]>
Commit | Line | Data |
---|---|---|
20c8ccb1 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
6aa8b732 AK |
2 | /* |
3 | * Kernel-based Virtual Machine driver for Linux | |
4 | * | |
5 | * AMD SVM support | |
6 | * | |
7 | * Copyright (C) 2006 Qumranet, Inc. | |
9611c187 | 8 | * Copyright 2010 Red Hat, Inc. and/or its affiliates. |
6aa8b732 AK |
9 | * |
10 | * Authors: | |
11 | * Yaniv Kamay <yaniv@qumranet.com> | |
12 | * Avi Kivity <avi@qumranet.com> | |
6aa8b732 | 13 | */ |
44a95dae SS |
14 | |
15 | #define pr_fmt(fmt) "SVM: " fmt | |
16 | ||
edf88417 AK |
17 | #include <linux/kvm_host.h> |
18 | ||
85f455f7 | 19 | #include "irq.h" |
1d737c8a | 20 | #include "mmu.h" |
5fdbf976 | 21 | #include "kvm_cache_regs.h" |
fe4c7b19 | 22 | #include "x86.h" |
66f7b72e | 23 | #include "cpuid.h" |
25462f7f | 24 | #include "pmu.h" |
e495606d | 25 | |
6aa8b732 | 26 | #include <linux/module.h> |
ae759544 | 27 | #include <linux/mod_devicetable.h> |
9d8f549d | 28 | #include <linux/kernel.h> |
6aa8b732 AK |
29 | #include <linux/vmalloc.h> |
30 | #include <linux/highmem.h> | |
e8edc6e0 | 31 | #include <linux/sched.h> |
af658dca | 32 | #include <linux/trace_events.h> |
5a0e3ad6 | 33 | #include <linux/slab.h> |
5881f737 SS |
34 | #include <linux/amd-iommu.h> |
35 | #include <linux/hashtable.h> | |
c207aee4 | 36 | #include <linux/frame.h> |
e9df0942 | 37 | #include <linux/psp-sev.h> |
1654efcb | 38 | #include <linux/file.h> |
89c50580 BS |
39 | #include <linux/pagemap.h> |
40 | #include <linux/swap.h> | |
6aa8b732 | 41 | |
8221c137 | 42 | #include <asm/apic.h> |
1018faa6 | 43 | #include <asm/perf_event.h> |
67ec6607 | 44 | #include <asm/tlbflush.h> |
e495606d | 45 | #include <asm/desc.h> |
facb0139 | 46 | #include <asm/debugreg.h> |
631bc487 | 47 | #include <asm/kvm_para.h> |
411b44ba | 48 | #include <asm/irq_remapping.h> |
28a27752 | 49 | #include <asm/spec-ctrl.h> |
6aa8b732 | 50 | |
63d1142f | 51 | #include <asm/virtext.h> |
229456fc | 52 | #include "trace.h" |
63d1142f | 53 | |
4ecac3fd AK |
54 | #define __ex(x) __kvm_handle_fault_on_reboot(x) |
55 | ||
6aa8b732 AK |
56 | MODULE_AUTHOR("Qumranet"); |
57 | MODULE_LICENSE("GPL"); | |
58 | ||
ae759544 JT |
59 | static const struct x86_cpu_id svm_cpu_id[] = { |
60 | X86_FEATURE_MATCH(X86_FEATURE_SVM), | |
61 | {} | |
62 | }; | |
63 | MODULE_DEVICE_TABLE(x86cpu, svm_cpu_id); | |
64 | ||
6aa8b732 AK |
65 | #define IOPM_ALLOC_ORDER 2 |
66 | #define MSRPM_ALLOC_ORDER 1 | |
67 | ||
6aa8b732 AK |
68 | #define SEG_TYPE_LDT 2 |
69 | #define SEG_TYPE_BUSY_TSS16 3 | |
70 | ||
6bc31bdc AP |
71 | #define SVM_FEATURE_NPT (1 << 0) |
72 | #define SVM_FEATURE_LBRV (1 << 1) | |
73 | #define SVM_FEATURE_SVML (1 << 2) | |
74 | #define SVM_FEATURE_NRIP (1 << 3) | |
ddce97aa AP |
75 | #define SVM_FEATURE_TSC_RATE (1 << 4) |
76 | #define SVM_FEATURE_VMCB_CLEAN (1 << 5) | |
77 | #define SVM_FEATURE_FLUSH_ASID (1 << 6) | |
78 | #define SVM_FEATURE_DECODE_ASSIST (1 << 7) | |
6bc31bdc | 79 | #define SVM_FEATURE_PAUSE_FILTER (1 << 10) |
80b7706e | 80 | |
340d3bc3 SS |
81 | #define SVM_AVIC_DOORBELL 0xc001011b |
82 | ||
410e4d57 JR |
83 | #define NESTED_EXIT_HOST 0 /* Exit handled on host level */ |
84 | #define NESTED_EXIT_DONE 1 /* Exit caused nested vmexit */ | |
85 | #define NESTED_EXIT_CONTINUE 2 /* Further checks needed */ | |
86 | ||
24e09cbf JR |
87 | #define DEBUGCTL_RESERVED_BITS (~(0x3fULL)) |
88 | ||
fbc0db76 | 89 | #define TSC_RATIO_RSVD 0xffffff0000000000ULL |
92a1f12d JR |
90 | #define TSC_RATIO_MIN 0x0000000000000001ULL |
91 | #define TSC_RATIO_MAX 0x000000ffffffffffULL | |
fbc0db76 | 92 | |
5446a979 | 93 | #define AVIC_HPA_MASK ~((0xFFFULL << 52) | 0xFFF) |
44a95dae SS |
94 | |
95 | /* | |
96 | * 0xff is broadcast, so the max index allowed for physical APIC ID | |
97 | * table is 0xfe. APIC IDs above 0xff are reserved. | |
98 | */ | |
99 | #define AVIC_MAX_PHYSICAL_ID_COUNT 255 | |
100 | ||
18f40c53 SS |
101 | #define AVIC_UNACCEL_ACCESS_WRITE_MASK 1 |
102 | #define AVIC_UNACCEL_ACCESS_OFFSET_MASK 0xFF0 | |
103 | #define AVIC_UNACCEL_ACCESS_VECTOR_MASK 0xFFFFFFFF | |
104 | ||
5ea11f2b SS |
105 | /* AVIC GATAG is encoded using VM and VCPU IDs */ |
106 | #define AVIC_VCPU_ID_BITS 8 | |
107 | #define AVIC_VCPU_ID_MASK ((1 << AVIC_VCPU_ID_BITS) - 1) | |
108 | ||
109 | #define AVIC_VM_ID_BITS 24 | |
110 | #define AVIC_VM_ID_NR (1 << AVIC_VM_ID_BITS) | |
111 | #define AVIC_VM_ID_MASK ((1 << AVIC_VM_ID_BITS) - 1) | |
112 | ||
113 | #define AVIC_GATAG(x, y) (((x & AVIC_VM_ID_MASK) << AVIC_VCPU_ID_BITS) | \ | |
114 | (y & AVIC_VCPU_ID_MASK)) | |
115 | #define AVIC_GATAG_TO_VMID(x) ((x >> AVIC_VCPU_ID_BITS) & AVIC_VM_ID_MASK) | |
116 | #define AVIC_GATAG_TO_VCPUID(x) (x & AVIC_VCPU_ID_MASK) | |
117 | ||
67ec6607 JR |
118 | static bool erratum_383_found __read_mostly; |
119 | ||
6c8166a7 AK |
120 | static const u32 host_save_user_msrs[] = { |
121 | #ifdef CONFIG_X86_64 | |
122 | MSR_STAR, MSR_LSTAR, MSR_CSTAR, MSR_SYSCALL_MASK, MSR_KERNEL_GS_BASE, | |
123 | MSR_FS_BASE, | |
124 | #endif | |
125 | MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP, | |
46896c73 | 126 | MSR_TSC_AUX, |
6c8166a7 AK |
127 | }; |
128 | ||
129 | #define NR_HOST_SAVE_USER_MSRS ARRAY_SIZE(host_save_user_msrs) | |
130 | ||
81811c16 SC |
131 | struct kvm_sev_info { |
132 | bool active; /* SEV enabled guest */ | |
133 | unsigned int asid; /* ASID used for this guest */ | |
134 | unsigned int handle; /* SEV firmware handle */ | |
135 | int fd; /* SEV device fd */ | |
136 | unsigned long pages_locked; /* Number of pages locked */ | |
137 | struct list_head regions_list; /* List of registered regions */ | |
138 | }; | |
139 | ||
140 | struct kvm_svm { | |
141 | struct kvm kvm; | |
142 | ||
143 | /* Struct members for AVIC */ | |
144 | u32 avic_vm_id; | |
81811c16 SC |
145 | struct page *avic_logical_id_table_page; |
146 | struct page *avic_physical_id_table_page; | |
147 | struct hlist_node hnode; | |
148 | ||
149 | struct kvm_sev_info sev_info; | |
150 | }; | |
151 | ||
6c8166a7 AK |
152 | struct kvm_vcpu; |
153 | ||
e6aa9abd JR |
154 | struct nested_state { |
155 | struct vmcb *hsave; | |
156 | u64 hsave_msr; | |
4a810181 | 157 | u64 vm_cr_msr; |
e6aa9abd JR |
158 | u64 vmcb; |
159 | ||
160 | /* These are the merged vectors */ | |
161 | u32 *msrpm; | |
162 | ||
163 | /* gpa pointers to the real vectors */ | |
164 | u64 vmcb_msrpm; | |
ce2ac085 | 165 | u64 vmcb_iopm; |
aad42c64 | 166 | |
cd3ff653 JR |
167 | /* A VMEXIT is required but not yet emulated */ |
168 | bool exit_required; | |
169 | ||
aad42c64 | 170 | /* cache for intercepts of the guest */ |
4ee546b4 | 171 | u32 intercept_cr; |
3aed041a | 172 | u32 intercept_dr; |
aad42c64 JR |
173 | u32 intercept_exceptions; |
174 | u64 intercept; | |
175 | ||
5bd2edc3 JR |
176 | /* Nested Paging related state */ |
177 | u64 nested_cr3; | |
e6aa9abd JR |
178 | }; |
179 | ||
323c3d80 JR |
180 | #define MSRPM_OFFSETS 16 |
181 | static u32 msrpm_offsets[MSRPM_OFFSETS] __read_mostly; | |
182 | ||
2b036c6b BO |
183 | /* |
184 | * Set osvw_len to higher value when updated Revision Guides | |
185 | * are published and we know what the new status bits are | |
186 | */ | |
187 | static uint64_t osvw_len = 4, osvw_status; | |
188 | ||
6c8166a7 AK |
189 | struct vcpu_svm { |
190 | struct kvm_vcpu vcpu; | |
191 | struct vmcb *vmcb; | |
192 | unsigned long vmcb_pa; | |
193 | struct svm_cpu_data *svm_data; | |
194 | uint64_t asid_generation; | |
195 | uint64_t sysenter_esp; | |
196 | uint64_t sysenter_eip; | |
46896c73 | 197 | uint64_t tsc_aux; |
6c8166a7 | 198 | |
d1d93fa9 TL |
199 | u64 msr_decfg; |
200 | ||
6c8166a7 AK |
201 | u64 next_rip; |
202 | ||
203 | u64 host_user_msrs[NR_HOST_SAVE_USER_MSRS]; | |
afe9e66f | 204 | struct { |
dacccfdd AK |
205 | u16 fs; |
206 | u16 gs; | |
207 | u16 ldt; | |
afe9e66f AK |
208 | u64 gs_base; |
209 | } host; | |
6c8166a7 | 210 | |
b2ac58f9 | 211 | u64 spec_ctrl; |
ccbcd267 TG |
212 | /* |
213 | * Contains guest-controlled bits of VIRT_SPEC_CTRL, which will be | |
214 | * translated into the appropriate L2_CFG bits on the host to | |
215 | * perform speculative control. | |
216 | */ | |
217 | u64 virt_spec_ctrl; | |
b2ac58f9 | 218 | |
6c8166a7 | 219 | u32 *msrpm; |
6c8166a7 | 220 | |
bd3d1ec3 AK |
221 | ulong nmi_iret_rip; |
222 | ||
e6aa9abd | 223 | struct nested_state nested; |
6be7d306 JK |
224 | |
225 | bool nmi_singlestep; | |
ab2f4d73 | 226 | u64 nmi_singlestep_guest_rflags; |
66b7138f JK |
227 | |
228 | unsigned int3_injected; | |
229 | unsigned long int3_rip; | |
fbc0db76 | 230 | |
6092d3d3 JR |
231 | /* cached guest cpuid flags for faster access */ |
232 | bool nrips_enabled : 1; | |
44a95dae | 233 | |
18f40c53 | 234 | u32 ldr_reg; |
98d90582 | 235 | u32 dfr_reg; |
44a95dae SS |
236 | struct page *avic_backing_page; |
237 | u64 *avic_physical_id_cache; | |
8221c137 | 238 | bool avic_is_running; |
411b44ba SS |
239 | |
240 | /* | |
241 | * Per-vcpu list of struct amd_svm_iommu_ir: | |
242 | * This is used mainly to store interrupt remapping information used | |
243 | * when update the vcpu affinity. This avoids the need to scan for | |
244 | * IRTE and try to match ga_tag in the IOMMU driver. | |
245 | */ | |
246 | struct list_head ir_list; | |
247 | spinlock_t ir_list_lock; | |
70cd94e6 BS |
248 | |
249 | /* which host CPU was used for running this vcpu */ | |
250 | unsigned int last_cpu; | |
411b44ba SS |
251 | }; |
252 | ||
253 | /* | |
254 | * This is a wrapper of struct amd_iommu_ir_data. | |
255 | */ | |
256 | struct amd_svm_iommu_ir { | |
257 | struct list_head node; /* Used by SVM for per-vcpu ir_list */ | |
258 | void *data; /* Storing pointer to struct amd_ir_data */ | |
6c8166a7 AK |
259 | }; |
260 | ||
44a95dae | 261 | #define AVIC_LOGICAL_ID_ENTRY_GUEST_PHYSICAL_ID_MASK (0xFF) |
e44e3eac | 262 | #define AVIC_LOGICAL_ID_ENTRY_VALID_BIT 31 |
44a95dae SS |
263 | #define AVIC_LOGICAL_ID_ENTRY_VALID_MASK (1 << 31) |
264 | ||
265 | #define AVIC_PHYSICAL_ID_ENTRY_HOST_PHYSICAL_ID_MASK (0xFFULL) | |
266 | #define AVIC_PHYSICAL_ID_ENTRY_BACKING_PAGE_MASK (0xFFFFFFFFFFULL << 12) | |
267 | #define AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK (1ULL << 62) | |
268 | #define AVIC_PHYSICAL_ID_ENTRY_VALID_MASK (1ULL << 63) | |
269 | ||
fbc0db76 JR |
270 | static DEFINE_PER_CPU(u64, current_tsc_ratio); |
271 | #define TSC_RATIO_DEFAULT 0x0100000000ULL | |
272 | ||
455716fa JR |
273 | #define MSR_INVALID 0xffffffffU |
274 | ||
09941fbb | 275 | static const struct svm_direct_access_msrs { |
ac72a9b7 JR |
276 | u32 index; /* Index of the MSR */ |
277 | bool always; /* True if intercept is always on */ | |
278 | } direct_access_msrs[] = { | |
8c06585d | 279 | { .index = MSR_STAR, .always = true }, |
ac72a9b7 JR |
280 | { .index = MSR_IA32_SYSENTER_CS, .always = true }, |
281 | #ifdef CONFIG_X86_64 | |
282 | { .index = MSR_GS_BASE, .always = true }, | |
283 | { .index = MSR_FS_BASE, .always = true }, | |
284 | { .index = MSR_KERNEL_GS_BASE, .always = true }, | |
285 | { .index = MSR_LSTAR, .always = true }, | |
286 | { .index = MSR_CSTAR, .always = true }, | |
287 | { .index = MSR_SYSCALL_MASK, .always = true }, | |
288 | #endif | |
b2ac58f9 | 289 | { .index = MSR_IA32_SPEC_CTRL, .always = false }, |
15d45071 | 290 | { .index = MSR_IA32_PRED_CMD, .always = false }, |
ac72a9b7 JR |
291 | { .index = MSR_IA32_LASTBRANCHFROMIP, .always = false }, |
292 | { .index = MSR_IA32_LASTBRANCHTOIP, .always = false }, | |
293 | { .index = MSR_IA32_LASTINTFROMIP, .always = false }, | |
294 | { .index = MSR_IA32_LASTINTTOIP, .always = false }, | |
295 | { .index = MSR_INVALID, .always = false }, | |
6c8166a7 AK |
296 | }; |
297 | ||
709ddebf JR |
298 | /* enable NPT for AMD64 and X86 with PAE */ |
299 | #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE) | |
300 | static bool npt_enabled = true; | |
301 | #else | |
e0231715 | 302 | static bool npt_enabled; |
709ddebf | 303 | #endif |
6c7dac72 | 304 | |
8566ac8b BM |
305 | /* |
306 | * These 2 parameters are used to config the controls for Pause-Loop Exiting: | |
307 | * pause_filter_count: On processors that support Pause filtering(indicated | |
308 | * by CPUID Fn8000_000A_EDX), the VMCB provides a 16 bit pause filter | |
309 | * count value. On VMRUN this value is loaded into an internal counter. | |
310 | * Each time a pause instruction is executed, this counter is decremented | |
311 | * until it reaches zero at which time a #VMEXIT is generated if pause | |
312 | * intercept is enabled. Refer to AMD APM Vol 2 Section 15.14.4 Pause | |
313 | * Intercept Filtering for more details. | |
314 | * This also indicate if ple logic enabled. | |
315 | * | |
316 | * pause_filter_thresh: In addition, some processor families support advanced | |
317 | * pause filtering (indicated by CPUID Fn8000_000A_EDX) upper bound on | |
318 | * the amount of time a guest is allowed to execute in a pause loop. | |
319 | * In this mode, a 16-bit pause filter threshold field is added in the | |
320 | * VMCB. The threshold value is a cycle count that is used to reset the | |
321 | * pause counter. As with simple pause filtering, VMRUN loads the pause | |
322 | * count value from VMCB into an internal counter. Then, on each pause | |
323 | * instruction the hardware checks the elapsed number of cycles since | |
324 | * the most recent pause instruction against the pause filter threshold. | |
325 | * If the elapsed cycle count is greater than the pause filter threshold, | |
326 | * then the internal pause count is reloaded from the VMCB and execution | |
327 | * continues. If the elapsed cycle count is less than the pause filter | |
328 | * threshold, then the internal pause count is decremented. If the count | |
329 | * value is less than zero and PAUSE intercept is enabled, a #VMEXIT is | |
330 | * triggered. If advanced pause filtering is supported and pause filter | |
331 | * threshold field is set to zero, the filter will operate in the simpler, | |
332 | * count only mode. | |
333 | */ | |
334 | ||
335 | static unsigned short pause_filter_thresh = KVM_DEFAULT_PLE_GAP; | |
336 | module_param(pause_filter_thresh, ushort, 0444); | |
337 | ||
338 | static unsigned short pause_filter_count = KVM_SVM_DEFAULT_PLE_WINDOW; | |
339 | module_param(pause_filter_count, ushort, 0444); | |
340 | ||
341 | /* Default doubles per-vcpu window every exit. */ | |
342 | static unsigned short pause_filter_count_grow = KVM_DEFAULT_PLE_WINDOW_GROW; | |
343 | module_param(pause_filter_count_grow, ushort, 0444); | |
344 | ||
345 | /* Default resets per-vcpu window every exit to pause_filter_count. */ | |
346 | static unsigned short pause_filter_count_shrink = KVM_DEFAULT_PLE_WINDOW_SHRINK; | |
347 | module_param(pause_filter_count_shrink, ushort, 0444); | |
348 | ||
349 | /* Default is to compute the maximum so we can never overflow. */ | |
350 | static unsigned short pause_filter_count_max = KVM_SVM_DEFAULT_PLE_WINDOW_MAX; | |
351 | module_param(pause_filter_count_max, ushort, 0444); | |
352 | ||
e2358851 DB |
353 | /* allow nested paging (virtualized MMU) for all guests */ |
354 | static int npt = true; | |
6c7dac72 | 355 | module_param(npt, int, S_IRUGO); |
e3da3acd | 356 | |
e2358851 DB |
357 | /* allow nested virtualization in KVM/SVM */ |
358 | static int nested = true; | |
236de055 AG |
359 | module_param(nested, int, S_IRUGO); |
360 | ||
44a95dae SS |
361 | /* enable / disable AVIC */ |
362 | static int avic; | |
5b8abf1f | 363 | #ifdef CONFIG_X86_LOCAL_APIC |
44a95dae | 364 | module_param(avic, int, S_IRUGO); |
5b8abf1f | 365 | #endif |
44a95dae | 366 | |
d647eb63 PB |
367 | /* enable/disable Next RIP Save */ |
368 | static int nrips = true; | |
369 | module_param(nrips, int, 0444); | |
370 | ||
89c8a498 JN |
371 | /* enable/disable Virtual VMLOAD VMSAVE */ |
372 | static int vls = true; | |
373 | module_param(vls, int, 0444); | |
374 | ||
640bd6e5 JN |
375 | /* enable/disable Virtual GIF */ |
376 | static int vgif = true; | |
377 | module_param(vgif, int, 0444); | |
5ea11f2b | 378 | |
e9df0942 BS |
379 | /* enable/disable SEV support */ |
380 | static int sev = IS_ENABLED(CONFIG_AMD_MEM_ENCRYPT_ACTIVE_BY_DEFAULT); | |
381 | module_param(sev, int, 0444); | |
382 | ||
6f2f8453 PB |
383 | static bool __read_mostly dump_invalid_vmcb = 0; |
384 | module_param(dump_invalid_vmcb, bool, 0644); | |
385 | ||
7607b717 BS |
386 | static u8 rsm_ins_bytes[] = "\x0f\xaa"; |
387 | ||
79a8059d | 388 | static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0); |
c2ba05cc | 389 | static void svm_flush_tlb(struct kvm_vcpu *vcpu, bool invalidate_gpa); |
a5c3832d | 390 | static void svm_complete_interrupts(struct vcpu_svm *svm); |
04d2cc77 | 391 | |
410e4d57 | 392 | static int nested_svm_exit_handled(struct vcpu_svm *svm); |
b8e88bc8 | 393 | static int nested_svm_intercept(struct vcpu_svm *svm); |
cf74a78b | 394 | static int nested_svm_vmexit(struct vcpu_svm *svm); |
cf74a78b AG |
395 | static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr, |
396 | bool has_error_code, u32 error_code); | |
397 | ||
8d28fec4 | 398 | enum { |
116a0a23 JR |
399 | VMCB_INTERCEPTS, /* Intercept vectors, TSC offset, |
400 | pause filter count */ | |
f56838e4 | 401 | VMCB_PERM_MAP, /* IOPM Base and MSRPM Base */ |
d48086d1 | 402 | VMCB_ASID, /* ASID */ |
decdbf6a | 403 | VMCB_INTR, /* int_ctl, int_vector */ |
b2747166 | 404 | VMCB_NPT, /* npt_en, nCR3, gPAT */ |
dcca1a65 | 405 | VMCB_CR, /* CR0, CR3, CR4, EFER */ |
72214b96 | 406 | VMCB_DR, /* DR6, DR7 */ |
17a703cb | 407 | VMCB_DT, /* GDT, IDT */ |
060d0c9a | 408 | VMCB_SEG, /* CS, DS, SS, ES, CPL */ |
0574dec0 | 409 | VMCB_CR2, /* CR2 only */ |
b53ba3f9 | 410 | VMCB_LBR, /* DBGCTL, BR_FROM, BR_TO, LAST_EX_FROM, LAST_EX_TO */ |
44a95dae SS |
411 | VMCB_AVIC, /* AVIC APIC_BAR, AVIC APIC_BACKING_PAGE, |
412 | * AVIC PHYSICAL_TABLE pointer, | |
413 | * AVIC LOGICAL_TABLE pointer | |
414 | */ | |
8d28fec4 RJ |
415 | VMCB_DIRTY_MAX, |
416 | }; | |
417 | ||
0574dec0 JR |
418 | /* TPR and CR2 are always written before VMRUN */ |
419 | #define VMCB_ALWAYS_DIRTY_MASK ((1U << VMCB_INTR) | (1U << VMCB_CR2)) | |
8d28fec4 | 420 | |
44a95dae SS |
421 | #define VMCB_AVIC_APIC_BAR_MASK 0xFFFFFFFFFF000ULL |
422 | ||
ed3cd233 | 423 | static unsigned int max_sev_asid; |
1654efcb BS |
424 | static unsigned int min_sev_asid; |
425 | static unsigned long *sev_asid_bitmap; | |
89c50580 | 426 | #define __sme_page_pa(x) __sme_set(page_to_pfn(x) << PAGE_SHIFT) |
1654efcb | 427 | |
1e80fdc0 BS |
428 | struct enc_region { |
429 | struct list_head list; | |
430 | unsigned long npages; | |
431 | struct page **pages; | |
432 | unsigned long uaddr; | |
433 | unsigned long size; | |
434 | }; | |
435 | ||
81811c16 SC |
436 | |
437 | static inline struct kvm_svm *to_kvm_svm(struct kvm *kvm) | |
438 | { | |
439 | return container_of(kvm, struct kvm_svm, kvm); | |
440 | } | |
441 | ||
1654efcb BS |
442 | static inline bool svm_sev_enabled(void) |
443 | { | |
853c1109 | 444 | return IS_ENABLED(CONFIG_KVM_AMD_SEV) ? max_sev_asid : 0; |
1654efcb BS |
445 | } |
446 | ||
447 | static inline bool sev_guest(struct kvm *kvm) | |
448 | { | |
853c1109 | 449 | #ifdef CONFIG_KVM_AMD_SEV |
81811c16 | 450 | struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info; |
1654efcb BS |
451 | |
452 | return sev->active; | |
853c1109 PB |
453 | #else |
454 | return false; | |
455 | #endif | |
1654efcb | 456 | } |
ed3cd233 | 457 | |
70cd94e6 BS |
458 | static inline int sev_get_asid(struct kvm *kvm) |
459 | { | |
81811c16 | 460 | struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info; |
70cd94e6 BS |
461 | |
462 | return sev->asid; | |
463 | } | |
464 | ||
8d28fec4 RJ |
465 | static inline void mark_all_dirty(struct vmcb *vmcb) |
466 | { | |
467 | vmcb->control.clean = 0; | |
468 | } | |
469 | ||
470 | static inline void mark_all_clean(struct vmcb *vmcb) | |
471 | { | |
472 | vmcb->control.clean = ((1 << VMCB_DIRTY_MAX) - 1) | |
473 | & ~VMCB_ALWAYS_DIRTY_MASK; | |
474 | } | |
475 | ||
476 | static inline void mark_dirty(struct vmcb *vmcb, int bit) | |
477 | { | |
478 | vmcb->control.clean &= ~(1 << bit); | |
479 | } | |
480 | ||
a2fa3e9f GH |
481 | static inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu) |
482 | { | |
fb3f0f51 | 483 | return container_of(vcpu, struct vcpu_svm, vcpu); |
a2fa3e9f GH |
484 | } |
485 | ||
44a95dae SS |
486 | static inline void avic_update_vapic_bar(struct vcpu_svm *svm, u64 data) |
487 | { | |
488 | svm->vmcb->control.avic_vapic_bar = data & VMCB_AVIC_APIC_BAR_MASK; | |
489 | mark_dirty(svm->vmcb, VMCB_AVIC); | |
490 | } | |
491 | ||
340d3bc3 SS |
492 | static inline bool avic_vcpu_is_running(struct kvm_vcpu *vcpu) |
493 | { | |
494 | struct vcpu_svm *svm = to_svm(vcpu); | |
495 | u64 *entry = svm->avic_physical_id_cache; | |
496 | ||
497 | if (!entry) | |
498 | return false; | |
499 | ||
500 | return (READ_ONCE(*entry) & AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK); | |
501 | } | |
502 | ||
384c6368 JR |
503 | static void recalc_intercepts(struct vcpu_svm *svm) |
504 | { | |
505 | struct vmcb_control_area *c, *h; | |
506 | struct nested_state *g; | |
507 | ||
116a0a23 JR |
508 | mark_dirty(svm->vmcb, VMCB_INTERCEPTS); |
509 | ||
384c6368 JR |
510 | if (!is_guest_mode(&svm->vcpu)) |
511 | return; | |
512 | ||
513 | c = &svm->vmcb->control; | |
514 | h = &svm->nested.hsave->control; | |
515 | g = &svm->nested; | |
516 | ||
4ee546b4 | 517 | c->intercept_cr = h->intercept_cr | g->intercept_cr; |
3aed041a | 518 | c->intercept_dr = h->intercept_dr | g->intercept_dr; |
bd89525a | 519 | c->intercept_exceptions = h->intercept_exceptions | g->intercept_exceptions; |
384c6368 JR |
520 | c->intercept = h->intercept | g->intercept; |
521 | } | |
522 | ||
4ee546b4 RJ |
523 | static inline struct vmcb *get_host_vmcb(struct vcpu_svm *svm) |
524 | { | |
525 | if (is_guest_mode(&svm->vcpu)) | |
526 | return svm->nested.hsave; | |
527 | else | |
528 | return svm->vmcb; | |
529 | } | |
530 | ||
531 | static inline void set_cr_intercept(struct vcpu_svm *svm, int bit) | |
532 | { | |
533 | struct vmcb *vmcb = get_host_vmcb(svm); | |
534 | ||
535 | vmcb->control.intercept_cr |= (1U << bit); | |
536 | ||
537 | recalc_intercepts(svm); | |
538 | } | |
539 | ||
540 | static inline void clr_cr_intercept(struct vcpu_svm *svm, int bit) | |
541 | { | |
542 | struct vmcb *vmcb = get_host_vmcb(svm); | |
543 | ||
544 | vmcb->control.intercept_cr &= ~(1U << bit); | |
545 | ||
546 | recalc_intercepts(svm); | |
547 | } | |
548 | ||
549 | static inline bool is_cr_intercept(struct vcpu_svm *svm, int bit) | |
550 | { | |
551 | struct vmcb *vmcb = get_host_vmcb(svm); | |
552 | ||
553 | return vmcb->control.intercept_cr & (1U << bit); | |
554 | } | |
555 | ||
5315c716 | 556 | static inline void set_dr_intercepts(struct vcpu_svm *svm) |
3aed041a JR |
557 | { |
558 | struct vmcb *vmcb = get_host_vmcb(svm); | |
559 | ||
5315c716 PB |
560 | vmcb->control.intercept_dr = (1 << INTERCEPT_DR0_READ) |
561 | | (1 << INTERCEPT_DR1_READ) | |
562 | | (1 << INTERCEPT_DR2_READ) | |
563 | | (1 << INTERCEPT_DR3_READ) | |
564 | | (1 << INTERCEPT_DR4_READ) | |
565 | | (1 << INTERCEPT_DR5_READ) | |
566 | | (1 << INTERCEPT_DR6_READ) | |
567 | | (1 << INTERCEPT_DR7_READ) | |
568 | | (1 << INTERCEPT_DR0_WRITE) | |
569 | | (1 << INTERCEPT_DR1_WRITE) | |
570 | | (1 << INTERCEPT_DR2_WRITE) | |
571 | | (1 << INTERCEPT_DR3_WRITE) | |
572 | | (1 << INTERCEPT_DR4_WRITE) | |
573 | | (1 << INTERCEPT_DR5_WRITE) | |
574 | | (1 << INTERCEPT_DR6_WRITE) | |
575 | | (1 << INTERCEPT_DR7_WRITE); | |
3aed041a JR |
576 | |
577 | recalc_intercepts(svm); | |
578 | } | |
579 | ||
5315c716 | 580 | static inline void clr_dr_intercepts(struct vcpu_svm *svm) |
3aed041a JR |
581 | { |
582 | struct vmcb *vmcb = get_host_vmcb(svm); | |
583 | ||
5315c716 | 584 | vmcb->control.intercept_dr = 0; |
3aed041a JR |
585 | |
586 | recalc_intercepts(svm); | |
587 | } | |
588 | ||
18c918c5 JR |
589 | static inline void set_exception_intercept(struct vcpu_svm *svm, int bit) |
590 | { | |
591 | struct vmcb *vmcb = get_host_vmcb(svm); | |
592 | ||
593 | vmcb->control.intercept_exceptions |= (1U << bit); | |
594 | ||
595 | recalc_intercepts(svm); | |
596 | } | |
597 | ||
598 | static inline void clr_exception_intercept(struct vcpu_svm *svm, int bit) | |
599 | { | |
600 | struct vmcb *vmcb = get_host_vmcb(svm); | |
601 | ||
602 | vmcb->control.intercept_exceptions &= ~(1U << bit); | |
603 | ||
604 | recalc_intercepts(svm); | |
605 | } | |
606 | ||
8a05a1b8 JR |
607 | static inline void set_intercept(struct vcpu_svm *svm, int bit) |
608 | { | |
609 | struct vmcb *vmcb = get_host_vmcb(svm); | |
610 | ||
611 | vmcb->control.intercept |= (1ULL << bit); | |
612 | ||
613 | recalc_intercepts(svm); | |
614 | } | |
615 | ||
616 | static inline void clr_intercept(struct vcpu_svm *svm, int bit) | |
617 | { | |
618 | struct vmcb *vmcb = get_host_vmcb(svm); | |
619 | ||
620 | vmcb->control.intercept &= ~(1ULL << bit); | |
621 | ||
622 | recalc_intercepts(svm); | |
623 | } | |
624 | ||
640bd6e5 JN |
625 | static inline bool vgif_enabled(struct vcpu_svm *svm) |
626 | { | |
627 | return !!(svm->vmcb->control.int_ctl & V_GIF_ENABLE_MASK); | |
628 | } | |
629 | ||
2af9194d JR |
630 | static inline void enable_gif(struct vcpu_svm *svm) |
631 | { | |
640bd6e5 JN |
632 | if (vgif_enabled(svm)) |
633 | svm->vmcb->control.int_ctl |= V_GIF_MASK; | |
634 | else | |
635 | svm->vcpu.arch.hflags |= HF_GIF_MASK; | |
2af9194d JR |
636 | } |
637 | ||
638 | static inline void disable_gif(struct vcpu_svm *svm) | |
639 | { | |
640bd6e5 JN |
640 | if (vgif_enabled(svm)) |
641 | svm->vmcb->control.int_ctl &= ~V_GIF_MASK; | |
642 | else | |
643 | svm->vcpu.arch.hflags &= ~HF_GIF_MASK; | |
2af9194d JR |
644 | } |
645 | ||
646 | static inline bool gif_set(struct vcpu_svm *svm) | |
647 | { | |
640bd6e5 JN |
648 | if (vgif_enabled(svm)) |
649 | return !!(svm->vmcb->control.int_ctl & V_GIF_MASK); | |
650 | else | |
651 | return !!(svm->vcpu.arch.hflags & HF_GIF_MASK); | |
2af9194d JR |
652 | } |
653 | ||
4866d5e3 | 654 | static unsigned long iopm_base; |
6aa8b732 AK |
655 | |
656 | struct kvm_ldttss_desc { | |
657 | u16 limit0; | |
658 | u16 base0; | |
e0231715 JR |
659 | unsigned base1:8, type:5, dpl:2, p:1; |
660 | unsigned limit1:4, zero0:3, g:1, base2:8; | |
6aa8b732 AK |
661 | u32 base3; |
662 | u32 zero1; | |
663 | } __attribute__((packed)); | |
664 | ||
665 | struct svm_cpu_data { | |
666 | int cpu; | |
667 | ||
5008fdf5 AK |
668 | u64 asid_generation; |
669 | u32 max_asid; | |
670 | u32 next_asid; | |
4faefff3 | 671 | u32 min_asid; |
6aa8b732 AK |
672 | struct kvm_ldttss_desc *tss_desc; |
673 | ||
674 | struct page *save_area; | |
15d45071 | 675 | struct vmcb *current_vmcb; |
70cd94e6 BS |
676 | |
677 | /* index = sev_asid, value = vmcb pointer */ | |
678 | struct vmcb **sev_vmcbs; | |
6aa8b732 AK |
679 | }; |
680 | ||
681 | static DEFINE_PER_CPU(struct svm_cpu_data *, svm_data); | |
682 | ||
09941fbb | 683 | static const u32 msrpm_ranges[] = {0, 0xc0000000, 0xc0010000}; |
6aa8b732 | 684 | |
9d8f549d | 685 | #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges) |
6aa8b732 AK |
686 | #define MSRS_RANGE_SIZE 2048 |
687 | #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2) | |
688 | ||
455716fa JR |
689 | static u32 svm_msrpm_offset(u32 msr) |
690 | { | |
691 | u32 offset; | |
692 | int i; | |
693 | ||
694 | for (i = 0; i < NUM_MSR_MAPS; i++) { | |
695 | if (msr < msrpm_ranges[i] || | |
696 | msr >= msrpm_ranges[i] + MSRS_IN_RANGE) | |
697 | continue; | |
698 | ||
699 | offset = (msr - msrpm_ranges[i]) / 4; /* 4 msrs per u8 */ | |
700 | offset += (i * MSRS_RANGE_SIZE); /* add range offset */ | |
701 | ||
702 | /* Now we have the u8 offset - but need the u32 offset */ | |
703 | return offset / 4; | |
704 | } | |
705 | ||
706 | /* MSR not in any range */ | |
707 | return MSR_INVALID; | |
708 | } | |
709 | ||
6aa8b732 AK |
710 | #define MAX_INST_SIZE 15 |
711 | ||
6aa8b732 AK |
712 | static inline void clgi(void) |
713 | { | |
ac5ffda2 | 714 | asm volatile (__ex("clgi")); |
6aa8b732 AK |
715 | } |
716 | ||
717 | static inline void stgi(void) | |
718 | { | |
ac5ffda2 | 719 | asm volatile (__ex("stgi")); |
6aa8b732 AK |
720 | } |
721 | ||
722 | static inline void invlpga(unsigned long addr, u32 asid) | |
723 | { | |
ac5ffda2 | 724 | asm volatile (__ex("invlpga %1, %0") : : "c"(asid), "a"(addr)); |
6aa8b732 AK |
725 | } |
726 | ||
855feb67 | 727 | static int get_npt_level(struct kvm_vcpu *vcpu) |
4b16184c JR |
728 | { |
729 | #ifdef CONFIG_X86_64 | |
2a7266a8 | 730 | return PT64_ROOT_4LEVEL; |
4b16184c JR |
731 | #else |
732 | return PT32E_ROOT_LEVEL; | |
733 | #endif | |
734 | } | |
735 | ||
6aa8b732 AK |
736 | static void svm_set_efer(struct kvm_vcpu *vcpu, u64 efer) |
737 | { | |
6dc696d4 | 738 | vcpu->arch.efer = efer; |
709ddebf | 739 | if (!npt_enabled && !(efer & EFER_LMA)) |
2b5203ee | 740 | efer &= ~EFER_LME; |
6aa8b732 | 741 | |
9962d032 | 742 | to_svm(vcpu)->vmcb->save.efer = efer | EFER_SVME; |
dcca1a65 | 743 | mark_dirty(to_svm(vcpu)->vmcb, VMCB_CR); |
6aa8b732 AK |
744 | } |
745 | ||
6aa8b732 AK |
746 | static int is_external_interrupt(u32 info) |
747 | { | |
748 | info &= SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID; | |
749 | return info == (SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR); | |
750 | } | |
751 | ||
37ccdcbe | 752 | static u32 svm_get_interrupt_shadow(struct kvm_vcpu *vcpu) |
2809f5d2 GC |
753 | { |
754 | struct vcpu_svm *svm = to_svm(vcpu); | |
755 | u32 ret = 0; | |
756 | ||
757 | if (svm->vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK) | |
37ccdcbe PB |
758 | ret = KVM_X86_SHADOW_INT_STI | KVM_X86_SHADOW_INT_MOV_SS; |
759 | return ret; | |
2809f5d2 GC |
760 | } |
761 | ||
762 | static void svm_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask) | |
763 | { | |
764 | struct vcpu_svm *svm = to_svm(vcpu); | |
765 | ||
766 | if (mask == 0) | |
767 | svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK; | |
768 | else | |
769 | svm->vmcb->control.int_state |= SVM_INTERRUPT_SHADOW_MASK; | |
770 | ||
771 | } | |
772 | ||
6aa8b732 AK |
773 | static void skip_emulated_instruction(struct kvm_vcpu *vcpu) |
774 | { | |
a2fa3e9f GH |
775 | struct vcpu_svm *svm = to_svm(vcpu); |
776 | ||
d647eb63 | 777 | if (nrips && svm->vmcb->control.next_rip != 0) { |
d2922422 | 778 | WARN_ON_ONCE(!static_cpu_has(X86_FEATURE_NRIPS)); |
6bc31bdc | 779 | svm->next_rip = svm->vmcb->control.next_rip; |
f104765b | 780 | } |
6bc31bdc | 781 | |
a2fa3e9f | 782 | if (!svm->next_rip) { |
0ce97a2b | 783 | if (kvm_emulate_instruction(vcpu, EMULTYPE_SKIP) != |
f629cf84 GN |
784 | EMULATE_DONE) |
785 | printk(KERN_DEBUG "%s: NOP\n", __func__); | |
6aa8b732 AK |
786 | return; |
787 | } | |
5fdbf976 MT |
788 | if (svm->next_rip - kvm_rip_read(vcpu) > MAX_INST_SIZE) |
789 | printk(KERN_ERR "%s: ip 0x%lx next 0x%llx\n", | |
790 | __func__, kvm_rip_read(vcpu), svm->next_rip); | |
6aa8b732 | 791 | |
5fdbf976 | 792 | kvm_rip_write(vcpu, svm->next_rip); |
2809f5d2 | 793 | svm_set_interrupt_shadow(vcpu, 0); |
6aa8b732 AK |
794 | } |
795 | ||
cfcd20e5 | 796 | static void svm_queue_exception(struct kvm_vcpu *vcpu) |
116a4752 JK |
797 | { |
798 | struct vcpu_svm *svm = to_svm(vcpu); | |
cfcd20e5 WL |
799 | unsigned nr = vcpu->arch.exception.nr; |
800 | bool has_error_code = vcpu->arch.exception.has_error_code; | |
664f8e26 | 801 | bool reinject = vcpu->arch.exception.injected; |
cfcd20e5 | 802 | u32 error_code = vcpu->arch.exception.error_code; |
116a4752 | 803 | |
e0231715 JR |
804 | /* |
805 | * If we are within a nested VM we'd better #VMEXIT and let the guest | |
806 | * handle the exception | |
807 | */ | |
ce7ddec4 JR |
808 | if (!reinject && |
809 | nested_svm_check_exception(svm, nr, has_error_code, error_code)) | |
116a4752 JK |
810 | return; |
811 | ||
da998b46 JM |
812 | kvm_deliver_exception_payload(&svm->vcpu); |
813 | ||
d647eb63 | 814 | if (nr == BP_VECTOR && !nrips) { |
66b7138f JK |
815 | unsigned long rip, old_rip = kvm_rip_read(&svm->vcpu); |
816 | ||
817 | /* | |
818 | * For guest debugging where we have to reinject #BP if some | |
819 | * INT3 is guest-owned: | |
820 | * Emulate nRIP by moving RIP forward. Will fail if injection | |
821 | * raises a fault that is not intercepted. Still better than | |
822 | * failing in all cases. | |
823 | */ | |
824 | skip_emulated_instruction(&svm->vcpu); | |
825 | rip = kvm_rip_read(&svm->vcpu); | |
826 | svm->int3_rip = rip + svm->vmcb->save.cs.base; | |
827 | svm->int3_injected = rip - old_rip; | |
828 | } | |
829 | ||
116a4752 JK |
830 | svm->vmcb->control.event_inj = nr |
831 | | SVM_EVTINJ_VALID | |
832 | | (has_error_code ? SVM_EVTINJ_VALID_ERR : 0) | |
833 | | SVM_EVTINJ_TYPE_EXEPT; | |
834 | svm->vmcb->control.event_inj_err = error_code; | |
835 | } | |
836 | ||
67ec6607 JR |
837 | static void svm_init_erratum_383(void) |
838 | { | |
839 | u32 low, high; | |
840 | int err; | |
841 | u64 val; | |
842 | ||
e6ee94d5 | 843 | if (!static_cpu_has_bug(X86_BUG_AMD_TLB_MMATCH)) |
67ec6607 JR |
844 | return; |
845 | ||
846 | /* Use _safe variants to not break nested virtualization */ | |
847 | val = native_read_msr_safe(MSR_AMD64_DC_CFG, &err); | |
848 | if (err) | |
849 | return; | |
850 | ||
851 | val |= (1ULL << 47); | |
852 | ||
853 | low = lower_32_bits(val); | |
854 | high = upper_32_bits(val); | |
855 | ||
856 | native_write_msr_safe(MSR_AMD64_DC_CFG, low, high); | |
857 | ||
858 | erratum_383_found = true; | |
859 | } | |
860 | ||
2b036c6b BO |
861 | static void svm_init_osvw(struct kvm_vcpu *vcpu) |
862 | { | |
863 | /* | |
864 | * Guests should see errata 400 and 415 as fixed (assuming that | |
865 | * HLT and IO instructions are intercepted). | |
866 | */ | |
867 | vcpu->arch.osvw.length = (osvw_len >= 3) ? (osvw_len) : 3; | |
868 | vcpu->arch.osvw.status = osvw_status & ~(6ULL); | |
869 | ||
870 | /* | |
871 | * By increasing VCPU's osvw.length to 3 we are telling the guest that | |
872 | * all osvw.status bits inside that length, including bit 0 (which is | |
873 | * reserved for erratum 298), are valid. However, if host processor's | |
874 | * osvw_len is 0 then osvw_status[0] carries no information. We need to | |
875 | * be conservative here and therefore we tell the guest that erratum 298 | |
876 | * is present (because we really don't know). | |
877 | */ | |
878 | if (osvw_len == 0 && boot_cpu_data.x86 == 0x10) | |
879 | vcpu->arch.osvw.status |= 1; | |
880 | } | |
881 | ||
6aa8b732 AK |
882 | static int has_svm(void) |
883 | { | |
63d1142f | 884 | const char *msg; |
6aa8b732 | 885 | |
63d1142f | 886 | if (!cpu_has_svm(&msg)) { |
ff81ff10 | 887 | printk(KERN_INFO "has_svm: %s\n", msg); |
6aa8b732 AK |
888 | return 0; |
889 | } | |
890 | ||
6aa8b732 AK |
891 | return 1; |
892 | } | |
893 | ||
13a34e06 | 894 | static void svm_hardware_disable(void) |
6aa8b732 | 895 | { |
fbc0db76 JR |
896 | /* Make sure we clean up behind us */ |
897 | if (static_cpu_has(X86_FEATURE_TSCRATEMSR)) | |
898 | wrmsrl(MSR_AMD64_TSC_RATIO, TSC_RATIO_DEFAULT); | |
899 | ||
2c8dceeb | 900 | cpu_svm_disable(); |
1018faa6 JR |
901 | |
902 | amd_pmu_disable_virt(); | |
6aa8b732 AK |
903 | } |
904 | ||
13a34e06 | 905 | static int svm_hardware_enable(void) |
6aa8b732 AK |
906 | { |
907 | ||
0fe1e009 | 908 | struct svm_cpu_data *sd; |
6aa8b732 | 909 | uint64_t efer; |
6aa8b732 AK |
910 | struct desc_struct *gdt; |
911 | int me = raw_smp_processor_id(); | |
912 | ||
10474ae8 AG |
913 | rdmsrl(MSR_EFER, efer); |
914 | if (efer & EFER_SVME) | |
915 | return -EBUSY; | |
916 | ||
6aa8b732 | 917 | if (!has_svm()) { |
1f5b77f5 | 918 | pr_err("%s: err EOPNOTSUPP on %d\n", __func__, me); |
10474ae8 | 919 | return -EINVAL; |
6aa8b732 | 920 | } |
0fe1e009 | 921 | sd = per_cpu(svm_data, me); |
0fe1e009 | 922 | if (!sd) { |
