blob: 699f8726539a2044616b82bd9cf792a8617f8001 [file] [log] [blame]
Carsten Otte043405e2007-10-10 17:16:19 +02001/*
2 * Kernel-based Virtual Machine driver for Linux
3 *
4 * derived from drivers/kvm/kvm_main.c
5 *
6 * Copyright (C) 2006 Qumranet, Inc.
Ben-Ami Yassour4d5c5d02008-07-28 19:26:26 +03007 * Copyright (C) 2008 Qumranet, Inc.
8 * Copyright IBM Corporation, 2008
Nicolas Kaiser9611c182010-10-06 14:23:22 +02009 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
Carsten Otte043405e2007-10-10 17:16:19 +020010 *
11 * Authors:
12 * Avi Kivity <avi@qumranet.com>
13 * Yaniv Kamay <yaniv@qumranet.com>
Ben-Ami Yassour4d5c5d02008-07-28 19:26:26 +030014 * Amit Shah <amit.shah@qumranet.com>
15 * Ben-Ami Yassour <benami@il.ibm.com>
Carsten Otte043405e2007-10-10 17:16:19 +020016 *
17 * This work is licensed under the terms of the GNU GPL, version 2. See
18 * the COPYING file in the top-level directory.
19 *
20 */
21
Avi Kivityedf88412007-12-16 11:02:48 +020022#include <linux/kvm_host.h>
Carsten Otte313a3dc2007-10-11 19:16:52 +020023#include "irq.h"
Zhang Xiantao1d737c82007-12-14 09:35:10 +080024#include "mmu.h"
Sheng Yang78376992008-01-28 05:10:22 +080025#include "i8254.h"
Izik Eidus37817f22008-03-24 23:14:53 +020026#include "tss.h"
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -030027#include "kvm_cache_regs.h"
Avi Kivity26eef702008-07-03 14:59:22 +030028#include "x86.h"
Avi Kivity00b27a32011-11-23 16:30:32 +020029#include "cpuid.h"
Paolo Bonzinic9eab582014-11-24 15:27:17 +010030#include "assigned-dev.h"
Wei Huang474a5bb2015-06-19 13:54:23 +020031#include "pmu.h"
Andrey Smetanine83d5882015-07-03 15:01:34 +030032#include "hyperv.h"
Carsten Otte313a3dc2007-10-11 19:16:52 +020033
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -020034#include <linux/clocksource.h>
Ben-Ami Yassour4d5c5d02008-07-28 19:26:26 +030035#include <linux/interrupt.h>
Carsten Otte313a3dc2007-10-11 19:16:52 +020036#include <linux/kvm.h>
37#include <linux/fs.h>
38#include <linux/vmalloc.h>
Paul Gortmaker1767e932016-07-13 20:19:00 -040039#include <linux/export.h>
40#include <linux/moduleparam.h>
Zhang Xiantao0de10342007-11-20 16:25:04 +080041#include <linux/mman.h>
Marcelo Tosatti2bacc552007-12-12 10:46:12 -050042#include <linux/highmem.h>
Joerg Roedel19de40a2008-12-03 14:43:34 +010043#include <linux/iommu.h>
Ben-Ami Yassour62c476c2008-09-14 03:48:28 +030044#include <linux/intel-iommu.h>
Gerd Hoffmannc8076602009-02-04 17:52:04 +010045#include <linux/cpufreq.h>
Avi Kivity18863bd2009-09-07 11:12:18 +030046#include <linux/user-return-notifier.h>
Marcelo Tosattia983fb22009-12-23 14:35:23 -020047#include <linux/srcu.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090048#include <linux/slab.h>
Zhang, Yanminff9d07a2010-04-19 13:32:45 +080049#include <linux/perf_event.h>
Lai Jiangshan7bee3422010-06-02 17:06:03 +080050#include <linux/uaccess.h>
Gleb Natapovaf585b92010-10-14 11:22:46 +020051#include <linux/hash.h>
Joerg Roedela1b60c12011-09-06 18:46:34 +020052#include <linux/pci.h>
Marcelo Tosatti16e8d742012-11-27 23:29:00 -020053#include <linux/timekeeper_internal.h>
54#include <linux/pvclock_gtod.h>
Feng Wu87276882015-09-18 22:29:40 +080055#include <linux/kvm_irqfd.h>
56#include <linux/irqbypass.h>
Avi Kivityaec51dc2009-07-01 16:01:02 +030057#include <trace/events/kvm.h>
Xiao Guangrong2ed152a2010-03-10 19:00:43 +080058
Frederic Weisbecker24f1e32c2009-09-09 19:22:48 +020059#include <asm/debugreg.h>
Zhang Xiantaod825ed02007-11-14 20:08:51 +080060#include <asm/msr.h>
Avi Kivitya5f61302008-02-20 17:57:21 +020061#include <asm/desc.h>
Huang Ying890ca9a2009-05-11 16:48:15 +080062#include <asm/mce.h>
Ingo Molnarf89e32e2015-04-22 10:58:10 +020063#include <linux/kernel_stat.h>
Ingo Molnar78f7f1e2015-04-24 02:54:44 +020064#include <asm/fpu/internal.h> /* Ugh! */
Zachary Amsden1d5f0662010-08-19 22:07:30 -100065#include <asm/pvclock.h>
Avi Kivity217fc9c2010-08-26 13:38:03 +030066#include <asm/div64.h>
Feng Wuefc64402015-09-18 22:29:51 +080067#include <asm/irq_remapping.h>
Carsten Otte043405e2007-10-10 17:16:19 +020068
Dave Hansend1898b72016-06-01 10:42:20 -070069#define CREATE_TRACE_POINTS
70#include "trace.h"
71
Carsten Otte313a3dc2007-10-11 19:16:52 +020072#define MAX_IO_MSRS 256
Huang Ying890ca9a2009-05-11 16:48:15 +080073#define KVM_MAX_MCE_BANKS 32
Ashok Rajc45dcc72016-06-22 14:59:56 +080074u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
75EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
Huang Ying890ca9a2009-05-11 16:48:15 +080076
Avi Kivity0f65dd72011-04-20 13:37:53 +030077#define emul_to_vcpu(ctxt) \
78 container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
79
Joerg Roedel50a37eb2008-01-31 14:57:38 +010080/* EFER defaults:
81 * - enable syscall per default because its emulated by KVM
82 * - enable LME and LMA per default on 64 bit KVM
83 */
84#ifdef CONFIG_X86_64
Lai Jiangshan1260edbe2011-02-21 11:51:35 +080085static
86u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
Joerg Roedel50a37eb2008-01-31 14:57:38 +010087#else
Lai Jiangshan1260edbe2011-02-21 11:51:35 +080088static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
Joerg Roedel50a37eb2008-01-31 14:57:38 +010089#endif
Carsten Otte313a3dc2007-10-11 19:16:52 +020090
Avi Kivityba1389b2007-11-18 16:24:12 +020091#define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
92#define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
Hollis Blanchard417bc302007-10-31 17:24:23 -050093
Radim Krčmářc5192652016-07-12 22:09:28 +020094#define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
95 KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
Radim Krčmář371313132016-07-12 22:09:27 +020096
Gleb Natapovcb142eb2009-08-09 15:17:40 +030097static void update_cr8_intercept(struct kvm_vcpu *vcpu);
Avi Kivity7460fb4a2011-09-20 13:43:14 +030098static void process_nmi(struct kvm_vcpu *vcpu);
Paolo Bonziniee2cd4b2016-06-01 22:26:01 +020099static void enter_smm(struct kvm_vcpu *vcpu);
Paolo Bonzini6addfc42014-03-27 11:29:28 +0100100static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
Avi Kivity674eea02008-02-11 18:37:23 +0200101
Paolo Bonzini893590c2015-11-06 11:46:24 +0100102struct kvm_x86_ops *kvm_x86_ops __read_mostly;
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -0300103EXPORT_SYMBOL_GPL(kvm_x86_ops);
Zhang Xiantao97896d02007-11-14 20:09:30 +0800104
Paolo Bonzini893590c2015-11-06 11:46:24 +0100105static bool __read_mostly ignore_msrs = 0;
Rusty Russell476bc002012-01-13 09:32:18 +1030106module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
Andre Przywaraed85c062009-06-25 12:36:49 +0200107
Marcelo Tosatti9ed96e82014-01-06 12:00:02 -0200108unsigned int min_timer_period_us = 500;
109module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
110
Marcelo Tosatti630994b2015-05-12 22:42:04 -0300111static bool __read_mostly kvmclock_periodic_sync = true;
112module_param(kvmclock_periodic_sync, bool, S_IRUGO);
113
Paolo Bonzini893590c2015-11-06 11:46:24 +0100114bool __read_mostly kvm_has_tsc_control;
Joerg Roedel92a1f122011-03-25 09:44:51 +0100115EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
Paolo Bonzini893590c2015-11-06 11:46:24 +0100116u32 __read_mostly kvm_max_guest_tsc_khz;
Joerg Roedel92a1f122011-03-25 09:44:51 +0100117EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
Haozhong Zhangbc9b9612015-10-20 15:39:01 +0800118u8 __read_mostly kvm_tsc_scaling_ratio_frac_bits;
119EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
120u64 __read_mostly kvm_max_tsc_scaling_ratio;
121EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
Yunhong Jiang64672c92016-06-13 14:19:59 -0700122u64 __read_mostly kvm_default_tsc_scaling_ratio;
123EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
Joerg Roedel92a1f122011-03-25 09:44:51 +0100124
Zachary Amsdencc578282012-02-03 15:43:50 -0200125/* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
Paolo Bonzini893590c2015-11-06 11:46:24 +0100126static u32 __read_mostly tsc_tolerance_ppm = 250;
Zachary Amsdencc578282012-02-03 15:43:50 -0200127module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
128
Marcelo Tosattid0659d92014-12-16 09:08:15 -0500129/* lapic timer advance (tscdeadline mode only) in nanoseconds */
Paolo Bonzini893590c2015-11-06 11:46:24 +0100130unsigned int __read_mostly lapic_timer_advance_ns = 0;
Marcelo Tosattid0659d92014-12-16 09:08:15 -0500131module_param(lapic_timer_advance_ns, uint, S_IRUGO | S_IWUSR);
132
Feng Wu520040142016-01-25 16:53:33 +0800133static bool __read_mostly vector_hashing = true;
134module_param(vector_hashing, bool, S_IRUGO);
135
Paolo Bonzini893590c2015-11-06 11:46:24 +0100136static bool __read_mostly backwards_tsc_observed = false;
Marcelo Tosatti16a96022014-05-14 12:43:24 -0300137
Avi Kivity18863bd2009-09-07 11:12:18 +0300138#define KVM_NR_SHARED_MSRS 16
139
140struct kvm_shared_msrs_global {
141 int nr;
Sheng Yang2bf78fa2009-12-18 16:48:44 +0800142 u32 msrs[KVM_NR_SHARED_MSRS];
Avi Kivity18863bd2009-09-07 11:12:18 +0300143};
144
145struct kvm_shared_msrs {
146 struct user_return_notifier urn;
147 bool registered;
Sheng Yang2bf78fa2009-12-18 16:48:44 +0800148 struct kvm_shared_msr_values {
149 u64 host;
150 u64 curr;
151 } values[KVM_NR_SHARED_MSRS];
Avi Kivity18863bd2009-09-07 11:12:18 +0300152};
153
154static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
Marcelo Tosatti013f6a52013-01-03 11:41:39 -0200155static struct kvm_shared_msrs __percpu *shared_msrs;
Avi Kivity18863bd2009-09-07 11:12:18 +0300156
Hollis Blanchard417bc302007-10-31 17:24:23 -0500157struct kvm_stats_debugfs_item debugfs_entries[] = {
Avi Kivityba1389b2007-11-18 16:24:12 +0200158 { "pf_fixed", VCPU_STAT(pf_fixed) },
159 { "pf_guest", VCPU_STAT(pf_guest) },
160 { "tlb_flush", VCPU_STAT(tlb_flush) },
161 { "invlpg", VCPU_STAT(invlpg) },
162 { "exits", VCPU_STAT(exits) },
163 { "io_exits", VCPU_STAT(io_exits) },
164 { "mmio_exits", VCPU_STAT(mmio_exits) },
165 { "signal_exits", VCPU_STAT(signal_exits) },
166 { "irq_window", VCPU_STAT(irq_window_exits) },
Sheng Yangf08864b2008-05-15 18:23:25 +0800167 { "nmi_window", VCPU_STAT(nmi_window_exits) },
Avi Kivityba1389b2007-11-18 16:24:12 +0200168 { "halt_exits", VCPU_STAT(halt_exits) },
Paolo Bonzinif7819512015-02-04 18:20:58 +0100169 { "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
Paolo Bonzini62bea5b2015-09-15 18:27:57 +0200170 { "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
Christian Borntraeger3491caf2016-05-13 12:16:35 +0200171 { "halt_poll_invalid", VCPU_STAT(halt_poll_invalid) },
Avi Kivityba1389b2007-11-18 16:24:12 +0200172 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
Amit Shahf11c3a82008-02-21 01:00:30 +0530173 { "hypercalls", VCPU_STAT(hypercalls) },
Avi Kivityba1389b2007-11-18 16:24:12 +0200174 { "request_irq", VCPU_STAT(request_irq_exits) },
175 { "irq_exits", VCPU_STAT(irq_exits) },
176 { "host_state_reload", VCPU_STAT(host_state_reload) },
177 { "efer_reload", VCPU_STAT(efer_reload) },
178 { "fpu_reload", VCPU_STAT(fpu_reload) },
179 { "insn_emulation", VCPU_STAT(insn_emulation) },
180 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
Avi Kivityfa89a812008-09-01 15:57:51 +0300181 { "irq_injections", VCPU_STAT(irq_injections) },
Jan Kiszkac4abb7c2008-09-26 09:30:55 +0200182 { "nmi_injections", VCPU_STAT(nmi_injections) },
Avi Kivity4cee5762007-11-18 16:37:07 +0200183 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
184 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
185 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
186 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
187 { "mmu_flooded", VM_STAT(mmu_flooded) },
188 { "mmu_recycled", VM_STAT(mmu_recycled) },
Avi Kivitydfc5aa02007-12-18 19:47:18 +0200189 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
Marcelo Tosatti4731d4c2008-09-23 13:18:39 -0300190 { "mmu_unsync", VM_STAT(mmu_unsync) },
Avi Kivity0f74a242007-11-20 23:01:14 +0200191 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
Marcelo Tosatti05da4552008-02-23 11:44:30 -0300192 { "largepages", VM_STAT(lpages) },
Hollis Blanchard417bc302007-10-31 17:24:23 -0500193 { NULL }
194};
195
Dexuan Cui2acf9232010-06-10 11:27:12 +0800196u64 __read_mostly host_xcr0;
197
Jan Kiszkab6785de2012-09-20 07:43:17 +0200198static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
Avi Kivityd6aa1002011-04-20 15:47:13 +0300199
Gleb Natapovaf585b92010-10-14 11:22:46 +0200200static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
201{
202 int i;
203 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
204 vcpu->arch.apf.gfns[i] = ~0;
205}
206
Avi Kivity18863bd2009-09-07 11:12:18 +0300207static void kvm_on_user_return(struct user_return_notifier *urn)
208{
209 unsigned slot;
Avi Kivity18863bd2009-09-07 11:12:18 +0300210 struct kvm_shared_msrs *locals
211 = container_of(urn, struct kvm_shared_msrs, urn);
Sheng Yang2bf78fa2009-12-18 16:48:44 +0800212 struct kvm_shared_msr_values *values;
Avi Kivity18863bd2009-09-07 11:12:18 +0300213
214 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
Sheng Yang2bf78fa2009-12-18 16:48:44 +0800215 values = &locals->values[slot];
216 if (values->host != values->curr) {
217 wrmsrl(shared_msrs_global.msrs[slot], values->host);
218 values->curr = values->host;
Avi Kivity18863bd2009-09-07 11:12:18 +0300219 }
220 }
221 locals->registered = false;
222 user_return_notifier_unregister(urn);
223}
224
Sheng Yang2bf78fa2009-12-18 16:48:44 +0800225static void shared_msr_update(unsigned slot, u32 msr)
Avi Kivity18863bd2009-09-07 11:12:18 +0300226{
Avi Kivity18863bd2009-09-07 11:12:18 +0300227 u64 value;
Marcelo Tosatti013f6a52013-01-03 11:41:39 -0200228 unsigned int cpu = smp_processor_id();
229 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
Avi Kivity18863bd2009-09-07 11:12:18 +0300230
Sheng Yang2bf78fa2009-12-18 16:48:44 +0800231 /* only read, and nobody should modify it at this time,
232 * so don't need lock */
233 if (slot >= shared_msrs_global.nr) {
234 printk(KERN_ERR "kvm: invalid MSR slot!");
235 return;
236 }
237 rdmsrl_safe(msr, &value);
238 smsr->values[slot].host = value;
239 smsr->values[slot].curr = value;
240}
241
242void kvm_define_shared_msr(unsigned slot, u32 msr)
243{
Nadav Amit0123be42014-07-24 15:06:56 +0300244 BUG_ON(slot >= KVM_NR_SHARED_MSRS);
Paolo Bonzinic847fe82015-07-29 11:06:34 +0200245 shared_msrs_global.msrs[slot] = msr;
Avi Kivity18863bd2009-09-07 11:12:18 +0300246 if (slot >= shared_msrs_global.nr)
247 shared_msrs_global.nr = slot + 1;
Avi Kivity18863bd2009-09-07 11:12:18 +0300248}
249EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
250
251static void kvm_shared_msr_cpu_online(void)
252{
253 unsigned i;
Avi Kivity18863bd2009-09-07 11:12:18 +0300254
255 for (i = 0; i < shared_msrs_global.nr; ++i)
Sheng Yang2bf78fa2009-12-18 16:48:44 +0800256 shared_msr_update(i, shared_msrs_global.msrs[i]);
Avi Kivity18863bd2009-09-07 11:12:18 +0300257}
258
Andy Honig8b3c3102014-08-27 11:16:44 -0700259int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
Avi Kivity18863bd2009-09-07 11:12:18 +0300260{
Marcelo Tosatti013f6a52013-01-03 11:41:39 -0200261 unsigned int cpu = smp_processor_id();
262 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
Andy Honig8b3c3102014-08-27 11:16:44 -0700263 int err;
Avi Kivity18863bd2009-09-07 11:12:18 +0300264
Sheng Yang2bf78fa2009-12-18 16:48:44 +0800265 if (((value ^ smsr->values[slot].curr) & mask) == 0)
Andy Honig8b3c3102014-08-27 11:16:44 -0700266 return 0;
Sheng Yang2bf78fa2009-12-18 16:48:44 +0800267 smsr->values[slot].curr = value;
Andy Honig8b3c3102014-08-27 11:16:44 -0700268 err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
269 if (err)
270 return 1;
271
Avi Kivity18863bd2009-09-07 11:12:18 +0300272 if (!smsr->registered) {
273 smsr->urn.on_user_return = kvm_on_user_return;
274 user_return_notifier_register(&smsr->urn);
275 smsr->registered = true;
276 }
Andy Honig8b3c3102014-08-27 11:16:44 -0700277 return 0;
Avi Kivity18863bd2009-09-07 11:12:18 +0300278}
279EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
280
Radim Krčmář13a34e02014-08-28 15:13:03 +0200281static void drop_user_return_notifiers(void)
Avi Kivity3548bab2009-11-28 14:18:47 +0200282{
Marcelo Tosatti013f6a52013-01-03 11:41:39 -0200283 unsigned int cpu = smp_processor_id();
284 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
Avi Kivity3548bab2009-11-28 14:18:47 +0200285
286 if (smsr->registered)
287 kvm_on_user_return(&smsr->urn);
288}
289
Carsten Otte6866b832007-10-29 16:09:10 +0100290u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
291{
Gleb Natapov8a5a87d2012-08-05 15:58:26 +0300292 return vcpu->arch.apic_base;
Carsten Otte6866b832007-10-29 16:09:10 +0100293}
294EXPORT_SYMBOL_GPL(kvm_get_apic_base);
295
Jan Kiszka58cb6282014-01-24 16:48:44 +0100296int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
Carsten Otte6866b832007-10-29 16:09:10 +0100297{
Jan Kiszka58cb6282014-01-24 16:48:44 +0100298 u64 old_state = vcpu->arch.apic_base &
299 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
300 u64 new_state = msr_info->data &
301 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
302 u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) |
303 0x2ff | (guest_cpuid_has_x2apic(vcpu) ? 0 : X2APIC_ENABLE);
304
305 if (!msr_info->host_initiated &&
306 ((msr_info->data & reserved_bits) != 0 ||
307 new_state == X2APIC_ENABLE ||
308 (new_state == MSR_IA32_APICBASE_ENABLE &&
309 old_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE)) ||
310 (new_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE) &&
311 old_state == 0)))
312 return 1;
313
314 kvm_lapic_set_base(vcpu, msr_info->data);
315 return 0;
Carsten Otte6866b832007-10-29 16:09:10 +0100316}
317EXPORT_SYMBOL_GPL(kvm_set_apic_base);
318
Andi Kleen2605fc22014-05-02 00:44:37 +0200319asmlinkage __visible void kvm_spurious_fault(void)
Geoff Levande3ba45b2013-04-05 19:20:30 +0000320{
321 /* Fault while not rebooting. We want the trace. */
322 BUG();
323}
324EXPORT_SYMBOL_GPL(kvm_spurious_fault);
325
Eddie Dong3fd28fc2009-11-19 17:54:07 +0200326#define EXCPT_BENIGN 0
327#define EXCPT_CONTRIBUTORY 1
328#define EXCPT_PF 2
329
330static int exception_class(int vector)
331{
332 switch (vector) {
333 case PF_VECTOR:
334 return EXCPT_PF;
335 case DE_VECTOR:
336 case TS_VECTOR:
337 case NP_VECTOR:
338 case SS_VECTOR:
339 case GP_VECTOR:
340 return EXCPT_CONTRIBUTORY;
341 default:
342 break;
343 }
344 return EXCPT_BENIGN;
345}
346
Nadav Amitd6e8c852014-07-24 14:51:24 +0300347#define EXCPT_FAULT 0
348#define EXCPT_TRAP 1
349#define EXCPT_ABORT 2
350#define EXCPT_INTERRUPT 3
351
352static int exception_type(int vector)
353{
354 unsigned int mask;
355
356 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
357 return EXCPT_INTERRUPT;
358
359 mask = 1 << vector;
360
361 /* #DB is trap, as instruction watchpoints are handled elsewhere */
362 if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
363 return EXCPT_TRAP;
364
365 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
366 return EXCPT_ABORT;
367
368 /* Reserved exceptions will result in fault */
369 return EXCPT_FAULT;
370}
371
Eddie Dong3fd28fc2009-11-19 17:54:07 +0200372static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
Joerg Roedelce7ddec2010-04-22 12:33:13 +0200373 unsigned nr, bool has_error, u32 error_code,
374 bool reinject)
Eddie Dong3fd28fc2009-11-19 17:54:07 +0200375{
376 u32 prev_nr;
377 int class1, class2;
378
Avi Kivity3842d132010-07-27 12:30:24 +0300379 kvm_make_request(KVM_REQ_EVENT, vcpu);
380
Eddie Dong3fd28fc2009-11-19 17:54:07 +0200381 if (!vcpu->arch.exception.pending) {
382 queue:
Nadav Amit3ffb2462014-11-02 11:54:42 +0200383 if (has_error && !is_protmode(vcpu))
384 has_error = false;
Eddie Dong3fd28fc2009-11-19 17:54:07 +0200385 vcpu->arch.exception.pending = true;
386 vcpu->arch.exception.has_error_code = has_error;
387 vcpu->arch.exception.nr = nr;
388 vcpu->arch.exception.error_code = error_code;
Joerg Roedel3f0fd292010-05-05 16:04:41 +0200389 vcpu->arch.exception.reinject = reinject;
Eddie Dong3fd28fc2009-11-19 17:54:07 +0200390 return;
391 }
392
393 /* to check exception */
394 prev_nr = vcpu->arch.exception.nr;
395 if (prev_nr == DF_VECTOR) {
396 /* triple fault -> shutdown */
Avi Kivitya8eeb042010-05-10 12:34:53 +0300397 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
Eddie Dong3fd28fc2009-11-19 17:54:07 +0200398 return;
399 }
400 class1 = exception_class(prev_nr);
401 class2 = exception_class(nr);
402 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
403 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
404 /* generate double fault per SDM Table 5-5 */
405 vcpu->arch.exception.pending = true;
406 vcpu->arch.exception.has_error_code = true;
407 vcpu->arch.exception.nr = DF_VECTOR;
408 vcpu->arch.exception.error_code = 0;
409 } else
410 /* replace previous exception with a new one in a hope
411 that instruction re-execution will regenerate lost
412 exception */
413 goto queue;
414}
415
Avi Kivity298101d2007-11-25 13:41:11 +0200416void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
417{
Joerg Roedelce7ddec2010-04-22 12:33:13 +0200418 kvm_multiple_exception(vcpu, nr, false, 0, false);
Avi Kivity298101d2007-11-25 13:41:11 +0200419}
420EXPORT_SYMBOL_GPL(kvm_queue_exception);
421
Joerg Roedelce7ddec2010-04-22 12:33:13 +0200422void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
423{
424 kvm_multiple_exception(vcpu, nr, false, 0, true);
425}
426EXPORT_SYMBOL_GPL(kvm_requeue_exception);
427
Andre Przywaradb8fcef2010-12-21 11:12:01 +0100428void kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
Avi Kivityc3c91fe2007-11-25 14:04:58 +0200429{
Andre Przywaradb8fcef2010-12-21 11:12:01 +0100430 if (err)
431 kvm_inject_gp(vcpu, 0);
432 else
433 kvm_x86_ops->skip_emulated_instruction(vcpu);
434}
435EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
Joerg Roedel8df25a32010-09-10 17:30:46 +0200436
Avi Kivity6389ee92010-11-29 16:12:30 +0200437void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
Avi Kivityc3c91fe2007-11-25 14:04:58 +0200438{
Avi Kivityc3c91fe2007-11-25 14:04:58 +0200439 ++vcpu->stat.pf_guest;
Avi Kivity6389ee92010-11-29 16:12:30 +0200440 vcpu->arch.cr2 = fault->address;
441 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
Avi Kivityc3c91fe2007-11-25 14:04:58 +0200442}
Nadav Har'El27d6c862011-05-25 23:06:59 +0300443EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
Avi Kivityc3c91fe2007-11-25 14:04:58 +0200444
Paolo Bonzinief54bcf2014-09-04 19:46:15 +0200445static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
Joerg Roedeld4f8cf62010-09-10 17:30:55 +0200446{
Avi Kivity6389ee92010-11-29 16:12:30 +0200447 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
448 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
Joerg Roedeld4f8cf62010-09-10 17:30:55 +0200449 else
Avi Kivity6389ee92010-11-29 16:12:30 +0200450 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
Paolo Bonzinief54bcf2014-09-04 19:46:15 +0200451
452 return fault->nested_page_fault;
Joerg Roedeld4f8cf62010-09-10 17:30:55 +0200453}
454
Sheng Yang3419ffc2008-05-15 09:52:48 +0800455void kvm_inject_nmi(struct kvm_vcpu *vcpu)
456{
Avi Kivity7460fb4a2011-09-20 13:43:14 +0300457 atomic_inc(&vcpu->arch.nmi_queued);
458 kvm_make_request(KVM_REQ_NMI, vcpu);
Sheng Yang3419ffc2008-05-15 09:52:48 +0800459}
460EXPORT_SYMBOL_GPL(kvm_inject_nmi);
461
Avi Kivity298101d2007-11-25 13:41:11 +0200462void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
463{
Joerg Roedelce7ddec2010-04-22 12:33:13 +0200464 kvm_multiple_exception(vcpu, nr, true, error_code, false);
Avi Kivity298101d2007-11-25 13:41:11 +0200465}
466EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
467
Joerg Roedelce7ddec2010-04-22 12:33:13 +0200468void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
469{
470 kvm_multiple_exception(vcpu, nr, true, error_code, true);
471}
472EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
473
Carsten Ottea03490e2007-10-29 16:09:35 +0100474/*
Avi Kivity0a79b002009-09-01 12:03:25 +0300475 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
476 * a #GP and return false.
477 */
478bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
Carsten Otte043405e2007-10-10 17:16:19 +0200479{
Avi Kivity0a79b002009-09-01 12:03:25 +0300480 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
481 return true;
482 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
483 return false;
Carsten Ottea03490e2007-10-29 16:09:35 +0100484}
Avi Kivity0a79b002009-09-01 12:03:25 +0300485EXPORT_SYMBOL_GPL(kvm_require_cpl);
Carsten Ottea03490e2007-10-29 16:09:35 +0100486
Nadav Amit16f8a6f2014-10-03 01:10:05 +0300487bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
488{
489 if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
490 return true;
491
492 kvm_queue_exception(vcpu, UD_VECTOR);
493 return false;
494}
495EXPORT_SYMBOL_GPL(kvm_require_dr);
496
Carsten Ottea03490e2007-10-29 16:09:35 +0100497/*
Joerg Roedelec92fe42010-09-10 17:30:51 +0200498 * This function will be used to read from the physical memory of the currently
Paolo Bonzini54bf36a2015-04-08 15:39:23 +0200499 * running guest. The difference to kvm_vcpu_read_guest_page is that this function
Joerg Roedelec92fe42010-09-10 17:30:51 +0200500 * can read from guest physical or from the guest's guest physical memory.
501 */
502int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
503 gfn_t ngfn, void *data, int offset, int len,
504 u32 access)
505{
Paolo Bonzini54987b72014-09-02 13:23:06 +0200506 struct x86_exception exception;
Joerg Roedelec92fe42010-09-10 17:30:51 +0200507 gfn_t real_gfn;
508 gpa_t ngpa;
509
510 ngpa = gfn_to_gpa(ngfn);
Paolo Bonzini54987b72014-09-02 13:23:06 +0200511 real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
Joerg Roedelec92fe42010-09-10 17:30:51 +0200512 if (real_gfn == UNMAPPED_GVA)
513 return -EFAULT;
514
515 real_gfn = gpa_to_gfn(real_gfn);
516
Paolo Bonzini54bf36a2015-04-08 15:39:23 +0200517 return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
Joerg Roedelec92fe42010-09-10 17:30:51 +0200518}
519EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
520
Fengguang Wu69b00492015-01-19 22:33:39 +0800521static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
Joerg Roedel3d06b8b2010-09-10 17:30:53 +0200522 void *data, int offset, int len, u32 access)
523{
524 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
525 data, offset, len, access);
526}
527
Joerg Roedelec92fe42010-09-10 17:30:51 +0200528/*
Carsten Ottea03490e2007-10-29 16:09:35 +0100529 * Load the pae pdptrs. Return true is they are all valid.
530 */
Joerg Roedelff03a072010-09-10 17:30:57 +0200531int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
Carsten Ottea03490e2007-10-29 16:09:35 +0100532{
533 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
534 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
535 int i;
536 int ret;
Joerg Roedelff03a072010-09-10 17:30:57 +0200537 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
Carsten Ottea03490e2007-10-29 16:09:35 +0100538
Joerg Roedelff03a072010-09-10 17:30:57 +0200539 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
540 offset * sizeof(u64), sizeof(pdpte),
541 PFERR_USER_MASK|PFERR_WRITE_MASK);
Carsten Ottea03490e2007-10-29 16:09:35 +0100542 if (ret < 0) {
543 ret = 0;
544 goto out;
545 }
546 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
Bandan Das812f30b2016-07-12 18:18:50 -0400547 if ((pdpte[i] & PT_PRESENT_MASK) &&
Xiao Guangronga0a64f52015-08-05 12:04:21 +0800548 (pdpte[i] &
549 vcpu->arch.mmu.guest_rsvd_check.rsvd_bits_mask[0][2])) {
Carsten Ottea03490e2007-10-29 16:09:35 +0100550 ret = 0;
551 goto out;
552 }
553 }
554 ret = 1;
555
Joerg Roedelff03a072010-09-10 17:30:57 +0200556 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
Avi Kivity6de4f3a2009-05-31 22:58:47 +0300557 __set_bit(VCPU_EXREG_PDPTR,
558 (unsigned long *)&vcpu->arch.regs_avail);
559 __set_bit(VCPU_EXREG_PDPTR,
560 (unsigned long *)&vcpu->arch.regs_dirty);
Carsten Ottea03490e2007-10-29 16:09:35 +0100561out:
Carsten Ottea03490e2007-10-29 16:09:35 +0100562
563 return ret;
564}
Joerg Roedelcc4b6872008-02-07 13:47:43 +0100565EXPORT_SYMBOL_GPL(load_pdptrs);
Carsten Ottea03490e2007-10-29 16:09:35 +0100566
Avi Kivityd835dfe2007-11-21 02:57:59 +0200567static bool pdptrs_changed(struct kvm_vcpu *vcpu)
568{
Joerg Roedelff03a072010-09-10 17:30:57 +0200569 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
Avi Kivityd835dfe2007-11-21 02:57:59 +0200570 bool changed = true;
Joerg Roedel3d06b8b2010-09-10 17:30:53 +0200571 int offset;
572 gfn_t gfn;
Avi Kivityd835dfe2007-11-21 02:57:59 +0200573 int r;
574
575 if (is_long_mode(vcpu) || !is_pae(vcpu))
576 return false;
577
Avi Kivity6de4f3a2009-05-31 22:58:47 +0300578 if (!test_bit(VCPU_EXREG_PDPTR,
579 (unsigned long *)&vcpu->arch.regs_avail))
580 return true;
581
Avi Kivity9f8fe502010-12-05 17:30:00 +0200582 gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT;
583 offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1);
Joerg Roedel3d06b8b2010-09-10 17:30:53 +0200584 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
585 PFERR_USER_MASK | PFERR_WRITE_MASK);
Avi Kivityd835dfe2007-11-21 02:57:59 +0200586 if (r < 0)
587 goto out;
Joerg Roedelff03a072010-09-10 17:30:57 +0200588 changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
Avi Kivityd835dfe2007-11-21 02:57:59 +0200589out:
Avi Kivityd835dfe2007-11-21 02:57:59 +0200590
591 return changed;
592}
593
Avi Kivity49a9b072010-06-10 17:02:14 +0300594int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
Carsten Ottea03490e2007-10-29 16:09:35 +0100595{
Sheng Yangaad82702010-05-12 16:40:42 +0800596 unsigned long old_cr0 = kvm_read_cr0(vcpu);
Xiao Guangrongd81135a2015-05-13 14:42:28 +0800597 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
Sheng Yangaad82702010-05-12 16:40:42 +0800598
Avi Kivityf9a48e62010-01-06 19:10:22 +0200599 cr0 |= X86_CR0_ET;
600
Gleb Natapovab344822010-01-21 15:28:46 +0200601#ifdef CONFIG_X86_64
Gleb Natapov0f122442010-04-28 19:15:31 +0300602 if (cr0 & 0xffffffff00000000UL)
603 return 1;
Gleb Natapovab344822010-01-21 15:28:46 +0200604#endif
605
606 cr0 &= ~CR0_RESERVED_BITS;
Carsten Ottea03490e2007-10-29 16:09:35 +0100607
Gleb Natapov0f122442010-04-28 19:15:31 +0300608 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
609 return 1;
Carsten Ottea03490e2007-10-29 16:09:35 +0100610
Gleb Natapov0f122442010-04-28 19:15:31 +0300611 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
612 return 1;
Carsten Ottea03490e2007-10-29 16:09:35 +0100613
614 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
615#ifdef CONFIG_X86_64
Avi Kivityf6801df2010-01-21 15:31:50 +0200616 if ((vcpu->arch.efer & EFER_LME)) {
Carsten Ottea03490e2007-10-29 16:09:35 +0100617 int cs_db, cs_l;
618
Gleb Natapov0f122442010-04-28 19:15:31 +0300619 if (!is_pae(vcpu))
620 return 1;
Carsten Ottea03490e2007-10-29 16:09:35 +0100621 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
Gleb Natapov0f122442010-04-28 19:15:31 +0300622 if (cs_l)
623 return 1;
Carsten Ottea03490e2007-10-29 16:09:35 +0100624 } else
625#endif
Joerg Roedelff03a072010-09-10 17:30:57 +0200626 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
Avi Kivity9f8fe502010-12-05 17:30:00 +0200627 kvm_read_cr3(vcpu)))
Gleb Natapov0f122442010-04-28 19:15:31 +0300628 return 1;
Carsten Ottea03490e2007-10-29 16:09:35 +0100629 }
630
Mao, Junjiead756a12012-07-02 01:18:48 +0000631 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
632 return 1;
633
Carsten Ottea03490e2007-10-29 16:09:35 +0100634 kvm_x86_ops->set_cr0(vcpu, cr0);
Carsten Ottea03490e2007-10-29 16:09:35 +0100635
Lai Jiangshand170c412011-02-21 11:21:30 +0800636 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
Xiao Guangronge5f3f022010-11-12 14:47:01 +0800637 kvm_clear_async_pf_completion_queue(vcpu);
Lai Jiangshand170c412011-02-21 11:21:30 +0800638 kvm_async_pf_hash_reset(vcpu);
639 }
Xiao Guangronge5f3f022010-11-12 14:47:01 +0800640
Sheng Yangaad82702010-05-12 16:40:42 +0800641 if ((cr0 ^ old_cr0) & update_bits)
642 kvm_mmu_reset_context(vcpu);
Xiao Guangrongb18d5432015-06-15 16:55:21 +0800643
Laszlo Ersek879ae182015-11-04 12:54:41 +0100644 if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
645 kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
646 !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
Xiao Guangrongb18d5432015-06-15 16:55:21 +0800647 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
648
Gleb Natapov0f122442010-04-28 19:15:31 +0300649 return 0;
650}
Avi Kivity2d3ad1f2008-02-24 11:20:43 +0200651EXPORT_SYMBOL_GPL(kvm_set_cr0);
Carsten Ottea03490e2007-10-29 16:09:35 +0100652
Avi Kivity2d3ad1f2008-02-24 11:20:43 +0200653void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
Carsten Ottea03490e2007-10-29 16:09:35 +0100654{
Avi Kivity49a9b072010-06-10 17:02:14 +0300655 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
Carsten Ottea03490e2007-10-29 16:09:35 +0100656}
Avi Kivity2d3ad1f2008-02-24 11:20:43 +0200657EXPORT_SYMBOL_GPL(kvm_lmsw);
Carsten Ottea03490e2007-10-29 16:09:35 +0100658
Marcelo Tosatti42bdf992013-04-15 23:30:13 -0300659static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
660{
661 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
662 !vcpu->guest_xcr0_loaded) {
663 /* kvm_set_xcr() also depends on this */
664 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
665 vcpu->guest_xcr0_loaded = 1;
666 }
667}
668
669static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
670{
671 if (vcpu->guest_xcr0_loaded) {
672 if (vcpu->arch.xcr0 != host_xcr0)
673 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
674 vcpu->guest_xcr0_loaded = 0;
675 }
676}
677
Fengguang Wu69b00492015-01-19 22:33:39 +0800678static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
Dexuan Cui2acf9232010-06-10 11:27:12 +0800679{
Liu, Jinsong56c103e2014-02-21 17:39:02 +0000680 u64 xcr0 = xcr;
681 u64 old_xcr0 = vcpu->arch.xcr0;
Paolo Bonzini46c34cb2013-10-17 16:50:46 +0200682 u64 valid_bits;
Dexuan Cui2acf9232010-06-10 11:27:12 +0800683
684 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
685 if (index != XCR_XFEATURE_ENABLED_MASK)
686 return 1;
Dave Hansend91cab72015-09-02 16:31:26 -0700687 if (!(xcr0 & XFEATURE_MASK_FP))
Dexuan Cui2acf9232010-06-10 11:27:12 +0800688 return 1;
Dave Hansend91cab72015-09-02 16:31:26 -0700689 if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
Dexuan Cui2acf9232010-06-10 11:27:12 +0800690 return 1;
Paolo Bonzini46c34cb2013-10-17 16:50:46 +0200691
692 /*
693 * Do not allow the guest to set bits that we do not support
694 * saving. However, xcr0 bit 0 is always set, even if the
695 * emulated CPU does not support XSAVE (see fx_init).
