Commit | Line | Data |
---|---|---|
6aa8b732 AK |
1 | /* |
2 | * Kernel-based Virtual Machine driver for Linux | |
3 | * | |
4 | * AMD SVM support | |
5 | * | |
6 | * Copyright (C) 2006 Qumranet, Inc. | |
9611c187 | 7 | * Copyright 2010 Red Hat, Inc. and/or its affiliates. |
6aa8b732 AK |
8 | * |
9 | * Authors: | |
10 | * Yaniv Kamay <yaniv@qumranet.com> | |
11 | * Avi Kivity <avi@qumranet.com> | |
12 | * | |
13 | * This work is licensed under the terms of the GNU GPL, version 2. See | |
14 | * the COPYING file in the top-level directory. | |
15 | * | |
16 | */ | |
44a95dae SS |
17 | |
18 | #define pr_fmt(fmt) "SVM: " fmt | |
19 | ||
edf88417 AK |
20 | #include <linux/kvm_host.h> |
21 | ||
85f455f7 | 22 | #include "irq.h" |
1d737c8a | 23 | #include "mmu.h" |
5fdbf976 | 24 | #include "kvm_cache_regs.h" |
fe4c7b19 | 25 | #include "x86.h" |
66f7b72e | 26 | #include "cpuid.h" |
25462f7f | 27 | #include "pmu.h" |
e495606d | 28 | |
6aa8b732 | 29 | #include <linux/module.h> |
ae759544 | 30 | #include <linux/mod_devicetable.h> |
9d8f549d | 31 | #include <linux/kernel.h> |
6aa8b732 AK |
32 | #include <linux/vmalloc.h> |
33 | #include <linux/highmem.h> | |
e8edc6e0 | 34 | #include <linux/sched.h> |
af658dca | 35 | #include <linux/trace_events.h> |
5a0e3ad6 | 36 | #include <linux/slab.h> |
5881f737 SS |
37 | #include <linux/amd-iommu.h> |
38 | #include <linux/hashtable.h> | |
c207aee4 | 39 | #include <linux/frame.h> |
e9df0942 | 40 | #include <linux/psp-sev.h> |
1654efcb | 41 | #include <linux/file.h> |
89c50580 BS |
42 | #include <linux/pagemap.h> |
43 | #include <linux/swap.h> | |
6aa8b732 | 44 | |
8221c137 | 45 | #include <asm/apic.h> |
1018faa6 | 46 | #include <asm/perf_event.h> |
67ec6607 | 47 | #include <asm/tlbflush.h> |
e495606d | 48 | #include <asm/desc.h> |
facb0139 | 49 | #include <asm/debugreg.h> |
631bc487 | 50 | #include <asm/kvm_para.h> |
411b44ba | 51 | #include <asm/irq_remapping.h> |
ecb586bd | 52 | #include <asm/microcode.h> |
117cc7a9 | 53 | #include <asm/nospec-branch.h> |
6aa8b732 | 54 | |
63d1142f | 55 | #include <asm/virtext.h> |
229456fc | 56 | #include "trace.h" |
63d1142f | 57 | |
4ecac3fd AK |
58 | #define __ex(x) __kvm_handle_fault_on_reboot(x) |
59 | ||
6aa8b732 AK |
60 | MODULE_AUTHOR("Qumranet"); |
61 | MODULE_LICENSE("GPL"); | |
62 | ||
ae759544 JT |
63 | static const struct x86_cpu_id svm_cpu_id[] = { |
64 | X86_FEATURE_MATCH(X86_FEATURE_SVM), | |
65 | {} | |
66 | }; | |
67 | MODULE_DEVICE_TABLE(x86cpu, svm_cpu_id); | |
68 | ||
6aa8b732 AK |
69 | #define IOPM_ALLOC_ORDER 2 |
70 | #define MSRPM_ALLOC_ORDER 1 | |
71 | ||
6aa8b732 AK |
72 | #define SEG_TYPE_LDT 2 |
73 | #define SEG_TYPE_BUSY_TSS16 3 | |
74 | ||
6bc31bdc AP |
75 | #define SVM_FEATURE_NPT (1 << 0) |
76 | #define SVM_FEATURE_LBRV (1 << 1) | |
77 | #define SVM_FEATURE_SVML (1 << 2) | |
78 | #define SVM_FEATURE_NRIP (1 << 3) | |
ddce97aa AP |
79 | #define SVM_FEATURE_TSC_RATE (1 << 4) |
80 | #define SVM_FEATURE_VMCB_CLEAN (1 << 5) | |
81 | #define SVM_FEATURE_FLUSH_ASID (1 << 6) | |
82 | #define SVM_FEATURE_DECODE_ASSIST (1 << 7) | |
6bc31bdc | 83 | #define SVM_FEATURE_PAUSE_FILTER (1 << 10) |
80b7706e | 84 | |
340d3bc3 SS |
85 | #define SVM_AVIC_DOORBELL 0xc001011b |
86 | ||
410e4d57 JR |
87 | #define NESTED_EXIT_HOST 0 /* Exit handled on host level */ |
88 | #define NESTED_EXIT_DONE 1 /* Exit caused nested vmexit */ | |
89 | #define NESTED_EXIT_CONTINUE 2 /* Further checks needed */ | |
90 | ||
24e09cbf JR |
91 | #define DEBUGCTL_RESERVED_BITS (~(0x3fULL)) |
92 | ||
fbc0db76 | 93 | #define TSC_RATIO_RSVD 0xffffff0000000000ULL |
92a1f12d JR |
94 | #define TSC_RATIO_MIN 0x0000000000000001ULL |
95 | #define TSC_RATIO_MAX 0x000000ffffffffffULL | |
fbc0db76 | 96 | |
5446a979 | 97 | #define AVIC_HPA_MASK ~((0xFFFULL << 52) | 0xFFF) |
44a95dae SS |
98 | |
99 | /* | |
100 | * 0xff is broadcast, so the max index allowed for physical APIC ID | |
101 | * table is 0xfe. APIC IDs above 0xff are reserved. | |
102 | */ | |
103 | #define AVIC_MAX_PHYSICAL_ID_COUNT 255 | |
104 | ||
18f40c53 SS |
105 | #define AVIC_UNACCEL_ACCESS_WRITE_MASK 1 |
106 | #define AVIC_UNACCEL_ACCESS_OFFSET_MASK 0xFF0 | |
107 | #define AVIC_UNACCEL_ACCESS_VECTOR_MASK 0xFFFFFFFF | |
108 | ||
5ea11f2b SS |
109 | /* AVIC GATAG is encoded using VM and VCPU IDs */ |
110 | #define AVIC_VCPU_ID_BITS 8 | |
111 | #define AVIC_VCPU_ID_MASK ((1 << AVIC_VCPU_ID_BITS) - 1) | |
112 | ||
113 | #define AVIC_VM_ID_BITS 24 | |
114 | #define AVIC_VM_ID_NR (1 << AVIC_VM_ID_BITS) | |
115 | #define AVIC_VM_ID_MASK ((1 << AVIC_VM_ID_BITS) - 1) | |
116 | ||
117 | #define AVIC_GATAG(x, y) (((x & AVIC_VM_ID_MASK) << AVIC_VCPU_ID_BITS) | \ | |
118 | (y & AVIC_VCPU_ID_MASK)) | |
119 | #define AVIC_GATAG_TO_VMID(x) ((x >> AVIC_VCPU_ID_BITS) & AVIC_VM_ID_MASK) | |
120 | #define AVIC_GATAG_TO_VCPUID(x) (x & AVIC_VCPU_ID_MASK) | |
121 | ||
67ec6607 JR |
122 | static bool erratum_383_found __read_mostly; |
123 | ||
6c8166a7 AK |
124 | static const u32 host_save_user_msrs[] = { |
125 | #ifdef CONFIG_X86_64 | |
126 | MSR_STAR, MSR_LSTAR, MSR_CSTAR, MSR_SYSCALL_MASK, MSR_KERNEL_GS_BASE, | |
127 | MSR_FS_BASE, | |
128 | #endif | |
129 | MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP, | |
46896c73 | 130 | MSR_TSC_AUX, |
6c8166a7 AK |
131 | }; |
132 | ||
133 | #define NR_HOST_SAVE_USER_MSRS ARRAY_SIZE(host_save_user_msrs) | |
134 | ||
81811c16 SC |
135 | struct kvm_sev_info { |
136 | bool active; /* SEV enabled guest */ | |
137 | unsigned int asid; /* ASID used for this guest */ | |
138 | unsigned int handle; /* SEV firmware handle */ | |
139 | int fd; /* SEV device fd */ | |
140 | unsigned long pages_locked; /* Number of pages locked */ | |
141 | struct list_head regions_list; /* List of registered regions */ | |
142 | }; | |
143 | ||
144 | struct kvm_svm { | |
145 | struct kvm kvm; | |
146 | ||
147 | /* Struct members for AVIC */ | |
148 | u32 avic_vm_id; | |
149 | u32 ldr_mode; | |
150 | struct page *avic_logical_id_table_page; | |
151 | struct page *avic_physical_id_table_page; | |
152 | struct hlist_node hnode; | |
153 | ||
154 | struct kvm_sev_info sev_info; | |
155 | }; | |
156 | ||
6c8166a7 AK |
157 | struct kvm_vcpu; |
158 | ||
e6aa9abd JR |
159 | struct nested_state { |
160 | struct vmcb *hsave; | |
161 | u64 hsave_msr; | |
4a810181 | 162 | u64 vm_cr_msr; |
e6aa9abd JR |
163 | u64 vmcb; |
164 | ||
165 | /* These are the merged vectors */ | |
166 | u32 *msrpm; | |
167 | ||
168 | /* gpa pointers to the real vectors */ | |
169 | u64 vmcb_msrpm; | |
ce2ac085 | 170 | u64 vmcb_iopm; |
aad42c64 | 171 | |
cd3ff653 JR |
172 | /* A VMEXIT is required but not yet emulated */ |
173 | bool exit_required; | |
174 | ||
aad42c64 | 175 | /* cache for intercepts of the guest */ |
4ee546b4 | 176 | u32 intercept_cr; |
3aed041a | 177 | u32 intercept_dr; |
aad42c64 JR |
178 | u32 intercept_exceptions; |
179 | u64 intercept; | |
180 | ||
5bd2edc3 JR |
181 | /* Nested Paging related state */ |
182 | u64 nested_cr3; | |
e6aa9abd JR |
183 | }; |
184 | ||
323c3d80 JR |
185 | #define MSRPM_OFFSETS 16 |
186 | static u32 msrpm_offsets[MSRPM_OFFSETS] __read_mostly; | |
187 | ||
2b036c6b BO |
188 | /* |
189 | * Set osvw_len to higher value when updated Revision Guides | |
190 | * are published and we know what the new status bits are | |
191 | */ | |
192 | static uint64_t osvw_len = 4, osvw_status; | |
193 | ||
6c8166a7 AK |
194 | struct vcpu_svm { |
195 | struct kvm_vcpu vcpu; | |
196 | struct vmcb *vmcb; | |
197 | unsigned long vmcb_pa; | |
198 | struct svm_cpu_data *svm_data; | |
199 | uint64_t asid_generation; | |
200 | uint64_t sysenter_esp; | |
201 | uint64_t sysenter_eip; | |
46896c73 | 202 | uint64_t tsc_aux; |
6c8166a7 | 203 | |
d1d93fa9 TL |
204 | u64 msr_decfg; |
205 | ||
6c8166a7 AK |
206 | u64 next_rip; |
207 | ||
208 | u64 host_user_msrs[NR_HOST_SAVE_USER_MSRS]; | |
afe9e66f | 209 | struct { |
dacccfdd AK |
210 | u16 fs; |
211 | u16 gs; | |
212 | u16 ldt; | |
afe9e66f AK |
213 | u64 gs_base; |
214 | } host; | |
6c8166a7 | 215 | |
b2ac58f9 KA |
216 | u64 spec_ctrl; |
217 | ||
6c8166a7 | 218 | u32 *msrpm; |
6c8166a7 | 219 | |
bd3d1ec3 AK |
220 | ulong nmi_iret_rip; |
221 | ||
e6aa9abd | 222 | struct nested_state nested; |
6be7d306 JK |
223 | |
224 | bool nmi_singlestep; | |
ab2f4d73 | 225 | u64 nmi_singlestep_guest_rflags; |
66b7138f JK |
226 | |
227 | unsigned int3_injected; | |
228 | unsigned long int3_rip; | |
fbc0db76 | 229 | |
6092d3d3 JR |
230 | /* cached guest cpuid flags for faster access */ |
231 | bool nrips_enabled : 1; | |
44a95dae | 232 | |
18f40c53 | 233 | u32 ldr_reg; |
44a95dae SS |
234 | struct page *avic_backing_page; |
235 | u64 *avic_physical_id_cache; | |
8221c137 | 236 | bool avic_is_running; |
411b44ba SS |
237 | |
238 | /* | |
239 | * Per-vcpu list of struct amd_svm_iommu_ir: | |
240 | * This is used mainly to store interrupt remapping information used | |
241 | * when update the vcpu affinity. This avoids the need to scan for | |
242 | * IRTE and try to match ga_tag in the IOMMU driver. | |
243 | */ | |
244 | struct list_head ir_list; | |
245 | spinlock_t ir_list_lock; | |
70cd94e6 BS |
246 | |
247 | /* which host CPU was used for running this vcpu */ | |
248 | unsigned int last_cpu; | |
411b44ba SS |
249 | }; |
250 | ||
251 | /* | |
252 | * This is a wrapper of struct amd_iommu_ir_data. | |
253 | */ | |
254 | struct amd_svm_iommu_ir { | |
255 | struct list_head node; /* Used by SVM for per-vcpu ir_list */ | |
256 | void *data; /* Storing pointer to struct amd_ir_data */ | |
6c8166a7 AK |
257 | }; |
258 | ||
44a95dae SS |
259 | #define AVIC_LOGICAL_ID_ENTRY_GUEST_PHYSICAL_ID_MASK (0xFF) |
260 | #define AVIC_LOGICAL_ID_ENTRY_VALID_MASK (1 << 31) | |
261 | ||
262 | #define AVIC_PHYSICAL_ID_ENTRY_HOST_PHYSICAL_ID_MASK (0xFFULL) | |
263 | #define AVIC_PHYSICAL_ID_ENTRY_BACKING_PAGE_MASK (0xFFFFFFFFFFULL << 12) | |
264 | #define AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK (1ULL << 62) | |
265 | #define AVIC_PHYSICAL_ID_ENTRY_VALID_MASK (1ULL << 63) | |
266 | ||
fbc0db76 JR |
267 | static DEFINE_PER_CPU(u64, current_tsc_ratio); |
268 | #define TSC_RATIO_DEFAULT 0x0100000000ULL | |
269 | ||
455716fa JR |
270 | #define MSR_INVALID 0xffffffffU |
271 | ||
09941fbb | 272 | static const struct svm_direct_access_msrs { |
ac72a9b7 JR |
273 | u32 index; /* Index of the MSR */ |
274 | bool always; /* True if intercept is always on */ | |
275 | } direct_access_msrs[] = { | |
8c06585d | 276 | { .index = MSR_STAR, .always = true }, |
ac72a9b7 JR |
277 | { .index = MSR_IA32_SYSENTER_CS, .always = true }, |
278 | #ifdef CONFIG_X86_64 | |
279 | { .index = MSR_GS_BASE, .always = true }, | |
280 | { .index = MSR_FS_BASE, .always = true }, | |
281 | { .index = MSR_KERNEL_GS_BASE, .always = true }, | |
282 | { .index = MSR_LSTAR, .always = true }, | |
283 | { .index = MSR_CSTAR, .always = true }, | |
284 | { .index = MSR_SYSCALL_MASK, .always = true }, | |
285 | #endif | |
b2ac58f9 | 286 | { .index = MSR_IA32_SPEC_CTRL, .always = false }, |
15d45071 | 287 | { .index = MSR_IA32_PRED_CMD, .always = false }, |
ac72a9b7 JR |
288 | { .index = MSR_IA32_LASTBRANCHFROMIP, .always = false }, |
289 | { .index = MSR_IA32_LASTBRANCHTOIP, .always = false }, | |
290 | { .index = MSR_IA32_LASTINTFROMIP, .always = false }, | |
291 | { .index = MSR_IA32_LASTINTTOIP, .always = false }, | |
292 | { .index = MSR_INVALID, .always = false }, | |
6c8166a7 AK |
293 | }; |
294 | ||
709ddebf JR |
295 | /* enable NPT for AMD64 and X86 with PAE */ |
296 | #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE) | |
297 | static bool npt_enabled = true; | |
298 | #else | |
e0231715 | 299 | static bool npt_enabled; |
709ddebf | 300 | #endif |
6c7dac72 | 301 | |
8566ac8b BM |
302 | /* |
303 | * These 2 parameters are used to config the controls for Pause-Loop Exiting: | |
304 | * pause_filter_count: On processors that support Pause filtering(indicated | |
305 | * by CPUID Fn8000_000A_EDX), the VMCB provides a 16 bit pause filter | |
306 | * count value. On VMRUN this value is loaded into an internal counter. | |
307 | * Each time a pause instruction is executed, this counter is decremented | |
308 | * until it reaches zero at which time a #VMEXIT is generated if pause | |
309 | * intercept is enabled. Refer to AMD APM Vol 2 Section 15.14.4 Pause | |
310 | * Intercept Filtering for more details. | |
311 | * This also indicate if ple logic enabled. | |
312 | * | |
313 | * pause_filter_thresh: In addition, some processor families support advanced | |
314 | * pause filtering (indicated by CPUID Fn8000_000A_EDX) upper bound on | |
315 | * the amount of time a guest is allowed to execute in a pause loop. | |
316 | * In this mode, a 16-bit pause filter threshold field is added in the | |
317 | * VMCB. The threshold value is a cycle count that is used to reset the | |
318 | * pause counter. As with simple pause filtering, VMRUN loads the pause | |
319 | * count value from VMCB into an internal counter. Then, on each pause | |
320 | * instruction the hardware checks the elapsed number of cycles since | |
321 | * the most recent pause instruction against the pause filter threshold. | |
322 | * If the elapsed cycle count is greater than the pause filter threshold, | |
323 | * then the internal pause count is reloaded from the VMCB and execution | |
324 | * continues. If the elapsed cycle count is less than the pause filter | |
325 | * threshold, then the internal pause count is decremented. If the count | |
326 | * value is less than zero and PAUSE intercept is enabled, a #VMEXIT is | |
327 | * triggered. If advanced pause filtering is supported and pause filter | |
328 | * threshold field is set to zero, the filter will operate in the simpler, | |
329 | * count only mode. | |
330 | */ | |
331 | ||
332 | static unsigned short pause_filter_thresh = KVM_DEFAULT_PLE_GAP; | |
333 | module_param(pause_filter_thresh, ushort, 0444); | |
334 | ||
335 | static unsigned short pause_filter_count = KVM_SVM_DEFAULT_PLE_WINDOW; | |
336 | module_param(pause_filter_count, ushort, 0444); | |
337 | ||
338 | /* Default doubles per-vcpu window every exit. */ | |
339 | static unsigned short pause_filter_count_grow = KVM_DEFAULT_PLE_WINDOW_GROW; | |
340 | module_param(pause_filter_count_grow, ushort, 0444); | |
341 | ||
342 | /* Default resets per-vcpu window every exit to pause_filter_count. */ | |
343 | static unsigned short pause_filter_count_shrink = KVM_DEFAULT_PLE_WINDOW_SHRINK; | |
344 | module_param(pause_filter_count_shrink, ushort, 0444); | |
345 | ||
346 | /* Default is to compute the maximum so we can never overflow. */ | |
347 | static unsigned short pause_filter_count_max = KVM_SVM_DEFAULT_PLE_WINDOW_MAX; | |
348 | module_param(pause_filter_count_max, ushort, 0444); | |
349 | ||
e2358851 DB |
350 | /* allow nested paging (virtualized MMU) for all guests */ |
351 | static int npt = true; | |
6c7dac72 | 352 | module_param(npt, int, S_IRUGO); |
e3da3acd | 353 | |
e2358851 DB |
354 | /* allow nested virtualization in KVM/SVM */ |
355 | static int nested = true; | |
236de055 AG |
356 | module_param(nested, int, S_IRUGO); |
357 | ||
44a95dae SS |
358 | /* enable / disable AVIC */ |
359 | static int avic; | |
5b8abf1f | 360 | #ifdef CONFIG_X86_LOCAL_APIC |
44a95dae | 361 | module_param(avic, int, S_IRUGO); |
5b8abf1f | 362 | #endif |
44a95dae | 363 | |
89c8a498 JN |
364 | /* enable/disable Virtual VMLOAD VMSAVE */ |
365 | static int vls = true; | |
366 | module_param(vls, int, 0444); | |
367 | ||
640bd6e5 JN |
368 | /* enable/disable Virtual GIF */ |
369 | static int vgif = true; | |
370 | module_param(vgif, int, 0444); | |
5ea11f2b | 371 | |
e9df0942 BS |
372 | /* enable/disable SEV support */ |
373 | static int sev = IS_ENABLED(CONFIG_AMD_MEM_ENCRYPT_ACTIVE_BY_DEFAULT); | |
374 | module_param(sev, int, 0444); | |
375 | ||
7607b717 BS |
376 | static u8 rsm_ins_bytes[] = "\x0f\xaa"; |
377 | ||
79a8059d | 378 | static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0); |
c2ba05cc | 379 | static void svm_flush_tlb(struct kvm_vcpu *vcpu, bool invalidate_gpa); |
a5c3832d | 380 | static void svm_complete_interrupts(struct vcpu_svm *svm); |
04d2cc77 | 381 | |
410e4d57 | 382 | static int nested_svm_exit_handled(struct vcpu_svm *svm); |
b8e88bc8 | 383 | static int nested_svm_intercept(struct vcpu_svm *svm); |
cf74a78b | 384 | static int nested_svm_vmexit(struct vcpu_svm *svm); |
cf74a78b AG |
385 | static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr, |
386 | bool has_error_code, u32 error_code); | |
387 | ||
8d28fec4 | 388 | enum { |
116a0a23 JR |
389 | VMCB_INTERCEPTS, /* Intercept vectors, TSC offset, |
390 | pause filter count */ | |
f56838e4 | 391 | VMCB_PERM_MAP, /* IOPM Base and MSRPM Base */ |
d48086d1 | 392 | VMCB_ASID, /* ASID */ |
decdbf6a | 393 | VMCB_INTR, /* int_ctl, int_vector */ |
b2747166 | 394 | VMCB_NPT, /* npt_en, nCR3, gPAT */ |
dcca1a65 | 395 | VMCB_CR, /* CR0, CR3, CR4, EFER */ |
72214b96 | 396 | VMCB_DR, /* DR6, DR7 */ |
17a703cb | 397 | VMCB_DT, /* GDT, IDT */ |
060d0c9a | 398 | VMCB_SEG, /* CS, DS, SS, ES, CPL */ |
0574dec0 | 399 | VMCB_CR2, /* CR2 only */ |
b53ba3f9 | 400 | VMCB_LBR, /* DBGCTL, BR_FROM, BR_TO, LAST_EX_FROM, LAST_EX_TO */ |
44a95dae SS |
401 | VMCB_AVIC, /* AVIC APIC_BAR, AVIC APIC_BACKING_PAGE, |
402 | * AVIC PHYSICAL_TABLE pointer, | |
403 | * AVIC LOGICAL_TABLE pointer | |
404 | */ | |
8d28fec4 RJ |
405 | VMCB_DIRTY_MAX, |
406 | }; | |
407 | ||
0574dec0 JR |
408 | /* TPR and CR2 are always written before VMRUN */ |
409 | #define VMCB_ALWAYS_DIRTY_MASK ((1U << VMCB_INTR) | (1U << VMCB_CR2)) | |
8d28fec4 | 410 | |
44a95dae SS |
411 | #define VMCB_AVIC_APIC_BAR_MASK 0xFFFFFFFFFF000ULL |
412 | ||
ed3cd233 | 413 | static unsigned int max_sev_asid; |
1654efcb BS |
414 | static unsigned int min_sev_asid; |
415 | static unsigned long *sev_asid_bitmap; | |
89c50580 | 416 | #define __sme_page_pa(x) __sme_set(page_to_pfn(x) << PAGE_SHIFT) |
1654efcb | 417 | |
1e80fdc0 BS |
418 | struct enc_region { |
419 | struct list_head list; | |
420 | unsigned long npages; | |
421 | struct page **pages; | |
422 | unsigned long uaddr; | |
423 | unsigned long size; | |
424 | }; | |
425 | ||
81811c16 SC |
426 | |
427 | static inline struct kvm_svm *to_kvm_svm(struct kvm *kvm) | |
428 | { | |
429 | return container_of(kvm, struct kvm_svm, kvm); | |
430 | } | |
431 | ||
1654efcb BS |
432 | static inline bool svm_sev_enabled(void) |
433 | { | |
434 | return max_sev_asid; | |
435 | } | |
436 | ||
437 | static inline bool sev_guest(struct kvm *kvm) | |
438 | { | |
81811c16 | 439 | struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info; |
1654efcb BS |
440 | |
441 | return sev->active; | |
442 | } | |
ed3cd233 | 443 | |
70cd94e6 BS |
444 | static inline int sev_get_asid(struct kvm *kvm) |
445 | { | |
81811c16 | 446 | struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info; |
70cd94e6 BS |
447 | |
448 | return sev->asid; | |
449 | } | |
450 | ||
8d28fec4 RJ |
451 | static inline void mark_all_dirty(struct vmcb *vmcb) |
452 | { | |
453 | vmcb->control.clean = 0; | |
454 | } | |
455 | ||
456 | static inline void mark_all_clean(struct vmcb *vmcb) | |
457 | { | |
458 | vmcb->control.clean = ((1 << VMCB_DIRTY_MAX) - 1) | |
459 | & ~VMCB_ALWAYS_DIRTY_MASK; | |
460 | } | |
461 | ||
462 | static inline void mark_dirty(struct vmcb *vmcb, int bit) | |
463 | { | |
464 | vmcb->control.clean &= ~(1 << bit); | |
465 | } | |
466 | ||
a2fa3e9f GH |
467 | static inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu) |
468 | { | |
fb3f0f51 | 469 | return container_of(vcpu, struct vcpu_svm, vcpu); |
a2fa3e9f GH |
470 | } |
471 | ||
44a95dae SS |
472 | static inline void avic_update_vapic_bar(struct vcpu_svm *svm, u64 data) |
473 | { | |
474 | svm->vmcb->control.avic_vapic_bar = data & VMCB_AVIC_APIC_BAR_MASK; | |
475 | mark_dirty(svm->vmcb, VMCB_AVIC); | |
476 | } | |
477 | ||
340d3bc3 SS |
478 | static inline bool avic_vcpu_is_running(struct kvm_vcpu *vcpu) |
479 | { | |
480 | struct vcpu_svm *svm = to_svm(vcpu); | |
481 | u64 *entry = svm->avic_physical_id_cache; | |
482 | ||
483 | if (!entry) | |
484 | return false; | |
485 | ||
486 | return (READ_ONCE(*entry) & AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK); | |
487 | } | |
488 | ||
384c6368 JR |
489 | static void recalc_intercepts(struct vcpu_svm *svm) |
490 | { | |
491 | struct vmcb_control_area *c, *h; | |
492 | struct nested_state *g; | |
493 | ||
116a0a23 JR |
494 | mark_dirty(svm->vmcb, VMCB_INTERCEPTS); |
495 | ||
384c6368 JR |
496 | if (!is_guest_mode(&svm->vcpu)) |
497 | return; | |
498 | ||
499 | c = &svm->vmcb->control; | |
500 | h = &svm->nested.hsave->control; | |
501 | g = &svm->nested; | |
502 | ||
4ee546b4 | 503 | c->intercept_cr = h->intercept_cr | g->intercept_cr; |
3aed041a | 504 | c->intercept_dr = h->intercept_dr | g->intercept_dr; |
bd89525a | 505 | c->intercept_exceptions = h->intercept_exceptions | g->intercept_exceptions; |
384c6368 JR |
506 | c->intercept = h->intercept | g->intercept; |
507 | } | |
508 | ||
4ee546b4 RJ |
509 | static inline struct vmcb *get_host_vmcb(struct vcpu_svm *svm) |
510 | { | |
511 | if (is_guest_mode(&svm->vcpu)) | |
512 | return svm->nested.hsave; | |
513 | else | |
514 | return svm->vmcb; | |
515 | } | |
516 | ||
517 | static inline void set_cr_intercept(struct vcpu_svm *svm, int bit) | |
518 | { | |
519 | struct vmcb *vmcb = get_host_vmcb(svm); | |
520 | ||
521 | vmcb->control.intercept_cr |= (1U << bit); | |
522 | ||
523 | recalc_intercepts(svm); | |
524 | } | |
525 | ||
526 | static inline void clr_cr_intercept(struct vcpu_svm *svm, int bit) | |
527 | { | |
528 | struct vmcb *vmcb = get_host_vmcb(svm); | |
529 | ||
530 | vmcb->control.intercept_cr &= ~(1U << bit); | |
531 | ||
532 | recalc_intercepts(svm); | |
533 | } | |
534 | ||
535 | static inline bool is_cr_intercept(struct vcpu_svm *svm, int bit) | |
536 | { | |
537 | struct vmcb *vmcb = get_host_vmcb(svm); | |
538 | ||
539 | return vmcb->control.intercept_cr & (1U << bit); | |
540 | } | |
541 | ||
5315c716 | 542 | static inline void set_dr_intercepts(struct vcpu_svm *svm) |
3aed041a JR |
543 | { |
544 | struct vmcb *vmcb = get_host_vmcb(svm); | |
545 | ||
5315c716 PB |
546 | vmcb->control.intercept_dr = (1 << INTERCEPT_DR0_READ) |
547 | | (1 << INTERCEPT_DR1_READ) | |
548 | | (1 << INTERCEPT_DR2_READ) | |
549 | | (1 << INTERCEPT_DR3_READ) | |
550 | | (1 << INTERCEPT_DR4_READ) | |
551 | | (1 << INTERCEPT_DR5_READ) | |
552 | | (1 << INTERCEPT_DR6_READ) | |
553 | | (1 << INTERCEPT_DR7_READ) | |
554 | | (1 << INTERCEPT_DR0_WRITE) | |
555 | | (1 << INTERCEPT_DR1_WRITE) | |
556 | | (1 << INTERCEPT_DR2_WRITE) | |
557 | | (1 << INTERCEPT_DR3_WRITE) | |
558 | | (1 << INTERCEPT_DR4_WRITE) | |
559 | | (1 << INTERCEPT_DR5_WRITE) | |
560 | | (1 << INTERCEPT_DR6_WRITE) | |
561 | | (1 << INTERCEPT_DR7_WRITE); | |
3aed041a JR |
562 | |
563 | recalc_intercepts(svm); | |
564 | } | |
565 | ||
5315c716 | 566 | static inline void clr_dr_intercepts(struct vcpu_svm *svm) |
3aed041a JR |
567 | { |
568 | struct vmcb *vmcb = get_host_vmcb(svm); | |
569 | ||
5315c716 | 570 | vmcb->control.intercept_dr = 0; |
3aed041a JR |
571 | |
572 | recalc_intercepts(svm); | |
573 | } | |
574 | ||
18c918c5 JR |
575 | static inline void set_exception_intercept(struct vcpu_svm *svm, int bit) |
576 | { | |
577 | struct vmcb *vmcb = get_host_vmcb(svm); | |
578 | ||
579 | vmcb->control.intercept_exceptions |= (1U << bit); | |
580 | ||
581 | recalc_intercepts(svm); | |
582 | } | |
583 | ||
584 | static inline void clr_exception_intercept(struct vcpu_svm *svm, int bit) | |
585 | { | |
586 | struct vmcb *vmcb = get_host_vmcb(svm); | |
587 | ||
588 | vmcb->control.intercept_exceptions &= ~(1U << bit); | |
589 | ||
590 | recalc_intercepts(svm); | |
591 | } | |
592 | ||
8a05a1b8 JR |
593 | static inline void set_intercept(struct vcpu_svm *svm, int bit) |
594 | { | |
595 | struct vmcb *vmcb = get_host_vmcb(svm); | |
596 | ||
597 | vmcb->control.intercept |= (1ULL << bit); | |
598 | ||
599 | recalc_intercepts(svm); | |
600 | } | |
601 | ||
602 | static inline void clr_intercept(struct vcpu_svm *svm, int bit) | |
603 | { | |
604 | struct vmcb *vmcb = get_host_vmcb(svm); | |
605 | ||
606 | vmcb->control.intercept &= ~(1ULL << bit); | |
607 | ||
608 | recalc_intercepts(svm); | |
609 | } | |
610 | ||
640bd6e5 JN |
611 | static inline bool vgif_enabled(struct vcpu_svm *svm) |
612 | { | |
613 | return !!(svm->vmcb->control.int_ctl & V_GIF_ENABLE_MASK); | |
614 | } | |
615 | ||
2af9194d JR |
616 | static inline void enable_gif(struct vcpu_svm *svm) |
617 | { | |
640bd6e5 JN |
618 | if (vgif_enabled(svm)) |
619 | svm->vmcb->control.int_ctl |= V_GIF_MASK; | |
620 | else | |
621 | svm->vcpu.arch.hflags |= HF_GIF_MASK; | |
2af9194d JR |
622 | } |
623 | ||
624 | static inline void disable_gif(struct vcpu_svm *svm) | |
625 | { | |
640bd6e5 JN |
626 | if (vgif_enabled(svm)) |
627 | svm->vmcb->control.int_ctl &= ~V_GIF_MASK; | |
628 | else | |
629 | svm->vcpu.arch.hflags &= ~HF_GIF_MASK; | |
2af9194d JR |
630 | } |
631 | ||
632 | static inline bool gif_set(struct vcpu_svm *svm) | |
633 | { | |
640bd6e5 JN |
634 | if (vgif_enabled(svm)) |
635 | return !!(svm->vmcb->control.int_ctl & V_GIF_MASK); | |
636 | else | |
637 | return !!(svm->vcpu.arch.hflags & HF_GIF_MASK); | |
2af9194d JR |
638 | } |
639 | ||
4866d5e3 | 640 | static unsigned long iopm_base; |
6aa8b732 AK |
641 | |
642 | struct kvm_ldttss_desc { | |
643 | u16 limit0; | |
644 | u16 base0; | |
e0231715 JR |
645 | unsigned base1:8, type:5, dpl:2, p:1; |
646 | unsigned limit1:4, zero0:3, g:1, base2:8; | |
6aa8b732 AK |
647 | u32 base3; |
648 | u32 zero1; | |
649 | } __attribute__((packed)); | |
650 | ||
651 | struct svm_cpu_data { | |
652 | int cpu; | |
653 | ||
5008fdf5 AK |
654 | u64 asid_generation; |
655 | u32 max_asid; | |
656 | u32 next_asid; | |
4faefff3 | 657 | u32 min_asid; |
6aa8b732 AK |
658 | struct kvm_ldttss_desc *tss_desc; |
659 | ||
660 | struct page *save_area; | |
15d45071 | 661 | struct vmcb *current_vmcb; |
70cd94e6 BS |
662 | |
663 | /* index = sev_asid, value = vmcb pointer */ | |
664 | struct vmcb **sev_vmcbs; | |
6aa8b732 AK |
665 | }; |
666 | ||
667 | static DEFINE_PER_CPU(struct svm_cpu_data *, svm_data); | |
668 | ||
669 | struct svm_init_data { | |
670 | int cpu; | |
671 | int r; | |
672 | }; | |
673 | ||
09941fbb | 674 | static const u32 msrpm_ranges[] = {0, 0xc0000000, 0xc0010000}; |
6aa8b732 | 675 | |
9d8f549d | 676 | #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges) |
6aa8b732 AK |
677 | #define MSRS_RANGE_SIZE 2048 |
678 | #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2) | |
679 | ||
455716fa JR |
680 | static u32 svm_msrpm_offset(u32 msr) |
681 | { | |
682 | u32 offset; | |
683 | int i; | |
684 | ||
685 | for (i = 0; i < NUM_MSR_MAPS; i++) { | |
686 | if (msr < msrpm_ranges[i] || | |
687 | msr >= msrpm_ranges[i] + MSRS_IN_RANGE) | |
688 | continue; | |
689 | ||
690 | offset = (msr - msrpm_ranges[i]) / 4; /* 4 msrs per u8 */ | |
691 | offset += (i * MSRS_RANGE_SIZE); /* add range offset */ | |
692 | ||
693 | /* Now we have the u8 offset - but need the u32 offset */ | |
694 | return offset / 4; | |
695 | } | |
696 | ||
697 | /* MSR not in any range */ | |
698 | return MSR_INVALID; | |
699 | } | |
700 | ||
6aa8b732 AK |
701 | #define MAX_INST_SIZE 15 |
702 | ||
6aa8b732 AK |
703 | static inline void clgi(void) |
704 | { | |
4ecac3fd | 705 | asm volatile (__ex(SVM_CLGI)); |
6aa8b732 AK |
706 | } |
707 | ||
708 | static inline void stgi(void) | |
709 | { | |
4ecac3fd | 710 | asm volatile (__ex(SVM_STGI)); |
6aa8b732 AK |
711 | } |
712 | ||
713 | static inline void invlpga(unsigned long addr, u32 asid) | |
714 | { | |
e0231715 | 715 | asm volatile (__ex(SVM_INVLPGA) : : "a"(addr), "c"(asid)); |
6aa8b732 AK |
716 | } |
717 | ||
855feb67 | 718 | static int get_npt_level(struct kvm_vcpu *vcpu) |
4b16184c JR |
719 | { |
720 | #ifdef CONFIG_X86_64 | |
2a7266a8 | 721 | return PT64_ROOT_4LEVEL; |
4b16184c JR |
722 | #else |
723 | return PT32E_ROOT_LEVEL; | |
724 | #endif | |
725 | } | |
726 | ||
6aa8b732 AK |
727 | static void svm_set_efer(struct kvm_vcpu *vcpu, u64 efer) |
728 | { | |
6dc696d4 | 729 | vcpu->arch.efer = efer; |
709ddebf | 730 | if (!npt_enabled && !(efer & EFER_LMA)) |
2b5203ee | 731 | efer &= ~EFER_LME; |
6aa8b732 | 732 | |
9962d032 | 733 | to_svm(vcpu)->vmcb->save.efer = efer | EFER_SVME; |
dcca1a65 | 734 | mark_dirty(to_svm(vcpu)->vmcb, VMCB_CR); |
6aa8b732 AK |
735 | } |
736 | ||
6aa8b732 AK |
737 | static int is_external_interrupt(u32 info) |
738 | { | |
739 | info &= SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID; | |
740 | return info == (SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR); | |
741 | } | |
742 | ||
37ccdcbe | 743 | static u32 svm_get_interrupt_shadow(struct kvm_vcpu *vcpu) |
2809f5d2 GC |
744 | { |
745 | struct vcpu_svm *svm = to_svm(vcpu); | |
746 | u32 ret = 0; | |
747 | ||
748 | if (svm->vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK) | |
37ccdcbe PB |
749 | ret = KVM_X86_SHADOW_INT_STI | KVM_X86_SHADOW_INT_MOV_SS; |
750 | return ret; | |
2809f5d2 GC |
751 | } |
752 | ||
753 | static void svm_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask) | |
754 | { | |
755 | struct vcpu_svm *svm = to_svm(vcpu); | |
756 | ||
757 | if (mask == 0) | |
758 | svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK; | |
759 | else | |
760 | svm->vmcb->control.int_state |= SVM_INTERRUPT_SHADOW_MASK; | |
761 | ||
762 | } | |
763 | ||
6aa8b732 AK |
764 | static void skip_emulated_instruction(struct kvm_vcpu *vcpu) |
765 | { | |
a2fa3e9f GH |
766 | struct vcpu_svm *svm = to_svm(vcpu); |
767 | ||
f104765b | 768 | if (svm->vmcb->control.next_rip != 0) { |
d2922422 | 769 | WARN_ON_ONCE(!static_cpu_has(X86_FEATURE_NRIPS)); |
6bc31bdc | 770 | svm->next_rip = svm->vmcb->control.next_rip; |
f104765b | 771 | } |
6bc31bdc | 772 | |
a2fa3e9f | 773 | if (!svm->next_rip) { |
51d8b661 | 774 | if (emulate_instruction(vcpu, EMULTYPE_SKIP) != |
f629cf84 GN |
775 | EMULATE_DONE) |
776 | printk(KERN_DEBUG "%s: NOP\n", __func__); | |
6aa8b732 AK |
777 | return; |
778 | } | |
5fdbf976 MT |
779 | if (svm->next_rip - kvm_rip_read(vcpu) > MAX_INST_SIZE) |
780 | printk(KERN_ERR "%s: ip 0x%lx next 0x%llx\n", | |
781 | __func__, kvm_rip_read(vcpu), svm->next_rip); | |
6aa8b732 | 782 | |
5fdbf976 | 783 | kvm_rip_write(vcpu, svm->next_rip); |
2809f5d2 | 784 | svm_set_interrupt_shadow(vcpu, 0); |
6aa8b732 AK |
785 | } |
786 | ||
cfcd20e5 | 787 | static void svm_queue_exception(struct kvm_vcpu *vcpu) |
116a4752 JK |
788 | { |
789 | struct vcpu_svm *svm = to_svm(vcpu); | |
cfcd20e5 WL |
790 | unsigned nr = vcpu->arch.exception.nr; |
791 | bool has_error_code = vcpu->arch.exception.has_error_code; | |
664f8e26 | 792 | bool reinject = vcpu->arch.exception.injected; |
cfcd20e5 | 793 | u32 error_code = vcpu->arch.exception.error_code; |
116a4752 | 794 | |
e0231715 JR |
795 | /* |
796 | * If we are within a nested VM we'd better #VMEXIT and let the guest | |
797 | * handle the exception | |
798 | */ | |
ce7ddec4 JR |
799 | if (!reinject && |
800 | nested_svm_check_exception(svm, nr, has_error_code, error_code)) | |
116a4752 JK |
801 | return; |
802 | ||
2a6b20b8 | 803 | if (nr == BP_VECTOR && !static_cpu_has(X86_FEATURE_NRIPS)) { |
66b7138f JK |
804 | unsigned long rip, old_rip = kvm_rip_read(&svm->vcpu); |
805 | ||
806 | /* | |
807 | * For guest debugging where we have to reinject #BP if some | |
808 | * INT3 is guest-owned: | |
809 | * Emulate nRIP by moving RIP forward. Will fail if injection | |
810 | * raises a fault that is not intercepted. Still better than | |
811 | * failing in all cases. | |
812 | */ | |
813 | skip_emulated_instruction(&svm->vcpu); | |
814 | rip = kvm_rip_read(&svm->vcpu); | |
815 | svm->int3_rip = rip + svm->vmcb->save.cs.base; | |
816 | svm->int3_injected = rip - old_rip; | |
817 | } | |
818 | ||
116a4752 JK |
819 | svm->vmcb->control.event_inj = nr |
820 | | SVM_EVTINJ_VALID | |
821 | | (has_error_code ? SVM_EVTINJ_VALID_ERR : 0) | |
822 | | SVM_EVTINJ_TYPE_EXEPT; | |
823 | svm->vmcb->control.event_inj_err = error_code; | |
824 | } | |
825 | ||
67ec6607 JR |
826 | static void svm_init_erratum_383(void) |
827 | { | |
828 | u32 low, high; | |
829 | int err; | |
830 | u64 val; | |
831 | ||
e6ee94d5 | 832 | if (!static_cpu_has_bug(X86_BUG_AMD_TLB_MMATCH)) |
67ec6607 JR |
833 | return; |
834 | ||
835 | /* Use _safe variants to not break nested virtualization */ | |
836 | val = native_read_msr_safe(MSR_AMD64_DC_CFG, &err); | |
837 | if (err) | |
838 | return; | |
839 | ||
840 | val |= (1ULL << 47); | |
841 | ||
842 | low = lower_32_bits(val); | |
843 | high = upper_32_bits(val); | |
844 | ||
845 | native_write_msr_safe(MSR_AMD64_DC_CFG, low, high); | |
846 | ||
847 | erratum_383_found = true; | |
848 | } | |
849 | ||
2b036c6b BO |
850 | static void svm_init_osvw(struct kvm_vcpu *vcpu) |
851 | { | |
852 | /* | |
853 | * Guests should see errata 400 and 415 as fixed (assuming that | |
854 | * HLT and IO instructions are intercepted). | |
855 | */ | |
856 | vcpu->arch.osvw.length = (osvw_len >= 3) ? (osvw_len) : 3; | |
857 | vcpu->arch.osvw.status = osvw_status & ~(6ULL); | |
858 | ||
859 | /* | |
860 | * By increasing VCPU's osvw.length to 3 we are telling the guest that | |
861 | * all osvw.status bits inside that length, including bit 0 (which is | |
862 | * reserved for erratum 298), are valid. However, if host processor's | |
863 | * osvw_len is 0 then osvw_status[0] carries no information. We need to | |
864 | * be conservative here and therefore we tell the guest that erratum 298 | |
865 | * is present (because we really don't know). | |
866 | */ | |
867 | if (osvw_len == 0 && boot_cpu_data.x86 == 0x10) | |
868 | vcpu->arch.osvw.status |= 1; | |
869 | } | |
870 | ||
6aa8b732 AK |
871 | static int has_svm(void) |
872 | { | |
63d1142f | 873 | const char *msg; |
6aa8b732 | 874 | |
63d1142f | 875 | if (!cpu_has_svm(&msg)) { |
ff81ff10 | 876 | printk(KERN_INFO "has_svm: %s\n", msg); |
6aa8b732 AK |
877 | return 0; |
878 | } | |
879 | ||
6aa8b732 AK |
880 | return 1; |
881 | } | |
882 | ||
13a34e06 | 883 | static void svm_hardware_disable(void) |
6aa8b732 | 884 | { |
fbc0db76 JR |
885 | /* Make sure we clean up behind us */ |
886 | if (static_cpu_has(X86_FEATURE_TSCRATEMSR)) | |
887 | wrmsrl(MSR_AMD64_TSC_RATIO, TSC_RATIO_DEFAULT); | |
888 | ||
2c8dceeb | 889 | cpu_svm_disable(); |
1018faa6 JR |
890 | |
891 | amd_pmu_disable_virt(); | |
