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_array(max_sev_asid + 1,
1011 per_cpu(svm_data, cpu) = sd;
1021 static bool valid_msr_intercept(u32 index)
1025 for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++)
1026 if (direct_access_msrs[i].index == index)
1032 static bool msr_write_intercepted(struct kvm_vcpu *vcpu, unsigned msr)
1039 msrpm = is_guest_mode(vcpu) ? to_svm(vcpu)->nested.msrpm:
1040 to_svm(vcpu)->msrpm;
1042 offset = svm_msrpm_offset(msr);
1043 bit_write = 2 * (msr & 0x0f) + 1;
1044 tmp = msrpm[offset];
1046 BUG_ON(offset == MSR_INVALID);
1048 return !!test_bit(bit_write, &tmp);
1051 static void set_msr_interception(u32 *msrpm, unsigned msr,
1052 int read, int write)
1054 u8 bit_read, bit_write;
1059 * If this warning triggers extend the direct_access_msrs list at the
1060 * beginning of the file
1062 WARN_ON(!valid_msr_intercept(msr));
1064 offset = svm_msrpm_offset(msr);
1065 bit_read = 2 * (msr & 0x0f);
1066 bit_write = 2 * (msr & 0x0f) + 1;
1067 tmp = msrpm[offset];
1069 BUG_ON(offset == MSR_INVALID);
1071 read ? clear_bit(bit_read, &tmp) : set_bit(bit_read, &tmp);
1072 write ? clear_bit(bit_write, &tmp) : set_bit(bit_write, &tmp);
1074 msrpm[offset] = tmp;
1077 static void svm_vcpu_init_msrpm(u32 *msrpm)
1081 memset(msrpm, 0xff, PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER));
1083 for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
1084 if (!direct_access_msrs[i].always)
1087 set_msr_interception(msrpm, direct_access_msrs[i].index, 1, 1);
1091 static void add_msr_offset(u32 offset)
1095 for (i = 0; i < MSRPM_OFFSETS; ++i) {
1097 /* Offset already in list? */
1098 if (msrpm_offsets[i] == offset)
1101 /* Slot used by another offset? */
1102 if (msrpm_offsets[i] != MSR_INVALID)
1105 /* Add offset to list */
1106 msrpm_offsets[i] = offset;
1112 * If this BUG triggers the msrpm_offsets table has an overflow. Just
1113 * increase MSRPM_OFFSETS in this case.
1118 static void init_msrpm_offsets(void)
1122 memset(msrpm_offsets, 0xff, sizeof(msrpm_offsets));
1124 for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
1127 offset = svm_msrpm_offset(direct_access_msrs[i].index);
1128 BUG_ON(offset == MSR_INVALID);
1130 add_msr_offset(offset);
1134 static void svm_enable_lbrv(struct vcpu_svm *svm)
1136 u32 *msrpm = svm->msrpm;
1138 svm->vmcb->control.virt_ext |= LBR_CTL_ENABLE_MASK;
1139 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 1, 1);
1140 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 1, 1);
1141 set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 1, 1);
1142 set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 1, 1);
1145 static void svm_disable_lbrv(struct vcpu_svm *svm)
1147 u32 *msrpm = svm->msrpm;
1149 svm->vmcb->control.virt_ext &= ~LBR_CTL_ENABLE_MASK;
1150 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 0, 0);
1151 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 0, 0);
1152 set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 0, 0);
1153 set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 0, 0);
1156 static void disable_nmi_singlestep(struct vcpu_svm *svm)
1158 svm->nmi_singlestep = false;
1160 if (!(svm->vcpu.guest_debug & KVM_GUESTDBG_SINGLESTEP)) {
1161 /* Clear our flags if they were not set by the guest */
1162 if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_TF))
1163 svm->vmcb->save.rflags &= ~X86_EFLAGS_TF;
1164 if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_RF))
1165 svm->vmcb->save.rflags &= ~X86_EFLAGS_RF;
1170 * This hash table is used to map VM_ID to a struct kvm_svm,
1171 * when handling AMD IOMMU GALOG notification to schedule in
1172 * a particular vCPU.
1174 #define SVM_VM_DATA_HASH_BITS 8
1175 static DEFINE_HASHTABLE(svm_vm_data_hash, SVM_VM_DATA_HASH_BITS);
1176 static u32 next_vm_id = 0;
1177 static bool next_vm_id_wrapped = 0;
1178 static DEFINE_SPINLOCK(svm_vm_data_hash_lock);
1181 * This function is called from IOMMU driver to notify
1182 * SVM to schedule in a particular vCPU of a particular VM.
1184 static int avic_ga_log_notifier(u32 ga_tag)
1186 unsigned long flags;
1187 struct kvm_svm *kvm_svm;
1188 struct kvm_vcpu *vcpu = NULL;
1189 u32 vm_id = AVIC_GATAG_TO_VMID(ga_tag);
1190 u32 vcpu_id = AVIC_GATAG_TO_VCPUID(ga_tag);
1192 pr_debug("SVM: %s: vm_id=%#x, vcpu_id=%#x\n", __func__, vm_id, vcpu_id);
1194 spin_lock_irqsave(&svm_vm_data_hash_lock, flags);
1195 hash_for_each_possible(svm_vm_data_hash, kvm_svm, hnode, vm_id) {
1196 if (kvm_svm->avic_vm_id != vm_id)
1198 vcpu = kvm_get_vcpu_by_id(&kvm_svm->kvm, vcpu_id);
1201 spin_unlock_irqrestore(&svm_vm_data_hash_lock, flags);
1204 * At this point, the IOMMU should have already set the pending
1205 * bit in the vAPIC backing page. So, we just need to schedule
1209 kvm_vcpu_wake_up(vcpu);
1214 static __init int sev_hardware_setup(void)
1216 struct sev_user_data_status *status;
1219 /* Maximum number of encrypted guests supported simultaneously */
1220 max_sev_asid = cpuid_ecx(0x8000001F);
1225 /* Minimum ASID value that should be used for SEV guest */
1226 min_sev_asid = cpuid_edx(0x8000001F);
1228 /* Initialize SEV ASID bitmap */
1229 sev_asid_bitmap = kcalloc(BITS_TO_LONGS(max_sev_asid),
1230 sizeof(unsigned long), GFP_KERNEL);
1231 if (!sev_asid_bitmap)
1234 status = kmalloc(sizeof(*status), GFP_KERNEL);
1239 * Check SEV platform status.
1241 * PLATFORM_STATUS can be called in any state, if we failed to query
1242 * the PLATFORM status then either PSP firmware does not support SEV
1243 * feature or SEV firmware is dead.
1245 rc = sev_platform_status(status, NULL);
1249 pr_info("SEV supported\n");
1256 static void grow_ple_window(struct kvm_vcpu *vcpu)
1258 struct vcpu_svm *svm = to_svm(vcpu);
1259 struct vmcb_control_area *control = &svm->vmcb->control;
1260 int old = control->pause_filter_count;
1262 control->pause_filter_count = __grow_ple_window(old,
1264 pause_filter_count_grow,
1265 pause_filter_count_max);
1267 if (control->pause_filter_count != old)
1268 mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
1270 trace_kvm_ple_window_grow(vcpu->vcpu_id,
1271 control->pause_filter_count, old);
1274 static void shrink_ple_window(struct kvm_vcpu *vcpu)
1276 struct vcpu_svm *svm = to_svm(vcpu);
1277 struct vmcb_control_area *control = &svm->vmcb->control;
1278 int old = control->pause_filter_count;
1280 control->pause_filter_count =
1281 __shrink_ple_window(old,
1283 pause_filter_count_shrink,
1284 pause_filter_count);
1285 if (control->pause_filter_count != old)
1286 mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
1288 trace_kvm_ple_window_shrink(vcpu->vcpu_id,
1289 control->pause_filter_count, old);
1292 static __init int svm_hardware_setup(void)
1295 struct page *iopm_pages;
1299 iopm_pages = alloc_pages(GFP_KERNEL, IOPM_ALLOC_ORDER);
1304 iopm_va = page_address(iopm_pages);
1305 memset(iopm_va, 0xff, PAGE_SIZE * (1 << IOPM_ALLOC_ORDER));
1306 iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT;
1308 init_msrpm_offsets();
1310 if (boot_cpu_has(X86_FEATURE_NX))
1311 kvm_enable_efer_bits(EFER_NX);
1313 if (boot_cpu_has(X86_FEATURE_FXSR_OPT))
1314 kvm_enable_efer_bits(EFER_FFXSR);
1316 if (boot_cpu_has(X86_FEATURE_TSCRATEMSR)) {
1317 kvm_has_tsc_control = true;
1318 kvm_max_tsc_scaling_ratio = TSC_RATIO_MAX;
1319 kvm_tsc_scaling_ratio_frac_bits = 32;
1322 /* Check for pause filtering support */
1323 if (!boot_cpu_has(X86_FEATURE_PAUSEFILTER)) {
1324 pause_filter_count = 0;
1325 pause_filter_thresh = 0;
1326 } else if (!boot_cpu_has(X86_FEATURE_PFTHRESHOLD)) {
1327 pause_filter_thresh = 0;
1331 printk(KERN_INFO "kvm: Nested Virtualization enabled\n");
1332 kvm_enable_efer_bits(EFER_SVME | EFER_LMSLE);
1336 if (boot_cpu_has(X86_FEATURE_SEV) &&
1337 IS_ENABLED(CONFIG_KVM_AMD_SEV)) {
1338 r = sev_hardware_setup();
1346 for_each_possible_cpu(cpu) {
1347 r = svm_cpu_init(cpu);
1352 if (!boot_cpu_has(X86_FEATURE_NPT))
1353 npt_enabled = false;
1355 if (npt_enabled && !npt) {
1356 printk(KERN_INFO "kvm: Nested Paging disabled\n");
1357 npt_enabled = false;
1361 printk(KERN_INFO "kvm: Nested Paging enabled\n");
1368 !boot_cpu_has(X86_FEATURE_AVIC) ||
1369 !IS_ENABLED(CONFIG_X86_LOCAL_APIC)) {
1372 pr_info("AVIC enabled\n");
1374 amd_iommu_register_ga_log_notifier(&avic_ga_log_notifier);
1380 !boot_cpu_has(X86_FEATURE_V_VMSAVE_VMLOAD) ||
1381 !IS_ENABLED(CONFIG_X86_64)) {
1384 pr_info("Virtual VMLOAD VMSAVE supported\n");
1389 if (!boot_cpu_has(X86_FEATURE_VGIF))
1392 pr_info("Virtual GIF supported\n");
1398 __free_pages(iopm_pages, IOPM_ALLOC_ORDER);
1403 static __exit void svm_hardware_unsetup(void)
1407 if (svm_sev_enabled())
1408 kfree(sev_asid_bitmap);
1410 for_each_possible_cpu(cpu)
1411 svm_cpu_uninit(cpu);
1413 __free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT), IOPM_ALLOC_ORDER);
1417 static void init_seg(struct vmcb_seg *seg)
1420 seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK |
1421 SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */
1422 seg->limit = 0xffff;
1426 static void init_sys_seg(struct vmcb_seg *seg, uint32_t type)
1429 seg->attrib = SVM_SELECTOR_P_MASK | type;
1430 seg->limit = 0xffff;
1434 static u64 svm_read_l1_tsc_offset(struct kvm_vcpu *vcpu)
1436 struct vcpu_svm *svm = to_svm(vcpu);
1438 if (is_guest_mode(vcpu))
1439 return svm->nested.hsave->control.tsc_offset;
1441 return vcpu->arch.tsc_offset;
1444 static void svm_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1446 struct vcpu_svm *svm = to_svm(vcpu);
1447 u64 g_tsc_offset = 0;
1449 if (is_guest_mode(vcpu)) {
1450 /* Write L1's TSC offset. */
1451 g_tsc_offset = svm->vmcb->control.tsc_offset -
1452 svm->nested.hsave->control.tsc_offset;
1453 svm->nested.hsave->control.tsc_offset = offset;
1455 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
1456 svm->vmcb->control.tsc_offset,
1459 svm->vmcb->control.tsc_offset = offset + g_tsc_offset;
1461 mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
1464 static void avic_init_vmcb(struct vcpu_svm *svm)
1466 struct vmcb *vmcb = svm->vmcb;
1467 struct kvm_svm *kvm_svm = to_kvm_svm(svm->vcpu.kvm);
1468 phys_addr_t bpa = __sme_set(page_to_phys(svm->avic_backing_page));
1469 phys_addr_t lpa = __sme_set(page_to_phys(kvm_svm->avic_logical_id_table_page));
1470 phys_addr_t ppa = __sme_set(page_to_phys(kvm_svm->avic_physical_id_table_page));
1472 vmcb->control.avic_backing_page = bpa & AVIC_HPA_MASK;
1473 vmcb->control.avic_logical_id = lpa & AVIC_HPA_MASK;
1474 vmcb->control.avic_physical_id = ppa & AVIC_HPA_MASK;
1475 vmcb->control.avic_physical_id |= AVIC_MAX_PHYSICAL_ID_COUNT;
1476 vmcb->control.int_ctl |= AVIC_ENABLE_MASK;
1479 static void init_vmcb(struct vcpu_svm *svm)
1481 struct vmcb_control_area *control = &svm->vmcb->control;
1482 struct vmcb_save_area *save = &svm->vmcb->save;
1484 svm->vcpu.arch.hflags = 0;
1486 set_cr_intercept(svm, INTERCEPT_CR0_READ);
1487 set_cr_intercept(svm, INTERCEPT_CR3_READ);
1488 set_cr_intercept(svm, INTERCEPT_CR4_READ);
1489 set_cr_intercept(svm, INTERCEPT_CR0_WRITE);
1490 set_cr_intercept(svm, INTERCEPT_CR3_WRITE);
1491 set_cr_intercept(svm, INTERCEPT_CR4_WRITE);
1492 if (!kvm_vcpu_apicv_active(&svm->vcpu))
1493 set_cr_intercept(svm, INTERCEPT_CR8_WRITE);
1495 set_dr_intercepts(svm);
1497 set_exception_intercept(svm, PF_VECTOR);
1498 set_exception_intercept(svm, UD_VECTOR);
1499 set_exception_intercept(svm, MC_VECTOR);
1500 set_exception_intercept(svm, AC_VECTOR);
1501 set_exception_intercept(svm, DB_VECTOR);
1503 * Guest access to VMware backdoor ports could legitimately
1504 * trigger #GP because of TSS I/O permission bitmap.
1505 * We intercept those #GP and allow access to them anyway
1508 if (enable_vmware_backdoor)
1509 set_exception_intercept(svm, GP_VECTOR);
1511 set_intercept(svm, INTERCEPT_INTR);
1512 set_intercept(svm, INTERCEPT_NMI);
1513 set_intercept(svm, INTERCEPT_SMI);
1514 set_intercept(svm, INTERCEPT_SELECTIVE_CR0);
1515 set_intercept(svm, INTERCEPT_RDPMC);
1516 set_intercept(svm, INTERCEPT_CPUID);
1517 set_intercept(svm, INTERCEPT_INVD);
1518 set_intercept(svm, INTERCEPT_INVLPG);
1519 set_intercept(svm, INTERCEPT_INVLPGA);
1520 set_intercept(svm, INTERCEPT_IOIO_PROT);
1521 set_intercept(svm, INTERCEPT_MSR_PROT);
1522 set_intercept(svm, INTERCEPT_TASK_SWITCH);
1523 set_intercept(svm, INTERCEPT_SHUTDOWN);
1524 set_intercept(svm, INTERCEPT_VMRUN);
1525 set_intercept(svm, INTERCEPT_VMMCALL);
1526 set_intercept(svm, INTERCEPT_VMLOAD);
1527 set_intercept(svm, INTERCEPT_VMSAVE);
1528 set_intercept(svm, INTERCEPT_STGI);
1529 set_intercept(svm, INTERCEPT_CLGI);
1530 set_intercept(svm, INTERCEPT_SKINIT);
1531 set_intercept(svm, INTERCEPT_WBINVD);
1532 set_intercept(svm, INTERCEPT_XSETBV);
1533 set_intercept(svm, INTERCEPT_RSM);
1535 if (!kvm_mwait_in_guest(svm->vcpu.kvm)) {
1536 set_intercept(svm, INTERCEPT_MONITOR);
1537 set_intercept(svm, INTERCEPT_MWAIT);
1540 if (!kvm_hlt_in_guest(svm->vcpu.kvm))
1541 set_intercept(svm, INTERCEPT_HLT);
1543 control->iopm_base_pa = __sme_set(iopm_base);
1544 control->msrpm_base_pa = __sme_set(__pa(svm->msrpm));
1545 control->int_ctl = V_INTR_MASKING_MASK;
1547 init_seg(&save->es);
1548 init_seg(&save->ss);
1549 init_seg(&save->ds);
1550 init_seg(&save->fs);
1551 init_seg(&save->gs);
1553 save->cs.selector = 0xf000;
1554 save->cs.base = 0xffff0000;
1555 /* Executable/Readable Code Segment */
1556 save->cs.attrib = SVM_SELECTOR_READ_MASK | SVM_SELECTOR_P_MASK |
1557 SVM_SELECTOR_S_MASK | SVM_SELECTOR_CODE_MASK;
1558 save->cs.limit = 0xffff;
1560 save->gdtr.limit = 0xffff;
1561 save->idtr.limit = 0xffff;
1563 init_sys_seg(&save->ldtr, SEG_TYPE_LDT);
1564 init_sys_seg(&save->tr, SEG_TYPE_BUSY_TSS16);
1566 svm_set_efer(&svm->vcpu, 0);
1567 save->dr6 = 0xffff0ff0;
1568 kvm_set_rflags(&svm->vcpu, 2);
1569 save->rip = 0x0000fff0;
1570 svm->vcpu.arch.regs[VCPU_REGS_RIP] = save->rip;
1573 * svm_set_cr0() sets PG and WP and clears NW and CD on save->cr0.
1574 * It also updates the guest-visible cr0 value.
1576 svm_set_cr0(&svm->vcpu, X86_CR0_NW | X86_CR0_CD | X86_CR0_ET);
1577 kvm_mmu_reset_context(&svm->vcpu);
1579 save->cr4 = X86_CR4_PAE;
1583 /* Setup VMCB for Nested Paging */
1584 control->nested_ctl |= SVM_NESTED_CTL_NP_ENABLE;
1585 clr_intercept(svm, INTERCEPT_INVLPG);
1586 clr_exception_intercept(svm, PF_VECTOR);
1587 clr_cr_intercept(svm, INTERCEPT_CR3_READ);
1588 clr_cr_intercept(svm, INTERCEPT_CR3_WRITE);
1589 save->g_pat = svm->vcpu.arch.pat;
1593 svm->asid_generation = 0;
1595 svm->nested.vmcb = 0;
1596 svm->vcpu.arch.hflags = 0;
1598 if (pause_filter_count) {
1599 control->pause_filter_count = pause_filter_count;
1600 if (pause_filter_thresh)
1601 control->pause_filter_thresh = pause_filter_thresh;
1602 set_intercept(svm, INTERCEPT_PAUSE);
1604 clr_intercept(svm, INTERCEPT_PAUSE);
1607 if (kvm_vcpu_apicv_active(&svm->vcpu))
1608 avic_init_vmcb(svm);
1611 * If hardware supports Virtual VMLOAD VMSAVE then enable it
1612 * in VMCB and clear intercepts to avoid #VMEXIT.
1615 clr_intercept(svm, INTERCEPT_VMLOAD);
1616 clr_intercept(svm, INTERCEPT_VMSAVE);
1617 svm->vmcb->control.virt_ext |= VIRTUAL_VMLOAD_VMSAVE_ENABLE_MASK;
1621 clr_intercept(svm, INTERCEPT_STGI);
1622 clr_intercept(svm, INTERCEPT_CLGI);
1623 svm->vmcb->control.int_ctl |= V_GIF_ENABLE_MASK;
1626 if (sev_guest(svm->vcpu.kvm)) {
1627 svm->vmcb->control.nested_ctl |= SVM_NESTED_CTL_SEV_ENABLE;
1628 clr_exception_intercept(svm, UD_VECTOR);
1631 mark_all_dirty(svm->vmcb);
1637 static u64 *avic_get_physical_id_entry(struct kvm_vcpu *vcpu,
1640 u64 *avic_physical_id_table;
1641 struct kvm_svm *kvm_svm = to_kvm_svm(vcpu->kvm);
1643 if (index >= AVIC_MAX_PHYSICAL_ID_COUNT)
1646 avic_physical_id_table = page_address(kvm_svm->avic_physical_id_table_page);
1648 return &avic_physical_id_table[index];
1653 * AVIC hardware walks the nested page table to check permissions,
1654 * but does not use the SPA address specified in the leaf page
1655 * table entry since it uses address in the AVIC_BACKING_PAGE pointer
1656 * field of the VMCB. Therefore, we set up the
1657 * APIC_ACCESS_PAGE_PRIVATE_MEMSLOT (4KB) here.
1659 static int avic_init_access_page(struct kvm_vcpu *vcpu)
1661 struct kvm *kvm = vcpu->kvm;
1664 if (kvm->arch.apic_access_page_done)
1667 ret = x86_set_memory_region(kvm,
1668 APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
1669 APIC_DEFAULT_PHYS_BASE,
1674 kvm->arch.apic_access_page_done = true;
1678 static int avic_init_backing_page(struct kvm_vcpu *vcpu)
1681 u64 *entry, new_entry;
1682 int id = vcpu->vcpu_id;
1683 struct vcpu_svm *svm = to_svm(vcpu);
1685 ret = avic_init_access_page(vcpu);
1689 if (id >= AVIC_MAX_PHYSICAL_ID_COUNT)
1692 if (!svm->vcpu.arch.apic->regs)
1695 svm->avic_backing_page = virt_to_page(svm->vcpu.arch.apic->regs);
1697 /* Setting AVIC backing page address in the phy APIC ID table */
1698 entry = avic_get_physical_id_entry(vcpu, id);
1702 new_entry = READ_ONCE(*entry);
1703 new_entry = __sme_set((page_to_phys(svm->avic_backing_page) &
1704 AVIC_PHYSICAL_ID_ENTRY_BACKING_PAGE_MASK) |
1705 AVIC_PHYSICAL_ID_ENTRY_VALID_MASK);
1706 WRITE_ONCE(*entry, new_entry);
1708 svm->avic_physical_id_cache = entry;
1713 static void __sev_asid_free(int asid)
1715 struct svm_cpu_data *sd;
1719 clear_bit(pos, sev_asid_bitmap);
1721 for_each_possible_cpu(cpu) {
1722 sd = per_cpu(svm_data, cpu);
1723 sd->sev_vmcbs[pos] = NULL;
1727 static void sev_asid_free(struct kvm *kvm)
1729 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1731 __sev_asid_free(sev->asid);
1734 static void sev_unbind_asid(struct kvm *kvm, unsigned int handle)
1736 struct sev_data_decommission *decommission;
1737 struct sev_data_deactivate *data;
1742 data = kzalloc(sizeof(*data), GFP_KERNEL);
1746 /* deactivate handle */
1747 data->handle = handle;
1748 sev_guest_deactivate(data, NULL);
1750 wbinvd_on_all_cpus();
1751 sev_guest_df_flush(NULL);
1754 decommission = kzalloc(sizeof(*decommission), GFP_KERNEL);
1758 /* decommission handle */
1759 decommission->handle = handle;
1760 sev_guest_decommission(decommission, NULL);
1762 kfree(decommission);
1765 static struct page **sev_pin_memory(struct kvm *kvm, unsigned long uaddr,
1766 unsigned long ulen, unsigned long *n,
1769 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1770 unsigned long npages, npinned, size;
1771 unsigned long locked, lock_limit;
1772 struct page **pages;
1773 unsigned long first, last;
1775 if (ulen == 0 || uaddr + ulen < uaddr)
1778 /* Calculate number of pages. */
1779 first = (uaddr & PAGE_MASK) >> PAGE_SHIFT;
1780 last = ((uaddr + ulen - 1) & PAGE_MASK) >> PAGE_SHIFT;
1781 npages = (last - first + 1);
1783 locked = sev->pages_locked + npages;
1784 lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1785 if (locked > lock_limit && !capable(CAP_IPC_LOCK)) {
1786 pr_err("SEV: %lu locked pages exceed the lock limit of %lu.\n", locked, lock_limit);
1790 /* Avoid using vmalloc for smaller buffers. */
1791 size = npages * sizeof(struct page *);
1792 if (size > PAGE_SIZE)
1793 pages = vmalloc(size);
1795 pages = kmalloc(size, GFP_KERNEL);
1800 /* Pin the user virtual address. */
1801 npinned = get_user_pages_fast(uaddr, npages, write ? FOLL_WRITE : 0, pages);
1802 if (npinned != npages) {
1803 pr_err("SEV: Failure locking %lu pages.\n", npages);
1808 sev->pages_locked = locked;
1814 release_pages(pages, npinned);
1820 static void sev_unpin_memory(struct kvm *kvm, struct page **pages,
1821 unsigned long npages)
1823 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1825 release_pages(pages, npages);
1827 sev->pages_locked -= npages;
1830 static void sev_clflush_pages(struct page *pages[], unsigned long npages)
1832 uint8_t *page_virtual;
1835 if (npages == 0 || pages == NULL)
1838 for (i = 0; i < npages; i++) {
1839 page_virtual = kmap_atomic(pages[i]);
1840 clflush_cache_range(page_virtual, PAGE_SIZE);
1841 kunmap_atomic(page_virtual);
1845 static void __unregister_enc_region_locked(struct kvm *kvm,
1846 struct enc_region *region)
1849 * The guest may change the memory encryption attribute from C=0 -> C=1
1850 * or vice versa for this memory range. Lets make sure caches are
1851 * flushed to ensure that guest data gets written into memory with
1854 sev_clflush_pages(region->pages, region->npages);
1856 sev_unpin_memory(kvm, region->pages, region->npages);
1857 list_del(®ion->list);
1861 static struct kvm *svm_vm_alloc(void)
1863 struct kvm_svm *kvm_svm = vzalloc(sizeof(struct kvm_svm));
1864 return &kvm_svm->kvm;
1867 static void svm_vm_free(struct kvm *kvm)
1869 vfree(to_kvm_svm(kvm));
1872 static void sev_vm_destroy(struct kvm *kvm)
1874 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1875 struct list_head *head = &sev->regions_list;
1876 struct list_head *pos, *q;
1878 if (!sev_guest(kvm))
1881 mutex_lock(&kvm->lock);
1884 * if userspace was terminated before unregistering the memory regions
1885 * then lets unpin all the registered memory.