696 */
Dave Hansend91cab72015-09-02 16:31:26 -0700697 valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
Paolo Bonzini46c34cb2013-10-17 16:50:46 +0200698 if (xcr0 & ~valid_bits)
Dexuan Cui2acf9232010-06-10 11:27:12 +0800699 return 1;
Paolo Bonzini46c34cb2013-10-17 16:50:46 +0200700
Dave Hansend91cab72015-09-02 16:31:26 -0700701 if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
702 (!(xcr0 & XFEATURE_MASK_BNDCSR)))
Liu, Jinsong390bd522014-02-24 10:58:09 +0000703 return 1;
704
Dave Hansend91cab72015-09-02 16:31:26 -0700705 if (xcr0 & XFEATURE_MASK_AVX512) {
706 if (!(xcr0 & XFEATURE_MASK_YMM))
Chao Peng612263b2014-10-22 17:35:24 +0800707 return 1;
Dave Hansend91cab72015-09-02 16:31:26 -0700708 if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
Chao Peng612263b2014-10-22 17:35:24 +0800709 return 1;
710 }
Dexuan Cui2acf9232010-06-10 11:27:12 +0800711 vcpu->arch.xcr0 = xcr0;
Liu, Jinsong56c103e2014-02-21 17:39:02 +0000712
Dave Hansend91cab72015-09-02 16:31:26 -0700713 if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
Liu, Jinsong56c103e2014-02-21 17:39:02 +0000714 kvm_update_cpuid(vcpu);
Dexuan Cui2acf9232010-06-10 11:27:12 +0800715 return 0;
716}
717
718int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
719{
Zhanghaoyu (A)764bcbc2013-06-14 07:36:13 +0000720 if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
721 __kvm_set_xcr(vcpu, index, xcr)) {
Dexuan Cui2acf9232010-06-10 11:27:12 +0800722 kvm_inject_gp(vcpu, 0);
723 return 1;
724 }
725 return 0;
726}
727EXPORT_SYMBOL_GPL(kvm_set_xcr);
728
Avi Kivitya83b29c2010-06-10 17:02:15 +0300729int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
Carsten Ottea03490e2007-10-29 16:09:35 +0100730{
Avi Kivityfc78f512009-12-07 12:16:48 +0200731 unsigned long old_cr4 = kvm_read_cr4(vcpu);
Xiao Guangrong0be02262015-05-11 22:55:21 +0800732 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
Huaitong Hanb9baba82016-03-22 16:51:21 +0800733 X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
Xiao Guangrong0be02262015-05-11 22:55:21 +0800734
Gleb Natapov0f122442010-04-28 19:15:31 +0300735 if (cr4 & CR4_RESERVED_BITS)
736 return 1;
Carsten Ottea03490e2007-10-29 16:09:35 +0100737
Dexuan Cui2acf9232010-06-10 11:27:12 +0800738 if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
739 return 1;
740
Yang, Wei Yc68b7342011-06-03 11:13:42 +0800741 if (!guest_cpuid_has_smep(vcpu) && (cr4 & X86_CR4_SMEP))
742 return 1;
743
Feng Wu97ec8c02014-04-01 17:46:34 +0800744 if (!guest_cpuid_has_smap(vcpu) && (cr4 & X86_CR4_SMAP))
745 return 1;
746
H. Peter Anvinafcbf132013-04-27 16:37:47 -0700747 if (!guest_cpuid_has_fsgsbase(vcpu) && (cr4 & X86_CR4_FSGSBASE))
Yang, Wei74dc2b42011-06-14 20:10:18 +0800748 return 1;
749
Huaitong Hanb9baba82016-03-22 16:51:21 +0800750 if (!guest_cpuid_has_pku(vcpu) && (cr4 & X86_CR4_PKE))
751 return 1;
752
Carsten Ottea03490e2007-10-29 16:09:35 +0100753 if (is_long_mode(vcpu)) {
Gleb Natapov0f122442010-04-28 19:15:31 +0300754 if (!(cr4 & X86_CR4_PAE))
755 return 1;
Avi Kivitya2edf572009-05-24 22:19:00 +0300756 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
757 && ((cr4 ^ old_cr4) & pdptr_bits)
Avi Kivity9f8fe502010-12-05 17:30:00 +0200758 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
759 kvm_read_cr3(vcpu)))
Gleb Natapov0f122442010-04-28 19:15:31 +0300760 return 1;
Carsten Ottea03490e2007-10-29 16:09:35 +0100761
Mao, Junjiead756a12012-07-02 01:18:48 +0000762 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
763 if (!guest_cpuid_has_pcid(vcpu))
764 return 1;
765
766 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
767 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
768 return 1;
769 }
770
Nadav Har'El5e1746d2011-05-25 23:03:24 +0300771 if (kvm_x86_ops->set_cr4(vcpu, cr4))
Gleb Natapov0f122442010-04-28 19:15:31 +0300772 return 1;
773
Mao, Junjiead756a12012-07-02 01:18:48 +0000774 if (((cr4 ^ old_cr4) & pdptr_bits) ||
775 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
Sheng Yangaad82702010-05-12 16:40:42 +0800776 kvm_mmu_reset_context(vcpu);
Gleb Natapov0f122442010-04-28 19:15:31 +0300777
Huaitong Hanb9baba82016-03-22 16:51:21 +0800778 if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
Avi Kivity00b27a32011-11-23 16:30:32 +0200779 kvm_update_cpuid(vcpu);
Dexuan Cui2acf9232010-06-10 11:27:12 +0800780
Gleb Natapov0f122442010-04-28 19:15:31 +0300781 return 0;
782}
Avi Kivity2d3ad1f2008-02-24 11:20:43 +0200783EXPORT_SYMBOL_GPL(kvm_set_cr4);
Carsten Ottea03490e2007-10-29 16:09:35 +0100784
Avi Kivity23902182010-06-10 17:02:16 +0300785int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
Carsten Ottea03490e2007-10-29 16:09:35 +0100786{
Paolo Bonziniac146232014-11-10 13:53:25 +0100787#ifdef CONFIG_X86_64
Nadav Amit9d88fca2014-11-02 11:54:52 +0200788 cr3 &= ~CR3_PCID_INVD;
Paolo Bonziniac146232014-11-10 13:53:25 +0100789#endif
Nadav Amit9d88fca2014-11-02 11:54:52 +0200790
Avi Kivity9f8fe502010-12-05 17:30:00 +0200791 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
Marcelo Tosatti0ba73cd2008-09-23 13:18:34 -0300792 kvm_mmu_sync_roots(vcpu);
Liang Chen77c39132014-09-18 12:38:37 -0400793 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
Gleb Natapov0f122442010-04-28 19:15:31 +0300794 return 0;
Avi Kivityd835dfe2007-11-21 02:57:59 +0200795 }
796
Carsten Ottea03490e2007-10-29 16:09:35 +0100797 if (is_long_mode(vcpu)) {
Jan Kiszkad9f89b82014-05-10 09:24:34 +0200798 if (cr3 & CR3_L_MODE_RESERVED_BITS)
799 return 1;
800 } else if (is_pae(vcpu) && is_paging(vcpu) &&
801 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
Nadav Amit346874c2014-04-18 03:35:09 +0300802 return 1;
Carsten Ottea03490e2007-10-29 16:09:35 +0100803
Gleb Natapov0f122442010-04-28 19:15:31 +0300804 vcpu->arch.cr3 = cr3;
Avi Kivityaff48ba2010-12-05 18:56:11 +0200805 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
Paolo Bonzinid8d173d2013-10-02 16:56:11 +0200806 kvm_mmu_new_cr3(vcpu);
Gleb Natapov0f122442010-04-28 19:15:31 +0300807 return 0;
808}
Avi Kivity2d3ad1f2008-02-24 11:20:43 +0200809EXPORT_SYMBOL_GPL(kvm_set_cr3);
Carsten Ottea03490e2007-10-29 16:09:35 +0100810
Andre Przywaraeea1cff2010-12-21 11:12:00 +0100811int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
Carsten Ottea03490e2007-10-29 16:09:35 +0100812{
Gleb Natapov0f122442010-04-28 19:15:31 +0300813 if (cr8 & CR8_RESERVED_BITS)
814 return 1;
Paolo Bonzini35754c92015-07-29 12:05:37 +0200815 if (lapic_in_kernel(vcpu))
Carsten Ottea03490e2007-10-29 16:09:35 +0100816 kvm_lapic_set_tpr(vcpu, cr8);
817 else
Zhang Xiantaoad312c72007-12-13 23:50:52 +0800818 vcpu->arch.cr8 = cr8;
Gleb Natapov0f122442010-04-28 19:15:31 +0300819 return 0;
820}
Avi Kivity2d3ad1f2008-02-24 11:20:43 +0200821EXPORT_SYMBOL_GPL(kvm_set_cr8);
Carsten Ottea03490e2007-10-29 16:09:35 +0100822
Avi Kivity2d3ad1f2008-02-24 11:20:43 +0200823unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
Carsten Ottea03490e2007-10-29 16:09:35 +0100824{
Paolo Bonzini35754c92015-07-29 12:05:37 +0200825 if (lapic_in_kernel(vcpu))
Carsten Ottea03490e2007-10-29 16:09:35 +0100826 return kvm_lapic_get_cr8(vcpu);
827 else
Zhang Xiantaoad312c72007-12-13 23:50:52 +0800828 return vcpu->arch.cr8;
Carsten Ottea03490e2007-10-29 16:09:35 +0100829}
Avi Kivity2d3ad1f2008-02-24 11:20:43 +0200830EXPORT_SYMBOL_GPL(kvm_get_cr8);
Carsten Ottea03490e2007-10-29 16:09:35 +0100831
Nadav Amitae561ed2015-04-02 03:10:37 +0300832static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
833{
834 int i;
835
836 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
837 for (i = 0; i < KVM_NR_DB_REGS; i++)
838 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
839 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
840 }
841}
842
Jan Kiszka73aaf249e2014-01-04 18:47:16 +0100843static void kvm_update_dr6(struct kvm_vcpu *vcpu)
844{
845 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
846 kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
847}
848
Jan Kiszkac8639012012-09-21 05:42:55 +0200849static void kvm_update_dr7(struct kvm_vcpu *vcpu)
850{
851 unsigned long dr7;
852
853 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
854 dr7 = vcpu->arch.guest_debug_dr7;
855 else
856 dr7 = vcpu->arch.dr7;
857 kvm_x86_ops->set_dr7(vcpu, dr7);
Paolo Bonzini360b9482014-02-21 09:55:56 +0100858 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
859 if (dr7 & DR7_BP_EN_MASK)
860 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
Jan Kiszkac8639012012-09-21 05:42:55 +0200861}
862
Nadav Amit6f43ed02014-07-15 17:37:46 +0300863static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
864{
865 u64 fixed = DR6_FIXED_1;
866
867 if (!guest_cpuid_has_rtm(vcpu))
868 fixed |= DR6_RTM;
869 return fixed;
870}
871
Gleb Natapov338dbc92010-04-28 19:15:32 +0300872static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
Gleb Natapov020df072010-04-13 10:05:23 +0300873{
874 switch (dr) {
875 case 0 ... 3:
876 vcpu->arch.db[dr] = val;
877 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
878 vcpu->arch.eff_db[dr] = val;
879 break;
880 case 4:
Gleb Natapov020df072010-04-13 10:05:23 +0300881 /* fall through */
882 case 6:
Gleb Natapov338dbc92010-04-28 19:15:32 +0300883 if (val & 0xffffffff00000000ULL)
884 return -1; /* #GP */
Nadav Amit6f43ed02014-07-15 17:37:46 +0300885 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
Jan Kiszka73aaf249e2014-01-04 18:47:16 +0100886 kvm_update_dr6(vcpu);
Gleb Natapov020df072010-04-13 10:05:23 +0300887 break;
888 case 5:
Gleb Natapov020df072010-04-13 10:05:23 +0300889 /* fall through */
890 default: /* 7 */
Gleb Natapov338dbc92010-04-28 19:15:32 +0300891 if (val & 0xffffffff00000000ULL)
892 return -1; /* #GP */
Gleb Natapov020df072010-04-13 10:05:23 +0300893 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
Jan Kiszkac8639012012-09-21 05:42:55 +0200894 kvm_update_dr7(vcpu);
Gleb Natapov020df072010-04-13 10:05:23 +0300895 break;
896 }
897
898 return 0;
899}
Gleb Natapov338dbc92010-04-28 19:15:32 +0300900
901int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
902{
Nadav Amit16f8a6f2014-10-03 01:10:05 +0300903 if (__kvm_set_dr(vcpu, dr, val)) {
Gleb Natapov338dbc92010-04-28 19:15:32 +0300904 kvm_inject_gp(vcpu, 0);
Nadav Amit16f8a6f2014-10-03 01:10:05 +0300905 return 1;
906 }
907 return 0;
Gleb Natapov338dbc92010-04-28 19:15:32 +0300908}
Gleb Natapov020df072010-04-13 10:05:23 +0300909EXPORT_SYMBOL_GPL(kvm_set_dr);
910
Nadav Amit16f8a6f2014-10-03 01:10:05 +0300911int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
Gleb Natapov020df072010-04-13 10:05:23 +0300912{
913 switch (dr) {
914 case 0 ... 3:
915 *val = vcpu->arch.db[dr];
916 break;
917 case 4:
Gleb Natapov020df072010-04-13 10:05:23 +0300918 /* fall through */
919 case 6:
Jan Kiszka73aaf249e2014-01-04 18:47:16 +0100920 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
921 *val = vcpu->arch.dr6;
922 else
923 *val = kvm_x86_ops->get_dr6(vcpu);
Gleb Natapov020df072010-04-13 10:05:23 +0300924 break;
925 case 5:
Gleb Natapov020df072010-04-13 10:05:23 +0300926 /* fall through */
927 default: /* 7 */
928 *val = vcpu->arch.dr7;
929 break;
930 }
Gleb Natapov338dbc92010-04-28 19:15:32 +0300931 return 0;
932}
Gleb Natapov020df072010-04-13 10:05:23 +0300933EXPORT_SYMBOL_GPL(kvm_get_dr);
934
Avi Kivity022cd0e2011-11-10 14:57:23 +0200935bool kvm_rdpmc(struct kvm_vcpu *vcpu)
936{
937 u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
938 u64 data;
939 int err;
940
Wei Huangc6702c92015-06-19 13:44:45 +0200941 err = kvm_pmu_rdpmc(vcpu, ecx, &data);
Avi Kivity022cd0e2011-11-10 14:57:23 +0200942 if (err)
943 return err;
944 kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
945 kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
946 return err;
947}
948EXPORT_SYMBOL_GPL(kvm_rdpmc);
949
Carsten Otte043405e2007-10-10 17:16:19 +0200950/*
951 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
952 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
953 *
954 * This list is modified at module load time to reflect the
Glauber Costae3267cb2009-10-06 13:24:50 -0400955 * capabilities of the host cpu. This capabilities test skips MSRs that are
Paolo Bonzini62ef68b2015-05-05 12:08:55 +0200956 * kvm-specific. Those are put in emulated_msrs; filtering of emulated_msrs
957 * may depend on host virtualization features rather than host cpu features.
Carsten Otte043405e2007-10-10 17:16:19 +0200958 */
Glauber Costae3267cb2009-10-06 13:24:50 -0400959
Carsten Otte043405e2007-10-10 17:16:19 +0200960static u32 msrs_to_save[] = {
961 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
Brian Gerst8c065852010-07-17 09:03:26 -0400962 MSR_STAR,
Carsten Otte043405e2007-10-10 17:16:19 +0200963#ifdef CONFIG_X86_64
964 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
965#endif
Nadav Har'Elb3897a42013-07-08 19:12:35 +0800966 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
Paolo Bonzini9dbe6cf2015-11-12 14:49:17 +0100967 MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
Carsten Otte043405e2007-10-10 17:16:19 +0200968};
969
970static unsigned num_msrs_to_save;
971
Paolo Bonzini62ef68b2015-05-05 12:08:55 +0200972static u32 emulated_msrs[] = {
973 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
974 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
975 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
976 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
Andrey Smetanine7d95132015-07-03 15:01:37 +0300977 HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
978 HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
Andrey Smetanine516ceb2015-09-16 12:29:48 +0300979 HV_X64_MSR_RESET,
Andrey Smetanin11c4b1c2015-09-16 12:29:49 +0300980 HV_X64_MSR_VP_INDEX,
Andrey Smetanin9eec50b2015-09-16 12:29:50 +0300981 HV_X64_MSR_VP_RUNTIME,
Andrey Smetanin5c9194122015-11-10 15:36:34 +0300982 HV_X64_MSR_SCONTROL,
Andrey Smetanin1f4b34f2015-11-30 19:22:21 +0300983 HV_X64_MSR_STIMER0_CONFIG,
Paolo Bonzini62ef68b2015-05-05 12:08:55 +0200984 HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
985 MSR_KVM_PV_EOI_EN,
986
Will Auldba904632012-11-29 12:42:50 -0800987 MSR_IA32_TSC_ADJUST,
Liu, Jinsonga3e06bb2011-09-22 16:55:52 +0800988 MSR_IA32_TSCDEADLINE,
Carsten Otte043405e2007-10-10 17:16:19 +0200989 MSR_IA32_MISC_ENABLE,
Avi Kivity908e75f2010-07-07 14:09:38 +0300990 MSR_IA32_MCG_STATUS,
991 MSR_IA32_MCG_CTL,
Ashok Rajc45dcc72016-06-22 14:59:56 +0800992 MSR_IA32_MCG_EXT_CTL,
Paolo Bonzini64d60672015-05-07 11:36:11 +0200993 MSR_IA32_SMBASE,
Carsten Otte043405e2007-10-10 17:16:19 +0200994};
995
Paolo Bonzini62ef68b2015-05-05 12:08:55 +0200996static unsigned num_emulated_msrs;
997
Jan Kiszka384bb782013-04-20 10:52:36 +0200998bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
Carsten Otte15c4a642007-10-30 18:44:17 +0100999{
Roedel, Joergb69e8ca2010-05-06 11:38:43 +02001000 if (efer & efer_reserved_bits)
Jan Kiszka384bb782013-04-20 10:52:36 +02001001 return false;
Carsten Otte15c4a642007-10-30 18:44:17 +01001002
Alexander Graf1b2fd702009-02-02 16:23:51 +01001003 if (efer & EFER_FFXSR) {
1004 struct kvm_cpuid_entry2 *feat;
1005
1006 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
Roedel, Joergb69e8ca2010-05-06 11:38:43 +02001007 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
Jan Kiszka384bb782013-04-20 10:52:36 +02001008 return false;
Alexander Graf1b2fd702009-02-02 16:23:51 +01001009 }
1010
Alexander Grafd8017472008-11-25 20:17:11 +01001011 if (efer & EFER_SVME) {
1012 struct kvm_cpuid_entry2 *feat;
1013
1014 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
Roedel, Joergb69e8ca2010-05-06 11:38:43 +02001015 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
Jan Kiszka384bb782013-04-20 10:52:36 +02001016 return false;
Alexander Grafd8017472008-11-25 20:17:11 +01001017 }
1018
Jan Kiszka384bb782013-04-20 10:52:36 +02001019 return true;
1020}
1021EXPORT_SYMBOL_GPL(kvm_valid_efer);
1022
1023static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
1024{
1025 u64 old_efer = vcpu->arch.efer;
1026
1027 if (!kvm_valid_efer(vcpu, efer))
1028 return 1;
1029
1030 if (is_paging(vcpu)
1031 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1032 return 1;
1033
Carsten Otte15c4a642007-10-30 18:44:17 +01001034 efer &= ~EFER_LMA;
Avi Kivityf6801df2010-01-21 15:31:50 +02001035 efer |= vcpu->arch.efer & EFER_LMA;
Carsten Otte15c4a642007-10-30 18:44:17 +01001036
Sheng Yanga3d204e2010-05-12 16:40:40 +08001037 kvm_x86_ops->set_efer(vcpu, efer);
1038
Sheng Yangaad82702010-05-12 16:40:42 +08001039 /* Update reserved bits */
1040 if ((efer ^ old_efer) & EFER_NX)
1041 kvm_mmu_reset_context(vcpu);
1042
Roedel, Joergb69e8ca2010-05-06 11:38:43 +02001043 return 0;
Carsten Otte15c4a642007-10-30 18:44:17 +01001044}
1045
Joerg Roedelf2b4b7d2008-01-31 14:57:37 +01001046void kvm_enable_efer_bits(u64 mask)
1047{
1048 efer_reserved_bits &= ~mask;
1049}
1050EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1051
Carsten Otte15c4a642007-10-30 18:44:17 +01001052/*
1053 * Writes msr value into into the appropriate "register".
1054 * Returns 0 on success, non-0 otherwise.
1055 * Assumes vcpu_load() was already called.
1056 */
Will Auld8fe8ab42012-11-29 12:42:12 -08001057int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
Carsten Otte15c4a642007-10-30 18:44:17 +01001058{
Nadav Amit854e8bb2014-09-16 03:24:05 +03001059 switch (msr->index) {
1060 case MSR_FS_BASE:
1061 case MSR_GS_BASE:
1062 case MSR_KERNEL_GS_BASE:
1063 case MSR_CSTAR:
1064 case MSR_LSTAR:
1065 if (is_noncanonical_address(msr->data))
1066 return 1;
1067 break;
1068 case MSR_IA32_SYSENTER_EIP:
1069 case MSR_IA32_SYSENTER_ESP:
1070 /*
1071 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1072 * non-canonical address is written on Intel but not on
1073 * AMD (which ignores the top 32-bits, because it does
1074 * not implement 64-bit SYSENTER).
1075 *
1076 * 64-bit code should hence be able to write a non-canonical
1077 * value on AMD. Making the address canonical ensures that
1078 * vmentry does not fail on Intel after writing a non-canonical
1079 * value, and that something deterministic happens if the guest
1080 * invokes 64-bit SYSENTER.
1081 */
1082 msr->data = get_canonical(msr->data);
1083 }
Will Auld8fe8ab42012-11-29 12:42:12 -08001084 return kvm_x86_ops->set_msr(vcpu, msr);
Carsten Otte15c4a642007-10-30 18:44:17 +01001085}
Nadav Amit854e8bb2014-09-16 03:24:05 +03001086EXPORT_SYMBOL_GPL(kvm_set_msr);
Carsten Otte15c4a642007-10-30 18:44:17 +01001087
Carsten Otte313a3dc2007-10-11 19:16:52 +02001088/*
1089 * Adapt set_msr() to msr_io()'s calling convention
1090 */
Paolo Bonzini609e36d2015-04-08 15:30:38 +02001091static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1092{
1093 struct msr_data msr;
1094 int r;
1095
1096 msr.index = index;
1097 msr.host_initiated = true;
1098 r = kvm_get_msr(vcpu, &msr);
1099 if (r)
1100 return r;
1101
1102 *data = msr.data;
1103 return 0;
1104}
1105
Carsten Otte313a3dc2007-10-11 19:16:52 +02001106static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1107{
Will Auld8fe8ab42012-11-29 12:42:12 -08001108 struct msr_data msr;
1109
1110 msr.data = *data;
1111 msr.index = index;
1112 msr.host_initiated = true;
1113 return kvm_set_msr(vcpu, &msr);
Carsten Otte313a3dc2007-10-11 19:16:52 +02001114}
1115
Marcelo Tosatti16e8d742012-11-27 23:29:00 -02001116#ifdef CONFIG_X86_64
1117struct pvclock_gtod_data {
1118 seqcount_t seq;
1119
1120 struct { /* extract of a clocksource struct */
1121 int vclock_mode;
1122 cycle_t cycle_last;
1123 cycle_t mask;
1124 u32 mult;
1125 u32 shift;
1126 } clock;
1127
Thomas Gleixnercbcf2dd2014-07-16 21:04:54 +00001128 u64 boot_ns;
1129 u64 nsec_base;
Marcelo Tosatti16e8d742012-11-27 23:29:00 -02001130};
1131
1132static struct pvclock_gtod_data pvclock_gtod_data;
1133
1134static void update_pvclock_gtod(struct timekeeper *tk)
1135{
1136 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
Thomas Gleixnercbcf2dd2014-07-16 21:04:54 +00001137 u64 boot_ns;
1138
Peter Zijlstra876e7882015-03-19 10:09:06 +01001139 boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
Marcelo Tosatti16e8d742012-11-27 23:29:00 -02001140
1141 write_seqcount_begin(&vdata->seq);
1142
1143 /* copy pvclock gtod data */
Peter Zijlstra876e7882015-03-19 10:09:06 +01001144 vdata->clock.vclock_mode = tk->tkr_mono.clock->archdata.vclock_mode;
1145 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
1146 vdata->clock.mask = tk->tkr_mono.mask;
1147 vdata->clock.mult = tk->tkr_mono.mult;
1148 vdata->clock.shift = tk->tkr_mono.shift;
Marcelo Tosatti16e8d742012-11-27 23:29:00 -02001149
Thomas Gleixnercbcf2dd2014-07-16 21:04:54 +00001150 vdata->boot_ns = boot_ns;
Peter Zijlstra876e7882015-03-19 10:09:06 +01001151 vdata->nsec_base = tk->tkr_mono.xtime_nsec;
Marcelo Tosatti16e8d742012-11-27 23:29:00 -02001152
1153 write_seqcount_end(&vdata->seq);
1154}
1155#endif
1156
Nicholas Krausebab5bb32015-01-01 22:05:18 -05001157void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1158{
1159 /*
1160 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1161 * vcpu_enter_guest. This function is only called from
1162 * the physical CPU that is running vcpu.
1163 */
1164 kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1165}
Marcelo Tosatti16e8d742012-11-27 23:29:00 -02001166
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001167static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1168{
Avi Kivity9ed3c442010-05-04 15:00:37 +03001169 int version;
1170 int r;
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +02001171 struct pvclock_wall_clock wc;
Arnd Bergmann87aeb542016-06-17 17:48:56 +02001172 struct timespec64 boot;
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001173
1174 if (!wall_clock)
1175 return;
1176
Avi Kivity9ed3c442010-05-04 15:00:37 +03001177 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1178 if (r)
1179 return;
1180
1181 if (version & 1)
1182 ++version; /* first time write, random junk */
1183
1184 ++version;
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001185
Nicholas Krause1dab1342015-12-30 13:08:46 -05001186 if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1187 return;
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001188
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +02001189 /*
1190 * The guest calculates current wall clock time by adding
Zachary Amsden34c238a2010-09-18 14:38:14 -10001191 * system time (updated by kvm_guest_time_update below) to the
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +02001192 * wall clock specified here. guest system time equals host
1193 * system time for us, thus we must fill in host boot time here.
1194 */
Arnd Bergmann87aeb542016-06-17 17:48:56 +02001195 getboottime64(&boot);
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +02001196
Bruce Rogers4b648662012-07-20 10:44:24 -06001197 if (kvm->arch.kvmclock_offset) {
Arnd Bergmann87aeb542016-06-17 17:48:56 +02001198 struct timespec64 ts = ns_to_timespec64(kvm->arch.kvmclock_offset);
1199 boot = timespec64_sub(boot, ts);
Bruce Rogers4b648662012-07-20 10:44:24 -06001200 }
Arnd Bergmann87aeb542016-06-17 17:48:56 +02001201 wc.sec = (u32)boot.tv_sec; /* overflow in 2106 guest time */
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +02001202 wc.nsec = boot.tv_nsec;
1203 wc.version = version;
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001204
1205 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1206
1207 version++;
1208 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001209}
1210
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +02001211static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1212{
Paolo Bonzinib51012d2016-01-22 11:39:22 +01001213 do_shl32_div32(dividend, divisor);
1214 return dividend;
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +02001215}
1216
Paolo Bonzini3ae13fa2016-02-08 15:11:15 +01001217static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
Zachary Amsden5f4e3f82010-09-18 14:38:13 -10001218 s8 *pshift, u32 *pmultiplier)
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +02001219{
Zachary Amsden5f4e3f82010-09-18 14:38:13 -10001220 uint64_t scaled64;
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +02001221 int32_t shift = 0;
1222 uint64_t tps64;
1223 uint32_t tps32;
1224
Paolo Bonzini3ae13fa2016-02-08 15:11:15 +01001225 tps64 = base_hz;
1226 scaled64 = scaled_hz;
Jan Kiszka50933622010-09-26 13:00:53 +02001227 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +02001228 tps64 >>= 1;
1229 shift--;
1230 }
1231
1232 tps32 = (uint32_t)tps64;
Jan Kiszka50933622010-09-26 13:00:53 +02001233 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1234 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
Zachary Amsden5f4e3f82010-09-18 14:38:13 -10001235 scaled64 >>= 1;
1236 else
1237 tps32 <<= 1;
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +02001238 shift++;
1239 }
1240
Zachary Amsden5f4e3f82010-09-18 14:38:13 -10001241 *pshift = shift;
1242 *pmultiplier = div_frac(scaled64, tps32);
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +02001243
Paolo Bonzini3ae13fa2016-02-08 15:11:15 +01001244 pr_debug("%s: base_hz %llu => %llu, shift %d, mul %u\n",
1245 __func__, base_hz, scaled_hz, shift, *pmultiplier);
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +02001246}
1247
Marcelo Tosattid8281992012-11-27 23:29:01 -02001248#ifdef CONFIG_X86_64
Marcelo Tosatti16e8d742012-11-27 23:29:00 -02001249static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
Marcelo Tosattid8281992012-11-27 23:29:01 -02001250#endif
Marcelo Tosatti16e8d742012-11-27 23:29:00 -02001251
Gerd Hoffmannc8076602009-02-04 17:52:04 +01001252static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
Fengguang Wu69b00492015-01-19 22:33:39 +08001253static unsigned long max_tsc_khz;
Gerd Hoffmannc8076602009-02-04 17:52:04 +01001254
Zachary Amsdencc578282012-02-03 15:43:50 -02001255static u32 adjust_tsc_khz(u32 khz, s32 ppm)
Zachary Amsdenc2855452010-09-18 14:38:15 -10001256{
Zachary Amsdencc578282012-02-03 15:43:50 -02001257 u64 v = (u64)khz * (1000000 + ppm);
1258 do_div(v, 1000000);
1259 return v;
1260}
1261
Haozhong Zhang381d5852015-10-20 15:39:04 +08001262static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1263{
1264 u64 ratio;
1265
1266 /* Guest TSC same frequency as host TSC? */
1267 if (!scale) {
1268 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1269 return 0;
1270 }
1271
1272 /* TSC scaling supported? */
1273 if (!kvm_has_tsc_control) {
1274 if (user_tsc_khz > tsc_khz) {
1275 vcpu->arch.tsc_catchup = 1;
1276 vcpu->arch.tsc_always_catchup = 1;
1277 return 0;
1278 } else {
1279 WARN(1, "user requested TSC rate below hardware speed\n");
1280 return -1;
1281 }
1282 }
1283
1284 /* TSC scaling required - calculate ratio */
1285 ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1286 user_tsc_khz, tsc_khz);
1287
1288 if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1289 WARN_ONCE(1, "Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1290 user_tsc_khz);
1291 return -1;
1292 }
1293
1294 vcpu->arch.tsc_scaling_ratio = ratio;
1295 return 0;
1296}
1297
Paolo Bonzini4941b8c2016-02-08 14:51:12 +01001298static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
Zachary Amsdencc578282012-02-03 15:43:50 -02001299{
1300 u32 thresh_lo, thresh_hi;
1301 int use_scaling = 0;
1302
Marcelo Tosatti03ba32c2013-03-11 23:10:24 -03001303 /* tsc_khz can be zero if TSC calibration fails */
Paolo Bonzini4941b8c2016-02-08 14:51:12 +01001304 if (user_tsc_khz == 0) {
Haozhong Zhangad721882015-10-20 15:39:02 +08001305 /* set tsc_scaling_ratio to a safe value */
1306 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
Haozhong Zhang381d5852015-10-20 15:39:04 +08001307 return -1;
Haozhong Zhangad721882015-10-20 15:39:02 +08001308 }
Marcelo Tosatti03ba32c2013-03-11 23:10:24 -03001309
Zachary Amsdenc2855452010-09-18 14:38:15 -10001310 /* Compute a scale to convert nanoseconds in TSC cycles */
Paolo Bonzini3ae13fa2016-02-08 15:11:15 +01001311 kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
Zachary Amsdencc578282012-02-03 15:43:50 -02001312 &vcpu->arch.virtual_tsc_shift,
1313 &vcpu->arch.virtual_tsc_mult);
Paolo Bonzini4941b8c2016-02-08 14:51:12 +01001314 vcpu->arch.virtual_tsc_khz = user_tsc_khz;
Zachary Amsdencc578282012-02-03 15:43:50 -02001315
1316 /*
1317 * Compute the variation in TSC rate which is acceptable
1318 * within the range of tolerance and decide if the
1319 * rate being applied is within that bounds of the hardware
1320 * rate. If so, no scaling or compensation need be done.