1f5b77f5 | 923 | pr_err("%s: svm_data is NULL on %d\n", __func__, me); |
10474ae8 | 924 | return -EINVAL; |
6aa8b732 AK |
925 | } |
926 | ||
0fe1e009 TH |
927 | sd->asid_generation = 1; |
928 | sd->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1; | |
929 | sd->next_asid = sd->max_asid + 1; | |
ed3cd233 | 930 | sd->min_asid = max_sev_asid + 1; |
6aa8b732 | 931 | |
45fc8757 | 932 | gdt = get_current_gdt_rw(); |
0fe1e009 | 933 | sd->tss_desc = (struct kvm_ldttss_desc *)(gdt + GDT_ENTRY_TSS); |
6aa8b732 | 934 | |
9962d032 | 935 | wrmsrl(MSR_EFER, efer | EFER_SVME); |
6aa8b732 | 936 | |
d0316554 | 937 | wrmsrl(MSR_VM_HSAVE_PA, page_to_pfn(sd->save_area) << PAGE_SHIFT); |
10474ae8 | 938 | |
fbc0db76 JR |
939 | if (static_cpu_has(X86_FEATURE_TSCRATEMSR)) { |
940 | wrmsrl(MSR_AMD64_TSC_RATIO, TSC_RATIO_DEFAULT); | |
89cbc767 | 941 | __this_cpu_write(current_tsc_ratio, TSC_RATIO_DEFAULT); |
fbc0db76 JR |
942 | } |
943 | ||
2b036c6b BO |
944 | |
945 | /* | |
946 | * Get OSVW bits. | |
947 | * | |
948 | * Note that it is possible to have a system with mixed processor | |
949 | * revisions and therefore different OSVW bits. If bits are not the same | |
950 | * on different processors then choose the worst case (i.e. if erratum | |
951 | * is present on one processor and not on another then assume that the | |
952 | * erratum is present everywhere). | |
953 | */ | |
954 | if (cpu_has(&boot_cpu_data, X86_FEATURE_OSVW)) { | |
955 | uint64_t len, status = 0; | |
956 | int err; | |
957 | ||
958 | len = native_read_msr_safe(MSR_AMD64_OSVW_ID_LENGTH, &err); | |
959 | if (!err) | |
960 | status = native_read_msr_safe(MSR_AMD64_OSVW_STATUS, | |
961 | &err); | |
962 | ||
963 | if (err) | |
964 | osvw_status = osvw_len = 0; | |
965 | else { | |
966 | if (len < osvw_len) | |
967 | osvw_len = len; | |
968 | osvw_status |= status; | |
969 | osvw_status &= (1ULL << osvw_len) - 1; | |
970 | } | |
971 | } else | |
972 | osvw_status = osvw_len = 0; | |
973 | ||
67ec6607 JR |
974 | svm_init_erratum_383(); |
975 | ||
1018faa6 JR |
976 | amd_pmu_enable_virt(); |
977 | ||
10474ae8 | 978 | return 0; |
6aa8b732 AK |
979 | } |
980 | ||
0da1db75 JR |
981 | static void svm_cpu_uninit(int cpu) |
982 | { | |
0fe1e009 | 983 | struct svm_cpu_data *sd = per_cpu(svm_data, raw_smp_processor_id()); |
0da1db75 | 984 | |
0fe1e009 | 985 | if (!sd) |
0da1db75 JR |
986 | return; |
987 | ||
988 | per_cpu(svm_data, raw_smp_processor_id()) = NULL; | |
70cd94e6 | 989 | kfree(sd->sev_vmcbs); |
0fe1e009 TH |
990 | __free_page(sd->save_area); |
991 | kfree(sd); | |
0da1db75 JR |
992 | } |
993 | ||
6aa8b732 AK |
994 | static int svm_cpu_init(int cpu) |
995 | { | |
0fe1e009 | 996 | struct svm_cpu_data *sd; |
6aa8b732 AK |
997 | int r; |
998 | ||
0fe1e009 TH |
999 | sd = kzalloc(sizeof(struct svm_cpu_data), GFP_KERNEL); |
1000 | if (!sd) | |
6aa8b732 | 1001 | return -ENOMEM; |
0fe1e009 | 1002 | sd->cpu = cpu; |
6aa8b732 | 1003 | r = -ENOMEM; |
70cd94e6 | 1004 | sd->save_area = alloc_page(GFP_KERNEL); |
0fe1e009 | 1005 | if (!sd->save_area) |
6aa8b732 AK |
1006 | goto err_1; |
1007 | ||
70cd94e6 BS |
1008 | if (svm_sev_enabled()) { |
1009 | r = -ENOMEM; | |
6da2ec56 KC |
1010 | sd->sev_vmcbs = kmalloc_array(max_sev_asid + 1, |
1011 | sizeof(void *), | |
1012 | GFP_KERNEL); | |
70cd94e6 BS |
1013 | if (!sd->sev_vmcbs) |
1014 | goto err_1; | |
1015 | } | |
1016 | ||
0fe1e009 | 1017 | per_cpu(svm_data, cpu) = sd; |
6aa8b732 AK |
1018 | |
1019 | return 0; | |
1020 | ||
1021 | err_1: | |
0fe1e009 | 1022 | kfree(sd); |
6aa8b732 AK |
1023 | return r; |
1024 | ||
1025 | } | |
1026 | ||
ac72a9b7 JR |
1027 | static bool valid_msr_intercept(u32 index) |
1028 | { | |
1029 | int i; | |
1030 | ||
1031 | for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) | |
1032 | if (direct_access_msrs[i].index == index) | |
1033 | return true; | |
1034 | ||
1035 | return false; | |
1036 | } | |
1037 | ||
b2ac58f9 KA |
1038 | static bool msr_write_intercepted(struct kvm_vcpu *vcpu, unsigned msr) |
1039 | { | |
1040 | u8 bit_write; | |
1041 | unsigned long tmp; | |
1042 | u32 offset; | |
1043 | u32 *msrpm; | |
1044 | ||
1045 | msrpm = is_guest_mode(vcpu) ? to_svm(vcpu)->nested.msrpm: | |
1046 | to_svm(vcpu)->msrpm; | |
1047 | ||
1048 | offset = svm_msrpm_offset(msr); | |
1049 | bit_write = 2 * (msr & 0x0f) + 1; | |
1050 | tmp = msrpm[offset]; | |
1051 | ||
1052 | BUG_ON(offset == MSR_INVALID); | |
1053 | ||
1054 | return !!test_bit(bit_write, &tmp); | |
1055 | } | |
1056 | ||
bfc733a7 RR |
1057 | static void set_msr_interception(u32 *msrpm, unsigned msr, |
1058 | int read, int write) | |
6aa8b732 | 1059 | { |
455716fa JR |
1060 | u8 bit_read, bit_write; |
1061 | unsigned long tmp; | |
1062 | u32 offset; | |
6aa8b732 | 1063 | |
ac72a9b7 JR |
1064 | /* |
1065 | * If this warning triggers extend the direct_access_msrs list at the | |
1066 | * beginning of the file | |
1067 | */ | |
1068 | WARN_ON(!valid_msr_intercept(msr)); | |
1069 | ||
455716fa JR |
1070 | offset = svm_msrpm_offset(msr); |
1071 | bit_read = 2 * (msr & 0x0f); | |
1072 | bit_write = 2 * (msr & 0x0f) + 1; | |
1073 | tmp = msrpm[offset]; | |
1074 | ||
1075 | BUG_ON(offset == MSR_INVALID); | |
1076 | ||
1077 | read ? clear_bit(bit_read, &tmp) : set_bit(bit_read, &tmp); | |
1078 | write ? clear_bit(bit_write, &tmp) : set_bit(bit_write, &tmp); | |
1079 | ||
1080 | msrpm[offset] = tmp; | |
6aa8b732 AK |
1081 | } |
1082 | ||
f65c229c | 1083 | static void svm_vcpu_init_msrpm(u32 *msrpm) |
6aa8b732 AK |
1084 | { |
1085 | int i; | |
1086 | ||
f65c229c JR |
1087 | memset(msrpm, 0xff, PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER)); |
1088 | ||
ac72a9b7 JR |
1089 | for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) { |
1090 | if (!direct_access_msrs[i].always) | |
1091 | continue; | |
1092 | ||
1093 | set_msr_interception(msrpm, direct_access_msrs[i].index, 1, 1); | |
1094 | } | |
f65c229c JR |
1095 | } |
1096 | ||
323c3d80 JR |
1097 | static void add_msr_offset(u32 offset) |
1098 | { | |
1099 | int i; | |
1100 | ||
1101 | for (i = 0; i < MSRPM_OFFSETS; ++i) { | |
1102 | ||
1103 | /* Offset already in list? */ | |
1104 | if (msrpm_offsets[i] == offset) | |
bfc733a7 | 1105 | return; |
323c3d80 JR |
1106 | |
1107 | /* Slot used by another offset? */ | |
1108 | if (msrpm_offsets[i] != MSR_INVALID) | |
1109 | continue; | |
1110 | ||
1111 | /* Add offset to list */ | |
1112 | msrpm_offsets[i] = offset; | |
1113 | ||
1114 | return; | |
6aa8b732 | 1115 | } |
323c3d80 JR |
1116 | |
1117 | /* | |
1118 | * If this BUG triggers the msrpm_offsets table has an overflow. Just | |
1119 | * increase MSRPM_OFFSETS in this case. | |
1120 | */ | |
bfc733a7 | 1121 | BUG(); |
6aa8b732 AK |
1122 | } |
1123 | ||
323c3d80 | 1124 | static void init_msrpm_offsets(void) |
f65c229c | 1125 | { |
323c3d80 | 1126 | int i; |
f65c229c | 1127 | |
323c3d80 JR |
1128 | memset(msrpm_offsets, 0xff, sizeof(msrpm_offsets)); |
1129 | ||
1130 | for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) { | |
1131 | u32 offset; | |
1132 | ||
1133 | offset = svm_msrpm_offset(direct_access_msrs[i].index); | |
1134 | BUG_ON(offset == MSR_INVALID); | |
1135 | ||
1136 | add_msr_offset(offset); | |
1137 | } | |
f65c229c JR |
1138 | } |
1139 | ||
24e09cbf JR |
1140 | static void svm_enable_lbrv(struct vcpu_svm *svm) |
1141 | { | |
1142 | u32 *msrpm = svm->msrpm; | |
1143 | ||
0dc92119 | 1144 | svm->vmcb->control.virt_ext |= LBR_CTL_ENABLE_MASK; |
24e09cbf JR |
1145 | set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 1, 1); |
1146 | set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 1, 1); | |
1147 | set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 1, 1); | |
1148 | set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 1, 1); | |
1149 | } | |
1150 | ||
1151 | static void svm_disable_lbrv(struct vcpu_svm *svm) | |
1152 | { | |
1153 | u32 *msrpm = svm->msrpm; | |
1154 | ||
0dc92119 | 1155 | svm->vmcb->control.virt_ext &= ~LBR_CTL_ENABLE_MASK; |
24e09cbf JR |
1156 | set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 0, 0); |
1157 | set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 0, 0); | |
1158 | set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 0, 0); | |
1159 | set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 0, 0); | |
1160 | } | |
1161 | ||
4aebd0e9 LP |
1162 | static void disable_nmi_singlestep(struct vcpu_svm *svm) |
1163 | { | |
1164 | svm->nmi_singlestep = false; | |
640bd6e5 | 1165 | |
ab2f4d73 LP |
1166 | if (!(svm->vcpu.guest_debug & KVM_GUESTDBG_SINGLESTEP)) { |
1167 | /* Clear our flags if they were not set by the guest */ | |
1168 | if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_TF)) | |
1169 | svm->vmcb->save.rflags &= ~X86_EFLAGS_TF; | |
1170 | if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_RF)) | |
1171 | svm->vmcb->save.rflags &= ~X86_EFLAGS_RF; | |
1172 | } | |
4aebd0e9 LP |
1173 | } |
1174 | ||
5881f737 | 1175 | /* Note: |
81811c16 | 1176 | * This hash table is used to map VM_ID to a struct kvm_svm, |
5881f737 SS |
1177 | * when handling AMD IOMMU GALOG notification to schedule in |
1178 | * a particular vCPU. | |
1179 | */ | |
1180 | #define SVM_VM_DATA_HASH_BITS 8 | |
681bcea8 | 1181 | static DEFINE_HASHTABLE(svm_vm_data_hash, SVM_VM_DATA_HASH_BITS); |
3f0d4db7 DV |
1182 | static u32 next_vm_id = 0; |
1183 | static bool next_vm_id_wrapped = 0; | |
681bcea8 | 1184 | static DEFINE_SPINLOCK(svm_vm_data_hash_lock); |
5881f737 SS |
1185 | |
1186 | /* Note: | |
1187 | * This function is called from IOMMU driver to notify | |
1188 | * SVM to schedule in a particular vCPU of a particular VM. | |
1189 | */ | |
1190 | static int avic_ga_log_notifier(u32 ga_tag) | |
1191 | { | |
1192 | unsigned long flags; | |
81811c16 | 1193 | struct kvm_svm *kvm_svm; |
5881f737 SS |
1194 | struct kvm_vcpu *vcpu = NULL; |
1195 | u32 vm_id = AVIC_GATAG_TO_VMID(ga_tag); | |
1196 | u32 vcpu_id = AVIC_GATAG_TO_VCPUID(ga_tag); | |
1197 | ||
1198 | pr_debug("SVM: %s: vm_id=%#x, vcpu_id=%#x\n", __func__, vm_id, vcpu_id); | |
1199 | ||
1200 | spin_lock_irqsave(&svm_vm_data_hash_lock, flags); | |
81811c16 SC |
1201 | hash_for_each_possible(svm_vm_data_hash, kvm_svm, hnode, vm_id) { |
1202 | if (kvm_svm->avic_vm_id != vm_id) | |
5881f737 | 1203 | continue; |
81811c16 | 1204 | vcpu = kvm_get_vcpu_by_id(&kvm_svm->kvm, vcpu_id); |
5881f737 SS |
1205 | break; |
1206 | } | |
1207 | spin_unlock_irqrestore(&svm_vm_data_hash_lock, flags); | |
1208 | ||
5881f737 SS |
1209 | /* Note: |
1210 | * At this point, the IOMMU should have already set the pending | |
1211 | * bit in the vAPIC backing page. So, we just need to schedule | |
1212 | * in the vcpu. | |
1213 | */ | |
1cf53587 | 1214 | if (vcpu) |
5881f737 SS |
1215 | kvm_vcpu_wake_up(vcpu); |
1216 | ||
1217 | return 0; | |
1218 | } | |
1219 | ||
e9df0942 BS |
1220 | static __init int sev_hardware_setup(void) |
1221 | { | |
1222 | struct sev_user_data_status *status; | |
1223 | int rc; | |
1224 | ||
1225 | /* Maximum number of encrypted guests supported simultaneously */ | |
1226 | max_sev_asid = cpuid_ecx(0x8000001F); | |
1227 | ||
1228 | if (!max_sev_asid) | |
1229 | return 1; | |
1230 | ||
1654efcb BS |
1231 | /* Minimum ASID value that should be used for SEV guest */ |
1232 | min_sev_asid = cpuid_edx(0x8000001F); | |
1233 | ||
1234 | /* Initialize SEV ASID bitmap */ | |
a101c9d6 | 1235 | sev_asid_bitmap = bitmap_zalloc(max_sev_asid, GFP_KERNEL); |
1654efcb BS |
1236 | if (!sev_asid_bitmap) |
1237 | return 1; | |
1238 | ||
e9df0942 BS |
1239 | status = kmalloc(sizeof(*status), GFP_KERNEL); |
1240 | if (!status) | |
1241 | return 1; | |
1242 | ||
1243 | /* | |
1244 | * Check SEV platform status. | |
1245 | * | |
1246 | * PLATFORM_STATUS can be called in any state, if we failed to query | |
1247 | * the PLATFORM status then either PSP firmware does not support SEV | |
1248 | * feature or SEV firmware is dead. | |
1249 | */ | |
1250 | rc = sev_platform_status(status, NULL); | |
1251 | if (rc) | |
1252 | goto err; | |
1253 | ||
1254 | pr_info("SEV supported\n"); | |
1255 | ||
1256 | err: | |
1257 | kfree(status); | |
1258 | return rc; | |
1259 | } | |
1260 | ||
8566ac8b BM |
1261 | static void grow_ple_window(struct kvm_vcpu *vcpu) |
1262 | { | |
1263 | struct vcpu_svm *svm = to_svm(vcpu); | |
1264 | struct vmcb_control_area *control = &svm->vmcb->control; | |
1265 | int old = control->pause_filter_count; | |
1266 | ||
1267 | control->pause_filter_count = __grow_ple_window(old, | |
1268 | pause_filter_count, | |
1269 | pause_filter_count_grow, | |
1270 | pause_filter_count_max); | |
1271 | ||
1272 | if (control->pause_filter_count != old) | |
1273 | mark_dirty(svm->vmcb, VMCB_INTERCEPTS); | |
1274 | ||
1275 | trace_kvm_ple_window_grow(vcpu->vcpu_id, | |
1276 | control->pause_filter_count, old); | |
1277 | } | |
1278 | ||
1279 | static void shrink_ple_window(struct kvm_vcpu *vcpu) | |
1280 | { | |
1281 | struct vcpu_svm *svm = to_svm(vcpu); | |
1282 | struct vmcb_control_area *control = &svm->vmcb->control; | |
1283 | int old = control->pause_filter_count; | |
1284 | ||
1285 | control->pause_filter_count = | |
1286 | __shrink_ple_window(old, | |
1287 | pause_filter_count, | |
1288 | pause_filter_count_shrink, | |
1289 | pause_filter_count); | |
1290 | if (control->pause_filter_count != old) | |
1291 | mark_dirty(svm->vmcb, VMCB_INTERCEPTS); | |
1292 | ||
1293 | trace_kvm_ple_window_shrink(vcpu->vcpu_id, | |
1294 | control->pause_filter_count, old); | |
1295 | } | |
1296 | ||
6aa8b732 AK |
1297 | static __init int svm_hardware_setup(void) |
1298 | { | |
1299 | int cpu; | |
1300 | struct page *iopm_pages; | |
f65c229c | 1301 | void *iopm_va; |
6aa8b732 AK |
1302 | int r; |
1303 | ||
6aa8b732 AK |
1304 | iopm_pages = alloc_pages(GFP_KERNEL, IOPM_ALLOC_ORDER); |
1305 | ||
1306 | if (!iopm_pages) | |
1307 | return -ENOMEM; | |
c8681339 AL |
1308 | |
1309 | iopm_va = page_address(iopm_pages); | |
1310 | memset(iopm_va, 0xff, PAGE_SIZE * (1 << IOPM_ALLOC_ORDER)); | |
6aa8b732 AK |
1311 | iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT; |
1312 | ||
323c3d80 JR |
1313 | init_msrpm_offsets(); |
1314 | ||
50a37eb4 JR |
1315 | if (boot_cpu_has(X86_FEATURE_NX)) |
1316 | kvm_enable_efer_bits(EFER_NX); | |
1317 | ||
1b2fd70c AG |
1318 | if (boot_cpu_has(X86_FEATURE_FXSR_OPT)) |
1319 | kvm_enable_efer_bits(EFER_FFXSR); | |
1320 | ||
92a1f12d | 1321 | if (boot_cpu_has(X86_FEATURE_TSCRATEMSR)) { |
92a1f12d | 1322 | kvm_has_tsc_control = true; |
bc9b961b HZ |
1323 | kvm_max_tsc_scaling_ratio = TSC_RATIO_MAX; |
1324 | kvm_tsc_scaling_ratio_frac_bits = 32; | |
92a1f12d JR |
1325 | } |
1326 | ||
8566ac8b BM |
1327 | /* Check for pause filtering support */ |
1328 | if (!boot_cpu_has(X86_FEATURE_PAUSEFILTER)) { | |
1329 | pause_filter_count = 0; | |
1330 | pause_filter_thresh = 0; | |
1331 | } else if (!boot_cpu_has(X86_FEATURE_PFTHRESHOLD)) { | |
1332 | pause_filter_thresh = 0; | |
1333 | } | |
1334 | ||
236de055 AG |
1335 | if (nested) { |
1336 | printk(KERN_INFO "kvm: Nested Virtualization enabled\n"); | |
eec4b140 | 1337 | kvm_enable_efer_bits(EFER_SVME | EFER_LMSLE); |
236de055 AG |
1338 | } |
1339 | ||
e9df0942 BS |
1340 | if (sev) { |
1341 | if (boot_cpu_has(X86_FEATURE_SEV) && | |
1342 | IS_ENABLED(CONFIG_KVM_AMD_SEV)) { | |
1343 | r = sev_hardware_setup(); | |
1344 | if (r) | |
1345 | sev = false; | |
1346 | } else { | |
1347 | sev = false; | |
1348 | } | |
1349 | } | |
1350 | ||
3230bb47 | 1351 | for_each_possible_cpu(cpu) { |
6aa8b732 AK |
1352 | r = svm_cpu_init(cpu); |
1353 | if (r) | |
f65c229c | 1354 | goto err; |
6aa8b732 | 1355 | } |
33bd6a0b | 1356 | |
2a6b20b8 | 1357 | if (!boot_cpu_has(X86_FEATURE_NPT)) |
e3da3acd JR |
1358 | npt_enabled = false; |
1359 | ||
6c7dac72 JR |
1360 | if (npt_enabled && !npt) { |
1361 | printk(KERN_INFO "kvm: Nested Paging disabled\n"); | |
1362 | npt_enabled = false; | |
1363 | } | |
1364 | ||
18552672 | 1365 | if (npt_enabled) { |
e3da3acd | 1366 | printk(KERN_INFO "kvm: Nested Paging enabled\n"); |
18552672 | 1367 | kvm_enable_tdp(); |
5f4cb662 JR |
1368 | } else |
1369 | kvm_disable_tdp(); | |
e3da3acd | 1370 | |
d647eb63 PB |
1371 | if (nrips) { |
1372 | if (!boot_cpu_has(X86_FEATURE_NRIPS)) | |
1373 | nrips = false; | |
1374 | } | |
1375 | ||
5b8abf1f SS |
1376 | if (avic) { |
1377 | if (!npt_enabled || | |
1378 | !boot_cpu_has(X86_FEATURE_AVIC) || | |
5881f737 | 1379 | !IS_ENABLED(CONFIG_X86_LOCAL_APIC)) { |
5b8abf1f | 1380 | avic = false; |
5881f737 | 1381 | } else { |
5b8abf1f | 1382 | pr_info("AVIC enabled\n"); |
5881f737 | 1383 | |
5881f737 SS |
1384 | amd_iommu_register_ga_log_notifier(&avic_ga_log_notifier); |
1385 | } | |
5b8abf1f | 1386 | } |
44a95dae | 1387 | |
89c8a498 JN |
1388 | if (vls) { |
1389 | if (!npt_enabled || | |
5442c269 | 1390 | !boot_cpu_has(X86_FEATURE_V_VMSAVE_VMLOAD) || |
89c8a498 JN |
1391 | !IS_ENABLED(CONFIG_X86_64)) { |
1392 | vls = false; | |
1393 | } else { | |
1394 | pr_info("Virtual VMLOAD VMSAVE supported\n"); | |
1395 | } | |
1396 | } | |
1397 | ||
640bd6e5 JN |
1398 | if (vgif) { |
1399 | if (!boot_cpu_has(X86_FEATURE_VGIF)) | |
1400 | vgif = false; | |
1401 | else | |
1402 | pr_info("Virtual GIF supported\n"); | |
1403 | } | |
1404 | ||
6aa8b732 AK |
1405 | return 0; |
1406 | ||
f65c229c | 1407 | err: |
6aa8b732 AK |
1408 | __free_pages(iopm_pages, IOPM_ALLOC_ORDER); |
1409 | iopm_base = 0; | |
1410 | return r; | |
1411 | } | |
1412 | ||
1413 | static __exit void svm_hardware_unsetup(void) | |
1414 | { | |
0da1db75 JR |
1415 | int cpu; |
1416 | ||
1654efcb | 1417 | if (svm_sev_enabled()) |
a101c9d6 | 1418 | bitmap_free(sev_asid_bitmap); |
1654efcb | 1419 | |
3230bb47 | 1420 | for_each_possible_cpu(cpu) |
0da1db75 JR |
1421 | svm_cpu_uninit(cpu); |
1422 | ||
6aa8b732 | 1423 | __free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT), IOPM_ALLOC_ORDER); |
f65c229c | 1424 | iopm_base = 0; |
6aa8b732 AK |
1425 | } |
1426 | ||
1427 | static void init_seg(struct vmcb_seg *seg) | |
1428 | { | |
1429 | seg->selector = 0; | |
1430 | seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK | | |
e0231715 | 1431 | SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */ |
6aa8b732 AK |
1432 | seg->limit = 0xffff; |
1433 | seg->base = 0; | |
1434 | } | |
1435 | ||
1436 | static void init_sys_seg(struct vmcb_seg *seg, uint32_t type) | |
1437 | { | |
1438 | seg->selector = 0; | |
1439 | seg->attrib = SVM_SELECTOR_P_MASK | type; | |
1440 | seg->limit = 0xffff; | |
1441 | seg->base = 0; | |
1442 | } | |
1443 | ||
e79f245d KA |
1444 | static u64 svm_read_l1_tsc_offset(struct kvm_vcpu *vcpu) |
1445 | { | |
1446 | struct vcpu_svm *svm = to_svm(vcpu); | |
1447 | ||
1448 | if (is_guest_mode(vcpu)) | |
1449 | return svm->nested.hsave->control.tsc_offset; | |
1450 | ||
1451 | return vcpu->arch.tsc_offset; | |
1452 | } | |
1453 | ||
326e7425 | 1454 | static u64 svm_write_l1_tsc_offset(struct kvm_vcpu *vcpu, u64 offset) |
f4e1b3c8 ZA |
1455 | { |
1456 | struct vcpu_svm *svm = to_svm(vcpu); | |
1457 | u64 g_tsc_offset = 0; | |
1458 | ||
2030753d | 1459 | if (is_guest_mode(vcpu)) { |
e79f245d | 1460 | /* Write L1's TSC offset. */ |
f4e1b3c8 ZA |
1461 | g_tsc_offset = svm->vmcb->control.tsc_offset - |
1462 | svm->nested.hsave->control.tsc_offset; | |
1463 | svm->nested.hsave->control.tsc_offset = offset; | |
45c3af97 PB |
1464 | } |
1465 | ||
1466 | trace_kvm_write_tsc_offset(vcpu->vcpu_id, | |
1467 | svm->vmcb->control.tsc_offset - g_tsc_offset, | |
1468 | offset); | |
f4e1b3c8 ZA |
1469 | |
1470 | svm->vmcb->control.tsc_offset = offset + g_tsc_offset; | |
116a0a23 JR |
1471 | |
1472 | mark_dirty(svm->vmcb, VMCB_INTERCEPTS); | |
326e7425 | 1473 | return svm->vmcb->control.tsc_offset; |
f4e1b3c8 ZA |
1474 | } |
1475 | ||
44a95dae SS |
1476 | static void avic_init_vmcb(struct vcpu_svm *svm) |
1477 | { | |
1478 | struct vmcb *vmcb = svm->vmcb; | |
81811c16 | 1479 | struct kvm_svm *kvm_svm = to_kvm_svm(svm->vcpu.kvm); |
d0ec49d4 | 1480 | phys_addr_t bpa = __sme_set(page_to_phys(svm->avic_backing_page)); |
81811c16 SC |
1481 | phys_addr_t lpa = __sme_set(page_to_phys(kvm_svm->avic_logical_id_table_page)); |
1482 | phys_addr_t ppa = __sme_set(page_to_phys(kvm_svm->avic_physical_id_table_page)); | |
44a95dae SS |
1483 | |
1484 | vmcb->control.avic_backing_page = bpa & AVIC_HPA_MASK; | |
1485 | vmcb->control.avic_logical_id = lpa & AVIC_HPA_MASK; | |
1486 | vmcb->control.avic_physical_id = ppa & AVIC_HPA_MASK; | |
1487 | vmcb->control.avic_physical_id |= AVIC_MAX_PHYSICAL_ID_COUNT; | |
1488 | vmcb->control.int_ctl |= AVIC_ENABLE_MASK; | |
44a95dae SS |
1489 | } |
1490 | ||
5690891b | 1491 | static void init_vmcb(struct vcpu_svm *svm) |
6aa8b732 | 1492 | { |
e6101a96 JR |
1493 | struct vmcb_control_area *control = &svm->vmcb->control; |
1494 | struct vmcb_save_area *save = &svm->vmcb->save; | |
6aa8b732 | 1495 | |
4ee546b4 | 1496 | svm->vcpu.arch.hflags = 0; |
bff78274 | 1497 | |
4ee546b4 RJ |
1498 | set_cr_intercept(svm, INTERCEPT_CR0_READ); |
1499 | set_cr_intercept(svm, INTERCEPT_CR3_READ); | |
1500 | set_cr_intercept(svm, INTERCEPT_CR4_READ); | |
1501 | set_cr_intercept(svm, INTERCEPT_CR0_WRITE); | |
1502 | set_cr_intercept(svm, INTERCEPT_CR3_WRITE); | |
1503 | set_cr_intercept(svm, INTERCEPT_CR4_WRITE); | |
3bbf3565 SS |
1504 | if (!kvm_vcpu_apicv_active(&svm->vcpu)) |
1505 | set_cr_intercept(svm, INTERCEPT_CR8_WRITE); | |
6aa8b732 | 1506 | |
5315c716 | 1507 | set_dr_intercepts(svm); |
6aa8b732 | 1508 | |
18c918c5 JR |
1509 | set_exception_intercept(svm, PF_VECTOR); |
1510 | set_exception_intercept(svm, UD_VECTOR); | |
1511 | set_exception_intercept(svm, MC_VECTOR); | |
54a20552 | 1512 | set_exception_intercept(svm, AC_VECTOR); |
cbdb967a | 1513 | set_exception_intercept(svm, DB_VECTOR); |
9718420e LA |
1514 | /* |
1515 | * Guest access to VMware backdoor ports could legitimately | |
1516 | * trigger #GP because of TSS I/O permission bitmap. | |
1517 | * We intercept those #GP and allow access to them anyway | |
1518 | * as VMware does. | |
1519 | */ | |
1520 | if (enable_vmware_backdoor) | |
1521 | set_exception_intercept(svm, GP_VECTOR); | |
6aa8b732 | 1522 | |
8a05a1b8 JR |
1523 | set_intercept(svm, INTERCEPT_INTR); |
1524 | set_intercept(svm, INTERCEPT_NMI); | |
1525 | set_intercept(svm, INTERCEPT_SMI); | |
1526 | set_intercept(svm, INTERCEPT_SELECTIVE_CR0); | |
332b56e4 | 1527 | set_intercept(svm, INTERCEPT_RDPMC); |
8a05a1b8 JR |
1528 | set_intercept(svm, INTERCEPT_CPUID); |
1529 | set_intercept(svm, INTERCEPT_INVD); | |
8a05a1b8 JR |
1530 | set_intercept(svm, INTERCEPT_INVLPG); |
1531 | set_intercept(svm, INTERCEPT_INVLPGA); | |
1532 | set_intercept(svm, INTERCEPT_IOIO_PROT); | |
1533 | set_intercept(svm, INTERCEPT_MSR_PROT); | |
1534 | set_intercept(svm, INTERCEPT_TASK_SWITCH); | |
1535 | set_intercept(svm, INTERCEPT_SHUTDOWN); | |
1536 | set_intercept(svm, INTERCEPT_VMRUN); | |
1537 | set_intercept(svm, INTERCEPT_VMMCALL); | |
1538 | set_intercept(svm, INTERCEPT_VMLOAD); | |
1539 | set_intercept(svm, INTERCEPT_VMSAVE); | |
1540 | set_intercept(svm, INTERCEPT_STGI); | |
1541 | set_intercept(svm, INTERCEPT_CLGI); | |
1542 | set_intercept(svm, INTERCEPT_SKINIT); | |
1543 | set_intercept(svm, INTERCEPT_WBINVD); | |
81dd35d4 | 1544 | set_intercept(svm, INTERCEPT_XSETBV); |
7607b717 | 1545 | set_intercept(svm, INTERCEPT_RSM); |
6aa8b732 | 1546 | |
4d5422ce | 1547 | if (!kvm_mwait_in_guest(svm->vcpu.kvm)) { |
668fffa3 MT |
1548 | set_intercept(svm, INTERCEPT_MONITOR); |
1549 | set_intercept(svm, INTERCEPT_MWAIT); | |
1550 | } | |
1551 | ||
caa057a2 WL |
1552 | if (!kvm_hlt_in_guest(svm->vcpu.kvm)) |
1553 | set_intercept(svm, INTERCEPT_HLT); | |
1554 | ||
d0ec49d4 TL |
1555 | control->iopm_base_pa = __sme_set(iopm_base); |
1556 | control->msrpm_base_pa = __sme_set(__pa(svm->msrpm)); | |
6aa8b732 AK |
1557 | control->int_ctl = V_INTR_MASKING_MASK; |
1558 | ||
1559 | init_seg(&save->es); | |
1560 | init_seg(&save->ss); | |
1561 | init_seg(&save->ds); | |
1562 | init_seg(&save->fs); | |
1563 | init_seg(&save->gs); | |
1564 | ||
1565 | save->cs.selector = 0xf000; | |
04b66839 | 1566 | save->cs.base = 0xffff0000; |
6aa8b732 AK |
1567 | /* Executable/Readable Code Segment */ |
1568 | save->cs.attrib = SVM_SELECTOR_READ_MASK | SVM_SELECTOR_P_MASK | | |
1569 | SVM_SELECTOR_S_MASK | SVM_SELECTOR_CODE_MASK; | |
1570 | save->cs.limit = 0xffff; | |
6aa8b732 AK |
1571 | |
1572 | save->gdtr.limit = 0xffff; | |
1573 | save->idtr.limit = 0xffff; | |
1574 | ||
1575 | init_sys_seg(&save->ldtr, SEG_TYPE_LDT); | |
1576 | init_sys_seg(&save->tr, SEG_TYPE_BUSY_TSS16); | |
1577 | ||
5690891b | 1578 | svm_set_efer(&svm->vcpu, 0); |
d77c26fc | 1579 | save->dr6 = 0xffff0ff0; |
f6e78475 | 1580 | kvm_set_rflags(&svm->vcpu, 2); |
6aa8b732 | 1581 | save->rip = 0x0000fff0; |
5fdbf976 | 1582 | svm->vcpu.arch.regs[VCPU_REGS_RIP] = save->rip; |
6aa8b732 | 1583 | |
e0231715 | 1584 | /* |
18fa000a | 1585 | * svm_set_cr0() sets PG and WP and clears NW and CD on save->cr0. |
d28bc9dd | 1586 | * It also updates the guest-visible cr0 value. |
6aa8b732 | 1587 | */ |
79a8059d | 1588 | svm_set_cr0(&svm->vcpu, X86_CR0_NW | X86_CR0_CD | X86_CR0_ET); |
ebae871a | 1589 | kvm_mmu_reset_context(&svm->vcpu); |
18fa000a | 1590 | |
66aee91a | 1591 | save->cr4 = X86_CR4_PAE; |
6aa8b732 | 1592 | /* rdx = ?? */ |
709ddebf JR |
1593 | |
1594 | if (npt_enabled) { | |
1595 | /* Setup VMCB for Nested Paging */ | |
cea3a19b | 1596 | control->nested_ctl |= SVM_NESTED_CTL_NP_ENABLE; |
8a05a1b8 | 1597 | clr_intercept(svm, INTERCEPT_INVLPG); |
18c918c5 | 1598 | clr_exception_intercept(svm, PF_VECTOR); |
4ee546b4 RJ |
1599 | clr_cr_intercept(svm, INTERCEPT_CR3_READ); |
1600 | clr_cr_intercept(svm, INTERCEPT_CR3_WRITE); | |
74545705 | 1601 | save->g_pat = svm->vcpu.arch.pat; |
709ddebf JR |
1602 | save->cr3 = 0; |
1603 | save->cr4 = 0; | |
1604 | } | |
f40f6a45 | 1605 | svm->asid_generation = 0; |
1371d904 | 1606 | |
e6aa9abd | 1607 | svm->nested.vmcb = 0; |
2af9194d JR |
1608 | svm->vcpu.arch.hflags = 0; |
1609 | ||
8566ac8b BM |
1610 | if (pause_filter_count) { |
1611 | control->pause_filter_count = pause_filter_count; | |
1612 | if (pause_filter_thresh) | |
1613 | control->pause_filter_thresh = pause_filter_thresh; | |
8a05a1b8 | 1614 | set_intercept(svm, INTERCEPT_PAUSE); |
8566ac8b BM |
1615 | } else { |
1616 | clr_intercept(svm, INTERCEPT_PAUSE); | |
565d0998 ML |
1617 | } |
1618 | ||
67034bb9 | 1619 | if (kvm_vcpu_apicv_active(&svm->vcpu)) |
44a95dae SS |
1620 | avic_init_vmcb(svm); |
1621 | ||
89c8a498 JN |
1622 | /* |
1623 | * If hardware supports Virtual VMLOAD VMSAVE then enable it | |
1624 | * in VMCB and clear intercepts to avoid #VMEXIT. | |
1625 | */ | |
1626 | if (vls) { | |
1627 | clr_intercept(svm, INTERCEPT_VMLOAD); | |
1628 | clr_intercept(svm, INTERCEPT_VMSAVE); | |
1629 | svm->vmcb->control.virt_ext |= VIRTUAL_VMLOAD_VMSAVE_ENABLE_MASK; | |
1630 | } | |
1631 | ||
640bd6e5 JN |
1632 | if (vgif) { |
1633 | clr_intercept(svm, INTERCEPT_STGI); | |
1634 | clr_intercept(svm, INTERCEPT_CLGI); | |
1635 | svm->vmcb->control.int_ctl |= V_GIF_ENABLE_MASK; | |
1636 | } | |
1637 | ||
35c6f649 | 1638 | if (sev_guest(svm->vcpu.kvm)) { |
1654efcb | 1639 | svm->vmcb->control.nested_ctl |= SVM_NESTED_CTL_SEV_ENABLE; |
35c6f649 BS |
1640 | clr_exception_intercept(svm, UD_VECTOR); |
1641 | } | |
1654efcb | 1642 | |
8d28fec4 RJ |
1643 | mark_all_dirty(svm->vmcb); |
1644 | ||
2af9194d | 1645 | enable_gif(svm); |
44a95dae SS |
1646 | |
1647 | } | |
1648 | ||
d3e7dec0 DC |
1649 | static u64 *avic_get_physical_id_entry(struct kvm_vcpu *vcpu, |
1650 | unsigned int index) | |
44a95dae SS |
1651 | { |
1652 | u64 *avic_physical_id_table; | |
81811c16 | 1653 | struct kvm_svm *kvm_svm = to_kvm_svm(vcpu->kvm); |
44a95dae SS |
1654 | |
1655 | if (index >= AVIC_MAX_PHYSICAL_ID_COUNT) | |
1656 | return NULL; | |
1657 | ||
81811c16 | 1658 | avic_physical_id_table = page_address(kvm_svm->avic_physical_id_table_page); |
44a95dae SS |
1659 | |
1660 | return &avic_physical_id_table[index]; | |
1661 | } | |
1662 | ||
1663 | /** | |
1664 | * Note: | |
1665 | * AVIC hardware walks the nested page table to check permissions, | |
1666 | * but does not use the SPA address specified in the leaf page | |
1667 | * table entry since it uses address in the AVIC_BACKING_PAGE pointer | |
1668 | * field of the VMCB. Therefore, we set up the | |
1669 | * APIC_ACCESS_PAGE_PRIVATE_MEMSLOT (4KB) here. | |
1670 | */ | |
1671 | static int avic_init_access_page(struct kvm_vcpu *vcpu) | |
1672 | { | |
1673 | struct kvm *kvm = vcpu->kvm; | |
30510387 | 1674 | int ret = 0; |
44a95dae | 1675 | |
30510387 | 1676 | mutex_lock(&kvm->slots_lock); |
44a95dae | 1677 | if (kvm->arch.apic_access_page_done) |
30510387 | 1678 | goto out; |
44a95dae | 1679 | |
30510387 WW |
1680 | ret = __x86_set_memory_region(kvm, |
1681 | APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, | |
1682 | APIC_DEFAULT_PHYS_BASE, | |
1683 | PAGE_SIZE); | |
44a95dae | 1684 | if (ret) |
30510387 | 1685 | goto out; |
44a95dae SS |
1686 | |
1687 | kvm->arch.apic_access_page_done = true; | |
30510387 WW |
1688 | out: |
1689 | mutex_unlock(&kvm->slots_lock); | |
1690 | return ret; | |
44a95dae SS |
1691 | } |
1692 | ||
1693 | static int avic_init_backing_page(struct kvm_vcpu *vcpu) | |
1694 | { | |
1695 | int ret; | |
1696 | u64 *entry, new_entry; | |
1697 | int id = vcpu->vcpu_id; | |
1698 | struct vcpu_svm *svm = to_svm(vcpu); | |
1699 | ||
1700 | ret = avic_init_access_page(vcpu); | |
1701 | if (ret) | |
1702 | return ret; | |
1703 | ||
1704 | if (id >= AVIC_MAX_PHYSICAL_ID_COUNT) | |
1705 | return -EINVAL; | |
1706 | ||
1707 | if (!svm->vcpu.arch.apic->regs) | |
1708 | return -EINVAL; | |
1709 | ||
1710 | svm->avic_backing_page = virt_to_page(svm->vcpu.arch.apic->regs); | |
1711 | ||
1712 | /* Setting AVIC backing page address in the phy APIC ID table */ | |
1713 | entry = avic_get_physical_id_entry(vcpu, id); | |
1714 | if (!entry) | |
1715 | return -EINVAL; | |
1716 | ||
1717 | new_entry = READ_ONCE(*entry); | |
d0ec49d4 TL |
1718 | new_entry = __sme_set((page_to_phys(svm->avic_backing_page) & |
1719 | AVIC_PHYSICAL_ID_ENTRY_BACKING_PAGE_MASK) | | |
1720 | AVIC_PHYSICAL_ID_ENTRY_VALID_MASK); | |
44a95dae SS |
1721 | WRITE_ONCE(*entry, new_entry); |
1722 | ||
1723 | svm->avic_physical_id_cache = entry; | |
1724 | ||
1725 | return 0; | |
1726 | } | |
1727 | ||
1654efcb BS |
1728 | static void __sev_asid_free(int asid) |
1729 | { | |
70cd94e6 BS |
1730 | struct svm_cpu_data *sd; |
1731 | int cpu, pos; | |
1654efcb BS |
1732 | |
1733 | pos = asid - 1; | |
1734 | clear_bit(pos, sev_asid_bitmap); | |
70cd94e6 BS |
1735 | |
1736 | for_each_possible_cpu(cpu) { | |
1737 | sd = per_cpu(svm_data, cpu); | |
1738 | sd->sev_vmcbs[pos] = NULL; | |
1739 | } | |
1654efcb BS |
1740 | } |
1741 | ||
1742 | static void sev_asid_free(struct kvm *kvm) | |
1743 | { | |
81811c16 | 1744 | struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info; |
1654efcb BS |
1745 | |
1746 | __sev_asid_free(sev->asid); | |
1747 | } | |
1748 | ||
59414c98 BS |
1749 | static void sev_unbind_asid(struct kvm *kvm, unsigned int handle) |
1750 | { | |
1751 | struct sev_data_decommission *decommission; | |
1752 | struct sev_data_deactivate *data; | |
1753 | ||
1754 | if (!handle) | |
1755 | return; | |
1756 | ||
1757 | data = kzalloc(sizeof(*data), GFP_KERNEL); | |
1758 | if (!data) | |
1759 | return; | |
1760 | ||
1761 | /* deactivate handle */ | |
1762 | data->handle = handle; | |
1763 | sev_guest_deactivate(data, NULL); | |
1764 | ||
1765 | wbinvd_on_all_cpus(); | |
1766 | sev_guest_df_flush(NULL); | |
1767 | kfree(data); | |
1768 | ||
1769 | decommission = kzalloc(sizeof(*decommission), GFP_KERNEL); | |
1770 | if (!decommission) | |
1771 | return; | |
1772 | ||
1773 | /* decommission handle */ | |
1774 | decommission->handle = handle; | |
1775 | sev_guest_decommission(decommission, NULL); | |
1776 | ||
1777 | kfree(decommission); | |
1778 | } | |
1779 | ||
89c50580 BS |
1780 | static struct page **sev_pin_memory(struct kvm *kvm, unsigned long uaddr, |
1781 | unsigned long ulen, unsigned long *n, | |
1782 | int write) | |
1783 | { | |
81811c16 | 1784 | struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info; |
89c50580 BS |
1785 | unsigned long npages, npinned, size; |
1786 | unsigned long locked, lock_limit; | |
1787 | struct page **pages; | |
86bf20cb DC |
1788 | unsigned long first, last; |
1789 | ||
1790 | if (ulen == 0 || uaddr + ulen < uaddr) | |
1791 | return NULL; | |
89c50580 BS |
1792 | |
1793 | /* Calculate number of pages. */ | |
1794 | first = (uaddr & PAGE_MASK) >> PAGE_SHIFT; | |
1795 | last = ((uaddr + ulen - 1) & PAGE_MASK) >> PAGE_SHIFT; | |
1796 | npages = (last - first + 1); | |
1797 | ||
1798 | locked = sev->pages_locked + npages; | |
1799 | lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
1800 | if (locked > lock_limit && !capable(CAP_IPC_LOCK)) { | |
1801 | pr_err("SEV: %lu locked pages exceed the lock limit of %lu.\n", locked, lock_limit); | |
1802 | return NULL; | |
1803 | } | |
1804 | ||
1805 | /* Avoid using vmalloc for smaller buffers. */ | |
1806 | size = npages * sizeof(struct page *); | |
1807 | if (size > PAGE_SIZE) | |
1ec69647 BG |
1808 | pages = __vmalloc(size, GFP_KERNEL_ACCOUNT | __GFP_ZERO, |
1809 | PAGE_KERNEL); | |
89c50580 | 1810 | else |
1ec69647 | 1811 | pages = kmalloc(size, GFP_KERNEL_ACCOUNT); |
89c50580 BS |
1812 | |
1813 | if (!pages) | |
1814 | return NULL; | |
1815 | ||
1816 | /* Pin the user virtual address. */ | |
73b0140b | 1817 | npinned = get_user_pages_fast(uaddr, npages, FOLL_WRITE, pages); |
89c50580 BS |
1818 | if (npinned != npages) { |
1819 | pr_err("SEV: Failure locking %lu pages.\n", npages); | |
1820 | goto err; | |
1821 | } | |
1822 | ||
1823 | *n = npages; | |
1824 | sev->pages_locked = locked; | |
1825 | ||
1826 | return pages; | |
1827 | ||
1828 | err: | |
1829 | if (npinned > 0) | |
1830 | release_pages(pages, npinned); | |
1831 | ||
1832 | kvfree(pages); | |
1833 | return NULL; | |
1834 | } | |
1835 | ||
1836 | static void sev_unpin_memory(struct kvm *kvm, struct page **pages, | |
1837 | unsigned long npages) | |
1838 | { | |
81811c16 | 1839 | struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info; |
89c50580 BS |
1840 | |
1841 | release_pages(pages, npages); | |
1842 | kvfree(pages); | |
1843 | sev->pages_locked -= npages; | |
1844 | } | |
1845 | ||
1846 | static void sev_clflush_pages(struct page *pages[], unsigned long npages) | |
1847 | { | |
1848 | uint8_t *page_virtual; | |
1849 | unsigned long i; | |
1850 | ||
1851 | if (npages == 0 || pages == NULL) | |
1852 | return; | |
1853 | ||
1854 | for (i = 0; i < npages; i++) { | |
1855 | page_virtual = kmap_atomic(pages[i]); | |
1856 | clflush_cache_range(page_virtual, PAGE_SIZE); | |
1857 | kunmap_atomic(page_virtual); | |
1858 | } | |
1859 | } | |
1860 | ||
1e80fdc0 BS |
1861 | static void __unregister_enc_region_locked(struct kvm *kvm, |
1862 | struct enc_region *region) | |
1863 | { | |
1864 | /* | |
1865 | * The guest may change the memory encryption attribute from C=0 -> C=1 | |
1866 | * or vice versa for this memory range. Lets make sure caches are | |
1867 | * flushed to ensure that guest data gets written into memory with | |
1868 | * correct C-bit. | |
1869 | */ | |
1870 | sev_clflush_pages(region->pages, region->npages); | |
1871 | ||
1872 | sev_unpin_memory(kvm, region->pages, region->npages); | |
1873 | list_del(®ion->list); | |
1874 | kfree(region); | |
1875 | } | |
1876 | ||
434a1e94 SC |
1877 | static struct kvm *svm_vm_alloc(void) |
1878 | { | |
1ec69647 BG |
1879 | struct kvm_svm *kvm_svm = __vmalloc(sizeof(struct kvm_svm), |
1880 | GFP_KERNEL_ACCOUNT | __GFP_ZERO, | |
1881 | PAGE_KERNEL); | |
81811c16 | 1882 | return &kvm_svm->kvm; |
434a1e94 SC |
1883 | } |
1884 | ||