6aa8b732 AK |
892 | } |
893 | ||
13a34e06 | 894 | static int svm_hardware_enable(void) |
6aa8b732 AK |
895 | { |
896 | ||
0fe1e009 | 897 | struct svm_cpu_data *sd; |
6aa8b732 | 898 | uint64_t efer; |
6aa8b732 AK |
899 | struct desc_struct *gdt; |
900 | int me = raw_smp_processor_id(); | |
901 | ||
10474ae8 AG |
902 | rdmsrl(MSR_EFER, efer); |
903 | if (efer & EFER_SVME) | |
904 | return -EBUSY; | |
905 | ||
6aa8b732 | 906 | if (!has_svm()) { |
1f5b77f5 | 907 | pr_err("%s: err EOPNOTSUPP on %d\n", __func__, me); |
10474ae8 | 908 | return -EINVAL; |
6aa8b732 | 909 | } |
0fe1e009 | 910 | sd = per_cpu(svm_data, me); |
0fe1e009 | 911 | if (!sd) { |
1f5b77f5 | 912 | pr_err("%s: svm_data is NULL on %d\n", __func__, me); |
10474ae8 | 913 | return -EINVAL; |
6aa8b732 AK |
914 | } |
915 | ||
0fe1e009 TH |
916 | sd->asid_generation = 1; |
917 | sd->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1; | |
918 | sd->next_asid = sd->max_asid + 1; | |
ed3cd233 | 919 | sd->min_asid = max_sev_asid + 1; |
6aa8b732 | 920 | |
45fc8757 | 921 | gdt = get_current_gdt_rw(); |
0fe1e009 | 922 | sd->tss_desc = (struct kvm_ldttss_desc *)(gdt + GDT_ENTRY_TSS); |
6aa8b732 | 923 | |
9962d032 | 924 | wrmsrl(MSR_EFER, efer | EFER_SVME); |
6aa8b732 | 925 | |
d0316554 | 926 | wrmsrl(MSR_VM_HSAVE_PA, page_to_pfn(sd->save_area) << PAGE_SHIFT); |
10474ae8 | 927 | |
fbc0db76 JR |
928 | if (static_cpu_has(X86_FEATURE_TSCRATEMSR)) { |
929 | wrmsrl(MSR_AMD64_TSC_RATIO, TSC_RATIO_DEFAULT); | |
89cbc767 | 930 | __this_cpu_write(current_tsc_ratio, TSC_RATIO_DEFAULT); |
fbc0db76 JR |
931 | } |
932 | ||
2b036c6b BO |
933 | |
934 | /* | |
935 | * Get OSVW bits. | |
936 | * | |
937 | * Note that it is possible to have a system with mixed processor | |
938 | * revisions and therefore different OSVW bits. If bits are not the same | |
939 | * on different processors then choose the worst case (i.e. if erratum | |
940 | * is present on one processor and not on another then assume that the | |
941 | * erratum is present everywhere). | |
942 | */ | |
943 | if (cpu_has(&boot_cpu_data, X86_FEATURE_OSVW)) { | |
944 | uint64_t len, status = 0; | |
945 | int err; | |
946 | ||
947 | len = native_read_msr_safe(MSR_AMD64_OSVW_ID_LENGTH, &err); | |
948 | if (!err) | |
949 | status = native_read_msr_safe(MSR_AMD64_OSVW_STATUS, | |
950 | &err); | |
951 | ||
952 | if (err) | |
953 | osvw_status = osvw_len = 0; | |
954 | else { | |
955 | if (len < osvw_len) | |
956 | osvw_len = len; | |
957 | osvw_status |= status; | |
958 | osvw_status &= (1ULL << osvw_len) - 1; | |
959 | } | |
960 | } else | |
961 | osvw_status = osvw_len = 0; | |
962 | ||
67ec6607 JR |
963 | svm_init_erratum_383(); |
964 | ||
1018faa6 JR |
965 | amd_pmu_enable_virt(); |
966 | ||
10474ae8 | 967 | return 0; |
6aa8b732 AK |
968 | } |
969 | ||
0da1db75 JR |
970 | static void svm_cpu_uninit(int cpu) |
971 | { | |
0fe1e009 | 972 | struct svm_cpu_data *sd = per_cpu(svm_data, raw_smp_processor_id()); |
0da1db75 | 973 | |
0fe1e009 | 974 | if (!sd) |
0da1db75 JR |
975 | return; |
976 | ||
977 | per_cpu(svm_data, raw_smp_processor_id()) = NULL; | |
70cd94e6 | 978 | kfree(sd->sev_vmcbs); |
0fe1e009 TH |
979 | __free_page(sd->save_area); |
980 | kfree(sd); | |
0da1db75 JR |
981 | } |
982 | ||
6aa8b732 AK |
983 | static int svm_cpu_init(int cpu) |
984 | { | |
0fe1e009 | 985 | struct svm_cpu_data *sd; |
6aa8b732 AK |
986 | int r; |
987 | ||
0fe1e009 TH |
988 | sd = kzalloc(sizeof(struct svm_cpu_data), GFP_KERNEL); |
989 | if (!sd) | |
6aa8b732 | 990 | return -ENOMEM; |
0fe1e009 | 991 | sd->cpu = cpu; |
6aa8b732 | 992 | r = -ENOMEM; |
70cd94e6 | 993 | sd->save_area = alloc_page(GFP_KERNEL); |
0fe1e009 | 994 | if (!sd->save_area) |
6aa8b732 AK |
995 | goto err_1; |
996 | ||
70cd94e6 BS |
997 | if (svm_sev_enabled()) { |
998 | r = -ENOMEM; | |
999 | sd->sev_vmcbs = kmalloc((max_sev_asid + 1) * sizeof(void *), GFP_KERNEL); | |
1000 | if (!sd->sev_vmcbs) | |
1001 | goto err_1; | |
1002 | } | |
1003 | ||
0fe1e009 | 1004 | per_cpu(svm_data, cpu) = sd; |
6aa8b732 AK |
1005 | |
1006 | return 0; | |
1007 | ||
1008 | err_1: | |
0fe1e009 | 1009 | kfree(sd); |
6aa8b732 AK |
1010 | return r; |
1011 | ||
1012 | } | |
1013 | ||
ac72a9b7 JR |
1014 | static bool valid_msr_intercept(u32 index) |
1015 | { | |
1016 | int i; | |
1017 | ||
1018 | for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) | |
1019 | if (direct_access_msrs[i].index == index) | |
1020 | return true; | |
1021 | ||
1022 | return false; | |
1023 | } | |
1024 | ||
b2ac58f9 KA |
1025 | static bool msr_write_intercepted(struct kvm_vcpu *vcpu, unsigned msr) |
1026 | { | |
1027 | u8 bit_write; | |
1028 | unsigned long tmp; | |
1029 | u32 offset; | |
1030 | u32 *msrpm; | |
1031 | ||
1032 | msrpm = is_guest_mode(vcpu) ? to_svm(vcpu)->nested.msrpm: | |
1033 | to_svm(vcpu)->msrpm; | |
1034 | ||
1035 | offset = svm_msrpm_offset(msr); | |
1036 | bit_write = 2 * (msr & 0x0f) + 1; | |
1037 | tmp = msrpm[offset]; | |
1038 | ||
1039 | BUG_ON(offset == MSR_INVALID); | |
1040 | ||
1041 | return !!test_bit(bit_write, &tmp); | |
1042 | } | |
1043 | ||
bfc733a7 RR |
1044 | static void set_msr_interception(u32 *msrpm, unsigned msr, |
1045 | int read, int write) | |
6aa8b732 | 1046 | { |
455716fa JR |
1047 | u8 bit_read, bit_write; |
1048 | unsigned long tmp; | |
1049 | u32 offset; | |
6aa8b732 | 1050 | |
ac72a9b7 JR |
1051 | /* |
1052 | * If this warning triggers extend the direct_access_msrs list at the | |
1053 | * beginning of the file | |
1054 | */ | |
1055 | WARN_ON(!valid_msr_intercept(msr)); | |
1056 | ||
455716fa JR |
1057 | offset = svm_msrpm_offset(msr); |
1058 | bit_read = 2 * (msr & 0x0f); | |
1059 | bit_write = 2 * (msr & 0x0f) + 1; | |
1060 | tmp = msrpm[offset]; | |
1061 | ||
1062 | BUG_ON(offset == MSR_INVALID); | |
1063 | ||
1064 | read ? clear_bit(bit_read, &tmp) : set_bit(bit_read, &tmp); | |
1065 | write ? clear_bit(bit_write, &tmp) : set_bit(bit_write, &tmp); | |
1066 | ||
1067 | msrpm[offset] = tmp; | |
6aa8b732 AK |
1068 | } |
1069 | ||
f65c229c | 1070 | static void svm_vcpu_init_msrpm(u32 *msrpm) |
6aa8b732 AK |
1071 | { |
1072 | int i; | |
1073 | ||
f65c229c JR |
1074 | memset(msrpm, 0xff, PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER)); |
1075 | ||
ac72a9b7 JR |
1076 | for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) { |
1077 | if (!direct_access_msrs[i].always) | |
1078 | continue; | |
1079 | ||
1080 | set_msr_interception(msrpm, direct_access_msrs[i].index, 1, 1); | |
1081 | } | |
f65c229c JR |
1082 | } |
1083 | ||
323c3d80 JR |
1084 | static void add_msr_offset(u32 offset) |
1085 | { | |
1086 | int i; | |
1087 | ||
1088 | for (i = 0; i < MSRPM_OFFSETS; ++i) { | |
1089 | ||
1090 | /* Offset already in list? */ | |
1091 | if (msrpm_offsets[i] == offset) | |
bfc733a7 | 1092 | return; |
323c3d80 JR |
1093 | |
1094 | /* Slot used by another offset? */ | |
1095 | if (msrpm_offsets[i] != MSR_INVALID) | |
1096 | continue; | |
1097 | ||
1098 | /* Add offset to list */ | |
1099 | msrpm_offsets[i] = offset; | |
1100 | ||
1101 | return; | |
6aa8b732 | 1102 | } |
323c3d80 JR |
1103 | |
1104 | /* | |
1105 | * If this BUG triggers the msrpm_offsets table has an overflow. Just | |
1106 | * increase MSRPM_OFFSETS in this case. | |
1107 | */ | |
bfc733a7 | 1108 | BUG(); |
6aa8b732 AK |
1109 | } |
1110 | ||
323c3d80 | 1111 | static void init_msrpm_offsets(void) |
f65c229c | 1112 | { |
323c3d80 | 1113 | int i; |
f65c229c | 1114 | |
323c3d80 JR |
1115 | memset(msrpm_offsets, 0xff, sizeof(msrpm_offsets)); |
1116 | ||
1117 | for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) { | |
1118 | u32 offset; | |
1119 | ||
1120 | offset = svm_msrpm_offset(direct_access_msrs[i].index); | |
1121 | BUG_ON(offset == MSR_INVALID); | |
1122 | ||
1123 | add_msr_offset(offset); | |
1124 | } | |
f65c229c JR |
1125 | } |
1126 | ||
24e09cbf JR |
1127 | static void svm_enable_lbrv(struct vcpu_svm *svm) |
1128 | { | |
1129 | u32 *msrpm = svm->msrpm; | |
1130 | ||
0dc92119 | 1131 | svm->vmcb->control.virt_ext |= LBR_CTL_ENABLE_MASK; |
24e09cbf JR |
1132 | set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 1, 1); |
1133 | set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 1, 1); | |
1134 | set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 1, 1); | |
1135 | set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 1, 1); | |
1136 | } | |
1137 | ||
1138 | static void svm_disable_lbrv(struct vcpu_svm *svm) | |
1139 | { | |
1140 | u32 *msrpm = svm->msrpm; | |
1141 | ||
0dc92119 | 1142 | svm->vmcb->control.virt_ext &= ~LBR_CTL_ENABLE_MASK; |
24e09cbf JR |
1143 | set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 0, 0); |
1144 | set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 0, 0); | |
1145 | set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 0, 0); | |
1146 | set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 0, 0); | |
1147 | } | |
1148 | ||
4aebd0e9 LP |
1149 | static void disable_nmi_singlestep(struct vcpu_svm *svm) |
1150 | { | |
1151 | svm->nmi_singlestep = false; | |
640bd6e5 | 1152 | |
ab2f4d73 LP |
1153 | if (!(svm->vcpu.guest_debug & KVM_GUESTDBG_SINGLESTEP)) { |
1154 | /* Clear our flags if they were not set by the guest */ | |
1155 | if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_TF)) | |
1156 | svm->vmcb->save.rflags &= ~X86_EFLAGS_TF; | |
1157 | if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_RF)) | |
1158 | svm->vmcb->save.rflags &= ~X86_EFLAGS_RF; | |
1159 | } | |
4aebd0e9 LP |
1160 | } |
1161 | ||
5881f737 | 1162 | /* Note: |
81811c16 | 1163 | * This hash table is used to map VM_ID to a struct kvm_svm, |
5881f737 SS |
1164 | * when handling AMD IOMMU GALOG notification to schedule in |
1165 | * a particular vCPU. | |
1166 | */ | |
1167 | #define SVM_VM_DATA_HASH_BITS 8 | |
681bcea8 | 1168 | static DEFINE_HASHTABLE(svm_vm_data_hash, SVM_VM_DATA_HASH_BITS); |
3f0d4db7 DV |
1169 | static u32 next_vm_id = 0; |
1170 | static bool next_vm_id_wrapped = 0; | |
681bcea8 | 1171 | static DEFINE_SPINLOCK(svm_vm_data_hash_lock); |
5881f737 SS |
1172 | |
1173 | /* Note: | |
1174 | * This function is called from IOMMU driver to notify | |
1175 | * SVM to schedule in a particular vCPU of a particular VM. | |
1176 | */ | |
1177 | static int avic_ga_log_notifier(u32 ga_tag) | |
1178 | { | |
1179 | unsigned long flags; | |
81811c16 | 1180 | struct kvm_svm *kvm_svm; |
5881f737 SS |
1181 | struct kvm_vcpu *vcpu = NULL; |
1182 | u32 vm_id = AVIC_GATAG_TO_VMID(ga_tag); | |
1183 | u32 vcpu_id = AVIC_GATAG_TO_VCPUID(ga_tag); | |
1184 | ||
1185 | pr_debug("SVM: %s: vm_id=%#x, vcpu_id=%#x\n", __func__, vm_id, vcpu_id); | |
1186 | ||
1187 | spin_lock_irqsave(&svm_vm_data_hash_lock, flags); | |
81811c16 SC |
1188 | hash_for_each_possible(svm_vm_data_hash, kvm_svm, hnode, vm_id) { |
1189 | if (kvm_svm->avic_vm_id != vm_id) | |
5881f737 | 1190 | continue; |
81811c16 | 1191 | vcpu = kvm_get_vcpu_by_id(&kvm_svm->kvm, vcpu_id); |
5881f737 SS |
1192 | break; |
1193 | } | |
1194 | spin_unlock_irqrestore(&svm_vm_data_hash_lock, flags); | |
1195 | ||
5881f737 SS |
1196 | /* Note: |
1197 | * At this point, the IOMMU should have already set the pending | |
1198 | * bit in the vAPIC backing page. So, we just need to schedule | |
1199 | * in the vcpu. | |
1200 | */ | |
1cf53587 | 1201 | if (vcpu) |
5881f737 SS |
1202 | kvm_vcpu_wake_up(vcpu); |
1203 | ||
1204 | return 0; | |
1205 | } | |
1206 | ||
e9df0942 BS |
1207 | static __init int sev_hardware_setup(void) |
1208 | { | |
1209 | struct sev_user_data_status *status; | |
1210 | int rc; | |
1211 | ||
1212 | /* Maximum number of encrypted guests supported simultaneously */ | |
1213 | max_sev_asid = cpuid_ecx(0x8000001F); | |
1214 | ||
1215 | if (!max_sev_asid) | |
1216 | return 1; | |
1217 | ||
1654efcb BS |
1218 | /* Minimum ASID value that should be used for SEV guest */ |
1219 | min_sev_asid = cpuid_edx(0x8000001F); | |
1220 | ||
1221 | /* Initialize SEV ASID bitmap */ | |
1222 | sev_asid_bitmap = kcalloc(BITS_TO_LONGS(max_sev_asid), | |
1223 | sizeof(unsigned long), GFP_KERNEL); | |
1224 | if (!sev_asid_bitmap) | |
1225 | return 1; | |
1226 | ||
e9df0942 BS |
1227 | status = kmalloc(sizeof(*status), GFP_KERNEL); |
1228 | if (!status) | |
1229 | return 1; | |
1230 | ||
1231 | /* | |
1232 | * Check SEV platform status. | |
1233 | * | |
1234 | * PLATFORM_STATUS can be called in any state, if we failed to query | |
1235 | * the PLATFORM status then either PSP firmware does not support SEV | |
1236 | * feature or SEV firmware is dead. | |
1237 | */ | |
1238 | rc = sev_platform_status(status, NULL); | |
1239 | if (rc) | |
1240 | goto err; | |
1241 | ||
1242 | pr_info("SEV supported\n"); | |
1243 | ||
1244 | err: | |
1245 | kfree(status); | |
1246 | return rc; | |
1247 | } | |
1248 | ||
8566ac8b BM |
1249 | static void grow_ple_window(struct kvm_vcpu *vcpu) |
1250 | { | |
1251 | struct vcpu_svm *svm = to_svm(vcpu); | |
1252 | struct vmcb_control_area *control = &svm->vmcb->control; | |
1253 | int old = control->pause_filter_count; | |
1254 | ||
1255 | control->pause_filter_count = __grow_ple_window(old, | |
1256 | pause_filter_count, | |
1257 | pause_filter_count_grow, | |
1258 | pause_filter_count_max); | |
1259 | ||
1260 | if (control->pause_filter_count != old) | |
1261 | mark_dirty(svm->vmcb, VMCB_INTERCEPTS); | |
1262 | ||
1263 | trace_kvm_ple_window_grow(vcpu->vcpu_id, | |
1264 | control->pause_filter_count, old); | |
1265 | } | |
1266 | ||
1267 | static void shrink_ple_window(struct kvm_vcpu *vcpu) | |
1268 | { | |
1269 | struct vcpu_svm *svm = to_svm(vcpu); | |
1270 | struct vmcb_control_area *control = &svm->vmcb->control; | |
1271 | int old = control->pause_filter_count; | |
1272 | ||
1273 | control->pause_filter_count = | |
1274 | __shrink_ple_window(old, | |
1275 | pause_filter_count, | |
1276 | pause_filter_count_shrink, | |
1277 | pause_filter_count); | |
1278 | if (control->pause_filter_count != old) | |
1279 | mark_dirty(svm->vmcb, VMCB_INTERCEPTS); | |
1280 | ||
1281 | trace_kvm_ple_window_shrink(vcpu->vcpu_id, | |
1282 | control->pause_filter_count, old); | |
1283 | } | |
1284 | ||
6aa8b732 AK |
1285 | static __init int svm_hardware_setup(void) |
1286 | { | |
1287 | int cpu; | |
1288 | struct page *iopm_pages; | |
f65c229c | 1289 | void *iopm_va; |
6aa8b732 AK |
1290 | int r; |
1291 | ||
6aa8b732 AK |
1292 | iopm_pages = alloc_pages(GFP_KERNEL, IOPM_ALLOC_ORDER); |
1293 | ||
1294 | if (!iopm_pages) | |
1295 | return -ENOMEM; | |
c8681339 AL |
1296 | |
1297 | iopm_va = page_address(iopm_pages); | |
1298 | memset(iopm_va, 0xff, PAGE_SIZE * (1 << IOPM_ALLOC_ORDER)); | |
6aa8b732 AK |
1299 | iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT; |
1300 | ||
323c3d80 JR |
1301 | init_msrpm_offsets(); |
1302 | ||
50a37eb4 JR |
1303 | if (boot_cpu_has(X86_FEATURE_NX)) |
1304 | kvm_enable_efer_bits(EFER_NX); | |
1305 | ||
1b2fd70c AG |
1306 | if (boot_cpu_has(X86_FEATURE_FXSR_OPT)) |
1307 | kvm_enable_efer_bits(EFER_FFXSR); | |
1308 | ||
92a1f12d | 1309 | if (boot_cpu_has(X86_FEATURE_TSCRATEMSR)) { |
92a1f12d | 1310 | kvm_has_tsc_control = true; |
bc9b961b HZ |
1311 | kvm_max_tsc_scaling_ratio = TSC_RATIO_MAX; |
1312 | kvm_tsc_scaling_ratio_frac_bits = 32; | |
92a1f12d JR |
1313 | } |
1314 | ||
8566ac8b BM |
1315 | /* Check for pause filtering support */ |
1316 | if (!boot_cpu_has(X86_FEATURE_PAUSEFILTER)) { | |
1317 | pause_filter_count = 0; | |
1318 | pause_filter_thresh = 0; | |
1319 | } else if (!boot_cpu_has(X86_FEATURE_PFTHRESHOLD)) { | |
1320 | pause_filter_thresh = 0; | |
1321 | } | |
1322 | ||
236de055 AG |
1323 | if (nested) { |
1324 | printk(KERN_INFO "kvm: Nested Virtualization enabled\n"); | |
eec4b140 | 1325 | kvm_enable_efer_bits(EFER_SVME | EFER_LMSLE); |
236de055 AG |
1326 | } |
1327 | ||
e9df0942 BS |
1328 | if (sev) { |
1329 | if (boot_cpu_has(X86_FEATURE_SEV) && | |
1330 | IS_ENABLED(CONFIG_KVM_AMD_SEV)) { | |
1331 | r = sev_hardware_setup(); | |
1332 | if (r) | |
1333 | sev = false; | |
1334 | } else { | |
1335 | sev = false; | |
1336 | } | |
1337 | } | |
1338 | ||
3230bb47 | 1339 | for_each_possible_cpu(cpu) { |
6aa8b732 AK |
1340 | r = svm_cpu_init(cpu); |
1341 | if (r) | |
f65c229c | 1342 | goto err; |
6aa8b732 | 1343 | } |
33bd6a0b | 1344 | |
2a6b20b8 | 1345 | if (!boot_cpu_has(X86_FEATURE_NPT)) |
e3da3acd JR |
1346 | npt_enabled = false; |
1347 | ||
6c7dac72 JR |
1348 | if (npt_enabled && !npt) { |
1349 | printk(KERN_INFO "kvm: Nested Paging disabled\n"); | |
1350 | npt_enabled = false; | |
1351 | } | |
1352 | ||
18552672 | 1353 | if (npt_enabled) { |
e3da3acd | 1354 | printk(KERN_INFO "kvm: Nested Paging enabled\n"); |
18552672 | 1355 | kvm_enable_tdp(); |
5f4cb662 JR |
1356 | } else |
1357 | kvm_disable_tdp(); | |
e3da3acd | 1358 | |
5b8abf1f SS |
1359 | if (avic) { |
1360 | if (!npt_enabled || | |
1361 | !boot_cpu_has(X86_FEATURE_AVIC) || | |
5881f737 | 1362 | !IS_ENABLED(CONFIG_X86_LOCAL_APIC)) { |
5b8abf1f | 1363 | avic = false; |
5881f737 | 1364 | } else { |
5b8abf1f | 1365 | pr_info("AVIC enabled\n"); |
5881f737 | 1366 | |
5881f737 SS |
1367 | amd_iommu_register_ga_log_notifier(&avic_ga_log_notifier); |
1368 | } | |
5b8abf1f | 1369 | } |
44a95dae | 1370 | |
89c8a498 JN |
1371 | if (vls) { |
1372 | if (!npt_enabled || | |
5442c269 | 1373 | !boot_cpu_has(X86_FEATURE_V_VMSAVE_VMLOAD) || |
89c8a498 JN |
1374 | !IS_ENABLED(CONFIG_X86_64)) { |
1375 | vls = false; | |
1376 | } else { | |
1377 | pr_info("Virtual VMLOAD VMSAVE supported\n"); | |
1378 | } | |
1379 | } | |
1380 | ||
640bd6e5 JN |
1381 | if (vgif) { |
1382 | if (!boot_cpu_has(X86_FEATURE_VGIF)) | |
1383 | vgif = false; | |
1384 | else | |
1385 | pr_info("Virtual GIF supported\n"); | |
1386 | } | |
1387 | ||
6aa8b732 AK |
1388 | return 0; |
1389 | ||
f65c229c | 1390 | err: |
6aa8b732 AK |
1391 | __free_pages(iopm_pages, IOPM_ALLOC_ORDER); |
1392 | iopm_base = 0; | |
1393 | return r; | |
1394 | } | |
1395 | ||
1396 | static __exit void svm_hardware_unsetup(void) | |
1397 | { | |
0da1db75 JR |
1398 | int cpu; |
1399 | ||
1654efcb BS |
1400 | if (svm_sev_enabled()) |
1401 | kfree(sev_asid_bitmap); | |
1402 | ||
3230bb47 | 1403 | for_each_possible_cpu(cpu) |
0da1db75 JR |
1404 | svm_cpu_uninit(cpu); |
1405 | ||
6aa8b732 | 1406 | __free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT), IOPM_ALLOC_ORDER); |
f65c229c | 1407 | iopm_base = 0; |
6aa8b732 AK |
1408 | } |
1409 | ||
1410 | static void init_seg(struct vmcb_seg *seg) | |
1411 | { | |
1412 | seg->selector = 0; | |
1413 | seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK | | |
e0231715 | 1414 | SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */ |
6aa8b732 AK |
1415 | seg->limit = 0xffff; |
1416 | seg->base = 0; | |
1417 | } | |
1418 | ||
1419 | static void init_sys_seg(struct vmcb_seg *seg, uint32_t type) | |
1420 | { | |
1421 | seg->selector = 0; | |
1422 | seg->attrib = SVM_SELECTOR_P_MASK | type; | |
1423 | seg->limit = 0xffff; | |
1424 | seg->base = 0; | |
1425 | } | |
1426 | ||
f4e1b3c8 ZA |
1427 | static void svm_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset) |
1428 | { | |
1429 | struct vcpu_svm *svm = to_svm(vcpu); | |
1430 | u64 g_tsc_offset = 0; | |
1431 | ||
2030753d | 1432 | if (is_guest_mode(vcpu)) { |
f4e1b3c8 ZA |
1433 | g_tsc_offset = svm->vmcb->control.tsc_offset - |
1434 | svm->nested.hsave->control.tsc_offset; | |
1435 | svm->nested.hsave->control.tsc_offset = offset; | |
489223ed YY |
1436 | } else |
1437 | trace_kvm_write_tsc_offset(vcpu->vcpu_id, | |
1438 | svm->vmcb->control.tsc_offset, | |
1439 | offset); | |
f4e1b3c8 ZA |
1440 | |
1441 | svm->vmcb->control.tsc_offset = offset + g_tsc_offset; | |
116a0a23 JR |
1442 | |
1443 | mark_dirty(svm->vmcb, VMCB_INTERCEPTS); | |
f4e1b3c8 ZA |
1444 | } |
1445 | ||
44a95dae SS |
1446 | static void avic_init_vmcb(struct vcpu_svm *svm) |
1447 | { | |
1448 | struct vmcb *vmcb = svm->vmcb; | |
81811c16 | 1449 | struct kvm_svm *kvm_svm = to_kvm_svm(svm->vcpu.kvm); |
d0ec49d4 | 1450 | phys_addr_t bpa = __sme_set(page_to_phys(svm->avic_backing_page)); |
81811c16 SC |
1451 | phys_addr_t lpa = __sme_set(page_to_phys(kvm_svm->avic_logical_id_table_page)); |
1452 | phys_addr_t ppa = __sme_set(page_to_phys(kvm_svm->avic_physical_id_table_page)); | |
44a95dae SS |
1453 | |
1454 | vmcb->control.avic_backing_page = bpa & AVIC_HPA_MASK; | |
1455 | vmcb->control.avic_logical_id = lpa & AVIC_HPA_MASK; | |
1456 | vmcb->control.avic_physical_id = ppa & AVIC_HPA_MASK; | |
1457 | vmcb->control.avic_physical_id |= AVIC_MAX_PHYSICAL_ID_COUNT; | |
1458 | vmcb->control.int_ctl |= AVIC_ENABLE_MASK; | |
44a95dae SS |
1459 | } |
1460 | ||
5690891b | 1461 | static void init_vmcb(struct vcpu_svm *svm) |
6aa8b732 | 1462 | { |
e6101a96 JR |
1463 | struct vmcb_control_area *control = &svm->vmcb->control; |
1464 | struct vmcb_save_area *save = &svm->vmcb->save; | |
6aa8b732 | 1465 | |
4ee546b4 | 1466 | svm->vcpu.arch.hflags = 0; |
bff78274 | 1467 | |
4ee546b4 RJ |
1468 | set_cr_intercept(svm, INTERCEPT_CR0_READ); |
1469 | set_cr_intercept(svm, INTERCEPT_CR3_READ); | |
1470 | set_cr_intercept(svm, INTERCEPT_CR4_READ); | |
1471 | set_cr_intercept(svm, INTERCEPT_CR0_WRITE); | |
1472 | set_cr_intercept(svm, INTERCEPT_CR3_WRITE); | |
1473 | set_cr_intercept(svm, INTERCEPT_CR4_WRITE); | |
3bbf3565 SS |
1474 | if (!kvm_vcpu_apicv_active(&svm->vcpu)) |
1475 | set_cr_intercept(svm, INTERCEPT_CR8_WRITE); | |
6aa8b732 | 1476 | |
5315c716 | 1477 | set_dr_intercepts(svm); |
6aa8b732 | 1478 | |
18c918c5 JR |
1479 | set_exception_intercept(svm, PF_VECTOR); |
1480 | set_exception_intercept(svm, UD_VECTOR); | |
1481 | set_exception_intercept(svm, MC_VECTOR); | |
54a20552 | 1482 | set_exception_intercept(svm, AC_VECTOR); |
cbdb967a | 1483 | set_exception_intercept(svm, DB_VECTOR); |
9718420e LA |
1484 | /* |
1485 | * Guest access to VMware backdoor ports could legitimately | |
1486 | * trigger #GP because of TSS I/O permission bitmap. | |
1487 | * We intercept those #GP and allow access to them anyway | |
1488 | * as VMware does. | |
1489 | */ | |
1490 | if (enable_vmware_backdoor) | |
1491 | set_exception_intercept(svm, GP_VECTOR); | |
6aa8b732 | 1492 | |
8a05a1b8 JR |
1493 | set_intercept(svm, INTERCEPT_INTR); |
1494 | set_intercept(svm, INTERCEPT_NMI); | |
1495 | set_intercept(svm, INTERCEPT_SMI); | |
1496 | set_intercept(svm, INTERCEPT_SELECTIVE_CR0); | |
332b56e4 | 1497 | set_intercept(svm, INTERCEPT_RDPMC); |
8a05a1b8 JR |
1498 | set_intercept(svm, INTERCEPT_CPUID); |
1499 | set_intercept(svm, INTERCEPT_INVD); | |
8a05a1b8 JR |
1500 | set_intercept(svm, INTERCEPT_INVLPG); |
1501 | set_intercept(svm, INTERCEPT_INVLPGA); | |
1502 | set_intercept(svm, INTERCEPT_IOIO_PROT); | |
1503 | set_intercept(svm, INTERCEPT_MSR_PROT); | |
1504 | set_intercept(svm, INTERCEPT_TASK_SWITCH); | |
1505 | set_intercept(svm, INTERCEPT_SHUTDOWN); | |
1506 | set_intercept(svm, INTERCEPT_VMRUN); | |
1507 | set_intercept(svm, INTERCEPT_VMMCALL); | |
1508 | set_intercept(svm, INTERCEPT_VMLOAD); | |
1509 | set_intercept(svm, INTERCEPT_VMSAVE); | |
1510 | set_intercept(svm, INTERCEPT_STGI); | |
1511 | set_intercept(svm, INTERCEPT_CLGI); | |
1512 | set_intercept(svm, INTERCEPT_SKINIT); | |
1513 | set_intercept(svm, INTERCEPT_WBINVD); | |
81dd35d4 | 1514 | set_intercept(svm, INTERCEPT_XSETBV); |
7607b717 | 1515 | set_intercept(svm, INTERCEPT_RSM); |
6aa8b732 | 1516 | |
4d5422ce | 1517 | if (!kvm_mwait_in_guest(svm->vcpu.kvm)) { |
668fffa3 MT |
1518 | set_intercept(svm, INTERCEPT_MONITOR); |
1519 | set_intercept(svm, INTERCEPT_MWAIT); | |
1520 | } | |
1521 | ||
caa057a2 WL |
1522 | if (!kvm_hlt_in_guest(svm->vcpu.kvm)) |
1523 | set_intercept(svm, INTERCEPT_HLT); | |
1524 | ||
d0ec49d4 TL |
1525 | control->iopm_base_pa = __sme_set(iopm_base); |
1526 | control->msrpm_base_pa = __sme_set(__pa(svm->msrpm)); | |
6aa8b732 AK |
1527 | control->int_ctl = V_INTR_MASKING_MASK; |
1528 | ||
1529 | init_seg(&save->es); | |
1530 | init_seg(&save->ss); | |
1531 | init_seg(&save->ds); | |
1532 | init_seg(&save->fs); | |
1533 | init_seg(&save->gs); | |
1534 | ||
1535 | save->cs.selector = 0xf000; | |
04b66839 | 1536 | save->cs.base = 0xffff0000; |
6aa8b732 AK |
1537 | /* Executable/Readable Code Segment */ |
1538 | save->cs.attrib = SVM_SELECTOR_READ_MASK | SVM_SELECTOR_P_MASK | | |
1539 | SVM_SELECTOR_S_MASK | SVM_SELECTOR_CODE_MASK; | |
1540 | save->cs.limit = 0xffff; | |
6aa8b732 AK |
1541 | |
1542 | save->gdtr.limit = 0xffff; | |
1543 | save->idtr.limit = 0xffff; | |
1544 | ||
1545 | init_sys_seg(&save->ldtr, SEG_TYPE_LDT); | |
1546 | init_sys_seg(&save->tr, SEG_TYPE_BUSY_TSS16); | |
1547 | ||
5690891b | 1548 | svm_set_efer(&svm->vcpu, 0); |
d77c26fc | 1549 | save->dr6 = 0xffff0ff0; |
f6e78475 | 1550 | kvm_set_rflags(&svm->vcpu, 2); |
6aa8b732 | 1551 | save->rip = 0x0000fff0; |
5fdbf976 | 1552 | svm->vcpu.arch.regs[VCPU_REGS_RIP] = save->rip; |
6aa8b732 | 1553 | |
e0231715 | 1554 | /* |
18fa000a | 1555 | * svm_set_cr0() sets PG and WP and clears NW and CD on save->cr0. |
d28bc9dd | 1556 | * It also updates the guest-visible cr0 value. |
6aa8b732 | 1557 | */ |
79a8059d | 1558 | svm_set_cr0(&svm->vcpu, X86_CR0_NW | X86_CR0_CD | X86_CR0_ET); |
ebae871a | 1559 | kvm_mmu_reset_context(&svm->vcpu); |
18fa000a | 1560 | |
66aee91a | 1561 | save->cr4 = X86_CR4_PAE; |
6aa8b732 | 1562 | /* rdx = ?? */ |
709ddebf JR |
1563 | |
1564 | if (npt_enabled) { | |
1565 | /* Setup VMCB for Nested Paging */ | |
cea3a19b | 1566 | control->nested_ctl |= SVM_NESTED_CTL_NP_ENABLE; |
8a05a1b8 | 1567 | clr_intercept(svm, INTERCEPT_INVLPG); |
18c918c5 | 1568 | clr_exception_intercept(svm, PF_VECTOR); |
4ee546b4 RJ |
1569 | clr_cr_intercept(svm, INTERCEPT_CR3_READ); |
1570 | clr_cr_intercept(svm, INTERCEPT_CR3_WRITE); | |
74545705 | 1571 | save->g_pat = svm->vcpu.arch.pat; |
709ddebf JR |
1572 | save->cr3 = 0; |
1573 | save->cr4 = 0; | |
1574 | } | |
f40f6a45 | 1575 | svm->asid_generation = 0; |
1371d904 | 1576 | |
e6aa9abd | 1577 | svm->nested.vmcb = 0; |
2af9194d JR |
1578 | svm->vcpu.arch.hflags = 0; |
1579 | ||
8566ac8b BM |
1580 | if (pause_filter_count) { |
1581 | control->pause_filter_count = pause_filter_count; | |
1582 | if (pause_filter_thresh) | |
1583 | control->pause_filter_thresh = pause_filter_thresh; | |
8a05a1b8 | 1584 | set_intercept(svm, INTERCEPT_PAUSE); |
8566ac8b BM |
1585 | } else { |
1586 | clr_intercept(svm, INTERCEPT_PAUSE); | |
565d0998 ML |
1587 | } |
1588 | ||
67034bb9 | 1589 | if (kvm_vcpu_apicv_active(&svm->vcpu)) |
44a95dae SS |
1590 | avic_init_vmcb(svm); |
1591 | ||
89c8a498 JN |
1592 | /* |
1593 | * If hardware supports Virtual VMLOAD VMSAVE then enable it | |
1594 | * in VMCB and clear intercepts to avoid #VMEXIT. | |
1595 | */ | |
1596 | if (vls) { | |
1597 | clr_intercept(svm, INTERCEPT_VMLOAD); | |
1598 | clr_intercept(svm, INTERCEPT_VMSAVE); | |
1599 | svm->vmcb->control.virt_ext |= VIRTUAL_VMLOAD_VMSAVE_ENABLE_MASK; | |
1600 | } | |
1601 | ||
640bd6e5 JN |
1602 | if (vgif) { |
1603 | clr_intercept(svm, INTERCEPT_STGI); | |
1604 | clr_intercept(svm, INTERCEPT_CLGI); | |
1605 | svm->vmcb->control.int_ctl |= V_GIF_ENABLE_MASK; | |
1606 | } | |
1607 | ||
35c6f649 | 1608 | if (sev_guest(svm->vcpu.kvm)) { |
1654efcb | 1609 | svm->vmcb->control.nested_ctl |= SVM_NESTED_CTL_SEV_ENABLE; |
35c6f649 BS |
1610 | clr_exception_intercept(svm, UD_VECTOR); |
1611 | } | |
1654efcb | 1612 | |
8d28fec4 RJ |
1613 | mark_all_dirty(svm->vmcb); |
1614 | ||
2af9194d | 1615 | enable_gif(svm); |
44a95dae SS |
1616 | |
1617 | } | |
1618 | ||
d3e7dec0 DC |
1619 | static u64 *avic_get_physical_id_entry(struct kvm_vcpu *vcpu, |
1620 | unsigned int index) | |
44a95dae SS |
1621 | { |
1622 | u64 *avic_physical_id_table; | |
81811c16 | 1623 | struct kvm_svm *kvm_svm = to_kvm_svm(vcpu->kvm); |
44a95dae SS |
1624 | |
1625 | if (index >= AVIC_MAX_PHYSICAL_ID_COUNT) | |
1626 | return NULL; | |
1627 | ||
81811c16 | 1628 | avic_physical_id_table = page_address(kvm_svm->avic_physical_id_table_page); |
44a95dae SS |
1629 | |
1630 | return &avic_physical_id_table[index]; | |
1631 | } | |
1632 | ||
1633 | /** | |
1634 | * Note: | |
1635 | * AVIC hardware walks the nested page table to check permissions, | |
1636 | * but does not use the SPA address specified in the leaf page | |
1637 | * table entry since it uses address in the AVIC_BACKING_PAGE pointer | |
1638 | * field of the VMCB. Therefore, we set up the | |
1639 | * APIC_ACCESS_PAGE_PRIVATE_MEMSLOT (4KB) here. | |
1640 | */ | |
1641 | static int avic_init_access_page(struct kvm_vcpu *vcpu) | |
1642 | { | |
1643 | struct kvm *kvm = vcpu->kvm; | |
1644 | int ret; | |
1645 | ||
1646 | if (kvm->arch.apic_access_page_done) | |
1647 | return 0; | |
1648 | ||
1649 | ret = x86_set_memory_region(kvm, | |
1650 | APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, | |
1651 | APIC_DEFAULT_PHYS_BASE, | |
1652 | PAGE_SIZE); | |
1653 | if (ret) | |
1654 | return ret; | |
1655 | ||
1656 | kvm->arch.apic_access_page_done = true; | |
1657 | return 0; | |
1658 | } | |
1659 | ||
1660 | static int avic_init_backing_page(struct kvm_vcpu *vcpu) | |
1661 | { | |
1662 | int ret; | |
1663 | u64 *entry, new_entry; | |
1664 | int id = vcpu->vcpu_id; | |
1665 | struct vcpu_svm *svm = to_svm(vcpu); | |
1666 | ||
1667 | ret = avic_init_access_page(vcpu); | |
1668 | if (ret) | |
1669 | return ret; | |
1670 | ||
1671 | if (id >= AVIC_MAX_PHYSICAL_ID_COUNT) | |
1672 | return -EINVAL; | |
1673 | ||
1674 | if (!svm->vcpu.arch.apic->regs) | |
1675 | return -EINVAL; | |
1676 | ||
1677 | svm->avic_backing_page = virt_to_page(svm->vcpu.arch.apic->regs); | |
1678 | ||
1679 | /* Setting AVIC backing page address in the phy APIC ID table */ | |
1680 | entry = avic_get_physical_id_entry(vcpu, id); | |
1681 | if (!entry) | |
1682 | return -EINVAL; | |
1683 | ||
1684 | new_entry = READ_ONCE(*entry); | |
d0ec49d4 TL |
1685 | new_entry = __sme_set((page_to_phys(svm->avic_backing_page) & |
1686 | AVIC_PHYSICAL_ID_ENTRY_BACKING_PAGE_MASK) | | |
1687 | AVIC_PHYSICAL_ID_ENTRY_VALID_MASK); | |
44a95dae SS |
1688 | WRITE_ONCE(*entry, new_entry); |
1689 | ||
1690 | svm->avic_physical_id_cache = entry; | |
1691 | ||
1692 | return 0; | |
1693 | } | |
1694 | ||
1654efcb BS |
1695 | static void __sev_asid_free(int asid) |
1696 | { | |
70cd94e6 BS |
1697 | struct svm_cpu_data *sd; |
1698 | int cpu, pos; | |
1654efcb BS |
1699 | |
1700 | pos = asid - 1; | |
1701 | clear_bit(pos, sev_asid_bitmap); | |
70cd94e6 BS |
1702 | |
1703 | for_each_possible_cpu(cpu) { | |
1704 | sd = per_cpu(svm_data, cpu); | |
1705 | sd->sev_vmcbs[pos] = NULL; | |
1706 | } | |
1654efcb BS |
1707 | } |
1708 | ||
1709 | static void sev_asid_free(struct kvm *kvm) | |
1710 | { | |
81811c16 | 1711 | struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info; |
1654efcb BS |
1712 | |
1713 | __sev_asid_free(sev->asid); | |
1714 | } | |
1715 | ||
59414c98 BS |
1716 | static void sev_unbind_asid(struct kvm *kvm, unsigned int handle) |
1717 | { | |
1718 | struct sev_data_decommission *decommission; | |
1719 | struct sev_data_deactivate *data; | |
1720 | ||
1721 | if (!handle) | |
1722 | return; | |
1723 | ||
1724 | data = kzalloc(sizeof(*data), GFP_KERNEL); | |
1725 | if (!data) | |
1726 | return; | |
1727 | ||
1728 | /* deactivate handle */ | |
1729 | data->handle = handle; | |
1730 | sev_guest_deactivate(data, NULL); | |
1731 | ||
1732 | wbinvd_on_all_cpus(); | |
1733 | sev_guest_df_flush(NULL); | |
1734 | kfree(data); | |
1735 | ||
1736 | decommission = kzalloc(sizeof(*decommission), GFP_KERNEL); | |
1737 | if (!decommission) | |
1738 | return; | |
1739 | ||
1740 | /* decommission handle */ | |
1741 | decommission->handle = handle; | |
1742 | sev_guest_decommission(decommission, NULL); | |
1743 | ||
1744 | kfree(decommission); | |
1745 | } | |
1746 | ||
89c50580 BS |
1747 | static struct page **sev_pin_memory(struct kvm *kvm, unsigned long uaddr, |
1748 | unsigned long ulen, unsigned long *n, | |
1749 | int write) | |
1750 | { | |
81811c16 | 1751 | struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info; |
89c50580 BS |
1752 | unsigned long npages, npinned, size; |
1753 | unsigned long locked, lock_limit; | |
1754 | struct page **pages; | |
1755 | int first, last; | |
1756 | ||
1757 | /* Calculate number of pages. */ | |
1758 | first = (uaddr & PAGE_MASK) >> PAGE_SHIFT; | |
1759 | last = ((uaddr + ulen - 1) & PAGE_MASK) >> PAGE_SHIFT; | |
1760 | npages = (last - first + 1); | |
1761 | ||
1762 | locked = sev->pages_locked + npages; | |
1763 | lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
1764 | if (locked > lock_limit && !capable(CAP_IPC_LOCK)) { | |
1765 | pr_err("SEV: %lu locked pages exceed the lock limit of %lu.\n", locked, lock_limit); | |
1766 | return NULL; | |
1767 | } | |
1768 | ||
1769 | /* Avoid using vmalloc for smaller buffers. */ | |
1770 | size = npages * sizeof(struct page *); | |
1771 | if (size > PAGE_SIZE) | |
1772 | pages = vmalloc(size); | |
1773 | else | |
1774 | pages = kmalloc(size, GFP_KERNEL); | |
1775 | ||
1776 | if (!pages) | |
1777 | return NULL; | |
1778 | ||
1779 | /* Pin the user virtual address. */ | |
1780 | npinned = get_user_pages_fast(uaddr, npages, write ? FOLL_WRITE : 0, pages); | |
1781 | if (npinned != npages) { | |
1782 | pr_err("SEV: Failure locking %lu pages.\n", npages); | |
1783 | goto err; | |
1784 | } | |
1785 | ||
1786 | *n = npages; | |
1787 | sev->pages_locked = locked; | |
1788 | ||
1789 | return pages; | |
1790 | ||
1791 | err: | |
1792 | if (npinned > 0) | |
1793 | release_pages(pages, npinned); | |
1794 | ||
1795 | kvfree(pages); | |
1796 | return NULL; | |
1797 | } | |
1798 | ||
1799 | static void sev_unpin_memory(struct kvm *kvm, struct page **pages, | |
1800 | unsigned long npages) | |
1801 | { | |
81811c16 | 1802 | struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info; |
89c50580 BS |
1803 | |
1804 | release_pages(pages, npages); | |
1805 | kvfree(pages); | |
1806 | sev->pages_locked -= npages; | |
1807 | } | |
1808 | ||
1809 | static void sev_clflush_pages(struct page *pages[], unsigned long npages) | |
1810 | { | |
1811 | uint8_t *page_virtual; | |
1812 | unsigned long i; | |
1813 | ||
1814 | if (npages == 0 || pages == NULL) | |
1815 | return; | |
1816 | ||
1817 | for (i = 0; i < npages; i++) { | |
1818 | page_virtual = kmap_atomic(pages[i]); | |
1819 | clflush_cache_range(page_virtual, PAGE_SIZE); | |
1820 | kunmap_atomic(page_virtual); | |
1821 | } | |
1822 | } | |
1823 | ||
1e80fdc0 BS |
1824 | static void __unregister_enc_region_locked(struct kvm *kvm, |