1887 if (!list_empty(head)) {
1888 list_for_each_safe(pos, q, head) {
1889 __unregister_enc_region_locked(kvm,
1890 list_entry(pos, struct enc_region, list));
1894 mutex_unlock(&kvm->lock);
1896 sev_unbind_asid(kvm, sev->handle);
1900 static void avic_vm_destroy(struct kvm *kvm)
1902 unsigned long flags;
1903 struct kvm_svm *kvm_svm = to_kvm_svm(kvm);
1908 if (kvm_svm->avic_logical_id_table_page)
1909 __free_page(kvm_svm->avic_logical_id_table_page);
1910 if (kvm_svm->avic_physical_id_table_page)
1911 __free_page(kvm_svm->avic_physical_id_table_page);
1913 spin_lock_irqsave(&svm_vm_data_hash_lock, flags);
1914 hash_del(&kvm_svm->hnode);
1915 spin_unlock_irqrestore(&svm_vm_data_hash_lock, flags);
1918 static void svm_vm_destroy(struct kvm *kvm)
1920 avic_vm_destroy(kvm);
1921 sev_vm_destroy(kvm);
1924 static int avic_vm_init(struct kvm *kvm)
1926 unsigned long flags;
1928 struct kvm_svm *kvm_svm = to_kvm_svm(kvm);
1930 struct page *p_page;
1931 struct page *l_page;
1937 /* Allocating physical APIC ID table (4KB) */
1938 p_page = alloc_page(GFP_KERNEL);
1942 kvm_svm->avic_physical_id_table_page = p_page;
1943 clear_page(page_address(p_page));
1945 /* Allocating logical APIC ID table (4KB) */
1946 l_page = alloc_page(GFP_KERNEL);
1950 kvm_svm->avic_logical_id_table_page = l_page;
1951 clear_page(page_address(l_page));
1953 spin_lock_irqsave(&svm_vm_data_hash_lock, flags);
1955 vm_id = next_vm_id = (next_vm_id + 1) & AVIC_VM_ID_MASK;
1956 if (vm_id == 0) { /* id is 1-based, zero is not okay */
1957 next_vm_id_wrapped = 1;
1960 /* Is it still in use? Only possible if wrapped at least once */
1961 if (next_vm_id_wrapped) {
1962 hash_for_each_possible(svm_vm_data_hash, k2, hnode, vm_id) {
1963 if (k2->avic_vm_id == vm_id)
1967 kvm_svm->avic_vm_id = vm_id;
1968 hash_add(svm_vm_data_hash, &kvm_svm->hnode, kvm_svm->avic_vm_id);
1969 spin_unlock_irqrestore(&svm_vm_data_hash_lock, flags);
1974 avic_vm_destroy(kvm);
1979 avic_update_iommu_vcpu_affinity(struct kvm_vcpu *vcpu, int cpu, bool r)
1982 unsigned long flags;
1983 struct amd_svm_iommu_ir *ir;
1984 struct vcpu_svm *svm = to_svm(vcpu);
1986 if (!kvm_arch_has_assigned_device(vcpu->kvm))
1990 * Here, we go through the per-vcpu ir_list to update all existing
1991 * interrupt remapping table entry targeting this vcpu.
1993 spin_lock_irqsave(&svm->ir_list_lock, flags);
1995 if (list_empty(&svm->ir_list))
1998 list_for_each_entry(ir, &svm->ir_list, node) {
1999 ret = amd_iommu_update_ga(cpu, r, ir->data);
2004 spin_unlock_irqrestore(&svm->ir_list_lock, flags);
2008 static void avic_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2011 /* ID = 0xff (broadcast), ID > 0xff (reserved) */
2012 int h_physical_id = kvm_cpu_get_apicid(cpu);
2013 struct vcpu_svm *svm = to_svm(vcpu);
2015 if (!kvm_vcpu_apicv_active(vcpu))
2018 if (WARN_ON(h_physical_id >= AVIC_MAX_PHYSICAL_ID_COUNT))
2021 entry = READ_ONCE(*(svm->avic_physical_id_cache));
2022 WARN_ON(entry & AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK);
2024 entry &= ~AVIC_PHYSICAL_ID_ENTRY_HOST_PHYSICAL_ID_MASK;
2025 entry |= (h_physical_id & AVIC_PHYSICAL_ID_ENTRY_HOST_PHYSICAL_ID_MASK);
2027 entry &= ~AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK;
2028 if (svm->avic_is_running)
2029 entry |= AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK;
2031 WRITE_ONCE(*(svm->avic_physical_id_cache), entry);
2032 avic_update_iommu_vcpu_affinity(vcpu, h_physical_id,
2033 svm->avic_is_running);
2036 static void avic_vcpu_put(struct kvm_vcpu *vcpu)
2039 struct vcpu_svm *svm = to_svm(vcpu);
2041 if (!kvm_vcpu_apicv_active(vcpu))
2044 entry = READ_ONCE(*(svm->avic_physical_id_cache));
2045 if (entry & AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK)
2046 avic_update_iommu_vcpu_affinity(vcpu, -1, 0);
2048 entry &= ~AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK;
2049 WRITE_ONCE(*(svm->avic_physical_id_cache), entry);
2053 * This function is called during VCPU halt/unhalt.
2055 static void avic_set_running(struct kvm_vcpu *vcpu, bool is_run)
2057 struct vcpu_svm *svm = to_svm(vcpu);
2059 svm->avic_is_running = is_run;
2061 avic_vcpu_load(vcpu, vcpu->cpu);
2063 avic_vcpu_put(vcpu);
2066 static void svm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
2068 struct vcpu_svm *svm = to_svm(vcpu);
2072 vcpu->arch.microcode_version = 0x01000065;
2074 svm->virt_spec_ctrl = 0;
2077 svm->vcpu.arch.apic_base = APIC_DEFAULT_PHYS_BASE |
2078 MSR_IA32_APICBASE_ENABLE;
2079 if (kvm_vcpu_is_reset_bsp(&svm->vcpu))
2080 svm->vcpu.arch.apic_base |= MSR_IA32_APICBASE_BSP;
2084 kvm_cpuid(vcpu, &eax, &dummy, &dummy, &dummy, true);
2085 kvm_register_write(vcpu, VCPU_REGS_RDX, eax);
2087 if (kvm_vcpu_apicv_active(vcpu) && !init_event)
2088 avic_update_vapic_bar(svm, APIC_DEFAULT_PHYS_BASE);
2091 static int avic_init_vcpu(struct vcpu_svm *svm)
2095 if (!kvm_vcpu_apicv_active(&svm->vcpu))
2098 ret = avic_init_backing_page(&svm->vcpu);
2102 INIT_LIST_HEAD(&svm->ir_list);
2103 spin_lock_init(&svm->ir_list_lock);
2108 static struct kvm_vcpu *svm_create_vcpu(struct kvm *kvm, unsigned int id)
2110 struct vcpu_svm *svm;
2112 struct page *msrpm_pages;
2113 struct page *hsave_page;
2114 struct page *nested_msrpm_pages;
2117 svm = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
2123 err = kvm_vcpu_init(&svm->vcpu, kvm, id);
2128 page = alloc_page(GFP_KERNEL);
2132 msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
2136 nested_msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
2137 if (!nested_msrpm_pages)
2140 hsave_page = alloc_page(GFP_KERNEL);
2144 err = avic_init_vcpu(svm);
2148 /* We initialize this flag to true to make sure that the is_running
2149 * bit would be set the first time the vcpu is loaded.
2151 svm->avic_is_running = true;
2153 svm->nested.hsave = page_address(hsave_page);
2155 svm->msrpm = page_address(msrpm_pages);
2156 svm_vcpu_init_msrpm(svm->msrpm);
2158 svm->nested.msrpm = page_address(nested_msrpm_pages);
2159 svm_vcpu_init_msrpm(svm->nested.msrpm);
2161 svm->vmcb = page_address(page);
2162 clear_page(svm->vmcb);
2163 svm->vmcb_pa = __sme_set(page_to_pfn(page) << PAGE_SHIFT);
2164 svm->asid_generation = 0;
2167 svm_init_osvw(&svm->vcpu);
2172 __free_page(hsave_page);
2174 __free_pages(nested_msrpm_pages, MSRPM_ALLOC_ORDER);
2176 __free_pages(msrpm_pages, MSRPM_ALLOC_ORDER);
2180 kvm_vcpu_uninit(&svm->vcpu);
2182 kmem_cache_free(kvm_vcpu_cache, svm);
2184 return ERR_PTR(err);
2187 static void svm_free_vcpu(struct kvm_vcpu *vcpu)
2189 struct vcpu_svm *svm = to_svm(vcpu);
2191 __free_page(pfn_to_page(__sme_clr(svm->vmcb_pa) >> PAGE_SHIFT));
2192 __free_pages(virt_to_page(svm->msrpm), MSRPM_ALLOC_ORDER);
2193 __free_page(virt_to_page(svm->nested.hsave));
2194 __free_pages(virt_to_page(svm->nested.msrpm), MSRPM_ALLOC_ORDER);
2195 kvm_vcpu_uninit(vcpu);
2196 kmem_cache_free(kvm_vcpu_cache, svm);
2198 * The vmcb page can be recycled, causing a false negative in
2199 * svm_vcpu_load(). So do a full IBPB now.
2201 indirect_branch_prediction_barrier();
2204 static void svm_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2206 struct vcpu_svm *svm = to_svm(vcpu);
2207 struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
2210 if (unlikely(cpu != vcpu->cpu)) {
2211 svm->asid_generation = 0;
2212 mark_all_dirty(svm->vmcb);
2215 #ifdef CONFIG_X86_64
2216 rdmsrl(MSR_GS_BASE, to_svm(vcpu)->host.gs_base);
2218 savesegment(fs, svm->host.fs);
2219 savesegment(gs, svm->host.gs);
2220 svm->host.ldt = kvm_read_ldt();
2222 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
2223 rdmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
2225 if (static_cpu_has(X86_FEATURE_TSCRATEMSR)) {
2226 u64 tsc_ratio = vcpu->arch.tsc_scaling_ratio;
2227 if (tsc_ratio != __this_cpu_read(current_tsc_ratio)) {
2228 __this_cpu_write(current_tsc_ratio, tsc_ratio);
2229 wrmsrl(MSR_AMD64_TSC_RATIO, tsc_ratio);
2232 /* This assumes that the kernel never uses MSR_TSC_AUX */
2233 if (static_cpu_has(X86_FEATURE_RDTSCP))
2234 wrmsrl(MSR_TSC_AUX, svm->tsc_aux);
2236 if (sd->current_vmcb != svm->vmcb) {
2237 sd->current_vmcb = svm->vmcb;
2238 indirect_branch_prediction_barrier();
2240 avic_vcpu_load(vcpu, cpu);
2243 static void svm_vcpu_put(struct kvm_vcpu *vcpu)
2245 struct vcpu_svm *svm = to_svm(vcpu);
2248 avic_vcpu_put(vcpu);
2250 ++vcpu->stat.host_state_reload;
2251 kvm_load_ldt(svm->host.ldt);
2252 #ifdef CONFIG_X86_64
2253 loadsegment(fs, svm->host.fs);
2254 wrmsrl(MSR_KERNEL_GS_BASE, current->thread.gsbase);
2255 load_gs_index(svm->host.gs);
2257 #ifdef CONFIG_X86_32_LAZY_GS
2258 loadsegment(gs, svm->host.gs);
2261 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
2262 wrmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
2265 static void svm_vcpu_blocking(struct kvm_vcpu *vcpu)
2267 avic_set_running(vcpu, false);
2270 static void svm_vcpu_unblocking(struct kvm_vcpu *vcpu)
2272 avic_set_running(vcpu, true);
2275 static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu)
2277 struct vcpu_svm *svm = to_svm(vcpu);
2278 unsigned long rflags = svm->vmcb->save.rflags;
2280 if (svm->nmi_singlestep) {
2281 /* Hide our flags if they were not set by the guest */
2282 if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_TF))
2283 rflags &= ~X86_EFLAGS_TF;
2284 if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_RF))
2285 rflags &= ~X86_EFLAGS_RF;
2290 static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
2292 if (to_svm(vcpu)->nmi_singlestep)
2293 rflags |= (X86_EFLAGS_TF | X86_EFLAGS_RF);
2296 * Any change of EFLAGS.VM is accompanied by a reload of SS
2297 * (caused by either a task switch or an inter-privilege IRET),
2298 * so we do not need to update the CPL here.
2300 to_svm(vcpu)->vmcb->save.rflags = rflags;
2303 static void svm_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
2306 case VCPU_EXREG_PDPTR:
2307 BUG_ON(!npt_enabled);
2308 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
2315 static void svm_set_vintr(struct vcpu_svm *svm)
2317 set_intercept(svm, INTERCEPT_VINTR);
2320 static void svm_clear_vintr(struct vcpu_svm *svm)
2322 clr_intercept(svm, INTERCEPT_VINTR);
2325 static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg)
2327 struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
2330 case VCPU_SREG_CS: return &save->cs;
2331 case VCPU_SREG_DS: return &save->ds;
2332 case VCPU_SREG_ES: return &save->es;
2333 case VCPU_SREG_FS: return &save->fs;
2334 case VCPU_SREG_GS: return &save->gs;
2335 case VCPU_SREG_SS: return &save->ss;
2336 case VCPU_SREG_TR: return &save->tr;
2337 case VCPU_SREG_LDTR: return &save->ldtr;
2343 static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg)
2345 struct vmcb_seg *s = svm_seg(vcpu, seg);
2350 static void svm_get_segment(struct kvm_vcpu *vcpu,
2351 struct kvm_segment *var, int seg)
2353 struct vmcb_seg *s = svm_seg(vcpu, seg);
2355 var->base = s->base;
2356 var->limit = s->limit;
2357 var->selector = s->selector;
2358 var->type = s->attrib & SVM_SELECTOR_TYPE_MASK;
2359 var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1;
2360 var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
2361 var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1;
2362 var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1;
2363 var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
2364 var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
2367 * AMD CPUs circa 2014 track the G bit for all segments except CS.
2368 * However, the SVM spec states that the G bit is not observed by the
2369 * CPU, and some VMware virtual CPUs drop the G bit for all segments.
2370 * So let's synthesize a legal G bit for all segments, this helps
2371 * running KVM nested. It also helps cross-vendor migration, because
2372 * Intel's vmentry has a check on the 'G' bit.
2374 var->g = s->limit > 0xfffff;
2377 * AMD's VMCB does not have an explicit unusable field, so emulate it
2378 * for cross vendor migration purposes by "not present"
2380 var->unusable = !var->present;
2385 * Work around a bug where the busy flag in the tr selector
2395 * The accessed bit must always be set in the segment
2396 * descriptor cache, although it can be cleared in the
2397 * descriptor, the cached bit always remains at 1. Since
2398 * Intel has a check on this, set it here to support
2399 * cross-vendor migration.
2406 * On AMD CPUs sometimes the DB bit in the segment
2407 * descriptor is left as 1, although the whole segment has
2408 * been made unusable. Clear it here to pass an Intel VMX
2409 * entry check when cross vendor migrating.
2413 /* This is symmetric with svm_set_segment() */
2414 var->dpl = to_svm(vcpu)->vmcb->save.cpl;
2419 static int svm_get_cpl(struct kvm_vcpu *vcpu)
2421 struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
2426 static void svm_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
2428 struct vcpu_svm *svm = to_svm(vcpu);
2430 dt->size = svm->vmcb->save.idtr.limit;
2431 dt->address = svm->vmcb->save.idtr.base;
2434 static void svm_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
2436 struct vcpu_svm *svm = to_svm(vcpu);
2438 svm->vmcb->save.idtr.limit = dt->size;
2439 svm->vmcb->save.idtr.base = dt->address ;
2440 mark_dirty(svm->vmcb, VMCB_DT);
2443 static void svm_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
2445 struct vcpu_svm *svm = to_svm(vcpu);
2447 dt->size = svm->vmcb->save.gdtr.limit;
2448 dt->address = svm->vmcb->save.gdtr.base;
2451 static void svm_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
2453 struct vcpu_svm *svm = to_svm(vcpu);
2455 svm->vmcb->save.gdtr.limit = dt->size;
2456 svm->vmcb->save.gdtr.base = dt->address ;
2457 mark_dirty(svm->vmcb, VMCB_DT);
2460 static void svm_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
2464 static void svm_decache_cr3(struct kvm_vcpu *vcpu)
2468 static void svm_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
2472 static void update_cr0_intercept(struct vcpu_svm *svm)
2474 ulong gcr0 = svm->vcpu.arch.cr0;
2475 u64 *hcr0 = &svm->vmcb->save.cr0;
2477 *hcr0 = (*hcr0 & ~SVM_CR0_SELECTIVE_MASK)
2478 | (gcr0 & SVM_CR0_SELECTIVE_MASK);
2480 mark_dirty(svm->vmcb, VMCB_CR);
2482 if (gcr0 == *hcr0) {
2483 clr_cr_intercept(svm, INTERCEPT_CR0_READ);
2484 clr_cr_intercept(svm, INTERCEPT_CR0_WRITE);
2486 set_cr_intercept(svm, INTERCEPT_CR0_READ);
2487 set_cr_intercept(svm, INTERCEPT_CR0_WRITE);
2491 static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
2493 struct vcpu_svm *svm = to_svm(vcpu);
2495 #ifdef CONFIG_X86_64
2496 if (vcpu->arch.efer & 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;
2502 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG)) {
2503 vcpu->arch.efer &= ~EFER_LMA;
2504 svm->vmcb->save.efer &= ~(EFER_LMA | EFER_LME);
2508 vcpu->arch.cr0 = cr0;
2511 cr0 |= X86_CR0_PG | X86_CR0_WP;
2514 * re-enable caching here because the QEMU bios
2515 * does not do it - this results in some delay at
2518 if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
2519 cr0 &= ~(X86_CR0_CD | X86_CR0_NW);
2520 svm->vmcb->save.cr0 = cr0;
2521 mark_dirty(svm->vmcb, VMCB_CR);
2522 update_cr0_intercept(svm);
2525 static int svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
2527 unsigned long host_cr4_mce = cr4_read_shadow() & X86_CR4_MCE;
2528 unsigned long old_cr4 = to_svm(vcpu)->vmcb->save.cr4;
2530 if (cr4 & X86_CR4_VMXE)
2533 if (npt_enabled && ((old_cr4 ^ cr4) & X86_CR4_PGE))
2534 svm_flush_tlb(vcpu, true);
2536 vcpu->arch.cr4 = cr4;
2539 cr4 |= host_cr4_mce;
2540 to_svm(vcpu)->vmcb->save.cr4 = cr4;
2541 mark_dirty(to_svm(vcpu)->vmcb, VMCB_CR);
2545 static void svm_set_segment(struct kvm_vcpu *vcpu,
2546 struct kvm_segment *var, int seg)
2548 struct vcpu_svm *svm = to_svm(vcpu);
2549 struct vmcb_seg *s = svm_seg(vcpu, seg);
2551 s->base = var->base;
2552 s->limit = var->limit;
2553 s->selector = var->selector;
2554 s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK);
2555 s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT;
2556 s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT;
2557 s->attrib |= ((var->present & 1) && !var->unusable) << SVM_SELECTOR_P_SHIFT;
2558 s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT;
2559 s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT;
2560 s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT;
2561 s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT;
2564 * This is always accurate, except if SYSRET returned to a segment
2565 * with SS.DPL != 3. Intel does not have this quirk, and always
2566 * forces SS.DPL to 3 on sysret, so we ignore that case; fixing it
2567 * would entail passing the CPL to userspace and back.
2569 if (seg == VCPU_SREG_SS)
2570 /* This is symmetric with svm_get_segment() */
2571 svm->vmcb->save.cpl = (var->dpl & 3);
2573 mark_dirty(svm->vmcb, VMCB_SEG);
2576 static void update_bp_intercept(struct kvm_vcpu *vcpu)
2578 struct vcpu_svm *svm = to_svm(vcpu);
2580 clr_exception_intercept(svm, BP_VECTOR);
2582 if (vcpu->guest_debug & KVM_GUESTDBG_ENABLE) {
2583 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
2584 set_exception_intercept(svm, BP_VECTOR);
2586 vcpu->guest_debug = 0;
2589 static void new_asid(struct vcpu_svm *svm, struct svm_cpu_data *sd)
2591 if (sd->next_asid > sd->max_asid) {
2592 ++sd->asid_generation;
2593 sd->next_asid = sd->min_asid;
2594 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID;
2597 svm->asid_generation = sd->asid_generation;
2598 svm->vmcb->control.asid = sd->next_asid++;
2600 mark_dirty(svm->vmcb, VMCB_ASID);
2603 static u64 svm_get_dr6(struct kvm_vcpu *vcpu)
2605 return to_svm(vcpu)->vmcb->save.dr6;
2608 static void svm_set_dr6(struct kvm_vcpu *vcpu, unsigned long value)
2610 struct vcpu_svm *svm = to_svm(vcpu);
2612 svm->vmcb->save.dr6 = value;
2613 mark_dirty(svm->vmcb, VMCB_DR);
2616 static void svm_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
2618 struct vcpu_svm *svm = to_svm(vcpu);
2620 get_debugreg(vcpu->arch.db[0], 0);
2621 get_debugreg(vcpu->arch.db[1], 1);
2622 get_debugreg(vcpu->arch.db[2], 2);
2623 get_debugreg(vcpu->arch.db[3], 3);
2624 vcpu->arch.dr6 = svm_get_dr6(vcpu);
2625 vcpu->arch.dr7 = svm->vmcb->save.dr7;
2627 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
2628 set_dr_intercepts(svm);
2631 static void svm_set_dr7(struct kvm_vcpu *vcpu, unsigned long value)
2633 struct vcpu_svm *svm = to_svm(vcpu);
2635 svm->vmcb->save.dr7 = value;
2636 mark_dirty(svm->vmcb, VMCB_DR);
2639 static int pf_interception(struct vcpu_svm *svm)
2641 u64 fault_address = __sme_clr(svm->vmcb->control.exit_info_2);
2642 u64 error_code = svm->vmcb->control.exit_info_1;
2644 return kvm_handle_page_fault(&svm->vcpu, error_code, fault_address,
2645 static_cpu_has(X86_FEATURE_DECODEASSISTS) ?
2646 svm->vmcb->control.insn_bytes : NULL,
2647 svm->vmcb->control.insn_len);
2650 static int npf_interception(struct vcpu_svm *svm)
2652 u64 fault_address = __sme_clr(svm->vmcb->control.exit_info_2);
2653 u64 error_code = svm->vmcb->control.exit_info_1;
2655 trace_kvm_page_fault(fault_address, error_code);
2656 return kvm_mmu_page_fault(&svm->vcpu, fault_address, error_code,
2657 static_cpu_has(X86_FEATURE_DECODEASSISTS) ?
