]>
Commit | Line | Data |
---|---|---|
043405e1 CO |
1 | /* |
2 | * Kernel-based Virtual Machine driver for Linux | |
3 | * | |
4 | * derived from drivers/kvm/kvm_main.c | |
5 | * | |
6 | * Copyright (C) 2006 Qumranet, Inc. | |
4d5c5d0f BAY |
7 | * Copyright (C) 2008 Qumranet, Inc. |
8 | * Copyright IBM Corporation, 2008 | |
9611c187 | 9 | * Copyright 2010 Red Hat, Inc. and/or its affiliates. |
043405e1 CO |
10 | * |
11 | * Authors: | |
12 | * Avi Kivity <avi@qumranet.com> | |
13 | * Yaniv Kamay <yaniv@qumranet.com> | |
4d5c5d0f BAY |
14 | * Amit Shah <amit.shah@qumranet.com> |
15 | * Ben-Ami Yassour <benami@il.ibm.com> | |
043405e1 CO |
16 | * |
17 | * This work is licensed under the terms of the GNU GPL, version 2. See | |
18 | * the COPYING file in the top-level directory. | |
19 | * | |
20 | */ | |
21 | ||
edf88417 | 22 | #include <linux/kvm_host.h> |
313a3dc7 | 23 | #include "irq.h" |
1d737c8a | 24 | #include "mmu.h" |
7837699f | 25 | #include "i8254.h" |
37817f29 | 26 | #include "tss.h" |
5fdbf976 | 27 | #include "kvm_cache_regs.h" |
26eef70c | 28 | #include "x86.h" |
00b27a3e | 29 | #include "cpuid.h" |
c9eab58f | 30 | #include "assigned-dev.h" |
474a5bb9 | 31 | #include "pmu.h" |
e83d5887 | 32 | #include "hyperv.h" |
313a3dc7 | 33 | |
18068523 | 34 | #include <linux/clocksource.h> |
4d5c5d0f | 35 | #include <linux/interrupt.h> |
313a3dc7 CO |
36 | #include <linux/kvm.h> |
37 | #include <linux/fs.h> | |
38 | #include <linux/vmalloc.h> | |
5fb76f9b | 39 | #include <linux/module.h> |
0de10343 | 40 | #include <linux/mman.h> |
2bacc55c | 41 | #include <linux/highmem.h> |
19de40a8 | 42 | #include <linux/iommu.h> |
62c476c7 | 43 | #include <linux/intel-iommu.h> |
c8076604 | 44 | #include <linux/cpufreq.h> |
18863bdd | 45 | #include <linux/user-return-notifier.h> |
a983fb23 | 46 | #include <linux/srcu.h> |
5a0e3ad6 | 47 | #include <linux/slab.h> |
ff9d07a0 | 48 | #include <linux/perf_event.h> |
7bee342a | 49 | #include <linux/uaccess.h> |
af585b92 | 50 | #include <linux/hash.h> |
a1b60c1c | 51 | #include <linux/pci.h> |
16e8d74d MT |
52 | #include <linux/timekeeper_internal.h> |
53 | #include <linux/pvclock_gtod.h> | |
87276880 FW |
54 | #include <linux/kvm_irqfd.h> |
55 | #include <linux/irqbypass.h> | |
aec51dc4 | 56 | #include <trace/events/kvm.h> |
2ed152af | 57 | |
229456fc MT |
58 | #define CREATE_TRACE_POINTS |
59 | #include "trace.h" | |
043405e1 | 60 | |
24f1e32c | 61 | #include <asm/debugreg.h> |
d825ed0a | 62 | #include <asm/msr.h> |
a5f61300 | 63 | #include <asm/desc.h> |
890ca9ae | 64 | #include <asm/mce.h> |
f89e32e0 | 65 | #include <linux/kernel_stat.h> |
78f7f1e5 | 66 | #include <asm/fpu/internal.h> /* Ugh! */ |
1d5f066e | 67 | #include <asm/pvclock.h> |
217fc9cf | 68 | #include <asm/div64.h> |
efc64404 | 69 | #include <asm/irq_remapping.h> |
043405e1 | 70 | |
313a3dc7 | 71 | #define MAX_IO_MSRS 256 |
890ca9ae | 72 | #define KVM_MAX_MCE_BANKS 32 |
5854dbca | 73 | #define KVM_MCE_CAP_SUPPORTED (MCG_CTL_P | MCG_SER_P) |
890ca9ae | 74 | |
0f65dd70 AK |
75 | #define emul_to_vcpu(ctxt) \ |
76 | container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt) | |
77 | ||
50a37eb4 JR |
78 | /* EFER defaults: |
79 | * - enable syscall per default because its emulated by KVM | |
80 | * - enable LME and LMA per default on 64 bit KVM | |
81 | */ | |
82 | #ifdef CONFIG_X86_64 | |
1260edbe LJ |
83 | static |
84 | u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA)); | |
50a37eb4 | 85 | #else |
1260edbe | 86 | static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE); |
50a37eb4 | 87 | #endif |
313a3dc7 | 88 | |
ba1389b7 AK |
89 | #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM |
90 | #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU | |
417bc304 | 91 | |
cb142eb7 | 92 | static void update_cr8_intercept(struct kvm_vcpu *vcpu); |
7460fb4a | 93 | static void process_nmi(struct kvm_vcpu *vcpu); |
6addfc42 | 94 | static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags); |
674eea0f | 95 | |
893590c7 | 96 | struct kvm_x86_ops *kvm_x86_ops __read_mostly; |
5fdbf976 | 97 | EXPORT_SYMBOL_GPL(kvm_x86_ops); |
97896d04 | 98 | |
893590c7 | 99 | static bool __read_mostly ignore_msrs = 0; |
476bc001 | 100 | module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR); |
ed85c068 | 101 | |
9ed96e87 MT |
102 | unsigned int min_timer_period_us = 500; |
103 | module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR); | |
104 | ||
630994b3 MT |
105 | static bool __read_mostly kvmclock_periodic_sync = true; |
106 | module_param(kvmclock_periodic_sync, bool, S_IRUGO); | |
107 | ||
893590c7 | 108 | bool __read_mostly kvm_has_tsc_control; |
92a1f12d | 109 | EXPORT_SYMBOL_GPL(kvm_has_tsc_control); |
893590c7 | 110 | u32 __read_mostly kvm_max_guest_tsc_khz; |
92a1f12d | 111 | EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz); |
bc9b961b HZ |
112 | u8 __read_mostly kvm_tsc_scaling_ratio_frac_bits; |
113 | EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits); | |
114 | u64 __read_mostly kvm_max_tsc_scaling_ratio; | |
115 | EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio); | |
ad721883 | 116 | static u64 __read_mostly kvm_default_tsc_scaling_ratio; |
92a1f12d | 117 | |
cc578287 | 118 | /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */ |
893590c7 | 119 | static u32 __read_mostly tsc_tolerance_ppm = 250; |
cc578287 ZA |
120 | module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR); |
121 | ||
d0659d94 | 122 | /* lapic timer advance (tscdeadline mode only) in nanoseconds */ |
893590c7 | 123 | unsigned int __read_mostly lapic_timer_advance_ns = 0; |
d0659d94 MT |
124 | module_param(lapic_timer_advance_ns, uint, S_IRUGO | S_IWUSR); |
125 | ||
52004014 FW |
126 | static bool __read_mostly vector_hashing = true; |
127 | module_param(vector_hashing, bool, S_IRUGO); | |
128 | ||
893590c7 | 129 | static bool __read_mostly backwards_tsc_observed = false; |
16a96021 | 130 | |
18863bdd AK |
131 | #define KVM_NR_SHARED_MSRS 16 |
132 | ||
133 | struct kvm_shared_msrs_global { | |
134 | int nr; | |
2bf78fa7 | 135 | u32 msrs[KVM_NR_SHARED_MSRS]; |
18863bdd AK |
136 | }; |
137 | ||
138 | struct kvm_shared_msrs { | |
139 | struct user_return_notifier urn; | |
140 | bool registered; | |
2bf78fa7 SY |
141 | struct kvm_shared_msr_values { |
142 | u64 host; | |
143 | u64 curr; | |
144 | } values[KVM_NR_SHARED_MSRS]; | |
18863bdd AK |
145 | }; |
146 | ||
147 | static struct kvm_shared_msrs_global __read_mostly shared_msrs_global; | |
013f6a5d | 148 | static struct kvm_shared_msrs __percpu *shared_msrs; |
18863bdd | 149 | |
417bc304 | 150 | struct kvm_stats_debugfs_item debugfs_entries[] = { |
ba1389b7 AK |
151 | { "pf_fixed", VCPU_STAT(pf_fixed) }, |
152 | { "pf_guest", VCPU_STAT(pf_guest) }, | |
153 | { "tlb_flush", VCPU_STAT(tlb_flush) }, | |
154 | { "invlpg", VCPU_STAT(invlpg) }, | |
155 | { "exits", VCPU_STAT(exits) }, | |
156 | { "io_exits", VCPU_STAT(io_exits) }, | |
157 | { "mmio_exits", VCPU_STAT(mmio_exits) }, | |
158 | { "signal_exits", VCPU_STAT(signal_exits) }, | |
159 | { "irq_window", VCPU_STAT(irq_window_exits) }, | |
f08864b4 | 160 | { "nmi_window", VCPU_STAT(nmi_window_exits) }, |
ba1389b7 | 161 | { "halt_exits", VCPU_STAT(halt_exits) }, |
f7819512 | 162 | { "halt_successful_poll", VCPU_STAT(halt_successful_poll) }, |
62bea5bf | 163 | { "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) }, |
ba1389b7 | 164 | { "halt_wakeup", VCPU_STAT(halt_wakeup) }, |
f11c3a8d | 165 | { "hypercalls", VCPU_STAT(hypercalls) }, |
ba1389b7 AK |
166 | { "request_irq", VCPU_STAT(request_irq_exits) }, |
167 | { "irq_exits", VCPU_STAT(irq_exits) }, | |
168 | { "host_state_reload", VCPU_STAT(host_state_reload) }, | |
169 | { "efer_reload", VCPU_STAT(efer_reload) }, | |
170 | { "fpu_reload", VCPU_STAT(fpu_reload) }, | |
171 | { "insn_emulation", VCPU_STAT(insn_emulation) }, | |
172 | { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) }, | |
fa89a817 | 173 | { "irq_injections", VCPU_STAT(irq_injections) }, |
c4abb7c9 | 174 | { "nmi_injections", VCPU_STAT(nmi_injections) }, |
4cee5764 AK |
175 | { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) }, |
176 | { "mmu_pte_write", VM_STAT(mmu_pte_write) }, | |
177 | { "mmu_pte_updated", VM_STAT(mmu_pte_updated) }, | |
178 | { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) }, | |
179 | { "mmu_flooded", VM_STAT(mmu_flooded) }, | |
180 | { "mmu_recycled", VM_STAT(mmu_recycled) }, | |
dfc5aa00 | 181 | { "mmu_cache_miss", VM_STAT(mmu_cache_miss) }, |
4731d4c7 | 182 | { "mmu_unsync", VM_STAT(mmu_unsync) }, |
0f74a24c | 183 | { "remote_tlb_flush", VM_STAT(remote_tlb_flush) }, |
05da4558 | 184 | { "largepages", VM_STAT(lpages) }, |
417bc304 HB |
185 | { NULL } |
186 | }; | |
187 | ||
2acf923e DC |
188 | u64 __read_mostly host_xcr0; |
189 | ||
b6785def | 190 | static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt); |
d6aa1000 | 191 | |
af585b92 GN |
192 | static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu) |
193 | { | |
194 | int i; | |
195 | for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++) | |
196 | vcpu->arch.apf.gfns[i] = ~0; | |
197 | } | |
198 | ||
18863bdd AK |
199 | static void kvm_on_user_return(struct user_return_notifier *urn) |
200 | { | |
201 | unsigned slot; | |
18863bdd AK |
202 | struct kvm_shared_msrs *locals |
203 | = container_of(urn, struct kvm_shared_msrs, urn); | |
2bf78fa7 | 204 | struct kvm_shared_msr_values *values; |
18863bdd AK |
205 | |
206 | for (slot = 0; slot < shared_msrs_global.nr; ++slot) { | |
2bf78fa7 SY |
207 | values = &locals->values[slot]; |
208 | if (values->host != values->curr) { | |
209 | wrmsrl(shared_msrs_global.msrs[slot], values->host); | |
210 | values->curr = values->host; | |
18863bdd AK |
211 | } |
212 | } | |
213 | locals->registered = false; | |
214 | user_return_notifier_unregister(urn); | |
215 | } | |
216 | ||
2bf78fa7 | 217 | static void shared_msr_update(unsigned slot, u32 msr) |
18863bdd | 218 | { |
18863bdd | 219 | u64 value; |
013f6a5d MT |
220 | unsigned int cpu = smp_processor_id(); |
221 | struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu); | |
18863bdd | 222 | |
2bf78fa7 SY |
223 | /* only read, and nobody should modify it at this time, |
224 | * so don't need lock */ | |
225 | if (slot >= shared_msrs_global.nr) { | |
226 | printk(KERN_ERR "kvm: invalid MSR slot!"); | |
227 | return; | |
228 | } | |
229 | rdmsrl_safe(msr, &value); | |
230 | smsr->values[slot].host = value; | |
231 | smsr->values[slot].curr = value; | |
232 | } | |
233 | ||
234 | void kvm_define_shared_msr(unsigned slot, u32 msr) | |
235 | { | |
0123be42 | 236 | BUG_ON(slot >= KVM_NR_SHARED_MSRS); |
c847fe88 | 237 | shared_msrs_global.msrs[slot] = msr; |
18863bdd AK |
238 | if (slot >= shared_msrs_global.nr) |
239 | shared_msrs_global.nr = slot + 1; | |
18863bdd AK |
240 | } |
241 | EXPORT_SYMBOL_GPL(kvm_define_shared_msr); | |
242 | ||
243 | static void kvm_shared_msr_cpu_online(void) | |
244 | { | |
245 | unsigned i; | |
18863bdd AK |
246 | |
247 | for (i = 0; i < shared_msrs_global.nr; ++i) | |
2bf78fa7 | 248 | shared_msr_update(i, shared_msrs_global.msrs[i]); |
18863bdd AK |
249 | } |
250 | ||
8b3c3104 | 251 | int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask) |
18863bdd | 252 | { |
013f6a5d MT |
253 | unsigned int cpu = smp_processor_id(); |
254 | struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu); | |
8b3c3104 | 255 | int err; |
18863bdd | 256 | |
2bf78fa7 | 257 | if (((value ^ smsr->values[slot].curr) & mask) == 0) |
8b3c3104 | 258 | return 0; |
2bf78fa7 | 259 | smsr->values[slot].curr = value; |
8b3c3104 AH |
260 | err = wrmsrl_safe(shared_msrs_global.msrs[slot], value); |
261 | if (err) | |
262 | return 1; | |
263 | ||
18863bdd AK |
264 | if (!smsr->registered) { |
265 | smsr->urn.on_user_return = kvm_on_user_return; | |
266 | user_return_notifier_register(&smsr->urn); | |
267 | smsr->registered = true; | |
268 | } | |
8b3c3104 | 269 | return 0; |
18863bdd AK |
270 | } |
271 | EXPORT_SYMBOL_GPL(kvm_set_shared_msr); | |
272 | ||
13a34e06 | 273 | static void drop_user_return_notifiers(void) |
3548bab5 | 274 | { |
013f6a5d MT |
275 | unsigned int cpu = smp_processor_id(); |
276 | struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu); | |
3548bab5 AK |
277 | |
278 | if (smsr->registered) | |
279 | kvm_on_user_return(&smsr->urn); | |
280 | } | |
281 | ||
6866b83e CO |
282 | u64 kvm_get_apic_base(struct kvm_vcpu *vcpu) |
283 | { | |
8a5a87d9 | 284 | return vcpu->arch.apic_base; |
6866b83e CO |
285 | } |
286 | EXPORT_SYMBOL_GPL(kvm_get_apic_base); | |
287 | ||
58cb628d JK |
288 | int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info) |
289 | { | |
290 | u64 old_state = vcpu->arch.apic_base & | |
291 | (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE); | |
292 | u64 new_state = msr_info->data & | |
293 | (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE); | |
294 | u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) | | |
295 | 0x2ff | (guest_cpuid_has_x2apic(vcpu) ? 0 : X2APIC_ENABLE); | |
296 | ||
297 | if (!msr_info->host_initiated && | |
298 | ((msr_info->data & reserved_bits) != 0 || | |
299 | new_state == X2APIC_ENABLE || | |
300 | (new_state == MSR_IA32_APICBASE_ENABLE && | |
301 | old_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE)) || | |
302 | (new_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE) && | |
303 | old_state == 0))) | |
304 | return 1; | |
305 | ||
306 | kvm_lapic_set_base(vcpu, msr_info->data); | |
307 | return 0; | |
6866b83e CO |
308 | } |
309 | EXPORT_SYMBOL_GPL(kvm_set_apic_base); | |
310 | ||
2605fc21 | 311 | asmlinkage __visible void kvm_spurious_fault(void) |
e3ba45b8 GL |
312 | { |
313 | /* Fault while not rebooting. We want the trace. */ | |
314 | BUG(); | |
315 | } | |
316 | EXPORT_SYMBOL_GPL(kvm_spurious_fault); | |
317 | ||
3fd28fce ED |
318 | #define EXCPT_BENIGN 0 |
319 | #define EXCPT_CONTRIBUTORY 1 | |
320 | #define EXCPT_PF 2 | |
321 | ||
322 | static int exception_class(int vector) | |
323 | { | |
324 | switch (vector) { | |
325 | case PF_VECTOR: | |
326 | return EXCPT_PF; | |
327 | case DE_VECTOR: | |
328 | case TS_VECTOR: | |
329 | case NP_VECTOR: | |
330 | case SS_VECTOR: | |
331 | case GP_VECTOR: | |
332 | return EXCPT_CONTRIBUTORY; | |
333 | default: | |
334 | break; | |
335 | } | |
336 | return EXCPT_BENIGN; | |
337 | } | |
338 | ||
d6e8c854 NA |
339 | #define EXCPT_FAULT 0 |
340 | #define EXCPT_TRAP 1 | |
341 | #define EXCPT_ABORT 2 | |
342 | #define EXCPT_INTERRUPT 3 | |
343 | ||
344 | static int exception_type(int vector) | |
345 | { | |
346 | unsigned int mask; | |
347 | ||
348 | if (WARN_ON(vector > 31 || vector == NMI_VECTOR)) | |
349 | return EXCPT_INTERRUPT; | |
350 | ||
351 | mask = 1 << vector; | |
352 | ||
353 | /* #DB is trap, as instruction watchpoints are handled elsewhere */ | |
354 | if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR))) | |
355 | return EXCPT_TRAP; | |
356 | ||
357 | if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR))) | |
358 | return EXCPT_ABORT; | |
359 | ||
360 | /* Reserved exceptions will result in fault */ | |
361 | return EXCPT_FAULT; | |
362 | } | |
363 | ||
3fd28fce | 364 | static void kvm_multiple_exception(struct kvm_vcpu *vcpu, |
ce7ddec4 JR |
365 | unsigned nr, bool has_error, u32 error_code, |
366 | bool reinject) | |
3fd28fce ED |
367 | { |
368 | u32 prev_nr; | |
369 | int class1, class2; | |
370 | ||
3842d135 AK |
371 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
372 | ||
3fd28fce ED |
373 | if (!vcpu->arch.exception.pending) { |
374 | queue: | |
3ffb2468 NA |
375 | if (has_error && !is_protmode(vcpu)) |
376 | has_error = false; | |
3fd28fce ED |
377 | vcpu->arch.exception.pending = true; |
378 | vcpu->arch.exception.has_error_code = has_error; | |
379 | vcpu->arch.exception.nr = nr; | |
380 | vcpu->arch.exception.error_code = error_code; | |
3f0fd292 | 381 | vcpu->arch.exception.reinject = reinject; |
3fd28fce ED |
382 | return; |
383 | } | |
384 | ||
385 | /* to check exception */ | |
386 | prev_nr = vcpu->arch.exception.nr; | |
387 | if (prev_nr == DF_VECTOR) { | |
388 | /* triple fault -> shutdown */ | |
a8eeb04a | 389 | kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu); |
3fd28fce ED |
390 | return; |
391 | } | |
392 | class1 = exception_class(prev_nr); | |
393 | class2 = exception_class(nr); | |
394 | if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY) | |
395 | || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) { | |
396 | /* generate double fault per SDM Table 5-5 */ | |
397 | vcpu->arch.exception.pending = true; | |
398 | vcpu->arch.exception.has_error_code = true; | |
399 | vcpu->arch.exception.nr = DF_VECTOR; | |
400 | vcpu->arch.exception.error_code = 0; | |
401 | } else | |
402 | /* replace previous exception with a new one in a hope | |
403 | that instruction re-execution will regenerate lost | |
404 | exception */ | |
405 | goto queue; | |
406 | } | |
407 | ||
298101da AK |
408 | void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr) |
409 | { | |
ce7ddec4 | 410 | kvm_multiple_exception(vcpu, nr, false, 0, false); |
298101da AK |
411 | } |
412 | EXPORT_SYMBOL_GPL(kvm_queue_exception); | |
413 | ||
ce7ddec4 JR |
414 | void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr) |
415 | { | |
416 | kvm_multiple_exception(vcpu, nr, false, 0, true); | |
417 | } | |
418 | EXPORT_SYMBOL_GPL(kvm_requeue_exception); | |
419 | ||
db8fcefa | 420 | void kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err) |
c3c91fee | 421 | { |
db8fcefa AP |
422 | if (err) |
423 | kvm_inject_gp(vcpu, 0); | |
424 | else | |
425 | kvm_x86_ops->skip_emulated_instruction(vcpu); | |
426 | } | |
427 | EXPORT_SYMBOL_GPL(kvm_complete_insn_gp); | |
8df25a32 | 428 | |
6389ee94 | 429 | void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault) |
c3c91fee AK |
430 | { |
431 | ++vcpu->stat.pf_guest; | |
6389ee94 AK |
432 | vcpu->arch.cr2 = fault->address; |
433 | kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code); | |
c3c91fee | 434 | } |
27d6c865 | 435 | EXPORT_SYMBOL_GPL(kvm_inject_page_fault); |
c3c91fee | 436 | |
ef54bcfe | 437 | static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault) |
d4f8cf66 | 438 | { |
6389ee94 AK |
439 | if (mmu_is_nested(vcpu) && !fault->nested_page_fault) |
440 | vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault); | |
d4f8cf66 | 441 | else |
6389ee94 | 442 | vcpu->arch.mmu.inject_page_fault(vcpu, fault); |
ef54bcfe PB |
443 | |
444 | return fault->nested_page_fault; | |
d4f8cf66 JR |
445 | } |
446 | ||
3419ffc8 SY |
447 | void kvm_inject_nmi(struct kvm_vcpu *vcpu) |
448 | { | |
7460fb4a AK |
449 | atomic_inc(&vcpu->arch.nmi_queued); |
450 | kvm_make_request(KVM_REQ_NMI, vcpu); | |
3419ffc8 SY |
451 | } |
452 | EXPORT_SYMBOL_GPL(kvm_inject_nmi); | |
453 | ||
298101da AK |
454 | void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code) |
455 | { | |
ce7ddec4 | 456 | kvm_multiple_exception(vcpu, nr, true, error_code, false); |
298101da AK |
457 | } |
458 | EXPORT_SYMBOL_GPL(kvm_queue_exception_e); | |
459 | ||
ce7ddec4 JR |
460 | void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code) |
461 | { | |
462 | kvm_multiple_exception(vcpu, nr, true, error_code, true); | |
463 | } | |
464 | EXPORT_SYMBOL_GPL(kvm_requeue_exception_e); | |
465 | ||
0a79b009 AK |
466 | /* |
467 | * Checks if cpl <= required_cpl; if true, return true. Otherwise queue | |
468 | * a #GP and return false. | |
469 | */ | |
470 | bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl) | |
298101da | 471 | { |
0a79b009 AK |
472 | if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl) |
473 | return true; | |
474 | kvm_queue_exception_e(vcpu, GP_VECTOR, 0); | |
475 | return false; | |
298101da | 476 | } |
0a79b009 | 477 | EXPORT_SYMBOL_GPL(kvm_require_cpl); |
298101da | 478 | |
16f8a6f9 NA |
479 | bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr) |
480 | { | |
481 | if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE)) | |
482 | return true; | |
483 | ||
484 | kvm_queue_exception(vcpu, UD_VECTOR); | |
485 | return false; | |
486 | } | |
487 | EXPORT_SYMBOL_GPL(kvm_require_dr); | |
488 | ||
ec92fe44 JR |
489 | /* |
490 | * This function will be used to read from the physical memory of the currently | |
54bf36aa | 491 | * running guest. The difference to kvm_vcpu_read_guest_page is that this function |
ec92fe44 JR |
492 | * can read from guest physical or from the guest's guest physical memory. |
493 | */ | |
494 | int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, | |
495 | gfn_t ngfn, void *data, int offset, int len, | |
496 | u32 access) | |
497 | { | |
54987b7a | 498 | struct x86_exception exception; |
ec92fe44 JR |
499 | gfn_t real_gfn; |
500 | gpa_t ngpa; | |
501 | ||
502 | ngpa = gfn_to_gpa(ngfn); | |
54987b7a | 503 | real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception); |
ec92fe44 JR |
504 | if (real_gfn == UNMAPPED_GVA) |
505 | return -EFAULT; | |
506 | ||
507 | real_gfn = gpa_to_gfn(real_gfn); | |
508 | ||
54bf36aa | 509 | return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len); |
ec92fe44 JR |
510 | } |
511 | EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu); | |
512 | ||
69b0049a | 513 | static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, |
3d06b8bf JR |
514 | void *data, int offset, int len, u32 access) |
515 | { | |
516 | return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn, | |
517 | data, offset, len, access); | |
518 | } | |
519 | ||
a03490ed CO |
520 | /* |
521 | * Load the pae pdptrs. Return true is they are all valid. | |
522 | */ | |
ff03a073 | 523 | int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3) |
a03490ed CO |
524 | { |
525 | gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT; | |
526 | unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2; | |
527 | int i; | |
528 | int ret; | |
ff03a073 | 529 | u64 pdpte[ARRAY_SIZE(mmu->pdptrs)]; |
a03490ed | 530 | |
ff03a073 JR |
531 | ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte, |
532 | offset * sizeof(u64), sizeof(pdpte), | |
533 | PFERR_USER_MASK|PFERR_WRITE_MASK); | |
a03490ed CO |
534 | if (ret < 0) { |
535 | ret = 0; | |
536 | goto out; | |
537 | } | |
538 | for (i = 0; i < ARRAY_SIZE(pdpte); ++i) { | |
43a3795a | 539 | if (is_present_gpte(pdpte[i]) && |
a0a64f50 XG |
540 | (pdpte[i] & |
541 | vcpu->arch.mmu.guest_rsvd_check.rsvd_bits_mask[0][2])) { | |
a03490ed CO |
542 | ret = 0; |
543 | goto out; | |
544 | } | |
545 | } | |
546 | ret = 1; | |
547 | ||
ff03a073 | 548 | memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs)); |
6de4f3ad AK |
549 | __set_bit(VCPU_EXREG_PDPTR, |
550 | (unsigned long *)&vcpu->arch.regs_avail); | |
551 | __set_bit(VCPU_EXREG_PDPTR, | |
552 | (unsigned long *)&vcpu->arch.regs_dirty); | |
a03490ed | 553 | out: |
a03490ed CO |
554 | |
555 | return ret; | |
556 | } | |
cc4b6871 | 557 | EXPORT_SYMBOL_GPL(load_pdptrs); |
a03490ed | 558 | |
d835dfec AK |
559 | static bool pdptrs_changed(struct kvm_vcpu *vcpu) |
560 | { | |
ff03a073 | 561 | u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)]; |
d835dfec | 562 | bool changed = true; |
3d06b8bf JR |
563 | int offset; |
564 | gfn_t gfn; | |
d835dfec AK |
565 | int r; |
566 | ||
567 | if (is_long_mode(vcpu) || !is_pae(vcpu)) | |
568 | return false; | |
569 | ||
6de4f3ad AK |
570 | if (!test_bit(VCPU_EXREG_PDPTR, |
571 | (unsigned long *)&vcpu->arch.regs_avail)) | |
572 | return true; | |
573 | ||
9f8fe504 AK |
574 | gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT; |
575 | offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1); | |
3d06b8bf JR |
576 | r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte), |
577 | PFERR_USER_MASK | PFERR_WRITE_MASK); | |
d835dfec AK |
578 | if (r < 0) |
579 | goto out; | |
ff03a073 | 580 | changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0; |
d835dfec | 581 | out: |
d835dfec AK |
582 | |
583 | return changed; | |
584 | } | |
585 | ||
49a9b07e | 586 | int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0) |
a03490ed | 587 | { |
aad82703 | 588 | unsigned long old_cr0 = kvm_read_cr0(vcpu); |
d81135a5 | 589 | unsigned long update_bits = X86_CR0_PG | X86_CR0_WP; |
aad82703 | 590 | |
f9a48e6a AK |
591 | cr0 |= X86_CR0_ET; |
592 | ||
ab344828 | 593 | #ifdef CONFIG_X86_64 |
0f12244f GN |
594 | if (cr0 & 0xffffffff00000000UL) |
595 | return 1; | |
ab344828 GN |
596 | #endif |
597 | ||
598 | cr0 &= ~CR0_RESERVED_BITS; | |
a03490ed | 599 | |
0f12244f GN |
600 | if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) |
601 | return 1; | |
a03490ed | 602 | |
0f12244f GN |
603 | if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) |
604 | return 1; | |
a03490ed CO |
605 | |
606 | if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) { | |
607 | #ifdef CONFIG_X86_64 | |
f6801dff | 608 | if ((vcpu->arch.efer & EFER_LME)) { |
a03490ed CO |
609 | int cs_db, cs_l; |
610 | ||
0f12244f GN |
611 | if (!is_pae(vcpu)) |
612 | return 1; | |
a03490ed | 613 | kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l); |
0f12244f GN |
614 | if (cs_l) |
615 | return 1; | |
a03490ed CO |
616 | } else |
617 | #endif | |
ff03a073 | 618 | if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu, |
9f8fe504 | 619 | kvm_read_cr3(vcpu))) |
0f12244f | 620 | return 1; |
a03490ed CO |
621 | } |
622 | ||
ad756a16 MJ |
623 | if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE)) |
624 | return 1; | |
625 | ||
a03490ed | 626 | kvm_x86_ops->set_cr0(vcpu, cr0); |
a03490ed | 627 | |
d170c419 | 628 | if ((cr0 ^ old_cr0) & X86_CR0_PG) { |
e5f3f027 | 629 | kvm_clear_async_pf_completion_queue(vcpu); |
d170c419 LJ |
630 | kvm_async_pf_hash_reset(vcpu); |
631 | } | |
e5f3f027 | 632 | |
aad82703 SY |
633 | if ((cr0 ^ old_cr0) & update_bits) |
634 | kvm_mmu_reset_context(vcpu); | |
b18d5431 | 635 | |
879ae188 LE |
636 | if (((cr0 ^ old_cr0) & X86_CR0_CD) && |
637 | kvm_arch_has_noncoherent_dma(vcpu->kvm) && | |
638 | !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED)) | |
b18d5431 XG |
639 | kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL); |
640 | ||
0f12244f GN |
641 | return 0; |
642 | } | |
2d3ad1f4 | 643 | EXPORT_SYMBOL_GPL(kvm_set_cr0); |
a03490ed | 644 | |
2d3ad1f4 | 645 | void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw) |
a03490ed | 646 | { |
49a9b07e | 647 | (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f)); |
a03490ed | 648 | } |
2d3ad1f4 | 649 | EXPORT_SYMBOL_GPL(kvm_lmsw); |
a03490ed | 650 | |
42bdf991 MT |
651 | static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu) |
652 | { | |
653 | if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) && | |
654 | !vcpu->guest_xcr0_loaded) { | |
655 | /* kvm_set_xcr() also depends on this */ | |
656 | xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0); | |
657 | vcpu->guest_xcr0_loaded = 1; | |
658 | } | |
659 | } | |
660 | ||
661 | static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu) | |
662 | { | |
663 | if (vcpu->guest_xcr0_loaded) { | |
664 | if (vcpu->arch.xcr0 != host_xcr0) | |
665 | xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0); | |
666 | vcpu->guest_xcr0_loaded = 0; | |
667 | } | |
668 | } | |
669 | ||
69b0049a | 670 | static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr) |
2acf923e | 671 | { |
56c103ec LJ |
672 | u64 xcr0 = xcr; |
673 | u64 old_xcr0 = vcpu->arch.xcr0; | |
46c34cb0 | 674 | u64 valid_bits; |
2acf923e DC |
675 | |
676 | /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */ | |
677 | if (index != XCR_XFEATURE_ENABLED_MASK) | |
678 | return 1; | |
d91cab78 | 679 | if (!(xcr0 & XFEATURE_MASK_FP)) |
2acf923e | 680 | return 1; |
d91cab78 | 681 | if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE)) |
2acf923e | 682 | return 1; |
46c34cb0 PB |
683 | |
684 | /* | |
685 | * Do not allow the guest to set bits that we do not support | |
686 | * saving. However, xcr0 bit 0 is always set, even if the | |
687 | * emulated CPU does not support XSAVE (see fx_init). | |
688 | */ | |
d91cab78 | 689 | valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP; |
46c34cb0 | 690 | if (xcr0 & ~valid_bits) |
2acf923e | 691 | return 1; |
46c34cb0 | 692 | |
d91cab78 DH |
693 | if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) != |
694 | (!(xcr0 & XFEATURE_MASK_BNDCSR))) | |
390bd528 LJ |
695 | return 1; |
696 | ||
d91cab78 DH |
697 | if (xcr0 & XFEATURE_MASK_AVX512) { |
698 | if (!(xcr0 & XFEATURE_MASK_YMM)) | |
612263b3 | 699 | return 1; |
d91cab78 | 700 | if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512) |
612263b3 CP |
701 | return 1; |
702 | } | |
42bdf991 | 703 | kvm_put_guest_xcr0(vcpu); |
2acf923e | 704 | vcpu->arch.xcr0 = xcr0; |
56c103ec | 705 | |
d91cab78 | 706 | if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND) |
56c103ec | 707 | kvm_update_cpuid(vcpu); |
2acf923e DC |
708 | return 0; |
709 | } | |
710 | ||
711 | int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr) | |
712 | { | |
764bcbc5 Z |
713 | if (kvm_x86_ops->get_cpl(vcpu) != 0 || |
714 | __kvm_set_xcr(vcpu, index, xcr)) { | |
2acf923e DC |
715 | kvm_inject_gp(vcpu, 0); |
716 | return 1; | |
717 | } | |
718 | return 0; | |
719 | } | |
720 | EXPORT_SYMBOL_GPL(kvm_set_xcr); | |
721 | ||
a83b29c6 | 722 | int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4) |
a03490ed | 723 | { |
fc78f519 | 724 | unsigned long old_cr4 = kvm_read_cr4(vcpu); |
0be0226f XG |
725 | unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE | |
726 | X86_CR4_SMEP | X86_CR4_SMAP; | |
727 | ||
0f12244f GN |
728 | if (cr4 & CR4_RESERVED_BITS) |
729 | return 1; | |
a03490ed | 730 | |
2acf923e DC |
731 | if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE)) |
732 | return 1; | |
733 | ||
c68b734f YW |
734 | if (!guest_cpuid_has_smep(vcpu) && (cr4 & X86_CR4_SMEP)) |
735 | return 1; | |
736 | ||
97ec8c06 FW |
737 | if (!guest_cpuid_has_smap(vcpu) && (cr4 & X86_CR4_SMAP)) |
738 | return 1; | |
739 | ||
afcbf13f | 740 | if (!guest_cpuid_has_fsgsbase(vcpu) && (cr4 & X86_CR4_FSGSBASE)) |
74dc2b4f YW |
741 | return 1; |
742 | ||
a03490ed | 743 | if (is_long_mode(vcpu)) { |
0f12244f GN |
744 | if (!(cr4 & X86_CR4_PAE)) |
745 | return 1; | |
a2edf57f AK |
746 | } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE) |
747 | && ((cr4 ^ old_cr4) & pdptr_bits) | |
9f8fe504 AK |
748 | && !load_pdptrs(vcpu, vcpu->arch.walk_mmu, |
749 | kvm_read_cr3(vcpu))) | |
0f12244f GN |
750 | return 1; |
751 | ||
ad756a16 MJ |
752 | if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) { |
753 | if (!guest_cpuid_has_pcid(vcpu)) | |
754 | return 1; | |
755 | ||
756 | /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */ | |
757 | if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu)) | |
758 | return 1; | |
759 | } | |
760 | ||
5e1746d6 | 761 | if (kvm_x86_ops->set_cr4(vcpu, cr4)) |
0f12244f | 762 | return 1; |
a03490ed | 763 | |
ad756a16 MJ |
764 | if (((cr4 ^ old_cr4) & pdptr_bits) || |
765 | (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE))) | |
aad82703 | 766 | kvm_mmu_reset_context(vcpu); |
0f12244f | 767 | |
2acf923e | 768 | if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE) |
00b27a3e | 769 | kvm_update_cpuid(vcpu); |
2acf923e | 770 | |
0f12244f GN |
771 | return 0; |
772 | } | |
2d3ad1f4 | 773 | EXPORT_SYMBOL_GPL(kvm_set_cr4); |
a03490ed | 774 | |
2390218b | 775 | int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3) |
a03490ed | 776 | { |
ac146235 | 777 | #ifdef CONFIG_X86_64 |
9d88fca7 | 778 | cr3 &= ~CR3_PCID_INVD; |
ac146235 | 779 | #endif |
9d88fca7 | 780 | |
9f8fe504 | 781 | if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) { |
0ba73cda | 782 | kvm_mmu_sync_roots(vcpu); |
77c3913b | 783 | kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu); |
0f12244f | 784 | return 0; |
d835dfec AK |
785 | } |
786 | ||
a03490ed | 787 | if (is_long_mode(vcpu)) { |
d9f89b88 JK |
788 | if (cr3 & CR3_L_MODE_RESERVED_BITS) |
789 | return 1; | |
790 | } else if (is_pae(vcpu) && is_paging(vcpu) && | |
791 | !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3)) | |
346874c9 | 792 | return 1; |
a03490ed | 793 | |
0f12244f | 794 | vcpu->arch.cr3 = cr3; |
aff48baa | 795 | __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail); |
d8d173da | 796 | kvm_mmu_new_cr3(vcpu); |
0f12244f GN |
797 | return 0; |
798 | } | |
2d3ad1f4 | 799 | EXPORT_SYMBOL_GPL(kvm_set_cr3); |
a03490ed | 800 | |
eea1cff9 | 801 | int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8) |
a03490ed | 802 | { |
0f12244f GN |
803 | if (cr8 & CR8_RESERVED_BITS) |
804 | return 1; | |
35754c98 | 805 | if (lapic_in_kernel(vcpu)) |
a03490ed CO |
806 | kvm_lapic_set_tpr(vcpu, cr8); |
807 | else | |
ad312c7c | 808 | vcpu->arch.cr8 = cr8; |
0f12244f GN |
809 | return 0; |
810 | } | |
2d3ad1f4 | 811 | EXPORT_SYMBOL_GPL(kvm_set_cr8); |
a03490ed | 812 | |
2d3ad1f4 | 813 | unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu) |
a03490ed | 814 | { |
35754c98 | 815 | if (lapic_in_kernel(vcpu)) |
a03490ed CO |
816 | return kvm_lapic_get_cr8(vcpu); |
817 | else | |
ad312c7c | 818 | return vcpu->arch.cr8; |
a03490ed | 819 | } |
2d3ad1f4 | 820 | EXPORT_SYMBOL_GPL(kvm_get_cr8); |
a03490ed | 821 | |
ae561ede NA |
822 | static void kvm_update_dr0123(struct kvm_vcpu *vcpu) |
823 | { | |
824 | int i; | |
825 | ||
826 | if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) { | |
827 | for (i = 0; i < KVM_NR_DB_REGS; i++) | |
828 | vcpu->arch.eff_db[i] = vcpu->arch.db[i]; | |
829 | vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD; | |
830 | } | |
831 | } | |
832 | ||
73aaf249 JK |
833 | static void kvm_update_dr6(struct kvm_vcpu *vcpu) |
834 | { | |
835 | if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) | |
836 | kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6); | |
837 | } | |
838 | ||
c8639010 JK |
839 | static void kvm_update_dr7(struct kvm_vcpu *vcpu) |
840 | { | |
841 | unsigned long dr7; | |
842 | ||
843 | if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) | |
844 | dr7 = vcpu->arch.guest_debug_dr7; | |
845 | else | |
846 | dr7 = vcpu->arch.dr7; | |
847 | kvm_x86_ops->set_dr7(vcpu, dr7); | |
360b948d PB |
848 | vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED; |
849 | if (dr7 & DR7_BP_EN_MASK) | |
850 | vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED; | |
c8639010 JK |
851 | } |
852 | ||
6f43ed01 NA |
853 | static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu) |
854 | { | |
855 | u64 fixed = DR6_FIXED_1; | |
856 | ||
857 | if (!guest_cpuid_has_rtm(vcpu)) | |
858 | fixed |= DR6_RTM; | |
859 | return fixed; | |
860 | } | |
861 | ||
338dbc97 | 862 | static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val) |
020df079 GN |
863 | { |
864 | switch (dr) { | |
865 | case 0 ... 3: | |
866 | vcpu->arch.db[dr] = val; | |
867 | if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) | |
868 | vcpu->arch.eff_db[dr] = val; | |
869 | break; | |
870 | case 4: | |
020df079 GN |
871 | /* fall through */ |
872 | case 6: | |
338dbc97 GN |
873 | if (val & 0xffffffff00000000ULL) |
874 | return -1; /* #GP */ | |
6f43ed01 | 875 | vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu); |
73aaf249 | 876 | kvm_update_dr6(vcpu); |
020df079 GN |
877 | break; |
878 | case 5: | |
020df079 GN |
879 | /* fall through */ |
880 | default: /* 7 */ | |
338dbc97 GN |
881 | if (val & 0xffffffff00000000ULL) |
882 | return -1; /* #GP */ | |
020df079 | 883 | vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1; |
c8639010 | 884 | kvm_update_dr7(vcpu); |
020df079 GN |
885 | break; |
886 | } | |
887 | ||
888 | return 0; | |
889 | } | |
338dbc97 GN |
890 | |
891 | int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val) | |
892 | { | |
16f8a6f9 | 893 | if (__kvm_set_dr(vcpu, dr, val)) { |
338dbc97 | 894 | kvm_inject_gp(vcpu, 0); |
16f8a6f9 NA |
895 | return 1; |
896 | } | |
897 | return 0; | |
338dbc97 | 898 | } |
020df079 GN |
899 | EXPORT_SYMBOL_GPL(kvm_set_dr); |
900 | ||
16f8a6f9 | 901 | int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val) |
020df079 GN |
902 | { |
903 | switch (dr) { | |
904 | case 0 ... 3: | |
905 | *val = vcpu->arch.db[dr]; | |
906 | break; | |
907 | case 4: | |
020df079 GN |
908 | /* fall through */ |
909 | case 6: | |
73aaf249 JK |
910 | if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) |
911 | *val = vcpu->arch.dr6; | |
912 | else | |
913 | *val = kvm_x86_ops->get_dr6(vcpu); | |
020df079 GN |
914 | break; |
915 | case 5: | |
020df079 GN |
916 | /* fall through */ |
917 | default: /* 7 */ | |
918 | *val = vcpu->arch.dr7; | |
919 | break; | |
920 | } | |
338dbc97 GN |
921 | return 0; |
922 | } | |
020df079 GN |
923 | EXPORT_SYMBOL_GPL(kvm_get_dr); |
924 | ||
022cd0e8 AK |
925 | bool kvm_rdpmc(struct kvm_vcpu *vcpu) |
926 | { | |
927 | u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX); | |
928 | u64 data; | |
929 | int err; | |
930 | ||
c6702c9d | 931 | err = kvm_pmu_rdpmc(vcpu, ecx, &data); |
022cd0e8 AK |
932 | if (err) |
933 | return err; | |
934 | kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data); | |
935 | kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32); | |
936 | return err; | |
937 | } | |
938 | EXPORT_SYMBOL_GPL(kvm_rdpmc); | |
939 | ||
043405e1 CO |
940 | /* |
941 | * List of msr numbers which we expose to userspace through KVM_GET_MSRS | |
942 | * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST. | |
943 | * | |
944 | * This list is modified at module load time to reflect the | |
e3267cbb | 945 | * capabilities of the host cpu. This capabilities test skips MSRs that are |
62ef68bb PB |
946 | * kvm-specific. Those are put in emulated_msrs; filtering of emulated_msrs |
947 | * may depend on host virtualization features rather than host cpu features. | |
043405e1 | 948 | */ |
e3267cbb | 949 | |
043405e1 CO |
950 | static u32 msrs_to_save[] = { |
951 | MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP, | |
8c06585d | 952 | MSR_STAR, |
043405e1 CO |
953 | #ifdef CONFIG_X86_64 |
954 | MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR, | |
955 | #endif | |
b3897a49 | 956 | MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA, |
9dbe6cf9 | 957 | MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS, MSR_TSC_AUX, |
043405e1 CO |
958 | }; |
959 | ||
960 | static unsigned num_msrs_to_save; | |
961 | ||
62ef68bb PB |
962 | static u32 emulated_msrs[] = { |
963 | MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK, | |
964 | MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW, | |
965 | HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL, | |
966 | HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC, | |
e7d9513b AS |
967 | HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2, |
968 | HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL, | |
e516cebb | 969 | HV_X64_MSR_RESET, |
11c4b1ca | 970 | HV_X64_MSR_VP_INDEX, |
9eec50b8 | 971 | HV_X64_MSR_VP_RUNTIME, |
5c919412 | 972 | HV_X64_MSR_SCONTROL, |
1f4b34f8 | 973 | HV_X64_MSR_STIMER0_CONFIG, |
62ef68bb PB |
974 | HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME, |
975 | MSR_KVM_PV_EOI_EN, | |
976 | ||
ba904635 | 977 | MSR_IA32_TSC_ADJUST, |
a3e06bbe | 978 | MSR_IA32_TSCDEADLINE, |
043405e1 | 979 | MSR_IA32_MISC_ENABLE, |
908e75f3 AK |
980 | MSR_IA32_MCG_STATUS, |
981 | MSR_IA32_MCG_CTL, | |
64d60670 | 982 | MSR_IA32_SMBASE, |
043405e1 CO |
983 | }; |
984 | ||
62ef68bb PB |
985 | static unsigned num_emulated_msrs; |
986 | ||
384bb783 | 987 | bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer) |
15c4a640 | 988 | { |
b69e8cae | 989 | if (efer & efer_reserved_bits) |
384bb783 | 990 | return false; |
15c4a640 | 991 | |
1b2fd70c AG |
992 | if (efer & EFER_FFXSR) { |
993 | struct kvm_cpuid_entry2 *feat; | |
994 | ||
995 | feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0); | |
b69e8cae | 996 | if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT))) |
384bb783 | 997 | return false; |
1b2fd70c AG |
998 | } |
999 | ||
d8017474 AG |
1000 | if (efer & EFER_SVME) { |
1001 | struct kvm_cpuid_entry2 *feat; | |
1002 | ||
1003 | feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0); | |
b69e8cae | 1004 | if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM))) |
384bb783 | 1005 | return false; |
d8017474 AG |
1006 | } |
1007 | ||
384bb783 JK |
1008 | return true; |
1009 | } | |
1010 | EXPORT_SYMBOL_GPL(kvm_valid_efer); | |
1011 | ||
1012 | static int set_efer(struct kvm_vcpu *vcpu, u64 efer) | |
1013 | { | |
1014 | u64 old_efer = vcpu->arch.efer; | |
1015 | ||
1016 | if (!kvm_valid_efer(vcpu, efer)) | |
1017 | return 1; | |
1018 | ||
1019 | if (is_paging(vcpu) | |
1020 | && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME)) | |
1021 | return 1; | |
1022 | ||
15c4a640 | 1023 | efer &= ~EFER_LMA; |
f6801dff | 1024 | efer |= vcpu->arch.efer & EFER_LMA; |
15c4a640 | 1025 | |
a3d204e2 SY |
1026 | kvm_x86_ops->set_efer(vcpu, efer); |
1027 | ||
aad82703 SY |
1028 | /* Update reserved bits */ |
1029 | if ((efer ^ old_efer) & EFER_NX) | |
1030 | kvm_mmu_reset_context(vcpu); | |
1031 | ||
b69e8cae | 1032 | return 0; |
15c4a640 CO |
1033 | } |
1034 | ||
f2b4b7dd JR |
1035 | void kvm_enable_efer_bits(u64 mask) |
1036 | { | |
1037 | efer_reserved_bits &= ~mask; | |
1038 | } | |
1039 | EXPORT_SYMBOL_GPL(kvm_enable_efer_bits); | |
1040 | ||
15c4a640 CO |
1041 | /* |
1042 | * Writes msr value into into the appropriate "register". | |
1043 | * Returns 0 on success, non-0 otherwise. | |
1044 | * Assumes vcpu_load() was already called. | |
1045 | */ | |
8fe8ab46 | 1046 | int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr) |
15c4a640 | 1047 | { |
854e8bb1 NA |
1048 | switch (msr->index) { |
1049 | case MSR_FS_BASE: | |
1050 | case MSR_GS_BASE: | |
