_t(size_t, (PAGE_SIZE - s_off), size);
6790 ret = __sev_dbg_decrypt_user(kvm,
6791 __sme_page_pa(src_p[0]) + s_off,
6793 __sme_page_pa(dst_p[0]) + d_off,
6796 ret = __sev_dbg_encrypt_user(kvm,
6797 __sme_page_pa(src_p[0]) + s_off,
6799 __sme_page_pa(dst_p[0]) + d_off,
6803 sev_unpin_memory(kvm, src_p, 1);
6804 sev_unpin_memory(kvm, dst_p, 1);
6809 next_vaddr = vaddr + len;
6810 dst_vaddr = dst_vaddr + len;
6817 static int sev_launch_secret(struct kvm *kvm, struct kvm_sev_cmd *argp)
6819 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
6820 struct sev_data_launch_secret *data;
6821 struct kvm_sev_launch_secret params;
6822 struct page **pages;
6827 if (!sev_guest(kvm))
6830 if (copy_from_user(¶ms, (void __user *)(uintptr_t)argp->data, sizeof(params)))
6833 pages = sev_pin_memory(kvm, params.guest_uaddr, params.guest_len, &n, 1);
6838 * The secret must be copied into contiguous memory region, lets verify
6839 * that userspace memory pages are contiguous before we issue command.
6841 if (get_num_contig_pages(0, pages, n) != n) {
6843 goto e_unpin_memory;
6847 data = kzalloc(sizeof(*data), GFP_KERNEL);
6849 goto e_unpin_memory;
6851 offset = params.guest_uaddr & (PAGE_SIZE - 1);
6852 data->guest_address = __sme_page_pa(pages[0]) + offset;
6853 data->guest_len = params.guest_len;
6855 blob = psp_copy_user_blob(params.trans_uaddr, params.trans_len);
6857 ret = PTR_ERR(blob);
6861 data->trans_address = __psp_pa(blob);
6862 data->trans_len = params.trans_len;
6864 hdr = psp_copy_user_blob(params.hdr_uaddr, params.hdr_len);
6869 data->hdr_address = __psp_pa(hdr);
6870 data->hdr_len = params.hdr_len;
6872 data->handle = sev->handle;
6873 ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_UPDATE_SECRET, data, &argp->error);
6882 sev_unpin_memory(kvm, pages, n);
6886 static int svm_mem_enc_op(struct kvm *kvm, void __user *argp)
6888 struct kvm_sev_cmd sev_cmd;
6891 if (!svm_sev_enabled())
6894 if (copy_from_user(&sev_cmd, argp, sizeof(struct kvm_sev_cmd)))
6897 mutex_lock(&kvm->lock);
6899 switch (sev_cmd.id) {
6901 r = sev_guest_init(kvm, &sev_cmd);
6903 case KVM_SEV_LAUNCH_START:
6904 r = sev_launch_start(kvm, &sev_cmd);
6906 case KVM_SEV_LAUNCH_UPDATE_DATA:
6907 r = sev_launch_update_data(kvm, &sev_cmd);
6909 case KVM_SEV_LAUNCH_MEASURE:
6910 r = sev_launch_measure(kvm, &sev_cmd);
6912 case KVM_SEV_LAUNCH_FINISH:
6913 r = sev_launch_finish(kvm, &sev_cmd);
6915 case KVM_SEV_GUEST_STATUS:
6916 r = sev_guest_status(kvm, &sev_cmd);
6918 case KVM_SEV_DBG_DECRYPT:
6919 r = sev_dbg_crypt(kvm, &sev_cmd, true);
6921 case KVM_SEV_DBG_ENCRYPT:
6922 r = sev_dbg_crypt(kvm, &sev_cmd, false);
6924 case KVM_SEV_LAUNCH_SECRET:
6925 r = sev_launch_secret(kvm, &sev_cmd);
6932 if (copy_to_user(argp, &sev_cmd, sizeof(struct kvm_sev_cmd)))
6936 mutex_unlock(&kvm->lock);
6940 static int svm_register_enc_region(struct kvm *kvm,
6941 struct kvm_enc_region *range)
6943 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
6944 struct enc_region *region;
6947 if (!sev_guest(kvm))