2658 svm->vmcb->control.insn_bytes : NULL,
2659 svm->vmcb->control.insn_len);
2662 static int db_interception(struct vcpu_svm *svm)
2664 struct kvm_run *kvm_run = svm->vcpu.run;
2666 if (!(svm->vcpu.guest_debug &
2667 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) &&
2668 !svm->nmi_singlestep) {
2669 kvm_queue_exception(&svm->vcpu, DB_VECTOR);
2673 if (svm->nmi_singlestep) {
2674 disable_nmi_singlestep(svm);
2677 if (svm->vcpu.guest_debug &
2678 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) {
2679 kvm_run->exit_reason = KVM_EXIT_DEBUG;
2680 kvm_run->debug.arch.pc =
2681 svm->vmcb->save.cs.base + svm->vmcb->save.rip;
2682 kvm_run->debug.arch.exception = DB_VECTOR;
2689 static int bp_interception(struct vcpu_svm *svm)
2691 struct kvm_run *kvm_run = svm->vcpu.run;
2693 kvm_run->exit_reason = KVM_EXIT_DEBUG;
2694 kvm_run->debug.arch.pc = svm->vmcb->save.cs.base + svm->vmcb->save.rip;
2695 kvm_run->debug.arch.exception = BP_VECTOR;
2699 static int ud_interception(struct vcpu_svm *svm)
2701 return handle_ud(&svm->vcpu);
2704 static int ac_interception(struct vcpu_svm *svm)
2706 kvm_queue_exception_e(&svm->vcpu, AC_VECTOR, 0);
2710 static int gp_interception(struct vcpu_svm *svm)
2712 struct kvm_vcpu *vcpu = &svm->vcpu;
2713 u32 error_code = svm->vmcb->control.exit_info_1;
2716 WARN_ON_ONCE(!enable_vmware_backdoor);
2718 er = emulate_instruction(vcpu,
2719 EMULTYPE_VMWARE | EMULTYPE_NO_UD_ON_FAIL);
2720 if (er == EMULATE_USER_EXIT)
2722 else if (er != EMULATE_DONE)
2723 kvm_queue_exception_e(vcpu, GP_VECTOR, error_code);
2727 static bool is_erratum_383(void)
2732 if (!erratum_383_found)
2735 value = native_read_msr_safe(MSR_IA32_MC0_STATUS, &err);
2739 /* Bit 62 may or may not be set for this mce */
2740 value &= ~(1ULL << 62);
2742 if (value != 0xb600000000010015ULL)
2745 /* Clear MCi_STATUS registers */
2746 for (i = 0; i < 6; ++i)
2747 native_write_msr_safe(MSR_IA32_MCx_STATUS(i), 0, 0);
2749 value = native_read_msr_safe(MSR_IA32_MCG_STATUS, &err);
2753 value &= ~(1ULL << 2);
2754 low = lower_32_bits(value);
2755 high = upper_32_bits(value);
2757 native_write_msr_safe(MSR_IA32_MCG_STATUS, low, high);
2760 /* Flush tlb to evict multi-match entries */
2766 static void svm_handle_mce(struct vcpu_svm *svm)
2768 if (is_erratum_383()) {
2770 * Erratum 383 triggered. Guest state is corrupt so kill the
2773 pr_err("KVM: Guest triggered AMD Erratum 383\n");
2775 kvm_make_request(KVM_REQ_TRIPLE_FAULT, &svm->vcpu);
2781 * On an #MC intercept the MCE handler is not called automatically in
2782 * the host. So do it by hand here.
2786 /* not sure if we ever come back to this point */
2791 static int mc_interception(struct vcpu_svm *svm)
2796 static int shutdown_interception(struct vcpu_svm *svm)
2798 struct kvm_run *kvm_run = svm->vcpu.run;
2801 * VMCB is undefined after a SHUTDOWN intercept
2802 * so reinitialize it.
2804 clear_page(svm->vmcb);
2807 kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
2811 static int io_interception(struct vcpu_svm *svm)
2813 struct kvm_vcpu *vcpu = &svm->vcpu;
2814 u32 io_info = svm->vmcb->control.exit_info_1; /* address size bug? */
2815 int size, in, string;
2818 ++svm->vcpu.stat.io_exits;
2819 string = (io_info & SVM_IOIO_STR_MASK) != 0;
2820 in = (io_info & SVM_IOIO_TYPE_MASK) != 0;
2822 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
2824 port = io_info >> 16;
2825 size = (io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT;
2826 svm->next_rip = svm->vmcb->control.exit_info_2;
2828 return kvm_fast_pio(&svm->vcpu, size, port, in);
2831 static int nmi_interception(struct vcpu_svm *svm)
2836 static int intr_interception(struct vcpu_svm *svm)
2838 ++svm->vcpu.stat.irq_exits;
2842 static int nop_on_interception(struct vcpu_svm *svm)
2847 static int halt_interception(struct vcpu_svm *svm)
2849 svm->next_rip = kvm_rip_read(&svm->vcpu) + 1;
2850 return kvm_emulate_halt(&svm->vcpu);
2853 static int vmmcall_interception(struct vcpu_svm *svm)
2855 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2856 return kvm_emulate_hypercall(&svm->vcpu);
2859 static unsigned long nested_svm_get_tdp_cr3(struct kvm_vcpu *vcpu)
2861 struct vcpu_svm *svm = to_svm(vcpu);
2863 return svm->nested.nested_cr3;
2866 static u64 nested_svm_get_tdp_pdptr(struct kvm_vcpu *vcpu, int index)
2868 struct vcpu_svm *svm = to_svm(vcpu);
2869 u64 cr3 = svm->nested.nested_cr3;
2873 ret = kvm_vcpu_read_guest_page(vcpu, gpa_to_gfn(__sme_clr(cr3)), &pdpte,
2874 offset_in_page(cr3) + index * 8, 8);
2880 static void nested_svm_set_tdp_cr3(struct kvm_vcpu *vcpu,
2883 struct vcpu_svm *svm = to_svm(vcpu);
2885 svm->vmcb->control.nested_cr3 = __sme_set(root);
2886 mark_dirty(svm->vmcb, VMCB_NPT);
2889 static void nested_svm_inject_npf_exit(struct kvm_vcpu *vcpu,
2890 struct x86_exception *fault)
2892 struct vcpu_svm *svm = to_svm(vcpu);
2894 if (svm->vmcb->control.exit_code != SVM_EXIT_NPF) {
2896 * TODO: track the cause of the nested page fault, and
2897 * correctly fill in the high bits of exit_info_1.
2899 svm->vmcb->control.exit_code = SVM_EXIT_NPF;
2900 svm->vmcb->control.exit_code_hi = 0;
2901 svm->vmcb->control.exit_info_1 = (1ULL << 32);
2902 svm->vmcb->control.exit_info_2 = fault->address;
2905 svm->vmcb->control.exit_info_1 &= ~0xffffffffULL;
2906 svm->vmcb->control.exit_info_1 |= fault->error_code;
2909 * The present bit is always zero for page structure faults on real
2912 if (svm->vmcb->control.exit_info_1 & (2ULL << 32))
2913 svm->vmcb->control.exit_info_1 &= ~1;
2915 nested_svm_vmexit(svm);
2918 static void nested_svm_init_mmu_context(struct kvm_vcpu *vcpu)
2920 WARN_ON(mmu_is_nested(vcpu));
2921 kvm_init_shadow_mmu(vcpu);
2922 vcpu->arch.mmu.set_cr3 = nested_svm_set_tdp_cr3;
2923 vcpu->arch.mmu.get_cr3 = nested_svm_get_tdp_cr3;
2924 vcpu->arch.mmu.get_pdptr = nested_svm_get_tdp_pdptr;
2925 vcpu->arch.mmu.inject_page_fault = nested_svm_inject_npf_exit;
2926 vcpu->arch.mmu.shadow_root_level = get_npt_level(vcpu);
2927 reset_shadow_zero_bits_mask(vcpu, &vcpu->arch.mmu);
2928 vcpu->arch.walk_mmu = &vcpu->arch.nested_mmu;
2931 static void nested_svm_uninit_mmu_context(struct kvm_vcpu *vcpu)
2933 vcpu->arch.walk_mmu = &vcpu->arch.mmu;
2936 static int nested_svm_check_permissions(struct vcpu_svm *svm)
2938 if (!(svm->vcpu.arch.efer & EFER_SVME) ||
2939 !is_paging(&svm->vcpu)) {
2940 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2944 if (svm->vmcb->save.cpl) {
2945 kvm_inject_gp(&svm->vcpu, 0);
2952 static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
2953 bool has_error_code, u32 error_code)
2957 if (!is_guest_mode(&svm->vcpu))
2960 vmexit = nested_svm_intercept(svm);
2961 if (vmexit != NESTED_EXIT_DONE)
2964 svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + nr;
2965 svm->vmcb->control.exit_code_hi = 0;
2966 svm->vmcb->control.exit_info_1 = error_code;
2969 * FIXME: we should not write CR2 when L1 intercepts an L2 #PF exception.
2970 * The fix is to add the ancillary datum (CR2 or DR6) to structs
2971 * kvm_queued_exception and kvm_vcpu_events, so that CR2 and DR6 can be
2972 * written only when inject_pending_event runs (DR6 would written here
2973 * too). This should be conditional on a new capability---if the
2974 * capability is disabled, kvm_multiple_exception would write the
2975 * ancillary information to CR2 or DR6, for backwards ABI-compatibility.
2977 if (svm->vcpu.arch.exception.nested_apf)
2978 svm->vmcb->control.exit_info_2 = svm->vcpu.arch.apf.nested_apf_token;
2980 svm->vmcb->control.exit_info_2 = svm->vcpu.arch.cr2;
2982 svm->nested.exit_required = true;
2986 /* This function returns true if it is save to enable the irq window */
2987 static inline bool nested_svm_intr(struct vcpu_svm *svm)
2989 if (!is_guest_mode(&svm->vcpu))
2992 if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
2995 if (!(svm->vcpu.arch.hflags & HF_HIF_MASK))
2999 * if vmexit was already requested (by intercepted exception
3000 * for instance) do not overwrite it with "external interrupt"
3003 if (svm->nested.exit_required)
3006 svm->vmcb->control.exit_code = SVM_EXIT_INTR;
3007 svm->vmcb->control.exit_info_1 = 0;
3008 svm->vmcb->control.exit_info_2 = 0;
3010 if (svm->nested.intercept & 1ULL) {
3012 * The #vmexit can't be emulated here directly because this
3013 * code path runs with irqs and preemption disabled. A
3014 * #vmexit emulation might sleep. Only signal request for
3017 svm->nested.exit_required = true;
3018 trace_kvm_nested_intr_vmexit(svm->vmcb->save.rip);
3025 /* This function returns true if it is save to enable the nmi window */
3026 static inline bool nested_svm_nmi(struct vcpu_svm *svm)
3028 if (!is_guest_mode(&svm->vcpu))
3031 if (!(svm->nested.intercept & (1ULL << INTERCEPT_NMI)))
3034 svm->vmcb->control.exit_code = SVM_EXIT_NMI;
3035 svm->nested.exit_required = true;
3040 static void *nested_svm_map(struct vcpu_svm *svm, u64 gpa, struct page **_page)
3046 page = kvm_vcpu_gfn_to_page(&svm->vcpu, gpa >> PAGE_SHIFT);
3047 if (is_error_page(page))
3055 kvm_inject_gp(&svm->vcpu, 0);
3060 static void nested_svm_unmap(struct page *page)
3063 kvm_release_page_dirty(page);
3066 static int nested_svm_intercept_ioio(struct vcpu_svm *svm)
3068 unsigned port, size, iopm_len;
3073 if (!(svm->nested.intercept & (1ULL << INTERCEPT_IOIO_PROT)))
3074 return NESTED_EXIT_HOST;
3076 port = svm->vmcb->control.exit_info_1 >> 16;
3077 size = (svm->vmcb->control.exit_info_1 & SVM_IOIO_SIZE_MASK) >>
3078 SVM_IOIO_SIZE_SHIFT;
3079 gpa = svm->nested.vmcb_iopm + (port / 8);
3080 start_bit = port % 8;
3081 iopm_len = (start_bit + size > 8) ? 2 : 1;
3082 mask = (0xf >> (4 - size)) << start_bit;
3085 if (kvm_vcpu_read_guest(&svm->vcpu, gpa, &val, iopm_len))
3086 return NESTED_EXIT_DONE;
3088 return (val & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
3091 static int nested_svm_exit_handled_msr(struct vcpu_svm *svm)
3093 u32 offset, msr, value;
3096 if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT)))
3097 return NESTED_EXIT_HOST;
3099 msr = svm->vcpu.arch.regs[VCPU_REGS_RCX];
3100 offset = svm_msrpm_offset(msr);
3101 write = svm->vmcb->control.exit_info_1 & 1;
3102 mask = 1 << ((2 * (msr & 0xf)) + write);
3104 if (offset == MSR_INVALID)
3105 return NESTED_EXIT_DONE;
3107 /* Offset is in 32 bit units but need in 8 bit units */
3110 if (kvm_vcpu_read_guest(&svm->vcpu, svm->nested.vmcb_msrpm + offset, &value, 4))
3111 return NESTED_EXIT_DONE;
3113 return (value & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
3116 /* DB exceptions for our internal use must not cause vmexit */
3117 static int nested_svm_intercept_db(struct vcpu_svm *svm)
3121 /* if we're not singlestepping, it's not ours */
3122 if (!svm->nmi_singlestep)
3123 return NESTED_EXIT_DONE;
3125 /* if it's not a singlestep exception, it's not ours */
3126 if (kvm_get_dr(&svm->vcpu, 6, &dr6))
3127 return NESTED_EXIT_DONE;
3128 if (!(dr6 & DR6_BS))
3129 return NESTED_EXIT_DONE;
3131 /* if the guest is singlestepping, it should get the vmexit */
3132 if (svm->nmi_singlestep_guest_rflags & X86_EFLAGS_TF) {
3133 disable_nmi_singlestep(svm);
3134 return NESTED_EXIT_DONE;
3137 /* it's ours, the nested hypervisor must not see this one */
3138 return NESTED_EXIT_HOST;
3141 static int nested_svm_exit_special(struct vcpu_svm *svm)
3143 u32 exit_code = svm->vmcb->control.exit_code;
3145 switch (exit_code) {
3148 case SVM_EXIT_EXCP_BASE + MC_VECTOR:
3149 return NESTED_EXIT_HOST;
3151 /* For now we are always handling NPFs when using them */
3153 return NESTED_EXIT_HOST;
3155 case SVM_EXIT_EXCP_BASE + PF_VECTOR:
3156 /* When we're shadowing, trap PFs, but not async PF */
3157 if (!npt_enabled && svm->vcpu.arch.apf.host_apf_reason == 0)
3158 return NESTED_EXIT_HOST;
3164 return NESTED_EXIT_CONTINUE;
3168 * If this function returns true, this #vmexit was already handled
3170 static int nested_svm_intercept(struct vcpu_svm *svm)
3172 u32 exit_code = svm->vmcb->control.exit_code;
3173 int vmexit = NESTED_EXIT_HOST;
3175 switch (exit_code) {
3177 vmexit = nested_svm_exit_handled_msr(svm);
3180 vmexit = nested_svm_intercept_ioio(svm);
3182 case SVM_EXIT_READ_CR0 ... SVM_EXIT_WRITE_CR8: {
3183 u32 bit = 1U << (exit_code - SVM_EXIT_READ_CR0);
3184 if (svm->nested.intercept_cr & bit)
3185 vmexit = NESTED_EXIT_DONE;
3188 case SVM_EXIT_READ_DR0 ... SVM_EXIT_WRITE_DR7: {
3189 u32 bit = 1U << (exit_code - SVM_EXIT_READ_DR0);
3190 if (svm->nested.intercept_dr & bit)
3191 vmexit = NESTED_EXIT_DONE;
3194 case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
3195 u32 excp_bits = 1 << (exit_code - SVM_EXIT_EXCP_BASE);
3196 if (svm->nested.intercept_exceptions & excp_bits) {
3197 if (exit_code == SVM_EXIT_EXCP_BASE + DB_VECTOR)
3198 vmexit = nested_svm_intercept_db(svm);
3200 vmexit = NESTED_EXIT_DONE;
3202 /* async page fault always cause vmexit */
3203 else if ((exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR) &&
3204 svm->vcpu.arch.exception.nested_apf != 0)
3205 vmexit = NESTED_EXIT_DONE;
3208 case SVM_EXIT_ERR: {
3209 vmexit = NESTED_EXIT_DONE;
3213 u64 exit_bits = 1ULL << (exit_code - SVM_EXIT_INTR);
3214 if (svm->nested.intercept & exit_bits)
3215 vmexit = NESTED_EXIT_DONE;
3222 static int nested_svm_exit_handled(struct vcpu_svm *svm)
3226 vmexit = nested_svm_intercept(svm);
3228 if (vmexit == NESTED_EXIT_DONE)
3229 nested_svm_vmexit(svm);
3234 static inline void copy_vmcb_control_area(struct vmcb *dst_vmcb, struct vmcb *from_vmcb)
3236 struct vmcb_control_area *dst = &dst_vmcb->control;
3237 struct vmcb_control_area *from = &from_vmcb->control;
3239 dst->intercept_cr = from->intercept_cr;
3240 dst->intercept_dr = from->intercept_dr;
3241 dst->intercept_exceptions = from->intercept_exceptions;
3242 dst->intercept = from->intercept;
3243 dst->iopm_base_pa = from->iopm_base_pa;
3244 dst->msrpm_base_pa = from->msrpm_base_pa;
3245 dst->tsc_offset = from->tsc_offset;
3246 dst->asid = from->asid;
3247 dst->tlb_ctl = from->tlb_ctl;
3248 dst->int_ctl = from->int_ctl;
3249 dst->int_vector = from->int_vector;
3250 dst->int_state = from->int_state;
3251 dst->exit_code = from->exit_code;
3252 dst->exit_code_hi = from->exit_code_hi;
3253 dst->exit_info_1 = from->exit_info_1;
3254 dst->exit_info_2 = from->exit_info_2;
3255 dst->exit_int_info = from->exit_int_info;
3256 dst->exit_int_info_err = from->exit_int_info_err;
3257 dst->nested_ctl = from->nested_ctl;
3258 dst->event_inj = from->event_inj;
3259 dst->event_inj_err = from->event_inj_err;
3260 dst->nested_cr3 = from->nested_cr3;
3261 dst->virt_ext = from->virt_ext;
3264 static int nested_svm_vmexit(struct vcpu_svm *svm)
3266 struct vmcb *nested_vmcb;
3267 struct vmcb *hsave = svm->nested.hsave;
3268 struct vmcb *vmcb = svm->vmcb;
3271 trace_kvm_nested_vmexit_inject(vmcb->control.exit_code,
3272 vmcb->control.exit_info_1,
3273 vmcb->control.exit_info_2,
3274 vmcb->control.exit_int_info,
3275 vmcb->control.exit_int_info_err,
3278 nested_vmcb = nested_svm_map(svm, svm->nested.vmcb, &page);
3282 /* Exit Guest-Mode */
3283 leave_guest_mode(&svm->vcpu);
3284 svm->nested.vmcb = 0;
3286 /* Give the current vmcb to the guest */
3289 nested_vmcb->save.es = vmcb->save.es;
3290 nested_vmcb->save.cs = vmcb->save.cs;
3291 nested_vmcb->save.ss = vmcb->save.ss;
3292 nested_vmcb->save.ds = vmcb->save.ds;
3293 nested_vmcb->save.gdtr = vmcb->save.gdtr;
3294 nested_vmcb->save.idtr = vmcb->save.idtr;
3295 nested_vmcb->save.efer = svm->vcpu.arch.efer;
3296 nested_vmcb->save.cr0 = kvm_read_cr0(&svm->vcpu);
3297 nested_vmcb->save.cr3 = kvm_read_cr3(&svm->vcpu);
3298 nested_vmcb->save.cr2 = vmcb->save.cr2;
3299 nested_vmcb->save.cr4 = svm->vcpu.arch.cr4;
3300 nested_vmcb->save.rflags = kvm_get_rflags(&svm->vcpu);
3301 nested_vmcb->save.rip = vmcb->save.rip;
3302 nested_vmcb->save.rsp = vmcb->save.rsp;
3303 nested_vmcb->save.rax = vmcb->save.rax;
3304 nested_vmcb->save.dr7 = vmcb->save.dr7;
3305 nested_vmcb->save.dr6 = vmcb->save.dr6;
3306 nested_vmcb->save.cpl = vmcb->save.cpl;
3308 nested_vmcb->control.int_ctl = vmcb->control.int_ctl;
3309 nested_vmcb->control.int_vector = vmcb->control.int_vector;
3310 nested_vmcb->control.int_state = vmcb->control.int_state;
3311 nested_vmcb->control.exit_code = vmcb->control.exit_code;
3312 nested_vmcb->control.exit_code_hi = vmcb->control.exit_code_hi;
3313 nested_vmcb->control.exit_info_1 = vmcb->control.exit_info_1;
3314 nested_vmcb->control.exit_info_2 = vmcb->control.exit_info_2;
3315 nested_vmcb->control.exit_int_info = vmcb->control.exit_int_info;
3316 nested_vmcb->control.exit_int_info_err = vmcb->control.exit_int_info_err;
3318 if (svm->nrips_enabled)
3319 nested_vmcb->control.next_rip = vmcb->control.next_rip;
3322 * If we emulate a VMRUN/#VMEXIT in the same host #vmexit cycle we have
3323 * to make sure that we do not lose injected events. So check event_inj
3324 * here and copy it to exit_int_info if it is valid.
3325 * Exit_int_info and event_inj can't be both valid because the case
3326 * below only happens on a VMRUN instruction intercept which has
3327 * no valid exit_int_info set.