1051 | case MSR_KERNEL_GS_BASE: | |
1052 | case MSR_CSTAR: | |
1053 | case MSR_LSTAR: | |
1054 | if (is_noncanonical_address(msr->data)) | |
1055 | return 1; | |
1056 | break; | |
1057 | case MSR_IA32_SYSENTER_EIP: | |
1058 | case MSR_IA32_SYSENTER_ESP: | |
1059 | /* | |
1060 | * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if | |
1061 | * non-canonical address is written on Intel but not on | |
1062 | * AMD (which ignores the top 32-bits, because it does | |
1063 | * not implement 64-bit SYSENTER). | |
1064 | * | |
1065 | * 64-bit code should hence be able to write a non-canonical | |
1066 | * value on AMD. Making the address canonical ensures that | |
1067 | * vmentry does not fail on Intel after writing a non-canonical | |
1068 | * value, and that something deterministic happens if the guest | |
1069 | * invokes 64-bit SYSENTER. | |
1070 | */ | |
1071 | msr->data = get_canonical(msr->data); | |
1072 | } | |
8fe8ab46 | 1073 | return kvm_x86_ops->set_msr(vcpu, msr); |
15c4a640 | 1074 | } |
854e8bb1 | 1075 | EXPORT_SYMBOL_GPL(kvm_set_msr); |
15c4a640 | 1076 | |
313a3dc7 CO |
1077 | /* |
1078 | * Adapt set_msr() to msr_io()'s calling convention | |
1079 | */ | |
609e36d3 PB |
1080 | static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data) |
1081 | { | |
1082 | struct msr_data msr; | |
1083 | int r; | |
1084 | ||
1085 | msr.index = index; | |
1086 | msr.host_initiated = true; | |
1087 | r = kvm_get_msr(vcpu, &msr); | |
1088 | if (r) | |
1089 | return r; | |
1090 | ||
1091 | *data = msr.data; | |
1092 | return 0; | |
1093 | } | |
1094 | ||
313a3dc7 CO |
1095 | static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data) |
1096 | { | |
8fe8ab46 WA |
1097 | struct msr_data msr; |
1098 | ||
1099 | msr.data = *data; | |
1100 | msr.index = index; | |
1101 | msr.host_initiated = true; | |
1102 | return kvm_set_msr(vcpu, &msr); | |
313a3dc7 CO |
1103 | } |
1104 | ||
16e8d74d MT |
1105 | #ifdef CONFIG_X86_64 |
1106 | struct pvclock_gtod_data { | |
1107 | seqcount_t seq; | |
1108 | ||
1109 | struct { /* extract of a clocksource struct */ | |
1110 | int vclock_mode; | |
1111 | cycle_t cycle_last; | |
1112 | cycle_t mask; | |
1113 | u32 mult; | |
1114 | u32 shift; | |
1115 | } clock; | |
1116 | ||
cbcf2dd3 TG |
1117 | u64 boot_ns; |
1118 | u64 nsec_base; | |
16e8d74d MT |
1119 | }; |
1120 | ||
1121 | static struct pvclock_gtod_data pvclock_gtod_data; | |
1122 | ||
1123 | static void update_pvclock_gtod(struct timekeeper *tk) | |
1124 | { | |
1125 | struct pvclock_gtod_data *vdata = &pvclock_gtod_data; | |
cbcf2dd3 TG |
1126 | u64 boot_ns; |
1127 | ||
876e7881 | 1128 | boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot)); |
16e8d74d MT |
1129 | |
1130 | write_seqcount_begin(&vdata->seq); | |
1131 | ||
1132 | /* copy pvclock gtod data */ | |
876e7881 PZ |
1133 | vdata->clock.vclock_mode = tk->tkr_mono.clock->archdata.vclock_mode; |
1134 | vdata->clock.cycle_last = tk->tkr_mono.cycle_last; | |
1135 | vdata->clock.mask = tk->tkr_mono.mask; | |
1136 | vdata->clock.mult = tk->tkr_mono.mult; | |
1137 | vdata->clock.shift = tk->tkr_mono.shift; | |
16e8d74d | 1138 | |
cbcf2dd3 | 1139 | vdata->boot_ns = boot_ns; |
876e7881 | 1140 | vdata->nsec_base = tk->tkr_mono.xtime_nsec; |
16e8d74d MT |
1141 | |
1142 | write_seqcount_end(&vdata->seq); | |
1143 | } | |
1144 | #endif | |
1145 | ||
bab5bb39 NK |
1146 | void kvm_set_pending_timer(struct kvm_vcpu *vcpu) |
1147 | { | |
1148 | /* | |
1149 | * Note: KVM_REQ_PENDING_TIMER is implicitly checked in | |
1150 | * vcpu_enter_guest. This function is only called from | |
1151 | * the physical CPU that is running vcpu. | |
1152 | */ | |
1153 | kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu); | |
1154 | } | |
16e8d74d | 1155 | |
18068523 GOC |
1156 | static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock) |
1157 | { | |
9ed3c444 AK |
1158 | int version; |
1159 | int r; | |
50d0a0f9 | 1160 | struct pvclock_wall_clock wc; |
923de3cf | 1161 | struct timespec boot; |
18068523 GOC |
1162 | |
1163 | if (!wall_clock) | |
1164 | return; | |
1165 | ||
9ed3c444 AK |
1166 | r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version)); |
1167 | if (r) | |
1168 | return; | |
1169 | ||
1170 | if (version & 1) | |
1171 | ++version; /* first time write, random junk */ | |
1172 | ||
1173 | ++version; | |
18068523 | 1174 | |
1dab1345 NK |
1175 | if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version))) |
1176 | return; | |
18068523 | 1177 | |
50d0a0f9 GH |
1178 | /* |
1179 | * The guest calculates current wall clock time by adding | |
34c238a1 | 1180 | * system time (updated by kvm_guest_time_update below) to the |
50d0a0f9 GH |
1181 | * wall clock specified here. guest system time equals host |
1182 | * system time for us, thus we must fill in host boot time here. | |
1183 | */ | |
923de3cf | 1184 | getboottime(&boot); |
50d0a0f9 | 1185 | |
4b648665 BR |
1186 | if (kvm->arch.kvmclock_offset) { |
1187 | struct timespec ts = ns_to_timespec(kvm->arch.kvmclock_offset); | |
1188 | boot = timespec_sub(boot, ts); | |
1189 | } | |
50d0a0f9 GH |
1190 | wc.sec = boot.tv_sec; |
1191 | wc.nsec = boot.tv_nsec; | |
1192 | wc.version = version; | |
18068523 GOC |
1193 | |
1194 | kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc)); | |
1195 | ||
1196 | version++; | |
1197 | kvm_write_guest(kvm, wall_clock, &version, sizeof(version)); | |
18068523 GOC |
1198 | } |
1199 | ||
50d0a0f9 GH |
1200 | static uint32_t div_frac(uint32_t dividend, uint32_t divisor) |
1201 | { | |
b51012de PB |
1202 | do_shl32_div32(dividend, divisor); |
1203 | return dividend; | |
50d0a0f9 GH |
1204 | } |
1205 | ||
3ae13faa | 1206 | static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz, |
5f4e3f88 | 1207 | s8 *pshift, u32 *pmultiplier) |
50d0a0f9 | 1208 | { |
5f4e3f88 | 1209 | uint64_t scaled64; |
50d0a0f9 GH |
1210 | int32_t shift = 0; |
1211 | uint64_t tps64; | |
1212 | uint32_t tps32; | |
1213 | ||
3ae13faa PB |
1214 | tps64 = base_hz; |
1215 | scaled64 = scaled_hz; | |
50933623 | 1216 | while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) { |
50d0a0f9 GH |
1217 | tps64 >>= 1; |
1218 | shift--; | |
1219 | } | |
1220 | ||
1221 | tps32 = (uint32_t)tps64; | |
50933623 JK |
1222 | while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) { |
1223 | if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000) | |
5f4e3f88 ZA |
1224 | scaled64 >>= 1; |
1225 | else | |
1226 | tps32 <<= 1; | |
50d0a0f9 GH |
1227 | shift++; |
1228 | } | |
1229 | ||
5f4e3f88 ZA |
1230 | *pshift = shift; |
1231 | *pmultiplier = div_frac(scaled64, tps32); | |
50d0a0f9 | 1232 | |
3ae13faa PB |
1233 | pr_debug("%s: base_hz %llu => %llu, shift %d, mul %u\n", |
1234 | __func__, base_hz, scaled_hz, shift, *pmultiplier); | |
50d0a0f9 GH |
1235 | } |
1236 | ||
d828199e | 1237 | #ifdef CONFIG_X86_64 |
16e8d74d | 1238 | static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0); |
d828199e | 1239 | #endif |
16e8d74d | 1240 | |
c8076604 | 1241 | static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz); |
69b0049a | 1242 | static unsigned long max_tsc_khz; |
c8076604 | 1243 | |
cc578287 | 1244 | static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec) |
8cfdc000 | 1245 | { |
cc578287 ZA |
1246 | return pvclock_scale_delta(nsec, vcpu->arch.virtual_tsc_mult, |
1247 | vcpu->arch.virtual_tsc_shift); | |
8cfdc000 ZA |
1248 | } |
1249 | ||
cc578287 | 1250 | static u32 adjust_tsc_khz(u32 khz, s32 ppm) |
1e993611 | 1251 | { |
cc578287 ZA |
1252 | u64 v = (u64)khz * (1000000 + ppm); |
1253 | do_div(v, 1000000); | |
1254 | return v; | |
1e993611 JR |
1255 | } |
1256 | ||
381d585c HZ |
1257 | static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale) |
1258 | { | |
1259 | u64 ratio; | |
1260 | ||
1261 | /* Guest TSC same frequency as host TSC? */ | |
1262 | if (!scale) { | |
1263 | vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio; | |
1264 | return 0; | |
1265 | } | |
1266 | ||
1267 | /* TSC scaling supported? */ | |
1268 | if (!kvm_has_tsc_control) { | |
1269 | if (user_tsc_khz > tsc_khz) { | |
1270 | vcpu->arch.tsc_catchup = 1; | |
1271 | vcpu->arch.tsc_always_catchup = 1; | |
1272 | return 0; | |
1273 | } else { | |
1274 | WARN(1, "user requested TSC rate below hardware speed\n"); | |
1275 | return -1; | |
1276 | } | |
1277 | } | |
1278 | ||
1279 | /* TSC scaling required - calculate ratio */ | |
1280 | ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits, | |
1281 | user_tsc_khz, tsc_khz); | |
1282 | ||
1283 | if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) { | |
1284 | WARN_ONCE(1, "Invalid TSC scaling ratio - virtual-tsc-khz=%u\n", | |
1285 | user_tsc_khz); | |
1286 | return -1; | |
1287 | } | |
1288 | ||
1289 | vcpu->arch.tsc_scaling_ratio = ratio; | |
1290 | return 0; | |
1291 | } | |
1292 | ||
4941b8cb | 1293 | static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz) |
759379dd | 1294 | { |
cc578287 ZA |
1295 | u32 thresh_lo, thresh_hi; |
1296 | int use_scaling = 0; | |
217fc9cf | 1297 | |
03ba32ca | 1298 | /* tsc_khz can be zero if TSC calibration fails */ |
4941b8cb | 1299 | if (user_tsc_khz == 0) { |
ad721883 HZ |
1300 | /* set tsc_scaling_ratio to a safe value */ |
1301 | vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio; | |
381d585c | 1302 | return -1; |
ad721883 | 1303 | } |
03ba32ca | 1304 | |
c285545f | 1305 | /* Compute a scale to convert nanoseconds in TSC cycles */ |
3ae13faa | 1306 | kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC, |
cc578287 ZA |
1307 | &vcpu->arch.virtual_tsc_shift, |
1308 | &vcpu->arch.virtual_tsc_mult); | |
4941b8cb | 1309 | vcpu->arch.virtual_tsc_khz = user_tsc_khz; |
cc578287 ZA |
1310 | |
1311 | /* | |
1312 | * Compute the variation in TSC rate which is acceptable | |
1313 | * within the range of tolerance and decide if the | |
1314 | * rate being applied is within that bounds of the hardware | |
1315 | * rate. If so, no scaling or compensation need be done. | |
1316 | */ | |
1317 | thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm); | |
1318 | thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm); | |
4941b8cb PB |
1319 | if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) { |
1320 | pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi); | |
cc578287 ZA |
1321 | use_scaling = 1; |
1322 | } | |
4941b8cb | 1323 | return set_tsc_khz(vcpu, user_tsc_khz, use_scaling); |
c285545f ZA |
1324 | } |
1325 | ||
1326 | static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns) | |
1327 | { | |
e26101b1 | 1328 | u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec, |
cc578287 ZA |
1329 | vcpu->arch.virtual_tsc_mult, |
1330 | vcpu->arch.virtual_tsc_shift); | |
e26101b1 | 1331 | tsc += vcpu->arch.this_tsc_write; |
c285545f ZA |
1332 | return tsc; |
1333 | } | |
1334 | ||
69b0049a | 1335 | static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu) |
b48aa97e MT |
1336 | { |
1337 | #ifdef CONFIG_X86_64 | |
1338 | bool vcpus_matched; | |
b48aa97e MT |
1339 | struct kvm_arch *ka = &vcpu->kvm->arch; |
1340 | struct pvclock_gtod_data *gtod = &pvclock_gtod_data; | |
1341 | ||
1342 | vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 == | |
1343 | atomic_read(&vcpu->kvm->online_vcpus)); | |
1344 | ||
7f187922 MT |
1345 | /* |
1346 | * Once the masterclock is enabled, always perform request in | |
1347 | * order to update it. | |
1348 | * | |
1349 | * In order to enable masterclock, the host clocksource must be TSC | |
1350 | * and the vcpus need to have matched TSCs. When that happens, | |
1351 | * perform request to enable masterclock. | |
1352 | */ | |
1353 | if (ka->use_master_clock || | |
1354 | (gtod->clock.vclock_mode == VCLOCK_TSC && vcpus_matched)) | |
b48aa97e MT |
1355 | kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu); |
1356 | ||
1357 | trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc, | |
1358 | atomic_read(&vcpu->kvm->online_vcpus), | |
1359 | ka->use_master_clock, gtod->clock.vclock_mode); | |
1360 | #endif | |
1361 | } | |
1362 | ||
ba904635 WA |
1363 | static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset) |
1364 | { | |
1365 | u64 curr_offset = kvm_x86_ops->read_tsc_offset(vcpu); | |
1366 | vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset; | |
1367 | } | |
1368 | ||
35181e86 HZ |
1369 | /* |
1370 | * Multiply tsc by a fixed point number represented by ratio. | |
1371 | * | |
1372 | * The most significant 64-N bits (mult) of ratio represent the | |
1373 | * integral part of the fixed point number; the remaining N bits | |
1374 | * (frac) represent the fractional part, ie. ratio represents a fixed | |
1375 | * point number (mult + frac * 2^(-N)). | |
1376 | * | |
1377 | * N equals to kvm_tsc_scaling_ratio_frac_bits. | |
1378 | */ | |
1379 | static inline u64 __scale_tsc(u64 ratio, u64 tsc) | |
1380 | { | |
1381 | return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits); | |
1382 | } | |
1383 | ||
1384 | u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc) | |
1385 | { | |
1386 | u64 _tsc = tsc; | |
1387 | u64 ratio = vcpu->arch.tsc_scaling_ratio; | |
1388 | ||
1389 | if (ratio != kvm_default_tsc_scaling_ratio) | |
1390 | _tsc = __scale_tsc(ratio, tsc); | |
1391 | ||
1392 | return _tsc; | |
1393 | } | |
1394 | EXPORT_SYMBOL_GPL(kvm_scale_tsc); | |
1395 | ||
07c1419a HZ |
1396 | static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc) |
1397 | { | |
1398 | u64 tsc; | |
1399 | ||
1400 | tsc = kvm_scale_tsc(vcpu, rdtsc()); | |
1401 | ||
1402 | return target_tsc - tsc; | |
1403 | } | |
1404 | ||
4ba76538 HZ |
1405 | u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc) |
1406 | { | |
1407 | return kvm_x86_ops->read_l1_tsc(vcpu, kvm_scale_tsc(vcpu, host_tsc)); | |
1408 | } | |
1409 | EXPORT_SYMBOL_GPL(kvm_read_l1_tsc); | |
1410 | ||
8fe8ab46 | 1411 | void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr) |
99e3e30a ZA |
1412 | { |
1413 | struct kvm *kvm = vcpu->kvm; | |
f38e098f | 1414 | u64 offset, ns, elapsed; |
99e3e30a | 1415 | unsigned long flags; |
02626b6a | 1416 | s64 usdiff; |
b48aa97e | 1417 | bool matched; |
0d3da0d2 | 1418 | bool already_matched; |
8fe8ab46 | 1419 | u64 data = msr->data; |
99e3e30a | 1420 | |
038f8c11 | 1421 | raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags); |
07c1419a | 1422 | offset = kvm_compute_tsc_offset(vcpu, data); |
759379dd | 1423 | ns = get_kernel_ns(); |
f38e098f | 1424 | elapsed = ns - kvm->arch.last_tsc_nsec; |
5d3cb0f6 | 1425 | |
03ba32ca | 1426 | if (vcpu->arch.virtual_tsc_khz) { |
8915aa27 MT |
1427 | int faulted = 0; |
1428 | ||
03ba32ca MT |
1429 | /* n.b - signed multiplication and division required */ |
1430 | usdiff = data - kvm->arch.last_tsc_write; | |
5d3cb0f6 | 1431 | #ifdef CONFIG_X86_64 |
03ba32ca | 1432 | usdiff = (usdiff * 1000) / vcpu->arch.virtual_tsc_khz; |
5d3cb0f6 | 1433 | #else |
03ba32ca | 1434 | /* do_div() only does unsigned */ |
8915aa27 MT |
1435 | asm("1: idivl %[divisor]\n" |
1436 | "2: xor %%edx, %%edx\n" | |
1437 | " movl $0, %[faulted]\n" | |
1438 | "3:\n" | |
1439 | ".section .fixup,\"ax\"\n" | |
1440 | "4: movl $1, %[faulted]\n" | |
1441 | " jmp 3b\n" | |
1442 | ".previous\n" | |
1443 | ||
1444 | _ASM_EXTABLE(1b, 4b) | |
1445 | ||
1446 | : "=A"(usdiff), [faulted] "=r" (faulted) | |
1447 | : "A"(usdiff * 1000), [divisor] "rm"(vcpu->arch.virtual_tsc_khz)); | |
1448 | ||
5d3cb0f6 | 1449 | #endif |
03ba32ca MT |
1450 | do_div(elapsed, 1000); |
1451 | usdiff -= elapsed; | |
1452 | if (usdiff < 0) | |
1453 | usdiff = -usdiff; | |
8915aa27 MT |
1454 | |
1455 | /* idivl overflow => difference is larger than USEC_PER_SEC */ | |
1456 | if (faulted) | |
1457 | usdiff = USEC_PER_SEC; | |
03ba32ca MT |
1458 | } else |
1459 | usdiff = USEC_PER_SEC; /* disable TSC match window below */ | |
f38e098f ZA |
1460 | |
1461 | /* | |
5d3cb0f6 ZA |
1462 | * Special case: TSC write with a small delta (1 second) of virtual |
1463 | * cycle time against real time is interpreted as an attempt to | |
1464 | * synchronize the CPU. | |
1465 | * | |
1466 | * For a reliable TSC, we can match TSC offsets, and for an unstable | |
1467 | * TSC, we add elapsed time in this computation. We could let the | |
1468 | * compensation code attempt to catch up if we fall behind, but | |
1469 | * it's better to try to match offsets from the beginning. | |
1470 | */ | |
02626b6a | 1471 | if (usdiff < USEC_PER_SEC && |
5d3cb0f6 | 1472 | vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) { |
f38e098f | 1473 | if (!check_tsc_unstable()) { |
e26101b1 | 1474 | offset = kvm->arch.cur_tsc_offset; |
f38e098f ZA |
1475 | pr_debug("kvm: matched tsc offset for %llu\n", data); |
1476 | } else { | |
857e4099 | 1477 | u64 delta = nsec_to_cycles(vcpu, elapsed); |
5d3cb0f6 | 1478 | data += delta; |
07c1419a | 1479 | offset = kvm_compute_tsc_offset(vcpu, data); |
759379dd | 1480 | pr_debug("kvm: adjusted tsc offset by %llu\n", delta); |
f38e098f | 1481 | } |
b48aa97e | 1482 | matched = true; |
0d3da0d2 | 1483 | already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation); |
e26101b1 ZA |
1484 | } else { |
1485 | /* | |
1486 | * We split periods of matched TSC writes into generations. | |
1487 | * For each generation, we track the original measured | |
1488 | * nanosecond time, offset, and write, so if TSCs are in | |
1489 | * sync, we can match exact offset, and if not, we can match | |
4a969980 | 1490 | * exact software computation in compute_guest_tsc() |
e26101b1 ZA |
1491 | * |
1492 | * These values are tracked in kvm->arch.cur_xxx variables. | |
1493 | */ | |
1494 | kvm->arch.cur_tsc_generation++; | |
1495 | kvm->arch.cur_tsc_nsec = ns; | |
1496 | kvm->arch.cur_tsc_write = data; | |
1497 | kvm->arch.cur_tsc_offset = offset; | |
b48aa97e | 1498 | matched = false; |
0d3da0d2 | 1499 | pr_debug("kvm: new tsc generation %llu, clock %llu\n", |
e26101b1 | 1500 | kvm->arch.cur_tsc_generation, data); |
f38e098f | 1501 | } |
e26101b1 ZA |
1502 | |
1503 | /* | |
1504 | * We also track th most recent recorded KHZ, write and time to | |
1505 | * allow the matching interval to be extended at each write. | |
1506 | */ | |
f38e098f ZA |
1507 | kvm->arch.last_tsc_nsec = ns; |
1508 | kvm->arch.last_tsc_write = data; | |
5d3cb0f6 | 1509 | kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz; |
99e3e30a | 1510 | |
b183aa58 | 1511 | vcpu->arch.last_guest_tsc = data; |
e26101b1 ZA |
1512 | |
1513 | /* Keep track of which generation this VCPU has synchronized to */ | |
1514 | vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation; | |
1515 | vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec; | |
1516 | vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write; | |
1517 | ||
ba904635 WA |
1518 | if (guest_cpuid_has_tsc_adjust(vcpu) && !msr->host_initiated) |
1519 | update_ia32_tsc_adjust_msr(vcpu, offset); | |
e26101b1 ZA |
1520 | kvm_x86_ops->write_tsc_offset(vcpu, offset); |
1521 | raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags); | |
b48aa97e MT |
1522 | |
1523 | spin_lock(&kvm->arch.pvclock_gtod_sync_lock); | |
0d3da0d2 | 1524 | if (!matched) { |
b48aa97e | 1525 | kvm->arch.nr_vcpus_matched_tsc = 0; |
0d3da0d2 TG |
1526 | } else if (!already_matched) { |
1527 | kvm->arch.nr_vcpus_matched_tsc++; | |
1528 | } | |
b48aa97e MT |
1529 | |
1530 | kvm_track_tsc_matching(vcpu); | |
1531 | spin_unlock(&kvm->arch.pvclock_gtod_sync_lock); | |
99e3e30a | 1532 | } |
e26101b1 | 1533 | |
99e3e30a ZA |
1534 | EXPORT_SYMBOL_GPL(kvm_write_tsc); |
1535 | ||
58ea6767 HZ |
1536 | static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu, |
1537 | s64 adjustment) | |
1538 | { | |
1539 | kvm_x86_ops->adjust_tsc_offset_guest(vcpu, adjustment); | |
1540 | } | |
1541 | ||
1542 | static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment) | |
1543 | { | |
1544 | if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio) | |
1545 | WARN_ON(adjustment < 0); | |
1546 | adjustment = kvm_scale_tsc(vcpu, (u64) adjustment); | |
1547 | kvm_x86_ops->adjust_tsc_offset_guest(vcpu, adjustment); | |
1548 | } | |
1549 | ||
d828199e MT |
1550 | #ifdef CONFIG_X86_64 |
1551 | ||
1552 | static cycle_t read_tsc(void) | |
1553 | { | |
03b9730b AL |
1554 | cycle_t ret = (cycle_t)rdtsc_ordered(); |
1555 | u64 last = pvclock_gtod_data.clock.cycle_last; | |
d828199e MT |
1556 | |
1557 | if (likely(ret >= last)) | |
1558 | return ret; | |
1559 | ||
1560 | /* | |
1561 | * GCC likes to generate cmov here, but this branch is extremely | |
1562 | * predictable (it's just a funciton of time and the likely is | |
1563 | * very likely) and there's a data dependence, so force GCC | |
1564 | * to generate a branch instead. I don't barrier() because | |
1565 | * we don't actually need a barrier, and if this function | |
1566 | * ever gets inlined it will generate worse code. | |
1567 | */ | |
1568 | asm volatile (""); | |
1569 | return last; | |
1570 | } | |
1571 | ||
1572 | static inline u64 vgettsc(cycle_t *cycle_now) | |
1573 | { | |
1574 | long v; | |
1575 | struct pvclock_gtod_data *gtod = &pvclock_gtod_data; | |
1576 | ||
1577 | *cycle_now = read_tsc(); | |
1578 | ||
1579 | v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask; | |
1580 | return v * gtod->clock.mult; | |
1581 | } | |
1582 | ||
cbcf2dd3 | 1583 | static int do_monotonic_boot(s64 *t, cycle_t *cycle_now) |
d828199e | 1584 | { |
cbcf2dd3 | 1585 | struct pvclock_gtod_data *gtod = &pvclock_gtod_data; |
d828199e | 1586 | unsigned long seq; |
d828199e | 1587 | int mode; |
cbcf2dd3 | 1588 | u64 ns; |
d828199e | 1589 | |
d828199e MT |
1590 | do { |
1591 | seq = read_seqcount_begin(>od->seq); | |
1592 | mode = gtod->clock.vclock_mode; | |
cbcf2dd3 | 1593 | ns = gtod->nsec_base; |
d828199e MT |
1594 | ns += vgettsc(cycle_now); |
1595 | ns >>= gtod->clock.shift; | |
cbcf2dd3 | 1596 | ns += gtod->boot_ns; |
d828199e | 1597 | } while (unlikely(read_seqcount_retry(>od->seq, seq))); |
cbcf2dd3 | 1598 | *t = ns; |
d828199e MT |
1599 | |
1600 | return mode; | |
1601 | } | |
1602 | ||
1603 | /* returns true if host is using tsc clocksource */ | |
1604 | static bool kvm_get_time_and_clockread(s64 *kernel_ns, cycle_t *cycle_now) | |
1605 | { | |
d828199e MT |
1606 | /* checked again under seqlock below */ |
1607 | if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC) | |
1608 | return false; | |
1609 | ||
cbcf2dd3 | 1610 | return do_monotonic_boot(kernel_ns, cycle_now) == VCLOCK_TSC; |
d828199e MT |
1611 | } |
1612 | #endif | |
1613 | ||
1614 | /* | |
1615 | * | |
b48aa97e MT |
1616 | * Assuming a stable TSC across physical CPUS, and a stable TSC |
1617 | * across virtual CPUs, the following condition is possible. | |
1618 | * Each numbered line represents an event visible to both | |
d828199e MT |
1619 | * CPUs at the next numbered event. |
1620 | * | |
1621 | * "timespecX" represents host monotonic time. "tscX" represents | |
1622 | * RDTSC value. | |
1623 | * | |
1624 | * VCPU0 on CPU0 | VCPU1 on CPU1 | |
1625 | * | |
1626 | * 1. read timespec0,tsc0 | |
1627 | * 2. | timespec1 = timespec0 + N | |
1628 | * | tsc1 = tsc0 + M | |
1629 | * 3. transition to guest | transition to guest | |
1630 | * 4. ret0 = timespec0 + (rdtsc - tsc0) | | |
1631 | * 5. | ret1 = timespec1 + (rdtsc - tsc1) | |
1632 | * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M)) | |
1633 | * | |
1634 | * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity: | |
1635 | * | |
1636 | * - ret0 < ret1 | |
1637 | * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M)) | |
1638 | * ... | |
1639 | * - 0 < N - M => M < N | |
1640 | * | |
1641 | * That is, when timespec0 != timespec1, M < N. Unfortunately that is not | |
1642 | * always the case (the difference between two distinct xtime instances | |
1643 | * might be smaller then the difference between corresponding TSC reads, | |
1644 | * when updating guest vcpus pvclock areas). | |
1645 | * | |
1646 | * To avoid that problem, do not allow visibility of distinct | |
1647 | * system_timestamp/tsc_timestamp values simultaneously: use a master | |
1648 | * copy of host monotonic time values. Update that master copy | |
1649 | * in lockstep. | |
1650 | * | |
b48aa97e | 1651 | * Rely on synchronization of host TSCs and guest TSCs for monotonicity. |
d828199e MT |
1652 | * |
1653 | */ | |
1654 | ||
1655 | static void pvclock_update_vm_gtod_copy(struct kvm *kvm) | |
1656 | { | |
1657 | #ifdef CONFIG_X86_64 | |
1658 | struct kvm_arch *ka = &kvm->arch; | |
1659 | int vclock_mode; | |
b48aa97e MT |
1660 | bool host_tsc_clocksource, vcpus_matched; |
1661 | ||
1662 | vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 == | |
1663 | atomic_read(&kvm->online_vcpus)); | |
d828199e MT |
1664 | |
1665 | /* | |
1666 | * If the host uses TSC clock, then passthrough TSC as stable | |
1667 | * to the guest. | |
1668 | */ | |
b48aa97e | 1669 | host_tsc_clocksource = kvm_get_time_and_clockread( |
d828199e MT |
1670 | &ka->master_kernel_ns, |
1671 | &ka->master_cycle_now); | |
1672 | ||
16a96021 | 1673 | ka->use_master_clock = host_tsc_clocksource && vcpus_matched |
54750f2c MT |
1674 | && !backwards_tsc_observed |
1675 | && !ka->boot_vcpu_runs_old_kvmclock; | |
b48aa97e | 1676 | |
d828199e MT |
1677 | if (ka->use_master_clock) |
1678 | atomic_set(&kvm_guest_has_master_clock, 1); | |
1679 | ||
1680 | vclock_mode = pvclock_gtod_data.clock.vclock_mode; | |
b48aa97e MT |
1681 | trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode, |
1682 | vcpus_matched); | |
d828199e MT |
1683 | #endif |
1684 | } | |
1685 | ||
2860c4b1 PB |
1686 | void kvm_make_mclock_inprogress_request(struct kvm *kvm) |
1687 | { | |
1688 | kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS); | |
1689 | } | |
1690 | ||
2e762ff7 MT |
1691 | static void kvm_gen_update_masterclock(struct kvm *kvm) |
1692 | { | |
1693 | #ifdef CONFIG_X86_64 | |
1694 | int i; | |
1695 | struct kvm_vcpu *vcpu; | |
1696 | struct kvm_arch *ka = &kvm->arch; | |
1697 | ||
1698 | spin_lock(&ka->pvclock_gtod_sync_lock); | |
1699 | kvm_make_mclock_inprogress_request(kvm); | |
1700 | /* no guest entries from this point */ | |
1701 | pvclock_update_vm_gtod_copy(kvm); | |
1702 | ||
1703 | kvm_for_each_vcpu(i, vcpu, kvm) | |
105b21bb | 1704 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu); |
2e762ff7 MT |
1705 | |
1706 | /* guest entries allowed */ | |
1707 | kvm_for_each_vcpu(i, vcpu, kvm) | |
1708 | clear_bit(KVM_REQ_MCLOCK_INPROGRESS, &vcpu->requests); | |
1709 | ||
1710 | spin_unlock(&ka->pvclock_gtod_sync_lock); | |
1711 | #endif | |
1712 | } | |
1713 | ||
34c238a1 | 1714 | static int kvm_guest_time_update(struct kvm_vcpu *v) |
18068523 | 1715 | { |
78db6a50 | 1716 | unsigned long flags, tgt_tsc_khz; |
18068523 | 1717 | struct kvm_vcpu_arch *vcpu = &v->arch; |
d828199e | 1718 | struct kvm_arch *ka = &v->kvm->arch; |
f25e656d | 1719 | s64 kernel_ns; |
d828199e | 1720 | u64 tsc_timestamp, host_tsc; |
0b79459b | 1721 | struct pvclock_vcpu_time_info guest_hv_clock; |
51d59c6b | 1722 | u8 pvclock_flags; |
d828199e MT |
1723 | bool use_master_clock; |
1724 | ||
1725 | kernel_ns = 0; | |
1726 | host_tsc = 0; | |
18068523 | 1727 | |
d828199e MT |
1728 | /* |
1729 | * If the host uses TSC clock, then passthrough TSC as stable | |
1730 | * to the guest. | |
1731 | */ | |
1732 | spin_lock(&ka->pvclock_gtod_sync_lock); | |
1733 | use_master_clock = ka->use_master_clock; | |
1734 | if (use_master_clock) { | |
1735 | host_tsc = ka->master_cycle_now; | |
1736 | kernel_ns = ka->master_kernel_ns; | |
1737 | } | |
1738 | spin_unlock(&ka->pvclock_gtod_sync_lock); | |
c09664bb MT |
1739 | |
1740 | /* Keep irq disabled to prevent changes to the clock */ | |
1741 | local_irq_save(flags); | |
78db6a50 PB |
1742 | tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz); |
1743 | if (unlikely(tgt_tsc_khz == 0)) { | |
c09664bb MT |
1744 | local_irq_restore(flags); |
1745 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, v); | |
1746 | return 1; | |
1747 | } | |
d828199e | 1748 | if (!use_master_clock) { |
4ea1636b | 1749 | host_tsc = rdtsc(); |
d828199e MT |
1750 | kernel_ns = get_kernel_ns(); |
1751 | } | |
1752 | ||
4ba76538 | 1753 | tsc_timestamp = kvm_read_l1_tsc(v, host_tsc); |
d828199e | 1754 | |
c285545f ZA |
1755 | /* |
1756 | * We may have to catch up the TSC to match elapsed wall clock | |
1757 | * time for two reasons, even if kvmclock is used. | |
1758 | * 1) CPU could have been running below the maximum TSC rate | |
1759 | * 2) Broken TSC compensation resets the base at each VCPU | |
1760 | * entry to avoid unknown leaps of TSC even when running | |
1761 | * again on the same CPU. This may cause apparent elapsed | |
1762 | * time to disappear, and the guest to stand still or run | |
1763 | * very slowly. | |
1764 | */ | |
1765 | if (vcpu->tsc_catchup) { | |
1766 | u64 tsc = compute_guest_tsc(v, kernel_ns); | |
1767 | if (tsc > tsc_timestamp) { | |
f1e2b260 | 1768 | adjust_tsc_offset_guest(v, tsc - tsc_timestamp); |
c285545f ZA |
1769 | tsc_timestamp = tsc; |
1770 | } | |
50d0a0f9 GH |
1771 | } |
1772 | ||
18068523 GOC |
1773 | local_irq_restore(flags); |
1774 | ||
0b79459b | 1775 | if (!vcpu->pv_time_enabled) |
c285545f | 1776 | return 0; |
18068523 | 1777 | |
78db6a50 PB |
1778 | if (kvm_has_tsc_control) |
1779 | tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz); | |
1780 | ||
1781 | if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) { | |
3ae13faa | 1782 | kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL, |
5f4e3f88 ZA |
1783 | &vcpu->hv_clock.tsc_shift, |
1784 | &vcpu->hv_clock.tsc_to_system_mul); | |
78db6a50 | 1785 | vcpu->hw_tsc_khz = tgt_tsc_khz; |
8cfdc000 ZA |
1786 | } |
1787 | ||
1788 | /* With all the info we got, fill in the values */ | |
1d5f066e | 1789 | vcpu->hv_clock.tsc_timestamp = tsc_timestamp; |
759379dd | 1790 | vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset; |
28e4639a | 1791 | vcpu->last_guest_tsc = tsc_timestamp; |
51d59c6b | 1792 | |
09a0c3f1 OH |
1793 | if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time, |
1794 | &guest_hv_clock, sizeof(guest_hv_clock)))) | |
1795 | return 0; | |
1796 | ||
5dca0d91 RK |
1797 | /* This VCPU is paused, but it's legal for a guest to read another |
1798 | * VCPU's kvmclock, so we really have to follow the specification where | |
1799 | * it says that version is odd if data is being modified, and even after | |
1800 | * it is consistent. | |
1801 | * | |
1802 | * Version field updates must be kept separate. This is because | |
1803 | * kvm_write_guest_cached might use a "rep movs" instruction, and | |
1804 | * writes within a string instruction are weakly ordered. So there | |
1805 | * are three writes overall. | |
1806 | * | |
1807 | * As a small optimization, only write the version field in the first | |
1808 | * and third write. The vcpu->pv_time cache is still valid, because the | |
1809 | * version field is the first in the struct. | |
18068523 | 1810 | */ |
5dca0d91 RK |
1811 | BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0); |
1812 | ||
1813 | vcpu->hv_clock.version = guest_hv_clock.version + 1; | |
1814 | kvm_write_guest_cached(v->kvm, &vcpu->pv_time, | |
1815 | &vcpu->hv_clock, | |
1816 | sizeof(vcpu->hv_clock.version)); | |
1817 | ||
1818 | smp_wmb(); | |
78c0337a MT |
1819 | |
1820 | /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */ | |
0b79459b | 1821 | pvclock_flags = (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED); |
78c0337a MT |
1822 | |
1823 | if (vcpu->pvclock_set_guest_stopped_request) { | |
1824 | pvclock_flags |= PVCLOCK_GUEST_STOPPED; | |
1825 | vcpu->pvclock_set_guest_stopped_request = false; | |
1826 | } | |
1827 | ||
d828199e MT |
1828 | /* If the host uses TSC clocksource, then it is stable */ |
1829 | if (use_master_clock) | |
1830 | pvclock_flags |= PVCLOCK_TSC_STABLE_BIT; | |
1831 | ||
78c0337a MT |
1832 | vcpu->hv_clock.flags = pvclock_flags; |
1833 | ||
ce1a5e60 DM |
1834 | trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock); |
1835 | ||
0b79459b AH |
1836 | kvm_write_guest_cached(v->kvm, &vcpu->pv_time, |
1837 | &vcpu->hv_clock, | |
1838 | sizeof(vcpu->hv_clock)); | |
5dca0d91 RK |
1839 | |
1840 | smp_wmb(); | |
1841 | ||
1842 | vcpu->hv_clock.version++; | |
1843 | kvm_write_guest_cached(v->kvm, &vcpu->pv_time, | |
1844 | &vcpu->hv_clock, | |
1845 | sizeof(vcpu->hv_clock.version)); | |
8cfdc000 | 1846 | return 0; |
c8076604 GH |
1847 | } |
1848 | ||
0061d53d MT |
1849 | /* |
1850 | * kvmclock updates which are isolated to a given vcpu, such as | |
1851 | * vcpu->cpu migration, should not allow system_timestamp from | |
1852 | * the rest of the vcpus to remain static. Otherwise ntp frequency | |
1853 | * correction applies to one vcpu's system_timestamp but not | |
1854 | * the others. | |
1855 | * | |
1856 | * So in those cases, request a kvmclock update for all vcpus. | |
7e44e449 AJ |
1857 | * We need to rate-limit these requests though, as they can |
1858 | * considerably slow guests that have a large number of vcpus. | |
1859 | * The time for a remote vcpu to update its kvmclock is bound | |
1860 | * by the delay we use to rate-limit the updates. | |
0061d53d MT |
1861 | */ |
1862 | ||
7e44e449 AJ |
1863 | #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100) |
1864 | ||
1865 | static void kvmclock_update_fn(struct work_struct *work) | |
0061d53d MT |
1866 | { |
1867 | int i; | |
7e44e449 AJ |
1868 | struct delayed_work *dwork = to_delayed_work(work); |
1869 | struct kvm_arch *ka = container_of(dwork, struct kvm_arch, | |
1870 | kvmclock_update_work); | |
1871 | struct kvm *kvm = container_of(ka, struct kvm, arch); | |
0061d53d MT |
1872 | struct kvm_vcpu *vcpu; |
1873 | ||
1874 | kvm_for_each_vcpu(i, vcpu, kvm) { | |
105b21bb | 1875 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu); |
0061d53d MT |
1876 | kvm_vcpu_kick(vcpu); |
1877 | } | |
1878 | } | |
1879 | ||
7e44e449 AJ |
1880 | static void kvm_gen_kvmclock_update(struct kvm_vcpu *v) |
1881 | { | |
1882 | struct kvm *kvm = v->kvm; | |
1883 | ||
105b21bb | 1884 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, v); |
7e44e449 AJ |
1885 | schedule_delayed_work(&kvm->arch.kvmclock_update_work, |
1886 | KVMCLOCK_UPDATE_DELAY); | |
1887 | } | |
1888 | ||
332967a3 AJ |
1889 | #define KVMCLOCK_SYNC_PERIOD (300 * HZ) |
1890 | ||
1891 | static void kvmclock_sync_fn(struct work_struct *work) | |
1892 | { | |
1893 | struct delayed_work *dwork = to_delayed_work(work); | |
1894 | struct kvm_arch *ka = container_of(dwork, struct kvm_arch, | |
1895 | kvmclock_sync_work); | |
1896 | struct kvm *kvm = container_of(ka, struct kvm, arch); | |
1897 | ||
630994b3 MT |
1898 | if (!kvmclock_periodic_sync) |
1899 | return; | |
1900 | ||
332967a3 AJ |
1901 | schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0); |
1902 | schedule_delayed_work(&kvm->arch.kvmclock_sync_work, | |
1903 | KVMCLOCK_SYNC_PERIOD); | |
1904 | } | |
1905 | ||
890ca9ae | 1906 | static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data) |
15c4a640 | 1907 | { |
890ca9ae HY |
1908 | u64 mcg_cap = vcpu->arch.mcg_cap; |
1909 | unsigned bank_num = mcg_cap & 0xff; | |
1910 | ||
15c4a640 | 1911 | switch (msr) { |
15c4a640 | 1912 | case MSR_IA32_MCG_STATUS: |
890ca9ae | 1913 | vcpu->arch.mcg_status = data; |
15c4a640 | 1914 | break; |
c7ac679c | 1915 | case MSR_IA32_MCG_CTL: |
890ca9ae HY |
1916 | if (!(mcg_cap & MCG_CTL_P)) |
1917 | return 1; | |
1918 | if (data != 0 && data != ~(u64)0) | |
1919 | return -1; | |
1920 | vcpu->arch.mcg_ctl = data; | |
1921 | break; | |
1922 | default: | |
1923 | if (msr >= MSR_IA32_MC0_CTL && | |
81760dcc | 1924 | msr < MSR_IA32_MCx_CTL(bank_num)) { |
890ca9ae | 1925 | u32 offset = msr - MSR_IA32_MC0_CTL; |
114be429 AP |
1926 | /* only 0 or all 1s can be written to IA32_MCi_CTL |
1927 | * some Linux kernels though clear bit 10 in bank 4 to | |
1928 | * workaround a BIOS/GART TBL issue on AMD K8s, ignore | |
1929 | * this to avoid an uncatched #GP in the guest | |
1930 | */ | |
890ca9ae | 1931 | if ((offset & 0x3) == 0 && |
114be429 | 1932 | data != 0 && (data | (1 << 10)) != ~(u64)0) |
890ca9ae HY |
1933 | return -1; |
1934 | vcpu->arch.mce_banks[offset] = data; | |
1935 | break; | |
1936 | } | |
1937 | return 1; | |
1938 | } | |
1939 | return 0; | |
1940 | } | |
1941 | ||
ffde22ac ES |
1942 | static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data) |
1943 | { | |
1944 | struct kvm *kvm = vcpu->kvm; | |
1945 | int lm = is_long_mode(vcpu); | |
1946 | u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64 | |
1947 | : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32; | |
1948 | u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64 | |
1949 | : kvm->arch.xen_hvm_config.blob_size_32; | |
1950 | u32 page_num = data & ~PAGE_MASK; | |
1951 | u64 page_addr = data & PAGE_MASK; | |
1952 | u8 *page; | |
1953 | int r; | |
1954 | ||
1955 | r = -E2BIG; | |
1956 | if (page_num >= blob_size) | |
1957 | goto out; | |
1958 | r = -ENOMEM; | |
ff5c2c03 SL |
1959 | page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE); |
1960 | if (IS_ERR(page)) { | |
1961 | r = PTR_ERR(page); | |
ffde22ac | 1962 | goto out; |
ff5c2c03 | 1963 | } |
54bf36aa | 1964 | if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE)) |
ffde22ac ES |
1965 | goto out_free; |
1966 | r = 0; | |
1967 | out_free: | |
1968 | kfree(page); | |
1969 | out: | |
1970 | return r; | |
1971 | } | |
1972 | ||
344d9588 GN |
1973 | static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data) |
1974 | { | |
1975 | gpa_t gpa = data & ~0x3f; | |
1976 | ||
4a969980 | 1977 | /* Bits 2:5 are reserved, Should be zero */ |
6adba527 | 1978 | if (data & 0x3c) |
344d9588 GN |
1979 | return 1; |
1980 | ||
1981 | vcpu->arch.apf.msr_val = data; | |
1982 | ||
1983 | if (!(data & KVM_ASYNC_PF_ENABLED)) { | |
1984 | kvm_clear_async_pf_completion_queue(vcpu); | |
1985 | kvm_async_pf_hash_reset(vcpu); | |
1986 | return 0; | |
1987 | } | |
1988 | ||
8f964525 AH |
1989 | if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa, |
1990 | sizeof(u32))) | |
344d9588 GN |
1991 | return 1; |
1992 | ||
6adba527 | 1993 | vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS); |
344d9588 GN |
1994 | kvm_async_pf_wakeup_all(vcpu); |
1995 | return 0; | |
1996 | } | |
1997 | ||
12f9a48f GC |
1998 | static void kvmclock_reset(struct kvm_vcpu *vcpu) |
1999 | { | |
0b79459b | 2000 | vcpu->arch.pv_time_enabled = false; |
12f9a48f GC |
2001 | } |
2002 | ||
c9aaa895 GC |
2003 | static void accumulate_steal_time(struct kvm_vcpu *vcpu) |
2004 | { | |
2005 | u64 delta; | |
2006 | ||
2007 | if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED)) | |
2008 | return; | |
2009 | ||
2010 | delta = current->sched_info.run_delay - vcpu->arch.st.last_steal; | |
2011 | vcpu->arch.st.last_steal = current->sched_info.run_delay; | |
2012 | vcpu->arch.st.accum_steal = delta; | |
2013 | } | |
2014 | ||
2015 | static void record_steal_time(struct kvm_vcpu *vcpu) | |
2016 | { | |
7cae2bed MT |
2017 | accumulate_steal_time(vcpu); |
2018 | ||
c9aaa895 GC |
2019 | if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED)) |
2020 | return; | |
2021 | ||
2022 | if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime, | |
2023 | &vcpu->arch.st.steal, sizeof(struct kvm_steal_time)))) | |
2024 | return; | |
2025 | ||
2026 | vcpu->arch.st.steal.steal += vcpu->arch.st.accum_steal; | |
2027 | vcpu->arch.st.steal.version += 2; | |
2028 | vcpu->arch.st.accum_steal = 0; | |
2029 | ||
2030 | kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime, | |
2031 | &vcpu->arch.st.steal, sizeof(struct kvm_steal_time)); | |
2032 | } | |
2033 | ||
8fe8ab46 | 2034 | int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info) |
15c4a640 | 2035 | { |
5753785f | 2036 | bool pr = false; |
8fe8ab46 WA |
2037 | u32 msr = msr_info->index; |
2038 | u64 data = msr_info->data; | |
5753785f | 2039 | |
15c4a640 | 2040 | switch (msr) { |
2e32b719 BP |
2041 | case MSR_AMD64_NB_CFG: |
2042 | case MSR_IA32_UCODE_REV: | |
2043 | case MSR_IA32_UCODE_WRITE: | |
2044 | case MSR_VM_HSAVE_PA: | |
2045 | case MSR_AMD64_PATCH_LOADER: | |
2046 | case MSR_AMD64_BU_CFG2: | |
2047 | break; | |
2048 | ||
15c4a640 | 2049 | case MSR_EFER: |
b69e8cae | 2050 | return set_efer(vcpu, data); |
8f1589d9 AP |
2051 | case MSR_K7_HWCR: |
2052 | data &= ~(u64)0x40; /* ignore flush filter disable */ | |
82494028 | 2053 | data &= ~(u64)0x100; /* ignore ignne emulation enable */ |
a223c313 | 2054 | data &= ~(u64)0x8; /* ignore TLB cache disable */ |
22d48b2d | 2055 | data &= ~(u64)0x40000; /* ignore Mc status write enable */ |
8f1589d9 | 2056 | if (data != 0) { |
a737f256 CD |
2057 | vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n", |
2058 | data); | |
8f1589d9 AP |
2059 | return 1; |
2060 | } | |
15c4a640 | 2061 | break; |
f7c6d140 AP |
2062 | case MSR_FAM10H_MMIO_CONF_BASE: |
2063 | if (data != 0) { | |
a737f256 CD |
2064 | vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: " |
2065 | "0x%llx\n", data); | |
f7c6d140 AP |
2066 | return 1; |
2067 | } | |
15c4a640 | 2068 | break; |
b5e2fec0 AG |
2069 | case MSR_IA32_DEBUGCTLMSR: |
2070 | if (!data) { | |
2071 | /* We support the non-activated case already */ | |
2072 | break; | |
2073 | } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) { | |
2074 | /* Values other than LBR and BTF are vendor-specific, | |
2075 | thus reserved and should throw a #GP */ | |
2076 | return 1; | |
2077 | } | |
a737f256 CD |
2078 | vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n", |
2079 | __func__, data); | |
b5e2fec0 | 2080 | break; |
9ba075a6 | 2081 | case 0x200 ... 0x2ff: |
ff53604b | 2082 | return kvm_mtrr_set_msr(vcpu, msr, data); |
15c4a640 | 2083 | case MSR_IA32_APICBASE: |
58cb628d | 2084 | return kvm_set_apic_base(vcpu, msr_info); |