1885 | static void svm_vm_free(struct kvm *kvm) | |
1886 | { | |
d1e5b0e9 | 1887 | vfree(to_kvm_svm(kvm)); |
434a1e94 SC |
1888 | } |
1889 | ||
1654efcb BS |
1890 | static void sev_vm_destroy(struct kvm *kvm) |
1891 | { | |
81811c16 | 1892 | struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info; |
1e80fdc0 BS |
1893 | struct list_head *head = &sev->regions_list; |
1894 | struct list_head *pos, *q; | |
59414c98 | 1895 | |
1654efcb BS |
1896 | if (!sev_guest(kvm)) |
1897 | return; | |
1898 | ||
1e80fdc0 BS |
1899 | mutex_lock(&kvm->lock); |
1900 | ||
1901 | /* | |
1902 | * if userspace was terminated before unregistering the memory regions | |
1903 | * then lets unpin all the registered memory. | |
1904 | */ | |
1905 | if (!list_empty(head)) { | |
1906 | list_for_each_safe(pos, q, head) { | |
1907 | __unregister_enc_region_locked(kvm, | |
1908 | list_entry(pos, struct enc_region, list)); | |
1909 | } | |
1910 | } | |
1911 | ||
1912 | mutex_unlock(&kvm->lock); | |
1913 | ||
59414c98 | 1914 | sev_unbind_asid(kvm, sev->handle); |
1654efcb BS |
1915 | sev_asid_free(kvm); |
1916 | } | |
1917 | ||
44a95dae SS |
1918 | static void avic_vm_destroy(struct kvm *kvm) |
1919 | { | |
5881f737 | 1920 | unsigned long flags; |
81811c16 | 1921 | struct kvm_svm *kvm_svm = to_kvm_svm(kvm); |
44a95dae | 1922 | |
3863dff0 DV |
1923 | if (!avic) |
1924 | return; | |
1925 | ||
81811c16 SC |
1926 | if (kvm_svm->avic_logical_id_table_page) |
1927 | __free_page(kvm_svm->avic_logical_id_table_page); | |
1928 | if (kvm_svm->avic_physical_id_table_page) | |
1929 | __free_page(kvm_svm->avic_physical_id_table_page); | |
5881f737 SS |
1930 | |
1931 | spin_lock_irqsave(&svm_vm_data_hash_lock, flags); | |
81811c16 | 1932 | hash_del(&kvm_svm->hnode); |
5881f737 | 1933 | spin_unlock_irqrestore(&svm_vm_data_hash_lock, flags); |
44a95dae SS |
1934 | } |
1935 | ||
1654efcb BS |
1936 | static void svm_vm_destroy(struct kvm *kvm) |
1937 | { | |
1938 | avic_vm_destroy(kvm); | |
1939 | sev_vm_destroy(kvm); | |
1940 | } | |
1941 | ||
44a95dae SS |
1942 | static int avic_vm_init(struct kvm *kvm) |
1943 | { | |
5881f737 | 1944 | unsigned long flags; |
3f0d4db7 | 1945 | int err = -ENOMEM; |
81811c16 SC |
1946 | struct kvm_svm *kvm_svm = to_kvm_svm(kvm); |
1947 | struct kvm_svm *k2; | |
44a95dae SS |
1948 | struct page *p_page; |
1949 | struct page *l_page; | |
3f0d4db7 | 1950 | u32 vm_id; |
44a95dae SS |
1951 | |
1952 | if (!avic) | |
1953 | return 0; | |
1954 | ||
1955 | /* Allocating physical APIC ID table (4KB) */ | |
1ec69647 | 1956 | p_page = alloc_page(GFP_KERNEL_ACCOUNT); |
44a95dae SS |
1957 | if (!p_page) |
1958 | goto free_avic; | |
1959 | ||
81811c16 | 1960 | kvm_svm->avic_physical_id_table_page = p_page; |
44a95dae SS |
1961 | clear_page(page_address(p_page)); |
1962 | ||
1963 | /* Allocating logical APIC ID table (4KB) */ | |
1ec69647 | 1964 | l_page = alloc_page(GFP_KERNEL_ACCOUNT); |
44a95dae SS |
1965 | if (!l_page) |
1966 | goto free_avic; | |
1967 | ||
81811c16 | 1968 | kvm_svm->avic_logical_id_table_page = l_page; |
44a95dae SS |
1969 | clear_page(page_address(l_page)); |
1970 | ||
5881f737 | 1971 | spin_lock_irqsave(&svm_vm_data_hash_lock, flags); |
3f0d4db7 DV |
1972 | again: |
1973 | vm_id = next_vm_id = (next_vm_id + 1) & AVIC_VM_ID_MASK; | |
1974 | if (vm_id == 0) { /* id is 1-based, zero is not okay */ | |
1975 | next_vm_id_wrapped = 1; | |
1976 | goto again; | |
1977 | } | |
1978 | /* Is it still in use? Only possible if wrapped at least once */ | |
1979 | if (next_vm_id_wrapped) { | |
81811c16 SC |
1980 | hash_for_each_possible(svm_vm_data_hash, k2, hnode, vm_id) { |
1981 | if (k2->avic_vm_id == vm_id) | |
3f0d4db7 DV |
1982 | goto again; |
1983 | } | |
1984 | } | |
81811c16 SC |
1985 | kvm_svm->avic_vm_id = vm_id; |
1986 | hash_add(svm_vm_data_hash, &kvm_svm->hnode, kvm_svm->avic_vm_id); | |
5881f737 SS |
1987 | spin_unlock_irqrestore(&svm_vm_data_hash_lock, flags); |
1988 | ||
44a95dae SS |
1989 | return 0; |
1990 | ||
1991 | free_avic: | |
1992 | avic_vm_destroy(kvm); | |
1993 | return err; | |
6aa8b732 AK |
1994 | } |
1995 | ||
411b44ba SS |
1996 | static inline int |
1997 | avic_update_iommu_vcpu_affinity(struct kvm_vcpu *vcpu, int cpu, bool r) | |
8221c137 | 1998 | { |
411b44ba SS |
1999 | int ret = 0; |
2000 | unsigned long flags; | |
2001 | struct amd_svm_iommu_ir *ir; | |
8221c137 SS |
2002 | struct vcpu_svm *svm = to_svm(vcpu); |
2003 | ||
411b44ba SS |
2004 | if (!kvm_arch_has_assigned_device(vcpu->kvm)) |
2005 | return 0; | |
8221c137 | 2006 | |
411b44ba SS |
2007 | /* |
2008 | * Here, we go through the per-vcpu ir_list to update all existing | |
2009 | * interrupt remapping table entry targeting this vcpu. | |
2010 | */ | |
2011 | spin_lock_irqsave(&svm->ir_list_lock, flags); | |
8221c137 | 2012 | |
411b44ba SS |
2013 | if (list_empty(&svm->ir_list)) |
2014 | goto out; | |
8221c137 | 2015 | |
411b44ba SS |
2016 | list_for_each_entry(ir, &svm->ir_list, node) { |
2017 | ret = amd_iommu_update_ga(cpu, r, ir->data); | |
2018 | if (ret) | |
2019 | break; | |
2020 | } | |
2021 | out: | |
2022 | spin_unlock_irqrestore(&svm->ir_list_lock, flags); | |
2023 | return ret; | |
8221c137 SS |
2024 | } |
2025 | ||
2026 | static void avic_vcpu_load(struct kvm_vcpu *vcpu, int cpu) | |
2027 | { | |
2028 | u64 entry; | |
2029 | /* ID = 0xff (broadcast), ID > 0xff (reserved) */ | |
7d669f50 | 2030 | int h_physical_id = kvm_cpu_get_apicid(cpu); |
8221c137 SS |
2031 | struct vcpu_svm *svm = to_svm(vcpu); |
2032 | ||
2033 | if (!kvm_vcpu_apicv_active(vcpu)) | |
2034 | return; | |
2035 | ||
c9bcd3e3 SS |
2036 | /* |
2037 | * Since the host physical APIC id is 8 bits, | |
2038 | * we can support host APIC ID upto 255. | |
2039 | */ | |
2040 | if (WARN_ON(h_physical_id > AVIC_PHYSICAL_ID_ENTRY_HOST_PHYSICAL_ID_MASK)) | |
8221c137 SS |
2041 | return; |
2042 | ||
2043 | entry = READ_ONCE(*(svm->avic_physical_id_cache)); | |
2044 | WARN_ON(entry & AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK); | |
2045 | ||
2046 | entry &= ~AVIC_PHYSICAL_ID_ENTRY_HOST_PHYSICAL_ID_MASK; | |
2047 | entry |= (h_physical_id & AVIC_PHYSICAL_ID_ENTRY_HOST_PHYSICAL_ID_MASK); | |
2048 | ||
2049 | entry &= ~AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK; | |
2050 | if (svm->avic_is_running) | |
2051 | entry |= AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK; | |
2052 | ||
2053 | WRITE_ONCE(*(svm->avic_physical_id_cache), entry); | |
411b44ba SS |
2054 | avic_update_iommu_vcpu_affinity(vcpu, h_physical_id, |
2055 | svm->avic_is_running); | |
8221c137 SS |
2056 | } |
2057 | ||
2058 | static void avic_vcpu_put(struct kvm_vcpu *vcpu) | |
2059 | { | |
2060 | u64 entry; | |
2061 | struct vcpu_svm *svm = to_svm(vcpu); | |
2062 | ||
2063 | if (!kvm_vcpu_apicv_active(vcpu)) | |
2064 | return; | |
2065 | ||
2066 | entry = READ_ONCE(*(svm->avic_physical_id_cache)); | |
411b44ba SS |
2067 | if (entry & AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK) |
2068 | avic_update_iommu_vcpu_affinity(vcpu, -1, 0); | |
2069 | ||
8221c137 SS |
2070 | entry &= ~AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK; |
2071 | WRITE_ONCE(*(svm->avic_physical_id_cache), entry); | |
6aa8b732 AK |
2072 | } |
2073 | ||
411b44ba SS |
2074 | /** |
2075 | * This function is called during VCPU halt/unhalt. | |
2076 | */ | |
2077 | static void avic_set_running(struct kvm_vcpu *vcpu, bool is_run) | |
2078 | { | |
2079 | struct vcpu_svm *svm = to_svm(vcpu); | |
2080 | ||
2081 | svm->avic_is_running = is_run; | |
2082 | if (is_run) | |
2083 | avic_vcpu_load(vcpu, vcpu->cpu); | |
2084 | else | |
2085 | avic_vcpu_put(vcpu); | |
2086 | } | |
2087 | ||
d28bc9dd | 2088 | static void svm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event) |
04d2cc77 AK |
2089 | { |
2090 | struct vcpu_svm *svm = to_svm(vcpu); | |
66f7b72e JS |
2091 | u32 dummy; |
2092 | u32 eax = 1; | |
04d2cc77 | 2093 | |
518e7b94 | 2094 | vcpu->arch.microcode_version = 0x01000065; |
b2ac58f9 | 2095 | svm->spec_ctrl = 0; |
ccbcd267 | 2096 | svm->virt_spec_ctrl = 0; |
b2ac58f9 | 2097 | |
d28bc9dd NA |
2098 | if (!init_event) { |
2099 | svm->vcpu.arch.apic_base = APIC_DEFAULT_PHYS_BASE | | |
2100 | MSR_IA32_APICBASE_ENABLE; | |
2101 | if (kvm_vcpu_is_reset_bsp(&svm->vcpu)) | |
2102 | svm->vcpu.arch.apic_base |= MSR_IA32_APICBASE_BSP; | |
2103 | } | |
5690891b | 2104 | init_vmcb(svm); |
70433389 | 2105 | |
e911eb3b | 2106 | kvm_cpuid(vcpu, &eax, &dummy, &dummy, &dummy, true); |
de3cd117 | 2107 | kvm_rdx_write(vcpu, eax); |
44a95dae SS |
2108 | |
2109 | if (kvm_vcpu_apicv_active(vcpu) && !init_event) | |
2110 | avic_update_vapic_bar(svm, APIC_DEFAULT_PHYS_BASE); | |
04d2cc77 AK |
2111 | } |
2112 | ||
dfa20099 SS |
2113 | static int avic_init_vcpu(struct vcpu_svm *svm) |
2114 | { | |
2115 | int ret; | |
2116 | ||
67034bb9 | 2117 | if (!kvm_vcpu_apicv_active(&svm->vcpu)) |
dfa20099 SS |
2118 | return 0; |
2119 | ||
2120 | ret = avic_init_backing_page(&svm->vcpu); | |
2121 | if (ret) | |
2122 | return ret; | |
2123 | ||
2124 | INIT_LIST_HEAD(&svm->ir_list); | |
2125 | spin_lock_init(&svm->ir_list_lock); | |
98d90582 | 2126 | svm->dfr_reg = APIC_DFR_FLAT; |
dfa20099 SS |
2127 | |
2128 | return ret; | |
2129 | } | |
2130 | ||
fb3f0f51 | 2131 | static struct kvm_vcpu *svm_create_vcpu(struct kvm *kvm, unsigned int id) |
6aa8b732 | 2132 | { |
a2fa3e9f | 2133 | struct vcpu_svm *svm; |
6aa8b732 | 2134 | struct page *page; |
f65c229c | 2135 | struct page *msrpm_pages; |
b286d5d8 | 2136 | struct page *hsave_page; |
3d6368ef | 2137 | struct page *nested_msrpm_pages; |
fb3f0f51 | 2138 | int err; |
6aa8b732 | 2139 | |
1ec69647 | 2140 | svm = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL_ACCOUNT); |
fb3f0f51 RR |
2141 | if (!svm) { |
2142 | err = -ENOMEM; | |
2143 | goto out; | |
2144 | } | |
2145 | ||
d9a710e5 WL |
2146 | svm->vcpu.arch.user_fpu = kmem_cache_zalloc(x86_fpu_cache, |
2147 | GFP_KERNEL_ACCOUNT); | |
2148 | if (!svm->vcpu.arch.user_fpu) { | |
2149 | printk(KERN_ERR "kvm: failed to allocate kvm userspace's fpu\n"); | |
2150 | err = -ENOMEM; | |
2151 | goto free_partial_svm; | |
2152 | } | |
2153 | ||
1ec69647 BG |
2154 | svm->vcpu.arch.guest_fpu = kmem_cache_zalloc(x86_fpu_cache, |
2155 | GFP_KERNEL_ACCOUNT); | |
b666a4b6 MO |
2156 | if (!svm->vcpu.arch.guest_fpu) { |
2157 | printk(KERN_ERR "kvm: failed to allocate vcpu's fpu\n"); | |
2158 | err = -ENOMEM; | |
d9a710e5 | 2159 | goto free_user_fpu; |
b666a4b6 MO |
2160 | } |
2161 | ||
fb3f0f51 RR |
2162 | err = kvm_vcpu_init(&svm->vcpu, kvm, id); |
2163 | if (err) | |
2164 | goto free_svm; | |
2165 | ||
b7af4043 | 2166 | err = -ENOMEM; |
1ec69647 | 2167 | page = alloc_page(GFP_KERNEL_ACCOUNT); |
b7af4043 | 2168 | if (!page) |
fb3f0f51 | 2169 | goto uninit; |
6aa8b732 | 2170 | |
1ec69647 | 2171 | msrpm_pages = alloc_pages(GFP_KERNEL_ACCOUNT, MSRPM_ALLOC_ORDER); |
f65c229c | 2172 | if (!msrpm_pages) |
b7af4043 | 2173 | goto free_page1; |
3d6368ef | 2174 | |
1ec69647 | 2175 | nested_msrpm_pages = alloc_pages(GFP_KERNEL_ACCOUNT, MSRPM_ALLOC_ORDER); |
3d6368ef | 2176 | if (!nested_msrpm_pages) |
b7af4043 | 2177 | goto free_page2; |
f65c229c | 2178 | |
1ec69647 | 2179 | hsave_page = alloc_page(GFP_KERNEL_ACCOUNT); |
b286d5d8 | 2180 | if (!hsave_page) |
b7af4043 TY |
2181 | goto free_page3; |
2182 | ||
dfa20099 SS |
2183 | err = avic_init_vcpu(svm); |
2184 | if (err) | |
2185 | goto free_page4; | |
44a95dae | 2186 | |
8221c137 SS |
2187 | /* We initialize this flag to true to make sure that the is_running |
2188 | * bit would be set the first time the vcpu is loaded. | |
2189 | */ | |
2190 | svm->avic_is_running = true; | |
2191 | ||
e6aa9abd | 2192 | svm->nested.hsave = page_address(hsave_page); |
b286d5d8 | 2193 | |
b7af4043 TY |
2194 | svm->msrpm = page_address(msrpm_pages); |
2195 | svm_vcpu_init_msrpm(svm->msrpm); | |
2196 | ||
e6aa9abd | 2197 | svm->nested.msrpm = page_address(nested_msrpm_pages); |
323c3d80 | 2198 | svm_vcpu_init_msrpm(svm->nested.msrpm); |
3d6368ef | 2199 | |
a2fa3e9f GH |
2200 | svm->vmcb = page_address(page); |
2201 | clear_page(svm->vmcb); | |
d0ec49d4 | 2202 | svm->vmcb_pa = __sme_set(page_to_pfn(page) << PAGE_SHIFT); |
a2fa3e9f | 2203 | svm->asid_generation = 0; |
5690891b | 2204 | init_vmcb(svm); |
6aa8b732 | 2205 | |
2b036c6b BO |
2206 | svm_init_osvw(&svm->vcpu); |
2207 | ||
fb3f0f51 | 2208 | return &svm->vcpu; |
36241b8c | 2209 | |
44a95dae SS |
2210 | free_page4: |
2211 | __free_page(hsave_page); | |
b7af4043 TY |
2212 | free_page3: |
2213 | __free_pages(nested_msrpm_pages, MSRPM_ALLOC_ORDER); | |
2214 | free_page2: | |
2215 | __free_pages(msrpm_pages, MSRPM_ALLOC_ORDER); | |
2216 | free_page1: | |
2217 | __free_page(page); | |
fb3f0f51 RR |
2218 | uninit: |
2219 | kvm_vcpu_uninit(&svm->vcpu); | |
2220 | free_svm: | |
b666a4b6 | 2221 | kmem_cache_free(x86_fpu_cache, svm->vcpu.arch.guest_fpu); |
d9a710e5 WL |
2222 | free_user_fpu: |
2223 | kmem_cache_free(x86_fpu_cache, svm->vcpu.arch.user_fpu); | |
b666a4b6 | 2224 | free_partial_svm: |
a4770347 | 2225 | kmem_cache_free(kvm_vcpu_cache, svm); |
fb3f0f51 RR |
2226 | out: |
2227 | return ERR_PTR(err); | |
6aa8b732 AK |
2228 | } |
2229 | ||
fd65d314 JM |
2230 | static void svm_clear_current_vmcb(struct vmcb *vmcb) |
2231 | { | |
2232 | int i; | |
2233 | ||
2234 | for_each_online_cpu(i) | |
2235 | cmpxchg(&per_cpu(svm_data, i)->current_vmcb, vmcb, NULL); | |
2236 | } | |
2237 | ||
6aa8b732 AK |
2238 | static void svm_free_vcpu(struct kvm_vcpu *vcpu) |
2239 | { | |
a2fa3e9f GH |
2240 | struct vcpu_svm *svm = to_svm(vcpu); |
2241 | ||
fd65d314 JM |
2242 | /* |
2243 | * The vmcb page can be recycled, causing a false negative in | |
2244 | * svm_vcpu_load(). So, ensure that no logical CPU has this | |
2245 | * vmcb page recorded as its current vmcb. | |
2246 | */ | |
2247 | svm_clear_current_vmcb(svm->vmcb); | |
2248 | ||
d0ec49d4 | 2249 | __free_page(pfn_to_page(__sme_clr(svm->vmcb_pa) >> PAGE_SHIFT)); |
f65c229c | 2250 | __free_pages(virt_to_page(svm->msrpm), MSRPM_ALLOC_ORDER); |
e6aa9abd JR |
2251 | __free_page(virt_to_page(svm->nested.hsave)); |
2252 | __free_pages(virt_to_page(svm->nested.msrpm), MSRPM_ALLOC_ORDER); | |
fb3f0f51 | 2253 | kvm_vcpu_uninit(vcpu); |
d9a710e5 | 2254 | kmem_cache_free(x86_fpu_cache, svm->vcpu.arch.user_fpu); |
b666a4b6 | 2255 | kmem_cache_free(x86_fpu_cache, svm->vcpu.arch.guest_fpu); |
a4770347 | 2256 | kmem_cache_free(kvm_vcpu_cache, svm); |
6aa8b732 AK |
2257 | } |
2258 | ||
15ad7146 | 2259 | static void svm_vcpu_load(struct kvm_vcpu *vcpu, int cpu) |
6aa8b732 | 2260 | { |
a2fa3e9f | 2261 | struct vcpu_svm *svm = to_svm(vcpu); |
15d45071 | 2262 | struct svm_cpu_data *sd = per_cpu(svm_data, cpu); |
15ad7146 | 2263 | int i; |
0cc5064d | 2264 | |
0cc5064d | 2265 | if (unlikely(cpu != vcpu->cpu)) { |
4b656b12 | 2266 | svm->asid_generation = 0; |
8d28fec4 | 2267 | mark_all_dirty(svm->vmcb); |
0cc5064d | 2268 | } |
94dfbdb3 | 2269 | |
82ca2d10 AK |
2270 | #ifdef CONFIG_X86_64 |
2271 | rdmsrl(MSR_GS_BASE, to_svm(vcpu)->host.gs_base); | |
2272 | #endif | |
dacccfdd AK |
2273 | savesegment(fs, svm->host.fs); |
2274 | savesegment(gs, svm->host.gs); | |
2275 | svm->host.ldt = kvm_read_ldt(); | |
2276 | ||
94dfbdb3 | 2277 | for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++) |
a2fa3e9f | 2278 | rdmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]); |
fbc0db76 | 2279 | |
ad721883 HZ |
2280 | if (static_cpu_has(X86_FEATURE_TSCRATEMSR)) { |
2281 | u64 tsc_ratio = vcpu->arch.tsc_scaling_ratio; | |
2282 | if (tsc_ratio != __this_cpu_read(current_tsc_ratio)) { | |
2283 | __this_cpu_write(current_tsc_ratio, tsc_ratio); | |
2284 | wrmsrl(MSR_AMD64_TSC_RATIO, tsc_ratio); | |
2285 | } | |
fbc0db76 | 2286 | } |
46896c73 PB |
2287 | /* This assumes that the kernel never uses MSR_TSC_AUX */ |
2288 | if (static_cpu_has(X86_FEATURE_RDTSCP)) | |
2289 | wrmsrl(MSR_TSC_AUX, svm->tsc_aux); | |
8221c137 | 2290 | |
15d45071 AR |
2291 | if (sd->current_vmcb != svm->vmcb) { |
2292 | sd->current_vmcb = svm->vmcb; | |
2293 | indirect_branch_prediction_barrier(); | |
2294 | } | |
8221c137 | 2295 | avic_vcpu_load(vcpu, cpu); |
6aa8b732 AK |
2296 | } |
2297 | ||
2298 | static void svm_vcpu_put(struct kvm_vcpu *vcpu) | |
2299 | { | |
a2fa3e9f | 2300 | struct vcpu_svm *svm = to_svm(vcpu); |
94dfbdb3 AL |
2301 | int i; |
2302 | ||
8221c137 SS |
2303 | avic_vcpu_put(vcpu); |
2304 | ||
e1beb1d3 | 2305 | ++vcpu->stat.host_state_reload; |
dacccfdd AK |
2306 | kvm_load_ldt(svm->host.ldt); |
2307 | #ifdef CONFIG_X86_64 | |
2308 | loadsegment(fs, svm->host.fs); | |
296f781a | 2309 | wrmsrl(MSR_KERNEL_GS_BASE, current->thread.gsbase); |
893a5ab6 | 2310 | load_gs_index(svm->host.gs); |
dacccfdd | 2311 | #else |
831ca609 | 2312 | #ifdef CONFIG_X86_32_LAZY_GS |
dacccfdd | 2313 | loadsegment(gs, svm->host.gs); |
831ca609 | 2314 | #endif |
dacccfdd | 2315 | #endif |
94dfbdb3 | 2316 | for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++) |
a2fa3e9f | 2317 | wrmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]); |
6aa8b732 AK |
2318 | } |
2319 | ||
8221c137 SS |
2320 | static void svm_vcpu_blocking(struct kvm_vcpu *vcpu) |
2321 | { | |
2322 | avic_set_running(vcpu, false); | |
2323 | } | |
2324 | ||
2325 | static void svm_vcpu_unblocking(struct kvm_vcpu *vcpu) | |
2326 | { | |
2327 | avic_set_running(vcpu, true); | |
2328 | } | |
2329 | ||
6aa8b732 AK |
2330 | static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu) |
2331 | { | |
9b611747 LP |
2332 | struct vcpu_svm *svm = to_svm(vcpu); |
2333 | unsigned long rflags = svm->vmcb->save.rflags; | |
2334 | ||
2335 | if (svm->nmi_singlestep) { | |
2336 | /* Hide our flags if they were not set by the guest */ | |
2337 | if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_TF)) | |
2338 | rflags &= ~X86_EFLAGS_TF; | |
2339 | if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_RF)) | |
2340 | rflags &= ~X86_EFLAGS_RF; | |
2341 | } | |
2342 | return rflags; | |
6aa8b732 AK |
2343 | } |
2344 | ||
2345 | static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags) | |
2346 | { | |
9b611747 LP |
2347 | if (to_svm(vcpu)->nmi_singlestep) |
2348 | rflags |= (X86_EFLAGS_TF | X86_EFLAGS_RF); | |
2349 | ||
ae9fedc7 | 2350 | /* |
bb3541f1 | 2351 | * Any change of EFLAGS.VM is accompanied by a reload of SS |
ae9fedc7 PB |
2352 | * (caused by either a task switch or an inter-privilege IRET), |
2353 | * so we do not need to update the CPL here. | |
2354 | */ | |
a2fa3e9f | 2355 | to_svm(vcpu)->vmcb->save.rflags = rflags; |
6aa8b732 AK |
2356 | } |
2357 | ||
6de4f3ad AK |
2358 | static void svm_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg) |
2359 | { | |
2360 | switch (reg) { | |
2361 | case VCPU_EXREG_PDPTR: | |
2362 | BUG_ON(!npt_enabled); | |
9f8fe504 | 2363 | load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu)); |
6de4f3ad AK |
2364 | break; |
2365 | default: | |
2366 | BUG(); | |
2367 | } | |
2368 | } | |
2369 | ||
f0b85051 AG |
2370 | static void svm_set_vintr(struct vcpu_svm *svm) |
2371 | { | |
8a05a1b8 | 2372 | set_intercept(svm, INTERCEPT_VINTR); |
f0b85051 AG |
2373 | } |
2374 | ||
2375 | static void svm_clear_vintr(struct vcpu_svm *svm) | |
2376 | { | |
8a05a1b8 | 2377 | clr_intercept(svm, INTERCEPT_VINTR); |
f0b85051 AG |
2378 | } |
2379 | ||
6aa8b732 AK |
2380 | static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg) |
2381 | { | |
a2fa3e9f | 2382 | struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save; |
6aa8b732 AK |
2383 | |
2384 | switch (seg) { | |
2385 | case VCPU_SREG_CS: return &save->cs; | |
2386 | case VCPU_SREG_DS: return &save->ds; | |
2387 | case VCPU_SREG_ES: return &save->es; | |
2388 | case VCPU_SREG_FS: return &save->fs; | |
2389 | case VCPU_SREG_GS: return &save->gs; | |
2390 | case VCPU_SREG_SS: return &save->ss; | |
2391 | case VCPU_SREG_TR: return &save->tr; | |
2392 | case VCPU_SREG_LDTR: return &save->ldtr; | |
2393 | } | |
2394 | BUG(); | |
8b6d44c7 | 2395 | return NULL; |
6aa8b732 AK |
2396 | } |
2397 | ||
2398 | static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg) | |
2399 | { | |
2400 | struct vmcb_seg *s = svm_seg(vcpu, seg); | |
2401 | ||
2402 | return s->base; | |
2403 | } | |
2404 | ||
2405 | static void svm_get_segment(struct kvm_vcpu *vcpu, | |
2406 | struct kvm_segment *var, int seg) | |
2407 | { | |
2408 | struct vmcb_seg *s = svm_seg(vcpu, seg); | |
2409 | ||
2410 | var->base = s->base; | |
2411 | var->limit = s->limit; | |
2412 | var->selector = s->selector; | |
2413 | var->type = s->attrib & SVM_SELECTOR_TYPE_MASK; | |
2414 | var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1; | |
2415 | var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3; | |
2416 | var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1; | |
2417 | var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1; | |
2418 | var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1; | |
2419 | var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1; | |
80112c89 JM |
2420 | |
2421 | /* | |
2422 | * AMD CPUs circa 2014 track the G bit for all segments except CS. | |
2423 | * However, the SVM spec states that the G bit is not observed by the | |
2424 | * CPU, and some VMware virtual CPUs drop the G bit for all segments. | |
2425 | * So let's synthesize a legal G bit for all segments, this helps | |
2426 | * running KVM nested. It also helps cross-vendor migration, because | |
2427 | * Intel's vmentry has a check on the 'G' bit. | |
2428 | */ | |
2429 | var->g = s->limit > 0xfffff; | |
25022acc | 2430 | |
e0231715 JR |
2431 | /* |
2432 | * AMD's VMCB does not have an explicit unusable field, so emulate it | |
19bca6ab AP |
2433 | * for cross vendor migration purposes by "not present" |
2434 | */ | |
8eae9570 | 2435 | var->unusable = !var->present; |
19bca6ab | 2436 | |
1fbdc7a5 | 2437 | switch (seg) { |
1fbdc7a5 AP |
2438 | case VCPU_SREG_TR: |
2439 | /* | |
2440 | * Work around a bug where the busy flag in the tr selector | |
2441 | * isn't exposed | |
2442 | */ | |
c0d09828 | 2443 | var->type |= 0x2; |
1fbdc7a5 AP |
2444 | break; |
2445 | case VCPU_SREG_DS: | |
2446 | case VCPU_SREG_ES: | |
2447 | case VCPU_SREG_FS: | |
2448 | case VCPU_SREG_GS: | |
2449 | /* | |
2450 | * The accessed bit must always be set in the segment | |
2451 | * descriptor cache, although it can be cleared in the | |
2452 | * descriptor, the cached bit always remains at 1. Since | |
2453 | * Intel has a check on this, set it here to support | |
2454 | * cross-vendor migration. | |
2455 | */ | |
2456 | if (!var->unusable) | |
2457 | var->type |= 0x1; | |
2458 | break; | |
b586eb02 | 2459 | case VCPU_SREG_SS: |
e0231715 JR |
2460 | /* |
2461 | * On AMD CPUs sometimes the DB bit in the segment | |
b586eb02 AP |
2462 | * descriptor is left as 1, although the whole segment has |
2463 | * been made unusable. Clear it here to pass an Intel VMX | |
2464 | * entry check when cross vendor migrating. | |
2465 | */ | |
2466 | if (var->unusable) | |
2467 | var->db = 0; | |
d9c1b543 | 2468 | /* This is symmetric with svm_set_segment() */ |
33b458d2 | 2469 | var->dpl = to_svm(vcpu)->vmcb->save.cpl; |
b586eb02 | 2470 | break; |
1fbdc7a5 | 2471 | } |
6aa8b732 AK |
2472 | } |
2473 | ||
2e4d2653 IE |
2474 | static int svm_get_cpl(struct kvm_vcpu *vcpu) |
2475 | { | |
2476 | struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save; | |
2477 | ||
2478 | return save->cpl; | |
2479 | } | |
2480 | ||
89a27f4d | 2481 | static void svm_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt) |
6aa8b732 | 2482 | { |
a2fa3e9f GH |
2483 | struct vcpu_svm *svm = to_svm(vcpu); |
2484 | ||
89a27f4d GN |
2485 | dt->size = svm->vmcb->save.idtr.limit; |
2486 | dt->address = svm->vmcb->save.idtr.base; | |
6aa8b732 AK |
2487 | } |
2488 | ||
89a27f4d | 2489 | static void svm_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt) |
6aa8b732 | 2490 | { |
a2fa3e9f GH |
2491 | struct vcpu_svm *svm = to_svm(vcpu); |
2492 | ||
89a27f4d GN |
2493 | svm->vmcb->save.idtr.limit = dt->size; |
2494 | svm->vmcb->save.idtr.base = dt->address ; | |
17a703cb | 2495 | mark_dirty(svm->vmcb, VMCB_DT); |
6aa8b732 AK |
2496 | } |
2497 | ||
89a27f4d | 2498 | static void svm_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt) |
6aa8b732 | 2499 | { |
a2fa3e9f GH |
2500 | struct vcpu_svm *svm = to_svm(vcpu); |
2501 | ||
89a27f4d GN |
2502 | dt->size = svm->vmcb->save.gdtr.limit; |
2503 | dt->address = svm->vmcb->save.gdtr.base; | |
6aa8b732 AK |
2504 | } |
2505 | ||
89a27f4d | 2506 | static void svm_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt) |
6aa8b732 | 2507 | { |
a2fa3e9f GH |
2508 | struct vcpu_svm *svm = to_svm(vcpu); |
2509 | ||
89a27f4d GN |
2510 | svm->vmcb->save.gdtr.limit = dt->size; |
2511 | svm->vmcb->save.gdtr.base = dt->address ; | |
17a703cb | 2512 | mark_dirty(svm->vmcb, VMCB_DT); |
6aa8b732 AK |
2513 | } |
2514 | ||
e8467fda AK |
2515 | static void svm_decache_cr0_guest_bits(struct kvm_vcpu *vcpu) |
2516 | { | |
2517 | } | |
2518 | ||
aff48baa AK |
2519 | static void svm_decache_cr3(struct kvm_vcpu *vcpu) |
2520 | { | |
2521 | } | |
2522 | ||
25c4c276 | 2523 | static void svm_decache_cr4_guest_bits(struct kvm_vcpu *vcpu) |
399badf3 AK |
2524 | { |
2525 | } | |
2526 | ||
d225157b AK |
2527 | static void update_cr0_intercept(struct vcpu_svm *svm) |
2528 | { | |
2529 | ulong gcr0 = svm->vcpu.arch.cr0; | |
2530 | u64 *hcr0 = &svm->vmcb->save.cr0; | |
2531 | ||
bd7e5b08 PB |
2532 | *hcr0 = (*hcr0 & ~SVM_CR0_SELECTIVE_MASK) |
2533 | | (gcr0 & SVM_CR0_SELECTIVE_MASK); | |
d225157b | 2534 | |
dcca1a65 | 2535 | mark_dirty(svm->vmcb, VMCB_CR); |
d225157b | 2536 | |
bd7e5b08 | 2537 | if (gcr0 == *hcr0) { |
4ee546b4 RJ |
2538 | clr_cr_intercept(svm, INTERCEPT_CR0_READ); |
2539 | clr_cr_intercept(svm, INTERCEPT_CR0_WRITE); | |
d225157b | 2540 | } else { |
4ee546b4 RJ |
2541 | set_cr_intercept(svm, INTERCEPT_CR0_READ); |
2542 | set_cr_intercept(svm, INTERCEPT_CR0_WRITE); | |
d225157b AK |
2543 | } |
2544 | } | |
2545 | ||
6aa8b732 AK |
2546 | static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0) |
2547 | { | |
a2fa3e9f GH |
2548 | struct vcpu_svm *svm = to_svm(vcpu); |
2549 | ||
05b3e0c2 | 2550 | #ifdef CONFIG_X86_64 |
f6801dff | 2551 | if (vcpu->arch.efer & EFER_LME) { |
707d92fa | 2552 | if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) { |
f6801dff | 2553 | vcpu->arch.efer |= EFER_LMA; |
2b5203ee | 2554 | svm->vmcb->save.efer |= EFER_LMA | EFER_LME; |
6aa8b732 AK |
2555 | } |
2556 | ||
d77c26fc | 2557 | if (is_paging(vcpu) && !(cr0 & X86_CR0_PG)) { |
f6801dff | 2558 | vcpu->arch.efer &= ~EFER_LMA; |
2b5203ee | 2559 | svm->vmcb->save.efer &= ~(EFER_LMA | EFER_LME); |
6aa8b732 AK |
2560 | } |
2561 | } | |
2562 | #endif | |
ad312c7c | 2563 | vcpu->arch.cr0 = cr0; |
888f9f3e AK |
2564 | |
2565 | if (!npt_enabled) | |
2566 | cr0 |= X86_CR0_PG | X86_CR0_WP; | |
02daab21 | 2567 | |
bcf166a9 PB |
2568 | /* |
2569 | * re-enable caching here because the QEMU bios | |
2570 | * does not do it - this results in some delay at | |
2571 | * reboot | |
2572 | */ | |
2573 | if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED)) | |
2574 | cr0 &= ~(X86_CR0_CD | X86_CR0_NW); | |
a2fa3e9f | 2575 | svm->vmcb->save.cr0 = cr0; |
dcca1a65 | 2576 | mark_dirty(svm->vmcb, VMCB_CR); |
d225157b | 2577 | update_cr0_intercept(svm); |
6aa8b732 AK |
2578 | } |
2579 | ||
5e1746d6 | 2580 | static int svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4) |
6aa8b732 | 2581 | { |
1e02ce4c | 2582 | unsigned long host_cr4_mce = cr4_read_shadow() & X86_CR4_MCE; |
e5eab0ce JR |
2583 | unsigned long old_cr4 = to_svm(vcpu)->vmcb->save.cr4; |
2584 | ||
5e1746d6 NHE |
2585 | if (cr4 & X86_CR4_VMXE) |
2586 | return 1; | |
2587 | ||
e5eab0ce | 2588 | if (npt_enabled && ((old_cr4 ^ cr4) & X86_CR4_PGE)) |
c2ba05cc | 2589 | svm_flush_tlb(vcpu, true); |
6394b649 | 2590 | |
ec077263 JR |
2591 | vcpu->arch.cr4 = cr4; |
2592 | if (!npt_enabled) | |
2593 | cr4 |= X86_CR4_PAE; | |
6394b649 | 2594 | cr4 |= host_cr4_mce; |
ec077263 | 2595 | to_svm(vcpu)->vmcb->save.cr4 = cr4; |
dcca1a65 | 2596 | mark_dirty(to_svm(vcpu)->vmcb, VMCB_CR); |
5e1746d6 | 2597 | return 0; |
6aa8b732 AK |
2598 | } |
2599 | ||
2600 | static void svm_set_segment(struct kvm_vcpu *vcpu, | |
2601 | struct kvm_segment *var, int seg) | |
2602 | { | |
a2fa3e9f | 2603 | struct vcpu_svm *svm = to_svm(vcpu); |
6aa8b732 AK |
2604 | struct vmcb_seg *s = svm_seg(vcpu, seg); |
2605 | ||
2606 | s->base = var->base; | |
2607 | s->limit = var->limit; | |
2608 | s->selector = var->selector; | |
d9c1b543 RP |
2609 | s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK); |
2610 | s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT; | |
2611 | s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT; | |
2612 | s->attrib |= ((var->present & 1) && !var->unusable) << SVM_SELECTOR_P_SHIFT; | |
2613 | s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT; | |
2614 | s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT; | |
2615 | s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT; | |
2616 | s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT; | |
ae9fedc7 PB |
2617 | |
2618 | /* | |
2619 | * This is always accurate, except if SYSRET returned to a segment | |
2620 | * with SS.DPL != 3. Intel does not have this quirk, and always | |
2621 | * forces SS.DPL to 3 on sysret, so we ignore that case; fixing it | |
2622 | * would entail passing the CPL to userspace and back. | |
2623 | */ | |
2624 | if (seg == VCPU_SREG_SS) | |
d9c1b543 RP |
2625 | /* This is symmetric with svm_get_segment() */ |
2626 | svm->vmcb->save.cpl = (var->dpl & 3); | |
6aa8b732 | 2627 | |
060d0c9a | 2628 | mark_dirty(svm->vmcb, VMCB_SEG); |
6aa8b732 AK |
2629 | } |
2630 | ||
cbdb967a | 2631 | static void update_bp_intercept(struct kvm_vcpu *vcpu) |
6aa8b732 | 2632 | { |
d0bfb940 JK |
2633 | struct vcpu_svm *svm = to_svm(vcpu); |
2634 | ||
18c918c5 | 2635 | clr_exception_intercept(svm, BP_VECTOR); |
44c11430 | 2636 | |
d0bfb940 | 2637 | if (vcpu->guest_debug & KVM_GUESTDBG_ENABLE) { |
d0bfb940 | 2638 | if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) |
18c918c5 | 2639 | set_exception_intercept(svm, BP_VECTOR); |
d0bfb940 JK |
2640 | } else |
2641 | vcpu->guest_debug = 0; | |
44c11430 GN |
2642 | } |
2643 | ||
0fe1e009 | 2644 | static void new_asid(struct vcpu_svm *svm, struct svm_cpu_data *sd) |
6aa8b732 | 2645 | { |
0fe1e009 TH |
2646 | if (sd->next_asid > sd->max_asid) { |
2647 | ++sd->asid_generation; | |
4faefff3 | 2648 | sd->next_asid = sd->min_asid; |
a2fa3e9f | 2649 | svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID; |
6aa8b732 AK |
2650 | } |
2651 | ||
0fe1e009 TH |
2652 | svm->asid_generation = sd->asid_generation; |
2653 | svm->vmcb->control.asid = sd->next_asid++; | |
d48086d1 JR |
2654 | |
2655 | mark_dirty(svm->vmcb, VMCB_ASID); | |
6aa8b732 AK |
2656 | } |
2657 | ||
73aaf249 JK |
2658 | static u64 svm_get_dr6(struct kvm_vcpu *vcpu) |
2659 | { | |
2660 | return to_svm(vcpu)->vmcb->save.dr6; | |
2661 | } | |
2662 | ||
2663 | static void svm_set_dr6(struct kvm_vcpu *vcpu, unsigned long value) | |
2664 | { | |
2665 | struct vcpu_svm *svm = to_svm(vcpu); | |
2666 | ||
2667 | svm->vmcb->save.dr6 = value; | |
2668 | mark_dirty(svm->vmcb, VMCB_DR); | |
2669 | } | |
2670 | ||
facb0139 PB |
2671 | static void svm_sync_dirty_debug_regs(struct kvm_vcpu *vcpu) |
2672 | { | |
2673 | struct vcpu_svm *svm = to_svm(vcpu); | |
2674 | ||
2675 | get_debugreg(vcpu->arch.db[0], 0); | |
2676 | get_debugreg(vcpu->arch.db[1], 1); | |
2677 | get_debugreg(vcpu->arch.db[2], 2); | |
2678 | get_debugreg(vcpu->arch.db[3], 3); | |
2679 | vcpu->arch.dr6 = svm_get_dr6(vcpu); | |
2680 | vcpu->arch.dr7 = svm->vmcb->save.dr7; | |
2681 | ||
2682 | vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT; | |
2683 | set_dr_intercepts(svm); | |
2684 | } | |
2685 | ||
020df079 | 2686 | static void svm_set_dr7(struct kvm_vcpu *vcpu, unsigned long value) |
6aa8b732 | 2687 | { |
42dbaa5a | 2688 | struct vcpu_svm *svm = to_svm(vcpu); |
42dbaa5a | 2689 | |
020df079 | 2690 | svm->vmcb->save.dr7 = value; |
72214b96 | 2691 | mark_dirty(svm->vmcb, VMCB_DR); |
6aa8b732 AK |
2692 | } |
2693 | ||
851ba692 | 2694 | static int pf_interception(struct vcpu_svm *svm) |
6aa8b732 | 2695 | { |
0ede79e1 | 2696 | u64 fault_address = __sme_clr(svm->vmcb->control.exit_info_2); |
1261bfa3 | 2697 | u64 error_code = svm->vmcb->control.exit_info_1; |
6aa8b732 | 2698 | |
1261bfa3 | 2699 | return kvm_handle_page_fault(&svm->vcpu, error_code, fault_address, |
00b10fe1 BS |
2700 | static_cpu_has(X86_FEATURE_DECODEASSISTS) ? |
2701 | svm->vmcb->control.insn_bytes : NULL, | |
d0006530 PB |
2702 | svm->vmcb->control.insn_len); |
2703 | } | |
2704 | ||
2705 | static int npf_interception(struct vcpu_svm *svm) | |
2706 | { | |
0ede79e1 | 2707 | u64 fault_address = __sme_clr(svm->vmcb->control.exit_info_2); |
d0006530 PB |
2708 | u64 error_code = svm->vmcb->control.exit_info_1; |
2709 | ||
2710 | trace_kvm_page_fault(fault_address, error_code); | |
2711 | return kvm_mmu_page_fault(&svm->vcpu, fault_address, error_code, | |
00b10fe1 BS |
2712 | static_cpu_has(X86_FEATURE_DECODEASSISTS) ? |
2713 | svm->vmcb->control.insn_bytes : NULL, | |
d0006530 | 2714 | svm->vmcb->control.insn_len); |
6aa8b732 AK |
2715 | } |
2716 | ||
851ba692 | 2717 | static int db_interception(struct vcpu_svm *svm) |
d0bfb940 | 2718 | { |
851ba692 | 2719 | struct kvm_run *kvm_run = svm->vcpu.run; |
99c22179 | 2720 | struct kvm_vcpu *vcpu = &svm->vcpu; |
851ba692 | 2721 | |
d0bfb940 | 2722 | if (!(svm->vcpu.guest_debug & |
44c11430 | 2723 | (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) && |
6be7d306 | 2724 | !svm->nmi_singlestep) { |
d0bfb940 JK |
2725 | kvm_queue_exception(&svm->vcpu, DB_VECTOR); |
2726 | return 1; | |
2727 | } | |
44c11430 | 2728 | |
6be7d306 | 2729 | if (svm->nmi_singlestep) { |
4aebd0e9 | 2730 | disable_nmi_singlestep(svm); |
99c22179 VK |
2731 | /* Make sure we check for pending NMIs upon entry */ |
2732 | kvm_make_request(KVM_REQ_EVENT, vcpu); | |
44c11430 GN |
2733 | } |
2734 | ||
2735 | if (svm->vcpu.guest_debug & | |
e0231715 | 2736 | (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) { |
44c11430 GN |
2737 | kvm_run->exit_reason = KVM_EXIT_DEBUG; |
2738 | kvm_run->debug.arch.pc = | |
2739 | svm->vmcb->save.cs.base + svm->vmcb->save.rip; | |
2740 | kvm_run->debug.arch.exception = DB_VECTOR; | |
2741 | return 0; | |
2742 | } | |
2743 | ||
2744 | return 1; | |
d0bfb940 JK |
2745 | } |
2746 | ||
851ba692 | 2747 | static int bp_interception(struct vcpu_svm *svm) |
d0bfb940 | 2748 | { |
851ba692 AK |
2749 | struct kvm_run *kvm_run = svm->vcpu.run; |
2750 | ||
d0bfb940 JK |
2751 | kvm_run->exit_reason = KVM_EXIT_DEBUG; |
2752 | kvm_run->debug.arch.pc = svm->vmcb->save.cs.base + svm->vmcb->save.rip; | |
2753 | kvm_run->debug.arch.exception = BP_VECTOR; | |
2754 | return 0; | |
2755 | } | |
2756 | ||
851ba692 | 2757 | static int ud_interception(struct vcpu_svm *svm) |
7aa81cc0 | 2758 | { |
082d06ed | 2759 | return handle_ud(&svm->vcpu); |
7aa81cc0 AL |
2760 | } |
2761 | ||
54a20552 EN |
2762 | static int ac_interception(struct vcpu_svm *svm) |
2763 | { | |
2764 | kvm_queue_exception_e(&svm->vcpu, AC_VECTOR, 0); | |
2765 | return 1; | |
2766 | } | |
2767 | ||
9718420e LA |
2768 | static int gp_interception(struct vcpu_svm *svm) |
2769 | { | |
2770 | struct kvm_vcpu *vcpu = &svm->vcpu; | |
2771 | u32 error_code = svm->vmcb->control.exit_info_1; | |
2772 | int er; | |
2773 | ||
2774 | WARN_ON_ONCE(!enable_vmware_backdoor); | |
2775 | ||
0ce97a2b | 2776 | er = kvm_emulate_instruction(vcpu, |
9718420e LA |
2777 | EMULTYPE_VMWARE | EMULTYPE_NO_UD_ON_FAIL); |
2778 | if (er == EMULATE_USER_EXIT) | |
2779 | return 0; | |
2780 | else if (er != EMULATE_DONE) | |
2781 | kvm_queue_exception_e(vcpu, GP_VECTOR, error_code); | |
