]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * Kernel-based Virtual Machine driver for Linux | |
3 | * | |
4 | * AMD SVM support | |
5 | * | |
6 | * Copyright (C) 2006 Qumranet, Inc. | |
7 | * Copyright 2010 Red Hat, Inc. and/or its affiliates. | |
8 | * | |
9 | * Authors: | |
10 | * Yaniv Kamay <yaniv@qumranet.com> | |
11 | * Avi Kivity <avi@qumranet.com> | |
12 | * | |
13 | * This work is licensed under the terms of the GNU GPL, version 2. See | |
14 | * the COPYING file in the top-level directory. | |
15 | * | |
16 | */ | |
17 | ||
18 | #define pr_fmt(fmt) "SVM: " fmt | |
19 | ||
20 | #include <linux/kvm_host.h> | |
21 | ||
22 | #include "irq.h" | |
23 | #include "mmu.h" | |
24 | #include "kvm_cache_regs.h" | |
25 | #include "x86.h" | |
26 | #include "cpuid.h" | |
27 | #include "pmu.h" | |
28 | ||
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 | ||
40 | #include <asm/apic.h> | |
41 | #include <asm/perf_event.h> | |
42 | #include <asm/tlbflush.h> | |
43 | #include <asm/desc.h> | |
44 | #include <asm/debugreg.h> | |
45 | #include <asm/kvm_para.h> | |
46 | #include <asm/irq_remapping.h> | |
47 | ||
48 | #include <asm/virtext.h> | |
49 | #include "trace.h" | |
50 | ||
51 | #define __ex(x) __kvm_handle_fault_on_reboot(x) | |
52 | ||
53 | MODULE_AUTHOR("Qumranet"); | |
54 | MODULE_LICENSE("GPL"); | |
55 | ||
56 | static const struct x86_cpu_id svm_cpu_id[] = { | |
57 | X86_FEATURE_MATCH(X86_FEATURE_SVM), | |
58 | {} | |
59 | }; | |
60 | MODULE_DEVICE_TABLE(x86cpu, svm_cpu_id); | |
61 | ||
62 | #define IOPM_ALLOC_ORDER 2 | |
63 | #define MSRPM_ALLOC_ORDER 1 | |
64 | ||
65 | #define SEG_TYPE_LDT 2 | |
66 | #define SEG_TYPE_BUSY_TSS16 3 | |
67 | ||
68 | #define SVM_FEATURE_NPT (1 << 0) | |
69 | #define SVM_FEATURE_LBRV (1 << 1) | |
70 | #define SVM_FEATURE_SVML (1 << 2) | |
71 | #define SVM_FEATURE_NRIP (1 << 3) | |
72 | #define SVM_FEATURE_TSC_RATE (1 << 4) | |
73 | #define SVM_FEATURE_VMCB_CLEAN (1 << 5) | |
74 | #define SVM_FEATURE_FLUSH_ASID (1 << 6) | |
75 | #define SVM_FEATURE_DECODE_ASSIST (1 << 7) | |
76 | #define SVM_FEATURE_PAUSE_FILTER (1 << 10) | |
77 | ||
78 | #define SVM_AVIC_DOORBELL 0xc001011b | |
79 | ||
80 | #define NESTED_EXIT_HOST 0 /* Exit handled on host level */ | |
81 | #define NESTED_EXIT_DONE 1 /* Exit caused nested vmexit */ | |
82 | #define NESTED_EXIT_CONTINUE 2 /* Further checks needed */ | |
83 | ||
84 | #define DEBUGCTL_RESERVED_BITS (~(0x3fULL)) | |
85 | ||
86 | #define TSC_RATIO_RSVD 0xffffff0000000000ULL | |
87 | #define TSC_RATIO_MIN 0x0000000000000001ULL | |
88 | #define TSC_RATIO_MAX 0x000000ffffffffffULL | |
89 | ||
90 | #define AVIC_HPA_MASK ~((0xFFFULL << 52) | 0xFFF) | |
91 | ||
92 | /* | |
93 | * 0xff is broadcast, so the max index allowed for physical APIC ID | |
94 | * table is 0xfe. APIC IDs above 0xff are reserved. | |
95 | */ | |
96 | #define AVIC_MAX_PHYSICAL_ID_COUNT 255 | |
97 | ||
98 | #define AVIC_UNACCEL_ACCESS_WRITE_MASK 1 | |
99 | #define AVIC_UNACCEL_ACCESS_OFFSET_MASK 0xFF0 | |
100 | #define AVIC_UNACCEL_ACCESS_VECTOR_MASK 0xFFFFFFFF | |
101 | ||
102 | /* AVIC GATAG is encoded using VM and VCPU IDs */ | |
103 | #define AVIC_VCPU_ID_BITS 8 | |
104 | #define AVIC_VCPU_ID_MASK ((1 << AVIC_VCPU_ID_BITS) - 1) | |
105 | ||
106 | #define AVIC_VM_ID_BITS 24 | |
107 | #define AVIC_VM_ID_NR (1 << AVIC_VM_ID_BITS) | |
108 | #define AVIC_VM_ID_MASK ((1 << AVIC_VM_ID_BITS) - 1) | |
109 | ||
110 | #define AVIC_GATAG(x, y) (((x & AVIC_VM_ID_MASK) << AVIC_VCPU_ID_BITS) | \ | |
111 | (y & AVIC_VCPU_ID_MASK)) | |
112 | #define AVIC_GATAG_TO_VMID(x) ((x >> AVIC_VCPU_ID_BITS) & AVIC_VM_ID_MASK) | |
113 | #define AVIC_GATAG_TO_VCPUID(x) (x & AVIC_VCPU_ID_MASK) | |
114 | ||
115 | static bool erratum_383_found __read_mostly; | |
116 | ||
117 | static const u32 host_save_user_msrs[] = { | |
118 | #ifdef CONFIG_X86_64 | |
119 | MSR_STAR, MSR_LSTAR, MSR_CSTAR, MSR_SYSCALL_MASK, MSR_KERNEL_GS_BASE, | |
120 | MSR_FS_BASE, | |
121 | #endif | |
122 | MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP, | |
123 | MSR_TSC_AUX, | |
124 | }; | |
125 | ||
126 | #define NR_HOST_SAVE_USER_MSRS ARRAY_SIZE(host_save_user_msrs) | |
127 | ||
128 | struct kvm_vcpu; | |
129 | ||
130 | struct nested_state { | |
131 | struct vmcb *hsave; | |
132 | u64 hsave_msr; | |
133 | u64 vm_cr_msr; | |
134 | u64 vmcb; | |
135 | ||
136 | /* These are the merged vectors */ | |
137 | u32 *msrpm; | |
138 | ||
139 | /* gpa pointers to the real vectors */ | |
140 | u64 vmcb_msrpm; | |
141 | u64 vmcb_iopm; | |
142 | ||
143 | /* A VMEXIT is required but not yet emulated */ | |
144 | bool exit_required; | |
145 | ||
146 | /* cache for intercepts of the guest */ | |
147 | u32 intercept_cr; | |
148 | u32 intercept_dr; | |
149 | u32 intercept_exceptions; | |
150 | u64 intercept; | |
151 | ||
152 | /* Nested Paging related state */ | |
153 | u64 nested_cr3; | |
154 | }; | |
155 | ||
156 | #define MSRPM_OFFSETS 16 | |
157 | static u32 msrpm_offsets[MSRPM_OFFSETS] __read_mostly; | |
158 | ||
159 | /* | |
160 | * Set osvw_len to higher value when updated Revision Guides | |
161 | * are published and we know what the new status bits are | |
162 | */ | |
163 | static uint64_t osvw_len = 4, osvw_status; | |
164 | ||
165 | struct vcpu_svm { | |
166 | struct kvm_vcpu vcpu; | |
167 | struct vmcb *vmcb; | |
168 | unsigned long vmcb_pa; | |
169 | struct svm_cpu_data *svm_data; | |
170 | uint64_t asid_generation; | |
171 | uint64_t sysenter_esp; | |
172 | uint64_t sysenter_eip; | |
173 | uint64_t tsc_aux; | |
174 | ||
175 | u64 next_rip; | |
176 | ||
177 | u64 host_user_msrs[NR_HOST_SAVE_USER_MSRS]; | |
178 | struct { | |
179 | u16 fs; | |
180 | u16 gs; | |
181 | u16 ldt; | |
182 | u64 gs_base; | |
183 | } host; | |
184 | ||
185 | u32 *msrpm; | |
186 | ||
187 | ulong nmi_iret_rip; | |
188 | ||
189 | struct nested_state nested; | |
190 | ||
191 | bool nmi_singlestep; | |
192 | ||
193 | unsigned int3_injected; | |
194 | unsigned long int3_rip; | |
195 | u32 apf_reason; | |
196 | ||
197 | /* cached guest cpuid flags for faster access */ | |
198 | bool nrips_enabled : 1; | |
199 | ||
200 | u32 ldr_reg; | |
201 | struct page *avic_backing_page; | |
202 | u64 *avic_physical_id_cache; | |
203 | bool avic_is_running; | |
204 | ||
205 | /* | |
206 | * Per-vcpu list of struct amd_svm_iommu_ir: | |
207 | * This is used mainly to store interrupt remapping information used | |
208 | * when update the vcpu affinity. This avoids the need to scan for | |
209 | * IRTE and try to match ga_tag in the IOMMU driver. | |
210 | */ | |
211 | struct list_head ir_list; | |
212 | spinlock_t ir_list_lock; | |
213 | }; | |
214 | ||
215 | /* | |
216 | * This is a wrapper of struct amd_iommu_ir_data. | |
217 | */ | |
218 | struct amd_svm_iommu_ir { | |
219 | struct list_head node; /* Used by SVM for per-vcpu ir_list */ | |
220 | void *data; /* Storing pointer to struct amd_ir_data */ | |
221 | }; | |
222 | ||
223 | #define AVIC_LOGICAL_ID_ENTRY_GUEST_PHYSICAL_ID_MASK (0xFF) | |
224 | #define AVIC_LOGICAL_ID_ENTRY_VALID_MASK (1 << 31) | |
225 | ||
226 | #define AVIC_PHYSICAL_ID_ENTRY_HOST_PHYSICAL_ID_MASK (0xFFULL) | |
227 | #define AVIC_PHYSICAL_ID_ENTRY_BACKING_PAGE_MASK (0xFFFFFFFFFFULL << 12) | |
228 | #define AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK (1ULL << 62) | |
229 | #define AVIC_PHYSICAL_ID_ENTRY_VALID_MASK (1ULL << 63) | |
230 | ||
231 | static DEFINE_PER_CPU(u64, current_tsc_ratio); | |
232 | #define TSC_RATIO_DEFAULT 0x0100000000ULL | |
233 | ||
234 | #define MSR_INVALID 0xffffffffU | |
235 | ||
236 | static const struct svm_direct_access_msrs { | |
237 | u32 index; /* Index of the MSR */ | |
238 | bool always; /* True if intercept is always on */ | |
239 | } direct_access_msrs[] = { | |
240 | { .index = MSR_STAR, .always = true }, | |
241 | { .index = MSR_IA32_SYSENTER_CS, .always = true }, | |
242 | #ifdef CONFIG_X86_64 | |
243 | { .index = MSR_GS_BASE, .always = true }, | |
244 | { .index = MSR_FS_BASE, .always = true }, | |
245 | { .index = MSR_KERNEL_GS_BASE, .always = true }, | |
246 | { .index = MSR_LSTAR, .always = true }, | |
247 | { .index = MSR_CSTAR, .always = true }, | |
248 | { .index = MSR_SYSCALL_MASK, .always = true }, | |
249 | #endif | |
250 | { .index = MSR_IA32_LASTBRANCHFROMIP, .always = false }, | |
251 | { .index = MSR_IA32_LASTBRANCHTOIP, .always = false }, | |
252 | { .index = MSR_IA32_LASTINTFROMIP, .always = false }, | |
253 | { .index = MSR_IA32_LASTINTTOIP, .always = false }, | |
254 | { .index = MSR_INVALID, .always = false }, | |
255 | }; | |
256 | ||
257 | /* enable NPT for AMD64 and X86 with PAE */ | |
258 | #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE) | |
259 | static bool npt_enabled = true; | |
260 | #else | |
261 | static bool npt_enabled; | |
262 | #endif | |
263 | ||
264 | /* allow nested paging (virtualized MMU) for all guests */ | |
265 | static int npt = true; | |
266 | module_param(npt, int, S_IRUGO); | |
267 | ||
268 | /* allow nested virtualization in KVM/SVM */ | |
269 | static int nested = true; | |
270 | module_param(nested, int, S_IRUGO); | |
271 | ||
272 | /* enable / disable AVIC */ | |
273 | static int avic; | |
274 | #ifdef CONFIG_X86_LOCAL_APIC | |
275 | module_param(avic, int, S_IRUGO); | |
276 | #endif | |
277 | ||
278 | /* AVIC VM ID bit masks and lock */ | |
279 | static DECLARE_BITMAP(avic_vm_id_bitmap, AVIC_VM_ID_NR); | |
280 | static DEFINE_SPINLOCK(avic_vm_id_lock); | |
281 | ||
282 | static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0); | |
283 | static void svm_flush_tlb(struct kvm_vcpu *vcpu); | |
284 | static void svm_complete_interrupts(struct vcpu_svm *svm); | |
285 | ||
286 | static int nested_svm_exit_handled(struct vcpu_svm *svm); | |
287 | static int nested_svm_intercept(struct vcpu_svm *svm); | |
288 | static int nested_svm_vmexit(struct vcpu_svm *svm); | |
289 | static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr, | |
290 | bool has_error_code, u32 error_code); | |
291 | ||
292 | enum { | |
293 | VMCB_INTERCEPTS, /* Intercept vectors, TSC offset, | |
294 | pause filter count */ | |
295 | VMCB_PERM_MAP, /* IOPM Base and MSRPM Base */ | |
296 | VMCB_ASID, /* ASID */ | |
297 | VMCB_INTR, /* int_ctl, int_vector */ | |
298 | VMCB_NPT, /* npt_en, nCR3, gPAT */ | |
299 | VMCB_CR, /* CR0, CR3, CR4, EFER */ | |
300 | VMCB_DR, /* DR6, DR7 */ | |
301 | VMCB_DT, /* GDT, IDT */ | |
302 | VMCB_SEG, /* CS, DS, SS, ES, CPL */ | |
303 | VMCB_CR2, /* CR2 only */ | |
304 | VMCB_LBR, /* DBGCTL, BR_FROM, BR_TO, LAST_EX_FROM, LAST_EX_TO */ | |
305 | VMCB_AVIC, /* AVIC APIC_BAR, AVIC APIC_BACKING_PAGE, | |
306 | * AVIC PHYSICAL_TABLE pointer, | |
307 | * AVIC LOGICAL_TABLE pointer | |
308 | */ | |
309 | VMCB_DIRTY_MAX, | |
310 | }; | |
311 | ||
312 | /* TPR and CR2 are always written before VMRUN */ | |
313 | #define VMCB_ALWAYS_DIRTY_MASK ((1U << VMCB_INTR) | (1U << VMCB_CR2)) | |
314 | ||
315 | #define VMCB_AVIC_APIC_BAR_MASK 0xFFFFFFFFFF000ULL | |
316 | ||
317 | static inline void mark_all_dirty(struct vmcb *vmcb) | |
318 | { | |
319 | vmcb->control.clean = 0; | |
320 | } | |
321 | ||
322 | static inline void mark_all_clean(struct vmcb *vmcb) | |
323 | { | |
324 | vmcb->control.clean = ((1 << VMCB_DIRTY_MAX) - 1) | |
325 | & ~VMCB_ALWAYS_DIRTY_MASK; | |
326 | } | |
327 | ||
328 | static inline void mark_dirty(struct vmcb *vmcb, int bit) | |
329 | { | |
330 | vmcb->control.clean &= ~(1 << bit); | |
331 | } | |
332 | ||
333 | static inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu) | |
334 | { | |
335 | return container_of(vcpu, struct vcpu_svm, vcpu); | |
336 | } | |
337 | ||
338 | static inline void avic_update_vapic_bar(struct vcpu_svm *svm, u64 data) | |
339 | { | |
340 | svm->vmcb->control.avic_vapic_bar = data & VMCB_AVIC_APIC_BAR_MASK; | |
341 | mark_dirty(svm->vmcb, VMCB_AVIC); | |
342 | } | |
343 | ||
344 | static inline bool avic_vcpu_is_running(struct kvm_vcpu *vcpu) | |
345 | { | |
346 | struct vcpu_svm *svm = to_svm(vcpu); | |
347 | u64 *entry = svm->avic_physical_id_cache; | |
348 | ||
349 | if (!entry) | |
350 | return false; | |
351 | ||
352 | return (READ_ONCE(*entry) & AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK); | |
353 | } | |
354 | ||
355 | static void recalc_intercepts(struct vcpu_svm *svm) | |
356 | { | |
357 | struct vmcb_control_area *c, *h; | |
358 | struct nested_state *g; | |
359 | ||
360 | mark_dirty(svm->vmcb, VMCB_INTERCEPTS); | |
361 | ||
362 | if (!is_guest_mode(&svm->vcpu)) | |
363 | return; | |
364 | ||
365 | c = &svm->vmcb->control; | |
366 | h = &svm->nested.hsave->control; | |
367 | g = &svm->nested; | |
368 | ||
369 | c->intercept_cr = h->intercept_cr | g->intercept_cr; | |
370 | c->intercept_dr = h->intercept_dr | g->intercept_dr; | |
371 | c->intercept_exceptions = h->intercept_exceptions | g->intercept_exceptions; | |
372 | c->intercept = h->intercept | g->intercept; | |
373 | } | |
374 | ||
375 | static inline struct vmcb *get_host_vmcb(struct vcpu_svm *svm) | |
376 | { | |
377 | if (is_guest_mode(&svm->vcpu)) | |
378 | return svm->nested.hsave; | |
379 | else | |
380 | return svm->vmcb; | |
381 | } | |
382 | ||
383 | static inline void set_cr_intercept(struct vcpu_svm *svm, int bit) | |
384 | { | |
385 | struct vmcb *vmcb = get_host_vmcb(svm); | |
386 | ||
387 | vmcb->control.intercept_cr |= (1U << bit); | |
388 | ||
389 | recalc_intercepts(svm); | |
390 | } | |
391 | ||
392 | static inline void clr_cr_intercept(struct vcpu_svm *svm, int bit) | |
393 | { | |
394 | struct vmcb *vmcb = get_host_vmcb(svm); | |
395 | ||
396 | vmcb->control.intercept_cr &= ~(1U << bit); | |
397 | ||
398 | recalc_intercepts(svm); | |
399 | } | |
400 | ||
401 | static inline bool is_cr_intercept(struct vcpu_svm *svm, int bit) | |
402 | { | |
403 | struct vmcb *vmcb = get_host_vmcb(svm); | |
404 | ||
405 | return vmcb->control.intercept_cr & (1U << bit); | |
406 | } | |
407 | ||
408 | static inline void set_dr_intercepts(struct vcpu_svm *svm) | |
409 | { | |
410 | struct vmcb *vmcb = get_host_vmcb(svm); | |
411 | ||
412 | vmcb->control.intercept_dr = (1 << INTERCEPT_DR0_READ) | |
413 | | (1 << INTERCEPT_DR1_READ) | |
414 | | (1 << INTERCEPT_DR2_READ) | |
415 | | (1 << INTERCEPT_DR3_READ) | |
416 | | (1 << INTERCEPT_DR4_READ) | |
417 | | (1 << INTERCEPT_DR5_READ) | |
418 | | (1 << INTERCEPT_DR6_READ) | |
419 | | (1 << INTERCEPT_DR7_READ) | |
420 | | (1 << INTERCEPT_DR0_WRITE) | |
421 | | (1 << INTERCEPT_DR1_WRITE) | |
422 | | (1 << INTERCEPT_DR2_WRITE) | |
423 | | (1 << INTERCEPT_DR3_WRITE) | |
424 | | (1 << INTERCEPT_DR4_WRITE) | |
425 | | (1 << INTERCEPT_DR5_WRITE) | |
426 | | (1 << INTERCEPT_DR6_WRITE) | |
427 | | (1 << INTERCEPT_DR7_WRITE); | |
428 | ||
429 | recalc_intercepts(svm); | |
430 | } | |
431 | ||
432 | static inline void clr_dr_intercepts(struct vcpu_svm *svm) | |
433 | { | |
434 | struct vmcb *vmcb = get_host_vmcb(svm); | |
435 | ||
436 | vmcb->control.intercept_dr = 0; | |
437 | ||
438 | recalc_intercepts(svm); | |
439 | } | |
440 | ||
441 | static inline void set_exception_intercept(struct vcpu_svm *svm, int bit) | |
442 | { | |
443 | struct vmcb *vmcb = get_host_vmcb(svm); | |
444 | ||
445 | vmcb->control.intercept_exceptions |= (1U << bit); | |
446 | ||
447 | recalc_intercepts(svm); | |
448 | } | |
449 | ||
450 | static inline void clr_exception_intercept(struct vcpu_svm *svm, int bit) | |
451 | { | |
452 | struct vmcb *vmcb = get_host_vmcb(svm); | |
453 | ||
454 | vmcb->control.intercept_exceptions &= ~(1U << bit); | |
455 | ||
456 | recalc_intercepts(svm); | |
457 | } | |
458 | ||
459 | static inline void set_intercept(struct vcpu_svm *svm, int bit) | |
460 | { | |
461 | struct vmcb *vmcb = get_host_vmcb(svm); | |
462 | ||
463 | vmcb->control.intercept |= (1ULL << bit); | |
464 | ||
465 | recalc_intercepts(svm); | |
466 | } | |
467 | ||
468 | static inline void clr_intercept(struct vcpu_svm *svm, int bit) | |
469 | { | |
470 | struct vmcb *vmcb = get_host_vmcb(svm); | |
471 | ||
472 | vmcb->control.intercept &= ~(1ULL << bit); | |
473 | ||
474 | recalc_intercepts(svm); | |
475 | } | |
476 | ||
477 | static inline void enable_gif(struct vcpu_svm *svm) | |
478 | { | |
479 | svm->vcpu.arch.hflags |= HF_GIF_MASK; | |
480 | } | |
481 | ||
482 | static inline void disable_gif(struct vcpu_svm *svm) | |
483 | { | |
484 | svm->vcpu.arch.hflags &= ~HF_GIF_MASK; | |
485 | } | |
486 | ||
487 | static inline bool gif_set(struct vcpu_svm *svm) | |
488 | { | |
489 | return !!(svm->vcpu.arch.hflags & HF_GIF_MASK); | |
490 | } | |
491 | ||
492 | static unsigned long iopm_base; | |
493 | ||
494 | struct kvm_ldttss_desc { | |
495 | u16 limit0; | |
496 | u16 base0; | |
497 | unsigned base1:8, type:5, dpl:2, p:1; | |
498 | unsigned limit1:4, zero0:3, g:1, base2:8; | |
499 | u32 base3; | |
500 | u32 zero1; | |
501 | } __attribute__((packed)); | |
502 | ||
503 | struct svm_cpu_data { | |
504 | int cpu; | |
505 | ||
506 | u64 asid_generation; | |
507 | u32 max_asid; | |
508 | u32 next_asid; | |
509 | struct kvm_ldttss_desc *tss_desc; | |
510 | ||
511 | struct page *save_area; | |
512 | }; | |
513 | ||
514 | static DEFINE_PER_CPU(struct svm_cpu_data *, svm_data); | |
515 | ||
516 | struct svm_init_data { | |
517 | int cpu; | |
518 | int r; | |
519 | }; | |
520 | ||
521 | static const u32 msrpm_ranges[] = {0, 0xc0000000, 0xc0010000}; | |
522 | ||
523 | #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges) | |
524 | #define MSRS_RANGE_SIZE 2048 | |
525 | #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2) | |
526 | ||
527 | static u32 svm_msrpm_offset(u32 msr) | |
528 | { | |
529 | u32 offset; | |
530 | int i; | |
531 | ||
532 | for (i = 0; i < NUM_MSR_MAPS; i++) { | |
533 | if (msr < msrpm_ranges[i] || | |
534 | msr >= msrpm_ranges[i] + MSRS_IN_RANGE) | |
535 | continue; | |
536 | ||
537 | offset = (msr - msrpm_ranges[i]) / 4; /* 4 msrs per u8 */ | |
538 | offset += (i * MSRS_RANGE_SIZE); /* add range offset */ | |
539 | ||
540 | /* Now we have the u8 offset - but need the u32 offset */ | |
541 | return offset / 4; | |
542 | } | |
543 | ||
544 | /* MSR not in any range */ | |
545 | return MSR_INVALID; | |
546 | } | |
547 | ||
548 | #define MAX_INST_SIZE 15 | |
549 | ||
550 | static inline void clgi(void) | |
551 | { | |
552 | asm volatile (__ex(SVM_CLGI)); | |
553 | } | |
554 | ||
555 | static inline void stgi(void) | |
556 | { | |
557 | asm volatile (__ex(SVM_STGI)); | |
558 | } | |
559 | ||
560 | static inline void invlpga(unsigned long addr, u32 asid) | |
561 | { | |
562 | asm volatile (__ex(SVM_INVLPGA) : : "a"(addr), "c"(asid)); | |
563 | } | |
564 | ||
565 | static int get_npt_level(void) | |
566 | { | |
567 | #ifdef CONFIG_X86_64 | |
568 | return PT64_ROOT_LEVEL; | |
569 | #else | |
570 | return PT32E_ROOT_LEVEL; | |
571 | #endif | |
572 | } | |
573 | ||
574 | static void svm_set_efer(struct kvm_vcpu *vcpu, u64 efer) | |
575 | { | |
576 | vcpu->arch.efer = efer; | |
577 | if (!npt_enabled && !(efer & EFER_LMA)) | |
578 | efer &= ~EFER_LME; | |
579 | ||
580 | to_svm(vcpu)->vmcb->save.efer = efer | EFER_SVME; | |
581 | mark_dirty(to_svm(vcpu)->vmcb, VMCB_CR); | |
582 | } | |
583 | ||
584 | static int is_external_interrupt(u32 info) | |
585 | { | |
586 | info &= SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID; | |
587 | return info == (SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR); | |
588 | } | |
589 | ||
590 | static u32 svm_get_interrupt_shadow(struct kvm_vcpu *vcpu) | |
591 | { | |
592 | struct vcpu_svm *svm = to_svm(vcpu); | |
593 | u32 ret = 0; | |
594 | ||
595 | if (svm->vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK) | |
596 | ret = KVM_X86_SHADOW_INT_STI | KVM_X86_SHADOW_INT_MOV_SS; | |
597 | return ret; | |
598 | } | |
599 | ||
600 | static void svm_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask) | |
601 | { | |
602 | struct vcpu_svm *svm = to_svm(vcpu); | |
603 | ||
604 | if (mask == 0) | |
605 | svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK; | |
606 | else | |
607 | svm->vmcb->control.int_state |= SVM_INTERRUPT_SHADOW_MASK; | |
608 | ||
609 | } | |
610 | ||
611 | static void skip_emulated_instruction(struct kvm_vcpu *vcpu) | |
612 | { | |
613 | struct vcpu_svm *svm = to_svm(vcpu); | |
614 | ||
615 | if (svm->vmcb->control.next_rip != 0) { | |
616 | WARN_ON_ONCE(!static_cpu_has(X86_FEATURE_NRIPS)); | |
617 | svm->next_rip = svm->vmcb->control.next_rip; | |
618 | } | |
619 | ||
620 | if (!svm->next_rip) { | |
621 | if (emulate_instruction(vcpu, EMULTYPE_SKIP) != | |
622 | EMULATE_DONE) | |
623 | printk(KERN_DEBUG "%s: NOP\n", __func__); | |
624 | return; | |
625 | } | |
626 | if (svm->next_rip - kvm_rip_read(vcpu) > MAX_INST_SIZE) | |
627 | printk(KERN_ERR "%s: ip 0x%lx next 0x%llx\n", | |
628 | __func__, kvm_rip_read(vcpu), svm->next_rip); | |
629 | ||
630 | kvm_rip_write(vcpu, svm->next_rip); | |
631 | svm_set_interrupt_shadow(vcpu, 0); | |
632 | } | |
633 | ||
634 | static void svm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr, | |
635 | bool has_error_code, u32 error_code, | |
636 | bool reinject) | |
637 | { | |
638 | struct vcpu_svm *svm = to_svm(vcpu); | |
639 | ||
640 | /* | |
641 | * If we are within a nested VM we'd better #VMEXIT and let the guest | |
642 | * handle the exception | |
643 | */ | |
644 | if (!reinject && | |
645 | nested_svm_check_exception(svm, nr, has_error_code, error_code)) | |
646 | return; | |
647 | ||
648 | if (nr == BP_VECTOR && !static_cpu_has(X86_FEATURE_NRIPS)) { | |
649 | unsigned long rip, old_rip = kvm_rip_read(&svm->vcpu); | |
650 | ||
651 | /* | |
652 | * For guest debugging where we have to reinject #BP if some | |
653 | * INT3 is guest-owned: | |
654 | * Emulate nRIP by moving RIP forward. Will fail if injection | |
655 | * raises a fault that is not intercepted. Still better than | |
656 | * failing in all cases. | |
657 | */ | |
658 | skip_emulated_instruction(&svm->vcpu); | |
659 | rip = kvm_rip_read(&svm->vcpu); | |
660 | svm->int3_rip = rip + svm->vmcb->save.cs.base; | |
661 | svm->int3_injected = rip - old_rip; | |
662 | } | |
663 | ||
664 | svm->vmcb->control.event_inj = nr | |
665 | | SVM_EVTINJ_VALID | |
666 | | (has_error_code ? SVM_EVTINJ_VALID_ERR : 0) | |
667 | | SVM_EVTINJ_TYPE_EXEPT; | |
668 | svm->vmcb->control.event_inj_err = error_code; | |
669 | } | |
670 | ||
671 | static void svm_init_erratum_383(void) | |
672 | { | |
673 | u32 low, high; | |
674 | int err; | |
675 | u64 val; | |
676 | ||
677 | if (!static_cpu_has_bug(X86_BUG_AMD_TLB_MMATCH)) | |
678 | return; | |
679 | ||
680 | /* Use _safe variants to not break nested virtualization */ | |
681 | val = native_read_msr_safe(MSR_AMD64_DC_CFG, &err); | |
682 | if (err) | |
683 | return; | |
684 | ||
685 | val |= (1ULL << 47); | |
686 | ||
687 | low = lower_32_bits(val); | |
688 | high = upper_32_bits(val); | |
689 | ||
690 | native_write_msr_safe(MSR_AMD64_DC_CFG, low, high); | |
691 | ||
692 | erratum_383_found = true; | |
693 | } | |
694 | ||
695 | static void svm_init_osvw(struct kvm_vcpu *vcpu) | |
696 | { | |
697 | /* | |
698 | * Guests should see errata 400 and 415 as fixed (assuming that | |
699 | * HLT and IO instructions are intercepted). | |
700 | */ | |
701 | vcpu->arch.osvw.length = (osvw_len >= 3) ? (osvw_len) : 3; | |
702 | vcpu->arch.osvw.status = osvw_status & ~(6ULL); | |
703 | ||
704 | /* | |
705 | * By increasing VCPU's osvw.length to 3 we are telling the guest that | |
706 | * all osvw.status bits inside that length, including bit 0 (which is | |
707 | * reserved for erratum 298), are valid. However, if host processor's | |
708 | * osvw_len is 0 then osvw_status[0] carries no information. We need to | |
709 | * be conservative here and therefore we tell the guest that erratum 298 | |
710 | * is present (because we really don't know). | |
711 | */ | |
712 | if (osvw_len == 0 && boot_cpu_data.x86 == 0x10) | |
713 | vcpu->arch.osvw.status |= 1; | |
714 | } | |
715 | ||
716 | static int has_svm(void) | |
717 | { | |
718 | const char *msg; | |
719 | ||
720 | if (!cpu_has_svm(&msg)) { | |