0105d1a5 GN |
2085 | case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff: |
2086 | return kvm_x2apic_msr_write(vcpu, msr, data); | |
a3e06bbe LJ |
2087 | case MSR_IA32_TSCDEADLINE: |
2088 | kvm_set_lapic_tscdeadline_msr(vcpu, data); | |
2089 | break; | |
ba904635 WA |
2090 | case MSR_IA32_TSC_ADJUST: |
2091 | if (guest_cpuid_has_tsc_adjust(vcpu)) { | |
2092 | if (!msr_info->host_initiated) { | |
d913b904 | 2093 | s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr; |
d7add054 | 2094 | adjust_tsc_offset_guest(vcpu, adj); |
ba904635 WA |
2095 | } |
2096 | vcpu->arch.ia32_tsc_adjust_msr = data; | |
2097 | } | |
2098 | break; | |
15c4a640 | 2099 | case MSR_IA32_MISC_ENABLE: |
ad312c7c | 2100 | vcpu->arch.ia32_misc_enable_msr = data; |
15c4a640 | 2101 | break; |
64d60670 PB |
2102 | case MSR_IA32_SMBASE: |
2103 | if (!msr_info->host_initiated) | |
2104 | return 1; | |
2105 | vcpu->arch.smbase = data; | |
2106 | break; | |
11c6bffa | 2107 | case MSR_KVM_WALL_CLOCK_NEW: |
18068523 GOC |
2108 | case MSR_KVM_WALL_CLOCK: |
2109 | vcpu->kvm->arch.wall_clock = data; | |
2110 | kvm_write_wall_clock(vcpu->kvm, data); | |
2111 | break; | |
11c6bffa | 2112 | case MSR_KVM_SYSTEM_TIME_NEW: |
18068523 | 2113 | case MSR_KVM_SYSTEM_TIME: { |
0b79459b | 2114 | u64 gpa_offset; |
54750f2c MT |
2115 | struct kvm_arch *ka = &vcpu->kvm->arch; |
2116 | ||
12f9a48f | 2117 | kvmclock_reset(vcpu); |
18068523 | 2118 | |
54750f2c MT |
2119 | if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) { |
2120 | bool tmp = (msr == MSR_KVM_SYSTEM_TIME); | |
2121 | ||
2122 | if (ka->boot_vcpu_runs_old_kvmclock != tmp) | |
2123 | set_bit(KVM_REQ_MASTERCLOCK_UPDATE, | |
2124 | &vcpu->requests); | |
2125 | ||
2126 | ka->boot_vcpu_runs_old_kvmclock = tmp; | |
2127 | } | |
2128 | ||
18068523 | 2129 | vcpu->arch.time = data; |
0061d53d | 2130 | kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu); |
18068523 GOC |
2131 | |
2132 | /* we verify if the enable bit is set... */ | |
2133 | if (!(data & 1)) | |
2134 | break; | |
2135 | ||
0b79459b | 2136 | gpa_offset = data & ~(PAGE_MASK | 1); |
18068523 | 2137 | |
0b79459b | 2138 | if (kvm_gfn_to_hva_cache_init(vcpu->kvm, |
8f964525 AH |
2139 | &vcpu->arch.pv_time, data & ~1ULL, |
2140 | sizeof(struct pvclock_vcpu_time_info))) | |
0b79459b AH |
2141 | vcpu->arch.pv_time_enabled = false; |
2142 | else | |
2143 | vcpu->arch.pv_time_enabled = true; | |
32cad84f | 2144 | |
18068523 GOC |
2145 | break; |
2146 | } | |
344d9588 GN |
2147 | case MSR_KVM_ASYNC_PF_EN: |
2148 | if (kvm_pv_enable_async_pf(vcpu, data)) | |
2149 | return 1; | |
2150 | break; | |
c9aaa895 GC |
2151 | case MSR_KVM_STEAL_TIME: |
2152 | ||
2153 | if (unlikely(!sched_info_on())) | |
2154 | return 1; | |
2155 | ||
2156 | if (data & KVM_STEAL_RESERVED_MASK) | |
2157 | return 1; | |
2158 | ||
2159 | if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime, | |
8f964525 AH |
2160 | data & KVM_STEAL_VALID_BITS, |
2161 | sizeof(struct kvm_steal_time))) | |
c9aaa895 GC |
2162 | return 1; |
2163 | ||
2164 | vcpu->arch.st.msr_val = data; | |
2165 | ||
2166 | if (!(data & KVM_MSR_ENABLED)) | |
2167 | break; | |
2168 | ||
c9aaa895 GC |
2169 | kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu); |
2170 | ||
2171 | break; | |
ae7a2a3f MT |
2172 | case MSR_KVM_PV_EOI_EN: |
2173 | if (kvm_lapic_enable_pv_eoi(vcpu, data)) | |
2174 | return 1; | |
2175 | break; | |
c9aaa895 | 2176 | |
890ca9ae HY |
2177 | case MSR_IA32_MCG_CTL: |
2178 | case MSR_IA32_MCG_STATUS: | |
81760dcc | 2179 | case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1: |
890ca9ae | 2180 | return set_msr_mce(vcpu, msr, data); |
71db6023 | 2181 | |
6912ac32 WH |
2182 | case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3: |
2183 | case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1: | |
2184 | pr = true; /* fall through */ | |
2185 | case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3: | |
2186 | case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1: | |
c6702c9d | 2187 | if (kvm_pmu_is_valid_msr(vcpu, msr)) |
afd80d85 | 2188 | return kvm_pmu_set_msr(vcpu, msr_info); |
5753785f GN |
2189 | |
2190 | if (pr || data != 0) | |
a737f256 CD |
2191 | vcpu_unimpl(vcpu, "disabled perfctr wrmsr: " |
2192 | "0x%x data 0x%llx\n", msr, data); | |
5753785f | 2193 | break; |
84e0cefa JS |
2194 | case MSR_K7_CLK_CTL: |
2195 | /* | |
2196 | * Ignore all writes to this no longer documented MSR. | |
2197 | * Writes are only relevant for old K7 processors, | |
2198 | * all pre-dating SVM, but a recommended workaround from | |
4a969980 | 2199 | * AMD for these chips. It is possible to specify the |
84e0cefa JS |
2200 | * affected processor models on the command line, hence |
2201 | * the need to ignore the workaround. | |
2202 | */ | |
2203 | break; | |
55cd8e5a | 2204 | case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15: |
e7d9513b AS |
2205 | case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4: |
2206 | case HV_X64_MSR_CRASH_CTL: | |
1f4b34f8 | 2207 | case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT: |
e7d9513b AS |
2208 | return kvm_hv_set_msr_common(vcpu, msr, data, |
2209 | msr_info->host_initiated); | |
91c9c3ed | 2210 | case MSR_IA32_BBL_CR_CTL3: |
2211 | /* Drop writes to this legacy MSR -- see rdmsr | |
2212 | * counterpart for further detail. | |
2213 | */ | |
a737f256 | 2214 | vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data); |
91c9c3ed | 2215 | break; |
2b036c6b BO |
2216 | case MSR_AMD64_OSVW_ID_LENGTH: |
2217 | if (!guest_cpuid_has_osvw(vcpu)) | |
2218 | return 1; | |
2219 | vcpu->arch.osvw.length = data; | |
2220 | break; | |
2221 | case MSR_AMD64_OSVW_STATUS: | |
2222 | if (!guest_cpuid_has_osvw(vcpu)) | |
2223 | return 1; | |
2224 | vcpu->arch.osvw.status = data; | |
2225 | break; | |
15c4a640 | 2226 | default: |
ffde22ac ES |
2227 | if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr)) |
2228 | return xen_hvm_config(vcpu, data); | |
c6702c9d | 2229 | if (kvm_pmu_is_valid_msr(vcpu, msr)) |
afd80d85 | 2230 | return kvm_pmu_set_msr(vcpu, msr_info); |
ed85c068 | 2231 | if (!ignore_msrs) { |
a737f256 CD |
2232 | vcpu_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n", |
2233 | msr, data); | |
ed85c068 AP |
2234 | return 1; |
2235 | } else { | |
a737f256 CD |
2236 | vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", |
2237 | msr, data); | |
ed85c068 AP |
2238 | break; |
2239 | } | |
15c4a640 CO |
2240 | } |
2241 | return 0; | |
2242 | } | |
2243 | EXPORT_SYMBOL_GPL(kvm_set_msr_common); | |
2244 | ||
2245 | ||
2246 | /* | |
2247 | * Reads an msr value (of 'msr_index') into 'pdata'. | |
2248 | * Returns 0 on success, non-0 otherwise. | |
2249 | * Assumes vcpu_load() was already called. | |
2250 | */ | |
609e36d3 | 2251 | int kvm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr) |
15c4a640 | 2252 | { |
609e36d3 | 2253 | return kvm_x86_ops->get_msr(vcpu, msr); |
15c4a640 | 2254 | } |
ff651cb6 | 2255 | EXPORT_SYMBOL_GPL(kvm_get_msr); |
15c4a640 | 2256 | |
890ca9ae | 2257 | static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata) |
15c4a640 CO |
2258 | { |
2259 | u64 data; | |
890ca9ae HY |
2260 | u64 mcg_cap = vcpu->arch.mcg_cap; |
2261 | unsigned bank_num = mcg_cap & 0xff; | |
15c4a640 CO |
2262 | |
2263 | switch (msr) { | |
15c4a640 CO |
2264 | case MSR_IA32_P5_MC_ADDR: |
2265 | case MSR_IA32_P5_MC_TYPE: | |
890ca9ae HY |
2266 | data = 0; |
2267 | break; | |
15c4a640 | 2268 | case MSR_IA32_MCG_CAP: |
890ca9ae HY |
2269 | data = vcpu->arch.mcg_cap; |
2270 | break; | |
c7ac679c | 2271 | case MSR_IA32_MCG_CTL: |
890ca9ae HY |
2272 | if (!(mcg_cap & MCG_CTL_P)) |
2273 | return 1; | |
2274 | data = vcpu->arch.mcg_ctl; | |
2275 | break; | |
2276 | case MSR_IA32_MCG_STATUS: | |
2277 | data = vcpu->arch.mcg_status; | |
2278 | break; | |
2279 | default: | |
2280 | if (msr >= MSR_IA32_MC0_CTL && | |
81760dcc | 2281 | msr < MSR_IA32_MCx_CTL(bank_num)) { |
890ca9ae HY |
2282 | u32 offset = msr - MSR_IA32_MC0_CTL; |
2283 | data = vcpu->arch.mce_banks[offset]; | |
2284 | break; | |
2285 | } | |
2286 | return 1; | |
2287 | } | |
2288 | *pdata = data; | |
2289 | return 0; | |
2290 | } | |
2291 | ||
609e36d3 | 2292 | int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info) |
890ca9ae | 2293 | { |
609e36d3 | 2294 | switch (msr_info->index) { |
890ca9ae | 2295 | case MSR_IA32_PLATFORM_ID: |
15c4a640 | 2296 | case MSR_IA32_EBL_CR_POWERON: |
b5e2fec0 AG |
2297 | case MSR_IA32_DEBUGCTLMSR: |
2298 | case MSR_IA32_LASTBRANCHFROMIP: | |
2299 | case MSR_IA32_LASTBRANCHTOIP: | |
2300 | case MSR_IA32_LASTINTFROMIP: | |
2301 | case MSR_IA32_LASTINTTOIP: | |
60af2ecd | 2302 | case MSR_K8_SYSCFG: |
3afb1121 PB |
2303 | case MSR_K8_TSEG_ADDR: |
2304 | case MSR_K8_TSEG_MASK: | |
60af2ecd | 2305 | case MSR_K7_HWCR: |
61a6bd67 | 2306 | case MSR_VM_HSAVE_PA: |
1fdbd48c | 2307 | case MSR_K8_INT_PENDING_MSG: |
c323c0e5 | 2308 | case MSR_AMD64_NB_CFG: |
f7c6d140 | 2309 | case MSR_FAM10H_MMIO_CONF_BASE: |
2e32b719 | 2310 | case MSR_AMD64_BU_CFG2: |
609e36d3 | 2311 | msr_info->data = 0; |
15c4a640 | 2312 | break; |
6912ac32 WH |
2313 | case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3: |
2314 | case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3: | |
2315 | case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1: | |
2316 | case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1: | |
c6702c9d | 2317 | if (kvm_pmu_is_valid_msr(vcpu, msr_info->index)) |
609e36d3 PB |
2318 | return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data); |
2319 | msr_info->data = 0; | |
5753785f | 2320 | break; |
742bc670 | 2321 | case MSR_IA32_UCODE_REV: |
609e36d3 | 2322 | msr_info->data = 0x100000000ULL; |
742bc670 | 2323 | break; |
9ba075a6 | 2324 | case MSR_MTRRcap: |
9ba075a6 | 2325 | case 0x200 ... 0x2ff: |
ff53604b | 2326 | return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data); |
15c4a640 | 2327 | case 0xcd: /* fsb frequency */ |
609e36d3 | 2328 | msr_info->data = 3; |
15c4a640 | 2329 | break; |
7b914098 JS |
2330 | /* |
2331 | * MSR_EBC_FREQUENCY_ID | |
2332 | * Conservative value valid for even the basic CPU models. | |
2333 | * Models 0,1: 000 in bits 23:21 indicating a bus speed of | |
2334 | * 100MHz, model 2 000 in bits 18:16 indicating 100MHz, | |
2335 | * and 266MHz for model 3, or 4. Set Core Clock | |
2336 | * Frequency to System Bus Frequency Ratio to 1 (bits | |
2337 | * 31:24) even though these are only valid for CPU | |
2338 | * models > 2, however guests may end up dividing or | |
2339 | * multiplying by zero otherwise. | |
2340 | */ | |
2341 | case MSR_EBC_FREQUENCY_ID: | |
609e36d3 | 2342 | msr_info->data = 1 << 24; |
7b914098 | 2343 | break; |
15c4a640 | 2344 | case MSR_IA32_APICBASE: |
609e36d3 | 2345 | msr_info->data = kvm_get_apic_base(vcpu); |
15c4a640 | 2346 | break; |
0105d1a5 | 2347 | case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff: |
609e36d3 | 2348 | return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data); |
0105d1a5 | 2349 | break; |
a3e06bbe | 2350 | case MSR_IA32_TSCDEADLINE: |
609e36d3 | 2351 | msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu); |
a3e06bbe | 2352 | break; |
ba904635 | 2353 | case MSR_IA32_TSC_ADJUST: |
609e36d3 | 2354 | msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr; |
ba904635 | 2355 | break; |
15c4a640 | 2356 | case MSR_IA32_MISC_ENABLE: |
609e36d3 | 2357 | msr_info->data = vcpu->arch.ia32_misc_enable_msr; |
15c4a640 | 2358 | break; |
64d60670 PB |
2359 | case MSR_IA32_SMBASE: |
2360 | if (!msr_info->host_initiated) | |
2361 | return 1; | |
2362 | msr_info->data = vcpu->arch.smbase; | |
15c4a640 | 2363 | break; |
847f0ad8 AG |
2364 | case MSR_IA32_PERF_STATUS: |
2365 | /* TSC increment by tick */ | |
609e36d3 | 2366 | msr_info->data = 1000ULL; |
847f0ad8 | 2367 | /* CPU multiplier */ |
b0996ae4 | 2368 | msr_info->data |= (((uint64_t)4ULL) << 40); |
847f0ad8 | 2369 | break; |
15c4a640 | 2370 | case MSR_EFER: |
609e36d3 | 2371 | msr_info->data = vcpu->arch.efer; |
15c4a640 | 2372 | break; |
18068523 | 2373 | case MSR_KVM_WALL_CLOCK: |
11c6bffa | 2374 | case MSR_KVM_WALL_CLOCK_NEW: |
609e36d3 | 2375 | msr_info->data = vcpu->kvm->arch.wall_clock; |
18068523 GOC |
2376 | break; |
2377 | case MSR_KVM_SYSTEM_TIME: | |
11c6bffa | 2378 | case MSR_KVM_SYSTEM_TIME_NEW: |
609e36d3 | 2379 | msr_info->data = vcpu->arch.time; |
18068523 | 2380 | break; |
344d9588 | 2381 | case MSR_KVM_ASYNC_PF_EN: |
609e36d3 | 2382 | msr_info->data = vcpu->arch.apf.msr_val; |
344d9588 | 2383 | break; |
c9aaa895 | 2384 | case MSR_KVM_STEAL_TIME: |
609e36d3 | 2385 | msr_info->data = vcpu->arch.st.msr_val; |
c9aaa895 | 2386 | break; |
1d92128f | 2387 | case MSR_KVM_PV_EOI_EN: |
609e36d3 | 2388 | msr_info->data = vcpu->arch.pv_eoi.msr_val; |
1d92128f | 2389 | break; |
890ca9ae HY |
2390 | case MSR_IA32_P5_MC_ADDR: |
2391 | case MSR_IA32_P5_MC_TYPE: | |
2392 | case MSR_IA32_MCG_CAP: | |
2393 | case MSR_IA32_MCG_CTL: | |
2394 | case MSR_IA32_MCG_STATUS: | |
81760dcc | 2395 | case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1: |
609e36d3 | 2396 | return get_msr_mce(vcpu, msr_info->index, &msr_info->data); |
84e0cefa JS |
2397 | case MSR_K7_CLK_CTL: |
2398 | /* | |
2399 | * Provide expected ramp-up count for K7. All other | |
2400 | * are set to zero, indicating minimum divisors for | |
2401 | * every field. | |
2402 | * | |
2403 | * This prevents guest kernels on AMD host with CPU | |
2404 | * type 6, model 8 and higher from exploding due to | |
2405 | * the rdmsr failing. | |
2406 | */ | |
609e36d3 | 2407 | msr_info->data = 0x20000000; |
84e0cefa | 2408 | break; |
55cd8e5a | 2409 | case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15: |
e7d9513b AS |
2410 | case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4: |
2411 | case HV_X64_MSR_CRASH_CTL: | |
1f4b34f8 | 2412 | case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT: |
e83d5887 AS |
2413 | return kvm_hv_get_msr_common(vcpu, |
2414 | msr_info->index, &msr_info->data); | |
55cd8e5a | 2415 | break; |
91c9c3ed | 2416 | case MSR_IA32_BBL_CR_CTL3: |
2417 | /* This legacy MSR exists but isn't fully documented in current | |
2418 | * silicon. It is however accessed by winxp in very narrow | |
2419 | * scenarios where it sets bit #19, itself documented as | |
2420 | * a "reserved" bit. Best effort attempt to source coherent | |
2421 | * read data here should the balance of the register be | |
2422 | * interpreted by the guest: | |
2423 | * | |
2424 | * L2 cache control register 3: 64GB range, 256KB size, | |
2425 | * enabled, latency 0x1, configured | |
2426 | */ | |
609e36d3 | 2427 | msr_info->data = 0xbe702111; |
91c9c3ed | 2428 | break; |
2b036c6b BO |
2429 | case MSR_AMD64_OSVW_ID_LENGTH: |
2430 | if (!guest_cpuid_has_osvw(vcpu)) | |
2431 | return 1; | |
609e36d3 | 2432 | msr_info->data = vcpu->arch.osvw.length; |
2b036c6b BO |
2433 | break; |
2434 | case MSR_AMD64_OSVW_STATUS: | |
2435 | if (!guest_cpuid_has_osvw(vcpu)) | |
2436 | return 1; | |
609e36d3 | 2437 | msr_info->data = vcpu->arch.osvw.status; |
2b036c6b | 2438 | break; |
15c4a640 | 2439 | default: |
c6702c9d | 2440 | if (kvm_pmu_is_valid_msr(vcpu, msr_info->index)) |
609e36d3 | 2441 | return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data); |
ed85c068 | 2442 | if (!ignore_msrs) { |
609e36d3 | 2443 | vcpu_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr_info->index); |
ed85c068 AP |
2444 | return 1; |
2445 | } else { | |
609e36d3 PB |
2446 | vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr_info->index); |
2447 | msr_info->data = 0; | |
ed85c068 AP |
2448 | } |
2449 | break; | |
15c4a640 | 2450 | } |
15c4a640 CO |
2451 | return 0; |
2452 | } | |
2453 | EXPORT_SYMBOL_GPL(kvm_get_msr_common); | |
2454 | ||
313a3dc7 CO |
2455 | /* |
2456 | * Read or write a bunch of msrs. All parameters are kernel addresses. | |
2457 | * | |
2458 | * @return number of msrs set successfully. | |
2459 | */ | |
2460 | static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs, | |
2461 | struct kvm_msr_entry *entries, | |
2462 | int (*do_msr)(struct kvm_vcpu *vcpu, | |
2463 | unsigned index, u64 *data)) | |
2464 | { | |
f656ce01 | 2465 | int i, idx; |
313a3dc7 | 2466 | |
f656ce01 | 2467 | idx = srcu_read_lock(&vcpu->kvm->srcu); |
313a3dc7 CO |
2468 | for (i = 0; i < msrs->nmsrs; ++i) |
2469 | if (do_msr(vcpu, entries[i].index, &entries[i].data)) | |
2470 | break; | |
f656ce01 | 2471 | srcu_read_unlock(&vcpu->kvm->srcu, idx); |
313a3dc7 | 2472 | |
313a3dc7 CO |
2473 | return i; |
2474 | } | |
2475 | ||
2476 | /* | |
2477 | * Read or write a bunch of msrs. Parameters are user addresses. | |
2478 | * | |
2479 | * @return number of msrs set successfully. | |
2480 | */ | |
2481 | static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs, | |
2482 | int (*do_msr)(struct kvm_vcpu *vcpu, | |
2483 | unsigned index, u64 *data), | |
2484 | int writeback) | |
2485 | { | |
2486 | struct kvm_msrs msrs; | |
2487 | struct kvm_msr_entry *entries; | |
2488 | int r, n; | |
2489 | unsigned size; | |
2490 | ||
2491 | r = -EFAULT; | |
2492 | if (copy_from_user(&msrs, user_msrs, sizeof msrs)) | |
2493 | goto out; | |
2494 | ||
2495 | r = -E2BIG; | |
2496 | if (msrs.nmsrs >= MAX_IO_MSRS) | |
2497 | goto out; | |
2498 | ||
313a3dc7 | 2499 | size = sizeof(struct kvm_msr_entry) * msrs.nmsrs; |
ff5c2c03 SL |
2500 | entries = memdup_user(user_msrs->entries, size); |
2501 | if (IS_ERR(entries)) { | |
2502 | r = PTR_ERR(entries); | |
313a3dc7 | 2503 | goto out; |
ff5c2c03 | 2504 | } |
313a3dc7 CO |
2505 | |
2506 | r = n = __msr_io(vcpu, &msrs, entries, do_msr); | |
2507 | if (r < 0) | |
2508 | goto out_free; | |
2509 | ||
2510 | r = -EFAULT; | |
2511 | if (writeback && copy_to_user(user_msrs->entries, entries, size)) | |
2512 | goto out_free; | |
2513 | ||
2514 | r = n; | |
2515 | ||
2516 | out_free: | |
7a73c028 | 2517 | kfree(entries); |
313a3dc7 CO |
2518 | out: |
2519 | return r; | |
2520 | } | |
2521 | ||
784aa3d7 | 2522 | int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext) |
018d00d2 ZX |
2523 | { |
2524 | int r; | |
2525 | ||
2526 | switch (ext) { | |
2527 | case KVM_CAP_IRQCHIP: | |
2528 | case KVM_CAP_HLT: | |
2529 | case KVM_CAP_MMU_SHADOW_CACHE_CONTROL: | |
018d00d2 | 2530 | case KVM_CAP_SET_TSS_ADDR: |
07716717 | 2531 | case KVM_CAP_EXT_CPUID: |
9c15bb1d | 2532 | case KVM_CAP_EXT_EMUL_CPUID: |
c8076604 | 2533 | case KVM_CAP_CLOCKSOURCE: |
7837699f | 2534 | case KVM_CAP_PIT: |
a28e4f5a | 2535 | case KVM_CAP_NOP_IO_DELAY: |
62d9f0db | 2536 | case KVM_CAP_MP_STATE: |
ed848624 | 2537 | case KVM_CAP_SYNC_MMU: |
a355c85c | 2538 | case KVM_CAP_USER_NMI: |
52d939a0 | 2539 | case KVM_CAP_REINJECT_CONTROL: |
4925663a | 2540 | case KVM_CAP_IRQ_INJECT_STATUS: |
d34e6b17 | 2541 | case KVM_CAP_IOEVENTFD: |
f848a5a8 | 2542 | case KVM_CAP_IOEVENTFD_NO_LENGTH: |
c5ff41ce | 2543 | case KVM_CAP_PIT2: |
e9f42757 | 2544 | case KVM_CAP_PIT_STATE2: |
b927a3ce | 2545 | case KVM_CAP_SET_IDENTITY_MAP_ADDR: |
ffde22ac | 2546 | case KVM_CAP_XEN_HVM: |
afbcf7ab | 2547 | case KVM_CAP_ADJUST_CLOCK: |
3cfc3092 | 2548 | case KVM_CAP_VCPU_EVENTS: |
55cd8e5a | 2549 | case KVM_CAP_HYPERV: |
10388a07 | 2550 | case KVM_CAP_HYPERV_VAPIC: |
c25bc163 | 2551 | case KVM_CAP_HYPERV_SPIN: |
5c919412 | 2552 | case KVM_CAP_HYPERV_SYNIC: |
ab9f4ecb | 2553 | case KVM_CAP_PCI_SEGMENT: |
a1efbe77 | 2554 | case KVM_CAP_DEBUGREGS: |
d2be1651 | 2555 | case KVM_CAP_X86_ROBUST_SINGLESTEP: |
2d5b5a66 | 2556 | case KVM_CAP_XSAVE: |
344d9588 | 2557 | case KVM_CAP_ASYNC_PF: |
92a1f12d | 2558 | case KVM_CAP_GET_TSC_KHZ: |
1c0b28c2 | 2559 | case KVM_CAP_KVMCLOCK_CTRL: |
4d8b81ab | 2560 | case KVM_CAP_READONLY_MEM: |
5f66b620 | 2561 | case KVM_CAP_HYPERV_TIME: |
100943c5 | 2562 | case KVM_CAP_IOAPIC_POLARITY_IGNORED: |
defcf51f | 2563 | case KVM_CAP_TSC_DEADLINE_TIMER: |
90de4a18 NA |
2564 | case KVM_CAP_ENABLE_CAP_VM: |
2565 | case KVM_CAP_DISABLE_QUIRKS: | |
d71ba788 | 2566 | case KVM_CAP_SET_BOOT_CPU_ID: |
49df6397 | 2567 | case KVM_CAP_SPLIT_IRQCHIP: |
2a5bab10 AW |
2568 | #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT |
2569 | case KVM_CAP_ASSIGN_DEV_IRQ: | |
2570 | case KVM_CAP_PCI_2_3: | |
2571 | #endif | |
018d00d2 ZX |
2572 | r = 1; |
2573 | break; | |
6d396b55 PB |
2574 | case KVM_CAP_X86_SMM: |
2575 | /* SMBASE is usually relocated above 1M on modern chipsets, | |
2576 | * and SMM handlers might indeed rely on 4G segment limits, | |
2577 | * so do not report SMM to be available if real mode is | |
2578 | * emulated via vm86 mode. Still, do not go to great lengths | |
2579 | * to avoid userspace's usage of the feature, because it is a | |
2580 | * fringe case that is not enabled except via specific settings | |
2581 | * of the module parameters. | |
2582 | */ | |
2583 | r = kvm_x86_ops->cpu_has_high_real_mode_segbase(); | |
2584 | break; | |
542472b5 LV |
2585 | case KVM_CAP_COALESCED_MMIO: |
2586 | r = KVM_COALESCED_MMIO_PAGE_OFFSET; | |
2587 | break; | |
774ead3a AK |
2588 | case KVM_CAP_VAPIC: |
2589 | r = !kvm_x86_ops->cpu_has_accelerated_tpr(); | |
2590 | break; | |
f725230a | 2591 | case KVM_CAP_NR_VCPUS: |
8c3ba334 SL |
2592 | r = KVM_SOFT_MAX_VCPUS; |
2593 | break; | |
2594 | case KVM_CAP_MAX_VCPUS: | |
f725230a AK |
2595 | r = KVM_MAX_VCPUS; |
2596 | break; | |
a988b910 | 2597 | case KVM_CAP_NR_MEMSLOTS: |
bbacc0c1 | 2598 | r = KVM_USER_MEM_SLOTS; |
a988b910 | 2599 | break; |
a68a6a72 MT |
2600 | case KVM_CAP_PV_MMU: /* obsolete */ |
2601 | r = 0; | |
2f333bcb | 2602 | break; |
4cee4b72 | 2603 | #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT |
62c476c7 | 2604 | case KVM_CAP_IOMMU: |
a1b60c1c | 2605 | r = iommu_present(&pci_bus_type); |
62c476c7 | 2606 | break; |
4cee4b72 | 2607 | #endif |
890ca9ae HY |
2608 | case KVM_CAP_MCE: |
2609 | r = KVM_MAX_MCE_BANKS; | |
2610 | break; | |
2d5b5a66 SY |
2611 | case KVM_CAP_XCRS: |
2612 | r = cpu_has_xsave; | |
2613 | break; | |
92a1f12d JR |
2614 | case KVM_CAP_TSC_CONTROL: |
2615 | r = kvm_has_tsc_control; | |
2616 | break; | |
018d00d2 ZX |
2617 | default: |
2618 | r = 0; | |
2619 | break; | |
2620 | } | |
2621 | return r; | |
2622 | ||
2623 | } | |
2624 | ||
043405e1 CO |
2625 | long kvm_arch_dev_ioctl(struct file *filp, |
2626 | unsigned int ioctl, unsigned long arg) | |
2627 | { | |
2628 | void __user *argp = (void __user *)arg; | |
2629 | long r; | |
2630 | ||
2631 | switch (ioctl) { | |
2632 | case KVM_GET_MSR_INDEX_LIST: { | |
2633 | struct kvm_msr_list __user *user_msr_list = argp; | |
2634 | struct kvm_msr_list msr_list; | |
2635 | unsigned n; | |
2636 | ||
2637 | r = -EFAULT; | |
2638 | if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list)) | |
2639 | goto out; | |
2640 | n = msr_list.nmsrs; | |
62ef68bb | 2641 | msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs; |
043405e1 CO |
2642 | if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list)) |
2643 | goto out; | |
2644 | r = -E2BIG; | |
e125e7b6 | 2645 | if (n < msr_list.nmsrs) |
043405e1 CO |
2646 | goto out; |
2647 | r = -EFAULT; | |
2648 | if (copy_to_user(user_msr_list->indices, &msrs_to_save, | |
2649 | num_msrs_to_save * sizeof(u32))) | |
2650 | goto out; | |
e125e7b6 | 2651 | if (copy_to_user(user_msr_list->indices + num_msrs_to_save, |
043405e1 | 2652 | &emulated_msrs, |
62ef68bb | 2653 | num_emulated_msrs * sizeof(u32))) |
043405e1 CO |
2654 | goto out; |
2655 | r = 0; | |
2656 | break; | |
2657 | } | |
9c15bb1d BP |
2658 | case KVM_GET_SUPPORTED_CPUID: |
2659 | case KVM_GET_EMULATED_CPUID: { | |
674eea0f AK |
2660 | struct kvm_cpuid2 __user *cpuid_arg = argp; |
2661 | struct kvm_cpuid2 cpuid; | |
2662 | ||
2663 | r = -EFAULT; | |
2664 | if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid)) | |
2665 | goto out; | |
9c15bb1d BP |
2666 | |
2667 | r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries, | |
2668 | ioctl); | |
674eea0f AK |
2669 | if (r) |
2670 | goto out; | |
2671 | ||
2672 | r = -EFAULT; | |
2673 | if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid)) | |
2674 | goto out; | |
2675 | r = 0; | |
2676 | break; | |
2677 | } | |
890ca9ae HY |
2678 | case KVM_X86_GET_MCE_CAP_SUPPORTED: { |
2679 | u64 mce_cap; | |
2680 | ||
2681 | mce_cap = KVM_MCE_CAP_SUPPORTED; | |
2682 | r = -EFAULT; | |
2683 | if (copy_to_user(argp, &mce_cap, sizeof mce_cap)) | |
2684 | goto out; | |
2685 | r = 0; | |
2686 | break; | |
2687 | } | |
043405e1 CO |
2688 | default: |
2689 | r = -EINVAL; | |
2690 | } | |
2691 | out: | |
2692 | return r; | |
2693 | } | |
2694 | ||
f5f48ee1 SY |
2695 | static void wbinvd_ipi(void *garbage) |
2696 | { | |
2697 | wbinvd(); | |
2698 | } | |
2699 | ||
2700 | static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu) | |
2701 | { | |
e0f0bbc5 | 2702 | return kvm_arch_has_noncoherent_dma(vcpu->kvm); |
f5f48ee1 SY |
2703 | } |
2704 | ||
2860c4b1 PB |
2705 | static inline void kvm_migrate_timers(struct kvm_vcpu *vcpu) |
2706 | { | |
2707 | set_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests); | |
2708 | } | |
2709 | ||
313a3dc7 CO |
2710 | void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu) |
2711 | { | |
f5f48ee1 SY |
2712 | /* Address WBINVD may be executed by guest */ |
2713 | if (need_emulate_wbinvd(vcpu)) { | |
2714 | if (kvm_x86_ops->has_wbinvd_exit()) | |
2715 | cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask); | |
2716 | else if (vcpu->cpu != -1 && vcpu->cpu != cpu) | |
2717 | smp_call_function_single(vcpu->cpu, | |
2718 | wbinvd_ipi, NULL, 1); | |
2719 | } | |
2720 | ||
313a3dc7 | 2721 | kvm_x86_ops->vcpu_load(vcpu, cpu); |
8f6055cb | 2722 | |
0dd6a6ed ZA |
2723 | /* Apply any externally detected TSC adjustments (due to suspend) */ |
2724 | if (unlikely(vcpu->arch.tsc_offset_adjustment)) { | |
2725 | adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment); | |
2726 | vcpu->arch.tsc_offset_adjustment = 0; | |
105b21bb | 2727 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu); |
0dd6a6ed | 2728 | } |
8f6055cb | 2729 | |
48434c20 | 2730 | if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) { |
6f526ec5 | 2731 | s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 : |
4ea1636b | 2732 | rdtsc() - vcpu->arch.last_host_tsc; |
e48672fa ZA |
2733 | if (tsc_delta < 0) |
2734 | mark_tsc_unstable("KVM discovered backwards TSC"); | |
c285545f | 2735 | if (check_tsc_unstable()) { |
07c1419a | 2736 | u64 offset = kvm_compute_tsc_offset(vcpu, |
b183aa58 ZA |
2737 | vcpu->arch.last_guest_tsc); |
2738 | kvm_x86_ops->write_tsc_offset(vcpu, offset); | |
c285545f | 2739 | vcpu->arch.tsc_catchup = 1; |
c285545f | 2740 | } |
d98d07ca MT |
2741 | /* |
2742 | * On a host with synchronized TSC, there is no need to update | |
2743 | * kvmclock on vcpu->cpu migration | |
2744 | */ | |
2745 | if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1) | |
0061d53d | 2746 | kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu); |
c285545f ZA |
2747 | if (vcpu->cpu != cpu) |
2748 | kvm_migrate_timers(vcpu); | |
e48672fa | 2749 | vcpu->cpu = cpu; |
6b7d7e76 | 2750 | } |
c9aaa895 | 2751 | |
c9aaa895 | 2752 | kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu); |
4e422bdd | 2753 | vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD; |
313a3dc7 CO |
2754 | } |
2755 | ||
2756 | void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu) | |
2757 | { | |
02daab21 | 2758 | kvm_x86_ops->vcpu_put(vcpu); |
1c11e713 | 2759 | kvm_put_guest_fpu(vcpu); |
4ea1636b | 2760 | vcpu->arch.last_host_tsc = rdtsc(); |
313a3dc7 CO |
2761 | } |
2762 | ||
313a3dc7 CO |
2763 | static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu, |
2764 | struct kvm_lapic_state *s) | |
2765 | { | |
d62caabb AS |
2766 | if (vcpu->arch.apicv_active) |
2767 | kvm_x86_ops->sync_pir_to_irr(vcpu); | |
2768 | ||
ad312c7c | 2769 | memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s); |
313a3dc7 CO |
2770 | |
2771 | return 0; | |
2772 | } | |
2773 | ||
2774 | static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu, | |
2775 | struct kvm_lapic_state *s) | |
2776 | { | |
64eb0620 | 2777 | kvm_apic_post_state_restore(vcpu, s); |
cb142eb7 | 2778 | update_cr8_intercept(vcpu); |
313a3dc7 CO |
2779 | |
2780 | return 0; | |
2781 | } | |
2782 | ||
127a457a MG |
2783 | static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu) |
2784 | { | |
2785 | return (!lapic_in_kernel(vcpu) || | |
2786 | kvm_apic_accept_pic_intr(vcpu)); | |
2787 | } | |
2788 | ||
782d422b MG |
2789 | /* |
2790 | * if userspace requested an interrupt window, check that the | |
2791 | * interrupt window is open. | |
2792 | * | |
2793 | * No need to exit to userspace if we already have an interrupt queued. | |
2794 | */ | |
2795 | static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu) | |
2796 | { | |
2797 | return kvm_arch_interrupt_allowed(vcpu) && | |
2798 | !kvm_cpu_has_interrupt(vcpu) && | |
2799 | !kvm_event_needs_reinjection(vcpu) && | |
2800 | kvm_cpu_accept_dm_intr(vcpu); | |
2801 | } | |
2802 | ||
f77bc6a4 ZX |
2803 | static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, |
2804 | struct kvm_interrupt *irq) | |
2805 | { | |
02cdb50f | 2806 | if (irq->irq >= KVM_NR_INTERRUPTS) |
f77bc6a4 | 2807 | return -EINVAL; |
1c1a9ce9 SR |
2808 | |
2809 | if (!irqchip_in_kernel(vcpu->kvm)) { | |
2810 | kvm_queue_interrupt(vcpu, irq->irq, false); | |
2811 | kvm_make_request(KVM_REQ_EVENT, vcpu); | |
2812 | return 0; | |
2813 | } | |
2814 | ||
2815 | /* | |
2816 | * With in-kernel LAPIC, we only use this to inject EXTINT, so | |
2817 | * fail for in-kernel 8259. | |
2818 | */ | |
2819 | if (pic_in_kernel(vcpu->kvm)) | |
f77bc6a4 | 2820 | return -ENXIO; |
f77bc6a4 | 2821 | |
1c1a9ce9 SR |
2822 | if (vcpu->arch.pending_external_vector != -1) |
2823 | return -EEXIST; | |
f77bc6a4 | 2824 | |
1c1a9ce9 | 2825 | vcpu->arch.pending_external_vector = irq->irq; |
934bf653 | 2826 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
f77bc6a4 ZX |
2827 | return 0; |
2828 | } | |
2829 | ||
c4abb7c9 JK |
2830 | static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu) |
2831 | { | |
c4abb7c9 | 2832 | kvm_inject_nmi(vcpu); |
c4abb7c9 JK |
2833 | |
2834 | return 0; | |
2835 | } | |
2836 | ||
f077825a PB |
2837 | static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu) |
2838 | { | |
64d60670 PB |
2839 | kvm_make_request(KVM_REQ_SMI, vcpu); |
2840 | ||
f077825a PB |
2841 | return 0; |
2842 | } | |
2843 | ||
b209749f AK |
2844 | static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu, |
2845 | struct kvm_tpr_access_ctl *tac) | |
2846 | { | |
2847 | if (tac->flags) | |
2848 | return -EINVAL; | |
2849 | vcpu->arch.tpr_access_reporting = !!tac->enabled; | |
2850 | return 0; | |
2851 | } | |
2852 | ||
890ca9ae HY |
2853 | static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu, |
2854 | u64 mcg_cap) | |
2855 | { | |
2856 | int r; | |
2857 | unsigned bank_num = mcg_cap & 0xff, bank; | |
2858 | ||
2859 | r = -EINVAL; | |
a9e38c3e | 2860 | if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS) |
890ca9ae HY |
2861 | goto out; |
2862 | if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000)) | |
2863 | goto out; | |
2864 | r = 0; | |
2865 | vcpu->arch.mcg_cap = mcg_cap; | |
2866 | /* Init IA32_MCG_CTL to all 1s */ | |
2867 | if (mcg_cap & MCG_CTL_P) | |
2868 | vcpu->arch.mcg_ctl = ~(u64)0; | |
2869 | /* Init IA32_MCi_CTL to all 1s */ | |
2870 | for (bank = 0; bank < bank_num; bank++) | |
2871 | vcpu->arch.mce_banks[bank*4] = ~(u64)0; | |
2872 | out: | |
2873 | return r; | |
2874 | } | |
2875 | ||
2876 | static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu, | |
2877 | struct kvm_x86_mce *mce) | |
2878 | { | |
2879 | u64 mcg_cap = vcpu->arch.mcg_cap; | |
2880 | unsigned bank_num = mcg_cap & 0xff; | |
2881 | u64 *banks = vcpu->arch.mce_banks; | |
2882 | ||
2883 | if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL)) | |
2884 | return -EINVAL; | |
2885 | /* | |
2886 | * if IA32_MCG_CTL is not all 1s, the uncorrected error | |
2887 | * reporting is disabled | |
2888 | */ | |
2889 | if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) && | |
2890 | vcpu->arch.mcg_ctl != ~(u64)0) | |
2891 | return 0; | |
2892 | banks += 4 * mce->bank; | |
2893 | /* | |
2894 | * if IA32_MCi_CTL is not all 1s, the uncorrected error | |
2895 | * reporting is disabled for the bank | |
2896 | */ | |
2897 | if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0) | |
2898 | return 0; | |
2899 | if (mce->status & MCI_STATUS_UC) { | |
2900 | if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) || | |
fc78f519 | 2901 | !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) { |
a8eeb04a | 2902 | kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu); |
890ca9ae HY |
2903 | return 0; |
2904 | } | |
2905 | if (banks[1] & MCI_STATUS_VAL) | |
2906 | mce->status |= MCI_STATUS_OVER; | |
2907 | banks[2] = mce->addr; | |
2908 | banks[3] = mce->misc; | |
2909 | vcpu->arch.mcg_status = mce->mcg_status; | |
2910 | banks[1] = mce->status; | |
2911 | kvm_queue_exception(vcpu, MC_VECTOR); | |
2912 | } else if (!(banks[1] & MCI_STATUS_VAL) | |
2913 | || !(banks[1] & MCI_STATUS_UC)) { | |
2914 | if (banks[1] & MCI_STATUS_VAL) | |
2915 | mce->status |= MCI_STATUS_OVER; | |
2916 | banks[2] = mce->addr; | |
2917 | banks[3] = mce->misc; | |
2918 | banks[1] = mce->status; | |
2919 | } else | |
2920 | banks[1] |= MCI_STATUS_OVER; | |
2921 | return 0; | |
2922 | } | |
2923 | ||
3cfc3092 JK |
2924 | static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu, |
2925 | struct kvm_vcpu_events *events) | |
2926 | { | |
7460fb4a | 2927 | process_nmi(vcpu); |
03b82a30 JK |
2928 | events->exception.injected = |
2929 | vcpu->arch.exception.pending && | |
2930 | !kvm_exception_is_soft(vcpu->arch.exception.nr); | |
3cfc3092 JK |
2931 | events->exception.nr = vcpu->arch.exception.nr; |
2932 | events->exception.has_error_code = vcpu->arch.exception.has_error_code; | |
97e69aa6 | 2933 | events->exception.pad = 0; |
3cfc3092 JK |
2934 | events->exception.error_code = vcpu->arch.exception.error_code; |
2935 | ||
03b82a30 JK |
2936 | events->interrupt.injected = |
2937 | vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft; | |
3cfc3092 | 2938 | events->interrupt.nr = vcpu->arch.interrupt.nr; |
03b82a30 | 2939 | events->interrupt.soft = 0; |
37ccdcbe | 2940 | events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu); |
3cfc3092 JK |
2941 | |
2942 | events->nmi.injected = vcpu->arch.nmi_injected; | |
7460fb4a | 2943 | events->nmi.pending = vcpu->arch.nmi_pending != 0; |
3cfc3092 | 2944 | events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu); |
97e69aa6 | 2945 | events->nmi.pad = 0; |
3cfc3092 | 2946 | |
66450a21 | 2947 | events->sipi_vector = 0; /* never valid when reporting to user space */ |
3cfc3092 | 2948 | |
f077825a PB |
2949 | events->smi.smm = is_smm(vcpu); |
2950 | events->smi.pending = vcpu->arch.smi_pending; | |
2951 | events->smi.smm_inside_nmi = | |
2952 | !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK); | |
2953 | events->smi.latched_init = kvm_lapic_latched_init(vcpu); | |
2954 | ||
dab4b911 | 2955 | events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING |
f077825a PB |
2956 | | KVM_VCPUEVENT_VALID_SHADOW |
2957 | | KVM_VCPUEVENT_VALID_SMM); | |
97e69aa6 | 2958 | memset(&events->reserved, 0, sizeof(events->reserved)); |
3cfc3092 JK |
2959 | } |
2960 | ||
2961 | static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu, | |
2962 | struct kvm_vcpu_events *events) | |
2963 | { | |
dab4b911 | 2964 | if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING |
48005f64 | 2965 | | KVM_VCPUEVENT_VALID_SIPI_VECTOR |
f077825a PB |
2966 | | KVM_VCPUEVENT_VALID_SHADOW |
2967 | | KVM_VCPUEVENT_VALID_SMM)) | |
3cfc3092 JK |
2968 | return -EINVAL; |
2969 | ||
7460fb4a | 2970 | process_nmi(vcpu); |
3cfc3092 JK |
2971 | vcpu->arch.exception.pending = events->exception.injected; |
2972 | vcpu->arch.exception.nr = events->exception.nr; | |
2973 | vcpu->arch.exception.has_error_code = events->exception.has_error_code; | |
2974 | vcpu->arch.exception.error_code = events->exception.error_code; | |
2975 | ||
2976 | vcpu->arch.interrupt.pending = events->interrupt.injected; | |
2977 | vcpu->arch.interrupt.nr = events->interrupt.nr; | |
2978 | vcpu->arch.interrupt.soft = events->interrupt.soft; | |
48005f64 JK |
2979 | if (events->flags & KVM_VCPUEVENT_VALID_SHADOW) |
2980 | kvm_x86_ops->set_interrupt_shadow(vcpu, | |
2981 | events->interrupt.shadow); | |
3cfc3092 JK |
2982 | |
2983 | vcpu->arch.nmi_injected = events->nmi.injected; | |
dab4b911 JK |
2984 | if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING) |
2985 | vcpu->arch.nmi_pending = events->nmi.pending; | |
3cfc3092 JK |
2986 | kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked); |
2987 | ||
66450a21 | 2988 | if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR && |
bce87cce | 2989 | lapic_in_kernel(vcpu)) |
66450a21 | 2990 | vcpu->arch.apic->sipi_vector = events->sipi_vector; |
3cfc3092 | 2991 | |
f077825a PB |
2992 | if (events->flags & KVM_VCPUEVENT_VALID_SMM) { |
2993 | if (events->smi.smm) | |
2994 | vcpu->arch.hflags |= HF_SMM_MASK; | |
2995 | else | |
2996 | vcpu->arch.hflags &= ~HF_SMM_MASK; | |
2997 | vcpu->arch.smi_pending = events->smi.pending; | |
2998 | if (events->smi.smm_inside_nmi) | |
2999 | vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK; | |
3000 | else | |
3001 | vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK; | |
bce87cce | 3002 | if (lapic_in_kernel(vcpu)) { |
f077825a PB |
3003 | if (events->smi.latched_init) |
3004 | set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events); | |
3005 | else | |
3006 | clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events); | |
3007 | } | |
3008 | } | |
3009 | ||
3842d135 AK |
3010 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
3011 | ||
3cfc3092 JK |
3012 | return 0; |
3013 | } | |
3014 | ||
a1efbe77 JK |
3015 | static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu, |
3016 | struct kvm_debugregs *dbgregs) | |
3017 | { | |
73aaf249 JK |
3018 | unsigned long val; |
3019 | ||
a1efbe77 | 3020 | memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db)); |
16f8a6f9 | 3021 | kvm_get_dr(vcpu, 6, &val); |
73aaf249 | 3022 | dbgregs->dr6 = val; |
a1efbe77 JK |
3023 | dbgregs->dr7 = vcpu->arch.dr7; |
3024 | dbgregs->flags = 0; | |
97e69aa6 | 3025 | memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved)); |
a1efbe77 JK |
3026 | } |
3027 | ||
3028 | static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu, | |
3029 | struct kvm_debugregs *dbgregs) | |
3030 | { | |
3031 | if (dbgregs->flags) | |
3032 | return -EINVAL; | |
3033 | ||
a1efbe77 | 3034 | memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db)); |
ae561ede | 3035 | kvm_update_dr0123(vcpu); |
a1efbe77 | 3036 | vcpu->arch.dr6 = dbgregs->dr6; |
73aaf249 | 3037 | kvm_update_dr6(vcpu); |
a1efbe77 | 3038 | vcpu->arch.dr7 = dbgregs->dr7; |
9926c9fd | 3039 | kvm_update_dr7(vcpu); |
a1efbe77 | 3040 | |
a1efbe77 JK |
3041 | return 0; |
3042 | } | |
3043 | ||
df1daba7 PB |
3044 | #define XSTATE_COMPACTION_ENABLED (1ULL << 63) |
3045 | ||
3046 | static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu) | |
3047 | { | |
c47ada30 | 3048 | struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave; |
400e4b20 | 3049 | u64 xstate_bv = xsave->header.xfeatures; |
df1daba7 PB |
3050 | u64 valid; |
3051 | ||
3052 | /* | |
3053 | * Copy legacy XSAVE area, to avoid complications with CPUID | |
3054 | * leaves 0 and 1 in the loop below. | |
3055 | */ | |
3056 | memcpy(dest, xsave, XSAVE_HDR_OFFSET); | |
3057 | ||
3058 | /* Set XSTATE_BV */ | |
3059 | *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv; | |
3060 | ||
3061 | /* | |
3062 | * Copy each region from the possibly compacted offset to the | |
3063 | * non-compacted offset. | |
3064 | */ | |
d91cab78 | 3065 | valid = xstate_bv & ~XFEATURE_MASK_FPSSE; |
df1daba7 PB |
3066 | while (valid) { |
3067 | u64 feature = valid & -valid; | |
3068 | int index = fls64(feature) - 1; | |
3069 | void *src = get_xsave_addr(xsave, feature); | |
3070 | ||
3071 | if (src) { | |
3072 | u32 size, offset, ecx, edx; | |
3073 | cpuid_count(XSTATE_CPUID, index, | |
3074 | &size, &offset, &ecx, &edx); | |
3075 | memcpy(dest + offset, src, size); | |
3076 | } | |
3077 | ||
3078 | valid -= feature; | |
3079 | } | |
3080 | } | |
3081 | ||
3082 | static void load_xsave(struct kvm_vcpu *vcpu, u8 *src) | |
3083 | { | |
c47ada30 | 3084 | struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave; |
df1daba7 PB |
3085 | u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET); |
3086 | u64 valid; | |
3087 | ||
3088 | /* | |
3089 | * Copy legacy XSAVE area, to avoid complications with CPUID | |
3090 | * leaves 0 and 1 in the loop below. | |
3091 | */ | |
3092 | memcpy(xsave, src, XSAVE_HDR_OFFSET); | |
3093 | ||
3094 | /* Set XSTATE_BV and possibly XCOMP_BV. */ | |
400e4b20 | 3095 | xsave->header.xfeatures = xstate_bv; |
df1daba7 | 3096 | if (cpu_has_xsaves) |
3a54450b | 3097 | xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED; |
df1daba7 PB |
3098 | |
3099 | /* | |
3100 | * Copy each region from the non-compacted offset to the | |
3101 | * possibly compacted offset. | |
3102 | */ | |
d91cab78 | 3103 | valid = xstate_bv & ~XFEATURE_MASK_FPSSE; |
df1daba7 PB |
3104 | while (valid) { |
3105 | u64 feature = valid & -valid; | |
3106 | int index = fls64(feature) - 1; | |
3107 | void *dest = get_xsave_addr(xsave, feature); | |
3108 | ||
3109 | if (dest) { | |
3110 | u32 size, offset, ecx, edx; | |
3111 | cpuid_count(XSTATE_CPUID, index, | |
3112 | &size, &offset, &ecx, &edx); | |
3113 | memcpy(dest, src + offset, size); | |
ee4100da | 3114 | } |
df1daba7 PB |
3115 | |