1825 | struct enc_region *region) | |
1826 | { | |
1827 | /* | |
1828 | * The guest may change the memory encryption attribute from C=0 -> C=1 | |
1829 | * or vice versa for this memory range. Lets make sure caches are | |
1830 | * flushed to ensure that guest data gets written into memory with | |
1831 | * correct C-bit. | |
1832 | */ | |
1833 | sev_clflush_pages(region->pages, region->npages); | |
1834 | ||
1835 | sev_unpin_memory(kvm, region->pages, region->npages); | |
1836 | list_del(®ion->list); | |
1837 | kfree(region); | |
1838 | } | |
1839 | ||
434a1e94 SC |
1840 | static struct kvm *svm_vm_alloc(void) |
1841 | { | |
81811c16 SC |
1842 | struct kvm_svm *kvm_svm = kzalloc(sizeof(struct kvm_svm), GFP_KERNEL); |
1843 | return &kvm_svm->kvm; | |
434a1e94 SC |
1844 | } |
1845 | ||
1846 | static void svm_vm_free(struct kvm *kvm) | |
1847 | { | |
81811c16 | 1848 | kfree(to_kvm_svm(kvm)); |
434a1e94 SC |
1849 | } |
1850 | ||
1654efcb BS |
1851 | static void sev_vm_destroy(struct kvm *kvm) |
1852 | { | |
81811c16 | 1853 | struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info; |
1e80fdc0 BS |
1854 | struct list_head *head = &sev->regions_list; |
1855 | struct list_head *pos, *q; | |
59414c98 | 1856 | |
1654efcb BS |
1857 | if (!sev_guest(kvm)) |
1858 | return; | |
1859 | ||
1e80fdc0 BS |
1860 | mutex_lock(&kvm->lock); |
1861 | ||
1862 | /* | |
1863 | * if userspace was terminated before unregistering the memory regions | |
1864 | * then lets unpin all the registered memory. | |
1865 | */ | |
1866 | if (!list_empty(head)) { | |
1867 | list_for_each_safe(pos, q, head) { | |
1868 | __unregister_enc_region_locked(kvm, | |
1869 | list_entry(pos, struct enc_region, list)); | |
1870 | } | |
1871 | } | |
1872 | ||
1873 | mutex_unlock(&kvm->lock); | |
1874 | ||
59414c98 | 1875 | sev_unbind_asid(kvm, sev->handle); |
1654efcb BS |
1876 | sev_asid_free(kvm); |
1877 | } | |
1878 | ||
44a95dae SS |
1879 | static void avic_vm_destroy(struct kvm *kvm) |
1880 | { | |
5881f737 | 1881 | unsigned long flags; |
81811c16 | 1882 | struct kvm_svm *kvm_svm = to_kvm_svm(kvm); |
44a95dae | 1883 | |
3863dff0 DV |
1884 | if (!avic) |
1885 | return; | |
1886 | ||
81811c16 SC |
1887 | if (kvm_svm->avic_logical_id_table_page) |
1888 | __free_page(kvm_svm->avic_logical_id_table_page); | |
1889 | if (kvm_svm->avic_physical_id_table_page) | |
1890 | __free_page(kvm_svm->avic_physical_id_table_page); | |
5881f737 SS |
1891 | |
1892 | spin_lock_irqsave(&svm_vm_data_hash_lock, flags); | |
81811c16 | 1893 | hash_del(&kvm_svm->hnode); |
5881f737 | 1894 | spin_unlock_irqrestore(&svm_vm_data_hash_lock, flags); |
44a95dae SS |
1895 | } |
1896 | ||
1654efcb BS |
1897 | static void svm_vm_destroy(struct kvm *kvm) |
1898 | { | |
1899 | avic_vm_destroy(kvm); | |
1900 | sev_vm_destroy(kvm); | |
1901 | } | |
1902 | ||
44a95dae SS |
1903 | static int avic_vm_init(struct kvm *kvm) |
1904 | { | |
5881f737 | 1905 | unsigned long flags; |
3f0d4db7 | 1906 | int err = -ENOMEM; |
81811c16 SC |
1907 | struct kvm_svm *kvm_svm = to_kvm_svm(kvm); |
1908 | struct kvm_svm *k2; | |
44a95dae SS |
1909 | struct page *p_page; |
1910 | struct page *l_page; | |
3f0d4db7 | 1911 | u32 vm_id; |
44a95dae SS |
1912 | |
1913 | if (!avic) | |
1914 | return 0; | |
1915 | ||
1916 | /* Allocating physical APIC ID table (4KB) */ | |
1917 | p_page = alloc_page(GFP_KERNEL); | |
1918 | if (!p_page) | |
1919 | goto free_avic; | |
1920 | ||
81811c16 | 1921 | kvm_svm->avic_physical_id_table_page = p_page; |
44a95dae SS |
1922 | clear_page(page_address(p_page)); |
1923 | ||
1924 | /* Allocating logical APIC ID table (4KB) */ | |
1925 | l_page = alloc_page(GFP_KERNEL); | |
1926 | if (!l_page) | |
1927 | goto free_avic; | |
1928 | ||
81811c16 | 1929 | kvm_svm->avic_logical_id_table_page = l_page; |
44a95dae SS |
1930 | clear_page(page_address(l_page)); |
1931 | ||
5881f737 | 1932 | spin_lock_irqsave(&svm_vm_data_hash_lock, flags); |
3f0d4db7 DV |
1933 | again: |
1934 | vm_id = next_vm_id = (next_vm_id + 1) & AVIC_VM_ID_MASK; | |
1935 | if (vm_id == 0) { /* id is 1-based, zero is not okay */ | |
1936 | next_vm_id_wrapped = 1; | |
1937 | goto again; | |
1938 | } | |
1939 | /* Is it still in use? Only possible if wrapped at least once */ | |
1940 | if (next_vm_id_wrapped) { | |
81811c16 SC |
1941 | hash_for_each_possible(svm_vm_data_hash, k2, hnode, vm_id) { |
1942 | if (k2->avic_vm_id == vm_id) | |
3f0d4db7 DV |
1943 | goto again; |
1944 | } | |
1945 | } | |
81811c16 SC |
1946 | kvm_svm->avic_vm_id = vm_id; |
1947 | hash_add(svm_vm_data_hash, &kvm_svm->hnode, kvm_svm->avic_vm_id); | |
5881f737 SS |
1948 | spin_unlock_irqrestore(&svm_vm_data_hash_lock, flags); |
1949 | ||
44a95dae SS |
1950 | return 0; |
1951 | ||
1952 | free_avic: | |
1953 | avic_vm_destroy(kvm); | |
1954 | return err; | |
6aa8b732 AK |
1955 | } |
1956 | ||
411b44ba SS |
1957 | static inline int |
1958 | avic_update_iommu_vcpu_affinity(struct kvm_vcpu *vcpu, int cpu, bool r) | |
8221c137 | 1959 | { |
411b44ba SS |
1960 | int ret = 0; |
1961 | unsigned long flags; | |
1962 | struct amd_svm_iommu_ir *ir; | |
8221c137 SS |
1963 | struct vcpu_svm *svm = to_svm(vcpu); |
1964 | ||
411b44ba SS |
1965 | if (!kvm_arch_has_assigned_device(vcpu->kvm)) |
1966 | return 0; | |
8221c137 | 1967 | |
411b44ba SS |
1968 | /* |
1969 | * Here, we go through the per-vcpu ir_list to update all existing | |
1970 | * interrupt remapping table entry targeting this vcpu. | |
1971 | */ | |
1972 | spin_lock_irqsave(&svm->ir_list_lock, flags); | |
8221c137 | 1973 | |
411b44ba SS |
1974 | if (list_empty(&svm->ir_list)) |
1975 | goto out; | |
8221c137 | 1976 | |
411b44ba SS |
1977 | list_for_each_entry(ir, &svm->ir_list, node) { |
1978 | ret = amd_iommu_update_ga(cpu, r, ir->data); | |
1979 | if (ret) | |
1980 | break; | |
1981 | } | |
1982 | out: | |
1983 | spin_unlock_irqrestore(&svm->ir_list_lock, flags); | |
1984 | return ret; | |
8221c137 SS |
1985 | } |
1986 | ||
1987 | static void avic_vcpu_load(struct kvm_vcpu *vcpu, int cpu) | |
1988 | { | |
1989 | u64 entry; | |
1990 | /* ID = 0xff (broadcast), ID > 0xff (reserved) */ | |
7d669f50 | 1991 | int h_physical_id = kvm_cpu_get_apicid(cpu); |
8221c137 SS |
1992 | struct vcpu_svm *svm = to_svm(vcpu); |
1993 | ||
1994 | if (!kvm_vcpu_apicv_active(vcpu)) | |
1995 | return; | |
1996 | ||
1997 | if (WARN_ON(h_physical_id >= AVIC_MAX_PHYSICAL_ID_COUNT)) | |
1998 | return; | |
1999 | ||
2000 | entry = READ_ONCE(*(svm->avic_physical_id_cache)); | |
2001 | WARN_ON(entry & AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK); | |
2002 | ||
2003 | entry &= ~AVIC_PHYSICAL_ID_ENTRY_HOST_PHYSICAL_ID_MASK; | |
2004 | entry |= (h_physical_id & AVIC_PHYSICAL_ID_ENTRY_HOST_PHYSICAL_ID_MASK); | |
2005 | ||
2006 | entry &= ~AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK; | |
2007 | if (svm->avic_is_running) | |
2008 | entry |= AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK; | |
2009 | ||
2010 | WRITE_ONCE(*(svm->avic_physical_id_cache), entry); | |
411b44ba SS |
2011 | avic_update_iommu_vcpu_affinity(vcpu, h_physical_id, |
2012 | svm->avic_is_running); | |
8221c137 SS |
2013 | } |
2014 | ||
2015 | static void avic_vcpu_put(struct kvm_vcpu *vcpu) | |
2016 | { | |
2017 | u64 entry; | |
2018 | struct vcpu_svm *svm = to_svm(vcpu); | |
2019 | ||
2020 | if (!kvm_vcpu_apicv_active(vcpu)) | |
2021 | return; | |
2022 | ||
2023 | entry = READ_ONCE(*(svm->avic_physical_id_cache)); | |
411b44ba SS |
2024 | if (entry & AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK) |
2025 | avic_update_iommu_vcpu_affinity(vcpu, -1, 0); | |
2026 | ||
8221c137 SS |
2027 | entry &= ~AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK; |
2028 | WRITE_ONCE(*(svm->avic_physical_id_cache), entry); | |
6aa8b732 AK |
2029 | } |
2030 | ||
411b44ba SS |
2031 | /** |
2032 | * This function is called during VCPU halt/unhalt. | |
2033 | */ | |
2034 | static void avic_set_running(struct kvm_vcpu *vcpu, bool is_run) | |
2035 | { | |
2036 | struct vcpu_svm *svm = to_svm(vcpu); | |
2037 | ||
2038 | svm->avic_is_running = is_run; | |
2039 | if (is_run) | |
2040 | avic_vcpu_load(vcpu, vcpu->cpu); | |
2041 | else | |
2042 | avic_vcpu_put(vcpu); | |
2043 | } | |
2044 | ||
d28bc9dd | 2045 | static void svm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event) |
04d2cc77 AK |
2046 | { |
2047 | struct vcpu_svm *svm = to_svm(vcpu); | |
66f7b72e JS |
2048 | u32 dummy; |
2049 | u32 eax = 1; | |
04d2cc77 | 2050 | |
518e7b94 | 2051 | vcpu->arch.microcode_version = 0x01000065; |
b2ac58f9 KA |
2052 | svm->spec_ctrl = 0; |
2053 | ||
d28bc9dd NA |
2054 | if (!init_event) { |
2055 | svm->vcpu.arch.apic_base = APIC_DEFAULT_PHYS_BASE | | |
2056 | MSR_IA32_APICBASE_ENABLE; | |
2057 | if (kvm_vcpu_is_reset_bsp(&svm->vcpu)) | |
2058 | svm->vcpu.arch.apic_base |= MSR_IA32_APICBASE_BSP; | |
2059 | } | |
5690891b | 2060 | init_vmcb(svm); |
70433389 | 2061 | |
e911eb3b | 2062 | kvm_cpuid(vcpu, &eax, &dummy, &dummy, &dummy, true); |
66f7b72e | 2063 | kvm_register_write(vcpu, VCPU_REGS_RDX, eax); |
44a95dae SS |
2064 | |
2065 | if (kvm_vcpu_apicv_active(vcpu) && !init_event) | |
2066 | avic_update_vapic_bar(svm, APIC_DEFAULT_PHYS_BASE); | |
04d2cc77 AK |
2067 | } |
2068 | ||
dfa20099 SS |
2069 | static int avic_init_vcpu(struct vcpu_svm *svm) |
2070 | { | |
2071 | int ret; | |
2072 | ||
67034bb9 | 2073 | if (!kvm_vcpu_apicv_active(&svm->vcpu)) |
dfa20099 SS |
2074 | return 0; |
2075 | ||
2076 | ret = avic_init_backing_page(&svm->vcpu); | |
2077 | if (ret) | |
2078 | return ret; | |
2079 | ||
2080 | INIT_LIST_HEAD(&svm->ir_list); | |
2081 | spin_lock_init(&svm->ir_list_lock); | |
2082 | ||
2083 | return ret; | |
2084 | } | |
2085 | ||
fb3f0f51 | 2086 | static struct kvm_vcpu *svm_create_vcpu(struct kvm *kvm, unsigned int id) |
6aa8b732 | 2087 | { |
a2fa3e9f | 2088 | struct vcpu_svm *svm; |
6aa8b732 | 2089 | struct page *page; |
f65c229c | 2090 | struct page *msrpm_pages; |
b286d5d8 | 2091 | struct page *hsave_page; |
3d6368ef | 2092 | struct page *nested_msrpm_pages; |
fb3f0f51 | 2093 | int err; |
6aa8b732 | 2094 | |
c16f862d | 2095 | svm = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL); |
fb3f0f51 RR |
2096 | if (!svm) { |
2097 | err = -ENOMEM; | |
2098 | goto out; | |
2099 | } | |
2100 | ||
2101 | err = kvm_vcpu_init(&svm->vcpu, kvm, id); | |
2102 | if (err) | |
2103 | goto free_svm; | |
2104 | ||
b7af4043 | 2105 | err = -ENOMEM; |
6aa8b732 | 2106 | page = alloc_page(GFP_KERNEL); |
b7af4043 | 2107 | if (!page) |
fb3f0f51 | 2108 | goto uninit; |
6aa8b732 | 2109 | |
f65c229c JR |
2110 | msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER); |
2111 | if (!msrpm_pages) | |
b7af4043 | 2112 | goto free_page1; |
3d6368ef AG |
2113 | |
2114 | nested_msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER); | |
2115 | if (!nested_msrpm_pages) | |
b7af4043 | 2116 | goto free_page2; |
f65c229c | 2117 | |
b286d5d8 AG |
2118 | hsave_page = alloc_page(GFP_KERNEL); |
2119 | if (!hsave_page) | |
b7af4043 TY |
2120 | goto free_page3; |
2121 | ||
dfa20099 SS |
2122 | err = avic_init_vcpu(svm); |
2123 | if (err) | |
2124 | goto free_page4; | |
44a95dae | 2125 | |
8221c137 SS |
2126 | /* We initialize this flag to true to make sure that the is_running |
2127 | * bit would be set the first time the vcpu is loaded. | |
2128 | */ | |
2129 | svm->avic_is_running = true; | |
2130 | ||
e6aa9abd | 2131 | svm->nested.hsave = page_address(hsave_page); |
b286d5d8 | 2132 | |
b7af4043 TY |
2133 | svm->msrpm = page_address(msrpm_pages); |
2134 | svm_vcpu_init_msrpm(svm->msrpm); | |
2135 | ||
e6aa9abd | 2136 | svm->nested.msrpm = page_address(nested_msrpm_pages); |
323c3d80 | 2137 | svm_vcpu_init_msrpm(svm->nested.msrpm); |
3d6368ef | 2138 | |
a2fa3e9f GH |
2139 | svm->vmcb = page_address(page); |
2140 | clear_page(svm->vmcb); | |
d0ec49d4 | 2141 | svm->vmcb_pa = __sme_set(page_to_pfn(page) << PAGE_SHIFT); |
a2fa3e9f | 2142 | svm->asid_generation = 0; |
5690891b | 2143 | init_vmcb(svm); |
6aa8b732 | 2144 | |
2b036c6b BO |
2145 | svm_init_osvw(&svm->vcpu); |
2146 | ||
fb3f0f51 | 2147 | return &svm->vcpu; |
36241b8c | 2148 | |
44a95dae SS |
2149 | free_page4: |
2150 | __free_page(hsave_page); | |
b7af4043 TY |
2151 | free_page3: |
2152 | __free_pages(nested_msrpm_pages, MSRPM_ALLOC_ORDER); | |
2153 | free_page2: | |
2154 | __free_pages(msrpm_pages, MSRPM_ALLOC_ORDER); | |
2155 | free_page1: | |
2156 | __free_page(page); | |
fb3f0f51 RR |
2157 | uninit: |
2158 | kvm_vcpu_uninit(&svm->vcpu); | |
2159 | free_svm: | |
a4770347 | 2160 | kmem_cache_free(kvm_vcpu_cache, svm); |
fb3f0f51 RR |
2161 | out: |
2162 | return ERR_PTR(err); | |
6aa8b732 AK |
2163 | } |
2164 | ||
2165 | static void svm_free_vcpu(struct kvm_vcpu *vcpu) | |
2166 | { | |
a2fa3e9f GH |
2167 | struct vcpu_svm *svm = to_svm(vcpu); |
2168 | ||
d0ec49d4 | 2169 | __free_page(pfn_to_page(__sme_clr(svm->vmcb_pa) >> PAGE_SHIFT)); |
f65c229c | 2170 | __free_pages(virt_to_page(svm->msrpm), MSRPM_ALLOC_ORDER); |
e6aa9abd JR |
2171 | __free_page(virt_to_page(svm->nested.hsave)); |
2172 | __free_pages(virt_to_page(svm->nested.msrpm), MSRPM_ALLOC_ORDER); | |
fb3f0f51 | 2173 | kvm_vcpu_uninit(vcpu); |
a4770347 | 2174 | kmem_cache_free(kvm_vcpu_cache, svm); |
15d45071 AR |
2175 | /* |
2176 | * The vmcb page can be recycled, causing a false negative in | |
2177 | * svm_vcpu_load(). So do a full IBPB now. | |
2178 | */ | |
2179 | indirect_branch_prediction_barrier(); | |
6aa8b732 AK |
2180 | } |
2181 | ||
15ad7146 | 2182 | static void svm_vcpu_load(struct kvm_vcpu *vcpu, int cpu) |
6aa8b732 | 2183 | { |
a2fa3e9f | 2184 | struct vcpu_svm *svm = to_svm(vcpu); |
15d45071 | 2185 | struct svm_cpu_data *sd = per_cpu(svm_data, cpu); |
15ad7146 | 2186 | int i; |
0cc5064d | 2187 | |
0cc5064d | 2188 | if (unlikely(cpu != vcpu->cpu)) { |
4b656b12 | 2189 | svm->asid_generation = 0; |
8d28fec4 | 2190 | mark_all_dirty(svm->vmcb); |
0cc5064d | 2191 | } |
94dfbdb3 | 2192 | |
82ca2d10 AK |
2193 | #ifdef CONFIG_X86_64 |
2194 | rdmsrl(MSR_GS_BASE, to_svm(vcpu)->host.gs_base); | |
2195 | #endif | |
dacccfdd AK |
2196 | savesegment(fs, svm->host.fs); |
2197 | savesegment(gs, svm->host.gs); | |
2198 | svm->host.ldt = kvm_read_ldt(); | |
2199 | ||
94dfbdb3 | 2200 | for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++) |
a2fa3e9f | 2201 | rdmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]); |
fbc0db76 | 2202 | |
ad721883 HZ |
2203 | if (static_cpu_has(X86_FEATURE_TSCRATEMSR)) { |
2204 | u64 tsc_ratio = vcpu->arch.tsc_scaling_ratio; | |
2205 | if (tsc_ratio != __this_cpu_read(current_tsc_ratio)) { | |
2206 | __this_cpu_write(current_tsc_ratio, tsc_ratio); | |
2207 | wrmsrl(MSR_AMD64_TSC_RATIO, tsc_ratio); | |
2208 | } | |
fbc0db76 | 2209 | } |
46896c73 PB |
2210 | /* This assumes that the kernel never uses MSR_TSC_AUX */ |
2211 | if (static_cpu_has(X86_FEATURE_RDTSCP)) | |
2212 | wrmsrl(MSR_TSC_AUX, svm->tsc_aux); | |
8221c137 | 2213 | |
15d45071 AR |
2214 | if (sd->current_vmcb != svm->vmcb) { |
2215 | sd->current_vmcb = svm->vmcb; | |
2216 | indirect_branch_prediction_barrier(); | |
2217 | } | |
8221c137 | 2218 | avic_vcpu_load(vcpu, cpu); |
6aa8b732 AK |
2219 | } |
2220 | ||
2221 | static void svm_vcpu_put(struct kvm_vcpu *vcpu) | |
2222 | { | |
a2fa3e9f | 2223 | struct vcpu_svm *svm = to_svm(vcpu); |
94dfbdb3 AL |
2224 | int i; |
2225 | ||
8221c137 SS |
2226 | avic_vcpu_put(vcpu); |
2227 | ||
e1beb1d3 | 2228 | ++vcpu->stat.host_state_reload; |
dacccfdd AK |
2229 | kvm_load_ldt(svm->host.ldt); |
2230 | #ifdef CONFIG_X86_64 | |
2231 | loadsegment(fs, svm->host.fs); | |
296f781a | 2232 | wrmsrl(MSR_KERNEL_GS_BASE, current->thread.gsbase); |
893a5ab6 | 2233 | load_gs_index(svm->host.gs); |
dacccfdd | 2234 | #else |
831ca609 | 2235 | #ifdef CONFIG_X86_32_LAZY_GS |
dacccfdd | 2236 | loadsegment(gs, svm->host.gs); |
831ca609 | 2237 | #endif |
dacccfdd | 2238 | #endif |
94dfbdb3 | 2239 | for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++) |
a2fa3e9f | 2240 | wrmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]); |
6aa8b732 AK |
2241 | } |
2242 | ||
8221c137 SS |
2243 | static void svm_vcpu_blocking(struct kvm_vcpu *vcpu) |
2244 | { | |
2245 | avic_set_running(vcpu, false); | |
2246 | } | |
2247 | ||
2248 | static void svm_vcpu_unblocking(struct kvm_vcpu *vcpu) | |
2249 | { | |
2250 | avic_set_running(vcpu, true); | |
2251 | } | |
2252 | ||
6aa8b732 AK |
2253 | static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu) |
2254 | { | |
9b611747 LP |
2255 | struct vcpu_svm *svm = to_svm(vcpu); |
2256 | unsigned long rflags = svm->vmcb->save.rflags; | |
2257 | ||
2258 | if (svm->nmi_singlestep) { | |
2259 | /* Hide our flags if they were not set by the guest */ | |
2260 | if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_TF)) | |
2261 | rflags &= ~X86_EFLAGS_TF; | |
2262 | if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_RF)) | |
2263 | rflags &= ~X86_EFLAGS_RF; | |
2264 | } | |
2265 | return rflags; | |
6aa8b732 AK |
2266 | } |
2267 | ||
2268 | static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags) | |
2269 | { | |
9b611747 LP |
2270 | if (to_svm(vcpu)->nmi_singlestep) |
2271 | rflags |= (X86_EFLAGS_TF | X86_EFLAGS_RF); | |
2272 | ||
ae9fedc7 | 2273 | /* |
bb3541f1 | 2274 | * Any change of EFLAGS.VM is accompanied by a reload of SS |
ae9fedc7 PB |
2275 | * (caused by either a task switch or an inter-privilege IRET), |
2276 | * so we do not need to update the CPL here. | |
2277 | */ | |
a2fa3e9f | 2278 | to_svm(vcpu)->vmcb->save.rflags = rflags; |
6aa8b732 AK |
2279 | } |
2280 | ||
6de4f3ad AK |
2281 | static void svm_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg) |
2282 | { | |
2283 | switch (reg) { | |
2284 | case VCPU_EXREG_PDPTR: | |
2285 | BUG_ON(!npt_enabled); | |
9f8fe504 | 2286 | load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu)); |
6de4f3ad AK |
2287 | break; |
2288 | default: | |
2289 | BUG(); | |
2290 | } | |
2291 | } | |
2292 | ||
f0b85051 AG |
2293 | static void svm_set_vintr(struct vcpu_svm *svm) |
2294 | { | |
8a05a1b8 | 2295 | set_intercept(svm, INTERCEPT_VINTR); |
f0b85051 AG |
2296 | } |
2297 | ||
2298 | static void svm_clear_vintr(struct vcpu_svm *svm) | |
2299 | { | |
8a05a1b8 | 2300 | clr_intercept(svm, INTERCEPT_VINTR); |
f0b85051 AG |
2301 | } |
2302 | ||
6aa8b732 AK |
2303 | static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg) |
2304 | { | |
a2fa3e9f | 2305 | struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save; |
6aa8b732 AK |
2306 | |
2307 | switch (seg) { | |
2308 | case VCPU_SREG_CS: return &save->cs; | |
2309 | case VCPU_SREG_DS: return &save->ds; | |
2310 | case VCPU_SREG_ES: return &save->es; | |
2311 | case VCPU_SREG_FS: return &save->fs; | |
2312 | case VCPU_SREG_GS: return &save->gs; | |
2313 | case VCPU_SREG_SS: return &save->ss; | |
2314 | case VCPU_SREG_TR: return &save->tr; | |
2315 | case VCPU_SREG_LDTR: return &save->ldtr; | |
2316 | } | |
2317 | BUG(); | |
8b6d44c7 | 2318 | return NULL; |
6aa8b732 AK |
2319 | } |
2320 | ||
2321 | static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg) | |
2322 | { | |
2323 | struct vmcb_seg *s = svm_seg(vcpu, seg); | |
2324 | ||
2325 | return s->base; | |
2326 | } | |
2327 | ||
2328 | static void svm_get_segment(struct kvm_vcpu *vcpu, | |
2329 | struct kvm_segment *var, int seg) | |
2330 | { | |
2331 | struct vmcb_seg *s = svm_seg(vcpu, seg); | |
2332 | ||
2333 | var->base = s->base; | |
2334 | var->limit = s->limit; | |
2335 | var->selector = s->selector; | |
2336 | var->type = s->attrib & SVM_SELECTOR_TYPE_MASK; | |
2337 | var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1; | |
2338 | var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3; | |
2339 | var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1; | |
2340 | var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1; | |
2341 | var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1; | |
2342 | var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1; | |
80112c89 JM |
2343 | |
2344 | /* | |
2345 | * AMD CPUs circa 2014 track the G bit for all segments except CS. | |
2346 | * However, the SVM spec states that the G bit is not observed by the | |
2347 | * CPU, and some VMware virtual CPUs drop the G bit for all segments. | |
2348 | * So let's synthesize a legal G bit for all segments, this helps | |
2349 | * running KVM nested. It also helps cross-vendor migration, because | |
2350 | * Intel's vmentry has a check on the 'G' bit. | |
2351 | */ | |
2352 | var->g = s->limit > 0xfffff; | |
25022acc | 2353 | |
e0231715 JR |
2354 | /* |
2355 | * AMD's VMCB does not have an explicit unusable field, so emulate it | |
19bca6ab AP |
2356 | * for cross vendor migration purposes by "not present" |
2357 | */ | |
8eae9570 | 2358 | var->unusable = !var->present; |
19bca6ab | 2359 | |
1fbdc7a5 | 2360 | switch (seg) { |
1fbdc7a5 AP |
2361 | case VCPU_SREG_TR: |
2362 | /* | |
2363 | * Work around a bug where the busy flag in the tr selector | |
2364 | * isn't exposed | |
2365 | */ | |
c0d09828 | 2366 | var->type |= 0x2; |
1fbdc7a5 AP |
2367 | break; |
2368 | case VCPU_SREG_DS: | |
2369 | case VCPU_SREG_ES: | |
2370 | case VCPU_SREG_FS: | |
2371 | case VCPU_SREG_GS: | |
2372 | /* | |
2373 | * The accessed bit must always be set in the segment | |
2374 | * descriptor cache, although it can be cleared in the | |
2375 | * descriptor, the cached bit always remains at 1. Since | |
2376 | * Intel has a check on this, set it here to support | |
2377 | * cross-vendor migration. | |
2378 | */ | |
2379 | if (!var->unusable) | |
2380 | var->type |= 0x1; | |
2381 | break; | |
b586eb02 | 2382 | case VCPU_SREG_SS: |
e0231715 JR |
2383 | /* |
2384 | * On AMD CPUs sometimes the DB bit in the segment | |
b586eb02 AP |
2385 | * descriptor is left as 1, although the whole segment has |
2386 | * been made unusable. Clear it here to pass an Intel VMX | |
2387 | * entry check when cross vendor migrating. | |
2388 | */ | |
2389 | if (var->unusable) | |
2390 | var->db = 0; | |
d9c1b543 | 2391 | /* This is symmetric with svm_set_segment() */ |
33b458d2 | 2392 | var->dpl = to_svm(vcpu)->vmcb->save.cpl; |
b586eb02 | 2393 | break; |
1fbdc7a5 | 2394 | } |
6aa8b732 AK |
2395 | } |
2396 | ||
2e4d2653 IE |
2397 | static int svm_get_cpl(struct kvm_vcpu *vcpu) |
2398 | { | |
2399 | struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save; | |
2400 | ||
2401 | return save->cpl; | |
2402 | } | |
2403 | ||
89a27f4d | 2404 | static void svm_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt) |
6aa8b732 | 2405 | { |
a2fa3e9f GH |
2406 | struct vcpu_svm *svm = to_svm(vcpu); |
2407 | ||
89a27f4d GN |
2408 | dt->size = svm->vmcb->save.idtr.limit; |
2409 | dt->address = svm->vmcb->save.idtr.base; | |
6aa8b732 AK |
2410 | } |
2411 | ||
89a27f4d | 2412 | static void svm_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt) |
6aa8b732 | 2413 | { |
a2fa3e9f GH |
2414 | struct vcpu_svm *svm = to_svm(vcpu); |
2415 | ||
89a27f4d GN |
2416 | svm->vmcb->save.idtr.limit = dt->size; |
2417 | svm->vmcb->save.idtr.base = dt->address ; | |
17a703cb | 2418 | mark_dirty(svm->vmcb, VMCB_DT); |
6aa8b732 AK |
2419 | } |
2420 | ||
89a27f4d | 2421 | static void svm_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt) |
6aa8b732 | 2422 | { |
a2fa3e9f GH |
2423 | struct vcpu_svm *svm = to_svm(vcpu); |
2424 | ||
89a27f4d GN |
2425 | dt->size = svm->vmcb->save.gdtr.limit; |
2426 | dt->address = svm->vmcb->save.gdtr.base; | |
6aa8b732 AK |
2427 | } |
2428 | ||
89a27f4d | 2429 | static void svm_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt) |
6aa8b732 | 2430 | { |
a2fa3e9f GH |
2431 | struct vcpu_svm *svm = to_svm(vcpu); |
2432 | ||
89a27f4d GN |
2433 | svm->vmcb->save.gdtr.limit = dt->size; |
2434 | svm->vmcb->save.gdtr.base = dt->address ; | |
17a703cb | 2435 | mark_dirty(svm->vmcb, VMCB_DT); |
6aa8b732 AK |
2436 | } |
2437 | ||
e8467fda AK |
2438 | static void svm_decache_cr0_guest_bits(struct kvm_vcpu *vcpu) |
2439 | { | |
2440 | } | |
2441 | ||
aff48baa AK |
2442 | static void svm_decache_cr3(struct kvm_vcpu *vcpu) |
2443 | { | |
2444 | } | |
2445 | ||
25c4c276 | 2446 | static void svm_decache_cr4_guest_bits(struct kvm_vcpu *vcpu) |
399badf3 AK |
2447 | { |
2448 | } | |
2449 | ||
d225157b AK |
2450 | static void update_cr0_intercept(struct vcpu_svm *svm) |
2451 | { | |
2452 | ulong gcr0 = svm->vcpu.arch.cr0; | |
2453 | u64 *hcr0 = &svm->vmcb->save.cr0; | |
2454 | ||
bd7e5b08 PB |
2455 | *hcr0 = (*hcr0 & ~SVM_CR0_SELECTIVE_MASK) |
2456 | | (gcr0 & SVM_CR0_SELECTIVE_MASK); | |
d225157b | 2457 | |
dcca1a65 | 2458 | mark_dirty(svm->vmcb, VMCB_CR); |
d225157b | 2459 | |
bd7e5b08 | 2460 | if (gcr0 == *hcr0) { |
4ee546b4 RJ |
2461 | clr_cr_intercept(svm, INTERCEPT_CR0_READ); |
2462 | clr_cr_intercept(svm, INTERCEPT_CR0_WRITE); | |
d225157b | 2463 | } else { |
4ee546b4 RJ |
2464 | set_cr_intercept(svm, INTERCEPT_CR0_READ); |
2465 | set_cr_intercept(svm, INTERCEPT_CR0_WRITE); | |
d225157b AK |
2466 | } |
2467 | } | |
2468 | ||
6aa8b732 AK |
2469 | static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0) |
2470 | { | |
a2fa3e9f GH |
2471 | struct vcpu_svm *svm = to_svm(vcpu); |
2472 | ||
05b3e0c2 | 2473 | #ifdef CONFIG_X86_64 |
f6801dff | 2474 | if (vcpu->arch.efer & EFER_LME) { |
707d92fa | 2475 | if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) { |
f6801dff | 2476 | vcpu->arch.efer |= EFER_LMA; |
2b5203ee | 2477 | svm->vmcb->save.efer |= EFER_LMA | EFER_LME; |
6aa8b732 AK |
2478 | } |
2479 | ||
d77c26fc | 2480 | if (is_paging(vcpu) && !(cr0 & X86_CR0_PG)) { |
f6801dff | 2481 | vcpu->arch.efer &= ~EFER_LMA; |
2b5203ee | 2482 | svm->vmcb->save.efer &= ~(EFER_LMA | EFER_LME); |
6aa8b732 AK |
2483 | } |
2484 | } | |
2485 | #endif | |
ad312c7c | 2486 | vcpu->arch.cr0 = cr0; |
888f9f3e AK |
2487 | |
2488 | if (!npt_enabled) | |
2489 | cr0 |= X86_CR0_PG | X86_CR0_WP; | |
02daab21 | 2490 | |
bcf166a9 PB |
2491 | /* |
2492 | * re-enable caching here because the QEMU bios | |
2493 | * does not do it - this results in some delay at | |
2494 | * reboot | |
2495 | */ | |
2496 | if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED)) | |
2497 | cr0 &= ~(X86_CR0_CD | X86_CR0_NW); | |
a2fa3e9f | 2498 | svm->vmcb->save.cr0 = cr0; |
dcca1a65 | 2499 | mark_dirty(svm->vmcb, VMCB_CR); |
d225157b | 2500 | update_cr0_intercept(svm); |
6aa8b732 AK |
2501 | } |
2502 | ||
5e1746d6 | 2503 | static int svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4) |
6aa8b732 | 2504 | { |
1e02ce4c | 2505 | unsigned long host_cr4_mce = cr4_read_shadow() & X86_CR4_MCE; |
e5eab0ce JR |
2506 | unsigned long old_cr4 = to_svm(vcpu)->vmcb->save.cr4; |
2507 | ||
5e1746d6 NHE |
2508 | if (cr4 & X86_CR4_VMXE) |
2509 | return 1; | |
2510 | ||
e5eab0ce | 2511 | if (npt_enabled && ((old_cr4 ^ cr4) & X86_CR4_PGE)) |
c2ba05cc | 2512 | svm_flush_tlb(vcpu, true); |
6394b649 | 2513 | |
ec077263 JR |
2514 | vcpu->arch.cr4 = cr4; |
2515 | if (!npt_enabled) | |
2516 | cr4 |= X86_CR4_PAE; | |
6394b649 | 2517 | cr4 |= host_cr4_mce; |
ec077263 | 2518 | to_svm(vcpu)->vmcb->save.cr4 = cr4; |
dcca1a65 | 2519 | mark_dirty(to_svm(vcpu)->vmcb, VMCB_CR); |
5e1746d6 | 2520 | return 0; |
6aa8b732 AK |
2521 | } |
2522 | ||
2523 | static void svm_set_segment(struct kvm_vcpu *vcpu, | |
2524 | struct kvm_segment *var, int seg) | |
2525 | { | |
a2fa3e9f | 2526 | struct vcpu_svm *svm = to_svm(vcpu); |
6aa8b732 AK |
2527 | struct vmcb_seg *s = svm_seg(vcpu, seg); |
2528 | ||
2529 | s->base = var->base; | |
2530 | s->limit = var->limit; | |
2531 | s->selector = var->selector; | |
d9c1b543 RP |
2532 | s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK); |
2533 | s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT; | |
2534 | s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT; | |
2535 | s->attrib |= ((var->present & 1) && !var->unusable) << SVM_SELECTOR_P_SHIFT; | |
2536 | s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT; | |
2537 | s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT; | |
2538 | s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT; | |
2539 | s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT; | |
ae9fedc7 PB |
2540 | |
2541 | /* | |
2542 | * This is always accurate, except if SYSRET returned to a segment | |
2543 | * with SS.DPL != 3. Intel does not have this quirk, and always | |
2544 | * forces SS.DPL to 3 on sysret, so we ignore that case; fixing it | |
2545 | * would entail passing the CPL to userspace and back. | |
2546 | */ | |
2547 | if (seg == VCPU_SREG_SS) | |
d9c1b543 RP |
2548 | /* This is symmetric with svm_get_segment() */ |
2549 | svm->vmcb->save.cpl = (var->dpl & 3); | |
6aa8b732 | 2550 | |
060d0c9a | 2551 | mark_dirty(svm->vmcb, VMCB_SEG); |
6aa8b732 AK |
2552 | } |
2553 | ||
cbdb967a | 2554 | static void update_bp_intercept(struct kvm_vcpu *vcpu) |
6aa8b732 | 2555 | { |
d0bfb940 JK |
2556 | struct vcpu_svm *svm = to_svm(vcpu); |
2557 | ||
18c918c5 | 2558 | clr_exception_intercept(svm, BP_VECTOR); |
44c11430 | 2559 | |
d0bfb940 | 2560 | if (vcpu->guest_debug & KVM_GUESTDBG_ENABLE) { |
d0bfb940 | 2561 | if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) |
18c918c5 | 2562 | set_exception_intercept(svm, BP_VECTOR); |
d0bfb940 JK |
2563 | } else |
2564 | vcpu->guest_debug = 0; | |
44c11430 GN |
2565 | } |
2566 | ||
0fe1e009 | 2567 | static void new_asid(struct vcpu_svm *svm, struct svm_cpu_data *sd) |
6aa8b732 | 2568 | { |
0fe1e009 TH |
2569 | if (sd->next_asid > sd->max_asid) { |
2570 | ++sd->asid_generation; | |
4faefff3 | 2571 | sd->next_asid = sd->min_asid; |
a2fa3e9f | 2572 | svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID; |
6aa8b732 AK |
2573 | } |
2574 | ||
0fe1e009 TH |
2575 | svm->asid_generation = sd->asid_generation; |
2576 | svm->vmcb->control.asid = sd->next_asid++; | |
d48086d1 JR |
2577 | |
2578 | mark_dirty(svm->vmcb, VMCB_ASID); | |
6aa8b732 AK |
2579 | } |
2580 | ||
73aaf249 JK |
2581 | static u64 svm_get_dr6(struct kvm_vcpu *vcpu) |
2582 | { | |
2583 | return to_svm(vcpu)->vmcb->save.dr6; | |
2584 | } | |
2585 | ||
2586 | static void svm_set_dr6(struct kvm_vcpu *vcpu, unsigned long value) | |
2587 | { | |
2588 | struct vcpu_svm *svm = to_svm(vcpu); | |
2589 | ||
2590 | svm->vmcb->save.dr6 = value; | |
2591 | mark_dirty(svm->vmcb, VMCB_DR); | |
2592 | } | |
2593 | ||
facb0139 PB |
2594 | static void svm_sync_dirty_debug_regs(struct kvm_vcpu *vcpu) |
2595 | { | |
2596 | struct vcpu_svm *svm = to_svm(vcpu); | |
2597 | ||
2598 | get_debugreg(vcpu->arch.db[0], 0); | |
2599 | get_debugreg(vcpu->arch.db[1], 1); | |
2600 | get_debugreg(vcpu->arch.db[2], 2); | |
2601 | get_debugreg(vcpu->arch.db[3], 3); | |
2602 | vcpu->arch.dr6 = svm_get_dr6(vcpu); | |
2603 | vcpu->arch.dr7 = svm->vmcb->save.dr7; | |
2604 | ||
2605 | vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT; | |
2606 | set_dr_intercepts(svm); | |
2607 | } | |
2608 | ||
020df079 | 2609 | static void svm_set_dr7(struct kvm_vcpu *vcpu, unsigned long value) |
6aa8b732 | 2610 | { |
42dbaa5a | 2611 | struct vcpu_svm *svm = to_svm(vcpu); |
42dbaa5a | 2612 | |
020df079 | 2613 | svm->vmcb->save.dr7 = value; |
72214b96 | 2614 | mark_dirty(svm->vmcb, VMCB_DR); |
6aa8b732 AK |
2615 | } |
2616 | ||
851ba692 | 2617 | static int pf_interception(struct vcpu_svm *svm) |
6aa8b732 | 2618 | { |
0ede79e1 | 2619 | u64 fault_address = __sme_clr(svm->vmcb->control.exit_info_2); |
1261bfa3 | 2620 | u64 error_code = svm->vmcb->control.exit_info_1; |
6aa8b732 | 2621 | |
1261bfa3 | 2622 | return kvm_handle_page_fault(&svm->vcpu, error_code, fault_address, |
00b10fe1 BS |
2623 | static_cpu_has(X86_FEATURE_DECODEASSISTS) ? |
2624 | svm->vmcb->control.insn_bytes : NULL, | |
d0006530 PB |
2625 | svm->vmcb->control.insn_len); |
2626 | } | |
2627 | ||
2628 | static int npf_interception(struct vcpu_svm *svm) | |
2629 | { | |
0ede79e1 | 2630 | u64 fault_address = __sme_clr(svm->vmcb->control.exit_info_2); |
d0006530 PB |
2631 | u64 error_code = svm->vmcb->control.exit_info_1; |
2632 | ||
2633 | trace_kvm_page_fault(fault_address, error_code); | |
2634 | return kvm_mmu_page_fault(&svm->vcpu, fault_address, error_code, | |
00b10fe1 BS |
2635 | static_cpu_has(X86_FEATURE_DECODEASSISTS) ? |
2636 | svm->vmcb->control.insn_bytes : NULL, | |
d0006530 | 2637 | svm->vmcb->control.insn_len); |
6aa8b732 AK |
2638 | } |
2639 | ||
851ba692 | 2640 | static int db_interception(struct vcpu_svm *svm) |
d0bfb940 | 2641 | { |
851ba692 AK |
2642 | struct kvm_run *kvm_run = svm->vcpu.run; |
2643 | ||
d0bfb940 | 2644 | if (!(svm->vcpu.guest_debug & |
44c11430 | 2645 | (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) && |
6be7d306 | 2646 | !svm->nmi_singlestep) { |
d0bfb940 JK |
2647 | kvm_queue_exception(&svm->vcpu, DB_VECTOR); |
2648 | return 1; | |
2649 | } | |
44c11430 | 2650 | |
6be7d306 | 2651 | if (svm->nmi_singlestep) { |
4aebd0e9 | 2652 | disable_nmi_singlestep(svm); |
44c11430 GN |
2653 | } |
2654 | ||
2655 | if (svm->vcpu.guest_debug & | |
e0231715 | 2656 | (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) { |
44c11430 GN |
2657 | kvm_run->exit_reason = KVM_EXIT_DEBUG; |
2658 | kvm_run->debug.arch.pc = | |
2659 | svm->vmcb->save.cs.base + svm->vmcb->save.rip; | |
2660 | kvm_run->debug.arch.exception = DB_VECTOR; | |
2661 | return 0; | |
2662 | } | |
2663 | ||
2664 | return 1; | |
d0bfb940 JK |
2665 | } |
2666 | ||
851ba692 | 2667 | static int bp_interception(struct vcpu_svm *svm) |
d0bfb940 | 2668 | { |
851ba692 AK |
2669 | struct kvm_run *kvm_run = svm->vcpu.run; |
2670 | ||
d0bfb940 JK |
2671 | kvm_run->exit_reason = KVM_EXIT_DEBUG; |
2672 | kvm_run->debug.arch.pc = svm->vmcb->save.cs.base + svm->vmcb->save.rip; | |
2673 | kvm_run->debug.arch.exception = BP_VECTOR; | |
2674 | return 0; | |
2675 | } | |
2676 | ||
851ba692 | 2677 | static int ud_interception(struct vcpu_svm *svm) |
7aa81cc0 | 2678 | { |
082d06ed | 2679 | return handle_ud(&svm->vcpu); |
7aa81cc0 AL |
2680 | } |
2681 | ||
54a20552 EN |
2682 | static int ac_interception(struct vcpu_svm *svm) |
2683 | { | |
2684 | kvm_queue_exception_e(&svm->vcpu, AC_VECTOR, 0); | |
2685 | return 1; | |
2686 | } | |
2687 | ||
9718420e LA |
2688 | static int gp_interception(struct vcpu_svm *svm) |
2689 | { | |
2690 | struct kvm_vcpu *vcpu = &svm->vcpu; | |
2691 | u32 error_code = svm->vmcb->control.exit_info_1; | |
2692 | int er; | |
2693 | ||
2694 | WARN_ON_ONCE(!enable_vmware_backdoor); | |
2695 | ||
2696 | er = emulate_instruction(vcpu, | |
2697 | EMULTYPE_VMWARE | EMULTYPE_NO_UD_ON_FAIL); | |
2698 | if (er == EMULATE_USER_EXIT) | |