721 | printk(KERN_INFO "has_svm: %s\n", msg); | |
722 | return 0; | |
723 | } | |
724 | ||
725 | return 1; | |
726 | } | |
727 | ||
728 | static void svm_hardware_disable(void) | |
729 | { | |
730 | /* Make sure we clean up behind us */ | |
731 | if (static_cpu_has(X86_FEATURE_TSCRATEMSR)) | |
732 | wrmsrl(MSR_AMD64_TSC_RATIO, TSC_RATIO_DEFAULT); | |
733 | ||
734 | cpu_svm_disable(); | |
735 | ||
736 | amd_pmu_disable_virt(); | |
737 | } | |
738 | ||
739 | static int svm_hardware_enable(void) | |
740 | { | |
741 | ||
742 | struct svm_cpu_data *sd; | |
743 | uint64_t efer; | |
744 | struct desc_ptr gdt_descr; | |
745 | struct desc_struct *gdt; | |
746 | int me = raw_smp_processor_id(); | |
747 | ||
748 | rdmsrl(MSR_EFER, efer); | |
749 | if (efer & EFER_SVME) | |
750 | return -EBUSY; | |
751 | ||
752 | if (!has_svm()) { | |
753 | pr_err("%s: err EOPNOTSUPP on %d\n", __func__, me); | |
754 | return -EINVAL; | |
755 | } | |
756 | sd = per_cpu(svm_data, me); | |
757 | if (!sd) { | |
758 | pr_err("%s: svm_data is NULL on %d\n", __func__, me); | |
759 | return -EINVAL; | |
760 | } | |
761 | ||
762 | sd->asid_generation = 1; | |
763 | sd->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1; | |
764 | sd->next_asid = sd->max_asid + 1; | |
765 | ||
766 | native_store_gdt(&gdt_descr); | |
767 | gdt = (struct desc_struct *)gdt_descr.address; | |
768 | sd->tss_desc = (struct kvm_ldttss_desc *)(gdt + GDT_ENTRY_TSS); | |
769 | ||
770 | wrmsrl(MSR_EFER, efer | EFER_SVME); | |
771 | ||
772 | wrmsrl(MSR_VM_HSAVE_PA, page_to_pfn(sd->save_area) << PAGE_SHIFT); | |
773 | ||
774 | if (static_cpu_has(X86_FEATURE_TSCRATEMSR)) { | |
775 | wrmsrl(MSR_AMD64_TSC_RATIO, TSC_RATIO_DEFAULT); | |
776 | __this_cpu_write(current_tsc_ratio, TSC_RATIO_DEFAULT); | |
777 | } | |
778 | ||
779 | ||
780 | /* | |
781 | * Get OSVW bits. | |
782 | * | |
783 | * Note that it is possible to have a system with mixed processor | |
784 | * revisions and therefore different OSVW bits. If bits are not the same | |
785 | * on different processors then choose the worst case (i.e. if erratum | |
786 | * is present on one processor and not on another then assume that the | |
787 | * erratum is present everywhere). | |
788 | */ | |
789 | if (cpu_has(&boot_cpu_data, X86_FEATURE_OSVW)) { | |
790 | uint64_t len, status = 0; | |
791 | int err; | |
792 | ||
793 | len = native_read_msr_safe(MSR_AMD64_OSVW_ID_LENGTH, &err); | |
794 | if (!err) | |
795 | status = native_read_msr_safe(MSR_AMD64_OSVW_STATUS, | |
796 | &err); | |
797 | ||
798 | if (err) | |
799 | osvw_status = osvw_len = 0; | |
800 | else { | |
801 | if (len < osvw_len) | |
802 | osvw_len = len; | |
803 | osvw_status |= status; | |
804 | osvw_status &= (1ULL << osvw_len) - 1; | |
805 | } | |
806 | } else | |
807 | osvw_status = osvw_len = 0; | |
808 | ||
809 | svm_init_erratum_383(); | |
810 | ||
811 | amd_pmu_enable_virt(); | |
812 | ||
813 | return 0; | |
814 | } | |
815 | ||
816 | static void svm_cpu_uninit(int cpu) | |
817 | { | |
818 | struct svm_cpu_data *sd = per_cpu(svm_data, raw_smp_processor_id()); | |
819 | ||
820 | if (!sd) | |
821 | return; | |
822 | ||
823 | per_cpu(svm_data, raw_smp_processor_id()) = NULL; | |
824 | __free_page(sd->save_area); | |
825 | kfree(sd); | |
826 | } | |
827 | ||
828 | static int svm_cpu_init(int cpu) | |
829 | { | |
830 | struct svm_cpu_data *sd; | |
831 | int r; | |
832 | ||
833 | sd = kzalloc(sizeof(struct svm_cpu_data), GFP_KERNEL); | |
834 | if (!sd) | |
835 | return -ENOMEM; | |
836 | sd->cpu = cpu; | |
837 | sd->save_area = alloc_page(GFP_KERNEL); | |
838 | r = -ENOMEM; | |
839 | if (!sd->save_area) | |
840 | goto err_1; | |
841 | ||
842 | per_cpu(svm_data, cpu) = sd; | |
843 | ||
844 | return 0; | |
845 | ||
846 | err_1: | |
847 | kfree(sd); | |
848 | return r; | |
849 | ||
850 | } | |
851 | ||
852 | static bool valid_msr_intercept(u32 index) | |
853 | { | |
854 | int i; | |
855 | ||
856 | for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) | |
857 | if (direct_access_msrs[i].index == index) | |
858 | return true; | |
859 | ||
860 | return false; | |
861 | } | |
862 | ||
863 | static void set_msr_interception(u32 *msrpm, unsigned msr, | |
864 | int read, int write) | |
865 | { | |
866 | u8 bit_read, bit_write; | |
867 | unsigned long tmp; | |
868 | u32 offset; | |
869 | ||
870 | /* | |
871 | * If this warning triggers extend the direct_access_msrs list at the | |
872 | * beginning of the file | |
873 | */ | |
874 | WARN_ON(!valid_msr_intercept(msr)); | |
875 | ||
876 | offset = svm_msrpm_offset(msr); | |
877 | bit_read = 2 * (msr & 0x0f); | |
878 | bit_write = 2 * (msr & 0x0f) + 1; | |
879 | tmp = msrpm[offset]; | |
880 | ||
881 | BUG_ON(offset == MSR_INVALID); | |
882 | ||
883 | read ? clear_bit(bit_read, &tmp) : set_bit(bit_read, &tmp); | |
884 | write ? clear_bit(bit_write, &tmp) : set_bit(bit_write, &tmp); | |
885 | ||
886 | msrpm[offset] = tmp; | |
887 | } | |
888 | ||
889 | static void svm_vcpu_init_msrpm(u32 *msrpm) | |
890 | { | |
891 | int i; | |
892 | ||
893 | memset(msrpm, 0xff, PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER)); | |
894 | ||
895 | for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) { | |
896 | if (!direct_access_msrs[i].always) | |
897 | continue; | |
898 | ||
899 | set_msr_interception(msrpm, direct_access_msrs[i].index, 1, 1); | |
900 | } | |
901 | } | |
902 | ||
903 | static void add_msr_offset(u32 offset) | |
904 | { | |
905 | int i; | |
906 | ||
907 | for (i = 0; i < MSRPM_OFFSETS; ++i) { | |
908 | ||
909 | /* Offset already in list? */ | |
910 | if (msrpm_offsets[i] == offset) | |
911 | return; | |
912 | ||
913 | /* Slot used by another offset? */ | |
914 | if (msrpm_offsets[i] != MSR_INVALID) | |
915 | continue; | |
916 | ||
917 | /* Add offset to list */ | |
918 | msrpm_offsets[i] = offset; | |
919 | ||
920 | return; | |
921 | } | |
922 | ||
923 | /* | |
924 | * If this BUG triggers the msrpm_offsets table has an overflow. Just | |
925 | * increase MSRPM_OFFSETS in this case. | |
926 | */ | |
927 | BUG(); | |
928 | } | |
929 | ||
930 | static void init_msrpm_offsets(void) | |
931 | { | |
932 | int i; | |
933 | ||
934 | memset(msrpm_offsets, 0xff, sizeof(msrpm_offsets)); | |
935 | ||
936 | for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) { | |
937 | u32 offset; | |
938 | ||
939 | offset = svm_msrpm_offset(direct_access_msrs[i].index); | |
940 | BUG_ON(offset == MSR_INVALID); | |
941 | ||
942 | add_msr_offset(offset); | |
943 | } | |
944 | } | |
945 | ||
946 | static void svm_enable_lbrv(struct vcpu_svm *svm) | |
947 | { | |
948 | u32 *msrpm = svm->msrpm; | |
949 | ||
950 | svm->vmcb->control.lbr_ctl = 1; | |
951 | set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 1, 1); | |
952 | set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 1, 1); | |
953 | set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 1, 1); | |
954 | set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 1, 1); | |
955 | } | |
956 | ||
957 | static void svm_disable_lbrv(struct vcpu_svm *svm) | |
958 | { | |
959 | u32 *msrpm = svm->msrpm; | |
960 | ||
961 | svm->vmcb->control.lbr_ctl = 0; | |
962 | set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 0, 0); | |
963 | set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 0, 0); | |
964 | set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 0, 0); | |
965 | set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 0, 0); | |
966 | } | |
967 | ||
968 | /* Note: | |
969 | * This hash table is used to map VM_ID to a struct kvm_arch, | |
970 | * when handling AMD IOMMU GALOG notification to schedule in | |
971 | * a particular vCPU. | |
972 | */ | |
973 | #define SVM_VM_DATA_HASH_BITS 8 | |
974 | DECLARE_HASHTABLE(svm_vm_data_hash, SVM_VM_DATA_HASH_BITS); | |
975 | static spinlock_t svm_vm_data_hash_lock; | |
976 | ||
977 | /* Note: | |
978 | * This function is called from IOMMU driver to notify | |
979 | * SVM to schedule in a particular vCPU of a particular VM. | |
980 | */ | |
981 | static int avic_ga_log_notifier(u32 ga_tag) | |
982 | { | |
983 | unsigned long flags; | |
984 | struct kvm_arch *ka = NULL; | |
985 | struct kvm_vcpu *vcpu = NULL; | |
986 | u32 vm_id = AVIC_GATAG_TO_VMID(ga_tag); | |
987 | u32 vcpu_id = AVIC_GATAG_TO_VCPUID(ga_tag); | |
988 | ||
989 | pr_debug("SVM: %s: vm_id=%#x, vcpu_id=%#x\n", __func__, vm_id, vcpu_id); | |
990 | ||
991 | spin_lock_irqsave(&svm_vm_data_hash_lock, flags); | |
992 | hash_for_each_possible(svm_vm_data_hash, ka, hnode, vm_id) { | |
993 | struct kvm *kvm = container_of(ka, struct kvm, arch); | |
994 | struct kvm_arch *vm_data = &kvm->arch; | |
995 | ||
996 | if (vm_data->avic_vm_id != vm_id) | |
997 | continue; | |
998 | vcpu = kvm_get_vcpu_by_id(kvm, vcpu_id); | |
999 | break; | |
1000 | } | |
1001 | spin_unlock_irqrestore(&svm_vm_data_hash_lock, flags); | |
1002 | ||
1003 | if (!vcpu) | |
1004 | return 0; | |
1005 | ||
1006 | /* Note: | |
1007 | * At this point, the IOMMU should have already set the pending | |
1008 | * bit in the vAPIC backing page. So, we just need to schedule | |
1009 | * in the vcpu. | |
1010 | */ | |
1011 | if (vcpu->mode == OUTSIDE_GUEST_MODE) | |
1012 | kvm_vcpu_wake_up(vcpu); | |
1013 | ||
1014 | return 0; | |
1015 | } | |
1016 | ||
1017 | static __init int svm_hardware_setup(void) | |
1018 | { | |
1019 | int cpu; | |
1020 | struct page *iopm_pages; | |
1021 | void *iopm_va; | |
1022 | int r; | |
1023 | ||
1024 | iopm_pages = alloc_pages(GFP_KERNEL, IOPM_ALLOC_ORDER); | |
1025 | ||
1026 | if (!iopm_pages) | |
1027 | return -ENOMEM; | |
1028 | ||
1029 | iopm_va = page_address(iopm_pages); | |
1030 | memset(iopm_va, 0xff, PAGE_SIZE * (1 << IOPM_ALLOC_ORDER)); | |
1031 | iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT; | |
1032 | ||
1033 | init_msrpm_offsets(); | |
1034 | ||
1035 | if (boot_cpu_has(X86_FEATURE_NX)) | |
1036 | kvm_enable_efer_bits(EFER_NX); | |
1037 | ||
1038 | if (boot_cpu_has(X86_FEATURE_FXSR_OPT)) | |
1039 | kvm_enable_efer_bits(EFER_FFXSR); | |
1040 | ||
1041 | if (boot_cpu_has(X86_FEATURE_TSCRATEMSR)) { | |
1042 | kvm_has_tsc_control = true; | |
1043 | kvm_max_tsc_scaling_ratio = TSC_RATIO_MAX; | |
1044 | kvm_tsc_scaling_ratio_frac_bits = 32; | |
1045 | } | |
1046 | ||
1047 | if (nested) { | |
1048 | printk(KERN_INFO "kvm: Nested Virtualization enabled\n"); | |
1049 | kvm_enable_efer_bits(EFER_SVME | EFER_LMSLE); | |
1050 | } | |
1051 | ||
1052 | for_each_possible_cpu(cpu) { | |
1053 | r = svm_cpu_init(cpu); | |
1054 | if (r) | |
1055 | goto err; | |
1056 | } | |
1057 | ||
1058 | if (!boot_cpu_has(X86_FEATURE_NPT)) | |
1059 | npt_enabled = false; | |
1060 | ||
1061 | if (npt_enabled && !npt) { | |
1062 | printk(KERN_INFO "kvm: Nested Paging disabled\n"); | |
1063 | npt_enabled = false; | |
1064 | } | |
1065 | ||
1066 | if (npt_enabled) { | |
1067 | printk(KERN_INFO "kvm: Nested Paging enabled\n"); | |
1068 | kvm_enable_tdp(); | |
1069 | } else | |
1070 | kvm_disable_tdp(); | |
1071 | ||
1072 | if (avic) { | |
1073 | if (!npt_enabled || | |
1074 | !boot_cpu_has(X86_FEATURE_AVIC) || | |
1075 | !IS_ENABLED(CONFIG_X86_LOCAL_APIC)) { | |
1076 | avic = false; | |
1077 | } else { | |
1078 | pr_info("AVIC enabled\n"); | |
1079 | ||
1080 | hash_init(svm_vm_data_hash); | |
1081 | spin_lock_init(&svm_vm_data_hash_lock); | |
1082 | amd_iommu_register_ga_log_notifier(&avic_ga_log_notifier); | |
1083 | } | |
1084 | } | |
1085 | ||
1086 | return 0; | |
1087 | ||
1088 | err: | |
1089 | __free_pages(iopm_pages, IOPM_ALLOC_ORDER); | |
1090 | iopm_base = 0; | |
1091 | return r; | |
1092 | } | |
1093 | ||
1094 | static __exit void svm_hardware_unsetup(void) | |
1095 | { | |
1096 | int cpu; | |
1097 | ||
1098 | for_each_possible_cpu(cpu) | |
1099 | svm_cpu_uninit(cpu); | |
1100 | ||
1101 | __free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT), IOPM_ALLOC_ORDER); | |
1102 | iopm_base = 0; | |
1103 | } | |
1104 | ||
1105 | static void init_seg(struct vmcb_seg *seg) | |
1106 | { | |
1107 | seg->selector = 0; | |
1108 | seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK | | |
1109 | SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */ | |
1110 | seg->limit = 0xffff; | |
1111 | seg->base = 0; | |
1112 | } | |
1113 | ||
1114 | static void init_sys_seg(struct vmcb_seg *seg, uint32_t type) | |
1115 | { | |
1116 | seg->selector = 0; | |
1117 | seg->attrib = SVM_SELECTOR_P_MASK | type; | |
1118 | seg->limit = 0xffff; | |
1119 | seg->base = 0; | |
1120 | } | |
1121 | ||
1122 | static void svm_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset) | |
1123 | { | |
1124 | struct vcpu_svm *svm = to_svm(vcpu); | |
1125 | u64 g_tsc_offset = 0; | |
1126 | ||
1127 | if (is_guest_mode(vcpu)) { | |
1128 | g_tsc_offset = svm->vmcb->control.tsc_offset - | |
1129 | svm->nested.hsave->control.tsc_offset; | |
1130 | svm->nested.hsave->control.tsc_offset = offset; | |
1131 | } else | |
1132 | trace_kvm_write_tsc_offset(vcpu->vcpu_id, | |
1133 | svm->vmcb->control.tsc_offset, | |
1134 | offset); | |
1135 | ||
1136 | svm->vmcb->control.tsc_offset = offset + g_tsc_offset; | |
1137 | ||
1138 | mark_dirty(svm->vmcb, VMCB_INTERCEPTS); | |
1139 | } | |
1140 | ||
1141 | static void avic_init_vmcb(struct vcpu_svm *svm) | |
1142 | { | |
1143 | struct vmcb *vmcb = svm->vmcb; | |
1144 | struct kvm_arch *vm_data = &svm->vcpu.kvm->arch; | |
1145 | phys_addr_t bpa = page_to_phys(svm->avic_backing_page); | |
1146 | phys_addr_t lpa = page_to_phys(vm_data->avic_logical_id_table_page); | |
1147 | phys_addr_t ppa = page_to_phys(vm_data->avic_physical_id_table_page); | |
1148 | ||
1149 | vmcb->control.avic_backing_page = bpa & AVIC_HPA_MASK; | |
1150 | vmcb->control.avic_logical_id = lpa & AVIC_HPA_MASK; | |
1151 | vmcb->control.avic_physical_id = ppa & AVIC_HPA_MASK; | |
1152 | vmcb->control.avic_physical_id |= AVIC_MAX_PHYSICAL_ID_COUNT; | |
1153 | vmcb->control.int_ctl |= AVIC_ENABLE_MASK; | |
1154 | svm->vcpu.arch.apicv_active = true; | |
1155 | } | |
1156 | ||
1157 | static void init_vmcb(struct vcpu_svm *svm) | |
1158 | { | |
1159 | struct vmcb_control_area *control = &svm->vmcb->control; | |
1160 | struct vmcb_save_area *save = &svm->vmcb->save; | |
1161 | ||
1162 | svm->vcpu.fpu_active = 1; | |
1163 | svm->vcpu.arch.hflags = 0; | |
1164 | ||
1165 | set_cr_intercept(svm, INTERCEPT_CR0_READ); | |
1166 | set_cr_intercept(svm, INTERCEPT_CR3_READ); | |
1167 | set_cr_intercept(svm, INTERCEPT_CR4_READ); | |
1168 | set_cr_intercept(svm, INTERCEPT_CR0_WRITE); | |
1169 | set_cr_intercept(svm, INTERCEPT_CR3_WRITE); | |
1170 | set_cr_intercept(svm, INTERCEPT_CR4_WRITE); | |
1171 | if (!kvm_vcpu_apicv_active(&svm->vcpu)) | |
1172 | set_cr_intercept(svm, INTERCEPT_CR8_WRITE); | |
1173 | ||
1174 | set_dr_intercepts(svm); | |
1175 | ||
1176 | set_exception_intercept(svm, PF_VECTOR); | |
1177 | set_exception_intercept(svm, UD_VECTOR); | |
1178 | set_exception_intercept(svm, MC_VECTOR); | |
1179 | set_exception_intercept(svm, AC_VECTOR); | |
1180 | set_exception_intercept(svm, DB_VECTOR); | |
1181 | ||
1182 | set_intercept(svm, INTERCEPT_INTR); | |
1183 | set_intercept(svm, INTERCEPT_NMI); | |
1184 | set_intercept(svm, INTERCEPT_SMI); | |
1185 | set_intercept(svm, INTERCEPT_SELECTIVE_CR0); | |
1186 | set_intercept(svm, INTERCEPT_RDPMC); | |
1187 | set_intercept(svm, INTERCEPT_CPUID); | |
1188 | set_intercept(svm, INTERCEPT_INVD); | |
1189 | set_intercept(svm, INTERCEPT_HLT); | |
1190 | set_intercept(svm, INTERCEPT_INVLPG); | |
1191 | set_intercept(svm, INTERCEPT_INVLPGA); | |
1192 | set_intercept(svm, INTERCEPT_IOIO_PROT); | |
1193 | set_intercept(svm, INTERCEPT_MSR_PROT); | |
1194 | set_intercept(svm, INTERCEPT_TASK_SWITCH); | |
1195 | set_intercept(svm, INTERCEPT_SHUTDOWN); | |
1196 | set_intercept(svm, INTERCEPT_VMRUN); | |
1197 | set_intercept(svm, INTERCEPT_VMMCALL); | |
1198 | set_intercept(svm, INTERCEPT_VMLOAD); | |
1199 | set_intercept(svm, INTERCEPT_VMSAVE); | |
1200 | set_intercept(svm, INTERCEPT_STGI); | |
1201 | set_intercept(svm, INTERCEPT_CLGI); | |
1202 | set_intercept(svm, INTERCEPT_SKINIT); | |
1203 | set_intercept(svm, INTERCEPT_WBINVD); | |
1204 | set_intercept(svm, INTERCEPT_MONITOR); | |
1205 | set_intercept(svm, INTERCEPT_MWAIT); | |
1206 | set_intercept(svm, INTERCEPT_XSETBV); | |
1207 | ||
1208 | control->iopm_base_pa = iopm_base; | |
1209 | control->msrpm_base_pa = __pa(svm->msrpm); | |
1210 | control->int_ctl = V_INTR_MASKING_MASK; | |
1211 | ||
1212 | init_seg(&save->es); | |
1213 | init_seg(&save->ss); | |
1214 | init_seg(&save->ds); | |
1215 | init_seg(&save->fs); | |
1216 | init_seg(&save->gs); | |
1217 | ||
1218 | save->cs.selector = 0xf000; | |
1219 | save->cs.base = 0xffff0000; | |
1220 | /* Executable/Readable Code Segment */ | |
1221 | save->cs.attrib = SVM_SELECTOR_READ_MASK | SVM_SELECTOR_P_MASK | | |
1222 | SVM_SELECTOR_S_MASK | SVM_SELECTOR_CODE_MASK; | |
1223 | save->cs.limit = 0xffff; | |
1224 | ||
1225 | save->gdtr.limit = 0xffff; | |
1226 | save->idtr.limit = 0xffff; | |
1227 | ||
1228 | init_sys_seg(&save->ldtr, SEG_TYPE_LDT); | |
1229 | init_sys_seg(&save->tr, SEG_TYPE_BUSY_TSS16); | |
1230 | ||
1231 | svm_set_efer(&svm->vcpu, 0); | |
1232 | save->dr6 = 0xffff0ff0; | |
1233 | kvm_set_rflags(&svm->vcpu, 2); | |
1234 | save->rip = 0x0000fff0; | |
1235 | svm->vcpu.arch.regs[VCPU_REGS_RIP] = save->rip; | |
1236 | ||
1237 | /* | |
1238 | * svm_set_cr0() sets PG and WP and clears NW and CD on save->cr0. | |
1239 | * It also updates the guest-visible cr0 value. | |
1240 | */ | |
1241 | svm_set_cr0(&svm->vcpu, X86_CR0_NW | X86_CR0_CD | X86_CR0_ET); | |
1242 | kvm_mmu_reset_context(&svm->vcpu); | |
1243 | ||
1244 | save->cr4 = X86_CR4_PAE; | |
1245 | /* rdx = ?? */ | |
1246 | ||
1247 | if (npt_enabled) { | |
1248 | /* Setup VMCB for Nested Paging */ | |
1249 | control->nested_ctl = 1; | |
1250 | clr_intercept(svm, INTERCEPT_INVLPG); | |
1251 | clr_exception_intercept(svm, PF_VECTOR); | |
1252 | clr_cr_intercept(svm, INTERCEPT_CR3_READ); | |
1253 | clr_cr_intercept(svm, INTERCEPT_CR3_WRITE); | |
1254 | save->g_pat = svm->vcpu.arch.pat; | |
1255 | save->cr3 = 0; | |
1256 | save->cr4 = 0; | |
1257 | } | |
1258 | svm->asid_generation = 0; | |
1259 | ||
1260 | svm->nested.vmcb = 0; | |
1261 | svm->vcpu.arch.hflags = 0; | |
1262 | ||
1263 | if (boot_cpu_has(X86_FEATURE_PAUSEFILTER)) { | |
1264 | control->pause_filter_count = 3000; | |
1265 | set_intercept(svm, INTERCEPT_PAUSE); | |
1266 | } | |
1267 | ||
1268 | if (avic) | |
1269 | avic_init_vmcb(svm); | |
1270 | ||
1271 | mark_all_dirty(svm->vmcb); | |
1272 | ||
1273 | enable_gif(svm); | |
1274 | ||
1275 | } | |
1276 | ||
1277 | static u64 *avic_get_physical_id_entry(struct kvm_vcpu *vcpu, int index) | |
1278 | { | |
1279 | u64 *avic_physical_id_table; | |
1280 | struct kvm_arch *vm_data = &vcpu->kvm->arch; | |
1281 | ||
1282 | if (index >= AVIC_MAX_PHYSICAL_ID_COUNT) | |
1283 | return NULL; | |
1284 | ||
1285 | avic_physical_id_table = page_address(vm_data->avic_physical_id_table_page); | |
1286 | ||
1287 | return &avic_physical_id_table[index]; | |
1288 | } | |
1289 | ||
1290 | /** | |
1291 | * Note: | |
1292 | * AVIC hardware walks the nested page table to check permissions, | |
1293 | * but does not use the SPA address specified in the leaf page | |
1294 | * table entry since it uses address in the AVIC_BACKING_PAGE pointer | |
1295 | * field of the VMCB. Therefore, we set up the | |
1296 | * APIC_ACCESS_PAGE_PRIVATE_MEMSLOT (4KB) here. | |
1297 | */ | |
1298 | static int avic_init_access_page(struct kvm_vcpu *vcpu) | |
1299 | { | |
1300 | struct kvm *kvm = vcpu->kvm; | |
1301 | int ret; | |
1302 | ||
1303 | if (kvm->arch.apic_access_page_done) | |
1304 | return 0; | |
1305 | ||
1306 | ret = x86_set_memory_region(kvm, | |
1307 | APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, | |
1308 | APIC_DEFAULT_PHYS_BASE, | |
1309 | PAGE_SIZE); | |
1310 | if (ret) | |
1311 | return ret; | |
1312 | ||
1313 | kvm->arch.apic_access_page_done = true; | |
1314 | return 0; | |
1315 | } | |
1316 | ||
1317 | static int avic_init_backing_page(struct kvm_vcpu *vcpu) | |
1318 | { | |
1319 | int ret; | |
1320 | u64 *entry, new_entry; | |
1321 | int id = vcpu->vcpu_id; | |
1322 | struct vcpu_svm *svm = to_svm(vcpu); | |
1323 | ||
1324 | ret = avic_init_access_page(vcpu); | |
1325 | if (ret) | |
1326 | return ret; | |
1327 | ||
1328 | if (id >= AVIC_MAX_PHYSICAL_ID_COUNT) | |
1329 | return -EINVAL; | |
1330 | ||
1331 | if (!svm->vcpu.arch.apic->regs) | |
1332 | return -EINVAL; | |
1333 | ||
1334 | svm->avic_backing_page = virt_to_page(svm->vcpu.arch.apic->regs); | |
1335 | ||
1336 | /* Setting AVIC backing page address in the phy APIC ID table */ | |
1337 | entry = avic_get_physical_id_entry(vcpu, id); | |
1338 | if (!entry) | |
1339 | return -EINVAL; | |
1340 | ||
1341 | new_entry = READ_ONCE(*entry); | |
1342 | new_entry = (page_to_phys(svm->avic_backing_page) & | |
1343 | AVIC_PHYSICAL_ID_ENTRY_BACKING_PAGE_MASK) | | |
1344 | AVIC_PHYSICAL_ID_ENTRY_VALID_MASK; | |
1345 | WRITE_ONCE(*entry, new_entry); | |
1346 | ||
1347 | svm->avic_physical_id_cache = entry; | |
1348 | ||
1349 | return 0; | |
1350 | } | |
1351 | ||
1352 | static inline int avic_get_next_vm_id(void) | |
1353 | { | |
1354 | int id; | |
1355 | ||
1356 | spin_lock(&avic_vm_id_lock); | |
1357 | ||
1358 | /* AVIC VM ID is one-based. */ | |
1359 | id = find_next_zero_bit(avic_vm_id_bitmap, AVIC_VM_ID_NR, 1); | |
1360 | if (id <= AVIC_VM_ID_MASK) | |
1361 | __set_bit(id, avic_vm_id_bitmap); | |
1362 | else | |
1363 | id = -EAGAIN; | |
1364 | ||
1365 | spin_unlock(&avic_vm_id_lock); | |
1366 | return id; | |
1367 | } | |
1368 | ||
1369 | static inline int avic_free_vm_id(int id) | |
1370 | { | |
1371 | if (id <= 0 || id > AVIC_VM_ID_MASK) | |
1372 | return -EINVAL; | |
1373 | ||
1374 | spin_lock(&avic_vm_id_lock); | |
1375 | __clear_bit(id, avic_vm_id_bitmap); | |
1376 | spin_unlock(&avic_vm_id_lock); | |
1377 | return 0; | |
1378 | } | |
1379 | ||
1380 | static void avic_vm_destroy(struct kvm *kvm) | |
1381 | { | |
1382 | unsigned long flags; | |
1383 | struct kvm_arch *vm_data = &kvm->arch; | |
1384 | ||
1385 | avic_free_vm_id(vm_data->avic_vm_id); | |
1386 | ||
1387 | if (vm_data->avic_logical_id_table_page) | |
1388 | __free_page(vm_data->avic_logical_id_table_page); | |
1389 | if (vm_data->avic_physical_id_table_page) | |
1390 | __free_page(vm_data->avic_physical_id_table_page); | |
1391 | ||
1392 | spin_lock_irqsave(&svm_vm_data_hash_lock, flags); | |
1393 | hash_del(&vm_data->hnode); | |
1394 | spin_unlock_irqrestore(&svm_vm_data_hash_lock, flags); | |
1395 | } | |
1396 | ||
1397 | static int avic_vm_init(struct kvm *kvm) | |
1398 | { | |
1399 | unsigned long flags; | |
1400 | int vm_id, err = -ENOMEM; | |
1401 | struct kvm_arch *vm_data = &kvm->arch; | |
1402 | struct page *p_page; | |
1403 | struct page *l_page; | |
1404 | ||
1405 | if (!avic) | |
1406 | return 0; | |
1407 | ||
1408 | vm_id = avic_get_next_vm_id(); | |
1409 | if (vm_id < 0) | |
1410 | return vm_id; | |
1411 | vm_data->avic_vm_id = (u32)vm_id; | |
1412 | ||
1413 | /* Allocating physical APIC ID table (4KB) */ | |
1414 | p_page = alloc_page(GFP_KERNEL); | |
1415 | if (!p_page) | |
1416 | goto free_avic; | |
1417 | ||
1418 | vm_data->avic_physical_id_table_page = p_page; | |
1419 | clear_page(page_address(p_page)); | |
1420 | ||
1421 | /* Allocating logical APIC ID table (4KB) */ | |
1422 | l_page = alloc_page(GFP_KERNEL); | |
1423 | if (!l_page) | |
1424 | goto free_avic; | |
1425 | ||
1426 | vm_data->avic_logical_id_table_page = l_page; | |
1427 | clear_page(page_address(l_page)); | |
1428 | ||
1429 | spin_lock_irqsave(&svm_vm_data_hash_lock, flags); | |
1430 | hash_add(svm_vm_data_hash, &vm_data->hnode, vm_data->avic_vm_id); | |
1431 | spin_unlock_irqrestore(&svm_vm_data_hash_lock, flags); | |
1432 | ||
1433 | return 0; | |
1434 | ||
1435 | free_avic: | |
1436 | avic_vm_destroy(kvm); | |
1437 | return err; | |
1438 | } | |
1439 | ||
1440 | static inline int | |
1441 | avic_update_iommu_vcpu_affinity(struct kvm_vcpu *vcpu, int cpu, bool r) | |
