]>
Commit | Line | Data |
---|---|---|
20c8ccb1 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
043405e1 CO |
2 | /* |
3 | * Kernel-based Virtual Machine driver for Linux | |
4 | * | |
5 | * derived from drivers/kvm/kvm_main.c | |
6 | * | |
7 | * Copyright (C) 2006 Qumranet, Inc. | |
4d5c5d0f BAY |
8 | * Copyright (C) 2008 Qumranet, Inc. |
9 | * Copyright IBM Corporation, 2008 | |
9611c187 | 10 | * Copyright 2010 Red Hat, Inc. and/or its affiliates. |
043405e1 CO |
11 | * |
12 | * Authors: | |
13 | * Avi Kivity <avi@qumranet.com> | |
14 | * Yaniv Kamay <yaniv@qumranet.com> | |
4d5c5d0f BAY |
15 | * Amit Shah <amit.shah@qumranet.com> |
16 | * Ben-Ami Yassour <benami@il.ibm.com> | |
043405e1 CO |
17 | */ |
18 | ||
edf88417 | 19 | #include <linux/kvm_host.h> |
313a3dc7 | 20 | #include "irq.h" |
1d737c8a | 21 | #include "mmu.h" |
7837699f | 22 | #include "i8254.h" |
37817f29 | 23 | #include "tss.h" |
5fdbf976 | 24 | #include "kvm_cache_regs.h" |
2f728d66 | 25 | #include "kvm_emulate.h" |
26eef70c | 26 | #include "x86.h" |
00b27a3e | 27 | #include "cpuid.h" |
474a5bb9 | 28 | #include "pmu.h" |
e83d5887 | 29 | #include "hyperv.h" |
8df14af4 | 30 | #include "lapic.h" |
313a3dc7 | 31 | |
18068523 | 32 | #include <linux/clocksource.h> |
4d5c5d0f | 33 | #include <linux/interrupt.h> |
313a3dc7 CO |
34 | #include <linux/kvm.h> |
35 | #include <linux/fs.h> | |
36 | #include <linux/vmalloc.h> | |
1767e931 PG |
37 | #include <linux/export.h> |
38 | #include <linux/moduleparam.h> | |
0de10343 | 39 | #include <linux/mman.h> |
2bacc55c | 40 | #include <linux/highmem.h> |
19de40a8 | 41 | #include <linux/iommu.h> |
62c476c7 | 42 | #include <linux/intel-iommu.h> |
c8076604 | 43 | #include <linux/cpufreq.h> |
18863bdd | 44 | #include <linux/user-return-notifier.h> |
a983fb23 | 45 | #include <linux/srcu.h> |
5a0e3ad6 | 46 | #include <linux/slab.h> |
ff9d07a0 | 47 | #include <linux/perf_event.h> |
7bee342a | 48 | #include <linux/uaccess.h> |
af585b92 | 49 | #include <linux/hash.h> |
a1b60c1c | 50 | #include <linux/pci.h> |
16e8d74d MT |
51 | #include <linux/timekeeper_internal.h> |
52 | #include <linux/pvclock_gtod.h> | |
87276880 FW |
53 | #include <linux/kvm_irqfd.h> |
54 | #include <linux/irqbypass.h> | |
3905f9ad | 55 | #include <linux/sched/stat.h> |
0c5f81da | 56 | #include <linux/sched/isolation.h> |
d0ec49d4 | 57 | #include <linux/mem_encrypt.h> |
3905f9ad | 58 | |
aec51dc4 | 59 | #include <trace/events/kvm.h> |
2ed152af | 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> |
b0c39dc6 | 70 | #include <asm/mshyperv.h> |
0092e434 | 71 | #include <asm/hypervisor.h> |
bf8c55d8 | 72 | #include <asm/intel_pt.h> |
b3dc0695 | 73 | #include <asm/emulate_prefix.h> |
dd2cb348 | 74 | #include <clocksource/hyperv_timer.h> |
043405e1 | 75 | |
d1898b73 DH |
76 | #define CREATE_TRACE_POINTS |
77 | #include "trace.h" | |
78 | ||
313a3dc7 | 79 | #define MAX_IO_MSRS 256 |
890ca9ae | 80 | #define KVM_MAX_MCE_BANKS 32 |
c45dcc71 AR |
81 | u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P; |
82 | EXPORT_SYMBOL_GPL(kvm_mce_cap_supported); | |
890ca9ae | 83 | |
0f65dd70 | 84 | #define emul_to_vcpu(ctxt) \ |
c9b8b07c | 85 | ((struct kvm_vcpu *)(ctxt)->vcpu) |
0f65dd70 | 86 | |
50a37eb4 JR |
87 | /* EFER defaults: |
88 | * - enable syscall per default because its emulated by KVM | |
89 | * - enable LME and LMA per default on 64 bit KVM | |
90 | */ | |
91 | #ifdef CONFIG_X86_64 | |
1260edbe LJ |
92 | static |
93 | u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA)); | |
50a37eb4 | 94 | #else |
1260edbe | 95 | static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE); |
50a37eb4 | 96 | #endif |
313a3dc7 | 97 | |
b11306b5 SC |
98 | static u64 __read_mostly cr4_reserved_bits = CR4_RESERVED_BITS; |
99 | ||
833b45de PB |
100 | #define VM_STAT(x, ...) offsetof(struct kvm, stat.x), KVM_STAT_VM, ## __VA_ARGS__ |
101 | #define VCPU_STAT(x, ...) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU, ## __VA_ARGS__ | |
417bc304 | 102 | |
c519265f RK |
103 | #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \ |
104 | KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK) | |
37131313 | 105 | |
cb142eb7 | 106 | static void update_cr8_intercept(struct kvm_vcpu *vcpu); |
7460fb4a | 107 | static void process_nmi(struct kvm_vcpu *vcpu); |
ee2cd4b7 | 108 | static void enter_smm(struct kvm_vcpu *vcpu); |
6addfc42 | 109 | static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags); |
01643c51 KH |
110 | static void store_regs(struct kvm_vcpu *vcpu); |
111 | static int sync_regs(struct kvm_vcpu *vcpu); | |
674eea0f | 112 | |
afaf0b2f | 113 | struct kvm_x86_ops kvm_x86_ops __read_mostly; |
5fdbf976 | 114 | EXPORT_SYMBOL_GPL(kvm_x86_ops); |
97896d04 | 115 | |
893590c7 | 116 | static bool __read_mostly ignore_msrs = 0; |
476bc001 | 117 | module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR); |
ed85c068 | 118 | |
fab0aa3b EM |
119 | static bool __read_mostly report_ignored_msrs = true; |
120 | module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR); | |
121 | ||
4c27625b | 122 | unsigned int min_timer_period_us = 200; |
9ed96e87 MT |
123 | module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR); |
124 | ||
630994b3 MT |
125 | static bool __read_mostly kvmclock_periodic_sync = true; |
126 | module_param(kvmclock_periodic_sync, bool, S_IRUGO); | |
127 | ||
893590c7 | 128 | bool __read_mostly kvm_has_tsc_control; |
92a1f12d | 129 | EXPORT_SYMBOL_GPL(kvm_has_tsc_control); |
893590c7 | 130 | u32 __read_mostly kvm_max_guest_tsc_khz; |
92a1f12d | 131 | EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz); |
bc9b961b HZ |
132 | u8 __read_mostly kvm_tsc_scaling_ratio_frac_bits; |
133 | EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits); | |
134 | u64 __read_mostly kvm_max_tsc_scaling_ratio; | |
135 | EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio); | |
64672c95 YJ |
136 | u64 __read_mostly kvm_default_tsc_scaling_ratio; |
137 | EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio); | |
92a1f12d | 138 | |
cc578287 | 139 | /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */ |
893590c7 | 140 | static u32 __read_mostly tsc_tolerance_ppm = 250; |
cc578287 ZA |
141 | module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR); |
142 | ||
c3941d9e SC |
143 | /* |
144 | * lapic timer advance (tscdeadline mode only) in nanoseconds. '-1' enables | |
145 | * adaptive tuning starting from default advancment of 1000ns. '0' disables | |
146 | * advancement entirely. Any other value is used as-is and disables adaptive | |
147 | * tuning, i.e. allows priveleged userspace to set an exact advancement time. | |
148 | */ | |
149 | static int __read_mostly lapic_timer_advance_ns = -1; | |
0e6edceb | 150 | module_param(lapic_timer_advance_ns, int, S_IRUGO | S_IWUSR); |
d0659d94 | 151 | |
52004014 FW |
152 | static bool __read_mostly vector_hashing = true; |
153 | module_param(vector_hashing, bool, S_IRUGO); | |
154 | ||
c4ae60e4 LA |
155 | bool __read_mostly enable_vmware_backdoor = false; |
156 | module_param(enable_vmware_backdoor, bool, S_IRUGO); | |
157 | EXPORT_SYMBOL_GPL(enable_vmware_backdoor); | |
158 | ||
6c86eedc WL |
159 | static bool __read_mostly force_emulation_prefix = false; |
160 | module_param(force_emulation_prefix, bool, S_IRUGO); | |
161 | ||
0c5f81da WL |
162 | int __read_mostly pi_inject_timer = -1; |
163 | module_param(pi_inject_timer, bint, S_IRUGO | S_IWUSR); | |
164 | ||
18863bdd AK |
165 | #define KVM_NR_SHARED_MSRS 16 |
166 | ||
167 | struct kvm_shared_msrs_global { | |
168 | int nr; | |
2bf78fa7 | 169 | u32 msrs[KVM_NR_SHARED_MSRS]; |
18863bdd AK |
170 | }; |
171 | ||
172 | struct kvm_shared_msrs { | |
173 | struct user_return_notifier urn; | |
174 | bool registered; | |
2bf78fa7 SY |
175 | struct kvm_shared_msr_values { |
176 | u64 host; | |
177 | u64 curr; | |
178 | } values[KVM_NR_SHARED_MSRS]; | |
18863bdd AK |
179 | }; |
180 | ||
181 | static struct kvm_shared_msrs_global __read_mostly shared_msrs_global; | |
013f6a5d | 182 | static struct kvm_shared_msrs __percpu *shared_msrs; |
18863bdd | 183 | |
cfc48181 SC |
184 | #define KVM_SUPPORTED_XCR0 (XFEATURE_MASK_FP | XFEATURE_MASK_SSE \ |
185 | | XFEATURE_MASK_YMM | XFEATURE_MASK_BNDREGS \ | |
186 | | XFEATURE_MASK_BNDCSR | XFEATURE_MASK_AVX512 \ | |
187 | | XFEATURE_MASK_PKRU) | |
188 | ||
91661989 SC |
189 | u64 __read_mostly host_efer; |
190 | EXPORT_SYMBOL_GPL(host_efer); | |
191 | ||
139a12cf | 192 | static u64 __read_mostly host_xss; |
408e9a31 PB |
193 | u64 __read_mostly supported_xss; |
194 | EXPORT_SYMBOL_GPL(supported_xss); | |
139a12cf | 195 | |
417bc304 | 196 | struct kvm_stats_debugfs_item debugfs_entries[] = { |
ba1389b7 AK |
197 | { "pf_fixed", VCPU_STAT(pf_fixed) }, |
198 | { "pf_guest", VCPU_STAT(pf_guest) }, | |
199 | { "tlb_flush", VCPU_STAT(tlb_flush) }, | |
200 | { "invlpg", VCPU_STAT(invlpg) }, | |
201 | { "exits", VCPU_STAT(exits) }, | |
202 | { "io_exits", VCPU_STAT(io_exits) }, | |
203 | { "mmio_exits", VCPU_STAT(mmio_exits) }, | |
204 | { "signal_exits", VCPU_STAT(signal_exits) }, | |
205 | { "irq_window", VCPU_STAT(irq_window_exits) }, | |
f08864b4 | 206 | { "nmi_window", VCPU_STAT(nmi_window_exits) }, |
ba1389b7 | 207 | { "halt_exits", VCPU_STAT(halt_exits) }, |
f7819512 | 208 | { "halt_successful_poll", VCPU_STAT(halt_successful_poll) }, |
62bea5bf | 209 | { "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) }, |
3491caf2 | 210 | { "halt_poll_invalid", VCPU_STAT(halt_poll_invalid) }, |
ba1389b7 | 211 | { "halt_wakeup", VCPU_STAT(halt_wakeup) }, |
f11c3a8d | 212 | { "hypercalls", VCPU_STAT(hypercalls) }, |
ba1389b7 AK |
213 | { "request_irq", VCPU_STAT(request_irq_exits) }, |
214 | { "irq_exits", VCPU_STAT(irq_exits) }, | |
215 | { "host_state_reload", VCPU_STAT(host_state_reload) }, | |
ba1389b7 AK |
216 | { "fpu_reload", VCPU_STAT(fpu_reload) }, |
217 | { "insn_emulation", VCPU_STAT(insn_emulation) }, | |
218 | { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) }, | |
fa89a817 | 219 | { "irq_injections", VCPU_STAT(irq_injections) }, |
c4abb7c9 | 220 | { "nmi_injections", VCPU_STAT(nmi_injections) }, |
0f1e261e | 221 | { "req_event", VCPU_STAT(req_event) }, |
c595ceee | 222 | { "l1d_flush", VCPU_STAT(l1d_flush) }, |
4cee5764 AK |
223 | { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) }, |
224 | { "mmu_pte_write", VM_STAT(mmu_pte_write) }, | |
225 | { "mmu_pte_updated", VM_STAT(mmu_pte_updated) }, | |
226 | { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) }, | |
227 | { "mmu_flooded", VM_STAT(mmu_flooded) }, | |
228 | { "mmu_recycled", VM_STAT(mmu_recycled) }, | |
dfc5aa00 | 229 | { "mmu_cache_miss", VM_STAT(mmu_cache_miss) }, |
4731d4c7 | 230 | { "mmu_unsync", VM_STAT(mmu_unsync) }, |
0f74a24c | 231 | { "remote_tlb_flush", VM_STAT(remote_tlb_flush) }, |
833b45de | 232 | { "largepages", VM_STAT(lpages, .mode = 0444) }, |
b8e8c830 | 233 | { "nx_largepages_splitted", VM_STAT(nx_lpage_splits, .mode = 0444) }, |
f3414bc7 DM |
234 | { "max_mmu_page_hash_collisions", |
235 | VM_STAT(max_mmu_page_hash_collisions) }, | |
417bc304 HB |
236 | { NULL } |
237 | }; | |
238 | ||
2acf923e | 239 | u64 __read_mostly host_xcr0; |
cfc48181 SC |
240 | u64 __read_mostly supported_xcr0; |
241 | EXPORT_SYMBOL_GPL(supported_xcr0); | |
2acf923e | 242 | |
b666a4b6 MO |
243 | struct kmem_cache *x86_fpu_cache; |
244 | EXPORT_SYMBOL_GPL(x86_fpu_cache); | |
245 | ||
c9b8b07c SC |
246 | static struct kmem_cache *x86_emulator_cache; |
247 | ||
248 | static struct kmem_cache *kvm_alloc_emulator_cache(void) | |
249 | { | |
06add254 SC |
250 | unsigned int useroffset = offsetof(struct x86_emulate_ctxt, src); |
251 | unsigned int size = sizeof(struct x86_emulate_ctxt); | |
252 | ||
253 | return kmem_cache_create_usercopy("x86_emulator", size, | |
c9b8b07c | 254 | __alignof__(struct x86_emulate_ctxt), |
06add254 SC |
255 | SLAB_ACCOUNT, useroffset, |
256 | size - useroffset, NULL); | |
c9b8b07c SC |
257 | } |
258 | ||
b6785def | 259 | static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt); |
d6aa1000 | 260 | |
af585b92 GN |
261 | static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu) |
262 | { | |
263 | int i; | |
264 | for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++) | |
265 | vcpu->arch.apf.gfns[i] = ~0; | |
266 | } | |
267 | ||
18863bdd AK |
268 | static void kvm_on_user_return(struct user_return_notifier *urn) |
269 | { | |
270 | unsigned slot; | |
18863bdd AK |
271 | struct kvm_shared_msrs *locals |
272 | = container_of(urn, struct kvm_shared_msrs, urn); | |
2bf78fa7 | 273 | struct kvm_shared_msr_values *values; |
1650b4eb IA |
274 | unsigned long flags; |
275 | ||
276 | /* | |
277 | * Disabling irqs at this point since the following code could be | |
278 | * interrupted and executed through kvm_arch_hardware_disable() | |
279 | */ | |
280 | local_irq_save(flags); | |
281 | if (locals->registered) { | |
282 | locals->registered = false; | |
283 | user_return_notifier_unregister(urn); | |
284 | } | |
285 | local_irq_restore(flags); | |
18863bdd | 286 | for (slot = 0; slot < shared_msrs_global.nr; ++slot) { |
2bf78fa7 SY |
287 | values = &locals->values[slot]; |
288 | if (values->host != values->curr) { | |
289 | wrmsrl(shared_msrs_global.msrs[slot], values->host); | |
290 | values->curr = values->host; | |
18863bdd AK |
291 | } |
292 | } | |
18863bdd AK |
293 | } |
294 | ||
2bf78fa7 SY |
295 | void kvm_define_shared_msr(unsigned slot, u32 msr) |
296 | { | |
0123be42 | 297 | BUG_ON(slot >= KVM_NR_SHARED_MSRS); |
c847fe88 | 298 | shared_msrs_global.msrs[slot] = msr; |
18863bdd AK |
299 | if (slot >= shared_msrs_global.nr) |
300 | shared_msrs_global.nr = slot + 1; | |
18863bdd AK |
301 | } |
302 | EXPORT_SYMBOL_GPL(kvm_define_shared_msr); | |
303 | ||
304 | static void kvm_shared_msr_cpu_online(void) | |
305 | { | |
05c19c2f SC |
306 | unsigned int cpu = smp_processor_id(); |
307 | struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu); | |
308 | u64 value; | |
309 | int i; | |
18863bdd | 310 | |
05c19c2f SC |
311 | for (i = 0; i < shared_msrs_global.nr; ++i) { |
312 | rdmsrl_safe(shared_msrs_global.msrs[i], &value); | |
313 | smsr->values[i].host = value; | |
314 | smsr->values[i].curr = value; | |
315 | } | |
18863bdd AK |
316 | } |
317 | ||
8b3c3104 | 318 | int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask) |
18863bdd | 319 | { |
013f6a5d MT |
320 | unsigned int cpu = smp_processor_id(); |
321 | struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu); | |
8b3c3104 | 322 | int err; |
18863bdd | 323 | |
de1fca5d PB |
324 | value = (value & mask) | (smsr->values[slot].host & ~mask); |
325 | if (value == smsr->values[slot].curr) | |
8b3c3104 | 326 | return 0; |
8b3c3104 AH |
327 | err = wrmsrl_safe(shared_msrs_global.msrs[slot], value); |
328 | if (err) | |
329 | return 1; | |
330 | ||
de1fca5d | 331 | smsr->values[slot].curr = value; |
18863bdd AK |
332 | if (!smsr->registered) { |
333 | smsr->urn.on_user_return = kvm_on_user_return; | |
334 | user_return_notifier_register(&smsr->urn); | |
335 | smsr->registered = true; | |
336 | } | |
8b3c3104 | 337 | return 0; |
18863bdd AK |
338 | } |
339 | EXPORT_SYMBOL_GPL(kvm_set_shared_msr); | |
340 | ||
13a34e06 | 341 | static void drop_user_return_notifiers(void) |
3548bab5 | 342 | { |
013f6a5d MT |
343 | unsigned int cpu = smp_processor_id(); |
344 | struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu); | |
3548bab5 AK |
345 | |
346 | if (smsr->registered) | |
347 | kvm_on_user_return(&smsr->urn); | |
348 | } | |
349 | ||
6866b83e CO |
350 | u64 kvm_get_apic_base(struct kvm_vcpu *vcpu) |
351 | { | |
8a5a87d9 | 352 | return vcpu->arch.apic_base; |
6866b83e CO |
353 | } |
354 | EXPORT_SYMBOL_GPL(kvm_get_apic_base); | |
355 | ||
58871649 JM |
356 | enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu) |
357 | { | |
358 | return kvm_apic_mode(kvm_get_apic_base(vcpu)); | |
359 | } | |
360 | EXPORT_SYMBOL_GPL(kvm_get_apic_mode); | |
361 | ||
58cb628d JK |
362 | int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info) |
363 | { | |
58871649 JM |
364 | enum lapic_mode old_mode = kvm_get_apic_mode(vcpu); |
365 | enum lapic_mode new_mode = kvm_apic_mode(msr_info->data); | |
d6321d49 RK |
366 | u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) | 0x2ff | |
367 | (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE); | |
58cb628d | 368 | |
58871649 | 369 | if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID) |
58cb628d | 370 | return 1; |
58871649 JM |
371 | if (!msr_info->host_initiated) { |
372 | if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC) | |
373 | return 1; | |
374 | if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC) | |
375 | return 1; | |
376 | } | |
58cb628d JK |
377 | |
378 | kvm_lapic_set_base(vcpu, msr_info->data); | |
4abaffce | 379 | kvm_recalculate_apic_map(vcpu->kvm); |
58cb628d | 380 | return 0; |
6866b83e CO |
381 | } |
382 | EXPORT_SYMBOL_GPL(kvm_set_apic_base); | |
383 | ||
2605fc21 | 384 | asmlinkage __visible void kvm_spurious_fault(void) |
e3ba45b8 GL |
385 | { |
386 | /* Fault while not rebooting. We want the trace. */ | |
b4fdcf60 | 387 | BUG_ON(!kvm_rebooting); |
e3ba45b8 GL |
388 | } |
389 | EXPORT_SYMBOL_GPL(kvm_spurious_fault); | |
390 | ||
3fd28fce ED |
391 | #define EXCPT_BENIGN 0 |
392 | #define EXCPT_CONTRIBUTORY 1 | |
393 | #define EXCPT_PF 2 | |
394 | ||
395 | static int exception_class(int vector) | |
396 | { | |
397 | switch (vector) { | |
398 | case PF_VECTOR: | |
399 | return EXCPT_PF; | |
400 | case DE_VECTOR: | |
401 | case TS_VECTOR: | |
402 | case NP_VECTOR: | |
403 | case SS_VECTOR: | |
404 | case GP_VECTOR: | |
405 | return EXCPT_CONTRIBUTORY; | |
406 | default: | |
407 | break; | |
408 | } | |
409 | return EXCPT_BENIGN; | |
410 | } | |
411 | ||
d6e8c854 NA |
412 | #define EXCPT_FAULT 0 |
413 | #define EXCPT_TRAP 1 | |
414 | #define EXCPT_ABORT 2 | |
415 | #define EXCPT_INTERRUPT 3 | |
416 | ||
417 | static int exception_type(int vector) | |
418 | { | |
419 | unsigned int mask; | |
420 | ||
421 | if (WARN_ON(vector > 31 || vector == NMI_VECTOR)) | |
422 | return EXCPT_INTERRUPT; | |
423 | ||
424 | mask = 1 << vector; | |
425 | ||
426 | /* #DB is trap, as instruction watchpoints are handled elsewhere */ | |
427 | if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR))) | |
428 | return EXCPT_TRAP; | |
429 | ||
430 | if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR))) | |
431 | return EXCPT_ABORT; | |
432 | ||
433 | /* Reserved exceptions will result in fault */ | |
434 | return EXCPT_FAULT; | |
435 | } | |
436 | ||
da998b46 JM |
437 | void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu) |
438 | { | |
439 | unsigned nr = vcpu->arch.exception.nr; | |
440 | bool has_payload = vcpu->arch.exception.has_payload; | |
441 | unsigned long payload = vcpu->arch.exception.payload; | |
442 | ||
443 | if (!has_payload) | |
444 | return; | |
445 | ||
446 | switch (nr) { | |
f10c729f JM |
447 | case DB_VECTOR: |
448 | /* | |
449 | * "Certain debug exceptions may clear bit 0-3. The | |
450 | * remaining contents of the DR6 register are never | |
451 | * cleared by the processor". | |
452 | */ | |
453 | vcpu->arch.dr6 &= ~DR_TRAP_BITS; | |
454 | /* | |
455 | * DR6.RTM is set by all #DB exceptions that don't clear it. | |
456 | */ | |
457 | vcpu->arch.dr6 |= DR6_RTM; | |
458 | vcpu->arch.dr6 |= payload; | |
459 | /* | |
460 | * Bit 16 should be set in the payload whenever the #DB | |
461 | * exception should clear DR6.RTM. This makes the payload | |
462 | * compatible with the pending debug exceptions under VMX. | |
463 | * Though not currently documented in the SDM, this also | |
464 | * makes the payload compatible with the exit qualification | |
465 | * for #DB exceptions under VMX. | |
466 | */ | |
467 | vcpu->arch.dr6 ^= payload & DR6_RTM; | |
307f1cfa OU |
468 | |
469 | /* | |
470 | * The #DB payload is defined as compatible with the 'pending | |
471 | * debug exceptions' field under VMX, not DR6. While bit 12 is | |
472 | * defined in the 'pending debug exceptions' field (enabled | |
473 | * breakpoint), it is reserved and must be zero in DR6. | |
474 | */ | |
475 | vcpu->arch.dr6 &= ~BIT(12); | |
f10c729f | 476 | break; |
da998b46 JM |
477 | case PF_VECTOR: |
478 | vcpu->arch.cr2 = payload; | |
479 | break; | |
480 | } | |
481 | ||
482 | vcpu->arch.exception.has_payload = false; | |
483 | vcpu->arch.exception.payload = 0; | |
484 | } | |
485 | EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload); | |
486 | ||
3fd28fce | 487 | static void kvm_multiple_exception(struct kvm_vcpu *vcpu, |
ce7ddec4 | 488 | unsigned nr, bool has_error, u32 error_code, |
91e86d22 | 489 | bool has_payload, unsigned long payload, bool reinject) |
3fd28fce ED |
490 | { |
491 | u32 prev_nr; | |
492 | int class1, class2; | |
493 | ||
3842d135 AK |
494 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
495 | ||
664f8e26 | 496 | if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) { |
3fd28fce | 497 | queue: |
3ffb2468 NA |
498 | if (has_error && !is_protmode(vcpu)) |
499 | has_error = false; | |
664f8e26 WL |
500 | if (reinject) { |
501 | /* | |
502 | * On vmentry, vcpu->arch.exception.pending is only | |
503 | * true if an event injection was blocked by | |
504 | * nested_run_pending. In that case, however, | |
505 | * vcpu_enter_guest requests an immediate exit, | |
506 | * and the guest shouldn't proceed far enough to | |
507 | * need reinjection. | |
508 | */ | |
509 | WARN_ON_ONCE(vcpu->arch.exception.pending); | |
510 | vcpu->arch.exception.injected = true; | |
91e86d22 JM |
511 | if (WARN_ON_ONCE(has_payload)) { |
512 | /* | |
513 | * A reinjected event has already | |
514 | * delivered its payload. | |
515 | */ | |
516 | has_payload = false; | |
517 | payload = 0; | |
518 | } | |
664f8e26 WL |
519 | } else { |
520 | vcpu->arch.exception.pending = true; | |
521 | vcpu->arch.exception.injected = false; | |
522 | } | |
3fd28fce ED |
523 | vcpu->arch.exception.has_error_code = has_error; |
524 | vcpu->arch.exception.nr = nr; | |
525 | vcpu->arch.exception.error_code = error_code; | |
91e86d22 JM |
526 | vcpu->arch.exception.has_payload = has_payload; |
527 | vcpu->arch.exception.payload = payload; | |
a06230b6 | 528 | if (!is_guest_mode(vcpu)) |
da998b46 | 529 | kvm_deliver_exception_payload(vcpu); |
3fd28fce ED |
530 | return; |
531 | } | |
532 | ||
533 | /* to check exception */ | |
534 | prev_nr = vcpu->arch.exception.nr; | |
535 | if (prev_nr == DF_VECTOR) { | |
536 | /* triple fault -> shutdown */ | |
a8eeb04a | 537 | kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu); |
3fd28fce ED |
538 | return; |
539 | } | |
540 | class1 = exception_class(prev_nr); | |
541 | class2 = exception_class(nr); | |
542 | if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY) | |
543 | || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) { | |
664f8e26 WL |
544 | /* |
545 | * Generate double fault per SDM Table 5-5. Set | |
546 | * exception.pending = true so that the double fault | |
547 | * can trigger a nested vmexit. | |
548 | */ | |
3fd28fce | 549 | vcpu->arch.exception.pending = true; |
664f8e26 | 550 | vcpu->arch.exception.injected = false; |
3fd28fce ED |
551 | vcpu->arch.exception.has_error_code = true; |
552 | vcpu->arch.exception.nr = DF_VECTOR; | |
553 | vcpu->arch.exception.error_code = 0; | |
c851436a JM |
554 | vcpu->arch.exception.has_payload = false; |
555 | vcpu->arch.exception.payload = 0; | |
3fd28fce ED |
556 | } else |
557 | /* replace previous exception with a new one in a hope | |
558 | that instruction re-execution will regenerate lost | |
559 | exception */ | |
560 | goto queue; | |
561 | } | |
562 | ||
298101da AK |
563 | void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr) |
564 | { | |
91e86d22 | 565 | kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false); |
298101da AK |
566 | } |
567 | EXPORT_SYMBOL_GPL(kvm_queue_exception); | |
568 | ||
ce7ddec4 JR |
569 | void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr) |
570 | { | |
91e86d22 | 571 | kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true); |
ce7ddec4 JR |
572 | } |
573 | EXPORT_SYMBOL_GPL(kvm_requeue_exception); | |
574 | ||
4d5523cf PB |
575 | void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr, |
576 | unsigned long payload) | |
f10c729f JM |
577 | { |
578 | kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false); | |
579 | } | |
4d5523cf | 580 | EXPORT_SYMBOL_GPL(kvm_queue_exception_p); |
f10c729f | 581 | |
da998b46 JM |
582 | static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr, |
583 | u32 error_code, unsigned long payload) | |
584 | { | |
585 | kvm_multiple_exception(vcpu, nr, true, error_code, | |
586 | true, payload, false); | |
587 | } | |
588 | ||
6affcbed | 589 | int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err) |
c3c91fee | 590 | { |
db8fcefa AP |
591 | if (err) |
592 | kvm_inject_gp(vcpu, 0); | |
593 | else | |
6affcbed KH |
594 | return kvm_skip_emulated_instruction(vcpu); |
595 | ||
596 | return 1; | |
db8fcefa AP |
597 | } |
598 | EXPORT_SYMBOL_GPL(kvm_complete_insn_gp); | |
8df25a32 | 599 | |
6389ee94 | 600 | void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault) |
c3c91fee AK |
601 | { |
602 | ++vcpu->stat.pf_guest; | |
adfe20fb WL |
603 | vcpu->arch.exception.nested_apf = |
604 | is_guest_mode(vcpu) && fault->async_page_fault; | |
da998b46 | 605 | if (vcpu->arch.exception.nested_apf) { |
adfe20fb | 606 | vcpu->arch.apf.nested_apf_token = fault->address; |
da998b46 JM |
607 | kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code); |
608 | } else { | |
609 | kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code, | |
610 | fault->address); | |
611 | } | |
c3c91fee | 612 | } |
27d6c865 | 613 | EXPORT_SYMBOL_GPL(kvm_inject_page_fault); |
c3c91fee | 614 | |
ef54bcfe | 615 | static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault) |
d4f8cf66 | 616 | { |
6389ee94 AK |
617 | if (mmu_is_nested(vcpu) && !fault->nested_page_fault) |
618 | vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault); | |
d4f8cf66 | 619 | else |
44dd3ffa | 620 | vcpu->arch.mmu->inject_page_fault(vcpu, fault); |
ef54bcfe PB |
621 | |
622 | return fault->nested_page_fault; | |
d4f8cf66 JR |
623 | } |
624 | ||
3419ffc8 SY |
625 | void kvm_inject_nmi(struct kvm_vcpu *vcpu) |
626 | { | |
7460fb4a AK |
627 | atomic_inc(&vcpu->arch.nmi_queued); |
628 | kvm_make_request(KVM_REQ_NMI, vcpu); | |
3419ffc8 SY |
629 | } |
630 | EXPORT_SYMBOL_GPL(kvm_inject_nmi); | |
631 | ||
298101da AK |
632 | void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code) |
633 | { | |
91e86d22 | 634 | kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false); |
298101da AK |
635 | } |
636 | EXPORT_SYMBOL_GPL(kvm_queue_exception_e); | |
637 | ||
ce7ddec4 JR |
638 | void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code) |
639 | { | |
91e86d22 | 640 | kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true); |
ce7ddec4 JR |
641 | } |
642 | EXPORT_SYMBOL_GPL(kvm_requeue_exception_e); | |
643 | ||
0a79b009 AK |
644 | /* |
645 | * Checks if cpl <= required_cpl; if true, return true. Otherwise queue | |
646 | * a #GP and return false. | |
647 | */ | |
648 | bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl) | |
298101da | 649 | { |
afaf0b2f | 650 | if (kvm_x86_ops.get_cpl(vcpu) <= required_cpl) |
0a79b009 AK |
651 | return true; |
652 | kvm_queue_exception_e(vcpu, GP_VECTOR, 0); | |
653 | return false; | |
298101da | 654 | } |
0a79b009 | 655 | EXPORT_SYMBOL_GPL(kvm_require_cpl); |
298101da | 656 | |
16f8a6f9 NA |
657 | bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr) |
658 | { | |
659 | if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE)) | |
660 | return true; | |
661 | ||
662 | kvm_queue_exception(vcpu, UD_VECTOR); | |
663 | return false; | |
664 | } | |
665 | EXPORT_SYMBOL_GPL(kvm_require_dr); | |
666 | ||
ec92fe44 JR |
667 | /* |
668 | * This function will be used to read from the physical memory of the currently | |
54bf36aa | 669 | * running guest. The difference to kvm_vcpu_read_guest_page is that this function |
ec92fe44 JR |
670 | * can read from guest physical or from the guest's guest physical memory. |
671 | */ | |
672 | int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, | |
673 | gfn_t ngfn, void *data, int offset, int len, | |
674 | u32 access) | |
675 | { | |
54987b7a | 676 | struct x86_exception exception; |
ec92fe44 JR |
677 | gfn_t real_gfn; |
678 | gpa_t ngpa; | |
679 | ||
680 | ngpa = gfn_to_gpa(ngfn); | |
54987b7a | 681 | real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception); |
ec92fe44 JR |
682 | if (real_gfn == UNMAPPED_GVA) |
683 | return -EFAULT; | |
684 | ||
685 | real_gfn = gpa_to_gfn(real_gfn); | |
686 | ||
54bf36aa | 687 | return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len); |
ec92fe44 JR |
688 | } |
689 | EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu); | |
690 | ||
69b0049a | 691 | static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, |
3d06b8bf JR |
692 | void *data, int offset, int len, u32 access) |
693 | { | |
694 | return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn, | |
695 | data, offset, len, access); | |
696 | } | |
697 | ||
16cfacc8 SC |
698 | static inline u64 pdptr_rsvd_bits(struct kvm_vcpu *vcpu) |
699 | { | |
700 | return rsvd_bits(cpuid_maxphyaddr(vcpu), 63) | rsvd_bits(5, 8) | | |
701 | rsvd_bits(1, 2); | |
702 | } | |
703 | ||
a03490ed | 704 | /* |
16cfacc8 | 705 | * Load the pae pdptrs. Return 1 if they are all valid, 0 otherwise. |
a03490ed | 706 | */ |
ff03a073 | 707 | int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3) |
a03490ed CO |
708 | { |
709 | gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT; | |
710 | unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2; | |
711 | int i; | |
712 | int ret; | |
ff03a073 | 713 | u64 pdpte[ARRAY_SIZE(mmu->pdptrs)]; |
a03490ed | 714 | |
ff03a073 JR |
715 | ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte, |
716 | offset * sizeof(u64), sizeof(pdpte), | |
717 | PFERR_USER_MASK|PFERR_WRITE_MASK); | |
a03490ed CO |
718 | if (ret < 0) { |
719 | ret = 0; | |
720 | goto out; | |
721 | } | |
722 | for (i = 0; i < ARRAY_SIZE(pdpte); ++i) { | |
812f30b2 | 723 | if ((pdpte[i] & PT_PRESENT_MASK) && |
16cfacc8 | 724 | (pdpte[i] & pdptr_rsvd_bits(vcpu))) { |
a03490ed CO |
725 | ret = 0; |
726 | goto out; | |
727 | } | |
728 | } | |
729 | ret = 1; | |
730 | ||
ff03a073 | 731 | memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs)); |
cb3c1e2f SC |
732 | kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR); |
733 | ||
a03490ed | 734 | out: |
a03490ed CO |
735 | |
736 | return ret; | |
737 | } | |
cc4b6871 | 738 | EXPORT_SYMBOL_GPL(load_pdptrs); |
a03490ed | 739 | |
9ed38ffa | 740 | bool pdptrs_changed(struct kvm_vcpu *vcpu) |
d835dfec | 741 | { |
ff03a073 | 742 | u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)]; |
3d06b8bf JR |
743 | int offset; |
744 | gfn_t gfn; | |
d835dfec AK |
745 | int r; |
746 | ||
bf03d4f9 | 747 | if (!is_pae_paging(vcpu)) |
d835dfec AK |
748 | return false; |
749 | ||
cb3c1e2f | 750 | if (!kvm_register_is_available(vcpu, VCPU_EXREG_PDPTR)) |
6de4f3ad AK |
751 | return true; |
752 | ||
a512177e PB |
753 | gfn = (kvm_read_cr3(vcpu) & 0xffffffe0ul) >> PAGE_SHIFT; |
754 | offset = (kvm_read_cr3(vcpu) & 0xffffffe0ul) & (PAGE_SIZE - 1); | |
3d06b8bf JR |
755 | r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte), |
756 | PFERR_USER_MASK | PFERR_WRITE_MASK); | |
d835dfec | 757 | if (r < 0) |
7f7f0d9c | 758 | return true; |
d835dfec | 759 | |
7f7f0d9c | 760 | return memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0; |
d835dfec | 761 | } |
9ed38ffa | 762 | EXPORT_SYMBOL_GPL(pdptrs_changed); |
d835dfec | 763 | |
49a9b07e | 764 | int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0) |
a03490ed | 765 | { |
aad82703 | 766 | unsigned long old_cr0 = kvm_read_cr0(vcpu); |
d81135a5 | 767 | unsigned long update_bits = X86_CR0_PG | X86_CR0_WP; |
aad82703 | 768 | |
f9a48e6a AK |
769 | cr0 |= X86_CR0_ET; |
770 | ||
ab344828 | 771 | #ifdef CONFIG_X86_64 |
0f12244f GN |
772 | if (cr0 & 0xffffffff00000000UL) |
773 | return 1; | |
ab344828 GN |
774 | #endif |
775 | ||
776 | cr0 &= ~CR0_RESERVED_BITS; | |
a03490ed | 777 | |
0f12244f GN |
778 | if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) |
779 | return 1; | |
a03490ed | 780 | |
0f12244f GN |
781 | if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) |
782 | return 1; | |
a03490ed CO |
783 | |
784 | if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) { | |
785 | #ifdef CONFIG_X86_64 | |
f6801dff | 786 | if ((vcpu->arch.efer & EFER_LME)) { |
a03490ed CO |
787 | int cs_db, cs_l; |
788 | ||
0f12244f GN |
789 | if (!is_pae(vcpu)) |
790 | return 1; | |
afaf0b2f | 791 | kvm_x86_ops.get_cs_db_l_bits(vcpu, &cs_db, &cs_l); |
0f12244f GN |
792 | if (cs_l) |
793 | return 1; | |
a03490ed CO |
794 | } else |
795 | #endif | |
ff03a073 | 796 | if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu, |
9f8fe504 | 797 | kvm_read_cr3(vcpu))) |
0f12244f | 798 | return 1; |
a03490ed CO |
799 | } |
800 | ||
ad756a16 MJ |
801 | if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE)) |
802 | return 1; | |
803 | ||
afaf0b2f | 804 | kvm_x86_ops.set_cr0(vcpu, cr0); |
a03490ed | 805 | |
d170c419 | 806 | if ((cr0 ^ old_cr0) & X86_CR0_PG) { |
e5f3f027 | 807 | kvm_clear_async_pf_completion_queue(vcpu); |
d170c419 LJ |
808 | kvm_async_pf_hash_reset(vcpu); |
809 | } | |
e5f3f027 | 810 | |
aad82703 SY |
811 | if ((cr0 ^ old_cr0) & update_bits) |
812 | kvm_mmu_reset_context(vcpu); | |
b18d5431 | 813 | |
879ae188 LE |
814 | if (((cr0 ^ old_cr0) & X86_CR0_CD) && |
815 | kvm_arch_has_noncoherent_dma(vcpu->kvm) && | |
816 | !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED)) | |
b18d5431 XG |
817 | kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL); |
818 | ||
0f12244f GN |
819 | return 0; |
820 | } | |
2d3ad1f4 | 821 | EXPORT_SYMBOL_GPL(kvm_set_cr0); |
a03490ed | 822 | |
2d3ad1f4 | 823 | void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw) |
a03490ed | 824 | { |
49a9b07e | 825 | (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f)); |
a03490ed | 826 | } |
2d3ad1f4 | 827 | EXPORT_SYMBOL_GPL(kvm_lmsw); |
a03490ed | 828 | |
139a12cf | 829 | void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu) |
42bdf991 | 830 | { |
139a12cf AL |
831 | if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) { |
832 | ||
833 | if (vcpu->arch.xcr0 != host_xcr0) | |
834 | xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0); | |
835 | ||
836 | if (vcpu->arch.xsaves_enabled && | |
837 | vcpu->arch.ia32_xss != host_xss) | |
838 | wrmsrl(MSR_IA32_XSS, vcpu->arch.ia32_xss); | |
839 | } | |
37486135 BM |
840 | |
841 | if (static_cpu_has(X86_FEATURE_PKU) && | |
842 | (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) || | |
843 | (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU)) && | |
844 | vcpu->arch.pkru != vcpu->arch.host_pkru) | |
845 | __write_pkru(vcpu->arch.pkru); | |
42bdf991 | 846 | } |
139a12cf | 847 | EXPORT_SYMBOL_GPL(kvm_load_guest_xsave_state); |
42bdf991 | 848 | |
139a12cf | 849 | void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu) |
42bdf991 | 850 | { |
37486135 BM |
851 | if (static_cpu_has(X86_FEATURE_PKU) && |
852 | (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) || | |
853 | (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU))) { | |
854 | vcpu->arch.pkru = rdpkru(); | |
855 | if (vcpu->arch.pkru != vcpu->arch.host_pkru) | |
856 | __write_pkru(vcpu->arch.host_pkru); | |
857 | } | |
858 | ||
139a12cf AL |
859 | if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) { |
860 | ||
861 | if (vcpu->arch.xcr0 != host_xcr0) | |
862 | xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0); | |
863 | ||
864 | if (vcpu->arch.xsaves_enabled && | |
865 | vcpu->arch.ia32_xss != host_xss) | |
866 | wrmsrl(MSR_IA32_XSS, host_xss); | |
867 | } | |
868 | ||
42bdf991 | 869 | } |
139a12cf | 870 | EXPORT_SYMBOL_GPL(kvm_load_host_xsave_state); |
42bdf991 | 871 | |
69b0049a | 872 | static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr) |
2acf923e | 873 | { |
56c103ec LJ |
874 | u64 xcr0 = xcr; |
875 | u64 old_xcr0 = vcpu->arch.xcr0; | |
46c34cb0 | 876 | u64 valid_bits; |
2acf923e DC |
877 | |
878 | /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */ | |
879 | if (index != XCR_XFEATURE_ENABLED_MASK) | |
880 | return 1; | |
d91cab78 | 881 | if (!(xcr0 & XFEATURE_MASK_FP)) |
2acf923e | 882 | return 1; |
d91cab78 | 883 | if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE)) |
2acf923e | 884 | return 1; |
46c34cb0 PB |
885 | |
886 | /* | |
887 | * Do not allow the guest to set bits that we do not support | |
888 | * saving. However, xcr0 bit 0 is always set, even if the | |
889 | * emulated CPU does not support XSAVE (see fx_init). | |
890 | */ | |
d91cab78 | 891 | valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP; |
46c34cb0 | 892 | if (xcr0 & ~valid_bits) |
2acf923e | 893 | return 1; |
46c34cb0 | 894 | |
d91cab78 DH |
895 | if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) != |
896 | (!(xcr0 & XFEATURE_MASK_BNDCSR))) | |
390bd528 LJ |
897 | return 1; |
898 | ||
d91cab78 DH |
899 | if (xcr0 & XFEATURE_MASK_AVX512) { |
900 | if (!(xcr0 & XFEATURE_MASK_YMM)) | |
612263b3 | 901 | return 1; |
d91cab78 | 902 | if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512) |
612263b3 CP |
903 | return 1; |
904 | } | |
2acf923e | 905 | vcpu->arch.xcr0 = xcr0; |
56c103ec | 906 | |
d91cab78 | 907 | if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND) |
56c103ec | 908 | kvm_update_cpuid(vcpu); |
2acf923e DC |
909 | return 0; |
910 | } | |
911 | ||
912 | int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr) | |
913 | { | |
afaf0b2f | 914 | if (kvm_x86_ops.get_cpl(vcpu) != 0 || |
764bcbc5 | 915 | __kvm_set_xcr(vcpu, index, xcr)) { |
2acf923e DC |
916 | kvm_inject_gp(vcpu, 0); |
917 | return 1; | |
918 | } | |
919 | return 0; | |
920 | } | |
921 | EXPORT_SYMBOL_GPL(kvm_set_xcr); | |
922 | ||
345599f9 SC |
923 | #define __cr4_reserved_bits(__cpu_has, __c) \ |
924 | ({ \ | |
925 | u64 __reserved_bits = CR4_RESERVED_BITS; \ | |
926 | \ | |
927 | if (!__cpu_has(__c, X86_FEATURE_XSAVE)) \ | |
928 | __reserved_bits |= X86_CR4_OSXSAVE; \ | |
929 | if (!__cpu_has(__c, X86_FEATURE_SMEP)) \ | |
930 | __reserved_bits |= X86_CR4_SMEP; \ | |
931 | if (!__cpu_has(__c, X86_FEATURE_SMAP)) \ | |
932 | __reserved_bits |= X86_CR4_SMAP; \ | |
933 | if (!__cpu_has(__c, X86_FEATURE_FSGSBASE)) \ | |
934 | __reserved_bits |= X86_CR4_FSGSBASE; \ | |
935 | if (!__cpu_has(__c, X86_FEATURE_PKU)) \ | |
936 | __reserved_bits |= X86_CR4_PKE; \ | |
937 | if (!__cpu_has(__c, X86_FEATURE_LA57)) \ | |
938 | __reserved_bits |= X86_CR4_LA57; \ | |
d76c7fbc SC |
939 | if (!__cpu_has(__c, X86_FEATURE_UMIP)) \ |
940 | __reserved_bits |= X86_CR4_UMIP; \ | |
345599f9 SC |
941 | __reserved_bits; \ |
942 | }) | |
a03490ed | 943 | |
3ca94192 | 944 | static int kvm_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4) |
a03490ed | 945 | { |
b11306b5 | 946 | if (cr4 & cr4_reserved_bits) |
3ca94192 | 947 | return -EINVAL; |
b9baba86 | 948 | |
345599f9 | 949 | if (cr4 & __cr4_reserved_bits(guest_cpuid_has, vcpu)) |
3ca94192 WL |
950 | return -EINVAL; |
951 | ||
952 | return 0; | |
953 | } | |
954 | ||
955 | int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4) | |
956 | { | |
957 | unsigned long old_cr4 = kvm_read_cr4(vcpu); | |
958 | unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE | | |
959 | X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE; | |
960 | ||
961 | if (kvm_valid_cr4(vcpu, cr4)) | |
ae3e61e1 PB |
962 | return 1; |
963 | ||
a03490ed | 964 | if (is_long_mode(vcpu)) { |
0f12244f GN |
965 | if (!(cr4 & X86_CR4_PAE)) |
966 | return 1; | |
a2edf57f AK |
967 | } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE) |
968 | && ((cr4 ^ old_cr4) & pdptr_bits) | |
9f8fe504 AK |
969 | && !load_pdptrs(vcpu, vcpu->arch.walk_mmu, |
970 | kvm_read_cr3(vcpu))) | |
0f12244f GN |
971 | return 1; |
972 | ||
ad756a16 | 973 | if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) { |
d6321d49 | 974 | if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID)) |
ad756a16 MJ |
975 | return 1; |
976 | ||
977 | /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */ | |
978 | if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu)) | |
979 | return 1; | |
980 | } | |
981 | ||
afaf0b2f | 982 | if (kvm_x86_ops.set_cr4(vcpu, cr4)) |
0f12244f | 983 | return 1; |
a03490ed | 984 | |
ad756a16 MJ |
985 | if (((cr4 ^ old_cr4) & pdptr_bits) || |
986 | (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE))) | |
aad82703 | 987 | kvm_mmu_reset_context(vcpu); |
0f12244f | 988 | |
b9baba86 | 989 | if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE)) |
00b27a3e | 990 | kvm_update_cpuid(vcpu); |
2acf923e | 991 | |
0f12244f GN |
992 | return 0; |
993 | } | |
2d3ad1f4 | 994 | EXPORT_SYMBOL_GPL(kvm_set_cr4); |
a03490ed | 995 | |
2390218b | 996 | int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3) |
a03490ed | 997 | { |
ade61e28 | 998 | bool skip_tlb_flush = false; |
ac146235 | 999 | #ifdef CONFIG_X86_64 |
c19986fe JS |
1000 | bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE); |
1001 | ||
ade61e28 | 1002 | if (pcid_enabled) { |
208320ba JS |
1003 | skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH; |
1004 | cr3 &= ~X86_CR3_PCID_NOFLUSH; | |
ade61e28 | 1005 | } |
ac146235 | 1006 | #endif |
9d88fca7 | 1007 | |
9f8fe504 | 1008 | if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) { |
956bf353 JS |
1009 | if (!skip_tlb_flush) { |
1010 | kvm_mmu_sync_roots(vcpu); | |
ade61e28 | 1011 | kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu); |
956bf353 | 1012 | } |
0f12244f | 1013 | return 0; |
d835dfec AK |
1014 | } |
1015 | ||
d1cd3ce9 | 1016 | if (is_long_mode(vcpu) && |
a780a3ea | 1017 | (cr3 & rsvd_bits(cpuid_maxphyaddr(vcpu), 63))) |
d1cd3ce9 | 1018 | return 1; |
bf03d4f9 PB |
1019 | else if (is_pae_paging(vcpu) && |
1020 | !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3)) | |
346874c9 | 1021 | return 1; |
a03490ed | 1022 | |
ade61e28 | 1023 | kvm_mmu_new_cr3(vcpu, cr3, skip_tlb_flush); |
0f12244f | 1024 | vcpu->arch.cr3 = cr3; |
cb3c1e2f | 1025 | kvm_register_mark_available(vcpu, VCPU_EXREG_CR3); |
7c390d35 | 1026 | |
0f12244f GN |
1027 | return 0; |
1028 | } | |
2d3ad1f4 | 1029 | EXPORT_SYMBOL_GPL(kvm_set_cr3); |
a03490ed | 1030 | |
eea1cff9 | 1031 | int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8) |
a03490ed | 1032 | { |
0f12244f GN |
1033 | if (cr8 & CR8_RESERVED_BITS) |
1034 | return 1; | |
35754c98 | 1035 | if (lapic_in_kernel(vcpu)) |
a03490ed CO |
1036 | kvm_lapic_set_tpr(vcpu, cr8); |
1037 | else | |
ad312c7c | 1038 | vcpu->arch.cr8 = cr8; |
0f12244f GN |
1039 | return 0; |
1040 | } | |
2d3ad1f4 | 1041 | EXPORT_SYMBOL_GPL(kvm_set_cr8); |
a03490ed | 1042 | |
2d3ad1f4 | 1043 | unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu) |
a03490ed | 1044 | { |
35754c98 | 1045 | if (lapic_in_kernel(vcpu)) |
a03490ed CO |
1046 | return kvm_lapic_get_cr8(vcpu); |
1047 | else | |
ad312c7c | 1048 | return vcpu->arch.cr8; |
a03490ed | 1049 | } |
2d3ad1f4 | 1050 | EXPORT_SYMBOL_GPL(kvm_get_cr8); |
a03490ed | 1051 | |
ae561ede NA |
1052 | static void kvm_update_dr0123(struct kvm_vcpu *vcpu) |
1053 | { | |
1054 | int i; | |
1055 | ||
1056 | if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) { | |
1057 | for (i = 0; i < KVM_NR_DB_REGS; i++) | |
1058 | vcpu->arch.eff_db[i] = vcpu->arch.db[i]; | |
1059 | vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD; | |
1060 | } | |
1061 | } | |
1062 | ||
c8639010 JK |
1063 | static void kvm_update_dr7(struct kvm_vcpu *vcpu) |
1064 | { | |
1065 | unsigned long dr7; | |
1066 | ||
1067 | if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) | |
1068 | dr7 = vcpu->arch.guest_debug_dr7; | |
1069 | else | |
1070 | dr7 = vcpu->arch.dr7; | |
afaf0b2f | 1071 | kvm_x86_ops.set_dr7(vcpu, dr7); |
360b948d PB |
1072 | vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED; |
1073 | if (dr7 & DR7_BP_EN_MASK) | |
1074 | vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED; | |
c8639010 JK |
1075 | } |
1076 | ||
6f43ed01 NA |
1077 | static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu) |
1078 | { | |
1079 | u64 fixed = DR6_FIXED_1; | |
1080 | ||
d6321d49 | 1081 | if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM)) |
6f43ed01 NA |
1082 | fixed |= DR6_RTM; |
1083 | return fixed; | |
1084 | } | |
1085 | ||
338dbc97 | 1086 | static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val) |
020df079 | 1087 | { |
ea740059 MP |
1088 | size_t size = ARRAY_SIZE(vcpu->arch.db); |
1089 | ||
020df079 GN |
1090 | switch (dr) { |
1091 | case 0 ... 3: | |
ea740059 | 1092 | vcpu->arch.db[array_index_nospec(dr, size)] = val; |
020df079 GN |
1093 | if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) |
1094 | vcpu->arch.eff_db[dr] = val; | |
1095 | break; | |
1096 | case 4: | |
020df079 GN |
1097 | /* fall through */ |
1098 | case 6: | |
338dbc97 GN |
1099 | if (val & 0xffffffff00000000ULL) |
1100 | return -1; /* #GP */ | |
6f43ed01 | 1101 | vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu); |
020df079 GN |
1102 | break; |
1103 | case 5: | |
020df079 GN |
1104 | /* fall through */ |
1105 | default: /* 7 */ | |
b91991bf | 1106 | if (!kvm_dr7_valid(val)) |
338dbc97 | 1107 | return -1; /* #GP */ |
020df079 | 1108 | vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1; |
c8639010 | 1109 | kvm_update_dr7(vcpu); |
020df079 GN |
1110 | break; |
1111 | } | |
1112 | ||
1113 | return 0; | |
1114 | } | |
338dbc97 GN |
1115 | |
1116 | int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val) | |
1117 | { | |
16f8a6f9 | 1118 | if (__kvm_set_dr(vcpu, dr, val)) { |
338dbc97 | 1119 | kvm_inject_gp(vcpu, 0); |
16f8a6f9 NA |
1120 | return 1; |
1121 | } | |
1122 | return 0; | |
338dbc97 | 1123 | } |
020df079 GN |
1124 | EXPORT_SYMBOL_GPL(kvm_set_dr); |
1125 | ||
16f8a6f9 | 1126 | int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val) |
020df079 | 1127 | { |
ea740059 MP |
1128 | size_t size = ARRAY_SIZE(vcpu->arch.db); |
1129 | ||
020df079 GN |
1130 | switch (dr) { |
1131 | case 0 ... 3: | |
ea740059 | 1132 | *val = vcpu->arch.db[array_index_nospec(dr, size)]; |
020df079 GN |
1133 | break; |
1134 | case 4: | |
020df079 GN |
1135 | /* fall through */ |
1136 | case 6: | |
5679b803 | 1137 | *val = vcpu->arch.dr6; |
020df079 GN |
1138 | break; |
1139 | case 5: | |
020df079 GN |
1140 | /* fall through */ |
1141 | default: /* 7 */ | |
1142 | *val = vcpu->arch.dr7; | |
1143 | break; | |
1144 | } | |
338dbc97 GN |
1145 | return 0; |
1146 | } | |
020df079 GN |
1147 | EXPORT_SYMBOL_GPL(kvm_get_dr); |
1148 | ||
022cd0e8 AK |
1149 | bool kvm_rdpmc(struct kvm_vcpu *vcpu) |
1150 | { | |
de3cd117 | 1151 | u32 ecx = kvm_rcx_read(vcpu); |
022cd0e8 AK |
1152 | u64 data; |
1153 | int err; | |
1154 | ||
c6702c9d | 1155 | err = kvm_pmu_rdpmc(vcpu, ecx, &data); |
022cd0e8 AK |
1156 | if (err) |
1157 | return err; | |
de3cd117 SC |
1158 | kvm_rax_write(vcpu, (u32)data); |
1159 | kvm_rdx_write(vcpu, data >> 32); | |
022cd0e8 AK |
1160 | return err; |
1161 | } | |
1162 | EXPORT_SYMBOL_GPL(kvm_rdpmc); | |
1163 | ||
043405e1 CO |
1164 | /* |
1165 | * List of msr numbers which we expose to userspace through KVM_GET_MSRS | |
1166 | * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST. | |
1167 | * | |
7a5ee6ed CQ |
1168 | * The three MSR lists(msrs_to_save, emulated_msrs, msr_based_features) |
1169 | * extract the supported MSRs from the related const lists. | |
1170 | * msrs_to_save is selected from the msrs_to_save_all to reflect the | |
e3267cbb | 1171 | * capabilities of the host cpu. This capabilities test skips MSRs that are |
7a5ee6ed | 1172 | * kvm-specific. Those are put in emulated_msrs_all; filtering of emulated_msrs |
62ef68bb | 1173 | * may depend on host virtualization features rather than host cpu features. |
043405e1 | 1174 | */ |
e3267cbb | 1175 | |
7a5ee6ed | 1176 | static const u32 msrs_to_save_all[] = { |
043405e1 | 1177 | MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP, |
8c06585d | 1178 | MSR_STAR, |
043405e1 CO |
1179 | #ifdef CONFIG_X86_64 |
1180 | MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR, | |
1181 | #endif | |
b3897a49 | 1182 | MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA, |
32ad73db | 1183 | MSR_IA32_FEAT_CTL, MSR_IA32_BNDCFGS, MSR_TSC_AUX, |
2bdb76c0 | 1184 | MSR_IA32_SPEC_CTRL, |
bf8c55d8 CP |
1185 | MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH, |
1186 | MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK, | |
1187 | MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B, | |
1188 | MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B, | |
1189 | MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B, | |
1190 | MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B, | |
6e3ba4ab TX |
1191 | MSR_IA32_UMWAIT_CONTROL, |
1192 | ||
e2ada66e JM |
1193 | MSR_ARCH_PERFMON_FIXED_CTR0, MSR_ARCH_PERFMON_FIXED_CTR1, |
1194 | MSR_ARCH_PERFMON_FIXED_CTR0 + 2, MSR_ARCH_PERFMON_FIXED_CTR0 + 3, | |
1195 | MSR_CORE_PERF_FIXED_CTR_CTRL, MSR_CORE_PERF_GLOBAL_STATUS, | |
1196 | MSR_CORE_PERF_GLOBAL_CTRL, MSR_CORE_PERF_GLOBAL_OVF_CTRL, | |
1197 | MSR_ARCH_PERFMON_PERFCTR0, MSR_ARCH_PERFMON_PERFCTR1, | |
1198 | MSR_ARCH_PERFMON_PERFCTR0 + 2, MSR_ARCH_PERFMON_PERFCTR0 + 3, | |
1199 | MSR_ARCH_PERFMON_PERFCTR0 + 4, MSR_ARCH_PERFMON_PERFCTR0 + 5, | |
1200 | MSR_ARCH_PERFMON_PERFCTR0 + 6, MSR_ARCH_PERFMON_PERFCTR0 + 7, | |
1201 | MSR_ARCH_PERFMON_PERFCTR0 + 8, MSR_ARCH_PERFMON_PERFCTR0 + 9, | |
1202 | MSR_ARCH_PERFMON_PERFCTR0 + 10, MSR_ARCH_PERFMON_PERFCTR0 + 11, | |
1203 | MSR_ARCH_PERFMON_PERFCTR0 + 12, MSR_ARCH_PERFMON_PERFCTR0 + 13, | |
1204 | MSR_ARCH_PERFMON_PERFCTR0 + 14, MSR_ARCH_PERFMON_PERFCTR0 + 15, | |
1205 | MSR_ARCH_PERFMON_PERFCTR0 + 16, MSR_ARCH_PERFMON_PERFCTR0 + 17, | |
e2ada66e JM |
1206 | MSR_ARCH_PERFMON_EVENTSEL0, MSR_ARCH_PERFMON_EVENTSEL1, |
1207 | MSR_ARCH_PERFMON_EVENTSEL0 + 2, MSR_ARCH_PERFMON_EVENTSEL0 + 3, | |
1208 | MSR_ARCH_PERFMON_EVENTSEL0 + 4, MSR_ARCH_PERFMON_EVENTSEL0 + 5, | |
1209 | MSR_ARCH_PERFMON_EVENTSEL0 + 6, MSR_ARCH_PERFMON_EVENTSEL0 + 7, | |
1210 | MSR_ARCH_PERFMON_EVENTSEL0 + 8, MSR_ARCH_PERFMON_EVENTSEL0 + 9, | |
1211 | MSR_ARCH_PERFMON_EVENTSEL0 + 10, MSR_ARCH_PERFMON_EVENTSEL0 + 11, | |
1212 | MSR_ARCH_PERFMON_EVENTSEL0 + 12, MSR_ARCH_PERFMON_EVENTSEL0 + 13, | |
1213 | MSR_ARCH_PERFMON_EVENTSEL0 + 14, MSR_ARCH_PERFMON_EVENTSEL0 + 15, | |
1214 | MSR_ARCH_PERFMON_EVENTSEL0 + 16, MSR_ARCH_PERFMON_EVENTSEL0 + 17, | |
043405e1 CO |
1215 | }; |
1216 | ||
7a5ee6ed | 1217 | static u32 msrs_to_save[ARRAY_SIZE(msrs_to_save_all)]; |
043405e1 CO |
1218 | static unsigned num_msrs_to_save; |
1219 | ||
7a5ee6ed | 1220 | static const u32 emulated_msrs_all[] = { |
62ef68bb PB |
1221 | MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK, |
1222 | MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW, | |
1223 | HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL, | |
1224 | HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC, | |
72c139ba | 1225 | HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY, |
e7d9513b AS |
1226 | HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2, |
1227 | HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL, | |
e516cebb | 1228 | HV_X64_MSR_RESET, |
11c4b1ca | 1229 | HV_X64_MSR_VP_INDEX, |
9eec50b8 | 1230 | HV_X64_MSR_VP_RUNTIME, |
5c919412 | 1231 | HV_X64_MSR_SCONTROL, |
1f4b34f8 | 1232 | HV_X64_MSR_STIMER0_CONFIG, |
d4abc577 | 1233 | HV_X64_MSR_VP_ASSIST_PAGE, |
a2e164e7 VK |
1234 | HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL, |
1235 | HV_X64_MSR_TSC_EMULATION_STATUS, | |
1236 | ||
1237 | MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME, | |
62ef68bb PB |
1238 | MSR_KVM_PV_EOI_EN, |
1239 | ||
ba904635 | 1240 | MSR_IA32_TSC_ADJUST, |
a3e06bbe | 1241 | MSR_IA32_TSCDEADLINE, |
2bdb76c0 | 1242 | MSR_IA32_ARCH_CAPABILITIES, |
043405e1 | 1243 | MSR_IA32_MISC_ENABLE, |
908e75f3 AK |
1244 | MSR_IA32_MCG_STATUS, |
1245 | MSR_IA32_MCG_CTL, | |
c45dcc71 | 1246 | MSR_IA32_MCG_EXT_CTL, |
64d60670 | 1247 | MSR_IA32_SMBASE, |
52797bf9 | 1248 | MSR_SMI_COUNT, |
db2336a8 KH |
1249 | MSR_PLATFORM_INFO, |
1250 | MSR_MISC_FEATURES_ENABLES, | |
bc226f07 | 1251 | MSR_AMD64_VIRT_SPEC_CTRL, |
6c6a2ab9 | 1252 | MSR_IA32_POWER_CTL, |
99634e3e | 1253 | MSR_IA32_UCODE_REV, |
191c8137 | 1254 | |
95c5c7c7 PB |
1255 | /* |
1256 | * The following list leaves out MSRs whose values are determined | |
1257 | * by arch/x86/kvm/vmx/nested.c based on CPUID or other MSRs. | |
1258 | * We always support the "true" VMX control MSRs, even if the host | |
1259 | * processor does not, so I am putting these registers here rather | |
7a5ee6ed | 1260 | * than in msrs_to_save_all. |
95c5c7c7 PB |
1261 | */ |
1262 | MSR_IA32_VMX_BASIC, | |
1263 | MSR_IA32_VMX_TRUE_PINBASED_CTLS, | |
1264 | MSR_IA32_VMX_TRUE_PROCBASED_CTLS, | |
1265 | MSR_IA32_VMX_TRUE_EXIT_CTLS, | |
1266 | MSR_IA32_VMX_TRUE_ENTRY_CTLS, | |
1267 | MSR_IA32_VMX_MISC, | |
1268 | MSR_IA32_VMX_CR0_FIXED0, | |
1269 | MSR_IA32_VMX_CR4_FIXED0, | |
1270 | MSR_IA32_VMX_VMCS_ENUM, | |
1271 | MSR_IA32_VMX_PROCBASED_CTLS2, | |
1272 | MSR_IA32_VMX_EPT_VPID_CAP, | |
1273 | MSR_IA32_VMX_VMFUNC, | |
1274 | ||
191c8137 | 1275 | MSR_K7_HWCR, |
2d5ba19b | 1276 | MSR_KVM_POLL_CONTROL, |
043405e1 CO |
1277 | }; |
1278 | ||
7a5ee6ed | 1279 | static u32 emulated_msrs[ARRAY_SIZE(emulated_msrs_all)]; |
62ef68bb PB |
1280 | static unsigned num_emulated_msrs; |
1281 | ||
801e459a TL |
1282 | /* |
1283 | * List of msr numbers which are used to expose MSR-based features that | |
1284 | * can be used by a hypervisor to validate requested CPU features. | |
1285 | */ | |
7a5ee6ed | 1286 | static const u32 msr_based_features_all[] = { |
1389309c PB |
1287 | MSR_IA32_VMX_BASIC, |
1288 | MSR_IA32_VMX_TRUE_PINBASED_CTLS, | |