2699 | return 0; | |
2700 | else if (er != EMULATE_DONE) | |
2701 | kvm_queue_exception_e(vcpu, GP_VECTOR, error_code); | |
2702 | return 1; | |
2703 | } | |
2704 | ||
67ec6607 JR |
2705 | static bool is_erratum_383(void) |
2706 | { | |
2707 | int err, i; | |
2708 | u64 value; | |
2709 | ||
2710 | if (!erratum_383_found) | |
2711 | return false; | |
2712 | ||
2713 | value = native_read_msr_safe(MSR_IA32_MC0_STATUS, &err); | |
2714 | if (err) | |
2715 | return false; | |
2716 | ||
2717 | /* Bit 62 may or may not be set for this mce */ | |
2718 | value &= ~(1ULL << 62); | |
2719 | ||
2720 | if (value != 0xb600000000010015ULL) | |
2721 | return false; | |
2722 | ||
2723 | /* Clear MCi_STATUS registers */ | |
2724 | for (i = 0; i < 6; ++i) | |
2725 | native_write_msr_safe(MSR_IA32_MCx_STATUS(i), 0, 0); | |
2726 | ||
2727 | value = native_read_msr_safe(MSR_IA32_MCG_STATUS, &err); | |
2728 | if (!err) { | |
2729 | u32 low, high; | |
2730 | ||
2731 | value &= ~(1ULL << 2); | |
2732 | low = lower_32_bits(value); | |
2733 | high = upper_32_bits(value); | |
2734 | ||
2735 | native_write_msr_safe(MSR_IA32_MCG_STATUS, low, high); | |
2736 | } | |
2737 | ||
2738 | /* Flush tlb to evict multi-match entries */ | |
2739 | __flush_tlb_all(); | |
2740 | ||
2741 | return true; | |
2742 | } | |
2743 | ||
fe5913e4 | 2744 | static void svm_handle_mce(struct vcpu_svm *svm) |
53371b50 | 2745 | { |
67ec6607 JR |
2746 | if (is_erratum_383()) { |
2747 | /* | |
2748 | * Erratum 383 triggered. Guest state is corrupt so kill the | |
2749 | * guest. | |
2750 | */ | |
2751 | pr_err("KVM: Guest triggered AMD Erratum 383\n"); | |
2752 | ||
a8eeb04a | 2753 | kvm_make_request(KVM_REQ_TRIPLE_FAULT, &svm->vcpu); |
67ec6607 JR |
2754 | |
2755 | return; | |
2756 | } | |
2757 | ||
53371b50 JR |
2758 | /* |
2759 | * On an #MC intercept the MCE handler is not called automatically in | |
2760 | * the host. So do it by hand here. | |
2761 | */ | |
2762 | asm volatile ( | |
2763 | "int $0x12\n"); | |
2764 | /* not sure if we ever come back to this point */ | |
2765 | ||
fe5913e4 JR |
2766 | return; |
2767 | } | |
2768 | ||
2769 | static int mc_interception(struct vcpu_svm *svm) | |
2770 | { | |
53371b50 JR |
2771 | return 1; |
2772 | } | |
2773 | ||
851ba692 | 2774 | static int shutdown_interception(struct vcpu_svm *svm) |
46fe4ddd | 2775 | { |
851ba692 AK |
2776 | struct kvm_run *kvm_run = svm->vcpu.run; |
2777 | ||
46fe4ddd JR |
2778 | /* |
2779 | * VMCB is undefined after a SHUTDOWN intercept | |
2780 | * so reinitialize it. | |
2781 | */ | |
a2fa3e9f | 2782 | clear_page(svm->vmcb); |
5690891b | 2783 | init_vmcb(svm); |
46fe4ddd JR |
2784 | |
2785 | kvm_run->exit_reason = KVM_EXIT_SHUTDOWN; | |
2786 | return 0; | |
2787 | } | |
2788 | ||
851ba692 | 2789 | static int io_interception(struct vcpu_svm *svm) |
6aa8b732 | 2790 | { |
cf8f70bf | 2791 | struct kvm_vcpu *vcpu = &svm->vcpu; |
d77c26fc | 2792 | u32 io_info = svm->vmcb->control.exit_info_1; /* address size bug? */ |
dca7f128 | 2793 | int size, in, string; |
039576c0 | 2794 | unsigned port; |
6aa8b732 | 2795 | |
e756fc62 | 2796 | ++svm->vcpu.stat.io_exits; |
e70669ab | 2797 | string = (io_info & SVM_IOIO_STR_MASK) != 0; |
039576c0 | 2798 | in = (io_info & SVM_IOIO_TYPE_MASK) != 0; |
8370c3d0 | 2799 | if (string) |
51d8b661 | 2800 | return emulate_instruction(vcpu, 0) == EMULATE_DONE; |
cf8f70bf | 2801 | |
039576c0 AK |
2802 | port = io_info >> 16; |
2803 | size = (io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT; | |
cf8f70bf | 2804 | svm->next_rip = svm->vmcb->control.exit_info_2; |
cf8f70bf | 2805 | |
dca7f128 | 2806 | return kvm_fast_pio(&svm->vcpu, size, port, in); |
6aa8b732 AK |
2807 | } |
2808 | ||
851ba692 | 2809 | static int nmi_interception(struct vcpu_svm *svm) |
c47f098d JR |
2810 | { |
2811 | return 1; | |
2812 | } | |
2813 | ||
851ba692 | 2814 | static int intr_interception(struct vcpu_svm *svm) |
a0698055 JR |
2815 | { |
2816 | ++svm->vcpu.stat.irq_exits; | |
2817 | return 1; | |
2818 | } | |
2819 | ||
851ba692 | 2820 | static int nop_on_interception(struct vcpu_svm *svm) |
6aa8b732 AK |
2821 | { |
2822 | return 1; | |
2823 | } | |
2824 | ||
851ba692 | 2825 | static int halt_interception(struct vcpu_svm *svm) |
6aa8b732 | 2826 | { |
5fdbf976 | 2827 | svm->next_rip = kvm_rip_read(&svm->vcpu) + 1; |
e756fc62 | 2828 | return kvm_emulate_halt(&svm->vcpu); |
6aa8b732 AK |
2829 | } |
2830 | ||
851ba692 | 2831 | static int vmmcall_interception(struct vcpu_svm *svm) |
02e235bc | 2832 | { |
5fdbf976 | 2833 | svm->next_rip = kvm_rip_read(&svm->vcpu) + 3; |
0d9c055e | 2834 | return kvm_emulate_hypercall(&svm->vcpu); |
02e235bc AK |
2835 | } |
2836 | ||
5bd2edc3 JR |
2837 | static unsigned long nested_svm_get_tdp_cr3(struct kvm_vcpu *vcpu) |
2838 | { | |
2839 | struct vcpu_svm *svm = to_svm(vcpu); | |
2840 | ||
2841 | return svm->nested.nested_cr3; | |
2842 | } | |
2843 | ||
e4e517b4 AK |
2844 | static u64 nested_svm_get_tdp_pdptr(struct kvm_vcpu *vcpu, int index) |
2845 | { | |
2846 | struct vcpu_svm *svm = to_svm(vcpu); | |
2847 | u64 cr3 = svm->nested.nested_cr3; | |
2848 | u64 pdpte; | |
2849 | int ret; | |
2850 | ||
d0ec49d4 | 2851 | ret = kvm_vcpu_read_guest_page(vcpu, gpa_to_gfn(__sme_clr(cr3)), &pdpte, |
54bf36aa | 2852 | offset_in_page(cr3) + index * 8, 8); |
e4e517b4 AK |
2853 | if (ret) |
2854 | return 0; | |
2855 | return pdpte; | |
2856 | } | |
2857 | ||
5bd2edc3 JR |
2858 | static void nested_svm_set_tdp_cr3(struct kvm_vcpu *vcpu, |
2859 | unsigned long root) | |
2860 | { | |
2861 | struct vcpu_svm *svm = to_svm(vcpu); | |
2862 | ||
d0ec49d4 | 2863 | svm->vmcb->control.nested_cr3 = __sme_set(root); |
b2747166 | 2864 | mark_dirty(svm->vmcb, VMCB_NPT); |
c2ba05cc | 2865 | svm_flush_tlb(vcpu, true); |
5bd2edc3 JR |
2866 | } |
2867 | ||
6389ee94 AK |
2868 | static void nested_svm_inject_npf_exit(struct kvm_vcpu *vcpu, |
2869 | struct x86_exception *fault) | |
5bd2edc3 JR |
2870 | { |
2871 | struct vcpu_svm *svm = to_svm(vcpu); | |
2872 | ||
5e352519 PB |
2873 | if (svm->vmcb->control.exit_code != SVM_EXIT_NPF) { |
2874 | /* | |
2875 | * TODO: track the cause of the nested page fault, and | |
2876 | * correctly fill in the high bits of exit_info_1. | |
2877 | */ | |
2878 | svm->vmcb->control.exit_code = SVM_EXIT_NPF; | |
2879 | svm->vmcb->control.exit_code_hi = 0; | |
2880 | svm->vmcb->control.exit_info_1 = (1ULL << 32); | |
2881 | svm->vmcb->control.exit_info_2 = fault->address; | |
2882 | } | |
2883 | ||
2884 | svm->vmcb->control.exit_info_1 &= ~0xffffffffULL; | |
2885 | svm->vmcb->control.exit_info_1 |= fault->error_code; | |
2886 | ||
2887 | /* | |
2888 | * The present bit is always zero for page structure faults on real | |
2889 | * hardware. | |
2890 | */ | |
2891 | if (svm->vmcb->control.exit_info_1 & (2ULL << 32)) | |
2892 | svm->vmcb->control.exit_info_1 &= ~1; | |
5bd2edc3 JR |
2893 | |
2894 | nested_svm_vmexit(svm); | |
2895 | } | |
2896 | ||
8a3c1a33 | 2897 | static void nested_svm_init_mmu_context(struct kvm_vcpu *vcpu) |
4b16184c | 2898 | { |
ad896af0 PB |
2899 | WARN_ON(mmu_is_nested(vcpu)); |
2900 | kvm_init_shadow_mmu(vcpu); | |
4b16184c JR |
2901 | vcpu->arch.mmu.set_cr3 = nested_svm_set_tdp_cr3; |
2902 | vcpu->arch.mmu.get_cr3 = nested_svm_get_tdp_cr3; | |
e4e517b4 | 2903 | vcpu->arch.mmu.get_pdptr = nested_svm_get_tdp_pdptr; |
4b16184c | 2904 | vcpu->arch.mmu.inject_page_fault = nested_svm_inject_npf_exit; |
855feb67 | 2905 | vcpu->arch.mmu.shadow_root_level = get_npt_level(vcpu); |
c258b62b | 2906 | reset_shadow_zero_bits_mask(vcpu, &vcpu->arch.mmu); |
4b16184c | 2907 | vcpu->arch.walk_mmu = &vcpu->arch.nested_mmu; |
4b16184c JR |
2908 | } |
2909 | ||
2910 | static void nested_svm_uninit_mmu_context(struct kvm_vcpu *vcpu) | |
2911 | { | |
2912 | vcpu->arch.walk_mmu = &vcpu->arch.mmu; | |
2913 | } | |
2914 | ||
c0725420 AG |
2915 | static int nested_svm_check_permissions(struct vcpu_svm *svm) |
2916 | { | |
e9196ceb DC |
2917 | if (!(svm->vcpu.arch.efer & EFER_SVME) || |
2918 | !is_paging(&svm->vcpu)) { | |
c0725420 AG |
2919 | kvm_queue_exception(&svm->vcpu, UD_VECTOR); |
2920 | return 1; | |
2921 | } | |
2922 | ||
2923 | if (svm->vmcb->save.cpl) { | |
2924 | kvm_inject_gp(&svm->vcpu, 0); | |
2925 | return 1; | |
2926 | } | |
2927 | ||
e9196ceb | 2928 | return 0; |
c0725420 AG |
2929 | } |
2930 | ||
cf74a78b AG |
2931 | static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr, |
2932 | bool has_error_code, u32 error_code) | |
2933 | { | |
b8e88bc8 JR |
2934 | int vmexit; |
2935 | ||
2030753d | 2936 | if (!is_guest_mode(&svm->vcpu)) |
0295ad7d | 2937 | return 0; |
cf74a78b | 2938 | |
adfe20fb WL |
2939 | vmexit = nested_svm_intercept(svm); |
2940 | if (vmexit != NESTED_EXIT_DONE) | |
2941 | return 0; | |
2942 | ||
0295ad7d JR |
2943 | svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + nr; |
2944 | svm->vmcb->control.exit_code_hi = 0; | |
2945 | svm->vmcb->control.exit_info_1 = error_code; | |
b96fb439 PB |
2946 | |
2947 | /* | |
2948 | * FIXME: we should not write CR2 when L1 intercepts an L2 #PF exception. | |
2949 | * The fix is to add the ancillary datum (CR2 or DR6) to structs | |
2950 | * kvm_queued_exception and kvm_vcpu_events, so that CR2 and DR6 can be | |
2951 | * written only when inject_pending_event runs (DR6 would written here | |
2952 | * too). This should be conditional on a new capability---if the | |
2953 | * capability is disabled, kvm_multiple_exception would write the | |
2954 | * ancillary information to CR2 or DR6, for backwards ABI-compatibility. | |
2955 | */ | |
adfe20fb WL |
2956 | if (svm->vcpu.arch.exception.nested_apf) |
2957 | svm->vmcb->control.exit_info_2 = svm->vcpu.arch.apf.nested_apf_token; | |
2958 | else | |
2959 | svm->vmcb->control.exit_info_2 = svm->vcpu.arch.cr2; | |
b8e88bc8 | 2960 | |
adfe20fb | 2961 | svm->nested.exit_required = true; |
b8e88bc8 | 2962 | return vmexit; |
cf74a78b AG |
2963 | } |
2964 | ||
8fe54654 JR |
2965 | /* This function returns true if it is save to enable the irq window */ |
2966 | static inline bool nested_svm_intr(struct vcpu_svm *svm) | |
cf74a78b | 2967 | { |
2030753d | 2968 | if (!is_guest_mode(&svm->vcpu)) |
8fe54654 | 2969 | return true; |
cf74a78b | 2970 | |
26666957 | 2971 | if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK)) |
8fe54654 | 2972 | return true; |
cf74a78b | 2973 | |
26666957 | 2974 | if (!(svm->vcpu.arch.hflags & HF_HIF_MASK)) |
8fe54654 | 2975 | return false; |
cf74a78b | 2976 | |
a0a07cd2 GN |
2977 | /* |
2978 | * if vmexit was already requested (by intercepted exception | |
2979 | * for instance) do not overwrite it with "external interrupt" | |
2980 | * vmexit. | |
2981 | */ | |
2982 | if (svm->nested.exit_required) | |
2983 | return false; | |
2984 | ||
197717d5 JR |
2985 | svm->vmcb->control.exit_code = SVM_EXIT_INTR; |
2986 | svm->vmcb->control.exit_info_1 = 0; | |
2987 | svm->vmcb->control.exit_info_2 = 0; | |
26666957 | 2988 | |
cd3ff653 JR |
2989 | if (svm->nested.intercept & 1ULL) { |
2990 | /* | |
2991 | * The #vmexit can't be emulated here directly because this | |
c5ec2e56 | 2992 | * code path runs with irqs and preemption disabled. A |
cd3ff653 JR |
2993 | * #vmexit emulation might sleep. Only signal request for |
2994 | * the #vmexit here. | |
2995 | */ | |
2996 | svm->nested.exit_required = true; | |
236649de | 2997 | trace_kvm_nested_intr_vmexit(svm->vmcb->save.rip); |
8fe54654 | 2998 | return false; |
cf74a78b AG |
2999 | } |
3000 | ||
8fe54654 | 3001 | return true; |
cf74a78b AG |
3002 | } |
3003 | ||
887f500c JR |
3004 | /* This function returns true if it is save to enable the nmi window */ |
3005 | static inline bool nested_svm_nmi(struct vcpu_svm *svm) | |
3006 | { | |
2030753d | 3007 | if (!is_guest_mode(&svm->vcpu)) |
887f500c JR |
3008 | return true; |
3009 | ||
3010 | if (!(svm->nested.intercept & (1ULL << INTERCEPT_NMI))) | |
3011 | return true; | |
3012 | ||
3013 | svm->vmcb->control.exit_code = SVM_EXIT_NMI; | |
3014 | svm->nested.exit_required = true; | |
3015 | ||
3016 | return false; | |
cf74a78b AG |
3017 | } |
3018 | ||
7597f129 | 3019 | static void *nested_svm_map(struct vcpu_svm *svm, u64 gpa, struct page **_page) |
34f80cfa JR |
3020 | { |
3021 | struct page *page; | |
3022 | ||
6c3bd3d7 JR |
3023 | might_sleep(); |
3024 | ||
54bf36aa | 3025 | page = kvm_vcpu_gfn_to_page(&svm->vcpu, gpa >> PAGE_SHIFT); |
34f80cfa JR |
3026 | if (is_error_page(page)) |
3027 | goto error; | |
3028 | ||
7597f129 JR |
3029 | *_page = page; |
3030 | ||
3031 | return kmap(page); | |
34f80cfa JR |
3032 | |
3033 | error: | |
34f80cfa JR |
3034 | kvm_inject_gp(&svm->vcpu, 0); |
3035 | ||
3036 | return NULL; | |
3037 | } | |
3038 | ||
7597f129 | 3039 | static void nested_svm_unmap(struct page *page) |
34f80cfa | 3040 | { |
7597f129 | 3041 | kunmap(page); |
34f80cfa JR |
3042 | kvm_release_page_dirty(page); |
3043 | } | |
34f80cfa | 3044 | |
ce2ac085 JR |
3045 | static int nested_svm_intercept_ioio(struct vcpu_svm *svm) |
3046 | { | |
9bf41833 JK |
3047 | unsigned port, size, iopm_len; |
3048 | u16 val, mask; | |
3049 | u8 start_bit; | |
ce2ac085 | 3050 | u64 gpa; |
34f80cfa | 3051 | |
ce2ac085 JR |
3052 | if (!(svm->nested.intercept & (1ULL << INTERCEPT_IOIO_PROT))) |
3053 | return NESTED_EXIT_HOST; | |
34f80cfa | 3054 | |
ce2ac085 | 3055 | port = svm->vmcb->control.exit_info_1 >> 16; |
9bf41833 JK |
3056 | size = (svm->vmcb->control.exit_info_1 & SVM_IOIO_SIZE_MASK) >> |
3057 | SVM_IOIO_SIZE_SHIFT; | |
ce2ac085 | 3058 | gpa = svm->nested.vmcb_iopm + (port / 8); |
9bf41833 JK |
3059 | start_bit = port % 8; |
3060 | iopm_len = (start_bit + size > 8) ? 2 : 1; | |
3061 | mask = (0xf >> (4 - size)) << start_bit; | |
3062 | val = 0; | |
ce2ac085 | 3063 | |
54bf36aa | 3064 | if (kvm_vcpu_read_guest(&svm->vcpu, gpa, &val, iopm_len)) |
9bf41833 | 3065 | return NESTED_EXIT_DONE; |
ce2ac085 | 3066 | |
9bf41833 | 3067 | return (val & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST; |
34f80cfa JR |
3068 | } |
3069 | ||
d2477826 | 3070 | static int nested_svm_exit_handled_msr(struct vcpu_svm *svm) |
4c2161ae | 3071 | { |
0d6b3537 JR |
3072 | u32 offset, msr, value; |
3073 | int write, mask; | |
4c2161ae | 3074 | |
3d62d9aa | 3075 | if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT))) |
d2477826 | 3076 | return NESTED_EXIT_HOST; |
3d62d9aa | 3077 | |
0d6b3537 JR |
3078 | msr = svm->vcpu.arch.regs[VCPU_REGS_RCX]; |
3079 | offset = svm_msrpm_offset(msr); | |
3080 | write = svm->vmcb->control.exit_info_1 & 1; | |
3081 | mask = 1 << ((2 * (msr & 0xf)) + write); | |
3d62d9aa | 3082 | |
0d6b3537 JR |
3083 | if (offset == MSR_INVALID) |
3084 | return NESTED_EXIT_DONE; | |
4c2161ae | 3085 | |
0d6b3537 JR |
3086 | /* Offset is in 32 bit units but need in 8 bit units */ |
3087 | offset *= 4; | |
4c2161ae | 3088 | |
54bf36aa | 3089 | if (kvm_vcpu_read_guest(&svm->vcpu, svm->nested.vmcb_msrpm + offset, &value, 4)) |
0d6b3537 | 3090 | return NESTED_EXIT_DONE; |
3d62d9aa | 3091 | |
0d6b3537 | 3092 | return (value & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST; |
4c2161ae JR |
3093 | } |
3094 | ||
ab2f4d73 LP |
3095 | /* DB exceptions for our internal use must not cause vmexit */ |
3096 | static int nested_svm_intercept_db(struct vcpu_svm *svm) | |
3097 | { | |
3098 | unsigned long dr6; | |
3099 | ||
3100 | /* if we're not singlestepping, it's not ours */ | |
3101 | if (!svm->nmi_singlestep) | |
3102 | return NESTED_EXIT_DONE; | |
3103 | ||
3104 | /* if it's not a singlestep exception, it's not ours */ | |
3105 | if (kvm_get_dr(&svm->vcpu, 6, &dr6)) | |
3106 | return NESTED_EXIT_DONE; | |
3107 | if (!(dr6 & DR6_BS)) | |
3108 | return NESTED_EXIT_DONE; | |
3109 | ||
3110 | /* if the guest is singlestepping, it should get the vmexit */ | |
3111 | if (svm->nmi_singlestep_guest_rflags & X86_EFLAGS_TF) { | |
3112 | disable_nmi_singlestep(svm); | |
3113 | return NESTED_EXIT_DONE; | |
3114 | } | |
3115 | ||
3116 | /* it's ours, the nested hypervisor must not see this one */ | |
3117 | return NESTED_EXIT_HOST; | |
3118 | } | |
3119 | ||
410e4d57 | 3120 | static int nested_svm_exit_special(struct vcpu_svm *svm) |
cf74a78b | 3121 | { |
cf74a78b | 3122 | u32 exit_code = svm->vmcb->control.exit_code; |
4c2161ae | 3123 | |
410e4d57 JR |
3124 | switch (exit_code) { |
3125 | case SVM_EXIT_INTR: | |
3126 | case SVM_EXIT_NMI: | |
ff47a49b | 3127 | case SVM_EXIT_EXCP_BASE + MC_VECTOR: |
410e4d57 | 3128 | return NESTED_EXIT_HOST; |
410e4d57 | 3129 | case SVM_EXIT_NPF: |
e0231715 | 3130 | /* For now we are always handling NPFs when using them */ |
410e4d57 JR |
3131 | if (npt_enabled) |
3132 | return NESTED_EXIT_HOST; | |
3133 | break; | |
410e4d57 | 3134 | case SVM_EXIT_EXCP_BASE + PF_VECTOR: |
631bc487 | 3135 | /* When we're shadowing, trap PFs, but not async PF */ |
1261bfa3 | 3136 | if (!npt_enabled && svm->vcpu.arch.apf.host_apf_reason == 0) |
410e4d57 JR |
3137 | return NESTED_EXIT_HOST; |
3138 | break; | |
3139 | default: | |
3140 | break; | |
cf74a78b AG |
3141 | } |
3142 | ||
410e4d57 JR |
3143 | return NESTED_EXIT_CONTINUE; |
3144 | } | |
3145 | ||
3146 | /* | |
3147 | * If this function returns true, this #vmexit was already handled | |
3148 | */ | |
b8e88bc8 | 3149 | static int nested_svm_intercept(struct vcpu_svm *svm) |
410e4d57 JR |
3150 | { |
3151 | u32 exit_code = svm->vmcb->control.exit_code; | |
3152 | int vmexit = NESTED_EXIT_HOST; | |
3153 | ||
cf74a78b | 3154 | switch (exit_code) { |
9c4e40b9 | 3155 | case SVM_EXIT_MSR: |
3d62d9aa | 3156 | vmexit = nested_svm_exit_handled_msr(svm); |
9c4e40b9 | 3157 | break; |
ce2ac085 JR |
3158 | case SVM_EXIT_IOIO: |
3159 | vmexit = nested_svm_intercept_ioio(svm); | |
3160 | break; | |
4ee546b4 RJ |
3161 | case SVM_EXIT_READ_CR0 ... SVM_EXIT_WRITE_CR8: { |
3162 | u32 bit = 1U << (exit_code - SVM_EXIT_READ_CR0); | |
3163 | if (svm->nested.intercept_cr & bit) | |
410e4d57 | 3164 | vmexit = NESTED_EXIT_DONE; |
cf74a78b AG |
3165 | break; |
3166 | } | |
3aed041a JR |
3167 | case SVM_EXIT_READ_DR0 ... SVM_EXIT_WRITE_DR7: { |
3168 | u32 bit = 1U << (exit_code - SVM_EXIT_READ_DR0); | |
3169 | if (svm->nested.intercept_dr & bit) | |
410e4d57 | 3170 | vmexit = NESTED_EXIT_DONE; |
cf74a78b AG |
3171 | break; |
3172 | } | |
3173 | case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: { | |
3174 | u32 excp_bits = 1 << (exit_code - SVM_EXIT_EXCP_BASE); | |
ab2f4d73 LP |
3175 | if (svm->nested.intercept_exceptions & excp_bits) { |
3176 | if (exit_code == SVM_EXIT_EXCP_BASE + DB_VECTOR) | |
3177 | vmexit = nested_svm_intercept_db(svm); | |
3178 | else | |
3179 | vmexit = NESTED_EXIT_DONE; | |
3180 | } | |
631bc487 GN |
3181 | /* async page fault always cause vmexit */ |
3182 | else if ((exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR) && | |
adfe20fb | 3183 | svm->vcpu.arch.exception.nested_apf != 0) |
631bc487 | 3184 | vmexit = NESTED_EXIT_DONE; |
cf74a78b AG |
3185 | break; |
3186 | } | |
228070b1 JR |
3187 | case SVM_EXIT_ERR: { |
3188 | vmexit = NESTED_EXIT_DONE; | |
3189 | break; | |
3190 | } | |
cf74a78b AG |
3191 | default: { |
3192 | u64 exit_bits = 1ULL << (exit_code - SVM_EXIT_INTR); | |
aad42c64 | 3193 | if (svm->nested.intercept & exit_bits) |
410e4d57 | 3194 | vmexit = NESTED_EXIT_DONE; |
cf74a78b AG |
3195 | } |
3196 | } | |
3197 | ||
b8e88bc8 JR |
3198 | return vmexit; |
3199 | } | |
3200 | ||
3201 | static int nested_svm_exit_handled(struct vcpu_svm *svm) | |
3202 | { | |
3203 | int vmexit; | |
3204 | ||
3205 | vmexit = nested_svm_intercept(svm); | |
3206 | ||
3207 | if (vmexit == NESTED_EXIT_DONE) | |
9c4e40b9 | 3208 | nested_svm_vmexit(svm); |
9c4e40b9 JR |
3209 | |
3210 | return vmexit; | |
cf74a78b AG |
3211 | } |
3212 | ||
0460a979 JR |
3213 | static inline void copy_vmcb_control_area(struct vmcb *dst_vmcb, struct vmcb *from_vmcb) |
3214 | { | |
3215 | struct vmcb_control_area *dst = &dst_vmcb->control; | |
3216 | struct vmcb_control_area *from = &from_vmcb->control; | |
3217 | ||
4ee546b4 | 3218 | dst->intercept_cr = from->intercept_cr; |
3aed041a | 3219 | dst->intercept_dr = from->intercept_dr; |
0460a979 JR |
3220 | dst->intercept_exceptions = from->intercept_exceptions; |
3221 | dst->intercept = from->intercept; | |
3222 | dst->iopm_base_pa = from->iopm_base_pa; | |
3223 | dst->msrpm_base_pa = from->msrpm_base_pa; | |
3224 | dst->tsc_offset = from->tsc_offset; | |
3225 | dst->asid = from->asid; | |
3226 | dst->tlb_ctl = from->tlb_ctl; | |
3227 | dst->int_ctl = from->int_ctl; | |
3228 | dst->int_vector = from->int_vector; | |
3229 | dst->int_state = from->int_state; | |
3230 | dst->exit_code = from->exit_code; | |
3231 | dst->exit_code_hi = from->exit_code_hi; | |
3232 | dst->exit_info_1 = from->exit_info_1; | |
3233 | dst->exit_info_2 = from->exit_info_2; | |
3234 | dst->exit_int_info = from->exit_int_info; | |
3235 | dst->exit_int_info_err = from->exit_int_info_err; | |
3236 | dst->nested_ctl = from->nested_ctl; | |
3237 | dst->event_inj = from->event_inj; | |
3238 | dst->event_inj_err = from->event_inj_err; | |
3239 | dst->nested_cr3 = from->nested_cr3; | |
0dc92119 | 3240 | dst->virt_ext = from->virt_ext; |
0460a979 JR |
3241 | } |
3242 | ||
34f80cfa | 3243 | static int nested_svm_vmexit(struct vcpu_svm *svm) |
cf74a78b | 3244 | { |
34f80cfa | 3245 | struct vmcb *nested_vmcb; |
e6aa9abd | 3246 | struct vmcb *hsave = svm->nested.hsave; |
33740e40 | 3247 | struct vmcb *vmcb = svm->vmcb; |
7597f129 | 3248 | struct page *page; |
cf74a78b | 3249 | |
17897f36 JR |
3250 | trace_kvm_nested_vmexit_inject(vmcb->control.exit_code, |
3251 | vmcb->control.exit_info_1, | |
3252 | vmcb->control.exit_info_2, | |
3253 | vmcb->control.exit_int_info, | |
e097e5ff SH |
3254 | vmcb->control.exit_int_info_err, |
3255 | KVM_ISA_SVM); | |
17897f36 | 3256 | |
7597f129 | 3257 | nested_vmcb = nested_svm_map(svm, svm->nested.vmcb, &page); |
34f80cfa JR |
3258 | if (!nested_vmcb) |
3259 | return 1; | |
3260 | ||
2030753d JR |
3261 | /* Exit Guest-Mode */ |
3262 | leave_guest_mode(&svm->vcpu); | |
06fc7772 JR |
3263 | svm->nested.vmcb = 0; |
3264 | ||
cf74a78b | 3265 | /* Give the current vmcb to the guest */ |
33740e40 JR |
3266 | disable_gif(svm); |
3267 | ||
3268 | nested_vmcb->save.es = vmcb->save.es; | |
3269 | nested_vmcb->save.cs = vmcb->save.cs; | |
3270 | nested_vmcb->save.ss = vmcb->save.ss; | |
3271 | nested_vmcb->save.ds = vmcb->save.ds; | |
3272 | nested_vmcb->save.gdtr = vmcb->save.gdtr; | |
3273 | nested_vmcb->save.idtr = vmcb->save.idtr; | |
3f6a9d16 | 3274 | nested_vmcb->save.efer = svm->vcpu.arch.efer; |
cdbbdc12 | 3275 | nested_vmcb->save.cr0 = kvm_read_cr0(&svm->vcpu); |
9f8fe504 | 3276 | nested_vmcb->save.cr3 = kvm_read_cr3(&svm->vcpu); |
33740e40 | 3277 | nested_vmcb->save.cr2 = vmcb->save.cr2; |
cdbbdc12 | 3278 | nested_vmcb->save.cr4 = svm->vcpu.arch.cr4; |
f6e78475 | 3279 | nested_vmcb->save.rflags = kvm_get_rflags(&svm->vcpu); |
33740e40 JR |
3280 | nested_vmcb->save.rip = vmcb->save.rip; |
3281 | nested_vmcb->save.rsp = vmcb->save.rsp; | |
3282 | nested_vmcb->save.rax = vmcb->save.rax; | |
3283 | nested_vmcb->save.dr7 = vmcb->save.dr7; | |
3284 | nested_vmcb->save.dr6 = vmcb->save.dr6; | |
3285 | nested_vmcb->save.cpl = vmcb->save.cpl; | |
3286 | ||
3287 | nested_vmcb->control.int_ctl = vmcb->control.int_ctl; | |
3288 | nested_vmcb->control.int_vector = vmcb->control.int_vector; | |
3289 | nested_vmcb->control.int_state = vmcb->control.int_state; | |
3290 | nested_vmcb->control.exit_code = vmcb->control.exit_code; | |
3291 | nested_vmcb->control.exit_code_hi = vmcb->control.exit_code_hi; | |
3292 | nested_vmcb->control.exit_info_1 = vmcb->control.exit_info_1; | |
3293 | nested_vmcb->control.exit_info_2 = vmcb->control.exit_info_2; | |
3294 | nested_vmcb->control.exit_int_info = vmcb->control.exit_int_info; | |
3295 | nested_vmcb->control.exit_int_info_err = vmcb->control.exit_int_info_err; | |
6092d3d3 JR |
3296 | |
3297 | if (svm->nrips_enabled) | |
3298 | nested_vmcb->control.next_rip = vmcb->control.next_rip; | |
8d23c466 AG |
3299 | |
3300 | /* | |
3301 | * If we emulate a VMRUN/#VMEXIT in the same host #vmexit cycle we have | |
3302 | * to make sure that we do not lose injected events. So check event_inj | |
3303 | * here and copy it to exit_int_info if it is valid. | |
3304 | * Exit_int_info and event_inj can't be both valid because the case | |
3305 | * below only happens on a VMRUN instruction intercept which has | |
3306 | * no valid exit_int_info set. | |
3307 | */ | |
3308 | if (vmcb->control.event_inj & SVM_EVTINJ_VALID) { | |
3309 | struct vmcb_control_area *nc = &nested_vmcb->control; | |
3310 | ||
3311 | nc->exit_int_info = vmcb->control.event_inj; | |
3312 | nc->exit_int_info_err = vmcb->control.event_inj_err; | |
3313 | } | |
3314 | ||
33740e40 JR |
3315 | nested_vmcb->control.tlb_ctl = 0; |
3316 | nested_vmcb->control.event_inj = 0; | |
3317 | nested_vmcb->control.event_inj_err = 0; | |
cf74a78b AG |
3318 | |
3319 | /* We always set V_INTR_MASKING and remember the old value in hflags */ | |
3320 | if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK)) | |
3321 | nested_vmcb->control.int_ctl &= ~V_INTR_MASKING_MASK; | |
3322 | ||
cf74a78b | 3323 | /* Restore the original control entries */ |
0460a979 | 3324 | copy_vmcb_control_area(vmcb, hsave); |
cf74a78b | 3325 | |
219b65dc AG |
3326 | kvm_clear_exception_queue(&svm->vcpu); |
3327 | kvm_clear_interrupt_queue(&svm->vcpu); | |
cf74a78b | 3328 | |
4b16184c JR |
3329 | svm->nested.nested_cr3 = 0; |
3330 | ||
cf74a78b AG |
3331 | /* Restore selected save entries */ |
3332 | svm->vmcb->save.es = hsave->save.es; | |
3333 | svm->vmcb->save.cs = hsave->save.cs; | |
3334 | svm->vmcb->save.ss = hsave->save.ss; | |
3335 | svm->vmcb->save.ds = hsave->save.ds; | |
3336 | svm->vmcb->save.gdtr = hsave->save.gdtr; | |
3337 | svm->vmcb->save.idtr = hsave->save.idtr; | |
f6e78475 | 3338 | kvm_set_rflags(&svm->vcpu, hsave->save.rflags); |
cf74a78b AG |
3339 | svm_set_efer(&svm->vcpu, hsave->save.efer); |
3340 | svm_set_cr0(&svm->vcpu, hsave->save.cr0 | X86_CR0_PE); | |
3341 | svm_set_cr4(&svm->vcpu, hsave->save.cr4); | |
3342 | if (npt_enabled) { | |
3343 | svm->vmcb->save.cr3 = hsave->save.cr3; | |
3344 | svm->vcpu.arch.cr3 = hsave->save.cr3; | |
3345 | } else { | |
2390218b | 3346 | (void)kvm_set_cr3(&svm->vcpu, hsave->save.cr3); |
cf74a78b AG |
3347 | } |
3348 | kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, hsave->save.rax); | |
3349 | kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, hsave->save.rsp); | |
3350 | kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, hsave->save.rip); | |
3351 | svm->vmcb->save.dr7 = 0; | |
3352 | svm->vmcb->save.cpl = 0; | |
3353 | svm->vmcb->control.exit_int_info = 0; | |
3354 | ||
8d28fec4 RJ |
3355 | mark_all_dirty(svm->vmcb); |
3356 | ||
7597f129 | 3357 | nested_svm_unmap(page); |
cf74a78b | 3358 | |
4b16184c | 3359 | nested_svm_uninit_mmu_context(&svm->vcpu); |
cf74a78b AG |
3360 | kvm_mmu_reset_context(&svm->vcpu); |
3361 | kvm_mmu_load(&svm->vcpu); | |
3362 | ||
3363 | return 0; | |
3364 | } | |
3d6368ef | 3365 | |
9738b2c9 | 3366 | static bool nested_svm_vmrun_msrpm(struct vcpu_svm *svm) |
3d6368ef | 3367 | { |
323c3d80 JR |
3368 | /* |
3369 | * This function merges the msr permission bitmaps of kvm and the | |
c5ec2e56 | 3370 | * nested vmcb. It is optimized in that it only merges the parts where |
323c3d80 JR |
3371 | * the kvm msr permission bitmap may contain zero bits |
3372 | */ | |
3d6368ef | 3373 | int i; |
9738b2c9 | 3374 | |
323c3d80 JR |
3375 | if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT))) |
3376 | return true; | |
9738b2c9 | 3377 | |
323c3d80 JR |
3378 | for (i = 0; i < MSRPM_OFFSETS; i++) { |
3379 | u32 value, p; | |
3380 | u64 offset; | |
9738b2c9 | 3381 | |
323c3d80 JR |
3382 | if (msrpm_offsets[i] == 0xffffffff) |
3383 | break; | |
3d6368ef | 3384 | |
0d6b3537 JR |
3385 | p = msrpm_offsets[i]; |
3386 | offset = svm->nested.vmcb_msrpm + (p * 4); | |
323c3d80 | 3387 | |
54bf36aa | 3388 | if (kvm_vcpu_read_guest(&svm->vcpu, offset, &value, 4)) |
323c3d80 JR |
3389 | return false; |
3390 | ||
3391 | svm->nested.msrpm[p] = svm->msrpm[p] | value; | |
3392 | } | |
3d6368ef | 3393 | |
d0ec49d4 | 3394 | svm->vmcb->control.msrpm_base_pa = __sme_set(__pa(svm->nested.msrpm)); |
9738b2c9 JR |
3395 | |
3396 | return true; | |
3d6368ef AG |
3397 | } |
3398 | ||
52c65a30 JR |
3399 | static bool nested_vmcb_checks(struct vmcb *vmcb) |
3400 | { | |
3401 | if ((vmcb->control.intercept & (1ULL << INTERCEPT_VMRUN)) == 0) | |
3402 | return false; | |
3403 | ||
dbe77584 JR |
3404 | if (vmcb->control.asid == 0) |
3405 | return false; | |
3406 | ||
cea3a19b TL |
3407 | if ((vmcb->control.nested_ctl & SVM_NESTED_CTL_NP_ENABLE) && |
3408 | !npt_enabled) | |
4b16184c JR |
3409 | return false; |
3410 | ||
52c65a30 JR |
3411 | return true; |
3412 | } | |
3413 | ||
c2634065 LP |
3414 | static void enter_svm_guest_mode(struct vcpu_svm *svm, u64 vmcb_gpa, |
3415 | struct vmcb *nested_vmcb, struct page *page) | |
3d6368ef | 3416 | { |
f6e78475 | 3417 | if (kvm_get_rflags(&svm->vcpu) & X86_EFLAGS_IF) |
3d6368ef AG |
3418 | svm->vcpu.arch.hflags |= HF_HIF_MASK; |
3419 | else | |
3420 | svm->vcpu.arch.hflags &= ~HF_HIF_MASK; | |
3421 | ||
cea3a19b | 3422 | if (nested_vmcb->control.nested_ctl & SVM_NESTED_CTL_NP_ENABLE) { |
4b16184c JR |
3423 | kvm_mmu_unload(&svm->vcpu); |
3424 | svm->nested.nested_cr3 = nested_vmcb->control.nested_cr3; | |
3425 | nested_svm_init_mmu_context(&svm->vcpu); | |
3426 | } | |
3427 | ||
3d6368ef AG |
3428 | /* Load the nested guest state */ |
3429 | svm->vmcb->save.es = nested_vmcb->save.es; | |
3430 | svm->vmcb->save.cs = nested_vmcb->save.cs; | |
3431 | svm->vmcb->save.ss = nested_vmcb->save.ss; | |
3432 | svm->vmcb->save.ds = nested_vmcb->save.ds; | |
3433 | svm->vmcb->save.gdtr = nested_vmcb->save.gdtr; | |
3434 | svm->vmcb->save.idtr = nested_vmcb->save.idtr; | |
f6e78475 | 3435 | kvm_set_rflags(&svm->vcpu, nested_vmcb->save.rflags); |
3d6368ef AG |
3436 | svm_set_efer(&svm->vcpu, nested_vmcb->save.efer); |
3437 | svm_set_cr0(&svm->vcpu, nested_vmcb->save.cr0); | |
3438 | svm_set_cr4(&svm->vcpu, nested_vmcb->save.cr4); | |
3439 | if (npt_enabled) { | |
3440 | svm->vmcb->save.cr3 = nested_vmcb->save.cr3; | |
3441 | svm->vcpu.arch.cr3 = nested_vmcb->save.cr3; | |
0e5cbe36 | 3442 | } else |
2390218b | 3443 | (void)kvm_set_cr3(&svm->vcpu, nested_vmcb->save.cr3); |
0e5cbe36 JR |
3444 | |
3445 | /* Guest paging mode is active - reset mmu */ | |
3446 | kvm_mmu_reset_context(&svm->vcpu); | |
3447 | ||
defbba56 | 3448 | svm->vmcb->save.cr2 = svm->vcpu.arch.cr2 = nested_vmcb->save.cr2; |
3d6368ef AG |
3449 | kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, nested_vmcb->save.rax); |
3450 | kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, nested_vmcb->save.rsp); | |
3451 | kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, nested_vmcb->save.rip); | |
e0231715 | 3452 | |
3d6368ef AG |
3453 | /* In case we don't even reach vcpu_run, the fields are not updated */ |
3454 | svm->vmcb->save.rax = nested_vmcb->save.rax; | |
3455 | svm->vmcb->save.rsp = nested_vmcb->save.rsp; | |
3456 | svm->vmcb->save.rip = nested_vmcb->save.rip; | |
3457 | svm->vmcb->save.dr7 = nested_vmcb->save.dr7; | |
3458 | svm->vmcb->save.dr6 = nested_vmcb->save.dr6; | |
3459 | svm->vmcb->save.cpl = nested_vmcb->save.cpl; | |
3460 | ||
f7138538 | 3461 | svm->nested.vmcb_msrpm = nested_vmcb->control.msrpm_base_pa & ~0x0fffULL; |
ce2ac085 | 3462 | svm->nested.vmcb_iopm = nested_vmcb->control.iopm_base_pa & ~0x0fffULL; |
3d6368ef | 3463 | |
aad42c64 | 3464 | /* cache intercepts */ |
4ee546b4 | 3465 | svm->nested.intercept_cr = nested_vmcb->control.intercept_cr; |
3aed041a | 3466 | svm->nested.intercept_dr = nested_vmcb->control.intercept_dr; |
aad42c64 JR |
3467 | svm->nested.intercept_exceptions = nested_vmcb->control.intercept_exceptions; |
3468 | svm->nested.intercept = nested_vmcb->control.intercept; | |
3469 | ||
c2ba05cc | 3470 | svm_flush_tlb(&svm->vcpu, true); |
3d6368ef | 3471 | svm->vmcb->control.int_ctl = nested_vmcb->control.int_ctl | V_INTR_MASKING_MASK; |
3d6368ef AG |
3472 | if (nested_vmcb->control.int_ctl & V_INTR_MASKING_MASK) |
3473 | svm->vcpu.arch.hflags |= HF_VINTR_MASK; | |
3474 | else | |
3475 | svm->vcpu.arch.hflags &= ~HF_VINTR_MASK; | |
3476 | ||
88ab24ad JR |
3477 | if (svm->vcpu.arch.hflags & HF_VINTR_MASK) { |
3478 | /* We only want the cr8 intercept bits of the guest */ | |
4ee546b4 RJ |
3479 | clr_cr_intercept(svm, INTERCEPT_CR8_READ); |
3480 | clr_cr_intercept(svm, INTERCEPT_CR8_WRITE); | |
88ab24ad JR |
3481 | } |
3482 | ||
0d945bd9 | 3483 | /* We don't want to see VMMCALLs from a nested guest */ |
8a05a1b8 | 3484 | clr_intercept(svm, INTERCEPT_VMMCALL); |
0d945bd9 | 3485 | |
0dc92119 | 3486 | svm->vmcb->control.virt_ext = nested_vmcb->control.virt_ext; |
3d6368ef AG |
3487 | svm->vmcb->control.int_vector = nested_vmcb->control.int_vector; |
3488 | svm->vmcb->control.int_state = nested_vmcb->control.int_state; | |
3489 | svm->vmcb->control.tsc_offset += nested_vmcb->control.tsc_offset; | |
3d6368ef AG |
3490 | svm->vmcb->control.event_inj = nested_vmcb->control.event_inj; |
3491 | svm->vmcb->control.event_inj_err = nested_vmcb->control.event_inj_err; | |
3492 | ||
7597f129 | 3493 | nested_svm_unmap(page); |
9738b2c9 | 3494 | |
2030753d JR |
3495 | /* Enter Guest-Mode */ |
3496 | enter_guest_mode(&svm->vcpu); | |
3497 | ||
384c6368 JR |
3498 | /* |
3499 | * Merge guest and host intercepts - must be called with vcpu in | |
3500 | * guest-mode to take affect here | |
3501 | */ | |
3502 | recalc_intercepts(svm); | |
3503 | ||
06fc7772 | 3504 | svm->nested.vmcb = vmcb_gpa; |
9738b2c9 | 3505 | |
2af9194d | 3506 | enable_gif(svm); |
3d6368ef | 3507 | |
8d28fec4 | 3508 | mark_all_dirty(svm->vmcb); |
c2634065 LP |
3509 | } |
3510 | ||
3511 | static bool nested_svm_vmrun(struct vcpu_svm *svm) | |
3512 | { | |
3513 | struct vmcb *nested_vmcb; | |
3514 | struct vmcb *hsave = svm->nested.hsave; | |
3515 | struct vmcb *vmcb = svm->vmcb; | |
3516 | struct page *page; | |
3517 | u64 vmcb_gpa; | |
3518 | ||
3519 | vmcb_gpa = svm->vmcb->save.rax; | |
3520 | ||
3521 | nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page); | |
3522 | if (!nested_vmcb) | |
3523 | return false; | |
3524 | ||
3525 | if (!nested_vmcb_checks(nested_vmcb)) { | |
3526 | nested_vmcb->control.exit_code = SVM_EXIT_ERR; | |
3527 | nested_vmcb->control.exit_code_hi = 0; | |
3528 | nested_vmcb->control.exit_info_1 = 0; | |