1321 */
1322 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1323 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
Paolo Bonzini4941b8c2016-02-08 14:51:12 +01001324 if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
1325 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
Zachary Amsdencc578282012-02-03 15:43:50 -02001326 use_scaling = 1;
1327 }
Paolo Bonzini4941b8c2016-02-08 14:51:12 +01001328 return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
Zachary Amsdenc2855452010-09-18 14:38:15 -10001329}
1330
1331static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1332{
Zachary Amsdene26101b2012-02-03 15:43:57 -02001333 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
Zachary Amsdencc578282012-02-03 15:43:50 -02001334 vcpu->arch.virtual_tsc_mult,
1335 vcpu->arch.virtual_tsc_shift);
Zachary Amsdene26101b2012-02-03 15:43:57 -02001336 tsc += vcpu->arch.this_tsc_write;
Zachary Amsdenc2855452010-09-18 14:38:15 -10001337 return tsc;
1338}
1339
Fengguang Wu69b00492015-01-19 22:33:39 +08001340static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
Marcelo Tosattib48aa972012-11-27 23:29:03 -02001341{
1342#ifdef CONFIG_X86_64
1343 bool vcpus_matched;
Marcelo Tosattib48aa972012-11-27 23:29:03 -02001344 struct kvm_arch *ka = &vcpu->kvm->arch;
1345 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1346
1347 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1348 atomic_read(&vcpu->kvm->online_vcpus));
1349
Marcelo Tosatti7f187922014-11-04 21:30:44 -02001350 /*
1351 * Once the masterclock is enabled, always perform request in
1352 * order to update it.
1353 *
1354 * In order to enable masterclock, the host clocksource must be TSC
1355 * and the vcpus need to have matched TSCs. When that happens,
1356 * perform request to enable masterclock.
1357 */
1358 if (ka->use_master_clock ||
1359 (gtod->clock.vclock_mode == VCLOCK_TSC && vcpus_matched))
Marcelo Tosattib48aa972012-11-27 23:29:03 -02001360 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1361
1362 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1363 atomic_read(&vcpu->kvm->online_vcpus),
1364 ka->use_master_clock, gtod->clock.vclock_mode);
1365#endif
1366}
1367
Will Auldba904632012-11-29 12:42:50 -08001368static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1369{
1370 u64 curr_offset = kvm_x86_ops->read_tsc_offset(vcpu);
1371 vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1372}
1373
Haozhong Zhang35181e82015-10-20 15:39:03 +08001374/*
1375 * Multiply tsc by a fixed point number represented by ratio.
1376 *
1377 * The most significant 64-N bits (mult) of ratio represent the
1378 * integral part of the fixed point number; the remaining N bits
1379 * (frac) represent the fractional part, ie. ratio represents a fixed
1380 * point number (mult + frac * 2^(-N)).
1381 *
1382 * N equals to kvm_tsc_scaling_ratio_frac_bits.
1383 */
1384static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1385{
1386 return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1387}
1388
1389u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1390{
1391 u64 _tsc = tsc;
1392 u64 ratio = vcpu->arch.tsc_scaling_ratio;
1393
1394 if (ratio != kvm_default_tsc_scaling_ratio)
1395 _tsc = __scale_tsc(ratio, tsc);
1396
1397 return _tsc;
1398}
1399EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1400
Haozhong Zhang07c14192015-10-20 15:39:05 +08001401static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1402{
1403 u64 tsc;
1404
1405 tsc = kvm_scale_tsc(vcpu, rdtsc());
1406
1407 return target_tsc - tsc;
1408}
1409
Haozhong Zhang4ba76532015-10-20 15:39:07 +08001410u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1411{
1412 return kvm_x86_ops->read_l1_tsc(vcpu, kvm_scale_tsc(vcpu, host_tsc));
1413}
1414EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1415
Will Auld8fe8ab42012-11-29 12:42:12 -08001416void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
Zachary Amsden99e3e302010-08-19 22:07:17 -10001417{
1418 struct kvm *kvm = vcpu->kvm;
Zachary Amsdenf38e0982010-08-19 22:07:20 -10001419 u64 offset, ns, elapsed;
Zachary Amsden99e3e302010-08-19 22:07:17 -10001420 unsigned long flags;
Marcelo Tosatti02626b62012-03-08 18:46:57 -03001421 s64 usdiff;
Marcelo Tosattib48aa972012-11-27 23:29:03 -02001422 bool matched;
Tomasz Grabiec0d3da0d2014-06-24 09:42:43 +02001423 bool already_matched;
Will Auld8fe8ab42012-11-29 12:42:12 -08001424 u64 data = msr->data;
Zachary Amsden99e3e302010-08-19 22:07:17 -10001425
Jan Kiszka038f8c12011-02-04 10:49:11 +01001426 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
Haozhong Zhang07c14192015-10-20 15:39:05 +08001427 offset = kvm_compute_tsc_offset(vcpu, data);
Zachary Amsden759379d2010-08-19 22:07:25 -10001428 ns = get_kernel_ns();
Zachary Amsdenf38e0982010-08-19 22:07:20 -10001429 elapsed = ns - kvm->arch.last_tsc_nsec;
Zachary Amsden5d3cb0f62012-02-03 15:43:51 -02001430
Marcelo Tosatti03ba32c2013-03-11 23:10:24 -03001431 if (vcpu->arch.virtual_tsc_khz) {
Marcelo Tosatti8915aa272013-06-11 23:31:12 -03001432 int faulted = 0;
1433
Marcelo Tosatti03ba32c2013-03-11 23:10:24 -03001434 /* n.b - signed multiplication and division required */
1435 usdiff = data - kvm->arch.last_tsc_write;
Zachary Amsden5d3cb0f62012-02-03 15:43:51 -02001436#ifdef CONFIG_X86_64
Marcelo Tosatti03ba32c2013-03-11 23:10:24 -03001437 usdiff = (usdiff * 1000) / vcpu->arch.virtual_tsc_khz;
Zachary Amsden5d3cb0f62012-02-03 15:43:51 -02001438#else
Marcelo Tosatti03ba32c2013-03-11 23:10:24 -03001439 /* do_div() only does unsigned */
Marcelo Tosatti8915aa272013-06-11 23:31:12 -03001440 asm("1: idivl %[divisor]\n"
1441 "2: xor %%edx, %%edx\n"
1442 " movl $0, %[faulted]\n"
1443 "3:\n"
1444 ".section .fixup,\"ax\"\n"
1445 "4: movl $1, %[faulted]\n"
1446 " jmp 3b\n"
1447 ".previous\n"
1448
1449 _ASM_EXTABLE(1b, 4b)
1450
1451 : "=A"(usdiff), [faulted] "=r" (faulted)
1452 : "A"(usdiff * 1000), [divisor] "rm"(vcpu->arch.virtual_tsc_khz));
1453
Zachary Amsden5d3cb0f62012-02-03 15:43:51 -02001454#endif
Marcelo Tosatti03ba32c2013-03-11 23:10:24 -03001455 do_div(elapsed, 1000);
1456 usdiff -= elapsed;
1457 if (usdiff < 0)
1458 usdiff = -usdiff;
Marcelo Tosatti8915aa272013-06-11 23:31:12 -03001459
1460 /* idivl overflow => difference is larger than USEC_PER_SEC */
1461 if (faulted)
1462 usdiff = USEC_PER_SEC;
Marcelo Tosatti03ba32c2013-03-11 23:10:24 -03001463 } else
1464 usdiff = USEC_PER_SEC; /* disable TSC match window below */
Zachary Amsdenf38e0982010-08-19 22:07:20 -10001465
1466 /*
Zachary Amsden5d3cb0f62012-02-03 15:43:51 -02001467 * Special case: TSC write with a small delta (1 second) of virtual
1468 * cycle time against real time is interpreted as an attempt to
1469 * synchronize the CPU.
1470 *
1471 * For a reliable TSC, we can match TSC offsets, and for an unstable
1472 * TSC, we add elapsed time in this computation. We could let the
1473 * compensation code attempt to catch up if we fall behind, but
1474 * it's better to try to match offsets from the beginning.
1475 */
Marcelo Tosatti02626b62012-03-08 18:46:57 -03001476 if (usdiff < USEC_PER_SEC &&
Zachary Amsden5d3cb0f62012-02-03 15:43:51 -02001477 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
Zachary Amsdenf38e0982010-08-19 22:07:20 -10001478 if (!check_tsc_unstable()) {
Zachary Amsdene26101b2012-02-03 15:43:57 -02001479 offset = kvm->arch.cur_tsc_offset;
Zachary Amsdenf38e0982010-08-19 22:07:20 -10001480 pr_debug("kvm: matched tsc offset for %llu\n", data);
1481 } else {
Joerg Roedel857e4092011-03-25 09:44:50 +01001482 u64 delta = nsec_to_cycles(vcpu, elapsed);
Zachary Amsden5d3cb0f62012-02-03 15:43:51 -02001483 data += delta;
Haozhong Zhang07c14192015-10-20 15:39:05 +08001484 offset = kvm_compute_tsc_offset(vcpu, data);
Zachary Amsden759379d2010-08-19 22:07:25 -10001485 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
Zachary Amsdenf38e0982010-08-19 22:07:20 -10001486 }
Marcelo Tosattib48aa972012-11-27 23:29:03 -02001487 matched = true;
Tomasz Grabiec0d3da0d2014-06-24 09:42:43 +02001488 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
Zachary Amsdene26101b2012-02-03 15:43:57 -02001489 } else {
1490 /*
1491 * We split periods of matched TSC writes into generations.
1492 * For each generation, we track the original measured
1493 * nanosecond time, offset, and write, so if TSCs are in
1494 * sync, we can match exact offset, and if not, we can match
Guo Chao4a969982012-06-28 15:17:27 +08001495 * exact software computation in compute_guest_tsc()
Zachary Amsdene26101b2012-02-03 15:43:57 -02001496 *
1497 * These values are tracked in kvm->arch.cur_xxx variables.
1498 */
1499 kvm->arch.cur_tsc_generation++;
1500 kvm->arch.cur_tsc_nsec = ns;
1501 kvm->arch.cur_tsc_write = data;
1502 kvm->arch.cur_tsc_offset = offset;
Marcelo Tosattib48aa972012-11-27 23:29:03 -02001503 matched = false;
Tomasz Grabiec0d3da0d2014-06-24 09:42:43 +02001504 pr_debug("kvm: new tsc generation %llu, clock %llu\n",
Zachary Amsdene26101b2012-02-03 15:43:57 -02001505 kvm->arch.cur_tsc_generation, data);
Zachary Amsdenf38e0982010-08-19 22:07:20 -10001506 }
Zachary Amsdene26101b2012-02-03 15:43:57 -02001507
1508 /*
1509 * We also track th most recent recorded KHZ, write and time to
1510 * allow the matching interval to be extended at each write.
1511 */
Zachary Amsdenf38e0982010-08-19 22:07:20 -10001512 kvm->arch.last_tsc_nsec = ns;
1513 kvm->arch.last_tsc_write = data;
Zachary Amsden5d3cb0f62012-02-03 15:43:51 -02001514 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
Zachary Amsden99e3e302010-08-19 22:07:17 -10001515
Zachary Amsdenb183aa52012-02-03 15:43:53 -02001516 vcpu->arch.last_guest_tsc = data;
Zachary Amsdene26101b2012-02-03 15:43:57 -02001517
1518 /* Keep track of which generation this VCPU has synchronized to */
1519 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1520 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1521 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1522
Will Auldba904632012-11-29 12:42:50 -08001523 if (guest_cpuid_has_tsc_adjust(vcpu) && !msr->host_initiated)
1524 update_ia32_tsc_adjust_msr(vcpu, offset);
Zachary Amsdene26101b2012-02-03 15:43:57 -02001525 kvm_x86_ops->write_tsc_offset(vcpu, offset);
1526 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
Marcelo Tosattib48aa972012-11-27 23:29:03 -02001527
1528 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
Tomasz Grabiec0d3da0d2014-06-24 09:42:43 +02001529 if (!matched) {
Marcelo Tosattib48aa972012-11-27 23:29:03 -02001530 kvm->arch.nr_vcpus_matched_tsc = 0;
Tomasz Grabiec0d3da0d2014-06-24 09:42:43 +02001531 } else if (!already_matched) {
1532 kvm->arch.nr_vcpus_matched_tsc++;
1533 }
Marcelo Tosattib48aa972012-11-27 23:29:03 -02001534
1535 kvm_track_tsc_matching(vcpu);
1536 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
Zachary Amsden99e3e302010-08-19 22:07:17 -10001537}
Zachary Amsdene26101b2012-02-03 15:43:57 -02001538
Zachary Amsden99e3e302010-08-19 22:07:17 -10001539EXPORT_SYMBOL_GPL(kvm_write_tsc);
1540
Haozhong Zhang58ea6762015-10-20 15:39:06 +08001541static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
1542 s64 adjustment)
1543{
1544 kvm_x86_ops->adjust_tsc_offset_guest(vcpu, adjustment);
1545}
1546
1547static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
1548{
1549 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
1550 WARN_ON(adjustment < 0);
1551 adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
1552 kvm_x86_ops->adjust_tsc_offset_guest(vcpu, adjustment);
1553}
1554
Marcelo Tosattid8281992012-11-27 23:29:01 -02001555#ifdef CONFIG_X86_64
1556
1557static cycle_t read_tsc(void)
1558{
Andy Lutomirski03b97302015-06-25 18:44:08 +02001559 cycle_t ret = (cycle_t)rdtsc_ordered();
1560 u64 last = pvclock_gtod_data.clock.cycle_last;
Marcelo Tosattid8281992012-11-27 23:29:01 -02001561
1562 if (likely(ret >= last))
1563 return ret;
1564
1565 /*
1566 * GCC likes to generate cmov here, but this branch is extremely
Adam Buchbinder6a6256f2016-02-23 15:34:30 -08001567 * predictable (it's just a function of time and the likely is
Marcelo Tosattid8281992012-11-27 23:29:01 -02001568 * very likely) and there's a data dependence, so force GCC
1569 * to generate a branch instead. I don't barrier() because
1570 * we don't actually need a barrier, and if this function
1571 * ever gets inlined it will generate worse code.
1572 */
1573 asm volatile ("");
1574 return last;
1575}
1576
1577static inline u64 vgettsc(cycle_t *cycle_now)
1578{
1579 long v;
1580 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1581
1582 *cycle_now = read_tsc();
1583
1584 v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask;
1585 return v * gtod->clock.mult;
1586}
1587
Thomas Gleixnercbcf2dd2014-07-16 21:04:54 +00001588static int do_monotonic_boot(s64 *t, cycle_t *cycle_now)
Marcelo Tosattid8281992012-11-27 23:29:01 -02001589{
Marcelo Tosattid8281992012-11-27 23:29:01 -02001590 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
Thomas Gleixnercbcf2dd2014-07-16 21:04:54 +00001591 unsigned long seq;
1592 int mode;
1593 u64 ns;
Marcelo Tosattid8281992012-11-27 23:29:01 -02001594
Marcelo Tosattid8281992012-11-27 23:29:01 -02001595 do {
1596 seq = read_seqcount_begin(&gtod->seq);
1597 mode = gtod->clock.vclock_mode;
Thomas Gleixnercbcf2dd2014-07-16 21:04:54 +00001598 ns = gtod->nsec_base;
Marcelo Tosattid8281992012-11-27 23:29:01 -02001599 ns += vgettsc(cycle_now);
1600 ns >>= gtod->clock.shift;
Thomas Gleixnercbcf2dd2014-07-16 21:04:54 +00001601 ns += gtod->boot_ns;
Marcelo Tosattid8281992012-11-27 23:29:01 -02001602 } while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
Thomas Gleixnercbcf2dd2014-07-16 21:04:54 +00001603 *t = ns;
Marcelo Tosattid8281992012-11-27 23:29:01 -02001604
1605 return mode;
1606}
1607
1608/* returns true if host is using tsc clocksource */
1609static bool kvm_get_time_and_clockread(s64 *kernel_ns, cycle_t *cycle_now)
1610{
Marcelo Tosattid8281992012-11-27 23:29:01 -02001611 /* checked again under seqlock below */
1612 if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1613 return false;
1614
Thomas Gleixnercbcf2dd2014-07-16 21:04:54 +00001615 return do_monotonic_boot(kernel_ns, cycle_now) == VCLOCK_TSC;
Marcelo Tosattid8281992012-11-27 23:29:01 -02001616}
1617#endif
1618
1619/*
1620 *
Marcelo Tosattib48aa972012-11-27 23:29:03 -02001621 * Assuming a stable TSC across physical CPUS, and a stable TSC
1622 * across virtual CPUs, the following condition is possible.
1623 * Each numbered line represents an event visible to both
Marcelo Tosattid8281992012-11-27 23:29:01 -02001624 * CPUs at the next numbered event.
1625 *
1626 * "timespecX" represents host monotonic time. "tscX" represents
1627 * RDTSC value.
1628 *
1629 * VCPU0 on CPU0 | VCPU1 on CPU1
1630 *
1631 * 1. read timespec0,tsc0
1632 * 2. | timespec1 = timespec0 + N
1633 * | tsc1 = tsc0 + M
1634 * 3. transition to guest | transition to guest
1635 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1636 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
1637 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1638 *
1639 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1640 *
1641 * - ret0 < ret1
1642 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1643 * ...
1644 * - 0 < N - M => M < N
1645 *
1646 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1647 * always the case (the difference between two distinct xtime instances
1648 * might be smaller then the difference between corresponding TSC reads,
1649 * when updating guest vcpus pvclock areas).
1650 *
1651 * To avoid that problem, do not allow visibility of distinct
1652 * system_timestamp/tsc_timestamp values simultaneously: use a master
1653 * copy of host monotonic time values. Update that master copy
1654 * in lockstep.
1655 *
Marcelo Tosattib48aa972012-11-27 23:29:03 -02001656 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
Marcelo Tosattid8281992012-11-27 23:29:01 -02001657 *
1658 */
1659
1660static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1661{
1662#ifdef CONFIG_X86_64
1663 struct kvm_arch *ka = &kvm->arch;
1664 int vclock_mode;
Marcelo Tosattib48aa972012-11-27 23:29:03 -02001665 bool host_tsc_clocksource, vcpus_matched;
1666
1667 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1668 atomic_read(&kvm->online_vcpus));
Marcelo Tosattid8281992012-11-27 23:29:01 -02001669
1670 /*
1671 * If the host uses TSC clock, then passthrough TSC as stable
1672 * to the guest.
1673 */
Marcelo Tosattib48aa972012-11-27 23:29:03 -02001674 host_tsc_clocksource = kvm_get_time_and_clockread(
Marcelo Tosattid8281992012-11-27 23:29:01 -02001675 &ka->master_kernel_ns,
1676 &ka->master_cycle_now);
1677
Marcelo Tosatti16a96022014-05-14 12:43:24 -03001678 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
Marcelo Tosatti54750f22015-01-20 15:54:52 -02001679 && !backwards_tsc_observed
1680 && !ka->boot_vcpu_runs_old_kvmclock;
Marcelo Tosattib48aa972012-11-27 23:29:03 -02001681
Marcelo Tosattid8281992012-11-27 23:29:01 -02001682 if (ka->use_master_clock)
1683 atomic_set(&kvm_guest_has_master_clock, 1);
1684
1685 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
Marcelo Tosattib48aa972012-11-27 23:29:03 -02001686 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1687 vcpus_matched);
Marcelo Tosattid8281992012-11-27 23:29:01 -02001688#endif
1689}
1690
Paolo Bonzini2860c4b2016-01-07 15:05:10 +01001691void kvm_make_mclock_inprogress_request(struct kvm *kvm)
1692{
1693 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
1694}
1695
Marcelo Tosatti2e762ff2013-08-27 23:55:29 -03001696static void kvm_gen_update_masterclock(struct kvm *kvm)
1697{
1698#ifdef CONFIG_X86_64
1699 int i;
1700 struct kvm_vcpu *vcpu;
1701 struct kvm_arch *ka = &kvm->arch;
1702
1703 spin_lock(&ka->pvclock_gtod_sync_lock);
1704 kvm_make_mclock_inprogress_request(kvm);
1705 /* no guest entries from this point */
1706 pvclock_update_vm_gtod_copy(kvm);
1707
1708 kvm_for_each_vcpu(i, vcpu, kvm)
Guo Hui Liu105b21b2014-09-12 13:43:19 +08001709 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
Marcelo Tosatti2e762ff2013-08-27 23:55:29 -03001710
1711 /* guest entries allowed */
1712 kvm_for_each_vcpu(i, vcpu, kvm)
1713 clear_bit(KVM_REQ_MCLOCK_INPROGRESS, &vcpu->requests);
1714
1715 spin_unlock(&ka->pvclock_gtod_sync_lock);
1716#endif
1717}
1718
Zachary Amsden34c238a2010-09-18 14:38:14 -10001719static int kvm_guest_time_update(struct kvm_vcpu *v)
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001720{
Paolo Bonzini78db6a52016-02-08 14:51:40 +01001721 unsigned long flags, tgt_tsc_khz;
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001722 struct kvm_vcpu_arch *vcpu = &v->arch;
Marcelo Tosattid8281992012-11-27 23:29:01 -02001723 struct kvm_arch *ka = &v->kvm->arch;
Marcelo Tosattif25e6562014-01-06 12:18:59 -02001724 s64 kernel_ns;
Marcelo Tosattid8281992012-11-27 23:29:01 -02001725 u64 tsc_timestamp, host_tsc;
Andy Honig0b794592013-02-20 14:48:10 -08001726 struct pvclock_vcpu_time_info guest_hv_clock;
Marcelo Tosatti51d59c62012-08-03 15:57:49 -03001727 u8 pvclock_flags;
Marcelo Tosattid8281992012-11-27 23:29:01 -02001728 bool use_master_clock;
1729
1730 kernel_ns = 0;
1731 host_tsc = 0;
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +02001732
Zachary Amsden1d5f0662010-08-19 22:07:30 -10001733 /*
Marcelo Tosattid8281992012-11-27 23:29:01 -02001734 * If the host uses TSC clock, then passthrough TSC as stable
1735 * to the guest.
1736 */
1737 spin_lock(&ka->pvclock_gtod_sync_lock);
1738 use_master_clock = ka->use_master_clock;
1739 if (use_master_clock) {
1740 host_tsc = ka->master_cycle_now;
1741 kernel_ns = ka->master_kernel_ns;
1742 }
1743 spin_unlock(&ka->pvclock_gtod_sync_lock);
Marcelo Tosattic09664b2013-03-18 13:54:32 -03001744
1745 /* Keep irq disabled to prevent changes to the clock */
1746 local_irq_save(flags);
Paolo Bonzini78db6a52016-02-08 14:51:40 +01001747 tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
1748 if (unlikely(tgt_tsc_khz == 0)) {
Marcelo Tosattic09664b2013-03-18 13:54:32 -03001749 local_irq_restore(flags);
1750 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1751 return 1;
1752 }
Marcelo Tosattid8281992012-11-27 23:29:01 -02001753 if (!use_master_clock) {
Andy Lutomirski4ea16362015-06-25 18:44:07 +02001754 host_tsc = rdtsc();
Marcelo Tosattid8281992012-11-27 23:29:01 -02001755 kernel_ns = get_kernel_ns();
1756 }
1757
Haozhong Zhang4ba76532015-10-20 15:39:07 +08001758 tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
Marcelo Tosattid8281992012-11-27 23:29:01 -02001759
1760 /*
Zachary Amsdenc2855452010-09-18 14:38:15 -10001761 * We may have to catch up the TSC to match elapsed wall clock
1762 * time for two reasons, even if kvmclock is used.
1763 * 1) CPU could have been running below the maximum TSC rate
1764 * 2) Broken TSC compensation resets the base at each VCPU
1765 * entry to avoid unknown leaps of TSC even when running
1766 * again on the same CPU. This may cause apparent elapsed
1767 * time to disappear, and the guest to stand still or run
1768 * very slowly.
1769 */
1770 if (vcpu->tsc_catchup) {
1771 u64 tsc = compute_guest_tsc(v, kernel_ns);
1772 if (tsc > tsc_timestamp) {
Marcelo Tosattif1e2b262012-02-03 15:43:55 -02001773 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
Zachary Amsdenc2855452010-09-18 14:38:15 -10001774 tsc_timestamp = tsc;
1775 }
1776 }
1777
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001778 local_irq_restore(flags);
1779
Andy Honig0b794592013-02-20 14:48:10 -08001780 if (!vcpu->pv_time_enabled)
Zachary Amsdenc2855452010-09-18 14:38:15 -10001781 return 0;
1782
Paolo Bonzini78db6a52016-02-08 14:51:40 +01001783 if (kvm_has_tsc_control)
1784 tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
1785
1786 if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
Paolo Bonzini3ae13fa2016-02-08 15:11:15 +01001787 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
Zachary Amsden5f4e3f82010-09-18 14:38:13 -10001788 &vcpu->hv_clock.tsc_shift,
1789 &vcpu->hv_clock.tsc_to_system_mul);
Paolo Bonzini78db6a52016-02-08 14:51:40 +01001790 vcpu->hw_tsc_khz = tgt_tsc_khz;
Zachary Amsden8cfdc002010-08-19 22:07:21 -10001791 }
1792
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001793 /* With all the info we got, fill in the values */
Zachary Amsden1d5f0662010-08-19 22:07:30 -10001794 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
Zachary Amsden759379d2010-08-19 22:07:25 -10001795 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
Zachary Amsden28e46392010-09-18 14:38:12 -10001796 vcpu->last_guest_tsc = tsc_timestamp;
Marcelo Tosatti51d59c62012-08-03 15:57:49 -03001797
Owen Hofmann09a0c3f2014-11-03 16:57:18 -08001798 if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
1799 &guest_hv_clock, sizeof(guest_hv_clock))))
1800 return 0;
1801
Radim Krčmář5dca0d92015-03-25 12:08:14 +01001802 /* This VCPU is paused, but it's legal for a guest to read another
1803 * VCPU's kvmclock, so we really have to follow the specification where
1804 * it says that version is odd if data is being modified, and even after
1805 * it is consistent.
1806 *
1807 * Version field updates must be kept separate. This is because
1808 * kvm_write_guest_cached might use a "rep movs" instruction, and
1809 * writes within a string instruction are weakly ordered. So there
1810 * are three writes overall.
1811 *
1812 * As a small optimization, only write the version field in the first
1813 * and third write. The vcpu->pv_time cache is still valid, because the
1814 * version field is the first in the struct.
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001815 */
Radim Krčmář5dca0d92015-03-25 12:08:14 +01001816 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
1817
1818 vcpu->hv_clock.version = guest_hv_clock.version + 1;
1819 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1820 &vcpu->hv_clock,
1821 sizeof(vcpu->hv_clock.version));
1822
1823 smp_wmb();
Marcelo Tosatti78c03372012-11-27 23:28:47 -02001824
1825 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
Andy Honig0b794592013-02-20 14:48:10 -08001826 pvclock_flags = (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
Marcelo Tosatti78c03372012-11-27 23:28:47 -02001827
1828 if (vcpu->pvclock_set_guest_stopped_request) {
1829 pvclock_flags |= PVCLOCK_GUEST_STOPPED;
1830 vcpu->pvclock_set_guest_stopped_request = false;
1831 }
1832
Marcelo Tosattid8281992012-11-27 23:29:01 -02001833 /* If the host uses TSC clocksource, then it is stable */
1834 if (use_master_clock)
1835 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
1836
Marcelo Tosatti78c03372012-11-27 23:28:47 -02001837 vcpu->hv_clock.flags = pvclock_flags;
1838
David Matlackce1a5e62014-11-05 11:46:42 -08001839 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
1840
Andy Honig0b794592013-02-20 14:48:10 -08001841 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1842 &vcpu->hv_clock,
1843 sizeof(vcpu->hv_clock));
Radim Krčmář5dca0d92015-03-25 12:08:14 +01001844
1845 smp_wmb();
1846
1847 vcpu->hv_clock.version++;
1848 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1849 &vcpu->hv_clock,
1850 sizeof(vcpu->hv_clock.version));
Zachary Amsden8cfdc002010-08-19 22:07:21 -10001851 return 0;
Gerd Hoffmannc8076602009-02-04 17:52:04 +01001852}
1853
Marcelo Tosatti0061d532013-05-09 20:21:41 -03001854/*
1855 * kvmclock updates which are isolated to a given vcpu, such as
1856 * vcpu->cpu migration, should not allow system_timestamp from
1857 * the rest of the vcpus to remain static. Otherwise ntp frequency
1858 * correction applies to one vcpu's system_timestamp but not
1859 * the others.
1860 *
1861 * So in those cases, request a kvmclock update for all vcpus.
Andrew Jones7e44e442014-02-28 12:52:54 +01001862 * We need to rate-limit these requests though, as they can
1863 * considerably slow guests that have a large number of vcpus.
1864 * The time for a remote vcpu to update its kvmclock is bound
1865 * by the delay we use to rate-limit the updates.