1289 | MSR_IA32_VMX_PINBASED_CTLS, | |
1290 | MSR_IA32_VMX_TRUE_PROCBASED_CTLS, | |
1291 | MSR_IA32_VMX_PROCBASED_CTLS, | |
1292 | MSR_IA32_VMX_TRUE_EXIT_CTLS, | |
1293 | MSR_IA32_VMX_EXIT_CTLS, | |
1294 | MSR_IA32_VMX_TRUE_ENTRY_CTLS, | |
1295 | MSR_IA32_VMX_ENTRY_CTLS, | |
1296 | MSR_IA32_VMX_MISC, | |
1297 | MSR_IA32_VMX_CR0_FIXED0, | |
1298 | MSR_IA32_VMX_CR0_FIXED1, | |
1299 | MSR_IA32_VMX_CR4_FIXED0, | |
1300 | MSR_IA32_VMX_CR4_FIXED1, | |
1301 | MSR_IA32_VMX_VMCS_ENUM, | |
1302 | MSR_IA32_VMX_PROCBASED_CTLS2, | |
1303 | MSR_IA32_VMX_EPT_VPID_CAP, | |
1304 | MSR_IA32_VMX_VMFUNC, | |
1305 | ||
d1d93fa9 | 1306 | MSR_F10H_DECFG, |
518e7b94 | 1307 | MSR_IA32_UCODE_REV, |
cd283252 | 1308 | MSR_IA32_ARCH_CAPABILITIES, |
801e459a TL |
1309 | }; |
1310 | ||
7a5ee6ed | 1311 | static u32 msr_based_features[ARRAY_SIZE(msr_based_features_all)]; |
801e459a TL |
1312 | static unsigned int num_msr_based_features; |
1313 | ||
4d22c17c | 1314 | static u64 kvm_get_arch_capabilities(void) |
5b76a3cf | 1315 | { |
4d22c17c | 1316 | u64 data = 0; |
5b76a3cf | 1317 | |
4d22c17c XL |
1318 | if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES)) |
1319 | rdmsrl(MSR_IA32_ARCH_CAPABILITIES, data); | |
5b76a3cf | 1320 | |
b8e8c830 PB |
1321 | /* |
1322 | * If nx_huge_pages is enabled, KVM's shadow paging will ensure that | |
1323 | * the nested hypervisor runs with NX huge pages. If it is not, | |
1324 | * L1 is anyway vulnerable to ITLB_MULTIHIT explots from other | |
1325 | * L1 guests, so it need not worry about its own (L2) guests. | |
1326 | */ | |
1327 | data |= ARCH_CAP_PSCHANGE_MC_NO; | |
1328 | ||
5b76a3cf PB |
1329 | /* |
1330 | * If we're doing cache flushes (either "always" or "cond") | |
1331 | * we will do one whenever the guest does a vmlaunch/vmresume. | |
1332 | * If an outer hypervisor is doing the cache flush for us | |
1333 | * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that | |
1334 | * capability to the guest too, and if EPT is disabled we're not | |
1335 | * vulnerable. Overall, only VMENTER_L1D_FLUSH_NEVER will | |
1336 | * require a nested hypervisor to do a flush of its own. | |
1337 | */ | |
1338 | if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER) | |
1339 | data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH; | |
1340 | ||
0c54914d PB |
1341 | if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN)) |
1342 | data |= ARCH_CAP_RDCL_NO; | |
1343 | if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS)) | |
1344 | data |= ARCH_CAP_SSB_NO; | |
1345 | if (!boot_cpu_has_bug(X86_BUG_MDS)) | |
1346 | data |= ARCH_CAP_MDS_NO; | |
1347 | ||
e1d38b63 | 1348 | /* |
c11f83e0 PB |
1349 | * On TAA affected systems: |
1350 | * - nothing to do if TSX is disabled on the host. | |
1351 | * - we emulate TSX_CTRL if present on the host. | |
1352 | * This lets the guest use VERW to clear CPU buffers. | |
e1d38b63 | 1353 | */ |
cbbaa272 | 1354 | if (!boot_cpu_has(X86_FEATURE_RTM)) |
c11f83e0 | 1355 | data &= ~(ARCH_CAP_TAA_NO | ARCH_CAP_TSX_CTRL_MSR); |
cbbaa272 PB |
1356 | else if (!boot_cpu_has_bug(X86_BUG_TAA)) |
1357 | data |= ARCH_CAP_TAA_NO; | |
e1d38b63 | 1358 | |
5b76a3cf PB |
1359 | return data; |
1360 | } | |
5b76a3cf | 1361 | |
66421c1e WL |
1362 | static int kvm_get_msr_feature(struct kvm_msr_entry *msr) |
1363 | { | |
1364 | switch (msr->index) { | |
cd283252 | 1365 | case MSR_IA32_ARCH_CAPABILITIES: |
5b76a3cf PB |
1366 | msr->data = kvm_get_arch_capabilities(); |
1367 | break; | |
1368 | case MSR_IA32_UCODE_REV: | |
cd283252 | 1369 | rdmsrl_safe(msr->index, &msr->data); |
518e7b94 | 1370 | break; |
66421c1e | 1371 | default: |
afaf0b2f | 1372 | if (kvm_x86_ops.get_msr_feature(msr)) |
66421c1e WL |
1373 | return 1; |
1374 | } | |
1375 | return 0; | |
1376 | } | |
1377 | ||
801e459a TL |
1378 | static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data) |
1379 | { | |
1380 | struct kvm_msr_entry msr; | |
66421c1e | 1381 | int r; |
801e459a TL |
1382 | |
1383 | msr.index = index; | |
66421c1e WL |
1384 | r = kvm_get_msr_feature(&msr); |
1385 | if (r) | |
1386 | return r; | |
801e459a TL |
1387 | |
1388 | *data = msr.data; | |
1389 | ||
1390 | return 0; | |
1391 | } | |
1392 | ||
11988499 | 1393 | static bool __kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer) |
15c4a640 | 1394 | { |
1b4d56b8 | 1395 | if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT)) |
11988499 | 1396 | return false; |
1b2fd70c | 1397 | |
1b4d56b8 | 1398 | if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM)) |
11988499 | 1399 | return false; |
d8017474 | 1400 | |
0a629563 SC |
1401 | if (efer & (EFER_LME | EFER_LMA) && |
1402 | !guest_cpuid_has(vcpu, X86_FEATURE_LM)) | |
1403 | return false; | |
1404 | ||
1405 | if (efer & EFER_NX && !guest_cpuid_has(vcpu, X86_FEATURE_NX)) | |
1406 | return false; | |
d8017474 | 1407 | |
384bb783 | 1408 | return true; |
11988499 SC |
1409 | |
1410 | } | |
1411 | bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer) | |
1412 | { | |
1413 | if (efer & efer_reserved_bits) | |
1414 | return false; | |
1415 | ||
1416 | return __kvm_valid_efer(vcpu, efer); | |
384bb783 JK |
1417 | } |
1418 | EXPORT_SYMBOL_GPL(kvm_valid_efer); | |
1419 | ||
11988499 | 1420 | static int set_efer(struct kvm_vcpu *vcpu, struct msr_data *msr_info) |
384bb783 JK |
1421 | { |
1422 | u64 old_efer = vcpu->arch.efer; | |
11988499 | 1423 | u64 efer = msr_info->data; |
384bb783 | 1424 | |
11988499 | 1425 | if (efer & efer_reserved_bits) |
66f61c92 | 1426 | return 1; |
384bb783 | 1427 | |
11988499 SC |
1428 | if (!msr_info->host_initiated) { |
1429 | if (!__kvm_valid_efer(vcpu, efer)) | |
1430 | return 1; | |
1431 | ||
1432 | if (is_paging(vcpu) && | |
1433 | (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME)) | |
1434 | return 1; | |
1435 | } | |
384bb783 | 1436 | |
15c4a640 | 1437 | efer &= ~EFER_LMA; |
f6801dff | 1438 | efer |= vcpu->arch.efer & EFER_LMA; |
15c4a640 | 1439 | |
afaf0b2f | 1440 | kvm_x86_ops.set_efer(vcpu, efer); |
a3d204e2 | 1441 | |
aad82703 SY |
1442 | /* Update reserved bits */ |
1443 | if ((efer ^ old_efer) & EFER_NX) | |
1444 | kvm_mmu_reset_context(vcpu); | |
1445 | ||
b69e8cae | 1446 | return 0; |
15c4a640 CO |
1447 | } |
1448 | ||
f2b4b7dd JR |
1449 | void kvm_enable_efer_bits(u64 mask) |
1450 | { | |
1451 | efer_reserved_bits &= ~mask; | |
1452 | } | |
1453 | EXPORT_SYMBOL_GPL(kvm_enable_efer_bits); | |
1454 | ||
15c4a640 | 1455 | /* |
f20935d8 SC |
1456 | * Write @data into the MSR specified by @index. Select MSR specific fault |
1457 | * checks are bypassed if @host_initiated is %true. | |
15c4a640 CO |
1458 | * Returns 0 on success, non-0 otherwise. |
1459 | * Assumes vcpu_load() was already called. | |
1460 | */ | |
f20935d8 SC |
1461 | static int __kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data, |
1462 | bool host_initiated) | |
15c4a640 | 1463 | { |
f20935d8 SC |
1464 | struct msr_data msr; |
1465 | ||
1466 | switch (index) { | |
854e8bb1 NA |
1467 | case MSR_FS_BASE: |
1468 | case MSR_GS_BASE: | |
1469 | case MSR_KERNEL_GS_BASE: | |
1470 | case MSR_CSTAR: | |
1471 | case MSR_LSTAR: | |
f20935d8 | 1472 | if (is_noncanonical_address(data, vcpu)) |
854e8bb1 NA |
1473 | return 1; |
1474 | break; | |
1475 | case MSR_IA32_SYSENTER_EIP: | |
1476 | case MSR_IA32_SYSENTER_ESP: | |
1477 | /* | |
1478 | * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if | |
1479 | * non-canonical address is written on Intel but not on | |
1480 | * AMD (which ignores the top 32-bits, because it does | |
1481 | * not implement 64-bit SYSENTER). | |
1482 | * | |
1483 | * 64-bit code should hence be able to write a non-canonical | |
1484 | * value on AMD. Making the address canonical ensures that | |
1485 | * vmentry does not fail on Intel after writing a non-canonical | |
1486 | * value, and that something deterministic happens if the guest | |
1487 | * invokes 64-bit SYSENTER. | |
1488 | */ | |
f20935d8 | 1489 | data = get_canonical(data, vcpu_virt_addr_bits(vcpu)); |
854e8bb1 | 1490 | } |
f20935d8 SC |
1491 | |
1492 | msr.data = data; | |
1493 | msr.index = index; | |
1494 | msr.host_initiated = host_initiated; | |
1495 | ||
afaf0b2f | 1496 | return kvm_x86_ops.set_msr(vcpu, &msr); |
15c4a640 CO |
1497 | } |
1498 | ||
313a3dc7 | 1499 | /* |
f20935d8 SC |
1500 | * Read the MSR specified by @index into @data. Select MSR specific fault |
1501 | * checks are bypassed if @host_initiated is %true. | |
1502 | * Returns 0 on success, non-0 otherwise. | |
1503 | * Assumes vcpu_load() was already called. | |
313a3dc7 | 1504 | */ |
edef5c36 PB |
1505 | int __kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data, |
1506 | bool host_initiated) | |
609e36d3 PB |
1507 | { |
1508 | struct msr_data msr; | |
f20935d8 | 1509 | int ret; |
609e36d3 PB |
1510 | |
1511 | msr.index = index; | |
f20935d8 | 1512 | msr.host_initiated = host_initiated; |
609e36d3 | 1513 | |
afaf0b2f | 1514 | ret = kvm_x86_ops.get_msr(vcpu, &msr); |
f20935d8 SC |
1515 | if (!ret) |
1516 | *data = msr.data; | |
1517 | return ret; | |
609e36d3 PB |
1518 | } |
1519 | ||
f20935d8 | 1520 | int kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data) |
313a3dc7 | 1521 | { |
f20935d8 SC |
1522 | return __kvm_get_msr(vcpu, index, data, false); |
1523 | } | |
1524 | EXPORT_SYMBOL_GPL(kvm_get_msr); | |
8fe8ab46 | 1525 | |
f20935d8 SC |
1526 | int kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data) |
1527 | { | |
1528 | return __kvm_set_msr(vcpu, index, data, false); | |
1529 | } | |
1530 | EXPORT_SYMBOL_GPL(kvm_set_msr); | |
1531 | ||
1edce0a9 SC |
1532 | int kvm_emulate_rdmsr(struct kvm_vcpu *vcpu) |
1533 | { | |
1534 | u32 ecx = kvm_rcx_read(vcpu); | |
1535 | u64 data; | |
1536 | ||
1537 | if (kvm_get_msr(vcpu, ecx, &data)) { | |
1538 | trace_kvm_msr_read_ex(ecx); | |
1539 | kvm_inject_gp(vcpu, 0); | |
1540 | return 1; | |
1541 | } | |
1542 | ||
1543 | trace_kvm_msr_read(ecx, data); | |
1544 | ||
1545 | kvm_rax_write(vcpu, data & -1u); | |
1546 | kvm_rdx_write(vcpu, (data >> 32) & -1u); | |
1547 | return kvm_skip_emulated_instruction(vcpu); | |
1548 | } | |
1549 | EXPORT_SYMBOL_GPL(kvm_emulate_rdmsr); | |
1550 | ||
1551 | int kvm_emulate_wrmsr(struct kvm_vcpu *vcpu) | |
1552 | { | |
1553 | u32 ecx = kvm_rcx_read(vcpu); | |
1554 | u64 data = kvm_read_edx_eax(vcpu); | |
1555 | ||
1556 | if (kvm_set_msr(vcpu, ecx, data)) { | |
1557 | trace_kvm_msr_write_ex(ecx, data); | |
1558 | kvm_inject_gp(vcpu, 0); | |
1559 | return 1; | |
1560 | } | |
1561 | ||
1562 | trace_kvm_msr_write(ecx, data); | |
1563 | return kvm_skip_emulated_instruction(vcpu); | |
1564 | } | |
1565 | EXPORT_SYMBOL_GPL(kvm_emulate_wrmsr); | |
1566 | ||
1e9e2622 WL |
1567 | /* |
1568 | * The fast path for frequent and performance sensitive wrmsr emulation, | |
1569 | * i.e. the sending of IPI, sending IPI early in the VM-Exit flow reduces | |
1570 | * the latency of virtual IPI by avoiding the expensive bits of transitioning | |
1571 | * from guest to host, e.g. reacquiring KVM's SRCU lock. In contrast to the | |
1572 | * other cases which must be called after interrupts are enabled on the host. | |
1573 | */ | |
1574 | static int handle_fastpath_set_x2apic_icr_irqoff(struct kvm_vcpu *vcpu, u64 data) | |
1575 | { | |
e1be9ac8 WL |
1576 | if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(vcpu->arch.apic)) |
1577 | return 1; | |
1578 | ||
1579 | if (((data & APIC_SHORT_MASK) == APIC_DEST_NOSHORT) && | |
1e9e2622 | 1580 | ((data & APIC_DEST_MASK) == APIC_DEST_PHYSICAL) && |
4064a4c6 WL |
1581 | ((data & APIC_MODE_MASK) == APIC_DM_FIXED) && |
1582 | ((u32)(data >> 32) != X2APIC_BROADCAST)) { | |
1e9e2622 | 1583 | |
d5361678 WL |
1584 | data &= ~(1 << 12); |
1585 | kvm_apic_send_ipi(vcpu->arch.apic, (u32)data, (u32)(data >> 32)); | |
1e9e2622 | 1586 | kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR2, (u32)(data >> 32)); |
d5361678 WL |
1587 | kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR, (u32)data); |
1588 | trace_kvm_apic_write(APIC_ICR, (u32)data); | |
1589 | return 0; | |
1e9e2622 WL |
1590 | } |
1591 | ||
1592 | return 1; | |
1593 | } | |
1594 | ||
1595 | enum exit_fastpath_completion handle_fastpath_set_msr_irqoff(struct kvm_vcpu *vcpu) | |
1596 | { | |
1597 | u32 msr = kvm_rcx_read(vcpu); | |
8a1038de | 1598 | u64 data; |
1e9e2622 WL |
1599 | int ret = 0; |
1600 | ||
1601 | switch (msr) { | |
1602 | case APIC_BASE_MSR + (APIC_ICR >> 4): | |
8a1038de | 1603 | data = kvm_read_edx_eax(vcpu); |
1e9e2622 WL |
1604 | ret = handle_fastpath_set_x2apic_icr_irqoff(vcpu, data); |
1605 | break; | |
1606 | default: | |
1607 | return EXIT_FASTPATH_NONE; | |
1608 | } | |
1609 | ||
1610 | if (!ret) { | |
1611 | trace_kvm_msr_write(msr, data); | |
1612 | return EXIT_FASTPATH_SKIP_EMUL_INS; | |
1613 | } | |
1614 | ||
1615 | return EXIT_FASTPATH_NONE; | |
1616 | } | |
1617 | EXPORT_SYMBOL_GPL(handle_fastpath_set_msr_irqoff); | |
1618 | ||
f20935d8 SC |
1619 | /* |
1620 | * Adapt set_msr() to msr_io()'s calling convention | |
1621 | */ | |
1622 | static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data) | |
1623 | { | |
1624 | return __kvm_get_msr(vcpu, index, data, true); | |
1625 | } | |
1626 | ||
1627 | static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data) | |
1628 | { | |
1629 | return __kvm_set_msr(vcpu, index, *data, true); | |
313a3dc7 CO |
1630 | } |
1631 | ||
16e8d74d | 1632 | #ifdef CONFIG_X86_64 |
53fafdbb MT |
1633 | struct pvclock_clock { |
1634 | int vclock_mode; | |
1635 | u64 cycle_last; | |
1636 | u64 mask; | |
1637 | u32 mult; | |
1638 | u32 shift; | |
917f9475 PB |
1639 | u64 base_cycles; |
1640 | u64 offset; | |
53fafdbb MT |
1641 | }; |
1642 | ||
16e8d74d MT |
1643 | struct pvclock_gtod_data { |
1644 | seqcount_t seq; | |
1645 | ||
53fafdbb MT |
1646 | struct pvclock_clock clock; /* extract of a clocksource struct */ |
1647 | struct pvclock_clock raw_clock; /* extract of a clocksource struct */ | |
16e8d74d | 1648 | |
917f9475 | 1649 | ktime_t offs_boot; |
55dd00a7 | 1650 | u64 wall_time_sec; |
16e8d74d MT |
1651 | }; |
1652 | ||
1653 | static struct pvclock_gtod_data pvclock_gtod_data; | |
1654 | ||
1655 | static void update_pvclock_gtod(struct timekeeper *tk) | |
1656 | { | |
1657 | struct pvclock_gtod_data *vdata = &pvclock_gtod_data; | |
1658 | ||
1659 | write_seqcount_begin(&vdata->seq); | |
1660 | ||
1661 | /* copy pvclock gtod data */ | |
b95a8a27 | 1662 | vdata->clock.vclock_mode = tk->tkr_mono.clock->vdso_clock_mode; |
876e7881 PZ |
1663 | vdata->clock.cycle_last = tk->tkr_mono.cycle_last; |
1664 | vdata->clock.mask = tk->tkr_mono.mask; | |
1665 | vdata->clock.mult = tk->tkr_mono.mult; | |
1666 | vdata->clock.shift = tk->tkr_mono.shift; | |
917f9475 PB |
1667 | vdata->clock.base_cycles = tk->tkr_mono.xtime_nsec; |
1668 | vdata->clock.offset = tk->tkr_mono.base; | |
16e8d74d | 1669 | |
b95a8a27 | 1670 | vdata->raw_clock.vclock_mode = tk->tkr_raw.clock->vdso_clock_mode; |
53fafdbb MT |
1671 | vdata->raw_clock.cycle_last = tk->tkr_raw.cycle_last; |
1672 | vdata->raw_clock.mask = tk->tkr_raw.mask; | |
1673 | vdata->raw_clock.mult = tk->tkr_raw.mult; | |
1674 | vdata->raw_clock.shift = tk->tkr_raw.shift; | |
917f9475 PB |
1675 | vdata->raw_clock.base_cycles = tk->tkr_raw.xtime_nsec; |
1676 | vdata->raw_clock.offset = tk->tkr_raw.base; | |
16e8d74d | 1677 | |
55dd00a7 MT |
1678 | vdata->wall_time_sec = tk->xtime_sec; |
1679 | ||
917f9475 | 1680 | vdata->offs_boot = tk->offs_boot; |
53fafdbb | 1681 | |
16e8d74d MT |
1682 | write_seqcount_end(&vdata->seq); |
1683 | } | |
8171cd68 PB |
1684 | |
1685 | static s64 get_kvmclock_base_ns(void) | |
1686 | { | |
1687 | /* Count up from boot time, but with the frequency of the raw clock. */ | |
1688 | return ktime_to_ns(ktime_add(ktime_get_raw(), pvclock_gtod_data.offs_boot)); | |
1689 | } | |
1690 | #else | |
1691 | static s64 get_kvmclock_base_ns(void) | |
1692 | { | |
1693 | /* Master clock not used, so we can just use CLOCK_BOOTTIME. */ | |
1694 | return ktime_get_boottime_ns(); | |
1695 | } | |
16e8d74d MT |
1696 | #endif |
1697 | ||
bab5bb39 NK |
1698 | void kvm_set_pending_timer(struct kvm_vcpu *vcpu) |
1699 | { | |
bab5bb39 | 1700 | kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu); |
4d151bf3 | 1701 | kvm_vcpu_kick(vcpu); |
bab5bb39 | 1702 | } |
16e8d74d | 1703 | |
18068523 GOC |
1704 | static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock) |
1705 | { | |
9ed3c444 AK |
1706 | int version; |
1707 | int r; | |
50d0a0f9 | 1708 | struct pvclock_wall_clock wc; |
8171cd68 | 1709 | u64 wall_nsec; |
18068523 GOC |
1710 | |
1711 | if (!wall_clock) | |
1712 | return; | |
1713 | ||
9ed3c444 AK |
1714 | r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version)); |
1715 | if (r) | |
1716 | return; | |
1717 | ||
1718 | if (version & 1) | |
1719 | ++version; /* first time write, random junk */ | |
1720 | ||
1721 | ++version; | |
18068523 | 1722 | |
1dab1345 NK |
1723 | if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version))) |
1724 | return; | |
18068523 | 1725 | |
50d0a0f9 GH |
1726 | /* |
1727 | * The guest calculates current wall clock time by adding | |
34c238a1 | 1728 | * system time (updated by kvm_guest_time_update below) to the |
8171cd68 | 1729 | * wall clock specified here. We do the reverse here. |
50d0a0f9 | 1730 | */ |
8171cd68 | 1731 | wall_nsec = ktime_get_real_ns() - get_kvmclock_ns(kvm); |
50d0a0f9 | 1732 | |
8171cd68 PB |
1733 | wc.nsec = do_div(wall_nsec, 1000000000); |
1734 | wc.sec = (u32)wall_nsec; /* overflow in 2106 guest time */ | |
50d0a0f9 | 1735 | wc.version = version; |
18068523 GOC |
1736 | |
1737 | kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc)); | |
1738 | ||
1739 | version++; | |
1740 | kvm_write_guest(kvm, wall_clock, &version, sizeof(version)); | |
18068523 GOC |
1741 | } |
1742 | ||
50d0a0f9 GH |
1743 | static uint32_t div_frac(uint32_t dividend, uint32_t divisor) |
1744 | { | |
b51012de PB |
1745 | do_shl32_div32(dividend, divisor); |
1746 | return dividend; | |
50d0a0f9 GH |
1747 | } |
1748 | ||
3ae13faa | 1749 | static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz, |
5f4e3f88 | 1750 | s8 *pshift, u32 *pmultiplier) |
50d0a0f9 | 1751 | { |
5f4e3f88 | 1752 | uint64_t scaled64; |
50d0a0f9 GH |
1753 | int32_t shift = 0; |
1754 | uint64_t tps64; | |
1755 | uint32_t tps32; | |
1756 | ||
3ae13faa PB |
1757 | tps64 = base_hz; |
1758 | scaled64 = scaled_hz; | |
50933623 | 1759 | while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) { |
50d0a0f9 GH |
1760 | tps64 >>= 1; |
1761 | shift--; | |
1762 | } | |
1763 | ||
1764 | tps32 = (uint32_t)tps64; | |
50933623 JK |
1765 | while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) { |
1766 | if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000) | |
5f4e3f88 ZA |
1767 | scaled64 >>= 1; |
1768 | else | |
1769 | tps32 <<= 1; | |
50d0a0f9 GH |
1770 | shift++; |
1771 | } | |
1772 | ||
5f4e3f88 ZA |
1773 | *pshift = shift; |
1774 | *pmultiplier = div_frac(scaled64, tps32); | |
50d0a0f9 GH |
1775 | } |
1776 | ||
d828199e | 1777 | #ifdef CONFIG_X86_64 |
16e8d74d | 1778 | static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0); |
d828199e | 1779 | #endif |
16e8d74d | 1780 | |
c8076604 | 1781 | static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz); |
69b0049a | 1782 | static unsigned long max_tsc_khz; |
c8076604 | 1783 | |
cc578287 | 1784 | static u32 adjust_tsc_khz(u32 khz, s32 ppm) |
1e993611 | 1785 | { |
cc578287 ZA |
1786 | u64 v = (u64)khz * (1000000 + ppm); |
1787 | do_div(v, 1000000); | |
1788 | return v; | |
1e993611 JR |
1789 | } |
1790 | ||
381d585c HZ |
1791 | static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale) |
1792 | { | |
1793 | u64 ratio; | |
1794 | ||
1795 | /* Guest TSC same frequency as host TSC? */ | |
1796 | if (!scale) { | |
1797 | vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio; | |
1798 | return 0; | |
1799 | } | |
1800 | ||
1801 | /* TSC scaling supported? */ | |
1802 | if (!kvm_has_tsc_control) { | |
1803 | if (user_tsc_khz > tsc_khz) { | |
1804 | vcpu->arch.tsc_catchup = 1; | |
1805 | vcpu->arch.tsc_always_catchup = 1; | |
1806 | return 0; | |
1807 | } else { | |
3f16a5c3 | 1808 | pr_warn_ratelimited("user requested TSC rate below hardware speed\n"); |
381d585c HZ |
1809 | return -1; |
1810 | } | |
1811 | } | |
1812 | ||
1813 | /* TSC scaling required - calculate ratio */ | |
1814 | ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits, | |
1815 | user_tsc_khz, tsc_khz); | |
1816 | ||
1817 | if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) { | |
3f16a5c3 PB |
1818 | pr_warn_ratelimited("Invalid TSC scaling ratio - virtual-tsc-khz=%u\n", |
1819 | user_tsc_khz); | |
381d585c HZ |
1820 | return -1; |
1821 | } | |
1822 | ||
1823 | vcpu->arch.tsc_scaling_ratio = ratio; | |
1824 | return 0; | |
1825 | } | |
1826 | ||
4941b8cb | 1827 | static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz) |
759379dd | 1828 | { |
cc578287 ZA |
1829 | u32 thresh_lo, thresh_hi; |
1830 | int use_scaling = 0; | |
217fc9cf | 1831 | |
03ba32ca | 1832 | /* tsc_khz can be zero if TSC calibration fails */ |
4941b8cb | 1833 | if (user_tsc_khz == 0) { |
ad721883 HZ |
1834 | /* set tsc_scaling_ratio to a safe value */ |
1835 | vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio; | |
381d585c | 1836 | return -1; |
ad721883 | 1837 | } |
03ba32ca | 1838 | |
c285545f | 1839 | /* Compute a scale to convert nanoseconds in TSC cycles */ |
3ae13faa | 1840 | kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC, |
cc578287 ZA |
1841 | &vcpu->arch.virtual_tsc_shift, |
1842 | &vcpu->arch.virtual_tsc_mult); | |
4941b8cb | 1843 | vcpu->arch.virtual_tsc_khz = user_tsc_khz; |
cc578287 ZA |
1844 | |
1845 | /* | |
1846 | * Compute the variation in TSC rate which is acceptable | |
1847 | * within the range of tolerance and decide if the | |
1848 | * rate being applied is within that bounds of the hardware | |
1849 | * rate. If so, no scaling or compensation need be done. | |
1850 | */ | |
1851 | thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm); | |
1852 | thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm); | |
4941b8cb PB |
1853 | if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) { |
1854 | pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi); | |
cc578287 ZA |
1855 | use_scaling = 1; |
1856 | } | |
4941b8cb | 1857 | return set_tsc_khz(vcpu, user_tsc_khz, use_scaling); |
c285545f ZA |
1858 | } |
1859 | ||
1860 | static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns) | |
1861 | { | |
e26101b1 | 1862 | u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec, |
cc578287 ZA |
1863 | vcpu->arch.virtual_tsc_mult, |
1864 | vcpu->arch.virtual_tsc_shift); | |
e26101b1 | 1865 | tsc += vcpu->arch.this_tsc_write; |
c285545f ZA |
1866 | return tsc; |
1867 | } | |
1868 | ||
b0c39dc6 VK |
1869 | static inline int gtod_is_based_on_tsc(int mode) |
1870 | { | |
b95a8a27 | 1871 | return mode == VDSO_CLOCKMODE_TSC || mode == VDSO_CLOCKMODE_HVCLOCK; |
b0c39dc6 VK |
1872 | } |
1873 | ||
69b0049a | 1874 | static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu) |
b48aa97e MT |
1875 | { |
1876 | #ifdef CONFIG_X86_64 | |
1877 | bool vcpus_matched; | |
b48aa97e MT |
1878 | struct kvm_arch *ka = &vcpu->kvm->arch; |
1879 | struct pvclock_gtod_data *gtod = &pvclock_gtod_data; | |
1880 | ||
1881 | vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 == | |
1882 | atomic_read(&vcpu->kvm->online_vcpus)); | |
1883 | ||
7f187922 MT |
1884 | /* |
1885 | * Once the masterclock is enabled, always perform request in | |
1886 | * order to update it. | |
1887 | * | |
1888 | * In order to enable masterclock, the host clocksource must be TSC | |
1889 | * and the vcpus need to have matched TSCs. When that happens, | |
1890 | * perform request to enable masterclock. | |
1891 | */ | |
1892 | if (ka->use_master_clock || | |
b0c39dc6 | 1893 | (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched)) |
b48aa97e MT |
1894 | kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu); |
1895 | ||
1896 | trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc, | |
1897 | atomic_read(&vcpu->kvm->online_vcpus), | |
1898 | ka->use_master_clock, gtod->clock.vclock_mode); | |
1899 | #endif | |
1900 | } | |
1901 | ||
ba904635 WA |
1902 | static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset) |
1903 | { | |
afaf0b2f | 1904 | u64 curr_offset = kvm_x86_ops.read_l1_tsc_offset(vcpu); |
ba904635 WA |
1905 | vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset; |
1906 | } | |
1907 | ||
35181e86 HZ |
1908 | /* |
1909 | * Multiply tsc by a fixed point number represented by ratio. | |
1910 | * | |
1911 | * The most significant 64-N bits (mult) of ratio represent the | |
1912 | * integral part of the fixed point number; the remaining N bits | |
1913 | * (frac) represent the fractional part, ie. ratio represents a fixed | |
1914 | * point number (mult + frac * 2^(-N)). | |
1915 | * | |
1916 | * N equals to kvm_tsc_scaling_ratio_frac_bits. | |
1917 | */ | |
1918 | static inline u64 __scale_tsc(u64 ratio, u64 tsc) | |
1919 | { | |
1920 | return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits); | |
1921 | } | |
1922 | ||
1923 | u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc) | |
1924 | { | |
1925 | u64 _tsc = tsc; | |
1926 | u64 ratio = vcpu->arch.tsc_scaling_ratio; | |
1927 | ||
1928 | if (ratio != kvm_default_tsc_scaling_ratio) | |
1929 | _tsc = __scale_tsc(ratio, tsc); | |
1930 | ||
1931 | return _tsc; | |
1932 | } | |
1933 | EXPORT_SYMBOL_GPL(kvm_scale_tsc); | |
1934 | ||
07c1419a HZ |
1935 | static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc) |
1936 | { | |
1937 | u64 tsc; | |
1938 | ||
1939 | tsc = kvm_scale_tsc(vcpu, rdtsc()); | |
1940 | ||
1941 | return target_tsc - tsc; | |
1942 | } | |
1943 | ||
4ba76538 HZ |
1944 | u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc) |
1945 | { | |
afaf0b2f | 1946 | u64 tsc_offset = kvm_x86_ops.read_l1_tsc_offset(vcpu); |
e79f245d KA |
1947 | |
1948 | return tsc_offset + kvm_scale_tsc(vcpu, host_tsc); | |
4ba76538 HZ |
1949 | } |
1950 | EXPORT_SYMBOL_GPL(kvm_read_l1_tsc); | |
1951 | ||
a545ab6a LC |
1952 | static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset) |
1953 | { | |
afaf0b2f | 1954 | vcpu->arch.tsc_offset = kvm_x86_ops.write_l1_tsc_offset(vcpu, offset); |
a545ab6a LC |
1955 | } |
1956 | ||
b0c39dc6 VK |
1957 | static inline bool kvm_check_tsc_unstable(void) |
1958 | { | |
1959 | #ifdef CONFIG_X86_64 | |
1960 | /* | |
1961 | * TSC is marked unstable when we're running on Hyper-V, | |
1962 | * 'TSC page' clocksource is good. | |
1963 | */ | |
b95a8a27 | 1964 | if (pvclock_gtod_data.clock.vclock_mode == VDSO_CLOCKMODE_HVCLOCK) |
b0c39dc6 VK |
1965 | return false; |
1966 | #endif | |
1967 | return check_tsc_unstable(); | |
1968 | } | |
1969 | ||
8fe8ab46 | 1970 | void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr) |
99e3e30a ZA |
1971 | { |
1972 | struct kvm *kvm = vcpu->kvm; | |
f38e098f | 1973 | u64 offset, ns, elapsed; |
99e3e30a | 1974 | unsigned long flags; |
b48aa97e | 1975 | bool matched; |
0d3da0d2 | 1976 | bool already_matched; |
8fe8ab46 | 1977 | u64 data = msr->data; |
c5e8ec8e | 1978 | bool synchronizing = false; |
99e3e30a | 1979 | |
038f8c11 | 1980 | raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags); |
07c1419a | 1981 | offset = kvm_compute_tsc_offset(vcpu, data); |
8171cd68 | 1982 | ns = get_kvmclock_base_ns(); |
f38e098f | 1983 | elapsed = ns - kvm->arch.last_tsc_nsec; |
5d3cb0f6 | 1984 | |
03ba32ca | 1985 | if (vcpu->arch.virtual_tsc_khz) { |
bd8fab39 DP |
1986 | if (data == 0 && msr->host_initiated) { |
1987 | /* | |
1988 | * detection of vcpu initialization -- need to sync | |
1989 | * with other vCPUs. This particularly helps to keep | |
1990 | * kvm_clock stable after CPU hotplug | |
1991 | */ | |
1992 | synchronizing = true; | |
1993 | } else { | |
1994 | u64 tsc_exp = kvm->arch.last_tsc_write + | |
1995 | nsec_to_cycles(vcpu, elapsed); | |
1996 | u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL; | |
1997 | /* | |
1998 | * Special case: TSC write with a small delta (1 second) | |
1999 | * of virtual cycle time against real time is | |
2000 | * interpreted as an attempt to synchronize the CPU. | |
2001 | */ | |
2002 | synchronizing = data < tsc_exp + tsc_hz && | |
2003 | data + tsc_hz > tsc_exp; | |
2004 | } | |
c5e8ec8e | 2005 | } |
f38e098f ZA |
2006 | |
2007 | /* | |
5d3cb0f6 ZA |
2008 | * For a reliable TSC, we can match TSC offsets, and for an unstable |
2009 | * TSC, we add elapsed time in this computation. We could let the | |
2010 | * compensation code attempt to catch up if we fall behind, but | |
2011 | * it's better to try to match offsets from the beginning. | |
2012 | */ | |
c5e8ec8e | 2013 | if (synchronizing && |
5d3cb0f6 | 2014 | vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) { |
b0c39dc6 | 2015 | if (!kvm_check_tsc_unstable()) { |
e26101b1 | 2016 | offset = kvm->arch.cur_tsc_offset; |
f38e098f | 2017 | } else { |
857e4099 | 2018 | u64 delta = nsec_to_cycles(vcpu, elapsed); |
5d3cb0f6 | 2019 | data += delta; |
07c1419a | 2020 | offset = kvm_compute_tsc_offset(vcpu, data); |
f38e098f | 2021 | } |
b48aa97e | 2022 | matched = true; |
0d3da0d2 | 2023 | already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation); |
e26101b1 ZA |
2024 | } else { |
2025 | /* | |
2026 | * We split periods of matched TSC writes into generations. | |
2027 | * For each generation, we track the original measured | |
2028 | * nanosecond time, offset, and write, so if TSCs are in | |
2029 | * sync, we can match exact offset, and if not, we can match | |
4a969980 | 2030 | * exact software computation in compute_guest_tsc() |
e26101b1 ZA |
2031 | * |
2032 | * These values are tracked in kvm->arch.cur_xxx variables. | |
2033 | */ | |
2034 | kvm->arch.cur_tsc_generation++; | |
2035 | kvm->arch.cur_tsc_nsec = ns; | |
2036 | kvm->arch.cur_tsc_write = data; | |
2037 | kvm->arch.cur_tsc_offset = offset; | |
b48aa97e | 2038 | matched = false; |
f38e098f | 2039 | } |
e26101b1 ZA |
2040 | |
2041 | /* | |
2042 | * We also track th most recent recorded KHZ, write and time to | |
2043 | * allow the matching interval to be extended at each write. | |
2044 | */ | |
f38e098f ZA |
2045 | kvm->arch.last_tsc_nsec = ns; |
2046 | kvm->arch.last_tsc_write = data; | |
5d3cb0f6 | 2047 | kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz; |
99e3e30a | 2048 | |
b183aa58 | 2049 | vcpu->arch.last_guest_tsc = data; |
e26101b1 ZA |
2050 | |
2051 | /* Keep track of which generation this VCPU has synchronized to */ | |
2052 | vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation; | |
2053 | vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec; | |
2054 | vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write; | |
2055 | ||
d6321d49 | 2056 | if (!msr->host_initiated && guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) |
ba904635 | 2057 | update_ia32_tsc_adjust_msr(vcpu, offset); |
d6321d49 | 2058 | |
a545ab6a | 2059 | kvm_vcpu_write_tsc_offset(vcpu, offset); |
e26101b1 | 2060 | raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags); |
b48aa97e MT |
2061 | |
2062 | spin_lock(&kvm->arch.pvclock_gtod_sync_lock); | |
0d3da0d2 | 2063 | if (!matched) { |
b48aa97e | 2064 | kvm->arch.nr_vcpus_matched_tsc = 0; |
0d3da0d2 TG |
2065 | } else if (!already_matched) { |
2066 | kvm->arch.nr_vcpus_matched_tsc++; | |
2067 | } | |
b48aa97e MT |
2068 | |
2069 | kvm_track_tsc_matching(vcpu); | |
2070 | spin_unlock(&kvm->arch.pvclock_gtod_sync_lock); | |
99e3e30a | 2071 | } |
e26101b1 | 2072 | |
99e3e30a ZA |
2073 | EXPORT_SYMBOL_GPL(kvm_write_tsc); |
2074 | ||
58ea6767 HZ |
2075 | static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu, |
2076 | s64 adjustment) | |
2077 | { | |
afaf0b2f | 2078 | u64 tsc_offset = kvm_x86_ops.read_l1_tsc_offset(vcpu); |
326e7425 | 2079 | kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment); |
58ea6767 HZ |
2080 | } |
2081 | ||
2082 | static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment) | |
2083 | { | |
2084 | if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio) | |
2085 | WARN_ON(adjustment < 0); | |
2086 | adjustment = kvm_scale_tsc(vcpu, (u64) adjustment); | |
ea26e4ec | 2087 | adjust_tsc_offset_guest(vcpu, adjustment); |
58ea6767 HZ |
2088 | } |
2089 | ||
d828199e MT |
2090 | #ifdef CONFIG_X86_64 |
2091 | ||
a5a1d1c2 | 2092 | static u64 read_tsc(void) |
d828199e | 2093 | { |
a5a1d1c2 | 2094 | u64 ret = (u64)rdtsc_ordered(); |
03b9730b | 2095 | u64 last = pvclock_gtod_data.clock.cycle_last; |
d828199e MT |
2096 | |
2097 | if (likely(ret >= last)) | |
2098 | return ret; | |
2099 | ||
2100 | /* | |
2101 | * GCC likes to generate cmov here, but this branch is extremely | |
6a6256f9 | 2102 | * predictable (it's just a function of time and the likely is |
d828199e MT |
2103 | * very likely) and there's a data dependence, so force GCC |
2104 | * to generate a branch instead. I don't barrier() because | |
2105 | * we don't actually need a barrier, and if this function | |
2106 | * ever gets inlined it will generate worse code. | |
2107 | */ | |
2108 | asm volatile (""); | |
2109 | return last; | |
2110 | } | |
2111 | ||
53fafdbb MT |
2112 | static inline u64 vgettsc(struct pvclock_clock *clock, u64 *tsc_timestamp, |
2113 | int *mode) | |
d828199e MT |
2114 | { |
2115 | long v; | |
b0c39dc6 VK |
2116 | u64 tsc_pg_val; |
2117 | ||
53fafdbb | 2118 | switch (clock->vclock_mode) { |
b95a8a27 | 2119 | case VDSO_CLOCKMODE_HVCLOCK: |
b0c39dc6 VK |
2120 | tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(), |
2121 | tsc_timestamp); | |
2122 | if (tsc_pg_val != U64_MAX) { | |
2123 | /* TSC page valid */ | |
b95a8a27 | 2124 | *mode = VDSO_CLOCKMODE_HVCLOCK; |
53fafdbb MT |
2125 | v = (tsc_pg_val - clock->cycle_last) & |
2126 | clock->mask; | |
b0c39dc6 VK |
2127 | } else { |
2128 | /* TSC page invalid */ | |
b95a8a27 | 2129 | *mode = VDSO_CLOCKMODE_NONE; |
b0c39dc6 VK |
2130 | } |
2131 | break; | |
b95a8a27 TG |
2132 | case VDSO_CLOCKMODE_TSC: |
2133 | *mode = VDSO_CLOCKMODE_TSC; | |
b0c39dc6 | 2134 | *tsc_timestamp = read_tsc(); |
53fafdbb MT |
2135 | v = (*tsc_timestamp - clock->cycle_last) & |
2136 | clock->mask; | |
b0c39dc6 VK |
2137 | break; |
2138 | default: | |
b95a8a27 | 2139 | *mode = VDSO_CLOCKMODE_NONE; |
b0c39dc6 | 2140 | } |
d828199e | 2141 | |
b95a8a27 | 2142 | if (*mode == VDSO_CLOCKMODE_NONE) |
b0c39dc6 | 2143 | *tsc_timestamp = v = 0; |
d828199e | 2144 | |
53fafdbb | 2145 | return v * clock->mult; |
d828199e MT |
2146 | } |
2147 | ||
53fafdbb | 2148 | static int do_monotonic_raw(s64 *t, u64 *tsc_timestamp) |
d828199e | 2149 | { |
cbcf2dd3 | 2150 | struct pvclock_gtod_data *gtod = &pvclock_gtod_data; |
d828199e | 2151 | unsigned long seq; |
d828199e | 2152 | int mode; |
cbcf2dd3 | 2153 | u64 ns; |
d828199e | 2154 | |
d828199e MT |
2155 | do { |
2156 | seq = read_seqcount_begin(>od->seq); | |
917f9475 | 2157 | ns = gtod->raw_clock.base_cycles; |
53fafdbb | 2158 | ns += vgettsc(>od->raw_clock, tsc_timestamp, &mode); |
917f9475 PB |
2159 | ns >>= gtod->raw_clock.shift; |
2160 | ns += ktime_to_ns(ktime_add(gtod->raw_clock.offset, gtod->offs_boot)); | |
d828199e | 2161 | } while (unlikely(read_seqcount_retry(>od->seq, seq))); |
cbcf2dd3 | 2162 | *t = ns; |
d828199e MT |
2163 | |
2164 | return mode; | |
2165 | } | |
2166 | ||
899a31f5 | 2167 | static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp) |
55dd00a7 MT |
2168 | { |
2169 | struct pvclock_gtod_data *gtod = &pvclock_gtod_data; | |
2170 | unsigned long seq; | |
2171 | int mode; | |
2172 | u64 ns; | |
2173 | ||
2174 | do { | |
2175 | seq = read_seqcount_begin(>od->seq); | |
55dd00a7 | 2176 | ts->tv_sec = gtod->wall_time_sec; |
917f9475 | 2177 | ns = gtod->clock.base_cycles; |
53fafdbb | 2178 | ns += vgettsc(>od->clock, tsc_timestamp, &mode); |
55dd00a7 MT |
2179 | ns >>= gtod->clock.shift; |
2180 | } while (unlikely(read_seqcount_retry(>od->seq, seq))); | |
2181 | ||
2182 | ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns); | |
2183 | ts->tv_nsec = ns; | |
2184 | ||
2185 | return mode; | |
2186 | } | |
2187 | ||
b0c39dc6 VK |
2188 | /* returns true if host is using TSC based clocksource */ |
2189 | static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp) | |
d828199e | 2190 | { |
d828199e | 2191 | /* checked again under seqlock below */ |
b0c39dc6 | 2192 | if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode)) |
d828199e MT |
2193 | return false; |
2194 | ||
53fafdbb | 2195 | return gtod_is_based_on_tsc(do_monotonic_raw(kernel_ns, |
b0c39dc6 | 2196 | tsc_timestamp)); |
d828199e | 2197 | } |
55dd00a7 | 2198 | |
b0c39dc6 | 2199 | /* returns true if host is using TSC based clocksource */ |
899a31f5 | 2200 | static bool kvm_get_walltime_and_clockread(struct timespec64 *ts, |
b0c39dc6 | 2201 | u64 *tsc_timestamp) |
55dd00a7 MT |
2202 | { |
2203 | /* checked again under seqlock below */ | |
b0c39dc6 | 2204 | if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode)) |
55dd00a7 MT |
2205 | return false; |
2206 | ||
b0c39dc6 | 2207 | return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp)); |
55dd00a7 | 2208 | } |
d828199e MT |
2209 | #endif |
2210 | ||
2211 | /* | |
2212 | * | |
b48aa97e MT |
2213 | * Assuming a stable TSC across physical CPUS, and a stable TSC |
2214 | * across virtual CPUs, the following condition is possible. | |
2215 | * Each numbered line represents an event visible to both | |
d828199e MT |
2216 | * CPUs at the next numbered event. |
2217 | * | |
2218 | * "timespecX" represents host monotonic time. "tscX" represents | |
2219 | * RDTSC value. | |
2220 | * | |
2221 | * VCPU0 on CPU0 | VCPU1 on CPU1 | |
2222 | * | |
2223 | * 1. read timespec0,tsc0 | |
2224 | * 2. | timespec1 = timespec0 + N | |
2225 | * | tsc1 = tsc0 + M | |
2226 | * 3. transition to guest | transition to guest | |
2227 | * 4. ret0 = timespec0 + (rdtsc - tsc0) | | |
2228 | * 5. | ret1 = timespec1 + (rdtsc - tsc1) | |
2229 | * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M)) | |
2230 | * | |
2231 | * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity: | |
2232 | * | |
2233 | * - ret0 < ret1 | |
2234 | * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M)) | |
2235 | * ... | |
2236 | * - 0 < N - M => M < N | |
2237 | * | |
2238 | * That is, when timespec0 != timespec1, M < N. Unfortunately that is not | |
2239 | * always the case (the difference between two distinct xtime instances | |
2240 | * might be smaller then the difference between corresponding TSC reads, | |
2241 | * when updating guest vcpus pvclock areas). | |
2242 | * | |
2243 | * To avoid that problem, do not allow visibility of distinct | |
2244 | * system_timestamp/tsc_timestamp values simultaneously: use a master | |
2245 | * copy of host monotonic time values. Update that master copy | |
2246 | * in lockstep. | |
2247 | * | |
b48aa97e | 2248 | * Rely on synchronization of host TSCs and guest TSCs for monotonicity. |
d828199e MT |
2249 | * |
2250 | */ | |
2251 | ||
2252 | static void pvclock_update_vm_gtod_copy(struct kvm *kvm) | |
2253 | { | |
2254 | #ifdef CONFIG_X86_64 | |
2255 | struct kvm_arch *ka = &kvm->arch; | |
2256 | int vclock_mode; | |
b48aa97e MT |
2257 | bool host_tsc_clocksource, vcpus_matched; |
2258 | ||
2259 | vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 == | |
2260 | atomic_read(&kvm->online_vcpus)); | |
d828199e MT |
2261 | |
2262 | /* | |
2263 | * If the host uses TSC clock, then passthrough TSC as stable | |
2264 | * to the guest. | |
2265 | */ | |
b48aa97e | 2266 | host_tsc_clocksource = kvm_get_time_and_clockread( |
d828199e MT |
2267 | &ka->master_kernel_ns, |
2268 | &ka->master_cycle_now); | |
2269 | ||
16a96021 | 2270 | ka->use_master_clock = host_tsc_clocksource && vcpus_matched |
a826faf1 | 2271 | && !ka->backwards_tsc_observed |
54750f2c | 2272 | && !ka->boot_vcpu_runs_old_kvmclock; |
b48aa97e | 2273 | |
d828199e MT |
2274 | if (ka->use_master_clock) |
2275 | atomic_set(&kvm_guest_has_master_clock, 1); | |
2276 | ||
2277 | vclock_mode = pvclock_gtod_data.clock.vclock_mode; | |
b48aa97e MT |
2278 | trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode, |
2279 | vcpus_matched); | |
d828199e MT |
2280 | #endif |
2281 | } | |
2282 | ||
2860c4b1 PB |
2283 | void kvm_make_mclock_inprogress_request(struct kvm *kvm) |
2284 | { | |
2285 | kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS); | |
2286 | } | |
2287 | ||
2e762ff7 MT |
2288 | static void kvm_gen_update_masterclock(struct kvm *kvm) |
2289 | { | |
2290 | #ifdef CONFIG_X86_64 | |
2291 | int i; | |
2292 | struct kvm_vcpu *vcpu; | |
2293 | struct kvm_arch *ka = &kvm->arch; | |
2294 | ||
2295 | spin_lock(&ka->pvclock_gtod_sync_lock); | |
2296 | kvm_make_mclock_inprogress_request(kvm); | |
2297 | /* no guest entries from this point */ | |
2298 | pvclock_update_vm_gtod_copy(kvm); | |
2299 | ||
2300 | kvm_for_each_vcpu(i, vcpu, kvm) | |
105b21bb | 2301 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu); |
2e762ff7 MT |
2302 | |
2303 | /* guest entries allowed */ | |
2304 | kvm_for_each_vcpu(i, vcpu, kvm) | |
72875d8a | 2305 | kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu); |
2e762ff7 MT |
2306 | |
2307 | spin_unlock(&ka->pvclock_gtod_sync_lock); | |
2308 | #endif | |
2309 | } | |
2310 | ||
e891a32e | 2311 | u64 get_kvmclock_ns(struct kvm *kvm) |
108b249c | 2312 | { |
108b249c | 2313 | struct kvm_arch *ka = &kvm->arch; |
8b953440 | 2314 | struct pvclock_vcpu_time_info hv_clock; |
e2c2206a | 2315 | u64 ret; |
108b249c | 2316 | |
8b953440 PB |
2317 | spin_lock(&ka->pvclock_gtod_sync_lock); |
2318 | if (!ka->use_master_clock) { | |
2319 | spin_unlock(&ka->pvclock_gtod_sync_lock); | |
8171cd68 | 2320 | return get_kvmclock_base_ns() + ka->kvmclock_offset; |
108b249c PB |
2321 | } |
2322 | ||
8b953440 PB |
2323 | hv_clock.tsc_timestamp = ka->master_cycle_now; |
2324 | hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset; | |
2325 | spin_unlock(&ka->pvclock_gtod_sync_lock); | |
2326 | ||
e2c2206a WL |
2327 | /* both __this_cpu_read() and rdtsc() should be on the same cpu */ |
2328 | get_cpu(); | |
2329 | ||
e70b57a6 WL |
2330 | if (__this_cpu_read(cpu_tsc_khz)) { |
2331 | kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL, | |
2332 | &hv_clock.tsc_shift, | |
2333 | &hv_clock.tsc_to_system_mul); | |
2334 | ret = __pvclock_read_cycles(&hv_clock, rdtsc()); | |
2335 | } else | |
8171cd68 | 2336 | ret = get_kvmclock_base_ns() + ka->kvmclock_offset; |
e2c2206a WL |
2337 | |
2338 | put_cpu(); | |
2339 | ||
2340 | return ret; | |
108b249c PB |
2341 | } |
2342 | ||
0d6dd2ff PB |
2343 | static void kvm_setup_pvclock_page(struct kvm_vcpu *v) |
2344 | { | |
2345 | struct kvm_vcpu_arch *vcpu = &v->arch; | |
2346 | struct pvclock_vcpu_time_info guest_hv_clock; | |
2347 | ||
4e335d9e | 2348 | if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time, |
0d6dd2ff PB |
2349 | &guest_hv_clock, sizeof(guest_hv_clock)))) |
2350 | return; | |
2351 | ||
2352 | /* This VCPU is paused, but it's legal for a guest to read another | |
2353 | * VCPU's kvmclock, so we really have to follow the specification where | |
2354 | * it says that version is odd if data is being modified, and even after | |
2355 | * it is consistent. | |
2356 | * | |
2357 | * Version field updates must be kept separate. This is because | |
2358 | * kvm_write_guest_cached might use a "rep movs" instruction, and | |
2359 | * writes within a string instruction are weakly ordered. So there | |
2360 | * are three writes overall. | |
2361 | * | |
2362 | * As a small optimization, only write the version field in the first | |
2363 | * and third write. The vcpu->pv_time cache is still valid, because the | |
2364 | * version field is the first in the struct. | |
2365 | */ | |
2366 | BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0); | |
2367 | ||
51c4b8bb LA |
2368 | if (guest_hv_clock.version & 1) |
2369 | ++guest_hv_clock.version; /* first time write, random junk */ | |
2370 | ||
0d6dd2ff | 2371 | vcpu->hv_clock.version = guest_hv_clock.version + 1; |
4e335d9e PB |
2372 | kvm_write_guest_cached(v->kvm, &vcpu->pv_time, |
2373 | &vcpu->hv_clock, | |
2374 | sizeof(vcpu->hv_clock.version)); | |
0d6dd2ff PB |
2375 | |
2376 | smp_wmb(); | |
2377 | ||
2378 | /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */ | |
2379 | vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED); | |
2380 | ||
2381 | if (vcpu->pvclock_set_guest_stopped_request) { | |
2382 | vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED; | |
2383 | vcpu->pvclock_set_guest_stopped_request = false; | |
2384 | } | |
2385 | ||
2386 | trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock); | |
2387 | ||
4e335d9e PB |
2388 | kvm_write_guest_cached(v->kvm, &vcpu->pv_time, |
2389 | &vcpu->hv_clock, | |
2390 | sizeof(vcpu->hv_clock)); | |
0d6dd2ff PB |
2391 | |
2392 | smp_wmb(); | |
2393 | ||
2394 | vcpu->hv_clock.version++; | |
4e335d9e PB |
2395 | kvm_write_guest_cached(v->kvm, &vcpu->pv_time, |
2396 | &vcpu->hv_clock, | |
2397 | sizeof(vcpu->hv_clock.version)); | |
0d6dd2ff PB |
2398 | } |
2399 | ||
34c238a1 | 2400 | static int kvm_guest_time_update(struct kvm_vcpu *v) |
18068523 | 2401 | { |
78db6a50 | 2402 | unsigned long flags, tgt_tsc_khz; |
18068523 | 2403 | struct kvm_vcpu_arch *vcpu = &v->arch; |
d828199e | 2404 | struct kvm_arch *ka = &v->kvm->arch; |
f25e656d | 2405 | s64 kernel_ns; |
d828199e | 2406 | u64 tsc_timestamp, host_tsc; |
51d59c6b | 2407 | u8 pvclock_flags; |
d828199e MT |
2408 | bool use_master_clock; |
2409 | ||
2410 | kernel_ns = 0; | |
2411 | host_tsc = 0; | |
18068523 | 2412 | |
d828199e MT |
2413 | /* |
2414 | * If the host uses TSC clock, then passthrough TSC as stable | |
2415 | * to the guest. | |
2416 | */ | |
2417 | spin_lock(&ka->pvclock_gtod_sync_lock); | |
2418 | use_master_clock = ka->use_master_clock; | |
2419 | if (use_master_clock) { | |
2420 | host_tsc = ka->master_cycle_now; | |
2421 | kernel_ns = ka->master_kernel_ns; | |
2422 | } | |
2423 | spin_unlock(&ka->pvclock_gtod_sync_lock); | |
c09664bb MT |
2424 | |
2425 | /* Keep irq disabled to prevent changes to the clock */ | |
2426 | local_irq_save(flags); | |
78db6a50 PB |
2427 | tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz); |
2428 | if (unlikely(tgt_tsc_khz == 0)) { | |
c09664bb MT |
2429 | local_irq_restore(flags); |
2430 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, v); | |
2431 | return 1; | |
2432 | } | |
d828199e | 2433 | if (!use_master_clock) { |
4ea1636b | 2434 | host_tsc = rdtsc(); |
8171cd68 | 2435 | kernel_ns = get_kvmclock_base_ns(); |
d828199e MT |
2436 | } |
2437 | ||
4ba76538 | 2438 | tsc_timestamp = kvm_read_l1_tsc(v, host_tsc); |
d828199e | 2439 | |
c285545f ZA |
2440 | /* |
2441 | * We may have to catch up the TSC to match elapsed wall clock | |
2442 | * time for two reasons, even if kvmclock is used. | |
2443 | * 1) CPU could have been running below the maximum TSC rate | |
2444 | * 2) Broken TSC compensation resets the base at each VCPU | |
2445 | * entry to avoid unknown leaps of TSC even when running | |
2446 | * again on the same CPU. This may cause apparent elapsed | |
2447 | * time to disappear, and the guest to stand still or run | |
2448 | * very slowly. | |
2449 | */ | |
2450 | if (vcpu->tsc_catchup) { | |
2451 | u64 tsc = compute_guest_tsc(v, kernel_ns); | |
2452 | if (tsc > tsc_timestamp) { | |
f1e2b260 | 2453 | adjust_tsc_offset_guest(v, tsc - tsc_timestamp); |
c285545f ZA |
2454 | tsc_timestamp = tsc; |
2455 | } | |
50d0a0f9 GH |
2456 | } |
2457 | ||
18068523 GOC |
2458 | local_irq_restore(flags); |
2459 | ||
0d6dd2ff | 2460 | /* With all the info we got, fill in the values */ |
18068523 | 2461 | |
78db6a50 PB |
2462 | if (kvm_has_tsc_control) |
2463 | tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz); | |
2464 | ||
2465 | if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) { | |
3ae13faa | 2466 | kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL, |
5f4e3f88 ZA |
2467 | &vcpu->hv_clock.tsc_shift, |
2468 | &vcpu->hv_clock.tsc_to_system_mul); | |
78db6a50 | 2469 | vcpu->hw_tsc_khz = tgt_tsc_khz; |
8cfdc000 ZA |
2470 | } |
2471 | ||
1d5f066e | 2472 | vcpu->hv_clock.tsc_timestamp = tsc_timestamp; |
759379dd | 2473 | vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset; |
28e4639a | 2474 | vcpu->last_guest_tsc = tsc_timestamp; |
51d59c6b | 2475 | |
d828199e | 2476 | /* If the host uses TSC clocksource, then it is stable */ |
0d6dd2ff | 2477 | pvclock_flags = 0; |
d828199e MT |
2478 | if (use_master_clock) |
2479 | pvclock_flags |= PVCLOCK_TSC_STABLE_BIT; | |
2480 | ||
78c0337a MT |
2481 | vcpu->hv_clock.flags = pvclock_flags; |
2482 | ||
095cf55d PB |
2483 | if (vcpu->pv_time_enabled) |
2484 | kvm_setup_pvclock_page(v); | |
2485 | if (v == kvm_get_vcpu(v->kvm, 0)) | |
2486 | kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock); | |
8cfdc000 | 2487 | return 0; |
c8076604 GH |
2488 | } |
2489 | ||
0061d53d MT |
2490 | /* |
2491 | * kvmclock updates which are isolated to a given vcpu, such as | |
2492 | * vcpu->cpu migration, should not allow system_timestamp from | |
2493 | * the rest of the vcpus to remain static. Otherwise ntp frequency | |
2494 | * correction applies to one vcpu's system_timestamp but not | |
2495 | * the others. | |
2496 | * | |
2497 | * So in those cases, request a kvmclock update for all vcpus. | |
7e44e449 AJ |
2498 | * We need to rate-limit these requests though, as they can |
2499 | * considerably slow guests that have a large number of vcpus. | |
2500 | * The time for a remote vcpu to update its kvmclock is bound | |
2501 | * by the delay we use to rate-limit the updates. | |
0061d53d MT |
2502 | */ |
2503 | ||
7e44e449 AJ |
2504 | #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100) |
2505 | ||
2506 | static void kvmclock_update_fn(struct work_struct *work) | |
0061d53d MT |
2507 | { |
2508 | int i; | |
7e44e449 AJ |
2509 | struct delayed_work *dwork = to_delayed_work(work); |
2510 | struct kvm_arch *ka = container_of(dwork, struct kvm_arch, | |
2511 | kvmclock_update_work); | |
2512 | struct kvm *kvm = container_of(ka, struct kvm, arch); | |
0061d53d MT |
2513 | struct kvm_vcpu *vcpu; |
2514 | ||
2515 | kvm_for_each_vcpu(i, vcpu, kvm) { | |
105b21bb | 2516 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu); |
0061d53d MT |
2517 | kvm_vcpu_kick(vcpu); |
2518 | } | |
2519 | } | |
2520 | ||
7e44e449 AJ |
2521 | static void kvm_gen_kvmclock_update(struct kvm_vcpu *v) |
2522 | { | |
2523 | struct kvm *kvm = v->kvm; | |
2524 | ||
105b21bb | 2525 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, v); |
7e44e449 AJ |
2526 | schedule_delayed_work(&kvm->arch.kvmclock_update_work, |
2527 | KVMCLOCK_UPDATE_DELAY); | |
2528 | } | |
2529 | ||
332967a3 AJ |
2530 | #define KVMCLOCK_SYNC_PERIOD (300 * HZ) |
2531 | ||
2532 | static void kvmclock_sync_fn(struct work_struct *work) | |
2533 | { | |
2534 | struct delayed_work *dwork = to_delayed_work(work); | |
2535 | struct kvm_arch *ka = container_of(dwork, struct kvm_arch, | |
2536 | kvmclock_sync_work); | |
2537 | struct kvm *kvm = container_of(ka, struct kvm, arch); | |
2538 | ||
630994b3 MT |
2539 | if (!kvmclock_periodic_sync) |
2540 | return; | |
2541 | ||
332967a3 AJ |
2542 | schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0); |
2543 | schedule_delayed_work(&kvm->arch.kvmclock_sync_work, | |
2544 | KVMCLOCK_SYNC_PERIOD); | |
2545 | } | |
2546 | ||
191c8137 BP |
2547 | /* |
2548 | * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP. | |
2549 | */ | |
2550 | static bool can_set_mci_status(struct kvm_vcpu *vcpu) | |
2551 | { | |
2552 | /* McStatusWrEn enabled? */ | |
23493d0a | 2553 | if (guest_cpuid_is_amd_or_hygon(vcpu)) |
191c8137 BP |
2554 | return !!(vcpu->arch.msr_hwcr & BIT_ULL(18)); |
2555 | ||
2556 | return false; | |
2557 | } | |
2558 | ||
9ffd986c | 2559 | static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info) |
15c4a640 | 2560 | { |
890ca9ae HY |
2561 | u64 mcg_cap = vcpu->arch.mcg_cap; |
2562 | unsigned bank_num = mcg_cap & 0xff; | |
9ffd986c WL |
2563 | u32 msr = msr_info->index; |
2564 | u64 data = msr_info->data; | |
890ca9ae | 2565 | |
15c4a640 | 2566 | switch (msr) { |
15c4a640 | 2567 | case MSR_IA32_MCG_STATUS: |
890ca9ae | 2568 | vcpu->arch.mcg_status = data; |
15c4a640 | 2569 | break; |
c7ac679c | 2570 | case MSR_IA32_MCG_CTL: |
44883f01 PB |
2571 | if (!(mcg_cap & MCG_CTL_P) && |
2572 | (data || !msr_info->host_initiated)) | |
890ca9ae HY |
2573 | return 1; |
2574 | if (data != 0 && data != ~(u64)0) | |
44883f01 | 2575 | return 1; |
890ca9ae HY |
2576 | vcpu->arch.mcg_ctl = data; |
2577 | break; | |
2578 | default: | |
2579 | if (msr >= MSR_IA32_MC0_CTL && | |
81760dcc | 2580 | msr < MSR_IA32_MCx_CTL(bank_num)) { |
6ec4c5ee MP |
2581 | u32 offset = array_index_nospec( |
2582 | msr - MSR_IA32_MC0_CTL, | |
2583 | MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL); | |
2584 | ||
114be429 AP |
2585 | /* only 0 or all 1s can be written to IA32_MCi_CTL |
2586 | * some Linux kernels though clear bit 10 in bank 4 to | |
2587 | * workaround a BIOS/GART TBL issue on AMD K8s, ignore | |
2588 | * this to avoid an uncatched #GP in the guest | |
2589 | */ | |
890ca9ae | 2590 | if ((offset & 0x3) == 0 && |
114be429 | 2591 | data != 0 && (data | (1 << 10)) != ~(u64)0) |
890ca9ae | 2592 | return -1; |
191c8137 BP |
2593 | |
2594 | /* MCi_STATUS */ | |
9ffd986c | 2595 | if (!msr_info->host_initiated && |
191c8137 BP |
2596 | (offset & 0x3) == 1 && data != 0) { |
2597 | if (!can_set_mci_status(vcpu)) | |
2598 | return -1; | |
2599 | } | |
2600 | ||
890ca9ae HY |
2601 | vcpu->arch.mce_banks[offset] = data; |
2602 | break; | |
2603 | } | |
2604 | return 1; | |
2605 | } | |
2606 | return 0; | |
2607 | } | |
2608 | ||
ffde22ac ES |
2609 | static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data) |
2610 | { | |
2611 | struct kvm *kvm = vcpu->kvm; | |
2612 | int lm = is_long_mode(vcpu); | |
2613 | u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64 | |