1442 | { | |
1443 | int ret = 0; | |
1444 | unsigned long flags; | |
1445 | struct amd_svm_iommu_ir *ir; | |
1446 | struct vcpu_svm *svm = to_svm(vcpu); | |
1447 | ||
1448 | if (!kvm_arch_has_assigned_device(vcpu->kvm)) | |
1449 | return 0; | |
1450 | ||
1451 | /* | |
1452 | * Here, we go through the per-vcpu ir_list to update all existing | |
1453 | * interrupt remapping table entry targeting this vcpu. | |
1454 | */ | |
1455 | spin_lock_irqsave(&svm->ir_list_lock, flags); | |
1456 | ||
1457 | if (list_empty(&svm->ir_list)) | |
1458 | goto out; | |
1459 | ||
1460 | list_for_each_entry(ir, &svm->ir_list, node) { | |
1461 | ret = amd_iommu_update_ga(cpu, r, ir->data); | |
1462 | if (ret) | |
1463 | break; | |
1464 | } | |
1465 | out: | |
1466 | spin_unlock_irqrestore(&svm->ir_list_lock, flags); | |
1467 | return ret; | |
1468 | } | |
1469 | ||
1470 | static void avic_vcpu_load(struct kvm_vcpu *vcpu, int cpu) | |
1471 | { | |
1472 | u64 entry; | |
1473 | /* ID = 0xff (broadcast), ID > 0xff (reserved) */ | |
1474 | int h_physical_id = kvm_cpu_get_apicid(cpu); | |
1475 | struct vcpu_svm *svm = to_svm(vcpu); | |
1476 | ||
1477 | if (!kvm_vcpu_apicv_active(vcpu)) | |
1478 | return; | |
1479 | ||
1480 | if (WARN_ON(h_physical_id >= AVIC_MAX_PHYSICAL_ID_COUNT)) | |
1481 | return; | |
1482 | ||
1483 | entry = READ_ONCE(*(svm->avic_physical_id_cache)); | |
1484 | WARN_ON(entry & AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK); | |
1485 | ||
1486 | entry &= ~AVIC_PHYSICAL_ID_ENTRY_HOST_PHYSICAL_ID_MASK; | |
1487 | entry |= (h_physical_id & AVIC_PHYSICAL_ID_ENTRY_HOST_PHYSICAL_ID_MASK); | |
1488 | ||
1489 | entry &= ~AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK; | |
1490 | if (svm->avic_is_running) | |
1491 | entry |= AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK; | |
1492 | ||
1493 | WRITE_ONCE(*(svm->avic_physical_id_cache), entry); | |
1494 | avic_update_iommu_vcpu_affinity(vcpu, h_physical_id, | |
1495 | svm->avic_is_running); | |
1496 | } | |
1497 | ||
1498 | static void avic_vcpu_put(struct kvm_vcpu *vcpu) | |
1499 | { | |
1500 | u64 entry; | |
1501 | struct vcpu_svm *svm = to_svm(vcpu); | |
1502 | ||
1503 | if (!kvm_vcpu_apicv_active(vcpu)) | |
1504 | return; | |
1505 | ||
1506 | entry = READ_ONCE(*(svm->avic_physical_id_cache)); | |
1507 | if (entry & AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK) | |
1508 | avic_update_iommu_vcpu_affinity(vcpu, -1, 0); | |
1509 | ||
1510 | entry &= ~AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK; | |
1511 | WRITE_ONCE(*(svm->avic_physical_id_cache), entry); | |
1512 | } | |
1513 | ||
1514 | /** | |
1515 | * This function is called during VCPU halt/unhalt. | |
1516 | */ | |
1517 | static void avic_set_running(struct kvm_vcpu *vcpu, bool is_run) | |
1518 | { | |
1519 | struct vcpu_svm *svm = to_svm(vcpu); | |
1520 | ||
1521 | svm->avic_is_running = is_run; | |
1522 | if (is_run) | |
1523 | avic_vcpu_load(vcpu, vcpu->cpu); | |
1524 | else | |
1525 | avic_vcpu_put(vcpu); | |
1526 | } | |
1527 | ||
1528 | static void svm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event) | |
1529 | { | |
1530 | struct vcpu_svm *svm = to_svm(vcpu); | |
1531 | u32 dummy; | |
1532 | u32 eax = 1; | |
1533 | ||
1534 | if (!init_event) { | |
1535 | svm->vcpu.arch.apic_base = APIC_DEFAULT_PHYS_BASE | | |
1536 | MSR_IA32_APICBASE_ENABLE; | |
1537 | if (kvm_vcpu_is_reset_bsp(&svm->vcpu)) | |
1538 | svm->vcpu.arch.apic_base |= MSR_IA32_APICBASE_BSP; | |
1539 | } | |
1540 | init_vmcb(svm); | |
1541 | ||
1542 | kvm_cpuid(vcpu, &eax, &dummy, &dummy, &dummy); | |
1543 | kvm_register_write(vcpu, VCPU_REGS_RDX, eax); | |
1544 | ||
1545 | if (kvm_vcpu_apicv_active(vcpu) && !init_event) | |
1546 | avic_update_vapic_bar(svm, APIC_DEFAULT_PHYS_BASE); | |
1547 | } | |
1548 | ||
1549 | static struct kvm_vcpu *svm_create_vcpu(struct kvm *kvm, unsigned int id) | |
1550 | { | |
1551 | struct vcpu_svm *svm; | |
1552 | struct page *page; | |
1553 | struct page *msrpm_pages; | |
1554 | struct page *hsave_page; | |
1555 | struct page *nested_msrpm_pages; | |
1556 | int err; | |
1557 | ||
1558 | svm = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL); | |
1559 | if (!svm) { | |
1560 | err = -ENOMEM; | |
1561 | goto out; | |
1562 | } | |
1563 | ||
1564 | err = kvm_vcpu_init(&svm->vcpu, kvm, id); | |
1565 | if (err) | |
1566 | goto free_svm; | |
1567 | ||
1568 | err = -ENOMEM; | |
1569 | page = alloc_page(GFP_KERNEL); | |
1570 | if (!page) | |
1571 | goto uninit; | |
1572 | ||
1573 | msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER); | |
1574 | if (!msrpm_pages) | |
1575 | goto free_page1; | |
1576 | ||
1577 | nested_msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER); | |
1578 | if (!nested_msrpm_pages) | |
1579 | goto free_page2; | |
1580 | ||
1581 | hsave_page = alloc_page(GFP_KERNEL); | |
1582 | if (!hsave_page) | |
1583 | goto free_page3; | |
1584 | ||
1585 | if (avic) { | |
1586 | err = avic_init_backing_page(&svm->vcpu); | |
1587 | if (err) | |
1588 | goto free_page4; | |
1589 | ||
1590 | INIT_LIST_HEAD(&svm->ir_list); | |
1591 | spin_lock_init(&svm->ir_list_lock); | |
1592 | } | |
1593 | ||
1594 | /* We initialize this flag to true to make sure that the is_running | |
1595 | * bit would be set the first time the vcpu is loaded. | |
1596 | */ | |
1597 | svm->avic_is_running = true; | |
1598 | ||
1599 | svm->nested.hsave = page_address(hsave_page); | |
1600 | ||
1601 | svm->msrpm = page_address(msrpm_pages); | |
1602 | svm_vcpu_init_msrpm(svm->msrpm); | |
1603 | ||
1604 | svm->nested.msrpm = page_address(nested_msrpm_pages); | |
1605 | svm_vcpu_init_msrpm(svm->nested.msrpm); | |
1606 | ||
1607 | svm->vmcb = page_address(page); | |
1608 | clear_page(svm->vmcb); | |
1609 | svm->vmcb_pa = page_to_pfn(page) << PAGE_SHIFT; | |
1610 | svm->asid_generation = 0; | |
1611 | init_vmcb(svm); | |
1612 | ||
1613 | svm_init_osvw(&svm->vcpu); | |
1614 | ||
1615 | return &svm->vcpu; | |
1616 | ||
1617 | free_page4: | |
1618 | __free_page(hsave_page); | |
1619 | free_page3: | |
1620 | __free_pages(nested_msrpm_pages, MSRPM_ALLOC_ORDER); | |
1621 | free_page2: | |
1622 | __free_pages(msrpm_pages, MSRPM_ALLOC_ORDER); | |
1623 | free_page1: | |
1624 | __free_page(page); | |
1625 | uninit: | |
1626 | kvm_vcpu_uninit(&svm->vcpu); | |
1627 | free_svm: | |
1628 | kmem_cache_free(kvm_vcpu_cache, svm); | |
1629 | out: | |
1630 | return ERR_PTR(err); | |
1631 | } | |
1632 | ||
1633 | static void svm_free_vcpu(struct kvm_vcpu *vcpu) | |
1634 | { | |
1635 | struct vcpu_svm *svm = to_svm(vcpu); | |
1636 | ||
1637 | __free_page(pfn_to_page(svm->vmcb_pa >> PAGE_SHIFT)); | |
1638 | __free_pages(virt_to_page(svm->msrpm), MSRPM_ALLOC_ORDER); | |
1639 | __free_page(virt_to_page(svm->nested.hsave)); | |
1640 | __free_pages(virt_to_page(svm->nested.msrpm), MSRPM_ALLOC_ORDER); | |
1641 | kvm_vcpu_uninit(vcpu); | |
1642 | kmem_cache_free(kvm_vcpu_cache, svm); | |
1643 | } | |
1644 | ||
1645 | static void svm_vcpu_load(struct kvm_vcpu *vcpu, int cpu) | |
1646 | { | |
1647 | struct vcpu_svm *svm = to_svm(vcpu); | |
1648 | int i; | |
1649 | ||
1650 | if (unlikely(cpu != vcpu->cpu)) { | |
1651 | svm->asid_generation = 0; | |
1652 | mark_all_dirty(svm->vmcb); | |
1653 | } | |
1654 | ||
1655 | #ifdef CONFIG_X86_64 | |
1656 | rdmsrl(MSR_GS_BASE, to_svm(vcpu)->host.gs_base); | |
1657 | #endif | |
1658 | savesegment(fs, svm->host.fs); | |
1659 | savesegment(gs, svm->host.gs); | |
1660 | svm->host.ldt = kvm_read_ldt(); | |
1661 | ||
1662 | for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++) | |
1663 | rdmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]); | |
1664 | ||
1665 | if (static_cpu_has(X86_FEATURE_TSCRATEMSR)) { | |
1666 | u64 tsc_ratio = vcpu->arch.tsc_scaling_ratio; | |
1667 | if (tsc_ratio != __this_cpu_read(current_tsc_ratio)) { | |
1668 | __this_cpu_write(current_tsc_ratio, tsc_ratio); | |
1669 | wrmsrl(MSR_AMD64_TSC_RATIO, tsc_ratio); | |
1670 | } | |
1671 | } | |
1672 | /* This assumes that the kernel never uses MSR_TSC_AUX */ | |
1673 | if (static_cpu_has(X86_FEATURE_RDTSCP)) | |
1674 | wrmsrl(MSR_TSC_AUX, svm->tsc_aux); | |
1675 | ||
1676 | avic_vcpu_load(vcpu, cpu); | |
1677 | } | |
1678 | ||
1679 | static void svm_vcpu_put(struct kvm_vcpu *vcpu) | |
1680 | { | |
1681 | struct vcpu_svm *svm = to_svm(vcpu); | |
1682 | int i; | |
1683 | ||
1684 | avic_vcpu_put(vcpu); | |
1685 | ||
1686 | ++vcpu->stat.host_state_reload; | |
1687 | kvm_load_ldt(svm->host.ldt); | |
1688 | #ifdef CONFIG_X86_64 | |
1689 | loadsegment(fs, svm->host.fs); | |
1690 | wrmsrl(MSR_KERNEL_GS_BASE, current->thread.gsbase); | |
1691 | load_gs_index(svm->host.gs); | |
1692 | #else | |
1693 | #ifdef CONFIG_X86_32_LAZY_GS | |
1694 | loadsegment(gs, svm->host.gs); | |
1695 | #endif | |
1696 | #endif | |
1697 | for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++) | |
1698 | wrmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]); | |
1699 | } | |
1700 | ||
1701 | static void svm_vcpu_blocking(struct kvm_vcpu *vcpu) | |
1702 | { | |
1703 | avic_set_running(vcpu, false); | |
1704 | } | |
1705 | ||
1706 | static void svm_vcpu_unblocking(struct kvm_vcpu *vcpu) | |
1707 | { | |
1708 | avic_set_running(vcpu, true); | |
1709 | } | |
1710 | ||
1711 | static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu) | |
1712 | { | |
1713 | return to_svm(vcpu)->vmcb->save.rflags; | |
1714 | } | |
1715 | ||
1716 | static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags) | |
1717 | { | |
1718 | /* | |
1719 | * Any change of EFLAGS.VM is accompanied by a reload of SS | |
1720 | * (caused by either a task switch or an inter-privilege IRET), | |
1721 | * so we do not need to update the CPL here. | |
1722 | */ | |
1723 | to_svm(vcpu)->vmcb->save.rflags = rflags; | |
1724 | } | |
1725 | ||
1726 | static u32 svm_get_pkru(struct kvm_vcpu *vcpu) | |
1727 | { | |
1728 | return 0; | |
1729 | } | |
1730 | ||
1731 | static void svm_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg) | |
1732 | { | |
1733 | switch (reg) { | |
1734 | case VCPU_EXREG_PDPTR: | |
1735 | BUG_ON(!npt_enabled); | |
1736 | load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu)); | |
1737 | break; | |
1738 | default: | |
1739 | BUG(); | |
1740 | } | |
1741 | } | |
1742 | ||
1743 | static void svm_set_vintr(struct vcpu_svm *svm) | |
1744 | { | |
1745 | set_intercept(svm, INTERCEPT_VINTR); | |
1746 | } | |
1747 | ||
1748 | static void svm_clear_vintr(struct vcpu_svm *svm) | |
1749 | { | |
1750 | clr_intercept(svm, INTERCEPT_VINTR); | |
1751 | } | |
1752 | ||
1753 | static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg) | |
1754 | { | |
1755 | struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save; | |
1756 | ||
1757 | switch (seg) { | |
1758 | case VCPU_SREG_CS: return &save->cs; | |
1759 | case VCPU_SREG_DS: return &save->ds; | |
1760 | case VCPU_SREG_ES: return &save->es; | |
1761 | case VCPU_SREG_FS: return &save->fs; | |
1762 | case VCPU_SREG_GS: return &save->gs; | |
1763 | case VCPU_SREG_SS: return &save->ss; | |
1764 | case VCPU_SREG_TR: return &save->tr; | |
1765 | case VCPU_SREG_LDTR: return &save->ldtr; | |
1766 | } | |
1767 | BUG(); | |
1768 | return NULL; | |
1769 | } | |
1770 | ||
1771 | static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg) | |
1772 | { | |
1773 | struct vmcb_seg *s = svm_seg(vcpu, seg); | |
1774 | ||
1775 | return s->base; | |
1776 | } | |
1777 | ||
1778 | static void svm_get_segment(struct kvm_vcpu *vcpu, | |
1779 | struct kvm_segment *var, int seg) | |
1780 | { | |
1781 | struct vmcb_seg *s = svm_seg(vcpu, seg); | |
1782 | ||
1783 | var->base = s->base; | |
1784 | var->limit = s->limit; | |
1785 | var->selector = s->selector; | |
1786 | var->type = s->attrib & SVM_SELECTOR_TYPE_MASK; | |
1787 | var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1; | |
1788 | var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3; | |
1789 | var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1; | |
1790 | var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1; | |
1791 | var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1; | |
1792 | var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1; | |
1793 | ||
1794 | /* | |
1795 | * AMD CPUs circa 2014 track the G bit for all segments except CS. | |
1796 | * However, the SVM spec states that the G bit is not observed by the | |
1797 | * CPU, and some VMware virtual CPUs drop the G bit for all segments. | |
1798 | * So let's synthesize a legal G bit for all segments, this helps | |
1799 | * running KVM nested. It also helps cross-vendor migration, because | |
1800 | * Intel's vmentry has a check on the 'G' bit. | |
1801 | */ | |
1802 | var->g = s->limit > 0xfffff; | |
1803 | ||
1804 | /* | |
1805 | * AMD's VMCB does not have an explicit unusable field, so emulate it | |
1806 | * for cross vendor migration purposes by "not present" | |
1807 | */ | |
1808 | var->unusable = !var->present || (var->type == 0); | |
1809 | ||
1810 | switch (seg) { | |
1811 | case VCPU_SREG_TR: | |
1812 | /* | |
1813 | * Work around a bug where the busy flag in the tr selector | |
1814 | * isn't exposed | |
1815 | */ | |
1816 | var->type |= 0x2; | |
1817 | break; | |
1818 | case VCPU_SREG_DS: | |
1819 | case VCPU_SREG_ES: | |
1820 | case VCPU_SREG_FS: | |
1821 | case VCPU_SREG_GS: | |
1822 | /* | |
1823 | * The accessed bit must always be set in the segment | |
1824 | * descriptor cache, although it can be cleared in the | |
1825 | * descriptor, the cached bit always remains at 1. Since | |
1826 | * Intel has a check on this, set it here to support | |
1827 | * cross-vendor migration. | |
1828 | */ | |
1829 | if (!var->unusable) | |
1830 | var->type |= 0x1; | |
1831 | break; | |
1832 | case VCPU_SREG_SS: | |
1833 | /* | |
1834 | * On AMD CPUs sometimes the DB bit in the segment | |
1835 | * descriptor is left as 1, although the whole segment has | |
1836 | * been made unusable. Clear it here to pass an Intel VMX | |
1837 | * entry check when cross vendor migrating. | |
1838 | */ | |
1839 | if (var->unusable) | |
1840 | var->db = 0; | |
1841 | var->dpl = to_svm(vcpu)->vmcb->save.cpl; | |
1842 | break; | |
1843 | } | |
1844 | } | |
1845 | ||
1846 | static int svm_get_cpl(struct kvm_vcpu *vcpu) | |
1847 | { | |
1848 | struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save; | |
1849 | ||
1850 | return save->cpl; | |
1851 | } | |
1852 | ||
1853 | static void svm_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt) | |
1854 | { | |
1855 | struct vcpu_svm *svm = to_svm(vcpu); | |
1856 | ||
1857 | dt->size = svm->vmcb->save.idtr.limit; | |
1858 | dt->address = svm->vmcb->save.idtr.base; | |
1859 | } | |
1860 | ||
1861 | static void svm_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt) | |
1862 | { | |
1863 | struct vcpu_svm *svm = to_svm(vcpu); | |
1864 | ||
1865 | svm->vmcb->save.idtr.limit = dt->size; | |
1866 | svm->vmcb->save.idtr.base = dt->address ; | |
1867 | mark_dirty(svm->vmcb, VMCB_DT); | |
1868 | } | |
1869 | ||
1870 | static void svm_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt) | |
1871 | { | |
1872 | struct vcpu_svm *svm = to_svm(vcpu); | |
1873 | ||
1874 | dt->size = svm->vmcb->save.gdtr.limit; | |
1875 | dt->address = svm->vmcb->save.gdtr.base; | |
1876 | } | |
1877 | ||
1878 | static void svm_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt) | |
1879 | { | |
1880 | struct vcpu_svm *svm = to_svm(vcpu); | |
1881 | ||
1882 | svm->vmcb->save.gdtr.limit = dt->size; | |
1883 | svm->vmcb->save.gdtr.base = dt->address ; | |
1884 | mark_dirty(svm->vmcb, VMCB_DT); | |
1885 | } | |
1886 | ||
1887 | static void svm_decache_cr0_guest_bits(struct kvm_vcpu *vcpu) | |
1888 | { | |
1889 | } | |
1890 | ||
1891 | static void svm_decache_cr3(struct kvm_vcpu *vcpu) | |
1892 | { | |
1893 | } | |
1894 | ||
1895 | static void svm_decache_cr4_guest_bits(struct kvm_vcpu *vcpu) | |
1896 | { | |
1897 | } | |
1898 | ||
1899 | static void update_cr0_intercept(struct vcpu_svm *svm) | |
1900 | { | |
1901 | ulong gcr0 = svm->vcpu.arch.cr0; | |
1902 | u64 *hcr0 = &svm->vmcb->save.cr0; | |
1903 | ||
1904 | if (!svm->vcpu.fpu_active) | |
1905 | *hcr0 |= SVM_CR0_SELECTIVE_MASK; | |
1906 | else | |
1907 | *hcr0 = (*hcr0 & ~SVM_CR0_SELECTIVE_MASK) | |
1908 | | (gcr0 & SVM_CR0_SELECTIVE_MASK); | |
1909 | ||
1910 | mark_dirty(svm->vmcb, VMCB_CR); | |
1911 | ||
1912 | if (gcr0 == *hcr0 && svm->vcpu.fpu_active) { | |
1913 | clr_cr_intercept(svm, INTERCEPT_CR0_READ); | |
1914 | clr_cr_intercept(svm, INTERCEPT_CR0_WRITE); | |
1915 | } else { | |
1916 | set_cr_intercept(svm, INTERCEPT_CR0_READ); | |
1917 | set_cr_intercept(svm, INTERCEPT_CR0_WRITE); | |
1918 | } | |
1919 | } | |
1920 | ||
1921 | static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0) | |
1922 | { | |
1923 | struct vcpu_svm *svm = to_svm(vcpu); | |
1924 | ||
1925 | #ifdef CONFIG_X86_64 | |
1926 | if (vcpu->arch.efer & EFER_LME) { | |
1927 | if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) { | |
1928 | vcpu->arch.efer |= EFER_LMA; | |
1929 | svm->vmcb->save.efer |= EFER_LMA | EFER_LME; | |
1930 | } | |
1931 | ||
1932 | if (is_paging(vcpu) && !(cr0 & X86_CR0_PG)) { | |
1933 | vcpu->arch.efer &= ~EFER_LMA; | |
1934 | svm->vmcb->save.efer &= ~(EFER_LMA | EFER_LME); | |
1935 | } | |
1936 | } | |
1937 | #endif | |
1938 | vcpu->arch.cr0 = cr0; | |
1939 | ||
1940 | if (!npt_enabled) | |
1941 | cr0 |= X86_CR0_PG | X86_CR0_WP; | |
1942 | ||
1943 | if (!vcpu->fpu_active) | |
1944 | cr0 |= X86_CR0_TS; | |
1945 | /* | |
1946 | * re-enable caching here because the QEMU bios | |
1947 | * does not do it - this results in some delay at | |
1948 | * reboot | |
1949 | */ | |
1950 | if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED)) | |
1951 | cr0 &= ~(X86_CR0_CD | X86_CR0_NW); | |
1952 | svm->vmcb->save.cr0 = cr0; | |
1953 | mark_dirty(svm->vmcb, VMCB_CR); | |
1954 | update_cr0_intercept(svm); | |
1955 | } | |
1956 | ||
1957 | static int svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4) | |
1958 | { | |
1959 | unsigned long host_cr4_mce = cr4_read_shadow() & X86_CR4_MCE; | |
1960 | unsigned long old_cr4 = to_svm(vcpu)->vmcb->save.cr4; | |
1961 | ||
1962 | if (cr4 & X86_CR4_VMXE) | |
1963 | return 1; | |
1964 | ||
1965 | if (npt_enabled && ((old_cr4 ^ cr4) & X86_CR4_PGE)) | |
1966 | svm_flush_tlb(vcpu); | |
1967 | ||
1968 | vcpu->arch.cr4 = cr4; | |
1969 | if (!npt_enabled) | |
1970 | cr4 |= X86_CR4_PAE; | |
1971 | cr4 |= host_cr4_mce; | |
1972 | to_svm(vcpu)->vmcb->save.cr4 = cr4; | |
1973 | mark_dirty(to_svm(vcpu)->vmcb, VMCB_CR); | |
1974 | return 0; | |
1975 | } | |
1976 | ||
1977 | static void svm_set_segment(struct kvm_vcpu *vcpu, | |
1978 | struct kvm_segment *var, int seg) | |
1979 | { | |
1980 | struct vcpu_svm *svm = to_svm(vcpu); | |
1981 | struct vmcb_seg *s = svm_seg(vcpu, seg); | |
1982 | ||
1983 | s->base = var->base; | |
1984 | s->limit = var->limit; | |
1985 | s->selector = var->selector; | |
1986 | if (var->unusable) | |
1987 | s->attrib = 0; | |
1988 | else { | |
1989 | s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK); | |
1990 | s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT; | |
1991 | s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT; | |
1992 | s->attrib |= (var->present & 1) << SVM_SELECTOR_P_SHIFT; | |
1993 | s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT; | |
1994 | s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT; | |
1995 | s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT; | |
1996 | s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT; | |
1997 | } | |
1998 | ||
1999 | /* | |
2000 | * This is always accurate, except if SYSRET returned to a segment | |
2001 | * with SS.DPL != 3. Intel does not have this quirk, and always | |
2002 | * forces SS.DPL to 3 on sysret, so we ignore that case; fixing it | |
2003 | * would entail passing the CPL to userspace and back. | |
2004 | */ | |
2005 | if (seg == VCPU_SREG_SS) | |
2006 | svm->vmcb->save.cpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3; | |
2007 | ||
2008 | mark_dirty(svm->vmcb, VMCB_SEG); | |
2009 | } | |
2010 | ||
2011 | static void update_bp_intercept(struct kvm_vcpu *vcpu) | |
2012 | { | |
2013 | struct vcpu_svm *svm = to_svm(vcpu); | |
2014 | ||
2015 | clr_exception_intercept(svm, BP_VECTOR); | |
2016 | ||
2017 | if (vcpu->guest_debug & KVM_GUESTDBG_ENABLE) { | |
2018 | if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) | |
2019 | set_exception_intercept(svm, BP_VECTOR); | |
2020 | } else | |
2021 | vcpu->guest_debug = 0; | |
2022 | } | |
2023 | ||
2024 | static void new_asid(struct vcpu_svm *svm, struct svm_cpu_data *sd) | |
2025 | { | |
2026 | if (sd->next_asid > sd->max_asid) { | |
2027 | ++sd->asid_generation; | |
2028 | sd->next_asid = 1; | |
2029 | svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID; | |
2030 | } | |
2031 | ||
2032 | svm->asid_generation = sd->asid_generation; | |
2033 | svm->vmcb->control.asid = sd->next_asid++; | |
2034 | ||
2035 | mark_dirty(svm->vmcb, VMCB_ASID); | |
2036 | } | |
2037 | ||
2038 | static u64 svm_get_dr6(struct kvm_vcpu *vcpu) | |
2039 | { | |
2040 | return to_svm(vcpu)->vmcb->save.dr6; | |
2041 | } | |
2042 | ||
2043 | static void svm_set_dr6(struct kvm_vcpu *vcpu, unsigned long value) | |
2044 | { | |
2045 | struct vcpu_svm *svm = to_svm(vcpu); | |
2046 | ||
2047 | svm->vmcb->save.dr6 = value; | |
2048 | mark_dirty(svm->vmcb, VMCB_DR); | |
2049 | } | |
2050 | ||
2051 | static void svm_sync_dirty_debug_regs(struct kvm_vcpu *vcpu) | |
2052 | { | |
2053 | struct vcpu_svm *svm = to_svm(vcpu); | |
2054 | ||
2055 | get_debugreg(vcpu->arch.db[0], 0); | |
2056 | get_debugreg(vcpu->arch.db[1], 1); | |
2057 | get_debugreg(vcpu->arch.db[2], 2); | |
2058 | get_debugreg(vcpu->arch.db[3], 3); | |
2059 | vcpu->arch.dr6 = svm_get_dr6(vcpu); | |
2060 | vcpu->arch.dr7 = svm->vmcb->save.dr7; | |
2061 | ||
2062 | vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT; | |
2063 | set_dr_intercepts(svm); | |
2064 | } | |
2065 | ||
2066 | static void svm_set_dr7(struct kvm_vcpu *vcpu, unsigned long value) | |
2067 | { | |
2068 | struct vcpu_svm *svm = to_svm(vcpu); | |
2069 | ||
2070 | svm->vmcb->save.dr7 = value; | |
2071 | mark_dirty(svm->vmcb, VMCB_DR); | |
2072 | } | |
2073 | ||
2074 | static int pf_interception(struct vcpu_svm *svm) | |
2075 | { | |
2076 | u64 fault_address = svm->vmcb->control.exit_info_2; | |
2077 | u64 error_code; | |
2078 | int r = 1; | |
2079 | ||
2080 | switch (svm->apf_reason) { | |
2081 | default: | |
2082 | error_code = svm->vmcb->control.exit_info_1; | |
2083 | ||
2084 | trace_kvm_page_fault(fault_address, error_code); | |
2085 | if (!npt_enabled && kvm_event_needs_reinjection(&svm->vcpu)) | |
2086 | kvm_mmu_unprotect_page_virt(&svm->vcpu, fault_address); | |
2087 | r = kvm_mmu_page_fault(&svm->vcpu, fault_address, error_code, | |
2088 | svm->vmcb->control.insn_bytes, | |
2089 | svm->vmcb->control.insn_len); | |
2090 | break; | |
2091 | case KVM_PV_REASON_PAGE_NOT_PRESENT: | |
2092 | svm->apf_reason = 0; | |
2093 | local_irq_disable(); | |
2094 | kvm_async_pf_task_wait(fault_address); | |
2095 | local_irq_enable(); | |
2096 | break; | |
2097 | case KVM_PV_REASON_PAGE_READY: | |
2098 | svm->apf_reason = 0; | |
2099 | local_irq_disable(); | |
2100 | kvm_async_pf_task_wake(fault_address); | |
2101 | local_irq_enable(); | |
2102 | break; | |
2103 | } | |
2104 | return r; | |
2105 | } | |
2106 | ||
2107 | static int db_interception(struct vcpu_svm *svm) | |
2108 | { | |
2109 | struct kvm_run *kvm_run = svm->vcpu.run; | |
2110 | ||
2111 | if (!(svm->vcpu.guest_debug & | |
2112 | (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) && | |
2113 | !svm->nmi_singlestep) { | |
2114 | kvm_queue_exception(&svm->vcpu, DB_VECTOR); | |
2115 | return 1; | |
2116 | } | |
2117 | ||
2118 | if (svm->nmi_singlestep) { | |
2119 | svm->nmi_singlestep = false; | |
2120 | if (!(svm->vcpu.guest_debug & KVM_GUESTDBG_SINGLESTEP)) | |
2121 | svm->vmcb->save.rflags &= | |
2122 | ~(X86_EFLAGS_TF | X86_EFLAGS_RF); | |