3529 | nested_vmcb->control.exit_info_2 = 0; | |
3530 | ||
3531 | nested_svm_unmap(page); | |
3532 | ||
3533 | return false; | |
3534 | } | |
3535 | ||
3536 | trace_kvm_nested_vmrun(svm->vmcb->save.rip, vmcb_gpa, | |
3537 | nested_vmcb->save.rip, | |
3538 | nested_vmcb->control.int_ctl, | |
3539 | nested_vmcb->control.event_inj, | |
3540 | nested_vmcb->control.nested_ctl); | |
3541 | ||
3542 | trace_kvm_nested_intercepts(nested_vmcb->control.intercept_cr & 0xffff, | |
3543 | nested_vmcb->control.intercept_cr >> 16, | |
3544 | nested_vmcb->control.intercept_exceptions, | |
3545 | nested_vmcb->control.intercept); | |
3546 | ||
3547 | /* Clear internal status */ | |
3548 | kvm_clear_exception_queue(&svm->vcpu); | |
3549 | kvm_clear_interrupt_queue(&svm->vcpu); | |
3550 | ||
3551 | /* | |
3552 | * Save the old vmcb, so we don't need to pick what we save, but can | |
3553 | * restore everything when a VMEXIT occurs | |
3554 | */ | |
3555 | hsave->save.es = vmcb->save.es; | |
3556 | hsave->save.cs = vmcb->save.cs; | |
3557 | hsave->save.ss = vmcb->save.ss; | |
3558 | hsave->save.ds = vmcb->save.ds; | |
3559 | hsave->save.gdtr = vmcb->save.gdtr; | |
3560 | hsave->save.idtr = vmcb->save.idtr; | |
3561 | hsave->save.efer = svm->vcpu.arch.efer; | |
3562 | hsave->save.cr0 = kvm_read_cr0(&svm->vcpu); | |
3563 | hsave->save.cr4 = svm->vcpu.arch.cr4; | |
3564 | hsave->save.rflags = kvm_get_rflags(&svm->vcpu); | |
3565 | hsave->save.rip = kvm_rip_read(&svm->vcpu); | |
3566 | hsave->save.rsp = vmcb->save.rsp; | |
3567 | hsave->save.rax = vmcb->save.rax; | |
3568 | if (npt_enabled) | |
3569 | hsave->save.cr3 = vmcb->save.cr3; | |
3570 | else | |
3571 | hsave->save.cr3 = kvm_read_cr3(&svm->vcpu); | |
3572 | ||
3573 | copy_vmcb_control_area(hsave, vmcb); | |
3574 | ||
3575 | enter_svm_guest_mode(svm, vmcb_gpa, nested_vmcb, page); | |
8d28fec4 | 3576 | |
9738b2c9 | 3577 | return true; |
3d6368ef AG |
3578 | } |
3579 | ||
9966bf68 | 3580 | static void nested_svm_vmloadsave(struct vmcb *from_vmcb, struct vmcb *to_vmcb) |
5542675b AG |
3581 | { |
3582 | to_vmcb->save.fs = from_vmcb->save.fs; | |
3583 | to_vmcb->save.gs = from_vmcb->save.gs; | |
3584 | to_vmcb->save.tr = from_vmcb->save.tr; | |
3585 | to_vmcb->save.ldtr = from_vmcb->save.ldtr; | |
3586 | to_vmcb->save.kernel_gs_base = from_vmcb->save.kernel_gs_base; | |
3587 | to_vmcb->save.star = from_vmcb->save.star; | |
3588 | to_vmcb->save.lstar = from_vmcb->save.lstar; | |
3589 | to_vmcb->save.cstar = from_vmcb->save.cstar; | |
3590 | to_vmcb->save.sfmask = from_vmcb->save.sfmask; | |
3591 | to_vmcb->save.sysenter_cs = from_vmcb->save.sysenter_cs; | |
3592 | to_vmcb->save.sysenter_esp = from_vmcb->save.sysenter_esp; | |
3593 | to_vmcb->save.sysenter_eip = from_vmcb->save.sysenter_eip; | |
5542675b AG |
3594 | } |
3595 | ||
851ba692 | 3596 | static int vmload_interception(struct vcpu_svm *svm) |
5542675b | 3597 | { |
9966bf68 | 3598 | struct vmcb *nested_vmcb; |
7597f129 | 3599 | struct page *page; |
b742c1e6 | 3600 | int ret; |
9966bf68 | 3601 | |
5542675b AG |
3602 | if (nested_svm_check_permissions(svm)) |
3603 | return 1; | |
3604 | ||
7597f129 | 3605 | nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page); |
9966bf68 JR |
3606 | if (!nested_vmcb) |
3607 | return 1; | |
3608 | ||
e3e9ed3d | 3609 | svm->next_rip = kvm_rip_read(&svm->vcpu) + 3; |
b742c1e6 | 3610 | ret = kvm_skip_emulated_instruction(&svm->vcpu); |
e3e9ed3d | 3611 | |
9966bf68 | 3612 | nested_svm_vmloadsave(nested_vmcb, svm->vmcb); |
7597f129 | 3613 | nested_svm_unmap(page); |
5542675b | 3614 | |
b742c1e6 | 3615 | return ret; |
5542675b AG |
3616 | } |
3617 | ||
851ba692 | 3618 | static int vmsave_interception(struct vcpu_svm *svm) |
5542675b | 3619 | { |
9966bf68 | 3620 | struct vmcb *nested_vmcb; |
7597f129 | 3621 | struct page *page; |
b742c1e6 | 3622 | int ret; |
9966bf68 | 3623 | |
5542675b AG |
3624 | if (nested_svm_check_permissions(svm)) |
3625 | return 1; | |
3626 | ||
7597f129 | 3627 | nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page); |
9966bf68 JR |
3628 | if (!nested_vmcb) |
3629 | return 1; | |
3630 | ||
e3e9ed3d | 3631 | svm->next_rip = kvm_rip_read(&svm->vcpu) + 3; |
b742c1e6 | 3632 | ret = kvm_skip_emulated_instruction(&svm->vcpu); |
e3e9ed3d | 3633 | |
9966bf68 | 3634 | nested_svm_vmloadsave(svm->vmcb, nested_vmcb); |
7597f129 | 3635 | nested_svm_unmap(page); |
5542675b | 3636 | |
b742c1e6 | 3637 | return ret; |
5542675b AG |
3638 | } |
3639 | ||
851ba692 | 3640 | static int vmrun_interception(struct vcpu_svm *svm) |
3d6368ef | 3641 | { |
3d6368ef AG |
3642 | if (nested_svm_check_permissions(svm)) |
3643 | return 1; | |
3644 | ||
b75f4eb3 RJ |
3645 | /* Save rip after vmrun instruction */ |
3646 | kvm_rip_write(&svm->vcpu, kvm_rip_read(&svm->vcpu) + 3); | |
3d6368ef | 3647 | |
9738b2c9 | 3648 | if (!nested_svm_vmrun(svm)) |
3d6368ef AG |
3649 | return 1; |
3650 | ||
9738b2c9 | 3651 | if (!nested_svm_vmrun_msrpm(svm)) |
1f8da478 JR |
3652 | goto failed; |
3653 | ||
3654 | return 1; | |
3655 | ||
3656 | failed: | |
3657 | ||
3658 | svm->vmcb->control.exit_code = SVM_EXIT_ERR; | |
3659 | svm->vmcb->control.exit_code_hi = 0; | |
3660 | svm->vmcb->control.exit_info_1 = 0; | |
3661 | svm->vmcb->control.exit_info_2 = 0; | |
3662 | ||
3663 | nested_svm_vmexit(svm); | |
3d6368ef AG |
3664 | |
3665 | return 1; | |
3666 | } | |
3667 | ||
851ba692 | 3668 | static int stgi_interception(struct vcpu_svm *svm) |
1371d904 | 3669 | { |
b742c1e6 LP |
3670 | int ret; |
3671 | ||
1371d904 AG |
3672 | if (nested_svm_check_permissions(svm)) |
3673 | return 1; | |
3674 | ||
640bd6e5 JN |
3675 | /* |
3676 | * If VGIF is enabled, the STGI intercept is only added to | |
cc3d967f | 3677 | * detect the opening of the SMI/NMI window; remove it now. |
640bd6e5 JN |
3678 | */ |
3679 | if (vgif_enabled(svm)) | |
3680 | clr_intercept(svm, INTERCEPT_STGI); | |
3681 | ||
1371d904 | 3682 | svm->next_rip = kvm_rip_read(&svm->vcpu) + 3; |
b742c1e6 | 3683 | ret = kvm_skip_emulated_instruction(&svm->vcpu); |
3842d135 | 3684 | kvm_make_request(KVM_REQ_EVENT, &svm->vcpu); |
1371d904 | 3685 | |
2af9194d | 3686 | enable_gif(svm); |
1371d904 | 3687 | |
b742c1e6 | 3688 | return ret; |
1371d904 AG |
3689 | } |
3690 | ||
851ba692 | 3691 | static int clgi_interception(struct vcpu_svm *svm) |
1371d904 | 3692 | { |
b742c1e6 LP |
3693 | int ret; |
3694 | ||
1371d904 AG |
3695 | if (nested_svm_check_permissions(svm)) |
3696 | return 1; | |
3697 | ||
3698 | svm->next_rip = kvm_rip_read(&svm->vcpu) + 3; | |
b742c1e6 | 3699 | ret = kvm_skip_emulated_instruction(&svm->vcpu); |
1371d904 | 3700 | |
2af9194d | 3701 | disable_gif(svm); |
1371d904 AG |
3702 | |
3703 | /* After a CLGI no interrupts should come */ | |
340d3bc3 SS |
3704 | if (!kvm_vcpu_apicv_active(&svm->vcpu)) { |
3705 | svm_clear_vintr(svm); | |
3706 | svm->vmcb->control.int_ctl &= ~V_IRQ_MASK; | |
3707 | mark_dirty(svm->vmcb, VMCB_INTR); | |
3708 | } | |
decdbf6a | 3709 | |
b742c1e6 | 3710 | return ret; |
1371d904 AG |
3711 | } |
3712 | ||
851ba692 | 3713 | static int invlpga_interception(struct vcpu_svm *svm) |
ff092385 AG |
3714 | { |
3715 | struct kvm_vcpu *vcpu = &svm->vcpu; | |
ff092385 | 3716 | |
668f198f DK |
3717 | trace_kvm_invlpga(svm->vmcb->save.rip, kvm_register_read(&svm->vcpu, VCPU_REGS_RCX), |
3718 | kvm_register_read(&svm->vcpu, VCPU_REGS_RAX)); | |
ec1ff790 | 3719 | |
ff092385 | 3720 | /* Let's treat INVLPGA the same as INVLPG (can be optimized!) */ |
668f198f | 3721 | kvm_mmu_invlpg(vcpu, kvm_register_read(&svm->vcpu, VCPU_REGS_RAX)); |
ff092385 AG |
3722 | |
3723 | svm->next_rip = kvm_rip_read(&svm->vcpu) + 3; | |
b742c1e6 | 3724 | return kvm_skip_emulated_instruction(&svm->vcpu); |
ff092385 AG |
3725 | } |
3726 | ||
532a46b9 JR |
3727 | static int skinit_interception(struct vcpu_svm *svm) |
3728 | { | |
668f198f | 3729 | trace_kvm_skinit(svm->vmcb->save.rip, kvm_register_read(&svm->vcpu, VCPU_REGS_RAX)); |
532a46b9 JR |
3730 | |
3731 | kvm_queue_exception(&svm->vcpu, UD_VECTOR); | |
3732 | return 1; | |
3733 | } | |
3734 | ||
dab429a7 DK |
3735 | static int wbinvd_interception(struct vcpu_svm *svm) |
3736 | { | |
6affcbed | 3737 | return kvm_emulate_wbinvd(&svm->vcpu); |
dab429a7 DK |
3738 | } |
3739 | ||
81dd35d4 JR |
3740 | static int xsetbv_interception(struct vcpu_svm *svm) |
3741 | { | |
3742 | u64 new_bv = kvm_read_edx_eax(&svm->vcpu); | |
3743 | u32 index = kvm_register_read(&svm->vcpu, VCPU_REGS_RCX); | |
3744 | ||
3745 | if (kvm_set_xcr(&svm->vcpu, index, new_bv) == 0) { | |
3746 | svm->next_rip = kvm_rip_read(&svm->vcpu) + 3; | |
b742c1e6 | 3747 | return kvm_skip_emulated_instruction(&svm->vcpu); |
81dd35d4 JR |
3748 | } |
3749 | ||
3750 | return 1; | |
3751 | } | |
3752 | ||
851ba692 | 3753 | static int task_switch_interception(struct vcpu_svm *svm) |
6aa8b732 | 3754 | { |
37817f29 | 3755 | u16 tss_selector; |
64a7ec06 GN |
3756 | int reason; |
3757 | int int_type = svm->vmcb->control.exit_int_info & | |
3758 | SVM_EXITINTINFO_TYPE_MASK; | |
8317c298 | 3759 | int int_vec = svm->vmcb->control.exit_int_info & SVM_EVTINJ_VEC_MASK; |
fe8e7f83 GN |
3760 | uint32_t type = |
3761 | svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_TYPE_MASK; | |
3762 | uint32_t idt_v = | |
3763 | svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_VALID; | |
e269fb21 JK |
3764 | bool has_error_code = false; |
3765 | u32 error_code = 0; | |
37817f29 IE |
3766 | |
3767 | tss_selector = (u16)svm->vmcb->control.exit_info_1; | |
64a7ec06 | 3768 | |
37817f29 IE |
3769 | if (svm->vmcb->control.exit_info_2 & |
3770 | (1ULL << SVM_EXITINFOSHIFT_TS_REASON_IRET)) | |
64a7ec06 GN |
3771 | reason = TASK_SWITCH_IRET; |
3772 | else if (svm->vmcb->control.exit_info_2 & | |
3773 | (1ULL << SVM_EXITINFOSHIFT_TS_REASON_JMP)) | |
3774 | reason = TASK_SWITCH_JMP; | |
fe8e7f83 | 3775 | else if (idt_v) |
64a7ec06 GN |
3776 | reason = TASK_SWITCH_GATE; |
3777 | else | |
3778 | reason = TASK_SWITCH_CALL; | |
3779 | ||
fe8e7f83 GN |
3780 | if (reason == TASK_SWITCH_GATE) { |
3781 | switch (type) { | |
3782 | case SVM_EXITINTINFO_TYPE_NMI: | |
3783 | svm->vcpu.arch.nmi_injected = false; | |
3784 | break; | |
3785 | case SVM_EXITINTINFO_TYPE_EXEPT: | |
e269fb21 JK |
3786 | if (svm->vmcb->control.exit_info_2 & |
3787 | (1ULL << SVM_EXITINFOSHIFT_TS_HAS_ERROR_CODE)) { | |
3788 | has_error_code = true; | |
3789 | error_code = | |
3790 | (u32)svm->vmcb->control.exit_info_2; | |
3791 | } | |
fe8e7f83 GN |
3792 | kvm_clear_exception_queue(&svm->vcpu); |
3793 | break; | |
3794 | case SVM_EXITINTINFO_TYPE_INTR: | |
3795 | kvm_clear_interrupt_queue(&svm->vcpu); | |
3796 | break; | |
3797 | default: | |
3798 | break; | |
3799 | } | |
3800 | } | |
64a7ec06 | 3801 | |
8317c298 GN |
3802 | if (reason != TASK_SWITCH_GATE || |
3803 | int_type == SVM_EXITINTINFO_TYPE_SOFT || | |
3804 | (int_type == SVM_EXITINTINFO_TYPE_EXEPT && | |
f629cf84 GN |
3805 | (int_vec == OF_VECTOR || int_vec == BP_VECTOR))) |
3806 | skip_emulated_instruction(&svm->vcpu); | |
64a7ec06 | 3807 | |
7f3d35fd KW |
3808 | if (int_type != SVM_EXITINTINFO_TYPE_SOFT) |
3809 | int_vec = -1; | |
3810 | ||
3811 | if (kvm_task_switch(&svm->vcpu, tss_selector, int_vec, reason, | |
acb54517 GN |
3812 | has_error_code, error_code) == EMULATE_FAIL) { |
3813 | svm->vcpu.run->exit_reason = KVM_EXIT_INTERNAL_ERROR; | |
3814 | svm->vcpu.run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION; | |
3815 | svm->vcpu.run->internal.ndata = 0; | |
3816 | return 0; | |
3817 | } | |
3818 | return 1; | |
6aa8b732 AK |
3819 | } |
3820 | ||
851ba692 | 3821 | static int cpuid_interception(struct vcpu_svm *svm) |
6aa8b732 | 3822 | { |
5fdbf976 | 3823 | svm->next_rip = kvm_rip_read(&svm->vcpu) + 2; |
6a908b62 | 3824 | return kvm_emulate_cpuid(&svm->vcpu); |
6aa8b732 AK |
3825 | } |
3826 | ||
851ba692 | 3827 | static int iret_interception(struct vcpu_svm *svm) |
95ba8273 GN |
3828 | { |
3829 | ++svm->vcpu.stat.nmi_window_exits; | |
8a05a1b8 | 3830 | clr_intercept(svm, INTERCEPT_IRET); |
44c11430 | 3831 | svm->vcpu.arch.hflags |= HF_IRET_MASK; |
bd3d1ec3 | 3832 | svm->nmi_iret_rip = kvm_rip_read(&svm->vcpu); |
f303b4ce | 3833 | kvm_make_request(KVM_REQ_EVENT, &svm->vcpu); |
95ba8273 GN |
3834 | return 1; |
3835 | } | |
3836 | ||
851ba692 | 3837 | static int invlpg_interception(struct vcpu_svm *svm) |
a7052897 | 3838 | { |
df4f3108 AP |
3839 | if (!static_cpu_has(X86_FEATURE_DECODEASSISTS)) |
3840 | return emulate_instruction(&svm->vcpu, 0) == EMULATE_DONE; | |
3841 | ||
3842 | kvm_mmu_invlpg(&svm->vcpu, svm->vmcb->control.exit_info_1); | |
b742c1e6 | 3843 | return kvm_skip_emulated_instruction(&svm->vcpu); |
a7052897 MT |
3844 | } |
3845 | ||
851ba692 | 3846 | static int emulate_on_interception(struct vcpu_svm *svm) |
6aa8b732 | 3847 | { |
51d8b661 | 3848 | return emulate_instruction(&svm->vcpu, 0) == EMULATE_DONE; |
6aa8b732 AK |
3849 | } |
3850 | ||
7607b717 BS |
3851 | static int rsm_interception(struct vcpu_svm *svm) |
3852 | { | |
3853 | return x86_emulate_instruction(&svm->vcpu, 0, 0, | |
3854 | rsm_ins_bytes, 2) == EMULATE_DONE; | |
3855 | } | |
3856 | ||
332b56e4 AK |
3857 | static int rdpmc_interception(struct vcpu_svm *svm) |
3858 | { | |
3859 | int err; | |
3860 | ||
3861 | if (!static_cpu_has(X86_FEATURE_NRIPS)) | |
3862 | return emulate_on_interception(svm); | |
3863 | ||
3864 | err = kvm_rdpmc(&svm->vcpu); | |
6affcbed | 3865 | return kvm_complete_insn_gp(&svm->vcpu, err); |
332b56e4 AK |
3866 | } |
3867 | ||
52eb5a6d XL |
3868 | static bool check_selective_cr0_intercepted(struct vcpu_svm *svm, |
3869 | unsigned long val) | |
628afd2a JR |
3870 | { |
3871 | unsigned long cr0 = svm->vcpu.arch.cr0; | |
3872 | bool ret = false; | |
3873 | u64 intercept; | |
3874 | ||
3875 | intercept = svm->nested.intercept; | |
3876 | ||
3877 | if (!is_guest_mode(&svm->vcpu) || | |
3878 | (!(intercept & (1ULL << INTERCEPT_SELECTIVE_CR0)))) | |
3879 | return false; | |
3880 | ||
3881 | cr0 &= ~SVM_CR0_SELECTIVE_MASK; | |
3882 | val &= ~SVM_CR0_SELECTIVE_MASK; | |
3883 | ||
3884 | if (cr0 ^ val) { | |
3885 | svm->vmcb->control.exit_code = SVM_EXIT_CR0_SEL_WRITE; | |
3886 | ret = (nested_svm_exit_handled(svm) == NESTED_EXIT_DONE); | |
3887 | } | |
3888 | ||
3889 | return ret; | |
3890 | } | |
3891 | ||
7ff76d58 AP |
3892 | #define CR_VALID (1ULL << 63) |
3893 | ||
3894 | static int cr_interception(struct vcpu_svm *svm) | |
3895 | { | |
3896 | int reg, cr; | |
3897 | unsigned long val; | |
3898 | int err; | |
3899 | ||
3900 | if (!static_cpu_has(X86_FEATURE_DECODEASSISTS)) | |
3901 | return emulate_on_interception(svm); | |
3902 | ||
3903 | if (unlikely((svm->vmcb->control.exit_info_1 & CR_VALID) == 0)) | |
3904 | return emulate_on_interception(svm); | |
3905 | ||
3906 | reg = svm->vmcb->control.exit_info_1 & SVM_EXITINFO_REG_MASK; | |
5e57518d DK |
3907 | if (svm->vmcb->control.exit_code == SVM_EXIT_CR0_SEL_WRITE) |
3908 | cr = SVM_EXIT_WRITE_CR0 - SVM_EXIT_READ_CR0; | |
3909 | else | |
3910 | cr = svm->vmcb->control.exit_code - SVM_EXIT_READ_CR0; | |
7ff76d58 AP |
3911 | |
3912 | err = 0; | |
3913 | if (cr >= 16) { /* mov to cr */ | |
3914 | cr -= 16; | |
3915 | val = kvm_register_read(&svm->vcpu, reg); | |
3916 | switch (cr) { | |
3917 | case 0: | |
628afd2a JR |
3918 | if (!check_selective_cr0_intercepted(svm, val)) |
3919 | err = kvm_set_cr0(&svm->vcpu, val); | |
977b2d03 JR |
3920 | else |
3921 | return 1; | |
3922 | ||
7ff76d58 AP |
3923 | break; |
3924 | case 3: | |
3925 | err = kvm_set_cr3(&svm->vcpu, val); | |
3926 | break; | |
3927 | case 4: | |
3928 | err = kvm_set_cr4(&svm->vcpu, val); | |
3929 | break; | |
3930 | case 8: | |
3931 | err = kvm_set_cr8(&svm->vcpu, val); | |
3932 | break; | |
3933 | default: | |
3934 | WARN(1, "unhandled write to CR%d", cr); | |
3935 | kvm_queue_exception(&svm->vcpu, UD_VECTOR); | |
3936 | return 1; | |
3937 | } | |
3938 | } else { /* mov from cr */ | |
3939 | switch (cr) { | |
3940 | case 0: | |
3941 | val = kvm_read_cr0(&svm->vcpu); | |
3942 | break; | |
3943 | case 2: | |
3944 | val = svm->vcpu.arch.cr2; | |
3945 | break; | |
3946 | case 3: | |
9f8fe504 | 3947 | val = kvm_read_cr3(&svm->vcpu); |
7ff76d58 AP |
3948 | break; |
3949 | case 4: | |
3950 | val = kvm_read_cr4(&svm->vcpu); | |
3951 | break; | |
3952 | case 8: | |
3953 | val = kvm_get_cr8(&svm->vcpu); | |
3954 | break; | |
3955 | default: | |
3956 | WARN(1, "unhandled read from CR%d", cr); | |
3957 | kvm_queue_exception(&svm->vcpu, UD_VECTOR); | |
3958 | return 1; | |
3959 | } | |
3960 | kvm_register_write(&svm->vcpu, reg, val); | |
3961 | } | |
6affcbed | 3962 | return kvm_complete_insn_gp(&svm->vcpu, err); |
7ff76d58 AP |
3963 | } |
3964 | ||
cae3797a AP |
3965 | static int dr_interception(struct vcpu_svm *svm) |
3966 | { | |
3967 | int reg, dr; | |
3968 | unsigned long val; | |
cae3797a | 3969 | |
facb0139 PB |
3970 | if (svm->vcpu.guest_debug == 0) { |
3971 | /* | |
3972 | * No more DR vmexits; force a reload of the debug registers | |
3973 | * and reenter on this instruction. The next vmexit will | |
3974 | * retrieve the full state of the debug registers. | |
3975 | */ | |
3976 | clr_dr_intercepts(svm); | |
3977 | svm->vcpu.arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT; | |
3978 | return 1; | |
3979 | } | |
3980 | ||
cae3797a AP |
3981 | if (!boot_cpu_has(X86_FEATURE_DECODEASSISTS)) |
3982 | return emulate_on_interception(svm); | |
3983 | ||
3984 | reg = svm->vmcb->control.exit_info_1 & SVM_EXITINFO_REG_MASK; | |
3985 | dr = svm->vmcb->control.exit_code - SVM_EXIT_READ_DR0; | |
3986 | ||
3987 | if (dr >= 16) { /* mov to DRn */ | |
16f8a6f9 NA |
3988 | if (!kvm_require_dr(&svm->vcpu, dr - 16)) |
3989 | return 1; | |
cae3797a AP |
3990 | val = kvm_register_read(&svm->vcpu, reg); |
3991 | kvm_set_dr(&svm->vcpu, dr - 16, val); | |
3992 | } else { | |
16f8a6f9 NA |
3993 | if (!kvm_require_dr(&svm->vcpu, dr)) |
3994 | return 1; | |
3995 | kvm_get_dr(&svm->vcpu, dr, &val); | |
3996 | kvm_register_write(&svm->vcpu, reg, val); | |
cae3797a AP |
3997 | } |
3998 | ||
b742c1e6 | 3999 | return kvm_skip_emulated_instruction(&svm->vcpu); |
cae3797a AP |
4000 | } |
4001 | ||
851ba692 | 4002 | static int cr8_write_interception(struct vcpu_svm *svm) |
1d075434 | 4003 | { |
851ba692 | 4004 | struct kvm_run *kvm_run = svm->vcpu.run; |
eea1cff9 | 4005 | int r; |
851ba692 | 4006 | |
0a5fff19 GN |
4007 | u8 cr8_prev = kvm_get_cr8(&svm->vcpu); |
4008 | /* instruction emulation calls kvm_set_cr8() */ | |
7ff76d58 | 4009 | r = cr_interception(svm); |
35754c98 | 4010 | if (lapic_in_kernel(&svm->vcpu)) |
7ff76d58 | 4011 | return r; |
0a5fff19 | 4012 | if (cr8_prev <= kvm_get_cr8(&svm->vcpu)) |
7ff76d58 | 4013 | return r; |
1d075434 JR |
4014 | kvm_run->exit_reason = KVM_EXIT_SET_TPR; |
4015 | return 0; | |
4016 | } | |
4017 | ||
801e459a TL |
4018 | static int svm_get_msr_feature(struct kvm_msr_entry *msr) |
4019 | { | |
d1d93fa9 TL |
4020 | msr->data = 0; |
4021 | ||
4022 | switch (msr->index) { | |
4023 | case MSR_F10H_DECFG: | |
4024 | if (boot_cpu_has(X86_FEATURE_LFENCE_RDTSC)) | |
4025 | msr->data |= MSR_F10H_DECFG_LFENCE_SERIALIZE; | |
4026 | break; | |
4027 | default: | |
4028 | return 1; | |
4029 | } | |
4030 | ||
4031 | return 0; | |
801e459a TL |
4032 | } |
4033 | ||
609e36d3 | 4034 | static int svm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info) |
6aa8b732 | 4035 | { |
a2fa3e9f GH |
4036 | struct vcpu_svm *svm = to_svm(vcpu); |
4037 | ||
609e36d3 | 4038 | switch (msr_info->index) { |
af24a4e4 | 4039 | case MSR_IA32_TSC: { |
609e36d3 | 4040 | msr_info->data = svm->vmcb->control.tsc_offset + |
35181e86 | 4041 | kvm_scale_tsc(vcpu, rdtsc()); |
fbc0db76 | 4042 | |
6aa8b732 AK |
4043 | break; |
4044 | } | |
8c06585d | 4045 | case MSR_STAR: |
609e36d3 | 4046 | msr_info->data = svm->vmcb->save.star; |
6aa8b732 | 4047 | break; |
0e859cac | 4048 | #ifdef CONFIG_X86_64 |
6aa8b732 | 4049 | case MSR_LSTAR: |
609e36d3 | 4050 | msr_info->data = svm->vmcb->save.lstar; |
6aa8b732 AK |
4051 | break; |
4052 | case MSR_CSTAR: | |
609e36d3 | 4053 | msr_info->data = svm->vmcb->save.cstar; |
6aa8b732 AK |
4054 | break; |
4055 | case MSR_KERNEL_GS_BASE: | |
609e36d3 | 4056 | msr_info->data = svm->vmcb->save.kernel_gs_base; |
6aa8b732 AK |
4057 | break; |
4058 | case MSR_SYSCALL_MASK: | |
609e36d3 | 4059 | msr_info->data = svm->vmcb->save.sfmask; |
6aa8b732 AK |
4060 | break; |
4061 | #endif | |
4062 | case MSR_IA32_SYSENTER_CS: | |
609e36d3 | 4063 | msr_info->data = svm->vmcb->save.sysenter_cs; |
6aa8b732 AK |
4064 | break; |
4065 | case MSR_IA32_SYSENTER_EIP: | |
609e36d3 | 4066 | msr_info->data = svm->sysenter_eip; |
6aa8b732 AK |
4067 | break; |
4068 | case MSR_IA32_SYSENTER_ESP: | |
609e36d3 | 4069 | msr_info->data = svm->sysenter_esp; |
6aa8b732 | 4070 | break; |
46896c73 PB |
4071 | case MSR_TSC_AUX: |
4072 | if (!boot_cpu_has(X86_FEATURE_RDTSCP)) | |
4073 | return 1; | |
4074 | msr_info->data = svm->tsc_aux; | |
4075 | break; | |
e0231715 JR |
4076 | /* |
4077 | * Nobody will change the following 5 values in the VMCB so we can | |
4078 | * safely return them on rdmsr. They will always be 0 until LBRV is | |
4079 | * implemented. | |
4080 | */ | |
a2938c80 | 4081 | case MSR_IA32_DEBUGCTLMSR: |
609e36d3 | 4082 | msr_info->data = svm->vmcb->save.dbgctl; |
a2938c80 JR |
4083 | break; |
4084 | case MSR_IA32_LASTBRANCHFROMIP: | |
609e36d3 | 4085 | msr_info->data = svm->vmcb->save.br_from; |
a2938c80 JR |
4086 | break; |
4087 | case MSR_IA32_LASTBRANCHTOIP: | |
609e36d3 | 4088 | msr_info->data = svm->vmcb->save.br_to; |
a2938c80 JR |
4089 | break; |
4090 | case MSR_IA32_LASTINTFROMIP: | |
609e36d3 | 4091 | msr_info->data = svm->vmcb->save.last_excp_from; |
a2938c80 JR |
4092 | break; |
4093 | case MSR_IA32_LASTINTTOIP: | |
609e36d3 | 4094 | msr_info->data = svm->vmcb->save.last_excp_to; |
a2938c80 | 4095 | break; |
b286d5d8 | 4096 | case MSR_VM_HSAVE_PA: |
609e36d3 | 4097 | msr_info->data = svm->nested.hsave_msr; |
b286d5d8 | 4098 | break; |
eb6f302e | 4099 | case MSR_VM_CR: |
609e36d3 | 4100 | msr_info->data = svm->nested.vm_cr_msr; |
eb6f302e | 4101 | break; |
b2ac58f9 KA |
4102 | case MSR_IA32_SPEC_CTRL: |
4103 | if (!msr_info->host_initiated && | |
4104 | !guest_cpuid_has(vcpu, X86_FEATURE_IBRS)) | |
4105 | return 1; | |
4106 | ||
4107 | msr_info->data = svm->spec_ctrl; | |
4108 | break; | |
ae8b7875 BP |
4109 | case MSR_F15H_IC_CFG: { |
4110 | ||
4111 | int family, model; | |
4112 | ||
4113 | family = guest_cpuid_family(vcpu); | |
4114 | model = guest_cpuid_model(vcpu); | |
4115 | ||
4116 | if (family < 0 || model < 0) | |
4117 | return kvm_get_msr_common(vcpu, msr_info); | |
4118 | ||
4119 | msr_info->data = 0; | |
4120 | ||
4121 | if (family == 0x15 && | |
4122 | (model >= 0x2 && model < 0x20)) | |
4123 | msr_info->data = 0x1E; | |
4124 | } | |
4125 | break; | |
d1d93fa9 TL |
4126 | case MSR_F10H_DECFG: |
4127 | msr_info->data = svm->msr_decfg; | |
4128 | break; | |
6aa8b732 | 4129 | default: |
609e36d3 | 4130 | return kvm_get_msr_common(vcpu, msr_info); |
6aa8b732 AK |
4131 | } |
4132 | return 0; | |
4133 | } | |
4134 | ||
851ba692 | 4135 | static int rdmsr_interception(struct vcpu_svm *svm) |
6aa8b732 | 4136 | { |
668f198f | 4137 | u32 ecx = kvm_register_read(&svm->vcpu, VCPU_REGS_RCX); |
609e36d3 | 4138 | struct msr_data msr_info; |
6aa8b732 | 4139 | |
609e36d3 PB |
4140 | msr_info.index = ecx; |
4141 | msr_info.host_initiated = false; | |
4142 | if (svm_get_msr(&svm->vcpu, &msr_info)) { | |
59200273 | 4143 | trace_kvm_msr_read_ex(ecx); |
c1a5d4f9 | 4144 | kvm_inject_gp(&svm->vcpu, 0); |
b742c1e6 | 4145 | return 1; |
59200273 | 4146 | } else { |
609e36d3 | 4147 | trace_kvm_msr_read(ecx, msr_info.data); |
af9ca2d7 | 4148 | |
609e36d3 PB |
4149 | kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, |
4150 | msr_info.data & 0xffffffff); | |
4151 | kvm_register_write(&svm->vcpu, VCPU_REGS_RDX, | |
4152 | msr_info.data >> 32); | |
5fdbf976 | 4153 | svm->next_rip = kvm_rip_read(&svm->vcpu) + 2; |
b742c1e6 | 4154 | return kvm_skip_emulated_instruction(&svm->vcpu); |
6aa8b732 | 4155 | } |
6aa8b732 AK |
4156 | } |
4157 | ||
4a810181 JR |
4158 | static int svm_set_vm_cr(struct kvm_vcpu *vcpu, u64 data) |
4159 | { | |
4160 | struct vcpu_svm *svm = to_svm(vcpu); | |
4161 | int svm_dis, chg_mask; | |
4162 | ||
4163 | if (data & ~SVM_VM_CR_VALID_MASK) | |
4164 | return 1; | |
4165 | ||
4166 | chg_mask = SVM_VM_CR_VALID_MASK; | |
4167 | ||
4168 | if (svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK) | |
4169 | chg_mask &= ~(SVM_VM_CR_SVM_LOCK_MASK | SVM_VM_CR_SVM_DIS_MASK); | |
4170 | ||
4171 | svm->nested.vm_cr_msr &= ~chg_mask; | |
4172 | svm->nested.vm_cr_msr |= (data & chg_mask); | |
4173 | ||
4174 | svm_dis = svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK; | |
4175 | ||
4176 | /* check for svm_disable while efer.svme is set */ | |
4177 | if (svm_dis && (vcpu->arch.efer & EFER_SVME)) | |
4178 | return 1; | |
4179 | ||
4180 | return 0; | |
4181 | } | |
4182 | ||
8fe8ab46 | 4183 | static int svm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr) |
6aa8b732 | 4184 | { |
a2fa3e9f GH |
4185 | struct vcpu_svm *svm = to_svm(vcpu); |
4186 | ||
8fe8ab46 WA |
4187 | u32 ecx = msr->index; |
4188 | u64 data = msr->data; | |
6aa8b732 | 4189 | switch (ecx) { |
15038e14 PB |
4190 | case MSR_IA32_CR_PAT: |
4191 | if (!kvm_mtrr_valid(vcpu, MSR_IA32_CR_PAT, data)) | |
4192 | return 1; | |
4193 | vcpu->arch.pat = data; | |
4194 | svm->vmcb->save.g_pat = data; | |
4195 | mark_dirty(svm->vmcb, VMCB_NPT); | |
4196 | break; | |
f4e1b3c8 | 4197 | case MSR_IA32_TSC: |
8fe8ab46 | 4198 | kvm_write_tsc(vcpu, msr); |
6aa8b732 | 4199 | break; |
b2ac58f9 KA |
4200 | case MSR_IA32_SPEC_CTRL: |
4201 | if (!msr->host_initiated && | |
4202 | !guest_cpuid_has(vcpu, X86_FEATURE_IBRS)) | |
4203 | return 1; | |
4204 | ||
4205 | /* The STIBP bit doesn't fault even if it's not advertised */ | |
4206 | if (data & ~(SPEC_CTRL_IBRS | SPEC_CTRL_STIBP)) | |
4207 | return 1; | |
4208 | ||
4209 | svm->spec_ctrl = data; | |
4210 | ||
4211 | if (!data) | |
4212 | break; | |
4213 | ||
4214 | /* | |
4215 | * For non-nested: | |
4216 | * When it's written (to non-zero) for the first time, pass | |
4217 | * it through. | |
4218 | * | |
4219 | * For nested: | |
4220 | * The handling of the MSR bitmap for L2 guests is done in | |
4221 | * nested_svm_vmrun_msrpm. | |
4222 | * We update the L1 MSR bit as well since it will end up | |
4223 | * touching the MSR anyway now. | |
4224 | */ | |
4225 | set_msr_interception(svm->msrpm, MSR_IA32_SPEC_CTRL, 1, 1); | |
4226 | break; | |
15d45071 AR |
4227 | case MSR_IA32_PRED_CMD: |
4228 | if (!msr->host_initiated && | |
4229 | !guest_cpuid_has(vcpu, X86_FEATURE_IBPB)) | |
4230 | return 1; | |
4231 | ||
4232 | if (data & ~PRED_CMD_IBPB) | |
4233 | return 1; | |
4234 | ||
4235 | if (!data) | |
4236 | break; | |
4237 | ||
4238 | wrmsrl(MSR_IA32_PRED_CMD, PRED_CMD_IBPB); | |
4239 | if (is_guest_mode(vcpu)) | |
4240 | break; | |
4241 | set_msr_interception(svm->msrpm, MSR_IA32_PRED_CMD, 0, 1); | |
4242 | break; | |
8c06585d | 4243 | case MSR_STAR: |
a2fa3e9f | 4244 | svm->vmcb->save.star = data; |
6aa8b732 | 4245 | break; |
49b14f24 | 4246 | #ifdef CONFIG_X86_64 |
6aa8b732 | 4247 | case MSR_LSTAR: |
a2fa3e9f | 4248 | svm->vmcb->save.lstar = data; |
6aa8b732 AK |
4249 | break; |
4250 | case MSR_CSTAR: | |
a2fa3e9f | 4251 | svm->vmcb->save.cstar = data; |
6aa8b732 AK |
4252 | break; |
4253 | case MSR_KERNEL_GS_BASE: | |
a2fa3e9f | 4254 | svm->vmcb->save.kernel_gs_base = data; |
6aa8b732 AK |
4255 | break; |
4256 | case MSR_SYSCALL_MASK: | |
a2fa3e9f | 4257 | svm->vmcb->save.sfmask = data; |
6aa8b732 AK |
4258 | break; |
4259 | #endif | |
4260 | case MSR_IA32_SYSENTER_CS: | |
a2fa3e9f | 4261 | svm->vmcb->save.sysenter_cs = data; |
6aa8b732 AK |
4262 | break; |
4263 | case MSR_IA32_SYSENTER_EIP: | |
017cb99e | 4264 | svm->sysenter_eip = data; |
a2fa3e9f | 4265 | svm->vmcb->save.sysenter_eip = data; |
6aa8b732 AK |
4266 | break; |
4267 | case MSR_IA32_SYSENTER_ESP: | |
017cb99e | 4268 | svm->sysenter_esp = data; |
a2fa3e9f | 4269 | svm->vmcb->save.sysenter_esp = data; |
6aa8b732 | 4270 | break; |
46896c73 PB |
4271 | case MSR_TSC_AUX: |
4272 | if (!boot_cpu_has(X86_FEATURE_RDTSCP)) | |
4273 | return 1; | |
4274 | ||
4275 | /* | |
4276 | * This is rare, so we update the MSR here instead of using | |
4277 | * direct_access_msrs. Doing that would require a rdmsr in | |
4278 | * svm_vcpu_put. | |
4279 | */ | |
4280 | svm->tsc_aux = data; | |
4281 | wrmsrl(MSR_TSC_AUX, svm->tsc_aux); | |
4282 | break; | |
a2938c80 | 4283 | case MSR_IA32_DEBUGCTLMSR: |
2a6b20b8 | 4284 | if (!boot_cpu_has(X86_FEATURE_LBRV)) { |
a737f256 CD |
4285 | vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTL 0x%llx, nop\n", |
4286 | __func__, data); | |
24e09cbf JR |
4287 | break; |
4288 | } | |
4289 | if (data & DEBUGCTL_RESERVED_BITS) | |
4290 | return 1; | |
4291 | ||
4292 | svm->vmcb->save.dbgctl = data; | |
b53ba3f9 | 4293 | mark_dirty(svm->vmcb, VMCB_LBR); |
24e09cbf JR |
4294 | if (data & (1ULL<<0)) |
4295 | svm_enable_lbrv(svm); | |
4296 | else | |
4297 | svm_disable_lbrv(svm); | |
a2938c80 | 4298 | break; |
b286d5d8 | 4299 | case MSR_VM_HSAVE_PA: |
e6aa9abd | 4300 | svm->nested.hsave_msr = data; |
62b9abaa | 4301 | break; |
3c5d0a44 | 4302 | case MSR_VM_CR: |
4a810181 | 4303 | return svm_set_vm_cr(vcpu, data); |
3c5d0a44 | 4304 | case MSR_VM_IGNNE: |
a737f256 | 4305 | vcpu_unimpl(vcpu, "unimplemented wrmsr: 0x%x data 0x%llx\n", ecx, data); |
3c5d0a44 | 4306 | break; |
d1d93fa9 TL |
4307 | case MSR_F10H_DECFG: { |
4308 | struct kvm_msr_entry msr_entry; | |
4309 | ||
4310 | msr_entry.index = msr->index; | |
4311 | if (svm_get_msr_feature(&msr_entry)) | |
4312 | return 1; | |
4313 | ||
4314 | /* Check the supported bits */ | |
4315 | if (data & ~msr_entry.data) | |
4316 | return 1; | |
4317 | ||
4318 | /* Don't allow the guest to change a bit, #GP */ | |
4319 | if (!msr->host_initiated && (data ^ msr_entry.data)) | |
4320 | return 1; | |
4321 | ||
4322 | svm->msr_decfg = data; | |
4323 | break; | |
4324 | } | |
44a95dae SS |
4325 | case MSR_IA32_APICBASE: |
4326 | if (kvm_vcpu_apicv_active(vcpu)) | |
4327 | avic_update_vapic_bar(to_svm(vcpu), data); | |
4328 | /* Follow through */ | |
6aa8b732 | 4329 | default: |
8fe8ab46 | 4330 | return kvm_set_msr_common(vcpu, msr); |
6aa8b732 AK |
4331 | } |
4332 | return 0; | |
4333 | } | |
4334 | ||
851ba692 | 4335 | static int wrmsr_interception(struct vcpu_svm *svm) |
6aa8b732 | 4336 | { |
8fe8ab46 | 4337 | struct msr_data msr; |
668f198f DK |
4338 | u32 ecx = kvm_register_read(&svm->vcpu, VCPU_REGS_RCX); |
4339 | u64 data = kvm_read_edx_eax(&svm->vcpu); | |
af9ca2d7 | 4340 | |
8fe8ab46 WA |
4341 | msr.data = data; |
4342 | msr.index = ecx; | |
4343 | msr.host_initiated = false; | |
af9ca2d7 | 4344 | |
5fdbf976 | 4345 | svm->next_rip = kvm_rip_read(&svm->vcpu) + 2; |
854e8bb1 | 4346 | if (kvm_set_msr(&svm->vcpu, &msr)) { |
59200273 | 4347 | trace_kvm_msr_write_ex(ecx, data); |
c1a5d4f9 | 4348 | kvm_inject_gp(&svm->vcpu, 0); |
b742c1e6 | 4349 | return 1; |
59200273 AK |
4350 | } else { |
4351 | trace_kvm_msr_write(ecx, data); | |
b742c1e6 | 4352 | return kvm_skip_emulated_instruction(&svm->vcpu); |
59200273 | 4353 | } |
6aa8b732 AK |
4354 | } |
4355 | ||
851ba692 | 4356 | static int msr_interception(struct vcpu_svm *svm) |
6aa8b732 | 4357 | { |
e756fc62 | 4358 | if (svm->vmcb->control.exit_info_1) |
851ba692 | 4359 | return wrmsr_interception(svm); |
6aa8b732 | 4360 | else |
851ba692 | 4361 | return rdmsr_interception(svm); |
6aa8b732 AK |
4362 | } |
4363 | ||
851ba692 | 4364 | static int interrupt_window_interception(struct vcpu_svm *svm) |
c1150d8c | 4365 | { |
3842d135 | 4366 | kvm_make_request(KVM_REQ_EVENT, &svm->vcpu); |
f0b85051 | 4367 | svm_clear_vintr(svm); |
85f455f7 | 4368 | svm->vmcb->control.int_ctl &= ~V_IRQ_MASK; |
decdbf6a | 4369 | mark_dirty(svm->vmcb, VMCB_INTR); |
675acb75 | 4370 | ++svm->vcpu.stat.irq_window_exits; |
c1150d8c DL |
4371 | return 1; |
4372 | } | |
4373 | ||
565d0998 ML |
4374 | static int pause_interception(struct vcpu_svm *svm) |
4375 | { | |
de63ad4c LM |
4376 | struct kvm_vcpu *vcpu = &svm->vcpu; |
4377 | bool in_kernel = (svm_get_cpl(vcpu) == 0); | |
4378 | ||
8566ac8b BM |
4379 | if (pause_filter_thresh) |
4380 | grow_ple_window(vcpu); | |
4381 | ||
de63ad4c | 4382 | kvm_vcpu_on_spin(vcpu, in_kernel); |
565d0998 ML |
4383 | return 1; |
4384 | } | |
4385 | ||
87c00572 GS |
4386 | static int nop_interception(struct vcpu_svm *svm) |
4387 | { | |
b742c1e6 | 4388 | return kvm_skip_emulated_instruction(&(svm->vcpu)); |
87c00572 GS |
4389 | } |
4390 | ||
4391 | static int monitor_interception(struct vcpu_svm *svm) | |
4392 | { | |
4393 | printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n"); | |
4394 | return nop_interception(svm); | |
4395 | } | |
4396 | ||
4397 | static int mwait_interception(struct vcpu_svm *svm) | |
4398 | { | |
4399 | printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n"); | |
4400 | return nop_interception(svm); | |
4401 | } | |
4402 | ||
18f40c53 SS |
4403 | enum avic_ipi_failure_cause { |
4404 | AVIC_IPI_FAILURE_INVALID_INT_TYPE, | |
4405 | AVIC_IPI_FAILURE_TARGET_NOT_RUNNING, | |
4406 | AVIC_IPI_FAILURE_INVALID_TARGET, | |
4407 | AVIC_IPI_FAILURE_INVALID_BACKING_PAGE, | |
4408 | }; | |
4409 | ||
4410 | static int avic_incomplete_ipi_interception(struct vcpu_svm *svm) | |
4411 | { | |
4412 | u32 icrh = svm->vmcb->control.exit_info_1 >> 32; | |
4413 | u32 icrl = svm->vmcb->control.exit_info_1; | |
4414 | u32 id = svm->vmcb->control.exit_info_2 >> 32; | |
5446a979 | 4415 | u32 index = svm->vmcb->control.exit_info_2 & 0xFF; |
18f40c53 SS |
4416 | struct kvm_lapic *apic = svm->vcpu.arch.apic; |
4417 | ||
4418 | trace_kvm_avic_incomplete_ipi(svm->vcpu.vcpu_id, icrh, icrl, id, index); | |
4419 | ||
4420 | switch (id) { | |
4421 | case AVIC_IPI_FAILURE_INVALID_INT_TYPE: | |
4422 | /* | |
4423 | * AVIC hardware handles the generation of | |
4424 | * IPIs when the specified Message Type is Fixed | |
4425 | * (also known as fixed delivery mode) and | |
4426 | * the Trigger Mode is edge-triggered. The hardware | |
4427 | * also supports self and broadcast delivery modes | |
4428 | * specified via the Destination Shorthand(DSH) | |
4429 | * field of the ICRL. Logical and physical APIC ID | |
4430 | * formats are supported. All other IPI types cause | |
4431 | * a #VMEXIT, which needs to emulated. | |
4432 | */ | |