2614 | : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32; | |
2615 | u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64 | |
2616 | : kvm->arch.xen_hvm_config.blob_size_32; | |
2617 | u32 page_num = data & ~PAGE_MASK; | |
2618 | u64 page_addr = data & PAGE_MASK; | |
2619 | u8 *page; | |
2620 | int r; | |
2621 | ||
2622 | r = -E2BIG; | |
2623 | if (page_num >= blob_size) | |
2624 | goto out; | |
2625 | r = -ENOMEM; | |
ff5c2c03 SL |
2626 | page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE); |
2627 | if (IS_ERR(page)) { | |
2628 | r = PTR_ERR(page); | |
ffde22ac | 2629 | goto out; |
ff5c2c03 | 2630 | } |
54bf36aa | 2631 | if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE)) |
ffde22ac ES |
2632 | goto out_free; |
2633 | r = 0; | |
2634 | out_free: | |
2635 | kfree(page); | |
2636 | out: | |
2637 | return r; | |
2638 | } | |
2639 | ||
344d9588 GN |
2640 | static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data) |
2641 | { | |
2642 | gpa_t gpa = data & ~0x3f; | |
2643 | ||
52a5c155 WL |
2644 | /* Bits 3:5 are reserved, Should be zero */ |
2645 | if (data & 0x38) | |
344d9588 GN |
2646 | return 1; |
2647 | ||
2648 | vcpu->arch.apf.msr_val = data; | |
2649 | ||
2650 | if (!(data & KVM_ASYNC_PF_ENABLED)) { | |
2651 | kvm_clear_async_pf_completion_queue(vcpu); | |
2652 | kvm_async_pf_hash_reset(vcpu); | |
2653 | return 0; | |
2654 | } | |
2655 | ||
4e335d9e | 2656 | if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa, |
8f964525 | 2657 | sizeof(u32))) |
344d9588 GN |
2658 | return 1; |
2659 | ||
6adba527 | 2660 | vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS); |
52a5c155 | 2661 | vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT; |
344d9588 GN |
2662 | kvm_async_pf_wakeup_all(vcpu); |
2663 | return 0; | |
2664 | } | |
2665 | ||
12f9a48f GC |
2666 | static void kvmclock_reset(struct kvm_vcpu *vcpu) |
2667 | { | |
0b79459b | 2668 | vcpu->arch.pv_time_enabled = false; |
49dedf0d | 2669 | vcpu->arch.time = 0; |
12f9a48f GC |
2670 | } |
2671 | ||
f38a7b75 WL |
2672 | static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu, bool invalidate_gpa) |
2673 | { | |
2674 | ++vcpu->stat.tlb_flush; | |
afaf0b2f | 2675 | kvm_x86_ops.tlb_flush(vcpu, invalidate_gpa); |
f38a7b75 WL |
2676 | } |
2677 | ||
c9aaa895 GC |
2678 | static void record_steal_time(struct kvm_vcpu *vcpu) |
2679 | { | |
b0431382 BO |
2680 | struct kvm_host_map map; |
2681 | struct kvm_steal_time *st; | |
2682 | ||
c9aaa895 GC |
2683 | if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED)) |
2684 | return; | |
2685 | ||
b0431382 BO |
2686 | /* -EAGAIN is returned in atomic context so we can just return. */ |
2687 | if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT, | |
2688 | &map, &vcpu->arch.st.cache, false)) | |
c9aaa895 GC |
2689 | return; |
2690 | ||
b0431382 BO |
2691 | st = map.hva + |
2692 | offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS); | |
2693 | ||
f38a7b75 WL |
2694 | /* |
2695 | * Doing a TLB flush here, on the guest's behalf, can avoid | |
2696 | * expensive IPIs. | |
2697 | */ | |
b382f44e | 2698 | trace_kvm_pv_tlb_flush(vcpu->vcpu_id, |
b0431382 BO |
2699 | st->preempted & KVM_VCPU_FLUSH_TLB); |
2700 | if (xchg(&st->preempted, 0) & KVM_VCPU_FLUSH_TLB) | |
f38a7b75 | 2701 | kvm_vcpu_flush_tlb(vcpu, false); |
0b9f6c46 | 2702 | |
a6bd811f | 2703 | vcpu->arch.st.preempted = 0; |
35f3fae1 | 2704 | |
b0431382 BO |
2705 | if (st->version & 1) |
2706 | st->version += 1; /* first time write, random junk */ | |
35f3fae1 | 2707 | |
b0431382 | 2708 | st->version += 1; |
35f3fae1 WL |
2709 | |
2710 | smp_wmb(); | |
2711 | ||
b0431382 | 2712 | st->steal += current->sched_info.run_delay - |
c54cdf14 LC |
2713 | vcpu->arch.st.last_steal; |
2714 | vcpu->arch.st.last_steal = current->sched_info.run_delay; | |
35f3fae1 | 2715 | |
35f3fae1 WL |
2716 | smp_wmb(); |
2717 | ||
b0431382 | 2718 | st->version += 1; |
c9aaa895 | 2719 | |
b0431382 | 2720 | kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, false); |
c9aaa895 GC |
2721 | } |
2722 | ||
8fe8ab46 | 2723 | int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info) |
15c4a640 | 2724 | { |
5753785f | 2725 | bool pr = false; |
8fe8ab46 WA |
2726 | u32 msr = msr_info->index; |
2727 | u64 data = msr_info->data; | |
5753785f | 2728 | |
15c4a640 | 2729 | switch (msr) { |
2e32b719 | 2730 | case MSR_AMD64_NB_CFG: |
2e32b719 BP |
2731 | case MSR_IA32_UCODE_WRITE: |
2732 | case MSR_VM_HSAVE_PA: | |
2733 | case MSR_AMD64_PATCH_LOADER: | |
2734 | case MSR_AMD64_BU_CFG2: | |
405a353a | 2735 | case MSR_AMD64_DC_CFG: |
0e1b869f | 2736 | case MSR_F15H_EX_CFG: |
2e32b719 BP |
2737 | break; |
2738 | ||
518e7b94 WL |
2739 | case MSR_IA32_UCODE_REV: |
2740 | if (msr_info->host_initiated) | |
2741 | vcpu->arch.microcode_version = data; | |
2742 | break; | |
0cf9135b SC |
2743 | case MSR_IA32_ARCH_CAPABILITIES: |
2744 | if (!msr_info->host_initiated) | |
2745 | return 1; | |
2746 | vcpu->arch.arch_capabilities = data; | |
2747 | break; | |
15c4a640 | 2748 | case MSR_EFER: |
11988499 | 2749 | return set_efer(vcpu, msr_info); |
8f1589d9 AP |
2750 | case MSR_K7_HWCR: |
2751 | data &= ~(u64)0x40; /* ignore flush filter disable */ | |
82494028 | 2752 | data &= ~(u64)0x100; /* ignore ignne emulation enable */ |
a223c313 | 2753 | data &= ~(u64)0x8; /* ignore TLB cache disable */ |
191c8137 BP |
2754 | |
2755 | /* Handle McStatusWrEn */ | |
2756 | if (data == BIT_ULL(18)) { | |
2757 | vcpu->arch.msr_hwcr = data; | |
2758 | } else if (data != 0) { | |
a737f256 CD |
2759 | vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n", |
2760 | data); | |
8f1589d9 AP |
2761 | return 1; |
2762 | } | |
15c4a640 | 2763 | break; |
f7c6d140 AP |
2764 | case MSR_FAM10H_MMIO_CONF_BASE: |
2765 | if (data != 0) { | |
a737f256 CD |
2766 | vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: " |
2767 | "0x%llx\n", data); | |
f7c6d140 AP |
2768 | return 1; |
2769 | } | |
15c4a640 | 2770 | break; |
b5e2fec0 AG |
2771 | case MSR_IA32_DEBUGCTLMSR: |
2772 | if (!data) { | |
2773 | /* We support the non-activated case already */ | |
2774 | break; | |
2775 | } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) { | |
2776 | /* Values other than LBR and BTF are vendor-specific, | |
2777 | thus reserved and should throw a #GP */ | |
2778 | return 1; | |
2779 | } | |
a737f256 CD |
2780 | vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n", |
2781 | __func__, data); | |
b5e2fec0 | 2782 | break; |
9ba075a6 | 2783 | case 0x200 ... 0x2ff: |
ff53604b | 2784 | return kvm_mtrr_set_msr(vcpu, msr, data); |
15c4a640 | 2785 | case MSR_IA32_APICBASE: |
58cb628d | 2786 | return kvm_set_apic_base(vcpu, msr_info); |
0105d1a5 GN |
2787 | case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff: |
2788 | return kvm_x2apic_msr_write(vcpu, msr, data); | |
a3e06bbe LJ |
2789 | case MSR_IA32_TSCDEADLINE: |
2790 | kvm_set_lapic_tscdeadline_msr(vcpu, data); | |
2791 | break; | |
ba904635 | 2792 | case MSR_IA32_TSC_ADJUST: |
d6321d49 | 2793 | if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) { |
ba904635 | 2794 | if (!msr_info->host_initiated) { |
d913b904 | 2795 | s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr; |
d7add054 | 2796 | adjust_tsc_offset_guest(vcpu, adj); |
ba904635 WA |
2797 | } |
2798 | vcpu->arch.ia32_tsc_adjust_msr = data; | |
2799 | } | |
2800 | break; | |
15c4a640 | 2801 | case MSR_IA32_MISC_ENABLE: |
511a8556 WL |
2802 | if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT) && |
2803 | ((vcpu->arch.ia32_misc_enable_msr ^ data) & MSR_IA32_MISC_ENABLE_MWAIT)) { | |
2804 | if (!guest_cpuid_has(vcpu, X86_FEATURE_XMM3)) | |
2805 | return 1; | |
2806 | vcpu->arch.ia32_misc_enable_msr = data; | |
2807 | kvm_update_cpuid(vcpu); | |
2808 | } else { | |
2809 | vcpu->arch.ia32_misc_enable_msr = data; | |
2810 | } | |
15c4a640 | 2811 | break; |
64d60670 PB |
2812 | case MSR_IA32_SMBASE: |
2813 | if (!msr_info->host_initiated) | |
2814 | return 1; | |
2815 | vcpu->arch.smbase = data; | |
2816 | break; | |
73f624f4 PB |
2817 | case MSR_IA32_POWER_CTL: |
2818 | vcpu->arch.msr_ia32_power_ctl = data; | |
2819 | break; | |
dd259935 PB |
2820 | case MSR_IA32_TSC: |
2821 | kvm_write_tsc(vcpu, msr_info); | |
2822 | break; | |
864e2ab2 AL |
2823 | case MSR_IA32_XSS: |
2824 | if (!msr_info->host_initiated && | |
2825 | !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES)) | |
2826 | return 1; | |
2827 | /* | |
a1bead2a SC |
2828 | * KVM supports exposing PT to the guest, but does not support |
2829 | * IA32_XSS[bit 8]. Guests have to use RDMSR/WRMSR rather than | |
2830 | * XSAVES/XRSTORS to save/restore PT MSRs. | |
864e2ab2 | 2831 | */ |
408e9a31 | 2832 | if (data & ~supported_xss) |
864e2ab2 AL |
2833 | return 1; |
2834 | vcpu->arch.ia32_xss = data; | |
2835 | break; | |
52797bf9 LA |
2836 | case MSR_SMI_COUNT: |
2837 | if (!msr_info->host_initiated) | |
2838 | return 1; | |
2839 | vcpu->arch.smi_count = data; | |
2840 | break; | |
11c6bffa | 2841 | case MSR_KVM_WALL_CLOCK_NEW: |
18068523 GOC |
2842 | case MSR_KVM_WALL_CLOCK: |
2843 | vcpu->kvm->arch.wall_clock = data; | |
2844 | kvm_write_wall_clock(vcpu->kvm, data); | |
2845 | break; | |
11c6bffa | 2846 | case MSR_KVM_SYSTEM_TIME_NEW: |
18068523 | 2847 | case MSR_KVM_SYSTEM_TIME: { |
54750f2c MT |
2848 | struct kvm_arch *ka = &vcpu->kvm->arch; |
2849 | ||
54750f2c MT |
2850 | if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) { |
2851 | bool tmp = (msr == MSR_KVM_SYSTEM_TIME); | |
2852 | ||
2853 | if (ka->boot_vcpu_runs_old_kvmclock != tmp) | |
1bd2009e | 2854 | kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu); |
54750f2c MT |
2855 | |
2856 | ka->boot_vcpu_runs_old_kvmclock = tmp; | |
2857 | } | |
2858 | ||
18068523 | 2859 | vcpu->arch.time = data; |
0061d53d | 2860 | kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu); |
18068523 GOC |
2861 | |
2862 | /* we verify if the enable bit is set... */ | |
49dedf0d | 2863 | vcpu->arch.pv_time_enabled = false; |
18068523 GOC |
2864 | if (!(data & 1)) |
2865 | break; | |
2866 | ||
49dedf0d | 2867 | if (!kvm_gfn_to_hva_cache_init(vcpu->kvm, |
8f964525 AH |
2868 | &vcpu->arch.pv_time, data & ~1ULL, |
2869 | sizeof(struct pvclock_vcpu_time_info))) | |
0b79459b | 2870 | vcpu->arch.pv_time_enabled = true; |
32cad84f | 2871 | |
18068523 GOC |
2872 | break; |
2873 | } | |
344d9588 GN |
2874 | case MSR_KVM_ASYNC_PF_EN: |
2875 | if (kvm_pv_enable_async_pf(vcpu, data)) | |
2876 | return 1; | |
2877 | break; | |
c9aaa895 GC |
2878 | case MSR_KVM_STEAL_TIME: |
2879 | ||
2880 | if (unlikely(!sched_info_on())) | |
2881 | return 1; | |
2882 | ||
2883 | if (data & KVM_STEAL_RESERVED_MASK) | |
2884 | return 1; | |
2885 | ||
c9aaa895 GC |
2886 | vcpu->arch.st.msr_val = data; |
2887 | ||
2888 | if (!(data & KVM_MSR_ENABLED)) | |
2889 | break; | |
2890 | ||
c9aaa895 GC |
2891 | kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu); |
2892 | ||
2893 | break; | |
ae7a2a3f | 2894 | case MSR_KVM_PV_EOI_EN: |
72bbf935 | 2895 | if (kvm_lapic_enable_pv_eoi(vcpu, data, sizeof(u8))) |
ae7a2a3f MT |
2896 | return 1; |
2897 | break; | |
c9aaa895 | 2898 | |
2d5ba19b MT |
2899 | case MSR_KVM_POLL_CONTROL: |
2900 | /* only enable bit supported */ | |
2901 | if (data & (-1ULL << 1)) | |
2902 | return 1; | |
2903 | ||
2904 | vcpu->arch.msr_kvm_poll_control = data; | |
2905 | break; | |
2906 | ||
890ca9ae HY |
2907 | case MSR_IA32_MCG_CTL: |
2908 | case MSR_IA32_MCG_STATUS: | |
81760dcc | 2909 | case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1: |
9ffd986c | 2910 | return set_msr_mce(vcpu, msr_info); |
71db6023 | 2911 | |
6912ac32 WH |
2912 | case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3: |
2913 | case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1: | |
2914 | pr = true; /* fall through */ | |
2915 | case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3: | |
2916 | case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1: | |
c6702c9d | 2917 | if (kvm_pmu_is_valid_msr(vcpu, msr)) |
afd80d85 | 2918 | return kvm_pmu_set_msr(vcpu, msr_info); |
5753785f GN |
2919 | |
2920 | if (pr || data != 0) | |
a737f256 CD |
2921 | vcpu_unimpl(vcpu, "disabled perfctr wrmsr: " |
2922 | "0x%x data 0x%llx\n", msr, data); | |
5753785f | 2923 | break; |
84e0cefa JS |
2924 | case MSR_K7_CLK_CTL: |
2925 | /* | |
2926 | * Ignore all writes to this no longer documented MSR. | |
2927 | * Writes are only relevant for old K7 processors, | |
2928 | * all pre-dating SVM, but a recommended workaround from | |
4a969980 | 2929 | * AMD for these chips. It is possible to specify the |
84e0cefa JS |
2930 | * affected processor models on the command line, hence |
2931 | * the need to ignore the workaround. | |
2932 | */ | |
2933 | break; | |
55cd8e5a | 2934 | case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15: |
e7d9513b AS |
2935 | case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4: |
2936 | case HV_X64_MSR_CRASH_CTL: | |
1f4b34f8 | 2937 | case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT: |
a2e164e7 VK |
2938 | case HV_X64_MSR_REENLIGHTENMENT_CONTROL: |
2939 | case HV_X64_MSR_TSC_EMULATION_CONTROL: | |
2940 | case HV_X64_MSR_TSC_EMULATION_STATUS: | |
e7d9513b AS |
2941 | return kvm_hv_set_msr_common(vcpu, msr, data, |
2942 | msr_info->host_initiated); | |
91c9c3ed | 2943 | case MSR_IA32_BBL_CR_CTL3: |
2944 | /* Drop writes to this legacy MSR -- see rdmsr | |
2945 | * counterpart for further detail. | |
2946 | */ | |
fab0aa3b EM |
2947 | if (report_ignored_msrs) |
2948 | vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n", | |
2949 | msr, data); | |
91c9c3ed | 2950 | break; |
2b036c6b | 2951 | case MSR_AMD64_OSVW_ID_LENGTH: |
d6321d49 | 2952 | if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW)) |
2b036c6b BO |
2953 | return 1; |
2954 | vcpu->arch.osvw.length = data; | |
2955 | break; | |
2956 | case MSR_AMD64_OSVW_STATUS: | |
d6321d49 | 2957 | if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW)) |
2b036c6b BO |
2958 | return 1; |
2959 | vcpu->arch.osvw.status = data; | |
2960 | break; | |
db2336a8 KH |
2961 | case MSR_PLATFORM_INFO: |
2962 | if (!msr_info->host_initiated || | |
db2336a8 KH |
2963 | (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) && |
2964 | cpuid_fault_enabled(vcpu))) | |
2965 | return 1; | |
2966 | vcpu->arch.msr_platform_info = data; | |
2967 | break; | |
2968 | case MSR_MISC_FEATURES_ENABLES: | |
2969 | if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT || | |
2970 | (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT && | |
2971 | !supports_cpuid_fault(vcpu))) | |
2972 | return 1; | |
2973 | vcpu->arch.msr_misc_features_enables = data; | |
2974 | break; | |
15c4a640 | 2975 | default: |
ffde22ac ES |
2976 | if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr)) |
2977 | return xen_hvm_config(vcpu, data); | |
c6702c9d | 2978 | if (kvm_pmu_is_valid_msr(vcpu, msr)) |
afd80d85 | 2979 | return kvm_pmu_set_msr(vcpu, msr_info); |
ed85c068 | 2980 | if (!ignore_msrs) { |
ae0f5499 | 2981 | vcpu_debug_ratelimited(vcpu, "unhandled wrmsr: 0x%x data 0x%llx\n", |
a737f256 | 2982 | msr, data); |
ed85c068 AP |
2983 | return 1; |
2984 | } else { | |
fab0aa3b EM |
2985 | if (report_ignored_msrs) |
2986 | vcpu_unimpl(vcpu, | |
2987 | "ignored wrmsr: 0x%x data 0x%llx\n", | |
2988 | msr, data); | |
ed85c068 AP |
2989 | break; |
2990 | } | |
15c4a640 CO |
2991 | } |
2992 | return 0; | |
2993 | } | |
2994 | EXPORT_SYMBOL_GPL(kvm_set_msr_common); | |
2995 | ||
44883f01 | 2996 | static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host) |
15c4a640 CO |
2997 | { |
2998 | u64 data; | |
890ca9ae HY |
2999 | u64 mcg_cap = vcpu->arch.mcg_cap; |
3000 | unsigned bank_num = mcg_cap & 0xff; | |
15c4a640 CO |
3001 | |
3002 | switch (msr) { | |
15c4a640 CO |
3003 | case MSR_IA32_P5_MC_ADDR: |
3004 | case MSR_IA32_P5_MC_TYPE: | |
890ca9ae HY |
3005 | data = 0; |
3006 | break; | |
15c4a640 | 3007 | case MSR_IA32_MCG_CAP: |
890ca9ae HY |
3008 | data = vcpu->arch.mcg_cap; |
3009 | break; | |
c7ac679c | 3010 | case MSR_IA32_MCG_CTL: |
44883f01 | 3011 | if (!(mcg_cap & MCG_CTL_P) && !host) |
890ca9ae HY |
3012 | return 1; |
3013 | data = vcpu->arch.mcg_ctl; | |
3014 | break; | |
3015 | case MSR_IA32_MCG_STATUS: | |
3016 | data = vcpu->arch.mcg_status; | |
3017 | break; | |
3018 | default: | |
3019 | if (msr >= MSR_IA32_MC0_CTL && | |
81760dcc | 3020 | msr < MSR_IA32_MCx_CTL(bank_num)) { |
6ec4c5ee MP |
3021 | u32 offset = array_index_nospec( |
3022 | msr - MSR_IA32_MC0_CTL, | |
3023 | MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL); | |
3024 | ||
890ca9ae HY |
3025 | data = vcpu->arch.mce_banks[offset]; |
3026 | break; | |
3027 | } | |
3028 | return 1; | |
3029 | } | |
3030 | *pdata = data; | |
3031 | return 0; | |
3032 | } | |
3033 | ||
609e36d3 | 3034 | int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info) |
890ca9ae | 3035 | { |
609e36d3 | 3036 | switch (msr_info->index) { |
890ca9ae | 3037 | case MSR_IA32_PLATFORM_ID: |
15c4a640 | 3038 | case MSR_IA32_EBL_CR_POWERON: |
b5e2fec0 AG |
3039 | case MSR_IA32_DEBUGCTLMSR: |
3040 | case MSR_IA32_LASTBRANCHFROMIP: | |
3041 | case MSR_IA32_LASTBRANCHTOIP: | |
3042 | case MSR_IA32_LASTINTFROMIP: | |
3043 | case MSR_IA32_LASTINTTOIP: | |
60af2ecd | 3044 | case MSR_K8_SYSCFG: |
3afb1121 PB |
3045 | case MSR_K8_TSEG_ADDR: |
3046 | case MSR_K8_TSEG_MASK: | |
61a6bd67 | 3047 | case MSR_VM_HSAVE_PA: |
1fdbd48c | 3048 | case MSR_K8_INT_PENDING_MSG: |
c323c0e5 | 3049 | case MSR_AMD64_NB_CFG: |
f7c6d140 | 3050 | case MSR_FAM10H_MMIO_CONF_BASE: |
2e32b719 | 3051 | case MSR_AMD64_BU_CFG2: |
0c2df2a1 | 3052 | case MSR_IA32_PERF_CTL: |
405a353a | 3053 | case MSR_AMD64_DC_CFG: |
0e1b869f | 3054 | case MSR_F15H_EX_CFG: |
2ca1a06a VS |
3055 | /* |
3056 | * Intel Sandy Bridge CPUs must support the RAPL (running average power | |
3057 | * limit) MSRs. Just return 0, as we do not want to expose the host | |
3058 | * data here. Do not conditionalize this on CPUID, as KVM does not do | |
3059 | * so for existing CPU-specific MSRs. | |
3060 | */ | |
3061 | case MSR_RAPL_POWER_UNIT: | |
3062 | case MSR_PP0_ENERGY_STATUS: /* Power plane 0 (core) */ | |
3063 | case MSR_PP1_ENERGY_STATUS: /* Power plane 1 (graphics uncore) */ | |
3064 | case MSR_PKG_ENERGY_STATUS: /* Total package */ | |
3065 | case MSR_DRAM_ENERGY_STATUS: /* DRAM controller */ | |
609e36d3 | 3066 | msr_info->data = 0; |
15c4a640 | 3067 | break; |
c51eb52b | 3068 | case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5: |
6912ac32 WH |
3069 | case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3: |
3070 | case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3: | |
3071 | case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1: | |
3072 | case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1: | |
c6702c9d | 3073 | if (kvm_pmu_is_valid_msr(vcpu, msr_info->index)) |
609e36d3 PB |
3074 | return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data); |
3075 | msr_info->data = 0; | |
5753785f | 3076 | break; |
742bc670 | 3077 | case MSR_IA32_UCODE_REV: |
518e7b94 | 3078 | msr_info->data = vcpu->arch.microcode_version; |
742bc670 | 3079 | break; |
0cf9135b SC |
3080 | case MSR_IA32_ARCH_CAPABILITIES: |
3081 | if (!msr_info->host_initiated && | |
3082 | !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES)) | |
3083 | return 1; | |
3084 | msr_info->data = vcpu->arch.arch_capabilities; | |
3085 | break; | |
73f624f4 PB |
3086 | case MSR_IA32_POWER_CTL: |
3087 | msr_info->data = vcpu->arch.msr_ia32_power_ctl; | |
3088 | break; | |
dd259935 PB |
3089 | case MSR_IA32_TSC: |
3090 | msr_info->data = kvm_scale_tsc(vcpu, rdtsc()) + vcpu->arch.tsc_offset; | |
3091 | break; | |
9ba075a6 | 3092 | case MSR_MTRRcap: |
9ba075a6 | 3093 | case 0x200 ... 0x2ff: |
ff53604b | 3094 | return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data); |
15c4a640 | 3095 | case 0xcd: /* fsb frequency */ |
609e36d3 | 3096 | msr_info->data = 3; |
15c4a640 | 3097 | break; |
7b914098 JS |
3098 | /* |
3099 | * MSR_EBC_FREQUENCY_ID | |
3100 | * Conservative value valid for even the basic CPU models. | |
3101 | * Models 0,1: 000 in bits 23:21 indicating a bus speed of | |
3102 | * 100MHz, model 2 000 in bits 18:16 indicating 100MHz, | |
3103 | * and 266MHz for model 3, or 4. Set Core Clock | |
3104 | * Frequency to System Bus Frequency Ratio to 1 (bits | |
3105 | * 31:24) even though these are only valid for CPU | |
3106 | * models > 2, however guests may end up dividing or | |
3107 | * multiplying by zero otherwise. | |
3108 | */ | |
3109 | case MSR_EBC_FREQUENCY_ID: | |
609e36d3 | 3110 | msr_info->data = 1 << 24; |
7b914098 | 3111 | break; |
15c4a640 | 3112 | case MSR_IA32_APICBASE: |
609e36d3 | 3113 | msr_info->data = kvm_get_apic_base(vcpu); |
15c4a640 | 3114 | break; |
0105d1a5 | 3115 | case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff: |
609e36d3 | 3116 | return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data); |
a3e06bbe | 3117 | case MSR_IA32_TSCDEADLINE: |
609e36d3 | 3118 | msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu); |
a3e06bbe | 3119 | break; |
ba904635 | 3120 | case MSR_IA32_TSC_ADJUST: |
609e36d3 | 3121 | msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr; |
ba904635 | 3122 | break; |
15c4a640 | 3123 | case MSR_IA32_MISC_ENABLE: |
609e36d3 | 3124 | msr_info->data = vcpu->arch.ia32_misc_enable_msr; |
15c4a640 | 3125 | break; |
64d60670 PB |
3126 | case MSR_IA32_SMBASE: |
3127 | if (!msr_info->host_initiated) | |
3128 | return 1; | |
3129 | msr_info->data = vcpu->arch.smbase; | |
15c4a640 | 3130 | break; |
52797bf9 LA |
3131 | case MSR_SMI_COUNT: |
3132 | msr_info->data = vcpu->arch.smi_count; | |
3133 | break; | |
847f0ad8 AG |
3134 | case MSR_IA32_PERF_STATUS: |
3135 | /* TSC increment by tick */ | |
609e36d3 | 3136 | msr_info->data = 1000ULL; |
847f0ad8 | 3137 | /* CPU multiplier */ |
b0996ae4 | 3138 | msr_info->data |= (((uint64_t)4ULL) << 40); |
847f0ad8 | 3139 | break; |
15c4a640 | 3140 | case MSR_EFER: |
609e36d3 | 3141 | msr_info->data = vcpu->arch.efer; |
15c4a640 | 3142 | break; |
18068523 | 3143 | case MSR_KVM_WALL_CLOCK: |
11c6bffa | 3144 | case MSR_KVM_WALL_CLOCK_NEW: |
609e36d3 | 3145 | msr_info->data = vcpu->kvm->arch.wall_clock; |
18068523 GOC |
3146 | break; |
3147 | case MSR_KVM_SYSTEM_TIME: | |
11c6bffa | 3148 | case MSR_KVM_SYSTEM_TIME_NEW: |
609e36d3 | 3149 | msr_info->data = vcpu->arch.time; |
18068523 | 3150 | break; |
344d9588 | 3151 | case MSR_KVM_ASYNC_PF_EN: |
609e36d3 | 3152 | msr_info->data = vcpu->arch.apf.msr_val; |
344d9588 | 3153 | break; |
c9aaa895 | 3154 | case MSR_KVM_STEAL_TIME: |
609e36d3 | 3155 | msr_info->data = vcpu->arch.st.msr_val; |
c9aaa895 | 3156 | break; |
1d92128f | 3157 | case MSR_KVM_PV_EOI_EN: |
609e36d3 | 3158 | msr_info->data = vcpu->arch.pv_eoi.msr_val; |
1d92128f | 3159 | break; |
2d5ba19b MT |
3160 | case MSR_KVM_POLL_CONTROL: |
3161 | msr_info->data = vcpu->arch.msr_kvm_poll_control; | |
3162 | break; | |
890ca9ae HY |
3163 | case MSR_IA32_P5_MC_ADDR: |
3164 | case MSR_IA32_P5_MC_TYPE: | |
3165 | case MSR_IA32_MCG_CAP: | |
3166 | case MSR_IA32_MCG_CTL: | |
3167 | case MSR_IA32_MCG_STATUS: | |
81760dcc | 3168 | case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1: |
44883f01 PB |
3169 | return get_msr_mce(vcpu, msr_info->index, &msr_info->data, |
3170 | msr_info->host_initiated); | |
864e2ab2 AL |
3171 | case MSR_IA32_XSS: |
3172 | if (!msr_info->host_initiated && | |
3173 | !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES)) | |
3174 | return 1; | |
3175 | msr_info->data = vcpu->arch.ia32_xss; | |
3176 | break; | |
84e0cefa JS |
3177 | case MSR_K7_CLK_CTL: |
3178 | /* | |
3179 | * Provide expected ramp-up count for K7. All other | |
3180 | * are set to zero, indicating minimum divisors for | |
3181 | * every field. | |
3182 | * | |
3183 | * This prevents guest kernels on AMD host with CPU | |
3184 | * type 6, model 8 and higher from exploding due to | |
3185 | * the rdmsr failing. | |
3186 | */ | |
609e36d3 | 3187 | msr_info->data = 0x20000000; |
84e0cefa | 3188 | break; |
55cd8e5a | 3189 | case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15: |
e7d9513b AS |
3190 | case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4: |
3191 | case HV_X64_MSR_CRASH_CTL: | |
1f4b34f8 | 3192 | case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT: |
a2e164e7 VK |
3193 | case HV_X64_MSR_REENLIGHTENMENT_CONTROL: |
3194 | case HV_X64_MSR_TSC_EMULATION_CONTROL: | |
3195 | case HV_X64_MSR_TSC_EMULATION_STATUS: | |
e83d5887 | 3196 | return kvm_hv_get_msr_common(vcpu, |
44883f01 PB |
3197 | msr_info->index, &msr_info->data, |
3198 | msr_info->host_initiated); | |
91c9c3ed | 3199 | case MSR_IA32_BBL_CR_CTL3: |
3200 | /* This legacy MSR exists but isn't fully documented in current | |
3201 | * silicon. It is however accessed by winxp in very narrow | |
3202 | * scenarios where it sets bit #19, itself documented as | |
3203 | * a "reserved" bit. Best effort attempt to source coherent | |
3204 | * read data here should the balance of the register be | |
3205 | * interpreted by the guest: | |
3206 | * | |
3207 | * L2 cache control register 3: 64GB range, 256KB size, | |
3208 | * enabled, latency 0x1, configured | |
3209 | */ | |
609e36d3 | 3210 | msr_info->data = 0xbe702111; |
91c9c3ed | 3211 | break; |
2b036c6b | 3212 | case MSR_AMD64_OSVW_ID_LENGTH: |
d6321d49 | 3213 | if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW)) |
2b036c6b | 3214 | return 1; |
609e36d3 | 3215 | msr_info->data = vcpu->arch.osvw.length; |
2b036c6b BO |
3216 | break; |
3217 | case MSR_AMD64_OSVW_STATUS: | |
d6321d49 | 3218 | if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW)) |
2b036c6b | 3219 | return 1; |
609e36d3 | 3220 | msr_info->data = vcpu->arch.osvw.status; |
2b036c6b | 3221 | break; |
db2336a8 | 3222 | case MSR_PLATFORM_INFO: |
6fbbde9a DS |
3223 | if (!msr_info->host_initiated && |
3224 | !vcpu->kvm->arch.guest_can_read_msr_platform_info) | |
3225 | return 1; | |
db2336a8 KH |
3226 | msr_info->data = vcpu->arch.msr_platform_info; |
3227 | break; | |
3228 | case MSR_MISC_FEATURES_ENABLES: | |
3229 | msr_info->data = vcpu->arch.msr_misc_features_enables; | |
3230 | break; | |
191c8137 BP |
3231 | case MSR_K7_HWCR: |
3232 | msr_info->data = vcpu->arch.msr_hwcr; | |
3233 | break; | |
15c4a640 | 3234 | default: |
c6702c9d | 3235 | if (kvm_pmu_is_valid_msr(vcpu, msr_info->index)) |
609e36d3 | 3236 | return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data); |
ed85c068 | 3237 | if (!ignore_msrs) { |
ae0f5499 BD |
3238 | vcpu_debug_ratelimited(vcpu, "unhandled rdmsr: 0x%x\n", |
3239 | msr_info->index); | |
ed85c068 AP |
3240 | return 1; |
3241 | } else { | |
fab0aa3b EM |
3242 | if (report_ignored_msrs) |
3243 | vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n", | |
3244 | msr_info->index); | |
609e36d3 | 3245 | msr_info->data = 0; |
ed85c068 AP |
3246 | } |
3247 | break; | |
15c4a640 | 3248 | } |
15c4a640 CO |
3249 | return 0; |
3250 | } | |
3251 | EXPORT_SYMBOL_GPL(kvm_get_msr_common); | |
3252 | ||
313a3dc7 CO |
3253 | /* |
3254 | * Read or write a bunch of msrs. All parameters are kernel addresses. | |
3255 | * | |
3256 | * @return number of msrs set successfully. | |
3257 | */ | |
3258 | static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs, | |
3259 | struct kvm_msr_entry *entries, | |
3260 | int (*do_msr)(struct kvm_vcpu *vcpu, | |
3261 | unsigned index, u64 *data)) | |
3262 | { | |
801e459a | 3263 | int i; |
313a3dc7 | 3264 | |
313a3dc7 CO |
3265 | for (i = 0; i < msrs->nmsrs; ++i) |
3266 | if (do_msr(vcpu, entries[i].index, &entries[i].data)) | |
3267 | break; | |
3268 | ||
313a3dc7 CO |
3269 | return i; |
3270 | } | |
3271 | ||
3272 | /* | |
3273 | * Read or write a bunch of msrs. Parameters are user addresses. | |
3274 | * | |
3275 | * @return number of msrs set successfully. | |
3276 | */ | |
3277 | static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs, | |
3278 | int (*do_msr)(struct kvm_vcpu *vcpu, | |
3279 | unsigned index, u64 *data), | |
3280 | int writeback) | |
3281 | { | |
3282 | struct kvm_msrs msrs; | |
3283 | struct kvm_msr_entry *entries; | |
3284 | int r, n; | |
3285 | unsigned size; | |
3286 | ||
3287 | r = -EFAULT; | |
0e96f31e | 3288 | if (copy_from_user(&msrs, user_msrs, sizeof(msrs))) |
313a3dc7 CO |
3289 | goto out; |
3290 | ||
3291 | r = -E2BIG; | |
3292 | if (msrs.nmsrs >= MAX_IO_MSRS) | |
3293 | goto out; | |
3294 | ||
313a3dc7 | 3295 | size = sizeof(struct kvm_msr_entry) * msrs.nmsrs; |
ff5c2c03 SL |
3296 | entries = memdup_user(user_msrs->entries, size); |
3297 | if (IS_ERR(entries)) { | |
3298 | r = PTR_ERR(entries); | |
313a3dc7 | 3299 | goto out; |
ff5c2c03 | 3300 | } |
313a3dc7 CO |
3301 | |
3302 | r = n = __msr_io(vcpu, &msrs, entries, do_msr); | |
3303 | if (r < 0) | |
3304 | goto out_free; | |
3305 | ||
3306 | r = -EFAULT; | |
3307 | if (writeback && copy_to_user(user_msrs->entries, entries, size)) | |
3308 | goto out_free; | |
3309 | ||
3310 | r = n; | |
3311 | ||
3312 | out_free: | |
7a73c028 | 3313 | kfree(entries); |
313a3dc7 CO |
3314 | out: |
3315 | return r; | |
3316 | } | |
3317 | ||
4d5422ce WL |
3318 | static inline bool kvm_can_mwait_in_guest(void) |
3319 | { | |
3320 | return boot_cpu_has(X86_FEATURE_MWAIT) && | |
8e9b29b6 KA |
3321 | !boot_cpu_has_bug(X86_BUG_MONITOR) && |
3322 | boot_cpu_has(X86_FEATURE_ARAT); | |
4d5422ce WL |
3323 | } |
3324 | ||
784aa3d7 | 3325 | int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext) |
018d00d2 | 3326 | { |
4d5422ce | 3327 | int r = 0; |
018d00d2 ZX |
3328 | |
3329 | switch (ext) { | |
3330 | case KVM_CAP_IRQCHIP: | |
3331 | case KVM_CAP_HLT: | |
3332 | case KVM_CAP_MMU_SHADOW_CACHE_CONTROL: | |
018d00d2 | 3333 | case KVM_CAP_SET_TSS_ADDR: |
07716717 | 3334 | case KVM_CAP_EXT_CPUID: |
9c15bb1d | 3335 | case KVM_CAP_EXT_EMUL_CPUID: |
c8076604 | 3336 | case KVM_CAP_CLOCKSOURCE: |
7837699f | 3337 | case KVM_CAP_PIT: |
a28e4f5a | 3338 | case KVM_CAP_NOP_IO_DELAY: |
62d9f0db | 3339 | case KVM_CAP_MP_STATE: |
ed848624 | 3340 | case KVM_CAP_SYNC_MMU: |
a355c85c | 3341 | case KVM_CAP_USER_NMI: |
52d939a0 | 3342 | case KVM_CAP_REINJECT_CONTROL: |
4925663a | 3343 | case KVM_CAP_IRQ_INJECT_STATUS: |
d34e6b17 | 3344 | case KVM_CAP_IOEVENTFD: |
f848a5a8 | 3345 | case KVM_CAP_IOEVENTFD_NO_LENGTH: |
c5ff41ce | 3346 | case KVM_CAP_PIT2: |
e9f42757 | 3347 | case KVM_CAP_PIT_STATE2: |
b927a3ce | 3348 | case KVM_CAP_SET_IDENTITY_MAP_ADDR: |
ffde22ac | 3349 | case KVM_CAP_XEN_HVM: |
3cfc3092 | 3350 | case KVM_CAP_VCPU_EVENTS: |
55cd8e5a | 3351 | case KVM_CAP_HYPERV: |
10388a07 | 3352 | case KVM_CAP_HYPERV_VAPIC: |
c25bc163 | 3353 | case KVM_CAP_HYPERV_SPIN: |
5c919412 | 3354 | case KVM_CAP_HYPERV_SYNIC: |
efc479e6 | 3355 | case KVM_CAP_HYPERV_SYNIC2: |
d3457c87 | 3356 | case KVM_CAP_HYPERV_VP_INDEX: |
faeb7833 | 3357 | case KVM_CAP_HYPERV_EVENTFD: |
c1aea919 | 3358 | case KVM_CAP_HYPERV_TLBFLUSH: |
214ff83d | 3359 | case KVM_CAP_HYPERV_SEND_IPI: |
2bc39970 | 3360 | case KVM_CAP_HYPERV_CPUID: |
ab9f4ecb | 3361 | case KVM_CAP_PCI_SEGMENT: |
a1efbe77 | 3362 | case KVM_CAP_DEBUGREGS: |
d2be1651 | 3363 | case KVM_CAP_X86_ROBUST_SINGLESTEP: |
2d5b5a66 | 3364 | case KVM_CAP_XSAVE: |
344d9588 | 3365 | case KVM_CAP_ASYNC_PF: |
92a1f12d | 3366 | case KVM_CAP_GET_TSC_KHZ: |
1c0b28c2 | 3367 | case KVM_CAP_KVMCLOCK_CTRL: |
4d8b81ab | 3368 | case KVM_CAP_READONLY_MEM: |
5f66b620 | 3369 | case KVM_CAP_HYPERV_TIME: |
100943c5 | 3370 | case KVM_CAP_IOAPIC_POLARITY_IGNORED: |
defcf51f | 3371 | case KVM_CAP_TSC_DEADLINE_TIMER: |
90de4a18 | 3372 | case KVM_CAP_DISABLE_QUIRKS: |
d71ba788 | 3373 | case KVM_CAP_SET_BOOT_CPU_ID: |
49df6397 | 3374 | case KVM_CAP_SPLIT_IRQCHIP: |
460df4c1 | 3375 | case KVM_CAP_IMMEDIATE_EXIT: |
66bb8a06 | 3376 | case KVM_CAP_PMU_EVENT_FILTER: |
801e459a | 3377 | case KVM_CAP_GET_MSR_FEATURES: |
6fbbde9a | 3378 | case KVM_CAP_MSR_PLATFORM_INFO: |
c4f55198 | 3379 | case KVM_CAP_EXCEPTION_PAYLOAD: |
b9b2782c | 3380 | case KVM_CAP_SET_GUEST_DEBUG: |
018d00d2 ZX |
3381 | r = 1; |
3382 | break; | |
01643c51 KH |
3383 | case KVM_CAP_SYNC_REGS: |
3384 | r = KVM_SYNC_X86_VALID_FIELDS; | |
3385 | break; | |
e3fd9a93 PB |
3386 | case KVM_CAP_ADJUST_CLOCK: |
3387 | r = KVM_CLOCK_TSC_STABLE; | |
3388 | break; | |
4d5422ce | 3389 | case KVM_CAP_X86_DISABLE_EXITS: |
b5170063 WL |
3390 | r |= KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE | |
3391 | KVM_X86_DISABLE_EXITS_CSTATE; | |
4d5422ce WL |
3392 | if(kvm_can_mwait_in_guest()) |
3393 | r |= KVM_X86_DISABLE_EXITS_MWAIT; | |
668fffa3 | 3394 | break; |
6d396b55 PB |
3395 | case KVM_CAP_X86_SMM: |
3396 | /* SMBASE is usually relocated above 1M on modern chipsets, | |
3397 | * and SMM handlers might indeed rely on 4G segment limits, | |
3398 | * so do not report SMM to be available if real mode is | |
3399 | * emulated via vm86 mode. Still, do not go to great lengths | |
3400 | * to avoid userspace's usage of the feature, because it is a | |
3401 | * fringe case that is not enabled except via specific settings | |
3402 | * of the module parameters. | |
3403 | */ | |
afaf0b2f | 3404 | r = kvm_x86_ops.has_emulated_msr(MSR_IA32_SMBASE); |
6d396b55 | 3405 | break; |
774ead3a | 3406 | case KVM_CAP_VAPIC: |
afaf0b2f | 3407 | r = !kvm_x86_ops.cpu_has_accelerated_tpr(); |
774ead3a | 3408 | break; |
f725230a | 3409 | case KVM_CAP_NR_VCPUS: |
8c3ba334 SL |
3410 | r = KVM_SOFT_MAX_VCPUS; |
3411 | break; | |
3412 | case KVM_CAP_MAX_VCPUS: | |
f725230a AK |
3413 | r = KVM_MAX_VCPUS; |
3414 | break; | |
a86cb413 TH |
3415 | case KVM_CAP_MAX_VCPU_ID: |
3416 | r = KVM_MAX_VCPU_ID; | |
3417 | break; | |
a68a6a72 MT |
3418 | case KVM_CAP_PV_MMU: /* obsolete */ |
3419 | r = 0; | |
2f333bcb | 3420 | break; |
890ca9ae HY |
3421 | case KVM_CAP_MCE: |
3422 | r = KVM_MAX_MCE_BANKS; | |
3423 | break; | |
2d5b5a66 | 3424 | case KVM_CAP_XCRS: |
d366bf7e | 3425 | r = boot_cpu_has(X86_FEATURE_XSAVE); |
2d5b5a66 | 3426 | break; |
92a1f12d JR |
3427 | case KVM_CAP_TSC_CONTROL: |
3428 | r = kvm_has_tsc_control; | |
3429 | break; | |
37131313 RK |
3430 | case KVM_CAP_X2APIC_API: |
3431 | r = KVM_X2APIC_API_VALID_FLAGS; | |
3432 | break; | |
8fcc4b59 | 3433 | case KVM_CAP_NESTED_STATE: |
afaf0b2f SC |
3434 | r = kvm_x86_ops.get_nested_state ? |
3435 | kvm_x86_ops.get_nested_state(NULL, NULL, 0) : 0; | |
8fcc4b59 | 3436 | break; |
344c6c80 | 3437 | case KVM_CAP_HYPERV_DIRECT_TLBFLUSH: |
afaf0b2f | 3438 | r = kvm_x86_ops.enable_direct_tlbflush != NULL; |
5a0165f6 VK |
3439 | break; |
3440 | case KVM_CAP_HYPERV_ENLIGHTENED_VMCS: | |
afaf0b2f | 3441 | r = kvm_x86_ops.nested_enable_evmcs != NULL; |
344c6c80 | 3442 | break; |
018d00d2 | 3443 | default: |
018d00d2 ZX |
3444 | break; |
3445 | } | |
3446 | return r; | |
3447 | ||
3448 | } | |
3449 | ||
043405e1 CO |
3450 | long kvm_arch_dev_ioctl(struct file *filp, |
3451 | unsigned int ioctl, unsigned long arg) | |
3452 | { | |
3453 | void __user *argp = (void __user *)arg; | |
3454 | long r; | |
3455 | ||
3456 | switch (ioctl) { | |
3457 | case KVM_GET_MSR_INDEX_LIST: { | |
3458 | struct kvm_msr_list __user *user_msr_list = argp; | |
3459 | struct kvm_msr_list msr_list; | |
3460 | unsigned n; | |
3461 | ||
3462 | r = -EFAULT; | |
0e96f31e | 3463 | if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list))) |
043405e1 CO |
3464 | goto out; |
3465 | n = msr_list.nmsrs; | |
62ef68bb | 3466 | msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs; |
0e96f31e | 3467 | if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list))) |
043405e1 CO |
3468 | goto out; |
3469 | r = -E2BIG; | |
e125e7b6 | 3470 | if (n < msr_list.nmsrs) |
043405e1 CO |
3471 | goto out; |
3472 | r = -EFAULT; | |
3473 | if (copy_to_user(user_msr_list->indices, &msrs_to_save, | |
3474 | num_msrs_to_save * sizeof(u32))) | |
3475 | goto out; | |
e125e7b6 | 3476 | if (copy_to_user(user_msr_list->indices + num_msrs_to_save, |
043405e1 | 3477 | &emulated_msrs, |
62ef68bb | 3478 | num_emulated_msrs * sizeof(u32))) |
043405e1 CO |
3479 | goto out; |
3480 | r = 0; | |
3481 | break; | |
3482 | } | |
9c15bb1d BP |
3483 | case KVM_GET_SUPPORTED_CPUID: |
3484 | case KVM_GET_EMULATED_CPUID: { | |
674eea0f AK |
3485 | struct kvm_cpuid2 __user *cpuid_arg = argp; |
3486 | struct kvm_cpuid2 cpuid; | |
3487 | ||
3488 | r = -EFAULT; | |
0e96f31e | 3489 | if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid))) |
674eea0f | 3490 | goto out; |
9c15bb1d BP |
3491 | |
3492 | r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries, | |
3493 | ioctl); | |
674eea0f AK |
3494 | if (r) |
3495 | goto out; | |
3496 | ||
3497 | r = -EFAULT; | |
0e96f31e | 3498 | if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid))) |
674eea0f AK |
3499 | goto out; |
3500 | r = 0; | |
3501 | break; | |
3502 | } | |
cf6c26ec | 3503 | case KVM_X86_GET_MCE_CAP_SUPPORTED: |
890ca9ae | 3504 | r = -EFAULT; |
c45dcc71 AR |
3505 | if (copy_to_user(argp, &kvm_mce_cap_supported, |
3506 | sizeof(kvm_mce_cap_supported))) | |
890ca9ae HY |
3507 | goto out; |
3508 | r = 0; | |
3509 | break; | |
801e459a TL |
3510 | case KVM_GET_MSR_FEATURE_INDEX_LIST: { |
3511 | struct kvm_msr_list __user *user_msr_list = argp; | |
3512 | struct kvm_msr_list msr_list; | |
3513 | unsigned int n; | |
3514 | ||
3515 | r = -EFAULT; | |
3516 | if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list))) | |
3517 | goto out; | |
3518 | n = msr_list.nmsrs; | |
3519 | msr_list.nmsrs = num_msr_based_features; | |
3520 | if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list))) | |
3521 | goto out; | |
3522 | r = -E2BIG; | |
3523 | if (n < msr_list.nmsrs) | |
3524 | goto out; | |
3525 | r = -EFAULT; | |
3526 | if (copy_to_user(user_msr_list->indices, &msr_based_features, | |
3527 | num_msr_based_features * sizeof(u32))) | |
3528 | goto out; | |
3529 | r = 0; | |
3530 | break; | |
3531 | } | |
3532 | case KVM_GET_MSRS: | |
3533 | r = msr_io(NULL, argp, do_get_msr_feature, 1); | |
3534 | break; | |
043405e1 CO |
3535 | default: |
3536 | r = -EINVAL; | |
cf6c26ec | 3537 | break; |
043405e1 CO |
3538 | } |
3539 | out: | |
3540 | return r; | |
3541 | } | |
3542 | ||
f5f48ee1 SY |
3543 | static void wbinvd_ipi(void *garbage) |
3544 | { | |
3545 | wbinvd(); | |
3546 | } | |
3547 | ||
3548 | static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu) | |
3549 | { | |
e0f0bbc5 | 3550 | return kvm_arch_has_noncoherent_dma(vcpu->kvm); |
f5f48ee1 SY |
3551 | } |
3552 | ||
313a3dc7 CO |
3553 | void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu) |
3554 | { | |
f5f48ee1 SY |
3555 | /* Address WBINVD may be executed by guest */ |
3556 | if (need_emulate_wbinvd(vcpu)) { | |
afaf0b2f | 3557 | if (kvm_x86_ops.has_wbinvd_exit()) |
f5f48ee1 SY |
3558 | cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask); |
3559 | else if (vcpu->cpu != -1 && vcpu->cpu != cpu) | |
3560 | smp_call_function_single(vcpu->cpu, | |
3561 | wbinvd_ipi, NULL, 1); | |
3562 | } | |
3563 | ||
afaf0b2f | 3564 | kvm_x86_ops.vcpu_load(vcpu, cpu); |
8f6055cb | 3565 | |
37486135 BM |
3566 | /* Save host pkru register if supported */ |
3567 | vcpu->arch.host_pkru = read_pkru(); | |
3568 | ||
0dd6a6ed ZA |
3569 | /* Apply any externally detected TSC adjustments (due to suspend) */ |
3570 | if (unlikely(vcpu->arch.tsc_offset_adjustment)) { | |
3571 | adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment); | |
3572 | vcpu->arch.tsc_offset_adjustment = 0; | |
105b21bb | 3573 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu); |
0dd6a6ed | 3574 | } |
8f6055cb | 3575 | |
b0c39dc6 | 3576 | if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) { |
6f526ec5 | 3577 | s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 : |
4ea1636b | 3578 | rdtsc() - vcpu->arch.last_host_tsc; |
e48672fa ZA |
3579 | if (tsc_delta < 0) |
3580 | mark_tsc_unstable("KVM discovered backwards TSC"); | |
ce7a058a | 3581 | |
b0c39dc6 | 3582 | if (kvm_check_tsc_unstable()) { |
07c1419a | 3583 | u64 offset = kvm_compute_tsc_offset(vcpu, |
b183aa58 | 3584 | vcpu->arch.last_guest_tsc); |
a545ab6a | 3585 | kvm_vcpu_write_tsc_offset(vcpu, offset); |
c285545f | 3586 | vcpu->arch.tsc_catchup = 1; |
c285545f | 3587 | } |
a749e247 PB |
3588 | |
3589 | if (kvm_lapic_hv_timer_in_use(vcpu)) | |
3590 | kvm_lapic_restart_hv_timer(vcpu); | |
3591 | ||
d98d07ca MT |
3592 | /* |
3593 | * On a host with synchronized TSC, there is no need to update | |
3594 | * kvmclock on vcpu->cpu migration | |
3595 | */ | |
3596 | if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1) | |
0061d53d | 3597 | kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu); |
c285545f | 3598 | if (vcpu->cpu != cpu) |
1bd2009e | 3599 | kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu); |
e48672fa | 3600 | vcpu->cpu = cpu; |
6b7d7e76 | 3601 | } |
c9aaa895 | 3602 | |
c9aaa895 | 3603 | kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu); |
313a3dc7 CO |
3604 | } |
3605 | ||
0b9f6c46 PX |
3606 | static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu) |
3607 | { | |
b0431382 BO |
3608 | struct kvm_host_map map; |
3609 | struct kvm_steal_time *st; | |
3610 | ||
0b9f6c46 PX |
3611 | if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED)) |
3612 | return; | |
3613 | ||
a6bd811f | 3614 | if (vcpu->arch.st.preempted) |
8c6de56a BO |
3615 | return; |
3616 | ||
b0431382 BO |
3617 | if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT, &map, |
3618 | &vcpu->arch.st.cache, true)) | |
3619 | return; | |
3620 | ||
3621 | st = map.hva + | |
3622 | offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS); | |
0b9f6c46 | 3623 | |
a6bd811f | 3624 | st->preempted = vcpu->arch.st.preempted = KVM_VCPU_PREEMPTED; |
0b9f6c46 | 3625 | |
b0431382 | 3626 | kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, true); |
0b9f6c46 PX |
3627 | } |
3628 | ||
313a3dc7 CO |
3629 | void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu) |
3630 | { | |
cc0d907c | 3631 | int idx; |
de63ad4c LM |
3632 | |
3633 | if (vcpu->preempted) | |
afaf0b2f | 3634 | vcpu->arch.preempted_in_kernel = !kvm_x86_ops.get_cpl(vcpu); |
de63ad4c | 3635 | |
931f261b AA |
3636 | /* |
3637 | * Disable page faults because we're in atomic context here. | |
3638 | * kvm_write_guest_offset_cached() would call might_fault() | |
3639 | * that relies on pagefault_disable() to tell if there's a | |
3640 | * bug. NOTE: the write to guest memory may not go through if | |
3641 | * during postcopy live migration or if there's heavy guest | |
3642 | * paging. | |
3643 | */ | |
3644 | pagefault_disable(); | |
cc0d907c AA |
3645 | /* |
3646 | * kvm_memslots() will be called by | |
3647 | * kvm_write_guest_offset_cached() so take the srcu lock. | |
3648 | */ | |
3649 | idx = srcu_read_lock(&vcpu->kvm->srcu); | |
0b9f6c46 | 3650 | kvm_steal_time_set_preempted(vcpu); |
cc0d907c | 3651 | srcu_read_unlock(&vcpu->kvm->srcu, idx); |
931f261b | 3652 | pagefault_enable(); |
afaf0b2f | 3653 | kvm_x86_ops.vcpu_put(vcpu); |
4ea1636b | 3654 | vcpu->arch.last_host_tsc = rdtsc(); |
efdab992 | 3655 | /* |
f9dcf08e RK |
3656 | * If userspace has set any breakpoints or watchpoints, dr6 is restored |
3657 | * on every vmexit, but if not, we might have a stale dr6 from the | |
3658 | * guest. do_debug expects dr6 to be cleared after it runs, do the same. | |
efdab992 | 3659 | */ |
f9dcf08e | 3660 | set_debugreg(0, 6); |
313a3dc7 CO |
3661 | } |
3662 | ||
313a3dc7 CO |
3663 | static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu, |
3664 | struct kvm_lapic_state *s) | |
3665 | { | |
fa59cc00 | 3666 | if (vcpu->arch.apicv_active) |
afaf0b2f | 3667 | kvm_x86_ops.sync_pir_to_irr(vcpu); |
d62caabb | 3668 | |
a92e2543 | 3669 | return kvm_apic_get_state(vcpu, s); |
313a3dc7 CO |
3670 | } |
3671 | ||
3672 | static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu, | |
3673 | struct kvm_lapic_state *s) | |
3674 | { | |
a92e2543 RK |
3675 | int r; |
3676 | ||
3677 | r = kvm_apic_set_state(vcpu, s); | |
3678 | if (r) | |
3679 | return r; | |
cb142eb7 | 3680 | update_cr8_intercept(vcpu); |
313a3dc7 CO |
3681 | |
3682 | return 0; | |
3683 | } | |
3684 | ||
127a457a MG |
3685 | static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu) |
3686 | { | |
3687 | return (!lapic_in_kernel(vcpu) || | |
3688 | kvm_apic_accept_pic_intr(vcpu)); | |
3689 | } | |
3690 | ||
782d422b MG |
3691 | /* |
3692 | * if userspace requested an interrupt window, check that the | |
3693 | * interrupt window is open. | |
3694 | * | |
3695 | * No need to exit to userspace if we already have an interrupt queued. | |
3696 | */ | |
3697 | static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu) | |
3698 | { | |
3699 | return kvm_arch_interrupt_allowed(vcpu) && | |
3700 | !kvm_cpu_has_interrupt(vcpu) && | |
3701 | !kvm_event_needs_reinjection(vcpu) && | |
3702 | kvm_cpu_accept_dm_intr(vcpu); | |
3703 | } | |
3704 | ||
f77bc6a4 ZX |
3705 | static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, |
3706 | struct kvm_interrupt *irq) | |
3707 | { | |
02cdb50f | 3708 | if (irq->irq >= KVM_NR_INTERRUPTS) |
f77bc6a4 | 3709 | return -EINVAL; |
1c1a9ce9 SR |
3710 | |
3711 | if (!irqchip_in_kernel(vcpu->kvm)) { | |
3712 | kvm_queue_interrupt(vcpu, irq->irq, false); | |
3713 | kvm_make_request(KVM_REQ_EVENT, vcpu); | |
3714 | return 0; | |
3715 | } | |
3716 | ||
3717 | /* | |
3718 | * With in-kernel LAPIC, we only use this to inject EXTINT, so | |
3719 | * fail for in-kernel 8259. | |
3720 | */ | |
3721 | if (pic_in_kernel(vcpu->kvm)) | |
f77bc6a4 | 3722 | return -ENXIO; |
f77bc6a4 | 3723 | |
1c1a9ce9 SR |
3724 | if (vcpu->arch.pending_external_vector != -1) |
3725 | return -EEXIST; | |
f77bc6a4 | 3726 | |
1c1a9ce9 | 3727 | vcpu->arch.pending_external_vector = irq->irq; |
934bf653 | 3728 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
f77bc6a4 ZX |
3729 | return 0; |
3730 | } | |
3731 | ||
c4abb7c9 JK |
3732 | static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu) |
3733 | { | |
c4abb7c9 | 3734 | kvm_inject_nmi(vcpu); |
c4abb7c9 JK |
3735 | |
3736 | return 0; | |
3737 | } | |
3738 | ||
f077825a PB |
3739 | static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu) |
3740 | { | |
64d60670 PB |
3741 | kvm_make_request(KVM_REQ_SMI, vcpu); |
3742 | ||
f077825a PB |
3743 | return 0; |
3744 | } | |
3745 | ||
b209749f AK |
3746 | static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu, |
3747 | struct kvm_tpr_access_ctl *tac) | |
3748 | { | |
3749 | if (tac->flags) | |
3750 | return -EINVAL; | |
3751 | vcpu->arch.tpr_access_reporting = !!tac->enabled; | |
3752 | return 0; | |
3753 | } | |
3754 | ||
890ca9ae HY |
3755 | static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu, |
3756 | u64 mcg_cap) | |
3757 | { | |
3758 | int r; | |
3759 | unsigned bank_num = mcg_cap & 0xff, bank; | |
3760 | ||
3761 | r = -EINVAL; | |
c4e0e4ab | 3762 | if (!bank_num || bank_num > KVM_MAX_MCE_BANKS) |
890ca9ae | 3763 | goto out; |
c45dcc71 | 3764 | if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000)) |
890ca9ae HY |
3765 | goto out; |
3766 | r = 0; | |
3767 | vcpu->arch.mcg_cap = mcg_cap; | |
3768 | /* Init IA32_MCG_CTL to all 1s */ | |
3769 | if (mcg_cap & MCG_CTL_P) | |
3770 | vcpu->arch.mcg_ctl = ~(u64)0; | |
3771 | /* Init IA32_MCi_CTL to all 1s */ | |
3772 | for (bank = 0; bank < bank_num; bank++) | |
3773 | vcpu->arch.mce_banks[bank*4] = ~(u64)0; | |
c45dcc71 | 3774 | |
afaf0b2f | 3775 | kvm_x86_ops.setup_mce(vcpu); |
890ca9ae HY |
3776 | out: |
3777 | return r; | |
3778 | } | |
3779 | ||
3780 | static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu, | |
3781 | struct kvm_x86_mce *mce) | |
3782 | { | |
3783 | u64 mcg_cap = vcpu->arch.mcg_cap; | |
3784 | unsigned bank_num = mcg_cap & 0xff; | |
3785 | u64 *banks = vcpu->arch.mce_banks; | |
3786 | ||
3787 | if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL)) | |
3788 | return -EINVAL; | |
3789 | /* | |
3790 | * if IA32_MCG_CTL is not all 1s, the uncorrected error | |
3791 | * reporting is disabled | |
3792 | */ | |
3793 | if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) && | |
3794 | vcpu->arch.mcg_ctl != ~(u64)0) | |
3795 | return 0; | |
3796 | banks += 4 * mce->bank; | |
3797 | /* | |
3798 | * if IA32_MCi_CTL is not all 1s, the uncorrected error | |
3799 | * reporting is disabled for the bank | |
3800 | */ | |
3801 | if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0) | |
3802 | return 0; | |
3803 | if (mce->status & MCI_STATUS_UC) { | |
3804 | if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) || | |
fc78f519 | 3805 | !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) { |
a8eeb04a | 3806 | kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu); |
890ca9ae HY |
3807 | return 0; |
3808 | } | |
3809 | if (banks[1] & MCI_STATUS_VAL) | |
3810 | mce->status |= MCI_STATUS_OVER; | |
3811 | banks[2] = mce->addr; | |
3812 | banks[3] = mce->misc; | |
3813 | vcpu->arch.mcg_status = mce->mcg_status; | |
3814 | banks[1] = mce->status; | |
3815 | kvm_queue_exception(vcpu, MC_VECTOR); | |
3816 | } else if (!(banks[1] & MCI_STATUS_VAL) | |
3817 | || !(banks[1] & MCI_STATUS_UC)) { | |
3818 | if (banks[1] & MCI_STATUS_VAL) | |
3819 | mce->status |= MCI_STATUS_OVER; | |
3820 | banks[2] = mce->addr; | |
3821 | banks[3] = mce->misc; | |
3822 | banks[1] = mce->status; | |
3823 | } else | |
3824 | banks[1] |= MCI_STATUS_OVER; | |
3825 | return 0; | |
3826 | } | |
3827 | ||
3cfc3092 JK |
3828 | static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu, |
3829 | struct kvm_vcpu_events *events) | |
3830 | { | |
7460fb4a | 3831 | process_nmi(vcpu); |
59073aaf | 3832 | |
a06230b6 OU |
3833 | /* |
3834 | * In guest mode, payload delivery should be deferred, | |
3835 | * so that the L1 hypervisor can intercept #PF before | |
3836 | * CR2 is modified (or intercept #DB before DR6 is | |
3837 | * modified under nVMX). Unless the per-VM capability, | |
3838 | * KVM_CAP_EXCEPTION_PAYLOAD, is set, we may not defer the delivery of | |
3839 | * an exception payload and handle after a KVM_GET_VCPU_EVENTS. Since we | |
3840 | * opportunistically defer the exception payload, deliver it if the | |
3841 | * capability hasn't been requested before processing a | |
3842 | * KVM_GET_VCPU_EVENTS. | |
3843 | */ | |
3844 | if (!vcpu->kvm->arch.exception_payload_enabled && | |
3845 | vcpu->arch.exception.pending && vcpu->arch.exception.has_payload) | |
3846 | kvm_deliver_exception_payload(vcpu); | |
3847 | ||
664f8e26 | 3848 | /* |
59073aaf JM |
3849 | * The API doesn't provide the instruction length for software |
3850 | * exceptions, so don't report them. As long as the guest RIP | |
3851 | * isn't advanced, we should expect to encounter the exception | |
3852 | * again. | |
664f8e26 | 3853 | */ |
59073aaf JM |
3854 | if (kvm_exception_is_soft(vcpu->arch.exception.nr)) { |
3855 | events->exception.injected = 0; | |
3856 | events->exception.pending = 0; | |
3857 | } else { | |
3858 | events->exception.injected = vcpu->arch.exception.injected; | |
3859 | events->exception.pending = vcpu->arch.exception.pending; | |
3860 | /* | |
3861 | * For ABI compatibility, deliberately conflate | |
3862 | * pending and injected exceptions when | |
3863 | * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled. | |
3864 | */ | |
3865 | if (!vcpu->kvm->arch.exception_payload_enabled) | |
3866 | events->exception.injected |= | |
3867 | vcpu->arch.exception.pending; | |
3868 | } | |
3cfc3092 JK |
3869 | events->exception.nr = vcpu->arch.exception.nr; |
3870 | events->exception.has_error_code = vcpu->arch.exception.has_error_code; | |
3871 | events->exception.error_code = vcpu->arch.exception.error_code; | |
59073aaf JM |
3872 | events->exception_has_payload = vcpu->arch.exception.has_payload; |
3873 | events->exception_payload = vcpu->arch.exception.payload; | |
3cfc3092 | 3874 | |
03b82a30 | 3875 | events->interrupt.injected = |
04140b41 | 3876 | vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft; |
3cfc3092 | 3877 | events->interrupt.nr = vcpu->arch.interrupt.nr; |
03b82a30 | 3878 | events->interrupt.soft = 0; |
afaf0b2f | 3879 | events->interrupt.shadow = kvm_x86_ops.get_interrupt_shadow(vcpu); |
3cfc3092 JK |
3880 | |
3881 | events->nmi.injected = vcpu->arch.nmi_injected; | |
7460fb4a | 3882 | events->nmi.pending = vcpu->arch.nmi_pending != 0; |
afaf0b2f | 3883 | events->nmi.masked = kvm_x86_ops.get_nmi_mask(vcpu); |
97e69aa6 | 3884 | events->nmi.pad = 0; |
3cfc3092 | 3885 | |
66450a21 | 3886 | events->sipi_vector = 0; /* never valid when reporting to user space */ |
3cfc3092 | 3887 | |
f077825a PB |
3888 | events->smi.smm = is_smm(vcpu); |
3889 | events->smi.pending = vcpu->arch.smi_pending; | |
3890 | events->smi.smm_inside_nmi = | |
3891 | !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK); | |
3892 | events->smi.latched_init = kvm_lapic_latched_init(vcpu); | |
3893 | ||
dab4b911 | 3894 | events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING |
f077825a PB |
3895 | | KVM_VCPUEVENT_VALID_SHADOW |
3896 | | KVM_VCPUEVENT_VALID_SMM); | |
59073aaf JM |
3897 | if (vcpu->kvm->arch.exception_payload_enabled) |
3898 | events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD; | |
3899 | ||
97e69aa6 | 3900 | memset(&events->reserved, 0, sizeof(events->reserved)); |
3cfc3092 JK |
3901 | } |
3902 | ||
c5833c7a | 3903 | static void kvm_smm_changed(struct kvm_vcpu *vcpu); |
6ef4e07e | 3904 | |
3cfc3092 JK |
3905 | static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu, |
3906 | struct kvm_vcpu_events *events) | |
3907 | { | |
dab4b911 | 3908 | if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING |
48005f64 | 3909 | | KVM_VCPUEVENT_VALID_SIPI_VECTOR |
f077825a | 3910 | | KVM_VCPUEVENT_VALID_SHADOW |
59073aaf JM |
3911 | | KVM_VCPUEVENT_VALID_SMM |
3912 | | KVM_VCPUEVENT_VALID_PAYLOAD)) | |
3cfc3092 JK |
3913 | return -EINVAL; |
3914 | ||
59073aaf JM |
3915 | if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) { |
3916 | if (!vcpu->kvm->arch.exception_payload_enabled) | |
3917 | return -EINVAL; | |
3918 | if (events->exception.pending) | |
3919 | events->exception.injected = 0; | |
3920 | else | |