2123 | } | |
2124 | ||
2125 | if (svm->vcpu.guest_debug & | |
2126 | (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) { | |
2127 | kvm_run->exit_reason = KVM_EXIT_DEBUG; | |
2128 | kvm_run->debug.arch.pc = | |
2129 | svm->vmcb->save.cs.base + svm->vmcb->save.rip; | |
2130 | kvm_run->debug.arch.exception = DB_VECTOR; | |
2131 | return 0; | |
2132 | } | |
2133 | ||
2134 | return 1; | |
2135 | } | |
2136 | ||
2137 | static int bp_interception(struct vcpu_svm *svm) | |
2138 | { | |
2139 | struct kvm_run *kvm_run = svm->vcpu.run; | |
2140 | ||
2141 | kvm_run->exit_reason = KVM_EXIT_DEBUG; | |
2142 | kvm_run->debug.arch.pc = svm->vmcb->save.cs.base + svm->vmcb->save.rip; | |
2143 | kvm_run->debug.arch.exception = BP_VECTOR; | |
2144 | return 0; | |
2145 | } | |
2146 | ||
2147 | static int ud_interception(struct vcpu_svm *svm) | |
2148 | { | |
2149 | int er; | |
2150 | ||
2151 | er = emulate_instruction(&svm->vcpu, EMULTYPE_TRAP_UD); | |
2152 | if (er != EMULATE_DONE) | |
2153 | kvm_queue_exception(&svm->vcpu, UD_VECTOR); | |
2154 | return 1; | |
2155 | } | |
2156 | ||
2157 | static int ac_interception(struct vcpu_svm *svm) | |
2158 | { | |
2159 | kvm_queue_exception_e(&svm->vcpu, AC_VECTOR, 0); | |
2160 | return 1; | |
2161 | } | |
2162 | ||
2163 | static void svm_fpu_activate(struct kvm_vcpu *vcpu) | |
2164 | { | |
2165 | struct vcpu_svm *svm = to_svm(vcpu); | |
2166 | ||
2167 | clr_exception_intercept(svm, NM_VECTOR); | |
2168 | ||
2169 | svm->vcpu.fpu_active = 1; | |
2170 | update_cr0_intercept(svm); | |
2171 | } | |
2172 | ||
2173 | static int nm_interception(struct vcpu_svm *svm) | |
2174 | { | |
2175 | svm_fpu_activate(&svm->vcpu); | |
2176 | return 1; | |
2177 | } | |
2178 | ||
2179 | static bool is_erratum_383(void) | |
2180 | { | |
2181 | int err, i; | |
2182 | u64 value; | |
2183 | ||
2184 | if (!erratum_383_found) | |
2185 | return false; | |
2186 | ||
2187 | value = native_read_msr_safe(MSR_IA32_MC0_STATUS, &err); | |
2188 | if (err) | |
2189 | return false; | |
2190 | ||
2191 | /* Bit 62 may or may not be set for this mce */ | |
2192 | value &= ~(1ULL << 62); | |
2193 | ||
2194 | if (value != 0xb600000000010015ULL) | |
2195 | return false; | |
2196 | ||
2197 | /* Clear MCi_STATUS registers */ | |
2198 | for (i = 0; i < 6; ++i) | |
2199 | native_write_msr_safe(MSR_IA32_MCx_STATUS(i), 0, 0); | |
2200 | ||
2201 | value = native_read_msr_safe(MSR_IA32_MCG_STATUS, &err); | |
2202 | if (!err) { | |
2203 | u32 low, high; | |
2204 | ||
2205 | value &= ~(1ULL << 2); | |
2206 | low = lower_32_bits(value); | |
2207 | high = upper_32_bits(value); | |
2208 | ||
2209 | native_write_msr_safe(MSR_IA32_MCG_STATUS, low, high); | |
2210 | } | |
2211 | ||
2212 | /* Flush tlb to evict multi-match entries */ | |
2213 | __flush_tlb_all(); | |
2214 | ||
2215 | return true; | |
2216 | } | |
2217 | ||
2218 | static void svm_handle_mce(struct vcpu_svm *svm) | |
2219 | { | |
2220 | if (is_erratum_383()) { | |
2221 | /* | |
2222 | * Erratum 383 triggered. Guest state is corrupt so kill the | |
2223 | * guest. | |
2224 | */ | |
2225 | pr_err("KVM: Guest triggered AMD Erratum 383\n"); | |
2226 | ||
2227 | kvm_make_request(KVM_REQ_TRIPLE_FAULT, &svm->vcpu); | |
2228 | ||
2229 | return; | |
2230 | } | |
2231 | ||
2232 | /* | |
2233 | * On an #MC intercept the MCE handler is not called automatically in | |
2234 | * the host. So do it by hand here. | |
2235 | */ | |
2236 | asm volatile ( | |
2237 | "int $0x12\n"); | |
2238 | /* not sure if we ever come back to this point */ | |
2239 | ||
2240 | return; | |
2241 | } | |
2242 | ||
2243 | static int mc_interception(struct vcpu_svm *svm) | |
2244 | { | |
2245 | return 1; | |
2246 | } | |
2247 | ||
2248 | static int shutdown_interception(struct vcpu_svm *svm) | |
2249 | { | |
2250 | struct kvm_run *kvm_run = svm->vcpu.run; | |
2251 | ||
2252 | /* | |
2253 | * VMCB is undefined after a SHUTDOWN intercept | |
2254 | * so reinitialize it. | |
2255 | */ | |
2256 | clear_page(svm->vmcb); | |
2257 | init_vmcb(svm); | |
2258 | ||
2259 | kvm_run->exit_reason = KVM_EXIT_SHUTDOWN; | |
2260 | return 0; | |
2261 | } | |
2262 | ||
2263 | static int io_interception(struct vcpu_svm *svm) | |
2264 | { | |
2265 | struct kvm_vcpu *vcpu = &svm->vcpu; | |
2266 | u32 io_info = svm->vmcb->control.exit_info_1; /* address size bug? */ | |
2267 | int size, in, string; | |
2268 | unsigned port; | |
2269 | ||
2270 | ++svm->vcpu.stat.io_exits; | |
2271 | string = (io_info & SVM_IOIO_STR_MASK) != 0; | |
2272 | in = (io_info & SVM_IOIO_TYPE_MASK) != 0; | |
2273 | if (string || in) | |
2274 | return emulate_instruction(vcpu, 0) == EMULATE_DONE; | |
2275 | ||
2276 | port = io_info >> 16; | |
2277 | size = (io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT; | |
2278 | svm->next_rip = svm->vmcb->control.exit_info_2; | |
2279 | skip_emulated_instruction(&svm->vcpu); | |
2280 | ||
2281 | return kvm_fast_pio_out(vcpu, size, port); | |
2282 | } | |
2283 | ||
2284 | static int nmi_interception(struct vcpu_svm *svm) | |
2285 | { | |
2286 | return 1; | |
2287 | } | |
2288 | ||
2289 | static int intr_interception(struct vcpu_svm *svm) | |
2290 | { | |
2291 | ++svm->vcpu.stat.irq_exits; | |
2292 | return 1; | |
2293 | } | |
2294 | ||
2295 | static int nop_on_interception(struct vcpu_svm *svm) | |
2296 | { | |
2297 | return 1; | |
2298 | } | |
2299 | ||
2300 | static int halt_interception(struct vcpu_svm *svm) | |
2301 | { | |
2302 | svm->next_rip = kvm_rip_read(&svm->vcpu) + 1; | |
2303 | return kvm_emulate_halt(&svm->vcpu); | |
2304 | } | |
2305 | ||
2306 | static int vmmcall_interception(struct vcpu_svm *svm) | |
2307 | { | |
2308 | svm->next_rip = kvm_rip_read(&svm->vcpu) + 3; | |
2309 | return kvm_emulate_hypercall(&svm->vcpu); | |
2310 | } | |
2311 | ||
2312 | static unsigned long nested_svm_get_tdp_cr3(struct kvm_vcpu *vcpu) | |
2313 | { | |
2314 | struct vcpu_svm *svm = to_svm(vcpu); | |
2315 | ||
2316 | return svm->nested.nested_cr3; | |
2317 | } | |
2318 | ||
2319 | static u64 nested_svm_get_tdp_pdptr(struct kvm_vcpu *vcpu, int index) | |
2320 | { | |
2321 | struct vcpu_svm *svm = to_svm(vcpu); | |
2322 | u64 cr3 = svm->nested.nested_cr3; | |
2323 | u64 pdpte; | |
2324 | int ret; | |
2325 | ||
2326 | ret = kvm_vcpu_read_guest_page(vcpu, gpa_to_gfn(cr3), &pdpte, | |
2327 | offset_in_page(cr3) + index * 8, 8); | |
2328 | if (ret) | |
2329 | return 0; | |
2330 | return pdpte; | |
2331 | } | |
2332 | ||
2333 | static void nested_svm_set_tdp_cr3(struct kvm_vcpu *vcpu, | |
2334 | unsigned long root) | |
2335 | { | |
2336 | struct vcpu_svm *svm = to_svm(vcpu); | |
2337 | ||
2338 | svm->vmcb->control.nested_cr3 = root; | |
2339 | mark_dirty(svm->vmcb, VMCB_NPT); | |
2340 | svm_flush_tlb(vcpu); | |
2341 | } | |
2342 | ||
2343 | static void nested_svm_inject_npf_exit(struct kvm_vcpu *vcpu, | |
2344 | struct x86_exception *fault) | |
2345 | { | |
2346 | struct vcpu_svm *svm = to_svm(vcpu); | |
2347 | ||
2348 | if (svm->vmcb->control.exit_code != SVM_EXIT_NPF) { | |
2349 | /* | |
2350 | * TODO: track the cause of the nested page fault, and | |
2351 | * correctly fill in the high bits of exit_info_1. | |
2352 | */ | |
2353 | svm->vmcb->control.exit_code = SVM_EXIT_NPF; | |
2354 | svm->vmcb->control.exit_code_hi = 0; | |
2355 | svm->vmcb->control.exit_info_1 = (1ULL << 32); | |
2356 | svm->vmcb->control.exit_info_2 = fault->address; | |
2357 | } | |
2358 | ||
2359 | svm->vmcb->control.exit_info_1 &= ~0xffffffffULL; | |
2360 | svm->vmcb->control.exit_info_1 |= fault->error_code; | |
2361 | ||
2362 | /* | |
2363 | * The present bit is always zero for page structure faults on real | |
2364 | * hardware. | |
2365 | */ | |
2366 | if (svm->vmcb->control.exit_info_1 & (2ULL << 32)) | |
2367 | svm->vmcb->control.exit_info_1 &= ~1; | |
2368 | ||
2369 | nested_svm_vmexit(svm); | |
2370 | } | |
2371 | ||
2372 | static void nested_svm_init_mmu_context(struct kvm_vcpu *vcpu) | |
2373 | { | |
2374 | WARN_ON(mmu_is_nested(vcpu)); | |
2375 | kvm_init_shadow_mmu(vcpu); | |
2376 | vcpu->arch.mmu.set_cr3 = nested_svm_set_tdp_cr3; | |
2377 | vcpu->arch.mmu.get_cr3 = nested_svm_get_tdp_cr3; | |
2378 | vcpu->arch.mmu.get_pdptr = nested_svm_get_tdp_pdptr; | |
2379 | vcpu->arch.mmu.inject_page_fault = nested_svm_inject_npf_exit; | |
2380 | vcpu->arch.mmu.shadow_root_level = get_npt_level(); | |
2381 | reset_shadow_zero_bits_mask(vcpu, &vcpu->arch.mmu); | |
2382 | vcpu->arch.walk_mmu = &vcpu->arch.nested_mmu; | |
2383 | } | |
2384 | ||
2385 | static void nested_svm_uninit_mmu_context(struct kvm_vcpu *vcpu) | |
2386 | { | |
2387 | vcpu->arch.walk_mmu = &vcpu->arch.mmu; | |
2388 | } | |
2389 | ||
2390 | static int nested_svm_check_permissions(struct vcpu_svm *svm) | |
2391 | { | |
2392 | if (!(svm->vcpu.arch.efer & EFER_SVME) | |
2393 | || !is_paging(&svm->vcpu)) { | |
2394 | kvm_queue_exception(&svm->vcpu, UD_VECTOR); | |
2395 | return 1; | |
2396 | } | |
2397 | ||
2398 | if (svm->vmcb->save.cpl) { | |
2399 | kvm_inject_gp(&svm->vcpu, 0); | |
2400 | return 1; | |
2401 | } | |
2402 | ||
2403 | return 0; | |
2404 | } | |
2405 | ||
2406 | static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr, | |
2407 | bool has_error_code, u32 error_code) | |
2408 | { | |
2409 | int vmexit; | |
2410 | ||
2411 | if (!is_guest_mode(&svm->vcpu)) | |
2412 | return 0; | |
2413 | ||
2414 | svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + nr; | |
2415 | svm->vmcb->control.exit_code_hi = 0; | |
2416 | svm->vmcb->control.exit_info_1 = error_code; | |
2417 | svm->vmcb->control.exit_info_2 = svm->vcpu.arch.cr2; | |
2418 | ||
2419 | vmexit = nested_svm_intercept(svm); | |
2420 | if (vmexit == NESTED_EXIT_DONE) | |
2421 | svm->nested.exit_required = true; | |
2422 | ||
2423 | return vmexit; | |
2424 | } | |
2425 | ||
2426 | /* This function returns true if it is save to enable the irq window */ | |
2427 | static inline bool nested_svm_intr(struct vcpu_svm *svm) | |
2428 | { | |
2429 | if (!is_guest_mode(&svm->vcpu)) | |
2430 | return true; | |
2431 | ||
2432 | if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK)) | |
2433 | return true; | |
2434 | ||
2435 | if (!(svm->vcpu.arch.hflags & HF_HIF_MASK)) | |
2436 | return false; | |
2437 | ||
2438 | /* | |
2439 | * if vmexit was already requested (by intercepted exception | |
2440 | * for instance) do not overwrite it with "external interrupt" | |
2441 | * vmexit. | |
2442 | */ | |
2443 | if (svm->nested.exit_required) | |
2444 | return false; | |
2445 | ||
2446 | svm->vmcb->control.exit_code = SVM_EXIT_INTR; | |
2447 | svm->vmcb->control.exit_info_1 = 0; | |
2448 | svm->vmcb->control.exit_info_2 = 0; | |
2449 | ||
2450 | if (svm->nested.intercept & 1ULL) { | |
2451 | /* | |
2452 | * The #vmexit can't be emulated here directly because this | |
2453 | * code path runs with irqs and preemption disabled. A | |
2454 | * #vmexit emulation might sleep. Only signal request for | |
2455 | * the #vmexit here. | |
2456 | */ | |
2457 | svm->nested.exit_required = true; | |
2458 | trace_kvm_nested_intr_vmexit(svm->vmcb->save.rip); | |
2459 | return false; | |
2460 | } | |
2461 | ||
2462 | return true; | |
2463 | } | |
2464 | ||
2465 | /* This function returns true if it is save to enable the nmi window */ | |
2466 | static inline bool nested_svm_nmi(struct vcpu_svm *svm) | |
2467 | { | |
2468 | if (!is_guest_mode(&svm->vcpu)) | |
2469 | return true; | |
2470 | ||
2471 | if (!(svm->nested.intercept & (1ULL << INTERCEPT_NMI))) | |
2472 | return true; | |
2473 | ||
2474 | svm->vmcb->control.exit_code = SVM_EXIT_NMI; | |
2475 | svm->nested.exit_required = true; | |
2476 | ||
2477 | return false; | |
2478 | } | |
2479 | ||
2480 | static void *nested_svm_map(struct vcpu_svm *svm, u64 gpa, struct page **_page) | |
2481 | { | |
2482 | struct page *page; | |
2483 | ||
2484 | might_sleep(); | |
2485 | ||
2486 | page = kvm_vcpu_gfn_to_page(&svm->vcpu, gpa >> PAGE_SHIFT); | |
2487 | if (is_error_page(page)) | |
2488 | goto error; | |
2489 | ||
2490 | *_page = page; | |
2491 | ||
2492 | return kmap(page); | |
2493 | ||
2494 | error: | |
2495 | kvm_inject_gp(&svm->vcpu, 0); | |
2496 | ||
2497 | return NULL; | |
2498 | } | |
2499 | ||
2500 | static void nested_svm_unmap(struct page *page) | |
2501 | { | |
2502 | kunmap(page); | |
2503 | kvm_release_page_dirty(page); | |
2504 | } | |
2505 | ||
2506 | static int nested_svm_intercept_ioio(struct vcpu_svm *svm) | |
2507 | { | |
2508 | unsigned port, size, iopm_len; | |
2509 | u16 val, mask; | |
2510 | u8 start_bit; | |
2511 | u64 gpa; | |
2512 | ||
2513 | if (!(svm->nested.intercept & (1ULL << INTERCEPT_IOIO_PROT))) | |
2514 | return NESTED_EXIT_HOST; | |
2515 | ||
2516 | port = svm->vmcb->control.exit_info_1 >> 16; | |
2517 | size = (svm->vmcb->control.exit_info_1 & SVM_IOIO_SIZE_MASK) >> | |
2518 | SVM_IOIO_SIZE_SHIFT; | |
2519 | gpa = svm->nested.vmcb_iopm + (port / 8); | |
2520 | start_bit = port % 8; | |
2521 | iopm_len = (start_bit + size > 8) ? 2 : 1; | |
2522 | mask = (0xf >> (4 - size)) << start_bit; | |
2523 | val = 0; | |
2524 | ||
2525 | if (kvm_vcpu_read_guest(&svm->vcpu, gpa, &val, iopm_len)) | |
2526 | return NESTED_EXIT_DONE; | |
2527 | ||
2528 | return (val & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST; | |
2529 | } | |
2530 | ||
2531 | static int nested_svm_exit_handled_msr(struct vcpu_svm *svm) | |
2532 | { | |
2533 | u32 offset, msr, value; | |
2534 | int write, mask; | |
2535 | ||
2536 | if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT))) | |
2537 | return NESTED_EXIT_HOST; | |
2538 | ||
2539 | msr = svm->vcpu.arch.regs[VCPU_REGS_RCX]; | |
2540 | offset = svm_msrpm_offset(msr); | |
2541 | write = svm->vmcb->control.exit_info_1 & 1; | |
2542 | mask = 1 << ((2 * (msr & 0xf)) + write); | |
2543 | ||
2544 | if (offset == MSR_INVALID) | |
2545 | return NESTED_EXIT_DONE; | |
2546 | ||
2547 | /* Offset is in 32 bit units but need in 8 bit units */ | |
2548 | offset *= 4; | |
2549 | ||
2550 | if (kvm_vcpu_read_guest(&svm->vcpu, svm->nested.vmcb_msrpm + offset, &value, 4)) | |
2551 | return NESTED_EXIT_DONE; | |
2552 | ||
2553 | return (value & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST; | |
2554 | } | |
2555 | ||
2556 | static int nested_svm_exit_special(struct vcpu_svm *svm) | |
2557 | { | |
2558 | u32 exit_code = svm->vmcb->control.exit_code; | |
2559 | ||
2560 | switch (exit_code) { | |
2561 | case SVM_EXIT_INTR: | |
2562 | case SVM_EXIT_NMI: | |
2563 | case SVM_EXIT_EXCP_BASE + MC_VECTOR: | |
2564 | return NESTED_EXIT_HOST; | |
2565 | case SVM_EXIT_NPF: | |
2566 | /* For now we are always handling NPFs when using them */ | |
2567 | if (npt_enabled) | |
2568 | return NESTED_EXIT_HOST; | |
2569 | break; | |
2570 | case SVM_EXIT_EXCP_BASE + PF_VECTOR: | |
2571 | /* When we're shadowing, trap PFs, but not async PF */ | |
2572 | if (!npt_enabled && svm->apf_reason == 0) | |
2573 | return NESTED_EXIT_HOST; | |
2574 | break; | |
2575 | case SVM_EXIT_EXCP_BASE + NM_VECTOR: | |
2576 | nm_interception(svm); | |
2577 | break; | |
2578 | default: | |
2579 | break; | |
2580 | } | |
2581 | ||
2582 | return NESTED_EXIT_CONTINUE; | |
2583 | } | |
2584 | ||
2585 | /* | |
2586 | * If this function returns true, this #vmexit was already handled | |
2587 | */ | |
2588 | static int nested_svm_intercept(struct vcpu_svm *svm) | |
2589 | { | |
2590 | u32 exit_code = svm->vmcb->control.exit_code; | |
2591 | int vmexit = NESTED_EXIT_HOST; | |
2592 | ||
2593 | switch (exit_code) { | |
2594 | case SVM_EXIT_MSR: | |
2595 | vmexit = nested_svm_exit_handled_msr(svm); | |
2596 | break; | |
2597 | case SVM_EXIT_IOIO: | |
2598 | vmexit = nested_svm_intercept_ioio(svm); | |
2599 | break; | |
2600 | case SVM_EXIT_READ_CR0 ... SVM_EXIT_WRITE_CR8: { | |
2601 | u32 bit = 1U << (exit_code - SVM_EXIT_READ_CR0); | |
2602 | if (svm->nested.intercept_cr & bit) | |
2603 | vmexit = NESTED_EXIT_DONE; | |
2604 | break; | |
2605 | } | |
2606 | case SVM_EXIT_READ_DR0 ... SVM_EXIT_WRITE_DR7: { | |
2607 | u32 bit = 1U << (exit_code - SVM_EXIT_READ_DR0); | |
2608 | if (svm->nested.intercept_dr & bit) | |
2609 | vmexit = NESTED_EXIT_DONE; | |
2610 | break; | |
2611 | } | |
2612 | case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: { | |
2613 | u32 excp_bits = 1 << (exit_code - SVM_EXIT_EXCP_BASE); | |
2614 | if (svm->nested.intercept_exceptions & excp_bits) | |
2615 | vmexit = NESTED_EXIT_DONE; | |
2616 | /* async page fault always cause vmexit */ | |
2617 | else if ((exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR) && | |
2618 | svm->apf_reason != 0) | |
2619 | vmexit = NESTED_EXIT_DONE; | |
2620 | break; | |
2621 | } | |
2622 | case SVM_EXIT_ERR: { | |
2623 | vmexit = NESTED_EXIT_DONE; | |
2624 | break; | |
2625 | } | |
2626 | default: { | |
2627 | u64 exit_bits = 1ULL << (exit_code - SVM_EXIT_INTR); | |
2628 | if (svm->nested.intercept & exit_bits) | |
2629 | vmexit = NESTED_EXIT_DONE; | |
2630 | } | |
2631 | } | |
2632 | ||
2633 | return vmexit; | |
2634 | } | |
2635 | ||
2636 | static int nested_svm_exit_handled(struct vcpu_svm *svm) | |
2637 | { | |
2638 | int vmexit; | |
2639 | ||
2640 | vmexit = nested_svm_intercept(svm); | |
2641 | ||
2642 | if (vmexit == NESTED_EXIT_DONE) | |
2643 | nested_svm_vmexit(svm); | |
2644 | ||
2645 | return vmexit; | |
2646 | } | |
2647 | ||
2648 | static inline void copy_vmcb_control_area(struct vmcb *dst_vmcb, struct vmcb *from_vmcb) | |
2649 | { | |
2650 | struct vmcb_control_area *dst = &dst_vmcb->control; | |
2651 | struct vmcb_control_area *from = &from_vmcb->control; | |
2652 | ||
2653 | dst->intercept_cr = from->intercept_cr; | |
2654 | dst->intercept_dr = from->intercept_dr; | |
2655 | dst->intercept_exceptions = from->intercept_exceptions; | |
2656 | dst->intercept = from->intercept; | |
2657 | dst->iopm_base_pa = from->iopm_base_pa; | |
2658 | dst->msrpm_base_pa = from->msrpm_base_pa; | |
2659 | dst->tsc_offset = from->tsc_offset; | |
2660 | dst->asid = from->asid; | |
2661 | dst->tlb_ctl = from->tlb_ctl; | |
2662 | dst->int_ctl = from->int_ctl; | |
2663 | dst->int_vector = from->int_vector; | |
2664 | dst->int_state = from->int_state; | |
2665 | dst->exit_code = from->exit_code; | |
2666 | dst->exit_code_hi = from->exit_code_hi; | |
2667 | dst->exit_info_1 = from->exit_info_1; | |
2668 | dst->exit_info_2 = from->exit_info_2; | |
2669 | dst->exit_int_info = from->exit_int_info; | |
2670 | dst->exit_int_info_err = from->exit_int_info_err; | |
2671 | dst->nested_ctl = from->nested_ctl; | |
2672 | dst->event_inj = from->event_inj; | |
2673 | dst->event_inj_err = from->event_inj_err; | |
2674 | dst->nested_cr3 = from->nested_cr3; | |
2675 | dst->lbr_ctl = from->lbr_ctl; | |
2676 | } | |
2677 | ||
2678 | static int nested_svm_vmexit(struct vcpu_svm *svm) | |
2679 | { | |
2680 | struct vmcb *nested_vmcb; | |
2681 | struct vmcb *hsave = svm->nested.hsave; | |
2682 | struct vmcb *vmcb = svm->vmcb; | |
2683 | struct page *page; | |
2684 | ||
2685 | trace_kvm_nested_vmexit_inject(vmcb->control.exit_code, | |
2686 | vmcb->control.exit_info_1, | |
2687 | vmcb->control.exit_info_2, | |
2688 | vmcb->control.exit_int_info, | |
2689 | vmcb->control.exit_int_info_err, | |
2690 | KVM_ISA_SVM); | |
2691 | ||
2692 | nested_vmcb = nested_svm_map(svm, svm->nested.vmcb, &page); | |
2693 | if (!nested_vmcb) | |
2694 | return 1; | |
2695 | ||
2696 | /* Exit Guest-Mode */ | |
2697 | leave_guest_mode(&svm->vcpu); | |
2698 | svm->nested.vmcb = 0; | |
2699 | ||
2700 | /* Give the current vmcb to the guest */ | |
2701 | disable_gif(svm); | |
2702 | ||
2703 | nested_vmcb->save.es = vmcb->save.es; | |
2704 | nested_vmcb->save.cs = vmcb->save.cs; | |
2705 | nested_vmcb->save.ss = vmcb->save.ss; | |
2706 | nested_vmcb->save.ds = vmcb->save.ds; | |
2707 | nested_vmcb->save.gdtr = vmcb->save.gdtr; | |
2708 | nested_vmcb->save.idtr = vmcb->save.idtr; | |
2709 | nested_vmcb->save.efer = svm->vcpu.arch.efer; | |
2710 | nested_vmcb->save.cr0 = kvm_read_cr0(&svm->vcpu); | |
2711 | nested_vmcb->save.cr3 = kvm_read_cr3(&svm->vcpu); | |
2712 | nested_vmcb->save.cr2 = vmcb->save.cr2; | |
2713 | nested_vmcb->save.cr4 = svm->vcpu.arch.cr4; | |
2714 | nested_vmcb->save.rflags = kvm_get_rflags(&svm->vcpu); | |
2715 | nested_vmcb->save.rip = vmcb->save.rip; | |
2716 | nested_vmcb->save.rsp = vmcb->save.rsp; | |
2717 | nested_vmcb->save.rax = vmcb->save.rax; | |
2718 | nested_vmcb->save.dr7 = vmcb->save.dr7; | |
2719 | nested_vmcb->save.dr6 = vmcb->save.dr6; | |
2720 | nested_vmcb->save.cpl = vmcb->save.cpl; | |
2721 | ||
2722 | nested_vmcb->control.int_ctl = vmcb->control.int_ctl; | |
2723 | nested_vmcb->control.int_vector = vmcb->control.int_vector; | |
2724 | nested_vmcb->control.int_state = vmcb->control.int_state; | |
2725 | nested_vmcb->control.exit_code = vmcb->control.exit_code; | |
2726 | nested_vmcb->control.exit_code_hi = vmcb->control.exit_code_hi; | |
2727 | nested_vmcb->control.exit_info_1 = vmcb->control.exit_info_1; | |
2728 | nested_vmcb->control.exit_info_2 = vmcb->control.exit_info_2; | |
2729 | nested_vmcb->control.exit_int_info = vmcb->control.exit_int_info; | |
2730 | nested_vmcb->control.exit_int_info_err = vmcb->control.exit_int_info_err; | |
2731 | ||
2732 | if (svm->nrips_enabled) | |
2733 | nested_vmcb->control.next_rip = vmcb->control.next_rip; | |
2734 | ||
2735 | /* | |
2736 | * If we emulate a VMRUN/#VMEXIT in the same host #vmexit cycle we have | |
2737 | * to make sure that we do not lose injected events. So check event_inj | |
2738 | * here and copy it to exit_int_info if it is valid. | |
2739 | * Exit_int_info and event_inj can't be both valid because the case | |
2740 | * below only happens on a VMRUN instruction intercept which has | |
2741 | * no valid exit_int_info set. | |
2742 | */ | |
2743 | if (vmcb->control.event_inj & SVM_EVTINJ_VALID) { | |
2744 | struct vmcb_control_area *nc = &nested_vmcb->control; | |
2745 | ||
2746 | nc->exit_int_info = vmcb->control.event_inj; | |
2747 | nc->exit_int_info_err = vmcb->control.event_inj_err; | |
2748 | } | |
2749 | ||
2750 | nested_vmcb->control.tlb_ctl = 0; | |
2751 | nested_vmcb->control.event_inj = 0; | |
2752 | nested_vmcb->control.event_inj_err = 0; | |
2753 | ||
2754 | /* We always set V_INTR_MASKING and remember the old value in hflags */ | |
2755 | if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK)) | |
2756 | nested_vmcb->control.int_ctl &= ~V_INTR_MASKING_MASK; | |
2757 | ||
2758 | /* Restore the original control entries */ | |
2759 | copy_vmcb_control_area(vmcb, hsave); | |
2760 | ||
2761 | kvm_clear_exception_queue(&svm->vcpu); | |
2762 | kvm_clear_interrupt_queue(&svm->vcpu); | |
2763 | ||
2764 | svm->nested.nested_cr3 = 0; | |
2765 | ||
2766 | /* Restore selected save entries */ | |
2767 | svm->vmcb->save.es = hsave->save.es; | |
2768 | svm->vmcb->save.cs = hsave->save.cs; | |
2769 | svm->vmcb->save.ss = hsave->save.ss; | |
2770 | svm->vmcb->save.ds = hsave->save.ds; | |
2771 | svm->vmcb->save.gdtr = hsave->save.gdtr; | |
2772 | svm->vmcb->save.idtr = hsave->save.idtr; | |
2773 | kvm_set_rflags(&svm->vcpu, hsave->save.rflags); | |
2774 | svm_set_efer(&svm->vcpu, hsave->save.efer); | |
2775 | svm_set_cr0(&svm->vcpu, hsave->save.cr0 | X86_CR0_PE); | |
2776 | svm_set_cr4(&svm->vcpu, hsave->save.cr4); | |
2777 | if (npt_enabled) { | |
2778 | svm->vmcb->save.cr3 = hsave->save.cr3; | |
2779 | svm->vcpu.arch.cr3 = hsave->save.cr3; | |
2780 | } else { | |
2781 | (void)kvm_set_cr3(&svm->vcpu, hsave->save.cr3); | |
2782 | } | |
2783 | kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, hsave->save.rax); | |
2784 | kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, hsave->save.rsp); | |
2785 | kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, hsave->save.rip); | |
2786 | svm->vmcb->save.dr7 = 0; | |
2787 | svm->vmcb->save.cpl = 0; | |
2788 | svm->vmcb->control.exit_int_info = 0; | |
2789 | ||
2790 | mark_all_dirty(svm->vmcb); | |
2791 | ||
2792 | nested_svm_unmap(page); | |
2793 | ||
2794 | nested_svm_uninit_mmu_context(&svm->vcpu); | |
2795 | kvm_mmu_reset_context(&svm->vcpu); | |