3329 if (vmcb->control.event_inj & SVM_EVTINJ_VALID) {
3330 struct vmcb_control_area *nc = &nested_vmcb->control;
3332 nc->exit_int_info = vmcb->control.event_inj;
3333 nc->exit_int_info_err = vmcb->control.event_inj_err;
3336 nested_vmcb->control.tlb_ctl = 0;
3337 nested_vmcb->control.event_inj = 0;
3338 nested_vmcb->control.event_inj_err = 0;
3340 /* We always set V_INTR_MASKING and remember the old value in hflags */
3341 if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
3342 nested_vmcb->control.int_ctl &= ~V_INTR_MASKING_MASK;
3344 /* Restore the original control entries */
3345 copy_vmcb_control_area(vmcb, hsave);
3347 svm->vcpu.arch.tsc_offset = svm->vmcb->control.tsc_offset;
3348 kvm_clear_exception_queue(&svm->vcpu);
3349 kvm_clear_interrupt_queue(&svm->vcpu);
3351 svm->nested.nested_cr3 = 0;
3353 /* Restore selected save entries */
3354 svm->vmcb->save.es = hsave->save.es;
3355 svm->vmcb->save.cs = hsave->save.cs;
3356 svm->vmcb->save.ss = hsave->save.ss;
3357 svm->vmcb->save.ds = hsave->save.ds;
3358 svm->vmcb->save.gdtr = hsave->save.gdtr;
3359 svm->vmcb->save.idtr = hsave->save.idtr;
3360 kvm_set_rflags(&svm->vcpu, hsave->save.rflags);
3361 svm_set_efer(&svm->vcpu, hsave->save.efer);
3362 svm_set_cr0(&svm->vcpu, hsave->save.cr0 | X86_CR0_PE);
3363 svm_set_cr4(&svm->vcpu, hsave->save.cr4);
3365 svm->vmcb->save.cr3 = hsave->save.cr3;
3366 svm->vcpu.arch.cr3 = hsave->save.cr3;
3368 (void)kvm_set_cr3(&svm->vcpu, hsave->save.cr3);
3370 kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, hsave->save.rax);
3371 kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, hsave->save.rsp);
3372 kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, hsave->save.rip);
3373 svm->vmcb->save.dr7 = 0;
3374 svm->vmcb->save.cpl = 0;
3375 svm->vmcb->control.exit_int_info = 0;
3377 mark_all_dirty(svm->vmcb);
3379 nested_svm_unmap(page);
3381 nested_svm_uninit_mmu_context(&svm->vcpu);
3382 kvm_mmu_reset_context(&svm->vcpu);
3383 kvm_mmu_load(&svm->vcpu);
3388 static bool nested_svm_vmrun_msrpm(struct vcpu_svm *svm)
3391 * This function merges the msr permission bitmaps of kvm and the
3392 * nested vmcb. It is optimized in that it only merges the parts where
3393 * the kvm msr permission bitmap may contain zero bits
3397 if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT)))
3400 for (i = 0; i < MSRPM_OFFSETS; i++) {
3404 if (msrpm_offsets[i] == 0xffffffff)
3407 p = msrpm_offsets[i];
3408 offset = svm->nested.vmcb_msrpm + (p * 4);
3410 if (kvm_vcpu_read_guest(&svm->vcpu, offset, &value, 4))
3413 svm->nested.msrpm[p] = svm->msrpm[p] | value;
3416 svm->vmcb->control.msrpm_base_pa = __sme_set(__pa(svm->nested.msrpm));
3421 static bool nested_vmcb_checks(struct vmcb *vmcb)
3423 if ((vmcb->control.intercept & (1ULL << INTERCEPT_VMRUN)) == 0)
3426 if (vmcb->control.asid == 0)
3429 if ((vmcb->control.nested_ctl & SVM_NESTED_CTL_NP_ENABLE) &&
3436 static void enter_svm_guest_mode(struct vcpu_svm *svm, u64 vmcb_gpa,
3437 struct vmcb *nested_vmcb, struct page *page)
3439 if (kvm_get_rflags(&svm->vcpu) & X86_EFLAGS_IF)
3440 svm->vcpu.arch.hflags |= HF_HIF_MASK;
3442 svm->vcpu.arch.hflags &= ~HF_HIF_MASK;
3444 if (nested_vmcb->control.nested_ctl & SVM_NESTED_CTL_NP_ENABLE) {
3445 kvm_mmu_unload(&svm->vcpu);
3446 svm->nested.nested_cr3 = nested_vmcb->control.nested_cr3;
3447 nested_svm_init_mmu_context(&svm->vcpu);
3450 /* Load the nested guest state */
3451 svm->vmcb->save.es = nested_vmcb->save.es;
3452 svm->vmcb->save.cs = nested_vmcb->save.cs;
3453 svm->vmcb->save.ss = nested_vmcb->save.ss;
3454 svm->vmcb->save.ds = nested_vmcb->save.ds;
3455 svm->vmcb->save.gdtr = nested_vmcb->save.gdtr;
3456 svm->vmcb->save.idtr = nested_vmcb->save.idtr;
3457 kvm_set_rflags(&svm->vcpu, nested_vmcb->save.rflags);
3458 svm_set_efer(&svm->vcpu, nested_vmcb->save.efer);
3459 svm_set_cr0(&svm->vcpu, nested_vmcb->save.cr0);
3460 svm_set_cr4(&svm->vcpu, nested_vmcb->save.cr4);
3462 svm->vmcb->save.cr3 = nested_vmcb->save.cr3;
3463 svm->vcpu.arch.cr3 = nested_vmcb->save.cr3;
3465 (void)kvm_set_cr3(&svm->vcpu, nested_vmcb->save.cr3);
3467 /* Guest paging mode is active - reset mmu */
3468 kvm_mmu_reset_context(&svm->vcpu);
3470 svm->vmcb->save.cr2 = svm->vcpu.arch.cr2 = nested_vmcb->save.cr2;
3471 kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, nested_vmcb->save.rax);
3472 kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, nested_vmcb->save.rsp);
3473 kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, nested_vmcb->save.rip);
3475 /* In case we don't even reach vcpu_run, the fields are not updated */
3476 svm->vmcb->save.rax = nested_vmcb->save.rax;
3477 svm->vmcb->save.rsp = nested_vmcb->save.rsp;
3478 svm->vmcb->save.rip = nested_vmcb->save.rip;
3479 svm->vmcb->save.dr7 = nested_vmcb->save.dr7;
3480 svm->vmcb->save.dr6 = nested_vmcb->save.dr6;
3481 svm->vmcb->save.cpl = nested_vmcb->save.cpl;
3483 svm->nested.vmcb_msrpm = nested_vmcb->control.msrpm_base_pa & ~0x0fffULL;
3484 svm->nested.vmcb_iopm = nested_vmcb->control.iopm_base_pa & ~0x0fffULL;
3486 /* cache intercepts */
3487 svm->nested.intercept_cr = nested_vmcb->control.intercept_cr;
3488 svm->nested.intercept_dr = nested_vmcb->control.intercept_dr;
3489 svm->nested.intercept_exceptions = nested_vmcb->control.intercept_exceptions;
3490 svm->nested.intercept = nested_vmcb->control.intercept;
3492 svm_flush_tlb(&svm->vcpu, true);
3493 svm->vmcb->control.int_ctl = nested_vmcb->control.int_ctl | V_INTR_MASKING_MASK;
3494 if (nested_vmcb->control.int_ctl & V_INTR_MASKING_MASK)
3495 svm->vcpu.arch.hflags |= HF_VINTR_MASK;
3497 svm->vcpu.arch.hflags &= ~HF_VINTR_MASK;
3499 if (svm->vcpu.arch.hflags & HF_VINTR_MASK) {
3500 /* We only want the cr8 intercept bits of the guest */
3501 clr_cr_intercept(svm, INTERCEPT_CR8_READ);
3502 clr_cr_intercept(svm, INTERCEPT_CR8_WRITE);
3505 /* We don't want to see VMMCALLs from a nested guest */
3506 clr_intercept(svm, INTERCEPT_VMMCALL);
3508 svm->vcpu.arch.tsc_offset += nested_vmcb->control.tsc_offset;
3509 svm->vmcb->control.tsc_offset = svm->vcpu.arch.tsc_offset;
3511 svm->vmcb->control.virt_ext = nested_vmcb->control.virt_ext;
3512 svm->vmcb->control.int_vector = nested_vmcb->control.int_vector;
3513 svm->vmcb->control.int_state = nested_vmcb->control.int_state;
3514 svm->vmcb->control.event_inj = nested_vmcb->control.event_inj;
3515 svm->vmcb->control.event_inj_err = nested_vmcb->control.event_inj_err;
3517 nested_svm_unmap(page);
3519 /* Enter Guest-Mode */
3520 enter_guest_mode(&svm->vcpu);
3523 * Merge guest and host intercepts - must be called with vcpu in
3524 * guest-mode to take affect here
3526 recalc_intercepts(svm);
3528 svm->nested.vmcb = vmcb_gpa;
3532 mark_all_dirty(svm->vmcb);
3535 static bool nested_svm_vmrun(struct vcpu_svm *svm)
3537 struct vmcb *nested_vmcb;
3538 struct vmcb *hsave = svm->nested.hsave;
3539 struct vmcb *vmcb = svm->vmcb;
3543 vmcb_gpa = svm->vmcb->save.rax;
3545 nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
3549 if (!nested_vmcb_checks(nested_vmcb)) {
3550 nested_vmcb->control.exit_code = SVM_EXIT_ERR;
3551 nested_vmcb->control.exit_code_hi = 0;
3552 nested_vmcb->control.exit_info_1 = 0;
3553 nested_vmcb->control.exit_info_2 = 0;
3555 nested_svm_unmap(page);
3560 trace_kvm_nested_vmrun(svm->vmcb->save.rip, vmcb_gpa,
3561 nested_vmcb->save.rip,
3562 nested_vmcb->control.int_ctl,
3563 nested_vmcb->control.event_inj,
3564 nested_vmcb->control.nested_ctl);
3566 trace_kvm_nested_intercepts(nested_vmcb->control.intercept_cr & 0xffff,
3567 nested_vmcb->control.intercept_cr >> 16,
3568 nested_vmcb->control.intercept_exceptions,
3569 nested_vmcb->control.intercept);
3571 /* Clear internal status */
3572 kvm_clear_exception_queue(&svm->vcpu);
3573 kvm_clear_interrupt_queue(&svm->vcpu);
3576 * Save the old vmcb, so we don't need to pick what we save, but can
3577 * restore everything when a VMEXIT occurs
3579 hsave->save.es = vmcb->save.es;
3580 hsave->save.cs = vmcb->save.cs;
3581 hsave->save.ss = vmcb->save.ss;
3582 hsave->save.ds = vmcb->save.ds;
3583 hsave->save.gdtr = vmcb->save.gdtr;
3584 hsave->save.idtr = vmcb->save.idtr;
3585 hsave->save.efer = svm->vcpu.arch.efer;
3586 hsave->save.cr0 = kvm_read_cr0(&svm->vcpu);
3587 hsave->save.cr4 = svm->vcpu.arch.cr4;
3588 hsave->save.rflags = kvm_get_rflags(&svm->vcpu);
3589 hsave->save.rip = kvm_rip_read(&svm->vcpu);
3590 hsave->save.rsp = vmcb->save.rsp;
3591 hsave->save.rax = vmcb->save.rax;
3593 hsave->save.cr3 = vmcb->save.cr3;
3595 hsave->save.cr3 = kvm_read_cr3(&svm->vcpu);
3597 copy_vmcb_control_area(hsave, vmcb);
3599 enter_svm_guest_mode(svm, vmcb_gpa, nested_vmcb, page);
3604 static void nested_svm_vmloadsave(struct vmcb *from_vmcb, struct vmcb *to_vmcb)
3606 to_vmcb->save.fs = from_vmcb->save.fs;
3607 to_vmcb->save.gs = from_vmcb->save.gs;
3608 to_vmcb->save.tr = from_vmcb->save.tr;
3609 to_vmcb->save.ldtr = from_vmcb->save.ldtr;
3610 to_vmcb->save.kernel_gs_base = from_vmcb->save.kernel_gs_base;
3611 to_vmcb->save.star = from_vmcb->save.star;
3612 to_vmcb->save.lstar = from_vmcb->save.lstar;
3613 to_vmcb->save.cstar = from_vmcb->save.cstar;
3614 to_vmcb->save.sfmask = from_vmcb->save.sfmask;
3615 to_vmcb->save.sysenter_cs = from_vmcb->save.sysenter_cs;
3616 to_vmcb->save.sysenter_esp = from_vmcb->save.sysenter_esp;
3617 to_vmcb->save.sysenter_eip = from_vmcb->save.sysenter_eip;
3620 static int vmload_interception(struct vcpu_svm *svm)
3622 struct vmcb *nested_vmcb;
3626 if (nested_svm_check_permissions(svm))
3629 nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
3633 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
3634 ret = kvm_skip_emulated_instruction(&svm->vcpu);
3636 nested_svm_vmloadsave(nested_vmcb, svm->vmcb);
3637 nested_svm_unmap(page);
3642 static int vmsave_interception(struct vcpu_svm *svm)
3644 struct vmcb *nested_vmcb;
3648 if (nested_svm_check_permissions(svm))
3651 nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
3655 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
3656 ret = kvm_skip_emulated_instruction(&svm->vcpu);
3658 nested_svm_vmloadsave(svm->vmcb, nested_vmcb);
3659 nested_svm_unmap(page);
3664 static int vmrun_interception(struct vcpu_svm *svm)
3666 if (nested_svm_check_permissions(svm))
3669 /* Save rip after vmrun instruction */
3670 kvm_rip_write(&svm->vcpu, kvm_rip_read(&svm->vcpu) + 3);
3672 if (!nested_svm_vmrun(svm))
3675 if (!nested_svm_vmrun_msrpm(svm))
3682 svm->vmcb->control.exit_code = SVM_EXIT_ERR;
3683 svm->vmcb->control.exit_code_hi = 0;
3684 svm->vmcb->control.exit_info_1 = 0;
3685 svm->vmcb->control.exit_info_2 = 0;
3687 nested_svm_vmexit(svm);
3692 static int stgi_interception(struct vcpu_svm *svm)
3696 if (nested_svm_check_permissions(svm))
3700 * If VGIF is enabled, the STGI intercept is only added to
3701 * detect the opening of the SMI/NMI window; remove it now.
3703 if (vgif_enabled(svm))
3704 clr_intercept(svm, INTERCEPT_STGI);
3706 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
3707 ret = kvm_skip_emulated_instruction(&svm->vcpu);
3708 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
3715 static int clgi_interception(struct vcpu_svm *svm)
3719 if (nested_svm_check_permissions(svm))
3722 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
3723 ret = kvm_skip_emulated_instruction(&svm->vcpu);
3727 /* After a CLGI no interrupts should come */
3728 if (!kvm_vcpu_apicv_active(&svm->vcpu)) {
3729 svm_clear_vintr(svm);
3730 svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
3731 mark_dirty(svm->vmcb, VMCB_INTR);
3737 static int invlpga_interception(struct vcpu_svm *svm)
3739 struct kvm_vcpu *vcpu = &svm->vcpu;
3741 trace_kvm_invlpga(svm->vmcb->save.rip, kvm_register_read(&svm->vcpu, VCPU_REGS_RCX),
3742 kvm_register_read(&svm->vcpu, VCPU_REGS_RAX));
3744 /* Let's treat INVLPGA the same as INVLPG (can be optimized!) */
3745 kvm_mmu_invlpg(vcpu, kvm_register_read(&svm->vcpu, VCPU_REGS_RAX));
3747 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
3748 return kvm_skip_emulated_instruction(&svm->vcpu);
3751 static int skinit_interception(struct vcpu_svm *svm)
3753 trace_kvm_skinit(svm->vmcb->save.rip, kvm_register_read(&svm->vcpu, VCPU_REGS_RAX));
3755 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
3759 static int wbinvd_interception(struct vcpu_svm *svm)
3761 return kvm_emulate_wbinvd(&svm->vcpu);
3764 static int xsetbv_interception(struct vcpu_svm *svm)
3766 u64 new_bv = kvm_read_edx_eax(&svm->vcpu);
3767 u32 index = kvm_register_read(&svm->vcpu, VCPU_REGS_RCX);
3769 if (kvm_set_xcr(&svm->vcpu, index, new_bv) == 0) {
3770 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
3771 return kvm_skip_emulated_instruction(&svm->vcpu);
3777 static int task_switch_interception(struct vcpu_svm *svm)
3781 int int_type = svm->vmcb->control.exit_int_info &
3782 SVM_EXITINTINFO_TYPE_MASK;
3783 int int_vec = svm->vmcb->control.exit_int_info & SVM_EVTINJ_VEC_MASK;
3785 svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_TYPE_MASK;
3787 svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_VALID;
3788 bool has_error_code = false;
3791 tss_selector = (u16)svm->vmcb->control.exit_info_1;
3793 if (svm->vmcb->control.exit_info_2 &
3794 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_IRET))
3795 reason = TASK_SWITCH_IRET;
3796 else if (svm->vmcb->control.exit_info_2 &
3797 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_JMP))
3798 reason = TASK_SWITCH_JMP;
3800 reason = TASK_SWITCH_GATE;
3802 reason = TASK_SWITCH_CALL;
3804 if (reason == TASK_SWITCH_GATE) {
3806 case SVM_EXITINTINFO_TYPE_NMI:
3807 svm->vcpu.arch.nmi_injected = false;
3809 case SVM_EXITINTINFO_TYPE_EXEPT:
3810 if (svm->vmcb->control.exit_info_2 &
3811 (1ULL << SVM_EXITINFOSHIFT_TS_HAS_ERROR_CODE)) {
3812 has_error_code = true;
3814 (u32)svm->vmcb->control.exit_info_2;
3816 kvm_clear_exception_queue(&svm->vcpu);
3818 case SVM_EXITINTINFO_TYPE_INTR:
3819 kvm_clear_interrupt_queue(&svm->vcpu);
3826 if (reason != TASK_SWITCH_GATE ||
3827 int_type == SVM_EXITINTINFO_TYPE_SOFT ||
3828 (int_type == SVM_EXITINTINFO_TYPE_EXEPT &&
3829 (int_vec == OF_VECTOR || int_vec == BP_VECTOR)))
3830 skip_emulated_instruction(&svm->vcpu);
3832 if (int_type != SVM_EXITINTINFO_TYPE_SOFT)
3835 if (kvm_task_switch(&svm->vcpu, tss_selector, int_vec, reason,
3836 has_error_code, error_code) == EMULATE_FAIL) {
3837 svm->vcpu.run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3838 svm->vcpu.run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
3839 svm->vcpu.run->internal.ndata = 0;
3845 static int cpuid_interception(struct vcpu_svm *svm)
3847 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
3848 return kvm_emulate_cpuid(&svm->vcpu);
3851 static int iret_interception(struct vcpu_svm *svm)
3853 ++svm->vcpu.stat.nmi_window_exits;
3854 clr_intercept(svm, INTERCEPT_IRET);
3855 svm->vcpu.arch.hflags |= HF_IRET_MASK;
3856 svm->nmi_iret_rip = kvm_rip_read(&svm->vcpu);
3857 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
3861 static int invlpg_interception(struct vcpu_svm *svm)
3863 if (!static_cpu_has(X86_FEATURE_DECODEASSISTS))
3864 return emulate_instruction(&svm->vcpu, 0) == EMULATE_DONE;
3866 kvm_mmu_invlpg(&svm->vcpu, svm->vmcb->control.exit_info_1);
3867 return kvm_skip_emulated_instruction(&svm->vcpu);
3870 static int emulate_on_interception(struct vcpu_svm *svm)
3872 return emulate_instruction(&svm->vcpu, 0) == EMULATE_DONE;
3875 static int rsm_interception(struct vcpu_svm *svm)
3877 return x86_emulate_instruction(&svm->vcpu, 0, 0,
3878 rsm_ins_bytes, 2) == EMULATE_DONE;
3881 static int rdpmc_interception(struct vcpu_svm *svm)
3885 if (!static_cpu_has(X86_FEATURE_NRIPS))
3886 return emulate_on_interception(svm);
3888 err = kvm_rdpmc(&svm->vcpu);
3889 return kvm_complete_insn_gp(&svm->vcpu, err);
3892 static bool check_selective_cr0_intercepted(struct vcpu_svm *svm,
3895 unsigned long cr0 = svm->vcpu.arch.cr0;
3899 intercept = svm->nested.intercept;
3901 if (!is_guest_mode(&svm->vcpu) ||
3902 (!(intercept & (1ULL << INTERCEPT_SELECTIVE_CR0))))
3905 cr0 &= ~SVM_CR0_SELECTIVE_MASK;
3906 val &= ~SVM_CR0_SELECTIVE_MASK;
3909 svm->vmcb->control.exit_code = SVM_EXIT_CR0_SEL_WRITE;
3910 ret = (nested_svm_exit_handled(svm) == NESTED_EXIT_DONE);
3916 #define CR_VALID (1ULL << 63)
3918 static int cr_interception(struct vcpu_svm *svm)
3924 if (!static_cpu_has(X86_FEATURE_DECODEASSISTS))
3925 return emulate_on_interception(svm);
3927 if (unlikely((svm->vmcb->control.exit_info_1 & CR_VALID) == 0))
3928 return emulate_on_interception(svm);
3930 reg = svm->vmcb->control.exit_info_1 & SVM_EXITINFO_REG_MASK;
3931 if (svm->vmcb->control.exit_code == SVM_EXIT_CR0_SEL_WRITE)
3932 cr = SVM_EXIT_WRITE_CR0 - SVM_EXIT_READ_CR0;
3934 cr = svm->vmcb->control.exit_code - SVM_EXIT_READ_CR0;
3937 if (cr >= 16) { /* mov to cr */
3939 val = kvm_register_read(&svm->vcpu, reg);
3942 if (!check_selective_cr0_intercepted(svm, val))
3943 err = kvm_set_cr0(&svm->vcpu, val);
3949 err = kvm_set_cr3(&svm->vcpu, val);
3952 err = kvm_set_cr4(&svm->vcpu, val);
3955 err = kvm_set_cr8(&svm->vcpu, val);
3958 WARN(1, "unhandled write to CR%d", cr);
3959 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
3962 } else { /* mov from cr */
3965 val = kvm_read_cr0(&svm->vcpu);
3968 val = svm->vcpu.arch.cr2;
3971 val = kvm_read_cr3(&svm->vcpu);
3974 val = kvm_read_cr4(&svm->vcpu);
3977 val = kvm_get_cr8(&svm->vcpu);
3980 WARN(1, "unhandled read from CR%d", cr);
3981 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
3984 kvm_register_write(&svm->vcpu, reg, val);
3986 return kvm_complete_insn_gp(&svm->vcpu, err);
3989 static int dr_interception(struct vcpu_svm *svm)
3994 if (svm->vcpu.guest_debug == 0) {
3996 * No more DR vmexits; force a reload of the debug registers
3997 * and reenter on this instruction. The next vmexit will
3998 * retrieve the full state of the debug registers.
4000 clr_dr_intercepts(svm);
4001 svm->vcpu.arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
4005 if (!boot_cpu_has(X86_FEATURE_DECODEASSISTS))
4006 return emulate_on_interception(svm);
4008 reg = svm->vmcb->control.exit_info_1 & SVM_EXITINFO_REG_MASK;
4009 dr = svm->vmcb->control.exit_code - SVM_EXIT_READ_DR0;
4011 if (dr >= 16) { /* mov to DRn */
4012 if (!kvm_require_dr(&svm->vcpu, dr - 16))
4014 val = kvm_register_read(&svm->vcpu, reg);
4015 kvm_set_dr(&svm->vcpu, dr - 16, val);
4017 if (!kvm_require_dr(&svm->vcpu, dr))
4019 kvm_get_dr(&svm->vcpu, dr, &val);
4020 kvm_register_write(&svm->vcpu, reg, val);
4023 return kvm_skip_emulated_instruction(&svm->vcpu);
4026 static int cr8_write_interception(struct vcpu_svm *svm)
4028 struct kvm_run *kvm_run = svm->vcpu.run;
4031 u8 cr8_prev = kvm_get_cr8(&svm->vcpu);
4032 /* instruction emulation calls kvm_set_cr8() */
4033 r = cr_interception(svm);
4034 if (lapic_in_kernel(&svm->vcpu))
4036 if (cr8_prev <= kvm_get_cr8(&svm->vcpu))
4038 kvm_run->exit_reason = KVM_EXIT_SET_TPR;
4042 static int svm_get_msr_feature(struct kvm_msr_entry *msr)
4046 switch (msr->index) {
4047 case MSR_F10H_DECFG:
4048 if (boot_cpu_has(X86_FEATURE_LFENCE_RDTSC))
4049 msr->data |= MSR_F10H_DECFG_LFENCE_SERIALIZE;
4058 static int svm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
4060 struct vcpu_svm *svm = to_svm(vcpu);
4062 switch (msr_info->index) {
4064 msr_info->data = svm->vmcb->save.star;
4066 #ifdef CONFIG_X86_64
4068 msr_info->data = svm->vmcb->save.lstar;
4071 msr_info->data = svm->vmcb->save.cstar;
4073 case MSR_KERNEL_GS_BASE:
4074 msr_info->data = svm->vmcb->save.kernel_gs_base;
4076 case MSR_SYSCALL_MASK:
4077 msr_info->data = svm->vmcb->save.sfmask;
4080 case MSR_IA32_SYSENTER_CS:
4081 msr_info->data = svm->vmcb->save.sysenter_cs;
4083 case MSR_IA32_SYSENTER_EIP:
4084 msr_info->data = svm->sysenter_eip;
4086 case MSR_IA32_SYSENTER_ESP:
4087 msr_info->data = svm->sysenter_esp;
4090 if (!boot_cpu_has(X86_FEATURE_RDTSCP))
4092 msr_info->data = svm->tsc_aux;
4095 * Nobody will change the following 5 values in the VMCB so we can
4096 * safely return them on rdmsr. They will always be 0 until LBRV is
4099 case MSR_IA32_DEBUGCTLMSR:
4100 msr_info->data = svm->vmcb->save.dbgctl;
4102 case MSR_IA32_LASTBRANCHFROMIP:
4103 msr_info->data = svm->vmcb->save.br_from;
4105 case MSR_IA32_LASTBRANCHTOIP:
4106 msr_info->data = svm->vmcb->save.br_to;
4108 case MSR_IA32_LASTINTFROMIP:
4109 msr_info->data = svm->vmcb->save.last_excp_from;
4111 case MSR_IA32_LASTINTTOIP:
4112 msr_info->data = svm->vmcb->save.last_excp_to;
4114 case MSR_VM_HSAVE_PA:
4115 msr_info->data = svm->nested.hsave_msr;
4118 msr_info->data = svm->nested.vm_cr_msr;
4120 case MSR_IA32_SPEC_CTRL:
4121 if (!msr_info->host_initiated &&
4122 !guest_cpuid_has(vcpu, X86_FEATURE_AMD_IBRS) &&
4123 !guest_cpuid_has(vcpu, X86_FEATURE_AMD_SSBD))
4126 msr_info->data = svm->spec_ctrl;
4128 case MSR_AMD64_VIRT_SPEC_CTRL:
4129 if (!msr_info->host_initiated &&
4130 !guest_cpuid_has(vcpu, X86_FEATURE_VIRT_SSBD))
4133 msr_info->data = svm->virt_spec_ctrl;
4135 case MSR_F15H_IC_CFG: {
4139 family = guest_cpuid_family(vcpu);
4140 model = guest_cpuid_model(vcpu);
4142 if (family < 0 || model < 0)
4143 return kvm_get_msr_common(vcpu, msr_info);
4147 if (family == 0x15 &&
4148 (model >= 0x2 && model < 0x20))
4149 msr_info->data = 0x1E;
4152 case MSR_F10H_DECFG:
4153 msr_info->data = svm->msr_decfg;
4156 return kvm_get_msr_common(vcpu, msr_info);
4161 static int rdmsr_interception(struct vcpu_svm *svm)
4163 u32 ecx = kvm_register_read(&svm->vcpu, VCPU_REGS_RCX);
4164 struct msr_data msr_info;
4166 msr_info.index = ecx;
4167 msr_info.host_initiated = false;
4168 if (svm_get_msr(&svm->vcpu, &msr_info)) {
4169 trace_kvm_msr_read_ex(ecx);
4170 kvm_inject_gp(&svm->vcpu, 0);
4173 trace_kvm_msr_read(ecx, msr_info.data);
4175 kvm_register_write(&svm->vcpu, VCPU_REGS_RAX,
4176 msr_info.data & 0xffffffff);
4177 kvm_register_write(&svm->vcpu, VCPU_REGS_RDX,
4178 msr_info.data >> 32);
4179 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
4180 return kvm_skip_emulated_instruction(&svm->vcpu);
4184 static int svm_set_vm_cr(struct kvm_vcpu *vcpu, u64 data)
4186 struct vcpu_svm *svm = to_svm(vcpu);
4187 int svm_dis, chg_mask;
4189 if (data & ~SVM_VM_CR_VALID_MASK)
4192 chg_mask = SVM_VM_CR_VALID_MASK;
4194 if (svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK)
4195 chg_mask &= ~(SVM_VM_CR_SVM_LOCK_MASK | SVM_VM_CR_SVM_DIS_MASK);
4197 svm->nested.vm_cr_msr &= ~chg_mask;
4198 svm->nested.vm_cr_msr |= (data & chg_mask);
4200 svm_dis = svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK;
4202 /* check for svm_disable while efer.svme is set */
4203 if (svm_dis && (vcpu->arch.efer & EFER_SVME))
4209 static int svm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
4211 struct vcpu_svm *svm = to_svm(vcpu);
4213 u32 ecx = msr->index;
4214 u64 data = msr->data;
4216 case MSR_IA32_CR_PAT:
4217 if (!kvm_mtrr_valid(vcpu, MSR_IA32_CR_PAT, data))
4219 vcpu->arch.pat = data;
4220 svm->vmcb->save.g_pat = data;
4221 mark_dirty(svm->vmcb, VMCB_NPT);
4223 case MSR_IA32_SPEC_CTRL:
4224 if (!msr->host_initiated &&
4225 !guest_cpuid_has(vcpu, X86_FEATURE_AMD_IBRS) &&
4226 !guest_cpuid_has(vcpu, X86_FEATURE_AMD_SSBD))
4229 /* The STIBP bit doesn't fault even if it's not advertised */
4230 if (data & ~(SPEC_CTRL_IBRS | SPEC_CTRL_STIBP | SPEC_CTRL_SSBD))
4233 svm->spec_ctrl = data;
4240 * When it's written (to non-zero) for the first time, pass
4244 * The handling of the MSR bitmap for L2 guests is done in
4245 * nested_svm_vmrun_msrpm.