3921 | events->exception_has_payload = 0; | |
3922 | } else { | |
3923 | events->exception.pending = 0; | |
3924 | events->exception_has_payload = 0; | |
3925 | } | |
3926 | ||
3927 | if ((events->exception.injected || events->exception.pending) && | |
3928 | (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR)) | |
78e546c8 PB |
3929 | return -EINVAL; |
3930 | ||
28bf2888 DH |
3931 | /* INITs are latched while in SMM */ |
3932 | if (events->flags & KVM_VCPUEVENT_VALID_SMM && | |
3933 | (events->smi.smm || events->smi.pending) && | |
3934 | vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) | |
3935 | return -EINVAL; | |
3936 | ||
7460fb4a | 3937 | process_nmi(vcpu); |
59073aaf JM |
3938 | vcpu->arch.exception.injected = events->exception.injected; |
3939 | vcpu->arch.exception.pending = events->exception.pending; | |
3cfc3092 JK |
3940 | vcpu->arch.exception.nr = events->exception.nr; |
3941 | vcpu->arch.exception.has_error_code = events->exception.has_error_code; | |
3942 | vcpu->arch.exception.error_code = events->exception.error_code; | |
59073aaf JM |
3943 | vcpu->arch.exception.has_payload = events->exception_has_payload; |
3944 | vcpu->arch.exception.payload = events->exception_payload; | |
3cfc3092 | 3945 | |
04140b41 | 3946 | vcpu->arch.interrupt.injected = events->interrupt.injected; |
3cfc3092 JK |
3947 | vcpu->arch.interrupt.nr = events->interrupt.nr; |
3948 | vcpu->arch.interrupt.soft = events->interrupt.soft; | |
48005f64 | 3949 | if (events->flags & KVM_VCPUEVENT_VALID_SHADOW) |
afaf0b2f | 3950 | kvm_x86_ops.set_interrupt_shadow(vcpu, |
48005f64 | 3951 | events->interrupt.shadow); |
3cfc3092 JK |
3952 | |
3953 | vcpu->arch.nmi_injected = events->nmi.injected; | |
dab4b911 JK |
3954 | if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING) |
3955 | vcpu->arch.nmi_pending = events->nmi.pending; | |
afaf0b2f | 3956 | kvm_x86_ops.set_nmi_mask(vcpu, events->nmi.masked); |
3cfc3092 | 3957 | |
66450a21 | 3958 | if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR && |
bce87cce | 3959 | lapic_in_kernel(vcpu)) |
66450a21 | 3960 | vcpu->arch.apic->sipi_vector = events->sipi_vector; |
3cfc3092 | 3961 | |
f077825a | 3962 | if (events->flags & KVM_VCPUEVENT_VALID_SMM) { |
c5833c7a SC |
3963 | if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm) { |
3964 | if (events->smi.smm) | |
3965 | vcpu->arch.hflags |= HF_SMM_MASK; | |
3966 | else | |
3967 | vcpu->arch.hflags &= ~HF_SMM_MASK; | |
3968 | kvm_smm_changed(vcpu); | |
3969 | } | |
6ef4e07e | 3970 | |
f077825a | 3971 | vcpu->arch.smi_pending = events->smi.pending; |
f4ef1910 WL |
3972 | |
3973 | if (events->smi.smm) { | |
3974 | if (events->smi.smm_inside_nmi) | |
3975 | vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK; | |
f077825a | 3976 | else |
f4ef1910 | 3977 | vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK; |
ff90afa7 LA |
3978 | } |
3979 | ||
3980 | if (lapic_in_kernel(vcpu)) { | |
3981 | if (events->smi.latched_init) | |
3982 | set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events); | |
3983 | else | |
3984 | clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events); | |
f077825a PB |
3985 | } |
3986 | } | |
3987 | ||
3842d135 AK |
3988 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
3989 | ||
3cfc3092 JK |
3990 | return 0; |
3991 | } | |
3992 | ||
a1efbe77 JK |
3993 | static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu, |
3994 | struct kvm_debugregs *dbgregs) | |
3995 | { | |
73aaf249 JK |
3996 | unsigned long val; |
3997 | ||
a1efbe77 | 3998 | memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db)); |
16f8a6f9 | 3999 | kvm_get_dr(vcpu, 6, &val); |
73aaf249 | 4000 | dbgregs->dr6 = val; |
a1efbe77 JK |
4001 | dbgregs->dr7 = vcpu->arch.dr7; |
4002 | dbgregs->flags = 0; | |
97e69aa6 | 4003 | memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved)); |
a1efbe77 JK |
4004 | } |
4005 | ||
4006 | static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu, | |
4007 | struct kvm_debugregs *dbgregs) | |
4008 | { | |
4009 | if (dbgregs->flags) | |
4010 | return -EINVAL; | |
4011 | ||
d14bdb55 PB |
4012 | if (dbgregs->dr6 & ~0xffffffffull) |
4013 | return -EINVAL; | |
4014 | if (dbgregs->dr7 & ~0xffffffffull) | |
4015 | return -EINVAL; | |
4016 | ||
a1efbe77 | 4017 | memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db)); |
ae561ede | 4018 | kvm_update_dr0123(vcpu); |
a1efbe77 JK |
4019 | vcpu->arch.dr6 = dbgregs->dr6; |
4020 | vcpu->arch.dr7 = dbgregs->dr7; | |
9926c9fd | 4021 | kvm_update_dr7(vcpu); |
a1efbe77 | 4022 | |
a1efbe77 JK |
4023 | return 0; |
4024 | } | |
4025 | ||
df1daba7 PB |
4026 | #define XSTATE_COMPACTION_ENABLED (1ULL << 63) |
4027 | ||
4028 | static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu) | |
4029 | { | |
b666a4b6 | 4030 | struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave; |
400e4b20 | 4031 | u64 xstate_bv = xsave->header.xfeatures; |
df1daba7 PB |
4032 | u64 valid; |
4033 | ||
4034 | /* | |
4035 | * Copy legacy XSAVE area, to avoid complications with CPUID | |
4036 | * leaves 0 and 1 in the loop below. | |
4037 | */ | |
4038 | memcpy(dest, xsave, XSAVE_HDR_OFFSET); | |
4039 | ||
4040 | /* Set XSTATE_BV */ | |
00c87e9a | 4041 | xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE; |
df1daba7 PB |
4042 | *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv; |
4043 | ||
4044 | /* | |
4045 | * Copy each region from the possibly compacted offset to the | |
4046 | * non-compacted offset. | |
4047 | */ | |
d91cab78 | 4048 | valid = xstate_bv & ~XFEATURE_MASK_FPSSE; |
df1daba7 | 4049 | while (valid) { |
abd16d68 SAS |
4050 | u64 xfeature_mask = valid & -valid; |
4051 | int xfeature_nr = fls64(xfeature_mask) - 1; | |
4052 | void *src = get_xsave_addr(xsave, xfeature_nr); | |
df1daba7 PB |
4053 | |
4054 | if (src) { | |
4055 | u32 size, offset, ecx, edx; | |
abd16d68 | 4056 | cpuid_count(XSTATE_CPUID, xfeature_nr, |
df1daba7 | 4057 | &size, &offset, &ecx, &edx); |
abd16d68 | 4058 | if (xfeature_nr == XFEATURE_PKRU) |
38cfd5e3 PB |
4059 | memcpy(dest + offset, &vcpu->arch.pkru, |
4060 | sizeof(vcpu->arch.pkru)); | |
4061 | else | |
4062 | memcpy(dest + offset, src, size); | |
4063 | ||
df1daba7 PB |
4064 | } |
4065 | ||
abd16d68 | 4066 | valid -= xfeature_mask; |
df1daba7 PB |
4067 | } |
4068 | } | |
4069 | ||
4070 | static void load_xsave(struct kvm_vcpu *vcpu, u8 *src) | |
4071 | { | |
b666a4b6 | 4072 | struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave; |
df1daba7 PB |
4073 | u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET); |
4074 | u64 valid; | |
4075 | ||
4076 | /* | |
4077 | * Copy legacy XSAVE area, to avoid complications with CPUID | |
4078 | * leaves 0 and 1 in the loop below. | |
4079 | */ | |
4080 | memcpy(xsave, src, XSAVE_HDR_OFFSET); | |
4081 | ||
4082 | /* Set XSTATE_BV and possibly XCOMP_BV. */ | |
400e4b20 | 4083 | xsave->header.xfeatures = xstate_bv; |
782511b0 | 4084 | if (boot_cpu_has(X86_FEATURE_XSAVES)) |
3a54450b | 4085 | xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED; |
df1daba7 PB |
4086 | |
4087 | /* | |
4088 | * Copy each region from the non-compacted offset to the | |
4089 | * possibly compacted offset. | |
4090 | */ | |
d91cab78 | 4091 | valid = xstate_bv & ~XFEATURE_MASK_FPSSE; |
df1daba7 | 4092 | while (valid) { |
abd16d68 SAS |
4093 | u64 xfeature_mask = valid & -valid; |
4094 | int xfeature_nr = fls64(xfeature_mask) - 1; | |
4095 | void *dest = get_xsave_addr(xsave, xfeature_nr); | |
df1daba7 PB |
4096 | |
4097 | if (dest) { | |
4098 | u32 size, offset, ecx, edx; | |
abd16d68 | 4099 | cpuid_count(XSTATE_CPUID, xfeature_nr, |
df1daba7 | 4100 | &size, &offset, &ecx, &edx); |
abd16d68 | 4101 | if (xfeature_nr == XFEATURE_PKRU) |
38cfd5e3 PB |
4102 | memcpy(&vcpu->arch.pkru, src + offset, |
4103 | sizeof(vcpu->arch.pkru)); | |
4104 | else | |
4105 | memcpy(dest, src + offset, size); | |
ee4100da | 4106 | } |
df1daba7 | 4107 | |
abd16d68 | 4108 | valid -= xfeature_mask; |
df1daba7 PB |
4109 | } |
4110 | } | |
4111 | ||
2d5b5a66 SY |
4112 | static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu, |
4113 | struct kvm_xsave *guest_xsave) | |
4114 | { | |
d366bf7e | 4115 | if (boot_cpu_has(X86_FEATURE_XSAVE)) { |
df1daba7 PB |
4116 | memset(guest_xsave, 0, sizeof(struct kvm_xsave)); |
4117 | fill_xsave((u8 *) guest_xsave->region, vcpu); | |
4344ee98 | 4118 | } else { |
2d5b5a66 | 4119 | memcpy(guest_xsave->region, |
b666a4b6 | 4120 | &vcpu->arch.guest_fpu->state.fxsave, |
c47ada30 | 4121 | sizeof(struct fxregs_state)); |
2d5b5a66 | 4122 | *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] = |
d91cab78 | 4123 | XFEATURE_MASK_FPSSE; |
2d5b5a66 SY |
4124 | } |
4125 | } | |
4126 | ||
a575813b WL |
4127 | #define XSAVE_MXCSR_OFFSET 24 |
4128 | ||
2d5b5a66 SY |
4129 | static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu, |
4130 | struct kvm_xsave *guest_xsave) | |
4131 | { | |
4132 | u64 xstate_bv = | |
4133 | *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)]; | |
a575813b | 4134 | u32 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)]; |
2d5b5a66 | 4135 | |
d366bf7e | 4136 | if (boot_cpu_has(X86_FEATURE_XSAVE)) { |
d7876f1b PB |
4137 | /* |
4138 | * Here we allow setting states that are not present in | |
4139 | * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility | |
4140 | * with old userspace. | |
4141 | */ | |
cfc48181 | 4142 | if (xstate_bv & ~supported_xcr0 || mxcsr & ~mxcsr_feature_mask) |
d7876f1b | 4143 | return -EINVAL; |
df1daba7 | 4144 | load_xsave(vcpu, (u8 *)guest_xsave->region); |
d7876f1b | 4145 | } else { |
a575813b WL |
4146 | if (xstate_bv & ~XFEATURE_MASK_FPSSE || |
4147 | mxcsr & ~mxcsr_feature_mask) | |
2d5b5a66 | 4148 | return -EINVAL; |
b666a4b6 | 4149 | memcpy(&vcpu->arch.guest_fpu->state.fxsave, |
c47ada30 | 4150 | guest_xsave->region, sizeof(struct fxregs_state)); |
2d5b5a66 SY |
4151 | } |
4152 | return 0; | |
4153 | } | |
4154 | ||
4155 | static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu, | |
4156 | struct kvm_xcrs *guest_xcrs) | |
4157 | { | |
d366bf7e | 4158 | if (!boot_cpu_has(X86_FEATURE_XSAVE)) { |
2d5b5a66 SY |
4159 | guest_xcrs->nr_xcrs = 0; |
4160 | return; | |
4161 | } | |
4162 | ||
4163 | guest_xcrs->nr_xcrs = 1; | |
4164 | guest_xcrs->flags = 0; | |
4165 | guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK; | |
4166 | guest_xcrs->xcrs[0].value = vcpu->arch.xcr0; | |
4167 | } | |
4168 | ||
4169 | static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu, | |
4170 | struct kvm_xcrs *guest_xcrs) | |
4171 | { | |
4172 | int i, r = 0; | |
4173 | ||
d366bf7e | 4174 | if (!boot_cpu_has(X86_FEATURE_XSAVE)) |
2d5b5a66 SY |
4175 | return -EINVAL; |
4176 | ||
4177 | if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags) | |
4178 | return -EINVAL; | |
4179 | ||
4180 | for (i = 0; i < guest_xcrs->nr_xcrs; i++) | |
4181 | /* Only support XCR0 currently */ | |
c67a04cb | 4182 | if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) { |
2d5b5a66 | 4183 | r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK, |
c67a04cb | 4184 | guest_xcrs->xcrs[i].value); |
2d5b5a66 SY |
4185 | break; |
4186 | } | |
4187 | if (r) | |
4188 | r = -EINVAL; | |
4189 | return r; | |
4190 | } | |
4191 | ||
1c0b28c2 EM |
4192 | /* |
4193 | * kvm_set_guest_paused() indicates to the guest kernel that it has been | |
4194 | * stopped by the hypervisor. This function will be called from the host only. | |
4195 | * EINVAL is returned when the host attempts to set the flag for a guest that | |
4196 | * does not support pv clocks. | |
4197 | */ | |
4198 | static int kvm_set_guest_paused(struct kvm_vcpu *vcpu) | |
4199 | { | |
0b79459b | 4200 | if (!vcpu->arch.pv_time_enabled) |
1c0b28c2 | 4201 | return -EINVAL; |
51d59c6b | 4202 | vcpu->arch.pvclock_set_guest_stopped_request = true; |
1c0b28c2 EM |
4203 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu); |
4204 | return 0; | |
4205 | } | |
4206 | ||
5c919412 AS |
4207 | static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu, |
4208 | struct kvm_enable_cap *cap) | |
4209 | { | |
57b119da VK |
4210 | int r; |
4211 | uint16_t vmcs_version; | |
4212 | void __user *user_ptr; | |
4213 | ||
5c919412 AS |
4214 | if (cap->flags) |
4215 | return -EINVAL; | |
4216 | ||
4217 | switch (cap->cap) { | |
efc479e6 RK |
4218 | case KVM_CAP_HYPERV_SYNIC2: |
4219 | if (cap->args[0]) | |
4220 | return -EINVAL; | |
b2869f28 GS |
4221 | /* fall through */ |
4222 | ||
5c919412 | 4223 | case KVM_CAP_HYPERV_SYNIC: |
546d87e5 WL |
4224 | if (!irqchip_in_kernel(vcpu->kvm)) |
4225 | return -EINVAL; | |
efc479e6 RK |
4226 | return kvm_hv_activate_synic(vcpu, cap->cap == |
4227 | KVM_CAP_HYPERV_SYNIC2); | |
57b119da | 4228 | case KVM_CAP_HYPERV_ENLIGHTENED_VMCS: |
afaf0b2f | 4229 | if (!kvm_x86_ops.nested_enable_evmcs) |
5158917c | 4230 | return -ENOTTY; |
afaf0b2f | 4231 | r = kvm_x86_ops.nested_enable_evmcs(vcpu, &vmcs_version); |
57b119da VK |
4232 | if (!r) { |
4233 | user_ptr = (void __user *)(uintptr_t)cap->args[0]; | |
4234 | if (copy_to_user(user_ptr, &vmcs_version, | |
4235 | sizeof(vmcs_version))) | |
4236 | r = -EFAULT; | |
4237 | } | |
4238 | return r; | |
344c6c80 | 4239 | case KVM_CAP_HYPERV_DIRECT_TLBFLUSH: |
afaf0b2f | 4240 | if (!kvm_x86_ops.enable_direct_tlbflush) |
344c6c80 TL |
4241 | return -ENOTTY; |
4242 | ||
afaf0b2f | 4243 | return kvm_x86_ops.enable_direct_tlbflush(vcpu); |
57b119da | 4244 | |
5c919412 AS |
4245 | default: |
4246 | return -EINVAL; | |
4247 | } | |
4248 | } | |
4249 | ||
313a3dc7 CO |
4250 | long kvm_arch_vcpu_ioctl(struct file *filp, |
4251 | unsigned int ioctl, unsigned long arg) | |
4252 | { | |
4253 | struct kvm_vcpu *vcpu = filp->private_data; | |
4254 | void __user *argp = (void __user *)arg; | |
4255 | int r; | |
d1ac91d8 AK |
4256 | union { |
4257 | struct kvm_lapic_state *lapic; | |
4258 | struct kvm_xsave *xsave; | |
4259 | struct kvm_xcrs *xcrs; | |
4260 | void *buffer; | |
4261 | } u; | |
4262 | ||
9b062471 CD |
4263 | vcpu_load(vcpu); |
4264 | ||
d1ac91d8 | 4265 | u.buffer = NULL; |
313a3dc7 CO |
4266 | switch (ioctl) { |
4267 | case KVM_GET_LAPIC: { | |
2204ae3c | 4268 | r = -EINVAL; |
bce87cce | 4269 | if (!lapic_in_kernel(vcpu)) |
2204ae3c | 4270 | goto out; |
254272ce BG |
4271 | u.lapic = kzalloc(sizeof(struct kvm_lapic_state), |
4272 | GFP_KERNEL_ACCOUNT); | |
313a3dc7 | 4273 | |
b772ff36 | 4274 | r = -ENOMEM; |
d1ac91d8 | 4275 | if (!u.lapic) |
b772ff36 | 4276 | goto out; |
d1ac91d8 | 4277 | r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic); |
313a3dc7 CO |
4278 | if (r) |
4279 | goto out; | |
4280 | r = -EFAULT; | |
d1ac91d8 | 4281 | if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state))) |
313a3dc7 CO |
4282 | goto out; |
4283 | r = 0; | |
4284 | break; | |
4285 | } | |
4286 | case KVM_SET_LAPIC: { | |
2204ae3c | 4287 | r = -EINVAL; |
bce87cce | 4288 | if (!lapic_in_kernel(vcpu)) |
2204ae3c | 4289 | goto out; |
ff5c2c03 | 4290 | u.lapic = memdup_user(argp, sizeof(*u.lapic)); |
9b062471 CD |
4291 | if (IS_ERR(u.lapic)) { |
4292 | r = PTR_ERR(u.lapic); | |
4293 | goto out_nofree; | |
4294 | } | |
ff5c2c03 | 4295 | |
d1ac91d8 | 4296 | r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic); |
313a3dc7 CO |
4297 | break; |
4298 | } | |
f77bc6a4 ZX |
4299 | case KVM_INTERRUPT: { |
4300 | struct kvm_interrupt irq; | |
4301 | ||
4302 | r = -EFAULT; | |
0e96f31e | 4303 | if (copy_from_user(&irq, argp, sizeof(irq))) |
f77bc6a4 ZX |
4304 | goto out; |
4305 | r = kvm_vcpu_ioctl_interrupt(vcpu, &irq); | |
f77bc6a4 ZX |
4306 | break; |
4307 | } | |
c4abb7c9 JK |
4308 | case KVM_NMI: { |
4309 | r = kvm_vcpu_ioctl_nmi(vcpu); | |
c4abb7c9 JK |
4310 | break; |
4311 | } | |
f077825a PB |
4312 | case KVM_SMI: { |
4313 | r = kvm_vcpu_ioctl_smi(vcpu); | |
4314 | break; | |
4315 | } | |
313a3dc7 CO |
4316 | case KVM_SET_CPUID: { |
4317 | struct kvm_cpuid __user *cpuid_arg = argp; | |
4318 | struct kvm_cpuid cpuid; | |
4319 | ||
4320 | r = -EFAULT; | |
0e96f31e | 4321 | if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid))) |
313a3dc7 CO |
4322 | goto out; |
4323 | r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries); | |
313a3dc7 CO |
4324 | break; |
4325 | } | |
07716717 DK |
4326 | case KVM_SET_CPUID2: { |
4327 | struct kvm_cpuid2 __user *cpuid_arg = argp; | |
4328 | struct kvm_cpuid2 cpuid; | |
4329 | ||
4330 | r = -EFAULT; | |
0e96f31e | 4331 | if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid))) |
07716717 DK |
4332 | goto out; |
4333 | r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid, | |
19355475 | 4334 | cpuid_arg->entries); |
07716717 DK |
4335 | break; |
4336 | } | |
4337 | case KVM_GET_CPUID2: { | |
4338 | struct kvm_cpuid2 __user *cpuid_arg = argp; | |
4339 | struct kvm_cpuid2 cpuid; | |
4340 | ||
4341 | r = -EFAULT; | |
0e96f31e | 4342 | if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid))) |
07716717 DK |
4343 | goto out; |
4344 | r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid, | |
19355475 | 4345 | cpuid_arg->entries); |
07716717 DK |
4346 | if (r) |
4347 | goto out; | |
4348 | r = -EFAULT; | |
0e96f31e | 4349 | if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid))) |
07716717 DK |
4350 | goto out; |
4351 | r = 0; | |
4352 | break; | |
4353 | } | |
801e459a TL |
4354 | case KVM_GET_MSRS: { |
4355 | int idx = srcu_read_lock(&vcpu->kvm->srcu); | |
609e36d3 | 4356 | r = msr_io(vcpu, argp, do_get_msr, 1); |
801e459a | 4357 | srcu_read_unlock(&vcpu->kvm->srcu, idx); |
313a3dc7 | 4358 | break; |
801e459a TL |
4359 | } |
4360 | case KVM_SET_MSRS: { | |
4361 | int idx = srcu_read_lock(&vcpu->kvm->srcu); | |
313a3dc7 | 4362 | r = msr_io(vcpu, argp, do_set_msr, 0); |
801e459a | 4363 | srcu_read_unlock(&vcpu->kvm->srcu, idx); |
313a3dc7 | 4364 | break; |
801e459a | 4365 | } |
b209749f AK |
4366 | case KVM_TPR_ACCESS_REPORTING: { |
4367 | struct kvm_tpr_access_ctl tac; | |
4368 | ||
4369 | r = -EFAULT; | |
0e96f31e | 4370 | if (copy_from_user(&tac, argp, sizeof(tac))) |
b209749f AK |
4371 | goto out; |
4372 | r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac); | |
4373 | if (r) | |
4374 | goto out; | |
4375 | r = -EFAULT; | |
0e96f31e | 4376 | if (copy_to_user(argp, &tac, sizeof(tac))) |
b209749f AK |
4377 | goto out; |
4378 | r = 0; | |
4379 | break; | |
4380 | }; | |
b93463aa AK |
4381 | case KVM_SET_VAPIC_ADDR: { |
4382 | struct kvm_vapic_addr va; | |
7301d6ab | 4383 | int idx; |
b93463aa AK |
4384 | |
4385 | r = -EINVAL; | |
35754c98 | 4386 | if (!lapic_in_kernel(vcpu)) |
b93463aa AK |
4387 | goto out; |
4388 | r = -EFAULT; | |
0e96f31e | 4389 | if (copy_from_user(&va, argp, sizeof(va))) |
b93463aa | 4390 | goto out; |
7301d6ab | 4391 | idx = srcu_read_lock(&vcpu->kvm->srcu); |
fda4e2e8 | 4392 | r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr); |
7301d6ab | 4393 | srcu_read_unlock(&vcpu->kvm->srcu, idx); |
b93463aa AK |
4394 | break; |
4395 | } | |
890ca9ae HY |
4396 | case KVM_X86_SETUP_MCE: { |
4397 | u64 mcg_cap; | |
4398 | ||
4399 | r = -EFAULT; | |
0e96f31e | 4400 | if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap))) |
890ca9ae HY |
4401 | goto out; |
4402 | r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap); | |
4403 | break; | |
4404 | } | |
4405 | case KVM_X86_SET_MCE: { | |
4406 | struct kvm_x86_mce mce; | |
4407 | ||
4408 | r = -EFAULT; | |
0e96f31e | 4409 | if (copy_from_user(&mce, argp, sizeof(mce))) |
890ca9ae HY |
4410 | goto out; |
4411 | r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce); | |
4412 | break; | |
4413 | } | |
3cfc3092 JK |
4414 | case KVM_GET_VCPU_EVENTS: { |
4415 | struct kvm_vcpu_events events; | |
4416 | ||
4417 | kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events); | |
4418 | ||
4419 | r = -EFAULT; | |
4420 | if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events))) | |
4421 | break; | |
4422 | r = 0; | |
4423 | break; | |
4424 | } | |
4425 | case KVM_SET_VCPU_EVENTS: { | |
4426 | struct kvm_vcpu_events events; | |
4427 | ||
4428 | r = -EFAULT; | |
4429 | if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events))) | |
4430 | break; | |
4431 | ||
4432 | r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events); | |
4433 | break; | |
4434 | } | |
a1efbe77 JK |
4435 | case KVM_GET_DEBUGREGS: { |
4436 | struct kvm_debugregs dbgregs; | |
4437 | ||
4438 | kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs); | |
4439 | ||
4440 | r = -EFAULT; | |
4441 | if (copy_to_user(argp, &dbgregs, | |
4442 | sizeof(struct kvm_debugregs))) | |
4443 | break; | |
4444 | r = 0; | |
4445 | break; | |
4446 | } | |
4447 | case KVM_SET_DEBUGREGS: { | |
4448 | struct kvm_debugregs dbgregs; | |
4449 | ||
4450 | r = -EFAULT; | |
4451 | if (copy_from_user(&dbgregs, argp, | |
4452 | sizeof(struct kvm_debugregs))) | |
4453 | break; | |
4454 | ||
4455 | r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs); | |
4456 | break; | |
4457 | } | |
2d5b5a66 | 4458 | case KVM_GET_XSAVE: { |
254272ce | 4459 | u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT); |
2d5b5a66 | 4460 | r = -ENOMEM; |
d1ac91d8 | 4461 | if (!u.xsave) |
2d5b5a66 SY |
4462 | break; |
4463 | ||
d1ac91d8 | 4464 | kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave); |
2d5b5a66 SY |
4465 | |
4466 | r = -EFAULT; | |
d1ac91d8 | 4467 | if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave))) |
2d5b5a66 SY |
4468 | break; |
4469 | r = 0; | |
4470 | break; | |
4471 | } | |
4472 | case KVM_SET_XSAVE: { | |
ff5c2c03 | 4473 | u.xsave = memdup_user(argp, sizeof(*u.xsave)); |
9b062471 CD |
4474 | if (IS_ERR(u.xsave)) { |
4475 | r = PTR_ERR(u.xsave); | |
4476 | goto out_nofree; | |
4477 | } | |
2d5b5a66 | 4478 | |
d1ac91d8 | 4479 | r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave); |
2d5b5a66 SY |
4480 | break; |
4481 | } | |
4482 | case KVM_GET_XCRS: { | |
254272ce | 4483 | u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT); |
2d5b5a66 | 4484 | r = -ENOMEM; |
d1ac91d8 | 4485 | if (!u.xcrs) |
2d5b5a66 SY |
4486 | break; |
4487 | ||
d1ac91d8 | 4488 | kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs); |
2d5b5a66 SY |
4489 | |
4490 | r = -EFAULT; | |
d1ac91d8 | 4491 | if (copy_to_user(argp, u.xcrs, |
2d5b5a66 SY |
4492 | sizeof(struct kvm_xcrs))) |
4493 | break; | |
4494 | r = 0; | |
4495 | break; | |
4496 | } | |
4497 | case KVM_SET_XCRS: { | |
ff5c2c03 | 4498 | u.xcrs = memdup_user(argp, sizeof(*u.xcrs)); |
9b062471 CD |
4499 | if (IS_ERR(u.xcrs)) { |
4500 | r = PTR_ERR(u.xcrs); | |
4501 | goto out_nofree; | |
4502 | } | |
2d5b5a66 | 4503 | |
d1ac91d8 | 4504 | r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs); |
2d5b5a66 SY |
4505 | break; |
4506 | } | |
92a1f12d JR |
4507 | case KVM_SET_TSC_KHZ: { |
4508 | u32 user_tsc_khz; | |
4509 | ||
4510 | r = -EINVAL; | |
92a1f12d JR |
4511 | user_tsc_khz = (u32)arg; |
4512 | ||
4513 | if (user_tsc_khz >= kvm_max_guest_tsc_khz) | |
4514 | goto out; | |
4515 | ||
cc578287 ZA |
4516 | if (user_tsc_khz == 0) |
4517 | user_tsc_khz = tsc_khz; | |
4518 | ||
381d585c HZ |
4519 | if (!kvm_set_tsc_khz(vcpu, user_tsc_khz)) |
4520 | r = 0; | |
92a1f12d | 4521 | |
92a1f12d JR |
4522 | goto out; |
4523 | } | |
4524 | case KVM_GET_TSC_KHZ: { | |
cc578287 | 4525 | r = vcpu->arch.virtual_tsc_khz; |
92a1f12d JR |
4526 | goto out; |
4527 | } | |
1c0b28c2 EM |
4528 | case KVM_KVMCLOCK_CTRL: { |
4529 | r = kvm_set_guest_paused(vcpu); | |
4530 | goto out; | |
4531 | } | |
5c919412 AS |
4532 | case KVM_ENABLE_CAP: { |
4533 | struct kvm_enable_cap cap; | |
4534 | ||
4535 | r = -EFAULT; | |
4536 | if (copy_from_user(&cap, argp, sizeof(cap))) | |
4537 | goto out; | |
4538 | r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap); | |
4539 | break; | |
4540 | } | |
8fcc4b59 JM |
4541 | case KVM_GET_NESTED_STATE: { |
4542 | struct kvm_nested_state __user *user_kvm_nested_state = argp; | |
4543 | u32 user_data_size; | |
4544 | ||
4545 | r = -EINVAL; | |
afaf0b2f | 4546 | if (!kvm_x86_ops.get_nested_state) |
8fcc4b59 JM |
4547 | break; |
4548 | ||
4549 | BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size)); | |
26b471c7 | 4550 | r = -EFAULT; |
8fcc4b59 | 4551 | if (get_user(user_data_size, &user_kvm_nested_state->size)) |
26b471c7 | 4552 | break; |
8fcc4b59 | 4553 | |
afaf0b2f | 4554 | r = kvm_x86_ops.get_nested_state(vcpu, user_kvm_nested_state, |
8fcc4b59 JM |
4555 | user_data_size); |
4556 | if (r < 0) | |
26b471c7 | 4557 | break; |
8fcc4b59 JM |
4558 | |
4559 | if (r > user_data_size) { | |
4560 | if (put_user(r, &user_kvm_nested_state->size)) | |
26b471c7 LA |
4561 | r = -EFAULT; |
4562 | else | |
4563 | r = -E2BIG; | |
4564 | break; | |
8fcc4b59 | 4565 | } |
26b471c7 | 4566 | |
8fcc4b59 JM |
4567 | r = 0; |
4568 | break; | |
4569 | } | |
4570 | case KVM_SET_NESTED_STATE: { | |
4571 | struct kvm_nested_state __user *user_kvm_nested_state = argp; | |
4572 | struct kvm_nested_state kvm_state; | |
ad5996d9 | 4573 | int idx; |
8fcc4b59 JM |
4574 | |
4575 | r = -EINVAL; | |
afaf0b2f | 4576 | if (!kvm_x86_ops.set_nested_state) |
8fcc4b59 JM |
4577 | break; |
4578 | ||
26b471c7 | 4579 | r = -EFAULT; |
8fcc4b59 | 4580 | if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state))) |
26b471c7 | 4581 | break; |
8fcc4b59 | 4582 | |
26b471c7 | 4583 | r = -EINVAL; |
8fcc4b59 | 4584 | if (kvm_state.size < sizeof(kvm_state)) |
26b471c7 | 4585 | break; |
8fcc4b59 JM |
4586 | |
4587 | if (kvm_state.flags & | |
8cab6507 VK |
4588 | ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE |
4589 | | KVM_STATE_NESTED_EVMCS)) | |
26b471c7 | 4590 | break; |
8fcc4b59 JM |
4591 | |
4592 | /* nested_run_pending implies guest_mode. */ | |
8cab6507 VK |
4593 | if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING) |
4594 | && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE)) | |
26b471c7 | 4595 | break; |
8fcc4b59 | 4596 | |
ad5996d9 | 4597 | idx = srcu_read_lock(&vcpu->kvm->srcu); |
afaf0b2f | 4598 | r = kvm_x86_ops.set_nested_state(vcpu, user_kvm_nested_state, &kvm_state); |
ad5996d9 | 4599 | srcu_read_unlock(&vcpu->kvm->srcu, idx); |
8fcc4b59 JM |
4600 | break; |
4601 | } | |
2bc39970 VK |
4602 | case KVM_GET_SUPPORTED_HV_CPUID: { |
4603 | struct kvm_cpuid2 __user *cpuid_arg = argp; | |
4604 | struct kvm_cpuid2 cpuid; | |
4605 | ||
4606 | r = -EFAULT; | |
4607 | if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid))) | |
4608 | goto out; | |
4609 | ||
4610 | r = kvm_vcpu_ioctl_get_hv_cpuid(vcpu, &cpuid, | |
4611 | cpuid_arg->entries); | |
4612 | if (r) | |
4613 | goto out; | |
4614 | ||
4615 | r = -EFAULT; | |
4616 | if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid))) | |
4617 | goto out; | |
4618 | r = 0; | |
4619 | break; | |
4620 | } | |
313a3dc7 CO |
4621 | default: |
4622 | r = -EINVAL; | |
4623 | } | |
4624 | out: | |
d1ac91d8 | 4625 | kfree(u.buffer); |
9b062471 CD |
4626 | out_nofree: |
4627 | vcpu_put(vcpu); | |
313a3dc7 CO |
4628 | return r; |
4629 | } | |
4630 | ||
1499fa80 | 4631 | vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf) |
5b1c1493 CO |
4632 | { |
4633 | return VM_FAULT_SIGBUS; | |
4634 | } | |
4635 | ||
1fe779f8 CO |
4636 | static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr) |
4637 | { | |
4638 | int ret; | |
4639 | ||
4640 | if (addr > (unsigned int)(-3 * PAGE_SIZE)) | |
951179ce | 4641 | return -EINVAL; |
afaf0b2f | 4642 | ret = kvm_x86_ops.set_tss_addr(kvm, addr); |
1fe779f8 CO |
4643 | return ret; |
4644 | } | |
4645 | ||
b927a3ce SY |
4646 | static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm, |
4647 | u64 ident_addr) | |
4648 | { | |
afaf0b2f | 4649 | return kvm_x86_ops.set_identity_map_addr(kvm, ident_addr); |
b927a3ce SY |
4650 | } |
4651 | ||
1fe779f8 | 4652 | static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm, |
bc8a3d89 | 4653 | unsigned long kvm_nr_mmu_pages) |
1fe779f8 CO |
4654 | { |
4655 | if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES) | |
4656 | return -EINVAL; | |
4657 | ||
79fac95e | 4658 | mutex_lock(&kvm->slots_lock); |
1fe779f8 CO |
4659 | |
4660 | kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages); | |
f05e70ac | 4661 | kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages; |
1fe779f8 | 4662 | |
79fac95e | 4663 | mutex_unlock(&kvm->slots_lock); |
1fe779f8 CO |
4664 | return 0; |
4665 | } | |
4666 | ||
bc8a3d89 | 4667 | static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm) |
1fe779f8 | 4668 | { |
39de71ec | 4669 | return kvm->arch.n_max_mmu_pages; |
1fe779f8 CO |
4670 | } |
4671 | ||
1fe779f8 CO |
4672 | static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip) |
4673 | { | |
90bca052 | 4674 | struct kvm_pic *pic = kvm->arch.vpic; |
1fe779f8 CO |
4675 | int r; |
4676 | ||
4677 | r = 0; | |
4678 | switch (chip->chip_id) { | |
4679 | case KVM_IRQCHIP_PIC_MASTER: | |
90bca052 | 4680 | memcpy(&chip->chip.pic, &pic->pics[0], |
1fe779f8 CO |
4681 | sizeof(struct kvm_pic_state)); |
4682 | break; | |
4683 | case KVM_IRQCHIP_PIC_SLAVE: | |
90bca052 | 4684 | memcpy(&chip->chip.pic, &pic->pics[1], |
1fe779f8 CO |
4685 | sizeof(struct kvm_pic_state)); |
4686 | break; | |
4687 | case KVM_IRQCHIP_IOAPIC: | |
33392b49 | 4688 | kvm_get_ioapic(kvm, &chip->chip.ioapic); |
1fe779f8 CO |
4689 | break; |
4690 | default: | |
4691 | r = -EINVAL; | |
4692 | break; | |
4693 | } | |
4694 | return r; | |
4695 | } | |
4696 | ||
4697 | static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip) | |
4698 | { | |
90bca052 | 4699 | struct kvm_pic *pic = kvm->arch.vpic; |
1fe779f8 CO |
4700 | int r; |
4701 | ||
4702 | r = 0; | |
4703 | switch (chip->chip_id) { | |
4704 | case KVM_IRQCHIP_PIC_MASTER: | |
90bca052 DH |
4705 | spin_lock(&pic->lock); |
4706 | memcpy(&pic->pics[0], &chip->chip.pic, | |
1fe779f8 | 4707 | sizeof(struct kvm_pic_state)); |
90bca052 | 4708 | spin_unlock(&pic->lock); |
1fe779f8 CO |
4709 | break; |
4710 | case KVM_IRQCHIP_PIC_SLAVE: | |
90bca052 DH |
4711 | spin_lock(&pic->lock); |
4712 | memcpy(&pic->pics[1], &chip->chip.pic, | |
1fe779f8 | 4713 | sizeof(struct kvm_pic_state)); |
90bca052 | 4714 | spin_unlock(&pic->lock); |
1fe779f8 CO |
4715 | break; |
4716 | case KVM_IRQCHIP_IOAPIC: | |
33392b49 | 4717 | kvm_set_ioapic(kvm, &chip->chip.ioapic); |
1fe779f8 CO |
4718 | break; |
4719 | default: | |
4720 | r = -EINVAL; | |
4721 | break; | |
4722 | } | |
90bca052 | 4723 | kvm_pic_update_irq(pic); |
1fe779f8 CO |
4724 | return r; |
4725 | } | |
4726 | ||
e0f63cb9 SY |
4727 | static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps) |
4728 | { | |
34f3941c RK |
4729 | struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state; |
4730 | ||
4731 | BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels)); | |
4732 | ||
4733 | mutex_lock(&kps->lock); | |
4734 | memcpy(ps, &kps->channels, sizeof(*ps)); | |
4735 | mutex_unlock(&kps->lock); | |
2da29bcc | 4736 | return 0; |
e0f63cb9 SY |
4737 | } |
4738 | ||
4739 | static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps) | |
4740 | { | |
0185604c | 4741 | int i; |
09edea72 RK |
4742 | struct kvm_pit *pit = kvm->arch.vpit; |
4743 | ||
4744 | mutex_lock(&pit->pit_state.lock); | |
34f3941c | 4745 | memcpy(&pit->pit_state.channels, ps, sizeof(*ps)); |
0185604c | 4746 | for (i = 0; i < 3; i++) |
09edea72 RK |
4747 | kvm_pit_load_count(pit, i, ps->channels[i].count, 0); |
4748 | mutex_unlock(&pit->pit_state.lock); | |
2da29bcc | 4749 | return 0; |
e9f42757 BK |
4750 | } |
4751 | ||
4752 | static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps) | |
4753 | { | |
e9f42757 BK |
4754 | mutex_lock(&kvm->arch.vpit->pit_state.lock); |
4755 | memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels, | |
4756 | sizeof(ps->channels)); | |
4757 | ps->flags = kvm->arch.vpit->pit_state.flags; | |
4758 | mutex_unlock(&kvm->arch.vpit->pit_state.lock); | |
97e69aa6 | 4759 | memset(&ps->reserved, 0, sizeof(ps->reserved)); |
2da29bcc | 4760 | return 0; |
e9f42757 BK |
4761 | } |
4762 | ||
4763 | static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps) | |
4764 | { | |
2da29bcc | 4765 | int start = 0; |
0185604c | 4766 | int i; |
e9f42757 | 4767 | u32 prev_legacy, cur_legacy; |
09edea72 RK |
4768 | struct kvm_pit *pit = kvm->arch.vpit; |
4769 | ||
4770 | mutex_lock(&pit->pit_state.lock); | |
4771 | prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY; | |
e9f42757 BK |
4772 | cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY; |
4773 | if (!prev_legacy && cur_legacy) | |
4774 | start = 1; | |
09edea72 RK |
4775 | memcpy(&pit->pit_state.channels, &ps->channels, |
4776 | sizeof(pit->pit_state.channels)); | |
4777 | pit->pit_state.flags = ps->flags; | |
0185604c | 4778 | for (i = 0; i < 3; i++) |
09edea72 | 4779 | kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count, |
e5e57e7a | 4780 | start && i == 0); |
09edea72 | 4781 | mutex_unlock(&pit->pit_state.lock); |
2da29bcc | 4782 | return 0; |
e0f63cb9 SY |
4783 | } |
4784 | ||
52d939a0 MT |
4785 | static int kvm_vm_ioctl_reinject(struct kvm *kvm, |
4786 | struct kvm_reinject_control *control) | |
4787 | { | |
71474e2f RK |
4788 | struct kvm_pit *pit = kvm->arch.vpit; |
4789 | ||
71474e2f RK |
4790 | /* pit->pit_state.lock was overloaded to prevent userspace from getting |
4791 | * an inconsistent state after running multiple KVM_REINJECT_CONTROL | |
4792 | * ioctls in parallel. Use a separate lock if that ioctl isn't rare. | |
4793 | */ | |
4794 | mutex_lock(&pit->pit_state.lock); | |
4795 | kvm_pit_set_reinject(pit, control->pit_reinject); | |
4796 | mutex_unlock(&pit->pit_state.lock); | |
b39c90b6 | 4797 | |
52d939a0 MT |
4798 | return 0; |
4799 | } | |
4800 | ||
0dff0846 | 4801 | void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot) |
5bb064dc | 4802 | { |
88178fd4 KH |
4803 | /* |
4804 | * Flush potentially hardware-cached dirty pages to dirty_bitmap. | |
4805 | */ | |
afaf0b2f SC |
4806 | if (kvm_x86_ops.flush_log_dirty) |
4807 | kvm_x86_ops.flush_log_dirty(kvm); | |
5bb064dc ZX |
4808 | } |
4809 | ||
aa2fbe6d YZ |
4810 | int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event, |
4811 | bool line_status) | |
23d43cf9 CD |
4812 | { |
4813 | if (!irqchip_in_kernel(kvm)) | |
4814 | return -ENXIO; | |
4815 | ||
4816 | irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID, | |
aa2fbe6d YZ |
4817 | irq_event->irq, irq_event->level, |
4818 | line_status); | |
23d43cf9 CD |
4819 | return 0; |
4820 | } | |
4821 | ||
e5d83c74 PB |
4822 | int kvm_vm_ioctl_enable_cap(struct kvm *kvm, |
4823 | struct kvm_enable_cap *cap) | |
90de4a18 NA |
4824 | { |
4825 | int r; | |
4826 | ||
4827 | if (cap->flags) | |
4828 | return -EINVAL; | |
4829 | ||
4830 | switch (cap->cap) { | |
4831 | case KVM_CAP_DISABLE_QUIRKS: | |
4832 | kvm->arch.disabled_quirks = cap->args[0]; | |
4833 | r = 0; | |
4834 | break; | |
49df6397 SR |
4835 | case KVM_CAP_SPLIT_IRQCHIP: { |
4836 | mutex_lock(&kvm->lock); | |
b053b2ae SR |
4837 | r = -EINVAL; |
4838 | if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS) | |
4839 | goto split_irqchip_unlock; | |
49df6397 SR |
4840 | r = -EEXIST; |
4841 | if (irqchip_in_kernel(kvm)) | |
4842 | goto split_irqchip_unlock; | |
557abc40 | 4843 | if (kvm->created_vcpus) |
49df6397 SR |
4844 | goto split_irqchip_unlock; |
4845 | r = kvm_setup_empty_irq_routing(kvm); | |
5c0aea0e | 4846 | if (r) |
49df6397 SR |
4847 | goto split_irqchip_unlock; |
4848 | /* Pairs with irqchip_in_kernel. */ | |
4849 | smp_wmb(); | |
49776faf | 4850 | kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT; |
b053b2ae | 4851 | kvm->arch.nr_reserved_ioapic_pins = cap->args[0]; |
49df6397 SR |
4852 | r = 0; |
4853 | split_irqchip_unlock: | |
4854 | mutex_unlock(&kvm->lock); | |
4855 | break; | |
4856 | } | |
37131313 RK |
4857 | case KVM_CAP_X2APIC_API: |
4858 | r = -EINVAL; | |
4859 | if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS) | |
4860 | break; | |
4861 | ||
4862 | if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS) | |
4863 | kvm->arch.x2apic_format = true; | |
c519265f RK |
4864 | if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK) |
4865 | kvm->arch.x2apic_broadcast_quirk_disabled = true; | |
37131313 RK |
4866 | |
4867 | r = 0; | |
4868 | break; | |
4d5422ce WL |
4869 | case KVM_CAP_X86_DISABLE_EXITS: |
4870 | r = -EINVAL; | |
4871 | if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS) | |
4872 | break; | |
4873 | ||
4874 | if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) && | |
4875 | kvm_can_mwait_in_guest()) | |
4876 | kvm->arch.mwait_in_guest = true; | |
766d3571 | 4877 | if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT) |
caa057a2 | 4878 | kvm->arch.hlt_in_guest = true; |
b31c114b WL |
4879 | if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE) |
4880 | kvm->arch.pause_in_guest = true; | |
b5170063 WL |
4881 | if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE) |
4882 | kvm->arch.cstate_in_guest = true; | |
4d5422ce WL |
4883 | r = 0; |
4884 | break; | |
6fbbde9a DS |
4885 | case KVM_CAP_MSR_PLATFORM_INFO: |
4886 | kvm->arch.guest_can_read_msr_platform_info = cap->args[0]; | |
4887 | r = 0; | |
c4f55198 JM |
4888 | break; |
4889 | case KVM_CAP_EXCEPTION_PAYLOAD: | |
4890 | kvm->arch.exception_payload_enabled = cap->args[0]; | |
4891 | r = 0; | |
6fbbde9a | 4892 | break; |
90de4a18 NA |
4893 | default: |
4894 | r = -EINVAL; | |
4895 | break; | |
4896 | } | |
4897 | return r; | |
4898 | } | |
4899 | ||
1fe779f8 CO |
4900 | long kvm_arch_vm_ioctl(struct file *filp, |
4901 | unsigned int ioctl, unsigned long arg) | |
4902 | { | |
4903 | struct kvm *kvm = filp->private_data; | |
4904 | void __user *argp = (void __user *)arg; | |
367e1319 | 4905 | int r = -ENOTTY; |
f0d66275 DH |
4906 | /* |
4907 | * This union makes it completely explicit to gcc-3.x | |
4908 | * that these two variables' stack usage should be | |
4909 | * combined, not added together. | |
4910 | */ | |
4911 | union { | |
4912 | struct kvm_pit_state ps; | |
e9f42757 | 4913 | struct kvm_pit_state2 ps2; |
c5ff41ce | 4914 | struct kvm_pit_config pit_config; |
f0d66275 | 4915 | } u; |
1fe779f8 CO |
4916 | |
4917 | switch (ioctl) { | |
4918 | case KVM_SET_TSS_ADDR: | |
4919 | r = kvm_vm_ioctl_set_tss_addr(kvm, arg); | |
1fe779f8 | 4920 | break; |
b927a3ce SY |
4921 | case KVM_SET_IDENTITY_MAP_ADDR: { |
4922 | u64 ident_addr; | |
4923 | ||
1af1ac91 DH |
4924 | mutex_lock(&kvm->lock); |
4925 | r = -EINVAL; | |
4926 | if (kvm->created_vcpus) | |
4927 | goto set_identity_unlock; | |
b927a3ce | 4928 | r = -EFAULT; |
0e96f31e | 4929 | if (copy_from_user(&ident_addr, argp, sizeof(ident_addr))) |
1af1ac91 | 4930 | goto set_identity_unlock; |
b927a3ce | 4931 | r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr); |
1af1ac91 DH |
4932 | set_identity_unlock: |
4933 | mutex_unlock(&kvm->lock); | |
b927a3ce SY |
4934 | break; |
4935 | } | |
1fe779f8 CO |
4936 | case KVM_SET_NR_MMU_PAGES: |
4937 | r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg); | |
1fe779f8 CO |
4938 | break; |
4939 | case KVM_GET_NR_MMU_PAGES: | |
4940 | r = kvm_vm_ioctl_get_nr_mmu_pages(kvm); | |
4941 | break; | |
3ddea128 | 4942 | case KVM_CREATE_IRQCHIP: { |
3ddea128 | 4943 | mutex_lock(&kvm->lock); |
09941366 | 4944 | |
3ddea128 | 4945 | r = -EEXIST; |
35e6eaa3 | 4946 | if (irqchip_in_kernel(kvm)) |
3ddea128 | 4947 | goto create_irqchip_unlock; |
09941366 | 4948 | |
3e515705 | 4949 | r = -EINVAL; |
557abc40 | 4950 | if (kvm->created_vcpus) |
3e515705 | 4951 | goto create_irqchip_unlock; |
09941366 RK |
4952 | |
4953 | r = kvm_pic_init(kvm); | |
4954 | if (r) | |
3ddea128 | 4955 | goto create_irqchip_unlock; |
09941366 RK |
4956 | |
4957 | r = kvm_ioapic_init(kvm); | |
4958 | if (r) { | |
09941366 | 4959 | kvm_pic_destroy(kvm); |
3ddea128 | 4960 | goto create_irqchip_unlock; |
09941366 RK |
4961 | } |
4962 | ||
399ec807 AK |
4963 | r = kvm_setup_default_irq_routing(kvm); |
4964 | if (r) { | |
72bb2fcd | 4965 | kvm_ioapic_destroy(kvm); |
09941366 | 4966 | kvm_pic_destroy(kvm); |
71ba994c | 4967 | goto create_irqchip_unlock; |
399ec807 | 4968 | } |
49776faf | 4969 | /* Write kvm->irq_routing before enabling irqchip_in_kernel. */ |
71ba994c | 4970 | smp_wmb(); |
49776faf | 4971 | kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL; |
3ddea128 MT |
4972 | create_irqchip_unlock: |
4973 | mutex_unlock(&kvm->lock); | |
1fe779f8 | 4974 | break; |
3ddea128 | 4975 | } |
7837699f | 4976 | case KVM_CREATE_PIT: |
c5ff41ce JK |
4977 | u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY; |
4978 | goto create_pit; | |
4979 | case KVM_CREATE_PIT2: | |
4980 | r = -EFAULT; | |
4981 | if (copy_from_user(&u.pit_config, argp, | |
4982 | sizeof(struct kvm_pit_config))) | |
4983 | goto out; | |
4984 | create_pit: | |
250715a6 | 4985 | mutex_lock(&kvm->lock); |
269e05e4 AK |
4986 | r = -EEXIST; |
4987 | if (kvm->arch.vpit) | |
4988 | goto create_pit_unlock; | |
7837699f | 4989 | r = -ENOMEM; |
c5ff41ce | 4990 | kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags); |
7837699f SY |
4991 | if (kvm->arch.vpit) |
4992 | r = 0; | |
269e05e4 | 4993 | create_pit_unlock: |
250715a6 | 4994 | mutex_unlock(&kvm->lock); |
7837699f | 4995 | break; |
1fe779f8 CO |
4996 | case KVM_GET_IRQCHIP: { |
4997 | /* 0: PIC master, 1: PIC slave, 2: IOAPIC */ | |
ff5c2c03 | 4998 | struct kvm_irqchip *chip; |
1fe779f8 | 4999 | |
ff5c2c03 SL |
5000 | chip = memdup_user(argp, sizeof(*chip)); |
5001 | if (IS_ERR(chip)) { | |
5002 | r = PTR_ERR(chip); | |
1fe779f8 | 5003 | goto out; |
ff5c2c03 SL |
5004 | } |
5005 | ||
1fe779f8 | 5006 | r = -ENXIO; |
826da321 | 5007 | if (!irqchip_kernel(kvm)) |
f0d66275 DH |
5008 | goto get_irqchip_out; |
5009 | r = kvm_vm_ioctl_get_irqchip(kvm, chip); | |
1fe779f8 | 5010 | if (r) |
f0d66275 | 5011 | goto get_irqchip_out; |
1fe779f8 | 5012 | r = -EFAULT; |
0e96f31e | 5013 | if (copy_to_user(argp, chip, sizeof(*chip))) |
f0d66275 | 5014 | goto get_irqchip_out; |
1fe779f8 | 5015 | r = 0; |
f0d66275 DH |
5016 | get_irqchip_out: |
5017 | kfree(chip); | |
1fe779f8 CO |
5018 | break; |
5019 | } | |
5020 | case KVM_SET_IRQCHIP: { | |
5021 | /* 0: PIC master, 1: PIC slave, 2: IOAPIC */ | |
ff5c2c03 | 5022 | struct kvm_irqchip *chip; |
1fe779f8 | 5023 | |
ff5c2c03 SL |
5024 | chip = memdup_user(argp, sizeof(*chip)); |
5025 | if (IS_ERR(chip)) { | |
5026 | r = PTR_ERR(chip); | |
1fe779f8 | 5027 | goto out; |
ff5c2c03 SL |
5028 | } |
5029 | ||
1fe779f8 | 5030 | r = -ENXIO; |
826da321 | 5031 | if (!irqchip_kernel(kvm)) |
f0d66275 DH |
5032 | goto set_irqchip_out; |
5033 | r = kvm_vm_ioctl_set_irqchip(kvm, chip); | |
f0d66275 DH |
5034 | set_irqchip_out: |
5035 | kfree(chip); | |
1fe779f8 CO |
5036 | break; |
5037 | } | |
e0f63cb9 | 5038 | case KVM_GET_PIT: { |
e0f63cb9 | 5039 | r = -EFAULT; |
f0d66275 | 5040 | if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state))) |
e0f63cb9 SY |
5041 | goto out; |
5042 | r = -ENXIO; | |
5043 | if (!kvm->arch.vpit) | |
5044 | goto out; | |
f0d66275 | 5045 | r = kvm_vm_ioctl_get_pit(kvm, &u.ps); |
e0f63cb9 SY |
5046 | if (r) |
5047 | goto out; | |
5048 | r = -EFAULT; | |
f0d66275 | 5049 | if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state))) |
e0f63cb9 SY |
5050 | goto out; |
5051 | r = 0; | |
5052 | break; | |
5053 | } | |
5054 | case KVM_SET_PIT: { | |
e0f63cb9 | 5055 | r = -EFAULT; |
0e96f31e | 5056 | if (copy_from_user(&u.ps, argp, sizeof(u.ps))) |
e0f63cb9 | 5057 | goto out; |
7289fdb5 | 5058 | mutex_lock(&kvm->lock); |
e0f63cb9 SY |
5059 | r = -ENXIO; |
5060 | if (!kvm->arch.vpit) | |
7289fdb5 | 5061 | goto set_pit_out; |
f0d66275 | 5062 | r = kvm_vm_ioctl_set_pit(kvm, &u.ps); |
7289fdb5 SR |
5063 | set_pit_out: |
5064 | mutex_unlock(&kvm->lock); | |
e0f63cb9 SY |
5065 | break; |
5066 | } | |
e9f42757 BK |
5067 | case KVM_GET_PIT2: { |
5068 | r = -ENXIO; | |
5069 | if (!kvm->arch.vpit) | |
5070 | goto out; | |
5071 | r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2); | |
5072 | if (r) | |
5073 | goto out; | |
5074 | r = -EFAULT; | |
5075 | if (copy_to_user(argp, &u.ps2, sizeof(u.ps2))) | |
5076 | goto out; | |
5077 | r = 0; | |
5078 | break; | |
5079 | } | |
5080 | case KVM_SET_PIT2: { | |
5081 | r = -EFAULT; | |
5082 | if (copy_from_user(&u.ps2, argp, sizeof(u.ps2))) | |
5083 | goto out; | |
7289fdb5 | 5084 | mutex_lock(&kvm->lock); |
e9f42757 BK |
5085 | r = -ENXIO; |
5086 | if (!kvm->arch.vpit) | |
7289fdb5 | 5087 | goto set_pit2_out; |
e9f42757 | 5088 | r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2); |
7289fdb5 SR |
5089 | set_pit2_out: |
5090 | mutex_unlock(&kvm->lock); | |
e9f42757 BK |
5091 | break; |
5092 | } | |
52d939a0 MT |
5093 | case KVM_REINJECT_CONTROL: { |
5094 | struct kvm_reinject_control control; | |
5095 | r = -EFAULT; | |
5096 | if (copy_from_user(&control, argp, sizeof(control))) | |
5097 | goto out; | |
cad23e72 ML |
5098 | r = -ENXIO; |
5099 | if (!kvm->arch.vpit) | |
5100 | goto out; | |
52d939a0 | 5101 | r = kvm_vm_ioctl_reinject(kvm, &control); |
52d939a0 MT |
5102 | break; |
5103 | } | |
d71ba788 PB |
5104 | case KVM_SET_BOOT_CPU_ID: |
5105 | r = 0; | |
5106 | mutex_lock(&kvm->lock); | |
557abc40 | 5107 | if (kvm->created_vcpus) |
d71ba788 PB |
5108 | r = -EBUSY; |
5109 | else | |
5110 | kvm->arch.bsp_vcpu_id = arg; | |
5111 | mutex_unlock(&kvm->lock); | |
5112 | break; | |
ffde22ac | 5113 | case KVM_XEN_HVM_CONFIG: { |
51776043 | 5114 | struct kvm_xen_hvm_config xhc; |
ffde22ac | 5115 | r = -EFAULT; |
51776043 | 5116 | if (copy_from_user(&xhc, argp, sizeof(xhc))) |
ffde22ac ES |
5117 | goto out; |
5118 | r = -EINVAL; | |
51776043 | 5119 | if (xhc.flags) |
ffde22ac | 5120 | goto out; |
51776043 | 5121 | memcpy(&kvm->arch.xen_hvm_config, &xhc, sizeof(xhc)); |
ffde22ac ES |
5122 | r = 0; |
5123 | break; | |
5124 | } | |
afbcf7ab | 5125 | case KVM_SET_CLOCK: { |
afbcf7ab GC |
5126 | struct kvm_clock_data user_ns; |
5127 | u64 now_ns; | |
afbcf7ab GC |
5128 | |
5129 | r = -EFAULT; | |
5130 | if (copy_from_user(&user_ns, argp, sizeof(user_ns))) | |
5131 | goto out; | |
5132 | ||
5133 | r = -EINVAL; | |
5134 | if (user_ns.flags) | |
5135 | goto out; | |
5136 | ||
5137 | r = 0; | |
0bc48bea RK |
5138 | /* |
5139 | * TODO: userspace has to take care of races with VCPU_RUN, so | |
5140 | * kvm_gen_update_masterclock() can be cut down to locked | |
5141 | * pvclock_update_vm_gtod_copy(). | |
5142 | */ | |
5143 | kvm_gen_update_masterclock(kvm); | |
e891a32e | 5144 | now_ns = get_kvmclock_ns(kvm); |
108b249c | 5145 | kvm->arch.kvmclock_offset += user_ns.clock - now_ns; |
0bc48bea | 5146 | kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE); |
afbcf7ab GC |
5147 | break; |
5148 | } | |
5149 | case KVM_GET_CLOCK: { | |
afbcf7ab GC |
5150 | struct kvm_clock_data user_ns; |
5151 | u64 now_ns; | |
5152 | ||
e891a32e | 5153 | now_ns = get_kvmclock_ns(kvm); |
108b249c | 5154 | user_ns.clock = now_ns; |
e3fd9a93 | 5155 | user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0; |
97e69aa6 | 5156 | memset(&user_ns.pad, 0, sizeof(user_ns.pad)); |
afbcf7ab GC |
5157 | |
5158 | r = -EFAULT; | |
5159 | if (copy_to_user(argp, &user_ns, sizeof(user_ns))) | |
5160 | goto out; | |
5161 | r = 0; | |
5162 | break; | |
5163 | } | |
5acc5c06 BS |
5164 | case KVM_MEMORY_ENCRYPT_OP: { |
5165 | r = -ENOTTY; | |
afaf0b2f SC |
5166 | if (kvm_x86_ops.mem_enc_op) |
5167 | r = kvm_x86_ops.mem_enc_op(kvm, argp); | |
5acc5c06 BS |
5168 | break; |
5169 | } | |
69eaedee BS |
5170 | case KVM_MEMORY_ENCRYPT_REG_REGION: { |
5171 | struct kvm_enc_region region; | |
5172 | ||
5173 | r = -EFAULT; | |
5174 | if (copy_from_user(®ion, argp, sizeof(region))) | |
5175 | goto out; | |
5176 | ||
5177 | r = -ENOTTY; | |
afaf0b2f SC |
5178 | if (kvm_x86_ops.mem_enc_reg_region) |
5179 | r = kvm_x86_ops.mem_enc_reg_region(kvm, ®ion); | |
69eaedee BS |
5180 | break; |
5181 | } | |
5182 | case KVM_MEMORY_ENCRYPT_UNREG_REGION: { | |
5183 | struct kvm_enc_region region; | |
5184 | ||
5185 | r = -EFAULT; | |
5186 | if (copy_from_user(®ion, argp, sizeof(region))) | |
5187 | goto out; | |
5188 | ||
5189 | r = -ENOTTY; | |
afaf0b2f SC |
5190 | if (kvm_x86_ops.mem_enc_unreg_region) |
5191 | r = kvm_x86_ops.mem_enc_unreg_region(kvm, ®ion); | |
69eaedee BS |
5192 | break; |
5193 | } | |
faeb7833 RK |
5194 | case KVM_HYPERV_EVENTFD: { |
5195 | struct kvm_hyperv_eventfd hvevfd; | |
5196 | ||
5197 | r = -EFAULT; | |
5198 | if (copy_from_user(&hvevfd, argp, sizeof(hvevfd))) | |
5199 | goto out; | |
5200 | r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd); | |
5201 | break; | |
5202 | } | |
66bb8a06 EH |
5203 | case KVM_SET_PMU_EVENT_FILTER: |
5204 | r = kvm_vm_ioctl_set_pmu_event_filter(kvm, argp); | |
5205 | break; | |
1fe779f8 | 5206 | default: |
ad6260da | 5207 | r = -ENOTTY; |
1fe779f8 CO |
5208 | } |
5209 | out: | |
5210 | return r; | |
5211 | } | |
5212 | ||
a16b043c | 5213 | static void kvm_init_msr_list(void) |
043405e1 | 5214 | { |
24c29b7a | 5215 | struct x86_pmu_capability x86_pmu; |
043405e1 | 5216 | u32 dummy[2]; |
7a5ee6ed | 5217 | unsigned i; |
043405e1 | 5218 | |
e2ada66e | 5219 | BUILD_BUG_ON_MSG(INTEL_PMC_MAX_FIXED != 4, |
7a5ee6ed | 5220 | "Please update the fixed PMCs in msrs_to_saved_all[]"); |
24c29b7a PB |
5221 | |
5222 | perf_get_x86_pmu_capability(&x86_pmu); | |
e2ada66e | 5223 | |
6cbee2b9 XL |
5224 | num_msrs_to_save = 0; |
5225 | num_emulated_msrs = 0; | |
5226 | num_msr_based_features = 0; | |
5227 | ||
7a5ee6ed CQ |
5228 | for (i = 0; i < ARRAY_SIZE(msrs_to_save_all); i++) { |
5229 | if (rdmsr_safe(msrs_to_save_all[i], &dummy[0], &dummy[1]) < 0) | |
043405e1 | 5230 | continue; |
93c4adc7 PB |
5231 | |
5232 | /* | |
5233 | * Even MSRs that are valid in the host may not be exposed | |
9dbe6cf9 | 5234 | * to the guests in some cases. |
93c4adc7 | 5235 | */ |
7a5ee6ed | 5236 | switch (msrs_to_save_all[i]) { |
93c4adc7 | 5237 | case MSR_IA32_BNDCFGS: |
503234b3 | 5238 | if (!kvm_mpx_supported()) |
93c4adc7 PB |
5239 | continue; |
5240 | break; | |
9dbe6cf9 | 5241 | case MSR_TSC_AUX: |
13908510 | 5242 | if (!kvm_cpu_cap_has(X86_FEATURE_RDTSCP)) |
9dbe6cf9 PB |
5243 | continue; |
5244 | break; | |
bf8c55d8 CP |
5245 | case MSR_IA32_RTIT_CTL: |
5246 | case MSR_IA32_RTIT_STATUS: | |
7b874c26 | 5247 | if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT)) |
bf8c55d8 CP |
5248 | continue; |
5249 | break; | |
5250 | case MSR_IA32_RTIT_CR3_MATCH: | |
7b874c26 | 5251 | if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) || |
bf8c55d8 CP |
5252 | !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering)) |
5253 | continue; | |
5254 | break; | |
5255 | case MSR_IA32_RTIT_OUTPUT_BASE: | |
5256 | case MSR_IA32_RTIT_OUTPUT_MASK: | |
7b874c26 | 5257 | if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) || |
bf8c55d8 CP |
5258 | (!intel_pt_validate_hw_cap(PT_CAP_topa_output) && |
5259 | !intel_pt_validate_hw_cap(PT_CAP_single_range_output))) | |
5260 | continue; | |
5261 | break; | |
5262 | case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B: { | |
7b874c26 | 5263 | if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) || |
7a5ee6ed | 5264 | msrs_to_save_all[i] - MSR_IA32_RTIT_ADDR0_A >= |
bf8c55d8 CP |
5265 | intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2) |
5266 | continue; | |
5267 | break; | |
cf05a67b | 5268 | case MSR_ARCH_PERFMON_PERFCTR0 ... MSR_ARCH_PERFMON_PERFCTR0 + 17: |
7a5ee6ed | 5269 | if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_PERFCTR0 >= |
24c29b7a PB |
5270 | min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp)) |
5271 | continue; | |
5272 | break; | |
cf05a67b | 5273 | case MSR_ARCH_PERFMON_EVENTSEL0 ... MSR_ARCH_PERFMON_EVENTSEL0 + 17: |
7a5ee6ed | 5274 | if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_EVENTSEL0 >= |
24c29b7a PB |
5275 | min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp)) |
5276 | continue; | |
bf8c55d8 | 5277 | } |
93c4adc7 PB |
5278 | default: |
5279 | break; | |
5280 | } | |
5281 | ||
7a5ee6ed | 5282 | msrs_to_save[num_msrs_to_save++] = msrs_to_save_all[i]; |
043405e1 | 5283 | } |
62ef68bb | 5284 | |
7a5ee6ed | 5285 | for (i = 0; i < ARRAY_SIZE(emulated_msrs_all); i++) { |
afaf0b2f | 5286 | if (!kvm_x86_ops.has_emulated_msr(emulated_msrs_all[i])) |
bc226f07 | 5287 | continue; |
62ef68bb | 5288 | |
7a5ee6ed | 5289 | emulated_msrs[num_emulated_msrs++] = emulated_msrs_all[i]; |
62ef68bb | 5290 | } |
801e459a | 5291 | |
7a5ee6ed | 5292 | for (i = 0; i < ARRAY_SIZE(msr_based_features_all); i++) { |
801e459a TL |
5293 | struct kvm_msr_entry msr; |
5294 | ||
7a5ee6ed | 5295 | msr.index = msr_based_features_all[i]; |
66421c1e | 5296 | if (kvm_get_msr_feature(&msr)) |
801e459a TL |
5297 | continue; |
5298 | ||
7a5ee6ed | 5299 | msr_based_features[num_msr_based_features++] = msr_based_features_all[i]; |
801e459a | 5300 | } |
043405e1 CO |
5301 | } |
5302 | ||
bda9020e MT |
5303 | static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len, |
5304 | const void *v) | |
bbd9b64e | 5305 | { |
70252a10 AK |
5306 | int handled = 0; |
5307 | int n; | |
5308 | ||
5309 | do { | |
5310 | n = min(len, 8); | |
bce87cce | 5311 | if (!(lapic_in_kernel(vcpu) && |
e32edf4f NN |
5312 | !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v)) |
5313 | && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v)) | |
70252a10 AK |
5314 | break; |
5315 | handled += n; | |
5316 | addr += n; | |
5317 | len -= n; | |
5318 | v += n; | |
5319 | } while (len); | |
bbd9b64e | 5320 | |
70252a10 | 5321 | return handled; |
bbd9b64e CO |
5322 | } |
5323 | ||
bda9020e | 5324 | static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v) |
bbd9b64e | 5325 | { |
70252a10 AK |
5326 | int handled = 0; |
5327 | int n; | |
5328 | ||
5329 | do { | |
5330 | n = min(len, 8); | |
bce87cce | 5331 | if (!(lapic_in_kernel(vcpu) && |
e32edf4f NN |
5332 | !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev, |
5333 | addr, n, v)) | |
5334 | && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v)) | |
70252a10 | 5335 | break; |
e39d200f | 5336 | trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v); |
70252a10 AK |
5337 | handled += n; |
5338 | addr += n; | |
5339 | len -= n; | |
5340 | v += n; | |
5341 | } while (len); | |
bbd9b64e | 5342 | |
70252a10 | 5343 | return handled; |