2796 | kvm_mmu_load(&svm->vcpu); | |
2797 | ||
2798 | return 0; | |
2799 | } | |
2800 | ||
2801 | static bool nested_svm_vmrun_msrpm(struct vcpu_svm *svm) | |
2802 | { | |
2803 | /* | |
2804 | * This function merges the msr permission bitmaps of kvm and the | |
2805 | * nested vmcb. It is optimized in that it only merges the parts where | |
2806 | * the kvm msr permission bitmap may contain zero bits | |
2807 | */ | |
2808 | int i; | |
2809 | ||
2810 | if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT))) | |
2811 | return true; | |
2812 | ||
2813 | for (i = 0; i < MSRPM_OFFSETS; i++) { | |
2814 | u32 value, p; | |
2815 | u64 offset; | |
2816 | ||
2817 | if (msrpm_offsets[i] == 0xffffffff) | |
2818 | break; | |
2819 | ||
2820 | p = msrpm_offsets[i]; | |
2821 | offset = svm->nested.vmcb_msrpm + (p * 4); | |
2822 | ||
2823 | if (kvm_vcpu_read_guest(&svm->vcpu, offset, &value, 4)) | |
2824 | return false; | |
2825 | ||
2826 | svm->nested.msrpm[p] = svm->msrpm[p] | value; | |
2827 | } | |
2828 | ||
2829 | svm->vmcb->control.msrpm_base_pa = __pa(svm->nested.msrpm); | |
2830 | ||
2831 | return true; | |
2832 | } | |
2833 | ||
2834 | static bool nested_vmcb_checks(struct vmcb *vmcb) | |
2835 | { | |
2836 | if ((vmcb->control.intercept & (1ULL << INTERCEPT_VMRUN)) == 0) | |
2837 | return false; | |
2838 | ||
2839 | if (vmcb->control.asid == 0) | |
2840 | return false; | |
2841 | ||
2842 | if (vmcb->control.nested_ctl && !npt_enabled) | |
2843 | return false; | |
2844 | ||
2845 | return true; | |
2846 | } | |
2847 | ||
2848 | static bool nested_svm_vmrun(struct vcpu_svm *svm) | |
2849 | { | |
2850 | struct vmcb *nested_vmcb; | |
2851 | struct vmcb *hsave = svm->nested.hsave; | |
2852 | struct vmcb *vmcb = svm->vmcb; | |
2853 | struct page *page; | |
2854 | u64 vmcb_gpa; | |
2855 | ||
2856 | vmcb_gpa = svm->vmcb->save.rax; | |
2857 | ||
2858 | nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page); | |
2859 | if (!nested_vmcb) | |
2860 | return false; | |
2861 | ||
2862 | if (!nested_vmcb_checks(nested_vmcb)) { | |
2863 | nested_vmcb->control.exit_code = SVM_EXIT_ERR; | |
2864 | nested_vmcb->control.exit_code_hi = 0; | |
2865 | nested_vmcb->control.exit_info_1 = 0; | |
2866 | nested_vmcb->control.exit_info_2 = 0; | |
2867 | ||
2868 | nested_svm_unmap(page); | |
2869 | ||
2870 | return false; | |
2871 | } | |
2872 | ||
2873 | trace_kvm_nested_vmrun(svm->vmcb->save.rip, vmcb_gpa, | |
2874 | nested_vmcb->save.rip, | |
2875 | nested_vmcb->control.int_ctl, | |
2876 | nested_vmcb->control.event_inj, | |
2877 | nested_vmcb->control.nested_ctl); | |
2878 | ||
2879 | trace_kvm_nested_intercepts(nested_vmcb->control.intercept_cr & 0xffff, | |
2880 | nested_vmcb->control.intercept_cr >> 16, | |
2881 | nested_vmcb->control.intercept_exceptions, | |
2882 | nested_vmcb->control.intercept); | |
2883 | ||
2884 | /* Clear internal status */ | |
2885 | kvm_clear_exception_queue(&svm->vcpu); | |
2886 | kvm_clear_interrupt_queue(&svm->vcpu); | |
2887 | ||
2888 | /* | |
2889 | * Save the old vmcb, so we don't need to pick what we save, but can | |
2890 | * restore everything when a VMEXIT occurs | |
2891 | */ | |
2892 | hsave->save.es = vmcb->save.es; | |
2893 | hsave->save.cs = vmcb->save.cs; | |
2894 | hsave->save.ss = vmcb->save.ss; | |
2895 | hsave->save.ds = vmcb->save.ds; | |
2896 | hsave->save.gdtr = vmcb->save.gdtr; | |
2897 | hsave->save.idtr = vmcb->save.idtr; | |
2898 | hsave->save.efer = svm->vcpu.arch.efer; | |
2899 | hsave->save.cr0 = kvm_read_cr0(&svm->vcpu); | |
2900 | hsave->save.cr4 = svm->vcpu.arch.cr4; | |
2901 | hsave->save.rflags = kvm_get_rflags(&svm->vcpu); | |
2902 | hsave->save.rip = kvm_rip_read(&svm->vcpu); | |
2903 | hsave->save.rsp = vmcb->save.rsp; | |
2904 | hsave->save.rax = vmcb->save.rax; | |
2905 | if (npt_enabled) | |
2906 | hsave->save.cr3 = vmcb->save.cr3; | |
2907 | else | |
2908 | hsave->save.cr3 = kvm_read_cr3(&svm->vcpu); | |
2909 | ||
2910 | copy_vmcb_control_area(hsave, vmcb); | |
2911 | ||
2912 | if (kvm_get_rflags(&svm->vcpu) & X86_EFLAGS_IF) | |
2913 | svm->vcpu.arch.hflags |= HF_HIF_MASK; | |
2914 | else | |
2915 | svm->vcpu.arch.hflags &= ~HF_HIF_MASK; | |
2916 | ||
2917 | if (nested_vmcb->control.nested_ctl) { | |
2918 | kvm_mmu_unload(&svm->vcpu); | |
2919 | svm->nested.nested_cr3 = nested_vmcb->control.nested_cr3; | |
2920 | nested_svm_init_mmu_context(&svm->vcpu); | |
2921 | } | |
2922 | ||
2923 | /* Load the nested guest state */ | |
2924 | svm->vmcb->save.es = nested_vmcb->save.es; | |
2925 | svm->vmcb->save.cs = nested_vmcb->save.cs; | |
2926 | svm->vmcb->save.ss = nested_vmcb->save.ss; | |
2927 | svm->vmcb->save.ds = nested_vmcb->save.ds; | |
2928 | svm->vmcb->save.gdtr = nested_vmcb->save.gdtr; | |
2929 | svm->vmcb->save.idtr = nested_vmcb->save.idtr; | |
2930 | kvm_set_rflags(&svm->vcpu, nested_vmcb->save.rflags); | |
2931 | svm_set_efer(&svm->vcpu, nested_vmcb->save.efer); | |
2932 | svm_set_cr0(&svm->vcpu, nested_vmcb->save.cr0); | |
2933 | svm_set_cr4(&svm->vcpu, nested_vmcb->save.cr4); | |
2934 | if (npt_enabled) { | |
2935 | svm->vmcb->save.cr3 = nested_vmcb->save.cr3; | |
2936 | svm->vcpu.arch.cr3 = nested_vmcb->save.cr3; | |
2937 | } else | |
2938 | (void)kvm_set_cr3(&svm->vcpu, nested_vmcb->save.cr3); | |
2939 | ||
2940 | /* Guest paging mode is active - reset mmu */ | |
2941 | kvm_mmu_reset_context(&svm->vcpu); | |
2942 | ||
2943 | svm->vmcb->save.cr2 = svm->vcpu.arch.cr2 = nested_vmcb->save.cr2; | |
2944 | kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, nested_vmcb->save.rax); | |
2945 | kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, nested_vmcb->save.rsp); | |
2946 | kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, nested_vmcb->save.rip); | |
2947 | ||
2948 | /* In case we don't even reach vcpu_run, the fields are not updated */ | |
2949 | svm->vmcb->save.rax = nested_vmcb->save.rax; | |
2950 | svm->vmcb->save.rsp = nested_vmcb->save.rsp; | |
2951 | svm->vmcb->save.rip = nested_vmcb->save.rip; | |
2952 | svm->vmcb->save.dr7 = nested_vmcb->save.dr7; | |
2953 | svm->vmcb->save.dr6 = nested_vmcb->save.dr6; | |
2954 | svm->vmcb->save.cpl = nested_vmcb->save.cpl; | |
2955 | ||
2956 | svm->nested.vmcb_msrpm = nested_vmcb->control.msrpm_base_pa & ~0x0fffULL; | |
2957 | svm->nested.vmcb_iopm = nested_vmcb->control.iopm_base_pa & ~0x0fffULL; | |
2958 | ||
2959 | /* cache intercepts */ | |
2960 | svm->nested.intercept_cr = nested_vmcb->control.intercept_cr; | |
2961 | svm->nested.intercept_dr = nested_vmcb->control.intercept_dr; | |
2962 | svm->nested.intercept_exceptions = nested_vmcb->control.intercept_exceptions; | |
2963 | svm->nested.intercept = nested_vmcb->control.intercept; | |
2964 | ||
2965 | svm_flush_tlb(&svm->vcpu); | |
2966 | svm->vmcb->control.int_ctl = nested_vmcb->control.int_ctl | V_INTR_MASKING_MASK; | |
2967 | if (nested_vmcb->control.int_ctl & V_INTR_MASKING_MASK) | |
2968 | svm->vcpu.arch.hflags |= HF_VINTR_MASK; | |
2969 | else | |
2970 | svm->vcpu.arch.hflags &= ~HF_VINTR_MASK; | |
2971 | ||
2972 | if (svm->vcpu.arch.hflags & HF_VINTR_MASK) { | |
2973 | /* We only want the cr8 intercept bits of the guest */ | |
2974 | clr_cr_intercept(svm, INTERCEPT_CR8_READ); | |
2975 | clr_cr_intercept(svm, INTERCEPT_CR8_WRITE); | |
2976 | } | |
2977 | ||
2978 | /* We don't want to see VMMCALLs from a nested guest */ | |
2979 | clr_intercept(svm, INTERCEPT_VMMCALL); | |
2980 | ||
2981 | svm->vmcb->control.lbr_ctl = nested_vmcb->control.lbr_ctl; | |
2982 | svm->vmcb->control.int_vector = nested_vmcb->control.int_vector; | |
2983 | svm->vmcb->control.int_state = nested_vmcb->control.int_state; | |
2984 | svm->vmcb->control.tsc_offset += nested_vmcb->control.tsc_offset; | |
2985 | svm->vmcb->control.event_inj = nested_vmcb->control.event_inj; | |
2986 | svm->vmcb->control.event_inj_err = nested_vmcb->control.event_inj_err; | |
2987 | ||
2988 | nested_svm_unmap(page); | |
2989 | ||
2990 | /* Enter Guest-Mode */ | |
2991 | enter_guest_mode(&svm->vcpu); | |
2992 | ||
2993 | /* | |
2994 | * Merge guest and host intercepts - must be called with vcpu in | |
2995 | * guest-mode to take affect here | |
2996 | */ | |
2997 | recalc_intercepts(svm); | |
2998 | ||
2999 | svm->nested.vmcb = vmcb_gpa; | |
3000 | ||
3001 | enable_gif(svm); | |
3002 | ||
3003 | mark_all_dirty(svm->vmcb); | |
3004 | ||
3005 | return true; | |
3006 | } | |
3007 | ||
3008 | static void nested_svm_vmloadsave(struct vmcb *from_vmcb, struct vmcb *to_vmcb) | |
3009 | { | |
3010 | to_vmcb->save.fs = from_vmcb->save.fs; | |
3011 | to_vmcb->save.gs = from_vmcb->save.gs; | |
3012 | to_vmcb->save.tr = from_vmcb->save.tr; | |
3013 | to_vmcb->save.ldtr = from_vmcb->save.ldtr; | |
3014 | to_vmcb->save.kernel_gs_base = from_vmcb->save.kernel_gs_base; | |
3015 | to_vmcb->save.star = from_vmcb->save.star; | |
3016 | to_vmcb->save.lstar = from_vmcb->save.lstar; | |
3017 | to_vmcb->save.cstar = from_vmcb->save.cstar; | |
3018 | to_vmcb->save.sfmask = from_vmcb->save.sfmask; | |
3019 | to_vmcb->save.sysenter_cs = from_vmcb->save.sysenter_cs; | |
3020 | to_vmcb->save.sysenter_esp = from_vmcb->save.sysenter_esp; | |
3021 | to_vmcb->save.sysenter_eip = from_vmcb->save.sysenter_eip; | |
3022 | } | |
3023 | ||
3024 | static int vmload_interception(struct vcpu_svm *svm) | |
3025 | { | |
3026 | struct vmcb *nested_vmcb; | |
3027 | struct page *page; | |
3028 | ||
3029 | if (nested_svm_check_permissions(svm)) | |
3030 | return 1; | |
3031 | ||
3032 | nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page); | |
3033 | if (!nested_vmcb) | |
3034 | return 1; | |
3035 | ||
3036 | svm->next_rip = kvm_rip_read(&svm->vcpu) + 3; | |
3037 | skip_emulated_instruction(&svm->vcpu); | |
3038 | ||
3039 | nested_svm_vmloadsave(nested_vmcb, svm->vmcb); | |
3040 | nested_svm_unmap(page); | |
3041 | ||
3042 | return 1; | |
3043 | } | |
3044 | ||
3045 | static int vmsave_interception(struct vcpu_svm *svm) | |
3046 | { | |
3047 | struct vmcb *nested_vmcb; | |
3048 | struct page *page; | |
3049 | ||
3050 | if (nested_svm_check_permissions(svm)) | |
3051 | return 1; | |
3052 | ||
3053 | nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page); | |
3054 | if (!nested_vmcb) | |
3055 | return 1; | |
3056 | ||
3057 | svm->next_rip = kvm_rip_read(&svm->vcpu) + 3; | |
3058 | skip_emulated_instruction(&svm->vcpu); | |
3059 | ||
3060 | nested_svm_vmloadsave(svm->vmcb, nested_vmcb); | |
3061 | nested_svm_unmap(page); | |
3062 | ||
3063 | return 1; | |
3064 | } | |
3065 | ||
3066 | static int vmrun_interception(struct vcpu_svm *svm) | |
3067 | { | |
3068 | if (nested_svm_check_permissions(svm)) | |
3069 | return 1; | |
3070 | ||
3071 | /* Save rip after vmrun instruction */ | |
3072 | kvm_rip_write(&svm->vcpu, kvm_rip_read(&svm->vcpu) + 3); | |
3073 | ||
3074 | if (!nested_svm_vmrun(svm)) | |
3075 | return 1; | |
3076 | ||
3077 | if (!nested_svm_vmrun_msrpm(svm)) | |
3078 | goto failed; | |
3079 | ||
3080 | return 1; | |
3081 | ||
3082 | failed: | |
3083 | ||
3084 | svm->vmcb->control.exit_code = SVM_EXIT_ERR; | |
3085 | svm->vmcb->control.exit_code_hi = 0; | |
3086 | svm->vmcb->control.exit_info_1 = 0; | |
3087 | svm->vmcb->control.exit_info_2 = 0; | |
3088 | ||
3089 | nested_svm_vmexit(svm); | |
3090 | ||
3091 | return 1; | |
3092 | } | |
3093 | ||
3094 | static int stgi_interception(struct vcpu_svm *svm) | |
3095 | { | |
3096 | if (nested_svm_check_permissions(svm)) | |
3097 | return 1; | |
3098 | ||
3099 | svm->next_rip = kvm_rip_read(&svm->vcpu) + 3; | |
3100 | skip_emulated_instruction(&svm->vcpu); | |
3101 | kvm_make_request(KVM_REQ_EVENT, &svm->vcpu); | |
3102 | ||
3103 | enable_gif(svm); | |
3104 | ||
3105 | return 1; | |
3106 | } | |
3107 | ||
3108 | static int clgi_interception(struct vcpu_svm *svm) | |
3109 | { | |
3110 | if (nested_svm_check_permissions(svm)) | |
3111 | return 1; | |
3112 | ||
3113 | svm->next_rip = kvm_rip_read(&svm->vcpu) + 3; | |
3114 | skip_emulated_instruction(&svm->vcpu); | |
3115 | ||
3116 | disable_gif(svm); | |
3117 | ||
3118 | /* After a CLGI no interrupts should come */ | |
3119 | if (!kvm_vcpu_apicv_active(&svm->vcpu)) { | |
3120 | svm_clear_vintr(svm); | |
3121 | svm->vmcb->control.int_ctl &= ~V_IRQ_MASK; | |
3122 | mark_dirty(svm->vmcb, VMCB_INTR); | |
3123 | } | |
3124 | ||
3125 | return 1; | |
3126 | } | |
3127 | ||
3128 | static int invlpga_interception(struct vcpu_svm *svm) | |
3129 | { | |
3130 | struct kvm_vcpu *vcpu = &svm->vcpu; | |
3131 | ||
3132 | trace_kvm_invlpga(svm->vmcb->save.rip, kvm_register_read(&svm->vcpu, VCPU_REGS_RCX), | |
3133 | kvm_register_read(&svm->vcpu, VCPU_REGS_RAX)); | |
3134 | ||
3135 | /* Let's treat INVLPGA the same as INVLPG (can be optimized!) */ | |
3136 | kvm_mmu_invlpg(vcpu, kvm_register_read(&svm->vcpu, VCPU_REGS_RAX)); | |
3137 | ||
3138 | svm->next_rip = kvm_rip_read(&svm->vcpu) + 3; | |
3139 | skip_emulated_instruction(&svm->vcpu); | |
3140 | return 1; | |
3141 | } | |
3142 | ||
3143 | static int skinit_interception(struct vcpu_svm *svm) | |
3144 | { | |
3145 | trace_kvm_skinit(svm->vmcb->save.rip, kvm_register_read(&svm->vcpu, VCPU_REGS_RAX)); | |
3146 | ||
3147 | kvm_queue_exception(&svm->vcpu, UD_VECTOR); | |
3148 | return 1; | |
3149 | } | |
3150 | ||
3151 | static int wbinvd_interception(struct vcpu_svm *svm) | |
3152 | { | |
3153 | kvm_emulate_wbinvd(&svm->vcpu); | |
3154 | return 1; | |
3155 | } | |
3156 | ||
3157 | static int xsetbv_interception(struct vcpu_svm *svm) | |
3158 | { | |
3159 | u64 new_bv = kvm_read_edx_eax(&svm->vcpu); | |
3160 | u32 index = kvm_register_read(&svm->vcpu, VCPU_REGS_RCX); | |
3161 | ||
3162 | if (kvm_set_xcr(&svm->vcpu, index, new_bv) == 0) { | |
3163 | svm->next_rip = kvm_rip_read(&svm->vcpu) + 3; | |
3164 | skip_emulated_instruction(&svm->vcpu); | |
3165 | } | |
3166 | ||
3167 | return 1; | |
3168 | } | |
3169 | ||
3170 | static int task_switch_interception(struct vcpu_svm *svm) | |
3171 | { | |
3172 | u16 tss_selector; | |
3173 | int reason; | |
3174 | int int_type = svm->vmcb->control.exit_int_info & | |
3175 | SVM_EXITINTINFO_TYPE_MASK; | |
3176 | int int_vec = svm->vmcb->control.exit_int_info & SVM_EVTINJ_VEC_MASK; | |
3177 | uint32_t type = | |
3178 | svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_TYPE_MASK; | |
3179 | uint32_t idt_v = | |
3180 | svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_VALID; | |
3181 | bool has_error_code = false; | |
3182 | u32 error_code = 0; | |
3183 | ||
3184 | tss_selector = (u16)svm->vmcb->control.exit_info_1; | |
3185 | ||
3186 | if (svm->vmcb->control.exit_info_2 & | |
3187 | (1ULL << SVM_EXITINFOSHIFT_TS_REASON_IRET)) | |
3188 | reason = TASK_SWITCH_IRET; | |
3189 | else if (svm->vmcb->control.exit_info_2 & | |
3190 | (1ULL << SVM_EXITINFOSHIFT_TS_REASON_JMP)) | |
3191 | reason = TASK_SWITCH_JMP; | |
3192 | else if (idt_v) | |
3193 | reason = TASK_SWITCH_GATE; | |
3194 | else | |
3195 | reason = TASK_SWITCH_CALL; | |
3196 | ||
3197 | if (reason == TASK_SWITCH_GATE) { | |
3198 | switch (type) { | |
3199 | case SVM_EXITINTINFO_TYPE_NMI: | |
3200 | svm->vcpu.arch.nmi_injected = false; | |
3201 | break; | |
3202 | case SVM_EXITINTINFO_TYPE_EXEPT: | |
3203 | if (svm->vmcb->control.exit_info_2 & | |
3204 | (1ULL << SVM_EXITINFOSHIFT_TS_HAS_ERROR_CODE)) { | |
3205 | has_error_code = true; | |
3206 | error_code = | |
3207 | (u32)svm->vmcb->control.exit_info_2; | |
3208 | } | |
3209 | kvm_clear_exception_queue(&svm->vcpu); | |
3210 | break; | |
3211 | case SVM_EXITINTINFO_TYPE_INTR: | |
3212 | kvm_clear_interrupt_queue(&svm->vcpu); | |
3213 | break; | |
3214 | default: | |
3215 | break; | |
3216 | } | |
3217 | } | |
3218 | ||
3219 | if (reason != TASK_SWITCH_GATE || | |
3220 | int_type == SVM_EXITINTINFO_TYPE_SOFT || | |
3221 | (int_type == SVM_EXITINTINFO_TYPE_EXEPT && | |
3222 | (int_vec == OF_VECTOR || int_vec == BP_VECTOR))) | |
3223 | skip_emulated_instruction(&svm->vcpu); | |
3224 | ||
3225 | if (int_type != SVM_EXITINTINFO_TYPE_SOFT) | |
3226 | int_vec = -1; | |
3227 | ||
3228 | if (kvm_task_switch(&svm->vcpu, tss_selector, int_vec, reason, | |
3229 | has_error_code, error_code) == EMULATE_FAIL) { | |
3230 | svm->vcpu.run->exit_reason = KVM_EXIT_INTERNAL_ERROR; | |
3231 | svm->vcpu.run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION; | |
3232 | svm->vcpu.run->internal.ndata = 0; | |
3233 | return 0; | |
3234 | } | |
3235 | return 1; | |
3236 | } | |
3237 | ||
3238 | static int cpuid_interception(struct vcpu_svm *svm) | |
3239 | { | |
3240 | svm->next_rip = kvm_rip_read(&svm->vcpu) + 2; | |
3241 | kvm_emulate_cpuid(&svm->vcpu); | |
3242 | return 1; | |
3243 | } | |
3244 | ||
3245 | static int iret_interception(struct vcpu_svm *svm) | |
3246 | { | |
3247 | ++svm->vcpu.stat.nmi_window_exits; | |
3248 | clr_intercept(svm, INTERCEPT_IRET); | |
3249 | svm->vcpu.arch.hflags |= HF_IRET_MASK; | |
3250 | svm->nmi_iret_rip = kvm_rip_read(&svm->vcpu); | |
3251 | kvm_make_request(KVM_REQ_EVENT, &svm->vcpu); | |
3252 | return 1; | |
3253 | } | |
3254 | ||
3255 | static int invlpg_interception(struct vcpu_svm *svm) | |
3256 | { | |
3257 | if (!static_cpu_has(X86_FEATURE_DECODEASSISTS)) | |
3258 | return emulate_instruction(&svm->vcpu, 0) == EMULATE_DONE; | |
3259 | ||
3260 | kvm_mmu_invlpg(&svm->vcpu, svm->vmcb->control.exit_info_1); | |
3261 | skip_emulated_instruction(&svm->vcpu); | |
3262 | return 1; | |
3263 | } | |
3264 | ||
3265 | static int emulate_on_interception(struct vcpu_svm *svm) | |
3266 | { | |
3267 | return emulate_instruction(&svm->vcpu, 0) == EMULATE_DONE; | |
3268 | } | |
3269 | ||
3270 | static int rdpmc_interception(struct vcpu_svm *svm) | |
3271 | { | |
3272 | int err; | |
3273 | ||
3274 | if (!static_cpu_has(X86_FEATURE_NRIPS)) | |
3275 | return emulate_on_interception(svm); | |
3276 | ||
3277 | err = kvm_rdpmc(&svm->vcpu); | |
3278 | kvm_complete_insn_gp(&svm->vcpu, err); | |
3279 | ||
3280 | return 1; | |
3281 | } | |
3282 | ||
3283 | static bool check_selective_cr0_intercepted(struct vcpu_svm *svm, | |
3284 | unsigned long val) | |
3285 | { | |
3286 | unsigned long cr0 = svm->vcpu.arch.cr0; | |
3287 | bool ret = false; | |
3288 | u64 intercept; | |
3289 | ||
3290 | intercept = svm->nested.intercept; | |
3291 | ||
3292 | if (!is_guest_mode(&svm->vcpu) || | |
3293 | (!(intercept & (1ULL << INTERCEPT_SELECTIVE_CR0)))) | |
3294 | return false; | |
3295 | ||
3296 | cr0 &= ~SVM_CR0_SELECTIVE_MASK; | |
3297 | val &= ~SVM_CR0_SELECTIVE_MASK; | |
3298 | ||
3299 | if (cr0 ^ val) { | |
3300 | svm->vmcb->control.exit_code = SVM_EXIT_CR0_SEL_WRITE; | |
3301 | ret = (nested_svm_exit_handled(svm) == NESTED_EXIT_DONE); | |
3302 | } | |
3303 | ||
3304 | return ret; | |
3305 | } | |
3306 | ||
3307 | #define CR_VALID (1ULL << 63) | |
3308 | ||
3309 | static int cr_interception(struct vcpu_svm *svm) | |
3310 | { | |
3311 | int reg, cr; | |
3312 | unsigned long val; | |
3313 | int err; | |
3314 | ||
3315 | if (!static_cpu_has(X86_FEATURE_DECODEASSISTS)) | |
3316 | return emulate_on_interception(svm); | |
3317 | ||
3318 | if (unlikely((svm->vmcb->control.exit_info_1 & CR_VALID) == 0)) | |
3319 | return emulate_on_interception(svm); | |
3320 | ||
3321 | reg = svm->vmcb->control.exit_info_1 & SVM_EXITINFO_REG_MASK; | |
3322 | if (svm->vmcb->control.exit_code == SVM_EXIT_CR0_SEL_WRITE) | |
3323 | cr = SVM_EXIT_WRITE_CR0 - SVM_EXIT_READ_CR0; | |
3324 | else | |
3325 | cr = svm->vmcb->control.exit_code - SVM_EXIT_READ_CR0; | |
3326 | ||
3327 | err = 0; | |
3328 | if (cr >= 16) { /* mov to cr */ | |
3329 | cr -= 16; | |
3330 | val = kvm_register_read(&svm->vcpu, reg); | |
3331 | switch (cr) { | |
3332 | case 0: | |
3333 | if (!check_selective_cr0_intercepted(svm, val)) | |
3334 | err = kvm_set_cr0(&svm->vcpu, val); | |
3335 | else | |
3336 | return 1; | |
3337 | ||
3338 | break; | |
3339 | case 3: | |
3340 | err = kvm_set_cr3(&svm->vcpu, val); | |
3341 | break; | |
3342 | case 4: | |
3343 | err = kvm_set_cr4(&svm->vcpu, val); | |
3344 | break; | |
3345 | case 8: | |
3346 | err = kvm_set_cr8(&svm->vcpu, val); | |
3347 | break; | |
3348 | default: | |
3349 | WARN(1, "unhandled write to CR%d", cr); | |
3350 | kvm_queue_exception(&svm->vcpu, UD_VECTOR); | |
3351 | return 1; | |
3352 | } | |
3353 | } else { /* mov from cr */ | |
3354 | switch (cr) { | |
3355 | case 0: | |
3356 | val = kvm_read_cr0(&svm->vcpu); | |
3357 | break; | |
3358 | case 2: | |
3359 | val = svm->vcpu.arch.cr2; | |
3360 | break; | |
3361 | case 3: | |
3362 | val = kvm_read_cr3(&svm->vcpu); | |
3363 | break; | |
3364 | case 4: | |
3365 | val = kvm_read_cr4(&svm->vcpu); | |
3366 | break; | |
3367 | case 8: | |
3368 | val = kvm_get_cr8(&svm->vcpu); | |
3369 | break; | |
3370 | default: | |
3371 | WARN(1, "unhandled read from CR%d", cr); | |
3372 | kvm_queue_exception(&svm->vcpu, UD_VECTOR); | |
3373 | return 1; | |
3374 | } | |
3375 | kvm_register_write(&svm->vcpu, reg, val); | |
3376 | } | |
3377 | kvm_complete_insn_gp(&svm->vcpu, err); | |
3378 | ||
3379 | return 1; | |
3380 | } | |
3381 | ||
3382 | static int dr_interception(struct vcpu_svm *svm) | |
3383 | { | |
3384 | int reg, dr; | |
3385 | unsigned long val; | |
3386 | ||
3387 | if (svm->vcpu.guest_debug == 0) { | |
3388 | /* | |
3389 | * No more DR vmexits; force a reload of the debug registers | |
3390 | * and reenter on this instruction. The next vmexit will | |
3391 | * retrieve the full state of the debug registers. | |
3392 | */ | |
3393 | clr_dr_intercepts(svm); | |
3394 | svm->vcpu.arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT; | |
3395 | return 1; | |
3396 | } | |
3397 | ||
3398 | if (!boot_cpu_has(X86_FEATURE_DECODEASSISTS)) | |
3399 | return emulate_on_interception(svm); | |
3400 | ||
3401 | reg = svm->vmcb->control.exit_info_1 & SVM_EXITINFO_REG_MASK; | |
3402 | dr = svm->vmcb->control.exit_code - SVM_EXIT_READ_DR0; | |
3403 | ||
3404 | if (dr >= 16) { /* mov to DRn */ | |
3405 | if (!kvm_require_dr(&svm->vcpu, dr - 16)) | |
3406 | return 1; | |
3407 | val = kvm_register_read(&svm->vcpu, reg); | |
3408 | kvm_set_dr(&svm->vcpu, dr - 16, val); | |
3409 | } else { | |
3410 | if (!kvm_require_dr(&svm->vcpu, dr)) | |
3411 | return 1; | |
3412 | kvm_get_dr(&svm->vcpu, dr, &val); | |
3413 | kvm_register_write(&svm->vcpu, reg, val); | |
3414 | } | |
3415 | ||
3416 | skip_emulated_instruction(&svm->vcpu); | |
3417 | ||
3418 | return 1; | |
3419 | } | |
3420 | ||
3421 | static int cr8_write_interception(struct vcpu_svm *svm) | |
3422 | { | |
3423 | struct kvm_run *kvm_run = svm->vcpu.run; | |
3424 | int r; | |
3425 | ||
3426 | u8 cr8_prev = kvm_get_cr8(&svm->vcpu); | |
3427 | /* instruction emulation calls kvm_set_cr8() */ | |
3428 | r = cr_interception(svm); | |
3429 | if (lapic_in_kernel(&svm->vcpu)) | |
3430 | return r; | |
3431 | if (cr8_prev <= kvm_get_cr8(&svm->vcpu)) | |
3432 | return r; | |
3433 | kvm_run->exit_reason = KVM_EXIT_SET_TPR; | |
3434 | return 0; | |
3435 | } | |
3436 | ||
3437 | static int svm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info) | |
3438 | { | |
3439 | struct vcpu_svm *svm = to_svm(vcpu); | |
3440 | ||
3441 | switch (msr_info->index) { | |
3442 | case MSR_IA32_TSC: { | |
3443 | msr_info->data = svm->vmcb->control.tsc_offset + | |
3444 | kvm_scale_tsc(vcpu, rdtsc()); | |
3445 | ||
3446 | break; | |
3447 | } | |
3448 | case MSR_STAR: | |
3449 | msr_info->data = svm->vmcb->save.star; | |
3450 | break; | |
3451 | #ifdef CONFIG_X86_64 | |
3452 | case MSR_LSTAR: | |
3453 | msr_info->data = svm->vmcb->save.lstar; | |
3454 | break; | |
3455 | case MSR_CSTAR: | |
3456 | msr_info->data = svm->vmcb->save.cstar; | |
3457 | break; | |
3458 | case MSR_KERNEL_GS_BASE: | |
3459 | msr_info->data = svm->vmcb->save.kernel_gs_base; | |
3460 | break; | |
3461 | case MSR_SYSCALL_MASK: | |
3462 | msr_info->data = svm->vmcb->save.sfmask; | |
3463 | break; | |
3464 | #endif | |
3465 | case MSR_IA32_SYSENTER_CS: | |
3466 | msr_info->data = svm->vmcb->save.sysenter_cs; | |
3467 | break; | |
3468 | case MSR_IA32_SYSENTER_EIP: | |
3469 | msr_info->data = svm->sysenter_eip; | |
3470 | break; | |
3471 | case MSR_IA32_SYSENTER_ESP: | |