4246 * We update the L1 MSR bit as well since it will end up
4247 * touching the MSR anyway now.
4249 set_msr_interception(svm->msrpm, MSR_IA32_SPEC_CTRL, 1, 1);
4251 case MSR_IA32_PRED_CMD:
4252 if (!msr->host_initiated &&
4253 !guest_cpuid_has(vcpu, X86_FEATURE_AMD_IBPB))
4256 if (data & ~PRED_CMD_IBPB)
4262 wrmsrl(MSR_IA32_PRED_CMD, PRED_CMD_IBPB);
4263 if (is_guest_mode(vcpu))
4265 set_msr_interception(svm->msrpm, MSR_IA32_PRED_CMD, 0, 1);
4267 case MSR_AMD64_VIRT_SPEC_CTRL:
4268 if (!msr->host_initiated &&
4269 !guest_cpuid_has(vcpu, X86_FEATURE_VIRT_SSBD))
4272 if (data & ~SPEC_CTRL_SSBD)
4275 svm->virt_spec_ctrl = data;
4278 svm->vmcb->save.star = data;
4280 #ifdef CONFIG_X86_64
4282 svm->vmcb->save.lstar = data;
4285 svm->vmcb->save.cstar = data;
4287 case MSR_KERNEL_GS_BASE:
4288 svm->vmcb->save.kernel_gs_base = data;
4290 case MSR_SYSCALL_MASK:
4291 svm->vmcb->save.sfmask = data;
4294 case MSR_IA32_SYSENTER_CS:
4295 svm->vmcb->save.sysenter_cs = data;
4297 case MSR_IA32_SYSENTER_EIP:
4298 svm->sysenter_eip = data;
4299 svm->vmcb->save.sysenter_eip = data;
4301 case MSR_IA32_SYSENTER_ESP:
4302 svm->sysenter_esp = data;
4303 svm->vmcb->save.sysenter_esp = data;
4306 if (!boot_cpu_has(X86_FEATURE_RDTSCP))
4310 * This is rare, so we update the MSR here instead of using
4311 * direct_access_msrs. Doing that would require a rdmsr in
4314 svm->tsc_aux = data;
4315 wrmsrl(MSR_TSC_AUX, svm->tsc_aux);
4317 case MSR_IA32_DEBUGCTLMSR:
4318 if (!boot_cpu_has(X86_FEATURE_LBRV)) {
4319 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTL 0x%llx, nop\n",
4323 if (data & DEBUGCTL_RESERVED_BITS)
4326 svm->vmcb->save.dbgctl = data;
4327 mark_dirty(svm->vmcb, VMCB_LBR);
4328 if (data & (1ULL<<0))
4329 svm_enable_lbrv(svm);
4331 svm_disable_lbrv(svm);
4333 case MSR_VM_HSAVE_PA:
4334 svm->nested.hsave_msr = data;
4337 return svm_set_vm_cr(vcpu, data);
4339 vcpu_unimpl(vcpu, "unimplemented wrmsr: 0x%x data 0x%llx\n", ecx, data);
4341 case MSR_F10H_DECFG: {
4342 struct kvm_msr_entry msr_entry;
4344 msr_entry.index = msr->index;
4345 if (svm_get_msr_feature(&msr_entry))
4348 /* Check the supported bits */
4349 if (data & ~msr_entry.data)
4352 /* Don't allow the guest to change a bit, #GP */
4353 if (!msr->host_initiated && (data ^ msr_entry.data))
4356 svm->msr_decfg = data;
4359 case MSR_IA32_APICBASE:
4360 if (kvm_vcpu_apicv_active(vcpu))
4361 avic_update_vapic_bar(to_svm(vcpu), data);
4362 /* Follow through */
4364 return kvm_set_msr_common(vcpu, msr);
4369 static int wrmsr_interception(struct vcpu_svm *svm)
4371 struct msr_data msr;
4372 u32 ecx = kvm_register_read(&svm->vcpu, VCPU_REGS_RCX);
4373 u64 data = kvm_read_edx_eax(&svm->vcpu);
4377 msr.host_initiated = false;
4379 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
4380 if (kvm_set_msr(&svm->vcpu, &msr)) {
4381 trace_kvm_msr_write_ex(ecx, data);
4382 kvm_inject_gp(&svm->vcpu, 0);
4385 trace_kvm_msr_write(ecx, data);
4386 return kvm_skip_emulated_instruction(&svm->vcpu);
4390 static int msr_interception(struct vcpu_svm *svm)
4392 if (svm->vmcb->control.exit_info_1)
4393 return wrmsr_interception(svm);
4395 return rdmsr_interception(svm);
4398 static int interrupt_window_interception(struct vcpu_svm *svm)
4400 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
4401 svm_clear_vintr(svm);
4402 svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
4403 mark_dirty(svm->vmcb, VMCB_INTR);
4404 ++svm->vcpu.stat.irq_window_exits;
4408 static int pause_interception(struct vcpu_svm *svm)
4410 struct kvm_vcpu *vcpu = &svm->vcpu;
4411 bool in_kernel = (svm_get_cpl(vcpu) == 0);
4413 if (pause_filter_thresh)
4414 grow_ple_window(vcpu);
4416 kvm_vcpu_on_spin(vcpu, in_kernel);
4420 static int nop_interception(struct vcpu_svm *svm)
4422 return kvm_skip_emulated_instruction(&(svm->vcpu));
4425 static int monitor_interception(struct vcpu_svm *svm)
4427 printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
4428 return nop_interception(svm);
4431 static int mwait_interception(struct vcpu_svm *svm)
4433 printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
4434 return nop_interception(svm);
4437 enum avic_ipi_failure_cause {
4438 AVIC_IPI_FAILURE_INVALID_INT_TYPE,
4439 AVIC_IPI_FAILURE_TARGET_NOT_RUNNING,
4440 AVIC_IPI_FAILURE_INVALID_TARGET,
4441 AVIC_IPI_FAILURE_INVALID_BACKING_PAGE,
4444 static int avic_incomplete_ipi_interception(struct vcpu_svm *svm)
4446 u32 icrh = svm->vmcb->control.exit_info_1 >> 32;
4447 u32 icrl = svm->vmcb->control.exit_info_1;
4448 u32 id = svm->vmcb->control.exit_info_2 >> 32;
4449 u32 index = svm->vmcb->control.exit_info_2 & 0xFF;
4450 struct kvm_lapic *apic = svm->vcpu.arch.apic;
4452 trace_kvm_avic_incomplete_ipi(svm->vcpu.vcpu_id, icrh, icrl, id, index);
4455 case AVIC_IPI_FAILURE_INVALID_INT_TYPE:
4457 * AVIC hardware handles the generation of
4458 * IPIs when the specified Message Type is Fixed
4459 * (also known as fixed delivery mode) and
4460 * the Trigger Mode is edge-triggered. The hardware
4461 * also supports self and broadcast delivery modes
4462 * specified via the Destination Shorthand(DSH)
4463 * field of the ICRL. Logical and physical APIC ID
4464 * formats are supported. All other IPI types cause
4465 * a #VMEXIT, which needs to emulated.
4467 kvm_lapic_reg_write(apic, APIC_ICR2, icrh);
4468 kvm_lapic_reg_write(apic, APIC_ICR, icrl);
4470 case AVIC_IPI_FAILURE_TARGET_NOT_RUNNING: {
4472 struct kvm_vcpu *vcpu;
4473 struct kvm *kvm = svm->vcpu.kvm;
4474 struct kvm_lapic *apic = svm->vcpu.arch.apic;
4477 * At this point, we expect that the AVIC HW has already
4478 * set the appropriate IRR bits on the valid target
4479 * vcpus. So, we just need to kick the appropriate vcpu.
4481 kvm_for_each_vcpu(i, vcpu, kvm) {
4482 bool m = kvm_apic_match_dest(vcpu, apic,
4483 icrl & KVM_APIC_SHORT_MASK,
4484 GET_APIC_DEST_FIELD(icrh),
4485 icrl & KVM_APIC_DEST_MASK);
4487 if (m && !avic_vcpu_is_running(vcpu))
4488 kvm_vcpu_wake_up(vcpu);
4492 case AVIC_IPI_FAILURE_INVALID_TARGET:
4494 case AVIC_IPI_FAILURE_INVALID_BACKING_PAGE:
4495 WARN_ONCE(1, "Invalid backing page\n");
4498 pr_err("Unknown IPI interception\n");
4504 static u32 *avic_get_logical_id_entry(struct kvm_vcpu *vcpu, u32 ldr, bool flat)
4506 struct kvm_svm *kvm_svm = to_kvm_svm(vcpu->kvm);
4508 u32 *logical_apic_id_table;
4509 int dlid = GET_APIC_LOGICAL_ID(ldr);
4514 if (flat) { /* flat */
4515 index = ffs(dlid) - 1;
4518 } else { /* cluster */
4519 int cluster = (dlid & 0xf0) >> 4;
4520 int apic = ffs(dlid & 0x0f) - 1;
4522 if ((apic < 0) || (apic > 7) ||
4525 index = (cluster << 2) + apic;
4528 logical_apic_id_table = (u32 *) page_address(kvm_svm->avic_logical_id_table_page);
4530 return &logical_apic_id_table[index];
4533 static int avic_ldr_write(struct kvm_vcpu *vcpu, u8 g_physical_id, u32 ldr,
4537 u32 *entry, new_entry;
4539 flat = kvm_lapic_get_reg(vcpu->arch.apic, APIC_DFR) == APIC_DFR_FLAT;
4540 entry = avic_get_logical_id_entry(vcpu, ldr, flat);
4544 new_entry = READ_ONCE(*entry);
4545 new_entry &= ~AVIC_LOGICAL_ID_ENTRY_GUEST_PHYSICAL_ID_MASK;
4546 new_entry |= (g_physical_id & AVIC_LOGICAL_ID_ENTRY_GUEST_PHYSICAL_ID_MASK);
4548 new_entry |= AVIC_LOGICAL_ID_ENTRY_VALID_MASK;
4550 new_entry &= ~AVIC_LOGICAL_ID_ENTRY_VALID_MASK;
4551 WRITE_ONCE(*entry, new_entry);
4556 static int avic_handle_ldr_update(struct kvm_vcpu *vcpu)
4559 struct vcpu_svm *svm = to_svm(vcpu);
4560 u32 ldr = kvm_lapic_get_reg(vcpu->arch.apic, APIC_LDR);
4565 ret = avic_ldr_write(vcpu, vcpu->vcpu_id, ldr, true);
4566 if (ret && svm->ldr_reg) {
4567 avic_ldr_write(vcpu, 0, svm->ldr_reg, false);
4575 static int avic_handle_apic_id_update(struct kvm_vcpu *vcpu)
4578 struct vcpu_svm *svm = to_svm(vcpu);
4579 u32 apic_id_reg = kvm_lapic_get_reg(vcpu->arch.apic, APIC_ID);
4580 u32 id = (apic_id_reg >> 24) & 0xff;
4582 if (vcpu->vcpu_id == id)
4585 old = avic_get_physical_id_entry(vcpu, vcpu->vcpu_id);
4586 new = avic_get_physical_id_entry(vcpu, id);
4590 /* We need to move physical_id_entry to new offset */
4593 to_svm(vcpu)->avic_physical_id_cache = new;
4596 * Also update the guest physical APIC ID in the logical
4597 * APIC ID table entry if already setup the LDR.
4600 avic_handle_ldr_update(vcpu);
4605 static int avic_handle_dfr_update(struct kvm_vcpu *vcpu)
4607 struct vcpu_svm *svm = to_svm(vcpu);
4608 struct kvm_svm *kvm_svm = to_kvm_svm(vcpu->kvm);
4609 u32 dfr = kvm_lapic_get_reg(vcpu->arch.apic, APIC_DFR);
4610 u32 mod = (dfr >> 28) & 0xf;
4613 * We assume that all local APICs are using the same type.
4614 * If this changes, we need to flush the AVIC logical
4617 if (kvm_svm->ldr_mode == mod)
4620 clear_page(page_address(kvm_svm->avic_logical_id_table_page));
4621 kvm_svm->ldr_mode = mod;
4624 avic_handle_ldr_update(vcpu);
4628 static int avic_unaccel_trap_write(struct vcpu_svm *svm)
4630 struct kvm_lapic *apic = svm->vcpu.arch.apic;
4631 u32 offset = svm->vmcb->control.exit_info_1 &
4632 AVIC_UNACCEL_ACCESS_OFFSET_MASK;
4636 if (avic_handle_apic_id_update(&svm->vcpu))
4640 if (avic_handle_ldr_update(&svm->vcpu))
4644 avic_handle_dfr_update(&svm->vcpu);
4650 kvm_lapic_reg_write(apic, offset, kvm_lapic_get_reg(apic, offset));
4655 static bool is_avic_unaccelerated_access_trap(u32 offset)
4684 static int avic_unaccelerated_access_interception(struct vcpu_svm *svm)
4687 u32 offset = svm->vmcb->control.exit_info_1 &
4688 AVIC_UNACCEL_ACCESS_OFFSET_MASK;
4689 u32 vector = svm->vmcb->control.exit_info_2 &
4690 AVIC_UNACCEL_ACCESS_VECTOR_MASK;
4691 bool write = (svm->vmcb->control.exit_info_1 >> 32) &
4692 AVIC_UNACCEL_ACCESS_WRITE_MASK;
4693 bool trap = is_avic_unaccelerated_access_trap(offset);
4695 trace_kvm_avic_unaccelerated_access(svm->vcpu.vcpu_id, offset,
4696 trap, write, vector);
4699 WARN_ONCE(!write, "svm: Handling trap read.\n");
4700 ret = avic_unaccel_trap_write(svm);
4702 /* Handling Fault */
4703 ret = (emulate_instruction(&svm->vcpu, 0) == EMULATE_DONE);
4709 static int (*const svm_exit_handlers[])(struct vcpu_svm *svm) = {
4710 [SVM_EXIT_READ_CR0] = cr_interception,
4711 [SVM_EXIT_READ_CR3] = cr_interception,
4712 [SVM_EXIT_READ_CR4] = cr_interception,
4713 [SVM_EXIT_READ_CR8] = cr_interception,
4714 [SVM_EXIT_CR0_SEL_WRITE] = cr_interception,
4715 [SVM_EXIT_WRITE_CR0] = cr_interception,
4716 [SVM_EXIT_WRITE_CR3] = cr_interception,
4717 [SVM_EXIT_WRITE_CR4] = cr_interception,
4718 [SVM_EXIT_WRITE_CR8] = cr8_write_interception,
4719 [SVM_EXIT_READ_DR0] = dr_interception,
4720 [SVM_EXIT_READ_DR1] = dr_interception,
4721 [SVM_EXIT_READ_DR2] = dr_interception,
4722 [SVM_EXIT_READ_DR3] = dr_interception,
4723 [SVM_EXIT_READ_DR4] = dr_interception,
4724 [SVM_EXIT_READ_DR5] = dr_interception,
4725 [SVM_EXIT_READ_DR6] = dr_interception,
4726 [SVM_EXIT_READ_DR7] = dr_interception,
4727 [SVM_EXIT_WRITE_DR0] = dr_interception,
4728 [SVM_EXIT_WRITE_DR1] = dr_interception,
4729 [SVM_EXIT_WRITE_DR2] = dr_interception,
4730 [SVM_EXIT_WRITE_DR3] = dr_interception,
4731 [SVM_EXIT_WRITE_DR4] = dr_interception,
4732 [SVM_EXIT_WRITE_DR5] = dr_interception,
4733 [SVM_EXIT_WRITE_DR6] = dr_interception,
4734 [SVM_EXIT_WRITE_DR7] = dr_interception,
4735 [SVM_EXIT_EXCP_BASE + DB_VECTOR] = db_interception,
4736 [SVM_EXIT_EXCP_BASE + BP_VECTOR] = bp_interception,
4737 [SVM_EXIT_EXCP_BASE + UD_VECTOR] = ud_interception,
4738 [SVM_EXIT_EXCP_BASE + PF_VECTOR] = pf_interception,
4739 [SVM_EXIT_EXCP_BASE + MC_VECTOR] = mc_interception,
4740 [SVM_EXIT_EXCP_BASE + AC_VECTOR] = ac_interception,
4741 [SVM_EXIT_EXCP_BASE + GP_VECTOR] = gp_interception,
4742 [SVM_EXIT_INTR] = intr_interception,
4743 [SVM_EXIT_NMI] = nmi_interception,
4744 [SVM_EXIT_SMI] = nop_on_interception,
4745 [SVM_EXIT_INIT] = nop_on_interception,
4746 [SVM_EXIT_VINTR] = interrupt_window_interception,
4747 [SVM_EXIT_RDPMC] = rdpmc_interception,
4748 [SVM_EXIT_CPUID] = cpuid_interception,
4749 [SVM_EXIT_IRET] = iret_interception,
4750 [SVM_EXIT_INVD] = emulate_on_interception,
4751 [SVM_EXIT_PAUSE] = pause_interception,
4752 [SVM_EXIT_HLT] = halt_interception,
4753 [SVM_EXIT_INVLPG] = invlpg_interception,
4754 [SVM_EXIT_INVLPGA] = invlpga_interception,
4755 [SVM_EXIT_IOIO] = io_interception,
4756 [SVM_EXIT_MSR] = msr_interception,
4757 [SVM_EXIT_TASK_SWITCH] = task_switch_interception,
4758 [SVM_EXIT_SHUTDOWN] = shutdown_interception,
4759 [SVM_EXIT_VMRUN] = vmrun_interception,
4760 [SVM_EXIT_VMMCALL] = vmmcall_interception,
4761 [SVM_EXIT_VMLOAD] = vmload_interception,
4762 [SVM_EXIT_VMSAVE] = vmsave_interception,
4763 [SVM_EXIT_STGI] = stgi_interception,
4764 [SVM_EXIT_CLGI] = clgi_interception,
4765 [SVM_EXIT_SKINIT] = skinit_interception,
4766 [SVM_EXIT_WBINVD] = wbinvd_interception,
4767 [SVM_EXIT_MONITOR] = monitor_interception,
4768 [SVM_EXIT_MWAIT] = mwait_interception,
4769 [SVM_EXIT_XSETBV] = xsetbv_interception,
4770 [SVM_EXIT_NPF] = npf_interception,
4771 [SVM_EXIT_RSM] = rsm_interception,
4772 [SVM_EXIT_AVIC_INCOMPLETE_IPI] = avic_incomplete_ipi_interception,
4773 [SVM_EXIT_AVIC_UNACCELERATED_ACCESS] = avic_unaccelerated_access_interception,
4776 static void dump_vmcb(struct kvm_vcpu *vcpu)
4778 struct vcpu_svm *svm = to_svm(vcpu);
4779 struct vmcb_control_area *control = &svm->vmcb->control;
4780 struct vmcb_save_area *save = &svm->vmcb->save;
4782 pr_err("VMCB Control Area:\n");
4783 pr_err("%-20s%04x\n", "cr_read:", control->intercept_cr & 0xffff);
4784 pr_err("%-20s%04x\n", "cr_write:", control->intercept_cr >> 16);
4785 pr_err("%-20s%04x\n", "dr_read:", control->intercept_dr & 0xffff);
4786 pr_err("%-20s%04x\n", "dr_write:", control->intercept_dr >> 16);
4787 pr_err("%-20s%08x\n", "exceptions:", control->intercept_exceptions);
4788 pr_err("%-20s%016llx\n", "intercepts:", control->intercept);
4789 pr_err("%-20s%d\n", "pause filter count:", control->pause_filter_count);
4790 pr_err("%-20s%d\n", "pause filter threshold:",
4791 control->pause_filter_thresh);
4792 pr_err("%-20s%016llx\n", "iopm_base_pa:", control->iopm_base_pa);
4793 pr_err("%-20s%016llx\n", "msrpm_base_pa:", control->msrpm_base_pa);
4794 pr_err("%-20s%016llx\n", "tsc_offset:", control->tsc_offset);
4795 pr_err("%-20s%d\n", "asid:", control->asid);
4796 pr_err("%-20s%d\n", "tlb_ctl:", control->tlb_ctl);
4797 pr_err("%-20s%08x\n", "int_ctl:", control->int_ctl);
4798 pr_err("%-20s%08x\n", "int_vector:", control->int_vector);
4799 pr_err("%-20s%08x\n", "int_state:", control->int_state);
4800 pr_err("%-20s%08x\n", "exit_code:", control->exit_code);
4801 pr_err("%-20s%016llx\n", "exit_info1:", control->exit_info_1);
4802 pr_err("%-20s%016llx\n", "exit_info2:", control->exit_info_2);
4803 pr_err("%-20s%08x\n", "exit_int_info:", control->exit_int_info);
4804 pr_err("%-20s%08x\n", "exit_int_info_err:", control->exit_int_info_err);
4805 pr_err("%-20s%lld\n", "nested_ctl:", control->nested_ctl);
4806 pr_err("%-20s%016llx\n", "nested_cr3:", control->nested_cr3);
4807 pr_err("%-20s%016llx\n", "avic_vapic_bar:", control->avic_vapic_bar);
4808 pr_err("%-20s%08x\n", "event_inj:", control->event_inj);
4809 pr_err("%-20s%08x\n", "event_inj_err:", control->event_inj_err);
4810 pr_err("%-20s%lld\n", "virt_ext:", control->virt_ext);
4811 pr_err("%-20s%016llx\n", "next_rip:", control->next_rip);
4812 pr_err("%-20s%016llx\n", "avic_backing_page:", control->avic_backing_page);
4813 pr_err("%-20s%016llx\n", "avic_logical_id:", control->avic_logical_id);
4814 pr_err("%-20s%016llx\n", "avic_physical_id:", control->avic_physical_id);
4815 pr_err("VMCB State Save Area:\n");
4816 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4818 save->es.selector, save->es.attrib,
4819 save->es.limit, save->es.base);
4820 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4822 save->cs.selector, save->cs.attrib,
4823 save->cs.limit, save->cs.base);
4824 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4826 save->ss.selector, save->ss.attrib,
4827 save->ss.limit, save->ss.base);
4828 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4830 save->ds.selector, save->ds.attrib,
4831 save->ds.limit, save->ds.base);
4832 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4834 save->fs.selector, save->fs.attrib,
4835 save->fs.limit, save->fs.base);
4836 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4838 save->gs.selector, save->gs.attrib,
4839 save->gs.limit, save->gs.base);
4840 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4842 save->gdtr.selector, save->gdtr.attrib,
4843 save->gdtr.limit, save->gdtr.base);
4844 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4846 save->ldtr.selector, save->ldtr.attrib,
4847 save->ldtr.limit, save->ldtr.base);
4848 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4850 save->idtr.selector, save->idtr.attrib,
4851 save->idtr.limit, save->idtr.base);
4852 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4854 save->tr.selector, save->tr.attrib,
4855 save->tr.limit, save->tr.base);
4856 pr_err("cpl: %d efer: %016llx\n",
4857 save->cpl, save->efer);
4858 pr_err("%-15s %016llx %-13s %016llx\n",
4859 "cr0:", save->cr0, "cr2:", save->cr2);
4860 pr_err("%-15s %016llx %-13s %016llx\n",
4861 "cr3:", save->cr3, "cr4:", save->cr4);
4862 pr_err("%-15s %016llx %-13s %016llx\n",
4863 "dr6:", save->dr6, "dr7:", save->dr7);
4864 pr_err("%-15s %016llx %-13s %016llx\n",
4865 "rip:", save->rip, "rflags:", save->rflags);
4866 pr_err("%-15s %016llx %-13s %016llx\n",
4867 "rsp:", save->rsp, "rax:", save->rax);
4868 pr_err("%-15s %016llx %-13s %016llx\n",
4869 "star:", save->star, "lstar:", save->lstar);
4870 pr_err("%-15s %016llx %-13s %016llx\n",
4871 "cstar:", save->cstar, "sfmask:", save->sfmask);
4872 pr_err("%-15s %016llx %-13s %016llx\n",
4873 "kernel_gs_base:", save->kernel_gs_base,
4874 "sysenter_cs:", save->sysenter_cs);
4875 pr_err("%-15s %016llx %-13s %016llx\n",
4876 "sysenter_esp:", save->sysenter_esp,
4877 "sysenter_eip:", save->sysenter_eip);
4878 pr_err("%-15s %016llx %-13s %016llx\n",
4879 "gpat:", save->g_pat, "dbgctl:", save->dbgctl);
4880 pr_err("%-15s %016llx %-13s %016llx\n",
4881 "br_from:", save->br_from, "br_to:", save->br_to);
4882 pr_err("%-15s %016llx %-13s %016llx\n",
4883 "excp_from:", save->last_excp_from,
4884 "excp_to:", save->last_excp_to);
4887 static void svm_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
4889 struct vmcb_control_area *control = &to_svm(vcpu)->vmcb->control;
4891 *info1 = control->exit_info_1;
4892 *info2 = control->exit_info_2;
4895 static int handle_exit(struct kvm_vcpu *vcpu)
4897 struct vcpu_svm *svm = to_svm(vcpu);
4898 struct kvm_run *kvm_run = vcpu->run;
4899 u32 exit_code = svm->vmcb->control.exit_code;
4901 trace_kvm_exit(exit_code, vcpu, KVM_ISA_SVM);
4903 if (!is_cr_intercept(svm, INTERCEPT_CR0_WRITE))
4904 vcpu->arch.cr0 = svm->vmcb->save.cr0;
4906 vcpu->arch.cr3 = svm->vmcb->save.cr3;
4908 if (unlikely(svm->nested.exit_required)) {
4909 nested_svm_vmexit(svm);
4910 svm->nested.exit_required = false;
4915 if (is_guest_mode(vcpu)) {
4918 trace_kvm_nested_vmexit(svm->vmcb->save.rip, exit_code,
4919 svm->vmcb->control.exit_info_1,
4920 svm->vmcb->control.exit_info_2,
4921 svm->vmcb->control.exit_int_info,
4922 svm->vmcb->control.exit_int_info_err,
4925 vmexit = nested_svm_exit_special(svm);
4927 if (vmexit == NESTED_EXIT_CONTINUE)
4928 vmexit = nested_svm_exit_handled(svm);
4930 if (vmexit == NESTED_EXIT_DONE)
4934 svm_complete_interrupts(svm);
4936 if (svm->vmcb->control.exit_code == SVM_EXIT_ERR) {
4937 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4938 kvm_run->fail_entry.hardware_entry_failure_reason
4939 = svm->vmcb->control.exit_code;
4940 pr_err("KVM: FAILED VMRUN WITH VMCB:\n");
4945 if (is_external_interrupt(svm->vmcb->control.exit_int_info) &&
4946 exit_code != SVM_EXIT_EXCP_BASE + PF_VECTOR &&
4947 exit_code != SVM_EXIT_NPF && exit_code != SVM_EXIT_TASK_SWITCH &&
4948 exit_code != SVM_EXIT_INTR && exit_code != SVM_EXIT_NMI)
4949 printk(KERN_ERR "%s: unexpected exit_int_info 0x%x "
4951 __func__, svm->vmcb->control.exit_int_info,
4954 if (exit_code >= ARRAY_SIZE(svm_exit_handlers)
4955 || !svm_exit_handlers[exit_code]) {
4956 WARN_ONCE(1, "svm: unexpected exit reason 0x%x\n", exit_code);
4957 kvm_queue_exception(vcpu, UD_VECTOR);
4961 return svm_exit_handlers[exit_code](svm);
4964 static void reload_tss(struct kvm_vcpu *vcpu)
4966 int cpu = raw_smp_processor_id();
4968 struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
4969 sd->tss_desc->type = 9; /* available 32/64-bit TSS */
4973 static void pre_sev_run(struct vcpu_svm *svm, int cpu)
4975 struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
4976 int asid = sev_get_asid(svm->vcpu.kvm);
4978 /* Assign the asid allocated with this SEV guest */
4979 svm->vmcb->control.asid = asid;
4984 * 1) when different VMCB for the same ASID is to be run on the same host CPU.