2782 | return 1; | |
2783 | } | |
2784 | ||
67ec6607 JR |
2785 | static bool is_erratum_383(void) |
2786 | { | |
2787 | int err, i; | |
2788 | u64 value; | |
2789 | ||
2790 | if (!erratum_383_found) | |
2791 | return false; | |
2792 | ||
2793 | value = native_read_msr_safe(MSR_IA32_MC0_STATUS, &err); | |
2794 | if (err) | |
2795 | return false; | |
2796 | ||
2797 | /* Bit 62 may or may not be set for this mce */ | |
2798 | value &= ~(1ULL << 62); | |
2799 | ||
2800 | if (value != 0xb600000000010015ULL) | |
2801 | return false; | |
2802 | ||
2803 | /* Clear MCi_STATUS registers */ | |
2804 | for (i = 0; i < 6; ++i) | |
2805 | native_write_msr_safe(MSR_IA32_MCx_STATUS(i), 0, 0); | |
2806 | ||
2807 | value = native_read_msr_safe(MSR_IA32_MCG_STATUS, &err); | |
2808 | if (!err) { | |
2809 | u32 low, high; | |
2810 | ||
2811 | value &= ~(1ULL << 2); | |
2812 | low = lower_32_bits(value); | |
2813 | high = upper_32_bits(value); | |
2814 | ||
2815 | native_write_msr_safe(MSR_IA32_MCG_STATUS, low, high); | |
2816 | } | |
2817 | ||
2818 | /* Flush tlb to evict multi-match entries */ | |
2819 | __flush_tlb_all(); | |
2820 | ||
2821 | return true; | |
2822 | } | |
2823 | ||
fe5913e4 | 2824 | static void svm_handle_mce(struct vcpu_svm *svm) |
53371b50 | 2825 | { |
67ec6607 JR |
2826 | if (is_erratum_383()) { |
2827 | /* | |
2828 | * Erratum 383 triggered. Guest state is corrupt so kill the | |
2829 | * guest. | |
2830 | */ | |
2831 | pr_err("KVM: Guest triggered AMD Erratum 383\n"); | |
2832 | ||
a8eeb04a | 2833 | kvm_make_request(KVM_REQ_TRIPLE_FAULT, &svm->vcpu); |
67ec6607 JR |
2834 | |
2835 | return; | |
2836 | } | |
2837 | ||
53371b50 JR |
2838 | /* |
2839 | * On an #MC intercept the MCE handler is not called automatically in | |
2840 | * the host. So do it by hand here. | |
2841 | */ | |
2842 | asm volatile ( | |
2843 | "int $0x12\n"); | |
2844 | /* not sure if we ever come back to this point */ | |
2845 | ||
fe5913e4 JR |
2846 | return; |
2847 | } | |
2848 | ||
2849 | static int mc_interception(struct vcpu_svm *svm) | |
2850 | { | |
53371b50 JR |
2851 | return 1; |
2852 | } | |
2853 | ||
851ba692 | 2854 | static int shutdown_interception(struct vcpu_svm *svm) |
46fe4ddd | 2855 | { |
851ba692 AK |
2856 | struct kvm_run *kvm_run = svm->vcpu.run; |
2857 | ||
46fe4ddd JR |
2858 | /* |
2859 | * VMCB is undefined after a SHUTDOWN intercept | |
2860 | * so reinitialize it. | |
2861 | */ | |
a2fa3e9f | 2862 | clear_page(svm->vmcb); |
5690891b | 2863 | init_vmcb(svm); |
46fe4ddd JR |
2864 | |
2865 | kvm_run->exit_reason = KVM_EXIT_SHUTDOWN; | |
2866 | return 0; | |
2867 | } | |
2868 | ||
851ba692 | 2869 | static int io_interception(struct vcpu_svm *svm) |
6aa8b732 | 2870 | { |
cf8f70bf | 2871 | struct kvm_vcpu *vcpu = &svm->vcpu; |
d77c26fc | 2872 | u32 io_info = svm->vmcb->control.exit_info_1; /* address size bug? */ |
dca7f128 | 2873 | int size, in, string; |
039576c0 | 2874 | unsigned port; |
6aa8b732 | 2875 | |
e756fc62 | 2876 | ++svm->vcpu.stat.io_exits; |
e70669ab | 2877 | string = (io_info & SVM_IOIO_STR_MASK) != 0; |
039576c0 | 2878 | in = (io_info & SVM_IOIO_TYPE_MASK) != 0; |
8370c3d0 | 2879 | if (string) |
0ce97a2b | 2880 | return kvm_emulate_instruction(vcpu, 0) == EMULATE_DONE; |
cf8f70bf | 2881 | |
039576c0 AK |
2882 | port = io_info >> 16; |
2883 | size = (io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT; | |
cf8f70bf | 2884 | svm->next_rip = svm->vmcb->control.exit_info_2; |
cf8f70bf | 2885 | |
dca7f128 | 2886 | return kvm_fast_pio(&svm->vcpu, size, port, in); |
6aa8b732 AK |
2887 | } |
2888 | ||
851ba692 | 2889 | static int nmi_interception(struct vcpu_svm *svm) |
c47f098d JR |
2890 | { |
2891 | return 1; | |
2892 | } | |
2893 | ||
851ba692 | 2894 | static int intr_interception(struct vcpu_svm *svm) |
a0698055 JR |
2895 | { |
2896 | ++svm->vcpu.stat.irq_exits; | |
2897 | return 1; | |
2898 | } | |
2899 | ||
851ba692 | 2900 | static int nop_on_interception(struct vcpu_svm *svm) |
6aa8b732 AK |
2901 | { |
2902 | return 1; | |
2903 | } | |
2904 | ||
851ba692 | 2905 | static int halt_interception(struct vcpu_svm *svm) |
6aa8b732 | 2906 | { |
5fdbf976 | 2907 | svm->next_rip = kvm_rip_read(&svm->vcpu) + 1; |
e756fc62 | 2908 | return kvm_emulate_halt(&svm->vcpu); |
6aa8b732 AK |
2909 | } |
2910 | ||
851ba692 | 2911 | static int vmmcall_interception(struct vcpu_svm *svm) |
02e235bc | 2912 | { |
5fdbf976 | 2913 | svm->next_rip = kvm_rip_read(&svm->vcpu) + 3; |
0d9c055e | 2914 | return kvm_emulate_hypercall(&svm->vcpu); |
02e235bc AK |
2915 | } |
2916 | ||
5bd2edc3 JR |
2917 | static unsigned long nested_svm_get_tdp_cr3(struct kvm_vcpu *vcpu) |
2918 | { | |
2919 | struct vcpu_svm *svm = to_svm(vcpu); | |
2920 | ||
2921 | return svm->nested.nested_cr3; | |
2922 | } | |
2923 | ||
e4e517b4 AK |
2924 | static u64 nested_svm_get_tdp_pdptr(struct kvm_vcpu *vcpu, int index) |
2925 | { | |
2926 | struct vcpu_svm *svm = to_svm(vcpu); | |
2927 | u64 cr3 = svm->nested.nested_cr3; | |
2928 | u64 pdpte; | |
2929 | int ret; | |
2930 | ||
d0ec49d4 | 2931 | ret = kvm_vcpu_read_guest_page(vcpu, gpa_to_gfn(__sme_clr(cr3)), &pdpte, |
54bf36aa | 2932 | offset_in_page(cr3) + index * 8, 8); |
e4e517b4 AK |
2933 | if (ret) |
2934 | return 0; | |
2935 | return pdpte; | |
2936 | } | |
2937 | ||
5bd2edc3 JR |
2938 | static void nested_svm_set_tdp_cr3(struct kvm_vcpu *vcpu, |
2939 | unsigned long root) | |
2940 | { | |
2941 | struct vcpu_svm *svm = to_svm(vcpu); | |
2942 | ||
d0ec49d4 | 2943 | svm->vmcb->control.nested_cr3 = __sme_set(root); |
b2747166 | 2944 | mark_dirty(svm->vmcb, VMCB_NPT); |
5bd2edc3 JR |
2945 | } |
2946 | ||
6389ee94 AK |
2947 | static void nested_svm_inject_npf_exit(struct kvm_vcpu *vcpu, |
2948 | struct x86_exception *fault) | |
5bd2edc3 JR |
2949 | { |
2950 | struct vcpu_svm *svm = to_svm(vcpu); | |
2951 | ||
5e352519 PB |
2952 | if (svm->vmcb->control.exit_code != SVM_EXIT_NPF) { |
2953 | /* | |
2954 | * TODO: track the cause of the nested page fault, and | |
2955 | * correctly fill in the high bits of exit_info_1. | |
2956 | */ | |
2957 | svm->vmcb->control.exit_code = SVM_EXIT_NPF; | |
2958 | svm->vmcb->control.exit_code_hi = 0; | |
2959 | svm->vmcb->control.exit_info_1 = (1ULL << 32); | |
2960 | svm->vmcb->control.exit_info_2 = fault->address; | |
2961 | } | |
2962 | ||
2963 | svm->vmcb->control.exit_info_1 &= ~0xffffffffULL; | |
2964 | svm->vmcb->control.exit_info_1 |= fault->error_code; | |
2965 | ||
2966 | /* | |
2967 | * The present bit is always zero for page structure faults on real | |
2968 | * hardware. | |
2969 | */ | |
2970 | if (svm->vmcb->control.exit_info_1 & (2ULL << 32)) | |
2971 | svm->vmcb->control.exit_info_1 &= ~1; | |
5bd2edc3 JR |
2972 | |
2973 | nested_svm_vmexit(svm); | |
2974 | } | |
2975 | ||
8a3c1a33 | 2976 | static void nested_svm_init_mmu_context(struct kvm_vcpu *vcpu) |
4b16184c | 2977 | { |
ad896af0 | 2978 | WARN_ON(mmu_is_nested(vcpu)); |
3cf85f9f VK |
2979 | |
2980 | vcpu->arch.mmu = &vcpu->arch.guest_mmu; | |
ad896af0 | 2981 | kvm_init_shadow_mmu(vcpu); |
44dd3ffa VK |
2982 | vcpu->arch.mmu->set_cr3 = nested_svm_set_tdp_cr3; |
2983 | vcpu->arch.mmu->get_cr3 = nested_svm_get_tdp_cr3; | |
2984 | vcpu->arch.mmu->get_pdptr = nested_svm_get_tdp_pdptr; | |
2985 | vcpu->arch.mmu->inject_page_fault = nested_svm_inject_npf_exit; | |
2986 | vcpu->arch.mmu->shadow_root_level = get_npt_level(vcpu); | |
2987 | reset_shadow_zero_bits_mask(vcpu, vcpu->arch.mmu); | |
4b16184c | 2988 | vcpu->arch.walk_mmu = &vcpu->arch.nested_mmu; |
4b16184c JR |
2989 | } |
2990 | ||
2991 | static void nested_svm_uninit_mmu_context(struct kvm_vcpu *vcpu) | |
2992 | { | |
3cf85f9f | 2993 | vcpu->arch.mmu = &vcpu->arch.root_mmu; |
44dd3ffa | 2994 | vcpu->arch.walk_mmu = &vcpu->arch.root_mmu; |
4b16184c JR |
2995 | } |
2996 | ||
c0725420 AG |
2997 | static int nested_svm_check_permissions(struct vcpu_svm *svm) |
2998 | { | |
e9196ceb DC |
2999 | if (!(svm->vcpu.arch.efer & EFER_SVME) || |
3000 | !is_paging(&svm->vcpu)) { | |
c0725420 AG |
3001 | kvm_queue_exception(&svm->vcpu, UD_VECTOR); |
3002 | return 1; | |
3003 | } | |
3004 | ||
3005 | if (svm->vmcb->save.cpl) { | |
3006 | kvm_inject_gp(&svm->vcpu, 0); | |
3007 | return 1; | |
3008 | } | |
3009 | ||
e9196ceb | 3010 | return 0; |
c0725420 AG |
3011 | } |
3012 | ||
cf74a78b AG |
3013 | static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr, |
3014 | bool has_error_code, u32 error_code) | |
3015 | { | |
b8e88bc8 JR |
3016 | int vmexit; |
3017 | ||
2030753d | 3018 | if (!is_guest_mode(&svm->vcpu)) |
0295ad7d | 3019 | return 0; |
cf74a78b | 3020 | |
adfe20fb WL |
3021 | vmexit = nested_svm_intercept(svm); |
3022 | if (vmexit != NESTED_EXIT_DONE) | |
3023 | return 0; | |
3024 | ||
0295ad7d JR |
3025 | svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + nr; |
3026 | svm->vmcb->control.exit_code_hi = 0; | |
3027 | svm->vmcb->control.exit_info_1 = error_code; | |
b96fb439 PB |
3028 | |
3029 | /* | |
da998b46 JM |
3030 | * EXITINFO2 is undefined for all exception intercepts other |
3031 | * than #PF. | |
b96fb439 | 3032 | */ |
adfe20fb WL |
3033 | if (svm->vcpu.arch.exception.nested_apf) |
3034 | svm->vmcb->control.exit_info_2 = svm->vcpu.arch.apf.nested_apf_token; | |
da998b46 JM |
3035 | else if (svm->vcpu.arch.exception.has_payload) |
3036 | svm->vmcb->control.exit_info_2 = svm->vcpu.arch.exception.payload; | |
adfe20fb WL |
3037 | else |
3038 | svm->vmcb->control.exit_info_2 = svm->vcpu.arch.cr2; | |
b8e88bc8 | 3039 | |
adfe20fb | 3040 | svm->nested.exit_required = true; |
b8e88bc8 | 3041 | return vmexit; |
cf74a78b AG |
3042 | } |
3043 | ||
8fe54654 JR |
3044 | /* This function returns true if it is save to enable the irq window */ |
3045 | static inline bool nested_svm_intr(struct vcpu_svm *svm) | |
cf74a78b | 3046 | { |
2030753d | 3047 | if (!is_guest_mode(&svm->vcpu)) |
8fe54654 | 3048 | return true; |
cf74a78b | 3049 | |
26666957 | 3050 | if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK)) |
8fe54654 | 3051 | return true; |
cf74a78b | 3052 | |
26666957 | 3053 | if (!(svm->vcpu.arch.hflags & HF_HIF_MASK)) |
8fe54654 | 3054 | return false; |
cf74a78b | 3055 | |
a0a07cd2 GN |
3056 | /* |
3057 | * if vmexit was already requested (by intercepted exception | |
3058 | * for instance) do not overwrite it with "external interrupt" | |
3059 | * vmexit. | |
3060 | */ | |
3061 | if (svm->nested.exit_required) | |
3062 | return false; | |
3063 | ||
197717d5 JR |
3064 | svm->vmcb->control.exit_code = SVM_EXIT_INTR; |
3065 | svm->vmcb->control.exit_info_1 = 0; | |
3066 | svm->vmcb->control.exit_info_2 = 0; | |
26666957 | 3067 | |
cd3ff653 JR |
3068 | if (svm->nested.intercept & 1ULL) { |
3069 | /* | |
3070 | * The #vmexit can't be emulated here directly because this | |
c5ec2e56 | 3071 | * code path runs with irqs and preemption disabled. A |
cd3ff653 JR |
3072 | * #vmexit emulation might sleep. Only signal request for |
3073 | * the #vmexit here. | |
3074 | */ | |
3075 | svm->nested.exit_required = true; | |
236649de | 3076 | trace_kvm_nested_intr_vmexit(svm->vmcb->save.rip); |
8fe54654 | 3077 | return false; |
cf74a78b AG |
3078 | } |
3079 | ||
8fe54654 | 3080 | return true; |
cf74a78b AG |
3081 | } |
3082 | ||
887f500c JR |
3083 | /* This function returns true if it is save to enable the nmi window */ |
3084 | static inline bool nested_svm_nmi(struct vcpu_svm *svm) | |
3085 | { | |
2030753d | 3086 | if (!is_guest_mode(&svm->vcpu)) |
887f500c JR |
3087 | return true; |
3088 | ||
3089 | if (!(svm->nested.intercept & (1ULL << INTERCEPT_NMI))) | |
3090 | return true; | |
3091 | ||
3092 | svm->vmcb->control.exit_code = SVM_EXIT_NMI; | |
3093 | svm->nested.exit_required = true; | |
3094 | ||
3095 | return false; | |
cf74a78b AG |
3096 | } |
3097 | ||
ce2ac085 JR |
3098 | static int nested_svm_intercept_ioio(struct vcpu_svm *svm) |
3099 | { | |
9bf41833 JK |
3100 | unsigned port, size, iopm_len; |
3101 | u16 val, mask; | |
3102 | u8 start_bit; | |
ce2ac085 | 3103 | u64 gpa; |
34f80cfa | 3104 | |
ce2ac085 JR |
3105 | if (!(svm->nested.intercept & (1ULL << INTERCEPT_IOIO_PROT))) |
3106 | return NESTED_EXIT_HOST; | |
34f80cfa | 3107 | |
ce2ac085 | 3108 | port = svm->vmcb->control.exit_info_1 >> 16; |
9bf41833 JK |
3109 | size = (svm->vmcb->control.exit_info_1 & SVM_IOIO_SIZE_MASK) >> |
3110 | SVM_IOIO_SIZE_SHIFT; | |
ce2ac085 | 3111 | gpa = svm->nested.vmcb_iopm + (port / 8); |
9bf41833 JK |
3112 | start_bit = port % 8; |
3113 | iopm_len = (start_bit + size > 8) ? 2 : 1; | |
3114 | mask = (0xf >> (4 - size)) << start_bit; | |
3115 | val = 0; | |
ce2ac085 | 3116 | |
54bf36aa | 3117 | if (kvm_vcpu_read_guest(&svm->vcpu, gpa, &val, iopm_len)) |
9bf41833 | 3118 | return NESTED_EXIT_DONE; |
ce2ac085 | 3119 | |
9bf41833 | 3120 | return (val & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST; |
34f80cfa JR |
3121 | } |
3122 | ||
d2477826 | 3123 | static int nested_svm_exit_handled_msr(struct vcpu_svm *svm) |
4c2161ae | 3124 | { |
0d6b3537 JR |
3125 | u32 offset, msr, value; |
3126 | int write, mask; | |
4c2161ae | 3127 | |
3d62d9aa | 3128 | if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT))) |
d2477826 | 3129 | return NESTED_EXIT_HOST; |
3d62d9aa | 3130 | |
0d6b3537 JR |
3131 | msr = svm->vcpu.arch.regs[VCPU_REGS_RCX]; |
3132 | offset = svm_msrpm_offset(msr); | |
3133 | write = svm->vmcb->control.exit_info_1 & 1; | |
3134 | mask = 1 << ((2 * (msr & 0xf)) + write); | |
3d62d9aa | 3135 | |
0d6b3537 JR |
3136 | if (offset == MSR_INVALID) |
3137 | return NESTED_EXIT_DONE; | |
4c2161ae | 3138 | |
0d6b3537 JR |
3139 | /* Offset is in 32 bit units but need in 8 bit units */ |
3140 | offset *= 4; | |
4c2161ae | 3141 | |
54bf36aa | 3142 | if (kvm_vcpu_read_guest(&svm->vcpu, svm->nested.vmcb_msrpm + offset, &value, 4)) |
0d6b3537 | 3143 | return NESTED_EXIT_DONE; |
3d62d9aa | 3144 | |
0d6b3537 | 3145 | return (value & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST; |
4c2161ae JR |
3146 | } |
3147 | ||
ab2f4d73 LP |
3148 | /* DB exceptions for our internal use must not cause vmexit */ |
3149 | static int nested_svm_intercept_db(struct vcpu_svm *svm) | |
3150 | { | |
3151 | unsigned long dr6; | |
3152 | ||
3153 | /* if we're not singlestepping, it's not ours */ | |
3154 | if (!svm->nmi_singlestep) | |
3155 | return NESTED_EXIT_DONE; | |
3156 | ||
3157 | /* if it's not a singlestep exception, it's not ours */ | |
3158 | if (kvm_get_dr(&svm->vcpu, 6, &dr6)) | |
3159 | return NESTED_EXIT_DONE; | |
3160 | if (!(dr6 & DR6_BS)) | |
3161 | return NESTED_EXIT_DONE; | |
3162 | ||
3163 | /* if the guest is singlestepping, it should get the vmexit */ | |
3164 | if (svm->nmi_singlestep_guest_rflags & X86_EFLAGS_TF) { | |
3165 | disable_nmi_singlestep(svm); | |
3166 | return NESTED_EXIT_DONE; | |
3167 | } | |
3168 | ||
3169 | /* it's ours, the nested hypervisor must not see this one */ | |
3170 | return NESTED_EXIT_HOST; | |
3171 | } | |
3172 | ||
410e4d57 | 3173 | static int nested_svm_exit_special(struct vcpu_svm *svm) |
cf74a78b | 3174 | { |
cf74a78b | 3175 | u32 exit_code = svm->vmcb->control.exit_code; |
4c2161ae | 3176 | |
410e4d57 JR |
3177 | switch (exit_code) { |
3178 | case SVM_EXIT_INTR: | |
3179 | case SVM_EXIT_NMI: | |
ff47a49b | 3180 | case SVM_EXIT_EXCP_BASE + MC_VECTOR: |
410e4d57 | 3181 | return NESTED_EXIT_HOST; |
410e4d57 | 3182 | case SVM_EXIT_NPF: |
e0231715 | 3183 | /* For now we are always handling NPFs when using them */ |
410e4d57 JR |
3184 | if (npt_enabled) |
3185 | return NESTED_EXIT_HOST; | |
3186 | break; | |
410e4d57 | 3187 | case SVM_EXIT_EXCP_BASE + PF_VECTOR: |
631bc487 | 3188 | /* When we're shadowing, trap PFs, but not async PF */ |
1261bfa3 | 3189 | if (!npt_enabled && svm->vcpu.arch.apf.host_apf_reason == 0) |
410e4d57 JR |
3190 | return NESTED_EXIT_HOST; |
3191 | break; | |
3192 | default: | |
3193 | break; | |
cf74a78b AG |
3194 | } |
3195 | ||
410e4d57 JR |
3196 | return NESTED_EXIT_CONTINUE; |
3197 | } | |
3198 | ||
3199 | /* | |
3200 | * If this function returns true, this #vmexit was already handled | |
3201 | */ | |
b8e88bc8 | 3202 | static int nested_svm_intercept(struct vcpu_svm *svm) |
410e4d57 JR |
3203 | { |
3204 | u32 exit_code = svm->vmcb->control.exit_code; | |
3205 | int vmexit = NESTED_EXIT_HOST; | |
3206 | ||
cf74a78b | 3207 | switch (exit_code) { |
9c4e40b9 | 3208 | case SVM_EXIT_MSR: |
3d62d9aa | 3209 | vmexit = nested_svm_exit_handled_msr(svm); |
9c4e40b9 | 3210 | break; |
ce2ac085 JR |
3211 | case SVM_EXIT_IOIO: |
3212 | vmexit = nested_svm_intercept_ioio(svm); | |
3213 | break; | |
4ee546b4 RJ |
3214 | case SVM_EXIT_READ_CR0 ... SVM_EXIT_WRITE_CR8: { |
3215 | u32 bit = 1U << (exit_code - SVM_EXIT_READ_CR0); | |
3216 | if (svm->nested.intercept_cr & bit) | |
410e4d57 | 3217 | vmexit = NESTED_EXIT_DONE; |
cf74a78b AG |
3218 | break; |
3219 | } | |
3aed041a JR |
3220 | case SVM_EXIT_READ_DR0 ... SVM_EXIT_WRITE_DR7: { |
3221 | u32 bit = 1U << (exit_code - SVM_EXIT_READ_DR0); | |
3222 | if (svm->nested.intercept_dr & bit) | |
410e4d57 | 3223 | vmexit = NESTED_EXIT_DONE; |
cf74a78b AG |
3224 | break; |
3225 | } | |
3226 | case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: { | |
3227 | u32 excp_bits = 1 << (exit_code - SVM_EXIT_EXCP_BASE); | |
ab2f4d73 LP |
3228 | if (svm->nested.intercept_exceptions & excp_bits) { |
3229 | if (exit_code == SVM_EXIT_EXCP_BASE + DB_VECTOR) | |
3230 | vmexit = nested_svm_intercept_db(svm); | |
3231 | else | |
3232 | vmexit = NESTED_EXIT_DONE; | |
3233 | } | |
631bc487 GN |
3234 | /* async page fault always cause vmexit */ |
3235 | else if ((exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR) && | |
adfe20fb | 3236 | svm->vcpu.arch.exception.nested_apf != 0) |
631bc487 | 3237 | vmexit = NESTED_EXIT_DONE; |
cf74a78b AG |
3238 | break; |
3239 | } | |
228070b1 JR |
3240 | case SVM_EXIT_ERR: { |
3241 | vmexit = NESTED_EXIT_DONE; | |
3242 | break; | |
3243 | } | |
cf74a78b AG |
3244 | default: { |
3245 | u64 exit_bits = 1ULL << (exit_code - SVM_EXIT_INTR); | |
aad42c64 | 3246 | if (svm->nested.intercept & exit_bits) |
410e4d57 | 3247 | vmexit = NESTED_EXIT_DONE; |
cf74a78b AG |
3248 | } |
3249 | } | |
3250 | ||
b8e88bc8 JR |
3251 | return vmexit; |
3252 | } | |
3253 | ||
3254 | static int nested_svm_exit_handled(struct vcpu_svm *svm) | |
3255 | { | |
3256 | int vmexit; | |
3257 | ||
3258 | vmexit = nested_svm_intercept(svm); | |
3259 | ||
3260 | if (vmexit == NESTED_EXIT_DONE) | |
9c4e40b9 | 3261 | nested_svm_vmexit(svm); |
9c4e40b9 JR |
3262 | |
3263 | return vmexit; | |
cf74a78b AG |
3264 | } |
3265 | ||
0460a979 JR |
3266 | static inline void copy_vmcb_control_area(struct vmcb *dst_vmcb, struct vmcb *from_vmcb) |
3267 | { | |
3268 | struct vmcb_control_area *dst = &dst_vmcb->control; | |
3269 | struct vmcb_control_area *from = &from_vmcb->control; | |
3270 | ||
4ee546b4 | 3271 | dst->intercept_cr = from->intercept_cr; |
3aed041a | 3272 | dst->intercept_dr = from->intercept_dr; |
0460a979 JR |
3273 | dst->intercept_exceptions = from->intercept_exceptions; |
3274 | dst->intercept = from->intercept; | |
3275 | dst->iopm_base_pa = from->iopm_base_pa; | |
3276 | dst->msrpm_base_pa = from->msrpm_base_pa; | |
3277 | dst->tsc_offset = from->tsc_offset; | |
3278 | dst->asid = from->asid; | |
3279 | dst->tlb_ctl = from->tlb_ctl; | |
3280 | dst->int_ctl = from->int_ctl; | |
3281 | dst->int_vector = from->int_vector; | |
3282 | dst->int_state = from->int_state; | |
3283 | dst->exit_code = from->exit_code; | |
3284 | dst->exit_code_hi = from->exit_code_hi; | |
3285 | dst->exit_info_1 = from->exit_info_1; | |
3286 | dst->exit_info_2 = from->exit_info_2; | |
3287 | dst->exit_int_info = from->exit_int_info; | |
3288 | dst->exit_int_info_err = from->exit_int_info_err; | |
3289 | dst->nested_ctl = from->nested_ctl; | |
3290 | dst->event_inj = from->event_inj; | |
3291 | dst->event_inj_err = from->event_inj_err; | |
3292 | dst->nested_cr3 = from->nested_cr3; | |
0dc92119 | 3293 | dst->virt_ext = from->virt_ext; |
e081354d TW |
3294 | dst->pause_filter_count = from->pause_filter_count; |
3295 | dst->pause_filter_thresh = from->pause_filter_thresh; | |
0460a979 JR |
3296 | } |
3297 | ||
34f80cfa | 3298 | static int nested_svm_vmexit(struct vcpu_svm *svm) |
cf74a78b | 3299 | { |
8c5fbf1a | 3300 | int rc; |
34f80cfa | 3301 | struct vmcb *nested_vmcb; |
e6aa9abd | 3302 | struct vmcb *hsave = svm->nested.hsave; |
33740e40 | 3303 | struct vmcb *vmcb = svm->vmcb; |
8c5fbf1a | 3304 | struct kvm_host_map map; |
cf74a78b | 3305 | |
17897f36 JR |
3306 | trace_kvm_nested_vmexit_inject(vmcb->control.exit_code, |
3307 | vmcb->control.exit_info_1, | |
3308 | vmcb->control.exit_info_2, | |
3309 | vmcb->control.exit_int_info, | |
e097e5ff SH |
3310 | vmcb->control.exit_int_info_err, |
3311 | KVM_ISA_SVM); | |
17897f36 | 3312 | |
8f38302c | 3313 | rc = kvm_vcpu_map(&svm->vcpu, gpa_to_gfn(svm->nested.vmcb), &map); |
8c5fbf1a KA |
3314 | if (rc) { |
3315 | if (rc == -EINVAL) | |
3316 | kvm_inject_gp(&svm->vcpu, 0); | |
34f80cfa | 3317 | return 1; |
8c5fbf1a KA |
3318 | } |
3319 | ||
3320 | nested_vmcb = map.hva; | |
34f80cfa | 3321 | |
2030753d JR |
3322 | /* Exit Guest-Mode */ |
3323 | leave_guest_mode(&svm->vcpu); | |
06fc7772 JR |
3324 | svm->nested.vmcb = 0; |
3325 | ||
cf74a78b | 3326 | /* Give the current vmcb to the guest */ |
33740e40 JR |
3327 | disable_gif(svm); |
3328 | ||
3329 | nested_vmcb->save.es = vmcb->save.es; | |
3330 | nested_vmcb->save.cs = vmcb->save.cs; | |
3331 | nested_vmcb->save.ss = vmcb->save.ss; | |
3332 | nested_vmcb->save.ds = vmcb->save.ds; | |
3333 | nested_vmcb->save.gdtr = vmcb->save.gdtr; | |
3334 | nested_vmcb->save.idtr = vmcb->save.idtr; | |
3f6a9d16 | 3335 | nested_vmcb->save.efer = svm->vcpu.arch.efer; |
cdbbdc12 | 3336 | nested_vmcb->save.cr0 = kvm_read_cr0(&svm->vcpu); |
9f8fe504 | 3337 | nested_vmcb->save.cr3 = kvm_read_cr3(&svm->vcpu); |
33740e40 | 3338 | nested_vmcb->save.cr2 = vmcb->save.cr2; |
cdbbdc12 | 3339 | nested_vmcb->save.cr4 = svm->vcpu.arch.cr4; |
f6e78475 | 3340 | nested_vmcb->save.rflags = kvm_get_rflags(&svm->vcpu); |
33740e40 JR |
3341 | nested_vmcb->save.rip = vmcb->save.rip; |
3342 | nested_vmcb->save.rsp = vmcb->save.rsp; | |
3343 | nested_vmcb->save.rax = vmcb->save.rax; | |
3344 | nested_vmcb->save.dr7 = vmcb->save.dr7; | |
3345 | nested_vmcb->save.dr6 = vmcb->save.dr6; | |
3346 | nested_vmcb->save.cpl = vmcb->save.cpl; | |
3347 | ||
3348 | nested_vmcb->control.int_ctl = vmcb->control.int_ctl; | |
3349 | nested_vmcb->control.int_vector = vmcb->control.int_vector; | |
3350 | nested_vmcb->control.int_state = vmcb->control.int_state; | |
3351 | nested_vmcb->control.exit_code = vmcb->control.exit_code; | |
3352 | nested_vmcb->control.exit_code_hi = vmcb->control.exit_code_hi; | |
3353 | nested_vmcb->control.exit_info_1 = vmcb->control.exit_info_1; | |
3354 | nested_vmcb->control.exit_info_2 = vmcb->control.exit_info_2; | |
3355 | nested_vmcb->control.exit_int_info = vmcb->control.exit_int_info; | |
3356 | nested_vmcb->control.exit_int_info_err = vmcb->control.exit_int_info_err; | |
6092d3d3 JR |
3357 | |
3358 | if (svm->nrips_enabled) | |
3359 | nested_vmcb->control.next_rip = vmcb->control.next_rip; | |
8d23c466 AG |
3360 | |
3361 | /* | |
3362 | * If we emulate a VMRUN/#VMEXIT in the same host #vmexit cycle we have | |
3363 | * to make sure that we do not lose injected events. So check event_inj | |
3364 | * here and copy it to exit_int_info if it is valid. | |
3365 | * Exit_int_info and event_inj can't be both valid because the case | |
3366 | * below only happens on a VMRUN instruction intercept which has | |
3367 | * no valid exit_int_info set. | |
3368 | */ | |
3369 | if (vmcb->control.event_inj & SVM_EVTINJ_VALID) { | |
3370 | struct vmcb_control_area *nc = &nested_vmcb->control; | |
3371 | ||
3372 | nc->exit_int_info = vmcb->control.event_inj; | |
3373 | nc->exit_int_info_err = vmcb->control.event_inj_err; | |
3374 | } | |
3375 | ||
33740e40 JR |
3376 | nested_vmcb->control.tlb_ctl = 0; |
3377 | nested_vmcb->control.event_inj = 0; | |
3378 | nested_vmcb->control.event_inj_err = 0; | |
cf74a78b | 3379 | |
e081354d TW |
3380 | nested_vmcb->control.pause_filter_count = |
3381 | svm->vmcb->control.pause_filter_count; | |
3382 | nested_vmcb->control.pause_filter_thresh = | |
3383 | svm->vmcb->control.pause_filter_thresh; | |
3384 | ||
cf74a78b AG |
3385 | /* We always set V_INTR_MASKING and remember the old value in hflags */ |
3386 | if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK)) | |
3387 | nested_vmcb->control.int_ctl &= ~V_INTR_MASKING_MASK; | |
3388 | ||
cf74a78b | 3389 | /* Restore the original control entries */ |
0460a979 | 3390 | copy_vmcb_control_area(vmcb, hsave); |
cf74a78b | 3391 | |
e79f245d | 3392 | svm->vcpu.arch.tsc_offset = svm->vmcb->control.tsc_offset; |
219b65dc AG |
3393 | kvm_clear_exception_queue(&svm->vcpu); |
3394 | kvm_clear_interrupt_queue(&svm->vcpu); | |
cf74a78b | 3395 | |
4b16184c JR |
3396 | svm->nested.nested_cr3 = 0; |
3397 | ||
cf74a78b AG |
3398 | /* Restore selected save entries */ |
3399 | svm->vmcb->save.es = hsave->save.es; | |
3400 | svm->vmcb->save.cs = hsave->save.cs; | |
3401 | svm->vmcb->save.ss = hsave->save.ss; | |
3402 | svm->vmcb->save.ds = hsave->save.ds; | |
3403 | svm->vmcb->save.gdtr = hsave->save.gdtr; | |
3404 | svm->vmcb->save.idtr = hsave->save.idtr; | |
f6e78475 | 3405 | kvm_set_rflags(&svm->vcpu, hsave->save.rflags); |
cf74a78b AG |
3406 | svm_set_efer(&svm->vcpu, hsave->save.efer); |
3407 | svm_set_cr0(&svm->vcpu, hsave->save.cr0 | X86_CR0_PE); | |
3408 | svm_set_cr4(&svm->vcpu, hsave->save.cr4); | |
3409 | if (npt_enabled) { | |
3410 | svm->vmcb->save.cr3 = hsave->save.cr3; | |
3411 | svm->vcpu.arch.cr3 = hsave->save.cr3; | |
3412 | } else { | |
2390218b | 3413 | (void)kvm_set_cr3(&svm->vcpu, hsave->save.cr3); |
cf74a78b | 3414 | } |
de3cd117 | 3415 | kvm_rax_write(&svm->vcpu, hsave->save.rax); |
e9c16c78 PB |
3416 | kvm_rsp_write(&svm->vcpu, hsave->save.rsp); |
3417 | kvm_rip_write(&svm->vcpu, hsave->save.rip); | |
cf74a78b AG |
3418 | svm->vmcb->save.dr7 = 0; |
3419 | svm->vmcb->save.cpl = 0; | |
3420 | svm->vmcb->control.exit_int_info = 0; | |
3421 | ||
8d28fec4 RJ |
3422 | mark_all_dirty(svm->vmcb); |
3423 | ||
8c5fbf1a | 3424 | kvm_vcpu_unmap(&svm->vcpu, &map, true); |
cf74a78b | 3425 | |
4b16184c | 3426 | nested_svm_uninit_mmu_context(&svm->vcpu); |
cf74a78b AG |
3427 | kvm_mmu_reset_context(&svm->vcpu); |
3428 | kvm_mmu_load(&svm->vcpu); | |
3429 | ||
619ad846 VK |
3430 | /* |
3431 | * Drop what we picked up for L2 via svm_complete_interrupts() so it | |
3432 | * doesn't end up in L1. | |
3433 | */ | |
3434 | svm->vcpu.arch.nmi_injected = false; | |
3435 | kvm_clear_exception_queue(&svm->vcpu); | |
3436 | kvm_clear_interrupt_queue(&svm->vcpu); | |
3437 | ||
cf74a78b AG |
3438 | return 0; |
3439 | } | |
3d6368ef | 3440 | |
9738b2c9 | 3441 | static bool nested_svm_vmrun_msrpm(struct vcpu_svm *svm) |
3d6368ef | 3442 | { |
323c3d80 JR |
3443 | /* |
3444 | * This function merges the msr permission bitmaps of kvm and the | |
c5ec2e56 | 3445 | * nested vmcb. It is optimized in that it only merges the parts where |
323c3d80 JR |
3446 | * the kvm msr permission bitmap may contain zero bits |
3447 | */ | |
3d6368ef | 3448 | int i; |
9738b2c9 | 3449 | |
323c3d80 JR |
3450 | if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT))) |
3451 | return true; | |
9738b2c9 | 3452 | |
323c3d80 JR |
3453 | for (i = 0; i < MSRPM_OFFSETS; i++) { |
3454 | u32 value, p; | |
3455 | u64 offset; | |
9738b2c9 | 3456 | |
323c3d80 JR |
3457 | if (msrpm_offsets[i] == 0xffffffff) |
3458 | break; | |
3d6368ef | 3459 | |
0d6b3537 JR |
3460 | p = msrpm_offsets[i]; |
3461 | offset = svm->nested.vmcb_msrpm + (p * 4); | |
323c3d80 | 3462 | |
54bf36aa | 3463 | if (kvm_vcpu_read_guest(&svm->vcpu, offset, &value, 4)) |
323c3d80 JR |
3464 | return false; |
3465 | ||
3466 | svm->nested.msrpm[p] = svm->msrpm[p] | value; | |
3467 | } | |
3d6368ef | 3468 | |
d0ec49d4 | 3469 | svm->vmcb->control.msrpm_base_pa = __sme_set(__pa(svm->nested.msrpm)); |
9738b2c9 JR |
3470 | |
3471 | return true; | |
3d6368ef AG |
3472 | } |
3473 | ||
52c65a30 JR |
3474 | static bool nested_vmcb_checks(struct vmcb *vmcb) |
3475 | { | |
3476 | if ((vmcb->control.intercept & (1ULL << INTERCEPT_VMRUN)) == 0) | |
3477 | return false; | |
3478 | ||
dbe77584 JR |
3479 | if (vmcb->control.asid == 0) |
3480 | return false; | |
3481 | ||
cea3a19b TL |
3482 | if ((vmcb->control.nested_ctl & SVM_NESTED_CTL_NP_ENABLE) && |
3483 | !npt_enabled) | |
4b16184c JR |
3484 | return false; |
3485 | ||
52c65a30 JR |
3486 | return true; |
3487 | } | |
3488 | ||
c2634065 | 3489 | static void enter_svm_guest_mode(struct vcpu_svm *svm, u64 vmcb_gpa, |
8c5fbf1a | 3490 | struct vmcb *nested_vmcb, struct kvm_host_map *map) |
3d6368ef | 3491 | { |
f6e78475 | 3492 | if (kvm_get_rflags(&svm->vcpu) & X86_EFLAGS_IF) |
3d6368ef AG |
3493 | svm->vcpu.arch.hflags |= HF_HIF_MASK; |
3494 | else | |
3495 | svm->vcpu.arch.hflags &= ~HF_HIF_MASK; | |
3496 | ||
cea3a19b | 3497 | if (nested_vmcb->control.nested_ctl & SVM_NESTED_CTL_NP_ENABLE) { |
4b16184c JR |
3498 | svm->nested.nested_cr3 = nested_vmcb->control.nested_cr3; |
3499 | nested_svm_init_mmu_context(&svm->vcpu); | |
3500 | } | |
3501 | ||
3d6368ef AG |
3502 | /* Load the nested guest state */ |
3503 | svm->vmcb->save.es = nested_vmcb->save.es; | |
3504 | svm->vmcb->save.cs = nested_vmcb->save.cs; | |
3505 | svm->vmcb->save.ss = nested_vmcb->save.ss; | |
3506 | svm->vmcb->save.ds = nested_vmcb->save.ds; | |
3507 | svm->vmcb->save.gdtr = nested_vmcb->save.gdtr; | |
3508 | svm->vmcb->save.idtr = nested_vmcb->save.idtr; | |
f6e78475 | 3509 | kvm_set_rflags(&svm->vcpu, nested_vmcb->save.rflags); |
3d6368ef AG |
3510 | svm_set_efer(&svm->vcpu, nested_vmcb->save.efer); |
3511 | svm_set_cr0(&svm->vcpu, nested_vmcb->save.cr0); | |
3512 | svm_set_cr4(&svm->vcpu, nested_vmcb->save.cr4); | |
3513 | if (npt_enabled) { | |
3514 | svm->vmcb->save.cr3 = nested_vmcb->save.cr3; | |
3515 | svm->vcpu.arch.cr3 = nested_vmcb->save.cr3; | |
0e5cbe36 | 3516 | } else |
2390218b | 3517 | (void)kvm_set_cr3(&svm->vcpu, nested_vmcb->save.cr3); |
0e5cbe36 JR |
3518 | |
3519 | /* Guest paging mode is active - reset mmu */ | |
3520 | kvm_mmu_reset_context(&svm->vcpu); | |
3521 | ||
defbba56 | 3522 | svm->vmcb->save.cr2 = svm->vcpu.arch.cr2 = nested_vmcb->save.cr2; |
de3cd117 | 3523 | kvm_rax_write(&svm->vcpu, nested_vmcb->save.rax); |
e9c16c78 PB |
3524 | kvm_rsp_write(&svm->vcpu, nested_vmcb->save.rsp); |
3525 | kvm_rip_write(&svm->vcpu, nested_vmcb->save.rip); | |
e0231715 | 3526 | |
3d6368ef AG |
3527 | /* In case we don't even reach vcpu_run, the fields are not updated */ |
3528 | svm->vmcb->save.rax = nested_vmcb->save.rax; | |
3529 | svm->vmcb->save.rsp = nested_vmcb->save.rsp; | |
3530 | svm->vmcb->save.rip = nested_vmcb->save.rip; | |
3531 | svm->vmcb->save.dr7 = nested_vmcb->save.dr7; | |
3532 | svm->vmcb->save.dr6 = nested_vmcb->save.dr6; | |
3533 | svm->vmcb->save.cpl = nested_vmcb->save.cpl; | |
3534 | ||
f7138538 | 3535 | svm->nested.vmcb_msrpm = nested_vmcb->control.msrpm_base_pa & ~0x0fffULL; |
ce2ac085 | 3536 | svm->nested.vmcb_iopm = nested_vmcb->control.iopm_base_pa & ~0x0fffULL; |
3d6368ef | 3537 | |
aad42c64 | 3538 | /* cache intercepts */ |
4ee546b4 | 3539 | svm->nested.intercept_cr = nested_vmcb->control.intercept_cr; |
3aed041a | 3540 | svm->nested.intercept_dr = nested_vmcb->control.intercept_dr; |
aad42c64 JR |
3541 | svm->nested.intercept_exceptions = nested_vmcb->control.intercept_exceptions; |
3542 | svm->nested.intercept = nested_vmcb->control.intercept; | |
3543 | ||
c2ba05cc | 3544 | svm_flush_tlb(&svm->vcpu, true); |
3d6368ef | 3545 | svm->vmcb->control.int_ctl = nested_vmcb->control.int_ctl | V_INTR_MASKING_MASK; |
3d6368ef AG |
3546 | if (nested_vmcb->control.int_ctl & V_INTR_MASKING_MASK) |
3547 | svm->vcpu.arch.hflags |= HF_VINTR_MASK; | |
3548 | else | |
3549 | svm->vcpu.arch.hflags &= ~HF_VINTR_MASK; | |
3550 | ||
88ab24ad JR |
3551 | if (svm->vcpu.arch.hflags & HF_VINTR_MASK) { |
3552 | /* We only want the cr8 intercept bits of the guest */ | |
4ee546b4 RJ |
3553 | clr_cr_intercept(svm, INTERCEPT_CR8_READ); |
3554 | clr_cr_intercept(svm, INTERCEPT_CR8_WRITE); | |
88ab24ad JR |
3555 | } |
3556 | ||
0d945bd9 | 3557 | /* We don't want to see VMMCALLs from a nested guest */ |
8a05a1b8 | 3558 | clr_intercept(svm, INTERCEPT_VMMCALL); |
0d945bd9 | 3559 | |
e79f245d KA |
3560 | svm->vcpu.arch.tsc_offset += nested_vmcb->control.tsc_offset; |
3561 | svm->vmcb->control.tsc_offset = svm->vcpu.arch.tsc_offset; | |
3562 | ||
0dc92119 | 3563 | svm->vmcb->control.virt_ext = nested_vmcb->control.virt_ext; |
3d6368ef AG |
3564 | svm->vmcb->control.int_vector = nested_vmcb->control.int_vector; |
3565 | svm->vmcb->control.int_state = nested_vmcb->control.int_state; | |
3d6368ef AG |
3566 | svm->vmcb->control.event_inj = nested_vmcb->control.event_inj; |
3567 | svm->vmcb->control.event_inj_err = nested_vmcb->control.event_inj_err; | |
3568 | ||
e081354d TW |
3569 | svm->vmcb->control.pause_filter_count = |
3570 | nested_vmcb->control.pause_filter_count; | |
3571 | svm->vmcb->control.pause_filter_thresh = | |
3572 | nested_vmcb->control.pause_filter_thresh; | |
3573 | ||
8c5fbf1a | 3574 | kvm_vcpu_unmap(&svm->vcpu, map, true); |
9738b2c9 | 3575 | |
2030753d JR |
3576 | /* Enter Guest-Mode */ |
3577 | enter_guest_mode(&svm->vcpu); | |
3578 | ||
384c6368 JR |
3579 | /* |
3580 | * Merge guest and host intercepts - must be called with vcpu in | |
3581 | * guest-mode to take affect here | |
3582 | */ | |
3583 | recalc_intercepts(svm); | |
3584 | ||
06fc7772 | 3585 | svm->nested.vmcb = vmcb_gpa; |
9738b2c9 | 3586 | |
2af9194d | 3587 | enable_gif(svm); |
3d6368ef | 3588 | |
8d28fec4 | 3589 | mark_all_dirty(svm->vmcb); |
c2634065 LP |
3590 | } |
3591 | ||
3592 | static bool nested_svm_vmrun(struct vcpu_svm *svm) | |
3593 | { | |
8c5fbf1a | 3594 | int rc; |
c2634065 LP |
3595 | struct vmcb *nested_vmcb; |
3596 | struct vmcb *hsave = svm->nested.hsave; | |
3597 | struct vmcb *vmcb = svm->vmcb; | |
8c5fbf1a | 3598 | struct kvm_host_map map; |
c2634065 LP |
3599 | u64 vmcb_gpa; |
3600 | ||
3601 | vmcb_gpa = svm->vmcb->save.rax; | |
3602 | ||
8f38302c | 3603 | rc = kvm_vcpu_map(&svm->vcpu, gpa_to_gfn(vmcb_gpa), &map); |
8c5fbf1a KA |
3604 | if (rc) { |
3605 | if (rc == -EINVAL) | |
3606 | kvm_inject_gp(&svm->vcpu, 0); | |
c2634065 | 3607 | return false; |
8c5fbf1a KA |
3608 | } |
3609 | ||
3610 | nested_vmcb = map.hva; | |
c2634065 LP |
3611 | |
3612 | if (!nested_vmcb_checks(nested_vmcb)) { | |
3613 | nested_vmcb->control.exit_code = SVM_EXIT_ERR; | |
3614 | nested_vmcb->control.exit_code_hi = 0; | |
3615 | nested_vmcb->control.exit_info_1 = 0; | |
3616 | nested_vmcb->control.exit_info_2 = 0; | |
3617 | ||
8c5fbf1a | 3618 | kvm_vcpu_unmap(&svm->vcpu, &map, true); |
c2634065 LP |
3619 | |
3620 | return false; | |
3621 | } | |
3622 | ||
3623 | trace_kvm_nested_vmrun(svm->vmcb->save.rip, vmcb_gpa, | |
3624 | nested_vmcb->save.rip, | |
3625 | nested_vmcb->control.int_ctl, | |
3626 | nested_vmcb->control.event_inj, | |
3627 | nested_vmcb->control.nested_ctl); | |
3628 | ||
3629 | trace_kvm_nested_intercepts(nested_vmcb->control.intercept_cr & 0xffff, | |
3630 | nested_vmcb->control.intercept_cr >> 16, | |
3631 | nested_vmcb->control.intercept_exceptions, | |
3632 | nested_vmcb->control.intercept); | |
3633 | ||
3634 | /* Clear internal status */ | |
3635 | kvm_clear_exception_queue(&svm->vcpu); | |
3636 | kvm_clear_interrupt_queue(&svm->vcpu); | |
3637 | ||
3638 | /* | |
3639 | * Save the old vmcb, so we don't need to pick what we save, but can | |
3640 | * restore everything when a VMEXIT occurs | |
3641 | */ | |
3642 | hsave->save.es = vmcb->save.es; | |