4433 | kvm_lapic_reg_write(apic, APIC_ICR2, icrh); | |
4434 | kvm_lapic_reg_write(apic, APIC_ICR, icrl); | |
4435 | break; | |
4436 | case AVIC_IPI_FAILURE_TARGET_NOT_RUNNING: { | |
4437 | int i; | |
4438 | struct kvm_vcpu *vcpu; | |
4439 | struct kvm *kvm = svm->vcpu.kvm; | |
4440 | struct kvm_lapic *apic = svm->vcpu.arch.apic; | |
4441 | ||
4442 | /* | |
4443 | * At this point, we expect that the AVIC HW has already | |
4444 | * set the appropriate IRR bits on the valid target | |
4445 | * vcpus. So, we just need to kick the appropriate vcpu. | |
4446 | */ | |
4447 | kvm_for_each_vcpu(i, vcpu, kvm) { | |
4448 | bool m = kvm_apic_match_dest(vcpu, apic, | |
4449 | icrl & KVM_APIC_SHORT_MASK, | |
4450 | GET_APIC_DEST_FIELD(icrh), | |
4451 | icrl & KVM_APIC_DEST_MASK); | |
4452 | ||
4453 | if (m && !avic_vcpu_is_running(vcpu)) | |
4454 | kvm_vcpu_wake_up(vcpu); | |
4455 | } | |
4456 | break; | |
4457 | } | |
4458 | case AVIC_IPI_FAILURE_INVALID_TARGET: | |
4459 | break; | |
4460 | case AVIC_IPI_FAILURE_INVALID_BACKING_PAGE: | |
4461 | WARN_ONCE(1, "Invalid backing page\n"); | |
4462 | break; | |
4463 | default: | |
4464 | pr_err("Unknown IPI interception\n"); | |
4465 | } | |
4466 | ||
4467 | return 1; | |
4468 | } | |
4469 | ||
4470 | static u32 *avic_get_logical_id_entry(struct kvm_vcpu *vcpu, u32 ldr, bool flat) | |
4471 | { | |
81811c16 | 4472 | struct kvm_svm *kvm_svm = to_kvm_svm(vcpu->kvm); |
18f40c53 SS |
4473 | int index; |
4474 | u32 *logical_apic_id_table; | |
4475 | int dlid = GET_APIC_LOGICAL_ID(ldr); | |
4476 | ||
4477 | if (!dlid) | |
4478 | return NULL; | |
4479 | ||
4480 | if (flat) { /* flat */ | |
4481 | index = ffs(dlid) - 1; | |
4482 | if (index > 7) | |
4483 | return NULL; | |
4484 | } else { /* cluster */ | |
4485 | int cluster = (dlid & 0xf0) >> 4; | |
4486 | int apic = ffs(dlid & 0x0f) - 1; | |
4487 | ||
4488 | if ((apic < 0) || (apic > 7) || | |
4489 | (cluster >= 0xf)) | |
4490 | return NULL; | |
4491 | index = (cluster << 2) + apic; | |
4492 | } | |
4493 | ||
81811c16 | 4494 | logical_apic_id_table = (u32 *) page_address(kvm_svm->avic_logical_id_table_page); |
18f40c53 SS |
4495 | |
4496 | return &logical_apic_id_table[index]; | |
4497 | } | |
4498 | ||
4499 | static int avic_ldr_write(struct kvm_vcpu *vcpu, u8 g_physical_id, u32 ldr, | |
4500 | bool valid) | |
4501 | { | |
4502 | bool flat; | |
4503 | u32 *entry, new_entry; | |
4504 | ||
4505 | flat = kvm_lapic_get_reg(vcpu->arch.apic, APIC_DFR) == APIC_DFR_FLAT; | |
4506 | entry = avic_get_logical_id_entry(vcpu, ldr, flat); | |
4507 | if (!entry) | |
4508 | return -EINVAL; | |
4509 | ||
4510 | new_entry = READ_ONCE(*entry); | |
4511 | new_entry &= ~AVIC_LOGICAL_ID_ENTRY_GUEST_PHYSICAL_ID_MASK; | |
4512 | new_entry |= (g_physical_id & AVIC_LOGICAL_ID_ENTRY_GUEST_PHYSICAL_ID_MASK); | |
4513 | if (valid) | |
4514 | new_entry |= AVIC_LOGICAL_ID_ENTRY_VALID_MASK; | |
4515 | else | |
4516 | new_entry &= ~AVIC_LOGICAL_ID_ENTRY_VALID_MASK; | |
4517 | WRITE_ONCE(*entry, new_entry); | |
4518 | ||
4519 | return 0; | |
4520 | } | |
4521 | ||
4522 | static int avic_handle_ldr_update(struct kvm_vcpu *vcpu) | |
4523 | { | |
4524 | int ret; | |
4525 | struct vcpu_svm *svm = to_svm(vcpu); | |
4526 | u32 ldr = kvm_lapic_get_reg(vcpu->arch.apic, APIC_LDR); | |
4527 | ||
4528 | if (!ldr) | |
4529 | return 1; | |
4530 | ||
4531 | ret = avic_ldr_write(vcpu, vcpu->vcpu_id, ldr, true); | |
4532 | if (ret && svm->ldr_reg) { | |
4533 | avic_ldr_write(vcpu, 0, svm->ldr_reg, false); | |
4534 | svm->ldr_reg = 0; | |
4535 | } else { | |
4536 | svm->ldr_reg = ldr; | |
4537 | } | |
4538 | return ret; | |
4539 | } | |
4540 | ||
4541 | static int avic_handle_apic_id_update(struct kvm_vcpu *vcpu) | |
4542 | { | |
4543 | u64 *old, *new; | |
4544 | struct vcpu_svm *svm = to_svm(vcpu); | |
4545 | u32 apic_id_reg = kvm_lapic_get_reg(vcpu->arch.apic, APIC_ID); | |
4546 | u32 id = (apic_id_reg >> 24) & 0xff; | |
4547 | ||
4548 | if (vcpu->vcpu_id == id) | |
4549 | return 0; | |
4550 | ||
4551 | old = avic_get_physical_id_entry(vcpu, vcpu->vcpu_id); | |
4552 | new = avic_get_physical_id_entry(vcpu, id); | |
4553 | if (!new || !old) | |
4554 | return 1; | |
4555 | ||
4556 | /* We need to move physical_id_entry to new offset */ | |
4557 | *new = *old; | |
4558 | *old = 0ULL; | |
4559 | to_svm(vcpu)->avic_physical_id_cache = new; | |
4560 | ||
4561 | /* | |
4562 | * Also update the guest physical APIC ID in the logical | |
4563 | * APIC ID table entry if already setup the LDR. | |
4564 | */ | |
4565 | if (svm->ldr_reg) | |
4566 | avic_handle_ldr_update(vcpu); | |
4567 | ||
4568 | return 0; | |
4569 | } | |
4570 | ||
4571 | static int avic_handle_dfr_update(struct kvm_vcpu *vcpu) | |
4572 | { | |
4573 | struct vcpu_svm *svm = to_svm(vcpu); | |
81811c16 | 4574 | struct kvm_svm *kvm_svm = to_kvm_svm(vcpu->kvm); |
18f40c53 SS |
4575 | u32 dfr = kvm_lapic_get_reg(vcpu->arch.apic, APIC_DFR); |
4576 | u32 mod = (dfr >> 28) & 0xf; | |
4577 | ||
4578 | /* | |
4579 | * We assume that all local APICs are using the same type. | |
4580 | * If this changes, we need to flush the AVIC logical | |
4581 | * APID id table. | |
4582 | */ | |
81811c16 | 4583 | if (kvm_svm->ldr_mode == mod) |
18f40c53 SS |
4584 | return 0; |
4585 | ||
81811c16 SC |
4586 | clear_page(page_address(kvm_svm->avic_logical_id_table_page)); |
4587 | kvm_svm->ldr_mode = mod; | |
18f40c53 SS |
4588 | |
4589 | if (svm->ldr_reg) | |
4590 | avic_handle_ldr_update(vcpu); | |
4591 | return 0; | |
4592 | } | |
4593 | ||
4594 | static int avic_unaccel_trap_write(struct vcpu_svm *svm) | |
4595 | { | |
4596 | struct kvm_lapic *apic = svm->vcpu.arch.apic; | |
4597 | u32 offset = svm->vmcb->control.exit_info_1 & | |
4598 | AVIC_UNACCEL_ACCESS_OFFSET_MASK; | |
4599 | ||
4600 | switch (offset) { | |
4601 | case APIC_ID: | |
4602 | if (avic_handle_apic_id_update(&svm->vcpu)) | |
4603 | return 0; | |
4604 | break; | |
4605 | case APIC_LDR: | |
4606 | if (avic_handle_ldr_update(&svm->vcpu)) | |
4607 | return 0; | |
4608 | break; | |
4609 | case APIC_DFR: | |
4610 | avic_handle_dfr_update(&svm->vcpu); | |
4611 | break; | |
4612 | default: | |
4613 | break; | |
4614 | } | |
4615 | ||
4616 | kvm_lapic_reg_write(apic, offset, kvm_lapic_get_reg(apic, offset)); | |
4617 | ||
4618 | return 1; | |
4619 | } | |
4620 | ||
4621 | static bool is_avic_unaccelerated_access_trap(u32 offset) | |
4622 | { | |
4623 | bool ret = false; | |
4624 | ||
4625 | switch (offset) { | |
4626 | case APIC_ID: | |
4627 | case APIC_EOI: | |
4628 | case APIC_RRR: | |
4629 | case APIC_LDR: | |
4630 | case APIC_DFR: | |
4631 | case APIC_SPIV: | |
4632 | case APIC_ESR: | |
4633 | case APIC_ICR: | |
4634 | case APIC_LVTT: | |
4635 | case APIC_LVTTHMR: | |
4636 | case APIC_LVTPC: | |
4637 | case APIC_LVT0: | |
4638 | case APIC_LVT1: | |
4639 | case APIC_LVTERR: | |
4640 | case APIC_TMICT: | |
4641 | case APIC_TDCR: | |
4642 | ret = true; | |
4643 | break; | |
4644 | default: | |
4645 | break; | |
4646 | } | |
4647 | return ret; | |
4648 | } | |
4649 | ||
4650 | static int avic_unaccelerated_access_interception(struct vcpu_svm *svm) | |
4651 | { | |
4652 | int ret = 0; | |
4653 | u32 offset = svm->vmcb->control.exit_info_1 & | |
4654 | AVIC_UNACCEL_ACCESS_OFFSET_MASK; | |
4655 | u32 vector = svm->vmcb->control.exit_info_2 & | |
4656 | AVIC_UNACCEL_ACCESS_VECTOR_MASK; | |
4657 | bool write = (svm->vmcb->control.exit_info_1 >> 32) & | |
4658 | AVIC_UNACCEL_ACCESS_WRITE_MASK; | |
4659 | bool trap = is_avic_unaccelerated_access_trap(offset); | |
4660 | ||
4661 | trace_kvm_avic_unaccelerated_access(svm->vcpu.vcpu_id, offset, | |
4662 | trap, write, vector); | |
4663 | if (trap) { | |
4664 | /* Handling Trap */ | |
4665 | WARN_ONCE(!write, "svm: Handling trap read.\n"); | |
4666 | ret = avic_unaccel_trap_write(svm); | |
4667 | } else { | |
4668 | /* Handling Fault */ | |
4669 | ret = (emulate_instruction(&svm->vcpu, 0) == EMULATE_DONE); | |
4670 | } | |
4671 | ||
4672 | return ret; | |
4673 | } | |
4674 | ||
09941fbb | 4675 | static int (*const svm_exit_handlers[])(struct vcpu_svm *svm) = { |
7ff76d58 AP |
4676 | [SVM_EXIT_READ_CR0] = cr_interception, |
4677 | [SVM_EXIT_READ_CR3] = cr_interception, | |
4678 | [SVM_EXIT_READ_CR4] = cr_interception, | |
4679 | [SVM_EXIT_READ_CR8] = cr_interception, | |
5e57518d | 4680 | [SVM_EXIT_CR0_SEL_WRITE] = cr_interception, |
628afd2a | 4681 | [SVM_EXIT_WRITE_CR0] = cr_interception, |
7ff76d58 AP |
4682 | [SVM_EXIT_WRITE_CR3] = cr_interception, |
4683 | [SVM_EXIT_WRITE_CR4] = cr_interception, | |
e0231715 | 4684 | [SVM_EXIT_WRITE_CR8] = cr8_write_interception, |
cae3797a AP |
4685 | [SVM_EXIT_READ_DR0] = dr_interception, |
4686 | [SVM_EXIT_READ_DR1] = dr_interception, | |
4687 | [SVM_EXIT_READ_DR2] = dr_interception, | |
4688 | [SVM_EXIT_READ_DR3] = dr_interception, | |
4689 | [SVM_EXIT_READ_DR4] = dr_interception, | |
4690 | [SVM_EXIT_READ_DR5] = dr_interception, | |
4691 | [SVM_EXIT_READ_DR6] = dr_interception, | |
4692 | [SVM_EXIT_READ_DR7] = dr_interception, | |
4693 | [SVM_EXIT_WRITE_DR0] = dr_interception, | |
4694 | [SVM_EXIT_WRITE_DR1] = dr_interception, | |
4695 | [SVM_EXIT_WRITE_DR2] = dr_interception, | |
4696 | [SVM_EXIT_WRITE_DR3] = dr_interception, | |
4697 | [SVM_EXIT_WRITE_DR4] = dr_interception, | |
4698 | [SVM_EXIT_WRITE_DR5] = dr_interception, | |
4699 | [SVM_EXIT_WRITE_DR6] = dr_interception, | |
4700 | [SVM_EXIT_WRITE_DR7] = dr_interception, | |
d0bfb940 JK |
4701 | [SVM_EXIT_EXCP_BASE + DB_VECTOR] = db_interception, |
4702 | [SVM_EXIT_EXCP_BASE + BP_VECTOR] = bp_interception, | |
7aa81cc0 | 4703 | [SVM_EXIT_EXCP_BASE + UD_VECTOR] = ud_interception, |
e0231715 | 4704 | [SVM_EXIT_EXCP_BASE + PF_VECTOR] = pf_interception, |
e0231715 | 4705 | [SVM_EXIT_EXCP_BASE + MC_VECTOR] = mc_interception, |
54a20552 | 4706 | [SVM_EXIT_EXCP_BASE + AC_VECTOR] = ac_interception, |
9718420e | 4707 | [SVM_EXIT_EXCP_BASE + GP_VECTOR] = gp_interception, |
e0231715 | 4708 | [SVM_EXIT_INTR] = intr_interception, |
c47f098d | 4709 | [SVM_EXIT_NMI] = nmi_interception, |
6aa8b732 AK |
4710 | [SVM_EXIT_SMI] = nop_on_interception, |
4711 | [SVM_EXIT_INIT] = nop_on_interception, | |
c1150d8c | 4712 | [SVM_EXIT_VINTR] = interrupt_window_interception, |
332b56e4 | 4713 | [SVM_EXIT_RDPMC] = rdpmc_interception, |
6aa8b732 | 4714 | [SVM_EXIT_CPUID] = cpuid_interception, |
95ba8273 | 4715 | [SVM_EXIT_IRET] = iret_interception, |
cf5a94d1 | 4716 | [SVM_EXIT_INVD] = emulate_on_interception, |
565d0998 | 4717 | [SVM_EXIT_PAUSE] = pause_interception, |
6aa8b732 | 4718 | [SVM_EXIT_HLT] = halt_interception, |
a7052897 | 4719 | [SVM_EXIT_INVLPG] = invlpg_interception, |
ff092385 | 4720 | [SVM_EXIT_INVLPGA] = invlpga_interception, |
e0231715 | 4721 | [SVM_EXIT_IOIO] = io_interception, |
6aa8b732 AK |
4722 | [SVM_EXIT_MSR] = msr_interception, |
4723 | [SVM_EXIT_TASK_SWITCH] = task_switch_interception, | |
46fe4ddd | 4724 | [SVM_EXIT_SHUTDOWN] = shutdown_interception, |
3d6368ef | 4725 | [SVM_EXIT_VMRUN] = vmrun_interception, |
02e235bc | 4726 | [SVM_EXIT_VMMCALL] = vmmcall_interception, |
5542675b AG |
4727 | [SVM_EXIT_VMLOAD] = vmload_interception, |
4728 | [SVM_EXIT_VMSAVE] = vmsave_interception, | |
1371d904 AG |
4729 | [SVM_EXIT_STGI] = stgi_interception, |
4730 | [SVM_EXIT_CLGI] = clgi_interception, | |
532a46b9 | 4731 | [SVM_EXIT_SKINIT] = skinit_interception, |
dab429a7 | 4732 | [SVM_EXIT_WBINVD] = wbinvd_interception, |
87c00572 GS |
4733 | [SVM_EXIT_MONITOR] = monitor_interception, |
4734 | [SVM_EXIT_MWAIT] = mwait_interception, | |
81dd35d4 | 4735 | [SVM_EXIT_XSETBV] = xsetbv_interception, |
d0006530 | 4736 | [SVM_EXIT_NPF] = npf_interception, |
7607b717 | 4737 | [SVM_EXIT_RSM] = rsm_interception, |
18f40c53 SS |
4738 | [SVM_EXIT_AVIC_INCOMPLETE_IPI] = avic_incomplete_ipi_interception, |
4739 | [SVM_EXIT_AVIC_UNACCELERATED_ACCESS] = avic_unaccelerated_access_interception, | |
6aa8b732 AK |
4740 | }; |
4741 | ||
ae8cc059 | 4742 | static void dump_vmcb(struct kvm_vcpu *vcpu) |
3f10c846 JR |
4743 | { |
4744 | struct vcpu_svm *svm = to_svm(vcpu); | |
4745 | struct vmcb_control_area *control = &svm->vmcb->control; | |
4746 | struct vmcb_save_area *save = &svm->vmcb->save; | |
4747 | ||
4748 | pr_err("VMCB Control Area:\n"); | |
ae8cc059 JP |
4749 | pr_err("%-20s%04x\n", "cr_read:", control->intercept_cr & 0xffff); |
4750 | pr_err("%-20s%04x\n", "cr_write:", control->intercept_cr >> 16); | |
4751 | pr_err("%-20s%04x\n", "dr_read:", control->intercept_dr & 0xffff); | |
4752 | pr_err("%-20s%04x\n", "dr_write:", control->intercept_dr >> 16); | |
4753 | pr_err("%-20s%08x\n", "exceptions:", control->intercept_exceptions); | |
4754 | pr_err("%-20s%016llx\n", "intercepts:", control->intercept); | |
4755 | pr_err("%-20s%d\n", "pause filter count:", control->pause_filter_count); | |
1d8fb44a BM |
4756 | pr_err("%-20s%d\n", "pause filter threshold:", |
4757 | control->pause_filter_thresh); | |
ae8cc059 JP |
4758 | pr_err("%-20s%016llx\n", "iopm_base_pa:", control->iopm_base_pa); |
4759 | pr_err("%-20s%016llx\n", "msrpm_base_pa:", control->msrpm_base_pa); | |
4760 | pr_err("%-20s%016llx\n", "tsc_offset:", control->tsc_offset); | |
4761 | pr_err("%-20s%d\n", "asid:", control->asid); | |
4762 | pr_err("%-20s%d\n", "tlb_ctl:", control->tlb_ctl); | |
4763 | pr_err("%-20s%08x\n", "int_ctl:", control->int_ctl); | |
4764 | pr_err("%-20s%08x\n", "int_vector:", control->int_vector); | |
4765 | pr_err("%-20s%08x\n", "int_state:", control->int_state); | |
4766 | pr_err("%-20s%08x\n", "exit_code:", control->exit_code); | |
4767 | pr_err("%-20s%016llx\n", "exit_info1:", control->exit_info_1); | |
4768 | pr_err("%-20s%016llx\n", "exit_info2:", control->exit_info_2); | |
4769 | pr_err("%-20s%08x\n", "exit_int_info:", control->exit_int_info); | |
4770 | pr_err("%-20s%08x\n", "exit_int_info_err:", control->exit_int_info_err); | |
4771 | pr_err("%-20s%lld\n", "nested_ctl:", control->nested_ctl); | |
4772 | pr_err("%-20s%016llx\n", "nested_cr3:", control->nested_cr3); | |
44a95dae | 4773 | pr_err("%-20s%016llx\n", "avic_vapic_bar:", control->avic_vapic_bar); |
ae8cc059 JP |
4774 | pr_err("%-20s%08x\n", "event_inj:", control->event_inj); |
4775 | pr_err("%-20s%08x\n", "event_inj_err:", control->event_inj_err); | |
0dc92119 | 4776 | pr_err("%-20s%lld\n", "virt_ext:", control->virt_ext); |
ae8cc059 | 4777 | pr_err("%-20s%016llx\n", "next_rip:", control->next_rip); |
44a95dae SS |
4778 | pr_err("%-20s%016llx\n", "avic_backing_page:", control->avic_backing_page); |
4779 | pr_err("%-20s%016llx\n", "avic_logical_id:", control->avic_logical_id); | |
4780 | pr_err("%-20s%016llx\n", "avic_physical_id:", control->avic_physical_id); | |
3f10c846 | 4781 | pr_err("VMCB State Save Area:\n"); |
ae8cc059 JP |
4782 | pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n", |
4783 | "es:", | |
4784 | save->es.selector, save->es.attrib, | |
4785 | save->es.limit, save->es.base); | |
4786 | pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n", | |
4787 | "cs:", | |
4788 | save->cs.selector, save->cs.attrib, | |
4789 | save->cs.limit, save->cs.base); | |
4790 | pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n", | |
4791 | "ss:", | |
4792 | save->ss.selector, save->ss.attrib, | |
4793 | save->ss.limit, save->ss.base); | |
4794 | pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n", | |
4795 | "ds:", | |
4796 | save->ds.selector, save->ds.attrib, | |
4797 | save->ds.limit, save->ds.base); | |
4798 | pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n", | |
4799 | "fs:", | |
4800 | save->fs.selector, save->fs.attrib, | |
4801 | save->fs.limit, save->fs.base); | |
4802 | pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n", | |
4803 | "gs:", | |
4804 | save->gs.selector, save->gs.attrib, | |
4805 | save->gs.limit, save->gs.base); | |
4806 | pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n", | |
4807 | "gdtr:", | |
4808 | save->gdtr.selector, save->gdtr.attrib, | |
4809 | save->gdtr.limit, save->gdtr.base); | |
4810 | pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n", | |
4811 | "ldtr:", | |
4812 | save->ldtr.selector, save->ldtr.attrib, | |
4813 | save->ldtr.limit, save->ldtr.base); | |
4814 | pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n", | |
4815 | "idtr:", | |
4816 | save->idtr.selector, save->idtr.attrib, | |
4817 | save->idtr.limit, save->idtr.base); | |
4818 | pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n", | |
4819 | "tr:", | |
4820 | save->tr.selector, save->tr.attrib, | |
4821 | save->tr.limit, save->tr.base); | |
3f10c846 JR |
4822 | pr_err("cpl: %d efer: %016llx\n", |
4823 | save->cpl, save->efer); | |
ae8cc059 JP |
4824 | pr_err("%-15s %016llx %-13s %016llx\n", |
4825 | "cr0:", save->cr0, "cr2:", save->cr2); | |
4826 | pr_err("%-15s %016llx %-13s %016llx\n", | |
4827 | "cr3:", save->cr3, "cr4:", save->cr4); | |
4828 | pr_err("%-15s %016llx %-13s %016llx\n", | |
4829 | "dr6:", save->dr6, "dr7:", save->dr7); | |
4830 | pr_err("%-15s %016llx %-13s %016llx\n", | |
4831 | "rip:", save->rip, "rflags:", save->rflags); | |
4832 | pr_err("%-15s %016llx %-13s %016llx\n", | |
4833 | "rsp:", save->rsp, "rax:", save->rax); | |
4834 | pr_err("%-15s %016llx %-13s %016llx\n", | |
4835 | "star:", save->star, "lstar:", save->lstar); | |
4836 | pr_err("%-15s %016llx %-13s %016llx\n", | |
4837 | "cstar:", save->cstar, "sfmask:", save->sfmask); | |
4838 | pr_err("%-15s %016llx %-13s %016llx\n", | |
4839 | "kernel_gs_base:", save->kernel_gs_base, | |
4840 | "sysenter_cs:", save->sysenter_cs); | |
4841 | pr_err("%-15s %016llx %-13s %016llx\n", | |
4842 | "sysenter_esp:", save->sysenter_esp, | |
4843 | "sysenter_eip:", save->sysenter_eip); | |
4844 | pr_err("%-15s %016llx %-13s %016llx\n", | |
4845 | "gpat:", save->g_pat, "dbgctl:", save->dbgctl); | |
4846 | pr_err("%-15s %016llx %-13s %016llx\n", | |
4847 | "br_from:", save->br_from, "br_to:", save->br_to); | |
4848 | pr_err("%-15s %016llx %-13s %016llx\n", | |
4849 | "excp_from:", save->last_excp_from, | |
4850 | "excp_to:", save->last_excp_to); | |
3f10c846 JR |
4851 | } |
4852 | ||
586f9607 AK |
4853 | static void svm_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2) |
4854 | { | |
4855 | struct vmcb_control_area *control = &to_svm(vcpu)->vmcb->control; | |
4856 | ||
4857 | *info1 = control->exit_info_1; | |
4858 | *info2 = control->exit_info_2; | |
4859 | } | |
4860 | ||
851ba692 | 4861 | static int handle_exit(struct kvm_vcpu *vcpu) |
6aa8b732 | 4862 | { |
04d2cc77 | 4863 | struct vcpu_svm *svm = to_svm(vcpu); |
851ba692 | 4864 | struct kvm_run *kvm_run = vcpu->run; |
a2fa3e9f | 4865 | u32 exit_code = svm->vmcb->control.exit_code; |
6aa8b732 | 4866 | |
8b89fe1f PB |
4867 | trace_kvm_exit(exit_code, vcpu, KVM_ISA_SVM); |
4868 | ||
4ee546b4 | 4869 | if (!is_cr_intercept(svm, INTERCEPT_CR0_WRITE)) |
2be4fc7a JR |
4870 | vcpu->arch.cr0 = svm->vmcb->save.cr0; |
4871 | if (npt_enabled) | |
4872 | vcpu->arch.cr3 = svm->vmcb->save.cr3; | |
af9ca2d7 | 4873 | |
cd3ff653 JR |
4874 | if (unlikely(svm->nested.exit_required)) { |
4875 | nested_svm_vmexit(svm); | |
4876 | svm->nested.exit_required = false; | |
4877 | ||
4878 | return 1; | |
4879 | } | |
4880 | ||
2030753d | 4881 | if (is_guest_mode(vcpu)) { |
410e4d57 JR |
4882 | int vmexit; |
4883 | ||
d8cabddf JR |
4884 | trace_kvm_nested_vmexit(svm->vmcb->save.rip, exit_code, |
4885 | svm->vmcb->control.exit_info_1, | |
4886 | svm->vmcb->control.exit_info_2, | |
4887 | svm->vmcb->control.exit_int_info, | |
e097e5ff SH |
4888 | svm->vmcb->control.exit_int_info_err, |
4889 | KVM_ISA_SVM); | |
d8cabddf | 4890 | |
410e4d57 JR |
4891 | vmexit = nested_svm_exit_special(svm); |
4892 | ||
4893 | if (vmexit == NESTED_EXIT_CONTINUE) | |
4894 | vmexit = nested_svm_exit_handled(svm); | |
4895 | ||
4896 | if (vmexit == NESTED_EXIT_DONE) | |
cf74a78b | 4897 | return 1; |
cf74a78b AG |
4898 | } |
4899 | ||
a5c3832d JR |
4900 | svm_complete_interrupts(svm); |
4901 | ||
04d2cc77 AK |
4902 | if (svm->vmcb->control.exit_code == SVM_EXIT_ERR) { |
4903 | kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY; | |
4904 | kvm_run->fail_entry.hardware_entry_failure_reason | |
4905 | = svm->vmcb->control.exit_code; | |
3f10c846 JR |
4906 | pr_err("KVM: FAILED VMRUN WITH VMCB:\n"); |
4907 | dump_vmcb(vcpu); | |
04d2cc77 AK |
4908 | return 0; |
4909 | } | |
4910 | ||
a2fa3e9f | 4911 | if (is_external_interrupt(svm->vmcb->control.exit_int_info) && |
709ddebf | 4912 | exit_code != SVM_EXIT_EXCP_BASE + PF_VECTOR && |
55c5e464 JR |
4913 | exit_code != SVM_EXIT_NPF && exit_code != SVM_EXIT_TASK_SWITCH && |
4914 | exit_code != SVM_EXIT_INTR && exit_code != SVM_EXIT_NMI) | |
6614c7d0 | 4915 | printk(KERN_ERR "%s: unexpected exit_int_info 0x%x " |
6aa8b732 | 4916 | "exit_code 0x%x\n", |
b8688d51 | 4917 | __func__, svm->vmcb->control.exit_int_info, |
6aa8b732 AK |
4918 | exit_code); |
4919 | ||
9d8f549d | 4920 | if (exit_code >= ARRAY_SIZE(svm_exit_handlers) |
56919c5c | 4921 | || !svm_exit_handlers[exit_code]) { |
faac2458 | 4922 | WARN_ONCE(1, "svm: unexpected exit reason 0x%x\n", exit_code); |
2bc19dc3 MT |
4923 | kvm_queue_exception(vcpu, UD_VECTOR); |
4924 | return 1; | |
6aa8b732 AK |
4925 | } |
4926 | ||
851ba692 | 4927 | return svm_exit_handlers[exit_code](svm); |
6aa8b732 AK |
4928 | } |
4929 | ||
4930 | static void reload_tss(struct kvm_vcpu *vcpu) | |
4931 | { | |
4932 | int cpu = raw_smp_processor_id(); | |
4933 | ||
0fe1e009 TH |
4934 | struct svm_cpu_data *sd = per_cpu(svm_data, cpu); |
4935 | sd->tss_desc->type = 9; /* available 32/64-bit TSS */ | |
6aa8b732 AK |
4936 | load_TR_desc(); |
4937 | } | |
4938 | ||
70cd94e6 BS |
4939 | static void pre_sev_run(struct vcpu_svm *svm, int cpu) |
4940 | { | |
4941 | struct svm_cpu_data *sd = per_cpu(svm_data, cpu); | |
4942 | int asid = sev_get_asid(svm->vcpu.kvm); | |
4943 | ||
4944 | /* Assign the asid allocated with this SEV guest */ | |
4945 | svm->vmcb->control.asid = asid; | |
4946 | ||
4947 | /* | |
4948 | * Flush guest TLB: | |
4949 | * | |
4950 | * 1) when different VMCB for the same ASID is to be run on the same host CPU. | |
4951 | * 2) or this VMCB was executed on different host CPU in previous VMRUNs. | |
4952 | */ | |
4953 | if (sd->sev_vmcbs[asid] == svm->vmcb && | |
4954 | svm->last_cpu == cpu) | |
4955 | return; | |
4956 | ||
4957 | svm->last_cpu = cpu; | |
4958 | sd->sev_vmcbs[asid] = svm->vmcb; | |
4959 | svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ASID; | |
4960 | mark_dirty(svm->vmcb, VMCB_ASID); | |
4961 | } | |
4962 | ||
e756fc62 | 4963 | static void pre_svm_run(struct vcpu_svm *svm) |
6aa8b732 AK |
4964 | { |
4965 | int cpu = raw_smp_processor_id(); | |
4966 | ||
0fe1e009 | 4967 | struct svm_cpu_data *sd = per_cpu(svm_data, cpu); |
6aa8b732 | 4968 | |
70cd94e6 BS |
4969 | if (sev_guest(svm->vcpu.kvm)) |
4970 | return pre_sev_run(svm, cpu); | |
4971 | ||
4b656b12 | 4972 | /* FIXME: handle wraparound of asid_generation */ |
0fe1e009 TH |
4973 | if (svm->asid_generation != sd->asid_generation) |
4974 | new_asid(svm, sd); | |
6aa8b732 AK |
4975 | } |
4976 | ||
95ba8273 GN |
4977 | static void svm_inject_nmi(struct kvm_vcpu *vcpu) |
4978 | { | |
4979 | struct vcpu_svm *svm = to_svm(vcpu); | |
4980 | ||
4981 | svm->vmcb->control.event_inj = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI; | |
4982 | vcpu->arch.hflags |= HF_NMI_MASK; | |
8a05a1b8 | 4983 | set_intercept(svm, INTERCEPT_IRET); |
95ba8273 GN |
4984 | ++vcpu->stat.nmi_injections; |
4985 | } | |
6aa8b732 | 4986 | |
85f455f7 | 4987 | static inline void svm_inject_irq(struct vcpu_svm *svm, int irq) |
6aa8b732 AK |
4988 | { |
4989 | struct vmcb_control_area *control; | |
4990 | ||
340d3bc3 | 4991 | /* The following fields are ignored when AVIC is enabled */ |
e756fc62 | 4992 | control = &svm->vmcb->control; |
85f455f7 | 4993 | control->int_vector = irq; |
6aa8b732 AK |
4994 | control->int_ctl &= ~V_INTR_PRIO_MASK; |
4995 | control->int_ctl |= V_IRQ_MASK | | |
4996 | ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT); | |
decdbf6a | 4997 | mark_dirty(svm->vmcb, VMCB_INTR); |
6aa8b732 AK |
4998 | } |
4999 | ||
66fd3f7f | 5000 | static void svm_set_irq(struct kvm_vcpu *vcpu) |
2a8067f1 ED |
5001 | { |
5002 | struct vcpu_svm *svm = to_svm(vcpu); | |
5003 | ||
2af9194d | 5004 | BUG_ON(!(gif_set(svm))); |
cf74a78b | 5005 | |
9fb2d2b4 GN |
5006 | trace_kvm_inj_virq(vcpu->arch.interrupt.nr); |
5007 | ++vcpu->stat.irq_injections; | |
5008 | ||
219b65dc AG |
5009 | svm->vmcb->control.event_inj = vcpu->arch.interrupt.nr | |
5010 | SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR; | |
2a8067f1 ED |
5011 | } |
5012 | ||
3bbf3565 SS |
5013 | static inline bool svm_nested_virtualize_tpr(struct kvm_vcpu *vcpu) |
5014 | { | |
5015 | return is_guest_mode(vcpu) && (vcpu->arch.hflags & HF_VINTR_MASK); | |
5016 | } | |
5017 | ||
95ba8273 | 5018 | static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr) |
aaacfc9a JR |
5019 | { |
5020 | struct vcpu_svm *svm = to_svm(vcpu); | |
aaacfc9a | 5021 | |
3bbf3565 SS |
5022 | if (svm_nested_virtualize_tpr(vcpu) || |
5023 | kvm_vcpu_apicv_active(vcpu)) | |
88ab24ad JR |
5024 | return; |
5025 | ||
596f3142 RK |
5026 | clr_cr_intercept(svm, INTERCEPT_CR8_WRITE); |
5027 | ||
95ba8273 | 5028 | if (irr == -1) |
aaacfc9a JR |
5029 | return; |
5030 | ||
95ba8273 | 5031 | if (tpr >= irr) |
4ee546b4 | 5032 | set_cr_intercept(svm, INTERCEPT_CR8_WRITE); |
95ba8273 | 5033 | } |
aaacfc9a | 5034 | |
8d14695f YZ |
5035 | static void svm_set_virtual_x2apic_mode(struct kvm_vcpu *vcpu, bool set) |
5036 | { | |
5037 | return; | |
5038 | } | |
5039 | ||
b2a05fef | 5040 | static bool svm_get_enable_apicv(struct kvm_vcpu *vcpu) |
d62caabb | 5041 | { |
67034bb9 | 5042 | return avic && irqchip_split(vcpu->kvm); |
44a95dae SS |
5043 | } |
5044 | ||
5045 | static void svm_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr) | |
5046 | { | |
d62caabb AS |
5047 | } |
5048 | ||
67c9dddc | 5049 | static void svm_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr) |
44a95dae | 5050 | { |
d62caabb AS |
5051 | } |
5052 | ||
44a95dae | 5053 | /* Note: Currently only used by Hyper-V. */ |
d62caabb | 5054 | static void svm_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu) |
c7c9c56c | 5055 | { |
44a95dae SS |
5056 | struct vcpu_svm *svm = to_svm(vcpu); |
5057 | struct vmcb *vmcb = svm->vmcb; | |
5058 | ||
67034bb9 | 5059 | if (!kvm_vcpu_apicv_active(&svm->vcpu)) |
44a95dae SS |
5060 | return; |
5061 | ||
5062 | vmcb->control.int_ctl &= ~AVIC_ENABLE_MASK; | |
5063 | mark_dirty(vmcb, VMCB_INTR); | |
c7c9c56c YZ |
5064 | } |
5065 | ||
6308630b | 5066 | static void svm_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap) |
c7c9c56c YZ |
5067 | { |
5068 | return; | |
5069 | } | |
5070 | ||
340d3bc3 SS |
5071 | static void svm_deliver_avic_intr(struct kvm_vcpu *vcpu, int vec) |
5072 | { | |
5073 | kvm_lapic_set_irr(vec, vcpu->arch.apic); | |
5074 | smp_mb__after_atomic(); | |
5075 | ||
5076 | if (avic_vcpu_is_running(vcpu)) | |
5077 | wrmsrl(SVM_AVIC_DOORBELL, | |
7d669f50 | 5078 | kvm_cpu_get_apicid(vcpu->cpu)); |
340d3bc3 SS |
5079 | else |
5080 | kvm_vcpu_wake_up(vcpu); | |
5081 | } | |
5082 | ||
411b44ba SS |
5083 | static void svm_ir_list_del(struct vcpu_svm *svm, struct amd_iommu_pi_data *pi) |
5084 | { | |
5085 | unsigned long flags; | |
5086 | struct amd_svm_iommu_ir *cur; | |
5087 | ||
5088 | spin_lock_irqsave(&svm->ir_list_lock, flags); | |
5089 | list_for_each_entry(cur, &svm->ir_list, node) { | |
5090 | if (cur->data != pi->ir_data) | |
5091 | continue; | |
5092 | list_del(&cur->node); | |
5093 | kfree(cur); | |
5094 | break; | |
5095 | } | |
5096 | spin_unlock_irqrestore(&svm->ir_list_lock, flags); | |
5097 | } | |
5098 | ||
5099 | static int svm_ir_list_add(struct vcpu_svm *svm, struct amd_iommu_pi_data *pi) | |
5100 | { | |
5101 | int ret = 0; | |
5102 | unsigned long flags; | |
5103 | struct amd_svm_iommu_ir *ir; | |
5104 | ||
5105 | /** | |
5106 | * In some cases, the existing irte is updaed and re-set, | |
5107 | * so we need to check here if it's already been * added | |
5108 | * to the ir_list. | |
5109 | */ | |
5110 | if (pi->ir_data && (pi->prev_ga_tag != 0)) { | |
5111 | struct kvm *kvm = svm->vcpu.kvm; | |
5112 | u32 vcpu_id = AVIC_GATAG_TO_VCPUID(pi->prev_ga_tag); | |
5113 | struct kvm_vcpu *prev_vcpu = kvm_get_vcpu_by_id(kvm, vcpu_id); | |
5114 | struct vcpu_svm *prev_svm; | |
5115 | ||
5116 | if (!prev_vcpu) { | |
5117 | ret = -EINVAL; | |
5118 | goto out; | |
5119 | } | |
5120 | ||
5121 | prev_svm = to_svm(prev_vcpu); | |
5122 | svm_ir_list_del(prev_svm, pi); | |
5123 | } | |
5124 | ||
5125 | /** | |
5126 | * Allocating new amd_iommu_pi_data, which will get | |
5127 | * add to the per-vcpu ir_list. | |
5128 | */ | |
5129 | ir = kzalloc(sizeof(struct amd_svm_iommu_ir), GFP_KERNEL); | |
5130 | if (!ir) { | |
5131 | ret = -ENOMEM; | |
5132 | goto out; | |
5133 | } | |
5134 | ir->data = pi->ir_data; | |
5135 | ||
5136 | spin_lock_irqsave(&svm->ir_list_lock, flags); | |
5137 | list_add(&ir->node, &svm->ir_list); | |
5138 | spin_unlock_irqrestore(&svm->ir_list_lock, flags); | |
5139 | out: | |
5140 | return ret; | |
5141 | } | |
5142 | ||
5143 | /** | |
5144 | * Note: | |
5145 | * The HW cannot support posting multicast/broadcast | |
5146 | * interrupts to a vCPU. So, we still use legacy interrupt | |
5147 | * remapping for these kind of interrupts. | |
5148 | * | |
5149 | * For lowest-priority interrupts, we only support | |
5150 | * those with single CPU as the destination, e.g. user | |
5151 | * configures the interrupts via /proc/irq or uses | |
5152 | * irqbalance to make the interrupts single-CPU. | |
5153 | */ | |
5154 | static int | |
5155 | get_pi_vcpu_info(struct kvm *kvm, struct kvm_kernel_irq_routing_entry *e, | |
5156 | struct vcpu_data *vcpu_info, struct vcpu_svm **svm) | |
5157 | { | |
5158 | struct kvm_lapic_irq irq; | |
5159 | struct kvm_vcpu *vcpu = NULL; | |
5160 | ||
5161 | kvm_set_msi_irq(kvm, e, &irq); | |
5162 | ||
5163 | if (!kvm_intr_is_single_vcpu(kvm, &irq, &vcpu)) { | |
5164 | pr_debug("SVM: %s: use legacy intr remap mode for irq %u\n", | |
5165 | __func__, irq.vector); | |
5166 | return -1; | |
5167 | } | |
5168 | ||
5169 | pr_debug("SVM: %s: use GA mode for irq %u\n", __func__, | |
5170 | irq.vector); | |
5171 | *svm = to_svm(vcpu); | |
d0ec49d4 | 5172 | vcpu_info->pi_desc_addr = __sme_set(page_to_phys((*svm)->avic_backing_page)); |
411b44ba SS |
5173 | vcpu_info->vector = irq.vector; |
5174 | ||
5175 | return 0; | |
5176 | } | |
5177 | ||
5178 | /* | |
5179 | * svm_update_pi_irte - set IRTE for Posted-Interrupts | |
5180 | * | |
5181 | * @kvm: kvm | |
5182 | * @host_irq: host irq of the interrupt | |
5183 | * @guest_irq: gsi of the interrupt | |
5184 | * @set: set or unset PI | |
5185 | * returns 0 on success, < 0 on failure | |
5186 | */ | |
5187 | static int svm_update_pi_irte(struct kvm *kvm, unsigned int host_irq, | |
5188 | uint32_t guest_irq, bool set) | |
5189 | { | |
5190 | struct kvm_kernel_irq_routing_entry *e; | |
5191 | struct kvm_irq_routing_table *irq_rt; | |
5192 | int idx, ret = -EINVAL; | |
5193 | ||
5194 | if (!kvm_arch_has_assigned_device(kvm) || | |
5195 | !irq_remapping_cap(IRQ_POSTING_CAP)) | |
5196 | return 0; | |
5197 | ||
5198 | pr_debug("SVM: %s: host_irq=%#x, guest_irq=%#x, set=%#x\n", | |
5199 | __func__, host_irq, guest_irq, set); | |
5200 | ||
5201 | idx = srcu_read_lock(&kvm->irq_srcu); | |
5202 | irq_rt = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu); | |
5203 | WARN_ON(guest_irq >= irq_rt->nr_rt_entries); | |
5204 | ||
5205 | hlist_for_each_entry(e, &irq_rt->map[guest_irq], link) { | |
5206 | struct vcpu_data vcpu_info; | |
5207 | struct vcpu_svm *svm = NULL; | |
5208 | ||
5209 | if (e->type != KVM_IRQ_ROUTING_MSI) | |
5210 | continue; | |
5211 | ||
5212 | /** | |
5213 | * Here, we setup with legacy mode in the following cases: | |
5214 | * 1. When cannot target interrupt to a specific vcpu. | |
5215 | * 2. Unsetting posted interrupt. | |
5216 | * 3. APIC virtialization is disabled for the vcpu. | |
5217 | */ | |
5218 | if (!get_pi_vcpu_info(kvm, e, &vcpu_info, &svm) && set && | |
5219 | kvm_vcpu_apicv_active(&svm->vcpu)) { | |
5220 | struct amd_iommu_pi_data pi; | |
5221 | ||
5222 | /* Try to enable guest_mode in IRTE */ | |
d0ec49d4 TL |
5223 | pi.base = __sme_set(page_to_phys(svm->avic_backing_page) & |
5224 | AVIC_HPA_MASK); | |
81811c16 | 5225 | pi.ga_tag = AVIC_GATAG(to_kvm_svm(kvm)->avic_vm_id, |
411b44ba SS |
5226 | svm->vcpu.vcpu_id); |
5227 | pi.is_guest_mode = true; | |
5228 | pi.vcpu_data = &vcpu_info; | |
5229 | ret = irq_set_vcpu_affinity(host_irq, &pi); | |
5230 | ||
5231 | /** | |
5232 | * Here, we successfully setting up vcpu affinity in | |
5233 | * IOMMU guest mode. Now, we need to store the posted | |
5234 | * interrupt information in a per-vcpu ir_list so that | |
5235 | * we can reference to them directly when we update vcpu | |
5236 | * scheduling information in IOMMU irte. | |
5237 | */ | |
5238 | if (!ret && pi.is_guest_mode) | |
5239 | svm_ir_list_add(svm, &pi); | |
5240 | } else { | |
5241 | /* Use legacy mode in IRTE */ | |
5242 | struct amd_iommu_pi_data pi; | |
5243 | ||
5244 | /** | |
5245 | * Here, pi is used to: | |
5246 | * - Tell IOMMU to use legacy mode for this interrupt. | |
5247 | * - Retrieve ga_tag of prior interrupt remapping data. | |
5248 | */ | |
5249 | pi.is_guest_mode = false; | |
5250 | ret = irq_set_vcpu_affinity(host_irq, &pi); | |
5251 | ||
5252 | /** | |
5253 | * Check if the posted interrupt was previously | |
5254 | * setup with the guest_mode by checking if the ga_tag | |
5255 | * was cached. If so, we need to clean up the per-vcpu | |
5256 | * ir_list. | |
5257 | */ | |
5258 | if (!ret && pi.prev_ga_tag) { | |
5259 | int id = AVIC_GATAG_TO_VCPUID(pi.prev_ga_tag); | |
5260 | struct kvm_vcpu *vcpu; | |
5261 | ||
5262 | vcpu = kvm_get_vcpu_by_id(kvm, id); | |
5263 | if (vcpu) | |
5264 | svm_ir_list_del(to_svm(vcpu), &pi); | |
5265 | } | |
5266 | } | |
5267 | ||
5268 | if (!ret && svm) { | |
2698d82e | 5269 | trace_kvm_pi_irte_update(host_irq, svm->vcpu.vcpu_id, |
5270 | e->gsi, vcpu_info.vector, | |
411b44ba SS |
5271 | vcpu_info.pi_desc_addr, set); |
5272 | } | |
5273 | ||
5274 | if (ret < 0) { | |
5275 | pr_err("%s: failed to update PI IRTE\n", __func__); | |
5276 | goto out; | |
5277 | } | |
5278 | } | |
5279 | ||
5280 | ret = 0; | |
5281 | out: | |
5282 | srcu_read_unlock(&kvm->irq_srcu, idx); | |
5283 | return ret; | |
5284 | } | |