Marcelo Tosatti0061d532013-05-09 20:21:41 -03001866 */
1867
Andrew Jones7e44e442014-02-28 12:52:54 +01001868#define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
1869
1870static void kvmclock_update_fn(struct work_struct *work)
Marcelo Tosatti0061d532013-05-09 20:21:41 -03001871{
1872 int i;
Andrew Jones7e44e442014-02-28 12:52:54 +01001873 struct delayed_work *dwork = to_delayed_work(work);
1874 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1875 kvmclock_update_work);
1876 struct kvm *kvm = container_of(ka, struct kvm, arch);
Marcelo Tosatti0061d532013-05-09 20:21:41 -03001877 struct kvm_vcpu *vcpu;
1878
1879 kvm_for_each_vcpu(i, vcpu, kvm) {
Guo Hui Liu105b21b2014-09-12 13:43:19 +08001880 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
Marcelo Tosatti0061d532013-05-09 20:21:41 -03001881 kvm_vcpu_kick(vcpu);
1882 }
1883}
1884
Andrew Jones7e44e442014-02-28 12:52:54 +01001885static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
1886{
1887 struct kvm *kvm = v->kvm;
1888
Guo Hui Liu105b21b2014-09-12 13:43:19 +08001889 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
Andrew Jones7e44e442014-02-28 12:52:54 +01001890 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
1891 KVMCLOCK_UPDATE_DELAY);
1892}
1893
Andrew Jones332967a2014-02-28 12:52:55 +01001894#define KVMCLOCK_SYNC_PERIOD (300 * HZ)
1895
1896static void kvmclock_sync_fn(struct work_struct *work)
1897{
1898 struct delayed_work *dwork = to_delayed_work(work);
1899 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1900 kvmclock_sync_work);
1901 struct kvm *kvm = container_of(ka, struct kvm, arch);
1902
Marcelo Tosatti630994b2015-05-12 22:42:04 -03001903 if (!kvmclock_periodic_sync)
1904 return;
1905
Andrew Jones332967a2014-02-28 12:52:55 +01001906 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
1907 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
1908 KVMCLOCK_SYNC_PERIOD);
1909}
1910
Huang Ying890ca9a2009-05-11 16:48:15 +08001911static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1912{
1913 u64 mcg_cap = vcpu->arch.mcg_cap;
1914 unsigned bank_num = mcg_cap & 0xff;
1915
1916 switch (msr) {
1917 case MSR_IA32_MCG_STATUS:
1918 vcpu->arch.mcg_status = data;
1919 break;
1920 case MSR_IA32_MCG_CTL:
1921 if (!(mcg_cap & MCG_CTL_P))
1922 return 1;
1923 if (data != 0 && data != ~(u64)0)
1924 return -1;
1925 vcpu->arch.mcg_ctl = data;
1926 break;
1927 default:
1928 if (msr >= MSR_IA32_MC0_CTL &&
Chen Yucong81760dc2014-09-23 10:44:35 +08001929 msr < MSR_IA32_MCx_CTL(bank_num)) {
Huang Ying890ca9a2009-05-11 16:48:15 +08001930 u32 offset = msr - MSR_IA32_MC0_CTL;
Andre Przywara114be422010-03-24 17:46:42 +01001931 /* only 0 or all 1s can be written to IA32_MCi_CTL
1932 * some Linux kernels though clear bit 10 in bank 4 to
1933 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1934 * this to avoid an uncatched #GP in the guest
1935 */
Huang Ying890ca9a2009-05-11 16:48:15 +08001936 if ((offset & 0x3) == 0 &&
Andre Przywara114be422010-03-24 17:46:42 +01001937 data != 0 && (data | (1 << 10)) != ~(u64)0)
Huang Ying890ca9a2009-05-11 16:48:15 +08001938 return -1;
1939 vcpu->arch.mce_banks[offset] = data;
1940 break;
1941 }
1942 return 1;
1943 }
1944 return 0;
1945}
1946
Ed Swierkffde22a2009-10-15 15:21:43 -07001947static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1948{
1949 struct kvm *kvm = vcpu->kvm;
1950 int lm = is_long_mode(vcpu);
1951 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1952 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1953 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1954 : kvm->arch.xen_hvm_config.blob_size_32;
1955 u32 page_num = data & ~PAGE_MASK;
1956 u64 page_addr = data & PAGE_MASK;
1957 u8 *page;
1958 int r;
1959
1960 r = -E2BIG;
1961 if (page_num >= blob_size)
1962 goto out;
1963 r = -ENOMEM;
Sasha Levinff5c2c02011-12-04 19:36:29 +02001964 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
1965 if (IS_ERR(page)) {
1966 r = PTR_ERR(page);
Ed Swierkffde22a2009-10-15 15:21:43 -07001967 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02001968 }
Paolo Bonzini54bf36a2015-04-08 15:39:23 +02001969 if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
Ed Swierkffde22a2009-10-15 15:21:43 -07001970 goto out_free;
1971 r = 0;
1972out_free:
1973 kfree(page);
1974out:
1975 return r;
1976}
1977
Gleb Natapov344d9582010-10-14 11:22:50 +02001978static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
1979{
1980 gpa_t gpa = data & ~0x3f;
1981
Guo Chao4a969982012-06-28 15:17:27 +08001982 /* Bits 2:5 are reserved, Should be zero */
Gleb Natapov6adba522010-10-14 11:22:55 +02001983 if (data & 0x3c)
Gleb Natapov344d9582010-10-14 11:22:50 +02001984 return 1;
1985
1986 vcpu->arch.apf.msr_val = data;
1987
1988 if (!(data & KVM_ASYNC_PF_ENABLED)) {
1989 kvm_clear_async_pf_completion_queue(vcpu);
1990 kvm_async_pf_hash_reset(vcpu);
1991 return 0;
1992 }
1993
Andrew Honig8f964522013-03-29 09:35:21 -07001994 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
1995 sizeof(u32)))
Gleb Natapov344d9582010-10-14 11:22:50 +02001996 return 1;
1997
Gleb Natapov6adba522010-10-14 11:22:55 +02001998 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
Gleb Natapov344d9582010-10-14 11:22:50 +02001999 kvm_async_pf_wakeup_all(vcpu);
2000 return 0;
2001}
2002
Glauber Costa12f9a482011-02-01 14:16:40 -05002003static void kvmclock_reset(struct kvm_vcpu *vcpu)
2004{
Andy Honig0b794592013-02-20 14:48:10 -08002005 vcpu->arch.pv_time_enabled = false;
Glauber Costa12f9a482011-02-01 14:16:40 -05002006}
2007
Glauber Costac9aaa892011-07-11 15:28:14 -04002008static void record_steal_time(struct kvm_vcpu *vcpu)
2009{
2010 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2011 return;
2012
2013 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2014 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2015 return;
2016
Wanpeng Li35f3fae12016-05-03 11:43:10 +08002017 if (vcpu->arch.st.steal.version & 1)
2018 vcpu->arch.st.steal.version += 1; /* first time write, random junk */
2019
2020 vcpu->arch.st.steal.version += 1;
2021
2022 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2023 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2024
2025 smp_wmb();
2026
Liang Chenc54cdf12016-03-16 19:33:16 +08002027 vcpu->arch.st.steal.steal += current->sched_info.run_delay -
2028 vcpu->arch.st.last_steal;
2029 vcpu->arch.st.last_steal = current->sched_info.run_delay;
Wanpeng Li35f3fae12016-05-03 11:43:10 +08002030
2031 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2032 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2033
2034 smp_wmb();
2035
2036 vcpu->arch.st.steal.version += 1;
Glauber Costac9aaa892011-07-11 15:28:14 -04002037
2038 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2039 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2040}
2041
Will Auld8fe8ab42012-11-29 12:42:12 -08002042int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
Carsten Otte15c4a642007-10-30 18:44:17 +01002043{
Gleb Natapov57537852012-01-15 14:17:22 +02002044 bool pr = false;
Will Auld8fe8ab42012-11-29 12:42:12 -08002045 u32 msr = msr_info->index;
2046 u64 data = msr_info->data;
Gleb Natapov57537852012-01-15 14:17:22 +02002047
Carsten Otte15c4a642007-10-30 18:44:17 +01002048 switch (msr) {
Borislav Petkov2e32b712013-02-19 19:33:13 +01002049 case MSR_AMD64_NB_CFG:
2050 case MSR_IA32_UCODE_REV:
2051 case MSR_IA32_UCODE_WRITE:
2052 case MSR_VM_HSAVE_PA:
2053 case MSR_AMD64_PATCH_LOADER:
2054 case MSR_AMD64_BU_CFG2:
2055 break;
2056
Carsten Otte15c4a642007-10-30 18:44:17 +01002057 case MSR_EFER:
Roedel, Joergb69e8ca2010-05-06 11:38:43 +02002058 return set_efer(vcpu, data);
Andre Przywara8f1589d2009-06-24 12:44:33 +02002059 case MSR_K7_HWCR:
2060 data &= ~(u64)0x40; /* ignore flush filter disable */
Joerg Roedel82494022010-02-24 18:59:16 +01002061 data &= ~(u64)0x100; /* ignore ignne emulation enable */
Nicolae Mogoreanua223c312012-02-21 13:44:21 -08002062 data &= ~(u64)0x8; /* ignore TLB cache disable */
Matthias Lange22d48b2d2014-06-26 13:50:15 +02002063 data &= ~(u64)0x40000; /* ignore Mc status write enable */
Andre Przywara8f1589d2009-06-24 12:44:33 +02002064 if (data != 0) {
Christoffer Dalla737f252012-06-03 21:17:48 +03002065 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2066 data);
Andre Przywara8f1589d2009-06-24 12:44:33 +02002067 return 1;
2068 }
Carsten Otte15c4a642007-10-30 18:44:17 +01002069 break;
Andre Przywaraf7c6d142009-07-02 15:04:14 +02002070 case MSR_FAM10H_MMIO_CONF_BASE:
2071 if (data != 0) {
Christoffer Dalla737f252012-06-03 21:17:48 +03002072 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2073 "0x%llx\n", data);
Andre Przywaraf7c6d142009-07-02 15:04:14 +02002074 return 1;
2075 }
Carsten Otte15c4a642007-10-30 18:44:17 +01002076 break;
Alexander Grafb5e2fec2008-07-22 08:00:45 +02002077 case MSR_IA32_DEBUGCTLMSR:
2078 if (!data) {
2079 /* We support the non-activated case already */
2080 break;
2081 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2082 /* Values other than LBR and BTF are vendor-specific,
2083 thus reserved and should throw a #GP */
2084 return 1;
2085 }
Christoffer Dalla737f252012-06-03 21:17:48 +03002086 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2087 __func__, data);
Alexander Grafb5e2fec2008-07-22 08:00:45 +02002088 break;
Avi Kivity9ba075a2008-05-26 20:06:35 +03002089 case 0x200 ... 0x2ff:
Xiao Guangrongff536042015-06-15 16:55:22 +08002090 return kvm_mtrr_set_msr(vcpu, msr, data);
Carsten Otte15c4a642007-10-30 18:44:17 +01002091 case MSR_IA32_APICBASE:
Jan Kiszka58cb6282014-01-24 16:48:44 +01002092 return kvm_set_apic_base(vcpu, msr_info);
Gleb Natapov0105d1a2009-07-05 17:39:36 +03002093 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2094 return kvm_x2apic_msr_write(vcpu, msr, data);
Liu, Jinsonga3e06bb2011-09-22 16:55:52 +08002095 case MSR_IA32_TSCDEADLINE:
2096 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2097 break;
Will Auldba904632012-11-29 12:42:50 -08002098 case MSR_IA32_TSC_ADJUST:
2099 if (guest_cpuid_has_tsc_adjust(vcpu)) {
2100 if (!msr_info->host_initiated) {
Chris J Argesd913b902014-11-12 21:00:39 -06002101 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
Haozhong Zhangd7add052015-08-07 11:24:32 +08002102 adjust_tsc_offset_guest(vcpu, adj);
Will Auldba904632012-11-29 12:42:50 -08002103 }
2104 vcpu->arch.ia32_tsc_adjust_msr = data;
2105 }
2106 break;
Carsten Otte15c4a642007-10-30 18:44:17 +01002107 case MSR_IA32_MISC_ENABLE:
Zhang Xiantaoad312c72007-12-13 23:50:52 +08002108 vcpu->arch.ia32_misc_enable_msr = data;
Carsten Otte15c4a642007-10-30 18:44:17 +01002109 break;
Paolo Bonzini64d60672015-05-07 11:36:11 +02002110 case MSR_IA32_SMBASE:
2111 if (!msr_info->host_initiated)
2112 return 1;
2113 vcpu->arch.smbase = data;
2114 break;
Glauber Costa11c6bff2010-05-11 12:17:41 -04002115 case MSR_KVM_WALL_CLOCK_NEW:
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02002116 case MSR_KVM_WALL_CLOCK:
2117 vcpu->kvm->arch.wall_clock = data;
2118 kvm_write_wall_clock(vcpu->kvm, data);
2119 break;
Glauber Costa11c6bff2010-05-11 12:17:41 -04002120 case MSR_KVM_SYSTEM_TIME_NEW:
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02002121 case MSR_KVM_SYSTEM_TIME: {
Andy Honig0b794592013-02-20 14:48:10 -08002122 u64 gpa_offset;
Marcelo Tosatti54750f22015-01-20 15:54:52 -02002123 struct kvm_arch *ka = &vcpu->kvm->arch;
2124
Glauber Costa12f9a482011-02-01 14:16:40 -05002125 kvmclock_reset(vcpu);
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02002126
Marcelo Tosatti54750f22015-01-20 15:54:52 -02002127 if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2128 bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2129
2130 if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2131 set_bit(KVM_REQ_MASTERCLOCK_UPDATE,
2132 &vcpu->requests);
2133
2134 ka->boot_vcpu_runs_old_kvmclock = tmp;
2135 }
2136
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02002137 vcpu->arch.time = data;
Marcelo Tosatti0061d532013-05-09 20:21:41 -03002138 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02002139
2140 /* we verify if the enable bit is set... */
2141 if (!(data & 1))
2142 break;
2143
Andy Honig0b794592013-02-20 14:48:10 -08002144 gpa_offset = data & ~(PAGE_MASK | 1);
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02002145
Andy Honig0b794592013-02-20 14:48:10 -08002146 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
Andrew Honig8f964522013-03-29 09:35:21 -07002147 &vcpu->arch.pv_time, data & ~1ULL,
2148 sizeof(struct pvclock_vcpu_time_info)))
Andy Honig0b794592013-02-20 14:48:10 -08002149 vcpu->arch.pv_time_enabled = false;
2150 else
2151 vcpu->arch.pv_time_enabled = true;
Xiao Guangrong32cad842012-08-03 15:42:52 +08002152
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02002153 break;
2154 }
Gleb Natapov344d9582010-10-14 11:22:50 +02002155 case MSR_KVM_ASYNC_PF_EN:
2156 if (kvm_pv_enable_async_pf(vcpu, data))
2157 return 1;
2158 break;
Glauber Costac9aaa892011-07-11 15:28:14 -04002159 case MSR_KVM_STEAL_TIME:
2160
2161 if (unlikely(!sched_info_on()))
2162 return 1;
2163
2164 if (data & KVM_STEAL_RESERVED_MASK)
2165 return 1;
2166
2167 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
Andrew Honig8f964522013-03-29 09:35:21 -07002168 data & KVM_STEAL_VALID_BITS,
2169 sizeof(struct kvm_steal_time)))
Glauber Costac9aaa892011-07-11 15:28:14 -04002170 return 1;
2171
2172 vcpu->arch.st.msr_val = data;
2173
2174 if (!(data & KVM_MSR_ENABLED))
2175 break;
2176
Glauber Costac9aaa892011-07-11 15:28:14 -04002177 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2178
2179 break;
Michael S. Tsirkinae7a2a32012-06-24 19:25:07 +03002180 case MSR_KVM_PV_EOI_EN:
2181 if (kvm_lapic_enable_pv_eoi(vcpu, data))
2182 return 1;
2183 break;
Glauber Costac9aaa892011-07-11 15:28:14 -04002184
Huang Ying890ca9a2009-05-11 16:48:15 +08002185 case MSR_IA32_MCG_CTL:
2186 case MSR_IA32_MCG_STATUS:
Chen Yucong81760dc2014-09-23 10:44:35 +08002187 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
Huang Ying890ca9a2009-05-11 16:48:15 +08002188 return set_msr_mce(vcpu, msr, data);
Andre Przywara71db6022009-06-12 22:01:29 +02002189
Wei Huang6912ac32015-06-12 01:34:56 -04002190 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2191 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2192 pr = true; /* fall through */
2193 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2194 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
Wei Huangc6702c92015-06-19 13:44:45 +02002195 if (kvm_pmu_is_valid_msr(vcpu, msr))
Paolo Bonziniafd80d82013-03-28 17:18:35 +01002196 return kvm_pmu_set_msr(vcpu, msr_info);
Gleb Natapov57537852012-01-15 14:17:22 +02002197
2198 if (pr || data != 0)
Christoffer Dalla737f252012-06-03 21:17:48 +03002199 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2200 "0x%x data 0x%llx\n", msr, data);
Gleb Natapov57537852012-01-15 14:17:22 +02002201 break;
Jes Sorensen84e0cef2010-09-01 11:42:04 +02002202 case MSR_K7_CLK_CTL:
2203 /*
2204 * Ignore all writes to this no longer documented MSR.
2205 * Writes are only relevant for old K7 processors,
2206 * all pre-dating SVM, but a recommended workaround from
Guo Chao4a969982012-06-28 15:17:27 +08002207 * AMD for these chips. It is possible to specify the
Jes Sorensen84e0cef2010-09-01 11:42:04 +02002208 * affected processor models on the command line, hence
2209 * the need to ignore the workaround.
2210 */
2211 break;
Gleb Natapov55cd8e52010-01-17 15:51:22 +02002212 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
Andrey Smetanine7d95132015-07-03 15:01:37 +03002213 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2214 case HV_X64_MSR_CRASH_CTL:
Andrey Smetanin1f4b34f2015-11-30 19:22:21 +03002215 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
Andrey Smetanine7d95132015-07-03 15:01:37 +03002216 return kvm_hv_set_msr_common(vcpu, msr, data,
2217 msr_info->host_initiated);
john cooper91c9c3e2011-01-21 00:21:00 -05002218 case MSR_IA32_BBL_CR_CTL3:
2219 /* Drop writes to this legacy MSR -- see rdmsr
2220 * counterpart for further detail.
2221 */
Christoffer Dalla737f252012-06-03 21:17:48 +03002222 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
john cooper91c9c3e2011-01-21 00:21:00 -05002223 break;
Boris Ostrovsky2b036c62012-01-09 14:00:35 -05002224 case MSR_AMD64_OSVW_ID_LENGTH:
2225 if (!guest_cpuid_has_osvw(vcpu))
2226 return 1;
2227 vcpu->arch.osvw.length = data;
2228 break;
2229 case MSR_AMD64_OSVW_STATUS:
2230 if (!guest_cpuid_has_osvw(vcpu))
2231 return 1;
2232 vcpu->arch.osvw.status = data;
2233 break;
Carsten Otte15c4a642007-10-30 18:44:17 +01002234 default:
Ed Swierkffde22a2009-10-15 15:21:43 -07002235 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2236 return xen_hvm_config(vcpu, data);
Wei Huangc6702c92015-06-19 13:44:45 +02002237 if (kvm_pmu_is_valid_msr(vcpu, msr))
Paolo Bonziniafd80d82013-03-28 17:18:35 +01002238 return kvm_pmu_set_msr(vcpu, msr_info);
Andre Przywaraed85c062009-06-25 12:36:49 +02002239 if (!ignore_msrs) {
Christoffer Dalla737f252012-06-03 21:17:48 +03002240 vcpu_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
2241 msr, data);
Andre Przywaraed85c062009-06-25 12:36:49 +02002242 return 1;
2243 } else {
Christoffer Dalla737f252012-06-03 21:17:48 +03002244 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
2245 msr, data);
Andre Przywaraed85c062009-06-25 12:36:49 +02002246 break;
2247 }
Carsten Otte15c4a642007-10-30 18:44:17 +01002248 }
2249 return 0;
2250}
2251EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2252
2253
2254/*
2255 * Reads an msr value (of 'msr_index') into 'pdata'.
2256 * Returns 0 on success, non-0 otherwise.
2257 * Assumes vcpu_load() was already called.
2258 */
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002259int kvm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
Carsten Otte15c4a642007-10-30 18:44:17 +01002260{
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002261 return kvm_x86_ops->get_msr(vcpu, msr);
Carsten Otte15c4a642007-10-30 18:44:17 +01002262}
Wincy Vanff651cb2014-12-11 08:52:58 +03002263EXPORT_SYMBOL_GPL(kvm_get_msr);
Carsten Otte15c4a642007-10-30 18:44:17 +01002264
Huang Ying890ca9a2009-05-11 16:48:15 +08002265static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2266{
2267 u64 data;
2268 u64 mcg_cap = vcpu->arch.mcg_cap;
2269 unsigned bank_num = mcg_cap & 0xff;
2270
2271 switch (msr) {
2272 case MSR_IA32_P5_MC_ADDR:
2273 case MSR_IA32_P5_MC_TYPE:
2274 data = 0;
2275 break;
2276 case MSR_IA32_MCG_CAP:
2277 data = vcpu->arch.mcg_cap;
2278 break;
2279 case MSR_IA32_MCG_CTL:
2280 if (!(mcg_cap & MCG_CTL_P))
2281 return 1;
2282 data = vcpu->arch.mcg_ctl;
2283 break;
2284 case MSR_IA32_MCG_STATUS:
2285 data = vcpu->arch.mcg_status;
2286 break;
2287 default:
2288 if (msr >= MSR_IA32_MC0_CTL &&
Chen Yucong81760dc2014-09-23 10:44:35 +08002289 msr < MSR_IA32_MCx_CTL(bank_num)) {
Huang Ying890ca9a2009-05-11 16:48:15 +08002290 u32 offset = msr - MSR_IA32_MC0_CTL;
2291 data = vcpu->arch.mce_banks[offset];
2292 break;
2293 }
2294 return 1;
2295 }
2296 *pdata = data;
2297 return 0;
2298}
2299
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002300int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
Carsten Otte15c4a642007-10-30 18:44:17 +01002301{
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002302 switch (msr_info->index) {
Carsten Otte15c4a642007-10-30 18:44:17 +01002303 case MSR_IA32_PLATFORM_ID:
Carsten Otte15c4a642007-10-30 18:44:17 +01002304 case MSR_IA32_EBL_CR_POWERON:
Alexander Grafb5e2fec2008-07-22 08:00:45 +02002305 case MSR_IA32_DEBUGCTLMSR:
2306 case MSR_IA32_LASTBRANCHFROMIP:
2307 case MSR_IA32_LASTBRANCHTOIP:
2308 case MSR_IA32_LASTINTFROMIP:
2309 case MSR_IA32_LASTINTTOIP:
Jaswinder Singh Rajput60af2ec2009-05-14 11:00:10 +05302310 case MSR_K8_SYSCFG:
Paolo Bonzini3afb1122015-09-18 17:33:04 +02002311 case MSR_K8_TSEG_ADDR:
2312 case MSR_K8_TSEG_MASK:
Jaswinder Singh Rajput60af2ec2009-05-14 11:00:10 +05302313 case MSR_K7_HWCR:
Avi Kivity61a6bd62008-12-29 17:32:28 +02002314 case MSR_VM_HSAVE_PA:
Andre Przywara1fdbd482009-06-24 12:44:34 +02002315 case MSR_K8_INT_PENDING_MSG:
Andre Przywarac323c0e2009-06-24 15:37:05 +02002316 case MSR_AMD64_NB_CFG:
Andre Przywaraf7c6d142009-07-02 15:04:14 +02002317 case MSR_FAM10H_MMIO_CONF_BASE:
Borislav Petkov2e32b712013-02-19 19:33:13 +01002318 case MSR_AMD64_BU_CFG2:
Dmitry Bilunov0c2df2a2016-05-31 17:38:24 +03002319 case MSR_IA32_PERF_CTL:
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002320 msr_info->data = 0;
Carsten Otte15c4a642007-10-30 18:44:17 +01002321 break;
Wei Huang6912ac32015-06-12 01:34:56 -04002322 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2323 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2324 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2325 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
Wei Huangc6702c92015-06-19 13:44:45 +02002326 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002327 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2328 msr_info->data = 0;
Gleb Natapov57537852012-01-15 14:17:22 +02002329 break;
Marcelo Tosatti742bc672011-07-29 19:44:21 -03002330 case MSR_IA32_UCODE_REV:
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002331 msr_info->data = 0x100000000ULL;
Marcelo Tosatti742bc672011-07-29 19:44:21 -03002332 break;
Avi Kivity9ba075a2008-05-26 20:06:35 +03002333 case MSR_MTRRcap:
Avi Kivity9ba075a2008-05-26 20:06:35 +03002334 case 0x200 ... 0x2ff:
Xiao Guangrongff536042015-06-15 16:55:22 +08002335 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
Carsten Otte15c4a642007-10-30 18:44:17 +01002336 case 0xcd: /* fsb frequency */
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002337 msr_info->data = 3;
Carsten Otte15c4a642007-10-30 18:44:17 +01002338 break;
Jes Sorensen7b914092010-09-09 12:06:46 +02002339 /*
2340 * MSR_EBC_FREQUENCY_ID
2341 * Conservative value valid for even the basic CPU models.
2342 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2343 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2344 * and 266MHz for model 3, or 4. Set Core Clock
2345 * Frequency to System Bus Frequency Ratio to 1 (bits
2346 * 31:24) even though these are only valid for CPU
2347 * models > 2, however guests may end up dividing or
2348 * multiplying by zero otherwise.
2349 */
2350 case MSR_EBC_FREQUENCY_ID:
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002351 msr_info->data = 1 << 24;
Jes Sorensen7b914092010-09-09 12:06:46 +02002352 break;
Carsten Otte15c4a642007-10-30 18:44:17 +01002353 case MSR_IA32_APICBASE:
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002354 msr_info->data = kvm_get_apic_base(vcpu);
Carsten Otte15c4a642007-10-30 18:44:17 +01002355 break;
Gleb Natapov0105d1a2009-07-05 17:39:36 +03002356 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002357 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
Gleb Natapov0105d1a2009-07-05 17:39:36 +03002358 break;
Liu, Jinsonga3e06bb2011-09-22 16:55:52 +08002359 case MSR_IA32_TSCDEADLINE:
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002360 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
Liu, Jinsonga3e06bb2011-09-22 16:55:52 +08002361 break;
Will Auldba904632012-11-29 12:42:50 -08002362 case MSR_IA32_TSC_ADJUST:
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002363 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
Will Auldba904632012-11-29 12:42:50 -08002364 break;
Carsten Otte15c4a642007-10-30 18:44:17 +01002365 case MSR_IA32_MISC_ENABLE:
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002366 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
Carsten Otte15c4a642007-10-30 18:44:17 +01002367 break;
Paolo Bonzini64d60672015-05-07 11:36:11 +02002368 case MSR_IA32_SMBASE:
2369 if (!msr_info->host_initiated)
2370 return 1;
2371 msr_info->data = vcpu->arch.smbase;
Carsten Otte15c4a642007-10-30 18:44:17 +01002372 break;
Alexander Graf847f0ad2008-02-21 12:11:01 +01002373 case MSR_IA32_PERF_STATUS:
2374 /* TSC increment by tick */
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002375 msr_info->data = 1000ULL;
Alexander Graf847f0ad2008-02-21 12:11:01 +01002376 /* CPU multiplier */
Nicolas Ioossb0996ae2015-06-29 18:39:23 +08002377 msr_info->data |= (((uint64_t)4ULL) << 40);
Alexander Graf847f0ad2008-02-21 12:11:01 +01002378 break;
Carsten Otte15c4a642007-10-30 18:44:17 +01002379 case MSR_EFER:
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002380 msr_info->data = vcpu->arch.efer;
Carsten Otte15c4a642007-10-30 18:44:17 +01002381 break;
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02002382 case MSR_KVM_WALL_CLOCK:
Glauber Costa11c6bff2010-05-11 12:17:41 -04002383 case MSR_KVM_WALL_CLOCK_NEW:
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002384 msr_info->data = vcpu->kvm->arch.wall_clock;
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02002385 break;
2386 case MSR_KVM_SYSTEM_TIME:
Glauber Costa11c6bff2010-05-11 12:17:41 -04002387 case MSR_KVM_SYSTEM_TIME_NEW:
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002388 msr_info->data = vcpu->arch.time;
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02002389 break;
Gleb Natapov344d9582010-10-14 11:22:50 +02002390 case MSR_KVM_ASYNC_PF_EN:
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002391 msr_info->data = vcpu->arch.apf.msr_val;
Gleb Natapov344d9582010-10-14 11:22:50 +02002392 break;
Glauber Costac9aaa892011-07-11 15:28:14 -04002393 case MSR_KVM_STEAL_TIME:
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002394 msr_info->data = vcpu->arch.st.msr_val;
Glauber Costac9aaa892011-07-11 15:28:14 -04002395 break;
Michael S. Tsirkin1d921282012-08-26 18:00:29 +03002396 case MSR_KVM_PV_EOI_EN:
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002397 msr_info->data = vcpu->arch.pv_eoi.msr_val;
Michael S. Tsirkin1d921282012-08-26 18:00:29 +03002398 break;
Huang Ying890ca9a2009-05-11 16:48:15 +08002399 case MSR_IA32_P5_MC_ADDR:
2400 case MSR_IA32_P5_MC_TYPE:
2401 case MSR_IA32_MCG_CAP:
2402 case MSR_IA32_MCG_CTL:
2403 case MSR_IA32_MCG_STATUS:
Chen Yucong81760dc2014-09-23 10:44:35 +08002404 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002405 return get_msr_mce(vcpu, msr_info->index, &msr_info->data);
Jes Sorensen84e0cef2010-09-01 11:42:04 +02002406 case MSR_K7_CLK_CTL:
2407 /*
2408 * Provide expected ramp-up count for K7. All other
2409 * are set to zero, indicating minimum divisors for
2410 * every field.
2411 *
2412 * This prevents guest kernels on AMD host with CPU
2413 * type 6, model 8 and higher from exploding due to
2414 * the rdmsr failing.
2415 */
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002416 msr_info->data = 0x20000000;
Jes Sorensen84e0cef2010-09-01 11:42:04 +02002417 break;
Gleb Natapov55cd8e52010-01-17 15:51:22 +02002418 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
Andrey Smetanine7d95132015-07-03 15:01:37 +03002419 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2420 case HV_X64_MSR_CRASH_CTL:
Andrey Smetanin1f4b34f2015-11-30 19:22:21 +03002421 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
Andrey Smetanine83d5882015-07-03 15:01:34 +03002422 return kvm_hv_get_msr_common(vcpu,
2423 msr_info->index, &msr_info->data);
Gleb Natapov55cd8e52010-01-17 15:51:22 +02002424 break;
john cooper91c9c3e2011-01-21 00:21:00 -05002425 case MSR_IA32_BBL_CR_CTL3:
2426 /* This legacy MSR exists but isn't fully documented in current
2427 * silicon. It is however accessed by winxp in very narrow
2428 * scenarios where it sets bit #19, itself documented as
2429 * a "reserved" bit. Best effort attempt to source coherent
2430 * read data here should the balance of the register be
2431 * interpreted by the guest:
2432 *
2433 * L2 cache control register 3: 64GB range, 256KB size,
2434 * enabled, latency 0x1, configured
2435 */
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002436 msr_info->data = 0xbe702111;
john cooper91c9c3e2011-01-21 00:21:00 -05002437 break;
Boris Ostrovsky2b036c62012-01-09 14:00:35 -05002438 case MSR_AMD64_OSVW_ID_LENGTH:
2439 if (!guest_cpuid_has_osvw(vcpu))
2440 return 1;
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002441 msr_info->data = vcpu->arch.osvw.length;
Boris Ostrovsky2b036c62012-01-09 14:00:35 -05002442 break;
2443 case MSR_AMD64_OSVW_STATUS:
2444 if (!guest_cpuid_has_osvw(vcpu))
2445 return 1;
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002446 msr_info->data = vcpu->arch.osvw.status;
Boris Ostrovsky2b036c62012-01-09 14:00:35 -05002447 break;
Carsten Otte15c4a642007-10-30 18:44:17 +01002448 default:
Wei Huangc6702c92015-06-19 13:44:45 +02002449 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002450 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
Andre Przywaraed85c062009-06-25 12:36:49 +02002451 if (!ignore_msrs) {
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002452 vcpu_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr_info->index);
Andre Przywaraed85c062009-06-25 12:36:49 +02002453 return 1;
2454 } else {
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002455 vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr_info->index);
2456 msr_info->data = 0;
Andre Przywaraed85c062009-06-25 12:36:49 +02002457 }
2458 break;
Carsten Otte15c4a642007-10-30 18:44:17 +01002459 }
Carsten Otte15c4a642007-10-30 18:44:17 +01002460 return 0;
2461}
2462EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2463
Carsten Otte313a3dc2007-10-11 19:16:52 +02002464/*
2465 * Read or write a bunch of msrs. All parameters are kernel addresses.
2466 *
2467 * @return number of msrs set successfully.
2468 */
2469static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2470 struct kvm_msr_entry *entries,
2471 int (*do_msr)(struct kvm_vcpu *vcpu,
2472 unsigned index, u64 *data))
2473{
Marcelo Tosattif656ce02009-12-23 14:35:25 -02002474 int i, idx;
Carsten Otte313a3dc2007-10-11 19:16:52 +02002475
Marcelo Tosattif656ce02009-12-23 14:35:25 -02002476 idx = srcu_read_lock(&vcpu->kvm->srcu);
Carsten Otte313a3dc2007-10-11 19:16:52 +02002477 for (i = 0; i < msrs->nmsrs; ++i)
2478 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2479 break;
Marcelo Tosattif656ce02009-12-23 14:35:25 -02002480 srcu_read_unlock(&vcpu->kvm->srcu, idx);
Carsten Otte313a3dc2007-10-11 19:16:52 +02002481
Carsten Otte313a3dc2007-10-11 19:16:52 +02002482 return i;
2483}
2484
2485/*
2486 * Read or write a bunch of msrs. Parameters are user addresses.
2487 *
2488 * @return number of msrs set successfully.
2489 */
2490static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2491 int (*do_msr)(struct kvm_vcpu *vcpu,
2492 unsigned index, u64 *data),
2493 int writeback)
2494{
2495 struct kvm_msrs msrs;
2496 struct kvm_msr_entry *entries;
2497 int r, n;
2498 unsigned size;
2499
2500 r = -EFAULT;
2501 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2502 goto out;
2503
2504 r = -E2BIG;
2505 if (msrs.nmsrs >= MAX_IO_MSRS)
2506 goto out;
2507
Carsten Otte313a3dc2007-10-11 19:16:52 +02002508 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002509 entries = memdup_user(user_msrs->entries, size);
2510 if (IS_ERR(entries)) {
2511 r = PTR_ERR(entries);
Carsten Otte313a3dc2007-10-11 19:16:52 +02002512 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002513 }
Carsten Otte313a3dc2007-10-11 19:16:52 +02002514
2515 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2516 if (r < 0)
2517 goto out_free;
2518
2519 r = -EFAULT;
2520 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2521 goto out_free;
2522
2523 r = n;
2524
2525out_free:
Avi Kivity7a73c022010-07-22 23:24:52 +03002526 kfree(entries);
Carsten Otte313a3dc2007-10-11 19:16:52 +02002527out:
2528 return r;
2529}
2530
Alexander Graf784aa3d2014-07-14 18:27:35 +02002531int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
Zhang Xiantao018d00d2007-11-15 23:07:47 +08002532{
2533 int r;
2534
2535 switch (ext) {
2536 case KVM_CAP_IRQCHIP:
2537 case KVM_CAP_HLT:
2538 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
Zhang Xiantao018d00d2007-11-15 23:07:47 +08002539 case KVM_CAP_SET_TSS_ADDR:
Dan Kenigsberg07716712007-11-21 17:10:04 +02002540 case KVM_CAP_EXT_CPUID:
Borislav Petkov9c15bb12013-09-22 16:44:50 +02002541 case KVM_CAP_EXT_EMUL_CPUID:
Gerd Hoffmannc8076602009-02-04 17:52:04 +01002542 case KVM_CAP_CLOCKSOURCE:
Sheng Yang78376992008-01-28 05:10:22 +08002543 case KVM_CAP_PIT:
Marcelo Tosattia28e4f52008-02-22 12:21:36 -05002544 case KVM_CAP_NOP_IO_DELAY:
Marcelo Tosatti62d9f0d2008-04-11 13:24:45 -03002545 case KVM_CAP_MP_STATE:
Avi Kivityed848622008-07-29 11:30:57 +03002546 case KVM_CAP_SYNC_MMU:
Lai Jiangshana355c852010-12-14 17:57:47 +08002547 case KVM_CAP_USER_NMI:
Marcelo Tosatti52d939a2008-12-30 15:55:06 -02002548 case KVM_CAP_REINJECT_CONTROL:
Gleb Natapov49256632009-02-04 17:28:14 +02002549 case KVM_CAP_IRQ_INJECT_STATUS:
Gregory Haskinsd34e6b12009-07-07 17:08:49 -04002550 case KVM_CAP_IOEVENTFD:
Michael S. Tsirkinf848a5a2014-03-31 21:50:38 +03002551 case KVM_CAP_IOEVENTFD_NO_LENGTH:
Jan Kiszkac5ff41c2009-05-14 22:42:53 +02002552 case KVM_CAP_PIT2:
Beth Kone9f42752009-07-07 11:50:38 -04002553 case KVM_CAP_PIT_STATE2:
Sheng Yangb927a3c2009-07-21 10:42:48 +08002554 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
Ed Swierkffde22a2009-10-15 15:21:43 -07002555 case KVM_CAP_XEN_HVM:
Glauber Costaafbcf7a2009-10-16 15:28:36 -04002556 case KVM_CAP_ADJUST_CLOCK:
Jan Kiszka3cfc3092009-11-12 01:04:25 +01002557 case KVM_CAP_VCPU_EVENTS:
Gleb Natapov55cd8e52010-01-17 15:51:22 +02002558 case KVM_CAP_HYPERV:
Gleb Natapov10388a02010-01-17 15:51:23 +02002559 case KVM_CAP_HYPERV_VAPIC:
Gleb Natapovc25bc162010-01-17 15:51:24 +02002560 case KVM_CAP_HYPERV_SPIN:
Andrey Smetanin5c9194122015-11-10 15:36:34 +03002561 case KVM_CAP_HYPERV_SYNIC:
Zhai, Edwinab9f4ec2010-01-29 14:38:44 +08002562 case KVM_CAP_PCI_SEGMENT:
Jan Kiszkaa1efbe72010-02-15 10:45:43 +01002563 case KVM_CAP_DEBUGREGS:
Jan Kiszkad2be1652010-02-23 17:47:57 +01002564 case KVM_CAP_X86_ROBUST_SINGLESTEP:
Sheng Yang2d5b5a62010-06-13 17:29:39 +08002565 case KVM_CAP_XSAVE:
Gleb Natapov344d9582010-10-14 11:22:50 +02002566 case KVM_CAP_ASYNC_PF:
Joerg Roedel92a1f122011-03-25 09:44:51 +01002567 case KVM_CAP_GET_TSC_KHZ:
Eric B Munson1c0b28c2012-03-10 14:37:27 -05002568 case KVM_CAP_KVMCLOCK_CTRL:
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08002569 case KVM_CAP_READONLY_MEM:
Paolo Bonzini5f66b622014-01-29 18:10:45 +01002570 case KVM_CAP_HYPERV_TIME:
Gabriel L. Somlo100943c2014-02-27 23:06:17 -05002571 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
Radim Krčmářdefcf512015-01-08 15:59:30 +01002572 case KVM_CAP_TSC_DEADLINE_TIMER:
Nadav Amit90de4a12015-04-13 01:53:41 +03002573 case KVM_CAP_ENABLE_CAP_VM:
2574 case KVM_CAP_DISABLE_QUIRKS:
Paolo Bonzinid71ba782015-07-29 11:56:48 +02002575 case KVM_CAP_SET_BOOT_CPU_ID:
Steve Rutherford49df6392015-07-29 23:21:40 -07002576 case KVM_CAP_SPLIT_IRQCHIP:
Alex Williamson2a5bab12013-04-16 13:49:18 -06002577#ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2578 case KVM_CAP_ASSIGN_DEV_IRQ:
2579 case KVM_CAP_PCI_2_3:
2580#endif
Zhang Xiantao018d00d2007-11-15 23:07:47 +08002581 r = 1;
2582 break;
Paolo Bonzini6d396b52015-04-01 14:25:33 +02002583 case KVM_CAP_X86_SMM:
2584 /* SMBASE is usually relocated above 1M on modern chipsets,
2585 * and SMM handlers might indeed rely on 4G segment limits,
2586 * so do not report SMM to be available if real mode is
2587 * emulated via vm86 mode. Still, do not go to great lengths
2588 * to avoid userspace's usage of the feature, because it is a
2589 * fringe case that is not enabled except via specific settings
2590 * of the module parameters.