3116 | valid -= feature; | |
3117 | } | |
3118 | } | |
3119 | ||
2d5b5a66 SY |
3120 | static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu, |
3121 | struct kvm_xsave *guest_xsave) | |
3122 | { | |
4344ee98 | 3123 | if (cpu_has_xsave) { |
df1daba7 PB |
3124 | memset(guest_xsave, 0, sizeof(struct kvm_xsave)); |
3125 | fill_xsave((u8 *) guest_xsave->region, vcpu); | |
4344ee98 | 3126 | } else { |
2d5b5a66 | 3127 | memcpy(guest_xsave->region, |
7366ed77 | 3128 | &vcpu->arch.guest_fpu.state.fxsave, |
c47ada30 | 3129 | sizeof(struct fxregs_state)); |
2d5b5a66 | 3130 | *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] = |
d91cab78 | 3131 | XFEATURE_MASK_FPSSE; |
2d5b5a66 SY |
3132 | } |
3133 | } | |
3134 | ||
3135 | static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu, | |
3136 | struct kvm_xsave *guest_xsave) | |
3137 | { | |
3138 | u64 xstate_bv = | |
3139 | *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)]; | |
3140 | ||
d7876f1b PB |
3141 | if (cpu_has_xsave) { |
3142 | /* | |
3143 | * Here we allow setting states that are not present in | |
3144 | * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility | |
3145 | * with old userspace. | |
3146 | */ | |
4ff41732 | 3147 | if (xstate_bv & ~kvm_supported_xcr0()) |
d7876f1b | 3148 | return -EINVAL; |
df1daba7 | 3149 | load_xsave(vcpu, (u8 *)guest_xsave->region); |
d7876f1b | 3150 | } else { |
d91cab78 | 3151 | if (xstate_bv & ~XFEATURE_MASK_FPSSE) |
2d5b5a66 | 3152 | return -EINVAL; |
7366ed77 | 3153 | memcpy(&vcpu->arch.guest_fpu.state.fxsave, |
c47ada30 | 3154 | guest_xsave->region, sizeof(struct fxregs_state)); |
2d5b5a66 SY |
3155 | } |
3156 | return 0; | |
3157 | } | |
3158 | ||
3159 | static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu, | |
3160 | struct kvm_xcrs *guest_xcrs) | |
3161 | { | |
3162 | if (!cpu_has_xsave) { | |
3163 | guest_xcrs->nr_xcrs = 0; | |
3164 | return; | |
3165 | } | |
3166 | ||
3167 | guest_xcrs->nr_xcrs = 1; | |
3168 | guest_xcrs->flags = 0; | |
3169 | guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK; | |
3170 | guest_xcrs->xcrs[0].value = vcpu->arch.xcr0; | |
3171 | } | |
3172 | ||
3173 | static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu, | |
3174 | struct kvm_xcrs *guest_xcrs) | |
3175 | { | |
3176 | int i, r = 0; | |
3177 | ||
3178 | if (!cpu_has_xsave) | |
3179 | return -EINVAL; | |
3180 | ||
3181 | if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags) | |
3182 | return -EINVAL; | |
3183 | ||
3184 | for (i = 0; i < guest_xcrs->nr_xcrs; i++) | |
3185 | /* Only support XCR0 currently */ | |
c67a04cb | 3186 | if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) { |
2d5b5a66 | 3187 | r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK, |
c67a04cb | 3188 | guest_xcrs->xcrs[i].value); |
2d5b5a66 SY |
3189 | break; |
3190 | } | |
3191 | if (r) | |
3192 | r = -EINVAL; | |
3193 | return r; | |
3194 | } | |
3195 | ||
1c0b28c2 EM |
3196 | /* |
3197 | * kvm_set_guest_paused() indicates to the guest kernel that it has been | |
3198 | * stopped by the hypervisor. This function will be called from the host only. | |
3199 | * EINVAL is returned when the host attempts to set the flag for a guest that | |
3200 | * does not support pv clocks. | |
3201 | */ | |
3202 | static int kvm_set_guest_paused(struct kvm_vcpu *vcpu) | |
3203 | { | |
0b79459b | 3204 | if (!vcpu->arch.pv_time_enabled) |
1c0b28c2 | 3205 | return -EINVAL; |
51d59c6b | 3206 | vcpu->arch.pvclock_set_guest_stopped_request = true; |
1c0b28c2 EM |
3207 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu); |
3208 | return 0; | |
3209 | } | |
3210 | ||
5c919412 AS |
3211 | static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu, |
3212 | struct kvm_enable_cap *cap) | |
3213 | { | |
3214 | if (cap->flags) | |
3215 | return -EINVAL; | |
3216 | ||
3217 | switch (cap->cap) { | |
3218 | case KVM_CAP_HYPERV_SYNIC: | |
3219 | return kvm_hv_activate_synic(vcpu); | |
3220 | default: | |
3221 | return -EINVAL; | |
3222 | } | |
3223 | } | |
3224 | ||
313a3dc7 CO |
3225 | long kvm_arch_vcpu_ioctl(struct file *filp, |
3226 | unsigned int ioctl, unsigned long arg) | |
3227 | { | |
3228 | struct kvm_vcpu *vcpu = filp->private_data; | |
3229 | void __user *argp = (void __user *)arg; | |
3230 | int r; | |
d1ac91d8 AK |
3231 | union { |
3232 | struct kvm_lapic_state *lapic; | |
3233 | struct kvm_xsave *xsave; | |
3234 | struct kvm_xcrs *xcrs; | |
3235 | void *buffer; | |
3236 | } u; | |
3237 | ||
3238 | u.buffer = NULL; | |
313a3dc7 CO |
3239 | switch (ioctl) { |
3240 | case KVM_GET_LAPIC: { | |
2204ae3c | 3241 | r = -EINVAL; |
bce87cce | 3242 | if (!lapic_in_kernel(vcpu)) |
2204ae3c | 3243 | goto out; |
d1ac91d8 | 3244 | u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL); |
313a3dc7 | 3245 | |
b772ff36 | 3246 | r = -ENOMEM; |
d1ac91d8 | 3247 | if (!u.lapic) |
b772ff36 | 3248 | goto out; |
d1ac91d8 | 3249 | r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic); |
313a3dc7 CO |
3250 | if (r) |
3251 | goto out; | |
3252 | r = -EFAULT; | |
d1ac91d8 | 3253 | if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state))) |
313a3dc7 CO |
3254 | goto out; |
3255 | r = 0; | |
3256 | break; | |
3257 | } | |
3258 | case KVM_SET_LAPIC: { | |
2204ae3c | 3259 | r = -EINVAL; |
bce87cce | 3260 | if (!lapic_in_kernel(vcpu)) |
2204ae3c | 3261 | goto out; |
ff5c2c03 | 3262 | u.lapic = memdup_user(argp, sizeof(*u.lapic)); |
18595411 GC |
3263 | if (IS_ERR(u.lapic)) |
3264 | return PTR_ERR(u.lapic); | |
ff5c2c03 | 3265 | |
d1ac91d8 | 3266 | r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic); |
313a3dc7 CO |
3267 | break; |
3268 | } | |
f77bc6a4 ZX |
3269 | case KVM_INTERRUPT: { |
3270 | struct kvm_interrupt irq; | |
3271 | ||
3272 | r = -EFAULT; | |
3273 | if (copy_from_user(&irq, argp, sizeof irq)) | |
3274 | goto out; | |
3275 | r = kvm_vcpu_ioctl_interrupt(vcpu, &irq); | |
f77bc6a4 ZX |
3276 | break; |
3277 | } | |
c4abb7c9 JK |
3278 | case KVM_NMI: { |
3279 | r = kvm_vcpu_ioctl_nmi(vcpu); | |
c4abb7c9 JK |
3280 | break; |
3281 | } | |
f077825a PB |
3282 | case KVM_SMI: { |
3283 | r = kvm_vcpu_ioctl_smi(vcpu); | |
3284 | break; | |
3285 | } | |
313a3dc7 CO |
3286 | case KVM_SET_CPUID: { |
3287 | struct kvm_cpuid __user *cpuid_arg = argp; | |
3288 | struct kvm_cpuid cpuid; | |
3289 | ||
3290 | r = -EFAULT; | |
3291 | if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid)) | |
3292 | goto out; | |
3293 | r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries); | |
313a3dc7 CO |
3294 | break; |
3295 | } | |
07716717 DK |
3296 | case KVM_SET_CPUID2: { |
3297 | struct kvm_cpuid2 __user *cpuid_arg = argp; | |
3298 | struct kvm_cpuid2 cpuid; | |
3299 | ||
3300 | r = -EFAULT; | |
3301 | if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid)) | |
3302 | goto out; | |
3303 | r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid, | |
19355475 | 3304 | cpuid_arg->entries); |
07716717 DK |
3305 | break; |
3306 | } | |
3307 | case KVM_GET_CPUID2: { | |
3308 | struct kvm_cpuid2 __user *cpuid_arg = argp; | |
3309 | struct kvm_cpuid2 cpuid; | |
3310 | ||
3311 | r = -EFAULT; | |
3312 | if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid)) | |
3313 | goto out; | |
3314 | r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid, | |
19355475 | 3315 | cpuid_arg->entries); |
07716717 DK |
3316 | if (r) |
3317 | goto out; | |
3318 | r = -EFAULT; | |
3319 | if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid)) | |
3320 | goto out; | |
3321 | r = 0; | |
3322 | break; | |
3323 | } | |
313a3dc7 | 3324 | case KVM_GET_MSRS: |
609e36d3 | 3325 | r = msr_io(vcpu, argp, do_get_msr, 1); |
313a3dc7 CO |
3326 | break; |
3327 | case KVM_SET_MSRS: | |
3328 | r = msr_io(vcpu, argp, do_set_msr, 0); | |
3329 | break; | |
b209749f AK |
3330 | case KVM_TPR_ACCESS_REPORTING: { |
3331 | struct kvm_tpr_access_ctl tac; | |
3332 | ||
3333 | r = -EFAULT; | |
3334 | if (copy_from_user(&tac, argp, sizeof tac)) | |
3335 | goto out; | |
3336 | r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac); | |
3337 | if (r) | |
3338 | goto out; | |
3339 | r = -EFAULT; | |
3340 | if (copy_to_user(argp, &tac, sizeof tac)) | |
3341 | goto out; | |
3342 | r = 0; | |
3343 | break; | |
3344 | }; | |
b93463aa AK |
3345 | case KVM_SET_VAPIC_ADDR: { |
3346 | struct kvm_vapic_addr va; | |
3347 | ||
3348 | r = -EINVAL; | |
35754c98 | 3349 | if (!lapic_in_kernel(vcpu)) |
b93463aa AK |
3350 | goto out; |
3351 | r = -EFAULT; | |
3352 | if (copy_from_user(&va, argp, sizeof va)) | |
3353 | goto out; | |
fda4e2e8 | 3354 | r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr); |
b93463aa AK |
3355 | break; |
3356 | } | |
890ca9ae HY |
3357 | case KVM_X86_SETUP_MCE: { |
3358 | u64 mcg_cap; | |
3359 | ||
3360 | r = -EFAULT; | |
3361 | if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap)) | |
3362 | goto out; | |
3363 | r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap); | |
3364 | break; | |
3365 | } | |
3366 | case KVM_X86_SET_MCE: { | |
3367 | struct kvm_x86_mce mce; | |
3368 | ||
3369 | r = -EFAULT; | |
3370 | if (copy_from_user(&mce, argp, sizeof mce)) | |
3371 | goto out; | |
3372 | r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce); | |
3373 | break; | |
3374 | } | |
3cfc3092 JK |
3375 | case KVM_GET_VCPU_EVENTS: { |
3376 | struct kvm_vcpu_events events; | |
3377 | ||
3378 | kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events); | |
3379 | ||
3380 | r = -EFAULT; | |
3381 | if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events))) | |
3382 | break; | |
3383 | r = 0; | |
3384 | break; | |
3385 | } | |
3386 | case KVM_SET_VCPU_EVENTS: { | |
3387 | struct kvm_vcpu_events events; | |
3388 | ||
3389 | r = -EFAULT; | |
3390 | if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events))) | |
3391 | break; | |
3392 | ||
3393 | r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events); | |
3394 | break; | |
3395 | } | |
a1efbe77 JK |
3396 | case KVM_GET_DEBUGREGS: { |
3397 | struct kvm_debugregs dbgregs; | |
3398 | ||
3399 | kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs); | |
3400 | ||
3401 | r = -EFAULT; | |
3402 | if (copy_to_user(argp, &dbgregs, | |
3403 | sizeof(struct kvm_debugregs))) | |
3404 | break; | |
3405 | r = 0; | |
3406 | break; | |
3407 | } | |
3408 | case KVM_SET_DEBUGREGS: { | |
3409 | struct kvm_debugregs dbgregs; | |
3410 | ||
3411 | r = -EFAULT; | |
3412 | if (copy_from_user(&dbgregs, argp, | |
3413 | sizeof(struct kvm_debugregs))) | |
3414 | break; | |
3415 | ||
3416 | r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs); | |
3417 | break; | |
3418 | } | |
2d5b5a66 | 3419 | case KVM_GET_XSAVE: { |
d1ac91d8 | 3420 | u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL); |
2d5b5a66 | 3421 | r = -ENOMEM; |
d1ac91d8 | 3422 | if (!u.xsave) |
2d5b5a66 SY |
3423 | break; |
3424 | ||
d1ac91d8 | 3425 | kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave); |
2d5b5a66 SY |
3426 | |
3427 | r = -EFAULT; | |
d1ac91d8 | 3428 | if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave))) |
2d5b5a66 SY |
3429 | break; |
3430 | r = 0; | |
3431 | break; | |
3432 | } | |
3433 | case KVM_SET_XSAVE: { | |
ff5c2c03 | 3434 | u.xsave = memdup_user(argp, sizeof(*u.xsave)); |
18595411 GC |
3435 | if (IS_ERR(u.xsave)) |
3436 | return PTR_ERR(u.xsave); | |
2d5b5a66 | 3437 | |
d1ac91d8 | 3438 | r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave); |
2d5b5a66 SY |
3439 | break; |
3440 | } | |
3441 | case KVM_GET_XCRS: { | |
d1ac91d8 | 3442 | u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL); |
2d5b5a66 | 3443 | r = -ENOMEM; |
d1ac91d8 | 3444 | if (!u.xcrs) |
2d5b5a66 SY |
3445 | break; |
3446 | ||
d1ac91d8 | 3447 | kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs); |
2d5b5a66 SY |
3448 | |
3449 | r = -EFAULT; | |
d1ac91d8 | 3450 | if (copy_to_user(argp, u.xcrs, |
2d5b5a66 SY |
3451 | sizeof(struct kvm_xcrs))) |
3452 | break; | |
3453 | r = 0; | |
3454 | break; | |
3455 | } | |
3456 | case KVM_SET_XCRS: { | |
ff5c2c03 | 3457 | u.xcrs = memdup_user(argp, sizeof(*u.xcrs)); |
18595411 GC |
3458 | if (IS_ERR(u.xcrs)) |
3459 | return PTR_ERR(u.xcrs); | |
2d5b5a66 | 3460 | |
d1ac91d8 | 3461 | r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs); |
2d5b5a66 SY |
3462 | break; |
3463 | } | |
92a1f12d JR |
3464 | case KVM_SET_TSC_KHZ: { |
3465 | u32 user_tsc_khz; | |
3466 | ||
3467 | r = -EINVAL; | |
92a1f12d JR |
3468 | user_tsc_khz = (u32)arg; |
3469 | ||
3470 | if (user_tsc_khz >= kvm_max_guest_tsc_khz) | |
3471 | goto out; | |
3472 | ||
cc578287 ZA |
3473 | if (user_tsc_khz == 0) |
3474 | user_tsc_khz = tsc_khz; | |
3475 | ||
381d585c HZ |
3476 | if (!kvm_set_tsc_khz(vcpu, user_tsc_khz)) |
3477 | r = 0; | |
92a1f12d | 3478 | |
92a1f12d JR |
3479 | goto out; |
3480 | } | |
3481 | case KVM_GET_TSC_KHZ: { | |
cc578287 | 3482 | r = vcpu->arch.virtual_tsc_khz; |
92a1f12d JR |
3483 | goto out; |
3484 | } | |
1c0b28c2 EM |
3485 | case KVM_KVMCLOCK_CTRL: { |
3486 | r = kvm_set_guest_paused(vcpu); | |
3487 | goto out; | |
3488 | } | |
5c919412 AS |
3489 | case KVM_ENABLE_CAP: { |
3490 | struct kvm_enable_cap cap; | |
3491 | ||
3492 | r = -EFAULT; | |
3493 | if (copy_from_user(&cap, argp, sizeof(cap))) | |
3494 | goto out; | |
3495 | r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap); | |
3496 | break; | |
3497 | } | |
313a3dc7 CO |
3498 | default: |
3499 | r = -EINVAL; | |
3500 | } | |
3501 | out: | |
d1ac91d8 | 3502 | kfree(u.buffer); |
313a3dc7 CO |
3503 | return r; |
3504 | } | |
3505 | ||
5b1c1493 CO |
3506 | int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf) |
3507 | { | |
3508 | return VM_FAULT_SIGBUS; | |
3509 | } | |
3510 | ||
1fe779f8 CO |
3511 | static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr) |
3512 | { | |
3513 | int ret; | |
3514 | ||
3515 | if (addr > (unsigned int)(-3 * PAGE_SIZE)) | |
951179ce | 3516 | return -EINVAL; |
1fe779f8 CO |
3517 | ret = kvm_x86_ops->set_tss_addr(kvm, addr); |
3518 | return ret; | |
3519 | } | |
3520 | ||
b927a3ce SY |
3521 | static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm, |
3522 | u64 ident_addr) | |
3523 | { | |
3524 | kvm->arch.ept_identity_map_addr = ident_addr; | |
3525 | return 0; | |
3526 | } | |
3527 | ||
1fe779f8 CO |
3528 | static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm, |
3529 | u32 kvm_nr_mmu_pages) | |
3530 | { | |
3531 | if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES) | |
3532 | return -EINVAL; | |
3533 | ||
79fac95e | 3534 | mutex_lock(&kvm->slots_lock); |
1fe779f8 CO |
3535 | |
3536 | kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages); | |
f05e70ac | 3537 | kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages; |
1fe779f8 | 3538 | |
79fac95e | 3539 | mutex_unlock(&kvm->slots_lock); |
1fe779f8 CO |
3540 | return 0; |
3541 | } | |
3542 | ||
3543 | static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm) | |
3544 | { | |
39de71ec | 3545 | return kvm->arch.n_max_mmu_pages; |
1fe779f8 CO |
3546 | } |
3547 | ||
1fe779f8 CO |
3548 | static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip) |
3549 | { | |
3550 | int r; | |
3551 | ||
3552 | r = 0; | |
3553 | switch (chip->chip_id) { | |
3554 | case KVM_IRQCHIP_PIC_MASTER: | |
3555 | memcpy(&chip->chip.pic, | |
3556 | &pic_irqchip(kvm)->pics[0], | |
3557 | sizeof(struct kvm_pic_state)); | |
3558 | break; | |
3559 | case KVM_IRQCHIP_PIC_SLAVE: | |
3560 | memcpy(&chip->chip.pic, | |
3561 | &pic_irqchip(kvm)->pics[1], | |
3562 | sizeof(struct kvm_pic_state)); | |
3563 | break; | |
3564 | case KVM_IRQCHIP_IOAPIC: | |
eba0226b | 3565 | r = kvm_get_ioapic(kvm, &chip->chip.ioapic); |
1fe779f8 CO |
3566 | break; |
3567 | default: | |
3568 | r = -EINVAL; | |
3569 | break; | |
3570 | } | |
3571 | return r; | |
3572 | } | |
3573 | ||
3574 | static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip) | |
3575 | { | |
3576 | int r; | |
3577 | ||
3578 | r = 0; | |
3579 | switch (chip->chip_id) { | |
3580 | case KVM_IRQCHIP_PIC_MASTER: | |
f4f51050 | 3581 | spin_lock(&pic_irqchip(kvm)->lock); |
1fe779f8 CO |
3582 | memcpy(&pic_irqchip(kvm)->pics[0], |
3583 | &chip->chip.pic, | |
3584 | sizeof(struct kvm_pic_state)); | |
f4f51050 | 3585 | spin_unlock(&pic_irqchip(kvm)->lock); |
1fe779f8 CO |
3586 | break; |
3587 | case KVM_IRQCHIP_PIC_SLAVE: | |
f4f51050 | 3588 | spin_lock(&pic_irqchip(kvm)->lock); |
1fe779f8 CO |
3589 | memcpy(&pic_irqchip(kvm)->pics[1], |
3590 | &chip->chip.pic, | |
3591 | sizeof(struct kvm_pic_state)); | |
f4f51050 | 3592 | spin_unlock(&pic_irqchip(kvm)->lock); |
1fe779f8 CO |
3593 | break; |
3594 | case KVM_IRQCHIP_IOAPIC: | |
eba0226b | 3595 | r = kvm_set_ioapic(kvm, &chip->chip.ioapic); |
1fe779f8 CO |
3596 | break; |
3597 | default: | |
3598 | r = -EINVAL; | |
3599 | break; | |
3600 | } | |
3601 | kvm_pic_update_irq(pic_irqchip(kvm)); | |
3602 | return r; | |
3603 | } | |
3604 | ||
e0f63cb9 SY |
3605 | static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps) |
3606 | { | |
34f3941c RK |
3607 | struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state; |
3608 | ||
3609 | BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels)); | |
3610 | ||
3611 | mutex_lock(&kps->lock); | |
3612 | memcpy(ps, &kps->channels, sizeof(*ps)); | |
3613 | mutex_unlock(&kps->lock); | |
2da29bcc | 3614 | return 0; |
e0f63cb9 SY |
3615 | } |
3616 | ||
3617 | static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps) | |
3618 | { | |
0185604c | 3619 | int i; |
09edea72 RK |
3620 | struct kvm_pit *pit = kvm->arch.vpit; |
3621 | ||
3622 | mutex_lock(&pit->pit_state.lock); | |
34f3941c | 3623 | memcpy(&pit->pit_state.channels, ps, sizeof(*ps)); |
0185604c | 3624 | for (i = 0; i < 3; i++) |
09edea72 RK |
3625 | kvm_pit_load_count(pit, i, ps->channels[i].count, 0); |
3626 | mutex_unlock(&pit->pit_state.lock); | |
2da29bcc | 3627 | return 0; |
e9f42757 BK |
3628 | } |
3629 | ||
3630 | static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps) | |
3631 | { | |
e9f42757 BK |
3632 | mutex_lock(&kvm->arch.vpit->pit_state.lock); |
3633 | memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels, | |
3634 | sizeof(ps->channels)); | |
3635 | ps->flags = kvm->arch.vpit->pit_state.flags; | |
3636 | mutex_unlock(&kvm->arch.vpit->pit_state.lock); | |
97e69aa6 | 3637 | memset(&ps->reserved, 0, sizeof(ps->reserved)); |
2da29bcc | 3638 | return 0; |
e9f42757 BK |
3639 | } |
3640 | ||
3641 | static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps) | |
3642 | { | |
2da29bcc | 3643 | int start = 0; |
0185604c | 3644 | int i; |
e9f42757 | 3645 | u32 prev_legacy, cur_legacy; |
09edea72 RK |
3646 | struct kvm_pit *pit = kvm->arch.vpit; |
3647 | ||
3648 | mutex_lock(&pit->pit_state.lock); | |
3649 | prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY; | |
e9f42757 BK |
3650 | cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY; |
3651 | if (!prev_legacy && cur_legacy) | |
3652 | start = 1; | |
09edea72 RK |
3653 | memcpy(&pit->pit_state.channels, &ps->channels, |
3654 | sizeof(pit->pit_state.channels)); | |
3655 | pit->pit_state.flags = ps->flags; | |
0185604c | 3656 | for (i = 0; i < 3; i++) |
09edea72 | 3657 | kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count, |
e5e57e7a | 3658 | start && i == 0); |
09edea72 | 3659 | mutex_unlock(&pit->pit_state.lock); |
2da29bcc | 3660 | return 0; |
e0f63cb9 SY |
3661 | } |
3662 | ||
52d939a0 MT |
3663 | static int kvm_vm_ioctl_reinject(struct kvm *kvm, |
3664 | struct kvm_reinject_control *control) | |
3665 | { | |
71474e2f RK |
3666 | struct kvm_pit *pit = kvm->arch.vpit; |
3667 | ||
3668 | if (!pit) | |
52d939a0 | 3669 | return -ENXIO; |
b39c90b6 | 3670 | |
71474e2f RK |
3671 | /* pit->pit_state.lock was overloaded to prevent userspace from getting |
3672 | * an inconsistent state after running multiple KVM_REINJECT_CONTROL | |
3673 | * ioctls in parallel. Use a separate lock if that ioctl isn't rare. | |
3674 | */ | |
3675 | mutex_lock(&pit->pit_state.lock); | |
3676 | kvm_pit_set_reinject(pit, control->pit_reinject); | |
3677 | mutex_unlock(&pit->pit_state.lock); | |
b39c90b6 | 3678 | |
52d939a0 MT |
3679 | return 0; |
3680 | } | |
3681 | ||
95d4c16c | 3682 | /** |
60c34612 TY |
3683 | * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot |
3684 | * @kvm: kvm instance | |
3685 | * @log: slot id and address to which we copy the log | |
95d4c16c | 3686 | * |
e108ff2f PB |
3687 | * Steps 1-4 below provide general overview of dirty page logging. See |
3688 | * kvm_get_dirty_log_protect() function description for additional details. | |
3689 | * | |
3690 | * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we | |
3691 | * always flush the TLB (step 4) even if previous step failed and the dirty | |
3692 | * bitmap may be corrupt. Regardless of previous outcome the KVM logging API | |
3693 | * does not preclude user space subsequent dirty log read. Flushing TLB ensures | |
3694 | * writes will be marked dirty for next log read. | |
95d4c16c | 3695 | * |
60c34612 TY |
3696 | * 1. Take a snapshot of the bit and clear it if needed. |
3697 | * 2. Write protect the corresponding page. | |
e108ff2f PB |
3698 | * 3. Copy the snapshot to the userspace. |
3699 | * 4. Flush TLB's if needed. | |
5bb064dc | 3700 | */ |
60c34612 | 3701 | int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log) |
5bb064dc | 3702 | { |
60c34612 | 3703 | bool is_dirty = false; |
e108ff2f | 3704 | int r; |
5bb064dc | 3705 | |
79fac95e | 3706 | mutex_lock(&kvm->slots_lock); |
5bb064dc | 3707 | |
88178fd4 KH |
3708 | /* |
3709 | * Flush potentially hardware-cached dirty pages to dirty_bitmap. | |
3710 | */ | |
3711 | if (kvm_x86_ops->flush_log_dirty) | |
3712 | kvm_x86_ops->flush_log_dirty(kvm); | |
3713 | ||
e108ff2f | 3714 | r = kvm_get_dirty_log_protect(kvm, log, &is_dirty); |
198c74f4 XG |
3715 | |
3716 | /* | |
3717 | * All the TLBs can be flushed out of mmu lock, see the comments in | |
3718 | * kvm_mmu_slot_remove_write_access(). | |
3719 | */ | |
e108ff2f | 3720 | lockdep_assert_held(&kvm->slots_lock); |
198c74f4 XG |
3721 | if (is_dirty) |
3722 | kvm_flush_remote_tlbs(kvm); | |
3723 | ||
79fac95e | 3724 | mutex_unlock(&kvm->slots_lock); |
5bb064dc ZX |
3725 | return r; |
3726 | } | |
3727 | ||
aa2fbe6d YZ |
3728 | int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event, |
3729 | bool line_status) | |
23d43cf9 CD |
3730 | { |
3731 | if (!irqchip_in_kernel(kvm)) | |
3732 | return -ENXIO; | |
3733 | ||
3734 | irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID, | |
aa2fbe6d YZ |
3735 | irq_event->irq, irq_event->level, |
3736 | line_status); | |
23d43cf9 CD |
3737 | return 0; |
3738 | } | |
3739 | ||
90de4a18 NA |
3740 | static int kvm_vm_ioctl_enable_cap(struct kvm *kvm, |
3741 | struct kvm_enable_cap *cap) | |
3742 | { | |
3743 | int r; | |
3744 | ||
3745 | if (cap->flags) | |
3746 | return -EINVAL; | |
3747 | ||
3748 | switch (cap->cap) { | |
3749 | case KVM_CAP_DISABLE_QUIRKS: | |
3750 | kvm->arch.disabled_quirks = cap->args[0]; | |
3751 | r = 0; | |
3752 | break; | |
49df6397 SR |
3753 | case KVM_CAP_SPLIT_IRQCHIP: { |
3754 | mutex_lock(&kvm->lock); | |
b053b2ae SR |
3755 | r = -EINVAL; |
3756 | if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS) | |
3757 | goto split_irqchip_unlock; | |
49df6397 SR |
3758 | r = -EEXIST; |
3759 | if (irqchip_in_kernel(kvm)) | |
3760 | goto split_irqchip_unlock; | |
3761 | if (atomic_read(&kvm->online_vcpus)) | |
3762 | goto split_irqchip_unlock; | |
3763 | r = kvm_setup_empty_irq_routing(kvm); | |
3764 | if (r) | |
3765 | goto split_irqchip_unlock; | |
3766 | /* Pairs with irqchip_in_kernel. */ | |
3767 | smp_wmb(); | |
3768 | kvm->arch.irqchip_split = true; | |
b053b2ae | 3769 | kvm->arch.nr_reserved_ioapic_pins = cap->args[0]; |
49df6397 SR |
3770 | r = 0; |
3771 | split_irqchip_unlock: | |
3772 | mutex_unlock(&kvm->lock); | |
3773 | break; | |
3774 | } | |
90de4a18 NA |
3775 | default: |
3776 | r = -EINVAL; | |
3777 | break; | |
3778 | } | |
3779 | return r; | |
3780 | } | |
3781 | ||
1fe779f8 CO |
3782 | long kvm_arch_vm_ioctl(struct file *filp, |
3783 | unsigned int ioctl, unsigned long arg) | |
3784 | { | |
3785 | struct kvm *kvm = filp->private_data; | |
3786 | void __user *argp = (void __user *)arg; | |
367e1319 | 3787 | int r = -ENOTTY; |
f0d66275 DH |
3788 | /* |
3789 | * This union makes it completely explicit to gcc-3.x | |
3790 | * that these two variables' stack usage should be | |
3791 | * combined, not added together. | |
3792 | */ | |
3793 | union { | |
3794 | struct kvm_pit_state ps; | |
e9f42757 | 3795 | struct kvm_pit_state2 ps2; |
c5ff41ce | 3796 | struct kvm_pit_config pit_config; |
f0d66275 | 3797 | } u; |
1fe779f8 CO |
3798 | |
3799 | switch (ioctl) { | |
3800 | case KVM_SET_TSS_ADDR: | |
3801 | r = kvm_vm_ioctl_set_tss_addr(kvm, arg); | |
1fe779f8 | 3802 | break; |
b927a3ce SY |
3803 | case KVM_SET_IDENTITY_MAP_ADDR: { |
3804 | u64 ident_addr; | |
3805 | ||
3806 | r = -EFAULT; | |
3807 | if (copy_from_user(&ident_addr, argp, sizeof ident_addr)) | |
3808 | goto out; | |
3809 | r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr); | |
b927a3ce SY |
3810 | break; |
3811 | } | |
1fe779f8 CO |
3812 | case KVM_SET_NR_MMU_PAGES: |
3813 | r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg); | |
1fe779f8 CO |
3814 | break; |
3815 | case KVM_GET_NR_MMU_PAGES: | |
3816 | r = kvm_vm_ioctl_get_nr_mmu_pages(kvm); | |
3817 | break; | |
3ddea128 MT |
3818 | case KVM_CREATE_IRQCHIP: { |
3819 | struct kvm_pic *vpic; | |
3820 | ||
3821 | mutex_lock(&kvm->lock); | |
3822 | r = -EEXIST; | |
3823 | if (kvm->arch.vpic) | |
3824 | goto create_irqchip_unlock; | |
3e515705 AK |
3825 | r = -EINVAL; |
3826 | if (atomic_read(&kvm->online_vcpus)) | |
3827 | goto create_irqchip_unlock; | |
1fe779f8 | 3828 | r = -ENOMEM; |
3ddea128 MT |
3829 | vpic = kvm_create_pic(kvm); |
3830 | if (vpic) { | |
1fe779f8 CO |
3831 | r = kvm_ioapic_init(kvm); |
3832 | if (r) { | |
175504cd | 3833 | mutex_lock(&kvm->slots_lock); |
71ba994c | 3834 | kvm_destroy_pic(vpic); |
175504cd | 3835 | mutex_unlock(&kvm->slots_lock); |
3ddea128 | 3836 | goto create_irqchip_unlock; |
1fe779f8 CO |
3837 | } |
3838 | } else | |
3ddea128 | 3839 | goto create_irqchip_unlock; |
399ec807 AK |
3840 | r = kvm_setup_default_irq_routing(kvm); |
3841 | if (r) { | |
175504cd | 3842 | mutex_lock(&kvm->slots_lock); |
3ddea128 | 3843 | mutex_lock(&kvm->irq_lock); |
72bb2fcd | 3844 | kvm_ioapic_destroy(kvm); |
71ba994c | 3845 | kvm_destroy_pic(vpic); |
3ddea128 | 3846 | mutex_unlock(&kvm->irq_lock); |
175504cd | 3847 | mutex_unlock(&kvm->slots_lock); |
71ba994c | 3848 | goto create_irqchip_unlock; |
399ec807 | 3849 | } |
71ba994c PB |
3850 | /* Write kvm->irq_routing before kvm->arch.vpic. */ |
3851 | smp_wmb(); | |
3852 | kvm->arch.vpic = vpic; | |
3ddea128 MT |
3853 | create_irqchip_unlock: |
3854 | mutex_unlock(&kvm->lock); | |
1fe779f8 | 3855 | break; |
3ddea128 | 3856 | } |
7837699f | 3857 | case KVM_CREATE_PIT: |
c5ff41ce JK |
3858 | u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY; |
3859 | goto create_pit; | |
3860 | case KVM_CREATE_PIT2: | |
3861 | r = -EFAULT; | |
3862 | if (copy_from_user(&u.pit_config, argp, | |
3863 | sizeof(struct kvm_pit_config))) | |
3864 | goto out; | |
3865 | create_pit: | |
79fac95e | 3866 | mutex_lock(&kvm->slots_lock); |
269e05e4 AK |
3867 | r = -EEXIST; |
3868 | if (kvm->arch.vpit) | |
3869 | goto create_pit_unlock; | |
7837699f | 3870 | r = -ENOMEM; |
c5ff41ce | 3871 | kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags); |
7837699f SY |
3872 | if (kvm->arch.vpit) |
3873 | r = 0; | |
269e05e4 | 3874 | create_pit_unlock: |
79fac95e | 3875 | mutex_unlock(&kvm->slots_lock); |
7837699f | 3876 | break; |
1fe779f8 CO |
3877 | case KVM_GET_IRQCHIP: { |
3878 | /* 0: PIC master, 1: PIC slave, 2: IOAPIC */ | |
ff5c2c03 | 3879 | struct kvm_irqchip *chip; |
1fe779f8 | 3880 | |
ff5c2c03 SL |
3881 | chip = memdup_user(argp, sizeof(*chip)); |
3882 | if (IS_ERR(chip)) { | |
3883 | r = PTR_ERR(chip); | |
1fe779f8 | 3884 | goto out; |
ff5c2c03 SL |
3885 | } |
3886 | ||
1fe779f8 | 3887 | r = -ENXIO; |
49df6397 | 3888 | if (!irqchip_in_kernel(kvm) || irqchip_split(kvm)) |
f0d66275 DH |
3889 | goto get_irqchip_out; |
3890 | r = kvm_vm_ioctl_get_irqchip(kvm, chip); | |
1fe779f8 | 3891 | if (r) |
f0d66275 | 3892 | goto get_irqchip_out; |
1fe779f8 | 3893 | r = -EFAULT; |
f0d66275 DH |
3894 | if (copy_to_user(argp, chip, sizeof *chip)) |
3895 | goto get_irqchip_out; | |
1fe779f8 | 3896 | r = 0; |
f0d66275 DH |
3897 | get_irqchip_out: |
3898 | kfree(chip); | |
1fe779f8 CO |
3899 | break; |
3900 | } | |
3901 | case KVM_SET_IRQCHIP: { | |
3902 | /* 0: PIC master, 1: PIC slave, 2: IOAPIC */ | |
ff5c2c03 | 3903 | struct kvm_irqchip *chip; |
1fe779f8 | 3904 | |
ff5c2c03 SL |
3905 | chip = memdup_user(argp, sizeof(*chip)); |
3906 | if (IS_ERR(chip)) { | |
3907 | r = PTR_ERR(chip); | |
1fe779f8 | 3908 | goto out; |
ff5c2c03 SL |
3909 | } |
3910 | ||
1fe779f8 | 3911 | r = -ENXIO; |
49df6397 | 3912 | if (!irqchip_in_kernel(kvm) || irqchip_split(kvm)) |
f0d66275 DH |
3913 | goto set_irqchip_out; |
3914 | r = kvm_vm_ioctl_set_irqchip(kvm, chip); | |
1fe779f8 | 3915 | if (r) |
f0d66275 | 3916 | goto set_irqchip_out; |
1fe779f8 | 3917 | r = 0; |
f0d66275 DH |
3918 | set_irqchip_out: |
3919 | kfree(chip); | |
1fe779f8 CO |
3920 | break; |
3921 | } | |
e0f63cb9 | 3922 | case KVM_GET_PIT: { |
e0f63cb9 | 3923 | r = -EFAULT; |
f0d66275 | 3924 | if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state))) |
e0f63cb9 SY |
3925 | goto out; |
3926 | r = -ENXIO; | |
3927 | if (!kvm->arch.vpit) | |
3928 | goto out; | |
f0d66275 | 3929 | r = kvm_vm_ioctl_get_pit(kvm, &u.ps); |
e0f63cb9 SY |
3930 | if (r) |
3931 | goto out; | |
3932 | r = -EFAULT; | |
f0d66275 | 3933 | if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state))) |
e0f63cb9 SY |
3934 | goto out; |
3935 | r = 0; | |
3936 | break; | |
3937 | } | |
3938 | case KVM_SET_PIT: { | |
e0f63cb9 | 3939 | r = -EFAULT; |
f0d66275 | 3940 | if (copy_from_user(&u.ps, argp, sizeof u.ps)) |
e0f63cb9 SY |
3941 | goto out; |
3942 | r = -ENXIO; | |
3943 | if (!kvm->arch.vpit) | |
3944 | goto out; | |
f0d66275 | 3945 | r = kvm_vm_ioctl_set_pit(kvm, &u.ps); |
e0f63cb9 SY |
3946 | break; |
3947 | } | |
e9f42757 BK |
3948 | case KVM_GET_PIT2: { |
3949 | r = -ENXIO; | |
3950 | if (!kvm->arch.vpit) | |
3951 | goto out; | |
3952 | r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2); | |
3953 | if (r) | |
3954 | goto out; | |
3955 | r = -EFAULT; | |
3956 | if (copy_to_user(argp, &u.ps2, sizeof(u.ps2))) | |
3957 | goto out; | |
3958 | r = 0; | |
3959 | break; | |
3960 | } | |
3961 | case KVM_SET_PIT2: { | |
3962 | r = -EFAULT; | |
3963 | if (copy_from_user(&u.ps2, argp, sizeof(u.ps2))) | |
3964 | goto out; | |
3965 | r = -ENXIO; | |
3966 | if (!kvm->arch.vpit) | |
3967 | goto out; | |
3968 | r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2); | |
e9f42757 BK |
3969 | break; |
3970 | } | |
52d939a0 MT |
3971 | case KVM_REINJECT_CONTROL: { |
3972 | struct kvm_reinject_control control; | |
3973 | r = -EFAULT; | |
3974 | if (copy_from_user(&control, argp, sizeof(control))) | |
3975 | goto out; | |
3976 | r = kvm_vm_ioctl_reinject(kvm, &control); | |
52d939a0 MT |
3977 | break; |
3978 | } | |
d71ba788 PB |
3979 | case KVM_SET_BOOT_CPU_ID: |
3980 | r = 0; | |
3981 | mutex_lock(&kvm->lock); | |
3982 | if (atomic_read(&kvm->online_vcpus) != 0) | |
3983 | r = -EBUSY; | |
3984 | else | |
3985 | kvm->arch.bsp_vcpu_id = arg; | |
3986 | mutex_unlock(&kvm->lock); | |
3987 | break; | |
ffde22ac ES |
3988 | case KVM_XEN_HVM_CONFIG: { |
3989 | r = -EFAULT; | |
3990 | if (copy_from_user(&kvm->arch.xen_hvm_config, argp, | |
3991 | sizeof(struct kvm_xen_hvm_config))) | |
3992 | goto out; | |
3993 | r = -EINVAL; | |
3994 | if (kvm->arch.xen_hvm_config.flags) | |
3995 | goto out; | |
3996 | r = 0; | |
3997 | break; | |
3998 | } | |
afbcf7ab | 3999 | case KVM_SET_CLOCK: { |
afbcf7ab GC |
4000 | struct kvm_clock_data user_ns; |
4001 | u64 now_ns; | |
4002 | s64 delta; | |
4003 | ||
4004 | r = -EFAULT; | |
4005 | if (copy_from_user(&user_ns, argp, sizeof(user_ns))) | |
4006 | goto out; | |
4007 | ||
4008 | r = -EINVAL; | |
4009 | if (user_ns.flags) | |
4010 | goto out; | |
4011 | ||
4012 | r = 0; | |
395c6b0a | 4013 | local_irq_disable(); |
759379dd | 4014 | now_ns = get_kernel_ns(); |
afbcf7ab | 4015 | delta = user_ns.clock - now_ns; |
395c6b0a | 4016 | local_irq_enable(); |
afbcf7ab | 4017 | kvm->arch.kvmclock_offset = delta; |
2e762ff7 | 4018 | kvm_gen_update_masterclock(kvm); |
afbcf7ab GC |
4019 | break; |
4020 | } | |
4021 | case KVM_GET_CLOCK: { | |
afbcf7ab GC |
4022 | struct kvm_clock_data user_ns; |
4023 | u64 now_ns; | |
4024 | ||
395c6b0a | 4025 | local_irq_disable(); |
759379dd | 4026 | now_ns = get_kernel_ns(); |
afbcf7ab | 4027 | user_ns.clock = kvm->arch.kvmclock_offset + now_ns; |
395c6b0a | 4028 | local_irq_enable(); |
afbcf7ab | 4029 | user_ns.flags = 0; |
97e69aa6 | 4030 | memset(&user_ns.pad, 0, sizeof(user_ns.pad)); |
afbcf7ab GC |
4031 | |
4032 | r = -EFAULT; | |
4033 | if (copy_to_user(argp, &user_ns, sizeof(user_ns))) | |
4034 | goto out; | |
4035 | r = 0; | |
4036 | break; | |
4037 | } | |
90de4a18 NA |
4038 | case KVM_ENABLE_CAP: { |
4039 | struct kvm_enable_cap cap; | |
afbcf7ab | 4040 | |
90de4a18 NA |
4041 | r = -EFAULT; |
4042 | if (copy_from_user(&cap, argp, sizeof(cap))) | |
4043 | goto out; | |
4044 | r = kvm_vm_ioctl_enable_cap(kvm, &cap); | |
4045 | break; | |
4046 | } | |
1fe779f8 | 4047 | default: |
c274e03a | 4048 | r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg); |
1fe779f8 CO |
4049 | } |
4050 | out: | |
4051 | return r; | |
4052 | } | |
4053 | ||
a16b043c | 4054 | static void kvm_init_msr_list(void) |
043405e1 CO |
4055 | { |
4056 | u32 dummy[2]; | |
4057 | unsigned i, j; | |
4058 | ||
62ef68bb | 4059 | for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) { |
043405e1 CO |
4060 | if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0) |
4061 | continue; | |
93c4adc7 PB |
4062 | |
4063 | /* | |
4064 | * Even MSRs that are valid in the host may not be exposed | |
9dbe6cf9 | 4065 | * to the guests in some cases. |
93c4adc7 PB |
4066 | */ |
4067 | switch (msrs_to_save[i]) { | |
4068 | case MSR_IA32_BNDCFGS: | |
4069 | if (!kvm_x86_ops->mpx_supported()) | |
4070 | continue; | |
4071 | break; | |
9dbe6cf9 PB |
4072 | case MSR_TSC_AUX: |
4073 | if (!kvm_x86_ops->rdtscp_supported()) | |
4074 | continue; | |
4075 | break; | |
93c4adc7 PB |
4076 | default: |
4077 | break; | |
4078 | } | |
4079 | ||
043405e1 CO |
4080 | if (j < i) |
4081 | msrs_to_save[j] = msrs_to_save[i]; | |
4082 | j++; | |
4083 | } | |
4084 | num_msrs_to_save = j; | |
62ef68bb PB |
4085 | |
4086 | for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) { | |
4087 | switch (emulated_msrs[i]) { | |
6d396b55 PB |
4088 | case MSR_IA32_SMBASE: |
4089 | if (!kvm_x86_ops->cpu_has_high_real_mode_segbase()) | |
4090 | continue; | |
4091 | break; | |
62ef68bb PB |
4092 | default: |
4093 | break; | |
4094 | } | |
4095 | ||
4096 | if (j < i) | |
4097 | emulated_msrs[j] = emulated_msrs[i]; | |
4098 | j++; | |
4099 | } | |
4100 | num_emulated_msrs = j; | |
043405e1 CO |
4101 | } |
4102 | ||
bda9020e MT |
4103 | static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len, |
4104 | const void *v) | |
bbd9b64e | 4105 | { |
70252a10 AK |
4106 | int handled = 0; |
4107 | int n; | |
4108 | ||
4109 | do { | |
4110 | n = min(len, 8); | |
bce87cce | 4111 | if (!(lapic_in_kernel(vcpu) && |
e32edf4f NN |
4112 | !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v)) |
4113 | && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v)) | |
70252a10 AK |
4114 | break; |
4115 | handled += n; | |
4116 | addr += n; | |
4117 | len -= n; | |
4118 | v += n; | |
4119 | } while (len); | |
bbd9b64e | 4120 | |
70252a10 | 4121 | return handled; |
bbd9b64e CO |
4122 | } |
4123 | ||
bda9020e | 4124 | static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v) |
bbd9b64e | 4125 | { |
70252a10 AK |
4126 | int handled = 0; |
4127 | int n; | |
4128 | ||
4129 | do { | |
4130 | n = min(len, 8); | |
bce87cce | 4131 | if (!(lapic_in_kernel(vcpu) && |
e32edf4f NN |
4132 | !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev, |
4133 | addr, n, v)) | |
4134 | && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v)) | |
70252a10 AK |
4135 | break; |
4136 | trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v); | |
4137 | handled += n; | |
4138 | addr += n; | |
4139 | len -= n; | |
4140 | v += n; | |
4141 | } while (len); | |
bbd9b64e | 4142 | |
70252a10 | 4143 | return handled; |
bbd9b64e CO |
4144 | } |
4145 | ||
2dafc6c2 GN |
4146 | static void kvm_set_segment(struct kvm_vcpu *vcpu, |
4147 | struct kvm_segment *var, int seg) | |
4148 | { | |
4149 | kvm_x86_ops->set_segment(vcpu, var, seg); | |
4150 | } | |
4151 | ||
4152 | void kvm_get_segment(struct kvm_vcpu *vcpu, | |
4153 | struct kvm_segment *var, int seg) | |
4154 | { | |
4155 | kvm_x86_ops->get_segment(vcpu, var, seg); | |
4156 | } | |
4157 | ||
54987b7a PB |
4158 | gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access, |
4159 | struct x86_exception *exception) | |
02f59dc9 JR |
4160 | { |
4161 | gpa_t t_gpa; | |
02f59dc9 JR |
4162 | |
4163 | BUG_ON(!mmu_is_nested(vcpu)); | |
4164 | ||
4165 | /* NPT walks are always user-walks */ | |
4166 | access |= PFERR_USER_MASK; | |
54987b7a | 4167 | t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, exception); |
02f59dc9 JR |
4168 | |
4169 | return t_gpa; | |
4170 | } | |
4171 | ||
ab9ae313 AK |
4172 | gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva, |
4173 | struct x86_exception *exception) | |
1871c602 GN |
4174 | { |
4175 | u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0; | |
ab9ae313 | 4176 | return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception); |
1871c602 GN |
4177 | } |
4178 | ||
ab9ae313 AK |
4179 | gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva, |
4180 | struct x86_exception *exception) | |
1871c602 GN |
4181 | { |
4182 | u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0; | |
4183 | access |= PFERR_FETCH_MASK; | |
ab9ae313 | 4184 | return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception); |
1871c602 GN |
4185 | } |
4186 | ||
ab9ae313 AK |
4187 | gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva, |
4188 | struct x86_exception *exception) | |
1871c602 GN |
4189 | { |
4190 | u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0; | |
4191 | access |= PFERR_WRITE_MASK; | |
ab9ae313 | 4192 | return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception); |
1871c602 GN |
4193 | } |
4194 | ||
4195 | /* uses this to access any guest's mapped memory without checking CPL */ | |
ab9ae313 AK |
4196 | gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva, |
4197 | struct x86_exception *exception) | |
1871c602 | 4198 | { |
ab9ae313 | 4199 | return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception); |
1871c602 GN |
4200 | } |
4201 | ||
4202 | static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes, | |
4203 | struct kvm_vcpu *vcpu, u32 access, | |
bcc55cba | 4204 | struct x86_exception *exception) |
bbd9b64e CO |
4205 | { |
4206 | void *data = val; | |
10589a46 | 4207 | int r = X86EMUL_CONTINUE; |
bbd9b64e CO |
4208 | |
4209 | while (bytes) { | |
14dfe855 | 4210 | gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access, |
ab9ae313 | 4211 | exception); |
bbd9b64e | 4212 | unsigned offset = addr & (PAGE_SIZE-1); |
77c2002e | 4213 | unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset); |
bbd9b64e CO |
4214 | int ret; |
4215 | ||
bcc55cba | 4216 | if (gpa == UNMAPPED_GVA) |
ab9ae313 | 4217 | return X86EMUL_PROPAGATE_FAULT; |
54bf36aa PB |
4218 | ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data, |
4219 | offset, toread); | |
10589a46 | 4220 | if (ret < 0) { |
c3cd7ffa | 4221 | r = X86EMUL_IO_NEEDED; |
10589a46 MT |
4222 | goto out; |
4223 | } | |
bbd9b64e | 4224 | |
77c2002e IE |
4225 | bytes -= toread; |
4226 | data += toread; | |
4227 | addr += toread; | |
bbd9b64e | 4228 | } |
10589a46 | 4229 | out: |
10589a46 | 4230 | return r; |
bbd9b64e | 4231 | } |
77c2002e | 4232 | |
1871c602 | 4233 | /* used for instruction fetching */ |
0f65dd70 AK |
4234 | static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt, |
4235 | gva_t addr, void *val, unsigned int bytes, | |
bcc55cba | 4236 | struct x86_exception *exception) |
1871c602 | 4237 | { |
0f65dd70 | 4238 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
1871c602 | 4239 | u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0; |
44583cba PB |
4240 | unsigned offset; |
4241 | int ret; | |
0f65dd70 | 4242 | |
44583cba PB |
4243 | /* Inline kvm_read_guest_virt_helper for speed. */ |
4244 | gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK, | |
4245 | exception); | |
4246 | if (unlikely(gpa == UNMAPPED_GVA)) | |
4247 | return X86EMUL_PROPAGATE_FAULT; | |
4248 | ||
4249 | offset = addr & (PAGE_SIZE-1); | |
4250 | if (WARN_ON(offset + bytes > PAGE_SIZE)) | |
4251 | bytes = (unsigned)PAGE_SIZE - offset; | |
54bf36aa PB |
4252 | ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val, |
4253 | offset, bytes); | |
44583cba PB |
4254 | if (unlikely(ret < 0)) |
4255 | return X86EMUL_IO_NEEDED; | |
4256 | ||
4257 | return X86EMUL_CONTINUE; | |
1871c602 GN |
4258 | } |
4259 | ||
064aea77 | 4260 | int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt, |
0f65dd70 | 4261 | gva_t addr, void *val, unsigned int bytes, |
bcc55cba | 4262 | struct x86_exception *exception) |
1871c602 | 4263 | { |
0f65dd70 | 4264 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
1871c602 | 4265 | u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0; |