3643 | hsave->save.cs = vmcb->save.cs; | |
3644 | hsave->save.ss = vmcb->save.ss; | |
3645 | hsave->save.ds = vmcb->save.ds; | |
3646 | hsave->save.gdtr = vmcb->save.gdtr; | |
3647 | hsave->save.idtr = vmcb->save.idtr; | |
3648 | hsave->save.efer = svm->vcpu.arch.efer; | |
3649 | hsave->save.cr0 = kvm_read_cr0(&svm->vcpu); | |
3650 | hsave->save.cr4 = svm->vcpu.arch.cr4; | |
3651 | hsave->save.rflags = kvm_get_rflags(&svm->vcpu); | |
3652 | hsave->save.rip = kvm_rip_read(&svm->vcpu); | |
3653 | hsave->save.rsp = vmcb->save.rsp; | |
3654 | hsave->save.rax = vmcb->save.rax; | |
3655 | if (npt_enabled) | |
3656 | hsave->save.cr3 = vmcb->save.cr3; | |
3657 | else | |
3658 | hsave->save.cr3 = kvm_read_cr3(&svm->vcpu); | |
3659 | ||
3660 | copy_vmcb_control_area(hsave, vmcb); | |
3661 | ||
8c5fbf1a | 3662 | enter_svm_guest_mode(svm, vmcb_gpa, nested_vmcb, &map); |
8d28fec4 | 3663 | |
9738b2c9 | 3664 | return true; |
3d6368ef AG |
3665 | } |
3666 | ||
9966bf68 | 3667 | static void nested_svm_vmloadsave(struct vmcb *from_vmcb, struct vmcb *to_vmcb) |
5542675b AG |
3668 | { |
3669 | to_vmcb->save.fs = from_vmcb->save.fs; | |
3670 | to_vmcb->save.gs = from_vmcb->save.gs; | |
3671 | to_vmcb->save.tr = from_vmcb->save.tr; | |
3672 | to_vmcb->save.ldtr = from_vmcb->save.ldtr; | |
3673 | to_vmcb->save.kernel_gs_base = from_vmcb->save.kernel_gs_base; | |
3674 | to_vmcb->save.star = from_vmcb->save.star; | |
3675 | to_vmcb->save.lstar = from_vmcb->save.lstar; | |
3676 | to_vmcb->save.cstar = from_vmcb->save.cstar; | |
3677 | to_vmcb->save.sfmask = from_vmcb->save.sfmask; | |
3678 | to_vmcb->save.sysenter_cs = from_vmcb->save.sysenter_cs; | |
3679 | to_vmcb->save.sysenter_esp = from_vmcb->save.sysenter_esp; | |
3680 | to_vmcb->save.sysenter_eip = from_vmcb->save.sysenter_eip; | |
5542675b AG |
3681 | } |
3682 | ||
851ba692 | 3683 | static int vmload_interception(struct vcpu_svm *svm) |
5542675b | 3684 | { |
9966bf68 | 3685 | struct vmcb *nested_vmcb; |
8c5fbf1a | 3686 | struct kvm_host_map map; |
b742c1e6 | 3687 | int ret; |
9966bf68 | 3688 | |
5542675b AG |
3689 | if (nested_svm_check_permissions(svm)) |
3690 | return 1; | |
3691 | ||
8c5fbf1a KA |
3692 | ret = kvm_vcpu_map(&svm->vcpu, gpa_to_gfn(svm->vmcb->save.rax), &map); |
3693 | if (ret) { | |
3694 | if (ret == -EINVAL) | |
3695 | kvm_inject_gp(&svm->vcpu, 0); | |
9966bf68 | 3696 | return 1; |
8c5fbf1a KA |
3697 | } |
3698 | ||
3699 | nested_vmcb = map.hva; | |
9966bf68 | 3700 | |
e3e9ed3d | 3701 | svm->next_rip = kvm_rip_read(&svm->vcpu) + 3; |
b742c1e6 | 3702 | ret = kvm_skip_emulated_instruction(&svm->vcpu); |
e3e9ed3d | 3703 | |
9966bf68 | 3704 | nested_svm_vmloadsave(nested_vmcb, svm->vmcb); |
8c5fbf1a | 3705 | kvm_vcpu_unmap(&svm->vcpu, &map, true); |
5542675b | 3706 | |
b742c1e6 | 3707 | return ret; |
5542675b AG |
3708 | } |
3709 | ||
851ba692 | 3710 | static int vmsave_interception(struct vcpu_svm *svm) |
5542675b | 3711 | { |
9966bf68 | 3712 | struct vmcb *nested_vmcb; |
8c5fbf1a | 3713 | struct kvm_host_map map; |
b742c1e6 | 3714 | int ret; |
9966bf68 | 3715 | |
5542675b AG |
3716 | if (nested_svm_check_permissions(svm)) |
3717 | return 1; | |
3718 | ||
8c5fbf1a KA |
3719 | ret = kvm_vcpu_map(&svm->vcpu, gpa_to_gfn(svm->vmcb->save.rax), &map); |
3720 | if (ret) { | |
3721 | if (ret == -EINVAL) | |
3722 | kvm_inject_gp(&svm->vcpu, 0); | |
9966bf68 | 3723 | return 1; |
8c5fbf1a KA |
3724 | } |
3725 | ||
3726 | nested_vmcb = map.hva; | |
9966bf68 | 3727 | |
e3e9ed3d | 3728 | svm->next_rip = kvm_rip_read(&svm->vcpu) + 3; |
b742c1e6 | 3729 | ret = kvm_skip_emulated_instruction(&svm->vcpu); |
e3e9ed3d | 3730 | |
9966bf68 | 3731 | nested_svm_vmloadsave(svm->vmcb, nested_vmcb); |
8c5fbf1a | 3732 | kvm_vcpu_unmap(&svm->vcpu, &map, true); |
5542675b | 3733 | |
b742c1e6 | 3734 | return ret; |
5542675b AG |
3735 | } |
3736 | ||
851ba692 | 3737 | static int vmrun_interception(struct vcpu_svm *svm) |
3d6368ef | 3738 | { |
3d6368ef AG |
3739 | if (nested_svm_check_permissions(svm)) |
3740 | return 1; | |
3741 | ||
b75f4eb3 RJ |
3742 | /* Save rip after vmrun instruction */ |
3743 | kvm_rip_write(&svm->vcpu, kvm_rip_read(&svm->vcpu) + 3); | |
3d6368ef | 3744 | |
9738b2c9 | 3745 | if (!nested_svm_vmrun(svm)) |
3d6368ef AG |
3746 | return 1; |
3747 | ||
9738b2c9 | 3748 | if (!nested_svm_vmrun_msrpm(svm)) |
1f8da478 JR |
3749 | goto failed; |
3750 | ||
3751 | return 1; | |
3752 | ||
3753 | failed: | |
3754 | ||
3755 | svm->vmcb->control.exit_code = SVM_EXIT_ERR; | |
3756 | svm->vmcb->control.exit_code_hi = 0; | |
3757 | svm->vmcb->control.exit_info_1 = 0; | |
3758 | svm->vmcb->control.exit_info_2 = 0; | |
3759 | ||
3760 | nested_svm_vmexit(svm); | |
3d6368ef AG |
3761 | |
3762 | return 1; | |
3763 | } | |
3764 | ||
851ba692 | 3765 | static int stgi_interception(struct vcpu_svm *svm) |
1371d904 | 3766 | { |
b742c1e6 LP |
3767 | int ret; |
3768 | ||
1371d904 AG |
3769 | if (nested_svm_check_permissions(svm)) |
3770 | return 1; | |
3771 | ||
640bd6e5 JN |
3772 | /* |
3773 | * If VGIF is enabled, the STGI intercept is only added to | |
cc3d967f | 3774 | * detect the opening of the SMI/NMI window; remove it now. |
640bd6e5 JN |
3775 | */ |
3776 | if (vgif_enabled(svm)) | |
3777 | clr_intercept(svm, INTERCEPT_STGI); | |
3778 | ||
1371d904 | 3779 | svm->next_rip = kvm_rip_read(&svm->vcpu) + 3; |
b742c1e6 | 3780 | ret = kvm_skip_emulated_instruction(&svm->vcpu); |
3842d135 | 3781 | kvm_make_request(KVM_REQ_EVENT, &svm->vcpu); |
1371d904 | 3782 | |
2af9194d | 3783 | enable_gif(svm); |
1371d904 | 3784 | |
b742c1e6 | 3785 | return ret; |
1371d904 AG |
3786 | } |
3787 | ||
851ba692 | 3788 | static int clgi_interception(struct vcpu_svm *svm) |
1371d904 | 3789 | { |
b742c1e6 LP |
3790 | int ret; |
3791 | ||
1371d904 AG |
3792 | if (nested_svm_check_permissions(svm)) |
3793 | return 1; | |
3794 | ||
3795 | svm->next_rip = kvm_rip_read(&svm->vcpu) + 3; | |
b742c1e6 | 3796 | ret = kvm_skip_emulated_instruction(&svm->vcpu); |
1371d904 | 3797 | |
2af9194d | 3798 | disable_gif(svm); |
1371d904 AG |
3799 | |
3800 | /* After a CLGI no interrupts should come */ | |
340d3bc3 SS |
3801 | if (!kvm_vcpu_apicv_active(&svm->vcpu)) { |
3802 | svm_clear_vintr(svm); | |
3803 | svm->vmcb->control.int_ctl &= ~V_IRQ_MASK; | |
3804 | mark_dirty(svm->vmcb, VMCB_INTR); | |
3805 | } | |
decdbf6a | 3806 | |
b742c1e6 | 3807 | return ret; |
1371d904 AG |
3808 | } |
3809 | ||
851ba692 | 3810 | static int invlpga_interception(struct vcpu_svm *svm) |
ff092385 AG |
3811 | { |
3812 | struct kvm_vcpu *vcpu = &svm->vcpu; | |
ff092385 | 3813 | |
de3cd117 SC |
3814 | trace_kvm_invlpga(svm->vmcb->save.rip, kvm_rcx_read(&svm->vcpu), |
3815 | kvm_rax_read(&svm->vcpu)); | |
ec1ff790 | 3816 | |
ff092385 | 3817 | /* Let's treat INVLPGA the same as INVLPG (can be optimized!) */ |
de3cd117 | 3818 | kvm_mmu_invlpg(vcpu, kvm_rax_read(&svm->vcpu)); |
ff092385 AG |
3819 | |
3820 | svm->next_rip = kvm_rip_read(&svm->vcpu) + 3; | |
b742c1e6 | 3821 | return kvm_skip_emulated_instruction(&svm->vcpu); |
ff092385 AG |
3822 | } |
3823 | ||
532a46b9 JR |
3824 | static int skinit_interception(struct vcpu_svm *svm) |
3825 | { | |
de3cd117 | 3826 | trace_kvm_skinit(svm->vmcb->save.rip, kvm_rax_read(&svm->vcpu)); |
532a46b9 JR |
3827 | |
3828 | kvm_queue_exception(&svm->vcpu, UD_VECTOR); | |
3829 | return 1; | |
3830 | } | |
3831 | ||
dab429a7 DK |
3832 | static int wbinvd_interception(struct vcpu_svm *svm) |
3833 | { | |
6affcbed | 3834 | return kvm_emulate_wbinvd(&svm->vcpu); |
dab429a7 DK |
3835 | } |
3836 | ||
81dd35d4 JR |
3837 | static int xsetbv_interception(struct vcpu_svm *svm) |
3838 | { | |
3839 | u64 new_bv = kvm_read_edx_eax(&svm->vcpu); | |
de3cd117 | 3840 | u32 index = kvm_rcx_read(&svm->vcpu); |
81dd35d4 JR |
3841 | |
3842 | if (kvm_set_xcr(&svm->vcpu, index, new_bv) == 0) { | |
3843 | svm->next_rip = kvm_rip_read(&svm->vcpu) + 3; | |
b742c1e6 | 3844 | return kvm_skip_emulated_instruction(&svm->vcpu); |
81dd35d4 JR |
3845 | } |
3846 | ||
3847 | return 1; | |
3848 | } | |
3849 | ||
851ba692 | 3850 | static int task_switch_interception(struct vcpu_svm *svm) |
6aa8b732 | 3851 | { |
37817f29 | 3852 | u16 tss_selector; |
64a7ec06 GN |
3853 | int reason; |
3854 | int int_type = svm->vmcb->control.exit_int_info & | |
3855 | SVM_EXITINTINFO_TYPE_MASK; | |
8317c298 | 3856 | int int_vec = svm->vmcb->control.exit_int_info & SVM_EVTINJ_VEC_MASK; |
fe8e7f83 GN |
3857 | uint32_t type = |
3858 | svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_TYPE_MASK; | |
3859 | uint32_t idt_v = | |
3860 | svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_VALID; | |
e269fb21 JK |
3861 | bool has_error_code = false; |
3862 | u32 error_code = 0; | |
37817f29 IE |
3863 | |
3864 | tss_selector = (u16)svm->vmcb->control.exit_info_1; | |
64a7ec06 | 3865 | |
37817f29 IE |
3866 | if (svm->vmcb->control.exit_info_2 & |
3867 | (1ULL << SVM_EXITINFOSHIFT_TS_REASON_IRET)) | |
64a7ec06 GN |
3868 | reason = TASK_SWITCH_IRET; |
3869 | else if (svm->vmcb->control.exit_info_2 & | |
3870 | (1ULL << SVM_EXITINFOSHIFT_TS_REASON_JMP)) | |
3871 | reason = TASK_SWITCH_JMP; | |
fe8e7f83 | 3872 | else if (idt_v) |
64a7ec06 GN |
3873 | reason = TASK_SWITCH_GATE; |
3874 | else | |
3875 | reason = TASK_SWITCH_CALL; | |
3876 | ||
fe8e7f83 GN |
3877 | if (reason == TASK_SWITCH_GATE) { |
3878 | switch (type) { | |
3879 | case SVM_EXITINTINFO_TYPE_NMI: | |
3880 | svm->vcpu.arch.nmi_injected = false; | |
3881 | break; | |
3882 | case SVM_EXITINTINFO_TYPE_EXEPT: | |
e269fb21 JK |
3883 | if (svm->vmcb->control.exit_info_2 & |
3884 | (1ULL << SVM_EXITINFOSHIFT_TS_HAS_ERROR_CODE)) { | |
3885 | has_error_code = true; | |
3886 | error_code = | |
3887 | (u32)svm->vmcb->control.exit_info_2; | |
3888 | } | |
fe8e7f83 GN |
3889 | kvm_clear_exception_queue(&svm->vcpu); |
3890 | break; | |
3891 | case SVM_EXITINTINFO_TYPE_INTR: | |
3892 | kvm_clear_interrupt_queue(&svm->vcpu); | |
3893 | break; | |
3894 | default: | |
3895 | break; | |
3896 | } | |
3897 | } | |
64a7ec06 | 3898 | |
8317c298 GN |
3899 | if (reason != TASK_SWITCH_GATE || |
3900 | int_type == SVM_EXITINTINFO_TYPE_SOFT || | |
3901 | (int_type == SVM_EXITINTINFO_TYPE_EXEPT && | |
f629cf84 GN |
3902 | (int_vec == OF_VECTOR || int_vec == BP_VECTOR))) |
3903 | skip_emulated_instruction(&svm->vcpu); | |
64a7ec06 | 3904 | |
7f3d35fd KW |
3905 | if (int_type != SVM_EXITINTINFO_TYPE_SOFT) |
3906 | int_vec = -1; | |
3907 | ||
3908 | if (kvm_task_switch(&svm->vcpu, tss_selector, int_vec, reason, | |
acb54517 GN |
3909 | has_error_code, error_code) == EMULATE_FAIL) { |
3910 | svm->vcpu.run->exit_reason = KVM_EXIT_INTERNAL_ERROR; | |
3911 | svm->vcpu.run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION; | |
3912 | svm->vcpu.run->internal.ndata = 0; | |
3913 | return 0; | |
3914 | } | |
3915 | return 1; | |
6aa8b732 AK |
3916 | } |
3917 | ||
851ba692 | 3918 | static int cpuid_interception(struct vcpu_svm *svm) |
6aa8b732 | 3919 | { |
5fdbf976 | 3920 | svm->next_rip = kvm_rip_read(&svm->vcpu) + 2; |
6a908b62 | 3921 | return kvm_emulate_cpuid(&svm->vcpu); |
6aa8b732 AK |
3922 | } |
3923 | ||
851ba692 | 3924 | static int iret_interception(struct vcpu_svm *svm) |
95ba8273 GN |
3925 | { |
3926 | ++svm->vcpu.stat.nmi_window_exits; | |
8a05a1b8 | 3927 | clr_intercept(svm, INTERCEPT_IRET); |
44c11430 | 3928 | svm->vcpu.arch.hflags |= HF_IRET_MASK; |
bd3d1ec3 | 3929 | svm->nmi_iret_rip = kvm_rip_read(&svm->vcpu); |
f303b4ce | 3930 | kvm_make_request(KVM_REQ_EVENT, &svm->vcpu); |
95ba8273 GN |
3931 | return 1; |
3932 | } | |
3933 | ||
851ba692 | 3934 | static int invlpg_interception(struct vcpu_svm *svm) |
a7052897 | 3935 | { |
df4f3108 | 3936 | if (!static_cpu_has(X86_FEATURE_DECODEASSISTS)) |
0ce97a2b | 3937 | return kvm_emulate_instruction(&svm->vcpu, 0) == EMULATE_DONE; |
df4f3108 AP |
3938 | |
3939 | kvm_mmu_invlpg(&svm->vcpu, svm->vmcb->control.exit_info_1); | |
b742c1e6 | 3940 | return kvm_skip_emulated_instruction(&svm->vcpu); |
a7052897 MT |
3941 | } |
3942 | ||
851ba692 | 3943 | static int emulate_on_interception(struct vcpu_svm *svm) |
6aa8b732 | 3944 | { |
0ce97a2b | 3945 | return kvm_emulate_instruction(&svm->vcpu, 0) == EMULATE_DONE; |
6aa8b732 AK |
3946 | } |
3947 | ||
7607b717 BS |
3948 | static int rsm_interception(struct vcpu_svm *svm) |
3949 | { | |
35be0ade SC |
3950 | return kvm_emulate_instruction_from_buffer(&svm->vcpu, |
3951 | rsm_ins_bytes, 2) == EMULATE_DONE; | |
7607b717 BS |
3952 | } |
3953 | ||
332b56e4 AK |
3954 | static int rdpmc_interception(struct vcpu_svm *svm) |
3955 | { | |
3956 | int err; | |
3957 | ||
d647eb63 | 3958 | if (!nrips) |
332b56e4 AK |
3959 | return emulate_on_interception(svm); |
3960 | ||
3961 | err = kvm_rdpmc(&svm->vcpu); | |
6affcbed | 3962 | return kvm_complete_insn_gp(&svm->vcpu, err); |
332b56e4 AK |
3963 | } |
3964 | ||
52eb5a6d XL |
3965 | static bool check_selective_cr0_intercepted(struct vcpu_svm *svm, |
3966 | unsigned long val) | |
628afd2a JR |
3967 | { |
3968 | unsigned long cr0 = svm->vcpu.arch.cr0; | |
3969 | bool ret = false; | |
3970 | u64 intercept; | |
3971 | ||
3972 | intercept = svm->nested.intercept; | |
3973 | ||
3974 | if (!is_guest_mode(&svm->vcpu) || | |
3975 | (!(intercept & (1ULL << INTERCEPT_SELECTIVE_CR0)))) | |
3976 | return false; | |
3977 | ||
3978 | cr0 &= ~SVM_CR0_SELECTIVE_MASK; | |
3979 | val &= ~SVM_CR0_SELECTIVE_MASK; | |
3980 | ||
3981 | if (cr0 ^ val) { | |
3982 | svm->vmcb->control.exit_code = SVM_EXIT_CR0_SEL_WRITE; | |
3983 | ret = (nested_svm_exit_handled(svm) == NESTED_EXIT_DONE); | |
3984 | } | |
3985 | ||
3986 | return ret; | |
3987 | } | |
3988 | ||
7ff76d58 AP |
3989 | #define CR_VALID (1ULL << 63) |
3990 | ||
3991 | static int cr_interception(struct vcpu_svm *svm) | |
3992 | { | |
3993 | int reg, cr; | |
3994 | unsigned long val; | |
3995 | int err; | |
3996 | ||
3997 | if (!static_cpu_has(X86_FEATURE_DECODEASSISTS)) | |
3998 | return emulate_on_interception(svm); | |
3999 | ||
4000 | if (unlikely((svm->vmcb->control.exit_info_1 & CR_VALID) == 0)) | |
4001 | return emulate_on_interception(svm); | |
4002 | ||
4003 | reg = svm->vmcb->control.exit_info_1 & SVM_EXITINFO_REG_MASK; | |
5e57518d DK |
4004 | if (svm->vmcb->control.exit_code == SVM_EXIT_CR0_SEL_WRITE) |
4005 | cr = SVM_EXIT_WRITE_CR0 - SVM_EXIT_READ_CR0; | |
4006 | else | |
4007 | cr = svm->vmcb->control.exit_code - SVM_EXIT_READ_CR0; | |
7ff76d58 AP |
4008 | |
4009 | err = 0; | |
4010 | if (cr >= 16) { /* mov to cr */ | |
4011 | cr -= 16; | |
4012 | val = kvm_register_read(&svm->vcpu, reg); | |
4013 | switch (cr) { | |
4014 | case 0: | |
628afd2a JR |
4015 | if (!check_selective_cr0_intercepted(svm, val)) |
4016 | err = kvm_set_cr0(&svm->vcpu, val); | |
977b2d03 JR |
4017 | else |
4018 | return 1; | |
4019 | ||
7ff76d58 AP |
4020 | break; |
4021 | case 3: | |
4022 | err = kvm_set_cr3(&svm->vcpu, val); | |
4023 | break; | |
4024 | case 4: | |
4025 | err = kvm_set_cr4(&svm->vcpu, val); | |
4026 | break; | |
4027 | case 8: | |
4028 | err = kvm_set_cr8(&svm->vcpu, val); | |
4029 | break; | |
4030 | default: | |
4031 | WARN(1, "unhandled write to CR%d", cr); | |
4032 | kvm_queue_exception(&svm->vcpu, UD_VECTOR); | |
4033 | return 1; | |
4034 | } | |
4035 | } else { /* mov from cr */ | |
4036 | switch (cr) { | |
4037 | case 0: | |
4038 | val = kvm_read_cr0(&svm->vcpu); | |
4039 | break; | |
4040 | case 2: | |
4041 | val = svm->vcpu.arch.cr2; | |
4042 | break; | |
4043 | case 3: | |
9f8fe504 | 4044 | val = kvm_read_cr3(&svm->vcpu); |
7ff76d58 AP |
4045 | break; |
4046 | case 4: | |
4047 | val = kvm_read_cr4(&svm->vcpu); | |
4048 | break; | |
4049 | case 8: | |
4050 | val = kvm_get_cr8(&svm->vcpu); | |
4051 | break; | |
4052 | default: | |
4053 | WARN(1, "unhandled read from CR%d", cr); | |
4054 | kvm_queue_exception(&svm->vcpu, UD_VECTOR); | |
4055 | return 1; | |
4056 | } | |
4057 | kvm_register_write(&svm->vcpu, reg, val); | |
4058 | } | |
6affcbed | 4059 | return kvm_complete_insn_gp(&svm->vcpu, err); |
7ff76d58 AP |
4060 | } |
4061 | ||
cae3797a AP |
4062 | static int dr_interception(struct vcpu_svm *svm) |
4063 | { | |
4064 | int reg, dr; | |
4065 | unsigned long val; | |
cae3797a | 4066 | |
facb0139 PB |
4067 | if (svm->vcpu.guest_debug == 0) { |
4068 | /* | |
4069 | * No more DR vmexits; force a reload of the debug registers | |
4070 | * and reenter on this instruction. The next vmexit will | |
4071 | * retrieve the full state of the debug registers. | |
4072 | */ | |
4073 | clr_dr_intercepts(svm); | |
4074 | svm->vcpu.arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT; | |
4075 | return 1; | |
4076 | } | |
4077 | ||
cae3797a AP |
4078 | if (!boot_cpu_has(X86_FEATURE_DECODEASSISTS)) |
4079 | return emulate_on_interception(svm); | |
4080 | ||
4081 | reg = svm->vmcb->control.exit_info_1 & SVM_EXITINFO_REG_MASK; | |
4082 | dr = svm->vmcb->control.exit_code - SVM_EXIT_READ_DR0; | |
4083 | ||
4084 | if (dr >= 16) { /* mov to DRn */ | |
16f8a6f9 NA |
4085 | if (!kvm_require_dr(&svm->vcpu, dr - 16)) |
4086 | return 1; | |
cae3797a AP |
4087 | val = kvm_register_read(&svm->vcpu, reg); |
4088 | kvm_set_dr(&svm->vcpu, dr - 16, val); | |
4089 | } else { | |
16f8a6f9 NA |
4090 | if (!kvm_require_dr(&svm->vcpu, dr)) |
4091 | return 1; | |
4092 | kvm_get_dr(&svm->vcpu, dr, &val); | |
4093 | kvm_register_write(&svm->vcpu, reg, val); | |
cae3797a AP |
4094 | } |
4095 | ||
b742c1e6 | 4096 | return kvm_skip_emulated_instruction(&svm->vcpu); |
cae3797a AP |
4097 | } |
4098 | ||
851ba692 | 4099 | static int cr8_write_interception(struct vcpu_svm *svm) |
1d075434 | 4100 | { |
851ba692 | 4101 | struct kvm_run *kvm_run = svm->vcpu.run; |
eea1cff9 | 4102 | int r; |
851ba692 | 4103 | |
0a5fff19 GN |
4104 | u8 cr8_prev = kvm_get_cr8(&svm->vcpu); |
4105 | /* instruction emulation calls kvm_set_cr8() */ | |
7ff76d58 | 4106 | r = cr_interception(svm); |
35754c98 | 4107 | if (lapic_in_kernel(&svm->vcpu)) |
7ff76d58 | 4108 | return r; |
0a5fff19 | 4109 | if (cr8_prev <= kvm_get_cr8(&svm->vcpu)) |
7ff76d58 | 4110 | return r; |
1d075434 JR |
4111 | kvm_run->exit_reason = KVM_EXIT_SET_TPR; |
4112 | return 0; | |
4113 | } | |
4114 | ||
801e459a TL |
4115 | static int svm_get_msr_feature(struct kvm_msr_entry *msr) |
4116 | { | |
d1d93fa9 TL |
4117 | msr->data = 0; |
4118 | ||
4119 | switch (msr->index) { | |
4120 | case MSR_F10H_DECFG: | |
4121 | if (boot_cpu_has(X86_FEATURE_LFENCE_RDTSC)) | |
4122 | msr->data |= MSR_F10H_DECFG_LFENCE_SERIALIZE; | |
4123 | break; | |
4124 | default: | |
4125 | return 1; | |
4126 | } | |
4127 | ||
4128 | return 0; | |
801e459a TL |
4129 | } |
4130 | ||
609e36d3 | 4131 | static int svm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info) |
6aa8b732 | 4132 | { |
a2fa3e9f GH |
4133 | struct vcpu_svm *svm = to_svm(vcpu); |
4134 | ||
609e36d3 | 4135 | switch (msr_info->index) { |
8c06585d | 4136 | case MSR_STAR: |
609e36d3 | 4137 | msr_info->data = svm->vmcb->save.star; |
6aa8b732 | 4138 | break; |
0e859cac | 4139 | #ifdef CONFIG_X86_64 |
6aa8b732 | 4140 | case MSR_LSTAR: |
609e36d3 | 4141 | msr_info->data = svm->vmcb->save.lstar; |
6aa8b732 AK |
4142 | break; |
4143 | case MSR_CSTAR: | |
609e36d3 | 4144 | msr_info->data = svm->vmcb->save.cstar; |
6aa8b732 AK |
4145 | break; |
4146 | case MSR_KERNEL_GS_BASE: | |
609e36d3 | 4147 | msr_info->data = svm->vmcb->save.kernel_gs_base; |
6aa8b732 AK |
4148 | break; |
4149 | case MSR_SYSCALL_MASK: | |
609e36d3 | 4150 | msr_info->data = svm->vmcb->save.sfmask; |
6aa8b732 AK |
4151 | break; |
4152 | #endif | |
4153 | case MSR_IA32_SYSENTER_CS: | |
609e36d3 | 4154 | msr_info->data = svm->vmcb->save.sysenter_cs; |
6aa8b732 AK |
4155 | break; |
4156 | case MSR_IA32_SYSENTER_EIP: | |
609e36d3 | 4157 | msr_info->data = svm->sysenter_eip; |
6aa8b732 AK |
4158 | break; |
4159 | case MSR_IA32_SYSENTER_ESP: | |
609e36d3 | 4160 | msr_info->data = svm->sysenter_esp; |
6aa8b732 | 4161 | break; |
46896c73 PB |
4162 | case MSR_TSC_AUX: |
4163 | if (!boot_cpu_has(X86_FEATURE_RDTSCP)) | |
4164 | return 1; | |
4165 | msr_info->data = svm->tsc_aux; | |
4166 | break; | |
e0231715 JR |
4167 | /* |
4168 | * Nobody will change the following 5 values in the VMCB so we can | |
4169 | * safely return them on rdmsr. They will always be 0 until LBRV is | |
4170 | * implemented. | |
4171 | */ | |
a2938c80 | 4172 | case MSR_IA32_DEBUGCTLMSR: |
609e36d3 | 4173 | msr_info->data = svm->vmcb->save.dbgctl; |
a2938c80 JR |
4174 | break; |
4175 | case MSR_IA32_LASTBRANCHFROMIP: | |
609e36d3 | 4176 | msr_info->data = svm->vmcb->save.br_from; |
a2938c80 JR |
4177 | break; |
4178 | case MSR_IA32_LASTBRANCHTOIP: | |
609e36d3 | 4179 | msr_info->data = svm->vmcb->save.br_to; |
a2938c80 JR |
4180 | break; |
4181 | case MSR_IA32_LASTINTFROMIP: | |
609e36d3 | 4182 | msr_info->data = svm->vmcb->save.last_excp_from; |
a2938c80 JR |
4183 | break; |
4184 | case MSR_IA32_LASTINTTOIP: | |
609e36d3 | 4185 | msr_info->data = svm->vmcb->save.last_excp_to; |
a2938c80 | 4186 | break; |
b286d5d8 | 4187 | case MSR_VM_HSAVE_PA: |
609e36d3 | 4188 | msr_info->data = svm->nested.hsave_msr; |
b286d5d8 | 4189 | break; |
eb6f302e | 4190 | case MSR_VM_CR: |
609e36d3 | 4191 | msr_info->data = svm->nested.vm_cr_msr; |
eb6f302e | 4192 | break; |
b2ac58f9 KA |
4193 | case MSR_IA32_SPEC_CTRL: |
4194 | if (!msr_info->host_initiated && | |
6ac2f49e KRW |
4195 | !guest_cpuid_has(vcpu, X86_FEATURE_AMD_IBRS) && |
4196 | !guest_cpuid_has(vcpu, X86_FEATURE_AMD_SSBD)) | |
b2ac58f9 KA |
4197 | return 1; |
4198 | ||
4199 | msr_info->data = svm->spec_ctrl; | |
4200 | break; | |
bc226f07 TL |
4201 | case MSR_AMD64_VIRT_SPEC_CTRL: |
4202 | if (!msr_info->host_initiated && | |
4203 | !guest_cpuid_has(vcpu, X86_FEATURE_VIRT_SSBD)) | |
4204 | return 1; | |
4205 | ||
4206 | msr_info->data = svm->virt_spec_ctrl; | |
4207 | break; | |
ae8b7875 BP |
4208 | case MSR_F15H_IC_CFG: { |
4209 | ||
4210 | int family, model; | |
4211 | ||
4212 | family = guest_cpuid_family(vcpu); | |
4213 | model = guest_cpuid_model(vcpu); | |
4214 | ||
4215 | if (family < 0 || model < 0) | |
4216 | return kvm_get_msr_common(vcpu, msr_info); | |
4217 | ||
4218 | msr_info->data = 0; | |
4219 | ||
4220 | if (family == 0x15 && | |
4221 | (model >= 0x2 && model < 0x20)) | |
4222 | msr_info->data = 0x1E; | |
4223 | } | |
4224 | break; | |
d1d93fa9 TL |
4225 | case MSR_F10H_DECFG: |
4226 | msr_info->data = svm->msr_decfg; | |
4227 | break; | |
6aa8b732 | 4228 | default: |
609e36d3 | 4229 | return kvm_get_msr_common(vcpu, msr_info); |
6aa8b732 AK |
4230 | } |
4231 | return 0; | |
4232 | } | |
4233 | ||
851ba692 | 4234 | static int rdmsr_interception(struct vcpu_svm *svm) |
6aa8b732 | 4235 | { |
de3cd117 | 4236 | u32 ecx = kvm_rcx_read(&svm->vcpu); |
609e36d3 | 4237 | struct msr_data msr_info; |
6aa8b732 | 4238 | |
609e36d3 PB |
4239 | msr_info.index = ecx; |
4240 | msr_info.host_initiated = false; | |
4241 | if (svm_get_msr(&svm->vcpu, &msr_info)) { | |
59200273 | 4242 | trace_kvm_msr_read_ex(ecx); |
c1a5d4f9 | 4243 | kvm_inject_gp(&svm->vcpu, 0); |
b742c1e6 | 4244 | return 1; |
59200273 | 4245 | } else { |
609e36d3 | 4246 | trace_kvm_msr_read(ecx, msr_info.data); |
af9ca2d7 | 4247 | |
de3cd117 SC |
4248 | kvm_rax_write(&svm->vcpu, msr_info.data & 0xffffffff); |
4249 | kvm_rdx_write(&svm->vcpu, msr_info.data >> 32); | |
5fdbf976 | 4250 | svm->next_rip = kvm_rip_read(&svm->vcpu) + 2; |
b742c1e6 | 4251 | return kvm_skip_emulated_instruction(&svm->vcpu); |
6aa8b732 | 4252 | } |
6aa8b732 AK |
4253 | } |
4254 | ||
4a810181 JR |
4255 | static int svm_set_vm_cr(struct kvm_vcpu *vcpu, u64 data) |
4256 | { | |
4257 | struct vcpu_svm *svm = to_svm(vcpu); | |
4258 | int svm_dis, chg_mask; | |
4259 | ||
4260 | if (data & ~SVM_VM_CR_VALID_MASK) | |
4261 | return 1; | |
4262 | ||
4263 | chg_mask = SVM_VM_CR_VALID_MASK; | |
4264 | ||
4265 | if (svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK) | |
4266 | chg_mask &= ~(SVM_VM_CR_SVM_LOCK_MASK | SVM_VM_CR_SVM_DIS_MASK); | |
4267 | ||
4268 | svm->nested.vm_cr_msr &= ~chg_mask; | |
4269 | svm->nested.vm_cr_msr |= (data & chg_mask); | |
4270 | ||
4271 | svm_dis = svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK; | |
4272 | ||
4273 | /* check for svm_disable while efer.svme is set */ | |
4274 | if (svm_dis && (vcpu->arch.efer & EFER_SVME)) | |
4275 | return 1; | |
4276 | ||
4277 | return 0; | |
4278 | } | |
4279 | ||
8fe8ab46 | 4280 | static int svm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr) |
6aa8b732 | 4281 | { |
a2fa3e9f GH |
4282 | struct vcpu_svm *svm = to_svm(vcpu); |
4283 | ||
8fe8ab46 WA |
4284 | u32 ecx = msr->index; |
4285 | u64 data = msr->data; | |
6aa8b732 | 4286 | switch (ecx) { |
15038e14 PB |
4287 | case MSR_IA32_CR_PAT: |
4288 | if (!kvm_mtrr_valid(vcpu, MSR_IA32_CR_PAT, data)) | |
4289 | return 1; | |
4290 | vcpu->arch.pat = data; | |
4291 | svm->vmcb->save.g_pat = data; | |
4292 | mark_dirty(svm->vmcb, VMCB_NPT); | |
4293 | break; | |
b2ac58f9 KA |
4294 | case MSR_IA32_SPEC_CTRL: |
4295 | if (!msr->host_initiated && | |
6ac2f49e KRW |
4296 | !guest_cpuid_has(vcpu, X86_FEATURE_AMD_IBRS) && |
4297 | !guest_cpuid_has(vcpu, X86_FEATURE_AMD_SSBD)) | |
b2ac58f9 KA |
4298 | return 1; |
4299 | ||
4300 | /* The STIBP bit doesn't fault even if it's not advertised */ | |
6ac2f49e | 4301 | if (data & ~(SPEC_CTRL_IBRS | SPEC_CTRL_STIBP | SPEC_CTRL_SSBD)) |
b2ac58f9 KA |
4302 | return 1; |
4303 | ||
4304 | svm->spec_ctrl = data; | |
4305 | ||
4306 | if (!data) | |
4307 | break; | |
4308 | ||
4309 | /* | |
4310 | * For non-nested: | |
4311 | * When it's written (to non-zero) for the first time, pass | |
4312 | * it through. | |
4313 | * | |
4314 | * For nested: | |
4315 | * The handling of the MSR bitmap for L2 guests is done in | |
4316 | * nested_svm_vmrun_msrpm. | |
4317 | * We update the L1 MSR bit as well since it will end up | |
4318 | * touching the MSR anyway now. | |
4319 | */ | |
4320 | set_msr_interception(svm->msrpm, MSR_IA32_SPEC_CTRL, 1, 1); | |
4321 | break; | |
15d45071 AR |
4322 | case MSR_IA32_PRED_CMD: |
4323 | if (!msr->host_initiated && | |
e7c587da | 4324 | !guest_cpuid_has(vcpu, X86_FEATURE_AMD_IBPB)) |
15d45071 AR |
4325 | return 1; |
4326 | ||
4327 | if (data & ~PRED_CMD_IBPB) | |
4328 | return 1; | |
4329 | ||
4330 | if (!data) | |
4331 | break; | |
4332 | ||
4333 | wrmsrl(MSR_IA32_PRED_CMD, PRED_CMD_IBPB); | |
4334 | if (is_guest_mode(vcpu)) | |
4335 | break; | |
4336 | set_msr_interception(svm->msrpm, MSR_IA32_PRED_CMD, 0, 1); | |
4337 | break; | |
bc226f07 TL |
4338 | case MSR_AMD64_VIRT_SPEC_CTRL: |
4339 | if (!msr->host_initiated && | |
4340 | !guest_cpuid_has(vcpu, X86_FEATURE_VIRT_SSBD)) | |
4341 | return 1; | |
4342 | ||
4343 | if (data & ~SPEC_CTRL_SSBD) | |
4344 | return 1; | |
4345 | ||
4346 | svm->virt_spec_ctrl = data; | |
4347 | break; | |
8c06585d | 4348 | case MSR_STAR: |
a2fa3e9f | 4349 | svm->vmcb->save.star = data; |
6aa8b732 | 4350 | break; |
49b14f24 | 4351 | #ifdef CONFIG_X86_64 |
6aa8b732 | 4352 | case MSR_LSTAR: |
a2fa3e9f | 4353 | svm->vmcb->save.lstar = data; |
6aa8b732 AK |
4354 | break; |
4355 | case MSR_CSTAR: | |
a2fa3e9f | 4356 | svm->vmcb->save.cstar = data; |
6aa8b732 AK |
4357 | break; |
4358 | case MSR_KERNEL_GS_BASE: | |
a2fa3e9f | 4359 | svm->vmcb->save.kernel_gs_base = data; |
6aa8b732 AK |
4360 | break; |
4361 | case MSR_SYSCALL_MASK: | |
a2fa3e9f | 4362 | svm->vmcb->save.sfmask = data; |
6aa8b732 AK |
4363 | break; |
4364 | #endif | |
4365 | case MSR_IA32_SYSENTER_CS: | |
a2fa3e9f | 4366 | svm->vmcb->save.sysenter_cs = data; |
6aa8b732 AK |
4367 | break; |
4368 | case MSR_IA32_SYSENTER_EIP: | |
017cb99e | 4369 | svm->sysenter_eip = data; |
a2fa3e9f | 4370 | svm->vmcb->save.sysenter_eip = data; |
6aa8b732 AK |
4371 | break; |
4372 | case MSR_IA32_SYSENTER_ESP: | |
017cb99e | 4373 | svm->sysenter_esp = data; |
a2fa3e9f | 4374 | svm->vmcb->save.sysenter_esp = data; |
6aa8b732 | 4375 | break; |
46896c73 PB |
4376 | case MSR_TSC_AUX: |
4377 | if (!boot_cpu_has(X86_FEATURE_RDTSCP)) | |
4378 | return 1; | |
4379 | ||
4380 | /* | |
4381 | * This is rare, so we update the MSR here instead of using | |
4382 | * direct_access_msrs. Doing that would require a rdmsr in | |
4383 | * svm_vcpu_put. | |
4384 | */ | |
4385 | svm->tsc_aux = data; | |
4386 | wrmsrl(MSR_TSC_AUX, svm->tsc_aux); | |
4387 | break; | |
a2938c80 | 4388 | case MSR_IA32_DEBUGCTLMSR: |
2a6b20b8 | 4389 | if (!boot_cpu_has(X86_FEATURE_LBRV)) { |
a737f256 CD |
4390 | vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTL 0x%llx, nop\n", |
4391 | __func__, data); | |
24e09cbf JR |
4392 | break; |
4393 | } | |
4394 | if (data & DEBUGCTL_RESERVED_BITS) | |
4395 | return 1; | |
4396 | ||
4397 | svm->vmcb->save.dbgctl = data; | |
b53ba3f9 | 4398 | mark_dirty(svm->vmcb, VMCB_LBR); |
24e09cbf JR |
4399 | if (data & (1ULL<<0)) |
4400 | svm_enable_lbrv(svm); | |
4401 | else | |
4402 | svm_disable_lbrv(svm); | |
a2938c80 | 4403 | break; |
b286d5d8 | 4404 | case MSR_VM_HSAVE_PA: |
e6aa9abd | 4405 | svm->nested.hsave_msr = data; |
62b9abaa | 4406 | break; |
3c5d0a44 | 4407 | case MSR_VM_CR: |
4a810181 | 4408 | return svm_set_vm_cr(vcpu, data); |
3c5d0a44 | 4409 | case MSR_VM_IGNNE: |
a737f256 | 4410 | vcpu_unimpl(vcpu, "unimplemented wrmsr: 0x%x data 0x%llx\n", ecx, data); |
3c5d0a44 | 4411 | break; |
d1d93fa9 TL |
4412 | case MSR_F10H_DECFG: { |
4413 | struct kvm_msr_entry msr_entry; | |
4414 | ||
4415 | msr_entry.index = msr->index; | |
4416 | if (svm_get_msr_feature(&msr_entry)) | |
4417 | return 1; | |
4418 | ||
4419 | /* Check the supported bits */ | |
4420 | if (data & ~msr_entry.data) | |
4421 | return 1; | |
4422 | ||
4423 | /* Don't allow the guest to change a bit, #GP */ | |
4424 | if (!msr->host_initiated && (data ^ msr_entry.data)) | |
4425 | return 1; | |
4426 | ||
4427 | svm->msr_decfg = data; | |
4428 | break; | |
4429 | } | |
44a95dae SS |
4430 | case MSR_IA32_APICBASE: |
4431 | if (kvm_vcpu_apicv_active(vcpu)) | |
4432 | avic_update_vapic_bar(to_svm(vcpu), data); | |
b2869f28 | 4433 | /* Fall through */ |
6aa8b732 | 4434 | default: |
8fe8ab46 | 4435 | return kvm_set_msr_common(vcpu, msr); |
6aa8b732 AK |
4436 | } |
4437 | return 0; | |
4438 | } | |
4439 | ||
851ba692 | 4440 | static int wrmsr_interception(struct vcpu_svm *svm) |
6aa8b732 | 4441 | { |
8fe8ab46 | 4442 | struct msr_data msr; |
de3cd117 | 4443 | u32 ecx = kvm_rcx_read(&svm->vcpu); |
668f198f | 4444 | u64 data = kvm_read_edx_eax(&svm->vcpu); |
af9ca2d7 | 4445 | |
8fe8ab46 WA |
4446 | msr.data = data; |
4447 | msr.index = ecx; | |
4448 | msr.host_initiated = false; | |
af9ca2d7 | 4449 | |
5fdbf976 | 4450 | svm->next_rip = kvm_rip_read(&svm->vcpu) + 2; |
854e8bb1 | 4451 | if (kvm_set_msr(&svm->vcpu, &msr)) { |
59200273 | 4452 | trace_kvm_msr_write_ex(ecx, data); |
c1a5d4f9 | 4453 | kvm_inject_gp(&svm->vcpu, 0); |
b742c1e6 | 4454 | return 1; |
59200273 AK |
4455 | } else { |
4456 | trace_kvm_msr_write(ecx, data); | |
b742c1e6 | 4457 | return kvm_skip_emulated_instruction(&svm->vcpu); |
59200273 | 4458 | } |
6aa8b732 AK |
4459 | } |
4460 | ||
851ba692 | 4461 | static int msr_interception(struct vcpu_svm *svm) |
6aa8b732 | 4462 | { |
e756fc62 | 4463 | if (svm->vmcb->control.exit_info_1) |
851ba692 | 4464 | return wrmsr_interception(svm); |
6aa8b732 | 4465 | else |
851ba692 | 4466 | return rdmsr_interception(svm); |
6aa8b732 AK |
4467 | } |
4468 | ||
851ba692 | 4469 | static int interrupt_window_interception(struct vcpu_svm *svm) |
c1150d8c | 4470 | { |
3842d135 | 4471 | kvm_make_request(KVM_REQ_EVENT, &svm->vcpu); |
f0b85051 | 4472 | svm_clear_vintr(svm); |
85f455f7 | 4473 | svm->vmcb->control.int_ctl &= ~V_IRQ_MASK; |
decdbf6a | 4474 | mark_dirty(svm->vmcb, VMCB_INTR); |
675acb75 | 4475 | ++svm->vcpu.stat.irq_window_exits; |
c1150d8c DL |
4476 | return 1; |
4477 | } | |
4478 | ||
565d0998 ML |
4479 | static int pause_interception(struct vcpu_svm *svm) |
4480 | { | |
de63ad4c LM |
4481 | struct kvm_vcpu *vcpu = &svm->vcpu; |
4482 | bool in_kernel = (svm_get_cpl(vcpu) == 0); | |
4483 | ||
8566ac8b BM |
4484 | if (pause_filter_thresh) |
4485 | grow_ple_window(vcpu); | |
4486 | ||
de63ad4c | 4487 | kvm_vcpu_on_spin(vcpu, in_kernel); |
565d0998 ML |
4488 | return 1; |
4489 | } | |
4490 | ||
87c00572 GS |
4491 | static int nop_interception(struct vcpu_svm *svm) |
4492 | { | |
b742c1e6 | 4493 | return kvm_skip_emulated_instruction(&(svm->vcpu)); |
87c00572 GS |
4494 | } |
4495 | ||
4496 | static int monitor_interception(struct vcpu_svm *svm) | |
4497 | { | |
4498 | printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n"); | |
4499 | return nop_interception(svm); | |
4500 | } | |
4501 | ||
4502 | static int mwait_interception(struct vcpu_svm *svm) | |
4503 | { | |
4504 | printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n"); | |
4505 | return nop_interception(svm); | |
4506 | } | |
4507 | ||
18f40c53 SS |
4508 | enum avic_ipi_failure_cause { |
4509 | AVIC_IPI_FAILURE_INVALID_INT_TYPE, | |
4510 | AVIC_IPI_FAILURE_TARGET_NOT_RUNNING, | |
4511 | AVIC_IPI_FAILURE_INVALID_TARGET, | |
4512 | AVIC_IPI_FAILURE_INVALID_BACKING_PAGE, | |
4513 | }; | |
4514 | ||
4515 | static int avic_incomplete_ipi_interception(struct vcpu_svm *svm) | |
4516 | { | |
4517 | u32 icrh = svm->vmcb->control.exit_info_1 >> 32; | |
4518 | u32 icrl = svm->vmcb->control.exit_info_1; | |
4519 | u32 id = svm->vmcb->control.exit_info_2 >> 32; | |
5446a979 | 4520 | u32 index = svm->vmcb->control.exit_info_2 & 0xFF; |
18f40c53 SS |
4521 | struct kvm_lapic *apic = svm->vcpu.arch.apic; |
4522 | ||
4523 | trace_kvm_avic_incomplete_ipi(svm->vcpu.vcpu_id, icrh, icrl, id, index); | |
4524 | ||
4525 | switch (id) { | |
4526 | case AVIC_IPI_FAILURE_INVALID_INT_TYPE: | |
4527 | /* | |
4528 | * AVIC hardware handles the generation of | |
4529 | * IPIs when the specified Message Type is Fixed | |
4530 | * (also known as fixed delivery mode) and | |
4531 | * the Trigger Mode is edge-triggered. The hardware | |
4532 | * also supports self and broadcast delivery modes | |
4533 | * specified via the Destination Shorthand(DSH) | |
4534 | * field of the ICRL. Logical and physical APIC ID | |
4535 | * formats are supported. All other IPI types cause | |
4536 | * a #VMEXIT, which needs to emulated. | |
4537 | */ | |
4538 | kvm_lapic_reg_write(apic, APIC_ICR2, icrh); | |
4539 | kvm_lapic_reg_write(apic, APIC_ICR, icrl); | |
4540 | break; | |
4541 | case AVIC_IPI_FAILURE_TARGET_NOT_RUNNING: { | |
4a58038b SS |
4542 | int i; |
4543 | struct kvm_vcpu *vcpu; | |
4544 | struct kvm *kvm = svm->vcpu.kvm; | |
18f40c53 SS |
4545 | struct kvm_lapic *apic = svm->vcpu.arch.apic; |
4546 | ||
4547 | /* | |
4a58038b SS |
4548 | * At this point, we expect that the AVIC HW has already |
4549 | * set the appropriate IRR bits on the valid target | |
4550 | * vcpus. So, we just need to kick the appropriate vcpu. | |
18f40c53 | 4551 | */ |
4a58038b SS |
4552 | kvm_for_each_vcpu(i, vcpu, kvm) { |
4553 | bool m = kvm_apic_match_dest(vcpu, apic, | |
4554 | icrl & KVM_APIC_SHORT_MASK, | |
4555 | GET_APIC_DEST_FIELD(icrh), | |
4556 | icrl & KVM_APIC_DEST_MASK); | |
4557 | ||
4558 | if (m && !avic_vcpu_is_running(vcpu)) | |
4559 | kvm_vcpu_wake_up(vcpu); | |
4560 | } | |
18f40c53 SS |
4561 | break; |
4562 | } | |
4563 | case AVIC_IPI_FAILURE_INVALID_TARGET: | |
37ef0c44 SS |
4564 | WARN_ONCE(1, "Invalid IPI target: index=%u, vcpu=%d, icr=%#0x:%#0x\n", |
4565 | index, svm->vcpu.vcpu_id, icrh, icrl); | |
18f40c53 SS |
4566 | break; |
4567 | case AVIC_IPI_FAILURE_INVALID_BACKING_PAGE: | |
4568 | WARN_ONCE(1, "Invalid backing page\n"); | |
4569 | break; | |
4570 | default: | |
4571 | pr_err("Unknown IPI interception\n"); | |
4572 | } | |
4573 | ||
4574 | return 1; | |
4575 | } | |
4576 | ||
4577 | static u32 *avic_get_logical_id_entry(struct kvm_vcpu *vcpu, u32 ldr, bool flat) | |
4578 | { | |
81811c16 | 4579 | struct kvm_svm *kvm_svm = to_kvm_svm(vcpu->kvm); |