bbd9b64e CO |
5344 | } |
5345 | ||
2dafc6c2 GN |
5346 | static void kvm_set_segment(struct kvm_vcpu *vcpu, |
5347 | struct kvm_segment *var, int seg) | |
5348 | { | |
afaf0b2f | 5349 | kvm_x86_ops.set_segment(vcpu, var, seg); |
2dafc6c2 GN |
5350 | } |
5351 | ||
5352 | void kvm_get_segment(struct kvm_vcpu *vcpu, | |
5353 | struct kvm_segment *var, int seg) | |
5354 | { | |
afaf0b2f | 5355 | kvm_x86_ops.get_segment(vcpu, var, seg); |
2dafc6c2 GN |
5356 | } |
5357 | ||
54987b7a PB |
5358 | gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access, |
5359 | struct x86_exception *exception) | |
02f59dc9 JR |
5360 | { |
5361 | gpa_t t_gpa; | |
02f59dc9 JR |
5362 | |
5363 | BUG_ON(!mmu_is_nested(vcpu)); | |
5364 | ||
5365 | /* NPT walks are always user-walks */ | |
5366 | access |= PFERR_USER_MASK; | |
44dd3ffa | 5367 | t_gpa = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception); |
02f59dc9 JR |
5368 | |
5369 | return t_gpa; | |
5370 | } | |
5371 | ||
ab9ae313 AK |
5372 | gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva, |
5373 | struct x86_exception *exception) | |
1871c602 | 5374 | { |
afaf0b2f | 5375 | u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0; |
ab9ae313 | 5376 | return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception); |
1871c602 GN |
5377 | } |
5378 | ||
ab9ae313 AK |
5379 | gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva, |
5380 | struct x86_exception *exception) | |
1871c602 | 5381 | { |
afaf0b2f | 5382 | u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0; |
1871c602 | 5383 | access |= PFERR_FETCH_MASK; |
ab9ae313 | 5384 | return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception); |
1871c602 GN |
5385 | } |
5386 | ||
ab9ae313 AK |
5387 | gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva, |
5388 | struct x86_exception *exception) | |
1871c602 | 5389 | { |
afaf0b2f | 5390 | u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0; |
1871c602 | 5391 | access |= PFERR_WRITE_MASK; |
ab9ae313 | 5392 | return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception); |
1871c602 GN |
5393 | } |
5394 | ||
5395 | /* uses this to access any guest's mapped memory without checking CPL */ | |
ab9ae313 AK |
5396 | gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva, |
5397 | struct x86_exception *exception) | |
1871c602 | 5398 | { |
ab9ae313 | 5399 | return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception); |
1871c602 GN |
5400 | } |
5401 | ||
5402 | static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes, | |
5403 | struct kvm_vcpu *vcpu, u32 access, | |
bcc55cba | 5404 | struct x86_exception *exception) |
bbd9b64e CO |
5405 | { |
5406 | void *data = val; | |
10589a46 | 5407 | int r = X86EMUL_CONTINUE; |
bbd9b64e CO |
5408 | |
5409 | while (bytes) { | |
14dfe855 | 5410 | gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access, |
ab9ae313 | 5411 | exception); |
bbd9b64e | 5412 | unsigned offset = addr & (PAGE_SIZE-1); |
77c2002e | 5413 | unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset); |
bbd9b64e CO |
5414 | int ret; |
5415 | ||
bcc55cba | 5416 | if (gpa == UNMAPPED_GVA) |
ab9ae313 | 5417 | return X86EMUL_PROPAGATE_FAULT; |
54bf36aa PB |
5418 | ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data, |
5419 | offset, toread); | |
10589a46 | 5420 | if (ret < 0) { |
c3cd7ffa | 5421 | r = X86EMUL_IO_NEEDED; |
10589a46 MT |
5422 | goto out; |
5423 | } | |
bbd9b64e | 5424 | |
77c2002e IE |
5425 | bytes -= toread; |
5426 | data += toread; | |
5427 | addr += toread; | |
bbd9b64e | 5428 | } |
10589a46 | 5429 | out: |
10589a46 | 5430 | return r; |
bbd9b64e | 5431 | } |
77c2002e | 5432 | |
1871c602 | 5433 | /* used for instruction fetching */ |
0f65dd70 AK |
5434 | static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt, |
5435 | gva_t addr, void *val, unsigned int bytes, | |
bcc55cba | 5436 | struct x86_exception *exception) |
1871c602 | 5437 | { |
0f65dd70 | 5438 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
afaf0b2f | 5439 | u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0; |
44583cba PB |
5440 | unsigned offset; |
5441 | int ret; | |
0f65dd70 | 5442 | |
44583cba PB |
5443 | /* Inline kvm_read_guest_virt_helper for speed. */ |
5444 | gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK, | |
5445 | exception); | |
5446 | if (unlikely(gpa == UNMAPPED_GVA)) | |
5447 | return X86EMUL_PROPAGATE_FAULT; | |
5448 | ||
5449 | offset = addr & (PAGE_SIZE-1); | |
5450 | if (WARN_ON(offset + bytes > PAGE_SIZE)) | |
5451 | bytes = (unsigned)PAGE_SIZE - offset; | |
54bf36aa PB |
5452 | ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val, |
5453 | offset, bytes); | |
44583cba PB |
5454 | if (unlikely(ret < 0)) |
5455 | return X86EMUL_IO_NEEDED; | |
5456 | ||
5457 | return X86EMUL_CONTINUE; | |
1871c602 GN |
5458 | } |
5459 | ||
ce14e868 | 5460 | int kvm_read_guest_virt(struct kvm_vcpu *vcpu, |
0f65dd70 | 5461 | gva_t addr, void *val, unsigned int bytes, |
bcc55cba | 5462 | struct x86_exception *exception) |
1871c602 | 5463 | { |
afaf0b2f | 5464 | u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0; |
0f65dd70 | 5465 | |
353c0956 PB |
5466 | /* |
5467 | * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED | |
5468 | * is returned, but our callers are not ready for that and they blindly | |
5469 | * call kvm_inject_page_fault. Ensure that they at least do not leak | |
5470 | * uninitialized kernel stack memory into cr2 and error code. | |
5471 | */ | |
5472 | memset(exception, 0, sizeof(*exception)); | |
1871c602 | 5473 | return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, |
bcc55cba | 5474 | exception); |
1871c602 | 5475 | } |
064aea77 | 5476 | EXPORT_SYMBOL_GPL(kvm_read_guest_virt); |
1871c602 | 5477 | |
ce14e868 PB |
5478 | static int emulator_read_std(struct x86_emulate_ctxt *ctxt, |
5479 | gva_t addr, void *val, unsigned int bytes, | |
3c9fa24c | 5480 | struct x86_exception *exception, bool system) |
1871c602 | 5481 | { |
0f65dd70 | 5482 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
3c9fa24c PB |
5483 | u32 access = 0; |
5484 | ||
afaf0b2f | 5485 | if (!system && kvm_x86_ops.get_cpl(vcpu) == 3) |
3c9fa24c PB |
5486 | access |= PFERR_USER_MASK; |
5487 | ||
5488 | return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception); | |
1871c602 GN |
5489 | } |
5490 | ||
7a036a6f RK |
5491 | static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt, |
5492 | unsigned long addr, void *val, unsigned int bytes) | |
5493 | { | |
5494 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); | |
5495 | int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes); | |
5496 | ||
5497 | return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE; | |
5498 | } | |
5499 | ||
ce14e868 PB |
5500 | static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes, |
5501 | struct kvm_vcpu *vcpu, u32 access, | |
5502 | struct x86_exception *exception) | |
77c2002e IE |
5503 | { |
5504 | void *data = val; | |
5505 | int r = X86EMUL_CONTINUE; | |
5506 | ||
5507 | while (bytes) { | |
14dfe855 | 5508 | gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, |
ce14e868 | 5509 | access, |
ab9ae313 | 5510 | exception); |
77c2002e IE |
5511 | unsigned offset = addr & (PAGE_SIZE-1); |
5512 | unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset); | |
5513 | int ret; | |
5514 | ||
bcc55cba | 5515 | if (gpa == UNMAPPED_GVA) |
ab9ae313 | 5516 | return X86EMUL_PROPAGATE_FAULT; |
54bf36aa | 5517 | ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite); |
77c2002e | 5518 | if (ret < 0) { |
c3cd7ffa | 5519 | r = X86EMUL_IO_NEEDED; |
77c2002e IE |
5520 | goto out; |
5521 | } | |
5522 | ||
5523 | bytes -= towrite; | |
5524 | data += towrite; | |
5525 | addr += towrite; | |
5526 | } | |
5527 | out: | |
5528 | return r; | |
5529 | } | |
ce14e868 PB |
5530 | |
5531 | static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val, | |
3c9fa24c PB |
5532 | unsigned int bytes, struct x86_exception *exception, |
5533 | bool system) | |
ce14e868 PB |
5534 | { |
5535 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); | |
3c9fa24c PB |
5536 | u32 access = PFERR_WRITE_MASK; |
5537 | ||
afaf0b2f | 5538 | if (!system && kvm_x86_ops.get_cpl(vcpu) == 3) |
3c9fa24c | 5539 | access |= PFERR_USER_MASK; |
ce14e868 PB |
5540 | |
5541 | return kvm_write_guest_virt_helper(addr, val, bytes, vcpu, | |
3c9fa24c | 5542 | access, exception); |
ce14e868 PB |
5543 | } |
5544 | ||
5545 | int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val, | |
5546 | unsigned int bytes, struct x86_exception *exception) | |
5547 | { | |
c595ceee PB |
5548 | /* kvm_write_guest_virt_system can pull in tons of pages. */ |
5549 | vcpu->arch.l1tf_flush_l1d = true; | |
5550 | ||
541ab2ae FH |
5551 | /* |
5552 | * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED | |
5553 | * is returned, but our callers are not ready for that and they blindly | |
5554 | * call kvm_inject_page_fault. Ensure that they at least do not leak | |
5555 | * uninitialized kernel stack memory into cr2 and error code. | |
5556 | */ | |
5557 | memset(exception, 0, sizeof(*exception)); | |
ce14e868 PB |
5558 | return kvm_write_guest_virt_helper(addr, val, bytes, vcpu, |
5559 | PFERR_WRITE_MASK, exception); | |
5560 | } | |
6a4d7550 | 5561 | EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system); |
77c2002e | 5562 | |
082d06ed WL |
5563 | int handle_ud(struct kvm_vcpu *vcpu) |
5564 | { | |
b3dc0695 | 5565 | static const char kvm_emulate_prefix[] = { __KVM_EMULATE_PREFIX }; |
6c86eedc | 5566 | int emul_type = EMULTYPE_TRAP_UD; |
6c86eedc WL |
5567 | char sig[5]; /* ud2; .ascii "kvm" */ |
5568 | struct x86_exception e; | |
5569 | ||
5570 | if (force_emulation_prefix && | |
3c9fa24c PB |
5571 | kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu), |
5572 | sig, sizeof(sig), &e) == 0 && | |
b3dc0695 | 5573 | memcmp(sig, kvm_emulate_prefix, sizeof(sig)) == 0) { |
6c86eedc | 5574 | kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig)); |
b4000606 | 5575 | emul_type = EMULTYPE_TRAP_UD_FORCED; |
6c86eedc | 5576 | } |
082d06ed | 5577 | |
60fc3d02 | 5578 | return kvm_emulate_instruction(vcpu, emul_type); |
082d06ed WL |
5579 | } |
5580 | EXPORT_SYMBOL_GPL(handle_ud); | |
5581 | ||
0f89b207 TL |
5582 | static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva, |
5583 | gpa_t gpa, bool write) | |
5584 | { | |
5585 | /* For APIC access vmexit */ | |
5586 | if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE) | |
5587 | return 1; | |
5588 | ||
5589 | if (vcpu_match_mmio_gpa(vcpu, gpa)) { | |
5590 | trace_vcpu_match_mmio(gva, gpa, write, true); | |
5591 | return 1; | |
5592 | } | |
5593 | ||
5594 | return 0; | |
5595 | } | |
5596 | ||
af7cc7d1 XG |
5597 | static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva, |
5598 | gpa_t *gpa, struct x86_exception *exception, | |
5599 | bool write) | |
5600 | { | |
afaf0b2f | 5601 | u32 access = ((kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0) |
97d64b78 | 5602 | | (write ? PFERR_WRITE_MASK : 0); |
af7cc7d1 | 5603 | |
be94f6b7 HH |
5604 | /* |
5605 | * currently PKRU is only applied to ept enabled guest so | |
5606 | * there is no pkey in EPT page table for L1 guest or EPT | |
5607 | * shadow page table for L2 guest. | |
5608 | */ | |
97d64b78 | 5609 | if (vcpu_match_mmio_gva(vcpu, gva) |
97ec8c06 | 5610 | && !permission_fault(vcpu, vcpu->arch.walk_mmu, |
871bd034 | 5611 | vcpu->arch.mmio_access, 0, access)) { |
bebb106a XG |
5612 | *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT | |
5613 | (gva & (PAGE_SIZE - 1)); | |
4f022648 | 5614 | trace_vcpu_match_mmio(gva, *gpa, write, false); |
bebb106a XG |
5615 | return 1; |
5616 | } | |
5617 | ||
af7cc7d1 XG |
5618 | *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception); |
5619 | ||
5620 | if (*gpa == UNMAPPED_GVA) | |
5621 | return -1; | |
5622 | ||
0f89b207 | 5623 | return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write); |
af7cc7d1 XG |
5624 | } |
5625 | ||
3200f405 | 5626 | int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa, |
bcc55cba | 5627 | const void *val, int bytes) |
bbd9b64e CO |
5628 | { |
5629 | int ret; | |
5630 | ||
54bf36aa | 5631 | ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes); |
9f811285 | 5632 | if (ret < 0) |
bbd9b64e | 5633 | return 0; |
0eb05bf2 | 5634 | kvm_page_track_write(vcpu, gpa, val, bytes); |
bbd9b64e CO |
5635 | return 1; |
5636 | } | |
5637 | ||
77d197b2 XG |
5638 | struct read_write_emulator_ops { |
5639 | int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val, | |
5640 | int bytes); | |
5641 | int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa, | |
5642 | void *val, int bytes); | |
5643 | int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa, | |
5644 | int bytes, void *val); | |
5645 | int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa, | |
5646 | void *val, int bytes); | |
5647 | bool write; | |
5648 | }; | |
5649 | ||
5650 | static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes) | |
5651 | { | |
5652 | if (vcpu->mmio_read_completed) { | |
77d197b2 | 5653 | trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes, |
e39d200f | 5654 | vcpu->mmio_fragments[0].gpa, val); |
77d197b2 XG |
5655 | vcpu->mmio_read_completed = 0; |
5656 | return 1; | |
5657 | } | |
5658 | ||
5659 | return 0; | |
5660 | } | |
5661 | ||
5662 | static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa, | |
5663 | void *val, int bytes) | |
5664 | { | |
54bf36aa | 5665 | return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes); |
77d197b2 XG |
5666 | } |
5667 | ||
5668 | static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa, | |
5669 | void *val, int bytes) | |
5670 | { | |
5671 | return emulator_write_phys(vcpu, gpa, val, bytes); | |
5672 | } | |
5673 | ||
5674 | static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val) | |
5675 | { | |
e39d200f | 5676 | trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val); |
77d197b2 XG |
5677 | return vcpu_mmio_write(vcpu, gpa, bytes, val); |
5678 | } | |
5679 | ||
5680 | static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, | |
5681 | void *val, int bytes) | |
5682 | { | |
e39d200f | 5683 | trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL); |
77d197b2 XG |
5684 | return X86EMUL_IO_NEEDED; |
5685 | } | |
5686 | ||
5687 | static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, | |
5688 | void *val, int bytes) | |
5689 | { | |
f78146b0 AK |
5690 | struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0]; |
5691 | ||
87da7e66 | 5692 | memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len)); |
77d197b2 XG |
5693 | return X86EMUL_CONTINUE; |
5694 | } | |
5695 | ||
0fbe9b0b | 5696 | static const struct read_write_emulator_ops read_emultor = { |
77d197b2 XG |
5697 | .read_write_prepare = read_prepare, |
5698 | .read_write_emulate = read_emulate, | |
5699 | .read_write_mmio = vcpu_mmio_read, | |
5700 | .read_write_exit_mmio = read_exit_mmio, | |
5701 | }; | |
5702 | ||
0fbe9b0b | 5703 | static const struct read_write_emulator_ops write_emultor = { |
77d197b2 XG |
5704 | .read_write_emulate = write_emulate, |
5705 | .read_write_mmio = write_mmio, | |
5706 | .read_write_exit_mmio = write_exit_mmio, | |
5707 | .write = true, | |
5708 | }; | |
5709 | ||
22388a3c XG |
5710 | static int emulator_read_write_onepage(unsigned long addr, void *val, |
5711 | unsigned int bytes, | |
5712 | struct x86_exception *exception, | |
5713 | struct kvm_vcpu *vcpu, | |
0fbe9b0b | 5714 | const struct read_write_emulator_ops *ops) |
bbd9b64e | 5715 | { |
af7cc7d1 XG |
5716 | gpa_t gpa; |
5717 | int handled, ret; | |
22388a3c | 5718 | bool write = ops->write; |
f78146b0 | 5719 | struct kvm_mmio_fragment *frag; |
c9b8b07c | 5720 | struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt; |
0f89b207 TL |
5721 | |
5722 | /* | |
5723 | * If the exit was due to a NPF we may already have a GPA. | |
5724 | * If the GPA is present, use it to avoid the GVA to GPA table walk. | |
5725 | * Note, this cannot be used on string operations since string | |
5726 | * operation using rep will only have the initial GPA from the NPF | |
5727 | * occurred. | |
5728 | */ | |
744e699c SC |
5729 | if (ctxt->gpa_available && emulator_can_use_gpa(ctxt) && |
5730 | (addr & ~PAGE_MASK) == (ctxt->gpa_val & ~PAGE_MASK)) { | |
5731 | gpa = ctxt->gpa_val; | |
618232e2 BS |
5732 | ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write); |
5733 | } else { | |
5734 | ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write); | |
5735 | if (ret < 0) | |
5736 | return X86EMUL_PROPAGATE_FAULT; | |
0f89b207 | 5737 | } |
10589a46 | 5738 | |
618232e2 | 5739 | if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes)) |
bbd9b64e CO |
5740 | return X86EMUL_CONTINUE; |
5741 | ||
bbd9b64e CO |
5742 | /* |
5743 | * Is this MMIO handled locally? | |
5744 | */ | |
22388a3c | 5745 | handled = ops->read_write_mmio(vcpu, gpa, bytes, val); |
70252a10 | 5746 | if (handled == bytes) |
bbd9b64e | 5747 | return X86EMUL_CONTINUE; |
bbd9b64e | 5748 | |
70252a10 AK |
5749 | gpa += handled; |
5750 | bytes -= handled; | |
5751 | val += handled; | |
5752 | ||
87da7e66 XG |
5753 | WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS); |
5754 | frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++]; | |
5755 | frag->gpa = gpa; | |
5756 | frag->data = val; | |
5757 | frag->len = bytes; | |
f78146b0 | 5758 | return X86EMUL_CONTINUE; |
bbd9b64e CO |
5759 | } |
5760 | ||
52eb5a6d XL |
5761 | static int emulator_read_write(struct x86_emulate_ctxt *ctxt, |
5762 | unsigned long addr, | |
22388a3c XG |
5763 | void *val, unsigned int bytes, |
5764 | struct x86_exception *exception, | |
0fbe9b0b | 5765 | const struct read_write_emulator_ops *ops) |
bbd9b64e | 5766 | { |
0f65dd70 | 5767 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
f78146b0 AK |
5768 | gpa_t gpa; |
5769 | int rc; | |
5770 | ||
5771 | if (ops->read_write_prepare && | |
5772 | ops->read_write_prepare(vcpu, val, bytes)) | |
5773 | return X86EMUL_CONTINUE; | |
5774 | ||
5775 | vcpu->mmio_nr_fragments = 0; | |
0f65dd70 | 5776 | |
bbd9b64e CO |
5777 | /* Crossing a page boundary? */ |
5778 | if (((addr + bytes - 1) ^ addr) & PAGE_MASK) { | |
f78146b0 | 5779 | int now; |
bbd9b64e CO |
5780 | |
5781 | now = -addr & ~PAGE_MASK; | |
22388a3c XG |
5782 | rc = emulator_read_write_onepage(addr, val, now, exception, |
5783 | vcpu, ops); | |
5784 | ||
bbd9b64e CO |
5785 | if (rc != X86EMUL_CONTINUE) |
5786 | return rc; | |
5787 | addr += now; | |
bac15531 NA |
5788 | if (ctxt->mode != X86EMUL_MODE_PROT64) |
5789 | addr = (u32)addr; | |
bbd9b64e CO |
5790 | val += now; |
5791 | bytes -= now; | |
5792 | } | |
22388a3c | 5793 | |
f78146b0 AK |
5794 | rc = emulator_read_write_onepage(addr, val, bytes, exception, |
5795 | vcpu, ops); | |
5796 | if (rc != X86EMUL_CONTINUE) | |
5797 | return rc; | |
5798 | ||
5799 | if (!vcpu->mmio_nr_fragments) | |
5800 | return rc; | |
5801 | ||
5802 | gpa = vcpu->mmio_fragments[0].gpa; | |
5803 | ||
5804 | vcpu->mmio_needed = 1; | |
5805 | vcpu->mmio_cur_fragment = 0; | |
5806 | ||
87da7e66 | 5807 | vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len); |
f78146b0 AK |
5808 | vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write; |
5809 | vcpu->run->exit_reason = KVM_EXIT_MMIO; | |
5810 | vcpu->run->mmio.phys_addr = gpa; | |
5811 | ||
5812 | return ops->read_write_exit_mmio(vcpu, gpa, val, bytes); | |
22388a3c XG |
5813 | } |
5814 | ||
5815 | static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt, | |
5816 | unsigned long addr, | |
5817 | void *val, | |
5818 | unsigned int bytes, | |
5819 | struct x86_exception *exception) | |
5820 | { | |
5821 | return emulator_read_write(ctxt, addr, val, bytes, | |
5822 | exception, &read_emultor); | |
5823 | } | |
5824 | ||
52eb5a6d | 5825 | static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt, |
22388a3c XG |
5826 | unsigned long addr, |
5827 | const void *val, | |
5828 | unsigned int bytes, | |
5829 | struct x86_exception *exception) | |
5830 | { | |
5831 | return emulator_read_write(ctxt, addr, (void *)val, bytes, | |
5832 | exception, &write_emultor); | |
bbd9b64e | 5833 | } |
bbd9b64e | 5834 | |
daea3e73 AK |
5835 | #define CMPXCHG_TYPE(t, ptr, old, new) \ |
5836 | (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old)) | |
5837 | ||
5838 | #ifdef CONFIG_X86_64 | |
5839 | # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new) | |
5840 | #else | |
5841 | # define CMPXCHG64(ptr, old, new) \ | |
9749a6c0 | 5842 | (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old)) |
daea3e73 AK |
5843 | #endif |
5844 | ||
0f65dd70 AK |
5845 | static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt, |
5846 | unsigned long addr, | |
bbd9b64e CO |
5847 | const void *old, |
5848 | const void *new, | |
5849 | unsigned int bytes, | |
0f65dd70 | 5850 | struct x86_exception *exception) |
bbd9b64e | 5851 | { |
42e35f80 | 5852 | struct kvm_host_map map; |
0f65dd70 | 5853 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
9de6fe3c | 5854 | u64 page_line_mask; |
daea3e73 | 5855 | gpa_t gpa; |
daea3e73 AK |
5856 | char *kaddr; |
5857 | bool exchanged; | |
2bacc55c | 5858 | |
daea3e73 AK |
5859 | /* guests cmpxchg8b have to be emulated atomically */ |
5860 | if (bytes > 8 || (bytes & (bytes - 1))) | |
5861 | goto emul_write; | |
10589a46 | 5862 | |
daea3e73 | 5863 | gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL); |
2bacc55c | 5864 | |
daea3e73 AK |
5865 | if (gpa == UNMAPPED_GVA || |
5866 | (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE) | |
5867 | goto emul_write; | |
2bacc55c | 5868 | |
9de6fe3c XL |
5869 | /* |
5870 | * Emulate the atomic as a straight write to avoid #AC if SLD is | |
5871 | * enabled in the host and the access splits a cache line. | |
5872 | */ | |
5873 | if (boot_cpu_has(X86_FEATURE_SPLIT_LOCK_DETECT)) | |
5874 | page_line_mask = ~(cache_line_size() - 1); | |
5875 | else | |
5876 | page_line_mask = PAGE_MASK; | |
5877 | ||
5878 | if (((gpa + bytes - 1) & page_line_mask) != (gpa & page_line_mask)) | |
daea3e73 | 5879 | goto emul_write; |
72dc67a6 | 5880 | |
42e35f80 | 5881 | if (kvm_vcpu_map(vcpu, gpa_to_gfn(gpa), &map)) |
c19b8bd6 | 5882 | goto emul_write; |
72dc67a6 | 5883 | |
42e35f80 KA |
5884 | kaddr = map.hva + offset_in_page(gpa); |
5885 | ||
daea3e73 AK |
5886 | switch (bytes) { |
5887 | case 1: | |
5888 | exchanged = CMPXCHG_TYPE(u8, kaddr, old, new); | |
5889 | break; | |
5890 | case 2: | |
5891 | exchanged = CMPXCHG_TYPE(u16, kaddr, old, new); | |
5892 | break; | |
5893 | case 4: | |
5894 | exchanged = CMPXCHG_TYPE(u32, kaddr, old, new); | |
5895 | break; | |
5896 | case 8: | |
5897 | exchanged = CMPXCHG64(kaddr, old, new); | |
5898 | break; | |
5899 | default: | |
5900 | BUG(); | |
2bacc55c | 5901 | } |
42e35f80 KA |
5902 | |
5903 | kvm_vcpu_unmap(vcpu, &map, true); | |
daea3e73 AK |
5904 | |
5905 | if (!exchanged) | |
5906 | return X86EMUL_CMPXCHG_FAILED; | |
5907 | ||
0eb05bf2 | 5908 | kvm_page_track_write(vcpu, gpa, new, bytes); |
8f6abd06 GN |
5909 | |
5910 | return X86EMUL_CONTINUE; | |
4a5f48f6 | 5911 | |
3200f405 | 5912 | emul_write: |
daea3e73 | 5913 | printk_once(KERN_WARNING "kvm: emulating exchange as write\n"); |
2bacc55c | 5914 | |
0f65dd70 | 5915 | return emulator_write_emulated(ctxt, addr, new, bytes, exception); |
bbd9b64e CO |
5916 | } |
5917 | ||
cf8f70bf GN |
5918 | static int kernel_pio(struct kvm_vcpu *vcpu, void *pd) |
5919 | { | |
cbfc6c91 | 5920 | int r = 0, i; |
cf8f70bf | 5921 | |
cbfc6c91 WL |
5922 | for (i = 0; i < vcpu->arch.pio.count; i++) { |
5923 | if (vcpu->arch.pio.in) | |
5924 | r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port, | |
5925 | vcpu->arch.pio.size, pd); | |
5926 | else | |
5927 | r = kvm_io_bus_write(vcpu, KVM_PIO_BUS, | |
5928 | vcpu->arch.pio.port, vcpu->arch.pio.size, | |
5929 | pd); | |
5930 | if (r) | |
5931 | break; | |
5932 | pd += vcpu->arch.pio.size; | |
5933 | } | |
cf8f70bf GN |
5934 | return r; |
5935 | } | |
5936 | ||
6f6fbe98 XG |
5937 | static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size, |
5938 | unsigned short port, void *val, | |
5939 | unsigned int count, bool in) | |
cf8f70bf | 5940 | { |
cf8f70bf | 5941 | vcpu->arch.pio.port = port; |
6f6fbe98 | 5942 | vcpu->arch.pio.in = in; |
7972995b | 5943 | vcpu->arch.pio.count = count; |
cf8f70bf GN |
5944 | vcpu->arch.pio.size = size; |
5945 | ||
5946 | if (!kernel_pio(vcpu, vcpu->arch.pio_data)) { | |
7972995b | 5947 | vcpu->arch.pio.count = 0; |
cf8f70bf GN |
5948 | return 1; |
5949 | } | |
5950 | ||
5951 | vcpu->run->exit_reason = KVM_EXIT_IO; | |
6f6fbe98 | 5952 | vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT; |
cf8f70bf GN |
5953 | vcpu->run->io.size = size; |
5954 | vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE; | |
5955 | vcpu->run->io.count = count; | |
5956 | vcpu->run->io.port = port; | |
5957 | ||
5958 | return 0; | |
5959 | } | |
5960 | ||
2e3bb4d8 SC |
5961 | static int emulator_pio_in(struct kvm_vcpu *vcpu, int size, |
5962 | unsigned short port, void *val, unsigned int count) | |
cf8f70bf | 5963 | { |
6f6fbe98 | 5964 | int ret; |
ca1d4a9e | 5965 | |
6f6fbe98 XG |
5966 | if (vcpu->arch.pio.count) |
5967 | goto data_avail; | |
cf8f70bf | 5968 | |
cbfc6c91 WL |
5969 | memset(vcpu->arch.pio_data, 0, size * count); |
5970 | ||
6f6fbe98 XG |
5971 | ret = emulator_pio_in_out(vcpu, size, port, val, count, true); |
5972 | if (ret) { | |
5973 | data_avail: | |
5974 | memcpy(val, vcpu->arch.pio_data, size * count); | |
1171903d | 5975 | trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data); |
7972995b | 5976 | vcpu->arch.pio.count = 0; |
cf8f70bf GN |
5977 | return 1; |
5978 | } | |
5979 | ||
cf8f70bf GN |
5980 | return 0; |
5981 | } | |
5982 | ||
2e3bb4d8 SC |
5983 | static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt, |
5984 | int size, unsigned short port, void *val, | |
5985 | unsigned int count) | |
6f6fbe98 | 5986 | { |
2e3bb4d8 | 5987 | return emulator_pio_in(emul_to_vcpu(ctxt), size, port, val, count); |
6f6fbe98 | 5988 | |
2e3bb4d8 | 5989 | } |
6f6fbe98 | 5990 | |
2e3bb4d8 SC |
5991 | static int emulator_pio_out(struct kvm_vcpu *vcpu, int size, |
5992 | unsigned short port, const void *val, | |
5993 | unsigned int count) | |
5994 | { | |
6f6fbe98 | 5995 | memcpy(vcpu->arch.pio_data, val, size * count); |
1171903d | 5996 | trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data); |
6f6fbe98 XG |
5997 | return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false); |
5998 | } | |
5999 | ||
2e3bb4d8 SC |
6000 | static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt, |
6001 | int size, unsigned short port, | |
6002 | const void *val, unsigned int count) | |
6003 | { | |
6004 | return emulator_pio_out(emul_to_vcpu(ctxt), size, port, val, count); | |
6005 | } | |
6006 | ||
bbd9b64e CO |
6007 | static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg) |
6008 | { | |
afaf0b2f | 6009 | return kvm_x86_ops.get_segment_base(vcpu, seg); |
bbd9b64e CO |
6010 | } |
6011 | ||
3cb16fe7 | 6012 | static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address) |
bbd9b64e | 6013 | { |
3cb16fe7 | 6014 | kvm_mmu_invlpg(emul_to_vcpu(ctxt), address); |
bbd9b64e CO |
6015 | } |
6016 | ||
ae6a2375 | 6017 | static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu) |
f5f48ee1 SY |
6018 | { |
6019 | if (!need_emulate_wbinvd(vcpu)) | |
6020 | return X86EMUL_CONTINUE; | |
6021 | ||
afaf0b2f | 6022 | if (kvm_x86_ops.has_wbinvd_exit()) { |
2eec7343 JK |
6023 | int cpu = get_cpu(); |
6024 | ||
6025 | cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask); | |
f5f48ee1 SY |
6026 | smp_call_function_many(vcpu->arch.wbinvd_dirty_mask, |
6027 | wbinvd_ipi, NULL, 1); | |
2eec7343 | 6028 | put_cpu(); |
f5f48ee1 | 6029 | cpumask_clear(vcpu->arch.wbinvd_dirty_mask); |
2eec7343 JK |
6030 | } else |
6031 | wbinvd(); | |
f5f48ee1 SY |
6032 | return X86EMUL_CONTINUE; |
6033 | } | |
5cb56059 JS |
6034 | |
6035 | int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu) | |
6036 | { | |
6affcbed KH |
6037 | kvm_emulate_wbinvd_noskip(vcpu); |
6038 | return kvm_skip_emulated_instruction(vcpu); | |
5cb56059 | 6039 | } |
f5f48ee1 SY |
6040 | EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd); |
6041 | ||
5cb56059 JS |
6042 | |
6043 | ||
bcaf5cc5 AK |
6044 | static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt) |
6045 | { | |
5cb56059 | 6046 | kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt)); |
bcaf5cc5 AK |
6047 | } |
6048 | ||
52eb5a6d XL |
6049 | static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, |
6050 | unsigned long *dest) | |
bbd9b64e | 6051 | { |
16f8a6f9 | 6052 | return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest); |
bbd9b64e CO |
6053 | } |
6054 | ||
52eb5a6d XL |
6055 | static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, |
6056 | unsigned long value) | |
bbd9b64e | 6057 | { |
338dbc97 | 6058 | |
717746e3 | 6059 | return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value); |
bbd9b64e CO |
6060 | } |
6061 | ||
52a46617 | 6062 | static u64 mk_cr_64(u64 curr_cr, u32 new_val) |
5fdbf976 | 6063 | { |
52a46617 | 6064 | return (curr_cr & ~((1ULL << 32) - 1)) | new_val; |
5fdbf976 MT |
6065 | } |
6066 | ||
717746e3 | 6067 | static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr) |
bbd9b64e | 6068 | { |
717746e3 | 6069 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
52a46617 GN |
6070 | unsigned long value; |
6071 | ||
6072 | switch (cr) { | |
6073 | case 0: | |
6074 | value = kvm_read_cr0(vcpu); | |
6075 | break; | |
6076 | case 2: | |
6077 | value = vcpu->arch.cr2; | |
6078 | break; | |
6079 | case 3: | |
9f8fe504 | 6080 | value = kvm_read_cr3(vcpu); |
52a46617 GN |
6081 | break; |
6082 | case 4: | |
6083 | value = kvm_read_cr4(vcpu); | |
6084 | break; | |
6085 | case 8: | |
6086 | value = kvm_get_cr8(vcpu); | |
6087 | break; | |
6088 | default: | |
a737f256 | 6089 | kvm_err("%s: unexpected cr %u\n", __func__, cr); |
52a46617 GN |
6090 | return 0; |
6091 | } | |
6092 | ||
6093 | return value; | |
6094 | } | |
6095 | ||
717746e3 | 6096 | static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val) |
52a46617 | 6097 | { |
717746e3 | 6098 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
0f12244f GN |
6099 | int res = 0; |
6100 | ||
52a46617 GN |
6101 | switch (cr) { |
6102 | case 0: | |
49a9b07e | 6103 | res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val)); |
52a46617 GN |
6104 | break; |
6105 | case 2: | |
6106 | vcpu->arch.cr2 = val; | |
6107 | break; | |
6108 | case 3: | |
2390218b | 6109 | res = kvm_set_cr3(vcpu, val); |
52a46617 GN |
6110 | break; |
6111 | case 4: | |
a83b29c6 | 6112 | res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val)); |
52a46617 GN |
6113 | break; |
6114 | case 8: | |
eea1cff9 | 6115 | res = kvm_set_cr8(vcpu, val); |
52a46617 GN |
6116 | break; |
6117 | default: | |
a737f256 | 6118 | kvm_err("%s: unexpected cr %u\n", __func__, cr); |
0f12244f | 6119 | res = -1; |
52a46617 | 6120 | } |
0f12244f GN |
6121 | |
6122 | return res; | |
52a46617 GN |
6123 | } |
6124 | ||
717746e3 | 6125 | static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt) |
9c537244 | 6126 | { |
afaf0b2f | 6127 | return kvm_x86_ops.get_cpl(emul_to_vcpu(ctxt)); |
9c537244 GN |
6128 | } |
6129 | ||
4bff1e86 | 6130 | static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt) |
2dafc6c2 | 6131 | { |
afaf0b2f | 6132 | kvm_x86_ops.get_gdt(emul_to_vcpu(ctxt), dt); |
2dafc6c2 GN |
6133 | } |
6134 | ||
4bff1e86 | 6135 | static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt) |
160ce1f1 | 6136 | { |
afaf0b2f | 6137 | kvm_x86_ops.get_idt(emul_to_vcpu(ctxt), dt); |
160ce1f1 MG |
6138 | } |
6139 | ||
1ac9d0cf AK |
6140 | static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt) |
6141 | { | |
afaf0b2f | 6142 | kvm_x86_ops.set_gdt(emul_to_vcpu(ctxt), dt); |
1ac9d0cf AK |
6143 | } |
6144 | ||
6145 | static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt) | |
6146 | { | |
afaf0b2f | 6147 | kvm_x86_ops.set_idt(emul_to_vcpu(ctxt), dt); |
1ac9d0cf AK |
6148 | } |
6149 | ||
4bff1e86 AK |
6150 | static unsigned long emulator_get_cached_segment_base( |
6151 | struct x86_emulate_ctxt *ctxt, int seg) | |
5951c442 | 6152 | { |
4bff1e86 | 6153 | return get_segment_base(emul_to_vcpu(ctxt), seg); |
5951c442 GN |
6154 | } |
6155 | ||
1aa36616 AK |
6156 | static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector, |
6157 | struct desc_struct *desc, u32 *base3, | |
6158 | int seg) | |
2dafc6c2 GN |
6159 | { |
6160 | struct kvm_segment var; | |
6161 | ||
4bff1e86 | 6162 | kvm_get_segment(emul_to_vcpu(ctxt), &var, seg); |
1aa36616 | 6163 | *selector = var.selector; |
2dafc6c2 | 6164 | |
378a8b09 GN |
6165 | if (var.unusable) { |
6166 | memset(desc, 0, sizeof(*desc)); | |
f0367ee1 RK |
6167 | if (base3) |
6168 | *base3 = 0; | |
2dafc6c2 | 6169 | return false; |
378a8b09 | 6170 | } |
2dafc6c2 GN |
6171 | |
6172 | if (var.g) | |
6173 | var.limit >>= 12; | |
6174 | set_desc_limit(desc, var.limit); | |
6175 | set_desc_base(desc, (unsigned long)var.base); | |
5601d05b GN |
6176 | #ifdef CONFIG_X86_64 |
6177 | if (base3) | |
6178 | *base3 = var.base >> 32; | |
6179 | #endif | |
2dafc6c2 GN |
6180 | desc->type = var.type; |
6181 | desc->s = var.s; | |
6182 | desc->dpl = var.dpl; | |
6183 | desc->p = var.present; | |
6184 | desc->avl = var.avl; | |
6185 | desc->l = var.l; | |
6186 | desc->d = var.db; | |
6187 | desc->g = var.g; | |
6188 | ||
6189 | return true; | |
6190 | } | |
6191 | ||
1aa36616 AK |
6192 | static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector, |
6193 | struct desc_struct *desc, u32 base3, | |
6194 | int seg) | |
2dafc6c2 | 6195 | { |
4bff1e86 | 6196 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
2dafc6c2 GN |
6197 | struct kvm_segment var; |
6198 | ||
1aa36616 | 6199 | var.selector = selector; |
2dafc6c2 | 6200 | var.base = get_desc_base(desc); |
5601d05b GN |
6201 | #ifdef CONFIG_X86_64 |
6202 | var.base |= ((u64)base3) << 32; | |
6203 | #endif | |
2dafc6c2 GN |
6204 | var.limit = get_desc_limit(desc); |
6205 | if (desc->g) | |
6206 | var.limit = (var.limit << 12) | 0xfff; | |
6207 | var.type = desc->type; | |
2dafc6c2 GN |
6208 | var.dpl = desc->dpl; |
6209 | var.db = desc->d; | |
6210 | var.s = desc->s; | |
6211 | var.l = desc->l; | |
6212 | var.g = desc->g; | |
6213 | var.avl = desc->avl; | |
6214 | var.present = desc->p; | |
6215 | var.unusable = !var.present; | |
6216 | var.padding = 0; | |
6217 | ||
6218 | kvm_set_segment(vcpu, &var, seg); | |
6219 | return; | |
6220 | } | |
6221 | ||
717746e3 AK |
6222 | static int emulator_get_msr(struct x86_emulate_ctxt *ctxt, |
6223 | u32 msr_index, u64 *pdata) | |
6224 | { | |
f20935d8 | 6225 | return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata); |
717746e3 AK |
6226 | } |
6227 | ||
6228 | static int emulator_set_msr(struct x86_emulate_ctxt *ctxt, | |
6229 | u32 msr_index, u64 data) | |
6230 | { | |
f20935d8 | 6231 | return kvm_set_msr(emul_to_vcpu(ctxt), msr_index, data); |
717746e3 AK |
6232 | } |
6233 | ||
64d60670 PB |
6234 | static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt) |
6235 | { | |
6236 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); | |
6237 | ||
6238 | return vcpu->arch.smbase; | |
6239 | } | |
6240 | ||
6241 | static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase) | |
6242 | { | |
6243 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); | |
6244 | ||
6245 | vcpu->arch.smbase = smbase; | |
6246 | } | |
6247 | ||
67f4d428 NA |
6248 | static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt, |
6249 | u32 pmc) | |
6250 | { | |
98ff80f5 | 6251 | return kvm_pmu_is_valid_rdpmc_ecx(emul_to_vcpu(ctxt), pmc); |
67f4d428 NA |
6252 | } |
6253 | ||
222d21aa AK |
6254 | static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt, |
6255 | u32 pmc, u64 *pdata) | |
6256 | { | |
c6702c9d | 6257 | return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata); |
222d21aa AK |
6258 | } |
6259 | ||
6c3287f7 AK |
6260 | static void emulator_halt(struct x86_emulate_ctxt *ctxt) |
6261 | { | |
6262 | emul_to_vcpu(ctxt)->arch.halt_request = 1; | |
6263 | } | |
6264 | ||
2953538e | 6265 | static int emulator_intercept(struct x86_emulate_ctxt *ctxt, |
8a76d7f2 | 6266 | struct x86_instruction_info *info, |
c4f035c6 AK |
6267 | enum x86_intercept_stage stage) |
6268 | { | |
afaf0b2f | 6269 | return kvm_x86_ops.check_intercept(emul_to_vcpu(ctxt), info, stage, |
21f1b8f2 | 6270 | &ctxt->exception); |
c4f035c6 AK |
6271 | } |
6272 | ||
e911eb3b | 6273 | static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt, |
f91af517 SC |
6274 | u32 *eax, u32 *ebx, u32 *ecx, u32 *edx, |
6275 | bool exact_only) | |
bdb42f5a | 6276 | { |
f91af517 | 6277 | return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, exact_only); |
bdb42f5a SB |
6278 | } |
6279 | ||
5ae78e95 SC |
6280 | static bool emulator_guest_has_long_mode(struct x86_emulate_ctxt *ctxt) |
6281 | { | |
6282 | return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_LM); | |
6283 | } | |
6284 | ||
6285 | static bool emulator_guest_has_movbe(struct x86_emulate_ctxt *ctxt) | |
6286 | { | |
6287 | return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_MOVBE); | |
6288 | } | |
6289 | ||
6290 | static bool emulator_guest_has_fxsr(struct x86_emulate_ctxt *ctxt) | |
6291 | { | |
6292 | return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_FXSR); | |
6293 | } | |
6294 | ||
dd856efa AK |
6295 | static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg) |
6296 | { | |
6297 | return kvm_register_read(emul_to_vcpu(ctxt), reg); | |
6298 | } | |
6299 | ||
6300 | static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val) | |
6301 | { | |
6302 | kvm_register_write(emul_to_vcpu(ctxt), reg, val); | |
6303 | } | |
6304 | ||
801806d9 NA |
6305 | static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked) |
6306 | { | |
afaf0b2f | 6307 | kvm_x86_ops.set_nmi_mask(emul_to_vcpu(ctxt), masked); |
801806d9 NA |
6308 | } |
6309 | ||
6ed071f0 LP |
6310 | static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt) |
6311 | { | |
6312 | return emul_to_vcpu(ctxt)->arch.hflags; | |
6313 | } | |
6314 | ||
6315 | static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags) | |
6316 | { | |
c5833c7a | 6317 | emul_to_vcpu(ctxt)->arch.hflags = emul_flags; |
6ed071f0 LP |
6318 | } |
6319 | ||
ed19321f SC |
6320 | static int emulator_pre_leave_smm(struct x86_emulate_ctxt *ctxt, |
6321 | const char *smstate) | |
0234bf88 | 6322 | { |
afaf0b2f | 6323 | return kvm_x86_ops.pre_leave_smm(emul_to_vcpu(ctxt), smstate); |
0234bf88 LP |
6324 | } |
6325 | ||
c5833c7a SC |
6326 | static void emulator_post_leave_smm(struct x86_emulate_ctxt *ctxt) |
6327 | { | |
6328 | kvm_smm_changed(emul_to_vcpu(ctxt)); | |
6329 | } | |
6330 | ||
02d4160f VK |
6331 | static int emulator_set_xcr(struct x86_emulate_ctxt *ctxt, u32 index, u64 xcr) |
6332 | { | |
6333 | return __kvm_set_xcr(emul_to_vcpu(ctxt), index, xcr); | |
6334 | } | |
6335 | ||
0225fb50 | 6336 | static const struct x86_emulate_ops emulate_ops = { |
dd856efa AK |
6337 | .read_gpr = emulator_read_gpr, |
6338 | .write_gpr = emulator_write_gpr, | |
ce14e868 PB |
6339 | .read_std = emulator_read_std, |
6340 | .write_std = emulator_write_std, | |
7a036a6f | 6341 | .read_phys = kvm_read_guest_phys_system, |
1871c602 | 6342 | .fetch = kvm_fetch_guest_virt, |
bbd9b64e CO |
6343 | .read_emulated = emulator_read_emulated, |
6344 | .write_emulated = emulator_write_emulated, | |
6345 | .cmpxchg_emulated = emulator_cmpxchg_emulated, | |
3cb16fe7 | 6346 | .invlpg = emulator_invlpg, |
cf8f70bf GN |
6347 | .pio_in_emulated = emulator_pio_in_emulated, |
6348 | .pio_out_emulated = emulator_pio_out_emulated, | |
1aa36616 AK |
6349 | .get_segment = emulator_get_segment, |
6350 | .set_segment = emulator_set_segment, | |
5951c442 | 6351 | .get_cached_segment_base = emulator_get_cached_segment_base, |
2dafc6c2 | 6352 | .get_gdt = emulator_get_gdt, |
160ce1f1 | 6353 | .get_idt = emulator_get_idt, |
1ac9d0cf AK |
6354 | .set_gdt = emulator_set_gdt, |
6355 | .set_idt = emulator_set_idt, | |
52a46617 GN |
6356 | .get_cr = emulator_get_cr, |
6357 | .set_cr = emulator_set_cr, | |
9c537244 | 6358 | .cpl = emulator_get_cpl, |
35aa5375 GN |
6359 | .get_dr = emulator_get_dr, |
6360 | .set_dr = emulator_set_dr, | |
64d60670 PB |
6361 | .get_smbase = emulator_get_smbase, |
6362 | .set_smbase = emulator_set_smbase, | |
717746e3 AK |
6363 | .set_msr = emulator_set_msr, |
6364 | .get_msr = emulator_get_msr, | |
67f4d428 | 6365 | .check_pmc = emulator_check_pmc, |
222d21aa | 6366 | .read_pmc = emulator_read_pmc, |
6c3287f7 | 6367 | .halt = emulator_halt, |
bcaf5cc5 | 6368 | .wbinvd = emulator_wbinvd, |
d6aa1000 | 6369 | .fix_hypercall = emulator_fix_hypercall, |
c4f035c6 | 6370 | .intercept = emulator_intercept, |
bdb42f5a | 6371 | .get_cpuid = emulator_get_cpuid, |
5ae78e95 SC |
6372 | .guest_has_long_mode = emulator_guest_has_long_mode, |
6373 | .guest_has_movbe = emulator_guest_has_movbe, | |
6374 | .guest_has_fxsr = emulator_guest_has_fxsr, | |
801806d9 | 6375 | .set_nmi_mask = emulator_set_nmi_mask, |
6ed071f0 LP |
6376 | .get_hflags = emulator_get_hflags, |
6377 | .set_hflags = emulator_set_hflags, | |
0234bf88 | 6378 | .pre_leave_smm = emulator_pre_leave_smm, |
c5833c7a | 6379 | .post_leave_smm = emulator_post_leave_smm, |
02d4160f | 6380 | .set_xcr = emulator_set_xcr, |
bbd9b64e CO |
6381 | }; |
6382 | ||
95cb2295 GN |
6383 | static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask) |
6384 | { | |
afaf0b2f | 6385 | u32 int_shadow = kvm_x86_ops.get_interrupt_shadow(vcpu); |
95cb2295 GN |
6386 | /* |
6387 | * an sti; sti; sequence only disable interrupts for the first | |
6388 | * instruction. So, if the last instruction, be it emulated or | |
6389 | * not, left the system with the INT_STI flag enabled, it | |
6390 | * means that the last instruction is an sti. We should not | |
6391 | * leave the flag on in this case. The same goes for mov ss | |
6392 | */ | |
37ccdcbe PB |
6393 | if (int_shadow & mask) |
6394 | mask = 0; | |
6addfc42 | 6395 | if (unlikely(int_shadow || mask)) { |
afaf0b2f | 6396 | kvm_x86_ops.set_interrupt_shadow(vcpu, mask); |
6addfc42 PB |
6397 | if (!mask) |
6398 | kvm_make_request(KVM_REQ_EVENT, vcpu); | |
6399 | } | |
95cb2295 GN |
6400 | } |
6401 | ||
ef54bcfe | 6402 | static bool inject_emulated_exception(struct kvm_vcpu *vcpu) |
54b8486f | 6403 | { |
c9b8b07c | 6404 | struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt; |
da9cb575 | 6405 | if (ctxt->exception.vector == PF_VECTOR) |
ef54bcfe PB |
6406 | return kvm_propagate_fault(vcpu, &ctxt->exception); |
6407 | ||
6408 | if (ctxt->exception.error_code_valid) | |
da9cb575 AK |
6409 | kvm_queue_exception_e(vcpu, ctxt->exception.vector, |
6410 | ctxt->exception.error_code); | |
54b8486f | 6411 | else |
da9cb575 | 6412 | kvm_queue_exception(vcpu, ctxt->exception.vector); |
ef54bcfe | 6413 | return false; |
54b8486f GN |
6414 | } |
6415 | ||
c9b8b07c SC |
6416 | static struct x86_emulate_ctxt *alloc_emulate_ctxt(struct kvm_vcpu *vcpu) |
6417 | { | |
6418 | struct x86_emulate_ctxt *ctxt; | |
6419 | ||
6420 | ctxt = kmem_cache_zalloc(x86_emulator_cache, GFP_KERNEL_ACCOUNT); | |
6421 | if (!ctxt) { | |
6422 | pr_err("kvm: failed to allocate vcpu's emulator\n"); | |
6423 | return NULL; | |
6424 | } | |
6425 | ||
6426 | ctxt->vcpu = vcpu; | |
6427 | ctxt->ops = &emulate_ops; | |
6428 | vcpu->arch.emulate_ctxt = ctxt; | |
6429 | ||
6430 | return ctxt; | |
6431 | } | |
6432 | ||
8ec4722d MG |
6433 | static void init_emulate_ctxt(struct kvm_vcpu *vcpu) |
6434 | { | |
c9b8b07c | 6435 | struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt; |
8ec4722d MG |
6436 | int cs_db, cs_l; |
6437 | ||
afaf0b2f | 6438 | kvm_x86_ops.get_cs_db_l_bits(vcpu, &cs_db, &cs_l); |
8ec4722d | 6439 | |
744e699c | 6440 | ctxt->gpa_available = false; |
adf52235 | 6441 | ctxt->eflags = kvm_get_rflags(vcpu); |
c8401dda PB |
6442 | ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0; |
6443 | ||
adf52235 TY |
6444 | ctxt->eip = kvm_rip_read(vcpu); |
6445 | ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL : | |
6446 | (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 : | |
42bf549f | 6447 | (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 : |
adf52235 TY |
6448 | cs_db ? X86EMUL_MODE_PROT32 : |
6449 | X86EMUL_MODE_PROT16; | |
a584539b | 6450 | BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK); |
64d60670 PB |
6451 | BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK); |
6452 | BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK); | |
adf52235 | 6453 | |
dd856efa | 6454 | init_decode_cache(ctxt); |
7ae441ea | 6455 | vcpu->arch.emulate_regs_need_sync_from_vcpu = false; |
8ec4722d MG |
6456 | } |
6457 | ||
9497e1f2 | 6458 | void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip) |
63995653 | 6459 | { |
c9b8b07c | 6460 | struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt; |
63995653 MG |
6461 | int ret; |
6462 | ||
6463 | init_emulate_ctxt(vcpu); | |
6464 | ||
9dac77fa AK |
6465 | ctxt->op_bytes = 2; |
6466 | ctxt->ad_bytes = 2; | |
6467 | ctxt->_eip = ctxt->eip + inc_eip; | |
9d74191a | 6468 | ret = emulate_int_real(ctxt, irq); |
63995653 | 6469 | |
9497e1f2 SC |
6470 | if (ret != X86EMUL_CONTINUE) { |
6471 | kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu); | |
6472 | } else { | |
6473 | ctxt->eip = ctxt->_eip; | |
6474 | kvm_rip_write(vcpu, ctxt->eip); | |
6475 | kvm_set_rflags(vcpu, ctxt->eflags); | |
6476 | } | |
63995653 MG |
6477 | } |
6478 | EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt); | |
6479 | ||
e2366171 | 6480 | static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type) |
6d77dbfc | 6481 | { |
6d77dbfc GN |
6482 | ++vcpu->stat.insn_emulation_fail; |
6483 | trace_kvm_emulate_insn_failed(vcpu); | |
e2366171 | 6484 | |
42cbf068 SC |
6485 | if (emulation_type & EMULTYPE_VMWARE_GP) { |
6486 | kvm_queue_exception_e(vcpu, GP_VECTOR, 0); | |
60fc3d02 | 6487 | return 1; |
42cbf068 | 6488 | } |
e2366171 | 6489 | |
738fece4 SC |
6490 | if (emulation_type & EMULTYPE_SKIP) { |
6491 | vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR; | |
6492 | vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION; | |
6493 | vcpu->run->internal.ndata = 0; | |
60fc3d02 | 6494 | return 0; |
738fece4 SC |
6495 | } |
6496 | ||
22da61c9 SC |
6497 | kvm_queue_exception(vcpu, UD_VECTOR); |
6498 | ||
afaf0b2f | 6499 | if (!is_guest_mode(vcpu) && kvm_x86_ops.get_cpl(vcpu) == 0) { |
fc3a9157 JR |
6500 | vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR; |
6501 | vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION; | |
6502 | vcpu->run->internal.ndata = 0; | |
60fc3d02 | 6503 | return 0; |
fc3a9157 | 6504 | } |
e2366171 | 6505 | |
60fc3d02 | 6506 | return 1; |
6d77dbfc GN |
6507 | } |
6508 | ||
736c291c | 6509 | static bool reexecute_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa, |
991eebf9 GN |
6510 | bool write_fault_to_shadow_pgtable, |
6511 | int emulation_type) | |
a6f177ef | 6512 | { |
736c291c | 6513 | gpa_t gpa = cr2_or_gpa; |
ba049e93 | 6514 | kvm_pfn_t pfn; |
a6f177ef | 6515 | |
92daa48b | 6516 | if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF)) |
991eebf9 GN |
6517 | return false; |
6518 | ||
92daa48b SC |
6519 | if (WARN_ON_ONCE(is_guest_mode(vcpu)) || |
6520 | WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF))) | |
6c3dfeb6 SC |
6521 | return false; |
6522 | ||
44dd3ffa | 6523 | if (!vcpu->arch.mmu->direct_map) { |
95b3cf69 XG |
6524 | /* |
6525 | * Write permission should be allowed since only | |
6526 | * write access need to be emulated. | |
6527 | */ | |
736c291c | 6528 | gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL); |
a6f177ef | 6529 | |
95b3cf69 XG |
6530 | /* |
6531 | * If the mapping is invalid in guest, let cpu retry | |
6532 | * it to generate fault. | |
6533 | */ | |
6534 | if (gpa == UNMAPPED_GVA) | |
6535 | return true; | |
6536 | } | |
a6f177ef | 6537 | |
8e3d9d06 XG |
6538 | /* |
6539 | * Do not retry the unhandleable instruction if it faults on the | |
6540 | * readonly host memory, otherwise it will goto a infinite loop: | |
6541 | * retry instruction -> write #PF -> emulation fail -> retry | |
6542 | * instruction -> ... | |
6543 | */ | |
6544 | pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa)); | |
95b3cf69 XG |
6545 | |
6546 | /* | |
6547 | * If the instruction failed on the error pfn, it can not be fixed, | |
6548 | * report the error to userspace. | |
6549 | */ | |
6550 | if (is_error_noslot_pfn(pfn)) | |
6551 | return false; | |
6552 | ||
6553 | kvm_release_pfn_clean(pfn); | |
6554 | ||
6555 | /* The instructions are well-emulated on direct mmu. */ | |
44dd3ffa | 6556 | if (vcpu->arch.mmu->direct_map) { |
95b3cf69 XG |
6557 | unsigned int indirect_shadow_pages; |
6558 | ||
6559 | spin_lock(&vcpu->kvm->mmu_lock); | |
6560 | indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages; | |
6561 | spin_unlock(&vcpu->kvm->mmu_lock); | |
6562 | ||
6563 | if (indirect_shadow_pages) | |
6564 | kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa)); | |
6565 | ||
a6f177ef | 6566 | return true; |
8e3d9d06 | 6567 | } |
a6f177ef | 6568 | |
95b3cf69 XG |
6569 | /* |
6570 | * if emulation was due to access to shadowed page table | |
6571 | * and it failed try to unshadow page and re-enter the | |
6572 | * guest to let CPU execute the instruction. | |
6573 | */ | |
6574 | kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa)); | |
93c05d3e XG |
6575 | |
6576 | /* | |
6577 | * If the access faults on its page table, it can not | |
6578 | * be fixed by unprotecting shadow page and it should | |
6579 | * be reported to userspace. | |
6580 | */ | |
6581 | return !write_fault_to_shadow_pgtable; | |
a6f177ef GN |
6582 | } |
6583 | ||
1cb3f3ae | 6584 | static bool retry_instruction(struct x86_emulate_ctxt *ctxt, |
736c291c | 6585 | gpa_t cr2_or_gpa, int emulation_type) |
1cb3f3ae XG |
6586 | { |
6587 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); | |
736c291c | 6588 | unsigned long last_retry_eip, last_retry_addr, gpa = cr2_or_gpa; |
1cb3f3ae XG |
6589 | |
6590 | last_retry_eip = vcpu->arch.last_retry_eip; | |
6591 | last_retry_addr = vcpu->arch.last_retry_addr; | |
6592 | ||
6593 | /* | |
6594 | * If the emulation is caused by #PF and it is non-page_table | |
6595 | * writing instruction, it means the VM-EXIT is caused by shadow | |
6596 | * page protected, we can zap the shadow page and retry this | |
6597 | * instruction directly. | |
6598 | * | |
6599 | * Note: if the guest uses a non-page-table modifying instruction | |
6600 | * on the PDE that points to the instruction, then we will unmap | |
6601 | * the instruction and go to an infinite loop. So, we cache the | |
6602 | * last retried eip and the last fault address, if we meet the eip | |
6603 | * and the address again, we can break out of the potential infinite | |
6604 | * loop. | |
6605 | */ | |
6606 | vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0; | |
6607 | ||
92daa48b | 6608 | if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF)) |
1cb3f3ae XG |
6609 | return false; |
6610 | ||
92daa48b SC |
6611 | if (WARN_ON_ONCE(is_guest_mode(vcpu)) || |
6612 | WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF))) | |
6c3dfeb6 SC |
6613 | return false; |
6614 | ||
1cb3f3ae XG |
6615 | if (x86_page_table_writing_insn(ctxt)) |
6616 | return false; | |
6617 | ||
736c291c | 6618 | if (ctxt->eip == last_retry_eip && last_retry_addr == cr2_or_gpa) |
1cb3f3ae XG |
6619 | return false; |
6620 | ||
6621 | vcpu->arch.last_retry_eip = ctxt->eip; | |
736c291c | 6622 | vcpu->arch.last_retry_addr = cr2_or_gpa; |
1cb3f3ae | 6623 | |
44dd3ffa | 6624 | if (!vcpu->arch.mmu->direct_map) |
736c291c | 6625 | gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL); |
1cb3f3ae | 6626 | |
22368028 | 6627 | kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa)); |
1cb3f3ae XG |
6628 | |
6629 | return true; | |
6630 | } | |
6631 | ||
716d51ab GN |
6632 | static int complete_emulated_mmio(struct kvm_vcpu *vcpu); |
6633 | static int complete_emulated_pio(struct kvm_vcpu *vcpu); | |
6634 | ||
64d60670 | 6635 | static void kvm_smm_changed(struct kvm_vcpu *vcpu) |
a584539b | 6636 | { |
64d60670 | 6637 | if (!(vcpu->arch.hflags & HF_SMM_MASK)) { |
660a5d51 PB |
6638 | /* This is a good place to trace that we are exiting SMM. */ |
6639 | trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false); | |
6640 | ||
c43203ca PB |
6641 | /* Process a latched INIT or SMI, if any. */ |
6642 | kvm_make_request(KVM_REQ_EVENT, vcpu); | |
64d60670 | 6643 | } |
699023e2 PB |
6644 | |
6645 | kvm_mmu_reset_context(vcpu); | |
64d60670 PB |
6646 | } |
6647 | ||
4a1e10d5 PB |
6648 | static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7, |
6649 | unsigned long *db) | |
6650 | { | |
6651 | u32 dr6 = 0; | |
6652 | int i; | |
6653 | u32 enable, rwlen; | |
6654 | ||
6655 | enable = dr7; | |
6656 | rwlen = dr7 >> 16; | |
6657 | for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4) | |
6658 | if ((enable & 3) && (rwlen & 15) == type && db[i] == addr) | |
6659 | dr6 |= (1 << i); | |
6660 | return dr6; | |
6661 | } | |
6662 | ||
120c2c4f | 6663 | static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu) |
663f4c61 PB |
6664 | { |
6665 | struct kvm_run *kvm_run = vcpu->run; | |
6666 | ||
c8401dda PB |
6667 | if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) { |
6668 | kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 | DR6_RTM; | |
d5d260c5 | 6669 | kvm_run->debug.arch.pc = kvm_get_linear_rip(vcpu); |
c8401dda PB |
6670 | kvm_run->debug.arch.exception = DB_VECTOR; |
6671 | kvm_run->exit_reason = KVM_EXIT_DEBUG; | |
60fc3d02 | 6672 | return 0; |
663f4c61 | 6673 | } |
120c2c4f | 6674 | kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS); |
60fc3d02 | 6675 | return 1; |
663f4c61 PB |
6676 | } |
6677 | ||