6950 region = kzalloc(sizeof(*region), GFP_KERNEL);
6954 region->pages = sev_pin_memory(kvm, range->addr, range->size, ®ion->npages, 1);
6955 if (!region->pages) {
6961 * The guest may change the memory encryption attribute from C=0 -> C=1
6962 * or vice versa for this memory range. Lets make sure caches are
6963 * flushed to ensure that guest data gets written into memory with
6966 sev_clflush_pages(region->pages, region->npages);
6968 region->uaddr = range->addr;
6969 region->size = range->size;
6971 mutex_lock(&kvm->lock);
6972 list_add_tail(®ion->list, &sev->regions_list);
6973 mutex_unlock(&kvm->lock);
6982 static struct enc_region *
6983 find_enc_region(struct kvm *kvm, struct kvm_enc_region *range)
6985 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
6986 struct list_head *head = &sev->regions_list;
6987 struct enc_region *i;
6989 list_for_each_entry(i, head, list) {
6990 if (i->uaddr == range->addr &&
6991 i->size == range->size)
6999 static int svm_unregister_enc_region(struct kvm *kvm,
7000 struct kvm_enc_region *range)
7002 struct enc_region *region;
7005 mutex_lock(&kvm->lock);
7007 if (!sev_guest(kvm)) {
7012 region = find_enc_region(kvm, range);
7018 __unregister_enc_region_locked(kvm, region);
7020 mutex_unlock(&kvm->lock);
7024 mutex_unlock(&kvm->lock);
7028 static struct kvm_x86_ops svm_x86_ops __ro_after_init = {
7029 .cpu_has_kvm_support = has_svm,
7030 .disabled_by_bios = is_disabled,
7031 .hardware_setup = svm_hardware_setup,
7032 .hardware_unsetup = svm_hardware_unsetup,
7033 .check_processor_compatibility = svm_check_processor_compat,
7034 .hardware_enable = svm_hardware_enable,
7035 .hardware_disable = svm_hardware_disable,
7036 .cpu_has_accelerated_tpr = svm_cpu_has_accelerated_tpr,
7037 .has_emulated_msr = svm_has_emulated_msr,
7039 .vcpu_create = svm_create_vcpu,
7040 .vcpu_free = svm_free_vcpu,
7041 .vcpu_reset = svm_vcpu_reset,
7043 .vm_alloc = svm_vm_alloc,
7044 .vm_free = svm_vm_free,
7045 .vm_init = avic_vm_init,
7046 .vm_destroy = svm_vm_destroy,
7048 .prepare_guest_switch = svm_prepare_guest_switch,
7049 .vcpu_load = svm_vcpu_load,
7050 .vcpu_put = svm_vcpu_put,
7051 .vcpu_blocking = svm_vcpu_blocking,
7052 .vcpu_unblocking = svm_vcpu_unblocking,
7054 .update_bp_intercept = update_bp_intercept,
7055 .get_msr_feature = svm_get_msr_feature,
7056 .get_msr = svm_get_msr,
7057 .set_msr = svm_set_msr,
7058 .get_segment_base = svm_get_segment_base,
7059 .get_segment = svm_get_segment,
7060 .set_segment = svm_set_segment,
7061 .get_cpl = svm_get_cpl,
7062 .get_cs_db_l_bits = kvm_get_cs_db_l_bits,
7063 .decache_cr0_guest_bits = svm_decache_cr0_guest_bits,
7064 .decache_cr3 = svm_decache_cr3,
7065 .decache_cr4_guest_bits = svm_decache_cr4_guest_bits,
7066 .set_cr0 = svm_set_cr0,
7067 .set_cr3 = svm_set_cr3,
7068 .set_cr4 = svm_set_cr4,
7069 .set_efer = svm_set_efer,
7070 .get_idt = svm_get_idt,
7071 .set_idt = svm_set_idt,
7072 .get_gdt = svm_get_gdt,
7073 .set_gdt = svm_set_gdt,
7074 .get_dr6 = svm_get_dr6,