5285 | ||
95ba8273 GN |
5286 | static int svm_nmi_allowed(struct kvm_vcpu *vcpu) |
5287 | { | |
5288 | struct vcpu_svm *svm = to_svm(vcpu); | |
5289 | struct vmcb *vmcb = svm->vmcb; | |
924584cc JR |
5290 | int ret; |
5291 | ret = !(vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK) && | |
5292 | !(svm->vcpu.arch.hflags & HF_NMI_MASK); | |
5293 | ret = ret && gif_set(svm) && nested_svm_nmi(svm); | |
5294 | ||
5295 | return ret; | |
aaacfc9a JR |
5296 | } |
5297 | ||
3cfc3092 JK |
5298 | static bool svm_get_nmi_mask(struct kvm_vcpu *vcpu) |
5299 | { | |
5300 | struct vcpu_svm *svm = to_svm(vcpu); | |
5301 | ||
5302 | return !!(svm->vcpu.arch.hflags & HF_NMI_MASK); | |
5303 | } | |
5304 | ||
5305 | static void svm_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked) | |
5306 | { | |
5307 | struct vcpu_svm *svm = to_svm(vcpu); | |
5308 | ||
5309 | if (masked) { | |
5310 | svm->vcpu.arch.hflags |= HF_NMI_MASK; | |
8a05a1b8 | 5311 | set_intercept(svm, INTERCEPT_IRET); |
3cfc3092 JK |
5312 | } else { |
5313 | svm->vcpu.arch.hflags &= ~HF_NMI_MASK; | |
8a05a1b8 | 5314 | clr_intercept(svm, INTERCEPT_IRET); |
3cfc3092 JK |
5315 | } |
5316 | } | |
5317 | ||
78646121 GN |
5318 | static int svm_interrupt_allowed(struct kvm_vcpu *vcpu) |
5319 | { | |
5320 | struct vcpu_svm *svm = to_svm(vcpu); | |
5321 | struct vmcb *vmcb = svm->vmcb; | |
7fcdb510 JR |
5322 | int ret; |
5323 | ||
5324 | if (!gif_set(svm) || | |
5325 | (vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK)) | |
5326 | return 0; | |
5327 | ||
f6e78475 | 5328 | ret = !!(kvm_get_rflags(vcpu) & X86_EFLAGS_IF); |
7fcdb510 | 5329 | |
2030753d | 5330 | if (is_guest_mode(vcpu)) |
7fcdb510 JR |
5331 | return ret && !(svm->vcpu.arch.hflags & HF_VINTR_MASK); |
5332 | ||
5333 | return ret; | |
78646121 GN |
5334 | } |
5335 | ||
c9a7953f | 5336 | static void enable_irq_window(struct kvm_vcpu *vcpu) |
6aa8b732 | 5337 | { |
219b65dc | 5338 | struct vcpu_svm *svm = to_svm(vcpu); |
219b65dc | 5339 | |
340d3bc3 SS |
5340 | if (kvm_vcpu_apicv_active(vcpu)) |
5341 | return; | |
5342 | ||
e0231715 JR |
5343 | /* |
5344 | * In case GIF=0 we can't rely on the CPU to tell us when GIF becomes | |
5345 | * 1, because that's a separate STGI/VMRUN intercept. The next time we | |
5346 | * get that intercept, this function will be called again though and | |
640bd6e5 JN |
5347 | * we'll get the vintr intercept. However, if the vGIF feature is |
5348 | * enabled, the STGI interception will not occur. Enable the irq | |
5349 | * window under the assumption that the hardware will set the GIF. | |
e0231715 | 5350 | */ |
640bd6e5 | 5351 | if ((vgif_enabled(svm) || gif_set(svm)) && nested_svm_intr(svm)) { |
219b65dc AG |
5352 | svm_set_vintr(svm); |
5353 | svm_inject_irq(svm, 0x0); | |
5354 | } | |
85f455f7 ED |
5355 | } |
5356 | ||
c9a7953f | 5357 | static void enable_nmi_window(struct kvm_vcpu *vcpu) |
c1150d8c | 5358 | { |
04d2cc77 | 5359 | struct vcpu_svm *svm = to_svm(vcpu); |
c1150d8c | 5360 | |
44c11430 GN |
5361 | if ((svm->vcpu.arch.hflags & (HF_NMI_MASK | HF_IRET_MASK)) |
5362 | == HF_NMI_MASK) | |
c9a7953f | 5363 | return; /* IRET will cause a vm exit */ |
44c11430 | 5364 | |
640bd6e5 JN |
5365 | if (!gif_set(svm)) { |
5366 | if (vgif_enabled(svm)) | |
5367 | set_intercept(svm, INTERCEPT_STGI); | |
1a5e1852 | 5368 | return; /* STGI will cause a vm exit */ |
640bd6e5 | 5369 | } |
1a5e1852 LP |
5370 | |
5371 | if (svm->nested.exit_required) | |
5372 | return; /* we're not going to run the guest yet */ | |
5373 | ||
e0231715 JR |
5374 | /* |
5375 | * Something prevents NMI from been injected. Single step over possible | |
5376 | * problem (IRET or exception injection or interrupt shadow) | |
5377 | */ | |
ab2f4d73 | 5378 | svm->nmi_singlestep_guest_rflags = svm_get_rflags(vcpu); |
6be7d306 | 5379 | svm->nmi_singlestep = true; |
44c11430 | 5380 | svm->vmcb->save.rflags |= (X86_EFLAGS_TF | X86_EFLAGS_RF); |
c1150d8c DL |
5381 | } |
5382 | ||
cbc94022 IE |
5383 | static int svm_set_tss_addr(struct kvm *kvm, unsigned int addr) |
5384 | { | |
5385 | return 0; | |
5386 | } | |
5387 | ||
2ac52ab8 SC |
5388 | static int svm_set_identity_map_addr(struct kvm *kvm, u64 ident_addr) |
5389 | { | |
5390 | return 0; | |
5391 | } | |
5392 | ||
c2ba05cc | 5393 | static void svm_flush_tlb(struct kvm_vcpu *vcpu, bool invalidate_gpa) |
d9e368d6 | 5394 | { |
38e5e92f JR |
5395 | struct vcpu_svm *svm = to_svm(vcpu); |
5396 | ||
5397 | if (static_cpu_has(X86_FEATURE_FLUSHBYASID)) | |
5398 | svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ASID; | |
5399 | else | |
5400 | svm->asid_generation--; | |
d9e368d6 AK |
5401 | } |
5402 | ||
04d2cc77 AK |
5403 | static void svm_prepare_guest_switch(struct kvm_vcpu *vcpu) |
5404 | { | |
5405 | } | |
5406 | ||
d7bf8221 JR |
5407 | static inline void sync_cr8_to_lapic(struct kvm_vcpu *vcpu) |
5408 | { | |
5409 | struct vcpu_svm *svm = to_svm(vcpu); | |
5410 | ||
3bbf3565 | 5411 | if (svm_nested_virtualize_tpr(vcpu)) |
88ab24ad JR |
5412 | return; |
5413 | ||
4ee546b4 | 5414 | if (!is_cr_intercept(svm, INTERCEPT_CR8_WRITE)) { |
d7bf8221 | 5415 | int cr8 = svm->vmcb->control.int_ctl & V_TPR_MASK; |
615d5193 | 5416 | kvm_set_cr8(vcpu, cr8); |
d7bf8221 JR |
5417 | } |
5418 | } | |
5419 | ||
649d6864 JR |
5420 | static inline void sync_lapic_to_cr8(struct kvm_vcpu *vcpu) |
5421 | { | |
5422 | struct vcpu_svm *svm = to_svm(vcpu); | |
5423 | u64 cr8; | |
5424 | ||
3bbf3565 SS |
5425 | if (svm_nested_virtualize_tpr(vcpu) || |
5426 | kvm_vcpu_apicv_active(vcpu)) | |
88ab24ad JR |
5427 | return; |
5428 | ||
649d6864 JR |
5429 | cr8 = kvm_get_cr8(vcpu); |
5430 | svm->vmcb->control.int_ctl &= ~V_TPR_MASK; | |
5431 | svm->vmcb->control.int_ctl |= cr8 & V_TPR_MASK; | |
5432 | } | |
5433 | ||
9222be18 GN |
5434 | static void svm_complete_interrupts(struct vcpu_svm *svm) |
5435 | { | |
5436 | u8 vector; | |
5437 | int type; | |
5438 | u32 exitintinfo = svm->vmcb->control.exit_int_info; | |
66b7138f JK |
5439 | unsigned int3_injected = svm->int3_injected; |
5440 | ||
5441 | svm->int3_injected = 0; | |
9222be18 | 5442 | |
bd3d1ec3 AK |
5443 | /* |
5444 | * If we've made progress since setting HF_IRET_MASK, we've | |
5445 | * executed an IRET and can allow NMI injection. | |
5446 | */ | |
5447 | if ((svm->vcpu.arch.hflags & HF_IRET_MASK) | |
5448 | && kvm_rip_read(&svm->vcpu) != svm->nmi_iret_rip) { | |
44c11430 | 5449 | svm->vcpu.arch.hflags &= ~(HF_NMI_MASK | HF_IRET_MASK); |
3842d135 AK |
5450 | kvm_make_request(KVM_REQ_EVENT, &svm->vcpu); |
5451 | } | |
44c11430 | 5452 | |
9222be18 GN |
5453 | svm->vcpu.arch.nmi_injected = false; |
5454 | kvm_clear_exception_queue(&svm->vcpu); | |
5455 | kvm_clear_interrupt_queue(&svm->vcpu); | |
5456 | ||
5457 | if (!(exitintinfo & SVM_EXITINTINFO_VALID)) | |
5458 | return; | |
5459 | ||
3842d135 AK |
5460 | kvm_make_request(KVM_REQ_EVENT, &svm->vcpu); |
5461 | ||
9222be18 GN |
5462 | vector = exitintinfo & SVM_EXITINTINFO_VEC_MASK; |
5463 | type = exitintinfo & SVM_EXITINTINFO_TYPE_MASK; | |
5464 | ||
5465 | switch (type) { | |
5466 | case SVM_EXITINTINFO_TYPE_NMI: | |
5467 | svm->vcpu.arch.nmi_injected = true; | |
5468 | break; | |
5469 | case SVM_EXITINTINFO_TYPE_EXEPT: | |
66b7138f JK |
5470 | /* |
5471 | * In case of software exceptions, do not reinject the vector, | |
5472 | * but re-execute the instruction instead. Rewind RIP first | |
5473 | * if we emulated INT3 before. | |
5474 | */ | |
5475 | if (kvm_exception_is_soft(vector)) { | |
5476 | if (vector == BP_VECTOR && int3_injected && | |
5477 | kvm_is_linear_rip(&svm->vcpu, svm->int3_rip)) | |
5478 | kvm_rip_write(&svm->vcpu, | |
5479 | kvm_rip_read(&svm->vcpu) - | |
5480 | int3_injected); | |
9222be18 | 5481 | break; |
66b7138f | 5482 | } |
9222be18 GN |
5483 | if (exitintinfo & SVM_EXITINTINFO_VALID_ERR) { |
5484 | u32 err = svm->vmcb->control.exit_int_info_err; | |
ce7ddec4 | 5485 | kvm_requeue_exception_e(&svm->vcpu, vector, err); |
9222be18 GN |
5486 | |
5487 | } else | |
ce7ddec4 | 5488 | kvm_requeue_exception(&svm->vcpu, vector); |
9222be18 GN |
5489 | break; |
5490 | case SVM_EXITINTINFO_TYPE_INTR: | |
66fd3f7f | 5491 | kvm_queue_interrupt(&svm->vcpu, vector, false); |
9222be18 GN |
5492 | break; |
5493 | default: | |
5494 | break; | |
5495 | } | |
5496 | } | |
5497 | ||
b463a6f7 AK |
5498 | static void svm_cancel_injection(struct kvm_vcpu *vcpu) |
5499 | { | |
5500 | struct vcpu_svm *svm = to_svm(vcpu); | |
5501 | struct vmcb_control_area *control = &svm->vmcb->control; | |
5502 | ||
5503 | control->exit_int_info = control->event_inj; | |
5504 | control->exit_int_info_err = control->event_inj_err; | |
5505 | control->event_inj = 0; | |
5506 | svm_complete_interrupts(svm); | |
5507 | } | |
5508 | ||
851ba692 | 5509 | static void svm_vcpu_run(struct kvm_vcpu *vcpu) |
6aa8b732 | 5510 | { |
a2fa3e9f | 5511 | struct vcpu_svm *svm = to_svm(vcpu); |
d9e368d6 | 5512 | |
2041a06a JR |
5513 | svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX]; |
5514 | svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP]; | |
5515 | svm->vmcb->save.rip = vcpu->arch.regs[VCPU_REGS_RIP]; | |
5516 | ||
cd3ff653 JR |
5517 | /* |
5518 | * A vmexit emulation is required before the vcpu can be executed | |
5519 | * again. | |
5520 | */ | |
5521 | if (unlikely(svm->nested.exit_required)) | |
5522 | return; | |
5523 | ||
a12713c2 LP |
5524 | /* |
5525 | * Disable singlestep if we're injecting an interrupt/exception. | |
5526 | * We don't want our modified rflags to be pushed on the stack where | |
5527 | * we might not be able to easily reset them if we disabled NMI | |
5528 | * singlestep later. | |
5529 | */ | |
5530 | if (svm->nmi_singlestep && svm->vmcb->control.event_inj) { | |
5531 | /* | |
5532 | * Event injection happens before external interrupts cause a | |
5533 | * vmexit and interrupts are disabled here, so smp_send_reschedule | |
5534 | * is enough to force an immediate vmexit. | |
5535 | */ | |
5536 | disable_nmi_singlestep(svm); | |
5537 | smp_send_reschedule(vcpu->cpu); | |
5538 | } | |
5539 | ||
e756fc62 | 5540 | pre_svm_run(svm); |
6aa8b732 | 5541 | |
649d6864 JR |
5542 | sync_lapic_to_cr8(vcpu); |
5543 | ||
cda0ffdd | 5544 | svm->vmcb->save.cr2 = vcpu->arch.cr2; |
6aa8b732 | 5545 | |
04d2cc77 AK |
5546 | clgi(); |
5547 | ||
5548 | local_irq_enable(); | |
36241b8c | 5549 | |
b2ac58f9 KA |
5550 | /* |
5551 | * If this vCPU has touched SPEC_CTRL, restore the guest's value if | |
5552 | * it's non-zero. Since vmentry is serialising on affected CPUs, there | |
5553 | * is no need to worry about the conditional branch over the wrmsr | |
5554 | * being speculatively taken. | |
5555 | */ | |
5556 | if (svm->spec_ctrl) | |
ecb586bd | 5557 | native_wrmsrl(MSR_IA32_SPEC_CTRL, svm->spec_ctrl); |
b2ac58f9 | 5558 | |
6aa8b732 | 5559 | asm volatile ( |
7454766f AK |
5560 | "push %%" _ASM_BP "; \n\t" |
5561 | "mov %c[rbx](%[svm]), %%" _ASM_BX " \n\t" | |
5562 | "mov %c[rcx](%[svm]), %%" _ASM_CX " \n\t" | |
5563 | "mov %c[rdx](%[svm]), %%" _ASM_DX " \n\t" | |
5564 | "mov %c[rsi](%[svm]), %%" _ASM_SI " \n\t" | |
5565 | "mov %c[rdi](%[svm]), %%" _ASM_DI " \n\t" | |
5566 | "mov %c[rbp](%[svm]), %%" _ASM_BP " \n\t" | |
05b3e0c2 | 5567 | #ifdef CONFIG_X86_64 |
fb3f0f51 RR |
5568 | "mov %c[r8](%[svm]), %%r8 \n\t" |
5569 | "mov %c[r9](%[svm]), %%r9 \n\t" | |
5570 | "mov %c[r10](%[svm]), %%r10 \n\t" | |
5571 | "mov %c[r11](%[svm]), %%r11 \n\t" | |
5572 | "mov %c[r12](%[svm]), %%r12 \n\t" | |
5573 | "mov %c[r13](%[svm]), %%r13 \n\t" | |
5574 | "mov %c[r14](%[svm]), %%r14 \n\t" | |
5575 | "mov %c[r15](%[svm]), %%r15 \n\t" | |
6aa8b732 AK |
5576 | #endif |
5577 | ||
6aa8b732 | 5578 | /* Enter guest mode */ |
7454766f AK |
5579 | "push %%" _ASM_AX " \n\t" |
5580 | "mov %c[vmcb](%[svm]), %%" _ASM_AX " \n\t" | |
4ecac3fd AK |
5581 | __ex(SVM_VMLOAD) "\n\t" |
5582 | __ex(SVM_VMRUN) "\n\t" | |
5583 | __ex(SVM_VMSAVE) "\n\t" | |
7454766f | 5584 | "pop %%" _ASM_AX " \n\t" |
6aa8b732 AK |
5585 | |
5586 | /* Save guest registers, load host registers */ | |
7454766f AK |
5587 | "mov %%" _ASM_BX ", %c[rbx](%[svm]) \n\t" |
5588 | "mov %%" _ASM_CX ", %c[rcx](%[svm]) \n\t" | |
5589 | "mov %%" _ASM_DX ", %c[rdx](%[svm]) \n\t" | |
5590 | "mov %%" _ASM_SI ", %c[rsi](%[svm]) \n\t" | |
5591 | "mov %%" _ASM_DI ", %c[rdi](%[svm]) \n\t" | |
5592 | "mov %%" _ASM_BP ", %c[rbp](%[svm]) \n\t" | |
05b3e0c2 | 5593 | #ifdef CONFIG_X86_64 |
fb3f0f51 RR |
5594 | "mov %%r8, %c[r8](%[svm]) \n\t" |
5595 | "mov %%r9, %c[r9](%[svm]) \n\t" | |
5596 | "mov %%r10, %c[r10](%[svm]) \n\t" | |
5597 | "mov %%r11, %c[r11](%[svm]) \n\t" | |
5598 | "mov %%r12, %c[r12](%[svm]) \n\t" | |
5599 | "mov %%r13, %c[r13](%[svm]) \n\t" | |
5600 | "mov %%r14, %c[r14](%[svm]) \n\t" | |
5601 | "mov %%r15, %c[r15](%[svm]) \n\t" | |
0cb5b306 JM |
5602 | #endif |
5603 | /* | |
5604 | * Clear host registers marked as clobbered to prevent | |
5605 | * speculative use. | |
5606 | */ | |
5607 | "xor %%" _ASM_BX ", %%" _ASM_BX " \n\t" | |
5608 | "xor %%" _ASM_CX ", %%" _ASM_CX " \n\t" | |
5609 | "xor %%" _ASM_DX ", %%" _ASM_DX " \n\t" | |
5610 | "xor %%" _ASM_SI ", %%" _ASM_SI " \n\t" | |
5611 | "xor %%" _ASM_DI ", %%" _ASM_DI " \n\t" | |
5612 | #ifdef CONFIG_X86_64 | |
5613 | "xor %%r8, %%r8 \n\t" | |
5614 | "xor %%r9, %%r9 \n\t" | |
5615 | "xor %%r10, %%r10 \n\t" | |
5616 | "xor %%r11, %%r11 \n\t" | |
5617 | "xor %%r12, %%r12 \n\t" | |
5618 | "xor %%r13, %%r13 \n\t" | |
5619 | "xor %%r14, %%r14 \n\t" | |
5620 | "xor %%r15, %%r15 \n\t" | |
6aa8b732 | 5621 | #endif |
7454766f | 5622 | "pop %%" _ASM_BP |
6aa8b732 | 5623 | : |
fb3f0f51 | 5624 | : [svm]"a"(svm), |
6aa8b732 | 5625 | [vmcb]"i"(offsetof(struct vcpu_svm, vmcb_pa)), |
ad312c7c ZX |
5626 | [rbx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBX])), |
5627 | [rcx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RCX])), | |
5628 | [rdx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDX])), | |
5629 | [rsi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RSI])), | |
5630 | [rdi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDI])), | |
5631 | [rbp]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBP])) | |
05b3e0c2 | 5632 | #ifdef CONFIG_X86_64 |
ad312c7c ZX |
5633 | , [r8]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R8])), |
5634 | [r9]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R9])), | |
5635 | [r10]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R10])), | |
5636 | [r11]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R11])), | |
5637 | [r12]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R12])), | |
5638 | [r13]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R13])), | |
5639 | [r14]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R14])), | |
5640 | [r15]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R15])) | |
6aa8b732 | 5641 | #endif |
54a08c04 LV |
5642 | : "cc", "memory" |
5643 | #ifdef CONFIG_X86_64 | |
7454766f | 5644 | , "rbx", "rcx", "rdx", "rsi", "rdi" |
54a08c04 | 5645 | , "r8", "r9", "r10", "r11" , "r12", "r13", "r14", "r15" |
7454766f AK |
5646 | #else |
5647 | , "ebx", "ecx", "edx", "esi", "edi" | |
54a08c04 LV |
5648 | #endif |
5649 | ); | |
6aa8b732 | 5650 | |
b2ac58f9 KA |
5651 | /* |
5652 | * We do not use IBRS in the kernel. If this vCPU has used the | |
5653 | * SPEC_CTRL MSR it may have left it on; save the value and | |
5654 | * turn it off. This is much more efficient than blindly adding | |
5655 | * it to the atomic save/restore list. Especially as the former | |
5656 | * (Saving guest MSRs on vmexit) doesn't even exist in KVM. | |
5657 | * | |
5658 | * For non-nested case: | |
5659 | * If the L01 MSR bitmap does not intercept the MSR, then we need to | |
5660 | * save it. | |
5661 | * | |
5662 | * For nested case: | |
5663 | * If the L02 MSR bitmap does not intercept the MSR, then we need to | |
5664 | * save it. | |
5665 | */ | |
946fbbc1 | 5666 | if (unlikely(!msr_write_intercepted(vcpu, MSR_IA32_SPEC_CTRL))) |
ecb586bd | 5667 | svm->spec_ctrl = native_read_msr(MSR_IA32_SPEC_CTRL); |
b2ac58f9 KA |
5668 | |
5669 | if (svm->spec_ctrl) | |
ecb586bd | 5670 | native_wrmsrl(MSR_IA32_SPEC_CTRL, 0); |
b2ac58f9 | 5671 | |
117cc7a9 DW |
5672 | /* Eliminate branch target predictions from guest mode */ |
5673 | vmexit_fill_RSB(); | |
5674 | ||
82ca2d10 AK |
5675 | #ifdef CONFIG_X86_64 |
5676 | wrmsrl(MSR_GS_BASE, svm->host.gs_base); | |
5677 | #else | |
dacccfdd | 5678 | loadsegment(fs, svm->host.fs); |
831ca609 AK |
5679 | #ifndef CONFIG_X86_32_LAZY_GS |
5680 | loadsegment(gs, svm->host.gs); | |
5681 | #endif | |
9581d442 | 5682 | #endif |
6aa8b732 AK |
5683 | |
5684 | reload_tss(vcpu); | |
5685 | ||
56ba47dd AK |
5686 | local_irq_disable(); |
5687 | ||
13c34e07 AK |
5688 | vcpu->arch.cr2 = svm->vmcb->save.cr2; |
5689 | vcpu->arch.regs[VCPU_REGS_RAX] = svm->vmcb->save.rax; | |
5690 | vcpu->arch.regs[VCPU_REGS_RSP] = svm->vmcb->save.rsp; | |
5691 | vcpu->arch.regs[VCPU_REGS_RIP] = svm->vmcb->save.rip; | |
5692 | ||
3781c01c | 5693 | if (unlikely(svm->vmcb->control.exit_code == SVM_EXIT_NMI)) |
dd60d217 | 5694 | kvm_before_interrupt(&svm->vcpu); |
3781c01c JR |
5695 | |
5696 | stgi(); | |
5697 | ||
5698 | /* Any pending NMI will happen here */ | |
5699 | ||
5700 | if (unlikely(svm->vmcb->control.exit_code == SVM_EXIT_NMI)) | |
dd60d217 | 5701 | kvm_after_interrupt(&svm->vcpu); |
3781c01c | 5702 | |
d7bf8221 JR |
5703 | sync_cr8_to_lapic(vcpu); |
5704 | ||
a2fa3e9f | 5705 | svm->next_rip = 0; |
9222be18 | 5706 | |
38e5e92f JR |
5707 | svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING; |
5708 | ||
631bc487 GN |
5709 | /* if exit due to PF check for async PF */ |
5710 | if (svm->vmcb->control.exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR) | |
1261bfa3 | 5711 | svm->vcpu.arch.apf.host_apf_reason = kvm_read_and_reset_pf_reason(); |
631bc487 | 5712 | |
6de4f3ad AK |
5713 | if (npt_enabled) { |
5714 | vcpu->arch.regs_avail &= ~(1 << VCPU_EXREG_PDPTR); | |
5715 | vcpu->arch.regs_dirty &= ~(1 << VCPU_EXREG_PDPTR); | |
5716 | } | |
fe5913e4 JR |
5717 | |
5718 | /* | |
5719 | * We need to handle MC intercepts here before the vcpu has a chance to | |
5720 | * change the physical cpu | |
5721 | */ | |
5722 | if (unlikely(svm->vmcb->control.exit_code == | |
5723 | SVM_EXIT_EXCP_BASE + MC_VECTOR)) | |
5724 | svm_handle_mce(svm); | |
8d28fec4 RJ |
5725 | |
5726 | mark_all_clean(svm->vmcb); | |
6aa8b732 | 5727 | } |
c207aee4 | 5728 | STACK_FRAME_NON_STANDARD(svm_vcpu_run); |
6aa8b732 | 5729 | |
6aa8b732 AK |
5730 | static void svm_set_cr3(struct kvm_vcpu *vcpu, unsigned long root) |
5731 | { | |
a2fa3e9f GH |
5732 | struct vcpu_svm *svm = to_svm(vcpu); |
5733 | ||
d0ec49d4 | 5734 | svm->vmcb->save.cr3 = __sme_set(root); |
dcca1a65 | 5735 | mark_dirty(svm->vmcb, VMCB_CR); |
c2ba05cc | 5736 | svm_flush_tlb(vcpu, true); |
6aa8b732 AK |
5737 | } |
5738 | ||
1c97f0a0 JR |
5739 | static void set_tdp_cr3(struct kvm_vcpu *vcpu, unsigned long root) |
5740 | { | |
5741 | struct vcpu_svm *svm = to_svm(vcpu); | |
5742 | ||
d0ec49d4 | 5743 | svm->vmcb->control.nested_cr3 = __sme_set(root); |
b2747166 | 5744 | mark_dirty(svm->vmcb, VMCB_NPT); |
1c97f0a0 JR |
5745 | |
5746 | /* Also sync guest cr3 here in case we live migrate */ | |
9f8fe504 | 5747 | svm->vmcb->save.cr3 = kvm_read_cr3(vcpu); |
dcca1a65 | 5748 | mark_dirty(svm->vmcb, VMCB_CR); |
1c97f0a0 | 5749 | |
c2ba05cc | 5750 | svm_flush_tlb(vcpu, true); |
1c97f0a0 JR |
5751 | } |
5752 | ||
6aa8b732 AK |
5753 | static int is_disabled(void) |
5754 | { | |
6031a61c JR |
5755 | u64 vm_cr; |
5756 | ||
5757 | rdmsrl(MSR_VM_CR, vm_cr); | |
5758 | if (vm_cr & (1 << SVM_VM_CR_SVM_DISABLE)) | |
5759 | return 1; | |
5760 | ||
6aa8b732 AK |
5761 | return 0; |
5762 | } | |
5763 | ||
102d8325 IM |
5764 | static void |
5765 | svm_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall) | |
5766 | { | |
5767 | /* | |
5768 | * Patch in the VMMCALL instruction: | |
5769 | */ | |
5770 | hypercall[0] = 0x0f; | |
5771 | hypercall[1] = 0x01; | |
5772 | hypercall[2] = 0xd9; | |
102d8325 IM |
5773 | } |
5774 | ||
002c7f7c YS |
5775 | static void svm_check_processor_compat(void *rtn) |
5776 | { | |
5777 | *(int *)rtn = 0; | |
5778 | } | |
5779 | ||
774ead3a AK |
5780 | static bool svm_cpu_has_accelerated_tpr(void) |
5781 | { | |
5782 | return false; | |
5783 | } | |
5784 | ||
6d396b55 PB |
5785 | static bool svm_has_high_real_mode_segbase(void) |
5786 | { | |
5787 | return true; | |
5788 | } | |
5789 | ||
fc07e76a PB |
5790 | static u64 svm_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio) |
5791 | { | |
5792 | return 0; | |
5793 | } | |
5794 | ||
0e851880 SY |
5795 | static void svm_cpuid_update(struct kvm_vcpu *vcpu) |
5796 | { | |
6092d3d3 JR |
5797 | struct vcpu_svm *svm = to_svm(vcpu); |
5798 | ||
5799 | /* Update nrips enabled cache */ | |
d6321d49 | 5800 | svm->nrips_enabled = !!guest_cpuid_has(&svm->vcpu, X86_FEATURE_NRIPS); |
46781eae SS |
5801 | |
5802 | if (!kvm_vcpu_apicv_active(vcpu)) | |
5803 | return; | |
5804 | ||
1b4d56b8 | 5805 | guest_cpuid_clear(vcpu, X86_FEATURE_X2APIC); |
0e851880 SY |
5806 | } |
5807 | ||
d4330ef2 JR |
5808 | static void svm_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry) |
5809 | { | |
c2c63a49 | 5810 | switch (func) { |
46781eae SS |
5811 | case 0x1: |
5812 | if (avic) | |
5813 | entry->ecx &= ~bit(X86_FEATURE_X2APIC); | |
5814 | break; | |
4c62a2dc JR |
5815 | case 0x80000001: |
5816 | if (nested) | |
5817 | entry->ecx |= (1 << 2); /* Set SVM bit */ | |
5818 | break; | |
c2c63a49 JR |
5819 | case 0x8000000A: |
5820 | entry->eax = 1; /* SVM revision 1 */ | |
5821 | entry->ebx = 8; /* Lets support 8 ASIDs in case we add proper | |
5822 | ASID emulation to nested SVM */ | |
5823 | entry->ecx = 0; /* Reserved */ | |
7a190667 JR |
5824 | entry->edx = 0; /* Per default do not support any |
5825 | additional features */ | |
5826 | ||
5827 | /* Support next_rip if host supports it */ | |
2a6b20b8 | 5828 | if (boot_cpu_has(X86_FEATURE_NRIPS)) |
7a190667 | 5829 | entry->edx |= SVM_FEATURE_NRIP; |
c2c63a49 | 5830 | |
3d4aeaad JR |
5831 | /* Support NPT for the guest if enabled */ |
5832 | if (npt_enabled) | |
5833 | entry->edx |= SVM_FEATURE_NPT; | |
5834 | ||
c2c63a49 | 5835 | break; |
8765d753 BS |
5836 | case 0x8000001F: |
5837 | /* Support memory encryption cpuid if host supports it */ | |
5838 | if (boot_cpu_has(X86_FEATURE_SEV)) | |
5839 | cpuid(0x8000001f, &entry->eax, &entry->ebx, | |
5840 | &entry->ecx, &entry->edx); | |
5841 | ||
c2c63a49 | 5842 | } |
d4330ef2 JR |
5843 | } |
5844 | ||
17cc3935 | 5845 | static int svm_get_lpage_level(void) |
344f414f | 5846 | { |
17cc3935 | 5847 | return PT_PDPE_LEVEL; |
344f414f JR |
5848 | } |
5849 | ||
4e47c7a6 SY |
5850 | static bool svm_rdtscp_supported(void) |
5851 | { | |
46896c73 | 5852 | return boot_cpu_has(X86_FEATURE_RDTSCP); |
4e47c7a6 SY |
5853 | } |
5854 | ||
ad756a16 MJ |
5855 | static bool svm_invpcid_supported(void) |
5856 | { | |
5857 | return false; | |
5858 | } | |
5859 | ||
93c4adc7 PB |
5860 | static bool svm_mpx_supported(void) |
5861 | { | |
5862 | return false; | |
5863 | } | |
5864 | ||
55412b2e WL |
5865 | static bool svm_xsaves_supported(void) |
5866 | { | |
5867 | return false; | |
5868 | } | |
5869 | ||
66336cab PB |
5870 | static bool svm_umip_emulated(void) |
5871 | { | |
5872 | return false; | |
5873 | } | |
5874 | ||
f5f48ee1 SY |
5875 | static bool svm_has_wbinvd_exit(void) |
5876 | { | |
5877 | return true; | |
5878 | } | |
5879 | ||
8061252e | 5880 | #define PRE_EX(exit) { .exit_code = (exit), \ |
40e19b51 | 5881 | .stage = X86_ICPT_PRE_EXCEPT, } |
cfec82cb | 5882 | #define POST_EX(exit) { .exit_code = (exit), \ |
40e19b51 | 5883 | .stage = X86_ICPT_POST_EXCEPT, } |
d7eb8203 | 5884 | #define POST_MEM(exit) { .exit_code = (exit), \ |
40e19b51 | 5885 | .stage = X86_ICPT_POST_MEMACCESS, } |
cfec82cb | 5886 | |
09941fbb | 5887 | static const struct __x86_intercept { |
cfec82cb JR |
5888 | u32 exit_code; |
5889 | enum x86_intercept_stage stage; | |
cfec82cb JR |
5890 | } x86_intercept_map[] = { |
5891 | [x86_intercept_cr_read] = POST_EX(SVM_EXIT_READ_CR0), | |
5892 | [x86_intercept_cr_write] = POST_EX(SVM_EXIT_WRITE_CR0), | |
5893 | [x86_intercept_clts] = POST_EX(SVM_EXIT_WRITE_CR0), | |
5894 | [x86_intercept_lmsw] = POST_EX(SVM_EXIT_WRITE_CR0), | |
5895 | [x86_intercept_smsw] = POST_EX(SVM_EXIT_READ_CR0), | |
3b88e41a JR |
5896 | [x86_intercept_dr_read] = POST_EX(SVM_EXIT_READ_DR0), |
5897 | [x86_intercept_dr_write] = POST_EX(SVM_EXIT_WRITE_DR0), | |
dee6bb70 JR |
5898 | [x86_intercept_sldt] = POST_EX(SVM_EXIT_LDTR_READ), |
5899 | [x86_intercept_str] = POST_EX(SVM_EXIT_TR_READ), | |
5900 | [x86_intercept_lldt] = POST_EX(SVM_EXIT_LDTR_WRITE), | |
5901 | [x86_intercept_ltr] = POST_EX(SVM_EXIT_TR_WRITE), | |
5902 | [x86_intercept_sgdt] = POST_EX(SVM_EXIT_GDTR_READ), | |
5903 | [x86_intercept_sidt] = POST_EX(SVM_EXIT_IDTR_READ), | |
5904 | [x86_intercept_lgdt] = POST_EX(SVM_EXIT_GDTR_WRITE), | |
5905 | [x86_intercept_lidt] = POST_EX(SVM_EXIT_IDTR_WRITE), | |
01de8b09 JR |
5906 | [x86_intercept_vmrun] = POST_EX(SVM_EXIT_VMRUN), |
5907 | [x86_intercept_vmmcall] = POST_EX(SVM_EXIT_VMMCALL), | |
5908 | [x86_intercept_vmload] = POST_EX(SVM_EXIT_VMLOAD), | |
5909 | [x86_intercept_vmsave] = POST_EX(SVM_EXIT_VMSAVE), | |
5910 | [x86_intercept_stgi] = POST_EX(SVM_EXIT_STGI), | |
5911 | [x86_intercept_clgi] = POST_EX(SVM_EXIT_CLGI), | |
5912 | [x86_intercept_skinit] = POST_EX(SVM_EXIT_SKINIT), | |
5913 | [x86_intercept_invlpga] = POST_EX(SVM_EXIT_INVLPGA), | |
d7eb8203 JR |
5914 | [x86_intercept_rdtscp] = POST_EX(SVM_EXIT_RDTSCP), |
5915 | [x86_intercept_monitor] = POST_MEM(SVM_EXIT_MONITOR), | |
5916 | [x86_intercept_mwait] = POST_EX(SVM_EXIT_MWAIT), | |
8061252e JR |
5917 | [x86_intercept_invlpg] = POST_EX(SVM_EXIT_INVLPG), |
5918 | [x86_intercept_invd] = POST_EX(SVM_EXIT_INVD), | |
5919 | [x86_intercept_wbinvd] = POST_EX(SVM_EXIT_WBINVD), | |
5920 | [x86_intercept_wrmsr] = POST_EX(SVM_EXIT_MSR), | |
5921 | [x86_intercept_rdtsc] = POST_EX(SVM_EXIT_RDTSC), | |
5922 | [x86_intercept_rdmsr] = POST_EX(SVM_EXIT_MSR), | |
5923 | [x86_intercept_rdpmc] = POST_EX(SVM_EXIT_RDPMC), | |
5924 | [x86_intercept_cpuid] = PRE_EX(SVM_EXIT_CPUID), | |
5925 | [x86_intercept_rsm] = PRE_EX(SVM_EXIT_RSM), | |
bf608f88 JR |
5926 | [x86_intercept_pause] = PRE_EX(SVM_EXIT_PAUSE), |
5927 | [x86_intercept_pushf] = PRE_EX(SVM_EXIT_PUSHF), | |
5928 | [x86_intercept_popf] = PRE_EX(SVM_EXIT_POPF), | |
5929 | [x86_intercept_intn] = PRE_EX(SVM_EXIT_SWINT), | |
5930 | [x86_intercept_iret] = PRE_EX(SVM_EXIT_IRET), | |
5931 | [x86_intercept_icebp] = PRE_EX(SVM_EXIT_ICEBP), | |
5932 | [x86_intercept_hlt] = POST_EX(SVM_EXIT_HLT), | |
f6511935 JR |
5933 | [x86_intercept_in] = POST_EX(SVM_EXIT_IOIO), |
5934 | [x86_intercept_ins] = POST_EX(SVM_EXIT_IOIO), | |
5935 | [x86_intercept_out] = POST_EX(SVM_EXIT_IOIO), | |
5936 | [x86_intercept_outs] = POST_EX(SVM_EXIT_IOIO), | |
cfec82cb JR |
5937 | }; |
5938 | ||
8061252e | 5939 | #undef PRE_EX |
cfec82cb | 5940 | #undef POST_EX |
d7eb8203 | 5941 | #undef POST_MEM |
cfec82cb | 5942 | |
8a76d7f2 JR |
5943 | static int svm_check_intercept(struct kvm_vcpu *vcpu, |
5944 | struct x86_instruction_info *info, | |
5945 | enum x86_intercept_stage stage) | |
5946 | { | |
cfec82cb JR |
5947 | struct vcpu_svm *svm = to_svm(vcpu); |
5948 | int vmexit, ret = X86EMUL_CONTINUE; | |
5949 | struct __x86_intercept icpt_info; | |
5950 | struct vmcb *vmcb = svm->vmcb; | |
5951 | ||
5952 | if (info->intercept >= ARRAY_SIZE(x86_intercept_map)) | |
5953 | goto out; | |
5954 | ||
5955 | icpt_info = x86_intercept_map[info->intercept]; | |
5956 | ||
40e19b51 | 5957 | if (stage != icpt_info.stage) |
cfec82cb JR |
5958 | goto out; |
5959 | ||
5960 | switch (icpt_info.exit_code) { | |
5961 | case SVM_EXIT_READ_CR0: | |
5962 | if (info->intercept == x86_intercept_cr_read) | |
5963 | icpt_info.exit_code += info->modrm_reg; | |
5964 | break; | |
5965 | case SVM_EXIT_WRITE_CR0: { | |
5966 | unsigned long cr0, val; | |
5967 | u64 intercept; | |
5968 | ||
5969 | if (info->intercept == x86_intercept_cr_write) | |
5970 | icpt_info.exit_code += info->modrm_reg; | |
5971 | ||
62baf44c JK |
5972 | if (icpt_info.exit_code != SVM_EXIT_WRITE_CR0 || |
5973 | info->intercept == x86_intercept_clts) | |
cfec82cb JR |
5974 | break; |
5975 | ||
5976 | intercept = svm->nested.intercept; | |
5977 | ||
5978 | if (!(intercept & (1ULL << INTERCEPT_SELECTIVE_CR0))) | |
5979 | break; | |
5980 | ||
5981 | cr0 = vcpu->arch.cr0 & ~SVM_CR0_SELECTIVE_MASK; | |
5982 | val = info->src_val & ~SVM_CR0_SELECTIVE_MASK; | |
5983 | ||
5984 | if (info->intercept == x86_intercept_lmsw) { | |
5985 | cr0 &= 0xfUL; | |
5986 | val &= 0xfUL; | |
5987 | /* lmsw can't clear PE - catch this here */ | |
5988 | if (cr0 & X86_CR0_PE) | |
5989 | val |= X86_CR0_PE; | |
5990 | } | |
5991 | ||
5992 | if (cr0 ^ val) | |
5993 | icpt_info.exit_code = SVM_EXIT_CR0_SEL_WRITE; | |
5994 | ||
5995 | break; | |
5996 | } | |
3b88e41a JR |
5997 | case SVM_EXIT_READ_DR0: |
5998 | case SVM_EXIT_WRITE_DR0: | |
5999 | icpt_info.exit_code += info->modrm_reg; | |
6000 | break; | |
8061252e JR |
6001 | case SVM_EXIT_MSR: |
6002 | if (info->intercept == x86_intercept_wrmsr) | |
6003 | vmcb->control.exit_info_1 = 1; | |
6004 | else | |
6005 | vmcb->control.exit_info_1 = 0; | |
6006 | break; | |
bf608f88 JR |
6007 | case SVM_EXIT_PAUSE: |
6008 | /* | |
6009 | * We get this for NOP only, but pause | |
6010 | * is rep not, check this here | |
6011 | */ | |
6012 | if (info->rep_prefix != REPE_PREFIX) | |
6013 | goto out; | |
49a8afca | 6014 | break; |
f6511935 JR |
6015 | case SVM_EXIT_IOIO: { |
6016 | u64 exit_info; | |
6017 | u32 bytes; | |
6018 | ||
f6511935 JR |
6019 | if (info->intercept == x86_intercept_in || |
6020 | info->intercept == x86_intercept_ins) { | |
6cbc5f5a JK |
6021 | exit_info = ((info->src_val & 0xffff) << 16) | |
6022 | SVM_IOIO_TYPE_MASK; | |
f6511935 | 6023 | bytes = info->dst_bytes; |
6493f157 | 6024 | } else { |
6cbc5f5a | 6025 | exit_info = (info->dst_val & 0xffff) << 16; |
6493f157 | 6026 | bytes = info->src_bytes; |
f6511935 JR |
6027 | } |
6028 | ||
6029 | if (info->intercept == x86_intercept_outs || | |
6030 | info->intercept == x86_intercept_ins) | |
6031 | exit_info |= SVM_IOIO_STR_MASK; | |
6032 | ||
6033 | if (info->rep_prefix) | |
6034 | exit_info |= SVM_IOIO_REP_MASK; | |
6035 | ||
6036 | bytes = min(bytes, 4u); | |
6037 | ||
6038 | exit_info |= bytes << SVM_IOIO_SIZE_SHIFT; | |
6039 | ||
6040 | exit_info |= (u32)info->ad_bytes << (SVM_IOIO_ASIZE_SHIFT - 1); | |
6041 | ||
6042 | vmcb->control.exit_info_1 = exit_info; | |
6043 | vmcb->control.exit_info_2 = info->next_rip; | |
6044 | ||
6045 | break; | |
6046 | } | |
cfec82cb JR |
6047 | default: |
6048 | break; | |
6049 | } | |
6050 | ||
f104765b BD |
6051 | /* TODO: Advertise NRIPS to guest hypervisor unconditionally */ |
6052 | if (static_cpu_has(X86_FEATURE_NRIPS)) | |
6053 | vmcb->control.next_rip = info->next_rip; | |
cfec82cb JR |
6054 | vmcb->control.exit_code = icpt_info.exit_code; |
6055 | vmexit = nested_svm_exit_handled(svm); | |
6056 | ||
6057 | ret = (vmexit == NESTED_EXIT_DONE) ? X86EMUL_INTERCEPTED | |
6058 | : X86EMUL_CONTINUE; | |
6059 | ||
6060 | out: | |
6061 | return ret; | |
8a76d7f2 JR |
6062 | } |
6063 | ||
a547c6db YZ |
6064 | static void svm_handle_external_intr(struct kvm_vcpu *vcpu) |
6065 | { | |
6066 | local_irq_enable(); | |
f2485b3e PB |
6067 | /* |
6068 | * We must have an instruction with interrupts enabled, so | |
6069 | * the timer interrupt isn't delayed by the interrupt shadow. | |
6070 | */ | |
6071 | asm("nop"); | |
6072 | local_irq_disable(); | |
a547c6db YZ |
6073 | } |
6074 | ||
ae97a3b8 RK |
6075 | static void svm_sched_in(struct kvm_vcpu *vcpu, int cpu) |
6076 | { | |
8566ac8b BM |
6077 | if (pause_filter_thresh) |
6078 | shrink_ple_window(vcpu); | |
ae97a3b8 RK |
6079 | } |
6080 | ||
be8ca170 SS |
6081 | static inline void avic_post_state_restore(struct kvm_vcpu *vcpu) |
6082 | { | |
6083 | if (avic_handle_apic_id_update(vcpu) != 0) | |
6084 | return; | |
6085 | if (avic_handle_dfr_update(vcpu) != 0) | |
6086 | return; | |
6087 | avic_handle_ldr_update(vcpu); | |
6088 | } | |
6089 | ||
74f16909 BP |
6090 | static void svm_setup_mce(struct kvm_vcpu *vcpu) |
6091 | { | |
6092 | /* [63:9] are reserved. */ | |
6093 | vcpu->arch.mcg_cap &= 0x1ff; | |
6094 | } | |
6095 | ||
72d7b374 LP |
6096 | static int svm_smi_allowed(struct kvm_vcpu *vcpu) |
6097 | { | |
05cade71 LP |
6098 | struct vcpu_svm *svm = to_svm(vcpu); |
6099 | ||
6100 | /* Per APM Vol.2 15.22.2 "Response to SMI" */ | |
6101 | if (!gif_set(svm)) | |
6102 | return 0; | |
6103 | ||
6104 | if (is_guest_mode(&svm->vcpu) && | |
6105 | svm->nested.intercept & (1ULL << INTERCEPT_SMI)) { | |
6106 | /* TODO: Might need to set exit_info_1 and exit_info_2 here */ | |
6107 | svm->vmcb->control.exit_code = SVM_EXIT_SMI; | |
6108 | svm->nested.exit_required = true; | |
6109 | return 0; | |
6110 | } | |
6111 | ||
72d7b374 LP |
6112 | return 1; |
6113 | } | |
6114 | ||
0234bf88 LP |
6115 | static int svm_pre_enter_smm(struct kvm_vcpu *vcpu, char *smstate) |
6116 | { | |
05cade71 LP |
6117 | struct vcpu_svm *svm = to_svm(vcpu); |
6118 | int ret; | |
6119 | ||
6120 | if (is_guest_mode(vcpu)) { | |
6121 | /* FED8h - SVM Guest */ | |
6122 | put_smstate(u64, smstate, 0x7ed8, 1); | |
6123 | /* FEE0h - SVM Guest VMCB Physical Address */ | |
6124 | put_smstate(u64, smstate, 0x7ee0, svm->nested.vmcb); | |
6125 | ||
6126 | svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX]; | |
6127 | svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP]; | |
6128 | svm->vmcb->save.rip = vcpu->arch.regs[VCPU_REGS_RIP]; | |
6129 | ||
6130 | ret = nested_svm_vmexit(svm); | |
6131 | if (ret) | |
6132 | return ret; | |
6133 | } | |
0234bf88 LP |
6134 | return 0; |
6135 | } | |
6136 | ||
6137 | static int svm_pre_leave_smm(struct kvm_vcpu *vcpu, u64 smbase) | |
6138 | { | |
05cade71 LP |
6139 | struct vcpu_svm *svm = to_svm(vcpu); |
6140 | struct vmcb *nested_vmcb; | |
6141 | struct page *page; | |
6142 | struct { | |
6143 | u64 guest; | |
6144 | u64 vmcb; | |
6145 | } svm_state_save; | |
6146 | int ret; | |
6147 | ||
6148 | ret = kvm_vcpu_read_guest(vcpu, smbase + 0xfed8, &svm_state_save, | |
6149 | sizeof(svm_state_save)); | |
6150 | if (ret) | |
6151 | return ret; | |
6152 | ||
6153 | if (svm_state_save.guest) { | |
6154 | vcpu->arch.hflags &= ~HF_SMM_MASK; | |