0f65dd70 | 4266 | |
1871c602 | 4267 | return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, |
bcc55cba | 4268 | exception); |
1871c602 | 4269 | } |
064aea77 | 4270 | EXPORT_SYMBOL_GPL(kvm_read_guest_virt); |
1871c602 | 4271 | |
0f65dd70 AK |
4272 | static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt, |
4273 | gva_t addr, void *val, unsigned int bytes, | |
bcc55cba | 4274 | struct x86_exception *exception) |
1871c602 | 4275 | { |
0f65dd70 | 4276 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
bcc55cba | 4277 | return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception); |
1871c602 GN |
4278 | } |
4279 | ||
7a036a6f RK |
4280 | static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt, |
4281 | unsigned long addr, void *val, unsigned int bytes) | |
4282 | { | |
4283 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); | |
4284 | int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes); | |
4285 | ||
4286 | return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE; | |
4287 | } | |
4288 | ||
6a4d7550 | 4289 | int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt, |
0f65dd70 | 4290 | gva_t addr, void *val, |
2dafc6c2 | 4291 | unsigned int bytes, |
bcc55cba | 4292 | struct x86_exception *exception) |
77c2002e | 4293 | { |
0f65dd70 | 4294 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
77c2002e IE |
4295 | void *data = val; |
4296 | int r = X86EMUL_CONTINUE; | |
4297 | ||
4298 | while (bytes) { | |
14dfe855 JR |
4299 | gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, |
4300 | PFERR_WRITE_MASK, | |
ab9ae313 | 4301 | exception); |
77c2002e IE |
4302 | unsigned offset = addr & (PAGE_SIZE-1); |
4303 | unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset); | |
4304 | int ret; | |
4305 | ||
bcc55cba | 4306 | if (gpa == UNMAPPED_GVA) |
ab9ae313 | 4307 | return X86EMUL_PROPAGATE_FAULT; |
54bf36aa | 4308 | ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite); |
77c2002e | 4309 | if (ret < 0) { |
c3cd7ffa | 4310 | r = X86EMUL_IO_NEEDED; |
77c2002e IE |
4311 | goto out; |
4312 | } | |
4313 | ||
4314 | bytes -= towrite; | |
4315 | data += towrite; | |
4316 | addr += towrite; | |
4317 | } | |
4318 | out: | |
4319 | return r; | |
4320 | } | |
6a4d7550 | 4321 | EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system); |
77c2002e | 4322 | |
af7cc7d1 XG |
4323 | static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva, |
4324 | gpa_t *gpa, struct x86_exception *exception, | |
4325 | bool write) | |
4326 | { | |
97d64b78 AK |
4327 | u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0) |
4328 | | (write ? PFERR_WRITE_MASK : 0); | |
af7cc7d1 | 4329 | |
97d64b78 | 4330 | if (vcpu_match_mmio_gva(vcpu, gva) |
97ec8c06 FW |
4331 | && !permission_fault(vcpu, vcpu->arch.walk_mmu, |
4332 | vcpu->arch.access, access)) { | |
bebb106a XG |
4333 | *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT | |
4334 | (gva & (PAGE_SIZE - 1)); | |
4f022648 | 4335 | trace_vcpu_match_mmio(gva, *gpa, write, false); |
bebb106a XG |
4336 | return 1; |
4337 | } | |
4338 | ||
af7cc7d1 XG |
4339 | *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception); |
4340 | ||
4341 | if (*gpa == UNMAPPED_GVA) | |
4342 | return -1; | |
4343 | ||
4344 | /* For APIC access vmexit */ | |
4345 | if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE) | |
4346 | return 1; | |
4347 | ||
4f022648 XG |
4348 | if (vcpu_match_mmio_gpa(vcpu, *gpa)) { |
4349 | trace_vcpu_match_mmio(gva, *gpa, write, true); | |
bebb106a | 4350 | return 1; |
4f022648 | 4351 | } |
bebb106a | 4352 | |
af7cc7d1 XG |
4353 | return 0; |
4354 | } | |
4355 | ||
3200f405 | 4356 | int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa, |
bcc55cba | 4357 | const void *val, int bytes) |
bbd9b64e CO |
4358 | { |
4359 | int ret; | |
4360 | ||
54bf36aa | 4361 | ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes); |
9f811285 | 4362 | if (ret < 0) |
bbd9b64e | 4363 | return 0; |
0eb05bf2 | 4364 | kvm_page_track_write(vcpu, gpa, val, bytes); |
bbd9b64e CO |
4365 | return 1; |
4366 | } | |
4367 | ||
77d197b2 XG |
4368 | struct read_write_emulator_ops { |
4369 | int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val, | |
4370 | int bytes); | |
4371 | int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa, | |
4372 | void *val, int bytes); | |
4373 | int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa, | |
4374 | int bytes, void *val); | |
4375 | int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa, | |
4376 | void *val, int bytes); | |
4377 | bool write; | |
4378 | }; | |
4379 | ||
4380 | static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes) | |
4381 | { | |
4382 | if (vcpu->mmio_read_completed) { | |
77d197b2 | 4383 | trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes, |
f78146b0 | 4384 | vcpu->mmio_fragments[0].gpa, *(u64 *)val); |
77d197b2 XG |
4385 | vcpu->mmio_read_completed = 0; |
4386 | return 1; | |
4387 | } | |
4388 | ||
4389 | return 0; | |
4390 | } | |
4391 | ||
4392 | static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa, | |
4393 | void *val, int bytes) | |
4394 | { | |
54bf36aa | 4395 | return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes); |
77d197b2 XG |
4396 | } |
4397 | ||
4398 | static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa, | |
4399 | void *val, int bytes) | |
4400 | { | |
4401 | return emulator_write_phys(vcpu, gpa, val, bytes); | |
4402 | } | |
4403 | ||
4404 | static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val) | |
4405 | { | |
4406 | trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val); | |
4407 | return vcpu_mmio_write(vcpu, gpa, bytes, val); | |
4408 | } | |
4409 | ||
4410 | static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, | |
4411 | void *val, int bytes) | |
4412 | { | |
4413 | trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0); | |
4414 | return X86EMUL_IO_NEEDED; | |
4415 | } | |
4416 | ||
4417 | static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, | |
4418 | void *val, int bytes) | |
4419 | { | |
f78146b0 AK |
4420 | struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0]; |
4421 | ||
87da7e66 | 4422 | memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len)); |
77d197b2 XG |
4423 | return X86EMUL_CONTINUE; |
4424 | } | |
4425 | ||
0fbe9b0b | 4426 | static const struct read_write_emulator_ops read_emultor = { |
77d197b2 XG |
4427 | .read_write_prepare = read_prepare, |
4428 | .read_write_emulate = read_emulate, | |
4429 | .read_write_mmio = vcpu_mmio_read, | |
4430 | .read_write_exit_mmio = read_exit_mmio, | |
4431 | }; | |
4432 | ||
0fbe9b0b | 4433 | static const struct read_write_emulator_ops write_emultor = { |
77d197b2 XG |
4434 | .read_write_emulate = write_emulate, |
4435 | .read_write_mmio = write_mmio, | |
4436 | .read_write_exit_mmio = write_exit_mmio, | |
4437 | .write = true, | |
4438 | }; | |
4439 | ||
22388a3c XG |
4440 | static int emulator_read_write_onepage(unsigned long addr, void *val, |
4441 | unsigned int bytes, | |
4442 | struct x86_exception *exception, | |
4443 | struct kvm_vcpu *vcpu, | |
0fbe9b0b | 4444 | const struct read_write_emulator_ops *ops) |
bbd9b64e | 4445 | { |
af7cc7d1 XG |
4446 | gpa_t gpa; |
4447 | int handled, ret; | |
22388a3c | 4448 | bool write = ops->write; |
f78146b0 | 4449 | struct kvm_mmio_fragment *frag; |
10589a46 | 4450 | |
22388a3c | 4451 | ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write); |
bbd9b64e | 4452 | |
af7cc7d1 | 4453 | if (ret < 0) |
bbd9b64e | 4454 | return X86EMUL_PROPAGATE_FAULT; |
bbd9b64e CO |
4455 | |
4456 | /* For APIC access vmexit */ | |
af7cc7d1 | 4457 | if (ret) |
bbd9b64e CO |
4458 | goto mmio; |
4459 | ||
22388a3c | 4460 | if (ops->read_write_emulate(vcpu, gpa, val, bytes)) |
bbd9b64e CO |
4461 | return X86EMUL_CONTINUE; |
4462 | ||
4463 | mmio: | |
4464 | /* | |
4465 | * Is this MMIO handled locally? | |
4466 | */ | |
22388a3c | 4467 | handled = ops->read_write_mmio(vcpu, gpa, bytes, val); |
70252a10 | 4468 | if (handled == bytes) |
bbd9b64e | 4469 | return X86EMUL_CONTINUE; |
bbd9b64e | 4470 | |
70252a10 AK |
4471 | gpa += handled; |
4472 | bytes -= handled; | |
4473 | val += handled; | |
4474 | ||
87da7e66 XG |
4475 | WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS); |
4476 | frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++]; | |
4477 | frag->gpa = gpa; | |
4478 | frag->data = val; | |
4479 | frag->len = bytes; | |
f78146b0 | 4480 | return X86EMUL_CONTINUE; |
bbd9b64e CO |
4481 | } |
4482 | ||
52eb5a6d XL |
4483 | static int emulator_read_write(struct x86_emulate_ctxt *ctxt, |
4484 | unsigned long addr, | |
22388a3c XG |
4485 | void *val, unsigned int bytes, |
4486 | struct x86_exception *exception, | |
0fbe9b0b | 4487 | const struct read_write_emulator_ops *ops) |
bbd9b64e | 4488 | { |
0f65dd70 | 4489 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
f78146b0 AK |
4490 | gpa_t gpa; |
4491 | int rc; | |
4492 | ||
4493 | if (ops->read_write_prepare && | |
4494 | ops->read_write_prepare(vcpu, val, bytes)) | |
4495 | return X86EMUL_CONTINUE; | |
4496 | ||
4497 | vcpu->mmio_nr_fragments = 0; | |
0f65dd70 | 4498 | |
bbd9b64e CO |
4499 | /* Crossing a page boundary? */ |
4500 | if (((addr + bytes - 1) ^ addr) & PAGE_MASK) { | |
f78146b0 | 4501 | int now; |
bbd9b64e CO |
4502 | |
4503 | now = -addr & ~PAGE_MASK; | |
22388a3c XG |
4504 | rc = emulator_read_write_onepage(addr, val, now, exception, |
4505 | vcpu, ops); | |
4506 | ||
bbd9b64e CO |
4507 | if (rc != X86EMUL_CONTINUE) |
4508 | return rc; | |
4509 | addr += now; | |
bac15531 NA |
4510 | if (ctxt->mode != X86EMUL_MODE_PROT64) |
4511 | addr = (u32)addr; | |
bbd9b64e CO |
4512 | val += now; |
4513 | bytes -= now; | |
4514 | } | |
22388a3c | 4515 | |
f78146b0 AK |
4516 | rc = emulator_read_write_onepage(addr, val, bytes, exception, |
4517 | vcpu, ops); | |
4518 | if (rc != X86EMUL_CONTINUE) | |
4519 | return rc; | |
4520 | ||
4521 | if (!vcpu->mmio_nr_fragments) | |
4522 | return rc; | |
4523 | ||
4524 | gpa = vcpu->mmio_fragments[0].gpa; | |
4525 | ||
4526 | vcpu->mmio_needed = 1; | |
4527 | vcpu->mmio_cur_fragment = 0; | |
4528 | ||
87da7e66 | 4529 | vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len); |
f78146b0 AK |
4530 | vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write; |
4531 | vcpu->run->exit_reason = KVM_EXIT_MMIO; | |
4532 | vcpu->run->mmio.phys_addr = gpa; | |
4533 | ||
4534 | return ops->read_write_exit_mmio(vcpu, gpa, val, bytes); | |
22388a3c XG |
4535 | } |
4536 | ||
4537 | static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt, | |
4538 | unsigned long addr, | |
4539 | void *val, | |
4540 | unsigned int bytes, | |
4541 | struct x86_exception *exception) | |
4542 | { | |
4543 | return emulator_read_write(ctxt, addr, val, bytes, | |
4544 | exception, &read_emultor); | |
4545 | } | |
4546 | ||
52eb5a6d | 4547 | static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt, |
22388a3c XG |
4548 | unsigned long addr, |
4549 | const void *val, | |
4550 | unsigned int bytes, | |
4551 | struct x86_exception *exception) | |
4552 | { | |
4553 | return emulator_read_write(ctxt, addr, (void *)val, bytes, | |
4554 | exception, &write_emultor); | |
bbd9b64e | 4555 | } |
bbd9b64e | 4556 | |
daea3e73 AK |
4557 | #define CMPXCHG_TYPE(t, ptr, old, new) \ |
4558 | (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old)) | |
4559 | ||
4560 | #ifdef CONFIG_X86_64 | |
4561 | # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new) | |
4562 | #else | |
4563 | # define CMPXCHG64(ptr, old, new) \ | |
9749a6c0 | 4564 | (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old)) |
daea3e73 AK |
4565 | #endif |
4566 | ||
0f65dd70 AK |
4567 | static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt, |
4568 | unsigned long addr, | |
bbd9b64e CO |
4569 | const void *old, |
4570 | const void *new, | |
4571 | unsigned int bytes, | |
0f65dd70 | 4572 | struct x86_exception *exception) |
bbd9b64e | 4573 | { |
0f65dd70 | 4574 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
daea3e73 AK |
4575 | gpa_t gpa; |
4576 | struct page *page; | |
4577 | char *kaddr; | |
4578 | bool exchanged; | |
2bacc55c | 4579 | |
daea3e73 AK |
4580 | /* guests cmpxchg8b have to be emulated atomically */ |
4581 | if (bytes > 8 || (bytes & (bytes - 1))) | |
4582 | goto emul_write; | |
10589a46 | 4583 | |
daea3e73 | 4584 | gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL); |
2bacc55c | 4585 | |
daea3e73 AK |
4586 | if (gpa == UNMAPPED_GVA || |
4587 | (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE) | |
4588 | goto emul_write; | |
2bacc55c | 4589 | |
daea3e73 AK |
4590 | if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK)) |
4591 | goto emul_write; | |
72dc67a6 | 4592 | |
54bf36aa | 4593 | page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT); |
32cad84f | 4594 | if (is_error_page(page)) |
c19b8bd6 | 4595 | goto emul_write; |
72dc67a6 | 4596 | |
8fd75e12 | 4597 | kaddr = kmap_atomic(page); |
daea3e73 AK |
4598 | kaddr += offset_in_page(gpa); |
4599 | switch (bytes) { | |
4600 | case 1: | |
4601 | exchanged = CMPXCHG_TYPE(u8, kaddr, old, new); | |
4602 | break; | |
4603 | case 2: | |
4604 | exchanged = CMPXCHG_TYPE(u16, kaddr, old, new); | |
4605 | break; | |
4606 | case 4: | |
4607 | exchanged = CMPXCHG_TYPE(u32, kaddr, old, new); | |
4608 | break; | |
4609 | case 8: | |
4610 | exchanged = CMPXCHG64(kaddr, old, new); | |
4611 | break; | |
4612 | default: | |
4613 | BUG(); | |
2bacc55c | 4614 | } |
8fd75e12 | 4615 | kunmap_atomic(kaddr); |
daea3e73 AK |
4616 | kvm_release_page_dirty(page); |
4617 | ||
4618 | if (!exchanged) | |
4619 | return X86EMUL_CMPXCHG_FAILED; | |
4620 | ||
54bf36aa | 4621 | kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT); |
0eb05bf2 | 4622 | kvm_page_track_write(vcpu, gpa, new, bytes); |
8f6abd06 GN |
4623 | |
4624 | return X86EMUL_CONTINUE; | |
4a5f48f6 | 4625 | |
3200f405 | 4626 | emul_write: |
daea3e73 | 4627 | printk_once(KERN_WARNING "kvm: emulating exchange as write\n"); |
2bacc55c | 4628 | |
0f65dd70 | 4629 | return emulator_write_emulated(ctxt, addr, new, bytes, exception); |
bbd9b64e CO |
4630 | } |
4631 | ||
cf8f70bf GN |
4632 | static int kernel_pio(struct kvm_vcpu *vcpu, void *pd) |
4633 | { | |
4634 | /* TODO: String I/O for in kernel device */ | |
4635 | int r; | |
4636 | ||
4637 | if (vcpu->arch.pio.in) | |
e32edf4f | 4638 | r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port, |
cf8f70bf GN |
4639 | vcpu->arch.pio.size, pd); |
4640 | else | |
e32edf4f | 4641 | r = kvm_io_bus_write(vcpu, KVM_PIO_BUS, |
cf8f70bf GN |
4642 | vcpu->arch.pio.port, vcpu->arch.pio.size, |
4643 | pd); | |
4644 | return r; | |
4645 | } | |
4646 | ||
6f6fbe98 XG |
4647 | static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size, |
4648 | unsigned short port, void *val, | |
4649 | unsigned int count, bool in) | |
cf8f70bf | 4650 | { |
cf8f70bf | 4651 | vcpu->arch.pio.port = port; |
6f6fbe98 | 4652 | vcpu->arch.pio.in = in; |
7972995b | 4653 | vcpu->arch.pio.count = count; |
cf8f70bf GN |
4654 | vcpu->arch.pio.size = size; |
4655 | ||
4656 | if (!kernel_pio(vcpu, vcpu->arch.pio_data)) { | |
7972995b | 4657 | vcpu->arch.pio.count = 0; |
cf8f70bf GN |
4658 | return 1; |
4659 | } | |
4660 | ||
4661 | vcpu->run->exit_reason = KVM_EXIT_IO; | |
6f6fbe98 | 4662 | vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT; |
cf8f70bf GN |
4663 | vcpu->run->io.size = size; |
4664 | vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE; | |
4665 | vcpu->run->io.count = count; | |
4666 | vcpu->run->io.port = port; | |
4667 | ||
4668 | return 0; | |
4669 | } | |
4670 | ||
6f6fbe98 XG |
4671 | static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt, |
4672 | int size, unsigned short port, void *val, | |
4673 | unsigned int count) | |
cf8f70bf | 4674 | { |
ca1d4a9e | 4675 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
6f6fbe98 | 4676 | int ret; |
ca1d4a9e | 4677 | |
6f6fbe98 XG |
4678 | if (vcpu->arch.pio.count) |
4679 | goto data_avail; | |
cf8f70bf | 4680 | |
6f6fbe98 XG |
4681 | ret = emulator_pio_in_out(vcpu, size, port, val, count, true); |
4682 | if (ret) { | |
4683 | data_avail: | |
4684 | memcpy(val, vcpu->arch.pio_data, size * count); | |
1171903d | 4685 | trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data); |
7972995b | 4686 | vcpu->arch.pio.count = 0; |
cf8f70bf GN |
4687 | return 1; |
4688 | } | |
4689 | ||
cf8f70bf GN |
4690 | return 0; |
4691 | } | |
4692 | ||
6f6fbe98 XG |
4693 | static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt, |
4694 | int size, unsigned short port, | |
4695 | const void *val, unsigned int count) | |
4696 | { | |
4697 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); | |
4698 | ||
4699 | memcpy(vcpu->arch.pio_data, val, size * count); | |
1171903d | 4700 | trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data); |
6f6fbe98 XG |
4701 | return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false); |
4702 | } | |
4703 | ||
bbd9b64e CO |
4704 | static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg) |
4705 | { | |
4706 | return kvm_x86_ops->get_segment_base(vcpu, seg); | |
4707 | } | |
4708 | ||
3cb16fe7 | 4709 | static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address) |
bbd9b64e | 4710 | { |
3cb16fe7 | 4711 | kvm_mmu_invlpg(emul_to_vcpu(ctxt), address); |
bbd9b64e CO |
4712 | } |
4713 | ||
5cb56059 | 4714 | int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu) |
f5f48ee1 SY |
4715 | { |
4716 | if (!need_emulate_wbinvd(vcpu)) | |
4717 | return X86EMUL_CONTINUE; | |
4718 | ||
4719 | if (kvm_x86_ops->has_wbinvd_exit()) { | |
2eec7343 JK |
4720 | int cpu = get_cpu(); |
4721 | ||
4722 | cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask); | |
f5f48ee1 SY |
4723 | smp_call_function_many(vcpu->arch.wbinvd_dirty_mask, |
4724 | wbinvd_ipi, NULL, 1); | |
2eec7343 | 4725 | put_cpu(); |
f5f48ee1 | 4726 | cpumask_clear(vcpu->arch.wbinvd_dirty_mask); |
2eec7343 JK |
4727 | } else |
4728 | wbinvd(); | |
f5f48ee1 SY |
4729 | return X86EMUL_CONTINUE; |
4730 | } | |
5cb56059 JS |
4731 | |
4732 | int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu) | |
4733 | { | |
4734 | kvm_x86_ops->skip_emulated_instruction(vcpu); | |
4735 | return kvm_emulate_wbinvd_noskip(vcpu); | |
4736 | } | |
f5f48ee1 SY |
4737 | EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd); |
4738 | ||
5cb56059 JS |
4739 | |
4740 | ||
bcaf5cc5 AK |
4741 | static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt) |
4742 | { | |
5cb56059 | 4743 | kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt)); |
bcaf5cc5 AK |
4744 | } |
4745 | ||
52eb5a6d XL |
4746 | static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, |
4747 | unsigned long *dest) | |
bbd9b64e | 4748 | { |
16f8a6f9 | 4749 | return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest); |
bbd9b64e CO |
4750 | } |
4751 | ||
52eb5a6d XL |
4752 | static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, |
4753 | unsigned long value) | |
bbd9b64e | 4754 | { |
338dbc97 | 4755 | |
717746e3 | 4756 | return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value); |
bbd9b64e CO |
4757 | } |
4758 | ||
52a46617 | 4759 | static u64 mk_cr_64(u64 curr_cr, u32 new_val) |
5fdbf976 | 4760 | { |
52a46617 | 4761 | return (curr_cr & ~((1ULL << 32) - 1)) | new_val; |
5fdbf976 MT |
4762 | } |
4763 | ||
717746e3 | 4764 | static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr) |
bbd9b64e | 4765 | { |
717746e3 | 4766 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
52a46617 GN |
4767 | unsigned long value; |
4768 | ||
4769 | switch (cr) { | |
4770 | case 0: | |
4771 | value = kvm_read_cr0(vcpu); | |
4772 | break; | |
4773 | case 2: | |
4774 | value = vcpu->arch.cr2; | |
4775 | break; | |
4776 | case 3: | |
9f8fe504 | 4777 | value = kvm_read_cr3(vcpu); |
52a46617 GN |
4778 | break; |
4779 | case 4: | |
4780 | value = kvm_read_cr4(vcpu); | |
4781 | break; | |
4782 | case 8: | |
4783 | value = kvm_get_cr8(vcpu); | |
4784 | break; | |
4785 | default: | |
a737f256 | 4786 | kvm_err("%s: unexpected cr %u\n", __func__, cr); |
52a46617 GN |
4787 | return 0; |
4788 | } | |
4789 | ||
4790 | return value; | |
4791 | } | |
4792 | ||
717746e3 | 4793 | static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val) |
52a46617 | 4794 | { |
717746e3 | 4795 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
0f12244f GN |
4796 | int res = 0; |
4797 | ||
52a46617 GN |
4798 | switch (cr) { |
4799 | case 0: | |
49a9b07e | 4800 | res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val)); |
52a46617 GN |
4801 | break; |
4802 | case 2: | |
4803 | vcpu->arch.cr2 = val; | |
4804 | break; | |
4805 | case 3: | |
2390218b | 4806 | res = kvm_set_cr3(vcpu, val); |
52a46617 GN |
4807 | break; |
4808 | case 4: | |
a83b29c6 | 4809 | res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val)); |
52a46617 GN |
4810 | break; |
4811 | case 8: | |
eea1cff9 | 4812 | res = kvm_set_cr8(vcpu, val); |
52a46617 GN |
4813 | break; |
4814 | default: | |
a737f256 | 4815 | kvm_err("%s: unexpected cr %u\n", __func__, cr); |
0f12244f | 4816 | res = -1; |
52a46617 | 4817 | } |
0f12244f GN |
4818 | |
4819 | return res; | |
52a46617 GN |
4820 | } |
4821 | ||
717746e3 | 4822 | static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt) |
9c537244 | 4823 | { |
717746e3 | 4824 | return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt)); |
9c537244 GN |
4825 | } |
4826 | ||
4bff1e86 | 4827 | static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt) |
2dafc6c2 | 4828 | { |
4bff1e86 | 4829 | kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt); |
2dafc6c2 GN |
4830 | } |
4831 | ||
4bff1e86 | 4832 | static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt) |
160ce1f1 | 4833 | { |
4bff1e86 | 4834 | kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt); |
160ce1f1 MG |
4835 | } |
4836 | ||
1ac9d0cf AK |
4837 | static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt) |
4838 | { | |
4839 | kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt); | |
4840 | } | |
4841 | ||
4842 | static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt) | |
4843 | { | |
4844 | kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt); | |
4845 | } | |
4846 | ||
4bff1e86 AK |
4847 | static unsigned long emulator_get_cached_segment_base( |
4848 | struct x86_emulate_ctxt *ctxt, int seg) | |
5951c442 | 4849 | { |
4bff1e86 | 4850 | return get_segment_base(emul_to_vcpu(ctxt), seg); |
5951c442 GN |
4851 | } |
4852 | ||
1aa36616 AK |
4853 | static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector, |
4854 | struct desc_struct *desc, u32 *base3, | |
4855 | int seg) | |
2dafc6c2 GN |
4856 | { |
4857 | struct kvm_segment var; | |
4858 | ||
4bff1e86 | 4859 | kvm_get_segment(emul_to_vcpu(ctxt), &var, seg); |
1aa36616 | 4860 | *selector = var.selector; |
2dafc6c2 | 4861 | |
378a8b09 GN |
4862 | if (var.unusable) { |
4863 | memset(desc, 0, sizeof(*desc)); | |
2dafc6c2 | 4864 | return false; |
378a8b09 | 4865 | } |
2dafc6c2 GN |
4866 | |
4867 | if (var.g) | |
4868 | var.limit >>= 12; | |
4869 | set_desc_limit(desc, var.limit); | |
4870 | set_desc_base(desc, (unsigned long)var.base); | |
5601d05b GN |
4871 | #ifdef CONFIG_X86_64 |
4872 | if (base3) | |
4873 | *base3 = var.base >> 32; | |
4874 | #endif | |
2dafc6c2 GN |
4875 | desc->type = var.type; |
4876 | desc->s = var.s; | |
4877 | desc->dpl = var.dpl; | |
4878 | desc->p = var.present; | |
4879 | desc->avl = var.avl; | |
4880 | desc->l = var.l; | |
4881 | desc->d = var.db; | |
4882 | desc->g = var.g; | |
4883 | ||
4884 | return true; | |
4885 | } | |
4886 | ||
1aa36616 AK |
4887 | static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector, |
4888 | struct desc_struct *desc, u32 base3, | |
4889 | int seg) | |
2dafc6c2 | 4890 | { |
4bff1e86 | 4891 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
2dafc6c2 GN |
4892 | struct kvm_segment var; |
4893 | ||
1aa36616 | 4894 | var.selector = selector; |
2dafc6c2 | 4895 | var.base = get_desc_base(desc); |
5601d05b GN |
4896 | #ifdef CONFIG_X86_64 |
4897 | var.base |= ((u64)base3) << 32; | |
4898 | #endif | |
2dafc6c2 GN |
4899 | var.limit = get_desc_limit(desc); |
4900 | if (desc->g) | |
4901 | var.limit = (var.limit << 12) | 0xfff; | |
4902 | var.type = desc->type; | |
2dafc6c2 GN |
4903 | var.dpl = desc->dpl; |
4904 | var.db = desc->d; | |
4905 | var.s = desc->s; | |
4906 | var.l = desc->l; | |
4907 | var.g = desc->g; | |
4908 | var.avl = desc->avl; | |
4909 | var.present = desc->p; | |
4910 | var.unusable = !var.present; | |
4911 | var.padding = 0; | |
4912 | ||
4913 | kvm_set_segment(vcpu, &var, seg); | |
4914 | return; | |
4915 | } | |
4916 | ||
717746e3 AK |
4917 | static int emulator_get_msr(struct x86_emulate_ctxt *ctxt, |
4918 | u32 msr_index, u64 *pdata) | |
4919 | { | |
609e36d3 PB |
4920 | struct msr_data msr; |
4921 | int r; | |
4922 | ||
4923 | msr.index = msr_index; | |
4924 | msr.host_initiated = false; | |
4925 | r = kvm_get_msr(emul_to_vcpu(ctxt), &msr); | |
4926 | if (r) | |
4927 | return r; | |
4928 | ||
4929 | *pdata = msr.data; | |
4930 | return 0; | |
717746e3 AK |
4931 | } |
4932 | ||
4933 | static int emulator_set_msr(struct x86_emulate_ctxt *ctxt, | |
4934 | u32 msr_index, u64 data) | |
4935 | { | |
8fe8ab46 WA |
4936 | struct msr_data msr; |
4937 | ||
4938 | msr.data = data; | |
4939 | msr.index = msr_index; | |
4940 | msr.host_initiated = false; | |
4941 | return kvm_set_msr(emul_to_vcpu(ctxt), &msr); | |
717746e3 AK |
4942 | } |
4943 | ||
64d60670 PB |
4944 | static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt) |
4945 | { | |
4946 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); | |
4947 | ||
4948 | return vcpu->arch.smbase; | |
4949 | } | |
4950 | ||
4951 | static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase) | |
4952 | { | |
4953 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); | |
4954 | ||
4955 | vcpu->arch.smbase = smbase; | |
4956 | } | |
4957 | ||
67f4d428 NA |
4958 | static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt, |
4959 | u32 pmc) | |
4960 | { | |
c6702c9d | 4961 | return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc); |
67f4d428 NA |
4962 | } |
4963 | ||
222d21aa AK |
4964 | static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt, |
4965 | u32 pmc, u64 *pdata) | |
4966 | { | |
c6702c9d | 4967 | return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata); |
222d21aa AK |
4968 | } |
4969 | ||
6c3287f7 AK |
4970 | static void emulator_halt(struct x86_emulate_ctxt *ctxt) |
4971 | { | |
4972 | emul_to_vcpu(ctxt)->arch.halt_request = 1; | |
4973 | } | |
4974 | ||
5037f6f3 AK |
4975 | static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt) |
4976 | { | |
4977 | preempt_disable(); | |
5197b808 | 4978 | kvm_load_guest_fpu(emul_to_vcpu(ctxt)); |
5037f6f3 AK |
4979 | /* |
4980 | * CR0.TS may reference the host fpu state, not the guest fpu state, | |
4981 | * so it may be clear at this point. | |
4982 | */ | |
4983 | clts(); | |
4984 | } | |
4985 | ||
4986 | static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt) | |
4987 | { | |
4988 | preempt_enable(); | |
4989 | } | |
4990 | ||
2953538e | 4991 | static int emulator_intercept(struct x86_emulate_ctxt *ctxt, |
8a76d7f2 | 4992 | struct x86_instruction_info *info, |
c4f035c6 AK |
4993 | enum x86_intercept_stage stage) |
4994 | { | |
2953538e | 4995 | return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage); |
c4f035c6 AK |
4996 | } |
4997 | ||
0017f93a | 4998 | static void emulator_get_cpuid(struct x86_emulate_ctxt *ctxt, |
bdb42f5a SB |
4999 | u32 *eax, u32 *ebx, u32 *ecx, u32 *edx) |
5000 | { | |
0017f93a | 5001 | kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx); |
bdb42f5a SB |
5002 | } |
5003 | ||
dd856efa AK |
5004 | static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg) |
5005 | { | |
5006 | return kvm_register_read(emul_to_vcpu(ctxt), reg); | |
5007 | } | |
5008 | ||
5009 | static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val) | |
5010 | { | |
5011 | kvm_register_write(emul_to_vcpu(ctxt), reg, val); | |
5012 | } | |
5013 | ||
801806d9 NA |
5014 | static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked) |
5015 | { | |
5016 | kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked); | |
5017 | } | |
5018 | ||
0225fb50 | 5019 | static const struct x86_emulate_ops emulate_ops = { |
dd856efa AK |
5020 | .read_gpr = emulator_read_gpr, |
5021 | .write_gpr = emulator_write_gpr, | |
1871c602 | 5022 | .read_std = kvm_read_guest_virt_system, |
2dafc6c2 | 5023 | .write_std = kvm_write_guest_virt_system, |
7a036a6f | 5024 | .read_phys = kvm_read_guest_phys_system, |
1871c602 | 5025 | .fetch = kvm_fetch_guest_virt, |
bbd9b64e CO |
5026 | .read_emulated = emulator_read_emulated, |
5027 | .write_emulated = emulator_write_emulated, | |
5028 | .cmpxchg_emulated = emulator_cmpxchg_emulated, | |
3cb16fe7 | 5029 | .invlpg = emulator_invlpg, |
cf8f70bf GN |
5030 | .pio_in_emulated = emulator_pio_in_emulated, |
5031 | .pio_out_emulated = emulator_pio_out_emulated, | |
1aa36616 AK |
5032 | .get_segment = emulator_get_segment, |
5033 | .set_segment = emulator_set_segment, | |
5951c442 | 5034 | .get_cached_segment_base = emulator_get_cached_segment_base, |
2dafc6c2 | 5035 | .get_gdt = emulator_get_gdt, |
160ce1f1 | 5036 | .get_idt = emulator_get_idt, |
1ac9d0cf AK |
5037 | .set_gdt = emulator_set_gdt, |
5038 | .set_idt = emulator_set_idt, | |
52a46617 GN |
5039 | .get_cr = emulator_get_cr, |
5040 | .set_cr = emulator_set_cr, | |
9c537244 | 5041 | .cpl = emulator_get_cpl, |
35aa5375 GN |
5042 | .get_dr = emulator_get_dr, |
5043 | .set_dr = emulator_set_dr, | |
64d60670 PB |
5044 | .get_smbase = emulator_get_smbase, |
5045 | .set_smbase = emulator_set_smbase, | |
717746e3 AK |
5046 | .set_msr = emulator_set_msr, |
5047 | .get_msr = emulator_get_msr, | |
67f4d428 | 5048 | .check_pmc = emulator_check_pmc, |
222d21aa | 5049 | .read_pmc = emulator_read_pmc, |
6c3287f7 | 5050 | .halt = emulator_halt, |
bcaf5cc5 | 5051 | .wbinvd = emulator_wbinvd, |
d6aa1000 | 5052 | .fix_hypercall = emulator_fix_hypercall, |
5037f6f3 AK |
5053 | .get_fpu = emulator_get_fpu, |
5054 | .put_fpu = emulator_put_fpu, | |
c4f035c6 | 5055 | .intercept = emulator_intercept, |
bdb42f5a | 5056 | .get_cpuid = emulator_get_cpuid, |
801806d9 | 5057 | .set_nmi_mask = emulator_set_nmi_mask, |
bbd9b64e CO |
5058 | }; |
5059 | ||
95cb2295 GN |
5060 | static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask) |
5061 | { | |
37ccdcbe | 5062 | u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu); |
95cb2295 GN |
5063 | /* |
5064 | * an sti; sti; sequence only disable interrupts for the first | |
5065 | * instruction. So, if the last instruction, be it emulated or | |
5066 | * not, left the system with the INT_STI flag enabled, it | |
5067 | * means that the last instruction is an sti. We should not | |
5068 | * leave the flag on in this case. The same goes for mov ss | |
5069 | */ | |
37ccdcbe PB |
5070 | if (int_shadow & mask) |
5071 | mask = 0; | |
6addfc42 | 5072 | if (unlikely(int_shadow || mask)) { |
95cb2295 | 5073 | kvm_x86_ops->set_interrupt_shadow(vcpu, mask); |
6addfc42 PB |
5074 | if (!mask) |
5075 | kvm_make_request(KVM_REQ_EVENT, vcpu); | |
5076 | } | |
95cb2295 GN |
5077 | } |
5078 | ||
ef54bcfe | 5079 | static bool inject_emulated_exception(struct kvm_vcpu *vcpu) |
54b8486f GN |
5080 | { |
5081 | struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt; | |
da9cb575 | 5082 | if (ctxt->exception.vector == PF_VECTOR) |
ef54bcfe PB |
5083 | return kvm_propagate_fault(vcpu, &ctxt->exception); |
5084 | ||
5085 | if (ctxt->exception.error_code_valid) | |
da9cb575 AK |
5086 | kvm_queue_exception_e(vcpu, ctxt->exception.vector, |
5087 | ctxt->exception.error_code); | |
54b8486f | 5088 | else |
da9cb575 | 5089 | kvm_queue_exception(vcpu, ctxt->exception.vector); |
ef54bcfe | 5090 | return false; |
54b8486f GN |
5091 | } |
5092 | ||
8ec4722d MG |
5093 | static void init_emulate_ctxt(struct kvm_vcpu *vcpu) |
5094 | { | |
adf52235 | 5095 | struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt; |
8ec4722d MG |
5096 | int cs_db, cs_l; |
5097 | ||
8ec4722d MG |
5098 | kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l); |
5099 | ||
adf52235 TY |
5100 | ctxt->eflags = kvm_get_rflags(vcpu); |
5101 | ctxt->eip = kvm_rip_read(vcpu); | |
5102 | ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL : | |
5103 | (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 : | |
42bf549f | 5104 | (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 : |
adf52235 TY |
5105 | cs_db ? X86EMUL_MODE_PROT32 : |
5106 | X86EMUL_MODE_PROT16; | |
a584539b | 5107 | BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK); |
64d60670 PB |
5108 | BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK); |
5109 | BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK); | |
a584539b | 5110 | ctxt->emul_flags = vcpu->arch.hflags; |
adf52235 | 5111 | |
dd856efa | 5112 | init_decode_cache(ctxt); |
7ae441ea | 5113 | vcpu->arch.emulate_regs_need_sync_from_vcpu = false; |
8ec4722d MG |
5114 | } |
5115 | ||
71f9833b | 5116 | int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip) |
63995653 | 5117 | { |
9d74191a | 5118 | struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt; |
63995653 MG |
5119 | int ret; |
5120 | ||
5121 | init_emulate_ctxt(vcpu); | |
5122 | ||
9dac77fa AK |
5123 | ctxt->op_bytes = 2; |
5124 | ctxt->ad_bytes = 2; | |
5125 | ctxt->_eip = ctxt->eip + inc_eip; | |
9d74191a | 5126 | ret = emulate_int_real(ctxt, irq); |
63995653 MG |
5127 | |
5128 | if (ret != X86EMUL_CONTINUE) | |
5129 | return EMULATE_FAIL; | |
5130 | ||
9dac77fa | 5131 | ctxt->eip = ctxt->_eip; |
9d74191a TY |
5132 | kvm_rip_write(vcpu, ctxt->eip); |
5133 | kvm_set_rflags(vcpu, ctxt->eflags); | |
63995653 MG |
5134 | |
5135 | if (irq == NMI_VECTOR) | |
7460fb4a | 5136 | vcpu->arch.nmi_pending = 0; |
63995653 MG |
5137 | else |
5138 | vcpu->arch.interrupt.pending = false; | |
5139 | ||
5140 | return EMULATE_DONE; | |
5141 | } | |
5142 | EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt); | |
5143 | ||
6d77dbfc GN |
5144 | static int handle_emulation_failure(struct kvm_vcpu *vcpu) |
5145 | { | |
fc3a9157 JR |
5146 | int r = EMULATE_DONE; |
5147 | ||
6d77dbfc GN |
5148 | ++vcpu->stat.insn_emulation_fail; |
5149 | trace_kvm_emulate_insn_failed(vcpu); | |
a2b9e6c1 | 5150 | if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) { |
fc3a9157 JR |
5151 | vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR; |
5152 | vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION; | |
5153 | vcpu->run->internal.ndata = 0; | |
5154 | r = EMULATE_FAIL; | |
5155 | } | |
6d77dbfc | 5156 | kvm_queue_exception(vcpu, UD_VECTOR); |
fc3a9157 JR |
5157 | |
5158 | return r; | |
6d77dbfc GN |
5159 | } |
5160 | ||
93c05d3e | 5161 | static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2, |
991eebf9 GN |
5162 | bool write_fault_to_shadow_pgtable, |
5163 | int emulation_type) | |
a6f177ef | 5164 | { |
95b3cf69 | 5165 | gpa_t gpa = cr2; |
ba049e93 | 5166 | kvm_pfn_t pfn; |
a6f177ef | 5167 | |
991eebf9 GN |
5168 | if (emulation_type & EMULTYPE_NO_REEXECUTE) |
5169 | return false; | |
5170 | ||
95b3cf69 XG |
5171 | if (!vcpu->arch.mmu.direct_map) { |
5172 | /* | |
5173 | * Write permission should be allowed since only | |
5174 | * write access need to be emulated. | |
5175 | */ | |
5176 | gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL); | |
a6f177ef | 5177 | |
95b3cf69 XG |
5178 | /* |
5179 | * If the mapping is invalid in guest, let cpu retry | |
5180 | * it to generate fault. | |
5181 | */ | |
5182 | if (gpa == UNMAPPED_GVA) | |
5183 | return true; | |
5184 | } | |
a6f177ef | 5185 | |
8e3d9d06 XG |
5186 | /* |
5187 | * Do not retry the unhandleable instruction if it faults on the | |
5188 | * readonly host memory, otherwise it will goto a infinite loop: | |
5189 | * retry instruction -> write #PF -> emulation fail -> retry | |
5190 | * instruction -> ... | |
5191 | */ | |
5192 | pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa)); | |
95b3cf69 XG |
5193 | |
5194 | /* | |
5195 | * If the instruction failed on the error pfn, it can not be fixed, | |
5196 | * report the error to userspace. | |
5197 | */ | |
5198 | if (is_error_noslot_pfn(pfn)) | |
5199 | return false; | |
5200 | ||
5201 | kvm_release_pfn_clean(pfn); | |
5202 | ||
5203 | /* The instructions are well-emulated on direct mmu. */ | |
5204 | if (vcpu->arch.mmu.direct_map) { | |
5205 | unsigned int indirect_shadow_pages; | |
5206 | ||
5207 | spin_lock(&vcpu->kvm->mmu_lock); | |
5208 | indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages; | |
5209 | spin_unlock(&vcpu->kvm->mmu_lock); | |
5210 | ||
5211 | if (indirect_shadow_pages) | |
5212 | kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa)); | |
5213 | ||
a6f177ef | 5214 | return true; |
8e3d9d06 | 5215 | } |
a6f177ef | 5216 | |
95b3cf69 XG |
5217 | /* |
5218 | * if emulation was due to access to shadowed page table | |
5219 | * and it failed try to unshadow page and re-enter the | |
5220 | * guest to let CPU execute the instruction. | |
5221 | */ | |
5222 | kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa)); | |
93c05d3e XG |
5223 | |
5224 | /* | |
5225 | * If the access faults on its page table, it can not | |
5226 | * be fixed by unprotecting shadow page and it should | |
5227 | * be reported to userspace. | |
5228 | */ | |
5229 | return !write_fault_to_shadow_pgtable; | |
a6f177ef GN |
5230 | } |
5231 | ||
1cb3f3ae XG |
5232 | static bool retry_instruction(struct x86_emulate_ctxt *ctxt, |
5233 | unsigned long cr2, int emulation_type) | |
5234 | { | |
5235 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); | |
5236 | unsigned long last_retry_eip, last_retry_addr, gpa = cr2; | |
5237 | ||
5238 | last_retry_eip = vcpu->arch.last_retry_eip; | |
5239 | last_retry_addr = vcpu->arch.last_retry_addr; | |
5240 | ||
5241 | /* | |
5242 | * If the emulation is caused by #PF and it is non-page_table | |
5243 | * writing instruction, it means the VM-EXIT is caused by shadow | |
5244 | * page protected, we can zap the shadow page and retry this | |
5245 | * instruction directly. | |
5246 | * | |
5247 | * Note: if the guest uses a non-page-table modifying instruction | |
5248 | * on the PDE that points to the instruction, then we will unmap | |
5249 | * the instruction and go to an infinite loop. So, we cache the | |
5250 | * last retried eip and the last fault address, if we meet the eip | |
5251 | * and the address again, we can break out of the potential infinite | |
5252 | * loop. | |
5253 | */ | |
5254 | vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0; | |
5255 | ||
5256 | if (!(emulation_type & EMULTYPE_RETRY)) | |
5257 | return false; | |
5258 | ||
5259 | if (x86_page_table_writing_insn(ctxt)) | |
5260 | return false; | |
5261 | ||
5262 | if (ctxt->eip == last_retry_eip && last_retry_addr == cr2) | |
5263 | return false; | |
5264 | ||
5265 | vcpu->arch.last_retry_eip = ctxt->eip; | |
5266 | vcpu->arch.last_retry_addr = cr2; | |
5267 | ||
5268 | if (!vcpu->arch.mmu.direct_map) | |
5269 | gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL); | |
5270 | ||
22368028 | 5271 | kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa)); |
1cb3f3ae XG |
5272 | |
5273 | return true; | |
5274 | } | |
5275 | ||
716d51ab GN |
5276 | static int complete_emulated_mmio(struct kvm_vcpu *vcpu); |
5277 | static int complete_emulated_pio(struct kvm_vcpu *vcpu); | |
5278 | ||
64d60670 | 5279 | static void kvm_smm_changed(struct kvm_vcpu *vcpu) |
a584539b | 5280 | { |
64d60670 | 5281 | if (!(vcpu->arch.hflags & HF_SMM_MASK)) { |
660a5d51 PB |
5282 | /* This is a good place to trace that we are exiting SMM. */ |
5283 | trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false); | |
5284 | ||
64d60670 PB |
5285 | if (unlikely(vcpu->arch.smi_pending)) { |
5286 | kvm_make_request(KVM_REQ_SMI, vcpu); | |
5287 | vcpu->arch.smi_pending = 0; | |
cd7764fe PB |
5288 | } else { |
5289 | /* Process a latched INIT, if any. */ | |
5290 | kvm_make_request(KVM_REQ_EVENT, vcpu); | |
64d60670 PB |
5291 | } |
5292 | } | |
699023e2 PB |
5293 | |
5294 | kvm_mmu_reset_context(vcpu); | |
64d60670 PB |
5295 | } |
5296 | ||
5297 | static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags) | |
5298 | { | |
5299 | unsigned changed = vcpu->arch.hflags ^ emul_flags; | |
5300 | ||
a584539b | 5301 | vcpu->arch.hflags = emul_flags; |
64d60670 PB |
5302 | |
5303 | if (changed & HF_SMM_MASK) | |
5304 | kvm_smm_changed(vcpu); | |
a584539b PB |
5305 | } |
5306 | ||
4a1e10d5 PB |
5307 | static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7, |
5308 | unsigned long *db) | |
5309 | { | |
5310 | u32 dr6 = 0; | |
5311 | int i; | |
5312 | u32 enable, rwlen; | |
5313 | ||
5314 | enable = dr7; | |
5315 | rwlen = dr7 >> 16; | |
5316 | for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4) | |
5317 | if ((enable & 3) && (rwlen & 15) == type && db[i] == addr) | |
5318 | dr6 |= (1 << i); | |
5319 | return dr6; | |
5320 | } | |
5321 | ||
6addfc42 | 5322 | static void kvm_vcpu_check_singlestep(struct kvm_vcpu *vcpu, unsigned long rflags, int *r) |
663f4c61 PB |
5323 | { |
5324 | struct kvm_run *kvm_run = vcpu->run; | |
5325 | ||
5326 | /* | |
6addfc42 PB |
5327 | * rflags is the old, "raw" value of the flags. The new value has |
5328 | * not been saved yet. | |
663f4c61 PB |
5329 | * |
5330 | * This is correct even for TF set by the guest, because "the | |
5331 | * processor will not generate this exception after the instruction | |
5332 | * that sets the TF flag". | |
5333 | */ | |
663f4c61 PB |
5334 | if (unlikely(rflags & X86_EFLAGS_TF)) { |