4985 * 2) or this VMCB was executed on different host CPU in previous VMRUNs.
4987 if (sd->sev_vmcbs[asid] == svm->vmcb &&
4988 svm->last_cpu == cpu)
4991 svm->last_cpu = cpu;
4992 sd->sev_vmcbs[asid] = svm->vmcb;
4993 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ASID;
4994 mark_dirty(svm->vmcb, VMCB_ASID);
4997 static void pre_svm_run(struct vcpu_svm *svm)
4999 int cpu = raw_smp_processor_id();
5001 struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
5003 if (sev_guest(svm->vcpu.kvm))
5004 return pre_sev_run(svm, cpu);
5006 /* FIXME: handle wraparound of asid_generation */
5007 if (svm->asid_generation != sd->asid_generation)
5011 static void svm_inject_nmi(struct kvm_vcpu *vcpu)
5013 struct vcpu_svm *svm = to_svm(vcpu);
5015 svm->vmcb->control.event_inj = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI;
5016 vcpu->arch.hflags |= HF_NMI_MASK;
5017 set_intercept(svm, INTERCEPT_IRET);
5018 ++vcpu->stat.nmi_injections;
5021 static inline void svm_inject_irq(struct vcpu_svm *svm, int irq)
5023 struct vmcb_control_area *control;
5025 /* The following fields are ignored when AVIC is enabled */
5026 control = &svm->vmcb->control;
5027 control->int_vector = irq;
5028 control->int_ctl &= ~V_INTR_PRIO_MASK;
5029 control->int_ctl |= V_IRQ_MASK |
5030 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT);
5031 mark_dirty(svm->vmcb, VMCB_INTR);
5034 static void svm_set_irq(struct kvm_vcpu *vcpu)
5036 struct vcpu_svm *svm = to_svm(vcpu);
5038 BUG_ON(!(gif_set(svm)));
5040 trace_kvm_inj_virq(vcpu->arch.interrupt.nr);
5041 ++vcpu->stat.irq_injections;
5043 svm->vmcb->control.event_inj = vcpu->arch.interrupt.nr |
5044 SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR;
5047 static inline bool svm_nested_virtualize_tpr(struct kvm_vcpu *vcpu)
5049 return is_guest_mode(vcpu) && (vcpu->arch.hflags & HF_VINTR_MASK);
5052 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
5054 struct vcpu_svm *svm = to_svm(vcpu);
5056 if (svm_nested_virtualize_tpr(vcpu) ||
5057 kvm_vcpu_apicv_active(vcpu))
5060 clr_cr_intercept(svm, INTERCEPT_CR8_WRITE);
5066 set_cr_intercept(svm, INTERCEPT_CR8_WRITE);
5069 static void svm_set_virtual_apic_mode(struct kvm_vcpu *vcpu)
5074 static bool svm_get_enable_apicv(struct kvm_vcpu *vcpu)
5076 return avic && irqchip_split(vcpu->kvm);
5079 static void svm_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
5083 static void svm_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr)
5087 /* Note: Currently only used by Hyper-V. */
5088 static void svm_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu)
5090 struct vcpu_svm *svm = to_svm(vcpu);
5091 struct vmcb *vmcb = svm->vmcb;
5093 if (!kvm_vcpu_apicv_active(&svm->vcpu))
5096 vmcb->control.int_ctl &= ~AVIC_ENABLE_MASK;
5097 mark_dirty(vmcb, VMCB_INTR);
5100 static void svm_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
5105 static void svm_deliver_avic_intr(struct kvm_vcpu *vcpu, int vec)
5107 kvm_lapic_set_irr(vec, vcpu->arch.apic);
5108 smp_mb__after_atomic();
5110 if (avic_vcpu_is_running(vcpu))
5111 wrmsrl(SVM_AVIC_DOORBELL,
5112 kvm_cpu_get_apicid(vcpu->cpu));
5114 kvm_vcpu_wake_up(vcpu);
5117 static void svm_ir_list_del(struct vcpu_svm *svm, struct amd_iommu_pi_data *pi)
5119 unsigned long flags;
5120 struct amd_svm_iommu_ir *cur;
5122 spin_lock_irqsave(&svm->ir_list_lock, flags);
5123 list_for_each_entry(cur, &svm->ir_list, node) {
5124 if (cur->data != pi->ir_data)
5126 list_del(&cur->node);
5130 spin_unlock_irqrestore(&svm->ir_list_lock, flags);
5133 static int svm_ir_list_add(struct vcpu_svm *svm, struct amd_iommu_pi_data *pi)
5136 unsigned long flags;
5137 struct amd_svm_iommu_ir *ir;
5140 * In some cases, the existing irte is updaed and re-set,
5141 * so we need to check here if it's already been * added
5144 if (pi->ir_data && (pi->prev_ga_tag != 0)) {
5145 struct kvm *kvm = svm->vcpu.kvm;
5146 u32 vcpu_id = AVIC_GATAG_TO_VCPUID(pi->prev_ga_tag);
5147 struct kvm_vcpu *prev_vcpu = kvm_get_vcpu_by_id(kvm, vcpu_id);
5148 struct vcpu_svm *prev_svm;
5155 prev_svm = to_svm(prev_vcpu);
5156 svm_ir_list_del(prev_svm, pi);
5160 * Allocating new amd_iommu_pi_data, which will get
5161 * add to the per-vcpu ir_list.
5163 ir = kzalloc(sizeof(struct amd_svm_iommu_ir), GFP_KERNEL);
5168 ir->data = pi->ir_data;
5170 spin_lock_irqsave(&svm->ir_list_lock, flags);
5171 list_add(&ir->node, &svm->ir_list);
5172 spin_unlock_irqrestore(&svm->ir_list_lock, flags);
5179 * The HW cannot support posting multicast/broadcast
5180 * interrupts to a vCPU. So, we still use legacy interrupt
5181 * remapping for these kind of interrupts.
5183 * For lowest-priority interrupts, we only support
5184 * those with single CPU as the destination, e.g. user
5185 * configures the interrupts via /proc/irq or uses
5186 * irqbalance to make the interrupts single-CPU.
5189 get_pi_vcpu_info(struct kvm *kvm, struct kvm_kernel_irq_routing_entry *e,
5190 struct vcpu_data *vcpu_info, struct vcpu_svm **svm)
5192 struct kvm_lapic_irq irq;
5193 struct kvm_vcpu *vcpu = NULL;
5195 kvm_set_msi_irq(kvm, e, &irq);
5197 if (!kvm_intr_is_single_vcpu(kvm, &irq, &vcpu)) {
5198 pr_debug("SVM: %s: use legacy intr remap mode for irq %u\n",
5199 __func__, irq.vector);
5203 pr_debug("SVM: %s: use GA mode for irq %u\n", __func__,
5205 *svm = to_svm(vcpu);
5206 vcpu_info->pi_desc_addr = __sme_set(page_to_phys((*svm)->avic_backing_page));
5207 vcpu_info->vector = irq.vector;
5213 * svm_update_pi_irte - set IRTE for Posted-Interrupts
5216 * @host_irq: host irq of the interrupt
5217 * @guest_irq: gsi of the interrupt
5218 * @set: set or unset PI
5219 * returns 0 on success, < 0 on failure
5221 static int svm_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
5222 uint32_t guest_irq, bool set)
5224 struct kvm_kernel_irq_routing_entry *e;
5225 struct kvm_irq_routing_table *irq_rt;
5226 int idx, ret = -EINVAL;
5228 if (!kvm_arch_has_assigned_device(kvm) ||
5229 !irq_remapping_cap(IRQ_POSTING_CAP))
5232 pr_debug("SVM: %s: host_irq=%#x, guest_irq=%#x, set=%#x\n",
5233 __func__, host_irq, guest_irq, set);
5235 idx = srcu_read_lock(&kvm->irq_srcu);
5236 irq_rt = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu);
5237 WARN_ON(guest_irq >= irq_rt->nr_rt_entries);
5239 hlist_for_each_entry(e, &irq_rt->map[guest_irq], link) {
5240 struct vcpu_data vcpu_info;
5241 struct vcpu_svm *svm = NULL;
5243 if (e->type != KVM_IRQ_ROUTING_MSI)
5247 * Here, we setup with legacy mode in the following cases:
5248 * 1. When cannot target interrupt to a specific vcpu.
5249 * 2. Unsetting posted interrupt.
5250 * 3. APIC virtialization is disabled for the vcpu.
5252 if (!get_pi_vcpu_info(kvm, e, &vcpu_info, &svm) && set &&
5253 kvm_vcpu_apicv_active(&svm->vcpu)) {
5254 struct amd_iommu_pi_data pi;
5256 /* Try to enable guest_mode in IRTE */
5257 pi.base = __sme_set(page_to_phys(svm->avic_backing_page) &
5259 pi.ga_tag = AVIC_GATAG(to_kvm_svm(kvm)->avic_vm_id,
5261 pi.is_guest_mode = true;
5262 pi.vcpu_data = &vcpu_info;
5263 ret = irq_set_vcpu_affinity(host_irq, &pi);
5266 * Here, we successfully setting up vcpu affinity in
5267 * IOMMU guest mode. Now, we need to store the posted
5268 * interrupt information in a per-vcpu ir_list so that
5269 * we can reference to them directly when we update vcpu
5270 * scheduling information in IOMMU irte.
5272 if (!ret && pi.is_guest_mode)
5273 svm_ir_list_add(svm, &pi);
5275 /* Use legacy mode in IRTE */
5276 struct amd_iommu_pi_data pi;
5279 * Here, pi is used to:
5280 * - Tell IOMMU to use legacy mode for this interrupt.
5281 * - Retrieve ga_tag of prior interrupt remapping data.
5283 pi.is_guest_mode = false;
5284 ret = irq_set_vcpu_affinity(host_irq, &pi);
5287 * Check if the posted interrupt was previously
5288 * setup with the guest_mode by checking if the ga_tag
5289 * was cached. If so, we need to clean up the per-vcpu
5292 if (!ret && pi.prev_ga_tag) {
5293 int id = AVIC_GATAG_TO_VCPUID(pi.prev_ga_tag);
5294 struct kvm_vcpu *vcpu;
5296 vcpu = kvm_get_vcpu_by_id(kvm, id);
5298 svm_ir_list_del(to_svm(vcpu), &pi);
5303 trace_kvm_pi_irte_update(host_irq, svm->vcpu.vcpu_id,
5304 e->gsi, vcpu_info.vector,
5305 vcpu_info.pi_desc_addr, set);
5309 pr_err("%s: failed to update PI IRTE\n", __func__);
5316 srcu_read_unlock(&kvm->irq_srcu, idx);
5320 static int svm_nmi_allowed(struct kvm_vcpu *vcpu)
5322 struct vcpu_svm *svm = to_svm(vcpu);
5323 struct vmcb *vmcb = svm->vmcb;
5325 ret = !(vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK) &&
5326 !(svm->vcpu.arch.hflags & HF_NMI_MASK);
5327 ret = ret && gif_set(svm) && nested_svm_nmi(svm);
5332 static bool svm_get_nmi_mask(struct kvm_vcpu *vcpu)
5334 struct vcpu_svm *svm = to_svm(vcpu);
5336 return !!(svm->vcpu.arch.hflags & HF_NMI_MASK);
5339 static void svm_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
5341 struct vcpu_svm *svm = to_svm(vcpu);
5344 svm->vcpu.arch.hflags |= HF_NMI_MASK;
5345 set_intercept(svm, INTERCEPT_IRET);
5347 svm->vcpu.arch.hflags &= ~HF_NMI_MASK;
5348 clr_intercept(svm, INTERCEPT_IRET);
5352 static int svm_interrupt_allowed(struct kvm_vcpu *vcpu)
5354 struct vcpu_svm *svm = to_svm(vcpu);
5355 struct vmcb *vmcb = svm->vmcb;
5358 if (!gif_set(svm) ||
5359 (vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK))
5362 ret = !!(kvm_get_rflags(vcpu) & X86_EFLAGS_IF);
5364 if (is_guest_mode(vcpu))
5365 return ret && !(svm->vcpu.arch.hflags & HF_VINTR_MASK);
5370 static void enable_irq_window(struct kvm_vcpu *vcpu)
5372 struct vcpu_svm *svm = to_svm(vcpu);
5374 if (kvm_vcpu_apicv_active(vcpu))
5378 * In case GIF=0 we can't rely on the CPU to tell us when GIF becomes
5379 * 1, because that's a separate STGI/VMRUN intercept. The next time we
5380 * get that intercept, this function will be called again though and
5381 * we'll get the vintr intercept. However, if the vGIF feature is
5382 * enabled, the STGI interception will not occur. Enable the irq
5383 * window under the assumption that the hardware will set the GIF.
5385 if ((vgif_enabled(svm) || gif_set(svm)) && nested_svm_intr(svm)) {
5387 svm_inject_irq(svm, 0x0);
5391 static void enable_nmi_window(struct kvm_vcpu *vcpu)
5393 struct vcpu_svm *svm = to_svm(vcpu);
5395 if ((svm->vcpu.arch.hflags & (HF_NMI_MASK | HF_IRET_MASK))
5397 return; /* IRET will cause a vm exit */
5399 if (!gif_set(svm)) {
5400 if (vgif_enabled(svm))
5401 set_intercept(svm, INTERCEPT_STGI);
5402 return; /* STGI will cause a vm exit */
5405 if (svm->nested.exit_required)
5406 return; /* we're not going to run the guest yet */
5409 * Something prevents NMI from been injected. Single step over possible
5410 * problem (IRET or exception injection or interrupt shadow)
5412 svm->nmi_singlestep_guest_rflags = svm_get_rflags(vcpu);
5413 svm->nmi_singlestep = true;
5414 svm->vmcb->save.rflags |= (X86_EFLAGS_TF | X86_EFLAGS_RF);
5417 static int svm_set_tss_addr(struct kvm *kvm, unsigned int addr)
5422 static int svm_set_identity_map_addr(struct kvm *kvm, u64 ident_addr)
5427 static void svm_flush_tlb(struct kvm_vcpu *vcpu, bool invalidate_gpa)
5429 struct vcpu_svm *svm = to_svm(vcpu);
5431 if (static_cpu_has(X86_FEATURE_FLUSHBYASID))
5432 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ASID;
5434 svm->asid_generation--;
5437 static void svm_flush_tlb_gva(struct kvm_vcpu *vcpu, gva_t gva)
5439 struct vcpu_svm *svm = to_svm(vcpu);
5441 invlpga(gva, svm->vmcb->control.asid);
5444 static void svm_prepare_guest_switch(struct kvm_vcpu *vcpu)
5448 static inline void sync_cr8_to_lapic(struct kvm_vcpu *vcpu)
5450 struct vcpu_svm *svm = to_svm(vcpu);
5452 if (svm_nested_virtualize_tpr(vcpu))
5455 if (!is_cr_intercept(svm, INTERCEPT_CR8_WRITE)) {
5456 int cr8 = svm->vmcb->control.int_ctl & V_TPR_MASK;
5457 kvm_set_cr8(vcpu, cr8);
5461 static inline void sync_lapic_to_cr8(struct kvm_vcpu *vcpu)
5463 struct vcpu_svm *svm = to_svm(vcpu);
5466 if (svm_nested_virtualize_tpr(vcpu) ||
5467 kvm_vcpu_apicv_active(vcpu))
5470 cr8 = kvm_get_cr8(vcpu);
5471 svm->vmcb->control.int_ctl &= ~V_TPR_MASK;
5472 svm->vmcb->control.int_ctl |= cr8 & V_TPR_MASK;
5475 static void svm_complete_interrupts(struct vcpu_svm *svm)
5479 u32 exitintinfo = svm->vmcb->control.exit_int_info;
5480 unsigned int3_injected = svm->int3_injected;
5482 svm->int3_injected = 0;
5485 * If we've made progress since setting HF_IRET_MASK, we've
5486 * executed an IRET and can allow NMI injection.
5488 if ((svm->vcpu.arch.hflags & HF_IRET_MASK)
5489 && kvm_rip_read(&svm->vcpu) != svm->nmi_iret_rip) {
5490 svm->vcpu.arch.hflags &= ~(HF_NMI_MASK | HF_IRET_MASK);
5491 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
5494 svm->vcpu.arch.nmi_injected = false;
5495 kvm_clear_exception_queue(&svm->vcpu);
5496 kvm_clear_interrupt_queue(&svm->vcpu);
5498 if (!(exitintinfo & SVM_EXITINTINFO_VALID))
5501 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
5503 vector = exitintinfo & SVM_EXITINTINFO_VEC_MASK;
5504 type = exitintinfo & SVM_EXITINTINFO_TYPE_MASK;
5507 case SVM_EXITINTINFO_TYPE_NMI:
5508 svm->vcpu.arch.nmi_injected = true;
5510 case SVM_EXITINTINFO_TYPE_EXEPT:
5512 * In case of software exceptions, do not reinject the vector,
5513 * but re-execute the instruction instead. Rewind RIP first
5514 * if we emulated INT3 before.
5516 if (kvm_exception_is_soft(vector)) {
5517 if (vector == BP_VECTOR && int3_injected &&
5518 kvm_is_linear_rip(&svm->vcpu, svm->int3_rip))
5519 kvm_rip_write(&svm->vcpu,
5520 kvm_rip_read(&svm->vcpu) -
5524 if (exitintinfo & SVM_EXITINTINFO_VALID_ERR) {
5525 u32 err = svm->vmcb->control.exit_int_info_err;
5526 kvm_requeue_exception_e(&svm->vcpu, vector, err);
5529 kvm_requeue_exception(&svm->vcpu, vector);
5531 case SVM_EXITINTINFO_TYPE_INTR:
5532 kvm_queue_interrupt(&svm->vcpu, vector, false);
5539 static void svm_cancel_injection(struct kvm_vcpu *vcpu)
5541 struct vcpu_svm *svm = to_svm(vcpu);
5542 struct vmcb_control_area *control = &svm->vmcb->control;
5544 control->exit_int_info = control->event_inj;
5545 control->exit_int_info_err = control->event_inj_err;
5546 control->event_inj = 0;
5547 svm_complete_interrupts(svm);
5550 static void svm_vcpu_run(struct kvm_vcpu *vcpu)
5552 struct vcpu_svm *svm = to_svm(vcpu);
5554 svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX];
5555 svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP];
5556 svm->vmcb->save.rip = vcpu->arch.regs[VCPU_REGS_RIP];
5559 * A vmexit emulation is required before the vcpu can be executed
5562 if (unlikely(svm->nested.exit_required))
5566 * Disable singlestep if we're injecting an interrupt/exception.