6155 | nested_vmcb = nested_svm_map(svm, svm_state_save.vmcb, &page); | |
6156 | if (nested_vmcb) | |
6157 | enter_svm_guest_mode(svm, svm_state_save.vmcb, nested_vmcb, page); | |
6158 | else | |
6159 | ret = 1; | |
6160 | vcpu->arch.hflags |= HF_SMM_MASK; | |
6161 | } | |
6162 | return ret; | |
0234bf88 LP |
6163 | } |
6164 | ||
cc3d967f LP |
6165 | static int enable_smi_window(struct kvm_vcpu *vcpu) |
6166 | { | |
6167 | struct vcpu_svm *svm = to_svm(vcpu); | |
6168 | ||
6169 | if (!gif_set(svm)) { | |
6170 | if (vgif_enabled(svm)) | |
6171 | set_intercept(svm, INTERCEPT_STGI); | |
6172 | /* STGI will cause a vm exit */ | |
6173 | return 1; | |
6174 | } | |
6175 | return 0; | |
6176 | } | |
6177 | ||
1654efcb BS |
6178 | static int sev_asid_new(void) |
6179 | { | |
6180 | int pos; | |
6181 | ||
6182 | /* | |
6183 | * SEV-enabled guest must use asid from min_sev_asid to max_sev_asid. | |
6184 | */ | |
6185 | pos = find_next_zero_bit(sev_asid_bitmap, max_sev_asid, min_sev_asid - 1); | |
6186 | if (pos >= max_sev_asid) | |
6187 | return -EBUSY; | |
6188 | ||
6189 | set_bit(pos, sev_asid_bitmap); | |
6190 | return pos + 1; | |
6191 | } | |
6192 | ||
6193 | static int sev_guest_init(struct kvm *kvm, struct kvm_sev_cmd *argp) | |
6194 | { | |
81811c16 | 6195 | struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info; |
1654efcb BS |
6196 | int asid, ret; |
6197 | ||
6198 | ret = -EBUSY; | |
6199 | asid = sev_asid_new(); | |
6200 | if (asid < 0) | |
6201 | return ret; | |
6202 | ||
6203 | ret = sev_platform_init(&argp->error); | |
6204 | if (ret) | |
6205 | goto e_free; | |
6206 | ||
6207 | sev->active = true; | |
6208 | sev->asid = asid; | |
1e80fdc0 | 6209 | INIT_LIST_HEAD(&sev->regions_list); |
1654efcb BS |
6210 | |
6211 | return 0; | |
6212 | ||
6213 | e_free: | |
6214 | __sev_asid_free(asid); | |
6215 | return ret; | |
6216 | } | |
6217 | ||
59414c98 BS |
6218 | static int sev_bind_asid(struct kvm *kvm, unsigned int handle, int *error) |
6219 | { | |
6220 | struct sev_data_activate *data; | |
6221 | int asid = sev_get_asid(kvm); | |
6222 | int ret; | |
6223 | ||
6224 | wbinvd_on_all_cpus(); | |
6225 | ||
6226 | ret = sev_guest_df_flush(error); | |
6227 | if (ret) | |
6228 | return ret; | |
6229 | ||
6230 | data = kzalloc(sizeof(*data), GFP_KERNEL); | |
6231 | if (!data) | |
6232 | return -ENOMEM; | |
6233 | ||
6234 | /* activate ASID on the given handle */ | |
6235 | data->handle = handle; | |
6236 | data->asid = asid; | |
6237 | ret = sev_guest_activate(data, error); | |
6238 | kfree(data); | |
6239 | ||
6240 | return ret; | |
6241 | } | |
6242 | ||
89c50580 | 6243 | static int __sev_issue_cmd(int fd, int id, void *data, int *error) |
59414c98 BS |
6244 | { |
6245 | struct fd f; | |
6246 | int ret; | |
6247 | ||
6248 | f = fdget(fd); | |
6249 | if (!f.file) | |
6250 | return -EBADF; | |
6251 | ||
6252 | ret = sev_issue_cmd_external_user(f.file, id, data, error); | |
6253 | ||
6254 | fdput(f); | |
6255 | return ret; | |
6256 | } | |
6257 | ||
89c50580 BS |
6258 | static int sev_issue_cmd(struct kvm *kvm, int id, void *data, int *error) |
6259 | { | |
81811c16 | 6260 | struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info; |
89c50580 BS |
6261 | |
6262 | return __sev_issue_cmd(sev->fd, id, data, error); | |
6263 | } | |
6264 | ||
59414c98 BS |
6265 | static int sev_launch_start(struct kvm *kvm, struct kvm_sev_cmd *argp) |
6266 | { | |
81811c16 | 6267 | struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info; |
59414c98 BS |
6268 | struct sev_data_launch_start *start; |
6269 | struct kvm_sev_launch_start params; | |
6270 | void *dh_blob, *session_blob; | |
6271 | int *error = &argp->error; | |
6272 | int ret; | |
6273 | ||
6274 | if (!sev_guest(kvm)) | |
6275 | return -ENOTTY; | |
6276 | ||
6277 | if (copy_from_user(¶ms, (void __user *)(uintptr_t)argp->data, sizeof(params))) | |
6278 | return -EFAULT; | |
6279 | ||
6280 | start = kzalloc(sizeof(*start), GFP_KERNEL); | |
6281 | if (!start) | |
6282 | return -ENOMEM; | |
6283 | ||
6284 | dh_blob = NULL; | |
6285 | if (params.dh_uaddr) { | |
6286 | dh_blob = psp_copy_user_blob(params.dh_uaddr, params.dh_len); | |
6287 | if (IS_ERR(dh_blob)) { | |
6288 | ret = PTR_ERR(dh_blob); | |
6289 | goto e_free; | |
6290 | } | |
6291 | ||
6292 | start->dh_cert_address = __sme_set(__pa(dh_blob)); | |
6293 | start->dh_cert_len = params.dh_len; | |
6294 | } | |
6295 | ||
6296 | session_blob = NULL; | |
6297 | if (params.session_uaddr) { | |
6298 | session_blob = psp_copy_user_blob(params.session_uaddr, params.session_len); | |
6299 | if (IS_ERR(session_blob)) { | |
6300 | ret = PTR_ERR(session_blob); | |
6301 | goto e_free_dh; | |
6302 | } | |
6303 | ||
6304 | start->session_address = __sme_set(__pa(session_blob)); | |
6305 | start->session_len = params.session_len; | |
6306 | } | |
6307 | ||
6308 | start->handle = params.handle; | |
6309 | start->policy = params.policy; | |
6310 | ||
6311 | /* create memory encryption context */ | |
89c50580 | 6312 | ret = __sev_issue_cmd(argp->sev_fd, SEV_CMD_LAUNCH_START, start, error); |
59414c98 BS |
6313 | if (ret) |
6314 | goto e_free_session; | |
6315 | ||
6316 | /* Bind ASID to this guest */ | |
6317 | ret = sev_bind_asid(kvm, start->handle, error); | |
6318 | if (ret) | |
6319 | goto e_free_session; | |
6320 | ||
6321 | /* return handle to userspace */ | |
6322 | params.handle = start->handle; | |
6323 | if (copy_to_user((void __user *)(uintptr_t)argp->data, ¶ms, sizeof(params))) { | |
6324 | sev_unbind_asid(kvm, start->handle); | |
6325 | ret = -EFAULT; | |
6326 | goto e_free_session; | |
6327 | } | |
6328 | ||
6329 | sev->handle = start->handle; | |
6330 | sev->fd = argp->sev_fd; | |
6331 | ||
6332 | e_free_session: | |
6333 | kfree(session_blob); | |
6334 | e_free_dh: | |
6335 | kfree(dh_blob); | |
6336 | e_free: | |
6337 | kfree(start); | |
6338 | return ret; | |
6339 | } | |
6340 | ||
89c50580 BS |
6341 | static int get_num_contig_pages(int idx, struct page **inpages, |
6342 | unsigned long npages) | |
6343 | { | |
6344 | unsigned long paddr, next_paddr; | |
6345 | int i = idx + 1, pages = 1; | |
6346 | ||
6347 | /* find the number of contiguous pages starting from idx */ | |
6348 | paddr = __sme_page_pa(inpages[idx]); | |
6349 | while (i < npages) { | |
6350 | next_paddr = __sme_page_pa(inpages[i++]); | |
6351 | if ((paddr + PAGE_SIZE) == next_paddr) { | |
6352 | pages++; | |
6353 | paddr = next_paddr; | |
6354 | continue; | |
6355 | } | |
6356 | break; | |
6357 | } | |
6358 | ||
6359 | return pages; | |
6360 | } | |
6361 | ||
6362 | static int sev_launch_update_data(struct kvm *kvm, struct kvm_sev_cmd *argp) | |
6363 | { | |
6364 | unsigned long vaddr, vaddr_end, next_vaddr, npages, size; | |
81811c16 | 6365 | struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info; |
89c50580 BS |
6366 | struct kvm_sev_launch_update_data params; |
6367 | struct sev_data_launch_update_data *data; | |
6368 | struct page **inpages; | |
6369 | int i, ret, pages; | |
6370 | ||
6371 | if (!sev_guest(kvm)) | |
6372 | return -ENOTTY; | |
6373 | ||
6374 | if (copy_from_user(¶ms, (void __user *)(uintptr_t)argp->data, sizeof(params))) | |
6375 | return -EFAULT; | |
6376 | ||
6377 | data = kzalloc(sizeof(*data), GFP_KERNEL); | |
6378 | if (!data) | |
6379 | return -ENOMEM; | |
6380 | ||
6381 | vaddr = params.uaddr; | |
6382 | size = params.len; | |
6383 | vaddr_end = vaddr + size; | |
6384 | ||
6385 | /* Lock the user memory. */ | |
6386 | inpages = sev_pin_memory(kvm, vaddr, size, &npages, 1); | |
6387 | if (!inpages) { | |
6388 | ret = -ENOMEM; | |
6389 | goto e_free; | |
6390 | } | |
6391 | ||
6392 | /* | |
6393 | * The LAUNCH_UPDATE command will perform in-place encryption of the | |
6394 | * memory content (i.e it will write the same memory region with C=1). | |
6395 | * It's possible that the cache may contain the data with C=0, i.e., | |
6396 | * unencrypted so invalidate it first. | |
6397 | */ | |
6398 | sev_clflush_pages(inpages, npages); | |
6399 | ||
6400 | for (i = 0; vaddr < vaddr_end; vaddr = next_vaddr, i += pages) { | |
6401 | int offset, len; | |
6402 | ||
6403 | /* | |
6404 | * If the user buffer is not page-aligned, calculate the offset | |
6405 | * within the page. | |
6406 | */ | |
6407 | offset = vaddr & (PAGE_SIZE - 1); | |
6408 | ||
6409 | /* Calculate the number of pages that can be encrypted in one go. */ | |
6410 | pages = get_num_contig_pages(i, inpages, npages); | |
6411 | ||
6412 | len = min_t(size_t, ((pages * PAGE_SIZE) - offset), size); | |
6413 | ||
6414 | data->handle = sev->handle; | |
6415 | data->len = len; | |
6416 | data->address = __sme_page_pa(inpages[i]) + offset; | |
6417 | ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_UPDATE_DATA, data, &argp->error); | |
6418 | if (ret) | |
6419 | goto e_unpin; | |
6420 | ||
6421 | size -= len; | |
6422 | next_vaddr = vaddr + len; | |
6423 | } | |
6424 | ||
6425 | e_unpin: | |
6426 | /* content of memory is updated, mark pages dirty */ | |
6427 | for (i = 0; i < npages; i++) { | |
6428 | set_page_dirty_lock(inpages[i]); | |
6429 | mark_page_accessed(inpages[i]); | |
6430 | } | |
6431 | /* unlock the user pages */ | |
6432 | sev_unpin_memory(kvm, inpages, npages); | |
6433 | e_free: | |
6434 | kfree(data); | |
6435 | return ret; | |
6436 | } | |
6437 | ||
0d0736f7 BS |
6438 | static int sev_launch_measure(struct kvm *kvm, struct kvm_sev_cmd *argp) |
6439 | { | |
3e233385 | 6440 | void __user *measure = (void __user *)(uintptr_t)argp->data; |
81811c16 | 6441 | struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info; |
0d0736f7 BS |
6442 | struct sev_data_launch_measure *data; |
6443 | struct kvm_sev_launch_measure params; | |
3e233385 | 6444 | void __user *p = NULL; |
0d0736f7 BS |
6445 | void *blob = NULL; |
6446 | int ret; | |
6447 | ||
6448 | if (!sev_guest(kvm)) | |
6449 | return -ENOTTY; | |
6450 | ||
3e233385 | 6451 | if (copy_from_user(¶ms, measure, sizeof(params))) |
0d0736f7 BS |
6452 | return -EFAULT; |
6453 | ||
6454 | data = kzalloc(sizeof(*data), GFP_KERNEL); | |
6455 | if (!data) | |
6456 | return -ENOMEM; | |
6457 | ||
6458 | /* User wants to query the blob length */ | |
6459 | if (!params.len) | |
6460 | goto cmd; | |
6461 | ||
3e233385 BS |
6462 | p = (void __user *)(uintptr_t)params.uaddr; |
6463 | if (p) { | |
0d0736f7 BS |
6464 | if (params.len > SEV_FW_BLOB_MAX_SIZE) { |
6465 | ret = -EINVAL; | |
6466 | goto e_free; | |
6467 | } | |
6468 | ||
0d0736f7 BS |
6469 | ret = -ENOMEM; |
6470 | blob = kmalloc(params.len, GFP_KERNEL); | |
6471 | if (!blob) | |
6472 | goto e_free; | |
6473 | ||
6474 | data->address = __psp_pa(blob); | |
6475 | data->len = params.len; | |
6476 | } | |
6477 | ||
6478 | cmd: | |
6479 | data->handle = sev->handle; | |
6480 | ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_MEASURE, data, &argp->error); | |
6481 | ||
6482 | /* | |
6483 | * If we query the session length, FW responded with expected data. | |
6484 | */ | |
6485 | if (!params.len) | |
6486 | goto done; | |
6487 | ||
6488 | if (ret) | |
6489 | goto e_free_blob; | |
6490 | ||
6491 | if (blob) { | |
3e233385 | 6492 | if (copy_to_user(p, blob, params.len)) |
0d0736f7 BS |
6493 | ret = -EFAULT; |
6494 | } | |
6495 | ||
6496 | done: | |
6497 | params.len = data->len; | |
3e233385 | 6498 | if (copy_to_user(measure, ¶ms, sizeof(params))) |
0d0736f7 BS |
6499 | ret = -EFAULT; |
6500 | e_free_blob: | |
6501 | kfree(blob); | |
6502 | e_free: | |
6503 | kfree(data); | |
6504 | return ret; | |
6505 | } | |
6506 | ||
5bdb0e2f BS |
6507 | static int sev_launch_finish(struct kvm *kvm, struct kvm_sev_cmd *argp) |
6508 | { | |
81811c16 | 6509 | struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info; |
5bdb0e2f BS |
6510 | struct sev_data_launch_finish *data; |
6511 | int ret; | |
6512 | ||
6513 | if (!sev_guest(kvm)) | |
6514 | return -ENOTTY; | |
6515 | ||
6516 | data = kzalloc(sizeof(*data), GFP_KERNEL); | |
6517 | if (!data) | |
6518 | return -ENOMEM; | |
6519 | ||
6520 | data->handle = sev->handle; | |
6521 | ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_FINISH, data, &argp->error); | |
6522 | ||
6523 | kfree(data); | |
6524 | return ret; | |
6525 | } | |
6526 | ||
255d9e75 BS |
6527 | static int sev_guest_status(struct kvm *kvm, struct kvm_sev_cmd *argp) |
6528 | { | |
81811c16 | 6529 | struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info; |
255d9e75 BS |
6530 | struct kvm_sev_guest_status params; |
6531 | struct sev_data_guest_status *data; | |
6532 | int ret; | |
6533 | ||
6534 | if (!sev_guest(kvm)) | |
6535 | return -ENOTTY; | |
6536 | ||
6537 | data = kzalloc(sizeof(*data), GFP_KERNEL); | |
6538 | if (!data) | |
6539 | return -ENOMEM; | |
6540 | ||
6541 | data->handle = sev->handle; | |
6542 | ret = sev_issue_cmd(kvm, SEV_CMD_GUEST_STATUS, data, &argp->error); | |
6543 | if (ret) | |
6544 | goto e_free; | |
6545 | ||
6546 | params.policy = data->policy; | |
6547 | params.state = data->state; | |
6548 | params.handle = data->handle; | |
6549 | ||
6550 | if (copy_to_user((void __user *)(uintptr_t)argp->data, ¶ms, sizeof(params))) | |
6551 | ret = -EFAULT; | |
6552 | e_free: | |
6553 | kfree(data); | |
6554 | return ret; | |
6555 | } | |
6556 | ||
24f41fb2 BS |
6557 | static int __sev_issue_dbg_cmd(struct kvm *kvm, unsigned long src, |
6558 | unsigned long dst, int size, | |
6559 | int *error, bool enc) | |
6560 | { | |
81811c16 | 6561 | struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info; |
24f41fb2 BS |
6562 | struct sev_data_dbg *data; |
6563 | int ret; | |
6564 | ||
6565 | data = kzalloc(sizeof(*data), GFP_KERNEL); | |
6566 | if (!data) | |
6567 | return -ENOMEM; | |
6568 | ||
6569 | data->handle = sev->handle; | |
6570 | data->dst_addr = dst; | |
6571 | data->src_addr = src; | |
6572 | data->len = size; | |
6573 | ||
6574 | ret = sev_issue_cmd(kvm, | |
6575 | enc ? SEV_CMD_DBG_ENCRYPT : SEV_CMD_DBG_DECRYPT, | |
6576 | data, error); | |
6577 | kfree(data); | |
6578 | return ret; | |
6579 | } | |
6580 | ||
6581 | static int __sev_dbg_decrypt(struct kvm *kvm, unsigned long src_paddr, | |
6582 | unsigned long dst_paddr, int sz, int *err) | |
6583 | { | |
6584 | int offset; | |
6585 | ||
6586 | /* | |
6587 | * Its safe to read more than we are asked, caller should ensure that | |
6588 | * destination has enough space. | |
6589 | */ | |
6590 | src_paddr = round_down(src_paddr, 16); | |
6591 | offset = src_paddr & 15; | |
6592 | sz = round_up(sz + offset, 16); | |
6593 | ||
6594 | return __sev_issue_dbg_cmd(kvm, src_paddr, dst_paddr, sz, err, false); | |
6595 | } | |
6596 | ||
6597 | static int __sev_dbg_decrypt_user(struct kvm *kvm, unsigned long paddr, | |
6598 | unsigned long __user dst_uaddr, | |
6599 | unsigned long dst_paddr, | |
6600 | int size, int *err) | |
6601 | { | |
6602 | struct page *tpage = NULL; | |
6603 | int ret, offset; | |
6604 | ||
6605 | /* if inputs are not 16-byte then use intermediate buffer */ | |
6606 | if (!IS_ALIGNED(dst_paddr, 16) || | |
6607 | !IS_ALIGNED(paddr, 16) || | |
6608 | !IS_ALIGNED(size, 16)) { | |
6609 | tpage = (void *)alloc_page(GFP_KERNEL); | |
6610 | if (!tpage) | |
6611 | return -ENOMEM; | |
6612 | ||
6613 | dst_paddr = __sme_page_pa(tpage); | |
6614 | } | |
6615 | ||
6616 | ret = __sev_dbg_decrypt(kvm, paddr, dst_paddr, size, err); | |
6617 | if (ret) | |
6618 | goto e_free; | |
6619 | ||
6620 | if (tpage) { | |
6621 | offset = paddr & 15; | |
6622 | if (copy_to_user((void __user *)(uintptr_t)dst_uaddr, | |
6623 | page_address(tpage) + offset, size)) | |
6624 | ret = -EFAULT; | |
6625 | } | |
6626 | ||
6627 | e_free: | |
6628 | if (tpage) | |
6629 | __free_page(tpage); | |
6630 | ||
6631 | return ret; | |
6632 | } | |
6633 | ||
7d1594f5 BS |
6634 | static int __sev_dbg_encrypt_user(struct kvm *kvm, unsigned long paddr, |
6635 | unsigned long __user vaddr, | |
6636 | unsigned long dst_paddr, | |
6637 | unsigned long __user dst_vaddr, | |
6638 | int size, int *error) | |
6639 | { | |
6640 | struct page *src_tpage = NULL; | |
6641 | struct page *dst_tpage = NULL; | |
6642 | int ret, len = size; | |
6643 | ||
6644 | /* If source buffer is not aligned then use an intermediate buffer */ | |
6645 | if (!IS_ALIGNED(vaddr, 16)) { | |
6646 | src_tpage = alloc_page(GFP_KERNEL); | |
6647 | if (!src_tpage) | |
6648 | return -ENOMEM; | |
6649 | ||
6650 | if (copy_from_user(page_address(src_tpage), | |
6651 | (void __user *)(uintptr_t)vaddr, size)) { | |
6652 | __free_page(src_tpage); | |
6653 | return -EFAULT; | |
6654 | } | |
6655 | ||
6656 | paddr = __sme_page_pa(src_tpage); | |
6657 | } | |
6658 | ||
6659 | /* | |
6660 | * If destination buffer or length is not aligned then do read-modify-write: | |
6661 | * - decrypt destination in an intermediate buffer | |
6662 | * - copy the source buffer in an intermediate buffer | |
6663 | * - use the intermediate buffer as source buffer | |
6664 | */ | |
6665 | if (!IS_ALIGNED(dst_vaddr, 16) || !IS_ALIGNED(size, 16)) { | |
6666 | int dst_offset; | |
6667 | ||
6668 | dst_tpage = alloc_page(GFP_KERNEL); | |
6669 | if (!dst_tpage) { | |
6670 | ret = -ENOMEM; | |
6671 | goto e_free; | |
6672 | } | |
6673 | ||
6674 | ret = __sev_dbg_decrypt(kvm, dst_paddr, | |
6675 | __sme_page_pa(dst_tpage), size, error); | |
6676 | if (ret) | |
6677 | goto e_free; | |
6678 | ||
6679 | /* | |
6680 | * If source is kernel buffer then use memcpy() otherwise | |
6681 | * copy_from_user(). | |
6682 | */ | |
6683 | dst_offset = dst_paddr & 15; | |
6684 | ||
6685 | if (src_tpage) | |
6686 | memcpy(page_address(dst_tpage) + dst_offset, | |
6687 | page_address(src_tpage), size); | |
6688 | else { | |
6689 | if (copy_from_user(page_address(dst_tpage) + dst_offset, | |
6690 | (void __user *)(uintptr_t)vaddr, size)) { | |
6691 | ret = -EFAULT; | |
6692 | goto e_free; | |
6693 | } | |
6694 | } | |
6695 | ||
6696 | paddr = __sme_page_pa(dst_tpage); | |
6697 | dst_paddr = round_down(dst_paddr, 16); | |
6698 | len = round_up(size, 16); | |
6699 | } | |
6700 | ||
6701 | ret = __sev_issue_dbg_cmd(kvm, paddr, dst_paddr, len, error, true); | |
6702 | ||
6703 | e_free: | |
6704 | if (src_tpage) | |
6705 | __free_page(src_tpage); | |
6706 | if (dst_tpage) | |
6707 | __free_page(dst_tpage); | |
6708 | return ret; | |
6709 | } | |
6710 | ||
24f41fb2 BS |
6711 | static int sev_dbg_crypt(struct kvm *kvm, struct kvm_sev_cmd *argp, bool dec) |
6712 | { | |
6713 | unsigned long vaddr, vaddr_end, next_vaddr; | |
6714 | unsigned long dst_vaddr, dst_vaddr_end; | |
6715 | struct page **src_p, **dst_p; | |
6716 | struct kvm_sev_dbg debug; | |
6717 | unsigned long n; | |
6718 | int ret, size; | |
6719 | ||
6720 | if (!sev_guest(kvm)) | |
6721 | return -ENOTTY; | |
6722 | ||
6723 | if (copy_from_user(&debug, (void __user *)(uintptr_t)argp->data, sizeof(debug))) | |
6724 | return -EFAULT; | |
6725 | ||
6726 | vaddr = debug.src_uaddr; | |
6727 | size = debug.len; | |
6728 | vaddr_end = vaddr + size; | |
6729 | dst_vaddr = debug.dst_uaddr; | |
6730 | dst_vaddr_end = dst_vaddr + size; | |
6731 | ||
6732 | for (; vaddr < vaddr_end; vaddr = next_vaddr) { | |
6733 | int len, s_off, d_off; | |
6734 | ||
6735 | /* lock userspace source and destination page */ | |
6736 | src_p = sev_pin_memory(kvm, vaddr & PAGE_MASK, PAGE_SIZE, &n, 0); | |
6737 | if (!src_p) | |
6738 | return -EFAULT; | |
6739 | ||
6740 | dst_p = sev_pin_memory(kvm, dst_vaddr & PAGE_MASK, PAGE_SIZE, &n, 1); | |
6741 | if (!dst_p) { | |
6742 | sev_unpin_memory(kvm, src_p, n); | |
6743 | return -EFAULT; | |
6744 | } | |
6745 | ||
6746 | /* | |
6747 | * The DBG_{DE,EN}CRYPT commands will perform {dec,en}cryption of the | |
6748 | * memory content (i.e it will write the same memory region with C=1). | |
6749 | * It's possible that the cache may contain the data with C=0, i.e., | |
6750 | * unencrypted so invalidate it first. | |
6751 | */ | |
6752 | sev_clflush_pages(src_p, 1); | |
6753 | sev_clflush_pages(dst_p, 1); | |
6754 | ||
6755 | /* | |
6756 | * Since user buffer may not be page aligned, calculate the | |
6757 | * offset within the page. | |
6758 | */ | |
6759 | s_off = vaddr & ~PAGE_MASK; | |
6760 | d_off = dst_vaddr & ~PAGE_MASK; | |
6761 | len = min_t(size_t, (PAGE_SIZE - s_off), size); | |
6762 | ||
7d1594f5 BS |
6763 | if (dec) |
6764 | ret = __sev_dbg_decrypt_user(kvm, | |
6765 | __sme_page_pa(src_p[0]) + s_off, | |
6766 | dst_vaddr, | |
6767 | __sme_page_pa(dst_p[0]) + d_off, | |
6768 | len, &argp->error); | |
6769 | else | |
6770 | ret = __sev_dbg_encrypt_user(kvm, | |
6771 | __sme_page_pa(src_p[0]) + s_off, | |
6772 | vaddr, | |
6773 | __sme_page_pa(dst_p[0]) + d_off, | |
6774 | dst_vaddr, | |
6775 | len, &argp->error); | |
24f41fb2 BS |
6776 | |
6777 | sev_unpin_memory(kvm, src_p, 1); | |
6778 | sev_unpin_memory(kvm, dst_p, 1); | |
6779 | ||
6780 | if (ret) | |
6781 | goto err; | |
6782 | ||
6783 | next_vaddr = vaddr + len; | |
6784 | dst_vaddr = dst_vaddr + len; | |
6785 | size -= len; | |
6786 | } | |
6787 | err: | |
6788 | return ret; | |
6789 | } | |
6790 | ||
9f5b5b95 BS |
6791 | static int sev_launch_secret(struct kvm *kvm, struct kvm_sev_cmd *argp) |
6792 | { | |
81811c16 | 6793 | struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info; |
9f5b5b95 BS |
6794 | struct sev_data_launch_secret *data; |
6795 | struct kvm_sev_launch_secret params; | |
6796 | struct page **pages; | |
6797 | void *blob, *hdr; | |
6798 | unsigned long n; | |
9c5e0afa | 6799 | int ret, offset; |
9f5b5b95 BS |
6800 | |
6801 | if (!sev_guest(kvm)) | |
6802 | return -ENOTTY; | |
6803 | ||
6804 | if (copy_from_user(¶ms, (void __user *)(uintptr_t)argp->data, sizeof(params))) | |
6805 | return -EFAULT; | |
6806 | ||
6807 | pages = sev_pin_memory(kvm, params.guest_uaddr, params.guest_len, &n, 1); | |
6808 | if (!pages) | |
6809 | return -ENOMEM; | |
6810 | ||
6811 | /* | |
6812 | * The secret must be copied into contiguous memory region, lets verify | |
6813 | * that userspace memory pages are contiguous before we issue command. | |
6814 | */ | |
6815 | if (get_num_contig_pages(0, pages, n) != n) { | |
6816 | ret = -EINVAL; | |
6817 | goto e_unpin_memory; | |
6818 | } | |
6819 | ||
6820 | ret = -ENOMEM; | |
6821 | data = kzalloc(sizeof(*data), GFP_KERNEL); | |
6822 | if (!data) | |
6823 | goto e_unpin_memory; | |
6824 | ||
9c5e0afa BS |
6825 | offset = params.guest_uaddr & (PAGE_SIZE - 1); |
6826 | data->guest_address = __sme_page_pa(pages[0]) + offset; | |
6827 | data->guest_len = params.guest_len; | |
6828 | ||
9f5b5b95 BS |
6829 | blob = psp_copy_user_blob(params.trans_uaddr, params.trans_len); |
6830 | if (IS_ERR(blob)) { | |
6831 | ret = PTR_ERR(blob); | |
6832 | goto e_free; | |
6833 | } | |
6834 | ||
6835 | data->trans_address = __psp_pa(blob); | |
6836 | data->trans_len = params.trans_len; | |
6837 | ||
6838 | hdr = psp_copy_user_blob(params.hdr_uaddr, params.hdr_len); | |
6839 | if (IS_ERR(hdr)) { | |
6840 | ret = PTR_ERR(hdr); | |
6841 | goto e_free_blob; | |
6842 | } | |
9c5e0afa BS |
6843 | data->hdr_address = __psp_pa(hdr); |
6844 | data->hdr_len = params.hdr_len; | |
9f5b5b95 BS |
6845 | |
6846 | data->handle = sev->handle; | |
6847 | ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_UPDATE_SECRET, data, &argp->error); | |
6848 | ||
6849 | kfree(hdr); | |
6850 | ||
6851 | e_free_blob: | |
6852 | kfree(blob); | |
6853 | e_free: | |
6854 | kfree(data); | |
6855 | e_unpin_memory: | |
6856 | sev_unpin_memory(kvm, pages, n); | |
6857 | return ret; | |
6858 | } | |
6859 | ||
1654efcb BS |
6860 | static int svm_mem_enc_op(struct kvm *kvm, void __user *argp) |
6861 | { | |
6862 | struct kvm_sev_cmd sev_cmd; | |
6863 | int r; | |
6864 | ||
6865 | if (!svm_sev_enabled()) | |
6866 | return -ENOTTY; | |
6867 | ||
6868 | if (copy_from_user(&sev_cmd, argp, sizeof(struct kvm_sev_cmd))) | |
6869 | return -EFAULT; | |
6870 | ||
6871 | mutex_lock(&kvm->lock); | |
6872 | ||
6873 | switch (sev_cmd.id) { | |
6874 | case KVM_SEV_INIT: | |
6875 | r = sev_guest_init(kvm, &sev_cmd); | |
6876 | break; | |
59414c98 BS |
6877 | case KVM_SEV_LAUNCH_START: |
6878 | r = sev_launch_start(kvm, &sev_cmd); | |
6879 | break; | |
89c50580 BS |
6880 | case KVM_SEV_LAUNCH_UPDATE_DATA: |
6881 | r = sev_launch_update_data(kvm, &sev_cmd); | |
6882 | break; | |
0d0736f7 BS |
6883 | case KVM_SEV_LAUNCH_MEASURE: |
6884 | r = sev_launch_measure(kvm, &sev_cmd); | |
6885 | break; | |
5bdb0e2f BS |
6886 | case KVM_SEV_LAUNCH_FINISH: |
6887 | r = sev_launch_finish(kvm, &sev_cmd); | |
6888 | break; | |
255d9e75 BS |
6889 | case KVM_SEV_GUEST_STATUS: |
6890 | r = sev_guest_status(kvm, &sev_cmd); | |
6891 | break; | |
24f41fb2 BS |
6892 | case KVM_SEV_DBG_DECRYPT: |
6893 | r = sev_dbg_crypt(kvm, &sev_cmd, true); | |
6894 | break; | |
7d1594f5 BS |
6895 | case KVM_SEV_DBG_ENCRYPT: |
6896 | r = sev_dbg_crypt(kvm, &sev_cmd, false); | |
6897 | break; | |
9f5b5b95 BS |
6898 | case KVM_SEV_LAUNCH_SECRET: |
6899 | r = sev_launch_secret(kvm, &sev_cmd); | |
6900 | break; | |
1654efcb BS |
6901 | default: |
6902 | r = -EINVAL; | |
6903 | goto out; | |
6904 | } | |
6905 | ||
6906 | if (copy_to_user(argp, &sev_cmd, sizeof(struct kvm_sev_cmd))) | |
6907 | r = -EFAULT; | |
6908 | ||
6909 | out: | |
6910 | mutex_unlock(&kvm->lock); | |
6911 | return r; | |
6912 | } | |
6913 | ||
1e80fdc0 BS |
6914 | static int svm_register_enc_region(struct kvm *kvm, |
6915 | struct kvm_enc_region *range) | |
6916 | { | |
81811c16 | 6917 | struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info; |
1e80fdc0 BS |
6918 | struct enc_region *region; |
6919 | int ret = 0; | |
6920 | ||
6921 | if (!sev_guest(kvm)) | |
6922 | return -ENOTTY; | |
6923 | ||
6924 | region = kzalloc(sizeof(*region), GFP_KERNEL); | |
6925 | if (!region) | |
6926 | return -ENOMEM; | |
6927 | ||
6928 | region->pages = sev_pin_memory(kvm, range->addr, range->size, ®ion->npages, 1); | |
6929 | if (!region->pages) { | |
6930 | ret = -ENOMEM; | |
6931 | goto e_free; | |
6932 | } | |
6933 | ||
6934 | /* | |
6935 | * The guest may change the memory encryption attribute from C=0 -> C=1 | |
6936 | * or vice versa for this memory range. Lets make sure caches are | |
6937 | * flushed to ensure that guest data gets written into memory with | |
6938 | * correct C-bit. | |
6939 | */ | |
6940 | sev_clflush_pages(region->pages, region->npages); | |
6941 | ||
6942 | region->uaddr = range->addr; | |
6943 | region->size = range->size; | |
6944 | ||
6945 | mutex_lock(&kvm->lock); | |
6946 | list_add_tail(®ion->list, &sev->regions_list); | |
6947 | mutex_unlock(&kvm->lock); | |
6948 | ||
6949 | return ret; | |
6950 | ||
6951 | e_free: | |
6952 | kfree(region); | |
6953 | return ret; | |
6954 | } | |
6955 | ||
6956 | static struct enc_region * | |
6957 | find_enc_region(struct kvm *kvm, struct kvm_enc_region *range) | |
6958 | { | |
81811c16 | 6959 | struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info; |
1e80fdc0 BS |
6960 | struct list_head *head = &sev->regions_list; |
6961 | struct enc_region *i; | |
6962 | ||
6963 | list_for_each_entry(i, head, list) { | |
6964 | if (i->uaddr == range->addr && | |
6965 | i->size == range->size) | |
6966 | return i; | |
6967 | } | |
6968 | ||
6969 | return NULL; | |
6970 | } | |
6971 | ||
6972 | ||
6973 | static int svm_unregister_enc_region(struct kvm *kvm, | |
6974 | struct kvm_enc_region *range) | |
6975 | { | |
6976 | struct enc_region *region; | |
6977 | int ret; | |
6978 | ||
6979 | mutex_lock(&kvm->lock); | |
6980 | ||
6981 | if (!sev_guest(kvm)) { | |
6982 | ret = -ENOTTY; | |
6983 | goto failed; | |
6984 | } | |
6985 | ||
6986 | region = find_enc_region(kvm, range); | |
6987 | if (!region) { | |
6988 | ret = -EINVAL; | |
6989 | goto failed; | |
6990 | } | |
6991 | ||
6992 | __unregister_enc_region_locked(kvm, region); | |
6993 | ||
6994 | mutex_unlock(&kvm->lock); | |
6995 | return 0; | |
6996 | ||
6997 | failed: | |
6998 | mutex_unlock(&kvm->lock); | |
6999 | return ret; | |
7000 | } | |
7001 | ||
404f6aac | 7002 | static struct kvm_x86_ops svm_x86_ops __ro_after_init = { |
6aa8b732 AK |
7003 | .cpu_has_kvm_support = has_svm, |
7004 | .disabled_by_bios = is_disabled, | |
7005 | .hardware_setup = svm_hardware_setup, | |
7006 | .hardware_unsetup = svm_hardware_unsetup, | |
002c7f7c | 7007 | .check_processor_compatibility = svm_check_processor_compat, |
6aa8b732 AK |
7008 | .hardware_enable = svm_hardware_enable, |
7009 | .hardware_disable = svm_hardware_disable, | |
774ead3a | 7010 | .cpu_has_accelerated_tpr = svm_cpu_has_accelerated_tpr, |
6d396b55 | 7011 | .cpu_has_high_real_mode_segbase = svm_has_high_real_mode_segbase, |
6aa8b732 AK |
7012 | |
7013 | .vcpu_create = svm_create_vcpu, | |
7014 | .vcpu_free = svm_free_vcpu, | |
04d2cc77 | 7015 | .vcpu_reset = svm_vcpu_reset, |
6aa8b732 | 7016 | |
434a1e94 SC |
7017 | .vm_alloc = svm_vm_alloc, |
7018 | .vm_free = svm_vm_free, | |
44a95dae | 7019 | .vm_init = avic_vm_init, |
1654efcb | 7020 | .vm_destroy = svm_vm_destroy, |
44a95dae | 7021 | |
04d2cc77 | 7022 | .prepare_guest_switch = svm_prepare_guest_switch, |
6aa8b732 AK |
7023 | .vcpu_load = svm_vcpu_load, |
7024 | .vcpu_put = svm_vcpu_put, | |
8221c137 SS |
7025 | .vcpu_blocking = svm_vcpu_blocking, |
7026 | .vcpu_unblocking = svm_vcpu_unblocking, | |
6aa8b732 | 7027 | |
a96036b8 | 7028 | .update_bp_intercept = update_bp_intercept, |
801e459a | 7029 | .get_msr_feature = svm_get_msr_feature, |
6aa8b732 AK |
7030 | .get_msr = svm_get_msr, |
7031 | .set_msr = svm_set_msr, | |
7032 | .get_segment_base = svm_get_segment_base, | |
7033 | .get_segment = svm_get_segment, | |
7034 | .set_segment = svm_set_segment, | |
2e4d2653 | 7035 | .get_cpl = svm_get_cpl, |
1747fb71 | 7036 | .get_cs_db_l_bits = kvm_get_cs_db_l_bits, |
e8467fda | 7037 | .decache_cr0_guest_bits = svm_decache_cr0_guest_bits, |
aff48baa | 7038 | .decache_cr3 = svm_decache_cr3, |
25c4c276 | 7039 | .decache_cr4_guest_bits = svm_decache_cr4_guest_bits, |
6aa8b732 | 7040 | .set_cr0 = svm_set_cr0, |
6aa8b732 AK |
7041 | .set_cr3 = svm_set_cr3, |
7042 | .set_cr4 = svm_set_cr4, | |
7043 | .set_efer = svm_set_efer, | |
7044 | .get_idt = svm_get_idt, | |
7045 | .set_idt = svm_set_idt, | |
7046 | .get_gdt = svm_get_gdt, | |
7047 | .set_gdt = svm_set_gdt, | |
73aaf249 JK |
7048 | .get_dr6 = svm_get_dr6, |
7049 | .set_dr6 = svm_set_dr6, | |
020df079 | 7050 | .set_dr7 = svm_set_dr7, |
facb0139 | 7051 | .sync_dirty_debug_regs = svm_sync_dirty_debug_regs, |
6de4f3ad | 7052 | .cache_reg = svm_cache_reg, |
6aa8b732 AK |
7053 | .get_rflags = svm_get_rflags, |
7054 | .set_rflags = svm_set_rflags, | |
be94f6b7 | 7055 | |
6aa8b732 | 7056 | .tlb_flush = svm_flush_tlb, |
6aa8b732 | 7057 | |
6aa8b732 | 7058 | .run = svm_vcpu_run, |
04d2cc77 | 7059 | .handle_exit = handle_exit, |
6aa8b732 | 7060 | .skip_emulated_instruction = skip_emulated_instruction, |
2809f5d2 GC |
7061 | .set_interrupt_shadow = svm_set_interrupt_shadow, |
7062 | .get_interrupt_shadow = svm_get_interrupt_shadow, | |
102d8325 | 7063 | .patch_hypercall = svm_patch_hypercall, |
2a8067f1 | 7064 | .set_irq = svm_set_irq, |
95ba8273 | 7065 | .set_nmi = svm_inject_nmi, |
298101da | 7066 | .queue_exception = svm_queue_exception, |
b463a6f7 | 7067 | .cancel_injection = svm_cancel_injection, |
78646121 | 7068 | .interrupt_allowed = svm_interrupt_allowed, |
95ba8273 | 7069 | .nmi_allowed = svm_nmi_allowed, |
3cfc3092 JK |
7070 | .get_nmi_mask = svm_get_nmi_mask, |
7071 | .set_nmi_mask = svm_set_nmi_mask, | |
95ba8273 GN |
7072 | .enable_nmi_window = enable_nmi_window, |
7073 | .enable_irq_window = enable_irq_window, | |
7074 | .update_cr8_intercept = update_cr8_intercept, | |
8d14695f | 7075 | .set_virtual_x2apic_mode = svm_set_virtual_x2apic_mode, |
d62caabb AS |
7076 | .get_enable_apicv = svm_get_enable_apicv, |
7077 | .refresh_apicv_exec_ctrl = svm_refresh_apicv_exec_ctrl, | |
c7c9c56c | 7078 | .load_eoi_exitmap = svm_load_eoi_exitmap, |
44a95dae SS |
7079 | .hwapic_irr_update = svm_hwapic_irr_update, |
7080 | .hwapic_isr_update = svm_hwapic_isr_update, | |
fa59cc00 | 7081 | .sync_pir_to_irr = kvm_lapic_find_highest_irr, |
be8ca170 | 7082 | .apicv_post_state_restore = avic_post_state_restore, |
cbc94022 IE |
7083 | |
7084 | .set_tss_addr = svm_set_tss_addr, | |
2ac52ab8 | 7085 | .set_identity_map_addr = svm_set_identity_map_addr, |
67253af5 | 7086 | .get_tdp_level = get_npt_level, |
4b12f0de | 7087 | .get_mt_mask = svm_get_mt_mask, |
229456fc | 7088 | |
586f9607 | 7089 | .get_exit_info = svm_get_exit_info, |
586f9607 | 7090 | |
17cc3935 | 7091 | .get_lpage_level = svm_get_lpage_level, |
0e851880 SY |
7092 | |
7093 | .cpuid_update = svm_cpuid_update, | |
4e47c7a6 SY |
7094 | |
7095 | .rdtscp_supported = svm_rdtscp_supported, | |
ad756a16 | 7096 | .invpcid_supported = svm_invpcid_supported, |
93c4adc7 | 7097 | .mpx_supported = svm_mpx_supported, |
55412b2e | 7098 | .xsaves_supported = svm_xsaves_supported, |
66336cab | 7099 | .umip_emulated = svm_umip_emulated, |
d4330ef2 JR |
7100 | |
7101 | .set_supported_cpuid = svm_set_supported_cpuid, | |
f5f48ee1 SY |
7102 | |
7103 | .has_wbinvd_exit = svm_has_wbinvd_exit, | |
99e3e30a ZA |
7104 | |
7105 | .write_tsc_offset = svm_write_tsc_offset, | |
1c97f0a0 JR |
7106 | |
7107 | .set_tdp_cr3 = set_tdp_cr3, | |
8a76d7f2 JR |
7108 | |
7109 | .check_intercept = svm_check_intercept, | |
a547c6db | 7110 | .handle_external_intr = svm_handle_external_intr, |
ae97a3b8 RK |
7111 | |
7112 | .sched_in = svm_sched_in, | |
25462f7f WH |
7113 | |
7114 | .pmu_ops = &amd_pmu_ops, | |
340d3bc3 | 7115 | .deliver_posted_interrupt = svm_deliver_avic_intr, |
411b44ba | 7116 | .update_pi_irte = svm_update_pi_irte, |
74f16909 | 7117 | .setup_mce = svm_setup_mce, |
0234bf88 | 7118 | |
72d7b374 | 7119 | .smi_allowed = svm_smi_allowed, |
0234bf88 LP |
7120 | .pre_enter_smm = svm_pre_enter_smm, |
7121 | .pre_leave_smm = svm_pre_leave_smm, | |
cc3d967f | 7122 | .enable_smi_window = enable_smi_window, |
1654efcb BS |
7123 | |
7124 | .mem_enc_op = svm_mem_enc_op, | |
1e80fdc0 BS |
7125 | .mem_enc_reg_region = svm_register_enc_region, |
7126 | .mem_enc_unreg_region = svm_unregister_enc_region, | |
6aa8b732 AK |
7127 | }; |
7128 | ||
7129 | static int __init svm_init(void) | |
7130 | { | |
cb498ea2 | 7131 | return kvm_init(&svm_x86_ops, sizeof(struct vcpu_svm), |
0ee75bea | 7132 | __alignof__(struct vcpu_svm), THIS_MODULE); |
6aa8b732 AK |
7133 | } |
7134 | ||
7135 | static void __exit svm_exit(void) | |
7136 | { | |
cb498ea2 | 7137 | kvm_exit(); |
6aa8b732 AK |
7138 | } |
7139 | ||
7140 | module_init(svm_init) | |
7141 | module_exit(svm_exit) |