7075 .set_dr6 = svm_set_dr6,
7076 .set_dr7 = svm_set_dr7,
7077 .sync_dirty_debug_regs = svm_sync_dirty_debug_regs,
7078 .cache_reg = svm_cache_reg,
7079 .get_rflags = svm_get_rflags,
7080 .set_rflags = svm_set_rflags,
7082 .tlb_flush = svm_flush_tlb,
7084 .run = svm_vcpu_run,
7085 .handle_exit = handle_exit,
7086 .skip_emulated_instruction = skip_emulated_instruction,
7087 .set_interrupt_shadow = svm_set_interrupt_shadow,
7088 .get_interrupt_shadow = svm_get_interrupt_shadow,
7089 .patch_hypercall = svm_patch_hypercall,
7090 .set_irq = svm_set_irq,
7091 .set_nmi = svm_inject_nmi,
7092 .queue_exception = svm_queue_exception,
7093 .cancel_injection = svm_cancel_injection,
7094 .interrupt_allowed = svm_interrupt_allowed,
7095 .nmi_allowed = svm_nmi_allowed,
7096 .get_nmi_mask = svm_get_nmi_mask,
7097 .set_nmi_mask = svm_set_nmi_mask,
7098 .enable_nmi_window = enable_nmi_window,
7099 .enable_irq_window = enable_irq_window,
7100 .update_cr8_intercept = update_cr8_intercept,
7101 .set_virtual_x2apic_mode = svm_set_virtual_x2apic_mode,
7102 .get_enable_apicv = svm_get_enable_apicv,
7103 .refresh_apicv_exec_ctrl = svm_refresh_apicv_exec_ctrl,
7104 .load_eoi_exitmap = svm_load_eoi_exitmap,
7105 .hwapic_irr_update = svm_hwapic_irr_update,
7106 .hwapic_isr_update = svm_hwapic_isr_update,
7107 .sync_pir_to_irr = kvm_lapic_find_highest_irr,
7108 .apicv_post_state_restore = avic_post_state_restore,
7110 .set_tss_addr = svm_set_tss_addr,
7111 .set_identity_map_addr = svm_set_identity_map_addr,
7112 .get_tdp_level = get_npt_level,
7113 .get_mt_mask = svm_get_mt_mask,
7115 .get_exit_info = svm_get_exit_info,
7117 .get_lpage_level = svm_get_lpage_level,
7119 .cpuid_update = svm_cpuid_update,
7121 .rdtscp_supported = svm_rdtscp_supported,
7122 .invpcid_supported = svm_invpcid_supported,
7123 .mpx_supported = svm_mpx_supported,
7124 .xsaves_supported = svm_xsaves_supported,
7125 .umip_emulated = svm_umip_emulated,
7127 .set_supported_cpuid = svm_set_supported_cpuid,
7129 .has_wbinvd_exit = svm_has_wbinvd_exit,
7131 .read_l1_tsc_offset = svm_read_l1_tsc_offset,
7132 .write_tsc_offset = svm_write_tsc_offset,
7134 .set_tdp_cr3 = set_tdp_cr3,
7136 .check_intercept = svm_check_intercept,
7137 .handle_external_intr = svm_handle_external_intr,
7139 .sched_in = svm_sched_in,
7141 .pmu_ops = &amd_pmu_ops,
7142 .deliver_posted_interrupt = svm_deliver_avic_intr,
7143 .update_pi_irte = svm_update_pi_irte,
7144 .setup_mce = svm_setup_mce,
7146 .smi_allowed = svm_smi_allowed,
7147 .pre_enter_smm = svm_pre_enter_smm,
7148 .pre_leave_smm = svm_pre_leave_smm,
7149 .enable_smi_window = enable_smi_window,
7151 .mem_enc_op = svm_mem_enc_op,
7152 .mem_enc_reg_region = svm_register_enc_region,
7153 .mem_enc_unreg_region = svm_unregister_enc_region,
7156 static int __init svm_init(void)
7158 return kvm_init(&svm_x86_ops, sizeof(struct vcpu_svm),
7159 __alignof__(struct vcpu_svm), THIS_MODULE);
7162 static void __exit svm_exit(void)
7167 module_init(svm_init)
7168 module_exit(svm_exit)