18f40c53 SS |
4580 | int index; |
4581 | u32 *logical_apic_id_table; | |
4582 | int dlid = GET_APIC_LOGICAL_ID(ldr); | |
4583 | ||
4584 | if (!dlid) | |
4585 | return NULL; | |
4586 | ||
4587 | if (flat) { /* flat */ | |
4588 | index = ffs(dlid) - 1; | |
4589 | if (index > 7) | |
4590 | return NULL; | |
4591 | } else { /* cluster */ | |
4592 | int cluster = (dlid & 0xf0) >> 4; | |
4593 | int apic = ffs(dlid & 0x0f) - 1; | |
4594 | ||
4595 | if ((apic < 0) || (apic > 7) || | |
4596 | (cluster >= 0xf)) | |
4597 | return NULL; | |
4598 | index = (cluster << 2) + apic; | |
4599 | } | |
4600 | ||
81811c16 | 4601 | logical_apic_id_table = (u32 *) page_address(kvm_svm->avic_logical_id_table_page); |
18f40c53 SS |
4602 | |
4603 | return &logical_apic_id_table[index]; | |
4604 | } | |
4605 | ||
98d90582 | 4606 | static int avic_ldr_write(struct kvm_vcpu *vcpu, u8 g_physical_id, u32 ldr) |
18f40c53 SS |
4607 | { |
4608 | bool flat; | |
4609 | u32 *entry, new_entry; | |
4610 | ||
4611 | flat = kvm_lapic_get_reg(vcpu->arch.apic, APIC_DFR) == APIC_DFR_FLAT; | |
4612 | entry = avic_get_logical_id_entry(vcpu, ldr, flat); | |
4613 | if (!entry) | |
4614 | return -EINVAL; | |
4615 | ||
4616 | new_entry = READ_ONCE(*entry); | |
4617 | new_entry &= ~AVIC_LOGICAL_ID_ENTRY_GUEST_PHYSICAL_ID_MASK; | |
4618 | new_entry |= (g_physical_id & AVIC_LOGICAL_ID_ENTRY_GUEST_PHYSICAL_ID_MASK); | |
98d90582 | 4619 | new_entry |= AVIC_LOGICAL_ID_ENTRY_VALID_MASK; |
18f40c53 SS |
4620 | WRITE_ONCE(*entry, new_entry); |
4621 | ||
4622 | return 0; | |
4623 | } | |
4624 | ||
98d90582 SS |
4625 | static void avic_invalidate_logical_id_entry(struct kvm_vcpu *vcpu) |
4626 | { | |
4627 | struct vcpu_svm *svm = to_svm(vcpu); | |
4628 | bool flat = svm->dfr_reg == APIC_DFR_FLAT; | |
4629 | u32 *entry = avic_get_logical_id_entry(vcpu, svm->ldr_reg, flat); | |
4630 | ||
4631 | if (entry) | |
e44e3eac | 4632 | clear_bit(AVIC_LOGICAL_ID_ENTRY_VALID_BIT, (unsigned long *)entry); |
98d90582 SS |
4633 | } |
4634 | ||
18f40c53 SS |
4635 | static int avic_handle_ldr_update(struct kvm_vcpu *vcpu) |
4636 | { | |
98d90582 | 4637 | int ret = 0; |
18f40c53 SS |
4638 | struct vcpu_svm *svm = to_svm(vcpu); |
4639 | u32 ldr = kvm_lapic_get_reg(vcpu->arch.apic, APIC_LDR); | |
4640 | ||
98d90582 SS |
4641 | if (ldr == svm->ldr_reg) |
4642 | return 0; | |
18f40c53 | 4643 | |
98d90582 SS |
4644 | avic_invalidate_logical_id_entry(vcpu); |
4645 | ||
4646 | if (ldr) | |
4647 | ret = avic_ldr_write(vcpu, vcpu->vcpu_id, ldr); | |
4648 | ||
4649 | if (!ret) | |
18f40c53 | 4650 | svm->ldr_reg = ldr; |
98d90582 | 4651 | |
18f40c53 SS |
4652 | return ret; |
4653 | } | |
4654 | ||
4655 | static int avic_handle_apic_id_update(struct kvm_vcpu *vcpu) | |
4656 | { | |
4657 | u64 *old, *new; | |
4658 | struct vcpu_svm *svm = to_svm(vcpu); | |
4659 | u32 apic_id_reg = kvm_lapic_get_reg(vcpu->arch.apic, APIC_ID); | |
4660 | u32 id = (apic_id_reg >> 24) & 0xff; | |
4661 | ||
4662 | if (vcpu->vcpu_id == id) | |
4663 | return 0; | |
4664 | ||
4665 | old = avic_get_physical_id_entry(vcpu, vcpu->vcpu_id); | |
4666 | new = avic_get_physical_id_entry(vcpu, id); | |
4667 | if (!new || !old) | |
4668 | return 1; | |
4669 | ||
4670 | /* We need to move physical_id_entry to new offset */ | |
4671 | *new = *old; | |
4672 | *old = 0ULL; | |
4673 | to_svm(vcpu)->avic_physical_id_cache = new; | |
4674 | ||
4675 | /* | |
4676 | * Also update the guest physical APIC ID in the logical | |
4677 | * APIC ID table entry if already setup the LDR. | |
4678 | */ | |
4679 | if (svm->ldr_reg) | |
4680 | avic_handle_ldr_update(vcpu); | |
4681 | ||
4682 | return 0; | |
4683 | } | |
4684 | ||
98d90582 | 4685 | static void avic_handle_dfr_update(struct kvm_vcpu *vcpu) |
18f40c53 SS |
4686 | { |
4687 | struct vcpu_svm *svm = to_svm(vcpu); | |
18f40c53 | 4688 | u32 dfr = kvm_lapic_get_reg(vcpu->arch.apic, APIC_DFR); |
18f40c53 | 4689 | |
98d90582 SS |
4690 | if (svm->dfr_reg == dfr) |
4691 | return; | |
18f40c53 | 4692 | |
98d90582 SS |
4693 | avic_invalidate_logical_id_entry(vcpu); |
4694 | svm->dfr_reg = dfr; | |
18f40c53 SS |
4695 | } |
4696 | ||
4697 | static int avic_unaccel_trap_write(struct vcpu_svm *svm) | |
4698 | { | |
4699 | struct kvm_lapic *apic = svm->vcpu.arch.apic; | |
4700 | u32 offset = svm->vmcb->control.exit_info_1 & | |
4701 | AVIC_UNACCEL_ACCESS_OFFSET_MASK; | |
4702 | ||
4703 | switch (offset) { | |
4704 | case APIC_ID: | |
4705 | if (avic_handle_apic_id_update(&svm->vcpu)) | |
4706 | return 0; | |
4707 | break; | |
4708 | case APIC_LDR: | |
4709 | if (avic_handle_ldr_update(&svm->vcpu)) | |
4710 | return 0; | |
4711 | break; | |
4712 | case APIC_DFR: | |
4713 | avic_handle_dfr_update(&svm->vcpu); | |
4714 | break; | |
4715 | default: | |
4716 | break; | |
4717 | } | |
4718 | ||
4719 | kvm_lapic_reg_write(apic, offset, kvm_lapic_get_reg(apic, offset)); | |
4720 | ||
4721 | return 1; | |
4722 | } | |
4723 | ||
4724 | static bool is_avic_unaccelerated_access_trap(u32 offset) | |
4725 | { | |
4726 | bool ret = false; | |
4727 | ||
4728 | switch (offset) { | |
4729 | case APIC_ID: | |
4730 | case APIC_EOI: | |
4731 | case APIC_RRR: | |
4732 | case APIC_LDR: | |
4733 | case APIC_DFR: | |
4734 | case APIC_SPIV: | |
4735 | case APIC_ESR: | |
4736 | case APIC_ICR: | |
4737 | case APIC_LVTT: | |
4738 | case APIC_LVTTHMR: | |
4739 | case APIC_LVTPC: | |
4740 | case APIC_LVT0: | |
4741 | case APIC_LVT1: | |
4742 | case APIC_LVTERR: | |
4743 | case APIC_TMICT: | |
4744 | case APIC_TDCR: | |
4745 | ret = true; | |
4746 | break; | |
4747 | default: | |
4748 | break; | |
4749 | } | |
4750 | return ret; | |
4751 | } | |
4752 | ||
4753 | static int avic_unaccelerated_access_interception(struct vcpu_svm *svm) | |
4754 | { | |
4755 | int ret = 0; | |
4756 | u32 offset = svm->vmcb->control.exit_info_1 & | |
4757 | AVIC_UNACCEL_ACCESS_OFFSET_MASK; | |
4758 | u32 vector = svm->vmcb->control.exit_info_2 & | |
4759 | AVIC_UNACCEL_ACCESS_VECTOR_MASK; | |
4760 | bool write = (svm->vmcb->control.exit_info_1 >> 32) & | |
4761 | AVIC_UNACCEL_ACCESS_WRITE_MASK; | |
4762 | bool trap = is_avic_unaccelerated_access_trap(offset); | |
4763 | ||
4764 | trace_kvm_avic_unaccelerated_access(svm->vcpu.vcpu_id, offset, | |
4765 | trap, write, vector); | |
4766 | if (trap) { | |
4767 | /* Handling Trap */ | |
4768 | WARN_ONCE(!write, "svm: Handling trap read.\n"); | |
4769 | ret = avic_unaccel_trap_write(svm); | |
4770 | } else { | |
4771 | /* Handling Fault */ | |
0ce97a2b | 4772 | ret = (kvm_emulate_instruction(&svm->vcpu, 0) == EMULATE_DONE); |
18f40c53 SS |
4773 | } |
4774 | ||
4775 | return ret; | |
4776 | } | |
4777 | ||
09941fbb | 4778 | static int (*const svm_exit_handlers[])(struct vcpu_svm *svm) = { |
7ff76d58 AP |
4779 | [SVM_EXIT_READ_CR0] = cr_interception, |
4780 | [SVM_EXIT_READ_CR3] = cr_interception, | |
4781 | [SVM_EXIT_READ_CR4] = cr_interception, | |
4782 | [SVM_EXIT_READ_CR8] = cr_interception, | |
5e57518d | 4783 | [SVM_EXIT_CR0_SEL_WRITE] = cr_interception, |
628afd2a | 4784 | [SVM_EXIT_WRITE_CR0] = cr_interception, |
7ff76d58 AP |
4785 | [SVM_EXIT_WRITE_CR3] = cr_interception, |
4786 | [SVM_EXIT_WRITE_CR4] = cr_interception, | |
e0231715 | 4787 | [SVM_EXIT_WRITE_CR8] = cr8_write_interception, |
cae3797a AP |
4788 | [SVM_EXIT_READ_DR0] = dr_interception, |
4789 | [SVM_EXIT_READ_DR1] = dr_interception, | |
4790 | [SVM_EXIT_READ_DR2] = dr_interception, | |
4791 | [SVM_EXIT_READ_DR3] = dr_interception, | |
4792 | [SVM_EXIT_READ_DR4] = dr_interception, | |
4793 | [SVM_EXIT_READ_DR5] = dr_interception, | |
4794 | [SVM_EXIT_READ_DR6] = dr_interception, | |
4795 | [SVM_EXIT_READ_DR7] = dr_interception, | |
4796 | [SVM_EXIT_WRITE_DR0] = dr_interception, | |
4797 | [SVM_EXIT_WRITE_DR1] = dr_interception, | |
4798 | [SVM_EXIT_WRITE_DR2] = dr_interception, | |
4799 | [SVM_EXIT_WRITE_DR3] = dr_interception, | |
4800 | [SVM_EXIT_WRITE_DR4] = dr_interception, | |
4801 | [SVM_EXIT_WRITE_DR5] = dr_interception, | |
4802 | [SVM_EXIT_WRITE_DR6] = dr_interception, | |
4803 | [SVM_EXIT_WRITE_DR7] = dr_interception, | |
d0bfb940 JK |
4804 | [SVM_EXIT_EXCP_BASE + DB_VECTOR] = db_interception, |
4805 | [SVM_EXIT_EXCP_BASE + BP_VECTOR] = bp_interception, | |
7aa81cc0 | 4806 | [SVM_EXIT_EXCP_BASE + UD_VECTOR] = ud_interception, |
e0231715 | 4807 | [SVM_EXIT_EXCP_BASE + PF_VECTOR] = pf_interception, |
e0231715 | 4808 | [SVM_EXIT_EXCP_BASE + MC_VECTOR] = mc_interception, |
54a20552 | 4809 | [SVM_EXIT_EXCP_BASE + AC_VECTOR] = ac_interception, |
9718420e | 4810 | [SVM_EXIT_EXCP_BASE + GP_VECTOR] = gp_interception, |
e0231715 | 4811 | [SVM_EXIT_INTR] = intr_interception, |
c47f098d | 4812 | [SVM_EXIT_NMI] = nmi_interception, |
6aa8b732 AK |
4813 | [SVM_EXIT_SMI] = nop_on_interception, |
4814 | [SVM_EXIT_INIT] = nop_on_interception, | |
c1150d8c | 4815 | [SVM_EXIT_VINTR] = interrupt_window_interception, |
332b56e4 | 4816 | [SVM_EXIT_RDPMC] = rdpmc_interception, |
6aa8b732 | 4817 | [SVM_EXIT_CPUID] = cpuid_interception, |
95ba8273 | 4818 | [SVM_EXIT_IRET] = iret_interception, |
cf5a94d1 | 4819 | [SVM_EXIT_INVD] = emulate_on_interception, |
565d0998 | 4820 | [SVM_EXIT_PAUSE] = pause_interception, |
6aa8b732 | 4821 | [SVM_EXIT_HLT] = halt_interception, |
a7052897 | 4822 | [SVM_EXIT_INVLPG] = invlpg_interception, |
ff092385 | 4823 | [SVM_EXIT_INVLPGA] = invlpga_interception, |
e0231715 | 4824 | [SVM_EXIT_IOIO] = io_interception, |
6aa8b732 AK |
4825 | [SVM_EXIT_MSR] = msr_interception, |
4826 | [SVM_EXIT_TASK_SWITCH] = task_switch_interception, | |
46fe4ddd | 4827 | [SVM_EXIT_SHUTDOWN] = shutdown_interception, |
3d6368ef | 4828 | [SVM_EXIT_VMRUN] = vmrun_interception, |
02e235bc | 4829 | [SVM_EXIT_VMMCALL] = vmmcall_interception, |
5542675b AG |
4830 | [SVM_EXIT_VMLOAD] = vmload_interception, |
4831 | [SVM_EXIT_VMSAVE] = vmsave_interception, | |
1371d904 AG |
4832 | [SVM_EXIT_STGI] = stgi_interception, |
4833 | [SVM_EXIT_CLGI] = clgi_interception, | |
532a46b9 | 4834 | [SVM_EXIT_SKINIT] = skinit_interception, |
dab429a7 | 4835 | [SVM_EXIT_WBINVD] = wbinvd_interception, |
87c00572 GS |
4836 | [SVM_EXIT_MONITOR] = monitor_interception, |
4837 | [SVM_EXIT_MWAIT] = mwait_interception, | |
81dd35d4 | 4838 | [SVM_EXIT_XSETBV] = xsetbv_interception, |
d0006530 | 4839 | [SVM_EXIT_NPF] = npf_interception, |
7607b717 | 4840 | [SVM_EXIT_RSM] = rsm_interception, |
18f40c53 SS |
4841 | [SVM_EXIT_AVIC_INCOMPLETE_IPI] = avic_incomplete_ipi_interception, |
4842 | [SVM_EXIT_AVIC_UNACCELERATED_ACCESS] = avic_unaccelerated_access_interception, | |
6aa8b732 AK |
4843 | }; |
4844 | ||
ae8cc059 | 4845 | static void dump_vmcb(struct kvm_vcpu *vcpu) |
3f10c846 JR |
4846 | { |
4847 | struct vcpu_svm *svm = to_svm(vcpu); | |
4848 | struct vmcb_control_area *control = &svm->vmcb->control; | |
4849 | struct vmcb_save_area *save = &svm->vmcb->save; | |
4850 | ||
6f2f8453 PB |
4851 | if (!dump_invalid_vmcb) { |
4852 | pr_warn_ratelimited("set kvm_amd.dump_invalid_vmcb=1 to dump internal KVM state.\n"); | |
4853 | return; | |
4854 | } | |
4855 | ||
3f10c846 | 4856 | pr_err("VMCB Control Area:\n"); |
ae8cc059 JP |
4857 | pr_err("%-20s%04x\n", "cr_read:", control->intercept_cr & 0xffff); |
4858 | pr_err("%-20s%04x\n", "cr_write:", control->intercept_cr >> 16); | |
4859 | pr_err("%-20s%04x\n", "dr_read:", control->intercept_dr & 0xffff); | |
4860 | pr_err("%-20s%04x\n", "dr_write:", control->intercept_dr >> 16); | |
4861 | pr_err("%-20s%08x\n", "exceptions:", control->intercept_exceptions); | |
4862 | pr_err("%-20s%016llx\n", "intercepts:", control->intercept); | |
4863 | pr_err("%-20s%d\n", "pause filter count:", control->pause_filter_count); | |
1d8fb44a BM |
4864 | pr_err("%-20s%d\n", "pause filter threshold:", |
4865 | control->pause_filter_thresh); | |
ae8cc059 JP |
4866 | pr_err("%-20s%016llx\n", "iopm_base_pa:", control->iopm_base_pa); |
4867 | pr_err("%-20s%016llx\n", "msrpm_base_pa:", control->msrpm_base_pa); | |
4868 | pr_err("%-20s%016llx\n", "tsc_offset:", control->tsc_offset); | |
4869 | pr_err("%-20s%d\n", "asid:", control->asid); | |
4870 | pr_err("%-20s%d\n", "tlb_ctl:", control->tlb_ctl); | |
4871 | pr_err("%-20s%08x\n", "int_ctl:", control->int_ctl); | |
4872 | pr_err("%-20s%08x\n", "int_vector:", control->int_vector); | |
4873 | pr_err("%-20s%08x\n", "int_state:", control->int_state); | |
4874 | pr_err("%-20s%08x\n", "exit_code:", control->exit_code); | |
4875 | pr_err("%-20s%016llx\n", "exit_info1:", control->exit_info_1); | |
4876 | pr_err("%-20s%016llx\n", "exit_info2:", control->exit_info_2); | |
4877 | pr_err("%-20s%08x\n", "exit_int_info:", control->exit_int_info); | |
4878 | pr_err("%-20s%08x\n", "exit_int_info_err:", control->exit_int_info_err); | |
4879 | pr_err("%-20s%lld\n", "nested_ctl:", control->nested_ctl); | |
4880 | pr_err("%-20s%016llx\n", "nested_cr3:", control->nested_cr3); | |
44a95dae | 4881 | pr_err("%-20s%016llx\n", "avic_vapic_bar:", control->avic_vapic_bar); |
ae8cc059 JP |
4882 | pr_err("%-20s%08x\n", "event_inj:", control->event_inj); |
4883 | pr_err("%-20s%08x\n", "event_inj_err:", control->event_inj_err); | |
0dc92119 | 4884 | pr_err("%-20s%lld\n", "virt_ext:", control->virt_ext); |
ae8cc059 | 4885 | pr_err("%-20s%016llx\n", "next_rip:", control->next_rip); |
44a95dae SS |
4886 | pr_err("%-20s%016llx\n", "avic_backing_page:", control->avic_backing_page); |
4887 | pr_err("%-20s%016llx\n", "avic_logical_id:", control->avic_logical_id); | |
4888 | pr_err("%-20s%016llx\n", "avic_physical_id:", control->avic_physical_id); | |
3f10c846 | 4889 | pr_err("VMCB State Save Area:\n"); |
ae8cc059 JP |
4890 | pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n", |
4891 | "es:", | |
4892 | save->es.selector, save->es.attrib, | |
4893 | save->es.limit, save->es.base); | |
4894 | pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n", | |
4895 | "cs:", | |
4896 | save->cs.selector, save->cs.attrib, | |
4897 | save->cs.limit, save->cs.base); | |
4898 | pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n", | |
4899 | "ss:", | |
4900 | save->ss.selector, save->ss.attrib, | |
4901 | save->ss.limit, save->ss.base); | |
4902 | pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n", | |
4903 | "ds:", | |
4904 | save->ds.selector, save->ds.attrib, | |
4905 | save->ds.limit, save->ds.base); | |
4906 | pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n", | |
4907 | "fs:", | |
4908 | save->fs.selector, save->fs.attrib, | |
4909 | save->fs.limit, save->fs.base); | |
4910 | pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n", | |
4911 | "gs:", | |
4912 | save->gs.selector, save->gs.attrib, | |
4913 | save->gs.limit, save->gs.base); | |
4914 | pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n", | |
4915 | "gdtr:", | |
4916 | save->gdtr.selector, save->gdtr.attrib, | |
4917 | save->gdtr.limit, save->gdtr.base); | |
4918 | pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n", | |
4919 | "ldtr:", | |
4920 | save->ldtr.selector, save->ldtr.attrib, | |
4921 | save->ldtr.limit, save->ldtr.base); | |
4922 | pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n", | |
4923 | "idtr:", | |
4924 | save->idtr.selector, save->idtr.attrib, | |
4925 | save->idtr.limit, save->idtr.base); | |
4926 | pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n", | |
4927 | "tr:", | |
4928 | save->tr.selector, save->tr.attrib, | |
4929 | save->tr.limit, save->tr.base); | |
3f10c846 JR |
4930 | pr_err("cpl: %d efer: %016llx\n", |
4931 | save->cpl, save->efer); | |
ae8cc059 JP |
4932 | pr_err("%-15s %016llx %-13s %016llx\n", |
4933 | "cr0:", save->cr0, "cr2:", save->cr2); | |
4934 | pr_err("%-15s %016llx %-13s %016llx\n", | |
4935 | "cr3:", save->cr3, "cr4:", save->cr4); | |
4936 | pr_err("%-15s %016llx %-13s %016llx\n", | |
4937 | "dr6:", save->dr6, "dr7:", save->dr7); | |
4938 | pr_err("%-15s %016llx %-13s %016llx\n", | |
4939 | "rip:", save->rip, "rflags:", save->rflags); | |
4940 | pr_err("%-15s %016llx %-13s %016llx\n", | |
4941 | "rsp:", save->rsp, "rax:", save->rax); | |
4942 | pr_err("%-15s %016llx %-13s %016llx\n", | |
4943 | "star:", save->star, "lstar:", save->lstar); | |
4944 | pr_err("%-15s %016llx %-13s %016llx\n", | |
4945 | "cstar:", save->cstar, "sfmask:", save->sfmask); | |
4946 | pr_err("%-15s %016llx %-13s %016llx\n", | |
4947 | "kernel_gs_base:", save->kernel_gs_base, | |
4948 | "sysenter_cs:", save->sysenter_cs); | |
4949 | pr_err("%-15s %016llx %-13s %016llx\n", | |
4950 | "sysenter_esp:", save->sysenter_esp, | |
4951 | "sysenter_eip:", save->sysenter_eip); | |
4952 | pr_err("%-15s %016llx %-13s %016llx\n", | |
4953 | "gpat:", save->g_pat, "dbgctl:", save->dbgctl); | |
4954 | pr_err("%-15s %016llx %-13s %016llx\n", | |
4955 | "br_from:", save->br_from, "br_to:", save->br_to); | |
4956 | pr_err("%-15s %016llx %-13s %016llx\n", | |
4957 | "excp_from:", save->last_excp_from, | |
4958 | "excp_to:", save->last_excp_to); | |
3f10c846 JR |
4959 | } |
4960 | ||
586f9607 AK |
4961 | static void svm_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2) |
4962 | { | |
4963 | struct vmcb_control_area *control = &to_svm(vcpu)->vmcb->control; | |
4964 | ||
4965 | *info1 = control->exit_info_1; | |
4966 | *info2 = control->exit_info_2; | |
4967 | } | |
4968 | ||
851ba692 | 4969 | static int handle_exit(struct kvm_vcpu *vcpu) |
6aa8b732 | 4970 | { |
04d2cc77 | 4971 | struct vcpu_svm *svm = to_svm(vcpu); |
851ba692 | 4972 | struct kvm_run *kvm_run = vcpu->run; |
a2fa3e9f | 4973 | u32 exit_code = svm->vmcb->control.exit_code; |
6aa8b732 | 4974 | |
8b89fe1f PB |
4975 | trace_kvm_exit(exit_code, vcpu, KVM_ISA_SVM); |
4976 | ||
4ee546b4 | 4977 | if (!is_cr_intercept(svm, INTERCEPT_CR0_WRITE)) |
2be4fc7a JR |
4978 | vcpu->arch.cr0 = svm->vmcb->save.cr0; |
4979 | if (npt_enabled) | |
4980 | vcpu->arch.cr3 = svm->vmcb->save.cr3; | |
af9ca2d7 | 4981 | |
cd3ff653 JR |
4982 | if (unlikely(svm->nested.exit_required)) { |
4983 | nested_svm_vmexit(svm); | |
4984 | svm->nested.exit_required = false; | |
4985 | ||
4986 | return 1; | |
4987 | } | |
4988 | ||
2030753d | 4989 | if (is_guest_mode(vcpu)) { |
410e4d57 JR |
4990 | int vmexit; |
4991 | ||
d8cabddf JR |
4992 | trace_kvm_nested_vmexit(svm->vmcb->save.rip, exit_code, |
4993 | svm->vmcb->control.exit_info_1, | |
4994 | svm->vmcb->control.exit_info_2, | |
4995 | svm->vmcb->control.exit_int_info, | |
e097e5ff SH |
4996 | svm->vmcb->control.exit_int_info_err, |
4997 | KVM_ISA_SVM); | |
d8cabddf | 4998 | |
410e4d57 JR |
4999 | vmexit = nested_svm_exit_special(svm); |
5000 | ||
5001 | if (vmexit == NESTED_EXIT_CONTINUE) | |
5002 | vmexit = nested_svm_exit_handled(svm); | |
5003 | ||
5004 | if (vmexit == NESTED_EXIT_DONE) | |
cf74a78b | 5005 | return 1; |
cf74a78b AG |
5006 | } |
5007 | ||
a5c3832d JR |
5008 | svm_complete_interrupts(svm); |
5009 | ||
04d2cc77 AK |
5010 | if (svm->vmcb->control.exit_code == SVM_EXIT_ERR) { |
5011 | kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY; | |
5012 | kvm_run->fail_entry.hardware_entry_failure_reason | |
5013 | = svm->vmcb->control.exit_code; | |
3f10c846 | 5014 | dump_vmcb(vcpu); |
04d2cc77 AK |
5015 | return 0; |
5016 | } | |
5017 | ||
a2fa3e9f | 5018 | if (is_external_interrupt(svm->vmcb->control.exit_int_info) && |
709ddebf | 5019 | exit_code != SVM_EXIT_EXCP_BASE + PF_VECTOR && |
55c5e464 JR |
5020 | exit_code != SVM_EXIT_NPF && exit_code != SVM_EXIT_TASK_SWITCH && |
5021 | exit_code != SVM_EXIT_INTR && exit_code != SVM_EXIT_NMI) | |
6614c7d0 | 5022 | printk(KERN_ERR "%s: unexpected exit_int_info 0x%x " |
6aa8b732 | 5023 | "exit_code 0x%x\n", |
b8688d51 | 5024 | __func__, svm->vmcb->control.exit_int_info, |
6aa8b732 AK |
5025 | exit_code); |
5026 | ||
9d8f549d | 5027 | if (exit_code >= ARRAY_SIZE(svm_exit_handlers) |
56919c5c | 5028 | || !svm_exit_handlers[exit_code]) { |
faac2458 | 5029 | WARN_ONCE(1, "svm: unexpected exit reason 0x%x\n", exit_code); |
2bc19dc3 MT |
5030 | kvm_queue_exception(vcpu, UD_VECTOR); |
5031 | return 1; | |
6aa8b732 AK |
5032 | } |
5033 | ||
851ba692 | 5034 | return svm_exit_handlers[exit_code](svm); |
6aa8b732 AK |
5035 | } |
5036 | ||
5037 | static void reload_tss(struct kvm_vcpu *vcpu) | |
5038 | { | |
5039 | int cpu = raw_smp_processor_id(); | |
5040 | ||
0fe1e009 TH |
5041 | struct svm_cpu_data *sd = per_cpu(svm_data, cpu); |
5042 | sd->tss_desc->type = 9; /* available 32/64-bit TSS */ | |
6aa8b732 AK |
5043 | load_TR_desc(); |
5044 | } | |
5045 | ||
70cd94e6 BS |
5046 | static void pre_sev_run(struct vcpu_svm *svm, int cpu) |
5047 | { | |
5048 | struct svm_cpu_data *sd = per_cpu(svm_data, cpu); | |
5049 | int asid = sev_get_asid(svm->vcpu.kvm); | |
5050 | ||
5051 | /* Assign the asid allocated with this SEV guest */ | |
5052 | svm->vmcb->control.asid = asid; | |
5053 | ||
5054 | /* | |
5055 | * Flush guest TLB: | |
5056 | * | |
5057 | * 1) when different VMCB for the same ASID is to be run on the same host CPU. | |
5058 | * 2) or this VMCB was executed on different host CPU in previous VMRUNs. | |
5059 | */ | |
5060 | if (sd->sev_vmcbs[asid] == svm->vmcb && | |
5061 | svm->last_cpu == cpu) | |
5062 | return; | |
5063 | ||
5064 | svm->last_cpu = cpu; | |
5065 | sd->sev_vmcbs[asid] = svm->vmcb; | |
5066 | svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ASID; | |
5067 | mark_dirty(svm->vmcb, VMCB_ASID); | |
5068 | } | |
5069 | ||
e756fc62 | 5070 | static void pre_svm_run(struct vcpu_svm *svm) |
6aa8b732 AK |
5071 | { |
5072 | int cpu = raw_smp_processor_id(); | |
5073 | ||
0fe1e009 | 5074 | struct svm_cpu_data *sd = per_cpu(svm_data, cpu); |
6aa8b732 | 5075 | |
70cd94e6 BS |
5076 | if (sev_guest(svm->vcpu.kvm)) |
5077 | return pre_sev_run(svm, cpu); | |
5078 | ||
4b656b12 | 5079 | /* FIXME: handle wraparound of asid_generation */ |
0fe1e009 TH |
5080 | if (svm->asid_generation != sd->asid_generation) |
5081 | new_asid(svm, sd); | |
6aa8b732 AK |
5082 | } |
5083 | ||
95ba8273 GN |
5084 | static void svm_inject_nmi(struct kvm_vcpu *vcpu) |
5085 | { | |
5086 | struct vcpu_svm *svm = to_svm(vcpu); | |
5087 | ||
5088 | svm->vmcb->control.event_inj = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI; | |
5089 | vcpu->arch.hflags |= HF_NMI_MASK; | |
8a05a1b8 | 5090 | set_intercept(svm, INTERCEPT_IRET); |
95ba8273 GN |
5091 | ++vcpu->stat.nmi_injections; |
5092 | } | |
6aa8b732 | 5093 | |
85f455f7 | 5094 | static inline void svm_inject_irq(struct vcpu_svm *svm, int irq) |
6aa8b732 AK |
5095 | { |
5096 | struct vmcb_control_area *control; | |
5097 | ||
340d3bc3 | 5098 | /* The following fields are ignored when AVIC is enabled */ |
e756fc62 | 5099 | control = &svm->vmcb->control; |
85f455f7 | 5100 | control->int_vector = irq; |
6aa8b732 AK |
5101 | control->int_ctl &= ~V_INTR_PRIO_MASK; |
5102 | control->int_ctl |= V_IRQ_MASK | | |
5103 | ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT); | |
decdbf6a | 5104 | mark_dirty(svm->vmcb, VMCB_INTR); |
6aa8b732 AK |
5105 | } |
5106 | ||
66fd3f7f | 5107 | static void svm_set_irq(struct kvm_vcpu *vcpu) |
2a8067f1 ED |
5108 | { |
5109 | struct vcpu_svm *svm = to_svm(vcpu); | |
5110 | ||
2af9194d | 5111 | BUG_ON(!(gif_set(svm))); |
cf74a78b | 5112 | |
9fb2d2b4 GN |
5113 | trace_kvm_inj_virq(vcpu->arch.interrupt.nr); |
5114 | ++vcpu->stat.irq_injections; | |
5115 | ||
219b65dc AG |
5116 | svm->vmcb->control.event_inj = vcpu->arch.interrupt.nr | |
5117 | SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR; | |
2a8067f1 ED |
5118 | } |
5119 | ||
3bbf3565 SS |
5120 | static inline bool svm_nested_virtualize_tpr(struct kvm_vcpu *vcpu) |
5121 | { | |
5122 | return is_guest_mode(vcpu) && (vcpu->arch.hflags & HF_VINTR_MASK); | |
5123 | } | |
5124 | ||
95ba8273 | 5125 | static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr) |
aaacfc9a JR |
5126 | { |
5127 | struct vcpu_svm *svm = to_svm(vcpu); | |
aaacfc9a | 5128 | |
3bbf3565 SS |
5129 | if (svm_nested_virtualize_tpr(vcpu) || |
5130 | kvm_vcpu_apicv_active(vcpu)) | |
88ab24ad JR |
5131 | return; |
5132 | ||
596f3142 RK |
5133 | clr_cr_intercept(svm, INTERCEPT_CR8_WRITE); |
5134 | ||
95ba8273 | 5135 | if (irr == -1) |
aaacfc9a JR |
5136 | return; |
5137 | ||
95ba8273 | 5138 | if (tpr >= irr) |
4ee546b4 | 5139 | set_cr_intercept(svm, INTERCEPT_CR8_WRITE); |
95ba8273 | 5140 | } |
aaacfc9a | 5141 | |
8d860bbe | 5142 | static void svm_set_virtual_apic_mode(struct kvm_vcpu *vcpu) |
8d14695f YZ |
5143 | { |
5144 | return; | |
5145 | } | |
5146 | ||
b2a05fef | 5147 | static bool svm_get_enable_apicv(struct kvm_vcpu *vcpu) |
d62caabb | 5148 | { |
67034bb9 | 5149 | return avic && irqchip_split(vcpu->kvm); |
44a95dae SS |
5150 | } |
5151 | ||
5152 | static void svm_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr) | |
5153 | { | |
d62caabb AS |
5154 | } |
5155 | ||
67c9dddc | 5156 | static void svm_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr) |
44a95dae | 5157 | { |
d62caabb AS |
5158 | } |
5159 | ||
44a95dae | 5160 | /* Note: Currently only used by Hyper-V. */ |
d62caabb | 5161 | static void svm_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu) |
c7c9c56c | 5162 | { |
44a95dae SS |
5163 | struct vcpu_svm *svm = to_svm(vcpu); |
5164 | struct vmcb *vmcb = svm->vmcb; | |
5165 | ||
c57cd3c8 SS |
5166 | if (kvm_vcpu_apicv_active(vcpu)) |
5167 | vmcb->control.int_ctl |= AVIC_ENABLE_MASK; | |
5168 | else | |
5169 | vmcb->control.int_ctl &= ~AVIC_ENABLE_MASK; | |
5170 | mark_dirty(vmcb, VMCB_AVIC); | |
c7c9c56c YZ |
5171 | } |
5172 | ||
6308630b | 5173 | static void svm_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap) |
c7c9c56c YZ |
5174 | { |
5175 | return; | |
5176 | } | |
5177 | ||
340d3bc3 SS |
5178 | static void svm_deliver_avic_intr(struct kvm_vcpu *vcpu, int vec) |
5179 | { | |
5180 | kvm_lapic_set_irr(vec, vcpu->arch.apic); | |
5181 | smp_mb__after_atomic(); | |
5182 | ||
0532dd52 SS |
5183 | if (avic_vcpu_is_running(vcpu)) { |
5184 | int cpuid = vcpu->cpu; | |
5185 | ||
5186 | if (cpuid != get_cpu()) | |
5187 | wrmsrl(SVM_AVIC_DOORBELL, kvm_cpu_get_apicid(cpuid)); | |
5188 | put_cpu(); | |
5189 | } else | |
340d3bc3 SS |
5190 | kvm_vcpu_wake_up(vcpu); |
5191 | } | |
5192 | ||
411b44ba SS |
5193 | static void svm_ir_list_del(struct vcpu_svm *svm, struct amd_iommu_pi_data *pi) |
5194 | { | |
5195 | unsigned long flags; | |
5196 | struct amd_svm_iommu_ir *cur; | |
5197 | ||
5198 | spin_lock_irqsave(&svm->ir_list_lock, flags); | |
5199 | list_for_each_entry(cur, &svm->ir_list, node) { | |
5200 | if (cur->data != pi->ir_data) | |
5201 | continue; | |
5202 | list_del(&cur->node); | |
5203 | kfree(cur); | |
5204 | break; | |
5205 | } | |
5206 | spin_unlock_irqrestore(&svm->ir_list_lock, flags); | |
5207 | } | |
5208 | ||
5209 | static int svm_ir_list_add(struct vcpu_svm *svm, struct amd_iommu_pi_data *pi) | |
5210 | { | |
5211 | int ret = 0; | |
5212 | unsigned long flags; | |
5213 | struct amd_svm_iommu_ir *ir; | |
5214 | ||
5215 | /** | |
5216 | * In some cases, the existing irte is updaed and re-set, | |
5217 | * so we need to check here if it's already been * added | |
5218 | * to the ir_list. | |
5219 | */ | |
5220 | if (pi->ir_data && (pi->prev_ga_tag != 0)) { | |
5221 | struct kvm *kvm = svm->vcpu.kvm; | |
5222 | u32 vcpu_id = AVIC_GATAG_TO_VCPUID(pi->prev_ga_tag); | |
5223 | struct kvm_vcpu *prev_vcpu = kvm_get_vcpu_by_id(kvm, vcpu_id); | |
5224 | struct vcpu_svm *prev_svm; | |
5225 | ||
5226 | if (!prev_vcpu) { | |
5227 | ret = -EINVAL; | |
5228 | goto out; | |
5229 | } | |
5230 | ||
5231 | prev_svm = to_svm(prev_vcpu); | |
5232 | svm_ir_list_del(prev_svm, pi); | |
5233 | } | |
5234 | ||
5235 | /** | |
5236 | * Allocating new amd_iommu_pi_data, which will get | |
5237 | * add to the per-vcpu ir_list. | |
5238 | */ | |
1ec69647 | 5239 | ir = kzalloc(sizeof(struct amd_svm_iommu_ir), GFP_KERNEL_ACCOUNT); |
411b44ba SS |
5240 | if (!ir) { |
5241 | ret = -ENOMEM; | |
5242 | goto out; | |
5243 | } | |
5244 | ir->data = pi->ir_data; | |
5245 | ||
5246 | spin_lock_irqsave(&svm->ir_list_lock, flags); | |
5247 | list_add(&ir->node, &svm->ir_list); | |
5248 | spin_unlock_irqrestore(&svm->ir_list_lock, flags); | |
5249 | out: | |
5250 | return ret; | |
5251 | } | |
5252 | ||
5253 | /** | |
5254 | * Note: | |
5255 | * The HW cannot support posting multicast/broadcast | |
5256 | * interrupts to a vCPU. So, we still use legacy interrupt | |
5257 | * remapping for these kind of interrupts. | |
5258 | * | |
5259 | * For lowest-priority interrupts, we only support | |
5260 | * those with single CPU as the destination, e.g. user | |
5261 | * configures the interrupts via /proc/irq or uses | |
5262 | * irqbalance to make the interrupts single-CPU. | |
5263 | */ | |
5264 | static int | |
5265 | get_pi_vcpu_info(struct kvm *kvm, struct kvm_kernel_irq_routing_entry *e, | |
5266 | struct vcpu_data *vcpu_info, struct vcpu_svm **svm) | |
5267 | { | |
5268 | struct kvm_lapic_irq irq; | |
5269 | struct kvm_vcpu *vcpu = NULL; | |
5270 | ||
5271 | kvm_set_msi_irq(kvm, e, &irq); | |
5272 | ||
5273 | if (!kvm_intr_is_single_vcpu(kvm, &irq, &vcpu)) { | |
5274 | pr_debug("SVM: %s: use legacy intr remap mode for irq %u\n", | |
5275 | __func__, irq.vector); | |
5276 | return -1; | |
5277 | } | |
5278 | ||
5279 | pr_debug("SVM: %s: use GA mode for irq %u\n", __func__, | |
5280 | irq.vector); | |
5281 | *svm = to_svm(vcpu); | |
d0ec49d4 | 5282 | vcpu_info->pi_desc_addr = __sme_set(page_to_phys((*svm)->avic_backing_page)); |
411b44ba SS |
5283 | vcpu_info->vector = irq.vector; |
5284 | ||
5285 | return 0; | |
5286 | } | |
5287 | ||
5288 | /* | |
5289 | * svm_update_pi_irte - set IRTE for Posted-Interrupts | |
5290 | * | |
5291 | * @kvm: kvm | |
5292 | * @host_irq: host irq of the interrupt | |
5293 | * @guest_irq: gsi of the interrupt | |
5294 | * @set: set or unset PI | |
5295 | * returns 0 on success, < 0 on failure | |
5296 | */ | |
5297 | static int svm_update_pi_irte(struct kvm *kvm, unsigned int host_irq, | |
5298 | uint32_t guest_irq, bool set) | |
5299 | { | |
5300 | struct kvm_kernel_irq_routing_entry *e; | |
5301 | struct kvm_irq_routing_table *irq_rt; | |
5302 | int idx, ret = -EINVAL; | |
5303 | ||
5304 | if (!kvm_arch_has_assigned_device(kvm) || | |
5305 | !irq_remapping_cap(IRQ_POSTING_CAP)) | |
5306 | return 0; | |
5307 | ||
5308 | pr_debug("SVM: %s: host_irq=%#x, guest_irq=%#x, set=%#x\n", | |
5309 | __func__, host_irq, guest_irq, set); | |
5310 | ||
5311 | idx = srcu_read_lock(&kvm->irq_srcu); | |
5312 | irq_rt = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu); | |
5313 | WARN_ON(guest_irq >= irq_rt->nr_rt_entries); | |
5314 | ||
5315 | hlist_for_each_entry(e, &irq_rt->map[guest_irq], link) { | |
5316 | struct vcpu_data vcpu_info; | |
5317 | struct vcpu_svm *svm = NULL; | |
5318 | ||
5319 | if (e->type != KVM_IRQ_ROUTING_MSI) | |
5320 | continue; | |
5321 | ||
5322 | /** | |
5323 | * Here, we setup with legacy mode in the following cases: | |
5324 | * 1. When cannot target interrupt to a specific vcpu. | |
5325 | * 2. Unsetting posted interrupt. | |
5326 | * 3. APIC virtialization is disabled for the vcpu. | |
5327 | */ | |
5328 | if (!get_pi_vcpu_info(kvm, e, &vcpu_info, &svm) && set && | |
5329 | kvm_vcpu_apicv_active(&svm->vcpu)) { | |
5330 | struct amd_iommu_pi_data pi; | |
5331 | ||
5332 | /* Try to enable guest_mode in IRTE */ | |
d0ec49d4 TL |
5333 | pi.base = __sme_set(page_to_phys(svm->avic_backing_page) & |
5334 | AVIC_HPA_MASK); | |
81811c16 | 5335 | pi.ga_tag = AVIC_GATAG(to_kvm_svm(kvm)->avic_vm_id, |
411b44ba SS |
5336 | svm->vcpu.vcpu_id); |
5337 | pi.is_guest_mode = true; | |
5338 | pi.vcpu_data = &vcpu_info; | |
5339 | ret = irq_set_vcpu_affinity(host_irq, &pi); | |
5340 | ||
5341 | /** | |
5342 | * Here, we successfully setting up vcpu affinity in | |
5343 | * IOMMU guest mode. Now, we need to store the posted | |
5344 | * interrupt information in a per-vcpu ir_list so that | |
5345 | * we can reference to them directly when we update vcpu | |
5346 | * scheduling information in IOMMU irte. | |
5347 | */ | |
5348 | if (!ret && pi.is_guest_mode) | |
5349 | svm_ir_list_add(svm, &pi); | |
5350 | } else { | |
5351 | /* Use legacy mode in IRTE */ | |
5352 | struct amd_iommu_pi_data pi; | |
5353 | ||
5354 | /** | |
5355 | * Here, pi is used to: | |
5356 | * - Tell IOMMU to use legacy mode for this interrupt. | |
5357 | * - Retrieve ga_tag of prior interrupt remapping data. | |
5358 | */ | |
5359 | pi.is_guest_mode = false; | |
5360 | ret = irq_set_vcpu_affinity(host_irq, &pi); | |
5361 | ||
5362 | /** | |
5363 | * Check if the posted interrupt was previously | |
5364 | * setup with the guest_mode by checking if the ga_tag | |
5365 | * was cached. If so, we need to clean up the per-vcpu | |
5366 | * ir_list. | |
5367 | */ | |
5368 | if (!ret && pi.prev_ga_tag) { | |
5369 | int id = AVIC_GATAG_TO_VCPUID(pi.prev_ga_tag); | |
5370 | struct kvm_vcpu *vcpu; | |
5371 | ||
5372 | vcpu = kvm_get_vcpu_by_id(kvm, id); | |
5373 | if (vcpu) | |
5374 | svm_ir_list_del(to_svm(vcpu), &pi); | |
5375 | } | |
5376 | } | |
5377 | ||
5378 | if (!ret && svm) { | |
2698d82e | 5379 | trace_kvm_pi_irte_update(host_irq, svm->vcpu.vcpu_id, |
5380 | e->gsi, vcpu_info.vector, | |
411b44ba SS |
5381 | vcpu_info.pi_desc_addr, set); |
5382 | } | |
5383 | ||
5384 | if (ret < 0) { | |
5385 | pr_err("%s: failed to update PI IRTE\n", __func__); | |
5386 | goto out; | |
5387 | } | |
5388 | } | |
5389 | ||
5390 | ret = 0; | |
5391 | out: | |
5392 | srcu_read_unlock(&kvm->irq_srcu, idx); | |
5393 | return ret; | |
5394 | } | |
5395 | ||
95ba8273 GN |
5396 | static int svm_nmi_allowed(struct kvm_vcpu *vcpu) |
5397 | { | |
5398 | struct vcpu_svm *svm = to_svm(vcpu); | |
5399 | struct vmcb *vmcb = svm->vmcb; | |
924584cc JR |
5400 | int ret; |
5401 | ret = !(vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK) && | |
5402 | !(svm->vcpu.arch.hflags & HF_NMI_MASK); | |
5403 | ret = ret && gif_set(svm) && nested_svm_nmi(svm); | |
5404 | ||
5405 | return ret; | |
aaacfc9a JR |
5406 | } |
5407 | ||
3cfc3092 JK |
5408 | static bool svm_get_nmi_mask(struct kvm_vcpu *vcpu) |
5409 | { | |
5410 | struct vcpu_svm *svm = to_svm(vcpu); | |
5411 | ||
5412 | return !!(svm->vcpu.arch.hflags & HF_NMI_MASK); | |
5413 | } | |
5414 | ||
5415 | static void svm_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked) | |
5416 | { | |
5417 | struct vcpu_svm *svm = to_svm(vcpu); | |
5418 | ||
5419 | if (masked) { | |
5420 | svm->vcpu.arch.hflags |= HF_NMI_MASK; | |
8a05a1b8 | 5421 | set_intercept(svm, INTERCEPT_IRET); |
3cfc3092 JK |
5422 | } else { |
5423 | svm->vcpu.arch.hflags &= ~HF_NMI_MASK; | |
8a05a1b8 | 5424 | clr_intercept(svm, INTERCEPT_IRET); |
3cfc3092 JK |
5425 | } |
5426 | } | |
5427 | ||
78646121 GN |
5428 | static int svm_interrupt_allowed(struct kvm_vcpu *vcpu) |
5429 | { | |
5430 | struct vcpu_svm *svm = to_svm(vcpu); | |
5431 | struct vmcb *vmcb = svm->vmcb; | |
7fcdb510 JR |
5432 | int ret; |
5433 | ||
5434 | if (!gif_set(svm) || | |
5435 | (vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK)) | |
5436 | return 0; | |
5437 | ||
f6e78475 | 5438 | ret = !!(kvm_get_rflags(vcpu) & X86_EFLAGS_IF); |
7fcdb510 | 5439 | |
2030753d | 5440 | if (is_guest_mode(vcpu)) |
7fcdb510 JR |
5441 | return ret && !(svm->vcpu.arch.hflags & HF_VINTR_MASK); |
5442 | ||
5443 | return ret; | |
78646121 GN |
5444 | } |
5445 | ||
c9a7953f | 5446 | static void enable_irq_window(struct kvm_vcpu *vcpu) |
6aa8b732 | 5447 | { |
219b65dc | 5448 | struct vcpu_svm *svm = to_svm(vcpu); |
219b65dc | 5449 | |
340d3bc3 SS |
5450 | if (kvm_vcpu_apicv_active(vcpu)) |
5451 | return; | |
5452 | ||
e0231715 JR |
5453 | /* |
5454 | * In case GIF=0 we can't rely on the CPU to tell us when GIF becomes | |
5455 | * 1, because that's a separate STGI/VMRUN intercept. The next time we | |
5456 | * get that intercept, this function will be called again though and | |
640bd6e5 JN |
5457 | * we'll get the vintr intercept. However, if the vGIF feature is |