5335 | if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) { | |
6f43ed01 NA |
5336 | kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 | |
5337 | DR6_RTM; | |
663f4c61 PB |
5338 | kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip; |
5339 | kvm_run->debug.arch.exception = DB_VECTOR; | |
5340 | kvm_run->exit_reason = KVM_EXIT_DEBUG; | |
5341 | *r = EMULATE_USER_EXIT; | |
5342 | } else { | |
5343 | vcpu->arch.emulate_ctxt.eflags &= ~X86_EFLAGS_TF; | |
5344 | /* | |
5345 | * "Certain debug exceptions may clear bit 0-3. The | |
5346 | * remaining contents of the DR6 register are never | |
5347 | * cleared by the processor". | |
5348 | */ | |
5349 | vcpu->arch.dr6 &= ~15; | |
6f43ed01 | 5350 | vcpu->arch.dr6 |= DR6_BS | DR6_RTM; |
663f4c61 PB |
5351 | kvm_queue_exception(vcpu, DB_VECTOR); |
5352 | } | |
5353 | } | |
5354 | } | |
5355 | ||
4a1e10d5 PB |
5356 | static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r) |
5357 | { | |
4a1e10d5 PB |
5358 | if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) && |
5359 | (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) { | |
82b32774 NA |
5360 | struct kvm_run *kvm_run = vcpu->run; |
5361 | unsigned long eip = kvm_get_linear_rip(vcpu); | |
5362 | u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0, | |
4a1e10d5 PB |
5363 | vcpu->arch.guest_debug_dr7, |
5364 | vcpu->arch.eff_db); | |
5365 | ||
5366 | if (dr6 != 0) { | |
6f43ed01 | 5367 | kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM; |
82b32774 | 5368 | kvm_run->debug.arch.pc = eip; |
4a1e10d5 PB |
5369 | kvm_run->debug.arch.exception = DB_VECTOR; |
5370 | kvm_run->exit_reason = KVM_EXIT_DEBUG; | |
5371 | *r = EMULATE_USER_EXIT; | |
5372 | return true; | |
5373 | } | |
5374 | } | |
5375 | ||
4161a569 NA |
5376 | if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) && |
5377 | !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) { | |
82b32774 NA |
5378 | unsigned long eip = kvm_get_linear_rip(vcpu); |
5379 | u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0, | |
4a1e10d5 PB |
5380 | vcpu->arch.dr7, |
5381 | vcpu->arch.db); | |
5382 | ||
5383 | if (dr6 != 0) { | |
5384 | vcpu->arch.dr6 &= ~15; | |
6f43ed01 | 5385 | vcpu->arch.dr6 |= dr6 | DR6_RTM; |
4a1e10d5 PB |
5386 | kvm_queue_exception(vcpu, DB_VECTOR); |
5387 | *r = EMULATE_DONE; | |
5388 | return true; | |
5389 | } | |
5390 | } | |
5391 | ||
5392 | return false; | |
5393 | } | |
5394 | ||
51d8b661 AP |
5395 | int x86_emulate_instruction(struct kvm_vcpu *vcpu, |
5396 | unsigned long cr2, | |
dc25e89e AP |
5397 | int emulation_type, |
5398 | void *insn, | |
5399 | int insn_len) | |
bbd9b64e | 5400 | { |
95cb2295 | 5401 | int r; |
9d74191a | 5402 | struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt; |
7ae441ea | 5403 | bool writeback = true; |
93c05d3e | 5404 | bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable; |
bbd9b64e | 5405 | |
93c05d3e XG |
5406 | /* |
5407 | * Clear write_fault_to_shadow_pgtable here to ensure it is | |
5408 | * never reused. | |
5409 | */ | |
5410 | vcpu->arch.write_fault_to_shadow_pgtable = false; | |
26eef70c | 5411 | kvm_clear_exception_queue(vcpu); |
8d7d8102 | 5412 | |
571008da | 5413 | if (!(emulation_type & EMULTYPE_NO_DECODE)) { |
8ec4722d | 5414 | init_emulate_ctxt(vcpu); |
4a1e10d5 PB |
5415 | |
5416 | /* | |
5417 | * We will reenter on the same instruction since | |
5418 | * we do not set complete_userspace_io. This does not | |
5419 | * handle watchpoints yet, those would be handled in | |
5420 | * the emulate_ops. | |
5421 | */ | |
5422 | if (kvm_vcpu_check_breakpoint(vcpu, &r)) | |
5423 | return r; | |
5424 | ||
9d74191a TY |
5425 | ctxt->interruptibility = 0; |
5426 | ctxt->have_exception = false; | |
e0ad0b47 | 5427 | ctxt->exception.vector = -1; |
9d74191a | 5428 | ctxt->perm_ok = false; |
bbd9b64e | 5429 | |
b51e974f | 5430 | ctxt->ud = emulation_type & EMULTYPE_TRAP_UD; |
4005996e | 5431 | |
9d74191a | 5432 | r = x86_decode_insn(ctxt, insn, insn_len); |
bbd9b64e | 5433 | |
e46479f8 | 5434 | trace_kvm_emulate_insn_start(vcpu); |
f2b5756b | 5435 | ++vcpu->stat.insn_emulation; |
1d2887e2 | 5436 | if (r != EMULATION_OK) { |
4005996e AK |
5437 | if (emulation_type & EMULTYPE_TRAP_UD) |
5438 | return EMULATE_FAIL; | |
991eebf9 GN |
5439 | if (reexecute_instruction(vcpu, cr2, write_fault_to_spt, |
5440 | emulation_type)) | |
bbd9b64e | 5441 | return EMULATE_DONE; |
6d77dbfc GN |
5442 | if (emulation_type & EMULTYPE_SKIP) |
5443 | return EMULATE_FAIL; | |
5444 | return handle_emulation_failure(vcpu); | |
bbd9b64e CO |
5445 | } |
5446 | } | |
5447 | ||
ba8afb6b | 5448 | if (emulation_type & EMULTYPE_SKIP) { |
9dac77fa | 5449 | kvm_rip_write(vcpu, ctxt->_eip); |
bb663c7a NA |
5450 | if (ctxt->eflags & X86_EFLAGS_RF) |
5451 | kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF); | |
ba8afb6b GN |
5452 | return EMULATE_DONE; |
5453 | } | |
5454 | ||
1cb3f3ae XG |
5455 | if (retry_instruction(ctxt, cr2, emulation_type)) |
5456 | return EMULATE_DONE; | |
5457 | ||
7ae441ea | 5458 | /* this is needed for vmware backdoor interface to work since it |
4d2179e1 | 5459 | changes registers values during IO operation */ |
7ae441ea GN |
5460 | if (vcpu->arch.emulate_regs_need_sync_from_vcpu) { |
5461 | vcpu->arch.emulate_regs_need_sync_from_vcpu = false; | |
dd856efa | 5462 | emulator_invalidate_register_cache(ctxt); |
7ae441ea | 5463 | } |
4d2179e1 | 5464 | |
5cd21917 | 5465 | restart: |
9d74191a | 5466 | r = x86_emulate_insn(ctxt); |
bbd9b64e | 5467 | |
775fde86 JR |
5468 | if (r == EMULATION_INTERCEPTED) |
5469 | return EMULATE_DONE; | |
5470 | ||
d2ddd1c4 | 5471 | if (r == EMULATION_FAILED) { |
991eebf9 GN |
5472 | if (reexecute_instruction(vcpu, cr2, write_fault_to_spt, |
5473 | emulation_type)) | |
c3cd7ffa GN |
5474 | return EMULATE_DONE; |
5475 | ||
6d77dbfc | 5476 | return handle_emulation_failure(vcpu); |
bbd9b64e CO |
5477 | } |
5478 | ||
9d74191a | 5479 | if (ctxt->have_exception) { |
d2ddd1c4 | 5480 | r = EMULATE_DONE; |
ef54bcfe PB |
5481 | if (inject_emulated_exception(vcpu)) |
5482 | return r; | |
d2ddd1c4 | 5483 | } else if (vcpu->arch.pio.count) { |
0912c977 PB |
5484 | if (!vcpu->arch.pio.in) { |
5485 | /* FIXME: return into emulator if single-stepping. */ | |
3457e419 | 5486 | vcpu->arch.pio.count = 0; |
0912c977 | 5487 | } else { |
7ae441ea | 5488 | writeback = false; |
716d51ab GN |
5489 | vcpu->arch.complete_userspace_io = complete_emulated_pio; |
5490 | } | |
ac0a48c3 | 5491 | r = EMULATE_USER_EXIT; |
7ae441ea GN |
5492 | } else if (vcpu->mmio_needed) { |
5493 | if (!vcpu->mmio_is_write) | |
5494 | writeback = false; | |
ac0a48c3 | 5495 | r = EMULATE_USER_EXIT; |
716d51ab | 5496 | vcpu->arch.complete_userspace_io = complete_emulated_mmio; |
7ae441ea | 5497 | } else if (r == EMULATION_RESTART) |
5cd21917 | 5498 | goto restart; |
d2ddd1c4 GN |
5499 | else |
5500 | r = EMULATE_DONE; | |
f850e2e6 | 5501 | |
7ae441ea | 5502 | if (writeback) { |
6addfc42 | 5503 | unsigned long rflags = kvm_x86_ops->get_rflags(vcpu); |
9d74191a | 5504 | toggle_interruptibility(vcpu, ctxt->interruptibility); |
7ae441ea | 5505 | vcpu->arch.emulate_regs_need_sync_to_vcpu = false; |
a584539b PB |
5506 | if (vcpu->arch.hflags != ctxt->emul_flags) |
5507 | kvm_set_hflags(vcpu, ctxt->emul_flags); | |
9d74191a | 5508 | kvm_rip_write(vcpu, ctxt->eip); |
663f4c61 | 5509 | if (r == EMULATE_DONE) |
6addfc42 | 5510 | kvm_vcpu_check_singlestep(vcpu, rflags, &r); |
38827dbd NA |
5511 | if (!ctxt->have_exception || |
5512 | exception_type(ctxt->exception.vector) == EXCPT_TRAP) | |
5513 | __kvm_set_rflags(vcpu, ctxt->eflags); | |
6addfc42 PB |
5514 | |
5515 | /* | |
5516 | * For STI, interrupts are shadowed; so KVM_REQ_EVENT will | |
5517 | * do nothing, and it will be requested again as soon as | |
5518 | * the shadow expires. But we still need to check here, | |
5519 | * because POPF has no interrupt shadow. | |
5520 | */ | |
5521 | if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF)) | |
5522 | kvm_make_request(KVM_REQ_EVENT, vcpu); | |
7ae441ea GN |
5523 | } else |
5524 | vcpu->arch.emulate_regs_need_sync_to_vcpu = true; | |
e85d28f8 GN |
5525 | |
5526 | return r; | |
de7d789a | 5527 | } |
51d8b661 | 5528 | EXPORT_SYMBOL_GPL(x86_emulate_instruction); |
de7d789a | 5529 | |
cf8f70bf | 5530 | int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port) |
de7d789a | 5531 | { |
cf8f70bf | 5532 | unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX); |
ca1d4a9e AK |
5533 | int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt, |
5534 | size, port, &val, 1); | |
cf8f70bf | 5535 | /* do not return to emulator after return from userspace */ |
7972995b | 5536 | vcpu->arch.pio.count = 0; |
de7d789a CO |
5537 | return ret; |
5538 | } | |
cf8f70bf | 5539 | EXPORT_SYMBOL_GPL(kvm_fast_pio_out); |
de7d789a | 5540 | |
8cfdc000 ZA |
5541 | static void tsc_bad(void *info) |
5542 | { | |
0a3aee0d | 5543 | __this_cpu_write(cpu_tsc_khz, 0); |
8cfdc000 ZA |
5544 | } |
5545 | ||
5546 | static void tsc_khz_changed(void *data) | |
c8076604 | 5547 | { |
8cfdc000 ZA |
5548 | struct cpufreq_freqs *freq = data; |
5549 | unsigned long khz = 0; | |
5550 | ||
5551 | if (data) | |
5552 | khz = freq->new; | |
5553 | else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) | |
5554 | khz = cpufreq_quick_get(raw_smp_processor_id()); | |
5555 | if (!khz) | |
5556 | khz = tsc_khz; | |
0a3aee0d | 5557 | __this_cpu_write(cpu_tsc_khz, khz); |
c8076604 GH |
5558 | } |
5559 | ||
c8076604 GH |
5560 | static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val, |
5561 | void *data) | |
5562 | { | |
5563 | struct cpufreq_freqs *freq = data; | |
5564 | struct kvm *kvm; | |
5565 | struct kvm_vcpu *vcpu; | |
5566 | int i, send_ipi = 0; | |
5567 | ||
8cfdc000 ZA |
5568 | /* |
5569 | * We allow guests to temporarily run on slowing clocks, | |
5570 | * provided we notify them after, or to run on accelerating | |
5571 | * clocks, provided we notify them before. Thus time never | |
5572 | * goes backwards. | |
5573 | * | |
5574 | * However, we have a problem. We can't atomically update | |
5575 | * the frequency of a given CPU from this function; it is | |
5576 | * merely a notifier, which can be called from any CPU. | |
5577 | * Changing the TSC frequency at arbitrary points in time | |
5578 | * requires a recomputation of local variables related to | |
5579 | * the TSC for each VCPU. We must flag these local variables | |
5580 | * to be updated and be sure the update takes place with the | |
5581 | * new frequency before any guests proceed. | |
5582 | * | |
5583 | * Unfortunately, the combination of hotplug CPU and frequency | |
5584 | * change creates an intractable locking scenario; the order | |
5585 | * of when these callouts happen is undefined with respect to | |
5586 | * CPU hotplug, and they can race with each other. As such, | |
5587 | * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is | |
5588 | * undefined; you can actually have a CPU frequency change take | |
5589 | * place in between the computation of X and the setting of the | |
5590 | * variable. To protect against this problem, all updates of | |
5591 | * the per_cpu tsc_khz variable are done in an interrupt | |
5592 | * protected IPI, and all callers wishing to update the value | |
5593 | * must wait for a synchronous IPI to complete (which is trivial | |
5594 | * if the caller is on the CPU already). This establishes the | |
5595 | * necessary total order on variable updates. | |
5596 | * | |
5597 | * Note that because a guest time update may take place | |
5598 | * anytime after the setting of the VCPU's request bit, the | |
5599 | * correct TSC value must be set before the request. However, | |
5600 | * to ensure the update actually makes it to any guest which | |
5601 | * starts running in hardware virtualization between the set | |
5602 | * and the acquisition of the spinlock, we must also ping the | |
5603 | * CPU after setting the request bit. | |
5604 | * | |
5605 | */ | |
5606 | ||
c8076604 GH |
5607 | if (val == CPUFREQ_PRECHANGE && freq->old > freq->new) |
5608 | return 0; | |
5609 | if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new) | |
5610 | return 0; | |
8cfdc000 ZA |
5611 | |
5612 | smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1); | |
c8076604 | 5613 | |
2f303b74 | 5614 | spin_lock(&kvm_lock); |
c8076604 | 5615 | list_for_each_entry(kvm, &vm_list, vm_list) { |
988a2cae | 5616 | kvm_for_each_vcpu(i, vcpu, kvm) { |
c8076604 GH |
5617 | if (vcpu->cpu != freq->cpu) |
5618 | continue; | |
c285545f | 5619 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu); |
c8076604 | 5620 | if (vcpu->cpu != smp_processor_id()) |
8cfdc000 | 5621 | send_ipi = 1; |
c8076604 GH |
5622 | } |
5623 | } | |
2f303b74 | 5624 | spin_unlock(&kvm_lock); |
c8076604 GH |
5625 | |
5626 | if (freq->old < freq->new && send_ipi) { | |
5627 | /* | |
5628 | * We upscale the frequency. Must make the guest | |
5629 | * doesn't see old kvmclock values while running with | |
5630 | * the new frequency, otherwise we risk the guest sees | |
5631 | * time go backwards. | |
5632 | * | |
5633 | * In case we update the frequency for another cpu | |
5634 | * (which might be in guest context) send an interrupt | |
5635 | * to kick the cpu out of guest context. Next time | |
5636 | * guest context is entered kvmclock will be updated, | |
5637 | * so the guest will not see stale values. | |
5638 | */ | |
8cfdc000 | 5639 | smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1); |
c8076604 GH |
5640 | } |
5641 | return 0; | |
5642 | } | |
5643 | ||
5644 | static struct notifier_block kvmclock_cpufreq_notifier_block = { | |
8cfdc000 ZA |
5645 | .notifier_call = kvmclock_cpufreq_notifier |
5646 | }; | |
5647 | ||
5648 | static int kvmclock_cpu_notifier(struct notifier_block *nfb, | |
5649 | unsigned long action, void *hcpu) | |
5650 | { | |
5651 | unsigned int cpu = (unsigned long)hcpu; | |
5652 | ||
5653 | switch (action) { | |
5654 | case CPU_ONLINE: | |
5655 | case CPU_DOWN_FAILED: | |
5656 | smp_call_function_single(cpu, tsc_khz_changed, NULL, 1); | |
5657 | break; | |
5658 | case CPU_DOWN_PREPARE: | |
5659 | smp_call_function_single(cpu, tsc_bad, NULL, 1); | |
5660 | break; | |
5661 | } | |
5662 | return NOTIFY_OK; | |
5663 | } | |
5664 | ||
5665 | static struct notifier_block kvmclock_cpu_notifier_block = { | |
5666 | .notifier_call = kvmclock_cpu_notifier, | |
5667 | .priority = -INT_MAX | |
c8076604 GH |
5668 | }; |
5669 | ||
b820cc0c ZA |
5670 | static void kvm_timer_init(void) |
5671 | { | |
5672 | int cpu; | |
5673 | ||
c285545f | 5674 | max_tsc_khz = tsc_khz; |
460dd42e SB |
5675 | |
5676 | cpu_notifier_register_begin(); | |
b820cc0c | 5677 | if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) { |
c285545f ZA |
5678 | #ifdef CONFIG_CPU_FREQ |
5679 | struct cpufreq_policy policy; | |
5680 | memset(&policy, 0, sizeof(policy)); | |
3e26f230 AK |
5681 | cpu = get_cpu(); |
5682 | cpufreq_get_policy(&policy, cpu); | |
c285545f ZA |
5683 | if (policy.cpuinfo.max_freq) |
5684 | max_tsc_khz = policy.cpuinfo.max_freq; | |
3e26f230 | 5685 | put_cpu(); |
c285545f | 5686 | #endif |
b820cc0c ZA |
5687 | cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block, |
5688 | CPUFREQ_TRANSITION_NOTIFIER); | |
5689 | } | |
c285545f | 5690 | pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz); |
8cfdc000 ZA |
5691 | for_each_online_cpu(cpu) |
5692 | smp_call_function_single(cpu, tsc_khz_changed, NULL, 1); | |
460dd42e SB |
5693 | |
5694 | __register_hotcpu_notifier(&kvmclock_cpu_notifier_block); | |
5695 | cpu_notifier_register_done(); | |
5696 | ||
b820cc0c ZA |
5697 | } |
5698 | ||
ff9d07a0 ZY |
5699 | static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu); |
5700 | ||
f5132b01 | 5701 | int kvm_is_in_guest(void) |
ff9d07a0 | 5702 | { |
086c9855 | 5703 | return __this_cpu_read(current_vcpu) != NULL; |
ff9d07a0 ZY |
5704 | } |
5705 | ||
5706 | static int kvm_is_user_mode(void) | |
5707 | { | |
5708 | int user_mode = 3; | |
dcf46b94 | 5709 | |
086c9855 AS |
5710 | if (__this_cpu_read(current_vcpu)) |
5711 | user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu)); | |
dcf46b94 | 5712 | |
ff9d07a0 ZY |
5713 | return user_mode != 0; |
5714 | } | |
5715 | ||
5716 | static unsigned long kvm_get_guest_ip(void) | |
5717 | { | |
5718 | unsigned long ip = 0; | |
dcf46b94 | 5719 | |
086c9855 AS |
5720 | if (__this_cpu_read(current_vcpu)) |
5721 | ip = kvm_rip_read(__this_cpu_read(current_vcpu)); | |
dcf46b94 | 5722 | |
ff9d07a0 ZY |
5723 | return ip; |
5724 | } | |
5725 | ||
5726 | static struct perf_guest_info_callbacks kvm_guest_cbs = { | |
5727 | .is_in_guest = kvm_is_in_guest, | |
5728 | .is_user_mode = kvm_is_user_mode, | |
5729 | .get_guest_ip = kvm_get_guest_ip, | |
5730 | }; | |
5731 | ||
5732 | void kvm_before_handle_nmi(struct kvm_vcpu *vcpu) | |
5733 | { | |
086c9855 | 5734 | __this_cpu_write(current_vcpu, vcpu); |
ff9d07a0 ZY |
5735 | } |
5736 | EXPORT_SYMBOL_GPL(kvm_before_handle_nmi); | |
5737 | ||
5738 | void kvm_after_handle_nmi(struct kvm_vcpu *vcpu) | |
5739 | { | |
086c9855 | 5740 | __this_cpu_write(current_vcpu, NULL); |
ff9d07a0 ZY |
5741 | } |
5742 | EXPORT_SYMBOL_GPL(kvm_after_handle_nmi); | |
5743 | ||
ce88decf XG |
5744 | static void kvm_set_mmio_spte_mask(void) |
5745 | { | |
5746 | u64 mask; | |
5747 | int maxphyaddr = boot_cpu_data.x86_phys_bits; | |
5748 | ||
5749 | /* | |
5750 | * Set the reserved bits and the present bit of an paging-structure | |
5751 | * entry to generate page fault with PFER.RSV = 1. | |
5752 | */ | |
885032b9 | 5753 | /* Mask the reserved physical address bits. */ |
d1431483 | 5754 | mask = rsvd_bits(maxphyaddr, 51); |
885032b9 XG |
5755 | |
5756 | /* Bit 62 is always reserved for 32bit host. */ | |
5757 | mask |= 0x3ull << 62; | |
5758 | ||
5759 | /* Set the present bit. */ | |
ce88decf XG |
5760 | mask |= 1ull; |
5761 | ||
5762 | #ifdef CONFIG_X86_64 | |
5763 | /* | |
5764 | * If reserved bit is not supported, clear the present bit to disable | |
5765 | * mmio page fault. | |
5766 | */ | |
5767 | if (maxphyaddr == 52) | |
5768 | mask &= ~1ull; | |
5769 | #endif | |
5770 | ||
5771 | kvm_mmu_set_mmio_spte_mask(mask); | |
5772 | } | |
5773 | ||
16e8d74d MT |
5774 | #ifdef CONFIG_X86_64 |
5775 | static void pvclock_gtod_update_fn(struct work_struct *work) | |
5776 | { | |
d828199e MT |
5777 | struct kvm *kvm; |
5778 | ||
5779 | struct kvm_vcpu *vcpu; | |
5780 | int i; | |
5781 | ||
2f303b74 | 5782 | spin_lock(&kvm_lock); |
d828199e MT |
5783 | list_for_each_entry(kvm, &vm_list, vm_list) |
5784 | kvm_for_each_vcpu(i, vcpu, kvm) | |
105b21bb | 5785 | kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu); |
d828199e | 5786 | atomic_set(&kvm_guest_has_master_clock, 0); |
2f303b74 | 5787 | spin_unlock(&kvm_lock); |
16e8d74d MT |
5788 | } |
5789 | ||
5790 | static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn); | |
5791 | ||
5792 | /* | |
5793 | * Notification about pvclock gtod data update. | |
5794 | */ | |
5795 | static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused, | |
5796 | void *priv) | |
5797 | { | |
5798 | struct pvclock_gtod_data *gtod = &pvclock_gtod_data; | |
5799 | struct timekeeper *tk = priv; | |
5800 | ||
5801 | update_pvclock_gtod(tk); | |
5802 | ||
5803 | /* disable master clock if host does not trust, or does not | |
5804 | * use, TSC clocksource | |
5805 | */ | |
5806 | if (gtod->clock.vclock_mode != VCLOCK_TSC && | |
5807 | atomic_read(&kvm_guest_has_master_clock) != 0) | |
5808 | queue_work(system_long_wq, &pvclock_gtod_work); | |
5809 | ||
5810 | return 0; | |
5811 | } | |
5812 | ||
5813 | static struct notifier_block pvclock_gtod_notifier = { | |
5814 | .notifier_call = pvclock_gtod_notify, | |
5815 | }; | |
5816 | #endif | |
5817 | ||
f8c16bba | 5818 | int kvm_arch_init(void *opaque) |
043405e1 | 5819 | { |
b820cc0c | 5820 | int r; |
6b61edf7 | 5821 | struct kvm_x86_ops *ops = opaque; |
f8c16bba | 5822 | |
f8c16bba ZX |
5823 | if (kvm_x86_ops) { |
5824 | printk(KERN_ERR "kvm: already loaded the other module\n"); | |
56c6d28a ZX |
5825 | r = -EEXIST; |
5826 | goto out; | |
f8c16bba ZX |
5827 | } |
5828 | ||
5829 | if (!ops->cpu_has_kvm_support()) { | |
5830 | printk(KERN_ERR "kvm: no hardware support\n"); | |
56c6d28a ZX |
5831 | r = -EOPNOTSUPP; |
5832 | goto out; | |
f8c16bba ZX |
5833 | } |
5834 | if (ops->disabled_by_bios()) { | |
5835 | printk(KERN_ERR "kvm: disabled by bios\n"); | |
56c6d28a ZX |
5836 | r = -EOPNOTSUPP; |
5837 | goto out; | |
f8c16bba ZX |
5838 | } |
5839 | ||
013f6a5d MT |
5840 | r = -ENOMEM; |
5841 | shared_msrs = alloc_percpu(struct kvm_shared_msrs); | |
5842 | if (!shared_msrs) { | |
5843 | printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n"); | |
5844 | goto out; | |
5845 | } | |
5846 | ||
97db56ce AK |
5847 | r = kvm_mmu_module_init(); |
5848 | if (r) | |
013f6a5d | 5849 | goto out_free_percpu; |
97db56ce | 5850 | |
ce88decf | 5851 | kvm_set_mmio_spte_mask(); |
97db56ce | 5852 | |
f8c16bba | 5853 | kvm_x86_ops = ops; |
920c8377 | 5854 | |
7b52345e | 5855 | kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK, |
4b12f0de | 5856 | PT_DIRTY_MASK, PT64_NX_MASK, 0); |
c8076604 | 5857 | |
b820cc0c | 5858 | kvm_timer_init(); |
c8076604 | 5859 | |
ff9d07a0 ZY |
5860 | perf_register_guest_info_callbacks(&kvm_guest_cbs); |
5861 | ||
2acf923e DC |
5862 | if (cpu_has_xsave) |
5863 | host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK); | |
5864 | ||
c5cc421b | 5865 | kvm_lapic_init(); |
16e8d74d MT |
5866 | #ifdef CONFIG_X86_64 |
5867 | pvclock_gtod_register_notifier(&pvclock_gtod_notifier); | |
5868 | #endif | |
5869 | ||
f8c16bba | 5870 | return 0; |
56c6d28a | 5871 | |
013f6a5d MT |
5872 | out_free_percpu: |
5873 | free_percpu(shared_msrs); | |
56c6d28a | 5874 | out: |
56c6d28a | 5875 | return r; |
043405e1 | 5876 | } |
8776e519 | 5877 | |
f8c16bba ZX |
5878 | void kvm_arch_exit(void) |
5879 | { | |
ff9d07a0 ZY |
5880 | perf_unregister_guest_info_callbacks(&kvm_guest_cbs); |
5881 | ||
888d256e JK |
5882 | if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) |
5883 | cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block, | |
5884 | CPUFREQ_TRANSITION_NOTIFIER); | |
8cfdc000 | 5885 | unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block); |
16e8d74d MT |
5886 | #ifdef CONFIG_X86_64 |
5887 | pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier); | |
5888 | #endif | |
f8c16bba | 5889 | kvm_x86_ops = NULL; |
56c6d28a | 5890 | kvm_mmu_module_exit(); |
013f6a5d | 5891 | free_percpu(shared_msrs); |
56c6d28a | 5892 | } |
f8c16bba | 5893 | |
5cb56059 | 5894 | int kvm_vcpu_halt(struct kvm_vcpu *vcpu) |
8776e519 HB |
5895 | { |
5896 | ++vcpu->stat.halt_exits; | |
35754c98 | 5897 | if (lapic_in_kernel(vcpu)) { |
a4535290 | 5898 | vcpu->arch.mp_state = KVM_MP_STATE_HALTED; |
8776e519 HB |
5899 | return 1; |
5900 | } else { | |
5901 | vcpu->run->exit_reason = KVM_EXIT_HLT; | |
5902 | return 0; | |
5903 | } | |
5904 | } | |
5cb56059 JS |
5905 | EXPORT_SYMBOL_GPL(kvm_vcpu_halt); |
5906 | ||
5907 | int kvm_emulate_halt(struct kvm_vcpu *vcpu) | |
5908 | { | |
5909 | kvm_x86_ops->skip_emulated_instruction(vcpu); | |
5910 | return kvm_vcpu_halt(vcpu); | |
5911 | } | |
8776e519 HB |
5912 | EXPORT_SYMBOL_GPL(kvm_emulate_halt); |
5913 | ||
6aef266c SV |
5914 | /* |
5915 | * kvm_pv_kick_cpu_op: Kick a vcpu. | |
5916 | * | |
5917 | * @apicid - apicid of vcpu to be kicked. | |
5918 | */ | |
5919 | static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid) | |
5920 | { | |
24d2166b | 5921 | struct kvm_lapic_irq lapic_irq; |
6aef266c | 5922 | |
24d2166b R |
5923 | lapic_irq.shorthand = 0; |
5924 | lapic_irq.dest_mode = 0; | |
5925 | lapic_irq.dest_id = apicid; | |
93bbf0b8 | 5926 | lapic_irq.msi_redir_hint = false; |
6aef266c | 5927 | |
24d2166b | 5928 | lapic_irq.delivery_mode = APIC_DM_REMRD; |
795a149e | 5929 | kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL); |
6aef266c SV |
5930 | } |
5931 | ||
d62caabb AS |
5932 | void kvm_vcpu_deactivate_apicv(struct kvm_vcpu *vcpu) |
5933 | { | |
5934 | vcpu->arch.apicv_active = false; | |
5935 | kvm_x86_ops->refresh_apicv_exec_ctrl(vcpu); | |
5936 | } | |
5937 | ||
8776e519 HB |
5938 | int kvm_emulate_hypercall(struct kvm_vcpu *vcpu) |
5939 | { | |
5940 | unsigned long nr, a0, a1, a2, a3, ret; | |
a449c7aa | 5941 | int op_64_bit, r = 1; |
8776e519 | 5942 | |
5cb56059 JS |
5943 | kvm_x86_ops->skip_emulated_instruction(vcpu); |
5944 | ||
55cd8e5a GN |
5945 | if (kvm_hv_hypercall_enabled(vcpu->kvm)) |
5946 | return kvm_hv_hypercall(vcpu); | |
5947 | ||
5fdbf976 MT |
5948 | nr = kvm_register_read(vcpu, VCPU_REGS_RAX); |
5949 | a0 = kvm_register_read(vcpu, VCPU_REGS_RBX); | |
5950 | a1 = kvm_register_read(vcpu, VCPU_REGS_RCX); | |
5951 | a2 = kvm_register_read(vcpu, VCPU_REGS_RDX); | |
5952 | a3 = kvm_register_read(vcpu, VCPU_REGS_RSI); | |
8776e519 | 5953 | |
229456fc | 5954 | trace_kvm_hypercall(nr, a0, a1, a2, a3); |
2714d1d3 | 5955 | |
a449c7aa NA |
5956 | op_64_bit = is_64_bit_mode(vcpu); |
5957 | if (!op_64_bit) { | |
8776e519 HB |
5958 | nr &= 0xFFFFFFFF; |
5959 | a0 &= 0xFFFFFFFF; | |
5960 | a1 &= 0xFFFFFFFF; | |
5961 | a2 &= 0xFFFFFFFF; | |
5962 | a3 &= 0xFFFFFFFF; | |
5963 | } | |
5964 | ||
07708c4a JK |
5965 | if (kvm_x86_ops->get_cpl(vcpu) != 0) { |
5966 | ret = -KVM_EPERM; | |
5967 | goto out; | |
5968 | } | |
5969 | ||
8776e519 | 5970 | switch (nr) { |
b93463aa AK |
5971 | case KVM_HC_VAPIC_POLL_IRQ: |
5972 | ret = 0; | |
5973 | break; | |
6aef266c SV |
5974 | case KVM_HC_KICK_CPU: |
5975 | kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1); | |
5976 | ret = 0; | |
5977 | break; | |
8776e519 HB |
5978 | default: |
5979 | ret = -KVM_ENOSYS; | |
5980 | break; | |
5981 | } | |
07708c4a | 5982 | out: |
a449c7aa NA |
5983 | if (!op_64_bit) |
5984 | ret = (u32)ret; | |
5fdbf976 | 5985 | kvm_register_write(vcpu, VCPU_REGS_RAX, ret); |
f11c3a8d | 5986 | ++vcpu->stat.hypercalls; |
2f333bcb | 5987 | return r; |
8776e519 HB |
5988 | } |
5989 | EXPORT_SYMBOL_GPL(kvm_emulate_hypercall); | |
5990 | ||
b6785def | 5991 | static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt) |
8776e519 | 5992 | { |
d6aa1000 | 5993 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
8776e519 | 5994 | char instruction[3]; |
5fdbf976 | 5995 | unsigned long rip = kvm_rip_read(vcpu); |
8776e519 | 5996 | |
8776e519 | 5997 | kvm_x86_ops->patch_hypercall(vcpu, instruction); |
8776e519 | 5998 | |
9d74191a | 5999 | return emulator_write_emulated(ctxt, rip, instruction, 3, NULL); |
8776e519 HB |
6000 | } |
6001 | ||
851ba692 | 6002 | static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu) |
b6c7a5dc | 6003 | { |
782d422b MG |
6004 | return vcpu->run->request_interrupt_window && |
6005 | likely(!pic_in_kernel(vcpu->kvm)); | |
b6c7a5dc HB |
6006 | } |
6007 | ||
851ba692 | 6008 | static void post_kvm_run_save(struct kvm_vcpu *vcpu) |
b6c7a5dc | 6009 | { |
851ba692 AK |
6010 | struct kvm_run *kvm_run = vcpu->run; |
6011 | ||
91586a3b | 6012 | kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0; |
f077825a | 6013 | kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0; |
2d3ad1f4 | 6014 | kvm_run->cr8 = kvm_get_cr8(vcpu); |
b6c7a5dc | 6015 | kvm_run->apic_base = kvm_get_apic_base(vcpu); |
127a457a MG |
6016 | kvm_run->ready_for_interrupt_injection = |
6017 | pic_in_kernel(vcpu->kvm) || | |
782d422b | 6018 | kvm_vcpu_ready_for_interrupt_injection(vcpu); |
b6c7a5dc HB |
6019 | } |
6020 | ||
95ba8273 GN |
6021 | static void update_cr8_intercept(struct kvm_vcpu *vcpu) |
6022 | { | |
6023 | int max_irr, tpr; | |
6024 | ||
6025 | if (!kvm_x86_ops->update_cr8_intercept) | |
6026 | return; | |
6027 | ||
bce87cce | 6028 | if (!lapic_in_kernel(vcpu)) |
88c808fd AK |
6029 | return; |
6030 | ||
d62caabb AS |
6031 | if (vcpu->arch.apicv_active) |
6032 | return; | |
6033 | ||
8db3baa2 GN |
6034 | if (!vcpu->arch.apic->vapic_addr) |
6035 | max_irr = kvm_lapic_find_highest_irr(vcpu); | |
6036 | else | |
6037 | max_irr = -1; | |
95ba8273 GN |
6038 | |
6039 | if (max_irr != -1) | |
6040 | max_irr >>= 4; | |
6041 | ||
6042 | tpr = kvm_lapic_get_cr8(vcpu); | |
6043 | ||
6044 | kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr); | |
6045 | } | |
6046 | ||
b6b8a145 | 6047 | static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win) |
95ba8273 | 6048 | { |
b6b8a145 JK |
6049 | int r; |
6050 | ||
95ba8273 | 6051 | /* try to reinject previous events if any */ |
b59bb7bd | 6052 | if (vcpu->arch.exception.pending) { |
5c1c85d0 AK |
6053 | trace_kvm_inj_exception(vcpu->arch.exception.nr, |
6054 | vcpu->arch.exception.has_error_code, | |
6055 | vcpu->arch.exception.error_code); | |
d6e8c854 NA |
6056 | |
6057 | if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT) | |
6058 | __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) | | |
6059 | X86_EFLAGS_RF); | |
6060 | ||
6bdf0662 NA |
6061 | if (vcpu->arch.exception.nr == DB_VECTOR && |
6062 | (vcpu->arch.dr7 & DR7_GD)) { | |
6063 | vcpu->arch.dr7 &= ~DR7_GD; | |
6064 | kvm_update_dr7(vcpu); | |
6065 | } | |
6066 | ||
b59bb7bd GN |
6067 | kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr, |
6068 | vcpu->arch.exception.has_error_code, | |
ce7ddec4 JR |
6069 | vcpu->arch.exception.error_code, |
6070 | vcpu->arch.exception.reinject); | |
b6b8a145 | 6071 | return 0; |
b59bb7bd GN |
6072 | } |
6073 | ||
95ba8273 GN |
6074 | if (vcpu->arch.nmi_injected) { |
6075 | kvm_x86_ops->set_nmi(vcpu); | |
b6b8a145 | 6076 | return 0; |
95ba8273 GN |
6077 | } |
6078 | ||
6079 | if (vcpu->arch.interrupt.pending) { | |
66fd3f7f | 6080 | kvm_x86_ops->set_irq(vcpu); |
b6b8a145 JK |
6081 | return 0; |
6082 | } | |
6083 | ||
6084 | if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) { | |
6085 | r = kvm_x86_ops->check_nested_events(vcpu, req_int_win); | |
6086 | if (r != 0) | |
6087 | return r; | |
95ba8273 GN |
6088 | } |
6089 | ||
6090 | /* try to inject new event if pending */ | |
6091 | if (vcpu->arch.nmi_pending) { | |
6092 | if (kvm_x86_ops->nmi_allowed(vcpu)) { | |
7460fb4a | 6093 | --vcpu->arch.nmi_pending; |
95ba8273 GN |
6094 | vcpu->arch.nmi_injected = true; |
6095 | kvm_x86_ops->set_nmi(vcpu); | |
6096 | } | |
c7c9c56c | 6097 | } else if (kvm_cpu_has_injectable_intr(vcpu)) { |
9242b5b6 BD |
6098 | /* |
6099 | * Because interrupts can be injected asynchronously, we are | |
6100 | * calling check_nested_events again here to avoid a race condition. | |
6101 | * See https://lkml.org/lkml/2014/7/2/60 for discussion about this | |
6102 | * proposal and current concerns. Perhaps we should be setting | |
6103 | * KVM_REQ_EVENT only on certain events and not unconditionally? | |
6104 | */ | |
6105 | if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) { | |
6106 | r = kvm_x86_ops->check_nested_events(vcpu, req_int_win); | |
6107 | if (r != 0) | |
6108 | return r; | |
6109 | } | |
95ba8273 | 6110 | if (kvm_x86_ops->interrupt_allowed(vcpu)) { |
66fd3f7f GN |
6111 | kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu), |
6112 | false); | |
6113 | kvm_x86_ops->set_irq(vcpu); | |
95ba8273 GN |
6114 | } |
6115 | } | |
b6b8a145 | 6116 | return 0; |
95ba8273 GN |
6117 | } |
6118 | ||
7460fb4a AK |
6119 | static void process_nmi(struct kvm_vcpu *vcpu) |
6120 | { | |
6121 | unsigned limit = 2; | |
6122 | ||
6123 | /* | |
6124 | * x86 is limited to one NMI running, and one NMI pending after it. | |
6125 | * If an NMI is already in progress, limit further NMIs to just one. | |
6126 | * Otherwise, allow two (and we'll inject the first one immediately). | |
6127 | */ | |
6128 | if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected) | |
6129 | limit = 1; | |
6130 | ||
6131 | vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0); | |
6132 | vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit); | |
6133 | kvm_make_request(KVM_REQ_EVENT, vcpu); | |
6134 | } | |
6135 | ||
660a5d51 PB |
6136 | #define put_smstate(type, buf, offset, val) \ |
6137 | *(type *)((buf) + (offset) - 0x7e00) = val | |
6138 | ||
6139 | static u32 process_smi_get_segment_flags(struct kvm_segment *seg) | |
6140 | { | |
6141 | u32 flags = 0; | |
6142 | flags |= seg->g << 23; | |
6143 | flags |= seg->db << 22; | |
6144 | flags |= seg->l << 21; | |
6145 | flags |= seg->avl << 20; | |
6146 | flags |= seg->present << 15; | |
6147 | flags |= seg->dpl << 13; | |
6148 | flags |= seg->s << 12; | |
6149 | flags |= seg->type << 8; | |
6150 | return flags; | |
6151 | } | |
6152 | ||
6153 | static void process_smi_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n) | |
6154 | { | |
6155 | struct kvm_segment seg; | |
6156 | int offset; | |
6157 | ||
6158 | kvm_get_segment(vcpu, &seg, n); | |
6159 | put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector); | |
6160 | ||
6161 | if (n < 3) | |
6162 | offset = 0x7f84 + n * 12; | |
6163 | else | |
6164 | offset = 0x7f2c + (n - 3) * 12; | |
6165 | ||
6166 | put_smstate(u32, buf, offset + 8, seg.base); | |
6167 | put_smstate(u32, buf, offset + 4, seg.limit); | |
6168 | put_smstate(u32, buf, offset, process_smi_get_segment_flags(&seg)); | |
6169 | } | |
6170 | ||
efbb288a | 6171 | #ifdef CONFIG_X86_64 |
660a5d51 PB |
6172 | static void process_smi_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n) |
6173 | { | |
6174 | struct kvm_segment seg; | |
6175 | int offset; | |
6176 | u16 flags; | |
6177 | ||
6178 | kvm_get_segment(vcpu, &seg, n); | |
6179 | offset = 0x7e00 + n * 16; | |
6180 | ||
6181 | flags = process_smi_get_segment_flags(&seg) >> 8; | |
6182 | put_smstate(u16, buf, offset, seg.selector); | |
6183 | put_smstate(u16, buf, offset + 2, flags); | |
6184 | put_smstate(u32, buf, offset + 4, seg.limit); | |
6185 | put_smstate(u64, buf, offset + 8, seg.base); | |
6186 | } | |
efbb288a | 6187 | #endif |
660a5d51 PB |
6188 | |
6189 | static void process_smi_save_state_32(struct kvm_vcpu *vcpu, char *buf) | |
6190 | { | |
6191 | struct desc_ptr dt; | |
6192 | struct kvm_segment seg; | |
6193 | unsigned long val; | |
6194 | int i; | |
6195 | ||
6196 | put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu)); | |
6197 | put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu)); | |
6198 | put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu)); | |
6199 | put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu)); | |
6200 | ||
6201 | for (i = 0; i < 8; i++) | |
6202 | put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i)); | |
6203 | ||
6204 | kvm_get_dr(vcpu, 6, &val); | |
6205 | put_smstate(u32, buf, 0x7fcc, (u32)val); | |
6206 | kvm_get_dr(vcpu, 7, &val); | |
6207 | put_smstate(u32, buf, 0x7fc8, (u32)val); | |
6208 | ||
6209 | kvm_get_segment(vcpu, &seg, VCPU_SREG_TR); | |
6210 | put_smstate(u32, buf, 0x7fc4, seg.selector); | |
6211 | put_smstate(u32, buf, 0x7f64, seg.base); | |
6212 | put_smstate(u32, buf, 0x7f60, seg.limit); | |
6213 | put_smstate(u32, buf, 0x7f5c, process_smi_get_segment_flags(&seg)); | |
6214 | ||
6215 | kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR); | |
6216 | put_smstate(u32, buf, 0x7fc0, seg.selector); | |
6217 | put_smstate(u32, buf, 0x7f80, seg.base); | |
6218 | put_smstate(u32, buf, 0x7f7c, seg.limit); | |
6219 | put_smstate(u32, buf, 0x7f78, process_smi_get_segment_flags(&seg)); | |
6220 | ||
6221 | kvm_x86_ops->get_gdt(vcpu, &dt); | |
6222 | put_smstate(u32, buf, 0x7f74, dt.address); | |
6223 | put_smstate(u32, buf, 0x7f70, dt.size); | |
6224 | ||
6225 | kvm_x86_ops->get_idt(vcpu, &dt); | |
6226 | put_smstate(u32, buf, 0x7f58, dt.address); | |
6227 | put_smstate(u32, buf, 0x7f54, dt.size); | |
6228 | ||
6229 | for (i = 0; i < 6; i++) | |
6230 | process_smi_save_seg_32(vcpu, buf, i); | |
6231 | ||
6232 | put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu)); | |
6233 | ||
6234 | /* revision id */ | |
6235 | put_smstate(u32, buf, 0x7efc, 0x00020000); | |
6236 | put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase); | |
6237 | } | |
6238 | ||
6239 | static void process_smi_save_state_64(struct kvm_vcpu *vcpu, char *buf) | |
6240 | { | |
6241 | #ifdef CONFIG_X86_64 | |
6242 | struct desc_ptr dt; | |
6243 | struct kvm_segment seg; | |
6244 | unsigned long val; | |
6245 | int i; | |
6246 | ||
6247 | for (i = 0; i < 16; i++) | |
6248 | put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i)); | |
6249 | ||
6250 | put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu)); | |
6251 | put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu)); | |
6252 | ||
6253 | kvm_get_dr(vcpu, 6, &val); | |
6254 | put_smstate(u64, buf, 0x7f68, val); | |
6255 | kvm_get_dr(vcpu, 7, &val); | |
6256 | put_smstate(u64, buf, 0x7f60, val); | |
6257 | ||
6258 | put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu)); | |
6259 | put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu)); | |
6260 | put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu)); | |
6261 | ||
6262 | put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase); | |
6263 | ||
6264 | /* revision id */ | |
6265 | put_smstate(u32, buf, 0x7efc, 0x00020064); | |
6266 | ||
6267 | put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer); | |
6268 | ||
6269 | kvm_get_segment(vcpu, &seg, VCPU_SREG_TR); | |
6270 | put_smstate(u16, buf, 0x7e90, seg.selector); | |
6271 | put_smstate(u16, buf, 0x7e92, process_smi_get_segment_flags(&seg) >> 8); | |
6272 | put_smstate(u32, buf, 0x7e94, seg.limit); | |
6273 | put_smstate(u64, buf, 0x7e98, seg.base); | |
6274 | ||
6275 | kvm_x86_ops->get_idt(vcpu, &dt); | |
6276 | put_smstate(u32, buf, 0x7e84, dt.size); | |
6277 | put_smstate(u64, buf, 0x7e88, dt.address); | |
6278 | ||
6279 | kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR); | |
6280 | put_smstate(u16, buf, 0x7e70, seg.selector); | |
6281 | put_smstate(u16, buf, 0x7e72, process_smi_get_segment_flags(&seg) >> 8); | |
6282 | put_smstate(u32, buf, 0x7e74, seg.limit); | |
6283 | put_smstate(u64, buf, 0x7e78, seg.base); | |
6284 | ||
6285 | kvm_x86_ops->get_gdt(vcpu, &dt); | |
6286 | put_smstate(u32, buf, 0x7e64, dt.size); | |
6287 | put_smstate(u64, buf, 0x7e68, dt.address); | |
6288 | ||
6289 | for (i = 0; i < 6; i++) | |
6290 | process_smi_save_seg_64(vcpu, buf, i); | |
6291 | #else | |
6292 | WARN_ON_ONCE(1); | |
6293 | #endif | |
6294 | } | |
6295 | ||
64d60670 PB |
6296 | static void process_smi(struct kvm_vcpu *vcpu) |
6297 | { | |
660a5d51 | 6298 | struct kvm_segment cs, ds; |
18c3626e | 6299 | struct desc_ptr dt; |
660a5d51 PB |
6300 | char buf[512]; |
6301 | u32 cr0; | |
6302 | ||
64d60670 PB |
6303 | if (is_smm(vcpu)) { |
6304 | vcpu->arch.smi_pending = true; | |
6305 | return; | |
6306 | } | |
6307 | ||
660a5d51 PB |
6308 | trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true); |
6309 | vcpu->arch.hflags |= HF_SMM_MASK; | |
6310 | memset(buf, 0, 512); | |
6311 | if (guest_cpuid_has_longmode(vcpu)) | |
6312 | process_smi_save_state_64(vcpu, buf); | |
6313 | else | |
6314 | process_smi_save_state_32(vcpu, buf); | |
6315 | ||
54bf36aa | 6316 | kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf)); |
660a5d51 PB |
6317 | |
6318 | if (kvm_x86_ops->get_nmi_mask(vcpu)) | |
6319 | vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK; | |
6320 | else | |
6321 | kvm_x86_ops->set_nmi_mask(vcpu, true); | |
6322 | ||
6323 | kvm_set_rflags(vcpu, X86_EFLAGS_FIXED); | |
6324 | kvm_rip_write(vcpu, 0x8000); | |
6325 | ||
6326 | cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG); | |
6327 | kvm_x86_ops->set_cr0(vcpu, cr0); | |
6328 | vcpu->arch.cr0 = cr0; | |
6329 | ||
6330 | kvm_x86_ops->set_cr4(vcpu, 0); | |
6331 | ||
18c3626e PB |
6332 | /* Undocumented: IDT limit is set to zero on entry to SMM. */ |
6333 | dt.address = dt.size = 0; | |
6334 | kvm_x86_ops->set_idt(vcpu, &dt); | |
6335 | ||
660a5d51 PB |
6336 | __kvm_set_dr(vcpu, 7, DR7_FIXED_1); |
6337 | ||
6338 | cs.selector = (vcpu->arch.smbase >> 4) & 0xffff; | |
6339 | cs.base = vcpu->arch.smbase; | |
6340 | ||
6341 | ds.selector = 0; | |
6342 | ds.base = 0; | |
6343 | ||
6344 | cs.limit = ds.limit = 0xffffffff; | |
6345 | cs.type = ds.type = 0x3; | |
6346 | cs.dpl = ds.dpl = 0; | |
6347 | cs.db = ds.db = 0; | |
6348 | cs.s = ds.s = 1; | |
6349 | cs.l = ds.l = 0; | |
6350 | cs.g = ds.g = 1; | |
6351 | cs.avl = ds.avl = 0; | |
6352 | cs.present = ds.present = 1; | |
6353 | cs.unusable = ds.unusable = 0; | |
6354 | cs.padding = ds.padding = 0; | |
6355 | ||
6356 | kvm_set_segment(vcpu, &cs, VCPU_SREG_CS); | |
6357 | kvm_set_segment(vcpu, &ds, VCPU_SREG_DS); | |
6358 | kvm_set_segment(vcpu, &ds, VCPU_SREG_ES); | |
6359 | kvm_set_segment(vcpu, &ds, VCPU_SREG_FS); | |
6360 | kvm_set_segment(vcpu, &ds, VCPU_SREG_GS); | |
6361 | kvm_set_segment(vcpu, &ds, VCPU_SREG_SS); | |
6362 | ||
6363 | if (guest_cpuid_has_longmode(vcpu)) | |
6364 | kvm_x86_ops->set_efer(vcpu, 0); | |
6365 | ||
6366 | kvm_update_cpuid(vcpu); | |
6367 | kvm_mmu_reset_context(vcpu); | |
64d60670 PB |
6368 | } |
6369 | ||
2860c4b1 PB |
6370 | void kvm_make_scan_ioapic_request(struct kvm *kvm) |
6371 | { | |
6372 | kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC); | |
6373 | } | |
6374 | ||
3d81bc7e | 6375 | static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu) |
c7c9c56c | 6376 | { |
5c919412 AS |
6377 | u64 eoi_exit_bitmap[4]; |
6378 | ||
3d81bc7e YZ |
6379 | if (!kvm_apic_hw_enabled(vcpu->arch.apic)) |
6380 | return; | |
c7c9c56c | 6381 | |
6308630b | 6382 | bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256); |
c7c9c56c | 6383 | |
b053b2ae | 6384 | if (irqchip_split(vcpu->kvm)) |
6308630b | 6385 | kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors); |
db2bdcbb | 6386 | else { |
d62caabb AS |
6387 | if (vcpu->arch.apicv_active) |