6affcbed KH |
6678 | int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu) |
6679 | { | |
afaf0b2f | 6680 | unsigned long rflags = kvm_x86_ops.get_rflags(vcpu); |
f8ea7c60 | 6681 | int r; |
6affcbed | 6682 | |
afaf0b2f | 6683 | r = kvm_x86_ops.skip_emulated_instruction(vcpu); |
60fc3d02 | 6684 | if (unlikely(!r)) |
f8ea7c60 | 6685 | return 0; |
c8401dda PB |
6686 | |
6687 | /* | |
6688 | * rflags is the old, "raw" value of the flags. The new value has | |
6689 | * not been saved yet. | |
6690 | * | |
6691 | * This is correct even for TF set by the guest, because "the | |
6692 | * processor will not generate this exception after the instruction | |
6693 | * that sets the TF flag". | |
6694 | */ | |
6695 | if (unlikely(rflags & X86_EFLAGS_TF)) | |
120c2c4f | 6696 | r = kvm_vcpu_do_singlestep(vcpu); |
60fc3d02 | 6697 | return r; |
6affcbed KH |
6698 | } |
6699 | EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction); | |
6700 | ||
4a1e10d5 PB |
6701 | static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r) |
6702 | { | |
4a1e10d5 PB |
6703 | if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) && |
6704 | (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) { | |
82b32774 NA |
6705 | struct kvm_run *kvm_run = vcpu->run; |
6706 | unsigned long eip = kvm_get_linear_rip(vcpu); | |
6707 | u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0, | |
4a1e10d5 PB |
6708 | vcpu->arch.guest_debug_dr7, |
6709 | vcpu->arch.eff_db); | |
6710 | ||
6711 | if (dr6 != 0) { | |
6f43ed01 | 6712 | kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM; |
82b32774 | 6713 | kvm_run->debug.arch.pc = eip; |
4a1e10d5 PB |
6714 | kvm_run->debug.arch.exception = DB_VECTOR; |
6715 | kvm_run->exit_reason = KVM_EXIT_DEBUG; | |
60fc3d02 | 6716 | *r = 0; |
4a1e10d5 PB |
6717 | return true; |
6718 | } | |
6719 | } | |
6720 | ||
4161a569 NA |
6721 | if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) && |
6722 | !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) { | |
82b32774 NA |
6723 | unsigned long eip = kvm_get_linear_rip(vcpu); |
6724 | u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0, | |
4a1e10d5 PB |
6725 | vcpu->arch.dr7, |
6726 | vcpu->arch.db); | |
6727 | ||
6728 | if (dr6 != 0) { | |
4d5523cf | 6729 | kvm_queue_exception_p(vcpu, DB_VECTOR, dr6); |
60fc3d02 | 6730 | *r = 1; |
4a1e10d5 PB |
6731 | return true; |
6732 | } | |
6733 | } | |
6734 | ||
6735 | return false; | |
6736 | } | |
6737 | ||
04789b66 LA |
6738 | static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt) |
6739 | { | |
2d7921c4 AM |
6740 | switch (ctxt->opcode_len) { |
6741 | case 1: | |
6742 | switch (ctxt->b) { | |
6743 | case 0xe4: /* IN */ | |
6744 | case 0xe5: | |
6745 | case 0xec: | |
6746 | case 0xed: | |
6747 | case 0xe6: /* OUT */ | |
6748 | case 0xe7: | |
6749 | case 0xee: | |
6750 | case 0xef: | |
6751 | case 0x6c: /* INS */ | |
6752 | case 0x6d: | |
6753 | case 0x6e: /* OUTS */ | |
6754 | case 0x6f: | |
6755 | return true; | |
6756 | } | |
6757 | break; | |
6758 | case 2: | |
6759 | switch (ctxt->b) { | |
6760 | case 0x33: /* RDPMC */ | |
6761 | return true; | |
6762 | } | |
6763 | break; | |
04789b66 LA |
6764 | } |
6765 | ||
6766 | return false; | |
6767 | } | |
6768 | ||
736c291c SC |
6769 | int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa, |
6770 | int emulation_type, void *insn, int insn_len) | |
bbd9b64e | 6771 | { |
95cb2295 | 6772 | int r; |
c9b8b07c | 6773 | struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt; |
7ae441ea | 6774 | bool writeback = true; |
93c05d3e | 6775 | bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable; |
bbd9b64e | 6776 | |
c595ceee PB |
6777 | vcpu->arch.l1tf_flush_l1d = true; |
6778 | ||
93c05d3e XG |
6779 | /* |
6780 | * Clear write_fault_to_shadow_pgtable here to ensure it is | |
6781 | * never reused. | |
6782 | */ | |
6783 | vcpu->arch.write_fault_to_shadow_pgtable = false; | |
26eef70c | 6784 | kvm_clear_exception_queue(vcpu); |
8d7d8102 | 6785 | |
571008da | 6786 | if (!(emulation_type & EMULTYPE_NO_DECODE)) { |
8ec4722d | 6787 | init_emulate_ctxt(vcpu); |
4a1e10d5 PB |
6788 | |
6789 | /* | |
6790 | * We will reenter on the same instruction since | |
6791 | * we do not set complete_userspace_io. This does not | |
6792 | * handle watchpoints yet, those would be handled in | |
6793 | * the emulate_ops. | |
6794 | */ | |
d391f120 VK |
6795 | if (!(emulation_type & EMULTYPE_SKIP) && |
6796 | kvm_vcpu_check_breakpoint(vcpu, &r)) | |
4a1e10d5 PB |
6797 | return r; |
6798 | ||
9d74191a TY |
6799 | ctxt->interruptibility = 0; |
6800 | ctxt->have_exception = false; | |
e0ad0b47 | 6801 | ctxt->exception.vector = -1; |
9d74191a | 6802 | ctxt->perm_ok = false; |
bbd9b64e | 6803 | |
b51e974f | 6804 | ctxt->ud = emulation_type & EMULTYPE_TRAP_UD; |
4005996e | 6805 | |
9d74191a | 6806 | r = x86_decode_insn(ctxt, insn, insn_len); |
bbd9b64e | 6807 | |
e46479f8 | 6808 | trace_kvm_emulate_insn_start(vcpu); |
f2b5756b | 6809 | ++vcpu->stat.insn_emulation; |
1d2887e2 | 6810 | if (r != EMULATION_OK) { |
b4000606 | 6811 | if ((emulation_type & EMULTYPE_TRAP_UD) || |
c83fad65 SC |
6812 | (emulation_type & EMULTYPE_TRAP_UD_FORCED)) { |
6813 | kvm_queue_exception(vcpu, UD_VECTOR); | |
60fc3d02 | 6814 | return 1; |
c83fad65 | 6815 | } |
736c291c SC |
6816 | if (reexecute_instruction(vcpu, cr2_or_gpa, |
6817 | write_fault_to_spt, | |
6818 | emulation_type)) | |
60fc3d02 | 6819 | return 1; |
8530a79c | 6820 | if (ctxt->have_exception) { |
c8848cee JD |
6821 | /* |
6822 | * #UD should result in just EMULATION_FAILED, and trap-like | |
6823 | * exception should not be encountered during decode. | |
6824 | */ | |
6825 | WARN_ON_ONCE(ctxt->exception.vector == UD_VECTOR || | |
6826 | exception_type(ctxt->exception.vector) == EXCPT_TRAP); | |
8530a79c | 6827 | inject_emulated_exception(vcpu); |
60fc3d02 | 6828 | return 1; |
8530a79c | 6829 | } |
e2366171 | 6830 | return handle_emulation_failure(vcpu, emulation_type); |
bbd9b64e CO |
6831 | } |
6832 | } | |
6833 | ||
42cbf068 SC |
6834 | if ((emulation_type & EMULTYPE_VMWARE_GP) && |
6835 | !is_vmware_backdoor_opcode(ctxt)) { | |
6836 | kvm_queue_exception_e(vcpu, GP_VECTOR, 0); | |
60fc3d02 | 6837 | return 1; |
42cbf068 | 6838 | } |
04789b66 | 6839 | |
1957aa63 SC |
6840 | /* |
6841 | * Note, EMULTYPE_SKIP is intended for use *only* by vendor callbacks | |
6842 | * for kvm_skip_emulated_instruction(). The caller is responsible for | |
6843 | * updating interruptibility state and injecting single-step #DBs. | |
6844 | */ | |
ba8afb6b | 6845 | if (emulation_type & EMULTYPE_SKIP) { |
9dac77fa | 6846 | kvm_rip_write(vcpu, ctxt->_eip); |
bb663c7a NA |
6847 | if (ctxt->eflags & X86_EFLAGS_RF) |
6848 | kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF); | |
60fc3d02 | 6849 | return 1; |
ba8afb6b GN |
6850 | } |
6851 | ||
736c291c | 6852 | if (retry_instruction(ctxt, cr2_or_gpa, emulation_type)) |
60fc3d02 | 6853 | return 1; |
1cb3f3ae | 6854 | |
7ae441ea | 6855 | /* this is needed for vmware backdoor interface to work since it |
4d2179e1 | 6856 | changes registers values during IO operation */ |
7ae441ea GN |
6857 | if (vcpu->arch.emulate_regs_need_sync_from_vcpu) { |
6858 | vcpu->arch.emulate_regs_need_sync_from_vcpu = false; | |
dd856efa | 6859 | emulator_invalidate_register_cache(ctxt); |
7ae441ea | 6860 | } |
4d2179e1 | 6861 | |
5cd21917 | 6862 | restart: |
92daa48b SC |
6863 | if (emulation_type & EMULTYPE_PF) { |
6864 | /* Save the faulting GPA (cr2) in the address field */ | |
6865 | ctxt->exception.address = cr2_or_gpa; | |
6866 | ||
6867 | /* With shadow page tables, cr2 contains a GVA or nGPA. */ | |
6868 | if (vcpu->arch.mmu->direct_map) { | |
744e699c SC |
6869 | ctxt->gpa_available = true; |
6870 | ctxt->gpa_val = cr2_or_gpa; | |
92daa48b SC |
6871 | } |
6872 | } else { | |
6873 | /* Sanitize the address out of an abundance of paranoia. */ | |
6874 | ctxt->exception.address = 0; | |
6875 | } | |
0f89b207 | 6876 | |
9d74191a | 6877 | r = x86_emulate_insn(ctxt); |
bbd9b64e | 6878 | |
775fde86 | 6879 | if (r == EMULATION_INTERCEPTED) |
60fc3d02 | 6880 | return 1; |
775fde86 | 6881 | |
d2ddd1c4 | 6882 | if (r == EMULATION_FAILED) { |
736c291c | 6883 | if (reexecute_instruction(vcpu, cr2_or_gpa, write_fault_to_spt, |
991eebf9 | 6884 | emulation_type)) |
60fc3d02 | 6885 | return 1; |
c3cd7ffa | 6886 | |
e2366171 | 6887 | return handle_emulation_failure(vcpu, emulation_type); |
bbd9b64e CO |
6888 | } |
6889 | ||
9d74191a | 6890 | if (ctxt->have_exception) { |
60fc3d02 | 6891 | r = 1; |
ef54bcfe PB |
6892 | if (inject_emulated_exception(vcpu)) |
6893 | return r; | |
d2ddd1c4 | 6894 | } else if (vcpu->arch.pio.count) { |
0912c977 PB |
6895 | if (!vcpu->arch.pio.in) { |
6896 | /* FIXME: return into emulator if single-stepping. */ | |
3457e419 | 6897 | vcpu->arch.pio.count = 0; |
0912c977 | 6898 | } else { |
7ae441ea | 6899 | writeback = false; |
716d51ab GN |
6900 | vcpu->arch.complete_userspace_io = complete_emulated_pio; |
6901 | } | |
60fc3d02 | 6902 | r = 0; |
7ae441ea | 6903 | } else if (vcpu->mmio_needed) { |
bc8a0aaf SC |
6904 | ++vcpu->stat.mmio_exits; |
6905 | ||
7ae441ea GN |
6906 | if (!vcpu->mmio_is_write) |
6907 | writeback = false; | |
60fc3d02 | 6908 | r = 0; |
716d51ab | 6909 | vcpu->arch.complete_userspace_io = complete_emulated_mmio; |
7ae441ea | 6910 | } else if (r == EMULATION_RESTART) |
5cd21917 | 6911 | goto restart; |
d2ddd1c4 | 6912 | else |
60fc3d02 | 6913 | r = 1; |
f850e2e6 | 6914 | |
7ae441ea | 6915 | if (writeback) { |
afaf0b2f | 6916 | unsigned long rflags = kvm_x86_ops.get_rflags(vcpu); |
9d74191a | 6917 | toggle_interruptibility(vcpu, ctxt->interruptibility); |
7ae441ea | 6918 | vcpu->arch.emulate_regs_need_sync_to_vcpu = false; |
38827dbd | 6919 | if (!ctxt->have_exception || |
75ee23b3 SC |
6920 | exception_type(ctxt->exception.vector) == EXCPT_TRAP) { |
6921 | kvm_rip_write(vcpu, ctxt->eip); | |
60fc3d02 | 6922 | if (r && ctxt->tf) |
120c2c4f | 6923 | r = kvm_vcpu_do_singlestep(vcpu); |
afaf0b2f SC |
6924 | if (kvm_x86_ops.update_emulated_instruction) |
6925 | kvm_x86_ops.update_emulated_instruction(vcpu); | |
38827dbd | 6926 | __kvm_set_rflags(vcpu, ctxt->eflags); |
75ee23b3 | 6927 | } |
6addfc42 PB |
6928 | |
6929 | /* | |
6930 | * For STI, interrupts are shadowed; so KVM_REQ_EVENT will | |
6931 | * do nothing, and it will be requested again as soon as | |
6932 | * the shadow expires. But we still need to check here, | |
6933 | * because POPF has no interrupt shadow. | |
6934 | */ | |
6935 | if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF)) | |
6936 | kvm_make_request(KVM_REQ_EVENT, vcpu); | |
7ae441ea GN |
6937 | } else |
6938 | vcpu->arch.emulate_regs_need_sync_to_vcpu = true; | |
e85d28f8 GN |
6939 | |
6940 | return r; | |
de7d789a | 6941 | } |
c60658d1 SC |
6942 | |
6943 | int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type) | |
6944 | { | |
6945 | return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0); | |
6946 | } | |
6947 | EXPORT_SYMBOL_GPL(kvm_emulate_instruction); | |
6948 | ||
6949 | int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu, | |
6950 | void *insn, int insn_len) | |
6951 | { | |
6952 | return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len); | |
6953 | } | |
6954 | EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer); | |
de7d789a | 6955 | |
8764ed55 SC |
6956 | static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu) |
6957 | { | |
6958 | vcpu->arch.pio.count = 0; | |
6959 | return 1; | |
6960 | } | |
6961 | ||
45def77e SC |
6962 | static int complete_fast_pio_out(struct kvm_vcpu *vcpu) |
6963 | { | |
6964 | vcpu->arch.pio.count = 0; | |
6965 | ||
6966 | if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) | |
6967 | return 1; | |
6968 | ||
6969 | return kvm_skip_emulated_instruction(vcpu); | |
6970 | } | |
6971 | ||
dca7f128 SC |
6972 | static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, |
6973 | unsigned short port) | |
de7d789a | 6974 | { |
de3cd117 | 6975 | unsigned long val = kvm_rax_read(vcpu); |
2e3bb4d8 SC |
6976 | int ret = emulator_pio_out(vcpu, size, port, &val, 1); |
6977 | ||
8764ed55 SC |
6978 | if (ret) |
6979 | return ret; | |
45def77e | 6980 | |
8764ed55 SC |
6981 | /* |
6982 | * Workaround userspace that relies on old KVM behavior of %rip being | |
6983 | * incremented prior to exiting to userspace to handle "OUT 0x7e". | |
6984 | */ | |
6985 | if (port == 0x7e && | |
6986 | kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) { | |
6987 | vcpu->arch.complete_userspace_io = | |
6988 | complete_fast_pio_out_port_0x7e; | |
6989 | kvm_skip_emulated_instruction(vcpu); | |
6990 | } else { | |
45def77e SC |
6991 | vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu); |
6992 | vcpu->arch.complete_userspace_io = complete_fast_pio_out; | |
6993 | } | |
8764ed55 | 6994 | return 0; |
de7d789a | 6995 | } |
de7d789a | 6996 | |
8370c3d0 TL |
6997 | static int complete_fast_pio_in(struct kvm_vcpu *vcpu) |
6998 | { | |
6999 | unsigned long val; | |
7000 | ||
7001 | /* We should only ever be called with arch.pio.count equal to 1 */ | |
7002 | BUG_ON(vcpu->arch.pio.count != 1); | |
7003 | ||
45def77e SC |
7004 | if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) { |
7005 | vcpu->arch.pio.count = 0; | |
7006 | return 1; | |
7007 | } | |
7008 | ||
8370c3d0 | 7009 | /* For size less than 4 we merge, else we zero extend */ |
de3cd117 | 7010 | val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0; |
8370c3d0 TL |
7011 | |
7012 | /* | |
2e3bb4d8 | 7013 | * Since vcpu->arch.pio.count == 1 let emulator_pio_in perform |
8370c3d0 TL |
7014 | * the copy and tracing |
7015 | */ | |
2e3bb4d8 | 7016 | emulator_pio_in(vcpu, vcpu->arch.pio.size, vcpu->arch.pio.port, &val, 1); |
de3cd117 | 7017 | kvm_rax_write(vcpu, val); |
8370c3d0 | 7018 | |
45def77e | 7019 | return kvm_skip_emulated_instruction(vcpu); |
8370c3d0 TL |
7020 | } |
7021 | ||
dca7f128 SC |
7022 | static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size, |
7023 | unsigned short port) | |
8370c3d0 TL |
7024 | { |
7025 | unsigned long val; | |
7026 | int ret; | |
7027 | ||
7028 | /* For size less than 4 we merge, else we zero extend */ | |
de3cd117 | 7029 | val = (size < 4) ? kvm_rax_read(vcpu) : 0; |
8370c3d0 | 7030 | |
2e3bb4d8 | 7031 | ret = emulator_pio_in(vcpu, size, port, &val, 1); |
8370c3d0 | 7032 | if (ret) { |
de3cd117 | 7033 | kvm_rax_write(vcpu, val); |
8370c3d0 TL |
7034 | return ret; |
7035 | } | |
7036 | ||
45def77e | 7037 | vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu); |
8370c3d0 TL |
7038 | vcpu->arch.complete_userspace_io = complete_fast_pio_in; |
7039 | ||
7040 | return 0; | |
7041 | } | |
dca7f128 SC |
7042 | |
7043 | int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in) | |
7044 | { | |
45def77e | 7045 | int ret; |
dca7f128 | 7046 | |
dca7f128 | 7047 | if (in) |
45def77e | 7048 | ret = kvm_fast_pio_in(vcpu, size, port); |
dca7f128 | 7049 | else |
45def77e SC |
7050 | ret = kvm_fast_pio_out(vcpu, size, port); |
7051 | return ret && kvm_skip_emulated_instruction(vcpu); | |
dca7f128 SC |
7052 | } |
7053 | EXPORT_SYMBOL_GPL(kvm_fast_pio); | |
8370c3d0 | 7054 | |
251a5fd6 | 7055 | static int kvmclock_cpu_down_prep(unsigned int cpu) |
8cfdc000 | 7056 | { |
0a3aee0d | 7057 | __this_cpu_write(cpu_tsc_khz, 0); |
251a5fd6 | 7058 | return 0; |
8cfdc000 ZA |
7059 | } |
7060 | ||
7061 | static void tsc_khz_changed(void *data) | |
c8076604 | 7062 | { |
8cfdc000 ZA |
7063 | struct cpufreq_freqs *freq = data; |
7064 | unsigned long khz = 0; | |
7065 | ||
7066 | if (data) | |
7067 | khz = freq->new; | |
7068 | else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) | |
7069 | khz = cpufreq_quick_get(raw_smp_processor_id()); | |
7070 | if (!khz) | |
7071 | khz = tsc_khz; | |
0a3aee0d | 7072 | __this_cpu_write(cpu_tsc_khz, khz); |
c8076604 GH |
7073 | } |
7074 | ||
5fa4ec9c | 7075 | #ifdef CONFIG_X86_64 |
0092e434 VK |
7076 | static void kvm_hyperv_tsc_notifier(void) |
7077 | { | |
0092e434 VK |
7078 | struct kvm *kvm; |
7079 | struct kvm_vcpu *vcpu; | |
7080 | int cpu; | |
7081 | ||
0d9ce162 | 7082 | mutex_lock(&kvm_lock); |
0092e434 VK |
7083 | list_for_each_entry(kvm, &vm_list, vm_list) |
7084 | kvm_make_mclock_inprogress_request(kvm); | |
7085 | ||
7086 | hyperv_stop_tsc_emulation(); | |
7087 | ||
7088 | /* TSC frequency always matches when on Hyper-V */ | |
7089 | for_each_present_cpu(cpu) | |
7090 | per_cpu(cpu_tsc_khz, cpu) = tsc_khz; | |
7091 | kvm_max_guest_tsc_khz = tsc_khz; | |
7092 | ||
7093 | list_for_each_entry(kvm, &vm_list, vm_list) { | |
7094 | struct kvm_arch *ka = &kvm->arch; | |
7095 | ||
7096 | spin_lock(&ka->pvclock_gtod_sync_lock); | |
7097 | ||
7098 | pvclock_update_vm_gtod_copy(kvm); | |
7099 | ||
7100 | kvm_for_each_vcpu(cpu, vcpu, kvm) | |
7101 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu); | |
7102 | ||
7103 | kvm_for_each_vcpu(cpu, vcpu, kvm) | |
7104 | kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu); | |
7105 | ||
7106 | spin_unlock(&ka->pvclock_gtod_sync_lock); | |
7107 | } | |
0d9ce162 | 7108 | mutex_unlock(&kvm_lock); |
0092e434 | 7109 | } |
5fa4ec9c | 7110 | #endif |
0092e434 | 7111 | |
df24014a | 7112 | static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu) |
c8076604 | 7113 | { |
c8076604 GH |
7114 | struct kvm *kvm; |
7115 | struct kvm_vcpu *vcpu; | |
7116 | int i, send_ipi = 0; | |
7117 | ||
8cfdc000 ZA |
7118 | /* |
7119 | * We allow guests to temporarily run on slowing clocks, | |
7120 | * provided we notify them after, or to run on accelerating | |
7121 | * clocks, provided we notify them before. Thus time never | |
7122 | * goes backwards. | |
7123 | * | |
7124 | * However, we have a problem. We can't atomically update | |
7125 | * the frequency of a given CPU from this function; it is | |
7126 | * merely a notifier, which can be called from any CPU. | |
7127 | * Changing the TSC frequency at arbitrary points in time | |
7128 | * requires a recomputation of local variables related to | |
7129 | * the TSC for each VCPU. We must flag these local variables | |
7130 | * to be updated and be sure the update takes place with the | |
7131 | * new frequency before any guests proceed. | |
7132 | * | |
7133 | * Unfortunately, the combination of hotplug CPU and frequency | |
7134 | * change creates an intractable locking scenario; the order | |
7135 | * of when these callouts happen is undefined with respect to | |
7136 | * CPU hotplug, and they can race with each other. As such, | |
7137 | * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is | |
7138 | * undefined; you can actually have a CPU frequency change take | |
7139 | * place in between the computation of X and the setting of the | |
7140 | * variable. To protect against this problem, all updates of | |
7141 | * the per_cpu tsc_khz variable are done in an interrupt | |
7142 | * protected IPI, and all callers wishing to update the value | |
7143 | * must wait for a synchronous IPI to complete (which is trivial | |
7144 | * if the caller is on the CPU already). This establishes the | |
7145 | * necessary total order on variable updates. | |
7146 | * | |
7147 | * Note that because a guest time update may take place | |
7148 | * anytime after the setting of the VCPU's request bit, the | |
7149 | * correct TSC value must be set before the request. However, | |
7150 | * to ensure the update actually makes it to any guest which | |
7151 | * starts running in hardware virtualization between the set | |
7152 | * and the acquisition of the spinlock, we must also ping the | |
7153 | * CPU after setting the request bit. | |
7154 | * | |
7155 | */ | |
7156 | ||
df24014a | 7157 | smp_call_function_single(cpu, tsc_khz_changed, freq, 1); |
c8076604 | 7158 | |
0d9ce162 | 7159 | mutex_lock(&kvm_lock); |
c8076604 | 7160 | list_for_each_entry(kvm, &vm_list, vm_list) { |
988a2cae | 7161 | kvm_for_each_vcpu(i, vcpu, kvm) { |
df24014a | 7162 | if (vcpu->cpu != cpu) |
c8076604 | 7163 | continue; |
c285545f | 7164 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu); |
0d9ce162 | 7165 | if (vcpu->cpu != raw_smp_processor_id()) |
8cfdc000 | 7166 | send_ipi = 1; |
c8076604 GH |
7167 | } |
7168 | } | |
0d9ce162 | 7169 | mutex_unlock(&kvm_lock); |
c8076604 GH |
7170 | |
7171 | if (freq->old < freq->new && send_ipi) { | |
7172 | /* | |
7173 | * We upscale the frequency. Must make the guest | |
7174 | * doesn't see old kvmclock values while running with | |
7175 | * the new frequency, otherwise we risk the guest sees | |
7176 | * time go backwards. | |
7177 | * | |
7178 | * In case we update the frequency for another cpu | |
7179 | * (which might be in guest context) send an interrupt | |
7180 | * to kick the cpu out of guest context. Next time | |
7181 | * guest context is entered kvmclock will be updated, | |
7182 | * so the guest will not see stale values. | |
7183 | */ | |
df24014a | 7184 | smp_call_function_single(cpu, tsc_khz_changed, freq, 1); |
c8076604 | 7185 | } |
df24014a VK |
7186 | } |
7187 | ||
7188 | static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val, | |
7189 | void *data) | |
7190 | { | |
7191 | struct cpufreq_freqs *freq = data; | |
7192 | int cpu; | |
7193 | ||
7194 | if (val == CPUFREQ_PRECHANGE && freq->old > freq->new) | |
7195 | return 0; | |
7196 | if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new) | |
7197 | return 0; | |
7198 | ||
7199 | for_each_cpu(cpu, freq->policy->cpus) | |
7200 | __kvmclock_cpufreq_notifier(freq, cpu); | |
7201 | ||
c8076604 GH |
7202 | return 0; |
7203 | } | |
7204 | ||
7205 | static struct notifier_block kvmclock_cpufreq_notifier_block = { | |
8cfdc000 ZA |
7206 | .notifier_call = kvmclock_cpufreq_notifier |
7207 | }; | |
7208 | ||
251a5fd6 | 7209 | static int kvmclock_cpu_online(unsigned int cpu) |
8cfdc000 | 7210 | { |
251a5fd6 SAS |
7211 | tsc_khz_changed(NULL); |
7212 | return 0; | |
8cfdc000 ZA |
7213 | } |
7214 | ||
b820cc0c ZA |
7215 | static void kvm_timer_init(void) |
7216 | { | |
c285545f | 7217 | max_tsc_khz = tsc_khz; |
460dd42e | 7218 | |
b820cc0c | 7219 | if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) { |
c285545f | 7220 | #ifdef CONFIG_CPU_FREQ |
aaec7c03 | 7221 | struct cpufreq_policy *policy; |
758f588d BP |
7222 | int cpu; |
7223 | ||
3e26f230 | 7224 | cpu = get_cpu(); |
aaec7c03 | 7225 | policy = cpufreq_cpu_get(cpu); |
9a11997e WL |
7226 | if (policy) { |
7227 | if (policy->cpuinfo.max_freq) | |
7228 | max_tsc_khz = policy->cpuinfo.max_freq; | |
7229 | cpufreq_cpu_put(policy); | |
7230 | } | |
3e26f230 | 7231 | put_cpu(); |
c285545f | 7232 | #endif |
b820cc0c ZA |
7233 | cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block, |
7234 | CPUFREQ_TRANSITION_NOTIFIER); | |
7235 | } | |
460dd42e | 7236 | |
73c1b41e | 7237 | cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online", |
251a5fd6 | 7238 | kvmclock_cpu_online, kvmclock_cpu_down_prep); |
b820cc0c ZA |
7239 | } |
7240 | ||
dd60d217 AK |
7241 | DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu); |
7242 | EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu); | |
ff9d07a0 | 7243 | |
f5132b01 | 7244 | int kvm_is_in_guest(void) |
ff9d07a0 | 7245 | { |
086c9855 | 7246 | return __this_cpu_read(current_vcpu) != NULL; |
ff9d07a0 ZY |
7247 | } |
7248 | ||
7249 | static int kvm_is_user_mode(void) | |
7250 | { | |
7251 | int user_mode = 3; | |
dcf46b94 | 7252 | |
086c9855 | 7253 | if (__this_cpu_read(current_vcpu)) |
afaf0b2f | 7254 | user_mode = kvm_x86_ops.get_cpl(__this_cpu_read(current_vcpu)); |
dcf46b94 | 7255 | |
ff9d07a0 ZY |
7256 | return user_mode != 0; |
7257 | } | |
7258 | ||
7259 | static unsigned long kvm_get_guest_ip(void) | |
7260 | { | |
7261 | unsigned long ip = 0; | |
dcf46b94 | 7262 | |
086c9855 AS |
7263 | if (__this_cpu_read(current_vcpu)) |
7264 | ip = kvm_rip_read(__this_cpu_read(current_vcpu)); | |
dcf46b94 | 7265 | |
ff9d07a0 ZY |
7266 | return ip; |
7267 | } | |
7268 | ||
8479e04e LK |
7269 | static void kvm_handle_intel_pt_intr(void) |
7270 | { | |
7271 | struct kvm_vcpu *vcpu = __this_cpu_read(current_vcpu); | |
7272 | ||
7273 | kvm_make_request(KVM_REQ_PMI, vcpu); | |
7274 | __set_bit(MSR_CORE_PERF_GLOBAL_OVF_CTRL_TRACE_TOPA_PMI_BIT, | |
7275 | (unsigned long *)&vcpu->arch.pmu.global_status); | |
7276 | } | |
7277 | ||
ff9d07a0 ZY |
7278 | static struct perf_guest_info_callbacks kvm_guest_cbs = { |
7279 | .is_in_guest = kvm_is_in_guest, | |
7280 | .is_user_mode = kvm_is_user_mode, | |
7281 | .get_guest_ip = kvm_get_guest_ip, | |
8479e04e | 7282 | .handle_intel_pt_intr = kvm_handle_intel_pt_intr, |
ff9d07a0 ZY |
7283 | }; |
7284 | ||
16e8d74d MT |
7285 | #ifdef CONFIG_X86_64 |
7286 | static void pvclock_gtod_update_fn(struct work_struct *work) | |
7287 | { | |
d828199e MT |
7288 | struct kvm *kvm; |
7289 | ||
7290 | struct kvm_vcpu *vcpu; | |
7291 | int i; | |
7292 | ||
0d9ce162 | 7293 | mutex_lock(&kvm_lock); |
d828199e MT |
7294 | list_for_each_entry(kvm, &vm_list, vm_list) |
7295 | kvm_for_each_vcpu(i, vcpu, kvm) | |
105b21bb | 7296 | kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu); |
d828199e | 7297 | atomic_set(&kvm_guest_has_master_clock, 0); |
0d9ce162 | 7298 | mutex_unlock(&kvm_lock); |
16e8d74d MT |
7299 | } |
7300 | ||
7301 | static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn); | |
7302 | ||
7303 | /* | |
7304 | * Notification about pvclock gtod data update. | |
7305 | */ | |
7306 | static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused, | |
7307 | void *priv) | |
7308 | { | |
7309 | struct pvclock_gtod_data *gtod = &pvclock_gtod_data; | |
7310 | struct timekeeper *tk = priv; | |
7311 | ||
7312 | update_pvclock_gtod(tk); | |
7313 | ||
7314 | /* disable master clock if host does not trust, or does not | |
b0c39dc6 | 7315 | * use, TSC based clocksource. |
16e8d74d | 7316 | */ |
b0c39dc6 | 7317 | if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) && |
16e8d74d MT |
7318 | atomic_read(&kvm_guest_has_master_clock) != 0) |
7319 | queue_work(system_long_wq, &pvclock_gtod_work); | |
7320 | ||
7321 | return 0; | |
7322 | } | |
7323 | ||
7324 | static struct notifier_block pvclock_gtod_notifier = { | |
7325 | .notifier_call = pvclock_gtod_notify, | |
7326 | }; | |
7327 | #endif | |
7328 | ||
f8c16bba | 7329 | int kvm_arch_init(void *opaque) |
043405e1 | 7330 | { |
d008dfdb | 7331 | struct kvm_x86_init_ops *ops = opaque; |
b820cc0c | 7332 | int r; |
f8c16bba | 7333 | |
afaf0b2f | 7334 | if (kvm_x86_ops.hardware_enable) { |
f8c16bba | 7335 | printk(KERN_ERR "kvm: already loaded the other module\n"); |
56c6d28a ZX |
7336 | r = -EEXIST; |
7337 | goto out; | |
f8c16bba ZX |
7338 | } |
7339 | ||
7340 | if (!ops->cpu_has_kvm_support()) { | |
ef935c25 | 7341 | pr_err_ratelimited("kvm: no hardware support\n"); |
56c6d28a ZX |
7342 | r = -EOPNOTSUPP; |
7343 | goto out; | |
f8c16bba ZX |
7344 | } |
7345 | if (ops->disabled_by_bios()) { | |
ef935c25 | 7346 | pr_err_ratelimited("kvm: disabled by bios\n"); |
56c6d28a ZX |
7347 | r = -EOPNOTSUPP; |
7348 | goto out; | |
f8c16bba ZX |
7349 | } |
7350 | ||
b666a4b6 MO |
7351 | /* |
7352 | * KVM explicitly assumes that the guest has an FPU and | |
7353 | * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the | |
7354 | * vCPU's FPU state as a fxregs_state struct. | |
7355 | */ | |
7356 | if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) { | |
7357 | printk(KERN_ERR "kvm: inadequate fpu\n"); | |
7358 | r = -EOPNOTSUPP; | |
7359 | goto out; | |
7360 | } | |
7361 | ||
013f6a5d | 7362 | r = -ENOMEM; |
ed8e4812 | 7363 | x86_fpu_cache = kmem_cache_create("x86_fpu", sizeof(struct fpu), |
b666a4b6 MO |
7364 | __alignof__(struct fpu), SLAB_ACCOUNT, |
7365 | NULL); | |
7366 | if (!x86_fpu_cache) { | |
7367 | printk(KERN_ERR "kvm: failed to allocate cache for x86 fpu\n"); | |
7368 | goto out; | |
7369 | } | |
7370 | ||
c9b8b07c SC |
7371 | x86_emulator_cache = kvm_alloc_emulator_cache(); |
7372 | if (!x86_emulator_cache) { | |
7373 | pr_err("kvm: failed to allocate cache for x86 emulator\n"); | |
7374 | goto out_free_x86_fpu_cache; | |
7375 | } | |
7376 | ||
013f6a5d MT |
7377 | shared_msrs = alloc_percpu(struct kvm_shared_msrs); |
7378 | if (!shared_msrs) { | |
7379 | printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n"); | |
c9b8b07c | 7380 | goto out_free_x86_emulator_cache; |
013f6a5d MT |
7381 | } |
7382 | ||
97db56ce AK |
7383 | r = kvm_mmu_module_init(); |
7384 | if (r) | |
013f6a5d | 7385 | goto out_free_percpu; |
97db56ce | 7386 | |
7b52345e | 7387 | kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK, |
ffb128c8 | 7388 | PT_DIRTY_MASK, PT64_NX_MASK, 0, |
d0ec49d4 | 7389 | PT_PRESENT_MASK, 0, sme_me_mask); |
b820cc0c | 7390 | kvm_timer_init(); |
c8076604 | 7391 | |
ff9d07a0 ZY |
7392 | perf_register_guest_info_callbacks(&kvm_guest_cbs); |
7393 | ||
cfc48181 | 7394 | if (boot_cpu_has(X86_FEATURE_XSAVE)) { |
2acf923e | 7395 | host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK); |
cfc48181 SC |
7396 | supported_xcr0 = host_xcr0 & KVM_SUPPORTED_XCR0; |
7397 | } | |
2acf923e | 7398 | |
c5cc421b | 7399 | kvm_lapic_init(); |
0c5f81da WL |
7400 | if (pi_inject_timer == -1) |
7401 | pi_inject_timer = housekeeping_enabled(HK_FLAG_TIMER); | |
16e8d74d MT |
7402 | #ifdef CONFIG_X86_64 |
7403 | pvclock_gtod_register_notifier(&pvclock_gtod_notifier); | |
0092e434 | 7404 | |
5fa4ec9c | 7405 | if (hypervisor_is_type(X86_HYPER_MS_HYPERV)) |
0092e434 | 7406 | set_hv_tscchange_cb(kvm_hyperv_tsc_notifier); |
16e8d74d MT |
7407 | #endif |
7408 | ||
f8c16bba | 7409 | return 0; |
56c6d28a | 7410 | |
013f6a5d MT |
7411 | out_free_percpu: |
7412 | free_percpu(shared_msrs); | |
c9b8b07c SC |
7413 | out_free_x86_emulator_cache: |
7414 | kmem_cache_destroy(x86_emulator_cache); | |
b666a4b6 MO |
7415 | out_free_x86_fpu_cache: |
7416 | kmem_cache_destroy(x86_fpu_cache); | |
56c6d28a | 7417 | out: |
56c6d28a | 7418 | return r; |
043405e1 | 7419 | } |
8776e519 | 7420 | |
f8c16bba ZX |
7421 | void kvm_arch_exit(void) |
7422 | { | |
0092e434 | 7423 | #ifdef CONFIG_X86_64 |
5fa4ec9c | 7424 | if (hypervisor_is_type(X86_HYPER_MS_HYPERV)) |
0092e434 VK |
7425 | clear_hv_tscchange_cb(); |
7426 | #endif | |
cef84c30 | 7427 | kvm_lapic_exit(); |
ff9d07a0 ZY |
7428 | perf_unregister_guest_info_callbacks(&kvm_guest_cbs); |
7429 | ||
888d256e JK |
7430 | if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) |
7431 | cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block, | |
7432 | CPUFREQ_TRANSITION_NOTIFIER); | |
251a5fd6 | 7433 | cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE); |
16e8d74d MT |
7434 | #ifdef CONFIG_X86_64 |
7435 | pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier); | |
7436 | #endif | |
afaf0b2f | 7437 | kvm_x86_ops.hardware_enable = NULL; |
56c6d28a | 7438 | kvm_mmu_module_exit(); |
013f6a5d | 7439 | free_percpu(shared_msrs); |
b666a4b6 | 7440 | kmem_cache_destroy(x86_fpu_cache); |
56c6d28a | 7441 | } |
f8c16bba | 7442 | |
5cb56059 | 7443 | int kvm_vcpu_halt(struct kvm_vcpu *vcpu) |
8776e519 HB |
7444 | { |
7445 | ++vcpu->stat.halt_exits; | |
35754c98 | 7446 | if (lapic_in_kernel(vcpu)) { |
a4535290 | 7447 | vcpu->arch.mp_state = KVM_MP_STATE_HALTED; |
8776e519 HB |
7448 | return 1; |
7449 | } else { | |
7450 | vcpu->run->exit_reason = KVM_EXIT_HLT; | |
7451 | return 0; | |
7452 | } | |
7453 | } | |
5cb56059 JS |
7454 | EXPORT_SYMBOL_GPL(kvm_vcpu_halt); |
7455 | ||
7456 | int kvm_emulate_halt(struct kvm_vcpu *vcpu) | |
7457 | { | |
6affcbed KH |
7458 | int ret = kvm_skip_emulated_instruction(vcpu); |
7459 | /* | |
7460 | * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered | |
7461 | * KVM_EXIT_DEBUG here. | |
7462 | */ | |
7463 | return kvm_vcpu_halt(vcpu) && ret; | |
5cb56059 | 7464 | } |
8776e519 HB |
7465 | EXPORT_SYMBOL_GPL(kvm_emulate_halt); |
7466 | ||
8ef81a9a | 7467 | #ifdef CONFIG_X86_64 |
55dd00a7 MT |
7468 | static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr, |
7469 | unsigned long clock_type) | |
7470 | { | |
7471 | struct kvm_clock_pairing clock_pairing; | |
899a31f5 | 7472 | struct timespec64 ts; |
80fbd89c | 7473 | u64 cycle; |
55dd00a7 MT |
7474 | int ret; |
7475 | ||
7476 | if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK) | |
7477 | return -KVM_EOPNOTSUPP; | |
7478 | ||
7479 | if (kvm_get_walltime_and_clockread(&ts, &cycle) == false) | |
7480 | return -KVM_EOPNOTSUPP; | |
7481 | ||
7482 | clock_pairing.sec = ts.tv_sec; | |
7483 | clock_pairing.nsec = ts.tv_nsec; | |
7484 | clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle); | |
7485 | clock_pairing.flags = 0; | |
bcbfbd8e | 7486 | memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad)); |
55dd00a7 MT |
7487 | |
7488 | ret = 0; | |
7489 | if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing, | |
7490 | sizeof(struct kvm_clock_pairing))) | |
7491 | ret = -KVM_EFAULT; | |
7492 | ||
7493 | return ret; | |
7494 | } | |
8ef81a9a | 7495 | #endif |
55dd00a7 | 7496 | |
6aef266c SV |
7497 | /* |
7498 | * kvm_pv_kick_cpu_op: Kick a vcpu. | |
7499 | * | |
7500 | * @apicid - apicid of vcpu to be kicked. | |
7501 | */ | |
7502 | static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid) | |
7503 | { | |
24d2166b | 7504 | struct kvm_lapic_irq lapic_irq; |
6aef266c | 7505 | |
150a84fe | 7506 | lapic_irq.shorthand = APIC_DEST_NOSHORT; |
c96001c5 | 7507 | lapic_irq.dest_mode = APIC_DEST_PHYSICAL; |
ebd28fcb | 7508 | lapic_irq.level = 0; |
24d2166b | 7509 | lapic_irq.dest_id = apicid; |
93bbf0b8 | 7510 | lapic_irq.msi_redir_hint = false; |
6aef266c | 7511 | |
24d2166b | 7512 | lapic_irq.delivery_mode = APIC_DM_REMRD; |
795a149e | 7513 | kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL); |
6aef266c SV |
7514 | } |
7515 | ||
4e19c36f SS |
7516 | bool kvm_apicv_activated(struct kvm *kvm) |
7517 | { | |
7518 | return (READ_ONCE(kvm->arch.apicv_inhibit_reasons) == 0); | |
7519 | } | |
7520 | EXPORT_SYMBOL_GPL(kvm_apicv_activated); | |
7521 | ||
7522 | void kvm_apicv_init(struct kvm *kvm, bool enable) | |
7523 | { | |
7524 | if (enable) | |
7525 | clear_bit(APICV_INHIBIT_REASON_DISABLE, | |
7526 | &kvm->arch.apicv_inhibit_reasons); | |
7527 | else | |
7528 | set_bit(APICV_INHIBIT_REASON_DISABLE, | |
7529 | &kvm->arch.apicv_inhibit_reasons); | |
7530 | } | |
7531 | EXPORT_SYMBOL_GPL(kvm_apicv_init); | |
7532 | ||
71506297 WL |
7533 | static void kvm_sched_yield(struct kvm *kvm, unsigned long dest_id) |
7534 | { | |
7535 | struct kvm_vcpu *target = NULL; | |
7536 | struct kvm_apic_map *map; | |
7537 | ||
7538 | rcu_read_lock(); | |
7539 | map = rcu_dereference(kvm->arch.apic_map); | |
7540 | ||
7541 | if (likely(map) && dest_id <= map->max_apic_id && map->phys_map[dest_id]) | |
7542 | target = map->phys_map[dest_id]->vcpu; | |
7543 | ||
7544 | rcu_read_unlock(); | |
7545 | ||
266e85a5 | 7546 | if (target && READ_ONCE(target->ready)) |
71506297 WL |
7547 | kvm_vcpu_yield_to(target); |
7548 | } | |
7549 | ||
8776e519 HB |
7550 | int kvm_emulate_hypercall(struct kvm_vcpu *vcpu) |
7551 | { | |
7552 | unsigned long nr, a0, a1, a2, a3, ret; | |
6356ee0c | 7553 | int op_64_bit; |
8776e519 | 7554 | |
696ca779 RK |
7555 | if (kvm_hv_hypercall_enabled(vcpu->kvm)) |
7556 | return kvm_hv_hypercall(vcpu); | |
55cd8e5a | 7557 | |
de3cd117 SC |
7558 | nr = kvm_rax_read(vcpu); |
7559 | a0 = kvm_rbx_read(vcpu); | |
7560 | a1 = kvm_rcx_read(vcpu); | |
7561 | a2 = kvm_rdx_read(vcpu); | |
7562 | a3 = kvm_rsi_read(vcpu); | |
8776e519 | 7563 | |
229456fc | 7564 | trace_kvm_hypercall(nr, a0, a1, a2, a3); |
2714d1d3 | 7565 | |
a449c7aa NA |
7566 | op_64_bit = is_64_bit_mode(vcpu); |
7567 | if (!op_64_bit) { | |
8776e519 HB |
7568 | nr &= 0xFFFFFFFF; |
7569 | a0 &= 0xFFFFFFFF; | |
7570 | a1 &= 0xFFFFFFFF; | |
7571 | a2 &= 0xFFFFFFFF; | |
7572 | a3 &= 0xFFFFFFFF; | |
7573 | } | |
7574 | ||
afaf0b2f | 7575 | if (kvm_x86_ops.get_cpl(vcpu) != 0) { |
07708c4a | 7576 | ret = -KVM_EPERM; |
696ca779 | 7577 | goto out; |
07708c4a JK |
7578 | } |
7579 | ||
8776e519 | 7580 | switch (nr) { |
b93463aa AK |
7581 | case KVM_HC_VAPIC_POLL_IRQ: |
7582 | ret = 0; | |
7583 | break; | |
6aef266c SV |
7584 | case KVM_HC_KICK_CPU: |
7585 | kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1); | |
266e85a5 | 7586 | kvm_sched_yield(vcpu->kvm, a1); |
6aef266c SV |
7587 | ret = 0; |
7588 | break; | |
8ef81a9a | 7589 | #ifdef CONFIG_X86_64 |
55dd00a7 MT |
7590 | case KVM_HC_CLOCK_PAIRING: |
7591 | ret = kvm_pv_clock_pairing(vcpu, a0, a1); | |
7592 | break; | |
1ed199a4 | 7593 | #endif |
4180bf1b WL |
7594 | case KVM_HC_SEND_IPI: |
7595 | ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit); | |
7596 | break; | |
71506297 WL |
7597 | case KVM_HC_SCHED_YIELD: |
7598 | kvm_sched_yield(vcpu->kvm, a0); | |
7599 | ret = 0; | |
7600 | break; | |
8776e519 HB |
7601 | default: |
7602 | ret = -KVM_ENOSYS; | |
7603 | break; | |
7604 | } | |
696ca779 | 7605 | out: |
a449c7aa NA |
7606 | if (!op_64_bit) |
7607 | ret = (u32)ret; | |
de3cd117 | 7608 | kvm_rax_write(vcpu, ret); |
6356ee0c | 7609 | |
f11c3a8d | 7610 | ++vcpu->stat.hypercalls; |
6356ee0c | 7611 | return kvm_skip_emulated_instruction(vcpu); |
8776e519 HB |
7612 | } |
7613 | EXPORT_SYMBOL_GPL(kvm_emulate_hypercall); | |
7614 | ||
b6785def | 7615 | static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt) |
8776e519 | 7616 | { |
d6aa1000 | 7617 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
8776e519 | 7618 | char instruction[3]; |
5fdbf976 | 7619 | unsigned long rip = kvm_rip_read(vcpu); |
8776e519 | 7620 | |
afaf0b2f | 7621 | kvm_x86_ops.patch_hypercall(vcpu, instruction); |
8776e519 | 7622 | |
ce2e852e DV |
7623 | return emulator_write_emulated(ctxt, rip, instruction, 3, |
7624 | &ctxt->exception); | |
8776e519 HB |
7625 | } |
7626 | ||
851ba692 | 7627 | static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu) |
b6c7a5dc | 7628 | { |
782d422b MG |
7629 | return vcpu->run->request_interrupt_window && |
7630 | likely(!pic_in_kernel(vcpu->kvm)); | |
b6c7a5dc HB |
7631 | } |
7632 | ||
851ba692 | 7633 | static void post_kvm_run_save(struct kvm_vcpu *vcpu) |
b6c7a5dc | 7634 | { |
851ba692 AK |
7635 | struct kvm_run *kvm_run = vcpu->run; |
7636 | ||
91586a3b | 7637 | kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0; |
f077825a | 7638 | kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0; |
2d3ad1f4 | 7639 | kvm_run->cr8 = kvm_get_cr8(vcpu); |
b6c7a5dc | 7640 | kvm_run->apic_base = kvm_get_apic_base(vcpu); |
127a457a MG |
7641 | kvm_run->ready_for_interrupt_injection = |
7642 | pic_in_kernel(vcpu->kvm) || | |
782d422b | 7643 | kvm_vcpu_ready_for_interrupt_injection(vcpu); |
b6c7a5dc HB |
7644 | } |
7645 | ||
95ba8273 GN |
7646 | static void update_cr8_intercept(struct kvm_vcpu *vcpu) |
7647 | { | |
7648 | int max_irr, tpr; | |
7649 | ||
afaf0b2f | 7650 | if (!kvm_x86_ops.update_cr8_intercept) |
95ba8273 GN |
7651 | return; |
7652 | ||
bce87cce | 7653 | if (!lapic_in_kernel(vcpu)) |
88c808fd AK |
7654 | return; |
7655 | ||
d62caabb AS |
7656 | if (vcpu->arch.apicv_active) |
7657 | return; | |
7658 | ||
8db3baa2 GN |
7659 | if (!vcpu->arch.apic->vapic_addr) |
7660 | max_irr = kvm_lapic_find_highest_irr(vcpu); | |
7661 | else | |
7662 | max_irr = -1; | |
95ba8273 GN |
7663 | |
7664 | if (max_irr != -1) | |
7665 | max_irr >>= 4; | |
7666 | ||
7667 | tpr = kvm_lapic_get_cr8(vcpu); | |
7668 | ||
afaf0b2f | 7669 | kvm_x86_ops.update_cr8_intercept(vcpu, tpr, max_irr); |
95ba8273 GN |
7670 | } |
7671 | ||
a1c77abb | 7672 | static int inject_pending_event(struct kvm_vcpu *vcpu) |
95ba8273 | 7673 | { |
b6b8a145 JK |
7674 | int r; |
7675 | ||
95ba8273 | 7676 | /* try to reinject previous events if any */ |
664f8e26 | 7677 | |
1a680e35 | 7678 | if (vcpu->arch.exception.injected) |
afaf0b2f | 7679 | kvm_x86_ops.queue_exception(vcpu); |
664f8e26 | 7680 | /* |
a042c26f LA |
7681 | * Do not inject an NMI or interrupt if there is a pending |
7682 | * exception. Exceptions and interrupts are recognized at | |
7683 | * instruction boundaries, i.e. the start of an instruction. | |
7684 | * Trap-like exceptions, e.g. #DB, have higher priority than | |
7685 | * NMIs and interrupts, i.e. traps are recognized before an | |
7686 | * NMI/interrupt that's pending on the same instruction. | |
7687 | * Fault-like exceptions, e.g. #GP and #PF, are the lowest | |
7688 | * priority, but are only generated (pended) during instruction | |
7689 | * execution, i.e. a pending fault-like exception means the | |
7690 | * fault occurred on the *previous* instruction and must be | |
7691 | * serviced prior to recognizing any new events in order to | |
7692 | * fully complete the previous instruction. | |
664f8e26 | 7693 | */ |
1a680e35 LA |
7694 | else if (!vcpu->arch.exception.pending) { |
7695 | if (vcpu->arch.nmi_injected) | |
afaf0b2f | 7696 | kvm_x86_ops.set_nmi(vcpu); |
1a680e35 | 7697 | else if (vcpu->arch.interrupt.injected) |
afaf0b2f | 7698 | kvm_x86_ops.set_irq(vcpu); |
664f8e26 WL |
7699 | } |
7700 | ||
1a680e35 LA |
7701 | /* |
7702 | * Call check_nested_events() even if we reinjected a previous event | |
7703 | * in order for caller to determine if it should require immediate-exit | |
7704 | * from L2 to L1 due to pending L1 events which require exit | |
7705 | * from L2 to L1. | |
7706 | */ | |
afaf0b2f SC |
7707 | if (is_guest_mode(vcpu) && kvm_x86_ops.check_nested_events) { |
7708 | r = kvm_x86_ops.check_nested_events(vcpu); | |
664f8e26 WL |
7709 | if (r != 0) |
7710 | return r; | |
7711 | } | |
7712 | ||
7713 | /* try to inject new event if pending */ | |
b59bb7bd | 7714 | if (vcpu->arch.exception.pending) { |
5c1c85d0 AK |
7715 | trace_kvm_inj_exception(vcpu->arch.exception.nr, |
7716 | vcpu->arch.exception.has_error_code, | |
7717 | vcpu->arch.exception.error_code); | |
d6e8c854 | 7718 | |
1a680e35 | 7719 | WARN_ON_ONCE(vcpu->arch.exception.injected); |
664f8e26 WL |
7720 | vcpu->arch.exception.pending = false; |
7721 | vcpu->arch.exception.injected = true; | |
7722 | ||
d6e8c854 NA |
7723 | if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT) |
7724 | __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) | | |
7725 | X86_EFLAGS_RF); | |
7726 | ||
f10c729f JM |
7727 | if (vcpu->arch.exception.nr == DB_VECTOR) { |
7728 | /* | |
7729 | * This code assumes that nSVM doesn't use | |
7730 | * check_nested_events(). If it does, the | |
7731 | * DR6/DR7 changes should happen before L1 | |
7732 | * gets a #VMEXIT for an intercepted #DB in | |
7733 | * L2. (Under VMX, on the other hand, the | |
7734 | * DR6/DR7 changes should not happen in the | |
7735 | * event of a VM-exit to L1 for an intercepted | |
7736 | * #DB in L2.) | |
7737 | */ | |
7738 | kvm_deliver_exception_payload(vcpu); | |
7739 | if (vcpu->arch.dr7 & DR7_GD) { | |
7740 | vcpu->arch.dr7 &= ~DR7_GD; | |
7741 | kvm_update_dr7(vcpu); | |
7742 | } | |
6bdf0662 NA |
7743 | } |
7744 | ||
afaf0b2f | 7745 | kvm_x86_ops.queue_exception(vcpu); |
1a680e35 LA |
7746 | } |
7747 | ||
7748 | /* Don't consider new event if we re-injected an event */ | |
7749 | if (kvm_event_needs_reinjection(vcpu)) | |
7750 | return 0; | |
7751 | ||
7752 | if (vcpu->arch.smi_pending && !is_smm(vcpu) && | |
afaf0b2f | 7753 | kvm_x86_ops.smi_allowed(vcpu)) { |
c43203ca | 7754 | vcpu->arch.smi_pending = false; |
52797bf9 | 7755 | ++vcpu->arch.smi_count; |
ee2cd4b7 | 7756 | enter_smm(vcpu); |
afaf0b2f | 7757 | } else if (vcpu->arch.nmi_pending && kvm_x86_ops.nmi_allowed(vcpu)) { |
321c5658 YS |
7758 | --vcpu->arch.nmi_pending; |
7759 | vcpu->arch.nmi_injected = true; | |
afaf0b2f | 7760 | kvm_x86_ops.set_nmi(vcpu); |
c7c9c56c | 7761 | } else if (kvm_cpu_has_injectable_intr(vcpu)) { |
9242b5b6 BD |
7762 | /* |
7763 | * Because interrupts can be injected asynchronously, we are | |
7764 | * calling check_nested_events again here to avoid a race condition. | |
7765 | * See https://lkml.org/lkml/2014/7/2/60 for discussion about this | |
7766 | * proposal and current concerns. Perhaps we should be setting | |
7767 | * KVM_REQ_EVENT only on certain events and not unconditionally? | |
7768 | */ | |
afaf0b2f SC |
7769 | if (is_guest_mode(vcpu) && kvm_x86_ops.check_nested_events) { |
7770 | r = kvm_x86_ops.check_nested_events(vcpu); | |
9242b5b6 BD |
7771 | if (r != 0) |
7772 | return r; | |
7773 | } | |
afaf0b2f | 7774 | if (kvm_x86_ops.interrupt_allowed(vcpu)) { |
66fd3f7f GN |
7775 | kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu), |
7776 | false); | |
afaf0b2f | 7777 | kvm_x86_ops.set_irq(vcpu); |
95ba8273 GN |
7778 | } |
7779 | } | |
ee2cd4b7 | 7780 | |
b6b8a145 | 7781 | return 0; |
95ba8273 GN |
7782 | } |
7783 | ||
7460fb4a AK |
7784 | static void process_nmi(struct kvm_vcpu *vcpu) |
7785 | { | |
7786 | unsigned limit = 2; | |
7787 | ||
7788 | /* | |
7789 | * x86 is limited to one NMI running, and one NMI pending after it. | |
7790 | * If an NMI is already in progress, limit further NMIs to just one. | |
7791 | * Otherwise, allow two (and we'll inject the first one immediately). | |
7792 | */ | |
afaf0b2f | 7793 | if (kvm_x86_ops.get_nmi_mask(vcpu) || vcpu->arch.nmi_injected) |
7460fb4a AK |
7794 | limit = 1; |
7795 | ||
7796 | vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0); | |
7797 | vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit); | |
7798 | kvm_make_request(KVM_REQ_EVENT, vcpu); | |
7799 | } | |
7800 | ||
ee2cd4b7 | 7801 | static u32 enter_smm_get_segment_flags(struct kvm_segment *seg) |
660a5d51 PB |
7802 | { |
7803 | u32 flags = 0; | |
7804 | flags |= seg->g << 23; | |
7805 | flags |= seg->db << 22; | |
7806 | flags |= seg->l << 21; | |
7807 | flags |= seg->avl << 20; | |
7808 | flags |= seg->present << 15; | |
7809 | flags |= seg->dpl << 13; | |
7810 | flags |= seg->s << 12; | |
7811 | flags |= seg->type << 8; | |
7812 | return flags; | |
7813 | } | |
7814 | ||
ee2cd4b7 | 7815 | static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n) |
660a5d51 PB |
7816 | { |
7817 | struct kvm_segment seg; | |
7818 | int offset; | |
7819 | ||
7820 | kvm_get_segment(vcpu, &seg, n); | |
7821 | put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector); | |
7822 | ||
7823 | if (n < 3) | |
7824 | offset = 0x7f84 + n * 12; | |
7825 | else | |
7826 | offset = 0x7f2c + (n - 3) * 12; | |
7827 | ||
7828 | put_smstate(u32, buf, offset + 8, seg.base); | |
7829 | put_smstate(u32, buf, offset + 4, seg.limit); | |
ee2cd4b7 | 7830 | put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg)); |
660a5d51 PB |
7831 | } |
7832 | ||
efbb288a | 7833 | #ifdef CONFIG_X86_64 |
ee2cd4b7 | 7834 | static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n) |
660a5d51 PB |
7835 | { |
7836 | struct kvm_segment seg; | |
7837 | int offset; | |
7838 | u16 flags; | |
7839 | ||
7840 | kvm_get_segment(vcpu, &seg, n); | |
7841 | offset = 0x7e00 + n * 16; | |
7842 | ||
ee2cd4b7 | 7843 | flags = enter_smm_get_segment_flags(&seg) >> 8; |
660a5d51 PB |
7844 | put_smstate(u16, buf, offset, seg.selector); |
7845 | put_smstate(u16, buf, offset + 2, flags); | |
7846 | put_smstate(u32, buf, offset + 4, seg.limit); | |
7847 | put_smstate(u64, buf, offset + 8, seg.base); | |
7848 | } | |
efbb288a | 7849 | #endif |
660a5d51 | 7850 | |
ee2cd4b7 | 7851 | static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf) |
660a5d51 PB |
7852 | { |
7853 | struct desc_ptr dt; | |
7854 | struct kvm_segment seg; | |
7855 | unsigned long val; | |
7856 | int i; | |
7857 | ||
7858 | put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu)); | |
7859 | put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu)); | |
7860 | put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu)); | |
7861 | put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu)); | |
7862 | ||
7863 | for (i = 0; i < 8; i++) | |
7864 | put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i)); | |
7865 | ||
7866 | kvm_get_dr(vcpu, 6, &val); | |
7867 | put_smstate(u32, buf, 0x7fcc, (u32)val); | |
7868 | kvm_get_dr(vcpu, 7, &val); | |
7869 | put_smstate(u32, buf, 0x7fc8, (u32)val); | |
7870 | ||
7871 | kvm_get_segment(vcpu, &seg, VCPU_SREG_TR); | |
7872 | put_smstate(u32, buf, 0x7fc4, seg.selector); | |
7873 | put_smstate(u32, buf, 0x7f64, seg.base); | |
7874 | put_smstate(u32, buf, 0x7f60, seg.limit); | |
ee2cd4b7 | 7875 | put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg)); |
660a5d51 PB |
7876 | |
7877 | kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR); | |
7878 | put_smstate(u32, buf, 0x7fc0, seg.selector); | |
7879 | put_smstate(u32, buf, 0x7f80, seg.base); | |
7880 | put_smstate(u32, buf, 0x7f7c, seg.limit); | |
ee2cd4b7 | 7881 | put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg)); |
660a5d51 | 7882 | |
afaf0b2f | 7883 | kvm_x86_ops.get_gdt(vcpu, &dt); |
660a5d51 PB |
7884 | put_smstate(u32, buf, 0x7f74, dt.address); |
7885 | put_smstate(u32, buf, 0x7f70, dt.size); | |
7886 | ||
afaf0b2f | 7887 | kvm_x86_ops.get_idt(vcpu, &dt); |
660a5d51 PB |
7888 | put_smstate(u32, buf, 0x7f58, dt.address); |
7889 | put_smstate(u32, buf, 0x7f54, dt.size); | |
7890 | ||
7891 | for (i = 0; i < 6; i++) | |
ee2cd4b7 | 7892 | enter_smm_save_seg_32(vcpu, buf, i); |
660a5d51 PB |
7893 | |
7894 | put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu)); | |
7895 | ||
7896 | /* revision id */ | |
7897 | put_smstate(u32, buf, 0x7efc, 0x00020000); | |
7898 | put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase); | |
7899 | } | |
7900 | ||
b68f3cc7 | 7901 | #ifdef CONFIG_X86_64 |
ee2cd4b7 | 7902 | static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf) |
660a5d51 | 7903 | { |
660a5d51 PB |
7904 | struct desc_ptr dt; |
7905 | struct kvm_segment seg; | |
7906 | unsigned long val; | |
7907 | int i; | |
7908 | ||
7909 | for (i = 0; i < 16; i++) | |
7910 | put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i)); | |
7911 | ||
7912 | put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu)); | |
7913 | put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu)); | |
7914 | ||
7915 | kvm_get_dr(vcpu, 6, &val); | |
7916 | put_smstate(u64, buf, 0x7f68, val); | |
7917 | kvm_get_dr(vcpu, 7, &val); | |
7918 | put_smstate(u64, buf, 0x7f60, val); | |
7919 | ||
7920 | put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu)); | |
7921 | put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu)); | |
7922 | put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu)); | |
7923 | ||
7924 | put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase); | |
7925 | ||
7926 | /* revision id */ | |
7927 | put_smstate(u32, buf, 0x7efc, 0x00020064); | |
7928 | ||
7929 | put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer); | |
7930 | ||
7931 | kvm_get_segment(vcpu, &seg, VCPU_SREG_TR); | |
7932 | put_smstate(u16, buf, 0x7e90, seg.selector); | |
ee2cd4b7 | 7933 | put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8); |
660a5d51 PB |
7934 | put_smstate(u32, buf, 0x7e94, seg.limit); |
7935 | put_smstate(u64, buf, 0x7e98, seg.base); | |
7936 | ||
afaf0b2f | 7937 | kvm_x86_ops.get_idt(vcpu, &dt); |
660a5d51 PB |
7938 | put_smstate(u32, buf, 0x7e84, dt.size); |
7939 | put_smstate(u64, buf, 0x7e88, dt.address); | |
7940 | ||
7941 | kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR); | |
7942 | put_smstate(u16, buf, 0x7e70, seg.selector); | |
ee2cd4b7 | 7943 | put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8); |
660a5d51 PB |
7944 | put_smstate(u32, buf, 0x7e74, seg.limit); |
7945 | put_smstate(u64, buf, 0x7e78, seg.base); | |
7946 | ||
afaf0b2f | 7947 | kvm_x86_ops.get_gdt(vcpu, &dt); |
660a5d51 PB |
7948 | put_smstate(u32, buf, 0x7e64, dt.size); |
7949 | put_smstate(u64, buf, 0x7e68, dt.address); | |
7950 | ||
7951 | for (i = 0; i < 6; i++) | |
ee2cd4b7 | 7952 | enter_smm_save_seg_64(vcpu, buf, i); |
660a5d51 | 7953 | } |
b68f3cc7 | 7954 | #endif |
660a5d51 | 7955 | |
ee2cd4b7 | 7956 | static void enter_smm(struct kvm_vcpu *vcpu) |
64d60670 | 7957 | { |
660a5d51 | 7958 | struct kvm_segment cs, ds; |
18c3626e | 7959 | struct desc_ptr dt; |
660a5d51 PB |
7960 | char buf[512]; |
7961 | u32 cr0; | |
7962 | ||
660a5d51 | 7963 | trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true); |
660a5d51 | 7964 | memset(buf, 0, 512); |
b68f3cc7 | 7965 | #ifdef CONFIG_X86_64 |
d6321d49 | 7966 | if (guest_cpuid_has(vcpu, X86_FEATURE_LM)) |
ee2cd4b7 | 7967 | enter_smm_save_state_64(vcpu, buf); |
660a5d51 | 7968 | else |
b68f3cc7 | 7969 | #endif |
ee2cd4b7 | 7970 | enter_smm_save_state_32(vcpu, buf); |
660a5d51 | 7971 | |
0234bf88 LP |
7972 | /* |
7973 | * Give pre_enter_smm() a chance to make ISA-specific changes to the | |
7974 | * vCPU state (e.g. leave guest mode) after we've saved the state into | |
7975 | * the SMM state-save area. | |
7976 | */ | |
afaf0b2f | 7977 | kvm_x86_ops.pre_enter_smm(vcpu, buf); |
0234bf88 LP |
7978 | |
7979 | vcpu->arch.hflags |= HF_SMM_MASK; | |
54bf36aa | 7980 | kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf)); |
660a5d51 | 7981 | |
afaf0b2f | 7982 | if (kvm_x86_ops.get_nmi_mask(vcpu)) |
660a5d51 PB |
7983 | vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK; |
7984 | else | |
afaf0b2f | 7985 | kvm_x86_ops.set_nmi_mask(vcpu, true); |
660a5d51 PB |
7986 | |
7987 | kvm_set_rflags(vcpu, X86_EFLAGS_FIXED); | |
7988 | kvm_rip_write(vcpu, 0x8000); | |
7989 | ||
7990 | cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG); | |
afaf0b2f | 7991 | kvm_x86_ops.set_cr0(vcpu, cr0); |
660a5d51 PB |
7992 | vcpu->arch.cr0 = cr0; |
7993 | ||
afaf0b2f | 7994 | kvm_x86_ops.set_cr4(vcpu, 0); |
660a5d51 | 7995 | |
18c3626e PB |
7996 | /* Undocumented: IDT limit is set to zero on entry to SMM. */ |
7997 | dt.address = dt.size = 0; | |
afaf0b2f | 7998 | kvm_x86_ops.set_idt(vcpu, &dt); |
18c3626e | 7999 | |
660a5d51 PB |
8000 | __kvm_set_dr(vcpu, 7, DR7_FIXED_1); |
8001 | ||
8002 | cs.selector = (vcpu->arch.smbase >> 4) & 0xffff; | |
8003 | cs.base = vcpu->arch.smbase; | |
8004 | ||
8005 | ds.selector = 0; | |
8006 | ds.base = 0; | |
8007 | ||
8008 | cs.limit = ds.limit = 0xffffffff; | |
8009 | cs.type = ds.type = 0x3; | |
8010 | cs.dpl = ds.dpl = 0; | |
8011 | cs.db = ds.db = 0; | |
8012 | cs.s = ds.s = 1; | |