2591 */
2592 r = kvm_x86_ops->cpu_has_high_real_mode_segbase();
2593 break;
Laurent Vivier542472b2008-05-30 16:05:55 +02002594 case KVM_CAP_COALESCED_MMIO:
2595 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2596 break;
Avi Kivity774ead32007-12-26 13:57:04 +02002597 case KVM_CAP_VAPIC:
2598 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2599 break;
Avi Kivityf7252302008-02-20 11:53:16 +02002600 case KVM_CAP_NR_VCPUS:
Sasha Levin8c3ba332011-07-18 17:17:15 +03002601 r = KVM_SOFT_MAX_VCPUS;
2602 break;
2603 case KVM_CAP_MAX_VCPUS:
Avi Kivityf7252302008-02-20 11:53:16 +02002604 r = KVM_MAX_VCPUS;
2605 break;
Avi Kivitya988b912008-02-20 11:59:20 +02002606 case KVM_CAP_NR_MEMSLOTS:
Alex Williamsonbbacc0c2012-12-10 10:33:09 -07002607 r = KVM_USER_MEM_SLOTS;
Avi Kivitya988b912008-02-20 11:59:20 +02002608 break;
Marcelo Tosattia68a6a72009-10-01 19:28:39 -03002609 case KVM_CAP_PV_MMU: /* obsolete */
2610 r = 0;
Marcelo Tosatti2f333bc2008-02-22 12:21:37 -05002611 break;
Alex Williamson4cee4b72013-04-29 10:54:08 -06002612#ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
Ben-Ami Yassour62c476c2008-09-14 03:48:28 +03002613 case KVM_CAP_IOMMU:
Joerg Roedela1b60c12011-09-06 18:46:34 +02002614 r = iommu_present(&pci_bus_type);
Ben-Ami Yassour62c476c2008-09-14 03:48:28 +03002615 break;
Alex Williamson4cee4b72013-04-29 10:54:08 -06002616#endif
Huang Ying890ca9a2009-05-11 16:48:15 +08002617 case KVM_CAP_MCE:
2618 r = KVM_MAX_MCE_BANKS;
2619 break;
Sheng Yang2d5b5a62010-06-13 17:29:39 +08002620 case KVM_CAP_XCRS:
Borislav Petkovd366bf72016-04-04 22:25:02 +02002621 r = boot_cpu_has(X86_FEATURE_XSAVE);
Sheng Yang2d5b5a62010-06-13 17:29:39 +08002622 break;
Joerg Roedel92a1f122011-03-25 09:44:51 +01002623 case KVM_CAP_TSC_CONTROL:
2624 r = kvm_has_tsc_control;
2625 break;
Radim Krčmář371313132016-07-12 22:09:27 +02002626 case KVM_CAP_X2APIC_API:
2627 r = KVM_X2APIC_API_VALID_FLAGS;
2628 break;
Zhang Xiantao018d00d2007-11-15 23:07:47 +08002629 default:
2630 r = 0;
2631 break;
2632 }
2633 return r;
2634
2635}
2636
Carsten Otte043405e2007-10-10 17:16:19 +02002637long kvm_arch_dev_ioctl(struct file *filp,
2638 unsigned int ioctl, unsigned long arg)
2639{
2640 void __user *argp = (void __user *)arg;
2641 long r;
2642
2643 switch (ioctl) {
2644 case KVM_GET_MSR_INDEX_LIST: {
2645 struct kvm_msr_list __user *user_msr_list = argp;
2646 struct kvm_msr_list msr_list;
2647 unsigned n;
2648
2649 r = -EFAULT;
2650 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2651 goto out;
2652 n = msr_list.nmsrs;
Paolo Bonzini62ef68b2015-05-05 12:08:55 +02002653 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
Carsten Otte043405e2007-10-10 17:16:19 +02002654 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2655 goto out;
2656 r = -E2BIG;
Jan Kiszkae125e7b2009-07-02 21:45:47 +02002657 if (n < msr_list.nmsrs)
Carsten Otte043405e2007-10-10 17:16:19 +02002658 goto out;
2659 r = -EFAULT;
2660 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2661 num_msrs_to_save * sizeof(u32)))
2662 goto out;
Jan Kiszkae125e7b2009-07-02 21:45:47 +02002663 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
Carsten Otte043405e2007-10-10 17:16:19 +02002664 &emulated_msrs,
Paolo Bonzini62ef68b2015-05-05 12:08:55 +02002665 num_emulated_msrs * sizeof(u32)))
Carsten Otte043405e2007-10-10 17:16:19 +02002666 goto out;
2667 r = 0;
2668 break;
2669 }
Borislav Petkov9c15bb12013-09-22 16:44:50 +02002670 case KVM_GET_SUPPORTED_CPUID:
2671 case KVM_GET_EMULATED_CPUID: {
Avi Kivity674eea02008-02-11 18:37:23 +02002672 struct kvm_cpuid2 __user *cpuid_arg = argp;
2673 struct kvm_cpuid2 cpuid;
2674
2675 r = -EFAULT;
2676 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2677 goto out;
Borislav Petkov9c15bb12013-09-22 16:44:50 +02002678
2679 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
2680 ioctl);
Avi Kivity674eea02008-02-11 18:37:23 +02002681 if (r)
2682 goto out;
2683
2684 r = -EFAULT;
2685 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2686 goto out;
2687 r = 0;
2688 break;
2689 }
Huang Ying890ca9a2009-05-11 16:48:15 +08002690 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
Huang Ying890ca9a2009-05-11 16:48:15 +08002691 r = -EFAULT;
Ashok Rajc45dcc72016-06-22 14:59:56 +08002692 if (copy_to_user(argp, &kvm_mce_cap_supported,
2693 sizeof(kvm_mce_cap_supported)))
Huang Ying890ca9a2009-05-11 16:48:15 +08002694 goto out;
2695 r = 0;
2696 break;
2697 }
Carsten Otte043405e2007-10-10 17:16:19 +02002698 default:
2699 r = -EINVAL;
2700 }
2701out:
2702 return r;
2703}
2704
Sheng Yangf5f48ee2010-06-30 12:25:15 +08002705static void wbinvd_ipi(void *garbage)
2706{
2707 wbinvd();
2708}
2709
2710static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2711{
Alex Williamsone0f0bbc2013-10-30 11:02:30 -06002712 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
Sheng Yangf5f48ee2010-06-30 12:25:15 +08002713}
2714
Paolo Bonzini2860c4b2016-01-07 15:05:10 +01002715static inline void kvm_migrate_timers(struct kvm_vcpu *vcpu)
2716{
2717 set_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests);
2718}
2719
Carsten Otte313a3dc2007-10-11 19:16:52 +02002720void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2721{
Sheng Yangf5f48ee2010-06-30 12:25:15 +08002722 /* Address WBINVD may be executed by guest */
2723 if (need_emulate_wbinvd(vcpu)) {
2724 if (kvm_x86_ops->has_wbinvd_exit())
2725 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2726 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2727 smp_call_function_single(vcpu->cpu,
2728 wbinvd_ipi, NULL, 1);
2729 }
2730
Carsten Otte313a3dc2007-10-11 19:16:52 +02002731 kvm_x86_ops->vcpu_load(vcpu, cpu);
Zachary Amsden0dd6a6e2012-02-03 15:43:56 -02002732
2733 /* Apply any externally detected TSC adjustments (due to suspend) */
2734 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
2735 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
2736 vcpu->arch.tsc_offset_adjustment = 0;
Guo Hui Liu105b21b2014-09-12 13:43:19 +08002737 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
Zachary Amsden0dd6a6e2012-02-03 15:43:56 -02002738 }
2739
Zachary Amsden48434c202010-08-19 22:07:24 -10002740 if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
Zachary Amsden6f526ec2012-02-03 15:43:54 -02002741 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
Andy Lutomirski4ea16362015-06-25 18:44:07 +02002742 rdtsc() - vcpu->arch.last_host_tsc;
Zachary Amsdene48672f2010-08-19 22:07:23 -10002743 if (tsc_delta < 0)
2744 mark_tsc_unstable("KVM discovered backwards TSC");
Yunhong Jiangce7a0582016-06-13 14:20:01 -07002745
Zachary Amsdenc2855452010-09-18 14:38:15 -10002746 if (check_tsc_unstable()) {
Haozhong Zhang07c14192015-10-20 15:39:05 +08002747 u64 offset = kvm_compute_tsc_offset(vcpu,
Zachary Amsdenb183aa52012-02-03 15:43:53 -02002748 vcpu->arch.last_guest_tsc);
2749 kvm_x86_ops->write_tsc_offset(vcpu, offset);
Zachary Amsdenc2855452010-09-18 14:38:15 -10002750 vcpu->arch.tsc_catchup = 1;
Zachary Amsdenc2855452010-09-18 14:38:15 -10002751 }
Wanpeng Lie12c8f32016-08-30 16:14:00 +08002752 if (kvm_lapic_hv_timer_in_use(vcpu) &&
2753 kvm_x86_ops->set_hv_timer(vcpu,
2754 kvm_get_lapic_tscdeadline_msr(vcpu)))
2755 kvm_lapic_switch_to_sw_timer(vcpu);
Marcelo Tosattid98d07c2012-11-27 23:29:04 -02002756 /*
2757 * On a host with synchronized TSC, there is no need to update
2758 * kvmclock on vcpu->cpu migration
2759 */
2760 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
Marcelo Tosatti0061d532013-05-09 20:21:41 -03002761 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
Zachary Amsdenc2855452010-09-18 14:38:15 -10002762 if (vcpu->cpu != cpu)
2763 kvm_migrate_timers(vcpu);
Zachary Amsdene48672f2010-08-19 22:07:23 -10002764 vcpu->cpu = cpu;
Zachary Amsden6b7d7e72009-10-09 16:26:08 -10002765 }
Glauber Costac9aaa892011-07-11 15:28:14 -04002766
Glauber Costac9aaa892011-07-11 15:28:14 -04002767 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
Carsten Otte313a3dc2007-10-11 19:16:52 +02002768}
2769
2770void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2771{
Avi Kivity02daab22009-12-30 12:40:26 +02002772 kvm_x86_ops->vcpu_put(vcpu);
Avi Kivity1c11e712010-05-03 16:05:44 +03002773 kvm_put_guest_fpu(vcpu);
Andy Lutomirski4ea16362015-06-25 18:44:07 +02002774 vcpu->arch.last_host_tsc = rdtsc();
Carsten Otte313a3dc2007-10-11 19:16:52 +02002775}
2776
Carsten Otte313a3dc2007-10-11 19:16:52 +02002777static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2778 struct kvm_lapic_state *s)
2779{
Andrey Smetanind62caab2015-11-10 15:36:33 +03002780 if (vcpu->arch.apicv_active)
2781 kvm_x86_ops->sync_pir_to_irr(vcpu);
2782
Radim Krčmářa92e2542016-07-12 22:09:22 +02002783 return kvm_apic_get_state(vcpu, s);
Carsten Otte313a3dc2007-10-11 19:16:52 +02002784}
2785
2786static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2787 struct kvm_lapic_state *s)
2788{
Radim Krčmářa92e2542016-07-12 22:09:22 +02002789 int r;
2790
2791 r = kvm_apic_set_state(vcpu, s);
2792 if (r)
2793 return r;
Gleb Natapovcb142eb2009-08-09 15:17:40 +03002794 update_cr8_intercept(vcpu);
Carsten Otte313a3dc2007-10-11 19:16:52 +02002795
2796 return 0;
2797}
2798
Matt Gingell127a4572015-11-17 17:32:05 +01002799static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
2800{
2801 return (!lapic_in_kernel(vcpu) ||
2802 kvm_apic_accept_pic_intr(vcpu));
2803}
2804
Matt Gingell782d4222015-11-16 15:26:00 -08002805/*
2806 * if userspace requested an interrupt window, check that the
2807 * interrupt window is open.
2808 *
2809 * No need to exit to userspace if we already have an interrupt queued.
2810 */
2811static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
2812{
2813 return kvm_arch_interrupt_allowed(vcpu) &&
2814 !kvm_cpu_has_interrupt(vcpu) &&
2815 !kvm_event_needs_reinjection(vcpu) &&
2816 kvm_cpu_accept_dm_intr(vcpu);
2817}
2818
Zhang Xiantaof77bc6a2007-11-21 04:36:41 +08002819static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2820 struct kvm_interrupt *irq)
2821{
Chen Gang02cdb502013-02-27 11:33:25 +08002822 if (irq->irq >= KVM_NR_INTERRUPTS)
Zhang Xiantaof77bc6a2007-11-21 04:36:41 +08002823 return -EINVAL;
Steve Rutherford1c1a9ce2015-07-30 11:27:16 +02002824
2825 if (!irqchip_in_kernel(vcpu->kvm)) {
2826 kvm_queue_interrupt(vcpu, irq->irq, false);
2827 kvm_make_request(KVM_REQ_EVENT, vcpu);
2828 return 0;
2829 }
2830
2831 /*
2832 * With in-kernel LAPIC, we only use this to inject EXTINT, so
2833 * fail for in-kernel 8259.
2834 */
2835 if (pic_in_kernel(vcpu->kvm))
Zhang Xiantaof77bc6a2007-11-21 04:36:41 +08002836 return -ENXIO;
Zhang Xiantaof77bc6a2007-11-21 04:36:41 +08002837
Steve Rutherford1c1a9ce2015-07-30 11:27:16 +02002838 if (vcpu->arch.pending_external_vector != -1)
2839 return -EEXIST;
Zhang Xiantaof77bc6a2007-11-21 04:36:41 +08002840
Steve Rutherford1c1a9ce2015-07-30 11:27:16 +02002841 vcpu->arch.pending_external_vector = irq->irq;
Matt Gingell934bf652015-11-16 15:26:05 -08002842 kvm_make_request(KVM_REQ_EVENT, vcpu);
Zhang Xiantaof77bc6a2007-11-21 04:36:41 +08002843 return 0;
2844}
2845
Jan Kiszkac4abb7c2008-09-26 09:30:55 +02002846static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2847{
Jan Kiszkac4abb7c2008-09-26 09:30:55 +02002848 kvm_inject_nmi(vcpu);
Jan Kiszkac4abb7c2008-09-26 09:30:55 +02002849
2850 return 0;
2851}
2852
Paolo Bonzinif0778252015-04-01 15:06:40 +02002853static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
2854{
Paolo Bonzini64d60672015-05-07 11:36:11 +02002855 kvm_make_request(KVM_REQ_SMI, vcpu);
2856
Paolo Bonzinif0778252015-04-01 15:06:40 +02002857 return 0;
2858}
2859
Avi Kivityb209749f2007-10-22 16:50:39 +02002860static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2861 struct kvm_tpr_access_ctl *tac)
2862{
2863 if (tac->flags)
2864 return -EINVAL;
2865 vcpu->arch.tpr_access_reporting = !!tac->enabled;
2866 return 0;
2867}
2868
Huang Ying890ca9a2009-05-11 16:48:15 +08002869static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2870 u64 mcg_cap)
2871{
2872 int r;
2873 unsigned bank_num = mcg_cap & 0xff, bank;
2874
2875 r = -EINVAL;
Jan Kiszkaa9e38c3e2009-10-23 09:37:00 +02002876 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
Huang Ying890ca9a2009-05-11 16:48:15 +08002877 goto out;
Ashok Rajc45dcc72016-06-22 14:59:56 +08002878 if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
Huang Ying890ca9a2009-05-11 16:48:15 +08002879 goto out;
2880 r = 0;
2881 vcpu->arch.mcg_cap = mcg_cap;
2882 /* Init IA32_MCG_CTL to all 1s */
2883 if (mcg_cap & MCG_CTL_P)
2884 vcpu->arch.mcg_ctl = ~(u64)0;
2885 /* Init IA32_MCi_CTL to all 1s */
2886 for (bank = 0; bank < bank_num; bank++)
2887 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
Ashok Rajc45dcc72016-06-22 14:59:56 +08002888
2889 if (kvm_x86_ops->setup_mce)
2890 kvm_x86_ops->setup_mce(vcpu);
Huang Ying890ca9a2009-05-11 16:48:15 +08002891out:
2892 return r;
2893}
2894
2895static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2896 struct kvm_x86_mce *mce)
2897{
2898 u64 mcg_cap = vcpu->arch.mcg_cap;
2899 unsigned bank_num = mcg_cap & 0xff;
2900 u64 *banks = vcpu->arch.mce_banks;
2901
2902 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2903 return -EINVAL;
2904 /*
2905 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2906 * reporting is disabled
2907 */
2908 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2909 vcpu->arch.mcg_ctl != ~(u64)0)
2910 return 0;
2911 banks += 4 * mce->bank;
2912 /*
2913 * if IA32_MCi_CTL is not all 1s, the uncorrected error
2914 * reporting is disabled for the bank
2915 */
2916 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2917 return 0;
2918 if (mce->status & MCI_STATUS_UC) {
2919 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
Avi Kivityfc78f512009-12-07 12:16:48 +02002920 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
Avi Kivitya8eeb042010-05-10 12:34:53 +03002921 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
Huang Ying890ca9a2009-05-11 16:48:15 +08002922 return 0;
2923 }
2924 if (banks[1] & MCI_STATUS_VAL)
2925 mce->status |= MCI_STATUS_OVER;
2926 banks[2] = mce->addr;
2927 banks[3] = mce->misc;
2928 vcpu->arch.mcg_status = mce->mcg_status;
2929 banks[1] = mce->status;
2930 kvm_queue_exception(vcpu, MC_VECTOR);
2931 } else if (!(banks[1] & MCI_STATUS_VAL)
2932 || !(banks[1] & MCI_STATUS_UC)) {
2933 if (banks[1] & MCI_STATUS_VAL)
2934 mce->status |= MCI_STATUS_OVER;
2935 banks[2] = mce->addr;
2936 banks[3] = mce->misc;
2937 banks[1] = mce->status;
2938 } else
2939 banks[1] |= MCI_STATUS_OVER;
2940 return 0;
2941}
2942
Jan Kiszka3cfc3092009-11-12 01:04:25 +01002943static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2944 struct kvm_vcpu_events *events)
2945{
Avi Kivity7460fb4a2011-09-20 13:43:14 +03002946 process_nmi(vcpu);
Jan Kiszka03b82a32010-02-15 10:45:41 +01002947 events->exception.injected =
2948 vcpu->arch.exception.pending &&
2949 !kvm_exception_is_soft(vcpu->arch.exception.nr);
Jan Kiszka3cfc3092009-11-12 01:04:25 +01002950 events->exception.nr = vcpu->arch.exception.nr;
2951 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
Vasiliy Kulikov97e69aa2010-10-30 22:54:47 +04002952 events->exception.pad = 0;
Jan Kiszka3cfc3092009-11-12 01:04:25 +01002953 events->exception.error_code = vcpu->arch.exception.error_code;
2954
Jan Kiszka03b82a32010-02-15 10:45:41 +01002955 events->interrupt.injected =
2956 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
Jan Kiszka3cfc3092009-11-12 01:04:25 +01002957 events->interrupt.nr = vcpu->arch.interrupt.nr;
Jan Kiszka03b82a32010-02-15 10:45:41 +01002958 events->interrupt.soft = 0;
Paolo Bonzini37ccdcb2014-05-20 14:29:47 +02002959 events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
Jan Kiszka3cfc3092009-11-12 01:04:25 +01002960
2961 events->nmi.injected = vcpu->arch.nmi_injected;
Avi Kivity7460fb4a2011-09-20 13:43:14 +03002962 events->nmi.pending = vcpu->arch.nmi_pending != 0;
Jan Kiszka3cfc3092009-11-12 01:04:25 +01002963 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
Vasiliy Kulikov97e69aa2010-10-30 22:54:47 +04002964 events->nmi.pad = 0;
Jan Kiszka3cfc3092009-11-12 01:04:25 +01002965
Jan Kiszka66450a22013-03-13 12:42:34 +01002966 events->sipi_vector = 0; /* never valid when reporting to user space */
Jan Kiszka3cfc3092009-11-12 01:04:25 +01002967
Paolo Bonzinif0778252015-04-01 15:06:40 +02002968 events->smi.smm = is_smm(vcpu);
2969 events->smi.pending = vcpu->arch.smi_pending;
2970 events->smi.smm_inside_nmi =
2971 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
2972 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
2973
Jan Kiszkadab4b912009-12-06 18:24:15 +01002974 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
Paolo Bonzinif0778252015-04-01 15:06:40 +02002975 | KVM_VCPUEVENT_VALID_SHADOW
2976 | KVM_VCPUEVENT_VALID_SMM);
Vasiliy Kulikov97e69aa2010-10-30 22:54:47 +04002977 memset(&events->reserved, 0, sizeof(events->reserved));
Jan Kiszka3cfc3092009-11-12 01:04:25 +01002978}
2979
2980static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2981 struct kvm_vcpu_events *events)
2982{
Jan Kiszkadab4b912009-12-06 18:24:15 +01002983 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
Jan Kiszka48005f62010-02-19 19:38:07 +01002984 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
Paolo Bonzinif0778252015-04-01 15:06:40 +02002985 | KVM_VCPUEVENT_VALID_SHADOW
2986 | KVM_VCPUEVENT_VALID_SMM))
Jan Kiszka3cfc3092009-11-12 01:04:25 +01002987 return -EINVAL;
2988
Paolo Bonzini78e546c2016-06-01 14:09:20 +02002989 if (events->exception.injected &&
2990 (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
2991 return -EINVAL;
2992
Avi Kivity7460fb4a2011-09-20 13:43:14 +03002993 process_nmi(vcpu);
Jan Kiszka3cfc3092009-11-12 01:04:25 +01002994 vcpu->arch.exception.pending = events->exception.injected;
2995 vcpu->arch.exception.nr = events->exception.nr;
2996 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2997 vcpu->arch.exception.error_code = events->exception.error_code;
2998
2999 vcpu->arch.interrupt.pending = events->interrupt.injected;
3000 vcpu->arch.interrupt.nr = events->interrupt.nr;
3001 vcpu->arch.interrupt.soft = events->interrupt.soft;
Jan Kiszka48005f62010-02-19 19:38:07 +01003002 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3003 kvm_x86_ops->set_interrupt_shadow(vcpu,
3004 events->interrupt.shadow);
Jan Kiszka3cfc3092009-11-12 01:04:25 +01003005
3006 vcpu->arch.nmi_injected = events->nmi.injected;
Jan Kiszkadab4b912009-12-06 18:24:15 +01003007 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3008 vcpu->arch.nmi_pending = events->nmi.pending;
Jan Kiszka3cfc3092009-11-12 01:04:25 +01003009 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3010
Jan Kiszka66450a22013-03-13 12:42:34 +01003011 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
Paolo Bonzinibce87cc2016-01-08 13:48:51 +01003012 lapic_in_kernel(vcpu))
Jan Kiszka66450a22013-03-13 12:42:34 +01003013 vcpu->arch.apic->sipi_vector = events->sipi_vector;
Jan Kiszka3cfc3092009-11-12 01:04:25 +01003014
Paolo Bonzinif0778252015-04-01 15:06:40 +02003015 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
3016 if (events->smi.smm)
3017 vcpu->arch.hflags |= HF_SMM_MASK;
3018 else
3019 vcpu->arch.hflags &= ~HF_SMM_MASK;
3020 vcpu->arch.smi_pending = events->smi.pending;
3021 if (events->smi.smm_inside_nmi)
3022 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
3023 else
3024 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
Paolo Bonzinibce87cc2016-01-08 13:48:51 +01003025 if (lapic_in_kernel(vcpu)) {
Paolo Bonzinif0778252015-04-01 15:06:40 +02003026 if (events->smi.latched_init)
3027 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3028 else
3029 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3030 }
3031 }
3032
Avi Kivity3842d132010-07-27 12:30:24 +03003033 kvm_make_request(KVM_REQ_EVENT, vcpu);
3034
Jan Kiszka3cfc3092009-11-12 01:04:25 +01003035 return 0;
3036}
3037
Jan Kiszkaa1efbe72010-02-15 10:45:43 +01003038static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3039 struct kvm_debugregs *dbgregs)
3040{
Jan Kiszka73aaf249e2014-01-04 18:47:16 +01003041 unsigned long val;
3042
Jan Kiszkaa1efbe72010-02-15 10:45:43 +01003043 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
Nadav Amit16f8a6f2014-10-03 01:10:05 +03003044 kvm_get_dr(vcpu, 6, &val);
Jan Kiszka73aaf249e2014-01-04 18:47:16 +01003045 dbgregs->dr6 = val;
Jan Kiszkaa1efbe72010-02-15 10:45:43 +01003046 dbgregs->dr7 = vcpu->arch.dr7;
3047 dbgregs->flags = 0;
Vasiliy Kulikov97e69aa2010-10-30 22:54:47 +04003048 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
Jan Kiszkaa1efbe72010-02-15 10:45:43 +01003049}
3050
3051static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3052 struct kvm_debugregs *dbgregs)
3053{
3054 if (dbgregs->flags)
3055 return -EINVAL;
3056
Paolo Bonzinid14bdb52016-06-01 14:09:23 +02003057 if (dbgregs->dr6 & ~0xffffffffull)
3058 return -EINVAL;
3059 if (dbgregs->dr7 & ~0xffffffffull)
3060 return -EINVAL;
3061
Jan Kiszkaa1efbe72010-02-15 10:45:43 +01003062 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
Nadav Amitae561ed2015-04-02 03:10:37 +03003063 kvm_update_dr0123(vcpu);
Jan Kiszkaa1efbe72010-02-15 10:45:43 +01003064 vcpu->arch.dr6 = dbgregs->dr6;
Jan Kiszka73aaf249e2014-01-04 18:47:16 +01003065 kvm_update_dr6(vcpu);
Jan Kiszkaa1efbe72010-02-15 10:45:43 +01003066 vcpu->arch.dr7 = dbgregs->dr7;
Jan Kiszka9926c9f2014-01-04 18:47:15 +01003067 kvm_update_dr7(vcpu);
Jan Kiszkaa1efbe72010-02-15 10:45:43 +01003068
Jan Kiszkaa1efbe72010-02-15 10:45:43 +01003069 return 0;
3070}
3071
Paolo Bonzinidf1daba2014-11-21 19:05:07 +01003072#define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3073
3074static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3075{
Ingo Molnarc47ada32015-04-30 17:15:32 +02003076 struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
Ingo Molnar400e4b22015-04-24 10:19:47 +02003077 u64 xstate_bv = xsave->header.xfeatures;
Paolo Bonzinidf1daba2014-11-21 19:05:07 +01003078 u64 valid;
3079
3080 /*
3081 * Copy legacy XSAVE area, to avoid complications with CPUID
3082 * leaves 0 and 1 in the loop below.
3083 */
3084 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3085
3086 /* Set XSTATE_BV */
3087 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3088
3089 /*
3090 * Copy each region from the possibly compacted offset to the
3091 * non-compacted offset.
3092 */
Dave Hansend91cab72015-09-02 16:31:26 -07003093 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
Paolo Bonzinidf1daba2014-11-21 19:05:07 +01003094 while (valid) {
3095 u64 feature = valid & -valid;
3096 int index = fls64(feature) - 1;
3097 void *src = get_xsave_addr(xsave, feature);
3098
3099 if (src) {
3100 u32 size, offset, ecx, edx;
3101 cpuid_count(XSTATE_CPUID, index,
3102 &size, &offset, &ecx, &edx);
3103 memcpy(dest + offset, src, size);
3104 }
3105
3106 valid -= feature;
3107 }
3108}
3109
3110static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3111{
Ingo Molnarc47ada32015-04-30 17:15:32 +02003112 struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
Paolo Bonzinidf1daba2014-11-21 19:05:07 +01003113 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3114 u64 valid;
3115
3116 /*
3117 * Copy legacy XSAVE area, to avoid complications with CPUID
3118 * leaves 0 and 1 in the loop below.
3119 */
3120 memcpy(xsave, src, XSAVE_HDR_OFFSET);
3121
3122 /* Set XSTATE_BV and possibly XCOMP_BV. */
Ingo Molnar400e4b22015-04-24 10:19:47 +02003123 xsave->header.xfeatures = xstate_bv;
Borislav Petkov782511b2016-04-04 22:25:03 +02003124 if (boot_cpu_has(X86_FEATURE_XSAVES))
Ingo Molnar3a544502015-04-24 10:14:36 +02003125 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
Paolo Bonzinidf1daba2014-11-21 19:05:07 +01003126
3127 /*
3128 * Copy each region from the non-compacted offset to the
3129 * possibly compacted offset.
3130 */
Dave Hansend91cab72015-09-02 16:31:26 -07003131 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
Paolo Bonzinidf1daba2014-11-21 19:05:07 +01003132 while (valid) {
3133 u64 feature = valid & -valid;
3134 int index = fls64(feature) - 1;
3135 void *dest = get_xsave_addr(xsave, feature);
3136
3137 if (dest) {
3138 u32 size, offset, ecx, edx;
3139 cpuid_count(XSTATE_CPUID, index,
3140 &size, &offset, &ecx, &edx);
3141 memcpy(dest, src + offset, size);
Wanpeng Liee4100d2015-07-09 15:44:52 +08003142 }
Paolo Bonzinidf1daba2014-11-21 19:05:07 +01003143
3144 valid -= feature;
3145 }
3146}
3147
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003148static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3149 struct kvm_xsave *guest_xsave)
3150{
Borislav Petkovd366bf72016-04-04 22:25:02 +02003151 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
Paolo Bonzinidf1daba2014-11-21 19:05:07 +01003152 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3153 fill_xsave((u8 *) guest_xsave->region, vcpu);
Paolo Bonzini4344ee92013-10-02 16:06:16 +02003154 } else {
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003155 memcpy(guest_xsave->region,
Ingo Molnar7366ed72015-04-27 04:19:39 +02003156 &vcpu->arch.guest_fpu.state.fxsave,
Ingo Molnarc47ada32015-04-30 17:15:32 +02003157 sizeof(struct fxregs_state));
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003158 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
Dave Hansend91cab72015-09-02 16:31:26 -07003159 XFEATURE_MASK_FPSSE;
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003160 }
3161}
3162
3163static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3164 struct kvm_xsave *guest_xsave)
3165{
3166 u64 xstate_bv =
3167 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3168
Borislav Petkovd366bf72016-04-04 22:25:02 +02003169 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
Paolo Bonzinid7876f12013-10-02 16:06:15 +02003170 /*
3171 * Here we allow setting states that are not present in
3172 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
3173 * with old userspace.
3174 */
Paolo Bonzini4ff41732014-02-24 12:15:16 +01003175 if (xstate_bv & ~kvm_supported_xcr0())
Paolo Bonzinid7876f12013-10-02 16:06:15 +02003176 return -EINVAL;
Paolo Bonzinidf1daba2014-11-21 19:05:07 +01003177 load_xsave(vcpu, (u8 *)guest_xsave->region);
Paolo Bonzinid7876f12013-10-02 16:06:15 +02003178 } else {
Dave Hansend91cab72015-09-02 16:31:26 -07003179 if (xstate_bv & ~XFEATURE_MASK_FPSSE)
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003180 return -EINVAL;
Ingo Molnar7366ed72015-04-27 04:19:39 +02003181 memcpy(&vcpu->arch.guest_fpu.state.fxsave,
Ingo Molnarc47ada32015-04-30 17:15:32 +02003182 guest_xsave->region, sizeof(struct fxregs_state));
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003183 }
3184 return 0;
3185}
3186
3187static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3188 struct kvm_xcrs *guest_xcrs)
3189{
Borislav Petkovd366bf72016-04-04 22:25:02 +02003190 if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003191 guest_xcrs->nr_xcrs = 0;
3192 return;
3193 }
3194
3195 guest_xcrs->nr_xcrs = 1;
3196 guest_xcrs->flags = 0;
3197 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3198 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3199}
3200
3201static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3202 struct kvm_xcrs *guest_xcrs)
3203{
3204 int i, r = 0;
3205
Borislav Petkovd366bf72016-04-04 22:25:02 +02003206 if (!boot_cpu_has(X86_FEATURE_XSAVE))
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003207 return -EINVAL;
3208
3209 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3210 return -EINVAL;
3211
3212 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3213 /* Only support XCR0 currently */
Paolo Bonzinic67a04c2013-10-17 16:50:47 +02003214 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003215 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
Paolo Bonzinic67a04c2013-10-17 16:50:47 +02003216 guest_xcrs->xcrs[i].value);
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003217 break;
3218 }
3219 if (r)
3220 r = -EINVAL;
3221 return r;
3222}
3223
Eric B Munson1c0b28c2012-03-10 14:37:27 -05003224/*
3225 * kvm_set_guest_paused() indicates to the guest kernel that it has been
3226 * stopped by the hypervisor. This function will be called from the host only.
3227 * EINVAL is returned when the host attempts to set the flag for a guest that
3228 * does not support pv clocks.