5567 * We don't want our modified rflags to be pushed on the stack where
5568 * we might not be able to easily reset them if we disabled NMI
5571 if (svm->nmi_singlestep && svm->vmcb->control.event_inj) {
5573 * Event injection happens before external interrupts cause a
5574 * vmexit and interrupts are disabled here, so smp_send_reschedule
5575 * is enough to force an immediate vmexit.
5577 disable_nmi_singlestep(svm);
5578 smp_send_reschedule(vcpu->cpu);
5583 sync_lapic_to_cr8(vcpu);
5585 svm->vmcb->save.cr2 = vcpu->arch.cr2;
5590 * If this vCPU has touched SPEC_CTRL, restore the guest's value if
5591 * it's non-zero. Since vmentry is serialising on affected CPUs, there
5592 * is no need to worry about the conditional branch over the wrmsr
5593 * being speculatively taken.
5595 x86_spec_ctrl_set_guest(svm->spec_ctrl, svm->virt_spec_ctrl);
5600 "push %%" _ASM_BP "; \n\t"
5601 "mov %c[rbx](%[svm]), %%" _ASM_BX " \n\t"
5602 "mov %c[rcx](%[svm]), %%" _ASM_CX " \n\t"
5603 "mov %c[rdx](%[svm]), %%" _ASM_DX " \n\t"
5604 "mov %c[rsi](%[svm]), %%" _ASM_SI " \n\t"
5605 "mov %c[rdi](%[svm]), %%" _ASM_DI " \n\t"
5606 "mov %c[rbp](%[svm]), %%" _ASM_BP " \n\t"
5607 #ifdef CONFIG_X86_64
5608 "mov %c[r8](%[svm]), %%r8 \n\t"
5609 "mov %c[r9](%[svm]), %%r9 \n\t"
5610 "mov %c[r10](%[svm]), %%r10 \n\t"
5611 "mov %c[r11](%[svm]), %%r11 \n\t"
5612 "mov %c[r12](%[svm]), %%r12 \n\t"
5613 "mov %c[r13](%[svm]), %%r13 \n\t"
5614 "mov %c[r14](%[svm]), %%r14 \n\t"
5615 "mov %c[r15](%[svm]), %%r15 \n\t"
5618 /* Enter guest mode */
5619 "push %%" _ASM_AX " \n\t"
5620 "mov %c[vmcb](%[svm]), %%" _ASM_AX " \n\t"
5621 __ex(SVM_VMLOAD) "\n\t"
5622 __ex(SVM_VMRUN) "\n\t"
5623 __ex(SVM_VMSAVE) "\n\t"
5624 "pop %%" _ASM_AX " \n\t"
5626 /* Save guest registers, load host registers */
5627 "mov %%" _ASM_BX ", %c[rbx](%[svm]) \n\t"
5628 "mov %%" _ASM_CX ", %c[rcx](%[svm]) \n\t"
5629 "mov %%" _ASM_DX ", %c[rdx](%[svm]) \n\t"
5630 "mov %%" _ASM_SI ", %c[rsi](%[svm]) \n\t"
5631 "mov %%" _ASM_DI ", %c[rdi](%[svm]) \n\t"
5632 "mov %%" _ASM_BP ", %c[rbp](%[svm]) \n\t"
5633 #ifdef CONFIG_X86_64
5634 "mov %%r8, %c[r8](%[svm]) \n\t"
5635 "mov %%r9, %c[r9](%[svm]) \n\t"
5636 "mov %%r10, %c[r10](%[svm]) \n\t"
5637 "mov %%r11, %c[r11](%[svm]) \n\t"
5638 "mov %%r12, %c[r12](%[svm]) \n\t"
5639 "mov %%r13, %c[r13](%[svm]) \n\t"
5640 "mov %%r14, %c[r14](%[svm]) \n\t"
5641 "mov %%r15, %c[r15](%[svm]) \n\t"
5644 * Clear host registers marked as clobbered to prevent
5647 "xor %%" _ASM_BX ", %%" _ASM_BX " \n\t"
5648 "xor %%" _ASM_CX ", %%" _ASM_CX " \n\t"
5649 "xor %%" _ASM_DX ", %%" _ASM_DX " \n\t"
5650 "xor %%" _ASM_SI ", %%" _ASM_SI " \n\t"
5651 "xor %%" _ASM_DI ", %%" _ASM_DI " \n\t"
5652 #ifdef CONFIG_X86_64
5653 "xor %%r8, %%r8 \n\t"
5654 "xor %%r9, %%r9 \n\t"
5655 "xor %%r10, %%r10 \n\t"
5656 "xor %%r11, %%r11 \n\t"
5657 "xor %%r12, %%r12 \n\t"
5658 "xor %%r13, %%r13 \n\t"
5659 "xor %%r14, %%r14 \n\t"
5660 "xor %%r15, %%r15 \n\t"
5665 [vmcb]"i"(offsetof(struct vcpu_svm, vmcb_pa)),
5666 [rbx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBX])),
5667 [rcx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RCX])),
5668 [rdx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDX])),
5669 [rsi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RSI])),
5670 [rdi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDI])),
5671 [rbp]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBP]))
5672 #ifdef CONFIG_X86_64
5673 , [r8]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R8])),
5674 [r9]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R9])),
5675 [r10]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R10])),
5676 [r11]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R11])),
5677 [r12]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R12])),
5678 [r13]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R13])),
5679 [r14]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R14])),
5680 [r15]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R15]))
5683 #ifdef CONFIG_X86_64
5684 , "rbx", "rcx", "rdx", "rsi", "rdi"
5685 , "r8", "r9", "r10", "r11" , "r12", "r13", "r14", "r15"
5687 , "ebx", "ecx", "edx", "esi", "edi"
5691 /* Eliminate branch target predictions from guest mode */
5694 #ifdef CONFIG_X86_64
5695 wrmsrl(MSR_GS_BASE, svm->host.gs_base);
5697 loadsegment(fs, svm->host.fs);
5698 #ifndef CONFIG_X86_32_LAZY_GS
5699 loadsegment(gs, svm->host.gs);
5704 * We do not use IBRS in the kernel. If this vCPU has used the
5705 * SPEC_CTRL MSR it may have left it on; save the value and
5706 * turn it off. This is much more efficient than blindly adding
5707 * it to the atomic save/restore list. Especially as the former
5708 * (Saving guest MSRs on vmexit) doesn't even exist in KVM.
5710 * For non-nested case:
5711 * If the L01 MSR bitmap does not intercept the MSR, then we need to
5715 * If the L02 MSR bitmap does not intercept the MSR, then we need to
5718 if (unlikely(!msr_write_intercepted(vcpu, MSR_IA32_SPEC_CTRL)))
5719 svm->spec_ctrl = native_read_msr(MSR_IA32_SPEC_CTRL);
5723 local_irq_disable();
5725 x86_spec_ctrl_restore_host(svm->spec_ctrl, svm->virt_spec_ctrl);
5727 vcpu->arch.cr2 = svm->vmcb->save.cr2;
5728 vcpu->arch.regs[VCPU_REGS_RAX] = svm->vmcb->save.rax;
5729 vcpu->arch.regs[VCPU_REGS_RSP] = svm->vmcb->save.rsp;
5730 vcpu->arch.regs[VCPU_REGS_RIP] = svm->vmcb->save.rip;
5732 if (unlikely(svm->vmcb->control.exit_code == SVM_EXIT_NMI))
5733 kvm_before_interrupt(&svm->vcpu);
5737 /* Any pending NMI will happen here */
5739 if (unlikely(svm->vmcb->control.exit_code == SVM_EXIT_NMI))
5740 kvm_after_interrupt(&svm->vcpu);
5742 sync_cr8_to_lapic(vcpu);
5746 svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
5748 /* if exit due to PF check for async PF */
5749 if (svm->vmcb->control.exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR)
5750 svm->vcpu.arch.apf.host_apf_reason = kvm_read_and_reset_pf_reason();
5753 vcpu->arch.regs_avail &= ~(1 << VCPU_EXREG_PDPTR);
5754 vcpu->arch.regs_dirty &= ~(1 << VCPU_EXREG_PDPTR);
5758 * We need to handle MC intercepts here before the vcpu has a chance to
5759 * change the physical cpu
5761 if (unlikely(svm->vmcb->control.exit_code ==
5762 SVM_EXIT_EXCP_BASE + MC_VECTOR))
5763 svm_handle_mce(svm);
5765 mark_all_clean(svm->vmcb);
5767 STACK_FRAME_NON_STANDARD(svm_vcpu_run);
5769 static void svm_set_cr3(struct kvm_vcpu *vcpu, unsigned long root)
5771 struct vcpu_svm *svm = to_svm(vcpu);
5773 svm->vmcb->save.cr3 = __sme_set(root);
5774 mark_dirty(svm->vmcb, VMCB_CR);
5777 static void set_tdp_cr3(struct kvm_vcpu *vcpu, unsigned long root)
5779 struct vcpu_svm *svm = to_svm(vcpu);
5781 svm->vmcb->control.nested_cr3 = __sme_set(root);
5782 mark_dirty(svm->vmcb, VMCB_NPT);
5784 /* Also sync guest cr3 here in case we live migrate */
5785 svm->vmcb->save.cr3 = kvm_read_cr3(vcpu);
5786 mark_dirty(svm->vmcb, VMCB_CR);
5789 static int is_disabled(void)
5793 rdmsrl(MSR_VM_CR, vm_cr);
5794 if (vm_cr & (1 << SVM_VM_CR_SVM_DISABLE))
5801 svm_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
5804 * Patch in the VMMCALL instruction:
5806 hypercall[0] = 0x0f;
5807 hypercall[1] = 0x01;
5808 hypercall[2] = 0xd9;
5811 static void svm_check_processor_compat(void *rtn)
5816 static bool svm_cpu_has_accelerated_tpr(void)
5821 static bool svm_has_emulated_msr(int index)
5826 static u64 svm_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
5831 static void svm_cpuid_update(struct kvm_vcpu *vcpu)
5833 struct vcpu_svm *svm = to_svm(vcpu);
5835 /* Update nrips enabled cache */
5836 svm->nrips_enabled = !!guest_cpuid_has(&svm->vcpu, X86_FEATURE_NRIPS);
5838 if (!kvm_vcpu_apicv_active(vcpu))
5841 guest_cpuid_clear(vcpu, X86_FEATURE_X2APIC);
5844 static void svm_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
5849 entry->ecx &= ~bit(X86_FEATURE_X2APIC);
5853 entry->ecx |= (1 << 2); /* Set SVM bit */
5856 entry->eax = 1; /* SVM revision 1 */
5857 entry->ebx = 8; /* Lets support 8 ASIDs in case we add proper
5858 ASID emulation to nested SVM */
5859 entry->ecx = 0; /* Reserved */
5860 entry->edx = 0; /* Per default do not support any
5861 additional features */
5863 /* Support next_rip if host supports it */
5864 if (boot_cpu_has(X86_FEATURE_NRIPS))
5865 entry->edx |= SVM_FEATURE_NRIP;
5867 /* Support NPT for the guest if enabled */
5869 entry->edx |= SVM_FEATURE_NPT;
5873 /* Support memory encryption cpuid if host supports it */
5874 if (boot_cpu_has(X86_FEATURE_SEV))
5875 cpuid(0x8000001f, &entry->eax, &entry->ebx,
5876 &entry->ecx, &entry->edx);
5881 static int svm_get_lpage_level(void)
5883 return PT_PDPE_LEVEL;
5886 static bool svm_rdtscp_supported(void)
5888 return boot_cpu_has(X86_FEATURE_RDTSCP);
5891 static bool svm_invpcid_supported(void)
5896 static bool svm_mpx_supported(void)
5901 static bool svm_xsaves_supported(void)
5906 static bool svm_umip_emulated(void)
5911 static bool svm_has_wbinvd_exit(void)
5916 #define PRE_EX(exit) { .exit_code = (exit), \
5917 .stage = X86_ICPT_PRE_EXCEPT, }
5918 #define POST_EX(exit) { .exit_code = (exit), \
5919 .stage = X86_ICPT_POST_EXCEPT, }
5920 #define POST_MEM(exit) { .exit_code = (exit), \
5921 .stage = X86_ICPT_POST_MEMACCESS, }
5923 static const struct __x86_intercept {
5925 enum x86_intercept_stage stage;
5926 } x86_intercept_map[] = {
5927 [x86_intercept_cr_read] = POST_EX(SVM_EXIT_READ_CR0),
5928 [x86_intercept_cr_write] = POST_EX(SVM_EXIT_WRITE_CR0),
5929 [x86_intercept_clts] = POST_EX(SVM_EXIT_WRITE_CR0),
5930 [x86_intercept_lmsw] = POST_EX(SVM_EXIT_WRITE_CR0),
5931 [x86_intercept_smsw] = POST_EX(SVM_EXIT_READ_CR0),
5932 [x86_intercept_dr_read] = POST_EX(SVM_EXIT_READ_DR0),
5933 [x86_intercept_dr_write] = POST_EX(SVM_EXIT_WRITE_DR0),
5934 [x86_intercept_sldt] = POST_EX(SVM_EXIT_LDTR_READ),
5935 [x86_intercept_str] = POST_EX(SVM_EXIT_TR_READ),
5936 [x86_intercept_lldt] = POST_EX(SVM_EXIT_LDTR_WRITE),
5937 [x86_intercept_ltr] = POST_EX(SVM_EXIT_TR_WRITE),
5938 [x86_intercept_sgdt] = POST_EX(SVM_EXIT_GDTR_READ),
5939 [x86_intercept_sidt] = POST_EX(SVM_EXIT_IDTR_READ),
5940 [x86_intercept_lgdt] = POST_EX(SVM_EXIT_GDTR_WRITE),
5941 [x86_intercept_lidt] = POST_EX(SVM_EXIT_IDTR_WRITE),
5942 [x86_intercept_vmrun] = POST_EX(SVM_EXIT_VMRUN),
5943 [x86_intercept_vmmcall] = POST_EX(SVM_EXIT_VMMCALL),
5944 [x86_intercept_vmload] = POST_EX(SVM_EXIT_VMLOAD),
5945 [x86_intercept_vmsave] = POST_EX(SVM_EXIT_VMSAVE),
5946 [x86_intercept_stgi] = POST_EX(SVM_EXIT_STGI),
5947 [x86_intercept_clgi] = POST_EX(SVM_EXIT_CLGI),
5948 [x86_intercept_skinit] = POST_EX(SVM_EXIT_SKINIT),
5949 [x86_intercept_invlpga] = POST_EX(SVM_EXIT_INVLPGA),
5950 [x86_intercept_rdtscp] = POST_EX(SVM_EXIT_RDTSCP),
5951 [x86_intercept_monitor] = POST_MEM(SVM_EXIT_MONITOR),
5952 [x86_intercept_mwait] = POST_EX(SVM_EXIT_MWAIT),
5953 [x86_intercept_invlpg] = POST_EX(SVM_EXIT_INVLPG),
5954 [x86_intercept_invd] = POST_EX(SVM_EXIT_INVD),
5955 [x86_intercept_wbinvd] = POST_EX(SVM_EXIT_WBINVD),
5956 [x86_intercept_wrmsr] = POST_EX(SVM_EXIT_MSR),
5957 [x86_intercept_rdtsc] = POST_EX(SVM_EXIT_RDTSC),
5958 [x86_intercept_rdmsr] = POST_EX(SVM_EXIT_MSR),
5959 [x86_intercept_rdpmc] = POST_EX(SVM_EXIT_RDPMC),
5960 [x86_intercept_cpuid] = PRE_EX(SVM_EXIT_CPUID),
5961 [x86_intercept_rsm] = PRE_EX(SVM_EXIT_RSM),
5962 [x86_intercept_pause] = PRE_EX(SVM_EXIT_PAUSE),
5963 [x86_intercept_pushf] = PRE_EX(SVM_EXIT_PUSHF),
5964 [x86_intercept_popf] = PRE_EX(SVM_EXIT_POPF),
5965 [x86_intercept_intn] = PRE_EX(SVM_EXIT_SWINT),
5966 [x86_intercept_iret] = PRE_EX(SVM_EXIT_IRET),
5967 [x86_intercept_icebp] = PRE_EX(SVM_EXIT_ICEBP),
5968 [x86_intercept_hlt] = POST_EX(SVM_EXIT_HLT),
5969 [x86_intercept_in] = POST_EX(SVM_EXIT_IOIO),
5970 [x86_intercept_ins] = POST_EX(SVM_EXIT_IOIO),
5971 [x86_intercept_out] = POST_EX(SVM_EXIT_IOIO),
5972 [x86_intercept_outs] = POST_EX(SVM_EXIT_IOIO),
5979 static int svm_check_intercept(struct kvm_vcpu *vcpu,
5980 struct x86_instruction_info *info,
5981 enum x86_intercept_stage stage)
5983 struct vcpu_svm *svm = to_svm(vcpu);
5984 int vmexit, ret = X86EMUL_CONTINUE;
5985 struct __x86_intercept icpt_info;
5986 struct vmcb *vmcb = svm->vmcb;
5988 if (info->intercept >= ARRAY_SIZE(x86_intercept_map))
5991 icpt_info = x86_intercept_map[info->intercept];
5993 if (stage != icpt_info.stage)
5996 switch (icpt_info.exit_code) {
5997 case SVM_EXIT_READ_CR0:
5998 if (info->intercept == x86_intercept_cr_read)
5999 icpt_info.exit_code += info->modrm_reg;
6001 case SVM_EXIT_WRITE_CR0: {
6002 unsigned long cr0, val;
6005 if (info->intercept == x86_intercept_cr_write)
6006 icpt_info.exit_code += info->modrm_reg;
6008 if (icpt_info.exit_code != SVM_EXIT_WRITE_CR0 ||
6009 info->intercept == x86_intercept_clts)
6012 intercept = svm->nested.intercept;
6014 if (!(intercept & (1ULL << INTERCEPT_SELECTIVE_CR0)))
6017 cr0 = vcpu->arch.cr0 & ~SVM_CR0_SELECTIVE_MASK;
6018 val = info->src_val & ~SVM_CR0_SELECTIVE_MASK;
6020 if (info->intercept == x86_intercept_lmsw) {
6023 /* lmsw can't clear PE - catch this here */
6024 if (cr0 & X86_CR0_PE)
6029 icpt_info.exit_code = SVM_EXIT_CR0_SEL_WRITE;
6033 case SVM_EXIT_READ_DR0:
6034 case SVM_EXIT_WRITE_DR0:
6035 icpt_info.exit_code += info->modrm_reg;
6038 if (info->intercept == x86_intercept_wrmsr)
6039 vmcb->control.exit_info_1 = 1;
6041 vmcb->control.exit_info_1 = 0;
6043 case SVM_EXIT_PAUSE:
6045 * We get this for NOP only, but pause
6046 * is rep not, check this here
6048 if (info->rep_prefix != REPE_PREFIX)
6051 case SVM_EXIT_IOIO: {
6055 if (info->intercept == x86_intercept_in ||
6056 info->intercept == x86_intercept_ins) {
6057 exit_info = ((info->src_val & 0xffff) << 16) |
6059 bytes = info->dst_bytes;
6061 exit_info = (info->dst_val & 0xffff) << 16;
6062 bytes = info->src_bytes;
6065 if (info->intercept == x86_intercept_outs ||
6066 info->intercept == x86_intercept_ins)
6067 exit_info |= SVM_IOIO_STR_MASK;
6069 if (info->rep_prefix)
6070 exit_info |= SVM_IOIO_REP_MASK;
6072 bytes = min(bytes, 4u);
6074 exit_info |= bytes << SVM_IOIO_SIZE_SHIFT;
6076 exit_info |= (u32)info->ad_bytes << (SVM_IOIO_ASIZE_SHIFT - 1);
6078 vmcb->control.exit_info_1 = exit_info;
6079 vmcb->control.exit_info_2 = info->next_rip;
6087 /* TODO: Advertise NRIPS to guest hypervisor unconditionally */
6088 if (static_cpu_has(X86_FEATURE_NRIPS))
6089 vmcb->control.next_rip = info->next_rip;
6090 vmcb->control.exit_code = icpt_info.exit_code;
6091 vmexit = nested_svm_exit_handled(svm);
6093 ret = (vmexit == NESTED_EXIT_DONE) ? X86EMUL_INTERCEPTED
6100 static void svm_handle_external_intr(struct kvm_vcpu *vcpu)
6104 * We must have an instruction with interrupts enabled, so
6105 * the timer interrupt isn't delayed by the interrupt shadow.
6108 local_irq_disable();
6111 static void svm_sched_in(struct kvm_vcpu *vcpu, int cpu)
6113 if (pause_filter_thresh)
6114 shrink_ple_window(vcpu);
6117 static inline void avic_post_state_restore(struct kvm_vcpu *vcpu)
6119 if (avic_handle_apic_id_update(vcpu) != 0)
6121 if (avic_handle_dfr_update(vcpu) != 0)
6123 avic_handle_ldr_update(vcpu);
6126 static void svm_setup_mce(struct kvm_vcpu *vcpu)
6128 /* [63:9] are reserved. */
6129 vcpu->arch.mcg_cap &= 0x1ff;
6132 static int svm_smi_allowed(struct kvm_vcpu *vcpu)
6134 struct vcpu_svm *svm = to_svm(vcpu);
6136 /* Per APM Vol.2 15.22.2 "Response to SMI" */
6140 if (is_guest_mode(&svm->vcpu) &&
6141 svm->nested.intercept & (1ULL << INTERCEPT_SMI)) {
6142 /* TODO: Might need to set exit_info_1 and exit_info_2 here */
6143 svm->vmcb->control.exit_code = SVM_EXIT_SMI;
6144 svm->nested.exit_required = true;
6151 static int svm_pre_enter_smm(struct kvm_vcpu *vcpu, char *smstate)
6153 struct vcpu_svm *svm = to_svm(vcpu);
6156 if (is_guest_mode(vcpu)) {
6157 /* FED8h - SVM Guest */
6158 put_smstate(u64, smstate, 0x7ed8, 1);
6159 /* FEE0h - SVM Guest VMCB Physical Address */
6160 put_smstate(u64, smstate, 0x7ee0, svm->nested.vmcb);
6162 svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX];
6163 svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP];
6164 svm->vmcb->save.rip = vcpu->arch.regs[VCPU_REGS_RIP];
6166 ret = nested_svm_vmexit(svm);
6173 static int svm_pre_leave_smm(struct kvm_vcpu *vcpu, u64 smbase)
6175 struct vcpu_svm *svm = to_svm(vcpu);
6176 struct vmcb *nested_vmcb;
6184 ret = kvm_vcpu_read_guest(vcpu, smbase + 0xfed8, &svm_state_save,
6185 sizeof(svm_state_save));
6189 if (svm_state_save.guest) {
6190 vcpu->arch.hflags &= ~HF_SMM_MASK;
6191 nested_vmcb = nested_svm_map(svm, svm_state_save.vmcb, &page);
6193 enter_svm_guest_mode(svm, svm_state_save.vmcb, nested_vmcb, page);
6196 vcpu->arch.hflags |= HF_SMM_MASK;
6201 static int enable_smi_window(struct kvm_vcpu *vcpu)
6203 struct vcpu_svm *svm = to_svm(vcpu);
6205 if (!gif_set(svm)) {
6206 if (vgif_enabled(svm))
6207 set_intercept(svm, INTERCEPT_STGI);
6208 /* STGI will cause a vm exit */
6214 static int sev_asid_new(void)
6219 * SEV-enabled guest must use asid from min_sev_asid to max_sev_asid.