3472 | msr_info->data = svm->sysenter_esp; | |
3473 | break; | |
3474 | case MSR_TSC_AUX: | |
3475 | if (!boot_cpu_has(X86_FEATURE_RDTSCP)) | |
3476 | return 1; | |
3477 | msr_info->data = svm->tsc_aux; | |
3478 | break; | |
3479 | /* | |
3480 | * Nobody will change the following 5 values in the VMCB so we can | |
3481 | * safely return them on rdmsr. They will always be 0 until LBRV is | |
3482 | * implemented. | |
3483 | */ | |
3484 | case MSR_IA32_DEBUGCTLMSR: | |
3485 | msr_info->data = svm->vmcb->save.dbgctl; | |
3486 | break; | |
3487 | case MSR_IA32_LASTBRANCHFROMIP: | |
3488 | msr_info->data = svm->vmcb->save.br_from; | |
3489 | break; | |
3490 | case MSR_IA32_LASTBRANCHTOIP: | |
3491 | msr_info->data = svm->vmcb->save.br_to; | |
3492 | break; | |
3493 | case MSR_IA32_LASTINTFROMIP: | |
3494 | msr_info->data = svm->vmcb->save.last_excp_from; | |
3495 | break; | |
3496 | case MSR_IA32_LASTINTTOIP: | |
3497 | msr_info->data = svm->vmcb->save.last_excp_to; | |
3498 | break; | |
3499 | case MSR_VM_HSAVE_PA: | |
3500 | msr_info->data = svm->nested.hsave_msr; | |
3501 | break; | |
3502 | case MSR_VM_CR: | |
3503 | msr_info->data = svm->nested.vm_cr_msr; | |
3504 | break; | |
3505 | case MSR_IA32_UCODE_REV: | |
3506 | msr_info->data = 0x01000065; | |
3507 | break; | |
3508 | case MSR_F15H_IC_CFG: { | |
3509 | ||
3510 | int family, model; | |
3511 | ||
3512 | family = guest_cpuid_family(vcpu); | |
3513 | model = guest_cpuid_model(vcpu); | |
3514 | ||
3515 | if (family < 0 || model < 0) | |
3516 | return kvm_get_msr_common(vcpu, msr_info); | |
3517 | ||
3518 | msr_info->data = 0; | |
3519 | ||
3520 | if (family == 0x15 && | |
3521 | (model >= 0x2 && model < 0x20)) | |
3522 | msr_info->data = 0x1E; | |
3523 | } | |
3524 | break; | |
3525 | default: | |
3526 | return kvm_get_msr_common(vcpu, msr_info); | |
3527 | } | |
3528 | return 0; | |
3529 | } | |
3530 | ||
3531 | static int rdmsr_interception(struct vcpu_svm *svm) | |
3532 | { | |
3533 | u32 ecx = kvm_register_read(&svm->vcpu, VCPU_REGS_RCX); | |
3534 | struct msr_data msr_info; | |
3535 | ||
3536 | msr_info.index = ecx; | |
3537 | msr_info.host_initiated = false; | |
3538 | if (svm_get_msr(&svm->vcpu, &msr_info)) { | |
3539 | trace_kvm_msr_read_ex(ecx); | |
3540 | kvm_inject_gp(&svm->vcpu, 0); | |
3541 | } else { | |
3542 | trace_kvm_msr_read(ecx, msr_info.data); | |
3543 | ||
3544 | kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, | |
3545 | msr_info.data & 0xffffffff); | |
3546 | kvm_register_write(&svm->vcpu, VCPU_REGS_RDX, | |
3547 | msr_info.data >> 32); | |
3548 | svm->next_rip = kvm_rip_read(&svm->vcpu) + 2; | |
3549 | skip_emulated_instruction(&svm->vcpu); | |
3550 | } | |
3551 | return 1; | |
3552 | } | |
3553 | ||
3554 | static int svm_set_vm_cr(struct kvm_vcpu *vcpu, u64 data) | |
3555 | { | |
3556 | struct vcpu_svm *svm = to_svm(vcpu); | |
3557 | int svm_dis, chg_mask; | |
3558 | ||
3559 | if (data & ~SVM_VM_CR_VALID_MASK) | |
3560 | return 1; | |
3561 | ||
3562 | chg_mask = SVM_VM_CR_VALID_MASK; | |
3563 | ||
3564 | if (svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK) | |
3565 | chg_mask &= ~(SVM_VM_CR_SVM_LOCK_MASK | SVM_VM_CR_SVM_DIS_MASK); | |
3566 | ||
3567 | svm->nested.vm_cr_msr &= ~chg_mask; | |
3568 | svm->nested.vm_cr_msr |= (data & chg_mask); | |
3569 | ||
3570 | svm_dis = svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK; | |
3571 | ||
3572 | /* check for svm_disable while efer.svme is set */ | |
3573 | if (svm_dis && (vcpu->arch.efer & EFER_SVME)) | |
3574 | return 1; | |
3575 | ||
3576 | return 0; | |
3577 | } | |
3578 | ||
3579 | static int svm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr) | |
3580 | { | |
3581 | struct vcpu_svm *svm = to_svm(vcpu); | |
3582 | ||
3583 | u32 ecx = msr->index; | |
3584 | u64 data = msr->data; | |
3585 | switch (ecx) { | |
3586 | case MSR_IA32_TSC: | |
3587 | kvm_write_tsc(vcpu, msr); | |
3588 | break; | |
3589 | case MSR_STAR: | |
3590 | svm->vmcb->save.star = data; | |
3591 | break; | |
3592 | #ifdef CONFIG_X86_64 | |
3593 | case MSR_LSTAR: | |
3594 | svm->vmcb->save.lstar = data; | |
3595 | break; | |
3596 | case MSR_CSTAR: | |
3597 | svm->vmcb->save.cstar = data; | |
3598 | break; | |
3599 | case MSR_KERNEL_GS_BASE: | |
3600 | svm->vmcb->save.kernel_gs_base = data; | |
3601 | break; | |
3602 | case MSR_SYSCALL_MASK: | |
3603 | svm->vmcb->save.sfmask = data; | |
3604 | break; | |
3605 | #endif | |
3606 | case MSR_IA32_SYSENTER_CS: | |
3607 | svm->vmcb->save.sysenter_cs = data; | |
3608 | break; | |
3609 | case MSR_IA32_SYSENTER_EIP: | |
3610 | svm->sysenter_eip = data; | |
3611 | svm->vmcb->save.sysenter_eip = data; | |
3612 | break; | |
3613 | case MSR_IA32_SYSENTER_ESP: | |
3614 | svm->sysenter_esp = data; | |
3615 | svm->vmcb->save.sysenter_esp = data; | |
3616 | break; | |
3617 | case MSR_TSC_AUX: | |
3618 | if (!boot_cpu_has(X86_FEATURE_RDTSCP)) | |
3619 | return 1; | |
3620 | ||
3621 | /* | |
3622 | * This is rare, so we update the MSR here instead of using | |
3623 | * direct_access_msrs. Doing that would require a rdmsr in | |
3624 | * svm_vcpu_put. | |
3625 | */ | |
3626 | svm->tsc_aux = data; | |
3627 | wrmsrl(MSR_TSC_AUX, svm->tsc_aux); | |
3628 | break; | |
3629 | case MSR_IA32_DEBUGCTLMSR: | |
3630 | if (!boot_cpu_has(X86_FEATURE_LBRV)) { | |
3631 | vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTL 0x%llx, nop\n", | |
3632 | __func__, data); | |
3633 | break; | |
3634 | } | |
3635 | if (data & DEBUGCTL_RESERVED_BITS) | |
3636 | return 1; | |
3637 | ||
3638 | svm->vmcb->save.dbgctl = data; | |
3639 | mark_dirty(svm->vmcb, VMCB_LBR); | |
3640 | if (data & (1ULL<<0)) | |
3641 | svm_enable_lbrv(svm); | |
3642 | else | |
3643 | svm_disable_lbrv(svm); | |
3644 | break; | |
3645 | case MSR_VM_HSAVE_PA: | |
3646 | svm->nested.hsave_msr = data; | |
3647 | break; | |
3648 | case MSR_VM_CR: | |
3649 | return svm_set_vm_cr(vcpu, data); | |
3650 | case MSR_VM_IGNNE: | |
3651 | vcpu_unimpl(vcpu, "unimplemented wrmsr: 0x%x data 0x%llx\n", ecx, data); | |
3652 | break; | |
3653 | case MSR_IA32_APICBASE: | |
3654 | if (kvm_vcpu_apicv_active(vcpu)) | |
3655 | avic_update_vapic_bar(to_svm(vcpu), data); | |
3656 | /* Follow through */ | |
3657 | default: | |
3658 | return kvm_set_msr_common(vcpu, msr); | |
3659 | } | |
3660 | return 0; | |
3661 | } | |
3662 | ||
3663 | static int wrmsr_interception(struct vcpu_svm *svm) | |
3664 | { | |
3665 | struct msr_data msr; | |
3666 | u32 ecx = kvm_register_read(&svm->vcpu, VCPU_REGS_RCX); | |
3667 | u64 data = kvm_read_edx_eax(&svm->vcpu); | |
3668 | ||
3669 | msr.data = data; | |
3670 | msr.index = ecx; | |
3671 | msr.host_initiated = false; | |
3672 | ||
3673 | svm->next_rip = kvm_rip_read(&svm->vcpu) + 2; | |
3674 | if (kvm_set_msr(&svm->vcpu, &msr)) { | |
3675 | trace_kvm_msr_write_ex(ecx, data); | |
3676 | kvm_inject_gp(&svm->vcpu, 0); | |
3677 | } else { | |
3678 | trace_kvm_msr_write(ecx, data); | |
3679 | skip_emulated_instruction(&svm->vcpu); | |
3680 | } | |
3681 | return 1; | |
3682 | } | |
3683 | ||
3684 | static int msr_interception(struct vcpu_svm *svm) | |
3685 | { | |
3686 | if (svm->vmcb->control.exit_info_1) | |
3687 | return wrmsr_interception(svm); | |
3688 | else | |
3689 | return rdmsr_interception(svm); | |
3690 | } | |
3691 | ||
3692 | static int interrupt_window_interception(struct vcpu_svm *svm) | |
3693 | { | |
3694 | kvm_make_request(KVM_REQ_EVENT, &svm->vcpu); | |
3695 | svm_clear_vintr(svm); | |
3696 | svm->vmcb->control.int_ctl &= ~V_IRQ_MASK; | |
3697 | mark_dirty(svm->vmcb, VMCB_INTR); | |
3698 | ++svm->vcpu.stat.irq_window_exits; | |
3699 | return 1; | |
3700 | } | |
3701 | ||
3702 | static int pause_interception(struct vcpu_svm *svm) | |
3703 | { | |
3704 | kvm_vcpu_on_spin(&(svm->vcpu)); | |
3705 | return 1; | |
3706 | } | |
3707 | ||
3708 | static int nop_interception(struct vcpu_svm *svm) | |
3709 | { | |
3710 | skip_emulated_instruction(&(svm->vcpu)); | |
3711 | return 1; | |
3712 | } | |
3713 | ||
3714 | static int monitor_interception(struct vcpu_svm *svm) | |
3715 | { | |
3716 | printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n"); | |
3717 | return nop_interception(svm); | |
3718 | } | |
3719 | ||
3720 | static int mwait_interception(struct vcpu_svm *svm) | |
3721 | { | |
3722 | printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n"); | |
3723 | return nop_interception(svm); | |
3724 | } | |
3725 | ||
3726 | enum avic_ipi_failure_cause { | |
3727 | AVIC_IPI_FAILURE_INVALID_INT_TYPE, | |
3728 | AVIC_IPI_FAILURE_TARGET_NOT_RUNNING, | |
3729 | AVIC_IPI_FAILURE_INVALID_TARGET, | |
3730 | AVIC_IPI_FAILURE_INVALID_BACKING_PAGE, | |
3731 | }; | |
3732 | ||
3733 | static int avic_incomplete_ipi_interception(struct vcpu_svm *svm) | |
3734 | { | |
3735 | u32 icrh = svm->vmcb->control.exit_info_1 >> 32; | |
3736 | u32 icrl = svm->vmcb->control.exit_info_1; | |
3737 | u32 id = svm->vmcb->control.exit_info_2 >> 32; | |
3738 | u32 index = svm->vmcb->control.exit_info_2 & 0xFF; | |
3739 | struct kvm_lapic *apic = svm->vcpu.arch.apic; | |
3740 | ||
3741 | trace_kvm_avic_incomplete_ipi(svm->vcpu.vcpu_id, icrh, icrl, id, index); | |
3742 | ||
3743 | switch (id) { | |
3744 | case AVIC_IPI_FAILURE_INVALID_INT_TYPE: | |
3745 | /* | |
3746 | * AVIC hardware handles the generation of | |
3747 | * IPIs when the specified Message Type is Fixed | |
3748 | * (also known as fixed delivery mode) and | |
3749 | * the Trigger Mode is edge-triggered. The hardware | |
3750 | * also supports self and broadcast delivery modes | |
3751 | * specified via the Destination Shorthand(DSH) | |
3752 | * field of the ICRL. Logical and physical APIC ID | |
3753 | * formats are supported. All other IPI types cause | |
3754 | * a #VMEXIT, which needs to emulated. | |
3755 | */ | |
3756 | kvm_lapic_reg_write(apic, APIC_ICR2, icrh); | |
3757 | kvm_lapic_reg_write(apic, APIC_ICR, icrl); | |
3758 | break; | |
3759 | case AVIC_IPI_FAILURE_TARGET_NOT_RUNNING: { | |
3760 | int i; | |
3761 | struct kvm_vcpu *vcpu; | |
3762 | struct kvm *kvm = svm->vcpu.kvm; | |
3763 | struct kvm_lapic *apic = svm->vcpu.arch.apic; | |
3764 | ||
3765 | /* | |
3766 | * At this point, we expect that the AVIC HW has already | |
3767 | * set the appropriate IRR bits on the valid target | |
3768 | * vcpus. So, we just need to kick the appropriate vcpu. | |
3769 | */ | |
3770 | kvm_for_each_vcpu(i, vcpu, kvm) { | |
3771 | bool m = kvm_apic_match_dest(vcpu, apic, | |
3772 | icrl & KVM_APIC_SHORT_MASK, | |
3773 | GET_APIC_DEST_FIELD(icrh), | |
3774 | icrl & KVM_APIC_DEST_MASK); | |
3775 | ||
3776 | if (m && !avic_vcpu_is_running(vcpu)) | |
3777 | kvm_vcpu_wake_up(vcpu); | |
3778 | } | |
3779 | break; | |
3780 | } | |
3781 | case AVIC_IPI_FAILURE_INVALID_TARGET: | |
3782 | break; | |
3783 | case AVIC_IPI_FAILURE_INVALID_BACKING_PAGE: | |
3784 | WARN_ONCE(1, "Invalid backing page\n"); | |
3785 | break; | |
3786 | default: | |
3787 | pr_err("Unknown IPI interception\n"); | |
3788 | } | |
3789 | ||
3790 | return 1; | |
3791 | } | |
3792 | ||
3793 | static u32 *avic_get_logical_id_entry(struct kvm_vcpu *vcpu, u32 ldr, bool flat) | |
3794 | { | |
3795 | struct kvm_arch *vm_data = &vcpu->kvm->arch; | |
3796 | int index; | |
3797 | u32 *logical_apic_id_table; | |
3798 | int dlid = GET_APIC_LOGICAL_ID(ldr); | |
3799 | ||
3800 | if (!dlid) | |
3801 | return NULL; | |
3802 | ||
3803 | if (flat) { /* flat */ | |
3804 | index = ffs(dlid) - 1; | |
3805 | if (index > 7) | |
3806 | return NULL; | |
3807 | } else { /* cluster */ | |
3808 | int cluster = (dlid & 0xf0) >> 4; | |
3809 | int apic = ffs(dlid & 0x0f) - 1; | |
3810 | ||
3811 | if ((apic < 0) || (apic > 7) || | |
3812 | (cluster >= 0xf)) | |
3813 | return NULL; | |
3814 | index = (cluster << 2) + apic; | |
3815 | } | |
3816 | ||
3817 | logical_apic_id_table = (u32 *) page_address(vm_data->avic_logical_id_table_page); | |
3818 | ||
3819 | return &logical_apic_id_table[index]; | |
3820 | } | |
3821 | ||
3822 | static int avic_ldr_write(struct kvm_vcpu *vcpu, u8 g_physical_id, u32 ldr, | |
3823 | bool valid) | |
3824 | { | |
3825 | bool flat; | |
3826 | u32 *entry, new_entry; | |
3827 | ||
3828 | flat = kvm_lapic_get_reg(vcpu->arch.apic, APIC_DFR) == APIC_DFR_FLAT; | |
3829 | entry = avic_get_logical_id_entry(vcpu, ldr, flat); | |
3830 | if (!entry) | |
3831 | return -EINVAL; | |
3832 | ||
3833 | new_entry = READ_ONCE(*entry); | |
3834 | new_entry &= ~AVIC_LOGICAL_ID_ENTRY_GUEST_PHYSICAL_ID_MASK; | |
3835 | new_entry |= (g_physical_id & AVIC_LOGICAL_ID_ENTRY_GUEST_PHYSICAL_ID_MASK); | |
3836 | if (valid) | |
3837 | new_entry |= AVIC_LOGICAL_ID_ENTRY_VALID_MASK; | |
3838 | else | |
3839 | new_entry &= ~AVIC_LOGICAL_ID_ENTRY_VALID_MASK; | |
3840 | WRITE_ONCE(*entry, new_entry); | |
3841 | ||
3842 | return 0; | |
3843 | } | |
3844 | ||
3845 | static int avic_handle_ldr_update(struct kvm_vcpu *vcpu) | |
3846 | { | |
3847 | int ret; | |
3848 | struct vcpu_svm *svm = to_svm(vcpu); | |
3849 | u32 ldr = kvm_lapic_get_reg(vcpu->arch.apic, APIC_LDR); | |
3850 | ||
3851 | if (!ldr) | |
3852 | return 1; | |
3853 | ||
3854 | ret = avic_ldr_write(vcpu, vcpu->vcpu_id, ldr, true); | |
3855 | if (ret && svm->ldr_reg) { | |
3856 | avic_ldr_write(vcpu, 0, svm->ldr_reg, false); | |
3857 | svm->ldr_reg = 0; | |
3858 | } else { | |
3859 | svm->ldr_reg = ldr; | |
3860 | } | |
3861 | return ret; | |
3862 | } | |
3863 | ||
3864 | static int avic_handle_apic_id_update(struct kvm_vcpu *vcpu) | |
3865 | { | |
3866 | u64 *old, *new; | |
3867 | struct vcpu_svm *svm = to_svm(vcpu); | |
3868 | u32 apic_id_reg = kvm_lapic_get_reg(vcpu->arch.apic, APIC_ID); | |
3869 | u32 id = (apic_id_reg >> 24) & 0xff; | |
3870 | ||
3871 | if (vcpu->vcpu_id == id) | |
3872 | return 0; | |
3873 | ||
3874 | old = avic_get_physical_id_entry(vcpu, vcpu->vcpu_id); | |
3875 | new = avic_get_physical_id_entry(vcpu, id); | |
3876 | if (!new || !old) | |
3877 | return 1; | |
3878 | ||
3879 | /* We need to move physical_id_entry to new offset */ | |
3880 | *new = *old; | |
3881 | *old = 0ULL; | |
3882 | to_svm(vcpu)->avic_physical_id_cache = new; | |
3883 | ||
3884 | /* | |
3885 | * Also update the guest physical APIC ID in the logical | |
3886 | * APIC ID table entry if already setup the LDR. | |
3887 | */ | |
3888 | if (svm->ldr_reg) | |
3889 | avic_handle_ldr_update(vcpu); | |
3890 | ||
3891 | return 0; | |
3892 | } | |
3893 | ||
3894 | static int avic_handle_dfr_update(struct kvm_vcpu *vcpu) | |
3895 | { | |
3896 | struct vcpu_svm *svm = to_svm(vcpu); | |
3897 | struct kvm_arch *vm_data = &vcpu->kvm->arch; | |
3898 | u32 dfr = kvm_lapic_get_reg(vcpu->arch.apic, APIC_DFR); | |
3899 | u32 mod = (dfr >> 28) & 0xf; | |
3900 | ||
3901 | /* | |
3902 | * We assume that all local APICs are using the same type. | |
3903 | * If this changes, we need to flush the AVIC logical | |
3904 | * APID id table. | |
3905 | */ | |
3906 | if (vm_data->ldr_mode == mod) | |
3907 | return 0; | |
3908 | ||
3909 | clear_page(page_address(vm_data->avic_logical_id_table_page)); | |
3910 | vm_data->ldr_mode = mod; | |
3911 | ||
3912 | if (svm->ldr_reg) | |
3913 | avic_handle_ldr_update(vcpu); | |
3914 | return 0; | |
3915 | } | |
3916 | ||
3917 | static int avic_unaccel_trap_write(struct vcpu_svm *svm) | |
3918 | { | |
3919 | struct kvm_lapic *apic = svm->vcpu.arch.apic; | |
3920 | u32 offset = svm->vmcb->control.exit_info_1 & | |
3921 | AVIC_UNACCEL_ACCESS_OFFSET_MASK; | |
3922 | ||
3923 | switch (offset) { | |
3924 | case APIC_ID: | |
3925 | if (avic_handle_apic_id_update(&svm->vcpu)) | |
3926 | return 0; | |
3927 | break; | |
3928 | case APIC_LDR: | |
3929 | if (avic_handle_ldr_update(&svm->vcpu)) | |
3930 | return 0; | |
3931 | break; | |
3932 | case APIC_DFR: | |
3933 | avic_handle_dfr_update(&svm->vcpu); | |
3934 | break; | |
3935 | default: | |
3936 | break; | |
3937 | } | |
3938 | ||
3939 | kvm_lapic_reg_write(apic, offset, kvm_lapic_get_reg(apic, offset)); | |
3940 | ||
3941 | return 1; | |
3942 | } | |
3943 | ||
3944 | static bool is_avic_unaccelerated_access_trap(u32 offset) | |
3945 | { | |
3946 | bool ret = false; | |
3947 | ||
3948 | switch (offset) { | |
3949 | case APIC_ID: | |
3950 | case APIC_EOI: | |
3951 | case APIC_RRR: | |
3952 | case APIC_LDR: | |
3953 | case APIC_DFR: | |
3954 | case APIC_SPIV: | |
3955 | case APIC_ESR: | |
3956 | case APIC_ICR: | |
3957 | case APIC_LVTT: | |
3958 | case APIC_LVTTHMR: | |
3959 | case APIC_LVTPC: | |
3960 | case APIC_LVT0: | |
3961 | case APIC_LVT1: | |
3962 | case APIC_LVTERR: | |
3963 | case APIC_TMICT: | |
3964 | case APIC_TDCR: | |
3965 | ret = true; | |
3966 | break; | |
3967 | default: | |
3968 | break; | |
3969 | } | |
3970 | return ret; | |
3971 | } | |
3972 | ||
3973 | static int avic_unaccelerated_access_interception(struct vcpu_svm *svm) | |
3974 | { | |
3975 | int ret = 0; | |
3976 | u32 offset = svm->vmcb->control.exit_info_1 & | |
3977 | AVIC_UNACCEL_ACCESS_OFFSET_MASK; | |
3978 | u32 vector = svm->vmcb->control.exit_info_2 & | |
3979 | AVIC_UNACCEL_ACCESS_VECTOR_MASK; | |
3980 | bool write = (svm->vmcb->control.exit_info_1 >> 32) & | |
3981 | AVIC_UNACCEL_ACCESS_WRITE_MASK; | |
3982 | bool trap = is_avic_unaccelerated_access_trap(offset); | |
3983 | ||
3984 | trace_kvm_avic_unaccelerated_access(svm->vcpu.vcpu_id, offset, | |
3985 | trap, write, vector); | |
3986 | if (trap) { | |
3987 | /* Handling Trap */ | |
3988 | WARN_ONCE(!write, "svm: Handling trap read.\n"); | |
3989 | ret = avic_unaccel_trap_write(svm); | |
3990 | } else { | |
3991 | /* Handling Fault */ | |
3992 | ret = (emulate_instruction(&svm->vcpu, 0) == EMULATE_DONE); | |
3993 | } | |
3994 | ||
3995 | return ret; | |
3996 | } | |
3997 | ||
3998 | static int (*const svm_exit_handlers[])(struct vcpu_svm *svm) = { | |
3999 | [SVM_EXIT_READ_CR0] = cr_interception, | |
4000 | [SVM_EXIT_READ_CR3] = cr_interception, | |
4001 | [SVM_EXIT_READ_CR4] = cr_interception, | |
4002 | [SVM_EXIT_READ_CR8] = cr_interception, | |
4003 | [SVM_EXIT_CR0_SEL_WRITE] = cr_interception, | |
4004 | [SVM_EXIT_WRITE_CR0] = cr_interception, | |
4005 | [SVM_EXIT_WRITE_CR3] = cr_interception, | |
4006 | [SVM_EXIT_WRITE_CR4] = cr_interception, | |
4007 | [SVM_EXIT_WRITE_CR8] = cr8_write_interception, | |
4008 | [SVM_EXIT_READ_DR0] = dr_interception, | |
4009 | [SVM_EXIT_READ_DR1] = dr_interception, | |
4010 | [SVM_EXIT_READ_DR2] = dr_interception, | |
4011 | [SVM_EXIT_READ_DR3] = dr_interception, | |
4012 | [SVM_EXIT_READ_DR4] = dr_interception, | |
4013 | [SVM_EXIT_READ_DR5] = dr_interception, | |
4014 | [SVM_EXIT_READ_DR6] = dr_interception, | |
4015 | [SVM_EXIT_READ_DR7] = dr_interception, | |
4016 | [SVM_EXIT_WRITE_DR0] = dr_interception, | |
4017 | [SVM_EXIT_WRITE_DR1] = dr_interception, | |
4018 | [SVM_EXIT_WRITE_DR2] = dr_interception, | |
4019 | [SVM_EXIT_WRITE_DR3] = dr_interception, | |
4020 | [SVM_EXIT_WRITE_DR4] = dr_interception, | |
4021 | [SVM_EXIT_WRITE_DR5] = dr_interception, | |
4022 | [SVM_EXIT_WRITE_DR6] = dr_interception, | |
4023 | [SVM_EXIT_WRITE_DR7] = dr_interception, | |
4024 | [SVM_EXIT_EXCP_BASE + DB_VECTOR] = db_interception, | |
4025 | [SVM_EXIT_EXCP_BASE + BP_VECTOR] = bp_interception, | |
4026 | [SVM_EXIT_EXCP_BASE + UD_VECTOR] = ud_interception, | |
4027 | [SVM_EXIT_EXCP_BASE + PF_VECTOR] = pf_interception, | |
4028 | [SVM_EXIT_EXCP_BASE + NM_VECTOR] = nm_interception, | |
4029 | [SVM_EXIT_EXCP_BASE + MC_VECTOR] = mc_interception, | |
4030 | [SVM_EXIT_EXCP_BASE + AC_VECTOR] = ac_interception, | |
4031 | [SVM_EXIT_INTR] = intr_interception, | |
4032 | [SVM_EXIT_NMI] = nmi_interception, | |
4033 | [SVM_EXIT_SMI] = nop_on_interception, | |
4034 | [SVM_EXIT_INIT] = nop_on_interception, | |
4035 | [SVM_EXIT_VINTR] = interrupt_window_interception, | |
4036 | [SVM_EXIT_RDPMC] = rdpmc_interception, | |
4037 | [SVM_EXIT_CPUID] = cpuid_interception, | |
4038 | [SVM_EXIT_IRET] = iret_interception, | |
4039 | [SVM_EXIT_INVD] = emulate_on_interception, | |
4040 | [SVM_EXIT_PAUSE] = pause_interception, | |
4041 | [SVM_EXIT_HLT] = halt_interception, | |
4042 | [SVM_EXIT_INVLPG] = invlpg_interception, | |
4043 | [SVM_EXIT_INVLPGA] = invlpga_interception, | |
4044 | [SVM_EXIT_IOIO] = io_interception, | |
4045 | [SVM_EXIT_MSR] = msr_interception, | |
4046 | [SVM_EXIT_TASK_SWITCH] = task_switch_interception, | |
4047 | [SVM_EXIT_SHUTDOWN] = shutdown_interception, | |
4048 | [SVM_EXIT_VMRUN] = vmrun_interception, | |
4049 | [SVM_EXIT_VMMCALL] = vmmcall_interception, | |
4050 | [SVM_EXIT_VMLOAD] = vmload_interception, | |
4051 | [SVM_EXIT_VMSAVE] = vmsave_interception, | |
4052 | [SVM_EXIT_STGI] = stgi_interception, | |
4053 | [SVM_EXIT_CLGI] = clgi_interception, | |
4054 | [SVM_EXIT_SKINIT] = skinit_interception, | |
4055 | [SVM_EXIT_WBINVD] = wbinvd_interception, | |
4056 | [SVM_EXIT_MONITOR] = monitor_interception, | |
4057 | [SVM_EXIT_MWAIT] = mwait_interception, | |
4058 | [SVM_EXIT_XSETBV] = xsetbv_interception, | |
4059 | [SVM_EXIT_NPF] = pf_interception, | |
4060 | [SVM_EXIT_RSM] = emulate_on_interception, | |
4061 | [SVM_EXIT_AVIC_INCOMPLETE_IPI] = avic_incomplete_ipi_interception, | |
4062 | [SVM_EXIT_AVIC_UNACCELERATED_ACCESS] = avic_unaccelerated_access_interception, | |
4063 | }; | |
4064 | ||
4065 | static void dump_vmcb(struct kvm_vcpu *vcpu) | |
4066 | { | |
4067 | struct vcpu_svm *svm = to_svm(vcpu); | |
4068 | struct vmcb_control_area *control = &svm->vmcb->control; | |
4069 | struct vmcb_save_area *save = &svm->vmcb->save; | |
4070 | ||
4071 | pr_err("VMCB Control Area:\n"); | |
4072 | pr_err("%-20s%04x\n", "cr_read:", control->intercept_cr & 0xffff); | |
4073 | pr_err("%-20s%04x\n", "cr_write:", control->intercept_cr >> 16); | |
4074 | pr_err("%-20s%04x\n", "dr_read:", control->intercept_dr & 0xffff); | |
4075 | pr_err("%-20s%04x\n", "dr_write:", control->intercept_dr >> 16); | |
4076 | pr_err("%-20s%08x\n", "exceptions:", control->intercept_exceptions); | |
4077 | pr_err("%-20s%016llx\n", "intercepts:", control->intercept); | |
4078 | pr_err("%-20s%d\n", "pause filter count:", control->pause_filter_count); | |
4079 | pr_err("%-20s%016llx\n", "iopm_base_pa:", control->iopm_base_pa); | |
4080 | pr_err("%-20s%016llx\n", "msrpm_base_pa:", control->msrpm_base_pa); | |
4081 | pr_err("%-20s%016llx\n", "tsc_offset:", control->tsc_offset); | |
4082 | pr_err("%-20s%d\n", "asid:", control->asid); | |
4083 | pr_err("%-20s%d\n", "tlb_ctl:", control->tlb_ctl); | |
4084 | pr_err("%-20s%08x\n", "int_ctl:", control->int_ctl); | |
4085 | pr_err("%-20s%08x\n", "int_vector:", control->int_vector); | |
4086 | pr_err("%-20s%08x\n", "int_state:", control->int_state); | |
4087 | pr_err("%-20s%08x\n", "exit_code:", control->exit_code); | |
4088 | pr_err("%-20s%016llx\n", "exit_info1:", control->exit_info_1); | |
4089 | pr_err("%-20s%016llx\n", "exit_info2:", control->exit_info_2); | |
4090 | pr_err("%-20s%08x\n", "exit_int_info:", control->exit_int_info); | |
4091 | pr_err("%-20s%08x\n", "exit_int_info_err:", control->exit_int_info_err); | |
4092 | pr_err("%-20s%lld\n", "nested_ctl:", control->nested_ctl); | |
4093 | pr_err("%-20s%016llx\n", "nested_cr3:", control->nested_cr3); | |
4094 | pr_err("%-20s%016llx\n", "avic_vapic_bar:", control->avic_vapic_bar); | |
4095 | pr_err("%-20s%08x\n", "event_inj:", control->event_inj); | |
4096 | pr_err("%-20s%08x\n", "event_inj_err:", control->event_inj_err); | |
4097 | pr_err("%-20s%lld\n", "lbr_ctl:", control->lbr_ctl); | |
4098 | pr_err("%-20s%016llx\n", "next_rip:", control->next_rip); | |
4099 | pr_err("%-20s%016llx\n", "avic_backing_page:", control->avic_backing_page); | |
4100 | pr_err("%-20s%016llx\n", "avic_logical_id:", control->avic_logical_id); | |
4101 | pr_err("%-20s%016llx\n", "avic_physical_id:", control->avic_physical_id); | |
4102 | pr_err("VMCB State Save Area:\n"); | |
4103 | pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n", | |
4104 | "es:", | |
4105 | save->es.selector, save->es.attrib, | |
4106 | save->es.limit, save->es.base); | |
4107 | pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n", | |
4108 | "cs:", | |