5458 | * enabled, the STGI interception will not occur. Enable the irq | |
5459 | * window under the assumption that the hardware will set the GIF. | |
e0231715 | 5460 | */ |
640bd6e5 | 5461 | if ((vgif_enabled(svm) || gif_set(svm)) && nested_svm_intr(svm)) { |
219b65dc AG |
5462 | svm_set_vintr(svm); |
5463 | svm_inject_irq(svm, 0x0); | |
5464 | } | |
85f455f7 ED |
5465 | } |
5466 | ||
c9a7953f | 5467 | static void enable_nmi_window(struct kvm_vcpu *vcpu) |
c1150d8c | 5468 | { |
04d2cc77 | 5469 | struct vcpu_svm *svm = to_svm(vcpu); |
c1150d8c | 5470 | |
44c11430 GN |
5471 | if ((svm->vcpu.arch.hflags & (HF_NMI_MASK | HF_IRET_MASK)) |
5472 | == HF_NMI_MASK) | |
c9a7953f | 5473 | return; /* IRET will cause a vm exit */ |
44c11430 | 5474 | |
640bd6e5 JN |
5475 | if (!gif_set(svm)) { |
5476 | if (vgif_enabled(svm)) | |
5477 | set_intercept(svm, INTERCEPT_STGI); | |
1a5e1852 | 5478 | return; /* STGI will cause a vm exit */ |
640bd6e5 | 5479 | } |
1a5e1852 LP |
5480 | |
5481 | if (svm->nested.exit_required) | |
5482 | return; /* we're not going to run the guest yet */ | |
5483 | ||
e0231715 JR |
5484 | /* |
5485 | * Something prevents NMI from been injected. Single step over possible | |
5486 | * problem (IRET or exception injection or interrupt shadow) | |
5487 | */ | |
ab2f4d73 | 5488 | svm->nmi_singlestep_guest_rflags = svm_get_rflags(vcpu); |
6be7d306 | 5489 | svm->nmi_singlestep = true; |
44c11430 | 5490 | svm->vmcb->save.rflags |= (X86_EFLAGS_TF | X86_EFLAGS_RF); |
c1150d8c DL |
5491 | } |
5492 | ||
cbc94022 IE |
5493 | static int svm_set_tss_addr(struct kvm *kvm, unsigned int addr) |
5494 | { | |
5495 | return 0; | |
5496 | } | |
5497 | ||
2ac52ab8 SC |
5498 | static int svm_set_identity_map_addr(struct kvm *kvm, u64 ident_addr) |
5499 | { | |
5500 | return 0; | |
5501 | } | |
5502 | ||
c2ba05cc | 5503 | static void svm_flush_tlb(struct kvm_vcpu *vcpu, bool invalidate_gpa) |
d9e368d6 | 5504 | { |
38e5e92f JR |
5505 | struct vcpu_svm *svm = to_svm(vcpu); |
5506 | ||
5507 | if (static_cpu_has(X86_FEATURE_FLUSHBYASID)) | |
5508 | svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ASID; | |
5509 | else | |
5510 | svm->asid_generation--; | |
d9e368d6 AK |
5511 | } |
5512 | ||
faff8758 JS |
5513 | static void svm_flush_tlb_gva(struct kvm_vcpu *vcpu, gva_t gva) |
5514 | { | |
5515 | struct vcpu_svm *svm = to_svm(vcpu); | |
5516 | ||
5517 | invlpga(gva, svm->vmcb->control.asid); | |
5518 | } | |
5519 | ||
04d2cc77 AK |
5520 | static void svm_prepare_guest_switch(struct kvm_vcpu *vcpu) |
5521 | { | |
5522 | } | |
5523 | ||
d7bf8221 JR |
5524 | static inline void sync_cr8_to_lapic(struct kvm_vcpu *vcpu) |
5525 | { | |
5526 | struct vcpu_svm *svm = to_svm(vcpu); | |
5527 | ||
3bbf3565 | 5528 | if (svm_nested_virtualize_tpr(vcpu)) |
88ab24ad JR |
5529 | return; |
5530 | ||
4ee546b4 | 5531 | if (!is_cr_intercept(svm, INTERCEPT_CR8_WRITE)) { |
d7bf8221 | 5532 | int cr8 = svm->vmcb->control.int_ctl & V_TPR_MASK; |
615d5193 | 5533 | kvm_set_cr8(vcpu, cr8); |
d7bf8221 JR |
5534 | } |
5535 | } | |
5536 | ||
649d6864 JR |
5537 | static inline void sync_lapic_to_cr8(struct kvm_vcpu *vcpu) |
5538 | { | |
5539 | struct vcpu_svm *svm = to_svm(vcpu); | |
5540 | u64 cr8; | |
5541 | ||
3bbf3565 SS |
5542 | if (svm_nested_virtualize_tpr(vcpu) || |
5543 | kvm_vcpu_apicv_active(vcpu)) | |
88ab24ad JR |
5544 | return; |
5545 | ||
649d6864 JR |
5546 | cr8 = kvm_get_cr8(vcpu); |
5547 | svm->vmcb->control.int_ctl &= ~V_TPR_MASK; | |
5548 | svm->vmcb->control.int_ctl |= cr8 & V_TPR_MASK; | |
5549 | } | |
5550 | ||
9222be18 GN |
5551 | static void svm_complete_interrupts(struct vcpu_svm *svm) |
5552 | { | |
5553 | u8 vector; | |
5554 | int type; | |
5555 | u32 exitintinfo = svm->vmcb->control.exit_int_info; | |
66b7138f JK |
5556 | unsigned int3_injected = svm->int3_injected; |
5557 | ||
5558 | svm->int3_injected = 0; | |
9222be18 | 5559 | |
bd3d1ec3 AK |
5560 | /* |
5561 | * If we've made progress since setting HF_IRET_MASK, we've | |
5562 | * executed an IRET and can allow NMI injection. | |
5563 | */ | |
5564 | if ((svm->vcpu.arch.hflags & HF_IRET_MASK) | |
5565 | && kvm_rip_read(&svm->vcpu) != svm->nmi_iret_rip) { | |
44c11430 | 5566 | svm->vcpu.arch.hflags &= ~(HF_NMI_MASK | HF_IRET_MASK); |
3842d135 AK |
5567 | kvm_make_request(KVM_REQ_EVENT, &svm->vcpu); |
5568 | } | |
44c11430 | 5569 | |
9222be18 GN |
5570 | svm->vcpu.arch.nmi_injected = false; |
5571 | kvm_clear_exception_queue(&svm->vcpu); | |
5572 | kvm_clear_interrupt_queue(&svm->vcpu); | |
5573 | ||
5574 | if (!(exitintinfo & SVM_EXITINTINFO_VALID)) | |
5575 | return; | |
5576 | ||
3842d135 AK |
5577 | kvm_make_request(KVM_REQ_EVENT, &svm->vcpu); |
5578 | ||
9222be18 GN |
5579 | vector = exitintinfo & SVM_EXITINTINFO_VEC_MASK; |
5580 | type = exitintinfo & SVM_EXITINTINFO_TYPE_MASK; | |
5581 | ||
5582 | switch (type) { | |
5583 | case SVM_EXITINTINFO_TYPE_NMI: | |
5584 | svm->vcpu.arch.nmi_injected = true; | |
5585 | break; | |
5586 | case SVM_EXITINTINFO_TYPE_EXEPT: | |
66b7138f JK |
5587 | /* |
5588 | * In case of software exceptions, do not reinject the vector, | |
5589 | * but re-execute the instruction instead. Rewind RIP first | |
5590 | * if we emulated INT3 before. | |
5591 | */ | |
5592 | if (kvm_exception_is_soft(vector)) { | |
5593 | if (vector == BP_VECTOR && int3_injected && | |
5594 | kvm_is_linear_rip(&svm->vcpu, svm->int3_rip)) | |
5595 | kvm_rip_write(&svm->vcpu, | |
5596 | kvm_rip_read(&svm->vcpu) - | |
5597 | int3_injected); | |
9222be18 | 5598 | break; |
66b7138f | 5599 | } |
9222be18 GN |
5600 | if (exitintinfo & SVM_EXITINTINFO_VALID_ERR) { |
5601 | u32 err = svm->vmcb->control.exit_int_info_err; | |
ce7ddec4 | 5602 | kvm_requeue_exception_e(&svm->vcpu, vector, err); |
9222be18 GN |
5603 | |
5604 | } else | |
ce7ddec4 | 5605 | kvm_requeue_exception(&svm->vcpu, vector); |
9222be18 GN |
5606 | break; |
5607 | case SVM_EXITINTINFO_TYPE_INTR: | |
66fd3f7f | 5608 | kvm_queue_interrupt(&svm->vcpu, vector, false); |
9222be18 GN |
5609 | break; |
5610 | default: | |
5611 | break; | |
5612 | } | |
5613 | } | |
5614 | ||
b463a6f7 AK |
5615 | static void svm_cancel_injection(struct kvm_vcpu *vcpu) |
5616 | { | |
5617 | struct vcpu_svm *svm = to_svm(vcpu); | |
5618 | struct vmcb_control_area *control = &svm->vmcb->control; | |
5619 | ||
5620 | control->exit_int_info = control->event_inj; | |
5621 | control->exit_int_info_err = control->event_inj_err; | |
5622 | control->event_inj = 0; | |
5623 | svm_complete_interrupts(svm); | |
5624 | } | |
5625 | ||
851ba692 | 5626 | static void svm_vcpu_run(struct kvm_vcpu *vcpu) |
6aa8b732 | 5627 | { |
a2fa3e9f | 5628 | struct vcpu_svm *svm = to_svm(vcpu); |
d9e368d6 | 5629 | |
2041a06a JR |
5630 | svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX]; |
5631 | svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP]; | |
5632 | svm->vmcb->save.rip = vcpu->arch.regs[VCPU_REGS_RIP]; | |
5633 | ||
cd3ff653 JR |
5634 | /* |
5635 | * A vmexit emulation is required before the vcpu can be executed | |
5636 | * again. | |
5637 | */ | |
5638 | if (unlikely(svm->nested.exit_required)) | |
5639 | return; | |
5640 | ||
a12713c2 LP |
5641 | /* |
5642 | * Disable singlestep if we're injecting an interrupt/exception. | |
5643 | * We don't want our modified rflags to be pushed on the stack where | |
5644 | * we might not be able to easily reset them if we disabled NMI | |
5645 | * singlestep later. | |
5646 | */ | |
5647 | if (svm->nmi_singlestep && svm->vmcb->control.event_inj) { | |
5648 | /* | |
5649 | * Event injection happens before external interrupts cause a | |
5650 | * vmexit and interrupts are disabled here, so smp_send_reschedule | |
5651 | * is enough to force an immediate vmexit. | |
5652 | */ | |
5653 | disable_nmi_singlestep(svm); | |
5654 | smp_send_reschedule(vcpu->cpu); | |
5655 | } | |
5656 | ||
e756fc62 | 5657 | pre_svm_run(svm); |
6aa8b732 | 5658 | |
649d6864 JR |
5659 | sync_lapic_to_cr8(vcpu); |
5660 | ||
cda0ffdd | 5661 | svm->vmcb->save.cr2 = vcpu->arch.cr2; |
6aa8b732 | 5662 | |
04d2cc77 | 5663 | clgi(); |
1811d979 | 5664 | kvm_load_guest_xcr0(vcpu); |
04d2cc77 | 5665 | |
b6c4bc65 WL |
5666 | if (lapic_in_kernel(vcpu) && |
5667 | vcpu->arch.apic->lapic_timer.timer_advance_ns) | |
5668 | kvm_wait_lapic_expire(vcpu); | |
5669 | ||
b2ac58f9 KA |
5670 | /* |
5671 | * If this vCPU has touched SPEC_CTRL, restore the guest's value if | |
5672 | * it's non-zero. Since vmentry is serialising on affected CPUs, there | |
5673 | * is no need to worry about the conditional branch over the wrmsr | |
5674 | * being speculatively taken. | |
5675 | */ | |
ccbcd267 | 5676 | x86_spec_ctrl_set_guest(svm->spec_ctrl, svm->virt_spec_ctrl); |
b2ac58f9 | 5677 | |
024d83ca TG |
5678 | local_irq_enable(); |
5679 | ||
6aa8b732 | 5680 | asm volatile ( |
7454766f AK |
5681 | "push %%" _ASM_BP "; \n\t" |
5682 | "mov %c[rbx](%[svm]), %%" _ASM_BX " \n\t" | |
5683 | "mov %c[rcx](%[svm]), %%" _ASM_CX " \n\t" | |
5684 | "mov %c[rdx](%[svm]), %%" _ASM_DX " \n\t" | |
5685 | "mov %c[rsi](%[svm]), %%" _ASM_SI " \n\t" | |
5686 | "mov %c[rdi](%[svm]), %%" _ASM_DI " \n\t" | |
5687 | "mov %c[rbp](%[svm]), %%" _ASM_BP " \n\t" | |
05b3e0c2 | 5688 | #ifdef CONFIG_X86_64 |
fb3f0f51 RR |
5689 | "mov %c[r8](%[svm]), %%r8 \n\t" |
5690 | "mov %c[r9](%[svm]), %%r9 \n\t" | |
5691 | "mov %c[r10](%[svm]), %%r10 \n\t" | |
5692 | "mov %c[r11](%[svm]), %%r11 \n\t" | |
5693 | "mov %c[r12](%[svm]), %%r12 \n\t" | |
5694 | "mov %c[r13](%[svm]), %%r13 \n\t" | |
5695 | "mov %c[r14](%[svm]), %%r14 \n\t" | |
5696 | "mov %c[r15](%[svm]), %%r15 \n\t" | |
6aa8b732 AK |
5697 | #endif |
5698 | ||
6aa8b732 | 5699 | /* Enter guest mode */ |
7454766f AK |
5700 | "push %%" _ASM_AX " \n\t" |
5701 | "mov %c[vmcb](%[svm]), %%" _ASM_AX " \n\t" | |
ac5ffda2 UB |
5702 | __ex("vmload %%" _ASM_AX) "\n\t" |
5703 | __ex("vmrun %%" _ASM_AX) "\n\t" | |
5704 | __ex("vmsave %%" _ASM_AX) "\n\t" | |
7454766f | 5705 | "pop %%" _ASM_AX " \n\t" |
6aa8b732 AK |
5706 | |
5707 | /* Save guest registers, load host registers */ | |
7454766f AK |
5708 | "mov %%" _ASM_BX ", %c[rbx](%[svm]) \n\t" |
5709 | "mov %%" _ASM_CX ", %c[rcx](%[svm]) \n\t" | |
5710 | "mov %%" _ASM_DX ", %c[rdx](%[svm]) \n\t" | |
5711 | "mov %%" _ASM_SI ", %c[rsi](%[svm]) \n\t" | |
5712 | "mov %%" _ASM_DI ", %c[rdi](%[svm]) \n\t" | |
5713 | "mov %%" _ASM_BP ", %c[rbp](%[svm]) \n\t" | |
05b3e0c2 | 5714 | #ifdef CONFIG_X86_64 |
fb3f0f51 RR |
5715 | "mov %%r8, %c[r8](%[svm]) \n\t" |
5716 | "mov %%r9, %c[r9](%[svm]) \n\t" | |
5717 | "mov %%r10, %c[r10](%[svm]) \n\t" | |
5718 | "mov %%r11, %c[r11](%[svm]) \n\t" | |
5719 | "mov %%r12, %c[r12](%[svm]) \n\t" | |
5720 | "mov %%r13, %c[r13](%[svm]) \n\t" | |
5721 | "mov %%r14, %c[r14](%[svm]) \n\t" | |
5722 | "mov %%r15, %c[r15](%[svm]) \n\t" | |
0cb5b306 JM |
5723 | /* |
5724 | * Clear host registers marked as clobbered to prevent | |
5725 | * speculative use. | |
5726 | */ | |
43ce76ce UB |
5727 | "xor %%r8d, %%r8d \n\t" |
5728 | "xor %%r9d, %%r9d \n\t" | |
5729 | "xor %%r10d, %%r10d \n\t" | |
5730 | "xor %%r11d, %%r11d \n\t" | |
5731 | "xor %%r12d, %%r12d \n\t" | |
5732 | "xor %%r13d, %%r13d \n\t" | |
5733 | "xor %%r14d, %%r14d \n\t" | |
5734 | "xor %%r15d, %%r15d \n\t" | |
6aa8b732 | 5735 | #endif |
43ce76ce UB |
5736 | "xor %%ebx, %%ebx \n\t" |
5737 | "xor %%ecx, %%ecx \n\t" | |
5738 | "xor %%edx, %%edx \n\t" | |
5739 | "xor %%esi, %%esi \n\t" | |
5740 | "xor %%edi, %%edi \n\t" | |
7454766f | 5741 | "pop %%" _ASM_BP |
6aa8b732 | 5742 | : |
fb3f0f51 | 5743 | : [svm]"a"(svm), |
6aa8b732 | 5744 | [vmcb]"i"(offsetof(struct vcpu_svm, vmcb_pa)), |
ad312c7c ZX |
5745 | [rbx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBX])), |
5746 | [rcx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RCX])), | |
5747 | [rdx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDX])), | |
5748 | [rsi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RSI])), | |
5749 | [rdi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDI])), | |
5750 | [rbp]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBP])) | |
05b3e0c2 | 5751 | #ifdef CONFIG_X86_64 |
ad312c7c ZX |
5752 | , [r8]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R8])), |
5753 | [r9]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R9])), | |
5754 | [r10]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R10])), | |
5755 | [r11]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R11])), | |
5756 | [r12]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R12])), | |
5757 | [r13]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R13])), | |
5758 | [r14]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R14])), | |
5759 | [r15]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R15])) | |
6aa8b732 | 5760 | #endif |
54a08c04 LV |
5761 | : "cc", "memory" |
5762 | #ifdef CONFIG_X86_64 | |
7454766f | 5763 | , "rbx", "rcx", "rdx", "rsi", "rdi" |
54a08c04 | 5764 | , "r8", "r9", "r10", "r11" , "r12", "r13", "r14", "r15" |
7454766f AK |
5765 | #else |
5766 | , "ebx", "ecx", "edx", "esi", "edi" | |
54a08c04 LV |
5767 | #endif |
5768 | ); | |
6aa8b732 | 5769 | |
15e6c22f TG |
5770 | /* Eliminate branch target predictions from guest mode */ |
5771 | vmexit_fill_RSB(); | |
5772 | ||
5773 | #ifdef CONFIG_X86_64 | |
5774 | wrmsrl(MSR_GS_BASE, svm->host.gs_base); | |
5775 | #else | |
5776 | loadsegment(fs, svm->host.fs); | |
5777 | #ifndef CONFIG_X86_32_LAZY_GS | |
5778 | loadsegment(gs, svm->host.gs); | |
5779 | #endif | |
5780 | #endif | |
5781 | ||
b2ac58f9 KA |
5782 | /* |
5783 | * We do not use IBRS in the kernel. If this vCPU has used the | |
5784 | * SPEC_CTRL MSR it may have left it on; save the value and | |
5785 | * turn it off. This is much more efficient than blindly adding | |
5786 | * it to the atomic save/restore list. Especially as the former | |
5787 | * (Saving guest MSRs on vmexit) doesn't even exist in KVM. | |
5788 | * | |
5789 | * For non-nested case: | |
5790 | * If the L01 MSR bitmap does not intercept the MSR, then we need to | |
5791 | * save it. | |
5792 | * | |
5793 | * For nested case: | |
5794 | * If the L02 MSR bitmap does not intercept the MSR, then we need to | |
5795 | * save it. | |
5796 | */ | |
946fbbc1 | 5797 | if (unlikely(!msr_write_intercepted(vcpu, MSR_IA32_SPEC_CTRL))) |
ecb586bd | 5798 | svm->spec_ctrl = native_read_msr(MSR_IA32_SPEC_CTRL); |
b2ac58f9 | 5799 | |
6aa8b732 AK |
5800 | reload_tss(vcpu); |
5801 | ||
56ba47dd AK |
5802 | local_irq_disable(); |
5803 | ||
024d83ca TG |
5804 | x86_spec_ctrl_restore_host(svm->spec_ctrl, svm->virt_spec_ctrl); |
5805 | ||
13c34e07 AK |
5806 | vcpu->arch.cr2 = svm->vmcb->save.cr2; |
5807 | vcpu->arch.regs[VCPU_REGS_RAX] = svm->vmcb->save.rax; | |
5808 | vcpu->arch.regs[VCPU_REGS_RSP] = svm->vmcb->save.rsp; | |
5809 | vcpu->arch.regs[VCPU_REGS_RIP] = svm->vmcb->save.rip; | |
5810 | ||
3781c01c | 5811 | if (unlikely(svm->vmcb->control.exit_code == SVM_EXIT_NMI)) |
dd60d217 | 5812 | kvm_before_interrupt(&svm->vcpu); |
3781c01c | 5813 | |
1811d979 | 5814 | kvm_put_guest_xcr0(vcpu); |
3781c01c JR |
5815 | stgi(); |
5816 | ||
5817 | /* Any pending NMI will happen here */ | |
5818 | ||
5819 | if (unlikely(svm->vmcb->control.exit_code == SVM_EXIT_NMI)) | |
dd60d217 | 5820 | kvm_after_interrupt(&svm->vcpu); |
3781c01c | 5821 | |
d7bf8221 JR |
5822 | sync_cr8_to_lapic(vcpu); |
5823 | ||
a2fa3e9f | 5824 | svm->next_rip = 0; |
9222be18 | 5825 | |
38e5e92f JR |
5826 | svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING; |
5827 | ||
631bc487 GN |
5828 | /* if exit due to PF check for async PF */ |
5829 | if (svm->vmcb->control.exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR) | |
1261bfa3 | 5830 | svm->vcpu.arch.apf.host_apf_reason = kvm_read_and_reset_pf_reason(); |
631bc487 | 5831 | |
6de4f3ad AK |
5832 | if (npt_enabled) { |
5833 | vcpu->arch.regs_avail &= ~(1 << VCPU_EXREG_PDPTR); | |
5834 | vcpu->arch.regs_dirty &= ~(1 << VCPU_EXREG_PDPTR); | |
5835 | } | |
fe5913e4 JR |
5836 | |
5837 | /* | |
5838 | * We need to handle MC intercepts here before the vcpu has a chance to | |
5839 | * change the physical cpu | |
5840 | */ | |
5841 | if (unlikely(svm->vmcb->control.exit_code == | |
5842 | SVM_EXIT_EXCP_BASE + MC_VECTOR)) | |
5843 | svm_handle_mce(svm); | |
8d28fec4 RJ |
5844 | |
5845 | mark_all_clean(svm->vmcb); | |
6aa8b732 | 5846 | } |
c207aee4 | 5847 | STACK_FRAME_NON_STANDARD(svm_vcpu_run); |
6aa8b732 | 5848 | |
6aa8b732 AK |
5849 | static void svm_set_cr3(struct kvm_vcpu *vcpu, unsigned long root) |
5850 | { | |
a2fa3e9f GH |
5851 | struct vcpu_svm *svm = to_svm(vcpu); |
5852 | ||
d0ec49d4 | 5853 | svm->vmcb->save.cr3 = __sme_set(root); |
dcca1a65 | 5854 | mark_dirty(svm->vmcb, VMCB_CR); |
6aa8b732 AK |
5855 | } |
5856 | ||
1c97f0a0 JR |
5857 | static void set_tdp_cr3(struct kvm_vcpu *vcpu, unsigned long root) |
5858 | { | |
5859 | struct vcpu_svm *svm = to_svm(vcpu); | |
5860 | ||
d0ec49d4 | 5861 | svm->vmcb->control.nested_cr3 = __sme_set(root); |
b2747166 | 5862 | mark_dirty(svm->vmcb, VMCB_NPT); |
1c97f0a0 JR |
5863 | |
5864 | /* Also sync guest cr3 here in case we live migrate */ | |
9f8fe504 | 5865 | svm->vmcb->save.cr3 = kvm_read_cr3(vcpu); |
dcca1a65 | 5866 | mark_dirty(svm->vmcb, VMCB_CR); |
1c97f0a0 JR |
5867 | } |
5868 | ||
6aa8b732 AK |
5869 | static int is_disabled(void) |
5870 | { | |
6031a61c JR |
5871 | u64 vm_cr; |
5872 | ||
5873 | rdmsrl(MSR_VM_CR, vm_cr); | |
5874 | if (vm_cr & (1 << SVM_VM_CR_SVM_DISABLE)) | |
5875 | return 1; | |
5876 | ||
6aa8b732 AK |
5877 | return 0; |
5878 | } | |
5879 | ||
102d8325 IM |
5880 | static void |
5881 | svm_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall) | |
5882 | { | |
5883 | /* | |
5884 | * Patch in the VMMCALL instruction: | |
5885 | */ | |
5886 | hypercall[0] = 0x0f; | |
5887 | hypercall[1] = 0x01; | |
5888 | hypercall[2] = 0xd9; | |
102d8325 IM |
5889 | } |
5890 | ||
f257d6dc | 5891 | static int __init svm_check_processor_compat(void) |
002c7f7c | 5892 | { |
f257d6dc | 5893 | return 0; |
002c7f7c YS |
5894 | } |
5895 | ||
774ead3a AK |
5896 | static bool svm_cpu_has_accelerated_tpr(void) |
5897 | { | |
5898 | return false; | |
5899 | } | |
5900 | ||
bc226f07 | 5901 | static bool svm_has_emulated_msr(int index) |
6d396b55 | 5902 | { |
e87555e5 VK |
5903 | switch (index) { |
5904 | case MSR_IA32_MCG_EXT_CTL: | |
95c5c7c7 | 5905 | case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC: |
e87555e5 VK |
5906 | return false; |
5907 | default: | |
5908 | break; | |
5909 | } | |
5910 | ||
6d396b55 PB |
5911 | return true; |
5912 | } | |
5913 | ||
fc07e76a PB |
5914 | static u64 svm_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio) |
5915 | { | |
5916 | return 0; | |
5917 | } | |
5918 | ||
0e851880 SY |
5919 | static void svm_cpuid_update(struct kvm_vcpu *vcpu) |
5920 | { | |
6092d3d3 JR |
5921 | struct vcpu_svm *svm = to_svm(vcpu); |
5922 | ||
5923 | /* Update nrips enabled cache */ | |
d6321d49 | 5924 | svm->nrips_enabled = !!guest_cpuid_has(&svm->vcpu, X86_FEATURE_NRIPS); |
46781eae SS |
5925 | |
5926 | if (!kvm_vcpu_apicv_active(vcpu)) | |
5927 | return; | |
5928 | ||
1b4d56b8 | 5929 | guest_cpuid_clear(vcpu, X86_FEATURE_X2APIC); |
0e851880 SY |
5930 | } |
5931 | ||
d4330ef2 JR |
5932 | static void svm_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry) |
5933 | { | |
c2c63a49 | 5934 | switch (func) { |
46781eae SS |
5935 | case 0x1: |
5936 | if (avic) | |
5937 | entry->ecx &= ~bit(X86_FEATURE_X2APIC); | |
5938 | break; | |
4c62a2dc JR |
5939 | case 0x80000001: |
5940 | if (nested) | |
5941 | entry->ecx |= (1 << 2); /* Set SVM bit */ | |
5942 | break; | |
c2c63a49 JR |
5943 | case 0x8000000A: |
5944 | entry->eax = 1; /* SVM revision 1 */ | |
5945 | entry->ebx = 8; /* Lets support 8 ASIDs in case we add proper | |
5946 | ASID emulation to nested SVM */ | |
5947 | entry->ecx = 0; /* Reserved */ | |
7a190667 JR |
5948 | entry->edx = 0; /* Per default do not support any |
5949 | additional features */ | |
5950 | ||
5951 | /* Support next_rip if host supports it */ | |
2a6b20b8 | 5952 | if (boot_cpu_has(X86_FEATURE_NRIPS)) |
7a190667 | 5953 | entry->edx |= SVM_FEATURE_NRIP; |
c2c63a49 | 5954 | |
3d4aeaad JR |
5955 | /* Support NPT for the guest if enabled */ |
5956 | if (npt_enabled) | |
5957 | entry->edx |= SVM_FEATURE_NPT; | |
5958 | ||
c2c63a49 | 5959 | break; |
8765d753 BS |
5960 | case 0x8000001F: |
5961 | /* Support memory encryption cpuid if host supports it */ | |
5962 | if (boot_cpu_has(X86_FEATURE_SEV)) | |
5963 | cpuid(0x8000001f, &entry->eax, &entry->ebx, | |
5964 | &entry->ecx, &entry->edx); | |
5965 | ||
c2c63a49 | 5966 | } |
d4330ef2 JR |
5967 | } |
5968 | ||
17cc3935 | 5969 | static int svm_get_lpage_level(void) |
344f414f | 5970 | { |
17cc3935 | 5971 | return PT_PDPE_LEVEL; |
344f414f JR |
5972 | } |
5973 | ||
4e47c7a6 SY |
5974 | static bool svm_rdtscp_supported(void) |
5975 | { | |
46896c73 | 5976 | return boot_cpu_has(X86_FEATURE_RDTSCP); |
4e47c7a6 SY |
5977 | } |
5978 | ||
ad756a16 MJ |
5979 | static bool svm_invpcid_supported(void) |
5980 | { | |
5981 | return false; | |
5982 | } | |
5983 | ||
93c4adc7 PB |
5984 | static bool svm_mpx_supported(void) |
5985 | { | |
5986 | return false; | |
5987 | } | |
5988 | ||
55412b2e WL |
5989 | static bool svm_xsaves_supported(void) |
5990 | { | |
5991 | return false; | |
5992 | } | |
5993 | ||
66336cab PB |
5994 | static bool svm_umip_emulated(void) |
5995 | { | |
5996 | return false; | |
5997 | } | |
5998 | ||
86f5201d CP |
5999 | static bool svm_pt_supported(void) |
6000 | { | |
6001 | return false; | |
6002 | } | |
6003 | ||
f5f48ee1 SY |
6004 | static bool svm_has_wbinvd_exit(void) |
6005 | { | |
6006 | return true; | |
6007 | } | |
6008 | ||
8061252e | 6009 | #define PRE_EX(exit) { .exit_code = (exit), \ |
40e19b51 | 6010 | .stage = X86_ICPT_PRE_EXCEPT, } |
cfec82cb | 6011 | #define POST_EX(exit) { .exit_code = (exit), \ |
40e19b51 | 6012 | .stage = X86_ICPT_POST_EXCEPT, } |
d7eb8203 | 6013 | #define POST_MEM(exit) { .exit_code = (exit), \ |
40e19b51 | 6014 | .stage = X86_ICPT_POST_MEMACCESS, } |
cfec82cb | 6015 | |
09941fbb | 6016 | static const struct __x86_intercept { |
cfec82cb JR |
6017 | u32 exit_code; |
6018 | enum x86_intercept_stage stage; | |
cfec82cb JR |
6019 | } x86_intercept_map[] = { |
6020 | [x86_intercept_cr_read] = POST_EX(SVM_EXIT_READ_CR0), | |
6021 | [x86_intercept_cr_write] = POST_EX(SVM_EXIT_WRITE_CR0), | |
6022 | [x86_intercept_clts] = POST_EX(SVM_EXIT_WRITE_CR0), | |
6023 | [x86_intercept_lmsw] = POST_EX(SVM_EXIT_WRITE_CR0), | |
6024 | [x86_intercept_smsw] = POST_EX(SVM_EXIT_READ_CR0), | |
3b88e41a JR |
6025 | [x86_intercept_dr_read] = POST_EX(SVM_EXIT_READ_DR0), |
6026 | [x86_intercept_dr_write] = POST_EX(SVM_EXIT_WRITE_DR0), | |
dee6bb70 JR |
6027 | [x86_intercept_sldt] = POST_EX(SVM_EXIT_LDTR_READ), |
6028 | [x86_intercept_str] = POST_EX(SVM_EXIT_TR_READ), | |
6029 | [x86_intercept_lldt] = POST_EX(SVM_EXIT_LDTR_WRITE), | |
6030 | [x86_intercept_ltr] = POST_EX(SVM_EXIT_TR_WRITE), | |
6031 | [x86_intercept_sgdt] = POST_EX(SVM_EXIT_GDTR_READ), | |
6032 | [x86_intercept_sidt] = POST_EX(SVM_EXIT_IDTR_READ), | |
6033 | [x86_intercept_lgdt] = POST_EX(SVM_EXIT_GDTR_WRITE), | |
6034 | [x86_intercept_lidt] = POST_EX(SVM_EXIT_IDTR_WRITE), | |
01de8b09 JR |
6035 | [x86_intercept_vmrun] = POST_EX(SVM_EXIT_VMRUN), |
6036 | [x86_intercept_vmmcall] = POST_EX(SVM_EXIT_VMMCALL), | |
6037 | [x86_intercept_vmload] = POST_EX(SVM_EXIT_VMLOAD), | |
6038 | [x86_intercept_vmsave] = POST_EX(SVM_EXIT_VMSAVE), | |
6039 | [x86_intercept_stgi] = POST_EX(SVM_EXIT_STGI), | |
6040 | [x86_intercept_clgi] = POST_EX(SVM_EXIT_CLGI), | |
6041 | [x86_intercept_skinit] = POST_EX(SVM_EXIT_SKINIT), | |
6042 | [x86_intercept_invlpga] = POST_EX(SVM_EXIT_INVLPGA), | |
d7eb8203 JR |
6043 | [x86_intercept_rdtscp] = POST_EX(SVM_EXIT_RDTSCP), |
6044 | [x86_intercept_monitor] = POST_MEM(SVM_EXIT_MONITOR), | |
6045 | [x86_intercept_mwait] = POST_EX(SVM_EXIT_MWAIT), | |
8061252e JR |
6046 | [x86_intercept_invlpg] = POST_EX(SVM_EXIT_INVLPG), |
6047 | [x86_intercept_invd] = POST_EX(SVM_EXIT_INVD), | |
6048 | [x86_intercept_wbinvd] = POST_EX(SVM_EXIT_WBINVD), | |
6049 | [x86_intercept_wrmsr] = POST_EX(SVM_EXIT_MSR), | |
6050 | [x86_intercept_rdtsc] = POST_EX(SVM_EXIT_RDTSC), | |
6051 | [x86_intercept_rdmsr] = POST_EX(SVM_EXIT_MSR), | |
6052 | [x86_intercept_rdpmc] = POST_EX(SVM_EXIT_RDPMC), | |
6053 | [x86_intercept_cpuid] = PRE_EX(SVM_EXIT_CPUID), | |
6054 | [x86_intercept_rsm] = PRE_EX(SVM_EXIT_RSM), | |
bf608f88 JR |
6055 | [x86_intercept_pause] = PRE_EX(SVM_EXIT_PAUSE), |
6056 | [x86_intercept_pushf] = PRE_EX(SVM_EXIT_PUSHF), | |
6057 | [x86_intercept_popf] = PRE_EX(SVM_EXIT_POPF), | |
6058 | [x86_intercept_intn] = PRE_EX(SVM_EXIT_SWINT), | |
6059 | [x86_intercept_iret] = PRE_EX(SVM_EXIT_IRET), | |
6060 | [x86_intercept_icebp] = PRE_EX(SVM_EXIT_ICEBP), | |
6061 | [x86_intercept_hlt] = POST_EX(SVM_EXIT_HLT), | |
f6511935 JR |
6062 | [x86_intercept_in] = POST_EX(SVM_EXIT_IOIO), |
6063 | [x86_intercept_ins] = POST_EX(SVM_EXIT_IOIO), | |
6064 | [x86_intercept_out] = POST_EX(SVM_EXIT_IOIO), | |
6065 | [x86_intercept_outs] = POST_EX(SVM_EXIT_IOIO), | |
cfec82cb JR |
6066 | }; |
6067 | ||
8061252e | 6068 | #undef PRE_EX |
cfec82cb | 6069 | #undef POST_EX |
d7eb8203 | 6070 | #undef POST_MEM |
cfec82cb | 6071 | |
8a76d7f2 JR |
6072 | static int svm_check_intercept(struct kvm_vcpu *vcpu, |
6073 | struct x86_instruction_info *info, | |
6074 | enum x86_intercept_stage stage) | |
6075 | { | |
cfec82cb JR |
6076 | struct vcpu_svm *svm = to_svm(vcpu); |
6077 | int vmexit, ret = X86EMUL_CONTINUE; | |
6078 | struct __x86_intercept icpt_info; | |
6079 | struct vmcb *vmcb = svm->vmcb; | |
6080 | ||
6081 | if (info->intercept >= ARRAY_SIZE(x86_intercept_map)) | |
6082 | goto out; | |
6083 | ||
6084 | icpt_info = x86_intercept_map[info->intercept]; | |
6085 | ||
40e19b51 | 6086 | if (stage != icpt_info.stage) |
cfec82cb JR |
6087 | goto out; |
6088 | ||
6089 | switch (icpt_info.exit_code) { | |
6090 | case SVM_EXIT_READ_CR0: | |
6091 | if (info->intercept == x86_intercept_cr_read) | |
6092 | icpt_info.exit_code += info->modrm_reg; | |
6093 | break; | |
6094 | case SVM_EXIT_WRITE_CR0: { | |
6095 | unsigned long cr0, val; | |
6096 | u64 intercept; | |
6097 | ||
6098 | if (info->intercept == x86_intercept_cr_write) | |
6099 | icpt_info.exit_code += info->modrm_reg; | |
6100 | ||
62baf44c JK |
6101 | if (icpt_info.exit_code != SVM_EXIT_WRITE_CR0 || |
6102 | info->intercept == x86_intercept_clts) | |
cfec82cb JR |
6103 | break; |
6104 | ||
6105 | intercept = svm->nested.intercept; | |
6106 | ||
6107 | if (!(intercept & (1ULL << INTERCEPT_SELECTIVE_CR0))) | |
6108 | break; | |
6109 | ||
6110 | cr0 = vcpu->arch.cr0 & ~SVM_CR0_SELECTIVE_MASK; | |
6111 | val = info->src_val & ~SVM_CR0_SELECTIVE_MASK; | |
6112 | ||
6113 | if (info->intercept == x86_intercept_lmsw) { | |
6114 | cr0 &= 0xfUL; | |
6115 | val &= 0xfUL; | |
6116 | /* lmsw can't clear PE - catch this here */ | |
6117 | if (cr0 & X86_CR0_PE) | |
6118 | val |= X86_CR0_PE; | |
6119 | } | |
6120 | ||
6121 | if (cr0 ^ val) | |
6122 | icpt_info.exit_code = SVM_EXIT_CR0_SEL_WRITE; | |
6123 | ||
6124 | break; | |
6125 | } | |
3b88e41a JR |
6126 | case SVM_EXIT_READ_DR0: |
6127 | case SVM_EXIT_WRITE_DR0: | |
6128 | icpt_info.exit_code += info->modrm_reg; | |
6129 | break; | |
8061252e JR |
6130 | case SVM_EXIT_MSR: |
6131 | if (info->intercept == x86_intercept_wrmsr) | |
6132 | vmcb->control.exit_info_1 = 1; | |
6133 | else | |
6134 | vmcb->control.exit_info_1 = 0; | |
6135 | break; | |
bf608f88 JR |
6136 | case SVM_EXIT_PAUSE: |
6137 | /* | |
6138 | * We get this for NOP only, but pause | |
6139 | * is rep not, check this here | |
6140 | */ | |
6141 | if (info->rep_prefix != REPE_PREFIX) | |
6142 | goto out; | |
49a8afca | 6143 | break; |
f6511935 JR |
6144 | case SVM_EXIT_IOIO: { |
6145 | u64 exit_info; | |
6146 | u32 bytes; | |
6147 | ||
f6511935 JR |
6148 | if (info->intercept == x86_intercept_in || |
6149 | info->intercept == x86_intercept_ins) { | |
6cbc5f5a JK |
6150 | exit_info = ((info->src_val & 0xffff) << 16) | |
6151 | SVM_IOIO_TYPE_MASK; | |
f6511935 | 6152 | bytes = info->dst_bytes; |
6493f157 | 6153 | } else { |
6cbc5f5a | 6154 | exit_info = (info->dst_val & 0xffff) << 16; |
6493f157 | 6155 | bytes = info->src_bytes; |
f6511935 JR |
6156 | } |
6157 | ||
6158 | if (info->intercept == x86_intercept_outs || | |
6159 | info->intercept == x86_intercept_ins) | |
6160 | exit_info |= SVM_IOIO_STR_MASK; | |
6161 | ||
6162 | if (info->rep_prefix) | |
6163 | exit_info |= SVM_IOIO_REP_MASK; | |
6164 | ||
6165 | bytes = min(bytes, 4u); | |
6166 | ||
6167 | exit_info |= bytes << SVM_IOIO_SIZE_SHIFT; | |
6168 | ||
6169 | exit_info |= (u32)info->ad_bytes << (SVM_IOIO_ASIZE_SHIFT - 1); | |
6170 | ||
6171 | vmcb->control.exit_info_1 = exit_info; | |
6172 | vmcb->control.exit_info_2 = info->next_rip; | |
6173 | ||
6174 | break; | |
6175 | } | |
cfec82cb JR |
6176 | default: |
6177 | break; | |
6178 | } | |
6179 | ||
f104765b BD |
6180 | /* TODO: Advertise NRIPS to guest hypervisor unconditionally */ |
6181 | if (static_cpu_has(X86_FEATURE_NRIPS)) | |
6182 | vmcb->control.next_rip = info->next_rip; | |
cfec82cb JR |
6183 | vmcb->control.exit_code = icpt_info.exit_code; |
6184 | vmexit = nested_svm_exit_handled(svm); | |
6185 | ||
6186 | ret = (vmexit == NESTED_EXIT_DONE) ? X86EMUL_INTERCEPTED | |
6187 | : X86EMUL_CONTINUE; | |
6188 | ||
6189 | out: | |
6190 | return ret; | |
8a76d7f2 JR |
6191 | } |
6192 | ||
95b5a48c | 6193 | static void svm_handle_exit_irqoff(struct kvm_vcpu *vcpu) |
a547c6db | 6194 | { |
d7a08882 | 6195 | |
a547c6db YZ |
6196 | } |
6197 | ||
ae97a3b8 RK |
6198 | static void svm_sched_in(struct kvm_vcpu *vcpu, int cpu) |
6199 | { | |
8566ac8b BM |
6200 | if (pause_filter_thresh) |
6201 | shrink_ple_window(vcpu); | |
ae97a3b8 RK |
6202 | } |
6203 | ||
be8ca170 SS |
6204 | static inline void avic_post_state_restore(struct kvm_vcpu *vcpu) |
6205 | { | |
6206 | if (avic_handle_apic_id_update(vcpu) != 0) | |
6207 | return; | |
98d90582 | 6208 | avic_handle_dfr_update(vcpu); |
be8ca170 SS |
6209 | avic_handle_ldr_update(vcpu); |
6210 | } | |
6211 | ||
74f16909 BP |
6212 | static void svm_setup_mce(struct kvm_vcpu *vcpu) |
6213 | { | |
6214 | /* [63:9] are reserved. */ | |
6215 | vcpu->arch.mcg_cap &= 0x1ff; | |
6216 | } | |
6217 | ||
72d7b374 LP |
6218 | static int svm_smi_allowed(struct kvm_vcpu *vcpu) |
6219 | { | |
05cade71 LP |
6220 | struct vcpu_svm *svm = to_svm(vcpu); |
6221 | ||
6222 | /* Per APM Vol.2 15.22.2 "Response to SMI" */ | |
6223 | if (!gif_set(svm)) | |
6224 | return 0; | |
6225 | ||
6226 | if (is_guest_mode(&svm->vcpu) && | |
6227 | svm->nested.intercept & (1ULL << INTERCEPT_SMI)) { | |
6228 | /* TODO: Might need to set exit_info_1 and exit_info_2 here */ | |
6229 | svm->vmcb->control.exit_code = SVM_EXIT_SMI; | |
6230 | svm->nested.exit_required = true; | |
6231 | return 0; | |
6232 | } | |
6233 | ||
72d7b374 LP |
6234 | return 1; |
6235 | } | |
6236 | ||
0234bf88 LP |
6237 | static int svm_pre_enter_smm(struct kvm_vcpu *vcpu, char *smstate) |
6238 | { | |
05cade71 LP |
6239 | struct vcpu_svm *svm = to_svm(vcpu); |
6240 | int ret; | |
6241 | ||
6242 | if (is_guest_mode(vcpu)) { | |
6243 | /* FED8h - SVM Guest */ | |
6244 | put_smstate(u64, smstate, 0x7ed8, 1); | |
6245 | /* FEE0h - SVM Guest VMCB Physical Address */ | |
6246 | put_smstate(u64, smstate, 0x7ee0, svm->nested.vmcb); | |
6247 | ||
6248 | svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX]; | |
6249 | svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP]; | |
6250 | svm->vmcb->save.rip = vcpu->arch.regs[VCPU_REGS_RIP]; | |
6251 | ||
6252 | ret = nested_svm_vmexit(svm); | |
6253 | if (ret) | |
6254 | return ret; | |
6255 | } | |
0234bf88 LP |
6256 | return 0; |
6257 | } | |
6258 | ||
ed19321f | 6259 | static int svm_pre_leave_smm(struct kvm_vcpu *vcpu, const char *smstate) |
0234bf88 | 6260 | { |
05cade71 LP |
6261 | struct vcpu_svm *svm = to_svm(vcpu); |
6262 | struct vmcb *nested_vmcb; | |
8c5fbf1a | 6263 | struct kvm_host_map map; |
ed19321f SC |
6264 | u64 guest; |
6265 | u64 vmcb; | |
05cade71 | 6266 | |
ed19321f SC |
6267 | guest = GET_SMSTATE(u64, smstate, 0x7ed8); |
6268 | vmcb = GET_SMSTATE(u64, smstate, 0x7ee0); | |
05cade71 | 6269 | |
ed19321f | 6270 | if (guest) { |
8c5fbf1a | 6271 | if (kvm_vcpu_map(&svm->vcpu, gpa_to_gfn(vmcb), &map) == -EINVAL) |
9ec19493 | 6272 | return 1; |
8c5fbf1a KA |
6273 | nested_vmcb = map.hva; |
6274 | enter_svm_guest_mode(svm, vmcb, nested_vmcb, &map); | |
05cade71 | 6275 | } |
9ec19493 | 6276 | return 0; |
0234bf88 LP |
6277 | } |
6278 | ||
cc3d967f LP |
6279 | static int enable_smi_window(struct kvm_vcpu *vcpu) |
6280 | { | |
6281 | struct vcpu_svm *svm = to_svm(vcpu); | |
6282 | ||
6283 | if (!gif_set(svm)) { | |
6284 | if (vgif_enabled(svm)) | |
6285 | set_intercept(svm, INTERCEPT_STGI); | |
6286 | /* STGI will cause a vm exit */ | |
6287 | return 1; | |
6288 | } | |
6289 | return 0; | |
6290 | } | |
6291 | ||
1654efcb BS |
6292 | static int sev_asid_new(void) |
6293 | { | |
6294 | int pos; | |
6295 | ||
6296 | /* | |
6297 | * SEV-enabled guest must use asid from min_sev_asid to max_sev_asid. | |
6298 | */ | |
6299 | pos = find_next_zero_bit(sev_asid_bitmap, max_sev_asid, min_sev_asid - 1); | |
6300 | if (pos >= max_sev_asid) | |
6301 | return -EBUSY; | |
6302 | ||
6303 | set_bit(pos, sev_asid_bitmap); | |
6304 | return pos + 1; | |
6305 | } | |
6306 | ||
6307 | static int sev_guest_init(struct kvm *kvm, struct kvm_sev_cmd *argp) | |
6308 | { | |
81811c16 | 6309 | struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info; |
1654efcb BS |
6310 | int asid, ret; |
6311 | ||
6312 | ret = -EBUSY; | |
3f14a89d DR |
6313 | if (unlikely(sev->active)) |
6314 | return ret; | |
6315 | ||
1654efcb BS |
6316 | asid = sev_asid_new(); |
6317 | if (asid < 0) | |
6318 | return ret; | |
6319 | ||
6320 | ret = sev_platform_init(&argp->error); | |
6321 | if (ret) | |
6322 | goto e_free; | |
6323 | ||
6324 | sev->active = true; | |
6325 | sev->asid = asid; | |
1e80fdc0 | 6326 | INIT_LIST_HEAD(&sev->regions_list); |
1654efcb BS |
6327 | |
6328 | return 0; | |
6329 | ||
6330 | e_free: | |
6331 | __sev_asid_free(asid); | |
6332 | return ret; | |
6333 | } | |
6334 | ||
59414c98 BS |
6335 | static int sev_bind_asid(struct kvm *kvm, unsigned int handle, int *error) |
6336 | { | |
6337 | struct sev_data_activate *data; | |
6338 | int asid = sev_get_asid(kvm); | |
6339 | int ret; | |
6340 | ||
6341 | wbinvd_on_all_cpus(); | |
6342 | ||
6343 | ret = sev_guest_df_flush(error); | |
6344 | if (ret) | |
6345 | return ret; | |
6346 | ||
1ec69647 | 6347 | data = kzalloc(sizeof(*data), GFP_KERNEL_ACCOUNT); |
59414c98 BS |
6348 | if (!data) |
6349 | return -ENOMEM; | |
6350 | ||
6351 | /* activate ASID on the given handle */ | |
6352 | data->handle = handle; | |
6353 | data->asid = asid; | |
6354 | ret = sev_guest_activate(data, error); | |
6355 | kfree(data); | |
6356 | ||
6357 | return ret; | |
6358 | } | |
6359 | ||
89c50580 | 6360 | static int __sev_issue_cmd(int fd, int id, void *data, int *error) |
59414c98 BS |
6361 | { |
6362 | struct fd f; | |
6363 | int ret; | |
6364 | ||
6365 | f = fdget(fd); | |
6366 | if (!f.file) | |
6367 | return -EBADF; | |
6368 | ||
6369 | ret = sev_issue_cmd_external_user(f.file, id, data, error); | |
6370 | ||
6371 | fdput(f); | |
6372 | return ret; | |
6373 | } | |
6374 | ||
89c50580 BS |
6375 | static int sev_issue_cmd(struct kvm *kvm, int id, void *data, int *error) |
6376 | { | |
81811c16 | 6377 | struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info; |
89c50580 BS |
6378 | |
6379 | return __sev_issue_cmd(sev->fd, id, data, error); | |
6380 | } | |
6381 | ||