8013 | cs.l = ds.l = 0; | |
8014 | cs.g = ds.g = 1; | |
8015 | cs.avl = ds.avl = 0; | |
8016 | cs.present = ds.present = 1; | |
8017 | cs.unusable = ds.unusable = 0; | |
8018 | cs.padding = ds.padding = 0; | |
8019 | ||
8020 | kvm_set_segment(vcpu, &cs, VCPU_SREG_CS); | |
8021 | kvm_set_segment(vcpu, &ds, VCPU_SREG_DS); | |
8022 | kvm_set_segment(vcpu, &ds, VCPU_SREG_ES); | |
8023 | kvm_set_segment(vcpu, &ds, VCPU_SREG_FS); | |
8024 | kvm_set_segment(vcpu, &ds, VCPU_SREG_GS); | |
8025 | kvm_set_segment(vcpu, &ds, VCPU_SREG_SS); | |
8026 | ||
b68f3cc7 | 8027 | #ifdef CONFIG_X86_64 |
d6321d49 | 8028 | if (guest_cpuid_has(vcpu, X86_FEATURE_LM)) |
afaf0b2f | 8029 | kvm_x86_ops.set_efer(vcpu, 0); |
b68f3cc7 | 8030 | #endif |
660a5d51 PB |
8031 | |
8032 | kvm_update_cpuid(vcpu); | |
8033 | kvm_mmu_reset_context(vcpu); | |
64d60670 PB |
8034 | } |
8035 | ||
ee2cd4b7 | 8036 | static void process_smi(struct kvm_vcpu *vcpu) |
c43203ca PB |
8037 | { |
8038 | vcpu->arch.smi_pending = true; | |
8039 | kvm_make_request(KVM_REQ_EVENT, vcpu); | |
8040 | } | |
8041 | ||
7ee30bc1 NNL |
8042 | void kvm_make_scan_ioapic_request_mask(struct kvm *kvm, |
8043 | unsigned long *vcpu_bitmap) | |
8044 | { | |
8045 | cpumask_var_t cpus; | |
7ee30bc1 NNL |
8046 | |
8047 | zalloc_cpumask_var(&cpus, GFP_ATOMIC); | |
8048 | ||
db5a95ec | 8049 | kvm_make_vcpus_request_mask(kvm, KVM_REQ_SCAN_IOAPIC, |
54163a34 | 8050 | NULL, vcpu_bitmap, cpus); |
7ee30bc1 NNL |
8051 | |
8052 | free_cpumask_var(cpus); | |
8053 | } | |
8054 | ||
2860c4b1 PB |
8055 | void kvm_make_scan_ioapic_request(struct kvm *kvm) |
8056 | { | |
8057 | kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC); | |
8058 | } | |
8059 | ||
8df14af4 SS |
8060 | void kvm_vcpu_update_apicv(struct kvm_vcpu *vcpu) |
8061 | { | |
8062 | if (!lapic_in_kernel(vcpu)) | |
8063 | return; | |
8064 | ||
8065 | vcpu->arch.apicv_active = kvm_apicv_activated(vcpu->kvm); | |
8066 | kvm_apic_update_apicv(vcpu); | |
afaf0b2f | 8067 | kvm_x86_ops.refresh_apicv_exec_ctrl(vcpu); |
8df14af4 SS |
8068 | } |
8069 | EXPORT_SYMBOL_GPL(kvm_vcpu_update_apicv); | |
8070 | ||
8071 | /* | |
8072 | * NOTE: Do not hold any lock prior to calling this. | |
8073 | * | |
8074 | * In particular, kvm_request_apicv_update() expects kvm->srcu not to be | |
8075 | * locked, because it calls __x86_set_memory_region() which does | |
8076 | * synchronize_srcu(&kvm->srcu). | |
8077 | */ | |
8078 | void kvm_request_apicv_update(struct kvm *kvm, bool activate, ulong bit) | |
8079 | { | |
7d611233 | 8080 | struct kvm_vcpu *except; |
8e205a6b PB |
8081 | unsigned long old, new, expected; |
8082 | ||
afaf0b2f SC |
8083 | if (!kvm_x86_ops.check_apicv_inhibit_reasons || |
8084 | !kvm_x86_ops.check_apicv_inhibit_reasons(bit)) | |
ef8efd7a SS |
8085 | return; |
8086 | ||
8e205a6b PB |
8087 | old = READ_ONCE(kvm->arch.apicv_inhibit_reasons); |
8088 | do { | |
8089 | expected = new = old; | |
8090 | if (activate) | |
8091 | __clear_bit(bit, &new); | |
8092 | else | |
8093 | __set_bit(bit, &new); | |
8094 | if (new == old) | |
8095 | break; | |
8096 | old = cmpxchg(&kvm->arch.apicv_inhibit_reasons, expected, new); | |
8097 | } while (old != expected); | |
8098 | ||
8099 | if (!!old == !!new) | |
8100 | return; | |
8df14af4 | 8101 | |
24bbf74c | 8102 | trace_kvm_apicv_update_request(activate, bit); |
afaf0b2f SC |
8103 | if (kvm_x86_ops.pre_update_apicv_exec_ctrl) |
8104 | kvm_x86_ops.pre_update_apicv_exec_ctrl(kvm, activate); | |
7d611233 SS |
8105 | |
8106 | /* | |
8107 | * Sending request to update APICV for all other vcpus, | |
8108 | * while update the calling vcpu immediately instead of | |
8109 | * waiting for another #VMEXIT to handle the request. | |
8110 | */ | |
8111 | except = kvm_get_running_vcpu(); | |
8112 | kvm_make_all_cpus_request_except(kvm, KVM_REQ_APICV_UPDATE, | |
8113 | except); | |
8114 | if (except) | |
8115 | kvm_vcpu_update_apicv(except); | |
8df14af4 SS |
8116 | } |
8117 | EXPORT_SYMBOL_GPL(kvm_request_apicv_update); | |
8118 | ||
3d81bc7e | 8119 | static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu) |
c7c9c56c | 8120 | { |
dcbd3e49 | 8121 | if (!kvm_apic_present(vcpu)) |
3d81bc7e | 8122 | return; |
c7c9c56c | 8123 | |
6308630b | 8124 | bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256); |
c7c9c56c | 8125 | |
b053b2ae | 8126 | if (irqchip_split(vcpu->kvm)) |
6308630b | 8127 | kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors); |
db2bdcbb | 8128 | else { |
fa59cc00 | 8129 | if (vcpu->arch.apicv_active) |
afaf0b2f | 8130 | kvm_x86_ops.sync_pir_to_irr(vcpu); |
e97f852f WL |
8131 | if (ioapic_in_kernel(vcpu->kvm)) |
8132 | kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors); | |
db2bdcbb | 8133 | } |
e40ff1d6 LA |
8134 | |
8135 | if (is_guest_mode(vcpu)) | |
8136 | vcpu->arch.load_eoi_exitmap_pending = true; | |
8137 | else | |
8138 | kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu); | |
8139 | } | |
8140 | ||
8141 | static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu) | |
8142 | { | |
8143 | u64 eoi_exit_bitmap[4]; | |
8144 | ||
8145 | if (!kvm_apic_hw_enabled(vcpu->arch.apic)) | |
8146 | return; | |
8147 | ||
5c919412 AS |
8148 | bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors, |
8149 | vcpu_to_synic(vcpu)->vec_bitmap, 256); | |
afaf0b2f | 8150 | kvm_x86_ops.load_eoi_exitmap(vcpu, eoi_exit_bitmap); |
c7c9c56c YZ |
8151 | } |
8152 | ||
93065ac7 MH |
8153 | int kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm, |
8154 | unsigned long start, unsigned long end, | |
8155 | bool blockable) | |
b1394e74 RK |
8156 | { |
8157 | unsigned long apic_address; | |
8158 | ||
8159 | /* | |
8160 | * The physical address of apic access page is stored in the VMCS. | |
8161 | * Update it when it becomes invalid. | |
8162 | */ | |
8163 | apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT); | |
8164 | if (start <= apic_address && apic_address < end) | |
8165 | kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD); | |
93065ac7 MH |
8166 | |
8167 | return 0; | |
b1394e74 RK |
8168 | } |
8169 | ||
4256f43f TC |
8170 | void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu) |
8171 | { | |
c24ae0dc TC |
8172 | struct page *page = NULL; |
8173 | ||
35754c98 | 8174 | if (!lapic_in_kernel(vcpu)) |
f439ed27 PB |
8175 | return; |
8176 | ||
afaf0b2f | 8177 | if (!kvm_x86_ops.set_apic_access_page_addr) |
4256f43f TC |
8178 | return; |
8179 | ||
c24ae0dc | 8180 | page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT); |
e8fd5e9e AA |
8181 | if (is_error_page(page)) |
8182 | return; | |
afaf0b2f | 8183 | kvm_x86_ops.set_apic_access_page_addr(vcpu, page_to_phys(page)); |
c24ae0dc TC |
8184 | |
8185 | /* | |
8186 | * Do not pin apic access page in memory, the MMU notifier | |
8187 | * will call us again if it is migrated or swapped out. | |
8188 | */ | |
8189 | put_page(page); | |
4256f43f | 8190 | } |
4256f43f | 8191 | |
d264ee0c SC |
8192 | void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu) |
8193 | { | |
8194 | smp_send_reschedule(vcpu->cpu); | |
8195 | } | |
8196 | EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit); | |
8197 | ||
9357d939 | 8198 | /* |
362c698f | 8199 | * Returns 1 to let vcpu_run() continue the guest execution loop without |
9357d939 TY |
8200 | * exiting to the userspace. Otherwise, the value will be returned to the |
8201 | * userspace. | |
8202 | */ | |
851ba692 | 8203 | static int vcpu_enter_guest(struct kvm_vcpu *vcpu) |
b6c7a5dc HB |
8204 | { |
8205 | int r; | |
62a193ed MG |
8206 | bool req_int_win = |
8207 | dm_request_for_irq_injection(vcpu) && | |
8208 | kvm_cpu_accept_dm_intr(vcpu); | |
1e9e2622 | 8209 | enum exit_fastpath_completion exit_fastpath = EXIT_FASTPATH_NONE; |
62a193ed | 8210 | |
730dca42 | 8211 | bool req_immediate_exit = false; |
b6c7a5dc | 8212 | |
2fa6e1e1 | 8213 | if (kvm_request_pending(vcpu)) { |
671ddc70 | 8214 | if (kvm_check_request(KVM_REQ_GET_VMCS12_PAGES, vcpu)) { |
afaf0b2f | 8215 | if (unlikely(!kvm_x86_ops.get_vmcs12_pages(vcpu))) { |
671ddc70 JM |
8216 | r = 0; |
8217 | goto out; | |
8218 | } | |
8219 | } | |
a8eeb04a | 8220 | if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu)) |
2e53d63a | 8221 | kvm_mmu_unload(vcpu); |
a8eeb04a | 8222 | if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu)) |
2f599714 | 8223 | __kvm_migrate_timers(vcpu); |
d828199e MT |
8224 | if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu)) |
8225 | kvm_gen_update_masterclock(vcpu->kvm); | |
0061d53d MT |
8226 | if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu)) |
8227 | kvm_gen_kvmclock_update(vcpu); | |
34c238a1 ZA |
8228 | if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) { |
8229 | r = kvm_guest_time_update(vcpu); | |
8cfdc000 ZA |
8230 | if (unlikely(r)) |
8231 | goto out; | |
8232 | } | |
a8eeb04a | 8233 | if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu)) |
4731d4c7 | 8234 | kvm_mmu_sync_roots(vcpu); |
727a7e27 PB |
8235 | if (kvm_check_request(KVM_REQ_LOAD_MMU_PGD, vcpu)) |
8236 | kvm_mmu_load_pgd(vcpu); | |
a8eeb04a | 8237 | if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) |
c2ba05cc | 8238 | kvm_vcpu_flush_tlb(vcpu, true); |
a8eeb04a | 8239 | if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) { |
851ba692 | 8240 | vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS; |
b93463aa AK |
8241 | r = 0; |
8242 | goto out; | |
8243 | } | |
a8eeb04a | 8244 | if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) { |
851ba692 | 8245 | vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN; |
bbeac283 | 8246 | vcpu->mmio_needed = 0; |
71c4dfaf JR |
8247 | r = 0; |
8248 | goto out; | |
8249 | } | |
af585b92 GN |
8250 | if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) { |
8251 | /* Page is swapped out. Do synthetic halt */ | |
8252 | vcpu->arch.apf.halted = true; | |
8253 | r = 1; | |
8254 | goto out; | |
8255 | } | |
c9aaa895 GC |
8256 | if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu)) |
8257 | record_steal_time(vcpu); | |
64d60670 PB |
8258 | if (kvm_check_request(KVM_REQ_SMI, vcpu)) |
8259 | process_smi(vcpu); | |
7460fb4a AK |
8260 | if (kvm_check_request(KVM_REQ_NMI, vcpu)) |
8261 | process_nmi(vcpu); | |
f5132b01 | 8262 | if (kvm_check_request(KVM_REQ_PMU, vcpu)) |
c6702c9d | 8263 | kvm_pmu_handle_event(vcpu); |
f5132b01 | 8264 | if (kvm_check_request(KVM_REQ_PMI, vcpu)) |
c6702c9d | 8265 | kvm_pmu_deliver_pmi(vcpu); |
7543a635 SR |
8266 | if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) { |
8267 | BUG_ON(vcpu->arch.pending_ioapic_eoi > 255); | |
8268 | if (test_bit(vcpu->arch.pending_ioapic_eoi, | |
6308630b | 8269 | vcpu->arch.ioapic_handled_vectors)) { |
7543a635 SR |
8270 | vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI; |
8271 | vcpu->run->eoi.vector = | |
8272 | vcpu->arch.pending_ioapic_eoi; | |
8273 | r = 0; | |
8274 | goto out; | |
8275 | } | |
8276 | } | |
3d81bc7e YZ |
8277 | if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu)) |
8278 | vcpu_scan_ioapic(vcpu); | |
e40ff1d6 LA |
8279 | if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu)) |
8280 | vcpu_load_eoi_exitmap(vcpu); | |
4256f43f TC |
8281 | if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu)) |
8282 | kvm_vcpu_reload_apic_access_page(vcpu); | |
2ce79189 AS |
8283 | if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) { |
8284 | vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT; | |
8285 | vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH; | |
8286 | r = 0; | |
8287 | goto out; | |
8288 | } | |
e516cebb AS |
8289 | if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) { |
8290 | vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT; | |
8291 | vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET; | |
8292 | r = 0; | |
8293 | goto out; | |
8294 | } | |
db397571 AS |
8295 | if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) { |
8296 | vcpu->run->exit_reason = KVM_EXIT_HYPERV; | |
8297 | vcpu->run->hyperv = vcpu->arch.hyperv.exit; | |
8298 | r = 0; | |
8299 | goto out; | |
8300 | } | |
f3b138c5 AS |
8301 | |
8302 | /* | |
8303 | * KVM_REQ_HV_STIMER has to be processed after | |
8304 | * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers | |
8305 | * depend on the guest clock being up-to-date | |
8306 | */ | |
1f4b34f8 AS |
8307 | if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu)) |
8308 | kvm_hv_process_stimers(vcpu); | |
8df14af4 SS |
8309 | if (kvm_check_request(KVM_REQ_APICV_UPDATE, vcpu)) |
8310 | kvm_vcpu_update_apicv(vcpu); | |
2f52d58c | 8311 | } |
b93463aa | 8312 | |
b463a6f7 | 8313 | if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) { |
0f1e261e | 8314 | ++vcpu->stat.req_event; |
66450a21 JK |
8315 | kvm_apic_accept_events(vcpu); |
8316 | if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) { | |
8317 | r = 1; | |
8318 | goto out; | |
8319 | } | |
8320 | ||
a1c77abb | 8321 | if (inject_pending_event(vcpu) != 0) |
b6b8a145 | 8322 | req_immediate_exit = true; |
321c5658 | 8323 | else { |
cc3d967f | 8324 | /* Enable SMI/NMI/IRQ window open exits if needed. |
c43203ca | 8325 | * |
cc3d967f LP |
8326 | * SMIs have three cases: |
8327 | * 1) They can be nested, and then there is nothing to | |
8328 | * do here because RSM will cause a vmexit anyway. | |
8329 | * 2) There is an ISA-specific reason why SMI cannot be | |
8330 | * injected, and the moment when this changes can be | |
8331 | * intercepted. | |
8332 | * 3) Or the SMI can be pending because | |
8333 | * inject_pending_event has completed the injection | |
8334 | * of an IRQ or NMI from the previous vmexit, and | |
8335 | * then we request an immediate exit to inject the | |
8336 | * SMI. | |
c43203ca PB |
8337 | */ |
8338 | if (vcpu->arch.smi_pending && !is_smm(vcpu)) | |
afaf0b2f | 8339 | if (!kvm_x86_ops.enable_smi_window(vcpu)) |
cc3d967f | 8340 | req_immediate_exit = true; |
321c5658 | 8341 | if (vcpu->arch.nmi_pending) |
afaf0b2f | 8342 | kvm_x86_ops.enable_nmi_window(vcpu); |
321c5658 | 8343 | if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win) |
afaf0b2f | 8344 | kvm_x86_ops.enable_irq_window(vcpu); |
664f8e26 | 8345 | WARN_ON(vcpu->arch.exception.pending); |
321c5658 | 8346 | } |
b463a6f7 AK |
8347 | |
8348 | if (kvm_lapic_enabled(vcpu)) { | |
8349 | update_cr8_intercept(vcpu); | |
8350 | kvm_lapic_sync_to_vapic(vcpu); | |
8351 | } | |
8352 | } | |
8353 | ||
d8368af8 AK |
8354 | r = kvm_mmu_reload(vcpu); |
8355 | if (unlikely(r)) { | |
d905c069 | 8356 | goto cancel_injection; |
d8368af8 AK |
8357 | } |
8358 | ||
b6c7a5dc HB |
8359 | preempt_disable(); |
8360 | ||
afaf0b2f | 8361 | kvm_x86_ops.prepare_guest_switch(vcpu); |
b95234c8 PB |
8362 | |
8363 | /* | |
8364 | * Disable IRQs before setting IN_GUEST_MODE. Posted interrupt | |
8365 | * IPI are then delayed after guest entry, which ensures that they | |
8366 | * result in virtual interrupt delivery. | |
8367 | */ | |
8368 | local_irq_disable(); | |
6b7e2d09 XG |
8369 | vcpu->mode = IN_GUEST_MODE; |
8370 | ||
01b71917 MT |
8371 | srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx); |
8372 | ||
0f127d12 | 8373 | /* |
b95234c8 | 8374 | * 1) We should set ->mode before checking ->requests. Please see |
cde9af6e | 8375 | * the comment in kvm_vcpu_exiting_guest_mode(). |
b95234c8 | 8376 | * |
81b01667 | 8377 | * 2) For APICv, we should set ->mode before checking PID.ON. This |
b95234c8 PB |
8378 | * pairs with the memory barrier implicit in pi_test_and_set_on |
8379 | * (see vmx_deliver_posted_interrupt). | |
8380 | * | |
8381 | * 3) This also orders the write to mode from any reads to the page | |
8382 | * tables done while the VCPU is running. Please see the comment | |
8383 | * in kvm_flush_remote_tlbs. | |
6b7e2d09 | 8384 | */ |
01b71917 | 8385 | smp_mb__after_srcu_read_unlock(); |
b6c7a5dc | 8386 | |
b95234c8 PB |
8387 | /* |
8388 | * This handles the case where a posted interrupt was | |
8389 | * notified with kvm_vcpu_kick. | |
8390 | */ | |
fa59cc00 | 8391 | if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active) |
afaf0b2f | 8392 | kvm_x86_ops.sync_pir_to_irr(vcpu); |
32f88400 | 8393 | |
2fa6e1e1 | 8394 | if (vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu) |
d94e1dc9 | 8395 | || need_resched() || signal_pending(current)) { |
6b7e2d09 | 8396 | vcpu->mode = OUTSIDE_GUEST_MODE; |
d94e1dc9 | 8397 | smp_wmb(); |
6c142801 AK |
8398 | local_irq_enable(); |
8399 | preempt_enable(); | |
01b71917 | 8400 | vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu); |
6c142801 | 8401 | r = 1; |
d905c069 | 8402 | goto cancel_injection; |
6c142801 AK |
8403 | } |
8404 | ||
c43203ca PB |
8405 | if (req_immediate_exit) { |
8406 | kvm_make_request(KVM_REQ_EVENT, vcpu); | |
afaf0b2f | 8407 | kvm_x86_ops.request_immediate_exit(vcpu); |
c43203ca | 8408 | } |
d6185f20 | 8409 | |
8b89fe1f | 8410 | trace_kvm_entry(vcpu->vcpu_id); |
6edaa530 | 8411 | guest_enter_irqoff(); |
b6c7a5dc | 8412 | |
2620fe26 SC |
8413 | fpregs_assert_state_consistent(); |
8414 | if (test_thread_flag(TIF_NEED_FPU_LOAD)) | |
8415 | switch_fpu_return(); | |
5f409e20 | 8416 | |
42dbaa5a | 8417 | if (unlikely(vcpu->arch.switch_db_regs)) { |
42dbaa5a JK |
8418 | set_debugreg(0, 7); |
8419 | set_debugreg(vcpu->arch.eff_db[0], 0); | |
8420 | set_debugreg(vcpu->arch.eff_db[1], 1); | |
8421 | set_debugreg(vcpu->arch.eff_db[2], 2); | |
8422 | set_debugreg(vcpu->arch.eff_db[3], 3); | |
c77fb5fe | 8423 | set_debugreg(vcpu->arch.dr6, 6); |
ae561ede | 8424 | vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD; |
42dbaa5a | 8425 | } |
b6c7a5dc | 8426 | |
afaf0b2f | 8427 | kvm_x86_ops.run(vcpu); |
b6c7a5dc | 8428 | |
c77fb5fe PB |
8429 | /* |
8430 | * Do this here before restoring debug registers on the host. And | |
8431 | * since we do this before handling the vmexit, a DR access vmexit | |
8432 | * can (a) read the correct value of the debug registers, (b) set | |
8433 | * KVM_DEBUGREG_WONT_EXIT again. | |
8434 | */ | |
8435 | if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) { | |
c77fb5fe | 8436 | WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP); |
afaf0b2f | 8437 | kvm_x86_ops.sync_dirty_debug_regs(vcpu); |
70e4da7a | 8438 | kvm_update_dr0123(vcpu); |
70e4da7a PB |
8439 | kvm_update_dr7(vcpu); |
8440 | vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD; | |
c77fb5fe PB |
8441 | } |
8442 | ||
24f1e32c FW |
8443 | /* |
8444 | * If the guest has used debug registers, at least dr7 | |
8445 | * will be disabled while returning to the host. | |
8446 | * If we don't have active breakpoints in the host, we don't | |
8447 | * care about the messed up debug address registers. But if | |
8448 | * we have some of them active, restore the old state. | |
8449 | */ | |
59d8eb53 | 8450 | if (hw_breakpoint_active()) |
24f1e32c | 8451 | hw_breakpoint_restore(); |
42dbaa5a | 8452 | |
4ba76538 | 8453 | vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc()); |
1d5f066e | 8454 | |
6b7e2d09 | 8455 | vcpu->mode = OUTSIDE_GUEST_MODE; |
d94e1dc9 | 8456 | smp_wmb(); |
a547c6db | 8457 | |
afaf0b2f | 8458 | kvm_x86_ops.handle_exit_irqoff(vcpu, &exit_fastpath); |
b6c7a5dc | 8459 | |
d7a08882 SC |
8460 | /* |
8461 | * Consume any pending interrupts, including the possible source of | |
8462 | * VM-Exit on SVM and any ticks that occur between VM-Exit and now. | |
8463 | * An instruction is required after local_irq_enable() to fully unblock | |
8464 | * interrupts on processors that implement an interrupt shadow, the | |
8465 | * stat.exits increment will do nicely. | |
8466 | */ | |
8467 | kvm_before_interrupt(vcpu); | |
8468 | local_irq_enable(); | |
b6c7a5dc | 8469 | ++vcpu->stat.exits; |
d7a08882 SC |
8470 | local_irq_disable(); |
8471 | kvm_after_interrupt(vcpu); | |
b6c7a5dc | 8472 | |
f2485b3e | 8473 | guest_exit_irqoff(); |
ec0671d5 WL |
8474 | if (lapic_in_kernel(vcpu)) { |
8475 | s64 delta = vcpu->arch.apic->lapic_timer.advance_expire_delta; | |
8476 | if (delta != S64_MIN) { | |
8477 | trace_kvm_wait_lapic_expire(vcpu->vcpu_id, delta); | |
8478 | vcpu->arch.apic->lapic_timer.advance_expire_delta = S64_MIN; | |
8479 | } | |
8480 | } | |
b6c7a5dc | 8481 | |
f2485b3e | 8482 | local_irq_enable(); |
b6c7a5dc HB |
8483 | preempt_enable(); |
8484 | ||
f656ce01 | 8485 | vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu); |
3200f405 | 8486 | |
b6c7a5dc HB |
8487 | /* |
8488 | * Profile KVM exit RIPs: | |
8489 | */ | |
8490 | if (unlikely(prof_on == KVM_PROFILING)) { | |
5fdbf976 MT |
8491 | unsigned long rip = kvm_rip_read(vcpu); |
8492 | profile_hit(KVM_PROFILING, (void *)rip); | |
b6c7a5dc HB |
8493 | } |
8494 | ||
cc578287 ZA |
8495 | if (unlikely(vcpu->arch.tsc_always_catchup)) |
8496 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu); | |
298101da | 8497 | |
5cfb1d5a MT |
8498 | if (vcpu->arch.apic_attention) |
8499 | kvm_lapic_sync_from_vapic(vcpu); | |
b93463aa | 8500 | |
afaf0b2f | 8501 | r = kvm_x86_ops.handle_exit(vcpu, exit_fastpath); |
d905c069 MT |
8502 | return r; |
8503 | ||
8504 | cancel_injection: | |
afaf0b2f | 8505 | kvm_x86_ops.cancel_injection(vcpu); |
ae7a2a3f MT |
8506 | if (unlikely(vcpu->arch.apic_attention)) |
8507 | kvm_lapic_sync_from_vapic(vcpu); | |
d7690175 MT |
8508 | out: |
8509 | return r; | |
8510 | } | |
b6c7a5dc | 8511 | |
362c698f PB |
8512 | static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu) |
8513 | { | |
bf9f6ac8 | 8514 | if (!kvm_arch_vcpu_runnable(vcpu) && |
afaf0b2f | 8515 | (!kvm_x86_ops.pre_block || kvm_x86_ops.pre_block(vcpu) == 0)) { |
9c8fd1ba PB |
8516 | srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx); |
8517 | kvm_vcpu_block(vcpu); | |
8518 | vcpu->srcu_idx = srcu_read_lock(&kvm->srcu); | |
bf9f6ac8 | 8519 | |
afaf0b2f SC |
8520 | if (kvm_x86_ops.post_block) |
8521 | kvm_x86_ops.post_block(vcpu); | |
bf9f6ac8 | 8522 | |
9c8fd1ba PB |
8523 | if (!kvm_check_request(KVM_REQ_UNHALT, vcpu)) |
8524 | return 1; | |
8525 | } | |
362c698f PB |
8526 | |
8527 | kvm_apic_accept_events(vcpu); | |
8528 | switch(vcpu->arch.mp_state) { | |
8529 | case KVM_MP_STATE_HALTED: | |
8530 | vcpu->arch.pv.pv_unhalted = false; | |
8531 | vcpu->arch.mp_state = | |
8532 | KVM_MP_STATE_RUNNABLE; | |
b2869f28 | 8533 | /* fall through */ |
362c698f PB |
8534 | case KVM_MP_STATE_RUNNABLE: |
8535 | vcpu->arch.apf.halted = false; | |
8536 | break; | |
8537 | case KVM_MP_STATE_INIT_RECEIVED: | |
8538 | break; | |
8539 | default: | |
8540 | return -EINTR; | |
362c698f PB |
8541 | } |
8542 | return 1; | |
8543 | } | |
09cec754 | 8544 | |
5d9bc648 PB |
8545 | static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu) |
8546 | { | |
afaf0b2f SC |
8547 | if (is_guest_mode(vcpu) && kvm_x86_ops.check_nested_events) |
8548 | kvm_x86_ops.check_nested_events(vcpu); | |
0ad3bed6 | 8549 | |
5d9bc648 PB |
8550 | return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE && |
8551 | !vcpu->arch.apf.halted); | |
8552 | } | |
8553 | ||
362c698f | 8554 | static int vcpu_run(struct kvm_vcpu *vcpu) |
d7690175 MT |
8555 | { |
8556 | int r; | |
f656ce01 | 8557 | struct kvm *kvm = vcpu->kvm; |
d7690175 | 8558 | |
f656ce01 | 8559 | vcpu->srcu_idx = srcu_read_lock(&kvm->srcu); |
c595ceee | 8560 | vcpu->arch.l1tf_flush_l1d = true; |
d7690175 | 8561 | |
362c698f | 8562 | for (;;) { |
58f800d5 | 8563 | if (kvm_vcpu_running(vcpu)) { |
851ba692 | 8564 | r = vcpu_enter_guest(vcpu); |
bf9f6ac8 | 8565 | } else { |
362c698f | 8566 | r = vcpu_block(kvm, vcpu); |
bf9f6ac8 FW |
8567 | } |
8568 | ||
09cec754 GN |
8569 | if (r <= 0) |
8570 | break; | |
8571 | ||
72875d8a | 8572 | kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu); |
09cec754 GN |
8573 | if (kvm_cpu_has_pending_timer(vcpu)) |
8574 | kvm_inject_pending_timer_irqs(vcpu); | |
8575 | ||
782d422b MG |
8576 | if (dm_request_for_irq_injection(vcpu) && |
8577 | kvm_vcpu_ready_for_interrupt_injection(vcpu)) { | |
4ca7dd8c PB |
8578 | r = 0; |
8579 | vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN; | |
09cec754 | 8580 | ++vcpu->stat.request_irq_exits; |
362c698f | 8581 | break; |
09cec754 | 8582 | } |
af585b92 GN |
8583 | |
8584 | kvm_check_async_pf_completion(vcpu); | |
8585 | ||
09cec754 GN |
8586 | if (signal_pending(current)) { |
8587 | r = -EINTR; | |
851ba692 | 8588 | vcpu->run->exit_reason = KVM_EXIT_INTR; |
09cec754 | 8589 | ++vcpu->stat.signal_exits; |
362c698f | 8590 | break; |
09cec754 GN |
8591 | } |
8592 | if (need_resched()) { | |
f656ce01 | 8593 | srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx); |
c08ac06a | 8594 | cond_resched(); |
f656ce01 | 8595 | vcpu->srcu_idx = srcu_read_lock(&kvm->srcu); |
d7690175 | 8596 | } |
b6c7a5dc HB |
8597 | } |
8598 | ||
f656ce01 | 8599 | srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx); |
b6c7a5dc HB |
8600 | |
8601 | return r; | |
8602 | } | |
8603 | ||
716d51ab GN |
8604 | static inline int complete_emulated_io(struct kvm_vcpu *vcpu) |
8605 | { | |
8606 | int r; | |
60fc3d02 | 8607 | |
716d51ab | 8608 | vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu); |
0ce97a2b | 8609 | r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE); |
716d51ab | 8610 | srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx); |
60fc3d02 | 8611 | return r; |
716d51ab GN |
8612 | } |
8613 | ||
8614 | static int complete_emulated_pio(struct kvm_vcpu *vcpu) | |
8615 | { | |
8616 | BUG_ON(!vcpu->arch.pio.count); | |
8617 | ||
8618 | return complete_emulated_io(vcpu); | |
8619 | } | |
8620 | ||
f78146b0 AK |
8621 | /* |
8622 | * Implements the following, as a state machine: | |
8623 | * | |
8624 | * read: | |
8625 | * for each fragment | |
87da7e66 XG |
8626 | * for each mmio piece in the fragment |
8627 | * write gpa, len | |
8628 | * exit | |
8629 | * copy data | |
f78146b0 AK |
8630 | * execute insn |
8631 | * | |
8632 | * write: | |
8633 | * for each fragment | |
87da7e66 XG |
8634 | * for each mmio piece in the fragment |
8635 | * write gpa, len | |
8636 | * copy data | |
8637 | * exit | |
f78146b0 | 8638 | */ |
716d51ab | 8639 | static int complete_emulated_mmio(struct kvm_vcpu *vcpu) |
5287f194 AK |
8640 | { |
8641 | struct kvm_run *run = vcpu->run; | |
f78146b0 | 8642 | struct kvm_mmio_fragment *frag; |
87da7e66 | 8643 | unsigned len; |
5287f194 | 8644 | |
716d51ab | 8645 | BUG_ON(!vcpu->mmio_needed); |
5287f194 | 8646 | |
716d51ab | 8647 | /* Complete previous fragment */ |
87da7e66 XG |
8648 | frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment]; |
8649 | len = min(8u, frag->len); | |
716d51ab | 8650 | if (!vcpu->mmio_is_write) |
87da7e66 XG |
8651 | memcpy(frag->data, run->mmio.data, len); |
8652 | ||
8653 | if (frag->len <= 8) { | |
8654 | /* Switch to the next fragment. */ | |
8655 | frag++; | |
8656 | vcpu->mmio_cur_fragment++; | |
8657 | } else { | |
8658 | /* Go forward to the next mmio piece. */ | |
8659 | frag->data += len; | |
8660 | frag->gpa += len; | |
8661 | frag->len -= len; | |
8662 | } | |
8663 | ||
a08d3b3b | 8664 | if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) { |
716d51ab | 8665 | vcpu->mmio_needed = 0; |
0912c977 PB |
8666 | |
8667 | /* FIXME: return into emulator if single-stepping. */ | |
cef4dea0 | 8668 | if (vcpu->mmio_is_write) |
716d51ab GN |
8669 | return 1; |
8670 | vcpu->mmio_read_completed = 1; | |
8671 | return complete_emulated_io(vcpu); | |
8672 | } | |
87da7e66 | 8673 | |
716d51ab GN |
8674 | run->exit_reason = KVM_EXIT_MMIO; |
8675 | run->mmio.phys_addr = frag->gpa; | |
8676 | if (vcpu->mmio_is_write) | |
87da7e66 XG |
8677 | memcpy(run->mmio.data, frag->data, min(8u, frag->len)); |
8678 | run->mmio.len = min(8u, frag->len); | |
716d51ab GN |
8679 | run->mmio.is_write = vcpu->mmio_is_write; |
8680 | vcpu->arch.complete_userspace_io = complete_emulated_mmio; | |
8681 | return 0; | |
5287f194 AK |
8682 | } |
8683 | ||
c9aef3b8 SC |
8684 | static void kvm_save_current_fpu(struct fpu *fpu) |
8685 | { | |
8686 | /* | |
8687 | * If the target FPU state is not resident in the CPU registers, just | |
8688 | * memcpy() from current, else save CPU state directly to the target. | |
8689 | */ | |
8690 | if (test_thread_flag(TIF_NEED_FPU_LOAD)) | |
8691 | memcpy(&fpu->state, ¤t->thread.fpu.state, | |
8692 | fpu_kernel_xstate_size); | |
8693 | else | |
8694 | copy_fpregs_to_fpstate(fpu); | |
8695 | } | |
8696 | ||
822f312d SAS |
8697 | /* Swap (qemu) user FPU context for the guest FPU context. */ |
8698 | static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu) | |
8699 | { | |
5f409e20 RR |
8700 | fpregs_lock(); |
8701 | ||
c9aef3b8 SC |
8702 | kvm_save_current_fpu(vcpu->arch.user_fpu); |
8703 | ||
afaf0b2f | 8704 | /* PKRU is separately restored in kvm_x86_ops.run. */ |
b666a4b6 | 8705 | __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu->state, |
822f312d | 8706 | ~XFEATURE_MASK_PKRU); |
5f409e20 RR |
8707 | |
8708 | fpregs_mark_activate(); | |
8709 | fpregs_unlock(); | |
8710 | ||
822f312d SAS |
8711 | trace_kvm_fpu(1); |
8712 | } | |
8713 | ||
8714 | /* When vcpu_run ends, restore user space FPU context. */ | |
8715 | static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu) | |
8716 | { | |
5f409e20 RR |
8717 | fpregs_lock(); |
8718 | ||
c9aef3b8 SC |
8719 | kvm_save_current_fpu(vcpu->arch.guest_fpu); |
8720 | ||
d9a710e5 | 8721 | copy_kernel_to_fpregs(&vcpu->arch.user_fpu->state); |
5f409e20 RR |
8722 | |
8723 | fpregs_mark_activate(); | |
8724 | fpregs_unlock(); | |
8725 | ||
822f312d SAS |
8726 | ++vcpu->stat.fpu_reload; |
8727 | trace_kvm_fpu(0); | |
8728 | } | |
8729 | ||
b6c7a5dc HB |
8730 | int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run) |
8731 | { | |
8732 | int r; | |
b6c7a5dc | 8733 | |
accb757d | 8734 | vcpu_load(vcpu); |
20b7035c | 8735 | kvm_sigset_activate(vcpu); |
5663d8f9 PX |
8736 | kvm_load_guest_fpu(vcpu); |
8737 | ||
a4535290 | 8738 | if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) { |
2f173d26 JS |
8739 | if (kvm_run->immediate_exit) { |
8740 | r = -EINTR; | |
8741 | goto out; | |
8742 | } | |
b6c7a5dc | 8743 | kvm_vcpu_block(vcpu); |
66450a21 | 8744 | kvm_apic_accept_events(vcpu); |
72875d8a | 8745 | kvm_clear_request(KVM_REQ_UNHALT, vcpu); |
ac9f6dc0 | 8746 | r = -EAGAIN; |
a0595000 JS |
8747 | if (signal_pending(current)) { |
8748 | r = -EINTR; | |
8749 | vcpu->run->exit_reason = KVM_EXIT_INTR; | |
8750 | ++vcpu->stat.signal_exits; | |
8751 | } | |
ac9f6dc0 | 8752 | goto out; |
b6c7a5dc HB |
8753 | } |
8754 | ||
01643c51 KH |
8755 | if (vcpu->run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) { |
8756 | r = -EINVAL; | |
8757 | goto out; | |
8758 | } | |
8759 | ||
8760 | if (vcpu->run->kvm_dirty_regs) { | |
8761 | r = sync_regs(vcpu); | |
8762 | if (r != 0) | |
8763 | goto out; | |
8764 | } | |
8765 | ||
b6c7a5dc | 8766 | /* re-sync apic's tpr */ |
35754c98 | 8767 | if (!lapic_in_kernel(vcpu)) { |
eea1cff9 AP |
8768 | if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) { |
8769 | r = -EINVAL; | |
8770 | goto out; | |
8771 | } | |
8772 | } | |
b6c7a5dc | 8773 | |
716d51ab GN |
8774 | if (unlikely(vcpu->arch.complete_userspace_io)) { |
8775 | int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io; | |
8776 | vcpu->arch.complete_userspace_io = NULL; | |
8777 | r = cui(vcpu); | |
8778 | if (r <= 0) | |
5663d8f9 | 8779 | goto out; |
716d51ab GN |
8780 | } else |
8781 | WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed); | |
5287f194 | 8782 | |
460df4c1 PB |
8783 | if (kvm_run->immediate_exit) |
8784 | r = -EINTR; | |
8785 | else | |
8786 | r = vcpu_run(vcpu); | |
b6c7a5dc HB |
8787 | |
8788 | out: | |
5663d8f9 | 8789 | kvm_put_guest_fpu(vcpu); |
01643c51 KH |
8790 | if (vcpu->run->kvm_valid_regs) |
8791 | store_regs(vcpu); | |
f1d86e46 | 8792 | post_kvm_run_save(vcpu); |
20b7035c | 8793 | kvm_sigset_deactivate(vcpu); |
b6c7a5dc | 8794 | |
accb757d | 8795 | vcpu_put(vcpu); |
b6c7a5dc HB |
8796 | return r; |
8797 | } | |
8798 | ||
01643c51 | 8799 | static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs) |
b6c7a5dc | 8800 | { |
7ae441ea GN |
8801 | if (vcpu->arch.emulate_regs_need_sync_to_vcpu) { |
8802 | /* | |
8803 | * We are here if userspace calls get_regs() in the middle of | |
8804 | * instruction emulation. Registers state needs to be copied | |
4a969980 | 8805 | * back from emulation context to vcpu. Userspace shouldn't do |
7ae441ea GN |
8806 | * that usually, but some bad designed PV devices (vmware |
8807 | * backdoor interface) need this to work | |
8808 | */ | |
c9b8b07c | 8809 | emulator_writeback_register_cache(vcpu->arch.emulate_ctxt); |
7ae441ea GN |
8810 | vcpu->arch.emulate_regs_need_sync_to_vcpu = false; |
8811 | } | |
de3cd117 SC |
8812 | regs->rax = kvm_rax_read(vcpu); |
8813 | regs->rbx = kvm_rbx_read(vcpu); | |
8814 | regs->rcx = kvm_rcx_read(vcpu); | |
8815 | regs->rdx = kvm_rdx_read(vcpu); | |
8816 | regs->rsi = kvm_rsi_read(vcpu); | |
8817 | regs->rdi = kvm_rdi_read(vcpu); | |
e9c16c78 | 8818 | regs->rsp = kvm_rsp_read(vcpu); |
de3cd117 | 8819 | regs->rbp = kvm_rbp_read(vcpu); |
b6c7a5dc | 8820 | #ifdef CONFIG_X86_64 |
de3cd117 SC |
8821 | regs->r8 = kvm_r8_read(vcpu); |
8822 | regs->r9 = kvm_r9_read(vcpu); | |
8823 | regs->r10 = kvm_r10_read(vcpu); | |
8824 | regs->r11 = kvm_r11_read(vcpu); | |
8825 | regs->r12 = kvm_r12_read(vcpu); | |
8826 | regs->r13 = kvm_r13_read(vcpu); | |
8827 | regs->r14 = kvm_r14_read(vcpu); | |
8828 | regs->r15 = kvm_r15_read(vcpu); | |
b6c7a5dc HB |
8829 | #endif |
8830 | ||
5fdbf976 | 8831 | regs->rip = kvm_rip_read(vcpu); |
91586a3b | 8832 | regs->rflags = kvm_get_rflags(vcpu); |
01643c51 | 8833 | } |
b6c7a5dc | 8834 | |
01643c51 KH |
8835 | int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs) |
8836 | { | |
8837 | vcpu_load(vcpu); | |
8838 | __get_regs(vcpu, regs); | |
1fc9b76b | 8839 | vcpu_put(vcpu); |
b6c7a5dc HB |
8840 | return 0; |
8841 | } | |
8842 | ||
01643c51 | 8843 | static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs) |
b6c7a5dc | 8844 | { |
7ae441ea GN |
8845 | vcpu->arch.emulate_regs_need_sync_from_vcpu = true; |
8846 | vcpu->arch.emulate_regs_need_sync_to_vcpu = false; | |
8847 | ||
de3cd117 SC |
8848 | kvm_rax_write(vcpu, regs->rax); |
8849 | kvm_rbx_write(vcpu, regs->rbx); | |
8850 | kvm_rcx_write(vcpu, regs->rcx); | |
8851 | kvm_rdx_write(vcpu, regs->rdx); | |
8852 | kvm_rsi_write(vcpu, regs->rsi); | |
8853 | kvm_rdi_write(vcpu, regs->rdi); | |
e9c16c78 | 8854 | kvm_rsp_write(vcpu, regs->rsp); |
de3cd117 | 8855 | kvm_rbp_write(vcpu, regs->rbp); |
b6c7a5dc | 8856 | #ifdef CONFIG_X86_64 |
de3cd117 SC |
8857 | kvm_r8_write(vcpu, regs->r8); |
8858 | kvm_r9_write(vcpu, regs->r9); | |
8859 | kvm_r10_write(vcpu, regs->r10); | |
8860 | kvm_r11_write(vcpu, regs->r11); | |
8861 | kvm_r12_write(vcpu, regs->r12); | |
8862 | kvm_r13_write(vcpu, regs->r13); | |
8863 | kvm_r14_write(vcpu, regs->r14); | |
8864 | kvm_r15_write(vcpu, regs->r15); | |
b6c7a5dc HB |
8865 | #endif |
8866 | ||
5fdbf976 | 8867 | kvm_rip_write(vcpu, regs->rip); |
d73235d1 | 8868 | kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED); |
b6c7a5dc | 8869 | |
b4f14abd JK |
8870 | vcpu->arch.exception.pending = false; |
8871 | ||
3842d135 | 8872 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
01643c51 | 8873 | } |
3842d135 | 8874 | |
01643c51 KH |
8875 | int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs) |
8876 | { | |
8877 | vcpu_load(vcpu); | |
8878 | __set_regs(vcpu, regs); | |
875656fe | 8879 | vcpu_put(vcpu); |
b6c7a5dc HB |
8880 | return 0; |
8881 | } | |
8882 | ||
b6c7a5dc HB |
8883 | void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l) |
8884 | { | |
8885 | struct kvm_segment cs; | |
8886 | ||
3e6e0aab | 8887 | kvm_get_segment(vcpu, &cs, VCPU_SREG_CS); |
b6c7a5dc HB |
8888 | *db = cs.db; |
8889 | *l = cs.l; | |
8890 | } | |
8891 | EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits); | |
8892 | ||
01643c51 | 8893 | static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs) |
b6c7a5dc | 8894 | { |
89a27f4d | 8895 | struct desc_ptr dt; |
b6c7a5dc | 8896 | |
3e6e0aab GT |
8897 | kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS); |
8898 | kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS); | |
8899 | kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES); | |
8900 | kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS); | |
8901 | kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS); | |
8902 | kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS); | |
b6c7a5dc | 8903 | |
3e6e0aab GT |
8904 | kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR); |
8905 | kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR); | |
b6c7a5dc | 8906 | |
afaf0b2f | 8907 | kvm_x86_ops.get_idt(vcpu, &dt); |
89a27f4d GN |
8908 | sregs->idt.limit = dt.size; |
8909 | sregs->idt.base = dt.address; | |
afaf0b2f | 8910 | kvm_x86_ops.get_gdt(vcpu, &dt); |
89a27f4d GN |
8911 | sregs->gdt.limit = dt.size; |
8912 | sregs->gdt.base = dt.address; | |
b6c7a5dc | 8913 | |
4d4ec087 | 8914 | sregs->cr0 = kvm_read_cr0(vcpu); |
ad312c7c | 8915 | sregs->cr2 = vcpu->arch.cr2; |
9f8fe504 | 8916 | sregs->cr3 = kvm_read_cr3(vcpu); |
fc78f519 | 8917 | sregs->cr4 = kvm_read_cr4(vcpu); |
2d3ad1f4 | 8918 | sregs->cr8 = kvm_get_cr8(vcpu); |
f6801dff | 8919 | sregs->efer = vcpu->arch.efer; |
b6c7a5dc HB |
8920 | sregs->apic_base = kvm_get_apic_base(vcpu); |
8921 | ||
0e96f31e | 8922 | memset(sregs->interrupt_bitmap, 0, sizeof(sregs->interrupt_bitmap)); |
b6c7a5dc | 8923 | |
04140b41 | 8924 | if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft) |
14d0bc1f GN |
8925 | set_bit(vcpu->arch.interrupt.nr, |
8926 | (unsigned long *)sregs->interrupt_bitmap); | |
01643c51 | 8927 | } |
16d7a191 | 8928 | |
01643c51 KH |
8929 | int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu, |
8930 | struct kvm_sregs *sregs) | |
8931 | { | |
8932 | vcpu_load(vcpu); | |
8933 | __get_sregs(vcpu, sregs); | |
bcdec41c | 8934 | vcpu_put(vcpu); |
b6c7a5dc HB |
8935 | return 0; |
8936 | } | |
8937 | ||
62d9f0db MT |
8938 | int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu, |
8939 | struct kvm_mp_state *mp_state) | |
8940 | { | |
fd232561 | 8941 | vcpu_load(vcpu); |
f958bd23 SC |
8942 | if (kvm_mpx_supported()) |
8943 | kvm_load_guest_fpu(vcpu); | |
fd232561 | 8944 | |
66450a21 | 8945 | kvm_apic_accept_events(vcpu); |
6aef266c SV |
8946 | if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED && |
8947 | vcpu->arch.pv.pv_unhalted) | |
8948 | mp_state->mp_state = KVM_MP_STATE_RUNNABLE; | |
8949 | else | |
8950 | mp_state->mp_state = vcpu->arch.mp_state; | |
8951 | ||
f958bd23 SC |
8952 | if (kvm_mpx_supported()) |
8953 | kvm_put_guest_fpu(vcpu); | |
fd232561 | 8954 | vcpu_put(vcpu); |
62d9f0db MT |
8955 | return 0; |
8956 | } | |
8957 | ||
8958 | int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu, | |
8959 | struct kvm_mp_state *mp_state) | |
8960 | { | |
e83dff5e CD |
8961 | int ret = -EINVAL; |
8962 | ||
8963 | vcpu_load(vcpu); | |
8964 | ||
bce87cce | 8965 | if (!lapic_in_kernel(vcpu) && |
66450a21 | 8966 | mp_state->mp_state != KVM_MP_STATE_RUNNABLE) |
e83dff5e | 8967 | goto out; |
66450a21 | 8968 | |
27cbe7d6 LA |
8969 | /* |
8970 | * KVM_MP_STATE_INIT_RECEIVED means the processor is in | |
8971 | * INIT state; latched init should be reported using | |
8972 | * KVM_SET_VCPU_EVENTS, so reject it here. | |
8973 | */ | |
8974 | if ((kvm_vcpu_latch_init(vcpu) || vcpu->arch.smi_pending) && | |
28bf2888 DH |
8975 | (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED || |
8976 | mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED)) | |
e83dff5e | 8977 | goto out; |
28bf2888 | 8978 | |
66450a21 JK |
8979 | if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) { |
8980 | vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED; | |
8981 | set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events); | |
8982 | } else | |
8983 | vcpu->arch.mp_state = mp_state->mp_state; | |
3842d135 | 8984 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
e83dff5e CD |
8985 | |
8986 | ret = 0; | |
8987 | out: | |
8988 | vcpu_put(vcpu); | |
8989 | return ret; | |
62d9f0db MT |
8990 | } |
8991 | ||
7f3d35fd KW |
8992 | int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index, |
8993 | int reason, bool has_error_code, u32 error_code) | |
b6c7a5dc | 8994 | { |
c9b8b07c | 8995 | struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt; |
8ec4722d | 8996 | int ret; |
e01c2426 | 8997 | |
8ec4722d | 8998 | init_emulate_ctxt(vcpu); |
c697518a | 8999 | |
7f3d35fd | 9000 | ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason, |
9d74191a | 9001 | has_error_code, error_code); |
1051778f SC |
9002 | if (ret) { |
9003 | vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR; | |
9004 | vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION; | |
9005 | vcpu->run->internal.ndata = 0; | |
60fc3d02 | 9006 | return 0; |
1051778f | 9007 | } |
37817f29 | 9008 | |
9d74191a TY |
9009 | kvm_rip_write(vcpu, ctxt->eip); |
9010 | kvm_set_rflags(vcpu, ctxt->eflags); | |
60fc3d02 | 9011 | return 1; |
37817f29 IE |
9012 | } |
9013 | EXPORT_SYMBOL_GPL(kvm_task_switch); | |
9014 | ||
3140c156 | 9015 | static int kvm_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs) |
f2981033 | 9016 | { |
37b95951 | 9017 | if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) { |
f2981033 LT |
9018 | /* |
9019 | * When EFER.LME and CR0.PG are set, the processor is in | |
9020 | * 64-bit mode (though maybe in a 32-bit code segment). | |
9021 | * CR4.PAE and EFER.LMA must be set. | |
9022 | */ | |
37b95951 | 9023 | if (!(sregs->cr4 & X86_CR4_PAE) |
f2981033 LT |
9024 | || !(sregs->efer & EFER_LMA)) |
9025 | return -EINVAL; | |
9026 | } else { | |
9027 | /* | |
9028 | * Not in 64-bit mode: EFER.LMA is clear and the code | |
9029 | * segment cannot be 64-bit. | |
9030 | */ | |
9031 | if (sregs->efer & EFER_LMA || sregs->cs.l) | |
9032 | return -EINVAL; | |
9033 | } | |
9034 | ||
3ca94192 | 9035 | return kvm_valid_cr4(vcpu, sregs->cr4); |
f2981033 LT |
9036 | } |
9037 | ||
01643c51 | 9038 | static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs) |
b6c7a5dc | 9039 | { |
58cb628d | 9040 | struct msr_data apic_base_msr; |
b6c7a5dc | 9041 | int mmu_reset_needed = 0; |
c4d21882 | 9042 | int cpuid_update_needed = 0; |
63f42e02 | 9043 | int pending_vec, max_bits, idx; |
89a27f4d | 9044 | struct desc_ptr dt; |
b4ef9d4e CD |
9045 | int ret = -EINVAL; |
9046 | ||
f2981033 | 9047 | if (kvm_valid_sregs(vcpu, sregs)) |
8dbfb2bf | 9048 | goto out; |
f2981033 | 9049 | |
d3802286 JM |
9050 | apic_base_msr.data = sregs->apic_base; |
9051 | apic_base_msr.host_initiated = true; | |
9052 | if (kvm_set_apic_base(vcpu, &apic_base_msr)) | |
b4ef9d4e | 9053 | goto out; |
6d1068b3 | 9054 | |
89a27f4d GN |
9055 | dt.size = sregs->idt.limit; |
9056 | dt.address = sregs->idt.base; | |
afaf0b2f | 9057 | kvm_x86_ops.set_idt(vcpu, &dt); |
89a27f4d GN |
9058 | dt.size = sregs->gdt.limit; |
9059 | dt.address = sregs->gdt.base; | |
afaf0b2f | 9060 | kvm_x86_ops.set_gdt(vcpu, &dt); |
b6c7a5dc | 9061 | |
ad312c7c | 9062 | vcpu->arch.cr2 = sregs->cr2; |
9f8fe504 | 9063 | mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3; |
dc7e795e | 9064 | vcpu->arch.cr3 = sregs->cr3; |
cb3c1e2f | 9065 | kvm_register_mark_available(vcpu, VCPU_EXREG_CR3); |
b6c7a5dc | 9066 | |
2d3ad1f4 | 9067 | kvm_set_cr8(vcpu, sregs->cr8); |
b6c7a5dc | 9068 | |
f6801dff | 9069 | mmu_reset_needed |= vcpu->arch.efer != sregs->efer; |
afaf0b2f | 9070 | kvm_x86_ops.set_efer(vcpu, sregs->efer); |
b6c7a5dc | 9071 | |
4d4ec087 | 9072 | mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0; |
afaf0b2f | 9073 | kvm_x86_ops.set_cr0(vcpu, sregs->cr0); |
d7306163 | 9074 | vcpu->arch.cr0 = sregs->cr0; |
b6c7a5dc | 9075 | |
fc78f519 | 9076 | mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4; |
c4d21882 WH |
9077 | cpuid_update_needed |= ((kvm_read_cr4(vcpu) ^ sregs->cr4) & |
9078 | (X86_CR4_OSXSAVE | X86_CR4_PKE)); | |
afaf0b2f | 9079 | kvm_x86_ops.set_cr4(vcpu, sregs->cr4); |
c4d21882 | 9080 | if (cpuid_update_needed) |
00b27a3e | 9081 | kvm_update_cpuid(vcpu); |
63f42e02 XG |
9082 | |
9083 | idx = srcu_read_lock(&vcpu->kvm->srcu); | |
bf03d4f9 | 9084 | if (is_pae_paging(vcpu)) { |
9f8fe504 | 9085 | load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu)); |
7c93be44 MT |
9086 | mmu_reset_needed = 1; |
9087 | } | |
63f42e02 | 9088 | srcu_read_unlock(&vcpu->kvm->srcu, idx); |
b6c7a5dc HB |
9089 | |
9090 | if (mmu_reset_needed) | |
9091 | kvm_mmu_reset_context(vcpu); | |
9092 | ||
a50abc3b | 9093 | max_bits = KVM_NR_INTERRUPTS; |
923c61bb GN |
9094 | pending_vec = find_first_bit( |
9095 | (const unsigned long *)sregs->interrupt_bitmap, max_bits); | |
9096 | if (pending_vec < max_bits) { | |
66fd3f7f | 9097 | kvm_queue_interrupt(vcpu, pending_vec, false); |
923c61bb | 9098 | pr_debug("Set back pending irq %d\n", pending_vec); |
b6c7a5dc HB |
9099 | } |
9100 | ||
3e6e0aab GT |
9101 | kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS); |
9102 | kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS); | |
9103 | kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES); | |
9104 | kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS); | |
9105 | kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS); | |
9106 | kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS); | |
b6c7a5dc | 9107 | |
3e6e0aab GT |
9108 | kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR); |
9109 | kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR); | |
b6c7a5dc | 9110 | |
5f0269f5 ME |
9111 | update_cr8_intercept(vcpu); |
9112 | ||
9c3e4aab | 9113 | /* Older userspace won't unhalt the vcpu on reset. */ |
c5af89b6 | 9114 | if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 && |
9c3e4aab | 9115 | sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 && |
3eeb3288 | 9116 | !is_protmode(vcpu)) |
9c3e4aab MT |
9117 | vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE; |
9118 | ||
3842d135 AK |
9119 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
9120 | ||
b4ef9d4e CD |
9121 | ret = 0; |
9122 | out: | |
01643c51 KH |
9123 | return ret; |
9124 | } | |
9125 | ||
9126 | int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu, | |
9127 | struct kvm_sregs *sregs) | |
9128 | { | |
9129 | int ret; | |
9130 | ||
9131 | vcpu_load(vcpu); | |
9132 | ret = __set_sregs(vcpu, sregs); | |
b4ef9d4e CD |
9133 | vcpu_put(vcpu); |
9134 | return ret; | |
b6c7a5dc HB |
9135 | } |
9136 | ||
d0bfb940 JK |
9137 | int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu, |
9138 | struct kvm_guest_debug *dbg) | |
b6c7a5dc | 9139 | { |
355be0b9 | 9140 | unsigned long rflags; |
ae675ef0 | 9141 | int i, r; |
b6c7a5dc | 9142 | |
66b56562 CD |
9143 | vcpu_load(vcpu); |
9144 | ||
4f926bf2 JK |
9145 | if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) { |
9146 | r = -EBUSY; | |
9147 | if (vcpu->arch.exception.pending) | |
2122ff5e | 9148 | goto out; |
4f926bf2 JK |
9149 | if (dbg->control & KVM_GUESTDBG_INJECT_DB) |
9150 | kvm_queue_exception(vcpu, DB_VECTOR); | |
9151 | else | |
9152 | kvm_queue_exception(vcpu, BP_VECTOR); | |
9153 | } | |
9154 | ||
91586a3b JK |
9155 | /* |
9156 | * Read rflags as long as potentially injected trace flags are still | |
9157 | * filtered out. | |
9158 | */ | |
9159 | rflags = kvm_get_rflags(vcpu); | |
355be0b9 JK |
9160 | |
9161 | vcpu->guest_debug = dbg->control; | |
9162 | if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE)) | |
9163 | vcpu->guest_debug = 0; | |
9164 | ||
9165 | if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) { | |
ae675ef0 JK |
9166 | for (i = 0; i < KVM_NR_DB_REGS; ++i) |
9167 | vcpu->arch.eff_db[i] = dbg->arch.debugreg[i]; | |
c8639010 | 9168 | vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7]; |
ae675ef0 JK |
9169 | } else { |
9170 | for (i = 0; i < KVM_NR_DB_REGS; i++) | |
9171 | vcpu->arch.eff_db[i] = vcpu->arch.db[i]; | |
ae675ef0 | 9172 | } |
c8639010 | 9173 | kvm_update_dr7(vcpu); |
ae675ef0 | 9174 | |
f92653ee JK |
9175 | if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) |
9176 | vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) + | |
9177 | get_segment_base(vcpu, VCPU_SREG_CS); | |
94fe45da | 9178 | |
91586a3b JK |
9179 | /* |
9180 | * Trigger an rflags update that will inject or remove the trace | |
9181 | * flags. | |
9182 | */ | |
9183 | kvm_set_rflags(vcpu, rflags); | |
b6c7a5dc | 9184 | |
afaf0b2f | 9185 | kvm_x86_ops.update_bp_intercept(vcpu); |
b6c7a5dc | 9186 | |
4f926bf2 | 9187 | r = 0; |
d0bfb940 | 9188 | |
2122ff5e | 9189 | out: |
66b56562 | 9190 | vcpu_put(vcpu); |
b6c7a5dc HB |
9191 | return r; |
9192 | } | |
9193 | ||
8b006791 ZX |
9194 | /* |
9195 | * Translate a guest virtual address to a guest physical address. | |
9196 | */ | |
9197 | int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu, | |
9198 | struct kvm_translation *tr) | |
9199 | { | |
9200 | unsigned long vaddr = tr->linear_address; | |
9201 | gpa_t gpa; | |
f656ce01 | 9202 | int idx; |
8b006791 | 9203 | |
1da5b61d CD |
9204 | vcpu_load(vcpu); |
9205 | ||
f656ce01 | 9206 | idx = srcu_read_lock(&vcpu->kvm->srcu); |
1871c602 | 9207 | gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL); |
f656ce01 | 9208 | srcu_read_unlock(&vcpu->kvm->srcu, idx); |
8b006791 ZX |
9209 | tr->physical_address = gpa; |
9210 | tr->valid = gpa != UNMAPPED_GVA; | |
9211 | tr->writeable = 1; | |
9212 | tr->usermode = 0; | |
8b006791 | 9213 | |
1da5b61d | 9214 | vcpu_put(vcpu); |
8b006791 ZX |
9215 | return 0; |
9216 | } | |
9217 | ||
d0752060 HB |
9218 | int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu) |
9219 | { | |
1393123e | 9220 | struct fxregs_state *fxsave; |
d0752060 | 9221 | |
1393123e | 9222 | vcpu_load(vcpu); |
d0752060 | 9223 | |
b666a4b6 | 9224 | fxsave = &vcpu->arch.guest_fpu->state.fxsave; |
d0752060 HB |
9225 | memcpy(fpu->fpr, fxsave->st_space, 128); |
9226 | fpu->fcw = fxsave->cwd; | |
9227 | fpu->fsw = fxsave->swd; | |
9228 | fpu->ftwx = fxsave->twd; | |
9229 | fpu->last_opcode = fxsave->fop; | |
9230 | fpu->last_ip = fxsave->rip; | |
9231 | fpu->last_dp = fxsave->rdp; | |
0e96f31e | 9232 | memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space)); |
d0752060 | 9233 | |
1393123e | 9234 | vcpu_put(vcpu); |
d0752060 HB |
9235 | return 0; |
9236 | } | |
9237 | ||
9238 | int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu) | |
9239 | { | |
6a96bc7f CD |
9240 | struct fxregs_state *fxsave; |
9241 | ||
9242 | vcpu_load(vcpu); | |
9243 | ||
b666a4b6 | 9244 | fxsave = &vcpu->arch.guest_fpu->state.fxsave; |
d0752060 | 9245 | |
d0752060 HB |
9246 | memcpy(fxsave->st_space, fpu->fpr, 128); |
9247 | fxsave->cwd = fpu->fcw; | |
9248 | fxsave->swd = fpu->fsw; | |
9249 | fxsave->twd = fpu->ftwx; | |
9250 | fxsave->fop = fpu->last_opcode; | |
9251 | fxsave->rip = fpu->last_ip; | |
9252 | fxsave->rdp = fpu->last_dp; | |
0e96f31e | 9253 | memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space)); |
d0752060 | 9254 | |
6a96bc7f | 9255 | vcpu_put(vcpu); |
d0752060 HB |
9256 | return 0; |
9257 | } | |
9258 | ||
01643c51 KH |
9259 | static void store_regs(struct kvm_vcpu *vcpu) |
9260 | { | |
9261 | BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES); | |
9262 | ||
9263 | if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS) | |
9264 | __get_regs(vcpu, &vcpu->run->s.regs.regs); | |
9265 | ||
9266 | if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS) | |
9267 | __get_sregs(vcpu, &vcpu->run->s.regs.sregs); | |
9268 | ||
9269 | if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS) | |
9270 | kvm_vcpu_ioctl_x86_get_vcpu_events( | |
9271 | vcpu, &vcpu->run->s.regs.events); | |
9272 | } | |
9273 | ||
9274 | static int sync_regs(struct kvm_vcpu *vcpu) | |
9275 | { | |
9276 | if (vcpu->run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS) | |
9277 | return -EINVAL; | |
9278 | ||
9279 | if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) { | |
9280 | __set_regs(vcpu, &vcpu->run->s.regs.regs); | |
9281 | vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS; | |
9282 | } | |
9283 | if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) { | |
9284 | if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs)) | |
9285 | return -EINVAL; | |
9286 | vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS; | |
9287 | } | |
9288 | if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) { | |
9289 | if (kvm_vcpu_ioctl_x86_set_vcpu_events( | |
9290 | vcpu, &vcpu->run->s.regs.events)) | |
9291 | return -EINVAL; | |
9292 | vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS; | |
9293 | } | |
9294 | ||
9295 | return 0; | |
9296 | } | |
9297 | ||
0ee6a517 | 9298 | static void fx_init(struct kvm_vcpu *vcpu) |
d0752060 | 9299 | { |
b666a4b6 | 9300 | fpstate_init(&vcpu->arch.guest_fpu->state); |
782511b0 | 9301 | if (boot_cpu_has(X86_FEATURE_XSAVES)) |
b666a4b6 | 9302 | vcpu->arch.guest_fpu->state.xsave.header.xcomp_bv = |
df1daba7 | 9303 | host_xcr0 | XSTATE_COMPACTION_ENABLED; |
d0752060 | 9304 | |
2acf923e DC |
9305 | /* |
9306 | * Ensure guest xcr0 is valid for loading | |
9307 | */ | |
d91cab78 | 9308 | vcpu->arch.xcr0 = XFEATURE_MASK_FP; |
2acf923e | 9309 | |
ad312c7c | 9310 | vcpu->arch.cr0 |= X86_CR0_ET; |
d0752060 | 9311 | } |
d0752060 | 9312 | |
897cc38e | 9313 | int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id) |
e9b11c17 | 9314 | { |
897cc38e SC |
9315 | if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0) |
9316 | pr_warn_once("kvm: SMP vm created on host with unstable TSC; " | |
9317 | "guest TSC will not be reliable\n"); | |
7f1ea208 | 9318 | |
897cc38e | 9319 | return 0; |
e9b11c17 ZX |
9320 | } |
9321 | ||
e529ef66 | 9322 | int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu) |
e9b11c17 | 9323 | { |
95a0d01e SC |
9324 | struct page *page; |
9325 | int r; | |
c447e76b | 9326 | |
95a0d01e SC |
9327 | if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu)) |
9328 | vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE; | |
9329 | else | |
9330 | vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED; | |
c447e76b | 9331 | |
95a0d01e | 9332 | kvm_set_tsc_khz(vcpu, max_tsc_khz); |
c447e76b | 9333 | |
95a0d01e SC |