6388 | kvm_x86_ops->sync_pir_to_irr(vcpu); | |
6308630b | 6389 | kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors); |
db2bdcbb | 6390 | } |
5c919412 AS |
6391 | bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors, |
6392 | vcpu_to_synic(vcpu)->vec_bitmap, 256); | |
6393 | kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap); | |
c7c9c56c YZ |
6394 | } |
6395 | ||
a70656b6 RK |
6396 | static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu) |
6397 | { | |
6398 | ++vcpu->stat.tlb_flush; | |
6399 | kvm_x86_ops->tlb_flush(vcpu); | |
6400 | } | |
6401 | ||
4256f43f TC |
6402 | void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu) |
6403 | { | |
c24ae0dc TC |
6404 | struct page *page = NULL; |
6405 | ||
35754c98 | 6406 | if (!lapic_in_kernel(vcpu)) |
f439ed27 PB |
6407 | return; |
6408 | ||
4256f43f TC |
6409 | if (!kvm_x86_ops->set_apic_access_page_addr) |
6410 | return; | |
6411 | ||
c24ae0dc | 6412 | page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT); |
e8fd5e9e AA |
6413 | if (is_error_page(page)) |
6414 | return; | |
c24ae0dc TC |
6415 | kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page)); |
6416 | ||
6417 | /* | |
6418 | * Do not pin apic access page in memory, the MMU notifier | |
6419 | * will call us again if it is migrated or swapped out. | |
6420 | */ | |
6421 | put_page(page); | |
4256f43f TC |
6422 | } |
6423 | EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page); | |
6424 | ||
fe71557a TC |
6425 | void kvm_arch_mmu_notifier_invalidate_page(struct kvm *kvm, |
6426 | unsigned long address) | |
6427 | { | |
c24ae0dc TC |
6428 | /* |
6429 | * The physical address of apic access page is stored in the VMCS. | |
6430 | * Update it when it becomes invalid. | |
6431 | */ | |
6432 | if (address == gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT)) | |
6433 | kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD); | |
fe71557a TC |
6434 | } |
6435 | ||
9357d939 | 6436 | /* |
362c698f | 6437 | * Returns 1 to let vcpu_run() continue the guest execution loop without |
9357d939 TY |
6438 | * exiting to the userspace. Otherwise, the value will be returned to the |
6439 | * userspace. | |
6440 | */ | |
851ba692 | 6441 | static int vcpu_enter_guest(struct kvm_vcpu *vcpu) |
b6c7a5dc HB |
6442 | { |
6443 | int r; | |
62a193ed MG |
6444 | bool req_int_win = |
6445 | dm_request_for_irq_injection(vcpu) && | |
6446 | kvm_cpu_accept_dm_intr(vcpu); | |
6447 | ||
730dca42 | 6448 | bool req_immediate_exit = false; |
b6c7a5dc | 6449 | |
3e007509 | 6450 | if (vcpu->requests) { |
a8eeb04a | 6451 | if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu)) |
2e53d63a | 6452 | kvm_mmu_unload(vcpu); |
a8eeb04a | 6453 | if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu)) |
2f599714 | 6454 | __kvm_migrate_timers(vcpu); |
d828199e MT |
6455 | if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu)) |
6456 | kvm_gen_update_masterclock(vcpu->kvm); | |
0061d53d MT |
6457 | if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu)) |
6458 | kvm_gen_kvmclock_update(vcpu); | |
34c238a1 ZA |
6459 | if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) { |
6460 | r = kvm_guest_time_update(vcpu); | |
8cfdc000 ZA |
6461 | if (unlikely(r)) |
6462 | goto out; | |
6463 | } | |
a8eeb04a | 6464 | if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu)) |
4731d4c7 | 6465 | kvm_mmu_sync_roots(vcpu); |
a8eeb04a | 6466 | if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) |
a70656b6 | 6467 | kvm_vcpu_flush_tlb(vcpu); |
a8eeb04a | 6468 | if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) { |
851ba692 | 6469 | vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS; |
b93463aa AK |
6470 | r = 0; |
6471 | goto out; | |
6472 | } | |
a8eeb04a | 6473 | if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) { |
851ba692 | 6474 | vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN; |
71c4dfaf JR |
6475 | r = 0; |
6476 | goto out; | |
6477 | } | |
a8eeb04a | 6478 | if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) { |
02daab21 AK |
6479 | vcpu->fpu_active = 0; |
6480 | kvm_x86_ops->fpu_deactivate(vcpu); | |
6481 | } | |
af585b92 GN |
6482 | if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) { |
6483 | /* Page is swapped out. Do synthetic halt */ | |
6484 | vcpu->arch.apf.halted = true; | |
6485 | r = 1; | |
6486 | goto out; | |
6487 | } | |
c9aaa895 GC |
6488 | if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu)) |
6489 | record_steal_time(vcpu); | |
64d60670 PB |
6490 | if (kvm_check_request(KVM_REQ_SMI, vcpu)) |
6491 | process_smi(vcpu); | |
7460fb4a AK |
6492 | if (kvm_check_request(KVM_REQ_NMI, vcpu)) |
6493 | process_nmi(vcpu); | |
f5132b01 | 6494 | if (kvm_check_request(KVM_REQ_PMU, vcpu)) |
c6702c9d | 6495 | kvm_pmu_handle_event(vcpu); |
f5132b01 | 6496 | if (kvm_check_request(KVM_REQ_PMI, vcpu)) |
c6702c9d | 6497 | kvm_pmu_deliver_pmi(vcpu); |
7543a635 SR |
6498 | if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) { |
6499 | BUG_ON(vcpu->arch.pending_ioapic_eoi > 255); | |
6500 | if (test_bit(vcpu->arch.pending_ioapic_eoi, | |
6308630b | 6501 | vcpu->arch.ioapic_handled_vectors)) { |
7543a635 SR |
6502 | vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI; |
6503 | vcpu->run->eoi.vector = | |
6504 | vcpu->arch.pending_ioapic_eoi; | |
6505 | r = 0; | |
6506 | goto out; | |
6507 | } | |
6508 | } | |
3d81bc7e YZ |
6509 | if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu)) |
6510 | vcpu_scan_ioapic(vcpu); | |
4256f43f TC |
6511 | if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu)) |
6512 | kvm_vcpu_reload_apic_access_page(vcpu); | |
2ce79189 AS |
6513 | if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) { |
6514 | vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT; | |
6515 | vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH; | |
6516 | r = 0; | |
6517 | goto out; | |
6518 | } | |
e516cebb AS |
6519 | if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) { |
6520 | vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT; | |
6521 | vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET; | |
6522 | r = 0; | |
6523 | goto out; | |
6524 | } | |
db397571 AS |
6525 | if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) { |
6526 | vcpu->run->exit_reason = KVM_EXIT_HYPERV; | |
6527 | vcpu->run->hyperv = vcpu->arch.hyperv.exit; | |
6528 | r = 0; | |
6529 | goto out; | |
6530 | } | |
f3b138c5 AS |
6531 | |
6532 | /* | |
6533 | * KVM_REQ_HV_STIMER has to be processed after | |
6534 | * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers | |
6535 | * depend on the guest clock being up-to-date | |
6536 | */ | |
1f4b34f8 AS |
6537 | if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu)) |
6538 | kvm_hv_process_stimers(vcpu); | |
2f52d58c | 6539 | } |
b93463aa | 6540 | |
bf9f6ac8 FW |
6541 | /* |
6542 | * KVM_REQ_EVENT is not set when posted interrupts are set by | |
6543 | * VT-d hardware, so we have to update RVI unconditionally. | |
6544 | */ | |
6545 | if (kvm_lapic_enabled(vcpu)) { | |
6546 | /* | |
6547 | * Update architecture specific hints for APIC | |
6548 | * virtual interrupt delivery. | |
6549 | */ | |
d62caabb | 6550 | if (vcpu->arch.apicv_active) |
bf9f6ac8 FW |
6551 | kvm_x86_ops->hwapic_irr_update(vcpu, |
6552 | kvm_lapic_find_highest_irr(vcpu)); | |
2f52d58c | 6553 | } |
b93463aa | 6554 | |
b463a6f7 | 6555 | if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) { |
66450a21 JK |
6556 | kvm_apic_accept_events(vcpu); |
6557 | if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) { | |
6558 | r = 1; | |
6559 | goto out; | |
6560 | } | |
6561 | ||
b6b8a145 JK |
6562 | if (inject_pending_event(vcpu, req_int_win) != 0) |
6563 | req_immediate_exit = true; | |
b463a6f7 | 6564 | /* enable NMI/IRQ window open exits if needed */ |
b6b8a145 | 6565 | else if (vcpu->arch.nmi_pending) |
c9a7953f | 6566 | kvm_x86_ops->enable_nmi_window(vcpu); |
c7c9c56c | 6567 | else if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win) |
c9a7953f | 6568 | kvm_x86_ops->enable_irq_window(vcpu); |
b463a6f7 AK |
6569 | |
6570 | if (kvm_lapic_enabled(vcpu)) { | |
6571 | update_cr8_intercept(vcpu); | |
6572 | kvm_lapic_sync_to_vapic(vcpu); | |
6573 | } | |
6574 | } | |
6575 | ||
d8368af8 AK |
6576 | r = kvm_mmu_reload(vcpu); |
6577 | if (unlikely(r)) { | |
d905c069 | 6578 | goto cancel_injection; |
d8368af8 AK |
6579 | } |
6580 | ||
b6c7a5dc HB |
6581 | preempt_disable(); |
6582 | ||
6583 | kvm_x86_ops->prepare_guest_switch(vcpu); | |
2608d7a1 AK |
6584 | if (vcpu->fpu_active) |
6585 | kvm_load_guest_fpu(vcpu); | |
2acf923e | 6586 | kvm_load_guest_xcr0(vcpu); |
b6c7a5dc | 6587 | |
6b7e2d09 XG |
6588 | vcpu->mode = IN_GUEST_MODE; |
6589 | ||
01b71917 MT |
6590 | srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx); |
6591 | ||
6b7e2d09 XG |
6592 | /* We should set ->mode before check ->requests, |
6593 | * see the comment in make_all_cpus_request. | |
6594 | */ | |
01b71917 | 6595 | smp_mb__after_srcu_read_unlock(); |
b6c7a5dc | 6596 | |
d94e1dc9 | 6597 | local_irq_disable(); |
32f88400 | 6598 | |
6b7e2d09 | 6599 | if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests |
d94e1dc9 | 6600 | || need_resched() || signal_pending(current)) { |
6b7e2d09 | 6601 | vcpu->mode = OUTSIDE_GUEST_MODE; |
d94e1dc9 | 6602 | smp_wmb(); |
6c142801 AK |
6603 | local_irq_enable(); |
6604 | preempt_enable(); | |
01b71917 | 6605 | vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu); |
6c142801 | 6606 | r = 1; |
d905c069 | 6607 | goto cancel_injection; |
6c142801 AK |
6608 | } |
6609 | ||
d6185f20 NHE |
6610 | if (req_immediate_exit) |
6611 | smp_send_reschedule(vcpu->cpu); | |
6612 | ||
8b89fe1f PB |
6613 | trace_kvm_entry(vcpu->vcpu_id); |
6614 | wait_lapic_expire(vcpu); | |
ccf73aaf | 6615 | __kvm_guest_enter(); |
b6c7a5dc | 6616 | |
42dbaa5a | 6617 | if (unlikely(vcpu->arch.switch_db_regs)) { |
42dbaa5a JK |
6618 | set_debugreg(0, 7); |
6619 | set_debugreg(vcpu->arch.eff_db[0], 0); | |
6620 | set_debugreg(vcpu->arch.eff_db[1], 1); | |
6621 | set_debugreg(vcpu->arch.eff_db[2], 2); | |
6622 | set_debugreg(vcpu->arch.eff_db[3], 3); | |
c77fb5fe | 6623 | set_debugreg(vcpu->arch.dr6, 6); |
ae561ede | 6624 | vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD; |
42dbaa5a | 6625 | } |
b6c7a5dc | 6626 | |
851ba692 | 6627 | kvm_x86_ops->run(vcpu); |
b6c7a5dc | 6628 | |
c77fb5fe PB |
6629 | /* |
6630 | * Do this here before restoring debug registers on the host. And | |
6631 | * since we do this before handling the vmexit, a DR access vmexit | |
6632 | * can (a) read the correct value of the debug registers, (b) set | |
6633 | * KVM_DEBUGREG_WONT_EXIT again. | |
6634 | */ | |
6635 | if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) { | |
6636 | int i; | |
6637 | ||
6638 | WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP); | |
6639 | kvm_x86_ops->sync_dirty_debug_regs(vcpu); | |
6640 | for (i = 0; i < KVM_NR_DB_REGS; i++) | |
6641 | vcpu->arch.eff_db[i] = vcpu->arch.db[i]; | |
6642 | } | |
6643 | ||
24f1e32c FW |
6644 | /* |
6645 | * If the guest has used debug registers, at least dr7 | |
6646 | * will be disabled while returning to the host. | |
6647 | * If we don't have active breakpoints in the host, we don't | |
6648 | * care about the messed up debug address registers. But if | |
6649 | * we have some of them active, restore the old state. | |
6650 | */ | |
59d8eb53 | 6651 | if (hw_breakpoint_active()) |
24f1e32c | 6652 | hw_breakpoint_restore(); |
42dbaa5a | 6653 | |
4ba76538 | 6654 | vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc()); |
1d5f066e | 6655 | |
6b7e2d09 | 6656 | vcpu->mode = OUTSIDE_GUEST_MODE; |
d94e1dc9 | 6657 | smp_wmb(); |
a547c6db YZ |
6658 | |
6659 | /* Interrupt is enabled by handle_external_intr() */ | |
6660 | kvm_x86_ops->handle_external_intr(vcpu); | |
b6c7a5dc HB |
6661 | |
6662 | ++vcpu->stat.exits; | |
6663 | ||
6664 | /* | |
6665 | * We must have an instruction between local_irq_enable() and | |
6666 | * kvm_guest_exit(), so the timer interrupt isn't delayed by | |
6667 | * the interrupt shadow. The stat.exits increment will do nicely. | |
6668 | * But we need to prevent reordering, hence this barrier(): | |
6669 | */ | |
6670 | barrier(); | |
6671 | ||
6672 | kvm_guest_exit(); | |
6673 | ||
6674 | preempt_enable(); | |
6675 | ||
f656ce01 | 6676 | vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu); |
3200f405 | 6677 | |
b6c7a5dc HB |
6678 | /* |
6679 | * Profile KVM exit RIPs: | |
6680 | */ | |
6681 | if (unlikely(prof_on == KVM_PROFILING)) { | |
5fdbf976 MT |
6682 | unsigned long rip = kvm_rip_read(vcpu); |
6683 | profile_hit(KVM_PROFILING, (void *)rip); | |
b6c7a5dc HB |
6684 | } |
6685 | ||
cc578287 ZA |
6686 | if (unlikely(vcpu->arch.tsc_always_catchup)) |
6687 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu); | |
298101da | 6688 | |
5cfb1d5a MT |
6689 | if (vcpu->arch.apic_attention) |
6690 | kvm_lapic_sync_from_vapic(vcpu); | |
b93463aa | 6691 | |
851ba692 | 6692 | r = kvm_x86_ops->handle_exit(vcpu); |
d905c069 MT |
6693 | return r; |
6694 | ||
6695 | cancel_injection: | |
6696 | kvm_x86_ops->cancel_injection(vcpu); | |
ae7a2a3f MT |
6697 | if (unlikely(vcpu->arch.apic_attention)) |
6698 | kvm_lapic_sync_from_vapic(vcpu); | |
d7690175 MT |
6699 | out: |
6700 | return r; | |
6701 | } | |
b6c7a5dc | 6702 | |
362c698f PB |
6703 | static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu) |
6704 | { | |
bf9f6ac8 FW |
6705 | if (!kvm_arch_vcpu_runnable(vcpu) && |
6706 | (!kvm_x86_ops->pre_block || kvm_x86_ops->pre_block(vcpu) == 0)) { | |
9c8fd1ba PB |
6707 | srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx); |
6708 | kvm_vcpu_block(vcpu); | |
6709 | vcpu->srcu_idx = srcu_read_lock(&kvm->srcu); | |
bf9f6ac8 FW |
6710 | |
6711 | if (kvm_x86_ops->post_block) | |
6712 | kvm_x86_ops->post_block(vcpu); | |
6713 | ||
9c8fd1ba PB |
6714 | if (!kvm_check_request(KVM_REQ_UNHALT, vcpu)) |
6715 | return 1; | |
6716 | } | |
362c698f PB |
6717 | |
6718 | kvm_apic_accept_events(vcpu); | |
6719 | switch(vcpu->arch.mp_state) { | |
6720 | case KVM_MP_STATE_HALTED: | |
6721 | vcpu->arch.pv.pv_unhalted = false; | |
6722 | vcpu->arch.mp_state = | |
6723 | KVM_MP_STATE_RUNNABLE; | |
6724 | case KVM_MP_STATE_RUNNABLE: | |
6725 | vcpu->arch.apf.halted = false; | |
6726 | break; | |
6727 | case KVM_MP_STATE_INIT_RECEIVED: | |
6728 | break; | |
6729 | default: | |
6730 | return -EINTR; | |
6731 | break; | |
6732 | } | |
6733 | return 1; | |
6734 | } | |
09cec754 | 6735 | |
5d9bc648 PB |
6736 | static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu) |
6737 | { | |
6738 | return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE && | |
6739 | !vcpu->arch.apf.halted); | |
6740 | } | |
6741 | ||
362c698f | 6742 | static int vcpu_run(struct kvm_vcpu *vcpu) |
d7690175 MT |
6743 | { |
6744 | int r; | |
f656ce01 | 6745 | struct kvm *kvm = vcpu->kvm; |
d7690175 | 6746 | |
f656ce01 | 6747 | vcpu->srcu_idx = srcu_read_lock(&kvm->srcu); |
d7690175 | 6748 | |
362c698f | 6749 | for (;;) { |
58f800d5 | 6750 | if (kvm_vcpu_running(vcpu)) { |
851ba692 | 6751 | r = vcpu_enter_guest(vcpu); |
bf9f6ac8 | 6752 | } else { |
362c698f | 6753 | r = vcpu_block(kvm, vcpu); |
bf9f6ac8 FW |
6754 | } |
6755 | ||
09cec754 GN |
6756 | if (r <= 0) |
6757 | break; | |
6758 | ||
6759 | clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests); | |
6760 | if (kvm_cpu_has_pending_timer(vcpu)) | |
6761 | kvm_inject_pending_timer_irqs(vcpu); | |
6762 | ||
782d422b MG |
6763 | if (dm_request_for_irq_injection(vcpu) && |
6764 | kvm_vcpu_ready_for_interrupt_injection(vcpu)) { | |
4ca7dd8c PB |
6765 | r = 0; |
6766 | vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN; | |
09cec754 | 6767 | ++vcpu->stat.request_irq_exits; |
362c698f | 6768 | break; |
09cec754 | 6769 | } |
af585b92 GN |
6770 | |
6771 | kvm_check_async_pf_completion(vcpu); | |
6772 | ||
09cec754 GN |
6773 | if (signal_pending(current)) { |
6774 | r = -EINTR; | |
851ba692 | 6775 | vcpu->run->exit_reason = KVM_EXIT_INTR; |
09cec754 | 6776 | ++vcpu->stat.signal_exits; |
362c698f | 6777 | break; |
09cec754 GN |
6778 | } |
6779 | if (need_resched()) { | |
f656ce01 | 6780 | srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx); |
c08ac06a | 6781 | cond_resched(); |
f656ce01 | 6782 | vcpu->srcu_idx = srcu_read_lock(&kvm->srcu); |
d7690175 | 6783 | } |
b6c7a5dc HB |
6784 | } |
6785 | ||
f656ce01 | 6786 | srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx); |
b6c7a5dc HB |
6787 | |
6788 | return r; | |
6789 | } | |
6790 | ||
716d51ab GN |
6791 | static inline int complete_emulated_io(struct kvm_vcpu *vcpu) |
6792 | { | |
6793 | int r; | |
6794 | vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu); | |
6795 | r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE); | |
6796 | srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx); | |
6797 | if (r != EMULATE_DONE) | |
6798 | return 0; | |
6799 | return 1; | |
6800 | } | |
6801 | ||
6802 | static int complete_emulated_pio(struct kvm_vcpu *vcpu) | |
6803 | { | |
6804 | BUG_ON(!vcpu->arch.pio.count); | |
6805 | ||
6806 | return complete_emulated_io(vcpu); | |
6807 | } | |
6808 | ||
f78146b0 AK |
6809 | /* |
6810 | * Implements the following, as a state machine: | |
6811 | * | |
6812 | * read: | |
6813 | * for each fragment | |
87da7e66 XG |
6814 | * for each mmio piece in the fragment |
6815 | * write gpa, len | |
6816 | * exit | |
6817 | * copy data | |
f78146b0 AK |
6818 | * execute insn |
6819 | * | |
6820 | * write: | |
6821 | * for each fragment | |
87da7e66 XG |
6822 | * for each mmio piece in the fragment |
6823 | * write gpa, len | |
6824 | * copy data | |
6825 | * exit | |
f78146b0 | 6826 | */ |
716d51ab | 6827 | static int complete_emulated_mmio(struct kvm_vcpu *vcpu) |
5287f194 AK |
6828 | { |
6829 | struct kvm_run *run = vcpu->run; | |
f78146b0 | 6830 | struct kvm_mmio_fragment *frag; |
87da7e66 | 6831 | unsigned len; |
5287f194 | 6832 | |
716d51ab | 6833 | BUG_ON(!vcpu->mmio_needed); |
5287f194 | 6834 | |
716d51ab | 6835 | /* Complete previous fragment */ |
87da7e66 XG |
6836 | frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment]; |
6837 | len = min(8u, frag->len); | |
716d51ab | 6838 | if (!vcpu->mmio_is_write) |
87da7e66 XG |
6839 | memcpy(frag->data, run->mmio.data, len); |
6840 | ||
6841 | if (frag->len <= 8) { | |
6842 | /* Switch to the next fragment. */ | |
6843 | frag++; | |
6844 | vcpu->mmio_cur_fragment++; | |
6845 | } else { | |
6846 | /* Go forward to the next mmio piece. */ | |
6847 | frag->data += len; | |
6848 | frag->gpa += len; | |
6849 | frag->len -= len; | |
6850 | } | |
6851 | ||
a08d3b3b | 6852 | if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) { |
716d51ab | 6853 | vcpu->mmio_needed = 0; |
0912c977 PB |
6854 | |
6855 | /* FIXME: return into emulator if single-stepping. */ | |
cef4dea0 | 6856 | if (vcpu->mmio_is_write) |
716d51ab GN |
6857 | return 1; |
6858 | vcpu->mmio_read_completed = 1; | |
6859 | return complete_emulated_io(vcpu); | |
6860 | } | |
87da7e66 | 6861 | |
716d51ab GN |
6862 | run->exit_reason = KVM_EXIT_MMIO; |
6863 | run->mmio.phys_addr = frag->gpa; | |
6864 | if (vcpu->mmio_is_write) | |
87da7e66 XG |
6865 | memcpy(run->mmio.data, frag->data, min(8u, frag->len)); |
6866 | run->mmio.len = min(8u, frag->len); | |
716d51ab GN |
6867 | run->mmio.is_write = vcpu->mmio_is_write; |
6868 | vcpu->arch.complete_userspace_io = complete_emulated_mmio; | |
6869 | return 0; | |
5287f194 AK |
6870 | } |
6871 | ||
716d51ab | 6872 | |
b6c7a5dc HB |
6873 | int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run) |
6874 | { | |
c5bedc68 | 6875 | struct fpu *fpu = ¤t->thread.fpu; |
b6c7a5dc HB |
6876 | int r; |
6877 | sigset_t sigsaved; | |
6878 | ||
c4d72e2d | 6879 | fpu__activate_curr(fpu); |
e5c30142 | 6880 | |
ac9f6dc0 AK |
6881 | if (vcpu->sigset_active) |
6882 | sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved); | |
6883 | ||
a4535290 | 6884 | if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) { |
b6c7a5dc | 6885 | kvm_vcpu_block(vcpu); |
66450a21 | 6886 | kvm_apic_accept_events(vcpu); |
d7690175 | 6887 | clear_bit(KVM_REQ_UNHALT, &vcpu->requests); |
ac9f6dc0 AK |
6888 | r = -EAGAIN; |
6889 | goto out; | |
b6c7a5dc HB |
6890 | } |
6891 | ||
b6c7a5dc | 6892 | /* re-sync apic's tpr */ |
35754c98 | 6893 | if (!lapic_in_kernel(vcpu)) { |
eea1cff9 AP |
6894 | if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) { |
6895 | r = -EINVAL; | |
6896 | goto out; | |
6897 | } | |
6898 | } | |
b6c7a5dc | 6899 | |
716d51ab GN |
6900 | if (unlikely(vcpu->arch.complete_userspace_io)) { |
6901 | int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io; | |
6902 | vcpu->arch.complete_userspace_io = NULL; | |
6903 | r = cui(vcpu); | |
6904 | if (r <= 0) | |
6905 | goto out; | |
6906 | } else | |
6907 | WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed); | |
5287f194 | 6908 | |
362c698f | 6909 | r = vcpu_run(vcpu); |
b6c7a5dc HB |
6910 | |
6911 | out: | |
f1d86e46 | 6912 | post_kvm_run_save(vcpu); |
b6c7a5dc HB |
6913 | if (vcpu->sigset_active) |
6914 | sigprocmask(SIG_SETMASK, &sigsaved, NULL); | |
6915 | ||
b6c7a5dc HB |
6916 | return r; |
6917 | } | |
6918 | ||
6919 | int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs) | |
6920 | { | |
7ae441ea GN |
6921 | if (vcpu->arch.emulate_regs_need_sync_to_vcpu) { |
6922 | /* | |
6923 | * We are here if userspace calls get_regs() in the middle of | |
6924 | * instruction emulation. Registers state needs to be copied | |
4a969980 | 6925 | * back from emulation context to vcpu. Userspace shouldn't do |
7ae441ea GN |
6926 | * that usually, but some bad designed PV devices (vmware |
6927 | * backdoor interface) need this to work | |
6928 | */ | |
dd856efa | 6929 | emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt); |
7ae441ea GN |
6930 | vcpu->arch.emulate_regs_need_sync_to_vcpu = false; |
6931 | } | |
5fdbf976 MT |
6932 | regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX); |
6933 | regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX); | |
6934 | regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX); | |
6935 | regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX); | |
6936 | regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI); | |
6937 | regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI); | |
6938 | regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP); | |
6939 | regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP); | |
b6c7a5dc | 6940 | #ifdef CONFIG_X86_64 |
5fdbf976 MT |
6941 | regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8); |
6942 | regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9); | |
6943 | regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10); | |
6944 | regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11); | |
6945 | regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12); | |
6946 | regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13); | |
6947 | regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14); | |
6948 | regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15); | |
b6c7a5dc HB |
6949 | #endif |
6950 | ||
5fdbf976 | 6951 | regs->rip = kvm_rip_read(vcpu); |
91586a3b | 6952 | regs->rflags = kvm_get_rflags(vcpu); |
b6c7a5dc | 6953 | |
b6c7a5dc HB |
6954 | return 0; |
6955 | } | |
6956 | ||
6957 | int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs) | |
6958 | { | |
7ae441ea GN |
6959 | vcpu->arch.emulate_regs_need_sync_from_vcpu = true; |
6960 | vcpu->arch.emulate_regs_need_sync_to_vcpu = false; | |
6961 | ||
5fdbf976 MT |
6962 | kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax); |
6963 | kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx); | |
6964 | kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx); | |
6965 | kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx); | |
6966 | kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi); | |
6967 | kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi); | |
6968 | kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp); | |
6969 | kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp); | |
b6c7a5dc | 6970 | #ifdef CONFIG_X86_64 |
5fdbf976 MT |
6971 | kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8); |
6972 | kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9); | |
6973 | kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10); | |
6974 | kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11); | |
6975 | kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12); | |
6976 | kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13); | |
6977 | kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14); | |
6978 | kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15); | |
b6c7a5dc HB |
6979 | #endif |
6980 | ||
5fdbf976 | 6981 | kvm_rip_write(vcpu, regs->rip); |
91586a3b | 6982 | kvm_set_rflags(vcpu, regs->rflags); |
b6c7a5dc | 6983 | |
b4f14abd JK |
6984 | vcpu->arch.exception.pending = false; |
6985 | ||
3842d135 AK |
6986 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
6987 | ||
b6c7a5dc HB |
6988 | return 0; |
6989 | } | |
6990 | ||
b6c7a5dc HB |
6991 | void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l) |
6992 | { | |
6993 | struct kvm_segment cs; | |
6994 | ||
3e6e0aab | 6995 | kvm_get_segment(vcpu, &cs, VCPU_SREG_CS); |
b6c7a5dc HB |
6996 | *db = cs.db; |
6997 | *l = cs.l; | |
6998 | } | |
6999 | EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits); | |
7000 | ||
7001 | int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu, | |
7002 | struct kvm_sregs *sregs) | |
7003 | { | |
89a27f4d | 7004 | struct desc_ptr dt; |
b6c7a5dc | 7005 | |
3e6e0aab GT |
7006 | kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS); |
7007 | kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS); | |
7008 | kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES); | |
7009 | kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS); | |
7010 | kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS); | |
7011 | kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS); | |
b6c7a5dc | 7012 | |
3e6e0aab GT |
7013 | kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR); |
7014 | kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR); | |
b6c7a5dc HB |
7015 | |
7016 | kvm_x86_ops->get_idt(vcpu, &dt); | |
89a27f4d GN |
7017 | sregs->idt.limit = dt.size; |
7018 | sregs->idt.base = dt.address; | |
b6c7a5dc | 7019 | kvm_x86_ops->get_gdt(vcpu, &dt); |
89a27f4d GN |
7020 | sregs->gdt.limit = dt.size; |
7021 | sregs->gdt.base = dt.address; | |
b6c7a5dc | 7022 | |
4d4ec087 | 7023 | sregs->cr0 = kvm_read_cr0(vcpu); |
ad312c7c | 7024 | sregs->cr2 = vcpu->arch.cr2; |
9f8fe504 | 7025 | sregs->cr3 = kvm_read_cr3(vcpu); |
fc78f519 | 7026 | sregs->cr4 = kvm_read_cr4(vcpu); |
2d3ad1f4 | 7027 | sregs->cr8 = kvm_get_cr8(vcpu); |
f6801dff | 7028 | sregs->efer = vcpu->arch.efer; |
b6c7a5dc HB |
7029 | sregs->apic_base = kvm_get_apic_base(vcpu); |
7030 | ||
923c61bb | 7031 | memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap); |
b6c7a5dc | 7032 | |
36752c9b | 7033 | if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft) |
14d0bc1f GN |
7034 | set_bit(vcpu->arch.interrupt.nr, |
7035 | (unsigned long *)sregs->interrupt_bitmap); | |
16d7a191 | 7036 | |
b6c7a5dc HB |
7037 | return 0; |
7038 | } | |
7039 | ||
62d9f0db MT |
7040 | int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu, |
7041 | struct kvm_mp_state *mp_state) | |
7042 | { | |
66450a21 | 7043 | kvm_apic_accept_events(vcpu); |
6aef266c SV |
7044 | if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED && |
7045 | vcpu->arch.pv.pv_unhalted) | |
7046 | mp_state->mp_state = KVM_MP_STATE_RUNNABLE; | |
7047 | else | |
7048 | mp_state->mp_state = vcpu->arch.mp_state; | |
7049 | ||
62d9f0db MT |
7050 | return 0; |
7051 | } | |
7052 | ||
7053 | int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu, | |
7054 | struct kvm_mp_state *mp_state) | |
7055 | { | |
bce87cce | 7056 | if (!lapic_in_kernel(vcpu) && |
66450a21 JK |
7057 | mp_state->mp_state != KVM_MP_STATE_RUNNABLE) |
7058 | return -EINVAL; | |
7059 | ||
7060 | if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) { | |
7061 | vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED; | |
7062 | set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events); | |
7063 | } else | |
7064 | vcpu->arch.mp_state = mp_state->mp_state; | |
3842d135 | 7065 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
62d9f0db MT |
7066 | return 0; |
7067 | } | |
7068 | ||
7f3d35fd KW |
7069 | int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index, |
7070 | int reason, bool has_error_code, u32 error_code) | |
b6c7a5dc | 7071 | { |
9d74191a | 7072 | struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt; |
8ec4722d | 7073 | int ret; |
e01c2426 | 7074 | |
8ec4722d | 7075 | init_emulate_ctxt(vcpu); |
c697518a | 7076 | |
7f3d35fd | 7077 | ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason, |
9d74191a | 7078 | has_error_code, error_code); |
c697518a | 7079 | |
c697518a | 7080 | if (ret) |
19d04437 | 7081 | return EMULATE_FAIL; |
37817f29 | 7082 | |
9d74191a TY |
7083 | kvm_rip_write(vcpu, ctxt->eip); |
7084 | kvm_set_rflags(vcpu, ctxt->eflags); | |
3842d135 | 7085 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
19d04437 | 7086 | return EMULATE_DONE; |
37817f29 IE |
7087 | } |
7088 | EXPORT_SYMBOL_GPL(kvm_task_switch); | |
7089 | ||
b6c7a5dc HB |
7090 | int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu, |
7091 | struct kvm_sregs *sregs) | |
7092 | { | |
58cb628d | 7093 | struct msr_data apic_base_msr; |
b6c7a5dc | 7094 | int mmu_reset_needed = 0; |
63f42e02 | 7095 | int pending_vec, max_bits, idx; |
89a27f4d | 7096 | struct desc_ptr dt; |
b6c7a5dc | 7097 | |
6d1068b3 PM |
7098 | if (!guest_cpuid_has_xsave(vcpu) && (sregs->cr4 & X86_CR4_OSXSAVE)) |
7099 | return -EINVAL; | |
7100 | ||
89a27f4d GN |
7101 | dt.size = sregs->idt.limit; |
7102 | dt.address = sregs->idt.base; | |
b6c7a5dc | 7103 | kvm_x86_ops->set_idt(vcpu, &dt); |
89a27f4d GN |
7104 | dt.size = sregs->gdt.limit; |
7105 | dt.address = sregs->gdt.base; | |
b6c7a5dc HB |
7106 | kvm_x86_ops->set_gdt(vcpu, &dt); |
7107 | ||
ad312c7c | 7108 | vcpu->arch.cr2 = sregs->cr2; |
9f8fe504 | 7109 | mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3; |
dc7e795e | 7110 | vcpu->arch.cr3 = sregs->cr3; |
aff48baa | 7111 | __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail); |
b6c7a5dc | 7112 | |
2d3ad1f4 | 7113 | kvm_set_cr8(vcpu, sregs->cr8); |
b6c7a5dc | 7114 | |
f6801dff | 7115 | mmu_reset_needed |= vcpu->arch.efer != sregs->efer; |
b6c7a5dc | 7116 | kvm_x86_ops->set_efer(vcpu, sregs->efer); |
58cb628d JK |
7117 | apic_base_msr.data = sregs->apic_base; |
7118 | apic_base_msr.host_initiated = true; | |
7119 | kvm_set_apic_base(vcpu, &apic_base_msr); | |
b6c7a5dc | 7120 | |
4d4ec087 | 7121 | mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0; |
b6c7a5dc | 7122 | kvm_x86_ops->set_cr0(vcpu, sregs->cr0); |
d7306163 | 7123 | vcpu->arch.cr0 = sregs->cr0; |
b6c7a5dc | 7124 | |
fc78f519 | 7125 | mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4; |
b6c7a5dc | 7126 | kvm_x86_ops->set_cr4(vcpu, sregs->cr4); |
3ea3aa8c | 7127 | if (sregs->cr4 & X86_CR4_OSXSAVE) |
00b27a3e | 7128 | kvm_update_cpuid(vcpu); |
63f42e02 XG |
7129 | |
7130 | idx = srcu_read_lock(&vcpu->kvm->srcu); | |
7c93be44 | 7131 | if (!is_long_mode(vcpu) && is_pae(vcpu)) { |
9f8fe504 | 7132 | load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu)); |
7c93be44 MT |
7133 | mmu_reset_needed = 1; |
7134 | } | |
63f42e02 | 7135 | srcu_read_unlock(&vcpu->kvm->srcu, idx); |
b6c7a5dc HB |
7136 | |
7137 | if (mmu_reset_needed) | |
7138 | kvm_mmu_reset_context(vcpu); | |
7139 | ||
a50abc3b | 7140 | max_bits = KVM_NR_INTERRUPTS; |
923c61bb GN |
7141 | pending_vec = find_first_bit( |
7142 | (const unsigned long *)sregs->interrupt_bitmap, max_bits); | |
7143 | if (pending_vec < max_bits) { | |
66fd3f7f | 7144 | kvm_queue_interrupt(vcpu, pending_vec, false); |
923c61bb | 7145 | pr_debug("Set back pending irq %d\n", pending_vec); |
b6c7a5dc HB |
7146 | } |
7147 | ||
3e6e0aab GT |
7148 | kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS); |
7149 | kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS); | |
7150 | kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES); | |
7151 | kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS); | |
7152 | kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS); | |
7153 | kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS); | |
b6c7a5dc | 7154 | |
3e6e0aab GT |
7155 | kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR); |
7156 | kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR); | |
b6c7a5dc | 7157 | |
5f0269f5 ME |
7158 | update_cr8_intercept(vcpu); |
7159 | ||
9c3e4aab | 7160 | /* Older userspace won't unhalt the vcpu on reset. */ |
c5af89b6 | 7161 | if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 && |
9c3e4aab | 7162 | sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 && |
3eeb3288 | 7163 | !is_protmode(vcpu)) |
9c3e4aab MT |
7164 | vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE; |
7165 | ||
3842d135 AK |
7166 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
7167 | ||
b6c7a5dc HB |
7168 | return 0; |
7169 | } | |
7170 | ||
d0bfb940 JK |
7171 | int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu, |
7172 | struct kvm_guest_debug *dbg) | |
b6c7a5dc | 7173 | { |
355be0b9 | 7174 | unsigned long rflags; |
ae675ef0 | 7175 | int i, r; |
b6c7a5dc | 7176 | |
4f926bf2 JK |
7177 | if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) { |
7178 | r = -EBUSY; | |
7179 | if (vcpu->arch.exception.pending) | |
2122ff5e | 7180 | goto out; |
4f926bf2 JK |
7181 | if (dbg->control & KVM_GUESTDBG_INJECT_DB) |
7182 | kvm_queue_exception(vcpu, DB_VECTOR); | |
7183 | else | |
7184 | kvm_queue_exception(vcpu, BP_VECTOR); | |
7185 | } | |
7186 | ||
91586a3b JK |
7187 | /* |
7188 | * Read rflags as long as potentially injected trace flags are still | |
7189 | * filtered out. | |
7190 | */ | |
7191 | rflags = kvm_get_rflags(vcpu); | |
355be0b9 JK |
7192 | |
7193 | vcpu->guest_debug = dbg->control; | |
7194 | if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE)) | |
7195 | vcpu->guest_debug = 0; | |
7196 | ||
7197 | if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) { | |
ae675ef0 JK |
7198 | for (i = 0; i < KVM_NR_DB_REGS; ++i) |
7199 | vcpu->arch.eff_db[i] = dbg->arch.debugreg[i]; | |
c8639010 | 7200 | vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7]; |
ae675ef0 JK |
7201 | } else { |
7202 | for (i = 0; i < KVM_NR_DB_REGS; i++) | |
7203 | vcpu->arch.eff_db[i] = vcpu->arch.db[i]; | |
ae675ef0 | 7204 | } |
c8639010 | 7205 | kvm_update_dr7(vcpu); |
ae675ef0 | 7206 | |
f92653ee JK |
7207 | if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) |
7208 | vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) + | |
7209 | get_segment_base(vcpu, VCPU_SREG_CS); | |
94fe45da | 7210 | |
91586a3b JK |
7211 | /* |
7212 | * Trigger an rflags update that will inject or remove the trace | |
7213 | * flags. | |
7214 | */ | |
7215 | kvm_set_rflags(vcpu, rflags); | |
b6c7a5dc | 7216 | |
a96036b8 | 7217 | kvm_x86_ops->update_bp_intercept(vcpu); |
b6c7a5dc | 7218 | |
4f926bf2 | 7219 | r = 0; |
d0bfb940 | 7220 | |
2122ff5e | 7221 | out: |
b6c7a5dc HB |
7222 | |
7223 | return r; | |
7224 | } | |
7225 | ||
8b006791 ZX |
7226 | /* |
7227 | * Translate a guest virtual address to a guest physical address. | |
7228 | */ | |
7229 | int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu, | |
7230 | struct kvm_translation *tr) | |
7231 | { | |
7232 | unsigned long vaddr = tr->linear_address; | |
7233 | gpa_t gpa; | |
f656ce01 | 7234 | int idx; |
8b006791 | 7235 | |
f656ce01 | 7236 | idx = srcu_read_lock(&vcpu->kvm->srcu); |
1871c602 | 7237 | gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL); |
f656ce01 | 7238 | srcu_read_unlock(&vcpu->kvm->srcu, idx); |
8b006791 ZX |
7239 | tr->physical_address = gpa; |
7240 | tr->valid = gpa != UNMAPPED_GVA; | |
7241 | tr->writeable = 1; | |
7242 | tr->usermode = 0; | |
8b006791 ZX |
7243 | |
7244 | return 0; | |
7245 | } | |
7246 | ||
d0752060 HB |
7247 | int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu) |
7248 | { | |
c47ada30 | 7249 | struct fxregs_state *fxsave = |
7366ed77 | 7250 | &vcpu->arch.guest_fpu.state.fxsave; |
d0752060 | 7251 | |
d0752060 HB |
7252 | memcpy(fpu->fpr, fxsave->st_space, 128); |
7253 | fpu->fcw = fxsave->cwd; | |
7254 | fpu->fsw = fxsave->swd; | |
7255 | fpu->ftwx = fxsave->twd; | |
7256 | fpu->last_opcode = fxsave->fop; | |
7257 | fpu->last_ip = fxsave->rip; | |
7258 | fpu->last_dp = fxsave->rdp; | |
7259 | memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space); | |
7260 | ||
d0752060 HB |
7261 | return 0; |
7262 | } | |
7263 | ||
7264 | int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu) | |
7265 | { | |
c47ada30 | 7266 | struct fxregs_state *fxsave = |
7366ed77 | 7267 | &vcpu->arch.guest_fpu.state.fxsave; |
d0752060 | 7268 | |
d0752060 HB |
7269 | memcpy(fxsave->st_space, fpu->fpr, 128); |
7270 | fxsave->cwd = fpu->fcw; | |
7271 | fxsave->swd = fpu->fsw; | |
7272 | fxsave->twd = fpu->ftwx; | |
7273 | fxsave->fop = fpu->last_opcode; | |
7274 | fxsave->rip = fpu->last_ip; | |
7275 | fxsave->rdp = fpu->last_dp; | |
7276 | memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space); | |
7277 | ||
d0752060 HB |
7278 | return 0; |
7279 | } | |
7280 | ||
0ee6a517 | 7281 | static void fx_init(struct kvm_vcpu *vcpu) |
d0752060 | 7282 | { |
bf935b0b | 7283 | fpstate_init(&vcpu->arch.guest_fpu.state); |
df1daba7 | 7284 | if (cpu_has_xsaves) |
7366ed77 | 7285 | vcpu->arch.guest_fpu.state.xsave.header.xcomp_bv = |
df1daba7 | 7286 | host_xcr0 | XSTATE_COMPACTION_ENABLED; |
d0752060 | 7287 | |
2acf923e DC |
7288 | /* |
7289 | * Ensure guest xcr0 is valid for loading | |
7290 | */ | |
d91cab78 | 7291 | vcpu->arch.xcr0 = XFEATURE_MASK_FP; |
2acf923e | 7292 | |
ad312c7c | 7293 | vcpu->arch.cr0 |= X86_CR0_ET; |
d0752060 | 7294 | } |
d0752060 HB |
7295 | |
7296 | void kvm_load_guest_fpu(struct kvm_vcpu *vcpu) | |
7297 | { | |
2608d7a1 | 7298 | if (vcpu->guest_fpu_loaded) |
d0752060 HB |
7299 | return; |
7300 | ||
2acf923e DC |
7301 | /* |
7302 | * Restore all possible states in the guest, | |
7303 | * and assume host would use all available bits. | |
7304 | * Guest xcr0 would be loaded later. | |
7305 | */ | |
7306 | kvm_put_guest_xcr0(vcpu); | |
d0752060 | 7307 | vcpu->guest_fpu_loaded = 1; |
b1a74bf8 | 7308 | __kernel_fpu_begin(); |
003e2e8b | 7309 | __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu.state); |
0c04851c | 7310 | trace_kvm_fpu(1); |
d0752060 | 7311 | } |
d0752060 HB |
7312 | |
7313 | void kvm_put_guest_fpu(struct kvm_vcpu *vcpu) | |
7314 | { | |
2acf923e DC |
7315 | kvm_put_guest_xcr0(vcpu); |
7316 | ||
653f52c3 RR |
7317 | if (!vcpu->guest_fpu_loaded) { |
7318 | vcpu->fpu_counter = 0; | |
d0752060 | 7319 | return; |
653f52c3 | 7320 | } |
d0752060 HB |
7321 | |
7322 | vcpu->guest_fpu_loaded = 0; | |
4f836347 | 7323 | copy_fpregs_to_fpstate(&vcpu->arch.guest_fpu); |
b1a74bf8 | 7324 | __kernel_fpu_end(); |
f096ed85 | 7325 | ++vcpu->stat.fpu_reload; |
653f52c3 RR |
7326 | /* |
7327 | * If using eager FPU mode, or if the guest is a frequent user | |
7328 | * of the FPU, just leave the FPU active for next time. | |
7329 | * Every 255 times fpu_counter rolls over to 0; a guest that uses | |
7330 | * the FPU in bursts will revert to loading it on demand. | |
7331 | */ | |
a9b4fb7e | 7332 | if (!vcpu->arch.eager_fpu) { |
653f52c3 RR |
7333 | if (++vcpu->fpu_counter < 5) |
7334 | kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu); | |
7335 | } | |
0c04851c | 7336 | trace_kvm_fpu(0); |
d0752060 | 7337 | } |
e9b11c17 ZX |
7338 | |
7339 | void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu) | |
7340 | { | |
12f9a48f | 7341 | kvmclock_reset(vcpu); |
7f1ea208 | 7342 | |
f5f48ee1 | 7343 | free_cpumask_var(vcpu->arch.wbinvd_dirty_mask); |
e9b11c17 ZX |
7344 | kvm_x86_ops->vcpu_free(vcpu); |
7345 | } | |
7346 | ||
7347 | struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, | |
7348 | unsigned int id) | |
7349 | { | |
c447e76b LL |
7350 | struct kvm_vcpu *vcpu; |
7351 | ||
6755bae8 ZA |
7352 | if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0) |
7353 | printk_once(KERN_WARNING | |
7354 | "kvm: SMP vm created on host with unstable TSC; " | |
7355 | "guest TSC will not be reliable\n"); | |
c447e76b LL |
7356 | |
7357 | vcpu = kvm_x86_ops->vcpu_create(kvm, id); | |
7358 | ||
c447e76b | 7359 | return vcpu; |
26e5215f | 7360 | } |
e9b11c17 | 7361 | |
26e5215f AK |
7362 | int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu) |
7363 | { | |
7364 | int r; | |
e9b11c17 | 7365 | |
19efffa2 | 7366 | kvm_vcpu_mtrr_init(vcpu); |
9fc77441 MT |
7367 | r = vcpu_load(vcpu); |
7368 | if (r) | |
7369 | return r; | |
d28bc9dd | 7370 | kvm_vcpu_reset(vcpu, false); |
8a3c1a33 | 7371 | kvm_mmu_setup(vcpu); |
e9b11c17 | 7372 | vcpu_put(vcpu); |
26e5215f | 7373 | return r; |
e9b11c17 ZX |
7374 | } |
7375 | ||
31928aa5 | 7376 | void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu) |
42897d86 | 7377 | { |
8fe8ab46 | 7378 | struct msr_data msr; |
332967a3 | 7379 | struct kvm *kvm = vcpu->kvm; |
42897d86 | 7380 | |
31928aa5 DD |
7381 | if (vcpu_load(vcpu)) |
7382 | return; | |
8fe8ab46 WA |
7383 | msr.data = 0x0; |
7384 | msr.index = MSR_IA32_TSC; | |
7385 | msr.host_initiated = true; | |
7386 | kvm_write_tsc(vcpu, &msr); | |
42897d86 MT |
7387 | vcpu_put(vcpu); |
7388 | ||
630994b3 MT |
7389 | if (!kvmclock_periodic_sync) |
7390 | return; | |
7391 | ||
332967a3 AJ |