59414c98 BS |
6382 | static int sev_launch_start(struct kvm *kvm, struct kvm_sev_cmd *argp) |
6383 | { | |
81811c16 | 6384 | struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info; |
59414c98 BS |
6385 | struct sev_data_launch_start *start; |
6386 | struct kvm_sev_launch_start params; | |
6387 | void *dh_blob, *session_blob; | |
6388 | int *error = &argp->error; | |
6389 | int ret; | |
6390 | ||
6391 | if (!sev_guest(kvm)) | |
6392 | return -ENOTTY; | |
6393 | ||
6394 | if (copy_from_user(¶ms, (void __user *)(uintptr_t)argp->data, sizeof(params))) | |
6395 | return -EFAULT; | |
6396 | ||
1ec69647 | 6397 | start = kzalloc(sizeof(*start), GFP_KERNEL_ACCOUNT); |
59414c98 BS |
6398 | if (!start) |
6399 | return -ENOMEM; | |
6400 | ||
6401 | dh_blob = NULL; | |
6402 | if (params.dh_uaddr) { | |
6403 | dh_blob = psp_copy_user_blob(params.dh_uaddr, params.dh_len); | |
6404 | if (IS_ERR(dh_blob)) { | |
6405 | ret = PTR_ERR(dh_blob); | |
6406 | goto e_free; | |
6407 | } | |
6408 | ||
6409 | start->dh_cert_address = __sme_set(__pa(dh_blob)); | |
6410 | start->dh_cert_len = params.dh_len; | |
6411 | } | |
6412 | ||
6413 | session_blob = NULL; | |
6414 | if (params.session_uaddr) { | |
6415 | session_blob = psp_copy_user_blob(params.session_uaddr, params.session_len); | |
6416 | if (IS_ERR(session_blob)) { | |
6417 | ret = PTR_ERR(session_blob); | |
6418 | goto e_free_dh; | |
6419 | } | |
6420 | ||
6421 | start->session_address = __sme_set(__pa(session_blob)); | |
6422 | start->session_len = params.session_len; | |
6423 | } | |
6424 | ||
6425 | start->handle = params.handle; | |
6426 | start->policy = params.policy; | |
6427 | ||
6428 | /* create memory encryption context */ | |
89c50580 | 6429 | ret = __sev_issue_cmd(argp->sev_fd, SEV_CMD_LAUNCH_START, start, error); |
59414c98 BS |
6430 | if (ret) |
6431 | goto e_free_session; | |
6432 | ||
6433 | /* Bind ASID to this guest */ | |
6434 | ret = sev_bind_asid(kvm, start->handle, error); | |
6435 | if (ret) | |
6436 | goto e_free_session; | |
6437 | ||
6438 | /* return handle to userspace */ | |
6439 | params.handle = start->handle; | |
6440 | if (copy_to_user((void __user *)(uintptr_t)argp->data, ¶ms, sizeof(params))) { | |
6441 | sev_unbind_asid(kvm, start->handle); | |
6442 | ret = -EFAULT; | |
6443 | goto e_free_session; | |
6444 | } | |
6445 | ||
6446 | sev->handle = start->handle; | |
6447 | sev->fd = argp->sev_fd; | |
6448 | ||
6449 | e_free_session: | |
6450 | kfree(session_blob); | |
6451 | e_free_dh: | |
6452 | kfree(dh_blob); | |
6453 | e_free: | |
6454 | kfree(start); | |
6455 | return ret; | |
6456 | } | |
6457 | ||
ede885ec DR |
6458 | static unsigned long get_num_contig_pages(unsigned long idx, |
6459 | struct page **inpages, unsigned long npages) | |
89c50580 BS |
6460 | { |
6461 | unsigned long paddr, next_paddr; | |
ede885ec | 6462 | unsigned long i = idx + 1, pages = 1; |
89c50580 BS |
6463 | |
6464 | /* find the number of contiguous pages starting from idx */ | |
6465 | paddr = __sme_page_pa(inpages[idx]); | |
6466 | while (i < npages) { | |
6467 | next_paddr = __sme_page_pa(inpages[i++]); | |
6468 | if ((paddr + PAGE_SIZE) == next_paddr) { | |
6469 | pages++; | |
6470 | paddr = next_paddr; | |
6471 | continue; | |
6472 | } | |
6473 | break; | |
6474 | } | |
6475 | ||
6476 | return pages; | |
6477 | } | |
6478 | ||
6479 | static int sev_launch_update_data(struct kvm *kvm, struct kvm_sev_cmd *argp) | |
6480 | { | |
ede885ec | 6481 | unsigned long vaddr, vaddr_end, next_vaddr, npages, pages, size, i; |
81811c16 | 6482 | struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info; |
89c50580 BS |
6483 | struct kvm_sev_launch_update_data params; |
6484 | struct sev_data_launch_update_data *data; | |
6485 | struct page **inpages; | |
ede885ec | 6486 | int ret; |
89c50580 BS |
6487 | |
6488 | if (!sev_guest(kvm)) | |
6489 | return -ENOTTY; | |
6490 | ||
6491 | if (copy_from_user(¶ms, (void __user *)(uintptr_t)argp->data, sizeof(params))) | |
6492 | return -EFAULT; | |
6493 | ||
1ec69647 | 6494 | data = kzalloc(sizeof(*data), GFP_KERNEL_ACCOUNT); |
89c50580 BS |
6495 | if (!data) |
6496 | return -ENOMEM; | |
6497 | ||
6498 | vaddr = params.uaddr; | |
6499 | size = params.len; | |
6500 | vaddr_end = vaddr + size; | |
6501 | ||
6502 | /* Lock the user memory. */ | |
6503 | inpages = sev_pin_memory(kvm, vaddr, size, &npages, 1); | |
6504 | if (!inpages) { | |
6505 | ret = -ENOMEM; | |
6506 | goto e_free; | |
6507 | } | |
6508 | ||
6509 | /* | |
6510 | * The LAUNCH_UPDATE command will perform in-place encryption of the | |
6511 | * memory content (i.e it will write the same memory region with C=1). | |
6512 | * It's possible that the cache may contain the data with C=0, i.e., | |
6513 | * unencrypted so invalidate it first. | |
6514 | */ | |
6515 | sev_clflush_pages(inpages, npages); | |
6516 | ||
6517 | for (i = 0; vaddr < vaddr_end; vaddr = next_vaddr, i += pages) { | |
6518 | int offset, len; | |
6519 | ||
6520 | /* | |
6521 | * If the user buffer is not page-aligned, calculate the offset | |
6522 | * within the page. | |
6523 | */ | |
6524 | offset = vaddr & (PAGE_SIZE - 1); | |
6525 | ||
6526 | /* Calculate the number of pages that can be encrypted in one go. */ | |
6527 | pages = get_num_contig_pages(i, inpages, npages); | |
6528 | ||
6529 | len = min_t(size_t, ((pages * PAGE_SIZE) - offset), size); | |
6530 | ||
6531 | data->handle = sev->handle; | |
6532 | data->len = len; | |
6533 | data->address = __sme_page_pa(inpages[i]) + offset; | |
6534 | ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_UPDATE_DATA, data, &argp->error); | |
6535 | if (ret) | |
6536 | goto e_unpin; | |
6537 | ||
6538 | size -= len; | |
6539 | next_vaddr = vaddr + len; | |
6540 | } | |
6541 | ||
6542 | e_unpin: | |
6543 | /* content of memory is updated, mark pages dirty */ | |
6544 | for (i = 0; i < npages; i++) { | |
6545 | set_page_dirty_lock(inpages[i]); | |
6546 | mark_page_accessed(inpages[i]); | |
6547 | } | |
6548 | /* unlock the user pages */ | |
6549 | sev_unpin_memory(kvm, inpages, npages); | |
6550 | e_free: | |
6551 | kfree(data); | |
6552 | return ret; | |
6553 | } | |
6554 | ||
0d0736f7 BS |
6555 | static int sev_launch_measure(struct kvm *kvm, struct kvm_sev_cmd *argp) |
6556 | { | |
3e233385 | 6557 | void __user *measure = (void __user *)(uintptr_t)argp->data; |
81811c16 | 6558 | struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info; |
0d0736f7 BS |
6559 | struct sev_data_launch_measure *data; |
6560 | struct kvm_sev_launch_measure params; | |
3e233385 | 6561 | void __user *p = NULL; |
0d0736f7 BS |
6562 | void *blob = NULL; |
6563 | int ret; | |
6564 | ||
6565 | if (!sev_guest(kvm)) | |
6566 | return -ENOTTY; | |
6567 | ||
3e233385 | 6568 | if (copy_from_user(¶ms, measure, sizeof(params))) |
0d0736f7 BS |
6569 | return -EFAULT; |
6570 | ||
1ec69647 | 6571 | data = kzalloc(sizeof(*data), GFP_KERNEL_ACCOUNT); |
0d0736f7 BS |
6572 | if (!data) |
6573 | return -ENOMEM; | |
6574 | ||
6575 | /* User wants to query the blob length */ | |
6576 | if (!params.len) | |
6577 | goto cmd; | |
6578 | ||
3e233385 BS |
6579 | p = (void __user *)(uintptr_t)params.uaddr; |
6580 | if (p) { | |
0d0736f7 BS |
6581 | if (params.len > SEV_FW_BLOB_MAX_SIZE) { |
6582 | ret = -EINVAL; | |
6583 | goto e_free; | |
6584 | } | |
6585 | ||
0d0736f7 BS |
6586 | ret = -ENOMEM; |
6587 | blob = kmalloc(params.len, GFP_KERNEL); | |
6588 | if (!blob) | |
6589 | goto e_free; | |
6590 | ||
6591 | data->address = __psp_pa(blob); | |
6592 | data->len = params.len; | |
6593 | } | |
6594 | ||
6595 | cmd: | |
6596 | data->handle = sev->handle; | |
6597 | ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_MEASURE, data, &argp->error); | |
6598 | ||
6599 | /* | |
6600 | * If we query the session length, FW responded with expected data. | |
6601 | */ | |
6602 | if (!params.len) | |
6603 | goto done; | |
6604 | ||
6605 | if (ret) | |
6606 | goto e_free_blob; | |
6607 | ||
6608 | if (blob) { | |
3e233385 | 6609 | if (copy_to_user(p, blob, params.len)) |
0d0736f7 BS |
6610 | ret = -EFAULT; |
6611 | } | |
6612 | ||
6613 | done: | |
6614 | params.len = data->len; | |
3e233385 | 6615 | if (copy_to_user(measure, ¶ms, sizeof(params))) |
0d0736f7 BS |
6616 | ret = -EFAULT; |
6617 | e_free_blob: | |
6618 | kfree(blob); | |
6619 | e_free: | |
6620 | kfree(data); | |
6621 | return ret; | |
6622 | } | |
6623 | ||
5bdb0e2f BS |
6624 | static int sev_launch_finish(struct kvm *kvm, struct kvm_sev_cmd *argp) |
6625 | { | |
81811c16 | 6626 | struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info; |
5bdb0e2f BS |
6627 | struct sev_data_launch_finish *data; |
6628 | int ret; | |
6629 | ||
6630 | if (!sev_guest(kvm)) | |
6631 | return -ENOTTY; | |
6632 | ||
1ec69647 | 6633 | data = kzalloc(sizeof(*data), GFP_KERNEL_ACCOUNT); |
5bdb0e2f BS |
6634 | if (!data) |
6635 | return -ENOMEM; | |
6636 | ||
6637 | data->handle = sev->handle; | |
6638 | ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_FINISH, data, &argp->error); | |
6639 | ||
6640 | kfree(data); | |
6641 | return ret; | |
6642 | } | |
6643 | ||
255d9e75 BS |
6644 | static int sev_guest_status(struct kvm *kvm, struct kvm_sev_cmd *argp) |
6645 | { | |
81811c16 | 6646 | struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info; |
255d9e75 BS |
6647 | struct kvm_sev_guest_status params; |
6648 | struct sev_data_guest_status *data; | |
6649 | int ret; | |
6650 | ||
6651 | if (!sev_guest(kvm)) | |
6652 | return -ENOTTY; | |
6653 | ||
1ec69647 | 6654 | data = kzalloc(sizeof(*data), GFP_KERNEL_ACCOUNT); |
255d9e75 BS |
6655 | if (!data) |
6656 | return -ENOMEM; | |
6657 | ||
6658 | data->handle = sev->handle; | |
6659 | ret = sev_issue_cmd(kvm, SEV_CMD_GUEST_STATUS, data, &argp->error); | |
6660 | if (ret) | |
6661 | goto e_free; | |
6662 | ||
6663 | params.policy = data->policy; | |
6664 | params.state = data->state; | |
6665 | params.handle = data->handle; | |
6666 | ||
6667 | if (copy_to_user((void __user *)(uintptr_t)argp->data, ¶ms, sizeof(params))) | |
6668 | ret = -EFAULT; | |
6669 | e_free: | |
6670 | kfree(data); | |
6671 | return ret; | |
6672 | } | |
6673 | ||
24f41fb2 BS |
6674 | static int __sev_issue_dbg_cmd(struct kvm *kvm, unsigned long src, |
6675 | unsigned long dst, int size, | |
6676 | int *error, bool enc) | |
6677 | { | |
81811c16 | 6678 | struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info; |
24f41fb2 BS |
6679 | struct sev_data_dbg *data; |
6680 | int ret; | |
6681 | ||
1ec69647 | 6682 | data = kzalloc(sizeof(*data), GFP_KERNEL_ACCOUNT); |
24f41fb2 BS |
6683 | if (!data) |
6684 | return -ENOMEM; | |
6685 | ||
6686 | data->handle = sev->handle; | |
6687 | data->dst_addr = dst; | |
6688 | data->src_addr = src; | |
6689 | data->len = size; | |
6690 | ||
6691 | ret = sev_issue_cmd(kvm, | |
6692 | enc ? SEV_CMD_DBG_ENCRYPT : SEV_CMD_DBG_DECRYPT, | |
6693 | data, error); | |
6694 | kfree(data); | |
6695 | return ret; | |
6696 | } | |
6697 | ||
6698 | static int __sev_dbg_decrypt(struct kvm *kvm, unsigned long src_paddr, | |
6699 | unsigned long dst_paddr, int sz, int *err) | |
6700 | { | |
6701 | int offset; | |
6702 | ||
6703 | /* | |
6704 | * Its safe to read more than we are asked, caller should ensure that | |
6705 | * destination has enough space. | |
6706 | */ | |
6707 | src_paddr = round_down(src_paddr, 16); | |
6708 | offset = src_paddr & 15; | |
6709 | sz = round_up(sz + offset, 16); | |
6710 | ||
6711 | return __sev_issue_dbg_cmd(kvm, src_paddr, dst_paddr, sz, err, false); | |
6712 | } | |
6713 | ||
6714 | static int __sev_dbg_decrypt_user(struct kvm *kvm, unsigned long paddr, | |
6715 | unsigned long __user dst_uaddr, | |
6716 | unsigned long dst_paddr, | |
6717 | int size, int *err) | |
6718 | { | |
6719 | struct page *tpage = NULL; | |
6720 | int ret, offset; | |
6721 | ||
6722 | /* if inputs are not 16-byte then use intermediate buffer */ | |
6723 | if (!IS_ALIGNED(dst_paddr, 16) || | |
6724 | !IS_ALIGNED(paddr, 16) || | |
6725 | !IS_ALIGNED(size, 16)) { | |
6726 | tpage = (void *)alloc_page(GFP_KERNEL); | |
6727 | if (!tpage) | |
6728 | return -ENOMEM; | |
6729 | ||
6730 | dst_paddr = __sme_page_pa(tpage); | |
6731 | } | |
6732 | ||
6733 | ret = __sev_dbg_decrypt(kvm, paddr, dst_paddr, size, err); | |
6734 | if (ret) | |
6735 | goto e_free; | |
6736 | ||
6737 | if (tpage) { | |
6738 | offset = paddr & 15; | |
6739 | if (copy_to_user((void __user *)(uintptr_t)dst_uaddr, | |
6740 | page_address(tpage) + offset, size)) | |
6741 | ret = -EFAULT; | |
6742 | } | |
6743 | ||
6744 | e_free: | |
6745 | if (tpage) | |
6746 | __free_page(tpage); | |
6747 | ||
6748 | return ret; | |
6749 | } | |
6750 | ||
7d1594f5 BS |
6751 | static int __sev_dbg_encrypt_user(struct kvm *kvm, unsigned long paddr, |
6752 | unsigned long __user vaddr, | |
6753 | unsigned long dst_paddr, | |
6754 | unsigned long __user dst_vaddr, | |
6755 | int size, int *error) | |
6756 | { | |
6757 | struct page *src_tpage = NULL; | |
6758 | struct page *dst_tpage = NULL; | |
6759 | int ret, len = size; | |
6760 | ||
6761 | /* If source buffer is not aligned then use an intermediate buffer */ | |
6762 | if (!IS_ALIGNED(vaddr, 16)) { | |
6763 | src_tpage = alloc_page(GFP_KERNEL); | |
6764 | if (!src_tpage) | |
6765 | return -ENOMEM; | |
6766 | ||
6767 | if (copy_from_user(page_address(src_tpage), | |
6768 | (void __user *)(uintptr_t)vaddr, size)) { | |
6769 | __free_page(src_tpage); | |
6770 | return -EFAULT; | |
6771 | } | |
6772 | ||
6773 | paddr = __sme_page_pa(src_tpage); | |
6774 | } | |
6775 | ||
6776 | /* | |
6777 | * If destination buffer or length is not aligned then do read-modify-write: | |
6778 | * - decrypt destination in an intermediate buffer | |
6779 | * - copy the source buffer in an intermediate buffer | |
6780 | * - use the intermediate buffer as source buffer | |
6781 | */ | |
6782 | if (!IS_ALIGNED(dst_vaddr, 16) || !IS_ALIGNED(size, 16)) { | |
6783 | int dst_offset; | |
6784 | ||
6785 | dst_tpage = alloc_page(GFP_KERNEL); | |
6786 | if (!dst_tpage) { | |
6787 | ret = -ENOMEM; | |
6788 | goto e_free; | |
6789 | } | |
6790 | ||
6791 | ret = __sev_dbg_decrypt(kvm, dst_paddr, | |
6792 | __sme_page_pa(dst_tpage), size, error); | |
6793 | if (ret) | |
6794 | goto e_free; | |
6795 | ||
6796 | /* | |
6797 | * If source is kernel buffer then use memcpy() otherwise | |
6798 | * copy_from_user(). | |
6799 | */ | |
6800 | dst_offset = dst_paddr & 15; | |
6801 | ||
6802 | if (src_tpage) | |
6803 | memcpy(page_address(dst_tpage) + dst_offset, | |
6804 | page_address(src_tpage), size); | |
6805 | else { | |
6806 | if (copy_from_user(page_address(dst_tpage) + dst_offset, | |
6807 | (void __user *)(uintptr_t)vaddr, size)) { | |
6808 | ret = -EFAULT; | |
6809 | goto e_free; | |
6810 | } | |
6811 | } | |
6812 | ||
6813 | paddr = __sme_page_pa(dst_tpage); | |
6814 | dst_paddr = round_down(dst_paddr, 16); | |
6815 | len = round_up(size, 16); | |
6816 | } | |
6817 | ||
6818 | ret = __sev_issue_dbg_cmd(kvm, paddr, dst_paddr, len, error, true); | |
6819 | ||
6820 | e_free: | |
6821 | if (src_tpage) | |
6822 | __free_page(src_tpage); | |
6823 | if (dst_tpage) | |
6824 | __free_page(dst_tpage); | |
6825 | return ret; | |
6826 | } | |
6827 | ||
24f41fb2 BS |
6828 | static int sev_dbg_crypt(struct kvm *kvm, struct kvm_sev_cmd *argp, bool dec) |
6829 | { | |
6830 | unsigned long vaddr, vaddr_end, next_vaddr; | |
0186ec82 | 6831 | unsigned long dst_vaddr; |
24f41fb2 BS |
6832 | struct page **src_p, **dst_p; |
6833 | struct kvm_sev_dbg debug; | |
6834 | unsigned long n; | |
b86bc285 DR |
6835 | unsigned int size; |
6836 | int ret; | |
24f41fb2 BS |
6837 | |
6838 | if (!sev_guest(kvm)) | |
6839 | return -ENOTTY; | |
6840 | ||
6841 | if (copy_from_user(&debug, (void __user *)(uintptr_t)argp->data, sizeof(debug))) | |
6842 | return -EFAULT; | |
6843 | ||
b86bc285 DR |
6844 | if (!debug.len || debug.src_uaddr + debug.len < debug.src_uaddr) |
6845 | return -EINVAL; | |
6846 | if (!debug.dst_uaddr) | |
6847 | return -EINVAL; | |
6848 | ||
24f41fb2 BS |
6849 | vaddr = debug.src_uaddr; |
6850 | size = debug.len; | |
6851 | vaddr_end = vaddr + size; | |
6852 | dst_vaddr = debug.dst_uaddr; | |
24f41fb2 BS |
6853 | |
6854 | for (; vaddr < vaddr_end; vaddr = next_vaddr) { | |
6855 | int len, s_off, d_off; | |
6856 | ||
6857 | /* lock userspace source and destination page */ | |
6858 | src_p = sev_pin_memory(kvm, vaddr & PAGE_MASK, PAGE_SIZE, &n, 0); | |
6859 | if (!src_p) | |
6860 | return -EFAULT; | |
6861 | ||
6862 | dst_p = sev_pin_memory(kvm, dst_vaddr & PAGE_MASK, PAGE_SIZE, &n, 1); | |
6863 | if (!dst_p) { | |
6864 | sev_unpin_memory(kvm, src_p, n); | |
6865 | return -EFAULT; | |
6866 | } | |
6867 | ||
6868 | /* | |
6869 | * The DBG_{DE,EN}CRYPT commands will perform {dec,en}cryption of the | |
6870 | * memory content (i.e it will write the same memory region with C=1). | |
6871 | * It's possible that the cache may contain the data with C=0, i.e., | |
6872 | * unencrypted so invalidate it first. | |
6873 | */ | |
6874 | sev_clflush_pages(src_p, 1); | |
6875 | sev_clflush_pages(dst_p, 1); | |
6876 | ||
6877 | /* | |
6878 | * Since user buffer may not be page aligned, calculate the | |
6879 | * offset within the page. | |
6880 | */ | |
6881 | s_off = vaddr & ~PAGE_MASK; | |
6882 | d_off = dst_vaddr & ~PAGE_MASK; | |
6883 | len = min_t(size_t, (PAGE_SIZE - s_off), size); | |
6884 | ||
7d1594f5 BS |
6885 | if (dec) |
6886 | ret = __sev_dbg_decrypt_user(kvm, | |
6887 | __sme_page_pa(src_p[0]) + s_off, | |
6888 | dst_vaddr, | |
6889 | __sme_page_pa(dst_p[0]) + d_off, | |
6890 | len, &argp->error); | |
6891 | else | |
6892 | ret = __sev_dbg_encrypt_user(kvm, | |
6893 | __sme_page_pa(src_p[0]) + s_off, | |
6894 | vaddr, | |
6895 | __sme_page_pa(dst_p[0]) + d_off, | |
6896 | dst_vaddr, | |
6897 | len, &argp->error); | |
24f41fb2 | 6898 | |
b86bc285 DR |
6899 | sev_unpin_memory(kvm, src_p, n); |
6900 | sev_unpin_memory(kvm, dst_p, n); | |
24f41fb2 BS |
6901 | |
6902 | if (ret) | |
6903 | goto err; | |
6904 | ||
6905 | next_vaddr = vaddr + len; | |
6906 | dst_vaddr = dst_vaddr + len; | |
6907 | size -= len; | |
6908 | } | |
6909 | err: | |
6910 | return ret; | |
6911 | } | |
6912 | ||
9f5b5b95 BS |
6913 | static int sev_launch_secret(struct kvm *kvm, struct kvm_sev_cmd *argp) |
6914 | { | |
81811c16 | 6915 | struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info; |
9f5b5b95 BS |
6916 | struct sev_data_launch_secret *data; |
6917 | struct kvm_sev_launch_secret params; | |
6918 | struct page **pages; | |
6919 | void *blob, *hdr; | |
6920 | unsigned long n; | |
9c5e0afa | 6921 | int ret, offset; |
9f5b5b95 BS |
6922 | |
6923 | if (!sev_guest(kvm)) | |
6924 | return -ENOTTY; | |
6925 | ||
6926 | if (copy_from_user(¶ms, (void __user *)(uintptr_t)argp->data, sizeof(params))) | |
6927 | return -EFAULT; | |
6928 | ||
6929 | pages = sev_pin_memory(kvm, params.guest_uaddr, params.guest_len, &n, 1); | |
6930 | if (!pages) | |
6931 | return -ENOMEM; | |
6932 | ||
6933 | /* | |
6934 | * The secret must be copied into contiguous memory region, lets verify | |
6935 | * that userspace memory pages are contiguous before we issue command. | |
6936 | */ | |
6937 | if (get_num_contig_pages(0, pages, n) != n) { | |
6938 | ret = -EINVAL; | |
6939 | goto e_unpin_memory; | |
6940 | } | |
6941 | ||
6942 | ret = -ENOMEM; | |
1ec69647 | 6943 | data = kzalloc(sizeof(*data), GFP_KERNEL_ACCOUNT); |
9f5b5b95 BS |
6944 | if (!data) |
6945 | goto e_unpin_memory; | |
6946 | ||
9c5e0afa BS |
6947 | offset = params.guest_uaddr & (PAGE_SIZE - 1); |
6948 | data->guest_address = __sme_page_pa(pages[0]) + offset; | |
6949 | data->guest_len = params.guest_len; | |
6950 | ||
9f5b5b95 BS |
6951 | blob = psp_copy_user_blob(params.trans_uaddr, params.trans_len); |
6952 | if (IS_ERR(blob)) { | |
6953 | ret = PTR_ERR(blob); | |
6954 | goto e_free; | |
6955 | } | |
6956 | ||
6957 | data->trans_address = __psp_pa(blob); | |
6958 | data->trans_len = params.trans_len; | |
6959 | ||
6960 | hdr = psp_copy_user_blob(params.hdr_uaddr, params.hdr_len); | |
6961 | if (IS_ERR(hdr)) { | |
6962 | ret = PTR_ERR(hdr); | |
6963 | goto e_free_blob; | |
6964 | } | |
9c5e0afa BS |
6965 | data->hdr_address = __psp_pa(hdr); |
6966 | data->hdr_len = params.hdr_len; | |
9f5b5b95 BS |
6967 | |
6968 | data->handle = sev->handle; | |
6969 | ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_UPDATE_SECRET, data, &argp->error); | |
6970 | ||
6971 | kfree(hdr); | |
6972 | ||
6973 | e_free_blob: | |
6974 | kfree(blob); | |
6975 | e_free: | |
6976 | kfree(data); | |
6977 | e_unpin_memory: | |
6978 | sev_unpin_memory(kvm, pages, n); | |
6979 | return ret; | |
6980 | } | |
6981 | ||
1654efcb BS |
6982 | static int svm_mem_enc_op(struct kvm *kvm, void __user *argp) |
6983 | { | |
6984 | struct kvm_sev_cmd sev_cmd; | |
6985 | int r; | |
6986 | ||
6987 | if (!svm_sev_enabled()) | |
6988 | return -ENOTTY; | |
6989 | ||
6990 | if (copy_from_user(&sev_cmd, argp, sizeof(struct kvm_sev_cmd))) | |
6991 | return -EFAULT; | |
6992 | ||
6993 | mutex_lock(&kvm->lock); | |
6994 | ||
6995 | switch (sev_cmd.id) { | |
6996 | case KVM_SEV_INIT: | |
6997 | r = sev_guest_init(kvm, &sev_cmd); | |
6998 | break; | |
59414c98 BS |
6999 | case KVM_SEV_LAUNCH_START: |
7000 | r = sev_launch_start(kvm, &sev_cmd); | |
7001 | break; | |
89c50580 BS |
7002 | case KVM_SEV_LAUNCH_UPDATE_DATA: |
7003 | r = sev_launch_update_data(kvm, &sev_cmd); | |
7004 | break; | |
0d0736f7 BS |
7005 | case KVM_SEV_LAUNCH_MEASURE: |
7006 | r = sev_launch_measure(kvm, &sev_cmd); | |
7007 | break; | |
5bdb0e2f BS |
7008 | case KVM_SEV_LAUNCH_FINISH: |
7009 | r = sev_launch_finish(kvm, &sev_cmd); | |
7010 | break; | |
255d9e75 BS |
7011 | case KVM_SEV_GUEST_STATUS: |
7012 | r = sev_guest_status(kvm, &sev_cmd); | |
7013 | break; | |
24f41fb2 BS |
7014 | case KVM_SEV_DBG_DECRYPT: |
7015 | r = sev_dbg_crypt(kvm, &sev_cmd, true); | |
7016 | break; | |
7d1594f5 BS |
7017 | case KVM_SEV_DBG_ENCRYPT: |
7018 | r = sev_dbg_crypt(kvm, &sev_cmd, false); | |
7019 | break; | |
9f5b5b95 BS |
7020 | case KVM_SEV_LAUNCH_SECRET: |
7021 | r = sev_launch_secret(kvm, &sev_cmd); | |
7022 | break; | |
1654efcb BS |
7023 | default: |
7024 | r = -EINVAL; | |
7025 | goto out; | |
7026 | } | |
7027 | ||
7028 | if (copy_to_user(argp, &sev_cmd, sizeof(struct kvm_sev_cmd))) | |
7029 | r = -EFAULT; | |
7030 | ||
7031 | out: | |
7032 | mutex_unlock(&kvm->lock); | |
7033 | return r; | |
7034 | } | |
7035 | ||
1e80fdc0 BS |
7036 | static int svm_register_enc_region(struct kvm *kvm, |
7037 | struct kvm_enc_region *range) | |
7038 | { | |
81811c16 | 7039 | struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info; |
1e80fdc0 BS |
7040 | struct enc_region *region; |
7041 | int ret = 0; | |
7042 | ||
7043 | if (!sev_guest(kvm)) | |
7044 | return -ENOTTY; | |
7045 | ||
86bf20cb DC |
7046 | if (range->addr > ULONG_MAX || range->size > ULONG_MAX) |
7047 | return -EINVAL; | |
7048 | ||
1ec69647 | 7049 | region = kzalloc(sizeof(*region), GFP_KERNEL_ACCOUNT); |
1e80fdc0 BS |
7050 | if (!region) |
7051 | return -ENOMEM; | |
7052 | ||
7053 | region->pages = sev_pin_memory(kvm, range->addr, range->size, ®ion->npages, 1); | |
7054 | if (!region->pages) { | |
7055 | ret = -ENOMEM; | |
7056 | goto e_free; | |
7057 | } | |
7058 | ||
7059 | /* | |
7060 | * The guest may change the memory encryption attribute from C=0 -> C=1 | |
7061 | * or vice versa for this memory range. Lets make sure caches are | |
7062 | * flushed to ensure that guest data gets written into memory with | |
7063 | * correct C-bit. | |
7064 | */ | |
7065 | sev_clflush_pages(region->pages, region->npages); | |
7066 | ||
7067 | region->uaddr = range->addr; | |
7068 | region->size = range->size; | |
7069 | ||
7070 | mutex_lock(&kvm->lock); | |
7071 | list_add_tail(®ion->list, &sev->regions_list); | |
7072 | mutex_unlock(&kvm->lock); | |
7073 | ||
7074 | return ret; | |
7075 | ||
7076 | e_free: | |
7077 | kfree(region); | |
7078 | return ret; | |
7079 | } | |
7080 | ||
7081 | static struct enc_region * | |
7082 | find_enc_region(struct kvm *kvm, struct kvm_enc_region *range) | |
7083 | { | |
81811c16 | 7084 | struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info; |
1e80fdc0 BS |
7085 | struct list_head *head = &sev->regions_list; |
7086 | struct enc_region *i; | |
7087 | ||
7088 | list_for_each_entry(i, head, list) { | |
7089 | if (i->uaddr == range->addr && | |
7090 | i->size == range->size) | |
7091 | return i; | |
7092 | } | |
7093 | ||
7094 | return NULL; | |
7095 | } | |
7096 | ||
7097 | ||
7098 | static int svm_unregister_enc_region(struct kvm *kvm, | |
7099 | struct kvm_enc_region *range) | |
7100 | { | |
7101 | struct enc_region *region; | |
7102 | int ret; | |
7103 | ||
7104 | mutex_lock(&kvm->lock); | |
7105 | ||
7106 | if (!sev_guest(kvm)) { | |
7107 | ret = -ENOTTY; | |
7108 | goto failed; | |
7109 | } | |
7110 | ||
7111 | region = find_enc_region(kvm, range); | |
7112 | if (!region) { | |
7113 | ret = -EINVAL; | |
7114 | goto failed; | |
7115 | } | |
7116 | ||
7117 | __unregister_enc_region_locked(kvm, region); | |
7118 | ||
7119 | mutex_unlock(&kvm->lock); | |
7120 | return 0; | |
7121 | ||
7122 | failed: | |
7123 | mutex_unlock(&kvm->lock); | |
7124 | return ret; | |
7125 | } | |
7126 | ||
e2e871ab VK |
7127 | static uint16_t nested_get_evmcs_version(struct kvm_vcpu *vcpu) |
7128 | { | |
7129 | /* Not supported */ | |
7130 | return 0; | |
7131 | } | |
7132 | ||
57b119da VK |
7133 | static int nested_enable_evmcs(struct kvm_vcpu *vcpu, |
7134 | uint16_t *vmcs_version) | |
7135 | { | |
7136 | /* Intel-only feature */ | |
7137 | return -ENODEV; | |
7138 | } | |
7139 | ||
05d5a486 SB |
7140 | static bool svm_need_emulation_on_page_fault(struct kvm_vcpu *vcpu) |
7141 | { | |
118154bd LA |
7142 | unsigned long cr4 = kvm_read_cr4(vcpu); |
7143 | bool smep = cr4 & X86_CR4_SMEP; | |
7144 | bool smap = cr4 & X86_CR4_SMAP; | |
7145 | bool is_user = svm_get_cpl(vcpu) == 3; | |
05d5a486 SB |
7146 | |
7147 | /* | |
118154bd LA |
7148 | * Detect and workaround Errata 1096 Fam_17h_00_0Fh. |
7149 | * | |
7150 | * Errata: | |
7151 | * When CPU raise #NPF on guest data access and vCPU CR4.SMAP=1, it is | |
7152 | * possible that CPU microcode implementing DecodeAssist will fail | |
7153 | * to read bytes of instruction which caused #NPF. In this case, | |
7154 | * GuestIntrBytes field of the VMCB on a VMEXIT will incorrectly | |
7155 | * return 0 instead of the correct guest instruction bytes. | |
7156 | * | |
7157 | * This happens because CPU microcode reading instruction bytes | |
7158 | * uses a special opcode which attempts to read data using CPL=0 | |
7159 | * priviledges. The microcode reads CS:RIP and if it hits a SMAP | |
7160 | * fault, it gives up and returns no instruction bytes. | |
7161 | * | |
7162 | * Detection: | |
7163 | * We reach here in case CPU supports DecodeAssist, raised #NPF and | |
7164 | * returned 0 in GuestIntrBytes field of the VMCB. | |
7165 | * First, errata can only be triggered in case vCPU CR4.SMAP=1. | |
7166 | * Second, if vCPU CR4.SMEP=1, errata could only be triggered | |
7167 | * in case vCPU CPL==3 (Because otherwise guest would have triggered | |
7168 | * a SMEP fault instead of #NPF). | |
7169 | * Otherwise, vCPU CR4.SMEP=0, errata could be triggered by any vCPU CPL. | |
7170 | * As most guests enable SMAP if they have also enabled SMEP, use above | |
7171 | * logic in order to attempt minimize false-positive of detecting errata | |
7172 | * while still preserving all cases semantic correctness. | |
7173 | * | |
7174 | * Workaround: | |
7175 | * To determine what instruction the guest was executing, the hypervisor | |
7176 | * will have to decode the instruction at the instruction pointer. | |
05d5a486 SB |
7177 | * |
7178 | * In non SEV guest, hypervisor will be able to read the guest | |
7179 | * memory to decode the instruction pointer when insn_len is zero | |
7180 | * so we return true to indicate that decoding is possible. | |
7181 | * | |
7182 | * But in the SEV guest, the guest memory is encrypted with the | |
7183 | * guest specific key and hypervisor will not be able to decode the | |
7184 | * instruction pointer so we will not able to workaround it. Lets | |
7185 | * print the error and request to kill the guest. | |
7186 | */ | |
118154bd | 7187 | if (smap && (!smep || is_user)) { |
05d5a486 SB |
7188 | if (!sev_guest(vcpu->kvm)) |
7189 | return true; | |
7190 | ||
118154bd | 7191 | pr_err_ratelimited("KVM: SEV Guest triggered AMD Erratum 1096\n"); |
05d5a486 SB |
7192 | kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu); |
7193 | } | |
7194 | ||
7195 | return false; | |
7196 | } | |
7197 | ||
404f6aac | 7198 | static struct kvm_x86_ops svm_x86_ops __ro_after_init = { |
6aa8b732 AK |
7199 | .cpu_has_kvm_support = has_svm, |
7200 | .disabled_by_bios = is_disabled, | |
7201 | .hardware_setup = svm_hardware_setup, | |
7202 | .hardware_unsetup = svm_hardware_unsetup, | |
002c7f7c | 7203 | .check_processor_compatibility = svm_check_processor_compat, |
6aa8b732 AK |
7204 | .hardware_enable = svm_hardware_enable, |
7205 | .hardware_disable = svm_hardware_disable, | |
774ead3a | 7206 | .cpu_has_accelerated_tpr = svm_cpu_has_accelerated_tpr, |
bc226f07 | 7207 | .has_emulated_msr = svm_has_emulated_msr, |
6aa8b732 AK |
7208 | |
7209 | .vcpu_create = svm_create_vcpu, | |
7210 | .vcpu_free = svm_free_vcpu, | |
04d2cc77 | 7211 | .vcpu_reset = svm_vcpu_reset, |
6aa8b732 | 7212 | |
434a1e94 SC |
7213 | .vm_alloc = svm_vm_alloc, |
7214 | .vm_free = svm_vm_free, | |
44a95dae | 7215 | .vm_init = avic_vm_init, |
1654efcb | 7216 | .vm_destroy = svm_vm_destroy, |
44a95dae | 7217 | |
04d2cc77 | 7218 | .prepare_guest_switch = svm_prepare_guest_switch, |
6aa8b732 AK |
7219 | .vcpu_load = svm_vcpu_load, |
7220 | .vcpu_put = svm_vcpu_put, | |
8221c137 SS |
7221 | .vcpu_blocking = svm_vcpu_blocking, |
7222 | .vcpu_unblocking = svm_vcpu_unblocking, | |
6aa8b732 | 7223 | |
a96036b8 | 7224 | .update_bp_intercept = update_bp_intercept, |
801e459a | 7225 | .get_msr_feature = svm_get_msr_feature, |
6aa8b732 AK |
7226 | .get_msr = svm_get_msr, |
7227 | .set_msr = svm_set_msr, | |
7228 | .get_segment_base = svm_get_segment_base, | |
7229 | .get_segment = svm_get_segment, | |
7230 | .set_segment = svm_set_segment, | |
2e4d2653 | 7231 | .get_cpl = svm_get_cpl, |
1747fb71 | 7232 | .get_cs_db_l_bits = kvm_get_cs_db_l_bits, |
e8467fda | 7233 | .decache_cr0_guest_bits = svm_decache_cr0_guest_bits, |
aff48baa | 7234 | .decache_cr3 = svm_decache_cr3, |
25c4c276 | 7235 | .decache_cr4_guest_bits = svm_decache_cr4_guest_bits, |
6aa8b732 | 7236 | .set_cr0 = svm_set_cr0, |
6aa8b732 AK |
7237 | .set_cr3 = svm_set_cr3, |
7238 | .set_cr4 = svm_set_cr4, | |
7239 | .set_efer = svm_set_efer, | |
7240 | .get_idt = svm_get_idt, | |
7241 | .set_idt = svm_set_idt, | |
7242 | .get_gdt = svm_get_gdt, | |
7243 | .set_gdt = svm_set_gdt, | |
73aaf249 JK |
7244 | .get_dr6 = svm_get_dr6, |
7245 | .set_dr6 = svm_set_dr6, | |
020df079 | 7246 | .set_dr7 = svm_set_dr7, |
facb0139 | 7247 | .sync_dirty_debug_regs = svm_sync_dirty_debug_regs, |
6de4f3ad | 7248 | .cache_reg = svm_cache_reg, |
6aa8b732 AK |
7249 | .get_rflags = svm_get_rflags, |
7250 | .set_rflags = svm_set_rflags, | |
be94f6b7 | 7251 | |
6aa8b732 | 7252 | .tlb_flush = svm_flush_tlb, |
faff8758 | 7253 | .tlb_flush_gva = svm_flush_tlb_gva, |
6aa8b732 | 7254 | |
6aa8b732 | 7255 | .run = svm_vcpu_run, |
04d2cc77 | 7256 | .handle_exit = handle_exit, |
6aa8b732 | 7257 | .skip_emulated_instruction = skip_emulated_instruction, |
2809f5d2 GC |
7258 | .set_interrupt_shadow = svm_set_interrupt_shadow, |
7259 | .get_interrupt_shadow = svm_get_interrupt_shadow, | |
102d8325 | 7260 | .patch_hypercall = svm_patch_hypercall, |
2a8067f1 | 7261 | .set_irq = svm_set_irq, |
95ba8273 | 7262 | .set_nmi = svm_inject_nmi, |
298101da | 7263 | .queue_exception = svm_queue_exception, |
b463a6f7 | 7264 | .cancel_injection = svm_cancel_injection, |
78646121 | 7265 | .interrupt_allowed = svm_interrupt_allowed, |
95ba8273 | 7266 | .nmi_allowed = svm_nmi_allowed, |
3cfc3092 JK |
7267 | .get_nmi_mask = svm_get_nmi_mask, |
7268 | .set_nmi_mask = svm_set_nmi_mask, | |
95ba8273 GN |
7269 | .enable_nmi_window = enable_nmi_window, |
7270 | .enable_irq_window = enable_irq_window, | |
7271 | .update_cr8_intercept = update_cr8_intercept, | |
8d860bbe | 7272 | .set_virtual_apic_mode = svm_set_virtual_apic_mode, |
d62caabb AS |
7273 | .get_enable_apicv = svm_get_enable_apicv, |
7274 | .refresh_apicv_exec_ctrl = svm_refresh_apicv_exec_ctrl, | |
c7c9c56c | 7275 | .load_eoi_exitmap = svm_load_eoi_exitmap, |
44a95dae SS |
7276 | .hwapic_irr_update = svm_hwapic_irr_update, |
7277 | .hwapic_isr_update = svm_hwapic_isr_update, | |
fa59cc00 | 7278 | .sync_pir_to_irr = kvm_lapic_find_highest_irr, |
be8ca170 | 7279 | .apicv_post_state_restore = avic_post_state_restore, |
cbc94022 IE |
7280 | |
7281 | .set_tss_addr = svm_set_tss_addr, | |
2ac52ab8 | 7282 | .set_identity_map_addr = svm_set_identity_map_addr, |
67253af5 | 7283 | .get_tdp_level = get_npt_level, |
4b12f0de | 7284 | .get_mt_mask = svm_get_mt_mask, |
229456fc | 7285 | |
586f9607 | 7286 | .get_exit_info = svm_get_exit_info, |
586f9607 | 7287 | |
17cc3935 | 7288 | .get_lpage_level = svm_get_lpage_level, |
0e851880 SY |
7289 | |
7290 | .cpuid_update = svm_cpuid_update, | |
4e47c7a6 SY |
7291 | |
7292 | .rdtscp_supported = svm_rdtscp_supported, | |
ad756a16 | 7293 | .invpcid_supported = svm_invpcid_supported, |
93c4adc7 | 7294 | .mpx_supported = svm_mpx_supported, |
55412b2e | 7295 | .xsaves_supported = svm_xsaves_supported, |
66336cab | 7296 | .umip_emulated = svm_umip_emulated, |
86f5201d | 7297 | .pt_supported = svm_pt_supported, |
d4330ef2 JR |
7298 | |
7299 | .set_supported_cpuid = svm_set_supported_cpuid, | |
f5f48ee1 SY |
7300 | |
7301 | .has_wbinvd_exit = svm_has_wbinvd_exit, | |
99e3e30a | 7302 | |
e79f245d | 7303 | .read_l1_tsc_offset = svm_read_l1_tsc_offset, |
326e7425 | 7304 | .write_l1_tsc_offset = svm_write_l1_tsc_offset, |
1c97f0a0 JR |
7305 | |
7306 | .set_tdp_cr3 = set_tdp_cr3, | |
8a76d7f2 JR |
7307 | |
7308 | .check_intercept = svm_check_intercept, | |
95b5a48c | 7309 | .handle_exit_irqoff = svm_handle_exit_irqoff, |
ae97a3b8 | 7310 | |
d264ee0c SC |
7311 | .request_immediate_exit = __kvm_request_immediate_exit, |
7312 | ||
ae97a3b8 | 7313 | .sched_in = svm_sched_in, |
25462f7f WH |
7314 | |
7315 | .pmu_ops = &amd_pmu_ops, | |
340d3bc3 | 7316 | .deliver_posted_interrupt = svm_deliver_avic_intr, |
411b44ba | 7317 | .update_pi_irte = svm_update_pi_irte, |
74f16909 | 7318 | .setup_mce = svm_setup_mce, |
0234bf88 | 7319 | |
72d7b374 | 7320 | .smi_allowed = svm_smi_allowed, |
0234bf88 LP |
7321 | .pre_enter_smm = svm_pre_enter_smm, |
7322 | .pre_leave_smm = svm_pre_leave_smm, | |
cc3d967f | 7323 | .enable_smi_window = enable_smi_window, |
1654efcb BS |
7324 | |
7325 | .mem_enc_op = svm_mem_enc_op, | |
1e80fdc0 BS |
7326 | .mem_enc_reg_region = svm_register_enc_region, |
7327 | .mem_enc_unreg_region = svm_unregister_enc_region, | |
57b119da VK |
7328 | |
7329 | .nested_enable_evmcs = nested_enable_evmcs, | |
e2e871ab | 7330 | .nested_get_evmcs_version = nested_get_evmcs_version, |
05d5a486 SB |
7331 | |
7332 | .need_emulation_on_page_fault = svm_need_emulation_on_page_fault, | |
6aa8b732 AK |
7333 | }; |
7334 | ||
7335 | static int __init svm_init(void) | |
7336 | { | |
cb498ea2 | 7337 | return kvm_init(&svm_x86_ops, sizeof(struct vcpu_svm), |
0ee75bea | 7338 | __alignof__(struct vcpu_svm), THIS_MODULE); |
6aa8b732 AK |
7339 | } |
7340 | ||
7341 | static void __exit svm_exit(void) | |
7342 | { | |
cb498ea2 | 7343 | kvm_exit(); |
6aa8b732 AK |
7344 | } |
7345 | ||
7346 | module_init(svm_init) | |
7347 | module_exit(svm_exit) |