9334 | r = kvm_mmu_create(vcpu); |
9335 | if (r < 0) | |
9336 | return r; | |
9337 | ||
9338 | if (irqchip_in_kernel(vcpu->kvm)) { | |
95a0d01e SC |
9339 | r = kvm_create_lapic(vcpu, lapic_timer_advance_ns); |
9340 | if (r < 0) | |
9341 | goto fail_mmu_destroy; | |
4e19c36f SS |
9342 | if (kvm_apicv_activated(vcpu->kvm)) |
9343 | vcpu->arch.apicv_active = true; | |
95a0d01e SC |
9344 | } else |
9345 | static_key_slow_inc(&kvm_no_apic_vcpu); | |
9346 | ||
9347 | r = -ENOMEM; | |
9348 | ||
9349 | page = alloc_page(GFP_KERNEL | __GFP_ZERO); | |
9350 | if (!page) | |
9351 | goto fail_free_lapic; | |
9352 | vcpu->arch.pio_data = page_address(page); | |
9353 | ||
9354 | vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4, | |
9355 | GFP_KERNEL_ACCOUNT); | |
9356 | if (!vcpu->arch.mce_banks) | |
9357 | goto fail_free_pio_data; | |
9358 | vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS; | |
9359 | ||
9360 | if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, | |
9361 | GFP_KERNEL_ACCOUNT)) | |
9362 | goto fail_free_mce_banks; | |
9363 | ||
c9b8b07c SC |
9364 | if (!alloc_emulate_ctxt(vcpu)) |
9365 | goto free_wbinvd_dirty_mask; | |
9366 | ||
95a0d01e SC |
9367 | vcpu->arch.user_fpu = kmem_cache_zalloc(x86_fpu_cache, |
9368 | GFP_KERNEL_ACCOUNT); | |
9369 | if (!vcpu->arch.user_fpu) { | |
9370 | pr_err("kvm: failed to allocate userspace's fpu\n"); | |
c9b8b07c | 9371 | goto free_emulate_ctxt; |
95a0d01e SC |
9372 | } |
9373 | ||
9374 | vcpu->arch.guest_fpu = kmem_cache_zalloc(x86_fpu_cache, | |
9375 | GFP_KERNEL_ACCOUNT); | |
9376 | if (!vcpu->arch.guest_fpu) { | |
9377 | pr_err("kvm: failed to allocate vcpu's fpu\n"); | |
9378 | goto free_user_fpu; | |
9379 | } | |
9380 | fx_init(vcpu); | |
9381 | ||
9382 | vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET; | |
9383 | ||
9384 | vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu); | |
9385 | ||
9386 | vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT; | |
9387 | ||
9388 | kvm_async_pf_hash_reset(vcpu); | |
9389 | kvm_pmu_init(vcpu); | |
9390 | ||
9391 | vcpu->arch.pending_external_vector = -1; | |
9392 | vcpu->arch.preempted_in_kernel = false; | |
9393 | ||
9394 | kvm_hv_vcpu_init(vcpu); | |
9395 | ||
afaf0b2f | 9396 | r = kvm_x86_ops.vcpu_create(vcpu); |
95a0d01e SC |
9397 | if (r) |
9398 | goto free_guest_fpu; | |
e9b11c17 | 9399 | |
0cf9135b | 9400 | vcpu->arch.arch_capabilities = kvm_get_arch_capabilities(); |
e53d88af | 9401 | vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT; |
19efffa2 | 9402 | kvm_vcpu_mtrr_init(vcpu); |
ec7660cc | 9403 | vcpu_load(vcpu); |
d28bc9dd | 9404 | kvm_vcpu_reset(vcpu, false); |
e1732991 | 9405 | kvm_init_mmu(vcpu, false); |
e9b11c17 | 9406 | vcpu_put(vcpu); |
ec7660cc | 9407 | return 0; |
95a0d01e SC |
9408 | |
9409 | free_guest_fpu: | |
9410 | kmem_cache_free(x86_fpu_cache, vcpu->arch.guest_fpu); | |
9411 | free_user_fpu: | |
9412 | kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu); | |
c9b8b07c SC |
9413 | free_emulate_ctxt: |
9414 | kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt); | |
95a0d01e SC |
9415 | free_wbinvd_dirty_mask: |
9416 | free_cpumask_var(vcpu->arch.wbinvd_dirty_mask); | |
9417 | fail_free_mce_banks: | |
9418 | kfree(vcpu->arch.mce_banks); | |
9419 | fail_free_pio_data: | |
9420 | free_page((unsigned long)vcpu->arch.pio_data); | |
9421 | fail_free_lapic: | |
9422 | kvm_free_lapic(vcpu); | |
9423 | fail_mmu_destroy: | |
9424 | kvm_mmu_destroy(vcpu); | |
9425 | return r; | |
e9b11c17 ZX |
9426 | } |
9427 | ||
31928aa5 | 9428 | void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu) |
42897d86 | 9429 | { |
8fe8ab46 | 9430 | struct msr_data msr; |
332967a3 | 9431 | struct kvm *kvm = vcpu->kvm; |
42897d86 | 9432 | |
d3457c87 RK |
9433 | kvm_hv_vcpu_postcreate(vcpu); |
9434 | ||
ec7660cc | 9435 | if (mutex_lock_killable(&vcpu->mutex)) |
31928aa5 | 9436 | return; |
ec7660cc | 9437 | vcpu_load(vcpu); |
8fe8ab46 WA |
9438 | msr.data = 0x0; |
9439 | msr.index = MSR_IA32_TSC; | |
9440 | msr.host_initiated = true; | |
9441 | kvm_write_tsc(vcpu, &msr); | |
42897d86 | 9442 | vcpu_put(vcpu); |
2d5ba19b MT |
9443 | |
9444 | /* poll control enabled by default */ | |
9445 | vcpu->arch.msr_kvm_poll_control = 1; | |
9446 | ||
ec7660cc | 9447 | mutex_unlock(&vcpu->mutex); |
42897d86 | 9448 | |
b34de572 WL |
9449 | if (kvmclock_periodic_sync && vcpu->vcpu_idx == 0) |
9450 | schedule_delayed_work(&kvm->arch.kvmclock_sync_work, | |
9451 | KVMCLOCK_SYNC_PERIOD); | |
42897d86 MT |
9452 | } |
9453 | ||
d40ccc62 | 9454 | void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu) |
e9b11c17 | 9455 | { |
4cbc418a | 9456 | struct gfn_to_pfn_cache *cache = &vcpu->arch.st.cache; |
95a0d01e | 9457 | int idx; |
344d9588 | 9458 | |
4cbc418a PB |
9459 | kvm_release_pfn(cache->pfn, cache->dirty, cache); |
9460 | ||
50b143e1 | 9461 | kvmclock_reset(vcpu); |
e9b11c17 | 9462 | |
afaf0b2f | 9463 | kvm_x86_ops.vcpu_free(vcpu); |
50b143e1 | 9464 | |
c9b8b07c | 9465 | kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt); |
50b143e1 SC |
9466 | free_cpumask_var(vcpu->arch.wbinvd_dirty_mask); |
9467 | kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu); | |
9468 | kmem_cache_free(x86_fpu_cache, vcpu->arch.guest_fpu); | |
95a0d01e SC |
9469 | |
9470 | kvm_hv_vcpu_uninit(vcpu); | |
9471 | kvm_pmu_destroy(vcpu); | |
9472 | kfree(vcpu->arch.mce_banks); | |
9473 | kvm_free_lapic(vcpu); | |
9474 | idx = srcu_read_lock(&vcpu->kvm->srcu); | |
9475 | kvm_mmu_destroy(vcpu); | |
9476 | srcu_read_unlock(&vcpu->kvm->srcu, idx); | |
9477 | free_page((unsigned long)vcpu->arch.pio_data); | |
9478 | if (!lapic_in_kernel(vcpu)) | |
9479 | static_key_slow_dec(&kvm_no_apic_vcpu); | |
e9b11c17 ZX |
9480 | } |
9481 | ||
d28bc9dd | 9482 | void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event) |
e9b11c17 | 9483 | { |
b7e31be3 RK |
9484 | kvm_lapic_reset(vcpu, init_event); |
9485 | ||
e69fab5d PB |
9486 | vcpu->arch.hflags = 0; |
9487 | ||
c43203ca | 9488 | vcpu->arch.smi_pending = 0; |
52797bf9 | 9489 | vcpu->arch.smi_count = 0; |
7460fb4a AK |
9490 | atomic_set(&vcpu->arch.nmi_queued, 0); |
9491 | vcpu->arch.nmi_pending = 0; | |
448fa4a9 | 9492 | vcpu->arch.nmi_injected = false; |
5f7552d4 NA |
9493 | kvm_clear_interrupt_queue(vcpu); |
9494 | kvm_clear_exception_queue(vcpu); | |
448fa4a9 | 9495 | |
42dbaa5a | 9496 | memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db)); |
ae561ede | 9497 | kvm_update_dr0123(vcpu); |
6f43ed01 | 9498 | vcpu->arch.dr6 = DR6_INIT; |
42dbaa5a | 9499 | vcpu->arch.dr7 = DR7_FIXED_1; |
c8639010 | 9500 | kvm_update_dr7(vcpu); |
42dbaa5a | 9501 | |
1119022c NA |
9502 | vcpu->arch.cr2 = 0; |
9503 | ||
3842d135 | 9504 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
344d9588 | 9505 | vcpu->arch.apf.msr_val = 0; |
c9aaa895 | 9506 | vcpu->arch.st.msr_val = 0; |
3842d135 | 9507 | |
12f9a48f GC |
9508 | kvmclock_reset(vcpu); |
9509 | ||
af585b92 GN |
9510 | kvm_clear_async_pf_completion_queue(vcpu); |
9511 | kvm_async_pf_hash_reset(vcpu); | |
9512 | vcpu->arch.apf.halted = false; | |
3842d135 | 9513 | |
a554d207 WL |
9514 | if (kvm_mpx_supported()) { |
9515 | void *mpx_state_buffer; | |
9516 | ||
9517 | /* | |
9518 | * To avoid have the INIT path from kvm_apic_has_events() that be | |
9519 | * called with loaded FPU and does not let userspace fix the state. | |
9520 | */ | |
f775b13e RR |
9521 | if (init_event) |
9522 | kvm_put_guest_fpu(vcpu); | |
b666a4b6 | 9523 | mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave, |
abd16d68 | 9524 | XFEATURE_BNDREGS); |
a554d207 WL |
9525 | if (mpx_state_buffer) |
9526 | memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state)); | |
b666a4b6 | 9527 | mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave, |
abd16d68 | 9528 | XFEATURE_BNDCSR); |
a554d207 WL |
9529 | if (mpx_state_buffer) |
9530 | memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr)); | |
f775b13e RR |
9531 | if (init_event) |
9532 | kvm_load_guest_fpu(vcpu); | |
a554d207 WL |
9533 | } |
9534 | ||
64d60670 | 9535 | if (!init_event) { |
d28bc9dd | 9536 | kvm_pmu_reset(vcpu); |
64d60670 | 9537 | vcpu->arch.smbase = 0x30000; |
db2336a8 | 9538 | |
db2336a8 | 9539 | vcpu->arch.msr_misc_features_enables = 0; |
a554d207 WL |
9540 | |
9541 | vcpu->arch.xcr0 = XFEATURE_MASK_FP; | |
64d60670 | 9542 | } |
f5132b01 | 9543 | |
66f7b72e JS |
9544 | memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs)); |
9545 | vcpu->arch.regs_avail = ~0; | |
9546 | vcpu->arch.regs_dirty = ~0; | |
9547 | ||
a554d207 WL |
9548 | vcpu->arch.ia32_xss = 0; |
9549 | ||
afaf0b2f | 9550 | kvm_x86_ops.vcpu_reset(vcpu, init_event); |
e9b11c17 ZX |
9551 | } |
9552 | ||
2b4a273b | 9553 | void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector) |
66450a21 JK |
9554 | { |
9555 | struct kvm_segment cs; | |
9556 | ||
9557 | kvm_get_segment(vcpu, &cs, VCPU_SREG_CS); | |
9558 | cs.selector = vector << 8; | |
9559 | cs.base = vector << 12; | |
9560 | kvm_set_segment(vcpu, &cs, VCPU_SREG_CS); | |
9561 | kvm_rip_write(vcpu, 0); | |
e9b11c17 ZX |
9562 | } |
9563 | ||
13a34e06 | 9564 | int kvm_arch_hardware_enable(void) |
e9b11c17 | 9565 | { |
ca84d1a2 ZA |
9566 | struct kvm *kvm; |
9567 | struct kvm_vcpu *vcpu; | |
9568 | int i; | |
0dd6a6ed ZA |
9569 | int ret; |
9570 | u64 local_tsc; | |
9571 | u64 max_tsc = 0; | |
9572 | bool stable, backwards_tsc = false; | |
18863bdd AK |
9573 | |
9574 | kvm_shared_msr_cpu_online(); | |
afaf0b2f | 9575 | ret = kvm_x86_ops.hardware_enable(); |
0dd6a6ed ZA |
9576 | if (ret != 0) |
9577 | return ret; | |
9578 | ||
4ea1636b | 9579 | local_tsc = rdtsc(); |
b0c39dc6 | 9580 | stable = !kvm_check_tsc_unstable(); |
0dd6a6ed ZA |
9581 | list_for_each_entry(kvm, &vm_list, vm_list) { |
9582 | kvm_for_each_vcpu(i, vcpu, kvm) { | |
9583 | if (!stable && vcpu->cpu == smp_processor_id()) | |
105b21bb | 9584 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu); |
0dd6a6ed ZA |
9585 | if (stable && vcpu->arch.last_host_tsc > local_tsc) { |
9586 | backwards_tsc = true; | |
9587 | if (vcpu->arch.last_host_tsc > max_tsc) | |
9588 | max_tsc = vcpu->arch.last_host_tsc; | |
9589 | } | |
9590 | } | |
9591 | } | |
9592 | ||
9593 | /* | |
9594 | * Sometimes, even reliable TSCs go backwards. This happens on | |
9595 | * platforms that reset TSC during suspend or hibernate actions, but | |
9596 | * maintain synchronization. We must compensate. Fortunately, we can | |
9597 | * detect that condition here, which happens early in CPU bringup, | |
9598 | * before any KVM threads can be running. Unfortunately, we can't | |
9599 | * bring the TSCs fully up to date with real time, as we aren't yet far | |
9600 | * enough into CPU bringup that we know how much real time has actually | |
9285ec4c | 9601 | * elapsed; our helper function, ktime_get_boottime_ns() will be using boot |
0dd6a6ed ZA |
9602 | * variables that haven't been updated yet. |
9603 | * | |
9604 | * So we simply find the maximum observed TSC above, then record the | |
9605 | * adjustment to TSC in each VCPU. When the VCPU later gets loaded, | |
9606 | * the adjustment will be applied. Note that we accumulate | |
9607 | * adjustments, in case multiple suspend cycles happen before some VCPU | |
9608 | * gets a chance to run again. In the event that no KVM threads get a | |
9609 | * chance to run, we will miss the entire elapsed period, as we'll have | |
9610 | * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may | |
9611 | * loose cycle time. This isn't too big a deal, since the loss will be | |
9612 | * uniform across all VCPUs (not to mention the scenario is extremely | |
9613 | * unlikely). It is possible that a second hibernate recovery happens | |
9614 | * much faster than a first, causing the observed TSC here to be | |
9615 | * smaller; this would require additional padding adjustment, which is | |
9616 | * why we set last_host_tsc to the local tsc observed here. | |
9617 | * | |
9618 | * N.B. - this code below runs only on platforms with reliable TSC, | |
9619 | * as that is the only way backwards_tsc is set above. Also note | |
9620 | * that this runs for ALL vcpus, which is not a bug; all VCPUs should | |
9621 | * have the same delta_cyc adjustment applied if backwards_tsc | |
9622 | * is detected. Note further, this adjustment is only done once, | |
9623 | * as we reset last_host_tsc on all VCPUs to stop this from being | |
9624 | * called multiple times (one for each physical CPU bringup). | |
9625 | * | |
4a969980 | 9626 | * Platforms with unreliable TSCs don't have to deal with this, they |
0dd6a6ed ZA |
9627 | * will be compensated by the logic in vcpu_load, which sets the TSC to |
9628 | * catchup mode. This will catchup all VCPUs to real time, but cannot | |
9629 | * guarantee that they stay in perfect synchronization. | |
9630 | */ | |
9631 | if (backwards_tsc) { | |
9632 | u64 delta_cyc = max_tsc - local_tsc; | |
9633 | list_for_each_entry(kvm, &vm_list, vm_list) { | |
a826faf1 | 9634 | kvm->arch.backwards_tsc_observed = true; |
0dd6a6ed ZA |
9635 | kvm_for_each_vcpu(i, vcpu, kvm) { |
9636 | vcpu->arch.tsc_offset_adjustment += delta_cyc; | |
9637 | vcpu->arch.last_host_tsc = local_tsc; | |
105b21bb | 9638 | kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu); |
0dd6a6ed ZA |
9639 | } |
9640 | ||
9641 | /* | |
9642 | * We have to disable TSC offset matching.. if you were | |
9643 | * booting a VM while issuing an S4 host suspend.... | |
9644 | * you may have some problem. Solving this issue is | |
9645 | * left as an exercise to the reader. | |
9646 | */ | |
9647 | kvm->arch.last_tsc_nsec = 0; | |
9648 | kvm->arch.last_tsc_write = 0; | |
9649 | } | |
9650 | ||
9651 | } | |
9652 | return 0; | |
e9b11c17 ZX |
9653 | } |
9654 | ||
13a34e06 | 9655 | void kvm_arch_hardware_disable(void) |
e9b11c17 | 9656 | { |
afaf0b2f | 9657 | kvm_x86_ops.hardware_disable(); |
13a34e06 | 9658 | drop_user_return_notifiers(); |
e9b11c17 ZX |
9659 | } |
9660 | ||
b9904085 | 9661 | int kvm_arch_hardware_setup(void *opaque) |
e9b11c17 | 9662 | { |
d008dfdb | 9663 | struct kvm_x86_init_ops *ops = opaque; |
9e9c3fe4 NA |
9664 | int r; |
9665 | ||
91661989 SC |
9666 | rdmsrl_safe(MSR_EFER, &host_efer); |
9667 | ||
408e9a31 PB |
9668 | if (boot_cpu_has(X86_FEATURE_XSAVES)) |
9669 | rdmsrl(MSR_IA32_XSS, host_xss); | |
9670 | ||
d008dfdb | 9671 | r = ops->hardware_setup(); |
9e9c3fe4 NA |
9672 | if (r != 0) |
9673 | return r; | |
9674 | ||
afaf0b2f | 9675 | memcpy(&kvm_x86_ops, ops->runtime_ops, sizeof(kvm_x86_ops)); |
69c6f69a | 9676 | |
408e9a31 PB |
9677 | if (!kvm_cpu_cap_has(X86_FEATURE_XSAVES)) |
9678 | supported_xss = 0; | |
9679 | ||
139f7425 PB |
9680 | #define __kvm_cpu_cap_has(UNUSED_, f) kvm_cpu_cap_has(f) |
9681 | cr4_reserved_bits = __cr4_reserved_bits(__kvm_cpu_cap_has, UNUSED_); | |
9682 | #undef __kvm_cpu_cap_has | |
b11306b5 | 9683 | |
35181e86 HZ |
9684 | if (kvm_has_tsc_control) { |
9685 | /* | |
9686 | * Make sure the user can only configure tsc_khz values that | |
9687 | * fit into a signed integer. | |
273ba457 | 9688 | * A min value is not calculated because it will always |
35181e86 HZ |
9689 | * be 1 on all machines. |
9690 | */ | |
9691 | u64 max = min(0x7fffffffULL, | |
9692 | __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz)); | |
9693 | kvm_max_guest_tsc_khz = max; | |
9694 | ||
ad721883 | 9695 | kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits; |
35181e86 | 9696 | } |
ad721883 | 9697 | |
9e9c3fe4 NA |
9698 | kvm_init_msr_list(); |
9699 | return 0; | |
e9b11c17 ZX |
9700 | } |
9701 | ||
9702 | void kvm_arch_hardware_unsetup(void) | |
9703 | { | |
afaf0b2f | 9704 | kvm_x86_ops.hardware_unsetup(); |
e9b11c17 ZX |
9705 | } |
9706 | ||
b9904085 | 9707 | int kvm_arch_check_processor_compat(void *opaque) |
e9b11c17 | 9708 | { |
f1cdecf5 | 9709 | struct cpuinfo_x86 *c = &cpu_data(smp_processor_id()); |
d008dfdb | 9710 | struct kvm_x86_init_ops *ops = opaque; |
f1cdecf5 SC |
9711 | |
9712 | WARN_ON(!irqs_disabled()); | |
9713 | ||
139f7425 PB |
9714 | if (__cr4_reserved_bits(cpu_has, c) != |
9715 | __cr4_reserved_bits(cpu_has, &boot_cpu_data)) | |
f1cdecf5 SC |
9716 | return -EIO; |
9717 | ||
d008dfdb | 9718 | return ops->check_processor_compatibility(); |
d71ba788 PB |
9719 | } |
9720 | ||
9721 | bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu) | |
9722 | { | |
9723 | return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id; | |
9724 | } | |
9725 | EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp); | |
9726 | ||
9727 | bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu) | |
9728 | { | |
9729 | return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0; | |
e9b11c17 ZX |
9730 | } |
9731 | ||
54e9818f | 9732 | struct static_key kvm_no_apic_vcpu __read_mostly; |
bce87cce | 9733 | EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu); |
54e9818f | 9734 | |
e790d9ef RK |
9735 | void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu) |
9736 | { | |
b35e5548 LX |
9737 | struct kvm_pmu *pmu = vcpu_to_pmu(vcpu); |
9738 | ||
c595ceee | 9739 | vcpu->arch.l1tf_flush_l1d = true; |
b35e5548 LX |
9740 | if (pmu->version && unlikely(pmu->event_count)) { |
9741 | pmu->need_cleanup = true; | |
9742 | kvm_make_request(KVM_REQ_PMU, vcpu); | |
9743 | } | |
afaf0b2f | 9744 | kvm_x86_ops.sched_in(vcpu, cpu); |
e790d9ef RK |
9745 | } |
9746 | ||
562b6b08 SC |
9747 | void kvm_arch_free_vm(struct kvm *kvm) |
9748 | { | |
9749 | kfree(kvm->arch.hyperv.hv_pa_pg); | |
9750 | vfree(kvm); | |
e790d9ef RK |
9751 | } |
9752 | ||
562b6b08 | 9753 | |
e08b9637 | 9754 | int kvm_arch_init_vm(struct kvm *kvm, unsigned long type) |
d19a9cd2 | 9755 | { |
e08b9637 CO |
9756 | if (type) |
9757 | return -EINVAL; | |
9758 | ||
6ef768fa | 9759 | INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list); |
f05e70ac | 9760 | INIT_LIST_HEAD(&kvm->arch.active_mmu_pages); |
10605204 | 9761 | INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages); |
1aa9b957 | 9762 | INIT_LIST_HEAD(&kvm->arch.lpage_disallowed_mmu_pages); |
4d5c5d0f | 9763 | INIT_LIST_HEAD(&kvm->arch.assigned_dev_head); |
e0f0bbc5 | 9764 | atomic_set(&kvm->arch.noncoherent_dma_count, 0); |
d19a9cd2 | 9765 | |
5550af4d SY |
9766 | /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */ |
9767 | set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap); | |
7a84428a AW |
9768 | /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */ |
9769 | set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID, | |
9770 | &kvm->arch.irq_sources_bitmap); | |
5550af4d | 9771 | |
038f8c11 | 9772 | raw_spin_lock_init(&kvm->arch.tsc_write_lock); |
1e08ec4a | 9773 | mutex_init(&kvm->arch.apic_map_lock); |
d828199e MT |
9774 | spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock); |
9775 | ||
8171cd68 | 9776 | kvm->arch.kvmclock_offset = -get_kvmclock_base_ns(); |
d828199e | 9777 | pvclock_update_vm_gtod_copy(kvm); |
53f658b3 | 9778 | |
6fbbde9a DS |
9779 | kvm->arch.guest_can_read_msr_platform_info = true; |
9780 | ||
7e44e449 | 9781 | INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn); |
332967a3 | 9782 | INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn); |
7e44e449 | 9783 | |
cbc0236a | 9784 | kvm_hv_init_vm(kvm); |
0eb05bf2 | 9785 | kvm_page_track_init(kvm); |
13d268ca | 9786 | kvm_mmu_init_vm(kvm); |
0eb05bf2 | 9787 | |
afaf0b2f | 9788 | return kvm_x86_ops.vm_init(kvm); |
d19a9cd2 ZX |
9789 | } |
9790 | ||
1aa9b957 JS |
9791 | int kvm_arch_post_init_vm(struct kvm *kvm) |
9792 | { | |
9793 | return kvm_mmu_post_init_vm(kvm); | |
9794 | } | |
9795 | ||
d19a9cd2 ZX |
9796 | static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu) |
9797 | { | |
ec7660cc | 9798 | vcpu_load(vcpu); |
d19a9cd2 ZX |
9799 | kvm_mmu_unload(vcpu); |
9800 | vcpu_put(vcpu); | |
9801 | } | |
9802 | ||
9803 | static void kvm_free_vcpus(struct kvm *kvm) | |
9804 | { | |
9805 | unsigned int i; | |
988a2cae | 9806 | struct kvm_vcpu *vcpu; |
d19a9cd2 ZX |
9807 | |
9808 | /* | |
9809 | * Unpin any mmu pages first. | |
9810 | */ | |
af585b92 GN |
9811 | kvm_for_each_vcpu(i, vcpu, kvm) { |
9812 | kvm_clear_async_pf_completion_queue(vcpu); | |
988a2cae | 9813 | kvm_unload_vcpu_mmu(vcpu); |
af585b92 | 9814 | } |
988a2cae | 9815 | kvm_for_each_vcpu(i, vcpu, kvm) |
4543bdc0 | 9816 | kvm_vcpu_destroy(vcpu); |
988a2cae GN |
9817 | |
9818 | mutex_lock(&kvm->lock); | |
9819 | for (i = 0; i < atomic_read(&kvm->online_vcpus); i++) | |
9820 | kvm->vcpus[i] = NULL; | |
d19a9cd2 | 9821 | |
988a2cae GN |
9822 | atomic_set(&kvm->online_vcpus, 0); |
9823 | mutex_unlock(&kvm->lock); | |
d19a9cd2 ZX |
9824 | } |
9825 | ||
ad8ba2cd SY |
9826 | void kvm_arch_sync_events(struct kvm *kvm) |
9827 | { | |
332967a3 | 9828 | cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work); |
7e44e449 | 9829 | cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work); |
aea924f6 | 9830 | kvm_free_pit(kvm); |
ad8ba2cd SY |
9831 | } |
9832 | ||
1d8007bd | 9833 | int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size) |
9da0e4d5 PB |
9834 | { |
9835 | int i, r; | |
0577d1ab | 9836 | unsigned long hva, uninitialized_var(old_npages); |
f0d648bd | 9837 | struct kvm_memslots *slots = kvm_memslots(kvm); |
0577d1ab | 9838 | struct kvm_memory_slot *slot; |
9da0e4d5 PB |
9839 | |
9840 | /* Called with kvm->slots_lock held. */ | |
1d8007bd PB |
9841 | if (WARN_ON(id >= KVM_MEM_SLOTS_NUM)) |
9842 | return -EINVAL; | |
9da0e4d5 | 9843 | |
f0d648bd PB |
9844 | slot = id_to_memslot(slots, id); |
9845 | if (size) { | |
0577d1ab | 9846 | if (slot && slot->npages) |
f0d648bd PB |
9847 | return -EEXIST; |
9848 | ||
9849 | /* | |
9850 | * MAP_SHARED to prevent internal slot pages from being moved | |
9851 | * by fork()/COW. | |
9852 | */ | |
9853 | hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE, | |
9854 | MAP_SHARED | MAP_ANONYMOUS, 0); | |
9855 | if (IS_ERR((void *)hva)) | |
9856 | return PTR_ERR((void *)hva); | |
9857 | } else { | |
0577d1ab | 9858 | if (!slot || !slot->npages) |
f0d648bd PB |
9859 | return 0; |
9860 | ||
abbed4fa SC |
9861 | /* |
9862 | * Stuff a non-canonical value to catch use-after-delete. This | |
9863 | * ends up being 0 on 32-bit KVM, but there's no better | |
9864 | * alternative. | |
9865 | */ | |
9866 | hva = (unsigned long)(0xdeadull << 48); | |
0577d1ab | 9867 | old_npages = slot->npages; |
f0d648bd PB |
9868 | } |
9869 | ||
9da0e4d5 | 9870 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) { |
1d8007bd | 9871 | struct kvm_userspace_memory_region m; |
9da0e4d5 | 9872 | |
1d8007bd PB |
9873 | m.slot = id | (i << 16); |
9874 | m.flags = 0; | |
9875 | m.guest_phys_addr = gpa; | |
f0d648bd | 9876 | m.userspace_addr = hva; |
1d8007bd | 9877 | m.memory_size = size; |
9da0e4d5 PB |
9878 | r = __kvm_set_memory_region(kvm, &m); |
9879 | if (r < 0) | |
9880 | return r; | |
9881 | } | |
9882 | ||
103c763c | 9883 | if (!size) |
0577d1ab | 9884 | vm_munmap(hva, old_npages * PAGE_SIZE); |
f0d648bd | 9885 | |
9da0e4d5 PB |
9886 | return 0; |
9887 | } | |
9888 | EXPORT_SYMBOL_GPL(__x86_set_memory_region); | |
9889 | ||
1aa9b957 JS |
9890 | void kvm_arch_pre_destroy_vm(struct kvm *kvm) |
9891 | { | |
9892 | kvm_mmu_pre_destroy_vm(kvm); | |
9893 | } | |
9894 | ||
d19a9cd2 ZX |
9895 | void kvm_arch_destroy_vm(struct kvm *kvm) |
9896 | { | |
27469d29 AH |
9897 | if (current->mm == kvm->mm) { |
9898 | /* | |
9899 | * Free memory regions allocated on behalf of userspace, | |
9900 | * unless the the memory map has changed due to process exit | |
9901 | * or fd copying. | |
9902 | */ | |
6a3c623b PX |
9903 | mutex_lock(&kvm->slots_lock); |
9904 | __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, | |
9905 | 0, 0); | |
9906 | __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, | |
9907 | 0, 0); | |
9908 | __x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0); | |
9909 | mutex_unlock(&kvm->slots_lock); | |
27469d29 | 9910 | } |
afaf0b2f SC |
9911 | if (kvm_x86_ops.vm_destroy) |
9912 | kvm_x86_ops.vm_destroy(kvm); | |
c761159c PX |
9913 | kvm_pic_destroy(kvm); |
9914 | kvm_ioapic_destroy(kvm); | |
d19a9cd2 | 9915 | kvm_free_vcpus(kvm); |
af1bae54 | 9916 | kvfree(rcu_dereference_check(kvm->arch.apic_map, 1)); |
66bb8a06 | 9917 | kfree(srcu_dereference_check(kvm->arch.pmu_event_filter, &kvm->srcu, 1)); |
13d268ca | 9918 | kvm_mmu_uninit_vm(kvm); |
2beb6dad | 9919 | kvm_page_track_cleanup(kvm); |
cbc0236a | 9920 | kvm_hv_destroy_vm(kvm); |
d19a9cd2 | 9921 | } |
0de10343 | 9922 | |
e96c81ee | 9923 | void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot) |
db3fe4eb TY |
9924 | { |
9925 | int i; | |
9926 | ||
d89cc617 | 9927 | for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) { |
e96c81ee SC |
9928 | kvfree(slot->arch.rmap[i]); |
9929 | slot->arch.rmap[i] = NULL; | |
9930 | ||
d89cc617 TY |
9931 | if (i == 0) |
9932 | continue; | |
9933 | ||
e96c81ee SC |
9934 | kvfree(slot->arch.lpage_info[i - 1]); |
9935 | slot->arch.lpage_info[i - 1] = NULL; | |
db3fe4eb | 9936 | } |
21ebbeda | 9937 | |
e96c81ee | 9938 | kvm_page_track_free_memslot(slot); |
db3fe4eb TY |
9939 | } |
9940 | ||
0dab98b7 SC |
9941 | static int kvm_alloc_memslot_metadata(struct kvm_memory_slot *slot, |
9942 | unsigned long npages) | |
db3fe4eb TY |
9943 | { |
9944 | int i; | |
9945 | ||
edd4fa37 SC |
9946 | /* |
9947 | * Clear out the previous array pointers for the KVM_MR_MOVE case. The | |
9948 | * old arrays will be freed by __kvm_set_memory_region() if installing | |
9949 | * the new memslot is successful. | |
9950 | */ | |
9951 | memset(&slot->arch, 0, sizeof(slot->arch)); | |
9952 | ||
d89cc617 | 9953 | for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) { |
92f94f1e | 9954 | struct kvm_lpage_info *linfo; |
db3fe4eb TY |
9955 | unsigned long ugfn; |
9956 | int lpages; | |
d89cc617 | 9957 | int level = i + 1; |
db3fe4eb TY |
9958 | |
9959 | lpages = gfn_to_index(slot->base_gfn + npages - 1, | |
9960 | slot->base_gfn, level) + 1; | |
9961 | ||
d89cc617 | 9962 | slot->arch.rmap[i] = |
778e1cdd | 9963 | kvcalloc(lpages, sizeof(*slot->arch.rmap[i]), |
254272ce | 9964 | GFP_KERNEL_ACCOUNT); |
d89cc617 | 9965 | if (!slot->arch.rmap[i]) |
77d11309 | 9966 | goto out_free; |
d89cc617 TY |
9967 | if (i == 0) |
9968 | continue; | |
77d11309 | 9969 | |
254272ce | 9970 | linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT); |
92f94f1e | 9971 | if (!linfo) |
db3fe4eb TY |
9972 | goto out_free; |
9973 | ||
92f94f1e XG |
9974 | slot->arch.lpage_info[i - 1] = linfo; |
9975 | ||
db3fe4eb | 9976 | if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1)) |
92f94f1e | 9977 | linfo[0].disallow_lpage = 1; |
db3fe4eb | 9978 | if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1)) |
92f94f1e | 9979 | linfo[lpages - 1].disallow_lpage = 1; |
db3fe4eb TY |
9980 | ugfn = slot->userspace_addr >> PAGE_SHIFT; |
9981 | /* | |
9982 | * If the gfn and userspace address are not aligned wrt each | |
600087b6 | 9983 | * other, disable large page support for this slot. |
db3fe4eb | 9984 | */ |
600087b6 | 9985 | if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1)) { |
db3fe4eb TY |
9986 | unsigned long j; |
9987 | ||
9988 | for (j = 0; j < lpages; ++j) | |
92f94f1e | 9989 | linfo[j].disallow_lpage = 1; |
db3fe4eb TY |
9990 | } |
9991 | } | |
9992 | ||
21ebbeda XG |
9993 | if (kvm_page_track_create_memslot(slot, npages)) |
9994 | goto out_free; | |
9995 | ||
db3fe4eb TY |
9996 | return 0; |
9997 | ||
9998 | out_free: | |
d89cc617 | 9999 | for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) { |
548ef284 | 10000 | kvfree(slot->arch.rmap[i]); |
d89cc617 TY |
10001 | slot->arch.rmap[i] = NULL; |
10002 | if (i == 0) | |
10003 | continue; | |
10004 | ||
548ef284 | 10005 | kvfree(slot->arch.lpage_info[i - 1]); |
d89cc617 | 10006 | slot->arch.lpage_info[i - 1] = NULL; |
db3fe4eb TY |
10007 | } |
10008 | return -ENOMEM; | |
10009 | } | |
10010 | ||
15248258 | 10011 | void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen) |
e59dbe09 | 10012 | { |
91724814 BO |
10013 | struct kvm_vcpu *vcpu; |
10014 | int i; | |
10015 | ||
e6dff7d1 TY |
10016 | /* |
10017 | * memslots->generation has been incremented. | |
10018 | * mmio generation may have reached its maximum value. | |
10019 | */ | |
15248258 | 10020 | kvm_mmu_invalidate_mmio_sptes(kvm, gen); |
91724814 BO |
10021 | |
10022 | /* Force re-initialization of steal_time cache */ | |
10023 | kvm_for_each_vcpu(i, vcpu, kvm) | |
10024 | kvm_vcpu_kick(vcpu); | |
e59dbe09 TY |
10025 | } |
10026 | ||
f7784b8e MT |
10027 | int kvm_arch_prepare_memory_region(struct kvm *kvm, |
10028 | struct kvm_memory_slot *memslot, | |
09170a49 | 10029 | const struct kvm_userspace_memory_region *mem, |
7b6195a9 | 10030 | enum kvm_mr_change change) |
0de10343 | 10031 | { |
0dab98b7 SC |
10032 | if (change == KVM_MR_CREATE || change == KVM_MR_MOVE) |
10033 | return kvm_alloc_memslot_metadata(memslot, | |
10034 | mem->memory_size >> PAGE_SHIFT); | |
f7784b8e MT |
10035 | return 0; |
10036 | } | |
10037 | ||
88178fd4 KH |
10038 | static void kvm_mmu_slot_apply_flags(struct kvm *kvm, |
10039 | struct kvm_memory_slot *new) | |
10040 | { | |
10041 | /* Still write protect RO slot */ | |
10042 | if (new->flags & KVM_MEM_READONLY) { | |
3c9bd400 | 10043 | kvm_mmu_slot_remove_write_access(kvm, new, PT_PAGE_TABLE_LEVEL); |
88178fd4 KH |
10044 | return; |
10045 | } | |
10046 | ||
10047 | /* | |
10048 | * Call kvm_x86_ops dirty logging hooks when they are valid. | |
10049 | * | |
afaf0b2f | 10050 | * kvm_x86_ops.slot_disable_log_dirty is called when: |
88178fd4 KH |
10051 | * |
10052 | * - KVM_MR_CREATE with dirty logging is disabled | |
10053 | * - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag | |
10054 | * | |
10055 | * The reason is, in case of PML, we need to set D-bit for any slots | |
10056 | * with dirty logging disabled in order to eliminate unnecessary GPA | |
0a03cbda | 10057 | * logging in PML buffer (and potential PML buffer full VMEXIT). This |
88178fd4 | 10058 | * guarantees leaving PML enabled during guest's lifetime won't have |
bdd303cb | 10059 | * any additional overhead from PML when guest is running with dirty |
88178fd4 KH |
10060 | * logging disabled for memory slots. |
10061 | * | |
afaf0b2f | 10062 | * kvm_x86_ops.slot_enable_log_dirty is called when switching new slot |
88178fd4 KH |
10063 | * to dirty logging mode. |
10064 | * | |
10065 | * If kvm_x86_ops dirty logging hooks are invalid, use write protect. | |
10066 | * | |
10067 | * In case of write protect: | |
10068 | * | |
10069 | * Write protect all pages for dirty logging. | |
10070 | * | |
10071 | * All the sptes including the large sptes which point to this | |
10072 | * slot are set to readonly. We can not create any new large | |
10073 | * spte on this slot until the end of the logging. | |
10074 | * | |
10075 | * See the comments in fast_page_fault(). | |
10076 | */ | |
10077 | if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) { | |
afaf0b2f SC |
10078 | if (kvm_x86_ops.slot_enable_log_dirty) { |
10079 | kvm_x86_ops.slot_enable_log_dirty(kvm, new); | |
3c9bd400 JZ |
10080 | } else { |
10081 | int level = | |
10082 | kvm_dirty_log_manual_protect_and_init_set(kvm) ? | |
10083 | PT_DIRECTORY_LEVEL : PT_PAGE_TABLE_LEVEL; | |
10084 | ||
10085 | /* | |
10086 | * If we're with initial-all-set, we don't need | |
10087 | * to write protect any small page because | |
10088 | * they're reported as dirty already. However | |
10089 | * we still need to write-protect huge pages | |
10090 | * so that the page split can happen lazily on | |
10091 | * the first write to the huge page. | |
10092 | */ | |
10093 | kvm_mmu_slot_remove_write_access(kvm, new, level); | |
10094 | } | |
88178fd4 | 10095 | } else { |
afaf0b2f SC |
10096 | if (kvm_x86_ops.slot_disable_log_dirty) |
10097 | kvm_x86_ops.slot_disable_log_dirty(kvm, new); | |
88178fd4 KH |
10098 | } |
10099 | } | |
10100 | ||
f7784b8e | 10101 | void kvm_arch_commit_memory_region(struct kvm *kvm, |
09170a49 | 10102 | const struct kvm_userspace_memory_region *mem, |
9d4c197c | 10103 | struct kvm_memory_slot *old, |
f36f3f28 | 10104 | const struct kvm_memory_slot *new, |
8482644a | 10105 | enum kvm_mr_change change) |
f7784b8e | 10106 | { |
48c0e4e9 | 10107 | if (!kvm->arch.n_requested_mmu_pages) |
4d66623c WY |
10108 | kvm_mmu_change_mmu_pages(kvm, |
10109 | kvm_mmu_calculate_default_mmu_pages(kvm)); | |
1c91cad4 | 10110 | |
3ea3b7fa WL |
10111 | /* |
10112 | * Dirty logging tracks sptes in 4k granularity, meaning that large | |
10113 | * sptes have to be split. If live migration is successful, the guest | |
10114 | * in the source machine will be destroyed and large sptes will be | |
10115 | * created in the destination. However, if the guest continues to run | |
10116 | * in the source machine (for example if live migration fails), small | |
10117 | * sptes will remain around and cause bad performance. | |
10118 | * | |
10119 | * Scan sptes if dirty logging has been stopped, dropping those | |
10120 | * which can be collapsed into a single large-page spte. Later | |
10121 | * page faults will create the large-page sptes. | |
319109a2 SC |
10122 | * |
10123 | * There is no need to do this in any of the following cases: | |
10124 | * CREATE: No dirty mappings will already exist. | |
10125 | * MOVE/DELETE: The old mappings will already have been cleaned up by | |
10126 | * kvm_arch_flush_shadow_memslot() | |
3ea3b7fa | 10127 | */ |
319109a2 | 10128 | if (change == KVM_MR_FLAGS_ONLY && |
3ea3b7fa WL |
10129 | (old->flags & KVM_MEM_LOG_DIRTY_PAGES) && |
10130 | !(new->flags & KVM_MEM_LOG_DIRTY_PAGES)) | |
10131 | kvm_mmu_zap_collapsible_sptes(kvm, new); | |
10132 | ||
c972f3b1 | 10133 | /* |
88178fd4 | 10134 | * Set up write protection and/or dirty logging for the new slot. |
c126d94f | 10135 | * |
88178fd4 KH |
10136 | * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have |
10137 | * been zapped so no dirty logging staff is needed for old slot. For | |
10138 | * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the | |
10139 | * new and it's also covered when dealing with the new slot. | |
f36f3f28 PB |
10140 | * |
10141 | * FIXME: const-ify all uses of struct kvm_memory_slot. | |
c972f3b1 | 10142 | */ |
88178fd4 | 10143 | if (change != KVM_MR_DELETE) |
f36f3f28 | 10144 | kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new); |
21198846 SC |
10145 | |
10146 | /* Free the arrays associated with the old memslot. */ | |
10147 | if (change == KVM_MR_MOVE) | |
e96c81ee | 10148 | kvm_arch_free_memslot(kvm, old); |
0de10343 | 10149 | } |
1d737c8a | 10150 | |
2df72e9b | 10151 | void kvm_arch_flush_shadow_all(struct kvm *kvm) |
34d4cb8f | 10152 | { |
7390de1e | 10153 | kvm_mmu_zap_all(kvm); |
34d4cb8f MT |
10154 | } |
10155 | ||
2df72e9b MT |
10156 | void kvm_arch_flush_shadow_memslot(struct kvm *kvm, |
10157 | struct kvm_memory_slot *slot) | |
10158 | { | |
ae7cd873 | 10159 | kvm_page_track_flush_slot(kvm, slot); |
2df72e9b MT |
10160 | } |
10161 | ||
e6c67d8c LA |
10162 | static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu) |
10163 | { | |
10164 | return (is_guest_mode(vcpu) && | |
afaf0b2f SC |
10165 | kvm_x86_ops.guest_apic_has_interrupt && |
10166 | kvm_x86_ops.guest_apic_has_interrupt(vcpu)); | |
e6c67d8c LA |
10167 | } |
10168 | ||
5d9bc648 PB |
10169 | static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu) |
10170 | { | |
10171 | if (!list_empty_careful(&vcpu->async_pf.done)) | |
10172 | return true; | |
10173 | ||
10174 | if (kvm_apic_has_events(vcpu)) | |
10175 | return true; | |
10176 | ||
10177 | if (vcpu->arch.pv.pv_unhalted) | |
10178 | return true; | |
10179 | ||
a5f01f8e WL |
10180 | if (vcpu->arch.exception.pending) |
10181 | return true; | |
10182 | ||
47a66eed Z |
10183 | if (kvm_test_request(KVM_REQ_NMI, vcpu) || |
10184 | (vcpu->arch.nmi_pending && | |
afaf0b2f | 10185 | kvm_x86_ops.nmi_allowed(vcpu))) |
5d9bc648 PB |
10186 | return true; |
10187 | ||
47a66eed Z |
10188 | if (kvm_test_request(KVM_REQ_SMI, vcpu) || |
10189 | (vcpu->arch.smi_pending && !is_smm(vcpu))) | |
73917739 PB |
10190 | return true; |
10191 | ||
5d9bc648 | 10192 | if (kvm_arch_interrupt_allowed(vcpu) && |
e6c67d8c LA |
10193 | (kvm_cpu_has_interrupt(vcpu) || |
10194 | kvm_guest_apic_has_interrupt(vcpu))) | |
5d9bc648 PB |
10195 | return true; |
10196 | ||
1f4b34f8 AS |
10197 | if (kvm_hv_has_stimer_pending(vcpu)) |
10198 | return true; | |
10199 | ||
5d9bc648 PB |
10200 | return false; |
10201 | } | |
10202 | ||
1d737c8a ZX |
10203 | int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu) |
10204 | { | |
5d9bc648 | 10205 | return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu); |
1d737c8a | 10206 | } |
5736199a | 10207 | |
17e433b5 WL |
10208 | bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu) |
10209 | { | |
10210 | if (READ_ONCE(vcpu->arch.pv.pv_unhalted)) | |
10211 | return true; | |
10212 | ||
10213 | if (kvm_test_request(KVM_REQ_NMI, vcpu) || | |
10214 | kvm_test_request(KVM_REQ_SMI, vcpu) || | |
10215 | kvm_test_request(KVM_REQ_EVENT, vcpu)) | |
10216 | return true; | |
10217 | ||
afaf0b2f | 10218 | if (vcpu->arch.apicv_active && kvm_x86_ops.dy_apicv_has_pending_interrupt(vcpu)) |
17e433b5 WL |
10219 | return true; |
10220 | ||
10221 | return false; | |
10222 | } | |
10223 | ||
199b5763 LM |
10224 | bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu) |
10225 | { | |
de63ad4c | 10226 | return vcpu->arch.preempted_in_kernel; |
199b5763 LM |
10227 | } |
10228 | ||
b6d33834 | 10229 | int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu) |
5736199a | 10230 | { |
b6d33834 | 10231 | return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE; |
5736199a | 10232 | } |
78646121 GN |
10233 | |
10234 | int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu) | |
10235 | { | |
afaf0b2f | 10236 | return kvm_x86_ops.interrupt_allowed(vcpu); |
78646121 | 10237 | } |
229456fc | 10238 | |
82b32774 | 10239 | unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu) |
f92653ee | 10240 | { |
82b32774 NA |
10241 | if (is_64_bit_mode(vcpu)) |
10242 | return kvm_rip_read(vcpu); | |
10243 | return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) + | |
10244 | kvm_rip_read(vcpu)); | |
10245 | } | |
10246 | EXPORT_SYMBOL_GPL(kvm_get_linear_rip); | |
f92653ee | 10247 | |
82b32774 NA |
10248 | bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip) |
10249 | { | |
10250 | return kvm_get_linear_rip(vcpu) == linear_rip; | |
f92653ee JK |
10251 | } |
10252 | EXPORT_SYMBOL_GPL(kvm_is_linear_rip); | |
10253 | ||
94fe45da JK |
10254 | unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu) |
10255 | { | |
10256 | unsigned long rflags; | |
10257 | ||
afaf0b2f | 10258 | rflags = kvm_x86_ops.get_rflags(vcpu); |
94fe45da | 10259 | if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) |
c310bac5 | 10260 | rflags &= ~X86_EFLAGS_TF; |
94fe45da JK |
10261 | return rflags; |
10262 | } | |
10263 | EXPORT_SYMBOL_GPL(kvm_get_rflags); | |
10264 | ||
6addfc42 | 10265 | static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags) |
94fe45da JK |
10266 | { |
10267 | if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP && | |
f92653ee | 10268 | kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip)) |
c310bac5 | 10269 | rflags |= X86_EFLAGS_TF; |
afaf0b2f | 10270 | kvm_x86_ops.set_rflags(vcpu, rflags); |
6addfc42 PB |
10271 | } |
10272 | ||
10273 | void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags) | |
10274 | { | |
10275 | __kvm_set_rflags(vcpu, rflags); | |
3842d135 | 10276 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
94fe45da JK |
10277 | } |
10278 | EXPORT_SYMBOL_GPL(kvm_set_rflags); | |
10279 | ||
56028d08 GN |
10280 | void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work) |
10281 | { | |
10282 | int r; | |
10283 | ||
44dd3ffa | 10284 | if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) || |
f2e10669 | 10285 | work->wakeup_all) |
56028d08 GN |
10286 | return; |
10287 | ||
10288 | r = kvm_mmu_reload(vcpu); | |
10289 | if (unlikely(r)) | |
10290 | return; | |
10291 | ||
44dd3ffa | 10292 | if (!vcpu->arch.mmu->direct_map && |
d8dd54e0 | 10293 | work->arch.cr3 != vcpu->arch.mmu->get_guest_pgd(vcpu)) |
fb67e14f XG |
10294 | return; |
10295 | ||
7a02674d | 10296 | kvm_mmu_do_page_fault(vcpu, work->cr2_or_gpa, 0, true); |
56028d08 GN |
10297 | } |
10298 | ||
af585b92 GN |
10299 | static inline u32 kvm_async_pf_hash_fn(gfn_t gfn) |
10300 | { | |
10301 | return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU)); | |
10302 | } | |
10303 | ||
10304 | static inline u32 kvm_async_pf_next_probe(u32 key) | |
10305 | { | |
10306 | return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1); | |
10307 | } | |
10308 | ||
10309 | static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn) | |
10310 | { | |
10311 | u32 key = kvm_async_pf_hash_fn(gfn); | |
10312 | ||
10313 | while (vcpu->arch.apf.gfns[key] != ~0) | |
10314 | key = kvm_async_pf_next_probe(key); | |
10315 | ||
10316 | vcpu->arch.apf.gfns[key] = gfn; | |
10317 | } | |
10318 | ||
10319 | static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn) | |
10320 | { | |
10321 | int i; | |
10322 | u32 key = kvm_async_pf_hash_fn(gfn); | |
10323 | ||
10324 | for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) && | |
c7d28c24 XG |
10325 | (vcpu->arch.apf.gfns[key] != gfn && |
10326 | vcpu->arch.apf.gfns[key] != ~0); i++) | |
af585b92 GN |
10327 | key = kvm_async_pf_next_probe(key); |
10328 | ||
10329 | return key; | |
10330 | } | |
10331 | ||
10332 | bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn) | |
10333 | { | |
10334 | return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn; | |
10335 | } | |
10336 | ||
10337 | static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn) | |
10338 | { | |
10339 | u32 i, j, k; | |
10340 | ||
10341 | i = j = kvm_async_pf_gfn_slot(vcpu, gfn); | |
10342 | while (true) { | |
10343 | vcpu->arch.apf.gfns[i] = ~0; | |
10344 | do { | |
10345 | j = kvm_async_pf_next_probe(j); | |
10346 | if (vcpu->arch.apf.gfns[j] == ~0) | |
10347 | return; | |
10348 | k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]); | |
10349 | /* | |
10350 | * k lies cyclically in ]i,j] | |
10351 | * | i.k.j | | |
10352 | * |....j i.k.| or |.k..j i...| | |
10353 | */ | |
10354 | } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j)); | |
10355 | vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j]; | |
10356 | i = j; | |
10357 | } | |
10358 | } | |
10359 | ||
7c90705b GN |
10360 | static int apf_put_user(struct kvm_vcpu *vcpu, u32 val) |
10361 | { | |
4e335d9e PB |
10362 | |
10363 | return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val, | |
10364 | sizeof(val)); | |
7c90705b GN |
10365 | } |
10366 | ||
9a6e7c39 WL |
10367 | static int apf_get_user(struct kvm_vcpu *vcpu, u32 *val) |
10368 | { | |
10369 | ||
10370 | return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, val, | |
10371 | sizeof(u32)); | |
10372 | } | |
10373 | ||
1dfdb45e PB |
10374 | static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu) |
10375 | { | |
10376 | if (!vcpu->arch.apf.delivery_as_pf_vmexit && is_guest_mode(vcpu)) | |
10377 | return false; | |
10378 | ||
10379 | if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) || | |
10380 | (vcpu->arch.apf.send_user_only && | |
afaf0b2f | 10381 | kvm_x86_ops.get_cpl(vcpu) == 0)) |
1dfdb45e PB |
10382 | return false; |
10383 | ||
10384 | return true; | |
10385 | } | |
10386 | ||
10387 | bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu) | |
10388 | { | |
10389 | if (unlikely(!lapic_in_kernel(vcpu) || | |
10390 | kvm_event_needs_reinjection(vcpu) || | |
10391 | vcpu->arch.exception.pending)) | |
10392 | return false; | |
10393 | ||
10394 | if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu)) | |
10395 | return false; | |
10396 | ||
10397 | /* | |
10398 | * If interrupts are off we cannot even use an artificial | |
10399 | * halt state. | |
10400 | */ | |
afaf0b2f | 10401 | return kvm_x86_ops.interrupt_allowed(vcpu); |
1dfdb45e PB |
10402 | } |
10403 | ||
af585b92 GN |
10404 | void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu, |
10405 | struct kvm_async_pf *work) | |
10406 | { | |
6389ee94 AK |
10407 | struct x86_exception fault; |
10408 | ||
736c291c | 10409 | trace_kvm_async_pf_not_present(work->arch.token, work->cr2_or_gpa); |
af585b92 | 10410 | kvm_add_async_pf_gfn(vcpu, work->arch.gfn); |
7c90705b | 10411 | |
1dfdb45e PB |
10412 | if (kvm_can_deliver_async_pf(vcpu) && |
10413 | !apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) { | |
6389ee94 AK |
10414 | fault.vector = PF_VECTOR; |
10415 | fault.error_code_valid = true; | |
10416 | fault.error_code = 0; | |
10417 | fault.nested_page_fault = false; | |
10418 | fault.address = work->arch.token; | |
adfe20fb | 10419 | fault.async_page_fault = true; |
6389ee94 | 10420 | kvm_inject_page_fault(vcpu, &fault); |
1dfdb45e PB |
10421 | } else { |
10422 | /* | |
10423 | * It is not possible to deliver a paravirtualized asynchronous | |
10424 | * page fault, but putting the guest in an artificial halt state | |
10425 | * can be beneficial nevertheless: if an interrupt arrives, we | |
10426 | * can deliver it timely and perhaps the guest will schedule | |
10427 | * another process. When the instruction that triggered a page | |
10428 | * fault is retried, hopefully the page will be ready in the host. | |
10429 | */ | |
10430 | kvm_make_request(KVM_REQ_APF_HALT, vcpu); | |
7c90705b | 10431 | } |
af585b92 GN |
10432 | } |
10433 | ||
10434 | void kvm_arch_async_page_present(struct kvm_vcpu *vcpu, | |
10435 | struct kvm_async_pf *work) | |
10436 | { | |
6389ee94 | 10437 | struct x86_exception fault; |
9a6e7c39 | 10438 | u32 val; |
6389ee94 | 10439 | |
f2e10669 | 10440 | if (work->wakeup_all) |
7c90705b GN |
10441 | work->arch.token = ~0; /* broadcast wakeup */ |
10442 | else | |
10443 | kvm_del_async_pf_gfn(vcpu, work->arch.gfn); | |
736c291c | 10444 | trace_kvm_async_pf_ready(work->arch.token, work->cr2_or_gpa); |
7c90705b | 10445 | |
9a6e7c39 WL |
10446 | if (vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED && |
10447 | !apf_get_user(vcpu, &val)) { | |
10448 | if (val == KVM_PV_REASON_PAGE_NOT_PRESENT && | |
10449 | vcpu->arch.exception.pending && | |
10450 | vcpu->arch.exception.nr == PF_VECTOR && | |
10451 | !apf_put_user(vcpu, 0)) { | |
10452 | vcpu->arch.exception.injected = false; | |
10453 | vcpu->arch.exception.pending = false; | |
10454 | vcpu->arch.exception.nr = 0; | |
10455 | vcpu->arch.exception.has_error_code = false; | |
10456 | vcpu->arch.exception.error_code = 0; | |
c851436a JM |
10457 | vcpu->arch.exception.has_payload = false; |
10458 | vcpu->arch.exception.payload = 0; | |
9a6e7c39 WL |
10459 | } else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) { |
10460 | fault.vector = PF_VECTOR; | |
10461 | fault.error_code_valid = true; | |
10462 | fault.error_code = 0; | |
10463 | fault.nested_page_fault = false; | |
10464 | fault.address = work->arch.token; | |
10465 | fault.async_page_fault = true; | |
10466 | kvm_inject_page_fault(vcpu, &fault); | |
10467 | } | |
7c90705b | 10468 | } |
e6d53e3b | 10469 | vcpu->arch.apf.halted = false; |
a4fa1635 | 10470 | vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE; |
7c90705b GN |
10471 | } |
10472 | ||
10473 | bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu) | |
10474 | { | |
10475 | if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED)) | |
10476 | return true; | |
10477 | else | |
9bc1f09f | 10478 | return kvm_can_do_async_pf(vcpu); |
af585b92 GN |
10479 | } |
10480 | ||
5544eb9b PB |
10481 | void kvm_arch_start_assignment(struct kvm *kvm) |
10482 | { | |
10483 | atomic_inc(&kvm->arch.assigned_device_count); | |
10484 | } | |
10485 | EXPORT_SYMBOL_GPL(kvm_arch_start_assignment); | |
10486 | ||
10487 | void kvm_arch_end_assignment(struct kvm *kvm) | |
10488 | { | |
10489 | atomic_dec(&kvm->arch.assigned_device_count); | |
10490 | } | |
10491 | EXPORT_SYMBOL_GPL(kvm_arch_end_assignment); | |
10492 | ||
10493 | bool kvm_arch_has_assigned_device(struct kvm *kvm) | |
10494 | { | |
10495 | return atomic_read(&kvm->arch.assigned_device_count); | |
10496 | } | |
10497 | EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device); | |
10498 | ||
e0f0bbc5 AW |
10499 | void kvm_arch_register_noncoherent_dma(struct kvm *kvm) |
10500 | { | |
10501 | atomic_inc(&kvm->arch.noncoherent_dma_count); | |
10502 | } | |
10503 | EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma); | |
10504 | ||
10505 | void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm) | |
10506 | { | |
10507 | atomic_dec(&kvm->arch.noncoherent_dma_count); | |
10508 | } | |
10509 | EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma); | |
10510 | ||
10511 | bool kvm_arch_has_noncoherent_dma(struct kvm *kvm) | |
10512 | { | |
10513 | return atomic_read(&kvm->arch.noncoherent_dma_count); | |
10514 | } | |
10515 | EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma); | |
10516 | ||
14717e20 AW |
10517 | bool kvm_arch_has_irq_bypass(void) |
10518 | { | |
92735b1b | 10519 | return true; |
14717e20 AW |
10520 | } |
10521 | ||
87276880 FW |
10522 | int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons, |
10523 | struct irq_bypass_producer *prod) | |
10524 | { | |
10525 | struct kvm_kernel_irqfd *irqfd = | |
10526 | container_of(cons, struct kvm_kernel_irqfd, consumer); | |
10527 | ||
14717e20 | 10528 | irqfd->producer = prod; |
87276880 | 10529 | |
afaf0b2f | 10530 | return kvm_x86_ops.update_pi_irte(irqfd->kvm, |
14717e20 | 10531 | prod->irq, irqfd->gsi, 1); |
87276880 FW |
10532 | } |
10533 | ||
10534 | void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons, | |
10535 | struct irq_bypass_producer *prod) | |
10536 | { | |
10537 | int ret; | |
10538 | struct kvm_kernel_irqfd *irqfd = | |
10539 | container_of(cons, struct kvm_kernel_irqfd, consumer); | |
10540 | ||
87276880 FW |
10541 | WARN_ON(irqfd->producer != prod); |
10542 | irqfd->producer = NULL; | |
10543 | ||
10544 | /* | |
10545 | * When producer of consumer is unregistered, we change back to | |
10546 | * remapped mode, so we can re-use the current implementation | |
bb3541f1 | 10547 | * when the irq is masked/disabled or the consumer side (KVM |
87276880 FW |
10548 | * int this case doesn't want to receive the interrupts. |
10549 | */ | |
afaf0b2f | 10550 | ret = kvm_x86_ops.update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0); |
87276880 FW |
10551 | if (ret) |
10552 | printk(KERN_INFO "irq bypass consumer (token %p) unregistration" | |
10553 | " fails: %d\n", irqfd->consumer.token, ret); | |
10554 | } | |
10555 | ||
10556 | int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq, | |
10557 | uint32_t guest_irq, bool set) | |
10558 | { | |
afaf0b2f | 10559 | return kvm_x86_ops.update_pi_irte(kvm, host_irq, guest_irq, set); |
87276880 FW |
10560 | } |
10561 | ||
52004014 FW |
10562 | bool kvm_vector_hashing_enabled(void) |
10563 | { | |
10564 | return vector_hashing; | |
10565 | } | |
52004014 | 10566 | |
2d5ba19b MT |
10567 | bool kvm_arch_no_poll(struct kvm_vcpu *vcpu) |
10568 | { | |
10569 | return (vcpu->arch.msr_kvm_poll_control & 1) == 0; | |
10570 | } | |
10571 | EXPORT_SYMBOL_GPL(kvm_arch_no_poll); | |
10572 | ||
6441fa61 PB |
10573 | u64 kvm_spec_ctrl_valid_bits(struct kvm_vcpu *vcpu) |
10574 | { | |
10575 | uint64_t bits = SPEC_CTRL_IBRS | SPEC_CTRL_STIBP | SPEC_CTRL_SSBD; | |
10576 | ||
10577 | /* The STIBP bit doesn't fault even if it's not advertised */ | |
10578 | if (!guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL) && | |
10579 | !guest_cpuid_has(vcpu, X86_FEATURE_AMD_IBRS)) | |
10580 | bits &= ~(SPEC_CTRL_IBRS | SPEC_CTRL_STIBP); | |
10581 | if (!boot_cpu_has(X86_FEATURE_SPEC_CTRL) && | |
10582 | !boot_cpu_has(X86_FEATURE_AMD_IBRS)) | |
10583 | bits &= ~(SPEC_CTRL_IBRS | SPEC_CTRL_STIBP); | |
10584 | ||
10585 | if (!guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL_SSBD) && | |
10586 | !guest_cpuid_has(vcpu, X86_FEATURE_AMD_SSBD)) | |
10587 | bits &= ~SPEC_CTRL_SSBD; | |
10588 | if (!boot_cpu_has(X86_FEATURE_SPEC_CTRL_SSBD) && | |
10589 | !boot_cpu_has(X86_FEATURE_AMD_SSBD)) | |
10590 | bits &= ~SPEC_CTRL_SSBD; | |
10591 | ||
10592 | return bits; | |
10593 | } | |
10594 | EXPORT_SYMBOL_GPL(kvm_spec_ctrl_valid_bits); | |
2d5ba19b | 10595 | |
229456fc | 10596 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit); |
931c33b1 | 10597 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio); |
229456fc MT |
10598 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq); |
10599 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault); | |
10600 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr); | |
10601 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr); | |
0ac406de | 10602 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun); |
d8cabddf | 10603 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit); |
17897f36 | 10604 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject); |
236649de | 10605 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit); |
5497b955 | 10606 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter_failed); |
ec1ff790 | 10607 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga); |
532a46b9 | 10608 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit); |
2e554e8d | 10609 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts); |
489223ed | 10610 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset); |
4f75bcc3 | 10611 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window_update); |
843e4330 | 10612 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full); |
efc64404 | 10613 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update); |
18f40c53 SS |
10614 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access); |
10615 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi); | |
ab56f8e6 | 10616 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_ga_log); |
24bbf74c | 10617 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_apicv_update_request); |