4109 | save->cs.selector, save->cs.attrib, | |
4110 | save->cs.limit, save->cs.base); | |
4111 | pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n", | |
4112 | "ss:", | |
4113 | save->ss.selector, save->ss.attrib, | |
4114 | save->ss.limit, save->ss.base); | |
4115 | pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n", | |
4116 | "ds:", | |
4117 | save->ds.selector, save->ds.attrib, | |
4118 | save->ds.limit, save->ds.base); | |
4119 | pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n", | |
4120 | "fs:", | |
4121 | save->fs.selector, save->fs.attrib, | |
4122 | save->fs.limit, save->fs.base); | |
4123 | pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n", | |
4124 | "gs:", | |
4125 | save->gs.selector, save->gs.attrib, | |
4126 | save->gs.limit, save->gs.base); | |
4127 | pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n", | |
4128 | "gdtr:", | |
4129 | save->gdtr.selector, save->gdtr.attrib, | |
4130 | save->gdtr.limit, save->gdtr.base); | |
4131 | pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n", | |
4132 | "ldtr:", | |
4133 | save->ldtr.selector, save->ldtr.attrib, | |
4134 | save->ldtr.limit, save->ldtr.base); | |
4135 | pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n", | |
4136 | "idtr:", | |
4137 | save->idtr.selector, save->idtr.attrib, | |
4138 | save->idtr.limit, save->idtr.base); | |
4139 | pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n", | |
4140 | "tr:", | |
4141 | save->tr.selector, save->tr.attrib, | |
4142 | save->tr.limit, save->tr.base); | |
4143 | pr_err("cpl: %d efer: %016llx\n", | |
4144 | save->cpl, save->efer); | |
4145 | pr_err("%-15s %016llx %-13s %016llx\n", | |
4146 | "cr0:", save->cr0, "cr2:", save->cr2); | |
4147 | pr_err("%-15s %016llx %-13s %016llx\n", | |
4148 | "cr3:", save->cr3, "cr4:", save->cr4); | |
4149 | pr_err("%-15s %016llx %-13s %016llx\n", | |
4150 | "dr6:", save->dr6, "dr7:", save->dr7); | |
4151 | pr_err("%-15s %016llx %-13s %016llx\n", | |
4152 | "rip:", save->rip, "rflags:", save->rflags); | |
4153 | pr_err("%-15s %016llx %-13s %016llx\n", | |
4154 | "rsp:", save->rsp, "rax:", save->rax); | |
4155 | pr_err("%-15s %016llx %-13s %016llx\n", | |
4156 | "star:", save->star, "lstar:", save->lstar); | |
4157 | pr_err("%-15s %016llx %-13s %016llx\n", | |
4158 | "cstar:", save->cstar, "sfmask:", save->sfmask); | |
4159 | pr_err("%-15s %016llx %-13s %016llx\n", | |
4160 | "kernel_gs_base:", save->kernel_gs_base, | |
4161 | "sysenter_cs:", save->sysenter_cs); | |
4162 | pr_err("%-15s %016llx %-13s %016llx\n", | |
4163 | "sysenter_esp:", save->sysenter_esp, | |
4164 | "sysenter_eip:", save->sysenter_eip); | |
4165 | pr_err("%-15s %016llx %-13s %016llx\n", | |
4166 | "gpat:", save->g_pat, "dbgctl:", save->dbgctl); | |
4167 | pr_err("%-15s %016llx %-13s %016llx\n", | |
4168 | "br_from:", save->br_from, "br_to:", save->br_to); | |
4169 | pr_err("%-15s %016llx %-13s %016llx\n", | |
4170 | "excp_from:", save->last_excp_from, | |
4171 | "excp_to:", save->last_excp_to); | |
4172 | } | |
4173 | ||
4174 | static void svm_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2) | |
4175 | { | |
4176 | struct vmcb_control_area *control = &to_svm(vcpu)->vmcb->control; | |
4177 | ||
4178 | *info1 = control->exit_info_1; | |
4179 | *info2 = control->exit_info_2; | |
4180 | } | |
4181 | ||
4182 | static int handle_exit(struct kvm_vcpu *vcpu) | |
4183 | { | |
4184 | struct vcpu_svm *svm = to_svm(vcpu); | |
4185 | struct kvm_run *kvm_run = vcpu->run; | |
4186 | u32 exit_code = svm->vmcb->control.exit_code; | |
4187 | ||
4188 | trace_kvm_exit(exit_code, vcpu, KVM_ISA_SVM); | |
4189 | ||
4190 | if (!is_cr_intercept(svm, INTERCEPT_CR0_WRITE)) | |
4191 | vcpu->arch.cr0 = svm->vmcb->save.cr0; | |
4192 | if (npt_enabled) | |
4193 | vcpu->arch.cr3 = svm->vmcb->save.cr3; | |
4194 | ||
4195 | if (unlikely(svm->nested.exit_required)) { | |
4196 | nested_svm_vmexit(svm); | |
4197 | svm->nested.exit_required = false; | |
4198 | ||
4199 | return 1; | |
4200 | } | |
4201 | ||
4202 | if (is_guest_mode(vcpu)) { | |
4203 | int vmexit; | |
4204 | ||
4205 | trace_kvm_nested_vmexit(svm->vmcb->save.rip, exit_code, | |
4206 | svm->vmcb->control.exit_info_1, | |
4207 | svm->vmcb->control.exit_info_2, | |
4208 | svm->vmcb->control.exit_int_info, | |
4209 | svm->vmcb->control.exit_int_info_err, | |
4210 | KVM_ISA_SVM); | |
4211 | ||
4212 | vmexit = nested_svm_exit_special(svm); | |
4213 | ||
4214 | if (vmexit == NESTED_EXIT_CONTINUE) | |
4215 | vmexit = nested_svm_exit_handled(svm); | |
4216 | ||
4217 | if (vmexit == NESTED_EXIT_DONE) | |
4218 | return 1; | |
4219 | } | |
4220 | ||
4221 | svm_complete_interrupts(svm); | |
4222 | ||
4223 | if (svm->vmcb->control.exit_code == SVM_EXIT_ERR) { | |
4224 | kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY; | |
4225 | kvm_run->fail_entry.hardware_entry_failure_reason | |
4226 | = svm->vmcb->control.exit_code; | |
4227 | pr_err("KVM: FAILED VMRUN WITH VMCB:\n"); | |
4228 | dump_vmcb(vcpu); | |
4229 | return 0; | |
4230 | } | |
4231 | ||
4232 | if (is_external_interrupt(svm->vmcb->control.exit_int_info) && | |
4233 | exit_code != SVM_EXIT_EXCP_BASE + PF_VECTOR && | |
4234 | exit_code != SVM_EXIT_NPF && exit_code != SVM_EXIT_TASK_SWITCH && | |
4235 | exit_code != SVM_EXIT_INTR && exit_code != SVM_EXIT_NMI) | |
4236 | printk(KERN_ERR "%s: unexpected exit_int_info 0x%x " | |
4237 | "exit_code 0x%x\n", | |
4238 | __func__, svm->vmcb->control.exit_int_info, | |
4239 | exit_code); | |
4240 | ||
4241 | if (exit_code >= ARRAY_SIZE(svm_exit_handlers) | |
4242 | || !svm_exit_handlers[exit_code]) { | |
4243 | WARN_ONCE(1, "svm: unexpected exit reason 0x%x\n", exit_code); | |
4244 | kvm_queue_exception(vcpu, UD_VECTOR); | |
4245 | return 1; | |
4246 | } | |
4247 | ||
4248 | return svm_exit_handlers[exit_code](svm); | |
4249 | } | |
4250 | ||
4251 | static void reload_tss(struct kvm_vcpu *vcpu) | |
4252 | { | |
4253 | int cpu = raw_smp_processor_id(); | |
4254 | ||
4255 | struct svm_cpu_data *sd = per_cpu(svm_data, cpu); | |
4256 | sd->tss_desc->type = 9; /* available 32/64-bit TSS */ | |
4257 | load_TR_desc(); | |
4258 | } | |
4259 | ||
4260 | static void pre_svm_run(struct vcpu_svm *svm) | |
4261 | { | |
4262 | int cpu = raw_smp_processor_id(); | |
4263 | ||
4264 | struct svm_cpu_data *sd = per_cpu(svm_data, cpu); | |
4265 | ||
4266 | /* FIXME: handle wraparound of asid_generation */ | |
4267 | if (svm->asid_generation != sd->asid_generation) | |
4268 | new_asid(svm, sd); | |
4269 | } | |
4270 | ||
4271 | static void svm_inject_nmi(struct kvm_vcpu *vcpu) | |
4272 | { | |
4273 | struct vcpu_svm *svm = to_svm(vcpu); | |
4274 | ||
4275 | svm->vmcb->control.event_inj = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI; | |
4276 | vcpu->arch.hflags |= HF_NMI_MASK; | |
4277 | set_intercept(svm, INTERCEPT_IRET); | |
4278 | ++vcpu->stat.nmi_injections; | |
4279 | } | |
4280 | ||
4281 | static inline void svm_inject_irq(struct vcpu_svm *svm, int irq) | |
4282 | { | |
4283 | struct vmcb_control_area *control; | |
4284 | ||
4285 | /* The following fields are ignored when AVIC is enabled */ | |
4286 | control = &svm->vmcb->control; | |
4287 | control->int_vector = irq; | |
4288 | control->int_ctl &= ~V_INTR_PRIO_MASK; | |
4289 | control->int_ctl |= V_IRQ_MASK | | |
4290 | ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT); | |
4291 | mark_dirty(svm->vmcb, VMCB_INTR); | |
4292 | } | |
4293 | ||
4294 | static void svm_set_irq(struct kvm_vcpu *vcpu) | |
4295 | { | |
4296 | struct vcpu_svm *svm = to_svm(vcpu); | |
4297 | ||
4298 | BUG_ON(!(gif_set(svm))); | |
4299 | ||
4300 | trace_kvm_inj_virq(vcpu->arch.interrupt.nr); | |
4301 | ++vcpu->stat.irq_injections; | |
4302 | ||
4303 | svm->vmcb->control.event_inj = vcpu->arch.interrupt.nr | | |
4304 | SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR; | |
4305 | } | |
4306 | ||
4307 | static inline bool svm_nested_virtualize_tpr(struct kvm_vcpu *vcpu) | |
4308 | { | |
4309 | return is_guest_mode(vcpu) && (vcpu->arch.hflags & HF_VINTR_MASK); | |
4310 | } | |
4311 | ||
4312 | static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr) | |
4313 | { | |
4314 | struct vcpu_svm *svm = to_svm(vcpu); | |
4315 | ||
4316 | if (svm_nested_virtualize_tpr(vcpu) || | |
4317 | kvm_vcpu_apicv_active(vcpu)) | |
4318 | return; | |
4319 | ||
4320 | clr_cr_intercept(svm, INTERCEPT_CR8_WRITE); | |
4321 | ||
4322 | if (irr == -1) | |
4323 | return; | |
4324 | ||
4325 | if (tpr >= irr) | |
4326 | set_cr_intercept(svm, INTERCEPT_CR8_WRITE); | |
4327 | } | |
4328 | ||
4329 | static void svm_set_virtual_x2apic_mode(struct kvm_vcpu *vcpu, bool set) | |
4330 | { | |
4331 | return; | |
4332 | } | |
4333 | ||
4334 | static bool svm_get_enable_apicv(void) | |
4335 | { | |
4336 | return avic; | |
4337 | } | |
4338 | ||
4339 | static void svm_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr) | |
4340 | { | |
4341 | } | |
4342 | ||
4343 | static void svm_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr) | |
4344 | { | |
4345 | } | |
4346 | ||
4347 | /* Note: Currently only used by Hyper-V. */ | |
4348 | static void svm_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu) | |
4349 | { | |
4350 | struct vcpu_svm *svm = to_svm(vcpu); | |
4351 | struct vmcb *vmcb = svm->vmcb; | |
4352 | ||
4353 | if (!avic) | |
4354 | return; | |
4355 | ||
4356 | vmcb->control.int_ctl &= ~AVIC_ENABLE_MASK; | |
4357 | mark_dirty(vmcb, VMCB_INTR); | |
4358 | } | |
4359 | ||
4360 | static void svm_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap) | |
4361 | { | |
4362 | return; | |
4363 | } | |
4364 | ||
4365 | static void svm_sync_pir_to_irr(struct kvm_vcpu *vcpu) | |
4366 | { | |
4367 | return; | |
4368 | } | |
4369 | ||
4370 | static void svm_deliver_avic_intr(struct kvm_vcpu *vcpu, int vec) | |
4371 | { | |
4372 | kvm_lapic_set_irr(vec, vcpu->arch.apic); | |
4373 | smp_mb__after_atomic(); | |
4374 | ||
4375 | if (avic_vcpu_is_running(vcpu)) | |
4376 | wrmsrl(SVM_AVIC_DOORBELL, | |
4377 | kvm_cpu_get_apicid(vcpu->cpu)); | |
4378 | else | |
4379 | kvm_vcpu_wake_up(vcpu); | |
4380 | } | |
4381 | ||
4382 | static void svm_ir_list_del(struct vcpu_svm *svm, struct amd_iommu_pi_data *pi) | |
4383 | { | |
4384 | unsigned long flags; | |
4385 | struct amd_svm_iommu_ir *cur; | |
4386 | ||
4387 | spin_lock_irqsave(&svm->ir_list_lock, flags); | |
4388 | list_for_each_entry(cur, &svm->ir_list, node) { | |
4389 | if (cur->data != pi->ir_data) | |
4390 | continue; | |
4391 | list_del(&cur->node); | |
4392 | kfree(cur); | |
4393 | break; | |
4394 | } | |
4395 | spin_unlock_irqrestore(&svm->ir_list_lock, flags); | |
4396 | } | |
4397 | ||
4398 | static int svm_ir_list_add(struct vcpu_svm *svm, struct amd_iommu_pi_data *pi) | |
4399 | { | |
4400 | int ret = 0; | |
4401 | unsigned long flags; | |
4402 | struct amd_svm_iommu_ir *ir; | |
4403 | ||
4404 | /** | |
4405 | * In some cases, the existing irte is updaed and re-set, | |
4406 | * so we need to check here if it's already been * added | |
4407 | * to the ir_list. | |
4408 | */ | |
4409 | if (pi->ir_data && (pi->prev_ga_tag != 0)) { | |
4410 | struct kvm *kvm = svm->vcpu.kvm; | |
4411 | u32 vcpu_id = AVIC_GATAG_TO_VCPUID(pi->prev_ga_tag); | |
4412 | struct kvm_vcpu *prev_vcpu = kvm_get_vcpu_by_id(kvm, vcpu_id); | |
4413 | struct vcpu_svm *prev_svm; | |
4414 | ||
4415 | if (!prev_vcpu) { | |
4416 | ret = -EINVAL; | |
4417 | goto out; | |
4418 | } | |
4419 | ||
4420 | prev_svm = to_svm(prev_vcpu); | |
4421 | svm_ir_list_del(prev_svm, pi); | |
4422 | } | |
4423 | ||
4424 | /** | |
4425 | * Allocating new amd_iommu_pi_data, which will get | |
4426 | * add to the per-vcpu ir_list. | |
4427 | */ | |
4428 | ir = kzalloc(sizeof(struct amd_svm_iommu_ir), GFP_KERNEL); | |
4429 | if (!ir) { | |
4430 | ret = -ENOMEM; | |
4431 | goto out; | |
4432 | } | |
4433 | ir->data = pi->ir_data; | |
4434 | ||
4435 | spin_lock_irqsave(&svm->ir_list_lock, flags); | |
4436 | list_add(&ir->node, &svm->ir_list); | |
4437 | spin_unlock_irqrestore(&svm->ir_list_lock, flags); | |
4438 | out: | |
4439 | return ret; | |
4440 | } | |
4441 | ||
4442 | /** | |
4443 | * Note: | |
4444 | * The HW cannot support posting multicast/broadcast | |
4445 | * interrupts to a vCPU. So, we still use legacy interrupt | |
4446 | * remapping for these kind of interrupts. | |
4447 | * | |
4448 | * For lowest-priority interrupts, we only support | |
4449 | * those with single CPU as the destination, e.g. user | |
4450 | * configures the interrupts via /proc/irq or uses | |
4451 | * irqbalance to make the interrupts single-CPU. | |
4452 | */ | |
4453 | static int | |
4454 | get_pi_vcpu_info(struct kvm *kvm, struct kvm_kernel_irq_routing_entry *e, | |
4455 | struct vcpu_data *vcpu_info, struct vcpu_svm **svm) | |
4456 | { | |
4457 | struct kvm_lapic_irq irq; | |
4458 | struct kvm_vcpu *vcpu = NULL; | |
4459 | ||
4460 | kvm_set_msi_irq(kvm, e, &irq); | |
4461 | ||
4462 | if (!kvm_intr_is_single_vcpu(kvm, &irq, &vcpu)) { | |
4463 | pr_debug("SVM: %s: use legacy intr remap mode for irq %u\n", | |
4464 | __func__, irq.vector); | |
4465 | return -1; | |
4466 | } | |
4467 | ||
4468 | pr_debug("SVM: %s: use GA mode for irq %u\n", __func__, | |
4469 | irq.vector); | |
4470 | *svm = to_svm(vcpu); | |
4471 | vcpu_info->pi_desc_addr = page_to_phys((*svm)->avic_backing_page); | |
4472 | vcpu_info->vector = irq.vector; | |
4473 | ||
4474 | return 0; | |
4475 | } | |
4476 | ||
4477 | /* | |
4478 | * svm_update_pi_irte - set IRTE for Posted-Interrupts | |
4479 | * | |
4480 | * @kvm: kvm | |
4481 | * @host_irq: host irq of the interrupt | |
4482 | * @guest_irq: gsi of the interrupt | |
4483 | * @set: set or unset PI | |
4484 | * returns 0 on success, < 0 on failure | |
4485 | */ | |
4486 | static int svm_update_pi_irte(struct kvm *kvm, unsigned int host_irq, | |
4487 | uint32_t guest_irq, bool set) | |
4488 | { | |
4489 | struct kvm_kernel_irq_routing_entry *e; | |
4490 | struct kvm_irq_routing_table *irq_rt; | |
4491 | int idx, ret = -EINVAL; | |
4492 | ||
4493 | if (!kvm_arch_has_assigned_device(kvm) || | |
4494 | !irq_remapping_cap(IRQ_POSTING_CAP)) | |
4495 | return 0; | |
4496 | ||
4497 | pr_debug("SVM: %s: host_irq=%#x, guest_irq=%#x, set=%#x\n", | |
4498 | __func__, host_irq, guest_irq, set); | |
4499 | ||
4500 | idx = srcu_read_lock(&kvm->irq_srcu); | |
4501 | irq_rt = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu); | |
4502 | WARN_ON(guest_irq >= irq_rt->nr_rt_entries); | |
4503 | ||
4504 | hlist_for_each_entry(e, &irq_rt->map[guest_irq], link) { | |
4505 | struct vcpu_data vcpu_info; | |
4506 | struct vcpu_svm *svm = NULL; | |
4507 | ||
4508 | if (e->type != KVM_IRQ_ROUTING_MSI) | |
4509 | continue; | |
4510 | ||
4511 | /** | |
4512 | * Here, we setup with legacy mode in the following cases: | |
4513 | * 1. When cannot target interrupt to a specific vcpu. | |
4514 | * 2. Unsetting posted interrupt. | |
4515 | * 3. APIC virtialization is disabled for the vcpu. | |
4516 | */ | |
4517 | if (!get_pi_vcpu_info(kvm, e, &vcpu_info, &svm) && set && | |
4518 | kvm_vcpu_apicv_active(&svm->vcpu)) { | |
4519 | struct amd_iommu_pi_data pi; | |
4520 | ||
4521 | /* Try to enable guest_mode in IRTE */ | |
4522 | pi.base = page_to_phys(svm->avic_backing_page) & AVIC_HPA_MASK; | |
4523 | pi.ga_tag = AVIC_GATAG(kvm->arch.avic_vm_id, | |
4524 | svm->vcpu.vcpu_id); | |
4525 | pi.is_guest_mode = true; | |
4526 | pi.vcpu_data = &vcpu_info; | |
4527 | ret = irq_set_vcpu_affinity(host_irq, &pi); | |
4528 | ||
4529 | /** | |
4530 | * Here, we successfully setting up vcpu affinity in | |
4531 | * IOMMU guest mode. Now, we need to store the posted | |
4532 | * interrupt information in a per-vcpu ir_list so that | |
4533 | * we can reference to them directly when we update vcpu | |
4534 | * scheduling information in IOMMU irte. | |
4535 | */ | |
4536 | if (!ret && pi.is_guest_mode) | |
4537 | svm_ir_list_add(svm, &pi); | |
4538 | } else { | |
4539 | /* Use legacy mode in IRTE */ | |
4540 | struct amd_iommu_pi_data pi; | |
4541 | ||
4542 | /** | |
4543 | * Here, pi is used to: | |
4544 | * - Tell IOMMU to use legacy mode for this interrupt. | |
4545 | * - Retrieve ga_tag of prior interrupt remapping data. | |
4546 | */ | |
4547 | pi.is_guest_mode = false; | |
4548 | ret = irq_set_vcpu_affinity(host_irq, &pi); | |
4549 | ||
4550 | /** | |
4551 | * Check if the posted interrupt was previously | |
4552 | * setup with the guest_mode by checking if the ga_tag | |
4553 | * was cached. If so, we need to clean up the per-vcpu | |
4554 | * ir_list. | |
4555 | */ | |
4556 | if (!ret && pi.prev_ga_tag) { | |
4557 | int id = AVIC_GATAG_TO_VCPUID(pi.prev_ga_tag); | |
4558 | struct kvm_vcpu *vcpu; | |
4559 | ||
4560 | vcpu = kvm_get_vcpu_by_id(kvm, id); | |
4561 | if (vcpu) | |
4562 | svm_ir_list_del(to_svm(vcpu), &pi); | |
4563 | } | |
4564 | } | |
4565 | ||
4566 | if (!ret && svm) { | |
4567 | trace_kvm_pi_irte_update(svm->vcpu.vcpu_id, | |
4568 | host_irq, e->gsi, | |
4569 | vcpu_info.vector, | |
4570 | vcpu_info.pi_desc_addr, set); | |
4571 | } | |
4572 | ||
4573 | if (ret < 0) { | |
4574 | pr_err("%s: failed to update PI IRTE\n", __func__); | |
4575 | goto out; | |
4576 | } | |
4577 | } | |
4578 | ||
4579 | ret = 0; | |
4580 | out: | |
4581 | srcu_read_unlock(&kvm->irq_srcu, idx); | |
4582 | return ret; | |
4583 | } | |
4584 | ||
4585 | static int svm_nmi_allowed(struct kvm_vcpu *vcpu) | |
4586 | { | |
4587 | struct vcpu_svm *svm = to_svm(vcpu); | |
4588 | struct vmcb *vmcb = svm->vmcb; | |
4589 | int ret; | |
4590 | ret = !(vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK) && | |
4591 | !(svm->vcpu.arch.hflags & HF_NMI_MASK); | |
4592 | ret = ret && gif_set(svm) && nested_svm_nmi(svm); | |
4593 | ||
4594 | return ret; | |
4595 | } | |
4596 | ||
4597 | static bool svm_get_nmi_mask(struct kvm_vcpu *vcpu) | |
4598 | { | |
4599 | struct vcpu_svm *svm = to_svm(vcpu); | |
4600 | ||
4601 | return !!(svm->vcpu.arch.hflags & HF_NMI_MASK); | |
4602 | } | |
4603 | ||
4604 | static void svm_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked) | |
4605 | { | |
4606 | struct vcpu_svm *svm = to_svm(vcpu); | |
4607 | ||
4608 | if (masked) { | |
4609 | svm->vcpu.arch.hflags |= HF_NMI_MASK; | |
4610 | set_intercept(svm, INTERCEPT_IRET); | |
4611 | } else { | |
4612 | svm->vcpu.arch.hflags &= ~HF_NMI_MASK; | |
4613 | clr_intercept(svm, INTERCEPT_IRET); | |
4614 | } | |
4615 | } | |
4616 | ||
4617 | static int svm_interrupt_allowed(struct kvm_vcpu *vcpu) | |
4618 | { | |
4619 | struct vcpu_svm *svm = to_svm(vcpu); | |
4620 | struct vmcb *vmcb = svm->vmcb; | |
4621 | int ret; | |
4622 | ||
4623 | if (!gif_set(svm) || | |
4624 | (vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK)) | |
4625 | return 0; | |
4626 | ||
4627 | ret = !!(kvm_get_rflags(vcpu) & X86_EFLAGS_IF); | |
4628 | ||
4629 | if (is_guest_mode(vcpu)) | |
4630 | return ret && !(svm->vcpu.arch.hflags & HF_VINTR_MASK); | |
4631 | ||
4632 | return ret; | |
4633 | } | |
4634 | ||
4635 | static void enable_irq_window(struct kvm_vcpu *vcpu) | |
4636 | { | |
4637 | struct vcpu_svm *svm = to_svm(vcpu); | |
4638 | ||
4639 | if (kvm_vcpu_apicv_active(vcpu)) | |
4640 | return; | |
4641 | ||
4642 | /* | |
4643 | * In case GIF=0 we can't rely on the CPU to tell us when GIF becomes | |
4644 | * 1, because that's a separate STGI/VMRUN intercept. The next time we | |
4645 | * get that intercept, this function will be called again though and | |
4646 | * we'll get the vintr intercept. | |
4647 | */ | |
4648 | if (gif_set(svm) && nested_svm_intr(svm)) { | |
4649 | svm_set_vintr(svm); | |
4650 | svm_inject_irq(svm, 0x0); | |
4651 | } | |
4652 | } | |
4653 | ||
4654 | static void enable_nmi_window(struct kvm_vcpu *vcpu) | |
4655 | { | |
4656 | struct vcpu_svm *svm = to_svm(vcpu); | |
4657 | ||
4658 | if ((svm->vcpu.arch.hflags & (HF_NMI_MASK | HF_IRET_MASK)) | |
4659 | == HF_NMI_MASK) | |
4660 | return; /* IRET will cause a vm exit */ | |
4661 | ||
4662 | /* | |
4663 | * Something prevents NMI from been injected. Single step over possible | |
4664 | * problem (IRET or exception injection or interrupt shadow) | |
4665 | */ | |
4666 | svm->nmi_singlestep = true; | |
4667 | svm->vmcb->save.rflags |= (X86_EFLAGS_TF | X86_EFLAGS_RF); | |
4668 | } | |
4669 | ||
4670 | static int svm_set_tss_addr(struct kvm *kvm, unsigned int addr) | |
4671 | { | |
4672 | return 0; | |
4673 | } | |
4674 | ||
4675 | static void svm_flush_tlb(struct kvm_vcpu *vcpu) | |
4676 | { | |
4677 | struct vcpu_svm *svm = to_svm(vcpu); | |
4678 | ||
4679 | if (static_cpu_has(X86_FEATURE_FLUSHBYASID)) | |
4680 | svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ASID; | |
4681 | else | |
4682 | svm->asid_generation--; | |
4683 | } | |
4684 | ||
4685 | static void svm_prepare_guest_switch(struct kvm_vcpu *vcpu) | |
4686 | { | |
4687 | } | |
4688 | ||
4689 | static inline void sync_cr8_to_lapic(struct kvm_vcpu *vcpu) | |
4690 | { | |
4691 | struct vcpu_svm *svm = to_svm(vcpu); | |
4692 | ||
4693 | if (svm_nested_virtualize_tpr(vcpu)) | |
4694 | return; | |
4695 | ||
4696 | if (!is_cr_intercept(svm, INTERCEPT_CR8_WRITE)) { | |
4697 | int cr8 = svm->vmcb->control.int_ctl & V_TPR_MASK; | |
4698 | kvm_set_cr8(vcpu, cr8); | |
4699 | } | |
4700 | } | |
4701 | ||
4702 | static inline void sync_lapic_to_cr8(struct kvm_vcpu *vcpu) | |
4703 | { | |
4704 | struct vcpu_svm *svm = to_svm(vcpu); | |
4705 | u64 cr8; | |
4706 | ||
4707 | if (svm_nested_virtualize_tpr(vcpu) || | |
4708 | kvm_vcpu_apicv_active(vcpu)) | |
4709 | return; | |
4710 | ||
4711 | cr8 = kvm_get_cr8(vcpu); | |
4712 | svm->vmcb->control.int_ctl &= ~V_TPR_MASK; | |
4713 | svm->vmcb->control.int_ctl |= cr8 & V_TPR_MASK; | |
4714 | } | |
4715 | ||
4716 | static void svm_complete_interrupts(struct vcpu_svm *svm) | |
4717 | { | |
4718 | u8 vector; | |
4719 | int type; | |
4720 | u32 exitintinfo = svm->vmcb->control.exit_int_info; | |
4721 | unsigned int3_injected = svm->int3_injected; | |
4722 | ||
4723 | svm->int3_injected = 0; | |
4724 | ||
4725 | /* | |
4726 | * If we've made progress since setting HF_IRET_MASK, we've | |
4727 | * executed an IRET and can allow NMI injection. | |
4728 | */ | |
4729 | if ((svm->vcpu.arch.hflags & HF_IRET_MASK) | |
4730 | && kvm_rip_read(&svm->vcpu) != svm->nmi_iret_rip) { | |
4731 | svm->vcpu.arch.hflags &= ~(HF_NMI_MASK | HF_IRET_MASK); | |
4732 | kvm_make_request(KVM_REQ_EVENT, &svm->vcpu); | |
4733 | } | |
4734 | ||
4735 | svm->vcpu.arch.nmi_injected = false; | |
4736 | kvm_clear_exception_queue(&svm->vcpu); | |
4737 | kvm_clear_interrupt_queue(&svm->vcpu); | |
4738 | ||
4739 | if (!(exitintinfo & SVM_EXITINTINFO_VALID)) | |
4740 | return; | |
4741 | ||
4742 | kvm_make_request(KVM_REQ_EVENT, &svm->vcpu); | |
4743 | ||
4744 | vector = exitintinfo & SVM_EXITINTINFO_VEC_MASK; | |
4745 | type = exitintinfo & SVM_EXITINTINFO_TYPE_MASK; | |
4746 | ||
4747 | switch (type) { | |
4748 | case SVM_EXITINTINFO_TYPE_NMI: | |
4749 | svm->vcpu.arch.nmi_injected = true; | |
4750 | break; | |
4751 | case SVM_EXITINTINFO_TYPE_EXEPT: | |
4752 | /* | |
4753 | * In case of software exceptions, do not reinject the vector, | |
4754 | * but re-execute the instruction instead. Rewind RIP first | |
4755 | * if we emulated INT3 before. | |
4756 | */ | |
4757 | if (kvm_exception_is_soft(vector)) { | |
4758 | if (vector == BP_VECTOR && int3_injected && | |
4759 | kvm_is_linear_rip(&svm->vcpu, svm->int3_rip)) | |
4760 | kvm_rip_write(&svm->vcpu, | |
4761 | kvm_rip_read(&svm->vcpu) - | |
4762 | int3_injected); | |
4763 | break; | |
4764 | } | |
4765 | if (exitintinfo & SVM_EXITINTINFO_VALID_ERR) { | |
4766 | u32 err = svm->vmcb->control.exit_int_info_err; | |
4767 | kvm_requeue_exception_e(&svm->vcpu, vector, err); | |
4768 | ||
4769 | } else | |
4770 | kvm_requeue_exception(&svm->vcpu, vector); | |
4771 | break; | |
4772 | case SVM_EXITINTINFO_TYPE_INTR: | |
4773 | kvm_queue_interrupt(&svm->vcpu, vector, false); | |
4774 | break; | |
4775 | default: | |
4776 | break; | |
4777 | } | |
4778 | } | |
4779 | ||