6221 pos = find_next_zero_bit(sev_asid_bitmap, max_sev_asid, min_sev_asid - 1);
6222 if (pos >= max_sev_asid)
6225 set_bit(pos, sev_asid_bitmap);
6229 static int sev_guest_init(struct kvm *kvm, struct kvm_sev_cmd *argp)
6231 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
6235 asid = sev_asid_new();
6239 ret = sev_platform_init(&argp->error);
6245 INIT_LIST_HEAD(&sev->regions_list);
6250 __sev_asid_free(asid);
6254 static int sev_bind_asid(struct kvm *kvm, unsigned int handle, int *error)
6256 struct sev_data_activate *data;
6257 int asid = sev_get_asid(kvm);
6260 wbinvd_on_all_cpus();
6262 ret = sev_guest_df_flush(error);
6266 data = kzalloc(sizeof(*data), GFP_KERNEL);
6270 /* activate ASID on the given handle */
6271 data->handle = handle;
6273 ret = sev_guest_activate(data, error);
6279 static int __sev_issue_cmd(int fd, int id, void *data, int *error)
6288 ret = sev_issue_cmd_external_user(f.file, id, data, error);
6294 static int sev_issue_cmd(struct kvm *kvm, int id, void *data, int *error)
6296 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
6298 return __sev_issue_cmd(sev->fd, id, data, error);
6301 static int sev_launch_start(struct kvm *kvm, struct kvm_sev_cmd *argp)
6303 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
6304 struct sev_data_launch_start *start;
6305 struct kvm_sev_launch_start params;
6306 void *dh_blob, *session_blob;
6307 int *error = &argp->error;
6310 if (!sev_guest(kvm))
6313 if (copy_from_user(¶ms, (void __user *)(uintptr_t)argp->data, sizeof(params)))
6316 start = kzalloc(sizeof(*start), GFP_KERNEL);
6321 if (params.dh_uaddr) {
6322 dh_blob = psp_copy_user_blob(params.dh_uaddr, params.dh_len);
6323 if (IS_ERR(dh_blob)) {
6324 ret = PTR_ERR(dh_blob);
6328 start->dh_cert_address = __sme_set(__pa(dh_blob));
6329 start->dh_cert_len = params.dh_len;
6332 session_blob = NULL;
6333 if (params.session_uaddr) {
6334 session_blob = psp_copy_user_blob(params.session_uaddr, params.session_len);
6335 if (IS_ERR(session_blob)) {
6336 ret = PTR_ERR(session_blob);
6340 start->session_address = __sme_set(__pa(session_blob));
6341 start->session_len = params.session_len;
6344 start->handle = params.handle;
6345 start->policy = params.policy;
6347 /* create memory encryption context */
6348 ret = __sev_issue_cmd(argp->sev_fd, SEV_CMD_LAUNCH_START, start, error);
6350 goto e_free_session;
6352 /* Bind ASID to this guest */
6353 ret = sev_bind_asid(kvm, start->handle, error);
6355 goto e_free_session;
6357 /* return handle to userspace */
6358 params.handle = start->handle;
6359 if (copy_to_user((void __user *)(uintptr_t)argp->data, ¶ms, sizeof(params))) {
6360 sev_unbind_asid(kvm, start->handle);
6362 goto e_free_session;
6365 sev->handle = start->handle;
6366 sev->fd = argp->sev_fd;
6369 kfree(session_blob);
6377 static int get_num_contig_pages(int idx, struct page **inpages,
6378 unsigned long npages)
6380 unsigned long paddr, next_paddr;
6381 int i = idx + 1, pages = 1;
6383 /* find the number of contiguous pages starting from idx */
6384 paddr = __sme_page_pa(inpages[idx]);
6385 while (i < npages) {
6386 next_paddr = __sme_page_pa(inpages[i++]);
6387 if ((paddr + PAGE_SIZE) == next_paddr) {
6398 static int sev_launch_update_data(struct kvm *kvm, struct kvm_sev_cmd *argp)
6400 unsigned long vaddr, vaddr_end, next_vaddr, npages, size;
6401 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
6402 struct kvm_sev_launch_update_data params;
6403 struct sev_data_launch_update_data *data;
6404 struct page **inpages;
6407 if (!sev_guest(kvm))
6410 if (copy_from_user(¶ms, (void __user *)(uintptr_t)argp->data, sizeof(params)))
6413 data = kzalloc(sizeof(*data), GFP_KERNEL);
6417 vaddr = params.uaddr;
6419 vaddr_end = vaddr + size;
6421 /* Lock the user memory. */
6422 inpages = sev_pin_memory(kvm, vaddr, size, &npages, 1);
6429 * The LAUNCH_UPDATE command will perform in-place encryption of the
6430 * memory content (i.e it will write the same memory region with C=1).
6431 * It's possible that the cache may contain the data with C=0, i.e.,
6432 * unencrypted so invalidate it first.
6434 sev_clflush_pages(inpages, npages);
6436 for (i = 0; vaddr < vaddr_end; vaddr = next_vaddr, i += pages) {
6440 * If the user buffer is not page-aligned, calculate the offset
6443 offset = vaddr & (PAGE_SIZE - 1);
6445 /* Calculate the number of pages that can be encrypted in one go. */
6446 pages = get_num_contig_pages(i, inpages, npages);
6448 len = min_t(size_t, ((pages * PAGE_SIZE) - offset), size);
6450 data->handle = sev->handle;
6452 data->address = __sme_page_pa(inpages[i]) + offset;
6453 ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_UPDATE_DATA, data, &argp->error);
6458 next_vaddr = vaddr + len;
6462 /* content of memory is updated, mark pages dirty */
6463 for (i = 0; i < npages; i++) {
6464 set_page_dirty_lock(inpages[i]);
6465 mark_page_accessed(inpages[i]);
6467 /* unlock the user pages */
6468 sev_unpin_memory(kvm, inpages, npages);
6474 static int sev_launch_measure(struct kvm *kvm, struct kvm_sev_cmd *argp)
6476 void __user *measure = (void __user *)(uintptr_t)argp->data;
6477 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
6478 struct sev_data_launch_measure *data;
6479 struct kvm_sev_launch_measure params;
6480 void __user *p = NULL;
6484 if (!sev_guest(kvm))
6487 if (copy_from_user(¶ms, measure, sizeof(params)))
6490 data = kzalloc(sizeof(*data), GFP_KERNEL);
6494 /* User wants to query the blob length */
6498 p = (void __user *)(uintptr_t)params.uaddr;
6500 if (params.len > SEV_FW_BLOB_MAX_SIZE) {
6506 blob = kmalloc(params.len, GFP_KERNEL);
6510 data->address = __psp_pa(blob);
6511 data->len = params.len;
6515 data->handle = sev->handle;
6516 ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_MEASURE, data, &argp->error);
6519 * If we query the session length, FW responded with expected data.
6528 if (copy_to_user(p, blob, params.len))
6533 params.len = data->len;
6534 if (copy_to_user(measure, ¶ms, sizeof(params)))
6543 static int sev_launch_finish(struct kvm *kvm, struct kvm_sev_cmd *argp)
6545 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
6546 struct sev_data_launch_finish *data;
6549 if (!sev_guest(kvm))
6552 data = kzalloc(sizeof(*data), GFP_KERNEL);
6556 data->handle = sev->handle;
6557 ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_FINISH, data, &argp->error);
6563 static int sev_guest_status(struct kvm *kvm, struct kvm_sev_cmd *argp)
6565 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
6566 struct kvm_sev_guest_status params;
6567 struct sev_data_guest_status *data;
6570 if (!sev_guest(kvm))
6573 data = kzalloc(sizeof(*data), GFP_KERNEL);
6577 data->handle = sev->handle;
6578 ret = sev_issue_cmd(kvm, SEV_CMD_GUEST_STATUS, data, &argp->error);
6582 params.policy = data->policy;
6583 params.state = data->state;
6584 params.handle = data->handle;
6586 if (copy_to_user((void __user *)(uintptr_t)argp->data, ¶ms, sizeof(params)))
6593 static int __sev_issue_dbg_cmd(struct kvm *kvm, unsigned long src,
6594 unsigned long dst, int size,
6595 int *error, bool enc)
6597 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
6598 struct sev_data_dbg *data;
6601 data = kzalloc(sizeof(*data), GFP_KERNEL);
6605 data->handle = sev->handle;
6606 data->dst_addr = dst;
6607 data->src_addr = src;
6610 ret = sev_issue_cmd(kvm,
6611 enc ? SEV_CMD_DBG_ENCRYPT : SEV_CMD_DBG_DECRYPT,
6617 static int __sev_dbg_decrypt(struct kvm *kvm, unsigned long src_paddr,
6618 unsigned long dst_paddr, int sz, int *err)
6623 * Its safe to read more than we are asked, caller should ensure that
6624 * destination has enough space.
6626 src_paddr = round_down(src_paddr, 16);
6627 offset = src_paddr & 15;
6628 sz = round_up(sz + offset, 16);
6630 return __sev_issue_dbg_cmd(kvm, src_paddr, dst_paddr, sz, err, false);
6633 static int __sev_dbg_decrypt_user(struct kvm *kvm, unsigned long paddr,
6634 unsigned long __user dst_uaddr,
6635 unsigned long dst_paddr,
6638 struct page *tpage = NULL;
6641 /* if inputs are not 16-byte then use intermediate buffer */
6642 if (!IS_ALIGNED(dst_paddr, 16) ||
6643 !IS_ALIGNED(paddr, 16) ||
6644 !IS_ALIGNED(size, 16)) {
6645 tpage = (void *)alloc_page(GFP_KERNEL);
6649 dst_paddr = __sme_page_pa(tpage);
6652 ret = __sev_dbg_decrypt(kvm, paddr, dst_paddr, size, err);
6657 offset = paddr & 15;
6658 if (copy_to_user((void __user *)(uintptr_t)dst_uaddr,
6659 page_address(tpage) + offset, size))
6670 static int __sev_dbg_encrypt_user(struct kvm *kvm, unsigned long paddr,
6671 unsigned long __user vaddr,
6672 unsigned long dst_paddr,
6673 unsigned long __user dst_vaddr,
6674 int size, int *error)
6676 struct page *src_tpage = NULL;
6677 struct page *dst_tpage = NULL;
6678 int ret, len = size;
6680 /* If source buffer is not aligned then use an intermediate buffer */
6681 if (!IS_ALIGNED(vaddr, 16)) {
6682 src_tpage = alloc_page(GFP_KERNEL);
6686 if (copy_from_user(page_address(src_tpage),
6687 (void __user *)(uintptr_t)vaddr, size)) {
6688 __free_page(src_tpage);
6692 paddr = __sme_page_pa(src_tpage);
6696 * If destination buffer or length is not aligned then do read-modify-write:
6697 * - decrypt destination in an intermediate buffer
6698 * - copy the source buffer in an intermediate buffer
6699 * - use the intermediate buffer as source buffer
6701 if (!IS_ALIGNED(dst_vaddr, 16) || !IS_ALIGNED(size, 16)) {
6704 dst_tpage = alloc_page(GFP_KERNEL);
6710 ret = __sev_dbg_decrypt(kvm, dst_paddr,
6711 __sme_page_pa(dst_tpage), size, error);
6716 * If source is kernel buffer then use memcpy() otherwise
6719 dst_offset = dst_paddr & 15;
6722 memcpy(page_address(dst_tpage) + dst_offset,
6723 page_address(src_tpage), size);
6725 if (copy_from_user(page_address(dst_tpage) + dst_offset,
6726 (void __user *)(uintptr_t)vaddr, size)) {
6732 paddr = __sme_page_pa(dst_tpage);
6733 dst_paddr = round_down(dst_paddr, 16);
6734 len = round_up(size, 16);
6737 ret = __sev_issue_dbg_cmd(kvm, paddr, dst_paddr, len, error, true);
6741 __free_page(src_tpage);
6743 __free_page(dst_tpage);
6747 static int sev_dbg_crypt(struct kvm *kvm, struct kvm_sev_cmd *argp, bool dec)
6749 unsigned long vaddr, vaddr_end, next_vaddr;
6750 unsigned long dst_vaddr, dst_vaddr_end;
6751 struct page **src_p, **dst_p;
6752 struct kvm_sev_dbg debug;
6756 if (!sev_guest(kvm))
6759 if (copy_from_user(&debug, (void __user *)(uintptr_t)argp->data, sizeof(debug)))
6762 vaddr = debug.src_uaddr;
6764 vaddr_end = vaddr + size;
6765 dst_vaddr = debug.dst_uaddr;
6766 dst_vaddr_end = dst_vaddr + size;
6768 for (; vaddr < vaddr_end; vaddr = next_vaddr) {
6769 int len, s_off, d_off;
6771 /* lock userspace source and destination page */
6772 src_p = sev_pin_memory(kvm, vaddr & PAGE_MASK, PAGE_SIZE, &n, 0);
6776 dst_p = sev_pin_memory(kvm, dst_vaddr & PAGE_MASK, PAGE_SIZE, &n, 1);
6778 sev_unpin_memory(kvm, src_p, n);
6783 * The DBG_{DE,EN}CRYPT commands will perform {dec,en}cryption of the
6784 * memory content (i.e it will write the same memory region with C=1).
6785 * It's possible that the cache may contain the data with C=0, i.e.,
6786 * unencrypted so invalidate it first.
6788 sev_clflush_pages(src_p, 1);
6789 sev_clflush_pages(dst_p, 1);
6792 * Since user buffer may not be page aligned, calculate the
6793 * offset within the page.
6795 s_off = vaddr & ~PAGE_MASK;
6796 d_off = dst_vaddr & ~PAGE_MASK;
6797 len = min_t(size_t, (PAGE_SIZE - s_off), size);
6800 ret = __sev_dbg_decrypt_user(kvm,
6801 __sme_page_pa(src_p[0]) + s_off,
6803 __sme_page_pa(dst_p[0]) + d_off,
6806 ret = __sev_dbg_encrypt_user(kvm,
6807 __sme_page_pa(src_p[0]) + s_off,
6809 __sme_page_pa(dst_p[0]) + d_off,
6813 sev_unpin_memory(kvm, src_p, 1);
6814 sev_unpin_memory(kvm, dst_p, 1);
6819 next_vaddr = vaddr + len;
6820 dst_vaddr = dst_vaddr + len;
6827 static int sev_launch_secret(struct kvm *kvm, struct kvm_sev_cmd *argp)
6829 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
6830 struct sev_data_launch_secret *data;
6831 struct kvm_sev_launch_secret params;
6832 struct page **pages;
6837 if (!sev_guest(kvm))
6840 if (copy_from_user(¶ms, (void __user *)(uintptr_t)argp->data, sizeof(params)))
6843 pages = sev_pin_memory(kvm, params.guest_uaddr, params.guest_len, &n, 1);
6848 * The secret must be copied into contiguous memory region, lets verify
6849 * that userspace memory pages are contiguous before we issue command.
6851 if (get_num_contig_pages(0, pages, n) != n) {
6853 goto e_unpin_memory;
6857 data = kzalloc(sizeof(*data), GFP_KERNEL);
6859 goto e_unpin_memory;
6861 offset = params.guest_uaddr & (PAGE_SIZE - 1);
6862 data->guest_address = __sme_page_pa(pages[0]) + offset;
6863 data->guest_len = params.guest_len;
6865 blob = psp_copy_user_blob(params.trans_uaddr, params.trans_len);
6867 ret = PTR_ERR(blob);
6871 data->trans_address = __psp_pa(blob);
6872 data->trans_len = params.trans_len;
6874 hdr = psp_copy_user_blob(params.hdr_uaddr, params.hdr_len);
6879 data->hdr_address = __psp_pa(hdr);
6880 data->hdr_len = params.hdr_len;
6882 data->handle = sev->handle;
6883 ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_UPDATE_SECRET, data, &argp->error);
6892 sev_unpin_memory(kvm, pages, n);
6896 static int svm_mem_enc_op(struct kvm *kvm, void __user *argp)
6898 struct kvm_sev_cmd sev_cmd;
6901 if (!svm_sev_enabled())
6904 if (copy_from_user(&sev_cmd, argp, sizeof(struct kvm_sev_cmd)))
6907 mutex_lock(&kvm->lock);
6909 switch (sev_cmd.id) {
6911 r = sev_guest_init(kvm, &sev_cmd);
6913 case KVM_SEV_LAUNCH_START:
6914 r = sev_launch_start(kvm, &sev_cmd);
6916 case KVM_SEV_LAUNCH_UPDATE_DATA:
6917 r = sev_launch_update_data(kvm, &sev_cmd);
6919 case KVM_SEV_LAUNCH_MEASURE:
6920 r = sev_launch_measure(kvm, &sev_cmd);
6922 case KVM_SEV_LAUNCH_FINISH:
6923 r = sev_launch_finish(kvm, &sev_cmd);
6925 case KVM_SEV_GUEST_STATUS:
6926 r = sev_guest_status(kvm, &sev_cmd);
6928 case KVM_SEV_DBG_DECRYPT:
6929 r = sev_dbg_crypt(kvm, &sev_cmd, true);
6931 case KVM_SEV_DBG_ENCRYPT:
6932 r = sev_dbg_crypt(kvm, &sev_cmd, false);
6934 case KVM_SEV_LAUNCH_SECRET:
6935 r = sev_launch_secret(kvm, &sev_cmd);
6942 if (copy_to_user(argp, &sev_cmd, sizeof(struct kvm_sev_cmd)))
6946 mutex_unlock(&kvm->lock);
6950 static int svm_register_enc_region(struct kvm *kvm,
6951 struct kvm_enc_region *range)
6953 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
6954 struct enc_region *region;
6957 if (!sev_guest(kvm))
6960 if (range->addr > ULONG_MAX || range->size > ULONG_MAX)
6963 region = kzalloc(sizeof(*region), GFP_KERNEL);
6967 region->pages = sev_pin_memory(kvm, range->addr, range->size, ®ion->npages, 1);
6968 if (!region->pages) {
6974 * The guest may change the memory encryption attribute from C=0 -> C=1
6975 * or vice versa for this memory range. Lets make sure caches are
6976 * flushed to ensure that guest data gets written into memory with
6979 sev_clflush_pages(region->pages, region->npages);
6981 region->uaddr = range->addr;
6982 region->size = range->size;
6984 mutex_lock(&kvm->lock);
6985 list_add_tail(®ion->list, &sev->regions_list);
6986 mutex_unlock(&kvm->lock);
6995 static struct enc_region *
6996 find_enc_region(struct kvm *kvm, struct kvm_enc_region *range)
6998 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
6999 struct list_head *head = &sev->regions_list;
7000 struct enc_region *i;
7002 list_for_each_entry(i, head, list) {
7003 if (i->uaddr == range->addr &&
7004 i->size == range->size)
7012 static int svm_unregister_enc_region(struct kvm *kvm,
7013 struct kvm_enc_region *range)
7015 struct enc_region *region;
7018 mutex_lock(&kvm->lock);
7020 if (!sev_guest(kvm)) {
7025 region = find_enc_region(kvm, range);
7031 __unregister_enc_region_locked(kvm, region);
7033 mutex_unlock(&kvm->lock);
7037 mutex_unlock(&kvm->lock);
7041 static struct kvm_x86_ops svm_x86_ops __ro_after_init = {
7042 .cpu_has_kvm_support = has_svm,
7043 .disabled_by_bios = is_disabled,
7044 .hardware_setup = svm_hardware_setup,
7045 .hardware_unsetup = svm_hardware_unsetup,
7046 .check_processor_compatibility = svm_check_processor_compat,
7047 .hardware_enable = svm_hardware_enable,
7048 .hardware_disable = svm_hardware_disable,
7049 .cpu_has_accelerated_tpr = svm_cpu_has_accelerated_tpr,
7050 .has_emulated_msr = svm_has_emulated_msr,
7052 .vcpu_create = svm_create_vcpu,
7053 .vcpu_free = svm_free_vcpu,
7054 .vcpu_reset = svm_vcpu_reset,
7056 .vm_alloc = svm_vm_alloc,
7057 .vm_free = svm_vm_free,
7058 .vm_init = avic_vm_init,
7059 .vm_destroy = svm_vm_destroy,
7061 .prepare_guest_switch = svm_prepare_guest_switch,
7062 .vcpu_load = svm_vcpu_load,
7063 .vcpu_put = svm_vcpu_put,
7064 .vcpu_blocking = svm_vcpu_blocking,
7065 .vcpu_unblocking = svm_vcpu_unblocking,
7067 .update_bp_intercept = update_bp_intercept,
7068 .get_msr_feature = svm_get_msr_feature,
7069 .get_msr = svm_get_msr,
7070 .set_msr = svm_set_msr,
7071 .get_segment_base = svm_get_segment_base,
7072 .get_segment = svm_get_segment,
7073 .set_segment = svm_set_segment,
7074 .get_cpl = svm_get_cpl,
7075 .get_cs_db_l_bits = kvm_get_cs_db_l_bits,
7076 .decache_cr0_guest_bits = svm_decache_cr0_guest_bits,
7077 .decache_cr3 = svm_decache_cr3,
7078 .decache_cr4_guest_bits = svm_decache_cr4_guest_bits,
7079 .set_cr0 = svm_set_cr0,
7080 .set_cr3 = svm_set_cr3,
7081 .set_cr4 = svm_set_cr4,
7082 .set_efer = svm_set_efer,
7083 .get_idt = svm_get_idt,
7084 .set_idt = svm_set_idt,
7085 .get_gdt = svm_get_gdt,
7086 .set_gdt = svm_set_gdt,
7087 .get_dr6 = svm_get_dr6,
7088 .set_dr6 = svm_set_dr6,
7089 .set_dr7 = svm_set_dr7,
7090 .sync_dirty_debug_regs = svm_sync_dirty_debug_regs,
7091 .cache_reg = svm_cache_reg,
7092 .get_rflags = svm_get_rflags,
7093 .set_rflags = svm_set_rflags,
7095 .tlb_flush = svm_flush_tlb,
7096 .tlb_flush_gva = svm_flush_tlb_gva,
7098 .run = svm_vcpu_run,
7099 .handle_exit = handle_exit,
7100 .skip_emulated_instruction = skip_emulated_instruction,
7101 .set_interrupt_shadow = svm_set_interrupt_shadow,
7102 .get_interrupt_shadow = svm_get_interrupt_shadow,
7103 .patch_hypercall = svm_patch_hypercall,
7104 .set_irq = svm_set_irq,
7105 .set_nmi = svm_inject_nmi,
7106 .queue_exception = svm_queue_exception,
7107 .cancel_injection = svm_cancel_injection,
7108 .interrupt_allowed = svm_interrupt_allowed,
7109 .nmi_allowed = svm_nmi_allowed,
7110 .get_nmi_mask = svm_get_nmi_mask,
7111 .set_nmi_mask = svm_set_nmi_mask,
7112 .enable_nmi_window = enable_nmi_window,
7113 .enable_irq_window = enable_irq_window,
7114 .update_cr8_intercept = update_cr8_intercept,
7115 .set_virtual_apic_mode = svm_set_virtual_apic_mode,
7116 .get_enable_apicv = svm_get_enable_apicv,
7117 .refresh_apicv_exec_ctrl = svm_refresh_apicv_exec_ctrl,
7118 .load_eoi_exitmap = svm_load_eoi_exitmap,
7119 .hwapic_irr_update = svm_hwapic_irr_update,
7120 .hwapic_isr_update = svm_hwapic_isr_update,
7121 .sync_pir_to_irr = kvm_lapic_find_highest_irr,
7122 .apicv_post_state_restore = avic_post_state_restore,
7124 .set_tss_addr = svm_set_tss_addr,
7125 .set_identity_map_addr = svm_set_identity_map_addr,
7126 .get_tdp_level = get_npt_level,
7127 .get_mt_mask = svm_get_mt_mask,
7129 .get_exit_info = svm_get_exit_info,
7131 .get_lpage_level = svm_get_lpage_level,
7133 .cpuid_update = svm_cpuid_update,
7135 .rdtscp_supported = svm_rdtscp_supported,
7136 .invpcid_supported = svm_invpcid_supported,
7137 .mpx_supported = svm_mpx_supported,
7138 .xsaves_supported = svm_xsaves_supported,
7139 .umip_emulated = svm_umip_emulated,
7141 .set_supported_cpuid = svm_set_supported_cpuid,
7143 .has_wbinvd_exit = svm_has_wbinvd_exit,
7145 .read_l1_tsc_offset = svm_read_l1_tsc_offset,
7146 .write_tsc_offset = svm_write_tsc_offset,
7148 .set_tdp_cr3 = set_tdp_cr3,
7150 .check_intercept = svm_check_intercept,
7151 .handle_external_intr = svm_handle_external_intr,
7153 .sched_in = svm_sched_in,
7155 .pmu_ops = &amd_pmu_ops,
7156 .deliver_posted_interrupt = svm_deliver_avic_intr,
7157 .update_pi_irte = svm_update_pi_irte,
7158 .setup_mce = svm_setup_mce,
7160 .smi_allowed = svm_smi_allowed,
7161 .pre_enter_smm = svm_pre_enter_smm,
7162 .pre_leave_smm = svm_pre_leave_smm,
7163 .enable_smi_window = enable_smi_window,
7165 .mem_enc_op = svm_mem_enc_op,
7166 .mem_enc_reg_region = svm_register_enc_region,
7167 .mem_enc_unreg_region = svm_unregister_enc_region,
7170 static int __init svm_init(void)
7172 return kvm_init(&svm_x86_ops, sizeof(struct vcpu_svm),
7173 __alignof__(struct vcpu_svm), THIS_MODULE);
7176 static void __exit svm_exit(void)
7181 module_init(svm_init)
7182 module_exit(svm_exit)