3229 */
3230static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3231{
Andy Honig0b794592013-02-20 14:48:10 -08003232 if (!vcpu->arch.pv_time_enabled)
Eric B Munson1c0b28c2012-03-10 14:37:27 -05003233 return -EINVAL;
Marcelo Tosatti51d59c62012-08-03 15:57:49 -03003234 vcpu->arch.pvclock_set_guest_stopped_request = true;
Eric B Munson1c0b28c2012-03-10 14:37:27 -05003235 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3236 return 0;
3237}
3238
Andrey Smetanin5c9194122015-11-10 15:36:34 +03003239static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
3240 struct kvm_enable_cap *cap)
3241{
3242 if (cap->flags)
3243 return -EINVAL;
3244
3245 switch (cap->cap) {
3246 case KVM_CAP_HYPERV_SYNIC:
3247 return kvm_hv_activate_synic(vcpu);
3248 default:
3249 return -EINVAL;
3250 }
3251}
3252
Carsten Otte313a3dc2007-10-11 19:16:52 +02003253long kvm_arch_vcpu_ioctl(struct file *filp,
3254 unsigned int ioctl, unsigned long arg)
3255{
3256 struct kvm_vcpu *vcpu = filp->private_data;
3257 void __user *argp = (void __user *)arg;
3258 int r;
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003259 union {
3260 struct kvm_lapic_state *lapic;
3261 struct kvm_xsave *xsave;
3262 struct kvm_xcrs *xcrs;
3263 void *buffer;
3264 } u;
Carsten Otte313a3dc2007-10-11 19:16:52 +02003265
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003266 u.buffer = NULL;
Carsten Otte313a3dc2007-10-11 19:16:52 +02003267 switch (ioctl) {
3268 case KVM_GET_LAPIC: {
Marcelo Tosatti2204ae32009-10-29 13:44:16 -02003269 r = -EINVAL;
Paolo Bonzinibce87cc2016-01-08 13:48:51 +01003270 if (!lapic_in_kernel(vcpu))
Marcelo Tosatti2204ae32009-10-29 13:44:16 -02003271 goto out;
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003272 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
Carsten Otte313a3dc2007-10-11 19:16:52 +02003273
Dave Hansenb772ff32008-08-11 10:01:47 -07003274 r = -ENOMEM;
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003275 if (!u.lapic)
Dave Hansenb772ff32008-08-11 10:01:47 -07003276 goto out;
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003277 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
Carsten Otte313a3dc2007-10-11 19:16:52 +02003278 if (r)
3279 goto out;
3280 r = -EFAULT;
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003281 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
Carsten Otte313a3dc2007-10-11 19:16:52 +02003282 goto out;
3283 r = 0;
3284 break;
3285 }
3286 case KVM_SET_LAPIC: {
Marcelo Tosatti2204ae32009-10-29 13:44:16 -02003287 r = -EINVAL;
Paolo Bonzinibce87cc2016-01-08 13:48:51 +01003288 if (!lapic_in_kernel(vcpu))
Marcelo Tosatti2204ae32009-10-29 13:44:16 -02003289 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02003290 u.lapic = memdup_user(argp, sizeof(*u.lapic));
Guo Chao18595412012-11-02 18:33:21 +08003291 if (IS_ERR(u.lapic))
3292 return PTR_ERR(u.lapic);
Sasha Levinff5c2c02011-12-04 19:36:29 +02003293
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003294 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
Carsten Otte313a3dc2007-10-11 19:16:52 +02003295 break;
3296 }
Zhang Xiantaof77bc6a2007-11-21 04:36:41 +08003297 case KVM_INTERRUPT: {
3298 struct kvm_interrupt irq;
3299
3300 r = -EFAULT;
3301 if (copy_from_user(&irq, argp, sizeof irq))
3302 goto out;
3303 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
Zhang Xiantaof77bc6a2007-11-21 04:36:41 +08003304 break;
3305 }
Jan Kiszkac4abb7c2008-09-26 09:30:55 +02003306 case KVM_NMI: {
3307 r = kvm_vcpu_ioctl_nmi(vcpu);
Jan Kiszkac4abb7c2008-09-26 09:30:55 +02003308 break;
3309 }
Paolo Bonzinif0778252015-04-01 15:06:40 +02003310 case KVM_SMI: {
3311 r = kvm_vcpu_ioctl_smi(vcpu);
3312 break;
3313 }
Carsten Otte313a3dc2007-10-11 19:16:52 +02003314 case KVM_SET_CPUID: {
3315 struct kvm_cpuid __user *cpuid_arg = argp;
3316 struct kvm_cpuid cpuid;
3317
3318 r = -EFAULT;
3319 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3320 goto out;
3321 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
Carsten Otte313a3dc2007-10-11 19:16:52 +02003322 break;
3323 }
Dan Kenigsberg07716712007-11-21 17:10:04 +02003324 case KVM_SET_CPUID2: {
3325 struct kvm_cpuid2 __user *cpuid_arg = argp;
3326 struct kvm_cpuid2 cpuid;
3327
3328 r = -EFAULT;
3329 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3330 goto out;
3331 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
Amit Shah19355472009-01-14 16:56:00 +00003332 cpuid_arg->entries);
Dan Kenigsberg07716712007-11-21 17:10:04 +02003333 break;
3334 }
3335 case KVM_GET_CPUID2: {
3336 struct kvm_cpuid2 __user *cpuid_arg = argp;
3337 struct kvm_cpuid2 cpuid;
3338
3339 r = -EFAULT;
3340 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3341 goto out;
3342 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
Amit Shah19355472009-01-14 16:56:00 +00003343 cpuid_arg->entries);
Dan Kenigsberg07716712007-11-21 17:10:04 +02003344 if (r)
3345 goto out;
3346 r = -EFAULT;
3347 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3348 goto out;
3349 r = 0;
3350 break;
3351 }
Carsten Otte313a3dc2007-10-11 19:16:52 +02003352 case KVM_GET_MSRS:
Paolo Bonzini609e36d2015-04-08 15:30:38 +02003353 r = msr_io(vcpu, argp, do_get_msr, 1);
Carsten Otte313a3dc2007-10-11 19:16:52 +02003354 break;
3355 case KVM_SET_MSRS:
3356 r = msr_io(vcpu, argp, do_set_msr, 0);
3357 break;
Avi Kivityb209749f2007-10-22 16:50:39 +02003358 case KVM_TPR_ACCESS_REPORTING: {
3359 struct kvm_tpr_access_ctl tac;
3360
3361 r = -EFAULT;
3362 if (copy_from_user(&tac, argp, sizeof tac))
3363 goto out;
3364 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3365 if (r)
3366 goto out;
3367 r = -EFAULT;
3368 if (copy_to_user(argp, &tac, sizeof tac))
3369 goto out;
3370 r = 0;
3371 break;
3372 };
Avi Kivityb93463a2007-10-25 16:52:32 +02003373 case KVM_SET_VAPIC_ADDR: {
3374 struct kvm_vapic_addr va;
3375
3376 r = -EINVAL;
Paolo Bonzini35754c92015-07-29 12:05:37 +02003377 if (!lapic_in_kernel(vcpu))
Avi Kivityb93463a2007-10-25 16:52:32 +02003378 goto out;
3379 r = -EFAULT;
3380 if (copy_from_user(&va, argp, sizeof va))
3381 goto out;
Andy Honigfda4e2e82013-11-20 10:23:22 -08003382 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
Avi Kivityb93463a2007-10-25 16:52:32 +02003383 break;
3384 }
Huang Ying890ca9a2009-05-11 16:48:15 +08003385 case KVM_X86_SETUP_MCE: {
3386 u64 mcg_cap;
3387
3388 r = -EFAULT;
3389 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3390 goto out;
3391 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3392 break;
3393 }
3394 case KVM_X86_SET_MCE: {
3395 struct kvm_x86_mce mce;
3396
3397 r = -EFAULT;
3398 if (copy_from_user(&mce, argp, sizeof mce))
3399 goto out;
3400 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3401 break;
3402 }
Jan Kiszka3cfc3092009-11-12 01:04:25 +01003403 case KVM_GET_VCPU_EVENTS: {
3404 struct kvm_vcpu_events events;
3405
3406 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3407
3408 r = -EFAULT;
3409 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3410 break;
3411 r = 0;
3412 break;
3413 }
3414 case KVM_SET_VCPU_EVENTS: {
3415 struct kvm_vcpu_events events;
3416
3417 r = -EFAULT;
3418 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3419 break;
3420
3421 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3422 break;
3423 }
Jan Kiszkaa1efbe72010-02-15 10:45:43 +01003424 case KVM_GET_DEBUGREGS: {
3425 struct kvm_debugregs dbgregs;
3426
3427 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3428
3429 r = -EFAULT;
3430 if (copy_to_user(argp, &dbgregs,
3431 sizeof(struct kvm_debugregs)))
3432 break;
3433 r = 0;
3434 break;
3435 }
3436 case KVM_SET_DEBUGREGS: {
3437 struct kvm_debugregs dbgregs;
3438
3439 r = -EFAULT;
3440 if (copy_from_user(&dbgregs, argp,
3441 sizeof(struct kvm_debugregs)))
3442 break;
3443
3444 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3445 break;
3446 }
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003447 case KVM_GET_XSAVE: {
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003448 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003449 r = -ENOMEM;
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003450 if (!u.xsave)
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003451 break;
3452
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003453 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003454
3455 r = -EFAULT;
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003456 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003457 break;
3458 r = 0;
3459 break;
3460 }
3461 case KVM_SET_XSAVE: {
Sasha Levinff5c2c02011-12-04 19:36:29 +02003462 u.xsave = memdup_user(argp, sizeof(*u.xsave));
Guo Chao18595412012-11-02 18:33:21 +08003463 if (IS_ERR(u.xsave))
3464 return PTR_ERR(u.xsave);
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003465
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003466 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003467 break;
3468 }
3469 case KVM_GET_XCRS: {
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003470 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003471 r = -ENOMEM;
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003472 if (!u.xcrs)
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003473 break;
3474
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003475 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003476
3477 r = -EFAULT;
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003478 if (copy_to_user(argp, u.xcrs,
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003479 sizeof(struct kvm_xcrs)))
3480 break;
3481 r = 0;
3482 break;
3483 }
3484 case KVM_SET_XCRS: {
Sasha Levinff5c2c02011-12-04 19:36:29 +02003485 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
Guo Chao18595412012-11-02 18:33:21 +08003486 if (IS_ERR(u.xcrs))
3487 return PTR_ERR(u.xcrs);
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003488
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003489 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003490 break;
3491 }
Joerg Roedel92a1f122011-03-25 09:44:51 +01003492 case KVM_SET_TSC_KHZ: {
3493 u32 user_tsc_khz;
3494
3495 r = -EINVAL;
Joerg Roedel92a1f122011-03-25 09:44:51 +01003496 user_tsc_khz = (u32)arg;
3497
3498 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3499 goto out;
3500
Zachary Amsdencc578282012-02-03 15:43:50 -02003501 if (user_tsc_khz == 0)
3502 user_tsc_khz = tsc_khz;
3503
Haozhong Zhang381d5852015-10-20 15:39:04 +08003504 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
3505 r = 0;
Joerg Roedel92a1f122011-03-25 09:44:51 +01003506
Joerg Roedel92a1f122011-03-25 09:44:51 +01003507 goto out;
3508 }
3509 case KVM_GET_TSC_KHZ: {
Zachary Amsdencc578282012-02-03 15:43:50 -02003510 r = vcpu->arch.virtual_tsc_khz;
Joerg Roedel92a1f122011-03-25 09:44:51 +01003511 goto out;
3512 }
Eric B Munson1c0b28c2012-03-10 14:37:27 -05003513 case KVM_KVMCLOCK_CTRL: {
3514 r = kvm_set_guest_paused(vcpu);
3515 goto out;
3516 }
Andrey Smetanin5c9194122015-11-10 15:36:34 +03003517 case KVM_ENABLE_CAP: {
3518 struct kvm_enable_cap cap;
3519
3520 r = -EFAULT;
3521 if (copy_from_user(&cap, argp, sizeof(cap)))
3522 goto out;
3523 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
3524 break;
3525 }
Carsten Otte313a3dc2007-10-11 19:16:52 +02003526 default:
3527 r = -EINVAL;
3528 }
3529out:
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003530 kfree(u.buffer);
Carsten Otte313a3dc2007-10-11 19:16:52 +02003531 return r;
3532}
3533
Carsten Otte5b1c1492012-01-04 10:25:23 +01003534int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3535{
3536 return VM_FAULT_SIGBUS;
3537}
3538
Carsten Otte1fe779f2007-10-29 16:08:35 +01003539static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3540{
3541 int ret;
3542
3543 if (addr > (unsigned int)(-3 * PAGE_SIZE))
Guo Chao951179c2012-11-02 18:33:22 +08003544 return -EINVAL;
Carsten Otte1fe779f2007-10-29 16:08:35 +01003545 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3546 return ret;
3547}
3548
Sheng Yangb927a3c2009-07-21 10:42:48 +08003549static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3550 u64 ident_addr)
3551{
3552 kvm->arch.ept_identity_map_addr = ident_addr;
3553 return 0;
3554}
3555
Carsten Otte1fe779f2007-10-29 16:08:35 +01003556static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3557 u32 kvm_nr_mmu_pages)
3558{
3559 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3560 return -EINVAL;
3561
Marcelo Tosatti79fac952009-12-23 14:35:26 -02003562 mutex_lock(&kvm->slots_lock);
Carsten Otte1fe779f2007-10-29 16:08:35 +01003563
3564 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
Zhang Xiantaof05e70a2007-12-14 10:01:48 +08003565 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
Carsten Otte1fe779f2007-10-29 16:08:35 +01003566
Marcelo Tosatti79fac952009-12-23 14:35:26 -02003567 mutex_unlock(&kvm->slots_lock);
Carsten Otte1fe779f2007-10-29 16:08:35 +01003568 return 0;
3569}
3570
3571static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3572{
Dave Hansen39de71e2010-08-19 18:11:14 -07003573 return kvm->arch.n_max_mmu_pages;
Carsten Otte1fe779f2007-10-29 16:08:35 +01003574}
3575
Carsten Otte1fe779f2007-10-29 16:08:35 +01003576static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3577{
3578 int r;
3579
3580 r = 0;
3581 switch (chip->chip_id) {
3582 case KVM_IRQCHIP_PIC_MASTER:
3583 memcpy(&chip->chip.pic,
3584 &pic_irqchip(kvm)->pics[0],
3585 sizeof(struct kvm_pic_state));
3586 break;
3587 case KVM_IRQCHIP_PIC_SLAVE:
3588 memcpy(&chip->chip.pic,
3589 &pic_irqchip(kvm)->pics[1],
3590 sizeof(struct kvm_pic_state));
3591 break;
3592 case KVM_IRQCHIP_IOAPIC:
Gleb Natapoveba02262009-08-24 11:54:25 +03003593 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
Carsten Otte1fe779f2007-10-29 16:08:35 +01003594 break;
3595 default:
3596 r = -EINVAL;
3597 break;
3598 }
3599 return r;
3600}
3601
3602static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3603{
3604 int r;
3605
3606 r = 0;
3607 switch (chip->chip_id) {
3608 case KVM_IRQCHIP_PIC_MASTER:
Avi Kivityf4f51052010-09-19 18:44:07 +02003609 spin_lock(&pic_irqchip(kvm)->lock);
Carsten Otte1fe779f2007-10-29 16:08:35 +01003610 memcpy(&pic_irqchip(kvm)->pics[0],
3611 &chip->chip.pic,
3612 sizeof(struct kvm_pic_state));
Avi Kivityf4f51052010-09-19 18:44:07 +02003613 spin_unlock(&pic_irqchip(kvm)->lock);
Carsten Otte1fe779f2007-10-29 16:08:35 +01003614 break;
3615 case KVM_IRQCHIP_PIC_SLAVE:
Avi Kivityf4f51052010-09-19 18:44:07 +02003616 spin_lock(&pic_irqchip(kvm)->lock);
Carsten Otte1fe779f2007-10-29 16:08:35 +01003617 memcpy(&pic_irqchip(kvm)->pics[1],
3618 &chip->chip.pic,
3619 sizeof(struct kvm_pic_state));
Avi Kivityf4f51052010-09-19 18:44:07 +02003620 spin_unlock(&pic_irqchip(kvm)->lock);
Carsten Otte1fe779f2007-10-29 16:08:35 +01003621 break;
3622 case KVM_IRQCHIP_IOAPIC:
Gleb Natapoveba02262009-08-24 11:54:25 +03003623 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
Carsten Otte1fe779f2007-10-29 16:08:35 +01003624 break;
3625 default:
3626 r = -EINVAL;
3627 break;
3628 }
3629 kvm_pic_update_irq(pic_irqchip(kvm));
3630 return r;
3631}
3632
Sheng Yange0f63cb2008-03-04 00:50:59 +08003633static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3634{
Radim Krčmář34f39412016-03-02 22:56:50 +01003635 struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
3636
3637 BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
3638
3639 mutex_lock(&kps->lock);
3640 memcpy(ps, &kps->channels, sizeof(*ps));
3641 mutex_unlock(&kps->lock);
Saurabh Sengar2da29bc2015-10-30 12:56:11 +05303642 return 0;
Sheng Yange0f63cb2008-03-04 00:50:59 +08003643}
3644
3645static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3646{
Andrew Honig01856042015-11-18 14:50:23 -08003647 int i;
Radim Krčmář09edea72016-03-02 22:56:43 +01003648 struct kvm_pit *pit = kvm->arch.vpit;
3649
3650 mutex_lock(&pit->pit_state.lock);
Radim Krčmář34f39412016-03-02 22:56:50 +01003651 memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
Andrew Honig01856042015-11-18 14:50:23 -08003652 for (i = 0; i < 3; i++)
Radim Krčmář09edea72016-03-02 22:56:43 +01003653 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
3654 mutex_unlock(&pit->pit_state.lock);
Saurabh Sengar2da29bc2015-10-30 12:56:11 +05303655 return 0;
Beth Kone9f42752009-07-07 11:50:38 -04003656}
3657
3658static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3659{
Beth Kone9f42752009-07-07 11:50:38 -04003660 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3661 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3662 sizeof(ps->channels));
3663 ps->flags = kvm->arch.vpit->pit_state.flags;
3664 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
Vasiliy Kulikov97e69aa2010-10-30 22:54:47 +04003665 memset(&ps->reserved, 0, sizeof(ps->reserved));
Saurabh Sengar2da29bc2015-10-30 12:56:11 +05303666 return 0;
Beth Kone9f42752009-07-07 11:50:38 -04003667}
3668
3669static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3670{
Saurabh Sengar2da29bc2015-10-30 12:56:11 +05303671 int start = 0;
Andrew Honig01856042015-11-18 14:50:23 -08003672 int i;
Beth Kone9f42752009-07-07 11:50:38 -04003673 u32 prev_legacy, cur_legacy;
Radim Krčmář09edea72016-03-02 22:56:43 +01003674 struct kvm_pit *pit = kvm->arch.vpit;
3675
3676 mutex_lock(&pit->pit_state.lock);
3677 prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
Beth Kone9f42752009-07-07 11:50:38 -04003678 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3679 if (!prev_legacy && cur_legacy)
3680 start = 1;
Radim Krčmář09edea72016-03-02 22:56:43 +01003681 memcpy(&pit->pit_state.channels, &ps->channels,
3682 sizeof(pit->pit_state.channels));
3683 pit->pit_state.flags = ps->flags;
Andrew Honig01856042015-11-18 14:50:23 -08003684 for (i = 0; i < 3; i++)
Radim Krčmář09edea72016-03-02 22:56:43 +01003685 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
Paolo Bonzinie5e57e72016-01-07 13:50:38 +01003686 start && i == 0);
Radim Krčmář09edea72016-03-02 22:56:43 +01003687 mutex_unlock(&pit->pit_state.lock);
Saurabh Sengar2da29bc2015-10-30 12:56:11 +05303688 return 0;
Sheng Yange0f63cb2008-03-04 00:50:59 +08003689}
3690
Marcelo Tosatti52d939a2008-12-30 15:55:06 -02003691static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3692 struct kvm_reinject_control *control)
3693{
Radim Krčmář71474e22016-03-02 22:56:45 +01003694 struct kvm_pit *pit = kvm->arch.vpit;
3695
3696 if (!pit)
Marcelo Tosatti52d939a2008-12-30 15:55:06 -02003697 return -ENXIO;
Radim Krčmářb39c90b2016-03-02 22:56:44 +01003698
Radim Krčmář71474e22016-03-02 22:56:45 +01003699 /* pit->pit_state.lock was overloaded to prevent userspace from getting
3700 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
3701 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
3702 */
3703 mutex_lock(&pit->pit_state.lock);
3704 kvm_pit_set_reinject(pit, control->pit_reinject);
3705 mutex_unlock(&pit->pit_state.lock);
Radim Krčmářb39c90b2016-03-02 22:56:44 +01003706
Marcelo Tosatti52d939a2008-12-30 15:55:06 -02003707 return 0;
3708}
3709
Takuya Yoshikawa95d4c162011-11-14 18:24:50 +09003710/**
Takuya Yoshikawa60c34612012-03-03 14:21:48 +09003711 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3712 * @kvm: kvm instance
3713 * @log: slot id and address to which we copy the log
Takuya Yoshikawa95d4c162011-11-14 18:24:50 +09003714 *
Paolo Bonzinie108ff22015-01-15 15:58:54 -08003715 * Steps 1-4 below provide general overview of dirty page logging. See
3716 * kvm_get_dirty_log_protect() function description for additional details.
3717 *
3718 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
3719 * always flush the TLB (step 4) even if previous step failed and the dirty
3720 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
3721 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
3722 * writes will be marked dirty for next log read.
Takuya Yoshikawa95d4c162011-11-14 18:24:50 +09003723 *
Takuya Yoshikawa60c34612012-03-03 14:21:48 +09003724 * 1. Take a snapshot of the bit and clear it if needed.
3725 * 2. Write protect the corresponding page.
Paolo Bonzinie108ff22015-01-15 15:58:54 -08003726 * 3. Copy the snapshot to the userspace.
3727 * 4. Flush TLB's if needed.
Takuya Yoshikawa95d4c162011-11-14 18:24:50 +09003728 */
Takuya Yoshikawa60c34612012-03-03 14:21:48 +09003729int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
Zhang Xiantao5bb064d2007-11-18 20:29:43 +08003730{
Takuya Yoshikawa60c34612012-03-03 14:21:48 +09003731 bool is_dirty = false;
Paolo Bonzinie108ff22015-01-15 15:58:54 -08003732 int r;
Zhang Xiantao5bb064d2007-11-18 20:29:43 +08003733
Marcelo Tosatti79fac952009-12-23 14:35:26 -02003734 mutex_lock(&kvm->slots_lock);
Zhang Xiantao5bb064d2007-11-18 20:29:43 +08003735
Kai Huang88178fd2015-01-28 10:54:27 +08003736 /*
3737 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
3738 */
3739 if (kvm_x86_ops->flush_log_dirty)
3740 kvm_x86_ops->flush_log_dirty(kvm);
3741
Paolo Bonzinie108ff22015-01-15 15:58:54 -08003742 r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
Xiao Guangrong198c74f2014-04-17 17:06:16 +08003743
3744 /*
3745 * All the TLBs can be flushed out of mmu lock, see the comments in
3746 * kvm_mmu_slot_remove_write_access().
3747 */
Paolo Bonzinie108ff22015-01-15 15:58:54 -08003748 lockdep_assert_held(&kvm->slots_lock);
Xiao Guangrong198c74f2014-04-17 17:06:16 +08003749 if (is_dirty)
3750 kvm_flush_remote_tlbs(kvm);
3751
Marcelo Tosatti79fac952009-12-23 14:35:26 -02003752 mutex_unlock(&kvm->slots_lock);
Zhang Xiantao5bb064d2007-11-18 20:29:43 +08003753 return r;
3754}
3755
Yang Zhangaa2fbe62013-04-11 19:21:40 +08003756int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
3757 bool line_status)
Christoffer Dall23d43cf2012-07-24 08:51:20 -04003758{
3759 if (!irqchip_in_kernel(kvm))
3760 return -ENXIO;
3761
3762 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
Yang Zhangaa2fbe62013-04-11 19:21:40 +08003763 irq_event->irq, irq_event->level,
3764 line_status);
Christoffer Dall23d43cf2012-07-24 08:51:20 -04003765 return 0;
3766}
3767
Nadav Amit90de4a12015-04-13 01:53:41 +03003768static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3769 struct kvm_enable_cap *cap)
3770{
3771 int r;
3772
3773 if (cap->flags)
3774 return -EINVAL;
3775
3776 switch (cap->cap) {
3777 case KVM_CAP_DISABLE_QUIRKS:
3778 kvm->arch.disabled_quirks = cap->args[0];
3779 r = 0;
3780 break;
Steve Rutherford49df6392015-07-29 23:21:40 -07003781 case KVM_CAP_SPLIT_IRQCHIP: {
3782 mutex_lock(&kvm->lock);
Steve Rutherfordb053b2a2015-07-29 23:32:35 -07003783 r = -EINVAL;
3784 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
3785 goto split_irqchip_unlock;
Steve Rutherford49df6392015-07-29 23:21:40 -07003786 r = -EEXIST;
3787 if (irqchip_in_kernel(kvm))
3788 goto split_irqchip_unlock;
Paolo Bonzini557abc42016-06-13 14:50:04 +02003789 if (kvm->created_vcpus)
Steve Rutherford49df6392015-07-29 23:21:40 -07003790 goto split_irqchip_unlock;
3791 r = kvm_setup_empty_irq_routing(kvm);
3792 if (r)
3793 goto split_irqchip_unlock;
3794 /* Pairs with irqchip_in_kernel. */
3795 smp_wmb();
3796 kvm->arch.irqchip_split = true;
Steve Rutherfordb053b2a2015-07-29 23:32:35 -07003797 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
Steve Rutherford49df6392015-07-29 23:21:40 -07003798 r = 0;
3799split_irqchip_unlock:
3800 mutex_unlock(&kvm->lock);
3801 break;
3802 }
Radim Krčmář371313132016-07-12 22:09:27 +02003803 case KVM_CAP_X2APIC_API:
3804 r = -EINVAL;
3805 if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
3806 break;
3807
3808 if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
3809 kvm->arch.x2apic_format = true;
Radim Krčmářc5192652016-07-12 22:09:28 +02003810 if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
3811 kvm->arch.x2apic_broadcast_quirk_disabled = true;
Radim Krčmář371313132016-07-12 22:09:27 +02003812
3813 r = 0;
3814 break;
Nadav Amit90de4a12015-04-13 01:53:41 +03003815 default:
3816 r = -EINVAL;
3817 break;
3818 }
3819 return r;
3820}
3821
Carsten Otte1fe779f2007-10-29 16:08:35 +01003822long kvm_arch_vm_ioctl(struct file *filp,
3823 unsigned int ioctl, unsigned long arg)
3824{
3825 struct kvm *kvm = filp->private_data;
3826 void __user *argp = (void __user *)arg;
Avi Kivity367e1312009-08-26 14:57:07 +03003827 int r = -ENOTTY;
Dave Hansenf0d66272008-08-11 10:01:45 -07003828 /*
3829 * This union makes it completely explicit to gcc-3.x
3830 * that these two variables' stack usage should be
3831 * combined, not added together.
3832 */
3833 union {
3834 struct kvm_pit_state ps;
Beth Kone9f42752009-07-07 11:50:38 -04003835 struct kvm_pit_state2 ps2;
Jan Kiszkac5ff41c2009-05-14 22:42:53 +02003836 struct kvm_pit_config pit_config;
Dave Hansenf0d66272008-08-11 10:01:45 -07003837 } u;
Carsten Otte1fe779f2007-10-29 16:08:35 +01003838
3839 switch (ioctl) {
3840 case KVM_SET_TSS_ADDR:
3841 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
Carsten Otte1fe779f2007-10-29 16:08:35 +01003842 break;
Sheng Yangb927a3c2009-07-21 10:42:48 +08003843 case KVM_SET_IDENTITY_MAP_ADDR: {
3844 u64 ident_addr;
3845
3846 r = -EFAULT;
3847 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3848 goto out;
3849 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
Sheng Yangb927a3c2009-07-21 10:42:48 +08003850 break;
3851 }
Carsten Otte1fe779f2007-10-29 16:08:35 +01003852 case KVM_SET_NR_MMU_PAGES:
3853 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
Carsten Otte1fe779f2007-10-29 16:08:35 +01003854 break;
3855 case KVM_GET_NR_MMU_PAGES:
3856 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3857 break;
Marcelo Tosatti3ddea122009-10-29 13:44:15 -02003858 case KVM_CREATE_IRQCHIP: {
3859 struct kvm_pic *vpic;
3860
3861 mutex_lock(&kvm->lock);
3862 r = -EEXIST;
3863 if (kvm->arch.vpic)
3864 goto create_irqchip_unlock;
Avi Kivity3e515702012-03-05 14:23:29 +02003865 r = -EINVAL;
Paolo Bonzini557abc42016-06-13 14:50:04 +02003866 if (kvm->created_vcpus)
Avi Kivity3e515702012-03-05 14:23:29 +02003867 goto create_irqchip_unlock;
Carsten Otte1fe779f2007-10-29 16:08:35 +01003868 r = -ENOMEM;
Marcelo Tosatti3ddea122009-10-29 13:44:15 -02003869 vpic = kvm_create_pic(kvm);
3870 if (vpic) {
Carsten Otte1fe779f2007-10-29 16:08:35 +01003871 r = kvm_ioapic_init(kvm);
3872 if (r) {
Takuya Yoshikawa175504c2010-12-16 01:41:37 +09003873 mutex_lock(&kvm->slots_lock);
Paolo Bonzini71ba9942015-07-29 12:31:15 +02003874 kvm_destroy_pic(vpic);
Takuya Yoshikawa175504c2010-12-16 01:41:37 +09003875 mutex_unlock(&kvm->slots_lock);
Marcelo Tosatti3ddea122009-10-29 13:44:15 -02003876 goto create_irqchip_unlock;
Carsten Otte1fe779f2007-10-29 16:08:35 +01003877 }
3878 } else
Marcelo Tosatti3ddea122009-10-29 13:44:15 -02003879 goto create_irqchip_unlock;
Avi Kivity399ec802008-11-19 13:58:46 +02003880 r = kvm_setup_default_irq_routing(kvm);
3881 if (r) {
Takuya Yoshikawa175504c2010-12-16 01:41:37 +09003882 mutex_lock(&kvm->slots_lock);
Marcelo Tosatti3ddea122009-10-29 13:44:15 -02003883 mutex_lock(&kvm->irq_lock);
Wei Yongjun72bb2fc2010-02-09 10:33:03 +08003884 kvm_ioapic_destroy(kvm);
Paolo Bonzini71ba9942015-07-29 12:31:15 +02003885 kvm_destroy_pic(vpic);
Marcelo Tosatti3ddea122009-10-29 13:44:15 -02003886 mutex_unlock(&kvm->irq_lock);
Takuya Yoshikawa175504c2010-12-16 01:41:37 +09003887 mutex_unlock(&kvm->slots_lock);
Paolo Bonzini71ba9942015-07-29 12:31:15 +02003888 goto create_irqchip_unlock;
Avi Kivity399ec802008-11-19 13:58:46 +02003889 }
Paolo Bonzini71ba9942015-07-29 12:31:15 +02003890 /* Write kvm->irq_routing before kvm->arch.vpic. */
3891 smp_wmb();
3892 kvm->arch.vpic = vpic;
Marcelo Tosatti3ddea122009-10-29 13:44:15 -02003893 create_irqchip_unlock:
3894 mutex_unlock(&kvm->lock);
Carsten Otte1fe779f2007-10-29 16:08:35 +01003895 break;
Marcelo Tosatti3ddea122009-10-29 13:44:15 -02003896 }
Sheng Yang78376992008-01-28 05:10:22 +08003897 case KVM_CREATE_PIT:
Jan Kiszkac5ff41c2009-05-14 22:42:53 +02003898 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3899 goto create_pit;
3900 case KVM_CREATE_PIT2:
3901 r = -EFAULT;
3902 if (copy_from_user(&u.pit_config, argp,
3903 sizeof(struct kvm_pit_config)))
3904 goto out;
3905 create_pit:
Paolo Bonzini250715a2016-06-01 14:09:24 +02003906 mutex_lock(&kvm->lock);
Avi Kivity269e05e2009-01-05 15:21:42 +02003907 r = -EEXIST;
3908 if (kvm->arch.vpit)
3909 goto create_pit_unlock;
Sheng Yang78376992008-01-28 05:10:22 +08003910 r = -ENOMEM;
Jan Kiszkac5ff41c2009-05-14 22:42:53 +02003911 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
Sheng Yang78376992008-01-28 05:10:22 +08003912 if (kvm->arch.vpit)
3913 r = 0;
Avi Kivity269e05e2009-01-05 15:21:42 +02003914 create_pit_unlock:
Paolo Bonzini250715a2016-06-01 14:09:24 +02003915 mutex_unlock(&kvm->lock);
Sheng Yang78376992008-01-28 05:10:22 +08003916 break;
Carsten Otte1fe779f2007-10-29 16:08:35 +01003917 case KVM_GET_IRQCHIP: {
3918 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
Sasha Levinff5c2c02011-12-04 19:36:29 +02003919 struct kvm_irqchip *chip;
Carsten Otte1fe779f2007-10-29 16:08:35 +01003920
Sasha Levinff5c2c02011-12-04 19:36:29 +02003921 chip = memdup_user(argp, sizeof(*chip));
3922 if (IS_ERR(chip)) {
3923 r = PTR_ERR(chip);
Carsten Otte1fe779f2007-10-29 16:08:35 +01003924 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02003925 }
3926
Carsten Otte1fe779f2007-10-29 16:08:35 +01003927 r = -ENXIO;
Steve Rutherford49df6392015-07-29 23:21:40 -07003928 if (!irqchip_in_kernel(kvm) || irqchip_split(kvm))
Dave Hansenf0d66272008-08-11 10:01:45 -07003929 goto get_irqchip_out;
3930 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3931 if (r)
3932 goto get_irqchip_out;
3933 r = -EFAULT;
3934 if (copy_to_user(argp, chip, sizeof *chip))
3935 goto get_irqchip_out;
3936 r = 0;
3937 get_irqchip_out:
3938 kfree(chip);
Carsten Otte1fe779f2007-10-29 16:08:35 +01003939 break;
3940 }
3941 case KVM_SET_IRQCHIP: {
3942 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
Sasha Levinff5c2c02011-12-04 19:36:29 +02003943 struct kvm_irqchip *chip;
Carsten Otte1fe779f2007-10-29 16:08:35 +01003944
Sasha Levinff5c2c02011-12-04 19:36:29 +02003945 chip = memdup_user(argp, sizeof(*chip));
3946 if (IS_ERR(chip)) {
3947 r = PTR_ERR(chip);
Carsten Otte1fe779f2007-10-29 16:08:35 +01003948 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02003949 }
3950
Carsten Otte1fe779f2007-10-29 16:08:35 +01003951 r = -ENXIO;
Steve Rutherford49df6392015-07-29 23:21:40 -07003952 if (!irqchip_in_kernel(kvm) || irqchip_split(kvm))
Dave Hansenf0d66272008-08-11 10:01:45 -07003953 goto set_irqchip_out;
3954 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3955 if (r)
3956 goto set_irqchip_out;
3957 r = 0;
3958 set_irqchip_out:
3959 kfree(chip);
Carsten Otte1fe779f2007-10-29 16:08:35 +01003960 break;
3961 }
Sheng Yange0f63cb2008-03-04 00:50:59 +08003962 case KVM_GET_PIT: {
Sheng Yange0f63cb2008-03-04 00:50:59 +08003963 r = -EFAULT;
Dave Hansenf0d66272008-08-11 10:01:45 -07003964 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
Sheng Yange0f63cb2008-03-04 00:50:59 +08003965 goto out;
3966 r = -ENXIO;
3967 if (!kvm->arch.vpit)
3968 goto out;
Dave Hansenf0d66272008-08-11 10:01:45 -07003969 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
Sheng Yange0f63cb2008-03-04 00:50:59 +08003970 if (r)
3971 goto out;
3972 r = -EFAULT;
Dave Hansenf0d66272008-08-11 10:01:45 -07003973 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
Sheng Yange0f63cb2008-03-04 00:50:59 +08003974 goto out;
3975 r = 0;
3976 break;
3977 }
3978 case KVM_SET_PIT: {
Sheng Yange0f63cb2008-03-04 00:50:59 +08003979 r = -EFAULT;
Dave Hansenf0d66272008-08-11 10:01:45 -07003980 if (copy_from_user(&u.ps, argp, sizeof u.ps))
Sheng Yange0f63cb2008-03-04 00:50:59 +08003981 goto out;
3982 r = -ENXIO;
3983 if (!kvm->arch.vpit)
3984 goto out;
Dave Hansenf0d66272008-08-11 10:01:45 -07003985 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
Sheng Yange0f63cb2008-03-04 00:50:59 +08003986 break;
3987 }
Beth Kone9f42752009-07-07 11:50:38 -04003988 case KVM_GET_PIT2: {
3989 r = -ENXIO;
3990 if (!kvm->arch.vpit)
3991 goto out;
3992 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3993 if (r)
3994 goto out;
3995 r = -EFAULT;
3996 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3997 goto out;
3998 r = 0;
3999 break;
4000 }
4001 case KVM_SET_PIT2: {
4002 r = -EFAULT;
4003 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
4004 goto out;
4005 r = -ENXIO;
4006 if (!kvm->arch.vpit)
4007 goto out;
4008 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
Beth Kone9f42752009-07-07 11:50:38 -04004009 break;
4010 }
Marcelo Tosatti52d939a2008-12-30 15:55:06 -02004011 case KVM_REINJECT_CONTROL: {
4012 struct kvm_reinject_control control;
4013 r = -EFAULT;
4014 if (copy_from_user(&control, argp, sizeof(control)))
4015 goto out;
4016 r = kvm_vm_ioctl_reinject(kvm, &control);
Marcelo Tosatti52d939a2008-12-30 15:55:06 -02004017 break;
4018 }
Paolo Bonzinid71ba782015-07-29 11:56:48 +02004019 case KVM_SET_BOOT_CPU_ID:
4020 r = 0;
4021 mutex_lock(&kvm->lock);
Paolo Bonzini557abc42016-06-13 14:50:04 +02004022 if (kvm->created_vcpus)
Paolo Bonzinid71ba782015-07-29 11:56:48 +02004023 r = -EBUSY;
4024 else
4025 kvm->arch.bsp_vcpu_id = arg;
4026 mutex_unlock(&kvm->lock);
4027 break;
Ed Swierkffde22a2009-10-15 15:21:43 -07004028 case KVM_XEN_HVM_CONFIG: {
4029 r = -EFAULT;
4030 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
4031 sizeof(struct kvm_xen_hvm_config)))
4032 goto out;
4033 r = -EINVAL;
4034 if (kvm->arch.xen_hvm_config.flags)
4035 goto out;
4036 r = 0;
4037 break;
4038 }
Glauber Costaafbcf7a2009-10-16 15:28:36 -04004039 case KVM_SET_CLOCK: {
Glauber Costaafbcf7a2009-10-16 15:28:36 -04004040 struct kvm_clock_data user_ns;
4041 u64 now_ns;
4042 s64 delta;
4043
4044 r = -EFAULT;
4045 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
4046 goto out;
4047
4048 r = -EINVAL;
4049 if (user_ns.flags)
4050 goto out;
4051
4052 r = 0;
Avi Kivity395c6b02010-10-04 12:55:49 +02004053 local_irq_disable();
Zachary Amsden759379d2010-08-19 22:07:25 -10004054 now_ns = get_kernel_ns();
Glauber Costaafbcf7a2009-10-16 15:28:36 -04004055 delta = user_ns.clock - now_ns;
Avi Kivity395c6b02010-10-04 12:55:49 +02004056 local_irq_enable();
Glauber Costaafbcf7a2009-10-16 15:28:36 -04004057 kvm->arch.kvmclock_offset = delta;
Marcelo Tosatti2e762ff2013-08-27 23:55:29 -03004058 kvm_gen_update_masterclock(kvm);
Glauber Costaafbcf7a2009-10-16 15:28:36 -04004059 break;
4060 }
4061 case KVM_GET_CLOCK: {
Glauber Costaafbcf7a2009-10-16 15:28:36 -04004062 struct kvm_clock_data user_ns;
4063 u64 now_ns;
4064
Avi Kivity395c6b02010-10-04 12:55:49 +02004065 local_irq_disable();
Zachary Amsden759379d2010-08-19 22:07:25 -10004066 now_ns = get_kernel_ns();
Glauber Costaafbcf7a2009-10-16 15:28:36 -04004067 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
Avi Kivity395c6b02010-10-04 12:55:49 +02004068 local_irq_enable();
Glauber Costaafbcf7a2009-10-16 15:28:36 -04004069 user_ns.flags = 0;
Vasiliy Kulikov97e69aa2010-10-30 22:54:47 +04004070 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
Glauber Costaafbcf7a2009-10-16 15:28:36 -04004071
4072 r = -EFAULT;
4073 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
4074 goto out;
4075 r = 0;
4076 break;
4077 }
Nadav Amit90de4a12015-04-13 01:53:41 +03004078 case KVM_ENABLE_CAP: {
4079 struct kvm_enable_cap cap;
Glauber Costaafbcf7a2009-10-16 15:28:36 -04004080
Nadav Amit90de4a12015-04-13 01:53:41 +03004081 r = -EFAULT;
4082 if (copy_from_user(&cap, argp, sizeof(cap)))
4083 goto out;
4084 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
4085 break;
4086 }
Carsten Otte1fe779f2007-10-29 16:08:35 +01004087 default:
Radim Krčmářc274e032014-11-21 22:21:50 +01004088 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
Carsten Otte1fe779f2007-10-29 16:08:35 +01004089 }
4090out:
4091 return r;
4092}
4093
Zhang Xiantaoa16b0432007-11-16 14:38:21 +08004094static void kvm_init_msr_list(void)
Carsten Otte043405e2007-10-10 17:16:19 +02004095{
4096 u32 dummy[2];
4097 unsigned i, j;
4098
Paolo Bonzini62ef68b2015-05-05 12:08:55 +02004099 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
Carsten Otte043405e2007-10-10 17:16:19 +02004100 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
4101 continue;
Paolo Bonzini93c4adc2014-03-05 23:19:52 +01004102
4103 /*
4104 * Even MSRs that are valid in the host may not be exposed
Paolo Bonzini9dbe6cf2015-11-12 14:49:17 +01004105 * to the guests in some cases.