4780 | static void svm_cancel_injection(struct kvm_vcpu *vcpu) | |
4781 | { | |
4782 | struct vcpu_svm *svm = to_svm(vcpu); | |
4783 | struct vmcb_control_area *control = &svm->vmcb->control; | |
4784 | ||
4785 | control->exit_int_info = control->event_inj; | |
4786 | control->exit_int_info_err = control->event_inj_err; | |
4787 | control->event_inj = 0; | |
4788 | svm_complete_interrupts(svm); | |
4789 | } | |
4790 | ||
4791 | static void svm_vcpu_run(struct kvm_vcpu *vcpu) | |
4792 | { | |
4793 | struct vcpu_svm *svm = to_svm(vcpu); | |
4794 | ||
4795 | svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX]; | |
4796 | svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP]; | |
4797 | svm->vmcb->save.rip = vcpu->arch.regs[VCPU_REGS_RIP]; | |
4798 | ||
4799 | /* | |
4800 | * A vmexit emulation is required before the vcpu can be executed | |
4801 | * again. | |
4802 | */ | |
4803 | if (unlikely(svm->nested.exit_required)) | |
4804 | return; | |
4805 | ||
4806 | pre_svm_run(svm); | |
4807 | ||
4808 | sync_lapic_to_cr8(vcpu); | |
4809 | ||
4810 | svm->vmcb->save.cr2 = vcpu->arch.cr2; | |
4811 | ||
4812 | clgi(); | |
4813 | ||
4814 | local_irq_enable(); | |
4815 | ||
4816 | asm volatile ( | |
4817 | "push %%" _ASM_BP "; \n\t" | |
4818 | "mov %c[rbx](%[svm]), %%" _ASM_BX " \n\t" | |
4819 | "mov %c[rcx](%[svm]), %%" _ASM_CX " \n\t" | |
4820 | "mov %c[rdx](%[svm]), %%" _ASM_DX " \n\t" | |
4821 | "mov %c[rsi](%[svm]), %%" _ASM_SI " \n\t" | |
4822 | "mov %c[rdi](%[svm]), %%" _ASM_DI " \n\t" | |
4823 | "mov %c[rbp](%[svm]), %%" _ASM_BP " \n\t" | |
4824 | #ifdef CONFIG_X86_64 | |
4825 | "mov %c[r8](%[svm]), %%r8 \n\t" | |
4826 | "mov %c[r9](%[svm]), %%r9 \n\t" | |
4827 | "mov %c[r10](%[svm]), %%r10 \n\t" | |
4828 | "mov %c[r11](%[svm]), %%r11 \n\t" | |
4829 | "mov %c[r12](%[svm]), %%r12 \n\t" | |
4830 | "mov %c[r13](%[svm]), %%r13 \n\t" | |
4831 | "mov %c[r14](%[svm]), %%r14 \n\t" | |
4832 | "mov %c[r15](%[svm]), %%r15 \n\t" | |
4833 | #endif | |
4834 | ||
4835 | /* Enter guest mode */ | |
4836 | "push %%" _ASM_AX " \n\t" | |
4837 | "mov %c[vmcb](%[svm]), %%" _ASM_AX " \n\t" | |
4838 | __ex(SVM_VMLOAD) "\n\t" | |
4839 | __ex(SVM_VMRUN) "\n\t" | |
4840 | __ex(SVM_VMSAVE) "\n\t" | |
4841 | "pop %%" _ASM_AX " \n\t" | |
4842 | ||
4843 | /* Save guest registers, load host registers */ | |
4844 | "mov %%" _ASM_BX ", %c[rbx](%[svm]) \n\t" | |
4845 | "mov %%" _ASM_CX ", %c[rcx](%[svm]) \n\t" | |
4846 | "mov %%" _ASM_DX ", %c[rdx](%[svm]) \n\t" | |
4847 | "mov %%" _ASM_SI ", %c[rsi](%[svm]) \n\t" | |
4848 | "mov %%" _ASM_DI ", %c[rdi](%[svm]) \n\t" | |
4849 | "mov %%" _ASM_BP ", %c[rbp](%[svm]) \n\t" | |
4850 | #ifdef CONFIG_X86_64 | |
4851 | "mov %%r8, %c[r8](%[svm]) \n\t" | |
4852 | "mov %%r9, %c[r9](%[svm]) \n\t" | |
4853 | "mov %%r10, %c[r10](%[svm]) \n\t" | |
4854 | "mov %%r11, %c[r11](%[svm]) \n\t" | |
4855 | "mov %%r12, %c[r12](%[svm]) \n\t" | |
4856 | "mov %%r13, %c[r13](%[svm]) \n\t" | |
4857 | "mov %%r14, %c[r14](%[svm]) \n\t" | |
4858 | "mov %%r15, %c[r15](%[svm]) \n\t" | |
4859 | #endif | |
4860 | "pop %%" _ASM_BP | |
4861 | : | |
4862 | : [svm]"a"(svm), | |
4863 | [vmcb]"i"(offsetof(struct vcpu_svm, vmcb_pa)), | |
4864 | [rbx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBX])), | |
4865 | [rcx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RCX])), | |
4866 | [rdx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDX])), | |
4867 | [rsi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RSI])), | |
4868 | [rdi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDI])), | |
4869 | [rbp]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBP])) | |
4870 | #ifdef CONFIG_X86_64 | |
4871 | , [r8]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R8])), | |
4872 | [r9]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R9])), | |
4873 | [r10]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R10])), | |
4874 | [r11]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R11])), | |
4875 | [r12]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R12])), | |
4876 | [r13]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R13])), | |
4877 | [r14]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R14])), | |
4878 | [r15]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R15])) | |
4879 | #endif | |
4880 | : "cc", "memory" | |
4881 | #ifdef CONFIG_X86_64 | |
4882 | , "rbx", "rcx", "rdx", "rsi", "rdi" | |
4883 | , "r8", "r9", "r10", "r11" , "r12", "r13", "r14", "r15" | |
4884 | #else | |
4885 | , "ebx", "ecx", "edx", "esi", "edi" | |
4886 | #endif | |
4887 | ); | |
4888 | ||
4889 | #ifdef CONFIG_X86_64 | |
4890 | wrmsrl(MSR_GS_BASE, svm->host.gs_base); | |
4891 | #else | |
4892 | loadsegment(fs, svm->host.fs); | |
4893 | #ifndef CONFIG_X86_32_LAZY_GS | |
4894 | loadsegment(gs, svm->host.gs); | |
4895 | #endif | |
4896 | #endif | |
4897 | ||
4898 | reload_tss(vcpu); | |
4899 | ||
4900 | local_irq_disable(); | |
4901 | ||
4902 | vcpu->arch.cr2 = svm->vmcb->save.cr2; | |
4903 | vcpu->arch.regs[VCPU_REGS_RAX] = svm->vmcb->save.rax; | |
4904 | vcpu->arch.regs[VCPU_REGS_RSP] = svm->vmcb->save.rsp; | |
4905 | vcpu->arch.regs[VCPU_REGS_RIP] = svm->vmcb->save.rip; | |
4906 | ||
4907 | if (unlikely(svm->vmcb->control.exit_code == SVM_EXIT_NMI)) | |
4908 | kvm_before_handle_nmi(&svm->vcpu); | |
4909 | ||
4910 | stgi(); | |
4911 | ||
4912 | /* Any pending NMI will happen here */ | |
4913 | ||
4914 | if (unlikely(svm->vmcb->control.exit_code == SVM_EXIT_NMI)) | |
4915 | kvm_after_handle_nmi(&svm->vcpu); | |
4916 | ||
4917 | sync_cr8_to_lapic(vcpu); | |
4918 | ||
4919 | svm->next_rip = 0; | |
4920 | ||
4921 | svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING; | |
4922 | ||
4923 | /* if exit due to PF check for async PF */ | |
4924 | if (svm->vmcb->control.exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR) | |
4925 | svm->apf_reason = kvm_read_and_reset_pf_reason(); | |
4926 | ||
4927 | if (npt_enabled) { | |
4928 | vcpu->arch.regs_avail &= ~(1 << VCPU_EXREG_PDPTR); | |
4929 | vcpu->arch.regs_dirty &= ~(1 << VCPU_EXREG_PDPTR); | |
4930 | } | |
4931 | ||
4932 | /* | |
4933 | * We need to handle MC intercepts here before the vcpu has a chance to | |
4934 | * change the physical cpu | |
4935 | */ | |
4936 | if (unlikely(svm->vmcb->control.exit_code == | |
4937 | SVM_EXIT_EXCP_BASE + MC_VECTOR)) | |
4938 | svm_handle_mce(svm); | |
4939 | ||
4940 | mark_all_clean(svm->vmcb); | |
4941 | } | |
4942 | ||
4943 | static void svm_set_cr3(struct kvm_vcpu *vcpu, unsigned long root) | |
4944 | { | |
4945 | struct vcpu_svm *svm = to_svm(vcpu); | |
4946 | ||
4947 | svm->vmcb->save.cr3 = root; | |
4948 | mark_dirty(svm->vmcb, VMCB_CR); | |
4949 | svm_flush_tlb(vcpu); | |
4950 | } | |
4951 | ||
4952 | static void set_tdp_cr3(struct kvm_vcpu *vcpu, unsigned long root) | |
4953 | { | |
4954 | struct vcpu_svm *svm = to_svm(vcpu); | |
4955 | ||
4956 | svm->vmcb->control.nested_cr3 = root; | |
4957 | mark_dirty(svm->vmcb, VMCB_NPT); | |
4958 | ||
4959 | /* Also sync guest cr3 here in case we live migrate */ | |
4960 | svm->vmcb->save.cr3 = kvm_read_cr3(vcpu); | |
4961 | mark_dirty(svm->vmcb, VMCB_CR); | |
4962 | ||
4963 | svm_flush_tlb(vcpu); | |
4964 | } | |
4965 | ||
4966 | static int is_disabled(void) | |
4967 | { | |
4968 | u64 vm_cr; | |
4969 | ||
4970 | rdmsrl(MSR_VM_CR, vm_cr); | |
4971 | if (vm_cr & (1 << SVM_VM_CR_SVM_DISABLE)) | |
4972 | return 1; | |
4973 | ||
4974 | return 0; | |
4975 | } | |
4976 | ||
4977 | static void | |
4978 | svm_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall) | |
4979 | { | |
4980 | /* | |
4981 | * Patch in the VMMCALL instruction: | |
4982 | */ | |
4983 | hypercall[0] = 0x0f; | |
4984 | hypercall[1] = 0x01; | |
4985 | hypercall[2] = 0xd9; | |
4986 | } | |
4987 | ||
4988 | static void svm_check_processor_compat(void *rtn) | |
4989 | { | |
4990 | *(int *)rtn = 0; | |
4991 | } | |
4992 | ||
4993 | static bool svm_cpu_has_accelerated_tpr(void) | |
4994 | { | |
4995 | return false; | |
4996 | } | |
4997 | ||
4998 | static bool svm_has_high_real_mode_segbase(void) | |
4999 | { | |
5000 | return true; | |
5001 | } | |
5002 | ||
5003 | static u64 svm_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio) | |
5004 | { | |
5005 | return 0; | |
5006 | } | |
5007 | ||
5008 | static void svm_cpuid_update(struct kvm_vcpu *vcpu) | |
5009 | { | |
5010 | struct vcpu_svm *svm = to_svm(vcpu); | |
5011 | struct kvm_cpuid_entry2 *entry; | |
5012 | ||
5013 | /* Update nrips enabled cache */ | |
5014 | svm->nrips_enabled = !!guest_cpuid_has_nrips(&svm->vcpu); | |
5015 | ||
5016 | if (!kvm_vcpu_apicv_active(vcpu)) | |
5017 | return; | |
5018 | ||
5019 | entry = kvm_find_cpuid_entry(vcpu, 1, 0); | |
5020 | if (entry) | |
5021 | entry->ecx &= ~bit(X86_FEATURE_X2APIC); | |
5022 | } | |
5023 | ||
5024 | static void svm_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry) | |
5025 | { | |
5026 | switch (func) { | |
5027 | case 0x1: | |
5028 | if (avic) | |
5029 | entry->ecx &= ~bit(X86_FEATURE_X2APIC); | |
5030 | break; | |
5031 | case 0x80000001: | |
5032 | if (nested) | |
5033 | entry->ecx |= (1 << 2); /* Set SVM bit */ | |
5034 | break; | |
5035 | case 0x8000000A: | |
5036 | entry->eax = 1; /* SVM revision 1 */ | |
5037 | entry->ebx = 8; /* Lets support 8 ASIDs in case we add proper | |
5038 | ASID emulation to nested SVM */ | |
5039 | entry->ecx = 0; /* Reserved */ | |
5040 | entry->edx = 0; /* Per default do not support any | |
5041 | additional features */ | |
5042 | ||
5043 | /* Support next_rip if host supports it */ | |
5044 | if (boot_cpu_has(X86_FEATURE_NRIPS)) | |
5045 | entry->edx |= SVM_FEATURE_NRIP; | |
5046 | ||
5047 | /* Support NPT for the guest if enabled */ | |
5048 | if (npt_enabled) | |
5049 | entry->edx |= SVM_FEATURE_NPT; | |
5050 | ||
5051 | break; | |
5052 | } | |
5053 | } | |
5054 | ||
5055 | static int svm_get_lpage_level(void) | |
5056 | { | |
5057 | return PT_PDPE_LEVEL; | |
5058 | } | |
5059 | ||
5060 | static bool svm_rdtscp_supported(void) | |
5061 | { | |
5062 | return boot_cpu_has(X86_FEATURE_RDTSCP); | |
5063 | } | |
5064 | ||
5065 | static bool svm_invpcid_supported(void) | |
5066 | { | |
5067 | return false; | |
5068 | } | |
5069 | ||
5070 | static bool svm_mpx_supported(void) | |
5071 | { | |
5072 | return false; | |
5073 | } | |
5074 | ||
5075 | static bool svm_xsaves_supported(void) | |
5076 | { | |
5077 | return false; | |
5078 | } | |
5079 | ||
5080 | static bool svm_has_wbinvd_exit(void) | |
5081 | { | |
5082 | return true; | |
5083 | } | |
5084 | ||
5085 | static void svm_fpu_deactivate(struct kvm_vcpu *vcpu) | |
5086 | { | |
5087 | struct vcpu_svm *svm = to_svm(vcpu); | |
5088 | ||
5089 | set_exception_intercept(svm, NM_VECTOR); | |
5090 | update_cr0_intercept(svm); | |
5091 | } | |
5092 | ||
5093 | #define PRE_EX(exit) { .exit_code = (exit), \ | |
5094 | .stage = X86_ICPT_PRE_EXCEPT, } | |
5095 | #define POST_EX(exit) { .exit_code = (exit), \ | |
5096 | .stage = X86_ICPT_POST_EXCEPT, } | |
5097 | #define POST_MEM(exit) { .exit_code = (exit), \ | |
5098 | .stage = X86_ICPT_POST_MEMACCESS, } | |
5099 | ||
5100 | static const struct __x86_intercept { | |
5101 | u32 exit_code; | |
5102 | enum x86_intercept_stage stage; | |
5103 | } x86_intercept_map[] = { | |
5104 | [x86_intercept_cr_read] = POST_EX(SVM_EXIT_READ_CR0), | |
5105 | [x86_intercept_cr_write] = POST_EX(SVM_EXIT_WRITE_CR0), | |
5106 | [x86_intercept_clts] = POST_EX(SVM_EXIT_WRITE_CR0), | |
5107 | [x86_intercept_lmsw] = POST_EX(SVM_EXIT_WRITE_CR0), | |
5108 | [x86_intercept_smsw] = POST_EX(SVM_EXIT_READ_CR0), | |
5109 | [x86_intercept_dr_read] = POST_EX(SVM_EXIT_READ_DR0), | |
5110 | [x86_intercept_dr_write] = POST_EX(SVM_EXIT_WRITE_DR0), | |
5111 | [x86_intercept_sldt] = POST_EX(SVM_EXIT_LDTR_READ), | |
5112 | [x86_intercept_str] = POST_EX(SVM_EXIT_TR_READ), | |
5113 | [x86_intercept_lldt] = POST_EX(SVM_EXIT_LDTR_WRITE), | |
5114 | [x86_intercept_ltr] = POST_EX(SVM_EXIT_TR_WRITE), | |
5115 | [x86_intercept_sgdt] = POST_EX(SVM_EXIT_GDTR_READ), | |
5116 | [x86_intercept_sidt] = POST_EX(SVM_EXIT_IDTR_READ), | |
5117 | [x86_intercept_lgdt] = POST_EX(SVM_EXIT_GDTR_WRITE), | |
5118 | [x86_intercept_lidt] = POST_EX(SVM_EXIT_IDTR_WRITE), | |
5119 | [x86_intercept_vmrun] = POST_EX(SVM_EXIT_VMRUN), | |
5120 | [x86_intercept_vmmcall] = POST_EX(SVM_EXIT_VMMCALL), | |
5121 | [x86_intercept_vmload] = POST_EX(SVM_EXIT_VMLOAD), | |
5122 | [x86_intercept_vmsave] = POST_EX(SVM_EXIT_VMSAVE), | |
5123 | [x86_intercept_stgi] = POST_EX(SVM_EXIT_STGI), | |
5124 | [x86_intercept_clgi] = POST_EX(SVM_EXIT_CLGI), | |
5125 | [x86_intercept_skinit] = POST_EX(SVM_EXIT_SKINIT), | |
5126 | [x86_intercept_invlpga] = POST_EX(SVM_EXIT_INVLPGA), | |
5127 | [x86_intercept_rdtscp] = POST_EX(SVM_EXIT_RDTSCP), | |
5128 | [x86_intercept_monitor] = POST_MEM(SVM_EXIT_MONITOR), | |
5129 | [x86_intercept_mwait] = POST_EX(SVM_EXIT_MWAIT), | |
5130 | [x86_intercept_invlpg] = POST_EX(SVM_EXIT_INVLPG), | |
5131 | [x86_intercept_invd] = POST_EX(SVM_EXIT_INVD), | |
5132 | [x86_intercept_wbinvd] = POST_EX(SVM_EXIT_WBINVD), | |
5133 | [x86_intercept_wrmsr] = POST_EX(SVM_EXIT_MSR), | |
5134 | [x86_intercept_rdtsc] = POST_EX(SVM_EXIT_RDTSC), | |
5135 | [x86_intercept_rdmsr] = POST_EX(SVM_EXIT_MSR), | |
5136 | [x86_intercept_rdpmc] = POST_EX(SVM_EXIT_RDPMC), | |
5137 | [x86_intercept_cpuid] = PRE_EX(SVM_EXIT_CPUID), | |
5138 | [x86_intercept_rsm] = PRE_EX(SVM_EXIT_RSM), | |
5139 | [x86_intercept_pause] = PRE_EX(SVM_EXIT_PAUSE), | |
5140 | [x86_intercept_pushf] = PRE_EX(SVM_EXIT_PUSHF), | |
5141 | [x86_intercept_popf] = PRE_EX(SVM_EXIT_POPF), | |
5142 | [x86_intercept_intn] = PRE_EX(SVM_EXIT_SWINT), | |
5143 | [x86_intercept_iret] = PRE_EX(SVM_EXIT_IRET), | |
5144 | [x86_intercept_icebp] = PRE_EX(SVM_EXIT_ICEBP), | |
5145 | [x86_intercept_hlt] = POST_EX(SVM_EXIT_HLT), | |
5146 | [x86_intercept_in] = POST_EX(SVM_EXIT_IOIO), | |
5147 | [x86_intercept_ins] = POST_EX(SVM_EXIT_IOIO), | |
5148 | [x86_intercept_out] = POST_EX(SVM_EXIT_IOIO), | |
5149 | [x86_intercept_outs] = POST_EX(SVM_EXIT_IOIO), | |
5150 | }; | |
5151 | ||
5152 | #undef PRE_EX | |
5153 | #undef POST_EX | |
5154 | #undef POST_MEM | |
5155 | ||
5156 | static int svm_check_intercept(struct kvm_vcpu *vcpu, | |
5157 | struct x86_instruction_info *info, | |
5158 | enum x86_intercept_stage stage) | |
5159 | { | |
5160 | struct vcpu_svm *svm = to_svm(vcpu); | |
5161 | int vmexit, ret = X86EMUL_CONTINUE; | |
5162 | struct __x86_intercept icpt_info; | |
5163 | struct vmcb *vmcb = svm->vmcb; | |
5164 | ||
5165 | if (info->intercept >= ARRAY_SIZE(x86_intercept_map)) | |
5166 | goto out; | |
5167 | ||
5168 | icpt_info = x86_intercept_map[info->intercept]; | |
5169 | ||
5170 | if (stage != icpt_info.stage) | |
5171 | goto out; | |
5172 | ||
5173 | switch (icpt_info.exit_code) { | |
5174 | case SVM_EXIT_READ_CR0: | |
5175 | if (info->intercept == x86_intercept_cr_read) | |
5176 | icpt_info.exit_code += info->modrm_reg; | |
5177 | break; | |
5178 | case SVM_EXIT_WRITE_CR0: { | |
5179 | unsigned long cr0, val; | |
5180 | u64 intercept; | |
5181 | ||
5182 | if (info->intercept == x86_intercept_cr_write) | |
5183 | icpt_info.exit_code += info->modrm_reg; | |
5184 | ||
5185 | if (icpt_info.exit_code != SVM_EXIT_WRITE_CR0 || | |
5186 | info->intercept == x86_intercept_clts) | |
5187 | break; | |
5188 | ||
5189 | intercept = svm->nested.intercept; | |
5190 | ||
5191 | if (!(intercept & (1ULL << INTERCEPT_SELECTIVE_CR0))) | |
5192 | break; | |
5193 | ||
5194 | cr0 = vcpu->arch.cr0 & ~SVM_CR0_SELECTIVE_MASK; | |
5195 | val = info->src_val & ~SVM_CR0_SELECTIVE_MASK; | |
5196 | ||
5197 | if (info->intercept == x86_intercept_lmsw) { | |
5198 | cr0 &= 0xfUL; | |
5199 | val &= 0xfUL; | |
5200 | /* lmsw can't clear PE - catch this here */ | |
5201 | if (cr0 & X86_CR0_PE) | |
5202 | val |= X86_CR0_PE; | |
5203 | } | |
5204 | ||
5205 | if (cr0 ^ val) | |
5206 | icpt_info.exit_code = SVM_EXIT_CR0_SEL_WRITE; | |
5207 | ||
5208 | break; | |
5209 | } | |
5210 | case SVM_EXIT_READ_DR0: | |
5211 | case SVM_EXIT_WRITE_DR0: | |
5212 | icpt_info.exit_code += info->modrm_reg; | |
5213 | break; | |
5214 | case SVM_EXIT_MSR: | |
5215 | if (info->intercept == x86_intercept_wrmsr) | |
5216 | vmcb->control.exit_info_1 = 1; | |
5217 | else | |
5218 | vmcb->control.exit_info_1 = 0; | |
5219 | break; | |
5220 | case SVM_EXIT_PAUSE: | |
5221 | /* | |
5222 | * We get this for NOP only, but pause | |
5223 | * is rep not, check this here | |
5224 | */ | |
5225 | if (info->rep_prefix != REPE_PREFIX) | |
5226 | goto out; | |
5227 | case SVM_EXIT_IOIO: { | |
5228 | u64 exit_info; | |
5229 | u32 bytes; | |
5230 | ||
5231 | if (info->intercept == x86_intercept_in || | |
5232 | info->intercept == x86_intercept_ins) { | |
5233 | exit_info = ((info->src_val & 0xffff) << 16) | | |
5234 | SVM_IOIO_TYPE_MASK; | |
5235 | bytes = info->dst_bytes; | |
5236 | } else { | |
5237 | exit_info = (info->dst_val & 0xffff) << 16; | |
5238 | bytes = info->src_bytes; | |
5239 | } | |
5240 | ||
5241 | if (info->intercept == x86_intercept_outs || | |
5242 | info->intercept == x86_intercept_ins) | |
5243 | exit_info |= SVM_IOIO_STR_MASK; | |
5244 | ||
5245 | if (info->rep_prefix) | |
5246 | exit_info |= SVM_IOIO_REP_MASK; | |
5247 | ||
5248 | bytes = min(bytes, 4u); | |
5249 | ||
5250 | exit_info |= bytes << SVM_IOIO_SIZE_SHIFT; | |
5251 | ||
5252 | exit_info |= (u32)info->ad_bytes << (SVM_IOIO_ASIZE_SHIFT - 1); | |
5253 | ||
5254 | vmcb->control.exit_info_1 = exit_info; | |
5255 | vmcb->control.exit_info_2 = info->next_rip; | |
5256 | ||
5257 | break; | |
5258 | } | |
5259 | default: | |
5260 | break; | |
5261 | } | |
5262 | ||
5263 | /* TODO: Advertise NRIPS to guest hypervisor unconditionally */ | |
5264 | if (static_cpu_has(X86_FEATURE_NRIPS)) | |
5265 | vmcb->control.next_rip = info->next_rip; | |
5266 | vmcb->control.exit_code = icpt_info.exit_code; | |
5267 | vmexit = nested_svm_exit_handled(svm); | |
5268 | ||
5269 | ret = (vmexit == NESTED_EXIT_DONE) ? X86EMUL_INTERCEPTED | |
5270 | : X86EMUL_CONTINUE; | |
5271 | ||
5272 | out: | |
5273 | return ret; | |
5274 | } | |
5275 | ||
5276 | static void svm_handle_external_intr(struct kvm_vcpu *vcpu) | |
5277 | { | |
5278 | local_irq_enable(); | |
5279 | /* | |
5280 | * We must have an instruction with interrupts enabled, so | |
5281 | * the timer interrupt isn't delayed by the interrupt shadow. | |
5282 | */ | |
5283 | asm("nop"); | |
5284 | local_irq_disable(); | |
5285 | } | |
5286 | ||
5287 | static void svm_sched_in(struct kvm_vcpu *vcpu, int cpu) | |
5288 | { | |
5289 | } | |
5290 | ||
5291 | static inline void avic_post_state_restore(struct kvm_vcpu *vcpu) | |
5292 | { | |
5293 | if (avic_handle_apic_id_update(vcpu) != 0) | |
5294 | return; | |
5295 | if (avic_handle_dfr_update(vcpu) != 0) | |
5296 | return; | |
5297 | avic_handle_ldr_update(vcpu); | |
5298 | } | |
5299 | ||
5300 | static struct kvm_x86_ops svm_x86_ops __ro_after_init = { | |
5301 | .cpu_has_kvm_support = has_svm, | |
5302 | .disabled_by_bios = is_disabled, | |
5303 | .hardware_setup = svm_hardware_setup, | |
5304 | .hardware_unsetup = svm_hardware_unsetup, | |
5305 | .check_processor_compatibility = svm_check_processor_compat, | |
5306 | .hardware_enable = svm_hardware_enable, | |
5307 | .hardware_disable = svm_hardware_disable, | |
5308 | .cpu_has_accelerated_tpr = svm_cpu_has_accelerated_tpr, | |
5309 | .cpu_has_high_real_mode_segbase = svm_has_high_real_mode_segbase, | |
5310 | ||
5311 | .vcpu_create = svm_create_vcpu, | |
5312 | .vcpu_free = svm_free_vcpu, | |
5313 | .vcpu_reset = svm_vcpu_reset, | |
5314 | ||
5315 | .vm_init = avic_vm_init, | |
5316 | .vm_destroy = avic_vm_destroy, | |
5317 | ||
5318 | .prepare_guest_switch = svm_prepare_guest_switch, | |
5319 | .vcpu_load = svm_vcpu_load, | |
5320 | .vcpu_put = svm_vcpu_put, | |
5321 | .vcpu_blocking = svm_vcpu_blocking, | |
5322 | .vcpu_unblocking = svm_vcpu_unblocking, | |
5323 | ||
5324 | .update_bp_intercept = update_bp_intercept, | |
5325 | .get_msr = svm_get_msr, | |
5326 | .set_msr = svm_set_msr, | |
5327 | .get_segment_base = svm_get_segment_base, | |
5328 | .get_segment = svm_get_segment, | |
5329 | .set_segment = svm_set_segment, | |
5330 | .get_cpl = svm_get_cpl, | |
5331 | .get_cs_db_l_bits = kvm_get_cs_db_l_bits, | |
5332 | .decache_cr0_guest_bits = svm_decache_cr0_guest_bits, | |
5333 | .decache_cr3 = svm_decache_cr3, | |
5334 | .decache_cr4_guest_bits = svm_decache_cr4_guest_bits, | |
5335 | .set_cr0 = svm_set_cr0, | |
5336 | .set_cr3 = svm_set_cr3, | |
5337 | .set_cr4 = svm_set_cr4, | |
5338 | .set_efer = svm_set_efer, | |
5339 | .get_idt = svm_get_idt, | |
5340 | .set_idt = svm_set_idt, | |
5341 | .get_gdt = svm_get_gdt, | |
5342 | .set_gdt = svm_set_gdt, | |
5343 | .get_dr6 = svm_get_dr6, | |
5344 | .set_dr6 = svm_set_dr6, | |
5345 | .set_dr7 = svm_set_dr7, | |
5346 | .sync_dirty_debug_regs = svm_sync_dirty_debug_regs, | |
5347 | .cache_reg = svm_cache_reg, | |
5348 | .get_rflags = svm_get_rflags, | |
5349 | .set_rflags = svm_set_rflags, | |
5350 | ||
5351 | .get_pkru = svm_get_pkru, | |
5352 | ||
5353 | .fpu_activate = svm_fpu_activate, | |
5354 | .fpu_deactivate = svm_fpu_deactivate, | |
5355 | ||
5356 | .tlb_flush = svm_flush_tlb, | |
5357 | ||
5358 | .run = svm_vcpu_run, | |
5359 | .handle_exit = handle_exit, | |
5360 | .skip_emulated_instruction = skip_emulated_instruction, | |
5361 | .set_interrupt_shadow = svm_set_interrupt_shadow, | |
5362 | .get_interrupt_shadow = svm_get_interrupt_shadow, | |
5363 | .patch_hypercall = svm_patch_hypercall, | |
5364 | .set_irq = svm_set_irq, | |
5365 | .set_nmi = svm_inject_nmi, | |
5366 | .queue_exception = svm_queue_exception, | |
5367 | .cancel_injection = svm_cancel_injection, | |
5368 | .interrupt_allowed = svm_interrupt_allowed, | |
5369 | .nmi_allowed = svm_nmi_allowed, | |
5370 | .get_nmi_mask = svm_get_nmi_mask, | |
5371 | .set_nmi_mask = svm_set_nmi_mask, | |
5372 | .enable_nmi_window = enable_nmi_window, | |
5373 | .enable_irq_window = enable_irq_window, | |
5374 | .update_cr8_intercept = update_cr8_intercept, | |
5375 | .set_virtual_x2apic_mode = svm_set_virtual_x2apic_mode, | |
5376 | .get_enable_apicv = svm_get_enable_apicv, | |
5377 | .refresh_apicv_exec_ctrl = svm_refresh_apicv_exec_ctrl, | |
5378 | .load_eoi_exitmap = svm_load_eoi_exitmap, | |
5379 | .sync_pir_to_irr = svm_sync_pir_to_irr, | |
5380 | .hwapic_irr_update = svm_hwapic_irr_update, | |
5381 | .hwapic_isr_update = svm_hwapic_isr_update, | |
5382 | .apicv_post_state_restore = avic_post_state_restore, | |
5383 | ||
5384 | .set_tss_addr = svm_set_tss_addr, | |
5385 | .get_tdp_level = get_npt_level, | |
5386 | .get_mt_mask = svm_get_mt_mask, | |
5387 | ||
5388 | .get_exit_info = svm_get_exit_info, | |
5389 | ||
5390 | .get_lpage_level = svm_get_lpage_level, | |
5391 | ||
5392 | .cpuid_update = svm_cpuid_update, | |
5393 | ||
5394 | .rdtscp_supported = svm_rdtscp_supported, | |
5395 | .invpcid_supported = svm_invpcid_supported, | |
5396 | .mpx_supported = svm_mpx_supported, | |
5397 | .xsaves_supported = svm_xsaves_supported, | |
5398 | ||
5399 | .set_supported_cpuid = svm_set_supported_cpuid, | |
5400 | ||
5401 | .has_wbinvd_exit = svm_has_wbinvd_exit, | |
5402 | ||
5403 | .write_tsc_offset = svm_write_tsc_offset, | |
5404 | ||
5405 | .set_tdp_cr3 = set_tdp_cr3, | |
5406 | ||
5407 | .check_intercept = svm_check_intercept, | |
5408 | .handle_external_intr = svm_handle_external_intr, | |
5409 | ||
5410 | .sched_in = svm_sched_in, | |
5411 | ||
5412 | .pmu_ops = &amd_pmu_ops, | |
5413 | .deliver_posted_interrupt = svm_deliver_avic_intr, | |
5414 | .update_pi_irte = svm_update_pi_irte, | |
5415 | }; | |
5416 | ||
5417 | static int __init svm_init(void) | |
5418 | { | |
5419 | return kvm_init(&svm_x86_ops, sizeof(struct vcpu_svm), | |
5420 | __alignof__(struct vcpu_svm), THIS_MODULE); | |
5421 | } | |
5422 | ||
5423 | static void __exit svm_exit(void) | |
5424 | { | |
5425 | kvm_exit(); | |
5426 | } | |
5427 | ||
5428 | module_init(svm_init) | |
5429 | module_exit(svm_exit) |