7392 | schedule_delayed_work(&kvm->arch.kvmclock_sync_work, |
7393 | KVMCLOCK_SYNC_PERIOD); | |
42897d86 MT |
7394 | } |
7395 | ||
d40ccc62 | 7396 | void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu) |
e9b11c17 | 7397 | { |
9fc77441 | 7398 | int r; |
344d9588 GN |
7399 | vcpu->arch.apf.msr_val = 0; |
7400 | ||
9fc77441 MT |
7401 | r = vcpu_load(vcpu); |
7402 | BUG_ON(r); | |
e9b11c17 ZX |
7403 | kvm_mmu_unload(vcpu); |
7404 | vcpu_put(vcpu); | |
7405 | ||
7406 | kvm_x86_ops->vcpu_free(vcpu); | |
7407 | } | |
7408 | ||
d28bc9dd | 7409 | void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event) |
e9b11c17 | 7410 | { |
e69fab5d PB |
7411 | vcpu->arch.hflags = 0; |
7412 | ||
7460fb4a AK |
7413 | atomic_set(&vcpu->arch.nmi_queued, 0); |
7414 | vcpu->arch.nmi_pending = 0; | |
448fa4a9 | 7415 | vcpu->arch.nmi_injected = false; |
5f7552d4 NA |
7416 | kvm_clear_interrupt_queue(vcpu); |
7417 | kvm_clear_exception_queue(vcpu); | |
448fa4a9 | 7418 | |
42dbaa5a | 7419 | memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db)); |
ae561ede | 7420 | kvm_update_dr0123(vcpu); |
6f43ed01 | 7421 | vcpu->arch.dr6 = DR6_INIT; |
73aaf249 | 7422 | kvm_update_dr6(vcpu); |
42dbaa5a | 7423 | vcpu->arch.dr7 = DR7_FIXED_1; |
c8639010 | 7424 | kvm_update_dr7(vcpu); |
42dbaa5a | 7425 | |
1119022c NA |
7426 | vcpu->arch.cr2 = 0; |
7427 | ||
3842d135 | 7428 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
344d9588 | 7429 | vcpu->arch.apf.msr_val = 0; |
c9aaa895 | 7430 | vcpu->arch.st.msr_val = 0; |
3842d135 | 7431 | |
12f9a48f GC |
7432 | kvmclock_reset(vcpu); |
7433 | ||
af585b92 GN |
7434 | kvm_clear_async_pf_completion_queue(vcpu); |
7435 | kvm_async_pf_hash_reset(vcpu); | |
7436 | vcpu->arch.apf.halted = false; | |
3842d135 | 7437 | |
64d60670 | 7438 | if (!init_event) { |
d28bc9dd | 7439 | kvm_pmu_reset(vcpu); |
64d60670 PB |
7440 | vcpu->arch.smbase = 0x30000; |
7441 | } | |
f5132b01 | 7442 | |
66f7b72e JS |
7443 | memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs)); |
7444 | vcpu->arch.regs_avail = ~0; | |
7445 | vcpu->arch.regs_dirty = ~0; | |
7446 | ||
d28bc9dd | 7447 | kvm_x86_ops->vcpu_reset(vcpu, init_event); |
e9b11c17 ZX |
7448 | } |
7449 | ||
2b4a273b | 7450 | void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector) |
66450a21 JK |
7451 | { |
7452 | struct kvm_segment cs; | |
7453 | ||
7454 | kvm_get_segment(vcpu, &cs, VCPU_SREG_CS); | |
7455 | cs.selector = vector << 8; | |
7456 | cs.base = vector << 12; | |
7457 | kvm_set_segment(vcpu, &cs, VCPU_SREG_CS); | |
7458 | kvm_rip_write(vcpu, 0); | |
e9b11c17 ZX |
7459 | } |
7460 | ||
13a34e06 | 7461 | int kvm_arch_hardware_enable(void) |
e9b11c17 | 7462 | { |
ca84d1a2 ZA |
7463 | struct kvm *kvm; |
7464 | struct kvm_vcpu *vcpu; | |
7465 | int i; | |
0dd6a6ed ZA |
7466 | int ret; |
7467 | u64 local_tsc; | |
7468 | u64 max_tsc = 0; | |
7469 | bool stable, backwards_tsc = false; | |
18863bdd AK |
7470 | |
7471 | kvm_shared_msr_cpu_online(); | |
13a34e06 | 7472 | ret = kvm_x86_ops->hardware_enable(); |
0dd6a6ed ZA |
7473 | if (ret != 0) |
7474 | return ret; | |
7475 | ||
4ea1636b | 7476 | local_tsc = rdtsc(); |
0dd6a6ed ZA |
7477 | stable = !check_tsc_unstable(); |
7478 | list_for_each_entry(kvm, &vm_list, vm_list) { | |
7479 | kvm_for_each_vcpu(i, vcpu, kvm) { | |
7480 | if (!stable && vcpu->cpu == smp_processor_id()) | |
105b21bb | 7481 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu); |
0dd6a6ed ZA |
7482 | if (stable && vcpu->arch.last_host_tsc > local_tsc) { |
7483 | backwards_tsc = true; | |
7484 | if (vcpu->arch.last_host_tsc > max_tsc) | |
7485 | max_tsc = vcpu->arch.last_host_tsc; | |
7486 | } | |
7487 | } | |
7488 | } | |
7489 | ||
7490 | /* | |
7491 | * Sometimes, even reliable TSCs go backwards. This happens on | |
7492 | * platforms that reset TSC during suspend or hibernate actions, but | |
7493 | * maintain synchronization. We must compensate. Fortunately, we can | |
7494 | * detect that condition here, which happens early in CPU bringup, | |
7495 | * before any KVM threads can be running. Unfortunately, we can't | |
7496 | * bring the TSCs fully up to date with real time, as we aren't yet far | |
7497 | * enough into CPU bringup that we know how much real time has actually | |
7498 | * elapsed; our helper function, get_kernel_ns() will be using boot | |
7499 | * variables that haven't been updated yet. | |
7500 | * | |
7501 | * So we simply find the maximum observed TSC above, then record the | |
7502 | * adjustment to TSC in each VCPU. When the VCPU later gets loaded, | |
7503 | * the adjustment will be applied. Note that we accumulate | |
7504 | * adjustments, in case multiple suspend cycles happen before some VCPU | |
7505 | * gets a chance to run again. In the event that no KVM threads get a | |
7506 | * chance to run, we will miss the entire elapsed period, as we'll have | |
7507 | * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may | |
7508 | * loose cycle time. This isn't too big a deal, since the loss will be | |
7509 | * uniform across all VCPUs (not to mention the scenario is extremely | |
7510 | * unlikely). It is possible that a second hibernate recovery happens | |
7511 | * much faster than a first, causing the observed TSC here to be | |
7512 | * smaller; this would require additional padding adjustment, which is | |
7513 | * why we set last_host_tsc to the local tsc observed here. | |
7514 | * | |
7515 | * N.B. - this code below runs only on platforms with reliable TSC, | |
7516 | * as that is the only way backwards_tsc is set above. Also note | |
7517 | * that this runs for ALL vcpus, which is not a bug; all VCPUs should | |
7518 | * have the same delta_cyc adjustment applied if backwards_tsc | |
7519 | * is detected. Note further, this adjustment is only done once, | |
7520 | * as we reset last_host_tsc on all VCPUs to stop this from being | |
7521 | * called multiple times (one for each physical CPU bringup). | |
7522 | * | |
4a969980 | 7523 | * Platforms with unreliable TSCs don't have to deal with this, they |
0dd6a6ed ZA |
7524 | * will be compensated by the logic in vcpu_load, which sets the TSC to |
7525 | * catchup mode. This will catchup all VCPUs to real time, but cannot | |
7526 | * guarantee that they stay in perfect synchronization. | |
7527 | */ | |
7528 | if (backwards_tsc) { | |
7529 | u64 delta_cyc = max_tsc - local_tsc; | |
16a96021 | 7530 | backwards_tsc_observed = true; |
0dd6a6ed ZA |
7531 | list_for_each_entry(kvm, &vm_list, vm_list) { |
7532 | kvm_for_each_vcpu(i, vcpu, kvm) { | |
7533 | vcpu->arch.tsc_offset_adjustment += delta_cyc; | |
7534 | vcpu->arch.last_host_tsc = local_tsc; | |
105b21bb | 7535 | kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu); |
0dd6a6ed ZA |
7536 | } |
7537 | ||
7538 | /* | |
7539 | * We have to disable TSC offset matching.. if you were | |
7540 | * booting a VM while issuing an S4 host suspend.... | |
7541 | * you may have some problem. Solving this issue is | |
7542 | * left as an exercise to the reader. | |
7543 | */ | |
7544 | kvm->arch.last_tsc_nsec = 0; | |
7545 | kvm->arch.last_tsc_write = 0; | |
7546 | } | |
7547 | ||
7548 | } | |
7549 | return 0; | |
e9b11c17 ZX |
7550 | } |
7551 | ||
13a34e06 | 7552 | void kvm_arch_hardware_disable(void) |
e9b11c17 | 7553 | { |
13a34e06 RK |
7554 | kvm_x86_ops->hardware_disable(); |
7555 | drop_user_return_notifiers(); | |
e9b11c17 ZX |
7556 | } |
7557 | ||
7558 | int kvm_arch_hardware_setup(void) | |
7559 | { | |
9e9c3fe4 NA |
7560 | int r; |
7561 | ||
7562 | r = kvm_x86_ops->hardware_setup(); | |
7563 | if (r != 0) | |
7564 | return r; | |
7565 | ||
35181e86 HZ |
7566 | if (kvm_has_tsc_control) { |
7567 | /* | |
7568 | * Make sure the user can only configure tsc_khz values that | |
7569 | * fit into a signed integer. | |
7570 | * A min value is not calculated needed because it will always | |
7571 | * be 1 on all machines. | |
7572 | */ | |
7573 | u64 max = min(0x7fffffffULL, | |
7574 | __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz)); | |
7575 | kvm_max_guest_tsc_khz = max; | |
7576 | ||
ad721883 | 7577 | kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits; |
35181e86 | 7578 | } |
ad721883 | 7579 | |
9e9c3fe4 NA |
7580 | kvm_init_msr_list(); |
7581 | return 0; | |
e9b11c17 ZX |
7582 | } |
7583 | ||
7584 | void kvm_arch_hardware_unsetup(void) | |
7585 | { | |
7586 | kvm_x86_ops->hardware_unsetup(); | |
7587 | } | |
7588 | ||
7589 | void kvm_arch_check_processor_compat(void *rtn) | |
7590 | { | |
7591 | kvm_x86_ops->check_processor_compatibility(rtn); | |
d71ba788 PB |
7592 | } |
7593 | ||
7594 | bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu) | |
7595 | { | |
7596 | return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id; | |
7597 | } | |
7598 | EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp); | |
7599 | ||
7600 | bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu) | |
7601 | { | |
7602 | return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0; | |
e9b11c17 ZX |
7603 | } |
7604 | ||
3e515705 AK |
7605 | bool kvm_vcpu_compatible(struct kvm_vcpu *vcpu) |
7606 | { | |
35754c98 | 7607 | return irqchip_in_kernel(vcpu->kvm) == lapic_in_kernel(vcpu); |
3e515705 AK |
7608 | } |
7609 | ||
54e9818f | 7610 | struct static_key kvm_no_apic_vcpu __read_mostly; |
bce87cce | 7611 | EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu); |
54e9818f | 7612 | |
e9b11c17 ZX |
7613 | int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu) |
7614 | { | |
7615 | struct page *page; | |
7616 | struct kvm *kvm; | |
7617 | int r; | |
7618 | ||
7619 | BUG_ON(vcpu->kvm == NULL); | |
7620 | kvm = vcpu->kvm; | |
7621 | ||
d62caabb | 7622 | vcpu->arch.apicv_active = kvm_x86_ops->get_enable_apicv(); |
6aef266c | 7623 | vcpu->arch.pv.pv_unhalted = false; |
9aabc88f | 7624 | vcpu->arch.emulate_ctxt.ops = &emulate_ops; |
58d269d8 | 7625 | if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_reset_bsp(vcpu)) |
a4535290 | 7626 | vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE; |
e9b11c17 | 7627 | else |
a4535290 | 7628 | vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED; |
e9b11c17 ZX |
7629 | |
7630 | page = alloc_page(GFP_KERNEL | __GFP_ZERO); | |
7631 | if (!page) { | |
7632 | r = -ENOMEM; | |
7633 | goto fail; | |
7634 | } | |
ad312c7c | 7635 | vcpu->arch.pio_data = page_address(page); |
e9b11c17 | 7636 | |
cc578287 | 7637 | kvm_set_tsc_khz(vcpu, max_tsc_khz); |
c285545f | 7638 | |
e9b11c17 ZX |
7639 | r = kvm_mmu_create(vcpu); |
7640 | if (r < 0) | |
7641 | goto fail_free_pio_data; | |
7642 | ||
7643 | if (irqchip_in_kernel(kvm)) { | |
7644 | r = kvm_create_lapic(vcpu); | |
7645 | if (r < 0) | |
7646 | goto fail_mmu_destroy; | |
54e9818f GN |
7647 | } else |
7648 | static_key_slow_inc(&kvm_no_apic_vcpu); | |
e9b11c17 | 7649 | |
890ca9ae HY |
7650 | vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4, |
7651 | GFP_KERNEL); | |
7652 | if (!vcpu->arch.mce_banks) { | |
7653 | r = -ENOMEM; | |
443c39bc | 7654 | goto fail_free_lapic; |
890ca9ae HY |
7655 | } |
7656 | vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS; | |
7657 | ||
f1797359 WY |
7658 | if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) { |
7659 | r = -ENOMEM; | |
f5f48ee1 | 7660 | goto fail_free_mce_banks; |
f1797359 | 7661 | } |
f5f48ee1 | 7662 | |
0ee6a517 | 7663 | fx_init(vcpu); |
66f7b72e | 7664 | |
ba904635 | 7665 | vcpu->arch.ia32_tsc_adjust_msr = 0x0; |
0b79459b | 7666 | vcpu->arch.pv_time_enabled = false; |
d7876f1b PB |
7667 | |
7668 | vcpu->arch.guest_supported_xcr0 = 0; | |
4344ee98 | 7669 | vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET; |
d7876f1b | 7670 | |
5a4f55cd EK |
7671 | vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu); |
7672 | ||
74545705 RK |
7673 | vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT; |
7674 | ||
af585b92 | 7675 | kvm_async_pf_hash_reset(vcpu); |
f5132b01 | 7676 | kvm_pmu_init(vcpu); |
af585b92 | 7677 | |
1c1a9ce9 SR |
7678 | vcpu->arch.pending_external_vector = -1; |
7679 | ||
5c919412 AS |
7680 | kvm_hv_vcpu_init(vcpu); |
7681 | ||
e9b11c17 | 7682 | return 0; |
0ee6a517 | 7683 | |
f5f48ee1 SY |
7684 | fail_free_mce_banks: |
7685 | kfree(vcpu->arch.mce_banks); | |
443c39bc WY |
7686 | fail_free_lapic: |
7687 | kvm_free_lapic(vcpu); | |
e9b11c17 ZX |
7688 | fail_mmu_destroy: |
7689 | kvm_mmu_destroy(vcpu); | |
7690 | fail_free_pio_data: | |
ad312c7c | 7691 | free_page((unsigned long)vcpu->arch.pio_data); |
e9b11c17 ZX |
7692 | fail: |
7693 | return r; | |
7694 | } | |
7695 | ||
7696 | void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu) | |
7697 | { | |
f656ce01 MT |
7698 | int idx; |
7699 | ||
1f4b34f8 | 7700 | kvm_hv_vcpu_uninit(vcpu); |
f5132b01 | 7701 | kvm_pmu_destroy(vcpu); |
36cb93fd | 7702 | kfree(vcpu->arch.mce_banks); |
e9b11c17 | 7703 | kvm_free_lapic(vcpu); |
f656ce01 | 7704 | idx = srcu_read_lock(&vcpu->kvm->srcu); |
e9b11c17 | 7705 | kvm_mmu_destroy(vcpu); |
f656ce01 | 7706 | srcu_read_unlock(&vcpu->kvm->srcu, idx); |
ad312c7c | 7707 | free_page((unsigned long)vcpu->arch.pio_data); |
35754c98 | 7708 | if (!lapic_in_kernel(vcpu)) |
54e9818f | 7709 | static_key_slow_dec(&kvm_no_apic_vcpu); |
e9b11c17 | 7710 | } |
d19a9cd2 | 7711 | |
e790d9ef RK |
7712 | void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu) |
7713 | { | |
ae97a3b8 | 7714 | kvm_x86_ops->sched_in(vcpu, cpu); |
e790d9ef RK |
7715 | } |
7716 | ||
e08b9637 | 7717 | int kvm_arch_init_vm(struct kvm *kvm, unsigned long type) |
d19a9cd2 | 7718 | { |
e08b9637 CO |
7719 | if (type) |
7720 | return -EINVAL; | |
7721 | ||
6ef768fa | 7722 | INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list); |
f05e70ac | 7723 | INIT_LIST_HEAD(&kvm->arch.active_mmu_pages); |
365c8868 | 7724 | INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages); |
4d5c5d0f | 7725 | INIT_LIST_HEAD(&kvm->arch.assigned_dev_head); |
e0f0bbc5 | 7726 | atomic_set(&kvm->arch.noncoherent_dma_count, 0); |
d19a9cd2 | 7727 | |
5550af4d SY |
7728 | /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */ |
7729 | set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap); | |
7a84428a AW |
7730 | /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */ |
7731 | set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID, | |
7732 | &kvm->arch.irq_sources_bitmap); | |
5550af4d | 7733 | |
038f8c11 | 7734 | raw_spin_lock_init(&kvm->arch.tsc_write_lock); |
1e08ec4a | 7735 | mutex_init(&kvm->arch.apic_map_lock); |
d828199e MT |
7736 | spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock); |
7737 | ||
7738 | pvclock_update_vm_gtod_copy(kvm); | |
53f658b3 | 7739 | |
7e44e449 | 7740 | INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn); |
332967a3 | 7741 | INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn); |
7e44e449 | 7742 | |
0eb05bf2 | 7743 | kvm_page_track_init(kvm); |
13d268ca | 7744 | kvm_mmu_init_vm(kvm); |
0eb05bf2 | 7745 | |
d89f5eff | 7746 | return 0; |
d19a9cd2 ZX |
7747 | } |
7748 | ||
7749 | static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu) | |
7750 | { | |
9fc77441 MT |
7751 | int r; |
7752 | r = vcpu_load(vcpu); | |
7753 | BUG_ON(r); | |
d19a9cd2 ZX |
7754 | kvm_mmu_unload(vcpu); |
7755 | vcpu_put(vcpu); | |
7756 | } | |
7757 | ||
7758 | static void kvm_free_vcpus(struct kvm *kvm) | |
7759 | { | |
7760 | unsigned int i; | |
988a2cae | 7761 | struct kvm_vcpu *vcpu; |
d19a9cd2 ZX |
7762 | |
7763 | /* | |
7764 | * Unpin any mmu pages first. | |
7765 | */ | |
af585b92 GN |
7766 | kvm_for_each_vcpu(i, vcpu, kvm) { |
7767 | kvm_clear_async_pf_completion_queue(vcpu); | |
988a2cae | 7768 | kvm_unload_vcpu_mmu(vcpu); |
af585b92 | 7769 | } |
988a2cae GN |
7770 | kvm_for_each_vcpu(i, vcpu, kvm) |
7771 | kvm_arch_vcpu_free(vcpu); | |
7772 | ||
7773 | mutex_lock(&kvm->lock); | |
7774 | for (i = 0; i < atomic_read(&kvm->online_vcpus); i++) | |
7775 | kvm->vcpus[i] = NULL; | |
d19a9cd2 | 7776 | |
988a2cae GN |
7777 | atomic_set(&kvm->online_vcpus, 0); |
7778 | mutex_unlock(&kvm->lock); | |
d19a9cd2 ZX |
7779 | } |
7780 | ||
ad8ba2cd SY |
7781 | void kvm_arch_sync_events(struct kvm *kvm) |
7782 | { | |
332967a3 | 7783 | cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work); |
7e44e449 | 7784 | cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work); |
ba4cef31 | 7785 | kvm_free_all_assigned_devices(kvm); |
aea924f6 | 7786 | kvm_free_pit(kvm); |
ad8ba2cd SY |
7787 | } |
7788 | ||
1d8007bd | 7789 | int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size) |
9da0e4d5 PB |
7790 | { |
7791 | int i, r; | |
25188b99 | 7792 | unsigned long hva; |
f0d648bd PB |
7793 | struct kvm_memslots *slots = kvm_memslots(kvm); |
7794 | struct kvm_memory_slot *slot, old; | |
9da0e4d5 PB |
7795 | |
7796 | /* Called with kvm->slots_lock held. */ | |
1d8007bd PB |
7797 | if (WARN_ON(id >= KVM_MEM_SLOTS_NUM)) |
7798 | return -EINVAL; | |
9da0e4d5 | 7799 | |
f0d648bd PB |
7800 | slot = id_to_memslot(slots, id); |
7801 | if (size) { | |
7802 | if (WARN_ON(slot->npages)) | |
7803 | return -EEXIST; | |
7804 | ||
7805 | /* | |
7806 | * MAP_SHARED to prevent internal slot pages from being moved | |
7807 | * by fork()/COW. | |
7808 | */ | |
7809 | hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE, | |
7810 | MAP_SHARED | MAP_ANONYMOUS, 0); | |
7811 | if (IS_ERR((void *)hva)) | |
7812 | return PTR_ERR((void *)hva); | |
7813 | } else { | |
7814 | if (!slot->npages) | |
7815 | return 0; | |
7816 | ||
7817 | hva = 0; | |
7818 | } | |
7819 | ||
7820 | old = *slot; | |
9da0e4d5 | 7821 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) { |
1d8007bd | 7822 | struct kvm_userspace_memory_region m; |
9da0e4d5 | 7823 | |
1d8007bd PB |
7824 | m.slot = id | (i << 16); |
7825 | m.flags = 0; | |
7826 | m.guest_phys_addr = gpa; | |
f0d648bd | 7827 | m.userspace_addr = hva; |
1d8007bd | 7828 | m.memory_size = size; |
9da0e4d5 PB |
7829 | r = __kvm_set_memory_region(kvm, &m); |
7830 | if (r < 0) | |
7831 | return r; | |
7832 | } | |
7833 | ||
f0d648bd PB |
7834 | if (!size) { |
7835 | r = vm_munmap(old.userspace_addr, old.npages * PAGE_SIZE); | |
7836 | WARN_ON(r < 0); | |
7837 | } | |
7838 | ||
9da0e4d5 PB |
7839 | return 0; |
7840 | } | |
7841 | EXPORT_SYMBOL_GPL(__x86_set_memory_region); | |
7842 | ||
1d8007bd | 7843 | int x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size) |
9da0e4d5 PB |
7844 | { |
7845 | int r; | |
7846 | ||
7847 | mutex_lock(&kvm->slots_lock); | |
1d8007bd | 7848 | r = __x86_set_memory_region(kvm, id, gpa, size); |
9da0e4d5 PB |
7849 | mutex_unlock(&kvm->slots_lock); |
7850 | ||
7851 | return r; | |
7852 | } | |
7853 | EXPORT_SYMBOL_GPL(x86_set_memory_region); | |
7854 | ||
d19a9cd2 ZX |
7855 | void kvm_arch_destroy_vm(struct kvm *kvm) |
7856 | { | |
27469d29 AH |
7857 | if (current->mm == kvm->mm) { |
7858 | /* | |
7859 | * Free memory regions allocated on behalf of userspace, | |
7860 | * unless the the memory map has changed due to process exit | |
7861 | * or fd copying. | |
7862 | */ | |
1d8007bd PB |
7863 | x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0); |
7864 | x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, 0, 0); | |
7865 | x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0); | |
27469d29 | 7866 | } |
6eb55818 | 7867 | kvm_iommu_unmap_guest(kvm); |
d7deeeb0 ZX |
7868 | kfree(kvm->arch.vpic); |
7869 | kfree(kvm->arch.vioapic); | |
d19a9cd2 | 7870 | kvm_free_vcpus(kvm); |
1e08ec4a | 7871 | kfree(rcu_dereference_check(kvm->arch.apic_map, 1)); |
13d268ca | 7872 | kvm_mmu_uninit_vm(kvm); |
d19a9cd2 | 7873 | } |
0de10343 | 7874 | |
5587027c | 7875 | void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free, |
db3fe4eb TY |
7876 | struct kvm_memory_slot *dont) |
7877 | { | |
7878 | int i; | |
7879 | ||
d89cc617 TY |
7880 | for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) { |
7881 | if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) { | |
548ef284 | 7882 | kvfree(free->arch.rmap[i]); |
d89cc617 | 7883 | free->arch.rmap[i] = NULL; |
77d11309 | 7884 | } |
d89cc617 TY |
7885 | if (i == 0) |
7886 | continue; | |
7887 | ||
7888 | if (!dont || free->arch.lpage_info[i - 1] != | |
7889 | dont->arch.lpage_info[i - 1]) { | |
548ef284 | 7890 | kvfree(free->arch.lpage_info[i - 1]); |
d89cc617 | 7891 | free->arch.lpage_info[i - 1] = NULL; |
db3fe4eb TY |
7892 | } |
7893 | } | |
21ebbeda XG |
7894 | |
7895 | kvm_page_track_free_memslot(free, dont); | |
db3fe4eb TY |
7896 | } |
7897 | ||
5587027c AK |
7898 | int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot, |
7899 | unsigned long npages) | |
db3fe4eb TY |
7900 | { |
7901 | int i; | |
7902 | ||
d89cc617 | 7903 | for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) { |
92f94f1e | 7904 | struct kvm_lpage_info *linfo; |
db3fe4eb TY |
7905 | unsigned long ugfn; |
7906 | int lpages; | |
d89cc617 | 7907 | int level = i + 1; |
db3fe4eb TY |
7908 | |
7909 | lpages = gfn_to_index(slot->base_gfn + npages - 1, | |
7910 | slot->base_gfn, level) + 1; | |
7911 | ||
d89cc617 TY |
7912 | slot->arch.rmap[i] = |
7913 | kvm_kvzalloc(lpages * sizeof(*slot->arch.rmap[i])); | |
7914 | if (!slot->arch.rmap[i]) | |
77d11309 | 7915 | goto out_free; |
d89cc617 TY |
7916 | if (i == 0) |
7917 | continue; | |
77d11309 | 7918 | |
92f94f1e XG |
7919 | linfo = kvm_kvzalloc(lpages * sizeof(*linfo)); |
7920 | if (!linfo) | |
db3fe4eb TY |
7921 | goto out_free; |
7922 | ||
92f94f1e XG |
7923 | slot->arch.lpage_info[i - 1] = linfo; |
7924 | ||
db3fe4eb | 7925 | if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1)) |
92f94f1e | 7926 | linfo[0].disallow_lpage = 1; |
db3fe4eb | 7927 | if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1)) |
92f94f1e | 7928 | linfo[lpages - 1].disallow_lpage = 1; |
db3fe4eb TY |
7929 | ugfn = slot->userspace_addr >> PAGE_SHIFT; |
7930 | /* | |
7931 | * If the gfn and userspace address are not aligned wrt each | |
7932 | * other, or if explicitly asked to, disable large page | |
7933 | * support for this slot | |
7934 | */ | |
7935 | if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) || | |
7936 | !kvm_largepages_enabled()) { | |
7937 | unsigned long j; | |
7938 | ||
7939 | for (j = 0; j < lpages; ++j) | |
92f94f1e | 7940 | linfo[j].disallow_lpage = 1; |
db3fe4eb TY |
7941 | } |
7942 | } | |
7943 | ||
21ebbeda XG |
7944 | if (kvm_page_track_create_memslot(slot, npages)) |
7945 | goto out_free; | |
7946 | ||
db3fe4eb TY |
7947 | return 0; |
7948 | ||
7949 | out_free: | |
d89cc617 | 7950 | for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) { |
548ef284 | 7951 | kvfree(slot->arch.rmap[i]); |
d89cc617 TY |
7952 | slot->arch.rmap[i] = NULL; |
7953 | if (i == 0) | |
7954 | continue; | |
7955 | ||
548ef284 | 7956 | kvfree(slot->arch.lpage_info[i - 1]); |
d89cc617 | 7957 | slot->arch.lpage_info[i - 1] = NULL; |
db3fe4eb TY |
7958 | } |
7959 | return -ENOMEM; | |
7960 | } | |
7961 | ||
15f46015 | 7962 | void kvm_arch_memslots_updated(struct kvm *kvm, struct kvm_memslots *slots) |
e59dbe09 | 7963 | { |
e6dff7d1 TY |
7964 | /* |
7965 | * memslots->generation has been incremented. | |
7966 | * mmio generation may have reached its maximum value. | |
7967 | */ | |
54bf36aa | 7968 | kvm_mmu_invalidate_mmio_sptes(kvm, slots); |
e59dbe09 TY |
7969 | } |
7970 | ||
f7784b8e MT |
7971 | int kvm_arch_prepare_memory_region(struct kvm *kvm, |
7972 | struct kvm_memory_slot *memslot, | |
09170a49 | 7973 | const struct kvm_userspace_memory_region *mem, |
7b6195a9 | 7974 | enum kvm_mr_change change) |
0de10343 | 7975 | { |
f7784b8e MT |
7976 | return 0; |
7977 | } | |
7978 | ||
88178fd4 KH |
7979 | static void kvm_mmu_slot_apply_flags(struct kvm *kvm, |
7980 | struct kvm_memory_slot *new) | |
7981 | { | |
7982 | /* Still write protect RO slot */ | |
7983 | if (new->flags & KVM_MEM_READONLY) { | |
7984 | kvm_mmu_slot_remove_write_access(kvm, new); | |
7985 | return; | |
7986 | } | |
7987 | ||
7988 | /* | |
7989 | * Call kvm_x86_ops dirty logging hooks when they are valid. | |
7990 | * | |
7991 | * kvm_x86_ops->slot_disable_log_dirty is called when: | |
7992 | * | |
7993 | * - KVM_MR_CREATE with dirty logging is disabled | |
7994 | * - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag | |
7995 | * | |
7996 | * The reason is, in case of PML, we need to set D-bit for any slots | |
7997 | * with dirty logging disabled in order to eliminate unnecessary GPA | |
7998 | * logging in PML buffer (and potential PML buffer full VMEXT). This | |
7999 | * guarantees leaving PML enabled during guest's lifetime won't have | |
8000 | * any additonal overhead from PML when guest is running with dirty | |
8001 | * logging disabled for memory slots. | |
8002 | * | |
8003 | * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot | |
8004 | * to dirty logging mode. | |
8005 | * | |
8006 | * If kvm_x86_ops dirty logging hooks are invalid, use write protect. | |
8007 | * | |
8008 | * In case of write protect: | |
8009 | * | |
8010 | * Write protect all pages for dirty logging. | |
8011 | * | |
8012 | * All the sptes including the large sptes which point to this | |
8013 | * slot are set to readonly. We can not create any new large | |
8014 | * spte on this slot until the end of the logging. | |
8015 | * | |
8016 | * See the comments in fast_page_fault(). | |
8017 | */ | |
8018 | if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) { | |
8019 | if (kvm_x86_ops->slot_enable_log_dirty) | |
8020 | kvm_x86_ops->slot_enable_log_dirty(kvm, new); | |
8021 | else | |
8022 | kvm_mmu_slot_remove_write_access(kvm, new); | |
8023 | } else { | |
8024 | if (kvm_x86_ops->slot_disable_log_dirty) | |
8025 | kvm_x86_ops->slot_disable_log_dirty(kvm, new); | |
8026 | } | |
8027 | } | |
8028 | ||
f7784b8e | 8029 | void kvm_arch_commit_memory_region(struct kvm *kvm, |
09170a49 | 8030 | const struct kvm_userspace_memory_region *mem, |
8482644a | 8031 | const struct kvm_memory_slot *old, |
f36f3f28 | 8032 | const struct kvm_memory_slot *new, |
8482644a | 8033 | enum kvm_mr_change change) |
f7784b8e | 8034 | { |
8482644a | 8035 | int nr_mmu_pages = 0; |
f7784b8e | 8036 | |
48c0e4e9 XG |
8037 | if (!kvm->arch.n_requested_mmu_pages) |
8038 | nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm); | |
8039 | ||
48c0e4e9 | 8040 | if (nr_mmu_pages) |
0de10343 | 8041 | kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages); |
1c91cad4 | 8042 | |
3ea3b7fa WL |
8043 | /* |
8044 | * Dirty logging tracks sptes in 4k granularity, meaning that large | |
8045 | * sptes have to be split. If live migration is successful, the guest | |
8046 | * in the source machine will be destroyed and large sptes will be | |
8047 | * created in the destination. However, if the guest continues to run | |
8048 | * in the source machine (for example if live migration fails), small | |
8049 | * sptes will remain around and cause bad performance. | |
8050 | * | |
8051 | * Scan sptes if dirty logging has been stopped, dropping those | |
8052 | * which can be collapsed into a single large-page spte. Later | |
8053 | * page faults will create the large-page sptes. | |
8054 | */ | |
8055 | if ((change != KVM_MR_DELETE) && | |
8056 | (old->flags & KVM_MEM_LOG_DIRTY_PAGES) && | |
8057 | !(new->flags & KVM_MEM_LOG_DIRTY_PAGES)) | |
8058 | kvm_mmu_zap_collapsible_sptes(kvm, new); | |
8059 | ||
c972f3b1 | 8060 | /* |
88178fd4 | 8061 | * Set up write protection and/or dirty logging for the new slot. |
c126d94f | 8062 | * |
88178fd4 KH |
8063 | * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have |
8064 | * been zapped so no dirty logging staff is needed for old slot. For | |
8065 | * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the | |
8066 | * new and it's also covered when dealing with the new slot. | |
f36f3f28 PB |
8067 | * |
8068 | * FIXME: const-ify all uses of struct kvm_memory_slot. | |
c972f3b1 | 8069 | */ |
88178fd4 | 8070 | if (change != KVM_MR_DELETE) |
f36f3f28 | 8071 | kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new); |
0de10343 | 8072 | } |
1d737c8a | 8073 | |
2df72e9b | 8074 | void kvm_arch_flush_shadow_all(struct kvm *kvm) |
34d4cb8f | 8075 | { |
6ca18b69 | 8076 | kvm_mmu_invalidate_zap_all_pages(kvm); |
34d4cb8f MT |
8077 | } |
8078 | ||
2df72e9b MT |
8079 | void kvm_arch_flush_shadow_memslot(struct kvm *kvm, |
8080 | struct kvm_memory_slot *slot) | |
8081 | { | |
6ca18b69 | 8082 | kvm_mmu_invalidate_zap_all_pages(kvm); |
2df72e9b MT |
8083 | } |
8084 | ||
5d9bc648 PB |
8085 | static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu) |
8086 | { | |
8087 | if (!list_empty_careful(&vcpu->async_pf.done)) | |
8088 | return true; | |
8089 | ||
8090 | if (kvm_apic_has_events(vcpu)) | |
8091 | return true; | |
8092 | ||
8093 | if (vcpu->arch.pv.pv_unhalted) | |
8094 | return true; | |
8095 | ||
8096 | if (atomic_read(&vcpu->arch.nmi_queued)) | |
8097 | return true; | |
8098 | ||
73917739 PB |
8099 | if (test_bit(KVM_REQ_SMI, &vcpu->requests)) |
8100 | return true; | |
8101 | ||
5d9bc648 PB |
8102 | if (kvm_arch_interrupt_allowed(vcpu) && |
8103 | kvm_cpu_has_interrupt(vcpu)) | |
8104 | return true; | |
8105 | ||
1f4b34f8 AS |
8106 | if (kvm_hv_has_stimer_pending(vcpu)) |
8107 | return true; | |
8108 | ||
5d9bc648 PB |
8109 | return false; |
8110 | } | |
8111 | ||
1d737c8a ZX |
8112 | int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu) |
8113 | { | |
b6b8a145 JK |
8114 | if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) |
8115 | kvm_x86_ops->check_nested_events(vcpu, false); | |
8116 | ||
5d9bc648 | 8117 | return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu); |
1d737c8a | 8118 | } |
5736199a | 8119 | |
b6d33834 | 8120 | int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu) |
5736199a | 8121 | { |
b6d33834 | 8122 | return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE; |
5736199a | 8123 | } |
78646121 GN |
8124 | |
8125 | int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu) | |
8126 | { | |
8127 | return kvm_x86_ops->interrupt_allowed(vcpu); | |
8128 | } | |
229456fc | 8129 | |
82b32774 | 8130 | unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu) |
f92653ee | 8131 | { |
82b32774 NA |
8132 | if (is_64_bit_mode(vcpu)) |
8133 | return kvm_rip_read(vcpu); | |
8134 | return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) + | |
8135 | kvm_rip_read(vcpu)); | |
8136 | } | |
8137 | EXPORT_SYMBOL_GPL(kvm_get_linear_rip); | |
f92653ee | 8138 | |
82b32774 NA |
8139 | bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip) |
8140 | { | |
8141 | return kvm_get_linear_rip(vcpu) == linear_rip; | |
f92653ee JK |
8142 | } |
8143 | EXPORT_SYMBOL_GPL(kvm_is_linear_rip); | |
8144 | ||
94fe45da JK |
8145 | unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu) |
8146 | { | |
8147 | unsigned long rflags; | |
8148 | ||
8149 | rflags = kvm_x86_ops->get_rflags(vcpu); | |
8150 | if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) | |
c310bac5 | 8151 | rflags &= ~X86_EFLAGS_TF; |
94fe45da JK |
8152 | return rflags; |
8153 | } | |
8154 | EXPORT_SYMBOL_GPL(kvm_get_rflags); | |
8155 | ||
6addfc42 | 8156 | static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags) |
94fe45da JK |
8157 | { |
8158 | if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP && | |
f92653ee | 8159 | kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip)) |
c310bac5 | 8160 | rflags |= X86_EFLAGS_TF; |
94fe45da | 8161 | kvm_x86_ops->set_rflags(vcpu, rflags); |
6addfc42 PB |
8162 | } |
8163 | ||
8164 | void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags) | |
8165 | { | |
8166 | __kvm_set_rflags(vcpu, rflags); | |
3842d135 | 8167 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
94fe45da JK |
8168 | } |
8169 | EXPORT_SYMBOL_GPL(kvm_set_rflags); | |
8170 | ||
56028d08 GN |
8171 | void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work) |
8172 | { | |
8173 | int r; | |
8174 | ||
fb67e14f | 8175 | if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) || |
f2e10669 | 8176 | work->wakeup_all) |
56028d08 GN |
8177 | return; |
8178 | ||
8179 | r = kvm_mmu_reload(vcpu); | |
8180 | if (unlikely(r)) | |
8181 | return; | |
8182 | ||
fb67e14f XG |
8183 | if (!vcpu->arch.mmu.direct_map && |
8184 | work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu)) | |
8185 | return; | |
8186 | ||
56028d08 GN |
8187 | vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true); |
8188 | } | |
8189 | ||
af585b92 GN |
8190 | static inline u32 kvm_async_pf_hash_fn(gfn_t gfn) |
8191 | { | |
8192 | return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU)); | |
8193 | } | |
8194 | ||
8195 | static inline u32 kvm_async_pf_next_probe(u32 key) | |
8196 | { | |
8197 | return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1); | |
8198 | } | |
8199 | ||
8200 | static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn) | |
8201 | { | |
8202 | u32 key = kvm_async_pf_hash_fn(gfn); | |
8203 | ||
8204 | while (vcpu->arch.apf.gfns[key] != ~0) | |
8205 | key = kvm_async_pf_next_probe(key); | |
8206 | ||
8207 | vcpu->arch.apf.gfns[key] = gfn; | |
8208 | } | |
8209 | ||
8210 | static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn) | |
8211 | { | |
8212 | int i; | |
8213 | u32 key = kvm_async_pf_hash_fn(gfn); | |
8214 | ||
8215 | for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) && | |
c7d28c24 XG |
8216 | (vcpu->arch.apf.gfns[key] != gfn && |
8217 | vcpu->arch.apf.gfns[key] != ~0); i++) | |
af585b92 GN |
8218 | key = kvm_async_pf_next_probe(key); |
8219 | ||
8220 | return key; | |
8221 | } | |
8222 | ||
8223 | bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn) | |
8224 | { | |
8225 | return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn; | |
8226 | } | |
8227 | ||
8228 | static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn) | |
8229 | { | |
8230 | u32 i, j, k; | |
8231 | ||
8232 | i = j = kvm_async_pf_gfn_slot(vcpu, gfn); | |
8233 | while (true) { | |
8234 | vcpu->arch.apf.gfns[i] = ~0; | |
8235 | do { | |
8236 | j = kvm_async_pf_next_probe(j); | |
8237 | if (vcpu->arch.apf.gfns[j] == ~0) | |
8238 | return; | |
8239 | k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]); | |
8240 | /* | |
8241 | * k lies cyclically in ]i,j] | |
8242 | * | i.k.j | | |
8243 | * |....j i.k.| or |.k..j i...| | |
8244 | */ | |
8245 | } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j)); | |
8246 | vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j]; | |
8247 | i = j; | |
8248 | } | |
8249 | } | |
8250 | ||
7c90705b GN |
8251 | static int apf_put_user(struct kvm_vcpu *vcpu, u32 val) |
8252 | { | |
8253 | ||
8254 | return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val, | |
8255 | sizeof(val)); | |
8256 | } | |
8257 | ||
af585b92 GN |
8258 | void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu, |
8259 | struct kvm_async_pf *work) | |
8260 | { | |
6389ee94 AK |
8261 | struct x86_exception fault; |
8262 | ||
7c90705b | 8263 | trace_kvm_async_pf_not_present(work->arch.token, work->gva); |
af585b92 | 8264 | kvm_add_async_pf_gfn(vcpu, work->arch.gfn); |
7c90705b GN |
8265 | |
8266 | if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) || | |
fc5f06fa GN |
8267 | (vcpu->arch.apf.send_user_only && |
8268 | kvm_x86_ops->get_cpl(vcpu) == 0)) | |
7c90705b GN |
8269 | kvm_make_request(KVM_REQ_APF_HALT, vcpu); |
8270 | else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) { | |
6389ee94 AK |
8271 | fault.vector = PF_VECTOR; |
8272 | fault.error_code_valid = true; | |
8273 | fault.error_code = 0; | |
8274 | fault.nested_page_fault = false; | |
8275 | fault.address = work->arch.token; | |
8276 | kvm_inject_page_fault(vcpu, &fault); | |
7c90705b | 8277 | } |
af585b92 GN |
8278 | } |
8279 | ||
8280 | void kvm_arch_async_page_present(struct kvm_vcpu *vcpu, | |
8281 | struct kvm_async_pf *work) | |
8282 | { | |
6389ee94 AK |
8283 | struct x86_exception fault; |
8284 | ||
7c90705b | 8285 | trace_kvm_async_pf_ready(work->arch.token, work->gva); |
f2e10669 | 8286 | if (work->wakeup_all) |
7c90705b GN |
8287 | work->arch.token = ~0; /* broadcast wakeup */ |
8288 | else | |
8289 | kvm_del_async_pf_gfn(vcpu, work->arch.gfn); | |
8290 | ||
8291 | if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) && | |
8292 | !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) { | |
6389ee94 AK |
8293 | fault.vector = PF_VECTOR; |
8294 | fault.error_code_valid = true; | |
8295 | fault.error_code = 0; | |
8296 | fault.nested_page_fault = false; | |
8297 | fault.address = work->arch.token; | |
8298 | kvm_inject_page_fault(vcpu, &fault); | |
7c90705b | 8299 | } |
e6d53e3b | 8300 | vcpu->arch.apf.halted = false; |
a4fa1635 | 8301 | vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE; |
7c90705b GN |
8302 | } |
8303 | ||
8304 | bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu) | |
8305 | { | |
8306 | if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED)) | |
8307 | return true; | |
8308 | else | |
8309 | return !kvm_event_needs_reinjection(vcpu) && | |
8310 | kvm_x86_ops->interrupt_allowed(vcpu); | |
af585b92 GN |
8311 | } |
8312 | ||
5544eb9b PB |
8313 | void kvm_arch_start_assignment(struct kvm *kvm) |
8314 | { | |
8315 | atomic_inc(&kvm->arch.assigned_device_count); | |
8316 | } | |
8317 | EXPORT_SYMBOL_GPL(kvm_arch_start_assignment); | |
8318 | ||
8319 | void kvm_arch_end_assignment(struct kvm *kvm) | |
8320 | { | |
8321 | atomic_dec(&kvm->arch.assigned_device_count); | |
8322 | } | |
8323 | EXPORT_SYMBOL_GPL(kvm_arch_end_assignment); | |
8324 | ||
8325 | bool kvm_arch_has_assigned_device(struct kvm *kvm) | |
8326 | { | |
8327 | return atomic_read(&kvm->arch.assigned_device_count); | |
8328 | } | |
8329 | EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device); | |
8330 | ||
e0f0bbc5 AW |
8331 | void kvm_arch_register_noncoherent_dma(struct kvm *kvm) |
8332 | { | |
8333 | atomic_inc(&kvm->arch.noncoherent_dma_count); | |
8334 | } | |
8335 | EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma); | |
8336 | ||
8337 | void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm) | |
8338 | { | |
8339 | atomic_dec(&kvm->arch.noncoherent_dma_count); | |
8340 | } | |
8341 | EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma); | |
8342 | ||
8343 | bool kvm_arch_has_noncoherent_dma(struct kvm *kvm) | |
8344 | { | |
8345 | return atomic_read(&kvm->arch.noncoherent_dma_count); | |
8346 | } | |
8347 | EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma); | |
8348 | ||
87276880 FW |
8349 | int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons, |
8350 | struct irq_bypass_producer *prod) | |
8351 | { | |
8352 | struct kvm_kernel_irqfd *irqfd = | |
8353 | container_of(cons, struct kvm_kernel_irqfd, consumer); | |
8354 | ||
8355 | if (kvm_x86_ops->update_pi_irte) { | |
8356 | irqfd->producer = prod; | |
8357 | return kvm_x86_ops->update_pi_irte(irqfd->kvm, | |
8358 | prod->irq, irqfd->gsi, 1); | |
8359 | } | |
8360 | ||
8361 | return -EINVAL; | |
8362 | } | |
8363 | ||
8364 | void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons, | |
8365 | struct irq_bypass_producer *prod) | |
8366 | { | |
8367 | int ret; | |
8368 | struct kvm_kernel_irqfd *irqfd = | |
8369 | container_of(cons, struct kvm_kernel_irqfd, consumer); | |
8370 | ||
8371 | if (!kvm_x86_ops->update_pi_irte) { | |
8372 | WARN_ON(irqfd->producer != NULL); | |
8373 | return; | |
8374 | } | |
8375 | ||
8376 | WARN_ON(irqfd->producer != prod); | |
8377 | irqfd->producer = NULL; | |
8378 | ||
8379 | /* | |
8380 | * When producer of consumer is unregistered, we change back to | |
8381 | * remapped mode, so we can re-use the current implementation | |
8382 | * when the irq is masked/disabed or the consumer side (KVM | |
8383 | * int this case doesn't want to receive the interrupts. | |
8384 | */ | |
8385 | ret = kvm_x86_ops->update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0); | |
8386 | if (ret) | |
8387 | printk(KERN_INFO "irq bypass consumer (token %p) unregistration" | |
8388 | " fails: %d\n", irqfd->consumer.token, ret); | |
8389 | } | |
8390 | ||
8391 | int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq, | |
8392 | uint32_t guest_irq, bool set) | |
8393 | { | |
8394 | if (!kvm_x86_ops->update_pi_irte) | |
8395 | return -EINVAL; | |
8396 | ||
8397 | return kvm_x86_ops->update_pi_irte(kvm, host_irq, guest_irq, set); | |
8398 | } | |
8399 | ||
52004014 FW |
8400 | bool kvm_vector_hashing_enabled(void) |
8401 | { | |
8402 | return vector_hashing; | |
8403 | } | |
8404 | EXPORT_SYMBOL_GPL(kvm_vector_hashing_enabled); | |
8405 | ||
229456fc | 8406 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit); |
931c33b1 | 8407 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio); |
229456fc MT |
8408 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq); |
8409 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault); | |
8410 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr); | |
8411 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr); | |
0ac406de | 8412 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun); |
d8cabddf | 8413 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit); |
17897f36 | 8414 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject); |
236649de | 8415 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit); |
ec1ff790 | 8416 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga); |
532a46b9 | 8417 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit); |
2e554e8d | 8418 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts); |
489223ed | 8419 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset); |
7b46268d | 8420 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window); |
843e4330 | 8421 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full); |
efc64404 | 8422 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update); |