Paolo Bonzini93c4adc2014-03-05 23:19:52 +01004106 */
4107 switch (msrs_to_save[i]) {
4108 case MSR_IA32_BNDCFGS:
4109 if (!kvm_x86_ops->mpx_supported())
4110 continue;
4111 break;
Paolo Bonzini9dbe6cf2015-11-12 14:49:17 +01004112 case MSR_TSC_AUX:
4113 if (!kvm_x86_ops->rdtscp_supported())
4114 continue;
4115 break;
Paolo Bonzini93c4adc2014-03-05 23:19:52 +01004116 default:
4117 break;
4118 }
4119
Carsten Otte043405e2007-10-10 17:16:19 +02004120 if (j < i)
4121 msrs_to_save[j] = msrs_to_save[i];
4122 j++;
4123 }
4124 num_msrs_to_save = j;
Paolo Bonzini62ef68b2015-05-05 12:08:55 +02004125
4126 for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
4127 switch (emulated_msrs[i]) {
Paolo Bonzini6d396b52015-04-01 14:25:33 +02004128 case MSR_IA32_SMBASE:
4129 if (!kvm_x86_ops->cpu_has_high_real_mode_segbase())
4130 continue;
4131 break;
Paolo Bonzini62ef68b2015-05-05 12:08:55 +02004132 default:
4133 break;
4134 }
4135
4136 if (j < i)
4137 emulated_msrs[j] = emulated_msrs[i];
4138 j++;
4139 }
4140 num_emulated_msrs = j;
Carsten Otte043405e2007-10-10 17:16:19 +02004141}
4142
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03004143static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
4144 const void *v)
Carsten Ottebbd9b642007-10-30 18:44:21 +01004145{
Avi Kivity70252a12010-01-19 12:51:22 +02004146 int handled = 0;
4147 int n;
Carsten Ottebbd9b642007-10-30 18:44:21 +01004148
Avi Kivity70252a12010-01-19 12:51:22 +02004149 do {
4150 n = min(len, 8);
Paolo Bonzinibce87cc2016-01-08 13:48:51 +01004151 if (!(lapic_in_kernel(vcpu) &&
Nikolay Nikolaeve32edf42015-03-26 14:39:28 +00004152 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
4153 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
Avi Kivity70252a12010-01-19 12:51:22 +02004154 break;
4155 handled += n;
4156 addr += n;
4157 len -= n;
4158 v += n;
4159 } while (len);
4160
4161 return handled;
Carsten Ottebbd9b642007-10-30 18:44:21 +01004162}
4163
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03004164static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
Carsten Ottebbd9b642007-10-30 18:44:21 +01004165{
Avi Kivity70252a12010-01-19 12:51:22 +02004166 int handled = 0;
4167 int n;
Carsten Ottebbd9b642007-10-30 18:44:21 +01004168
Avi Kivity70252a12010-01-19 12:51:22 +02004169 do {
4170 n = min(len, 8);
Paolo Bonzinibce87cc2016-01-08 13:48:51 +01004171 if (!(lapic_in_kernel(vcpu) &&
Nikolay Nikolaeve32edf42015-03-26 14:39:28 +00004172 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
4173 addr, n, v))
4174 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
Avi Kivity70252a12010-01-19 12:51:22 +02004175 break;
4176 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
4177 handled += n;
4178 addr += n;
4179 len -= n;
4180 v += n;
4181 } while (len);
4182
4183 return handled;
Carsten Ottebbd9b642007-10-30 18:44:21 +01004184}
4185
Gleb Natapov2dafc6c2010-03-18 15:20:16 +02004186static void kvm_set_segment(struct kvm_vcpu *vcpu,
4187 struct kvm_segment *var, int seg)
4188{
4189 kvm_x86_ops->set_segment(vcpu, var, seg);
4190}
4191
4192void kvm_get_segment(struct kvm_vcpu *vcpu,
4193 struct kvm_segment *var, int seg)
4194{
4195 kvm_x86_ops->get_segment(vcpu, var, seg);
4196}
4197
Paolo Bonzini54987b72014-09-02 13:23:06 +02004198gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
4199 struct x86_exception *exception)
Joerg Roedel02f59dc2010-09-10 17:30:54 +02004200{
4201 gpa_t t_gpa;
Joerg Roedel02f59dc2010-09-10 17:30:54 +02004202
4203 BUG_ON(!mmu_is_nested(vcpu));
4204
4205 /* NPT walks are always user-walks */
4206 access |= PFERR_USER_MASK;
Paolo Bonzini54987b72014-09-02 13:23:06 +02004207 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, exception);
Joerg Roedel02f59dc2010-09-10 17:30:54 +02004208
4209 return t_gpa;
4210}
4211
Avi Kivityab9ae312010-11-22 17:53:26 +02004212gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
4213 struct x86_exception *exception)
Gleb Natapov1871c602010-02-10 14:21:32 +02004214{
4215 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
Avi Kivityab9ae312010-11-22 17:53:26 +02004216 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
Gleb Natapov1871c602010-02-10 14:21:32 +02004217}
4218
Avi Kivityab9ae312010-11-22 17:53:26 +02004219 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
4220 struct x86_exception *exception)
Gleb Natapov1871c602010-02-10 14:21:32 +02004221{
4222 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4223 access |= PFERR_FETCH_MASK;
Avi Kivityab9ae312010-11-22 17:53:26 +02004224 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
Gleb Natapov1871c602010-02-10 14:21:32 +02004225}
4226
Avi Kivityab9ae312010-11-22 17:53:26 +02004227gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
4228 struct x86_exception *exception)
Gleb Natapov1871c602010-02-10 14:21:32 +02004229{
4230 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4231 access |= PFERR_WRITE_MASK;
Avi Kivityab9ae312010-11-22 17:53:26 +02004232 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
Gleb Natapov1871c602010-02-10 14:21:32 +02004233}
4234
4235/* uses this to access any guest's mapped memory without checking CPL */
Avi Kivityab9ae312010-11-22 17:53:26 +02004236gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
4237 struct x86_exception *exception)
Gleb Natapov1871c602010-02-10 14:21:32 +02004238{
Avi Kivityab9ae312010-11-22 17:53:26 +02004239 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
Gleb Natapov1871c602010-02-10 14:21:32 +02004240}
4241
4242static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
4243 struct kvm_vcpu *vcpu, u32 access,
Avi Kivitybcc55cb2010-11-22 17:53:22 +02004244 struct x86_exception *exception)
Carsten Ottebbd9b642007-10-30 18:44:21 +01004245{
4246 void *data = val;
Marcelo Tosatti10589a42007-12-20 19:18:22 -05004247 int r = X86EMUL_CONTINUE;
Carsten Ottebbd9b642007-10-30 18:44:21 +01004248
4249 while (bytes) {
Joerg Roedel14dfe852010-09-10 17:30:49 +02004250 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
Avi Kivityab9ae312010-11-22 17:53:26 +02004251 exception);
Carsten Ottebbd9b642007-10-30 18:44:21 +01004252 unsigned offset = addr & (PAGE_SIZE-1);
Izik Eidus77c20022008-12-29 01:42:19 +02004253 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
Carsten Ottebbd9b642007-10-30 18:44:21 +01004254 int ret;
4255
Avi Kivitybcc55cb2010-11-22 17:53:22 +02004256 if (gpa == UNMAPPED_GVA)
Avi Kivityab9ae312010-11-22 17:53:26 +02004257 return X86EMUL_PROPAGATE_FAULT;
Paolo Bonzini54bf36a2015-04-08 15:39:23 +02004258 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
4259 offset, toread);
Marcelo Tosatti10589a42007-12-20 19:18:22 -05004260 if (ret < 0) {
Gleb Natapovc3cd7ff2010-04-28 19:15:35 +03004261 r = X86EMUL_IO_NEEDED;
Marcelo Tosatti10589a42007-12-20 19:18:22 -05004262 goto out;
4263 }
Carsten Ottebbd9b642007-10-30 18:44:21 +01004264
Izik Eidus77c20022008-12-29 01:42:19 +02004265 bytes -= toread;
4266 data += toread;
4267 addr += toread;
Carsten Ottebbd9b642007-10-30 18:44:21 +01004268 }
Marcelo Tosatti10589a42007-12-20 19:18:22 -05004269out:
Marcelo Tosatti10589a42007-12-20 19:18:22 -05004270 return r;
Carsten Ottebbd9b642007-10-30 18:44:21 +01004271}
Izik Eidus77c20022008-12-29 01:42:19 +02004272
Gleb Natapov1871c602010-02-10 14:21:32 +02004273/* used for instruction fetching */
Avi Kivity0f65dd72011-04-20 13:37:53 +03004274static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
4275 gva_t addr, void *val, unsigned int bytes,
Avi Kivitybcc55cb2010-11-22 17:53:22 +02004276 struct x86_exception *exception)
Gleb Natapov1871c602010-02-10 14:21:32 +02004277{
Avi Kivity0f65dd72011-04-20 13:37:53 +03004278 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
Gleb Natapov1871c602010-02-10 14:21:32 +02004279 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
Paolo Bonzini44583cb2014-05-13 14:02:13 +02004280 unsigned offset;
4281 int ret;
Avi Kivity0f65dd72011-04-20 13:37:53 +03004282
Paolo Bonzini44583cb2014-05-13 14:02:13 +02004283 /* Inline kvm_read_guest_virt_helper for speed. */
4284 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
4285 exception);
4286 if (unlikely(gpa == UNMAPPED_GVA))
4287 return X86EMUL_PROPAGATE_FAULT;
4288
4289 offset = addr & (PAGE_SIZE-1);
4290 if (WARN_ON(offset + bytes > PAGE_SIZE))
4291 bytes = (unsigned)PAGE_SIZE - offset;
Paolo Bonzini54bf36a2015-04-08 15:39:23 +02004292 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
4293 offset, bytes);
Paolo Bonzini44583cb2014-05-13 14:02:13 +02004294 if (unlikely(ret < 0))
4295 return X86EMUL_IO_NEEDED;
4296
4297 return X86EMUL_CONTINUE;
Gleb Natapov1871c602010-02-10 14:21:32 +02004298}
4299
Nadav Har'El064aea72011-05-25 23:04:56 +03004300int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
Avi Kivity0f65dd72011-04-20 13:37:53 +03004301 gva_t addr, void *val, unsigned int bytes,
Avi Kivitybcc55cb2010-11-22 17:53:22 +02004302 struct x86_exception *exception)
Gleb Natapov1871c602010-02-10 14:21:32 +02004303{
Avi Kivity0f65dd72011-04-20 13:37:53 +03004304 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
Gleb Natapov1871c602010-02-10 14:21:32 +02004305 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
Avi Kivity0f65dd72011-04-20 13:37:53 +03004306
Gleb Natapov1871c602010-02-10 14:21:32 +02004307 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
Avi Kivitybcc55cb2010-11-22 17:53:22 +02004308 exception);
Gleb Natapov1871c602010-02-10 14:21:32 +02004309}
Nadav Har'El064aea72011-05-25 23:04:56 +03004310EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
Gleb Natapov1871c602010-02-10 14:21:32 +02004311
Avi Kivity0f65dd72011-04-20 13:37:53 +03004312static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4313 gva_t addr, void *val, unsigned int bytes,
Avi Kivitybcc55cb2010-11-22 17:53:22 +02004314 struct x86_exception *exception)
Gleb Natapov1871c602010-02-10 14:21:32 +02004315{
Avi Kivity0f65dd72011-04-20 13:37:53 +03004316 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
Avi Kivitybcc55cb2010-11-22 17:53:22 +02004317 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
Gleb Natapov1871c602010-02-10 14:21:32 +02004318}
4319
Radim Krčmář7a036a62015-10-30 16:36:24 +01004320static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
4321 unsigned long addr, void *val, unsigned int bytes)
4322{
4323 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4324 int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
4325
4326 return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
4327}
4328
Nadav Har'El6a4d7552011-05-25 23:08:00 +03004329int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
Avi Kivity0f65dd72011-04-20 13:37:53 +03004330 gva_t addr, void *val,
Gleb Natapov2dafc6c2010-03-18 15:20:16 +02004331 unsigned int bytes,
Avi Kivitybcc55cb2010-11-22 17:53:22 +02004332 struct x86_exception *exception)
Izik Eidus77c20022008-12-29 01:42:19 +02004333{
Avi Kivity0f65dd72011-04-20 13:37:53 +03004334 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
Izik Eidus77c20022008-12-29 01:42:19 +02004335 void *data = val;
4336 int r = X86EMUL_CONTINUE;
4337
4338 while (bytes) {
Joerg Roedel14dfe852010-09-10 17:30:49 +02004339 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4340 PFERR_WRITE_MASK,
Avi Kivityab9ae312010-11-22 17:53:26 +02004341 exception);
Izik Eidus77c20022008-12-29 01:42:19 +02004342 unsigned offset = addr & (PAGE_SIZE-1);
4343 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4344 int ret;
4345
Avi Kivitybcc55cb2010-11-22 17:53:22 +02004346 if (gpa == UNMAPPED_GVA)
Avi Kivityab9ae312010-11-22 17:53:26 +02004347 return X86EMUL_PROPAGATE_FAULT;
Paolo Bonzini54bf36a2015-04-08 15:39:23 +02004348 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
Izik Eidus77c20022008-12-29 01:42:19 +02004349 if (ret < 0) {
Gleb Natapovc3cd7ff2010-04-28 19:15:35 +03004350 r = X86EMUL_IO_NEEDED;
Izik Eidus77c20022008-12-29 01:42:19 +02004351 goto out;
4352 }
4353
4354 bytes -= towrite;
4355 data += towrite;
4356 addr += towrite;
4357 }
4358out:
4359 return r;
4360}
Nadav Har'El6a4d7552011-05-25 23:08:00 +03004361EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
Izik Eidus77c20022008-12-29 01:42:19 +02004362
Xiao Guangrongaf7cc7d2011-07-12 03:22:46 +08004363static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4364 gpa_t *gpa, struct x86_exception *exception,
4365 bool write)
4366{
Avi Kivity97d64b72012-09-12 14:52:00 +03004367 u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
4368 | (write ? PFERR_WRITE_MASK : 0);
Xiao Guangrongaf7cc7d2011-07-12 03:22:46 +08004369
Huaitong Hanbe94f6b2016-03-22 16:51:20 +08004370 /*
4371 * currently PKRU is only applied to ept enabled guest so
4372 * there is no pkey in EPT page table for L1 guest or EPT
4373 * shadow page table for L2 guest.
4374 */
Avi Kivity97d64b72012-09-12 14:52:00 +03004375 if (vcpu_match_mmio_gva(vcpu, gva)
Feng Wu97ec8c02014-04-01 17:46:34 +08004376 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
Huaitong Hanbe94f6b2016-03-22 16:51:20 +08004377 vcpu->arch.access, 0, access)) {
Xiao Guangrongbebb1062011-07-12 03:23:20 +08004378 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4379 (gva & (PAGE_SIZE - 1));
Xiao Guangrong4f022642011-07-12 03:34:24 +08004380 trace_vcpu_match_mmio(gva, *gpa, write, false);
Xiao Guangrongbebb1062011-07-12 03:23:20 +08004381 return 1;
4382 }
4383
Xiao Guangrongaf7cc7d2011-07-12 03:22:46 +08004384 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4385
4386 if (*gpa == UNMAPPED_GVA)
4387 return -1;
4388
4389 /* For APIC access vmexit */
4390 if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4391 return 1;
4392
Xiao Guangrong4f022642011-07-12 03:34:24 +08004393 if (vcpu_match_mmio_gpa(vcpu, *gpa)) {
4394 trace_vcpu_match_mmio(gva, *gpa, write, true);
Xiao Guangrongbebb1062011-07-12 03:23:20 +08004395 return 1;
Xiao Guangrong4f022642011-07-12 03:34:24 +08004396 }
Xiao Guangrongbebb1062011-07-12 03:23:20 +08004397
Xiao Guangrongaf7cc7d2011-07-12 03:22:46 +08004398 return 0;
4399}
4400
Marcelo Tosatti3200f402008-03-29 20:17:59 -03004401int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
Avi Kivitybcc55cb2010-11-22 17:53:22 +02004402 const void *val, int bytes)
Avi Kivity9f811282008-03-02 14:06:05 +02004403{
4404 int ret;
4405
Paolo Bonzini54bf36a2015-04-08 15:39:23 +02004406 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
Avi Kivity9f811282008-03-02 14:06:05 +02004407 if (ret < 0)
4408 return 0;
Xiao Guangrong0eb05bf2016-02-24 17:51:13 +08004409 kvm_page_track_write(vcpu, gpa, val, bytes);
Avi Kivity9f811282008-03-02 14:06:05 +02004410 return 1;
4411}
4412
Xiao Guangrong77d197b2011-07-13 14:31:50 +08004413struct read_write_emulator_ops {
4414 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4415 int bytes);
4416 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4417 void *val, int bytes);
4418 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4419 int bytes, void *val);
4420 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4421 void *val, int bytes);
4422 bool write;
4423};
4424
4425static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4426{
4427 if (vcpu->mmio_read_completed) {
Xiao Guangrong77d197b2011-07-13 14:31:50 +08004428 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
Avi Kivityf78146b2012-04-18 19:22:47 +03004429 vcpu->mmio_fragments[0].gpa, *(u64 *)val);
Xiao Guangrong77d197b2011-07-13 14:31:50 +08004430 vcpu->mmio_read_completed = 0;
4431 return 1;
4432 }
4433
4434 return 0;
4435}
4436
4437static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4438 void *val, int bytes)
4439{
Paolo Bonzini54bf36a2015-04-08 15:39:23 +02004440 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
Xiao Guangrong77d197b2011-07-13 14:31:50 +08004441}
4442
4443static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4444 void *val, int bytes)
4445{
4446 return emulator_write_phys(vcpu, gpa, val, bytes);
4447}
4448
4449static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4450{
4451 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4452 return vcpu_mmio_write(vcpu, gpa, bytes, val);
4453}
4454
4455static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4456 void *val, int bytes)
4457{
4458 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4459 return X86EMUL_IO_NEEDED;
4460}
4461
4462static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4463 void *val, int bytes)
4464{
Avi Kivityf78146b2012-04-18 19:22:47 +03004465 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4466
Xiao Guangrong87da7e62012-10-24 14:07:59 +08004467 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
Xiao Guangrong77d197b2011-07-13 14:31:50 +08004468 return X86EMUL_CONTINUE;
4469}
4470
Mathias Krause0fbe9b02012-08-30 01:30:17 +02004471static const struct read_write_emulator_ops read_emultor = {
Xiao Guangrong77d197b2011-07-13 14:31:50 +08004472 .read_write_prepare = read_prepare,
4473 .read_write_emulate = read_emulate,
4474 .read_write_mmio = vcpu_mmio_read,
4475 .read_write_exit_mmio = read_exit_mmio,
4476};
4477
Mathias Krause0fbe9b02012-08-30 01:30:17 +02004478static const struct read_write_emulator_ops write_emultor = {
Xiao Guangrong77d197b2011-07-13 14:31:50 +08004479 .read_write_emulate = write_emulate,
4480 .read_write_mmio = write_mmio,
4481 .read_write_exit_mmio = write_exit_mmio,
4482 .write = true,
4483};
4484
Xiao Guangrong22388a32011-07-13 14:32:31 +08004485static int emulator_read_write_onepage(unsigned long addr, void *val,
4486 unsigned int bytes,
4487 struct x86_exception *exception,
4488 struct kvm_vcpu *vcpu,
Mathias Krause0fbe9b02012-08-30 01:30:17 +02004489 const struct read_write_emulator_ops *ops)
Carsten Ottebbd9b642007-10-30 18:44:21 +01004490{
Xiao Guangrongaf7cc7d2011-07-12 03:22:46 +08004491 gpa_t gpa;
4492 int handled, ret;
Xiao Guangrong22388a32011-07-13 14:32:31 +08004493 bool write = ops->write;
Avi Kivityf78146b2012-04-18 19:22:47 +03004494 struct kvm_mmio_fragment *frag;
Xiao Guangrong22388a32011-07-13 14:32:31 +08004495
4496 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
Carsten Ottebbd9b642007-10-30 18:44:21 +01004497
Xiao Guangrongaf7cc7d2011-07-12 03:22:46 +08004498 if (ret < 0)
Carsten Ottebbd9b642007-10-30 18:44:21 +01004499 return X86EMUL_PROPAGATE_FAULT;
Carsten Ottebbd9b642007-10-30 18:44:21 +01004500
4501 /* For APIC access vmexit */
Xiao Guangrongaf7cc7d2011-07-12 03:22:46 +08004502 if (ret)
Carsten Ottebbd9b642007-10-30 18:44:21 +01004503 goto mmio;
4504
Xiao Guangrong22388a32011-07-13 14:32:31 +08004505 if (ops->read_write_emulate(vcpu, gpa, val, bytes))
Carsten Ottebbd9b642007-10-30 18:44:21 +01004506 return X86EMUL_CONTINUE;
4507
4508mmio:
4509 /*
4510 * Is this MMIO handled locally?
4511 */
Xiao Guangrong22388a32011-07-13 14:32:31 +08004512 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
Avi Kivity70252a12010-01-19 12:51:22 +02004513 if (handled == bytes)
Carsten Ottebbd9b642007-10-30 18:44:21 +01004514 return X86EMUL_CONTINUE;
Carsten Ottebbd9b642007-10-30 18:44:21 +01004515
Avi Kivity70252a12010-01-19 12:51:22 +02004516 gpa += handled;
4517 bytes -= handled;
4518 val += handled;
4519
Xiao Guangrong87da7e62012-10-24 14:07:59 +08004520 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
4521 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
4522 frag->gpa = gpa;
4523 frag->data = val;
4524 frag->len = bytes;
Avi Kivityf78146b2012-04-18 19:22:47 +03004525 return X86EMUL_CONTINUE;
Carsten Ottebbd9b642007-10-30 18:44:21 +01004526}
4527
Xiubo Li52eb5a62015-03-13 17:39:45 +08004528static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
4529 unsigned long addr,
Xiao Guangrong22388a32011-07-13 14:32:31 +08004530 void *val, unsigned int bytes,
4531 struct x86_exception *exception,
Mathias Krause0fbe9b02012-08-30 01:30:17 +02004532 const struct read_write_emulator_ops *ops)
Carsten Ottebbd9b642007-10-30 18:44:21 +01004533{
Avi Kivity0f65dd72011-04-20 13:37:53 +03004534 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
Avi Kivityf78146b2012-04-18 19:22:47 +03004535 gpa_t gpa;
4536 int rc;
4537
4538 if (ops->read_write_prepare &&
4539 ops->read_write_prepare(vcpu, val, bytes))
4540 return X86EMUL_CONTINUE;
4541
4542 vcpu->mmio_nr_fragments = 0;
Avi Kivity0f65dd72011-04-20 13:37:53 +03004543
Carsten Ottebbd9b642007-10-30 18:44:21 +01004544 /* Crossing a page boundary? */
4545 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
Avi Kivityf78146b2012-04-18 19:22:47 +03004546 int now;
Carsten Ottebbd9b642007-10-30 18:44:21 +01004547
4548 now = -addr & ~PAGE_MASK;
Xiao Guangrong22388a32011-07-13 14:32:31 +08004549 rc = emulator_read_write_onepage(addr, val, now, exception,
4550 vcpu, ops);
4551
Carsten Ottebbd9b642007-10-30 18:44:21 +01004552 if (rc != X86EMUL_CONTINUE)
4553 return rc;
4554 addr += now;
Nadav Amitbac155312015-01-26 09:32:26 +02004555 if (ctxt->mode != X86EMUL_MODE_PROT64)
4556 addr = (u32)addr;
Carsten Ottebbd9b642007-10-30 18:44:21 +01004557 val += now;
4558 bytes -= now;
4559 }
Xiao Guangrong22388a32011-07-13 14:32:31 +08004560
Avi Kivityf78146b2012-04-18 19:22:47 +03004561 rc = emulator_read_write_onepage(addr, val, bytes, exception,
4562 vcpu, ops);
4563 if (rc != X86EMUL_CONTINUE)
4564 return rc;
4565
4566 if (!vcpu->mmio_nr_fragments)
4567 return rc;
4568
4569 gpa = vcpu->mmio_fragments[0].gpa;
4570
4571 vcpu->mmio_needed = 1;
4572 vcpu->mmio_cur_fragment = 0;
4573
Xiao Guangrong87da7e62012-10-24 14:07:59 +08004574 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
Avi Kivityf78146b2012-04-18 19:22:47 +03004575 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
4576 vcpu->run->exit_reason = KVM_EXIT_MMIO;
4577 vcpu->run->mmio.phys_addr = gpa;
4578
4579 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
Xiao Guangrong22388a32011-07-13 14:32:31 +08004580}
4581
4582static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4583 unsigned long addr,
4584 void *val,
4585 unsigned int bytes,
4586 struct x86_exception *exception)
4587{
4588 return emulator_read_write(ctxt, addr, val, bytes,
4589 exception, &read_emultor);
4590}
4591
Xiubo Li52eb5a62015-03-13 17:39:45 +08004592static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
Xiao Guangrong22388a32011-07-13 14:32:31 +08004593 unsigned long addr,
4594 const void *val,
4595 unsigned int bytes,
4596 struct x86_exception *exception)
4597{
4598 return emulator_read_write(ctxt, addr, (void *)val, bytes,
4599 exception, &write_emultor);
Carsten Ottebbd9b642007-10-30 18:44:21 +01004600}
Carsten Ottebbd9b642007-10-30 18:44:21 +01004601
Avi Kivitydaea3e72010-03-15 13:59:54 +02004602#define CMPXCHG_TYPE(t, ptr, old, new) \
4603 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4604
4605#ifdef CONFIG_X86_64
4606# define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4607#else
4608# define CMPXCHG64(ptr, old, new) \
Jan Kiszka9749a6c2010-03-20 10:14:13 +01004609 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
Avi Kivitydaea3e72010-03-15 13:59:54 +02004610#endif
4611
Avi Kivity0f65dd72011-04-20 13:37:53 +03004612static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4613 unsigned long addr,
Carsten Ottebbd9b642007-10-30 18:44:21 +01004614 const void *old,
4615 const void *new,
4616 unsigned int bytes,
Avi Kivity0f65dd72011-04-20 13:37:53 +03004617 struct x86_exception *exception)
Carsten Ottebbd9b642007-10-30 18:44:21 +01004618{
Avi Kivity0f65dd72011-04-20 13:37:53 +03004619 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
Avi Kivitydaea3e72010-03-15 13:59:54 +02004620 gpa_t gpa;
4621 struct page *page;
4622 char *kaddr;
4623 bool exchanged;
4624
Marcelo Tosatti2bacc552007-12-12 10:46:12 -05004625 /* guests cmpxchg8b have to be emulated atomically */
Avi Kivitydaea3e72010-03-15 13:59:54 +02004626 if (bytes > 8 || (bytes & (bytes - 1)))
4627 goto emul_write;
Marcelo Tosatti2bacc552007-12-12 10:46:12 -05004628
Avi Kivitydaea3e72010-03-15 13:59:54 +02004629 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
Marcelo Tosatti10589a42007-12-20 19:18:22 -05004630
Avi Kivitydaea3e72010-03-15 13:59:54 +02004631 if (gpa == UNMAPPED_GVA ||
4632 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4633 goto emul_write;
Marcelo Tosatti2bacc552007-12-12 10:46:12 -05004634
Avi Kivitydaea3e72010-03-15 13:59:54 +02004635 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4636 goto emul_write;
Marcelo Tosatti2bacc552007-12-12 10:46:12 -05004637
Paolo Bonzini54bf36a2015-04-08 15:39:23 +02004638 page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
Xiao Guangrong32cad842012-08-03 15:42:52 +08004639 if (is_error_page(page))
Wei Yongjunc19b8bd62010-07-15 08:51:58 +08004640 goto emul_write;
Izik Eidus72dc67a2008-02-10 18:04:15 +02004641
Cong Wang8fd75e12011-11-25 23:14:17 +08004642 kaddr = kmap_atomic(page);
Avi Kivitydaea3e72010-03-15 13:59:54 +02004643 kaddr += offset_in_page(gpa);
4644 switch (bytes) {
4645 case 1:
4646 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4647 break;
4648 case 2:
4649 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4650 break;
4651 case 4:
4652 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4653 break;
4654 case 8:
4655 exchanged = CMPXCHG64(kaddr, old, new);
4656 break;
4657 default:
4658 BUG();
Marcelo Tosatti2bacc552007-12-12 10:46:12 -05004659 }
Cong Wang8fd75e12011-11-25 23:14:17 +08004660 kunmap_atomic(kaddr);
Avi Kivitydaea3e72010-03-15 13:59:54 +02004661 kvm_release_page_dirty(page);
4662
4663 if (!exchanged)
4664 return X86EMUL_CMPXCHG_FAILED;
4665
Paolo Bonzini54bf36a2015-04-08 15:39:23 +02004666 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
Xiao Guangrong0eb05bf2016-02-24 17:51:13 +08004667 kvm_page_track_write(vcpu, gpa, new, bytes);
Gleb Natapov8f6abd02010-04-13 10:21:56 +03004668
4669 return X86EMUL_CONTINUE;
Avi Kivity4a5f48f2010-03-15 13:59:55 +02004670
Marcelo Tosatti3200f402008-03-29 20:17:59 -03004671emul_write:
Avi Kivitydaea3e72010-03-15 13:59:54 +02004672 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
Marcelo Tosatti2bacc552007-12-12 10:46:12 -05004673
Avi Kivity0f65dd72011-04-20 13:37:53 +03004674 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
Carsten Ottebbd9b642007-10-30 18:44:21 +01004675}
4676
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02004677static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4678{
4679 /* TODO: String I/O for in kernel device */
4680 int r;
4681
4682 if (vcpu->arch.pio.in)
Nikolay Nikolaeve32edf42015-03-26 14:39:28 +00004683 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02004684 vcpu->arch.pio.size, pd);
4685 else
Nikolay Nikolaeve32edf42015-03-26 14:39:28 +00004686 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02004687 vcpu->arch.pio.port, vcpu->arch.pio.size,
4688 pd);
4689 return r;
4690}
4691
Xiao Guangrong6f6fbe92011-09-22 16:55:10 +08004692static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
4693 unsigned short port, void *val,
4694 unsigned int count, bool in)
4695{
Xiao Guangrong6f6fbe92011-09-22 16:55:10 +08004696 vcpu->arch.pio.port = port;
4697 vcpu->arch.pio.in = in;
4698 vcpu->arch.pio.count = count;
4699 vcpu->arch.pio.size = size;
4700
4701 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4702 vcpu->arch.pio.count = 0;
4703 return 1;
4704 }
4705
4706 vcpu->run->exit_reason = KVM_EXIT_IO;
4707 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
4708 vcpu->run->io.size = size;
4709 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4710 vcpu->run->io.count = count;
4711 vcpu->run->io.port = port;
4712
4713 return 0;
4714}
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02004715
Avi Kivityca1d4a92011-04-20 13:37:53 +03004716static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4717 int size, unsigned short port, void *val,
4718 unsigned int count)
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02004719{
Avi Kivityca1d4a92011-04-20 13:37:53 +03004720 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
Xiao Guangrong6f6fbe92011-09-22 16:55:10 +08004721 int ret;
Avi Kivityca1d4a92011-04-20 13:37:53 +03004722
Gleb Natapov79729952010-03-18 15:20:24 +02004723 if (vcpu->arch.pio.count)
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02004724 goto data_avail;
4725
Xiao Guangrong6f6fbe92011-09-22 16:55:10 +08004726 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
4727 if (ret) {
4728data_avail:
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02004729 memcpy(val, vcpu->arch.pio_data, size * count);
Ulrich Obergfell11719032014-05-02 17:57:47 +02004730 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
Gleb Natapov79729952010-03-18 15:20:24 +02004731 vcpu->arch.pio.count = 0;
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02004732 return 1;
4733 }
4734
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02004735 return 0;
4736}
4737
Avi Kivityca1d4a92011-04-20 13:37:53 +03004738static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4739 int size, unsigned short port,
4740 const void *val, unsigned int count)
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02004741{
Avi Kivityca1d4a92011-04-20 13:37:53 +03004742 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4743
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02004744 memcpy(vcpu->arch.pio_data, val, size * count);
Ulrich Obergfell11719032014-05-02 17:57:47 +02004745 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
Xiao Guangrong6f6fbe92011-09-22 16:55:10 +08004746 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02004747}
4748
Carsten Ottebbd9b642007-10-30 18:44:21 +01004749static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4750{
4751 return kvm_x86_ops->get_segment_base(vcpu, seg);
4752}
4753
Avi Kivity3cb16fe2011-04-20 15:38:44 +03004754static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
Carsten Ottebbd9b642007-10-30 18:44:21 +01004755{
Avi Kivity3cb16fe2011-04-20 15:38:44 +03004756 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
Carsten Ottebbd9b642007-10-30 18:44:21 +01004757}
4758
Joel Schopp5cb56052015-03-02 13:43:31 -06004759int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
Sheng Yangf5f48ee2010-06-30 12:25:15 +08004760{
4761 if (!need_emulate_wbinvd(vcpu))
4762 return X86EMUL_CONTINUE;
4763
4764 if (kvm_x86_ops->has_wbinvd_exit()) {
Jan Kiszka2eec7342010-11-01 14:01:29 +01004765 int cpu = get_cpu();
4766
4767 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
Sheng Yangf5f48ee2010-06-30 12:25:15 +08004768 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4769 wbinvd_ipi, NULL, 1);
Jan Kiszka2eec7342010-11-01 14:01:29 +01004770 put_cpu();
Sheng Yangf5f48ee2010-06-30 12:25:15 +08004771 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
Jan Kiszka2eec7342010-11-01 14:01:29 +01004772 } else
4773 wbinvd();
Sheng Yangf5f48ee2010-06-30 12:25:15 +08004774 return X86EMUL_CONTINUE;
4775}
Joel Schopp5cb56052015-03-02 13:43:31 -06004776
4777int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4778{
4779 kvm_x86_ops->skip_emulated_instruction(vcpu);
4780 return kvm_emulate_wbinvd_noskip(vcpu);
4781}
Sheng Yangf5f48ee2010-06-30 12:25:15 +08004782EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4783
Joel Schopp5cb56052015-03-02 13:43:31 -06004784
4785
Avi Kivitybcaf5cc2011-04-20 15:53:23 +03004786static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4787{
Joel Schopp5cb56052015-03-02 13:43:31 -06004788 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
Avi Kivitybcaf5cc2011-04-20 15:53:23 +03004789}
4790
Xiubo Li52eb5a62015-03-13 17:39:45 +08004791static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
4792 unsigned long *dest)
Carsten Ottebbd9b642007-10-30 18:44:21 +01004793{
Nadav Amit16f8a6f2014-10-03 01:10:05 +03004794 return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
Carsten Ottebbd9b642007-10-30 18:44:21 +01004795}
4796
Xiubo Li52eb5a62015-03-13 17:39:45 +08004797static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
4798 unsigned long value)
Carsten Ottebbd9b642007-10-30 18:44:21 +01004799{
Gleb Natapov338dbc92010-04-28 19:15:32 +03004800
Avi Kivity717746e2011-04-20 13:37:53 +03004801 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
Carsten Ottebbd9b642007-10-30 18:44:21 +01004802}
4803
Gleb Natapov52a46612010-03-18 15:20:03 +02004804static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4805{
4806 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4807}
4808
Avi Kivity717746e2011-04-20 13:37:53 +03004809static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
Gleb Natapov52a46612010-03-18 15:20:03 +02004810{
Avi Kivity717746e2011-04-20 13:37:53 +03004811 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
Gleb Natapov52a46612010-03-18 15:20:03 +02004812 unsigned long value;
4813
4814 switch (cr) {
4815 case 0:
4816 value = kvm_read_cr0(vcpu);
4817 break;
4818 case 2:
4819 value = vcpu->arch.cr2;
4820 break;
4821 case 3:
Avi Kivity9f8fe502010-12-05 17:30:00 +02004822 value = kvm_read_cr3(vcpu);
Gleb Natapov52a46612010-03-18 15:20:03 +02004823 break;
4824 case 4:
4825 value = kvm_read_cr4(vcpu);
4826 break;
4827 case 8:
4828 value = kvm_get_cr8(vcpu);
4829 break;
4830 default:
Christoffer Dalla737f252012-06-03 21:17:48 +03004831 kvm_err("%s: unexpected cr %u\n", __func__, cr);
Gleb Natapov52a46612010-03-18 15:20:03 +02004832 return 0;
4833 }
4834
4835 return value;
4836}
4837
Avi Kivity717746e2011-04-20 13:37:53 +03004838static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
Gleb Natapov52a46612010-03-18 15:20:03 +02004839{
Avi Kivity717746e2011-04-20 13:37:53 +03004840 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
Gleb Natapov0f122442010-04-28 19:15:31 +03004841 int res = 0;
4842
Gleb Natapov52a46612010-03-18 15:20:03 +02004843 switch (cr) {
4844 case 0:
Avi Kivity49a9b072010-06-10 17:02:14 +03004845 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
Gleb Natapov52a46612010-03-18 15:20:03 +02004846 break;
4847 case 2:
4848 vcpu->arch.cr2 = val;
4849 break;
4850 case 3:
Avi Kivity23902182010-06-10 17:02:16 +03004851 res = kvm_set_cr3(vcpu, val);
Gleb Natapov52a46612010-03-18 15:20:03 +02004852 break;
4853 case 4:
Avi Kivitya83b29c2010-06-10 17:02:15 +03004854 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
Gleb Natapov52a46612010-03-18 15:20:03 +02004855 break;
4856 case 8:
Andre Przywaraeea1cff2010-12-21 11:12:00 +01004857 res = kvm_set_cr8(vcpu, val);
Gleb Natapov52a46612010-03-18 15:20:03 +02004858 break;
4859 default:
Christoffer Dalla737f252012-06-03 21:17:48 +03004860 kvm_err("%s: unexpected cr %u\n", __func__, cr);
Gleb Natapov0f122442010-04-28 19:15:31 +03004861 res = -1;
Gleb Natapov52a46612010-03-18 15:20:03 +02004862 }
Gleb Natapov0f122442010-04-28 19:15:31 +03004863
4864 return res;
Gleb Natapov52a46612010-03-18 15:20:03 +02004865}
4866
Avi Kivity717746e2011-04-20 13:37:53 +03004867static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
Gleb Natapov9c537242010-03-18 15:20:05 +02004868{
Avi Kivity717746e2011-04-20 13:37:53 +03004869 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
Gleb Natapov9c537242010-03-18 15:20:05 +02004870}
4871
Avi Kivity4bff1e862011-04-20 13:37:53 +03004872static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
Gleb Natapov2dafc6c2010-03-18 15:20:16 +02004873{
Avi Kivity4bff1e862011-04-20 13:37:53 +03004874 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
Gleb Natapov2dafc6c2010-03-18 15:20:16 +02004875}
4876
Avi Kivity4bff1e862011-04-20 13:37:53 +03004877static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
Mohammed Gamal160ce1f2010-08-04 05:44:24 +03004878{
Avi Kivity4bff1e862011-04-20 13:37:53 +03004879 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
Mohammed Gamal160ce1f2010-08-04 05:44:24 +03004880}
4881
Avi Kivity1ac9d0c2011-04-20 15:12:00 +03004882static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4883{
4884 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4885}
4886
4887static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4888{
4889 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4890}
4891
Avi Kivity4bff1e862011-04-20 13:37:53 +03004892static unsigned long emulator_get_cached_segment_base(
4893 struct x86_emulate_ctxt *ctxt, int seg)
Gleb Natapov5951c442010-04-28 19:15:29 +03004894{
Avi Kivity4bff1e862011-04-20 13:37:53 +03004895 return get_segment_base(emul_to_vcpu(ctxt), seg);
Gleb Natapov5951c442010-04-28 19:15:29 +03004896}
4897
Avi Kivity1aa36612011-04-27 13:20:30 +03004898static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
4899 struct desc_struct *desc, u32 *base3,
4900 int seg)
Gleb Natapov2dafc6c2010-03-18 15:20:16 +02004901{
4902 struct kvm_segment var;
4903
Avi Kivity4bff1e862011-04-20 13:37:53 +03004904 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
Avi Kivity1aa36612011-04-27 13:20:30 +03004905 *selector = var.selector;
Gleb Natapov2dafc6c2010-03-18 15:20:16 +02004906
Gleb Natapov378a8b02013-01-21 15:36:48 +02004907 if (var.unusable) {
4908 memset(desc, 0, sizeof(*desc));
Gleb Natapov2dafc6c2010-03-18 15:20:16 +02004909 return false;
Gleb Natapov378a8b02013-01-21 15:36:48 +02004910 }
Gleb Natapov2dafc6c2010-03-18 15:20:16 +02004911
4912 if (var.g)
4913 var.limit >>= 12;
4914 set_desc_limit(desc, var.limit);
4915 set_desc_base(desc, (unsigned long)var.base);
Gleb Natapov5601d052011-03-07 14:55:06 +02004916#ifdef CONFIG_X86_64
4917 if (base3)
4918 *base3 = var.base >> 32;
4919#endif
Gleb Natapov2dafc6c2010-03-18 15:20:16 +02004920 desc->type = var.type;
4921 desc->s = var.s;
4922 desc->dpl = var.dpl;
4923 desc->p = var.present;
4924 desc->avl = var.avl;
4925 desc->l = var.l;
4926 desc->d = var.db;
4927 desc->g = var.g;
4928
4929 return true;
4930}
4931
Avi Kivity1aa36612011-04-27 13:20:30 +03004932static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
4933 struct desc_struct *desc, u32 base3,
4934 int seg)
Gleb Natapov2dafc6c2010-03-18 15:20:16 +02004935{
Avi Kivity4bff1e862011-04-20 13:37:53 +03004936 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
Gleb Natapov2dafc6c2010-03-18 15:20:16 +02004937 struct kvm_segment var;
4938
Avi Kivity1aa36612011-04-27 13:20:30 +03004939 var.selector = selector;
Gleb Natapov2dafc6c2010-03-18 15:20:16 +02004940 var.base = get_desc_base(desc);
Gleb Natapov5601d052011-03-07 14:55:06 +02004941#ifdef CONFIG_X86_64
4942 var.base |= ((u64)base3) << 32;
4943#endif
Gleb Natapov2dafc6c2010-03-18 15:20:16 +02004944 var.limit = get_desc_limit(desc);
4945 if (desc->g)
4946 var.limit = (var.limit << 12) | 0xfff;
4947 var.type = desc->type;
Gleb Natapov2dafc6c2010-03-18 15:20:16 +02004948 var.dpl = desc->dpl;
4949 var.db = desc->d;
4950 var.s = desc->s;
4951 var.l = desc->l;
4952 var.g = desc->g;
4953 var.avl = desc->avl;
4954 var.present = desc->p;
4955 var.unusable = !var.present;
4956 var.padding = 0;
4957
4958 kvm_set_segment(vcpu, &var, seg);
4959 return;
4960}
4961
Avi Kivity717746e2011-04-20 13:37:53 +03004962static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
4963 u32 msr_index, u64 *pdata)
4964{
Paolo Bonzini609e36d2015-04-08 15:30:38 +02004965 struct msr_data msr;
4966 int r;
4967
4968 msr.index = msr_index;
4969 msr.host_initiated = false;
4970 r = kvm_get_msr(emul_to_vcpu(ctxt), &msr);
4971 if (r)
4972 return r;
4973
4974 *pdata = msr.data;
4975 return 0;
Avi Kivity717746e2011-04-20 13:37:53 +03004976}
4977
4978static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
4979 u32 msr_index, u64 data)
4980{
Will Auld8fe8ab42012-11-29 12:42:12 -08004981 struct msr_data msr;
4982
4983 msr.data = data;
4984 msr.index = msr_index;
4985 msr.host_initiated = false;
4986 return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
Avi Kivity717746e2011-04-20 13:37:53 +03004987}
4988
Paolo Bonzini64d60672015-05-07 11:36:11 +02004989static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
4990{
4991 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4992
4993 return vcpu->arch.smbase;
4994}
4995
4996static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
4997{
4998 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4999
5000 vcpu->arch.smbase = smbase;
5001}
5002
Nadav Amit67f4d422014-06-02 18:34:09 +03005003static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
5004 u32 pmc)
5005{
Wei Huangc6702c92015-06-19 13:44:45 +02005006 return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc);
Nadav Amit67f4d422014-06-02 18:34:09 +03005007}
5008
Avi Kivity222d21a2011-11-10 14:57:30 +02005009static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
5010 u32 pmc, u64 *pdata)
5011{
Wei Huangc6702c92015-06-19 13:44:45 +02005012 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
Avi Kivity222d21a2011-11-10 14:57:30 +02005013}
5014
Avi Kivity6c3287f2011-04-20 15:43:05 +03005015static void emulator_halt(struct x86_emulate_ctxt *ctxt)
5016{
5017 emul_to_vcpu(ctxt)->arch.halt_request = 1;
5018}
5019
Avi Kivity5037f6f2011-03-28 16:53:59 +02005020static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
5021{
5022 preempt_disable();
Avi Kivity5197b802011-04-20 15:55:40 +03005023 kvm_load_guest_fpu(emul_to_vcpu(ctxt));
Avi Kivity5037f6f2011-03-28 16:53:59 +02005024 /*
5025 * CR0.TS may reference the host fpu state, not the guest fpu state,
5026 * so it may be clear at this point.
5027 */
5028 clts();
5029}
5030
5031static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
5032{
5033 preempt_enable();
5034}
5035
Avi Kivity29535382011-04-20 13:37:53 +03005036static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
Joerg Roedel8a76d7f2011-04-04 12:39:27 +02005037 struct x86_instruction_info *info,
Avi Kivityc4f035c2011-04-04 12:39:22 +02005038 enum x86_intercept_stage stage)
5039{
Avi Kivity29535382011-04-20 13:37:53 +03005040 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
Avi Kivityc4f035c2011-04-04 12:39:22 +02005041}
5042
Avi Kivity0017f932012-06-07 14:10:16 +03005043static void emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
Stephan Bärwolfbdb42f52012-01-12 16:43:03 +01005044 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
5045{
Avi Kivity0017f932012-06-07 14:10:16 +03005046 kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx);
Stephan Bärwolfbdb42f52012-01-12 16:43:03 +01005047}
5048
Avi Kivitydd856ef2012-08-27 23:46:17 +03005049static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
5050{
5051 return kvm_register_read(emul_to_vcpu(ctxt), reg);
5052}
5053
5054static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
5055{
5056 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
5057}
5058
Nadav Amit801806d2015-01-26 09:32:23 +02005059static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
5060{
5061 kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
5062}
5063
Mathias Krause0225fb52012-08-30 01:30:16 +02005064static const struct x86_emulate_ops emulate_ops = {
Avi Kivitydd856ef2012-08-27 23:46:17 +03005065 .read_gpr = emulator_read_gpr,
5066 .write_gpr = emulator_write_gpr,
Gleb Natapov1871c602010-02-10 14:21:32 +02005067 .read_std = kvm_read_guest_virt_system,
Gleb Natapov2dafc6c2010-03-18 15:20:16 +02005068 .write_std = kvm_write_guest_virt_system,
Radim Krčmář7a036a62015-10-30 16:36:24 +01005069 .read_phys = kvm_read_guest_phys_system,
Gleb Natapov1871c602010-02-10 14:21:32 +02005070 .fetch = kvm_fetch_guest_virt,
Carsten Ottebbd9b642007-10-30 18:44:21 +01005071 .read_emulated = emulator_read_emulated,
5072 .write_emulated = emulator_write_emulated,
5073 .cmpxchg_emulated = emulator_cmpxchg_emulated,
Avi Kivity3cb16fe2011-04-20 15:38:44 +03005074 .invlpg = emulator_invlpg,
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02005075 .pio_in_emulated = emulator_pio_in_emulated,
5076 .pio_out_emulated = emulator_pio_out_emulated,
Avi Kivity1aa36612011-04-27 13:20:30 +03005077 .get_segment = emulator_get_segment,
5078 .set_segment = emulator_set_segment,
Gleb Natapov5951c442010-04-28 19:15:29 +03005079 .get_cached_segment_base = emulator_get_cached_segment_base,
Gleb Natapov2dafc6c2010-03-18 15:20:16 +02005080 .get_gdt = emulator_get_gdt,
Mohammed Gamal160ce1f2010-08-04 05:44:24 +03005081 .get_idt = emulator_get_idt,
Avi Kivity1ac9d0c2011-04-20 15:12:00 +03005082 .set_gdt = emulator_set_gdt,
5083 .set_idt = emulator_set_idt,
Gleb Natapov52a46612010-03-18 15:20:03 +02005084 .get_cr = emulator_get_cr,
5085 .set_cr = emulator_set_cr,
Gleb Natapov9c537242010-03-18 15:20:05 +02005086 .cpl = emulator_get_cpl,
Gleb Natapov35aa5372010-04-28 19:15:27 +03005087 .get_dr = emulator_get_dr,
5088 .set_dr = emulator_set_dr,
Paolo Bonzini64d60672015-05-07 11:36:11 +02005089 .get_smbase = emulator_get_smbase,
5090 .set_smbase = emulator_set_smbase,
Avi Kivity717746e2011-04-20 13:37:53 +03005091 .set_msr = emulator_set_msr,
5092 .get_msr = emulator_get_msr,
Nadav Amit67f4d422014-06-02 18:34:09 +03005093 .check_pmc = emulator_check_pmc,
Avi Kivity222d21a2011-11-10 14:57:30 +02005094 .read_pmc = emulator_read_pmc,
Avi Kivity6c3287f2011-04-20 15:43:05 +03005095 .halt = emulator_halt,
Avi Kivitybcaf5cc2011-04-20 15:53:23 +03005096 .wbinvd = emulator_wbinvd,
Avi Kivityd6aa1002011-04-20 15:47:13 +03005097 .fix_hypercall = emulator_fix_hypercall,
Avi Kivity5037f6f2011-03-28 16:53:59 +02005098 .get_fpu = emulator_get_fpu,
5099 .put_fpu = emulator_put_fpu,
Avi Kivityc4f035c2011-04-04 12:39:22 +02005100 .intercept = emulator_intercept,
Stephan Bärwolfbdb42f52012-01-12 16:43:03 +01005101 .get_cpuid = emulator_get_cpuid,
Nadav Amit801806d2015-01-26 09:32:23 +02005102 .set_nmi_mask = emulator_set_nmi_mask,
Carsten Ottebbd9b642007-10-30 18:44:21 +01005103};
5104
Gleb Natapov95cb2292010-04-28 19:15:43 +03005105static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
5106{
Paolo Bonzini37ccdcb2014-05-20 14:29:47 +02005107 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
Gleb Natapov95cb2292010-04-28 19:15:43 +03005108 /*
5109 * an sti; sti; sequence only disable interrupts for the first
5110 * instruction. So, if the last instruction, be it emulated or
5111 * not, left the system with the INT_STI flag enabled, it
5112 * means that the last instruction is an sti. We should not
5113 * leave the flag on in this case. The same goes for mov ss
5114 */
Paolo Bonzini37ccdcb2014-05-20 14:29:47 +02005115 if (int_shadow & mask)
5116 mask = 0;
Paolo Bonzini6addfc42014-03-27 11:29:28 +01005117 if (unlikely(int_shadow || mask)) {
Gleb Natapov95cb2292010-04-28 19:15:43 +03005118 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
Paolo Bonzini6addfc42014-03-27 11:29:28 +01005119 if (!mask)
5120 kvm_make_request(KVM_REQ_EVENT, vcpu);
5121 }
Gleb Natapov95cb2292010-04-28 19:15:43 +03005122}
5123
Paolo Bonzinief54bcf2014-09-04 19:46:15 +02005124static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
Gleb Natapov54b84862010-04-28 19:15:44 +03005125{
5126 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
Avi Kivityda9cb572010-11-22 17:53:21 +02005127 if (ctxt->exception.vector == PF_VECTOR)
Paolo Bonzinief54bcf2014-09-04 19:46:15 +02005128 return kvm_propagate_fault(vcpu, &ctxt->exception);
5129
5130 if (ctxt->exception.error_code_valid)
Avi Kivityda9cb572010-11-22 17:53:21 +02005131 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
5132 ctxt->exception.error_code);
Gleb Natapov54b84862010-04-28 19:15:44 +03005133 else
Avi Kivityda9cb572010-11-22 17:53:21 +02005134 kvm_queue_exception(vcpu, ctxt->exception.vector);
Paolo Bonzinief54bcf2014-09-04 19:46:15 +02005135 return false;
Gleb Natapov54b84862010-04-28 19:15:44 +03005136}
5137
Mohammed Gamal8ec47222010-08-16 00:47:01 +03005138static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
5139{
Takuya Yoshikawaadf52232011-05-25 11:06:16 +09005140 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
Mohammed Gamal8ec47222010-08-16 00:47:01 +03005141 int cs_db, cs_l;
5142
Mohammed Gamal8ec47222010-08-16 00:47:01 +03005143 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5144
Takuya Yoshikawaadf52232011-05-25 11:06:16 +09005145 ctxt->eflags = kvm_get_rflags(vcpu);
5146 ctxt->eip = kvm_rip_read(vcpu);
5147 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
5148 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
Nadav Amit42bf5492014-04-18 07:11:34 +03005149 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
Takuya Yoshikawaadf52232011-05-25 11:06:16 +09005150 cs_db ? X86EMUL_MODE_PROT32 :
5151 X86EMUL_MODE_PROT16;
Paolo Bonzinia5845392015-04-01 18:18:53 +02005152 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
Paolo Bonzini64d60672015-05-07 11:36:11 +02005153 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
5154 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
Paolo Bonzinia5845392015-04-01 18:18:53 +02005155 ctxt->emul_flags = vcpu->arch.hflags;
Takuya Yoshikawaadf52232011-05-25 11:06:16 +09005156
Avi Kivitydd856ef2012-08-27 23:46:17 +03005157 init_decode_cache(ctxt);
Gleb Natapov7ae441e2011-03-31 12:06:41 +02005158 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
Mohammed Gamal8ec47222010-08-16 00:47:01 +03005159}
5160
Serge E. Hallyn71f98332011-04-13 09:12:54 -05005161int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
Mohammed Gamal63995652010-09-19 14:34:06 +02005162{
Takuya Yoshikawa9d741912011-05-29 21:53:48 +09005163 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
Mohammed Gamal63995652010-09-19 14:34:06 +02005164 int ret;
5165
5166 init_emulate_ctxt(vcpu);
5167
Avi Kivity9dac77f2011-06-01 15:34:25 +03005168 ctxt->op_bytes = 2;
5169 ctxt->ad_bytes = 2;
5170 ctxt->_eip = ctxt->eip + inc_eip;
Takuya Yoshikawa9d741912011-05-29 21:53:48 +09005171 ret = emulate_int_real(ctxt, irq);
Mohammed Gamal63995652010-09-19 14:34:06 +02005172
5173 if (ret != X86EMUL_CONTINUE)
5174 return EMULATE_FAIL;
5175
Avi Kivity9dac77f2011-06-01 15:34:25 +03005176 ctxt->eip = ctxt->_eip;
Takuya Yoshikawa9d741912011-05-29 21:53:48 +09005177 kvm_rip_write(vcpu, ctxt->eip);
5178 kvm_set_rflags(vcpu, ctxt->eflags);
Mohammed Gamal63995652010-09-19 14:34:06 +02005179
5180 if (irq == NMI_VECTOR)
Avi Kivity7460fb4a2011-09-20 13:43:14 +03005181 vcpu->arch.nmi_pending = 0;
Mohammed Gamal63995652010-09-19 14:34:06 +02005182 else
5183 vcpu->arch.interrupt.pending = false;
5184
5185 return EMULATE_DONE;
5186}
5187EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
5188
Gleb Natapov6d77dbf2010-05-10 11:16:56 +03005189static int handle_emulation_failure(struct kvm_vcpu *vcpu)
5190{
Joerg Roedelfc3a9152010-11-29 17:51:49 +01005191 int r = EMULATE_DONE;
5192
Gleb Natapov6d77dbf2010-05-10 11:16:56 +03005193 ++vcpu->stat.insn_emulation_fail;
5194 trace_kvm_emulate_insn_failed(vcpu);
Nadav Amita2b9e6c2014-09-17 02:50:50 +03005195 if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
Joerg Roedelfc3a9152010-11-29 17:51:49 +01005196 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5197 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5198 vcpu->run->internal.ndata = 0;
5199 r = EMULATE_FAIL;
5200 }
Gleb Natapov6d77dbf2010-05-10 11:16:56 +03005201 kvm_queue_exception(vcpu, UD_VECTOR);
Joerg Roedelfc3a9152010-11-29 17:51:49 +01005202
5203 return r;
Gleb Natapov6d77dbf2010-05-10 11:16:56 +03005204}
5205
Xiao Guangrong93c05d32013-01-13 23:49:07 +08005206static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
Gleb Natapov991eebf2013-04-11 12:10:51 +03005207 bool write_fault_to_shadow_pgtable,
5208 int emulation_type)
Gleb Natapova6f177e2010-07-08 12:41:12 +03005209{
Xiao Guangrong95b3cf62013-01-13 23:46:52 +08005210 gpa_t gpa = cr2;
Dan Williamsba049e92016-01-15 16:56:11 -08005211 kvm_pfn_t pfn;
Gleb Natapova6f177e2010-07-08 12:41:12 +03005212
Gleb Natapov991eebf2013-04-11 12:10:51 +03005213 if (emulation_type & EMULTYPE_NO_REEXECUTE)
5214 return false;
5215
Xiao Guangrong95b3cf62013-01-13 23:46:52 +08005216 if (!vcpu->arch.mmu.direct_map) {
5217 /*
5218 * Write permission should be allowed since only
5219 * write access need to be emulated.
5220 */
5221 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
Gleb Natapov68be0802010-07-14 19:05:45 +03005222
Xiao Guangrong95b3cf62013-01-13 23:46:52 +08005223 /*
5224 * If the mapping is invalid in guest, let cpu retry
5225 * it to generate fault.
5226 */
5227 if (gpa == UNMAPPED_GVA)
5228 return true;
5229 }
Gleb Natapova6f177e2010-07-08 12:41:12 +03005230
Xiao Guangrong8e3d9d02012-08-21 10:57:42 +08005231 /*
5232 * Do not retry the unhandleable instruction if it faults on the
5233 * readonly host memory, otherwise it will goto a infinite loop:
5234 * retry instruction -> write #PF -> emulation fail -> retry
5235 * instruction -> ...
5236 */
5237 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
Xiao Guangrong95b3cf62013-01-13 23:46:52 +08005238
5239 /*
5240 * If the instruction failed on the error pfn, it can not be fixed,
5241 * report the error to userspace.
5242 */
5243 if (is_error_noslot_pfn(pfn))
5244 return false;
5245
5246 kvm_release_pfn_clean(pfn);
5247
5248 /* The instructions are well-emulated on direct mmu. */
5249 if (vcpu->arch.mmu.direct_map) {
5250 unsigned int indirect_shadow_pages;
5251
5252 spin_lock(&vcpu->kvm->mmu_lock);
5253 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
5254 spin_unlock(&vcpu->kvm->mmu_lock);
5255
5256 if (indirect_shadow_pages)
5257 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5258
Gleb Natapova6f177e2010-07-08 12:41:12 +03005259 return true;
Xiao Guangrong8e3d9d02012-08-21 10:57:42 +08005260 }
Gleb Natapova6f177e2010-07-08 12:41:12 +03005261
Xiao Guangrong95b3cf62013-01-13 23:46:52 +08005262 /*
5263 * if emulation was due to access to shadowed page table
5264 * and it failed try to unshadow page and re-enter the
5265 * guest to let CPU execute the instruction.
5266 */
5267 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
Xiao Guangrong93c05d32013-01-13 23:49:07 +08005268
5269 /*
5270 * If the access faults on its page table, it can not
5271 * be fixed by unprotecting shadow page and it should
5272 * be reported to userspace.
5273 */
5274 return !write_fault_to_shadow_pgtable;
Gleb Natapova6f177e2010-07-08 12:41:12 +03005275}
5276
Xiao Guangrong1cb3f3a2011-09-22 17:02:48 +08005277static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
5278 unsigned long cr2, int emulation_type)
5279{
5280 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5281 unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
5282
5283 last_retry_eip = vcpu->arch.last_retry_eip;
5284 last_retry_addr = vcpu->arch.last_retry_addr;
5285
5286 /*
5287 * If the emulation is caused by #PF and it is non-page_table
5288 * writing instruction, it means the VM-EXIT is caused by shadow
5289 * page protected, we can zap the shadow page and retry this
5290 * instruction directly.
5291 *
5292 * Note: if the guest uses a non-page-table modifying instruction
5293 * on the PDE that points to the instruction, then we will unmap
5294 * the instruction and go to an infinite loop. So, we cache the
5295 * last retried eip and the last fault address, if we meet the eip
5296 * and the address again, we can break out of the potential infinite
5297 * loop.
5298 */
5299 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
5300
5301 if (!(emulation_type & EMULTYPE_RETRY))
5302 return false;
5303
5304 if (x86_page_table_writing_insn(ctxt))
5305 return false;
5306
5307 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
5308 return false;
5309
5310 vcpu->arch.last_retry_eip = ctxt->eip;
5311 vcpu->arch.last_retry_addr = cr2;
5312
5313 if (!vcpu->arch.mmu.direct_map)
5314 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5315
Xiao Guangrong22368022013-01-13 23:44:12 +08005316 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
Xiao Guangrong1cb3f3a2011-09-22 17:02:48 +08005317
5318 return true;
5319}
5320
Gleb Natapov716d51a2012-09-03 15:24:26 +03005321static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
5322static int complete_emulated_pio(struct kvm_vcpu *vcpu);
5323
Paolo Bonzini64d60672015-05-07 11:36:11 +02005324static void kvm_smm_changed(struct kvm_vcpu *vcpu)
Paolo Bonzinia5845392015-04-01 18:18:53 +02005325{
Paolo Bonzini64d60672015-05-07 11:36:11 +02005326 if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
Paolo Bonzini660a5d52015-05-05 11:50:23 +02005327 /* This is a good place to trace that we are exiting SMM. */
5328 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
5329
Paolo Bonzinic43203c2016-06-01 22:26:00 +02005330 /* Process a latched INIT or SMI, if any. */
5331 kvm_make_request(KVM_REQ_EVENT, vcpu);
Paolo Bonzini64d60672015-05-07 11:36:11 +02005332 }
Paolo Bonzini699023e2015-05-18 15:03:39 +02005333
5334 kvm_mmu_reset_context(vcpu);
Paolo Bonzini64d60672015-05-07 11:36:11 +02005335}
5336
5337static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags)
5338{
5339 unsigned changed = vcpu->arch.hflags ^ emul_flags;
5340
Paolo Bonzinia5845392015-04-01 18:18:53 +02005341 vcpu->arch.hflags = emul_flags;
Paolo Bonzini64d60672015-05-07 11:36:11 +02005342
5343 if (changed & HF_SMM_MASK)
5344 kvm_smm_changed(vcpu);
Paolo Bonzinia5845392015-04-01 18:18:53 +02005345}
5346
Paolo Bonzini4a1e10d2013-05-30 11:48:30 +02005347static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
5348 unsigned long *db)
5349{
5350 u32 dr6 = 0;
5351 int i;
5352 u32 enable, rwlen;
5353
5354 enable = dr7;
5355 rwlen = dr7 >> 16;
5356 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
5357 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
5358 dr6 |= (1 << i);
5359 return dr6;
5360}
5361
Paolo Bonzini6addfc42014-03-27 11:29:28 +01005362static void kvm_vcpu_check_singlestep(struct kvm_vcpu *vcpu, unsigned long rflags, int *r)
Paolo Bonzini663f4c62013-06-25 18:32:07 +02005363{
5364 struct kvm_run *kvm_run = vcpu->run;
5365
5366 /*
Paolo Bonzini6addfc42014-03-27 11:29:28 +01005367 * rflags is the old, "raw" value of the flags. The new value has
5368 * not been saved yet.
Paolo Bonzini663f4c62013-06-25 18:32:07 +02005369 *
5370 * This is correct even for TF set by the guest, because "the
5371 * processor will not generate this exception after the instruction
5372 * that sets the TF flag".
5373 */
Paolo Bonzini663f4c62013-06-25 18:32:07 +02005374 if (unlikely(rflags & X86_EFLAGS_TF)) {
5375 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
Nadav Amit6f43ed02014-07-15 17:37:46 +03005376 kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 |
5377 DR6_RTM;
Paolo Bonzini663f4c62013-06-25 18:32:07 +02005378 kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
5379 kvm_run->debug.arch.exception = DB_VECTOR;
5380 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5381 *r = EMULATE_USER_EXIT;
5382 } else {
5383 vcpu->arch.emulate_ctxt.eflags &= ~X86_EFLAGS_TF;
5384 /*
5385 * "Certain debug exceptions may clear bit 0-3. The
5386 * remaining contents of the DR6 register are never
5387 * cleared by the processor".
5388 */
5389 vcpu->arch.dr6 &= ~15;
Nadav Amit6f43ed02014-07-15 17:37:46 +03005390 vcpu->arch.dr6 |= DR6_BS | DR6_RTM;
Paolo Bonzini663f4c62013-06-25 18:32:07 +02005391 kvm_queue_exception(vcpu, DB_VECTOR);
5392 }
5393 }
5394}
5395
Paolo Bonzini4a1e10d2013-05-30 11:48:30 +02005396static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
5397{
Paolo Bonzini4a1e10d2013-05-30 11:48:30 +02005398 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
5399 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
Nadav Amit82b32772014-11-02 11:54:45 +02005400 struct kvm_run *kvm_run = vcpu->run;
5401 unsigned long eip = kvm_get_linear_rip(vcpu);
5402 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
Paolo Bonzini4a1e10d2013-05-30 11:48:30 +02005403 vcpu->arch.guest_debug_dr7,
5404 vcpu->arch.eff_db);
5405
5406 if (dr6 != 0) {
Nadav Amit6f43ed02014-07-15 17:37:46 +03005407 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
Nadav Amit82b32772014-11-02 11:54:45 +02005408 kvm_run->debug.arch.pc = eip;
Paolo Bonzini4a1e10d2013-05-30 11:48:30 +02005409 kvm_run->debug.arch.exception = DB_VECTOR;
5410 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5411 *r = EMULATE_USER_EXIT;
5412 return true;
5413 }
5414 }
5415
Nadav Amit4161a562014-07-17 01:19:31 +03005416 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
5417 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
Nadav Amit82b32772014-11-02 11:54:45 +02005418 unsigned long eip = kvm_get_linear_rip(vcpu);
5419 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
Paolo Bonzini4a1e10d2013-05-30 11:48:30 +02005420 vcpu->arch.dr7,
5421 vcpu->arch.db);
5422
5423 if (dr6 != 0) {
5424 vcpu->arch.dr6 &= ~15;
Nadav Amit6f43ed02014-07-15 17:37:46 +03005425 vcpu->arch.dr6 |= dr6 | DR6_RTM;
Paolo Bonzini4a1e10d2013-05-30 11:48:30 +02005426 kvm_queue_exception(vcpu, DB_VECTOR);
5427 *r = EMULATE_DONE;
5428 return true;
5429 }
5430 }
5431
5432 return false;
5433}
5434
Andre Przywara51d8b662010-12-21 11:12:02 +01005435int x86_emulate_instruction(struct kvm_vcpu *vcpu,
5436 unsigned long cr2,
Andre Przywaradc25e892010-12-21 11:12:07 +01005437 int emulation_type,
5438 void *insn,
5439 int insn_len)
Carsten Ottebbd9b642007-10-30 18:44:21 +01005440{
Gleb Natapov95cb2292010-04-28 19:15:43 +03005441 int r;
Takuya Yoshikawa9d741912011-05-29 21:53:48 +09005442 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
Gleb Natapov7ae441e2011-03-31 12:06:41 +02005443 bool writeback = true;
Xiao Guangrong93c05d32013-01-13 23:49:07 +08005444 bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
Carsten Ottebbd9b642007-10-30 18:44:21 +01005445
Xiao Guangrong93c05d32013-01-13 23:49:07 +08005446 /*
5447 * Clear write_fault_to_shadow_pgtable here to ensure it is
5448 * never reused.
5449 */
5450 vcpu->arch.write_fault_to_shadow_pgtable = false;
Avi Kivity26eef702008-07-03 14:59:22 +03005451 kvm_clear_exception_queue(vcpu);
Gleb Natapov8d7d81022011-04-12 12:36:21 +03005452
Sheng Yang571008d2008-01-02 14:49:22 +08005453 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
Mohammed Gamal8ec47222010-08-16 00:47:01 +03005454 init_emulate_ctxt(vcpu);
Paolo Bonzini4a1e10d2013-05-30 11:48:30 +02005455
5456 /*
5457 * We will reenter on the same instruction since
5458 * we do not set complete_userspace_io. This does not
5459 * handle watchpoints yet, those would be handled in
5460 * the emulate_ops.
5461 */
5462 if (kvm_vcpu_check_breakpoint(vcpu, &r))
5463 return r;
5464
Takuya Yoshikawa9d741912011-05-29 21:53:48 +09005465 ctxt->interruptibility = 0;
5466 ctxt->have_exception = false;
Paolo Bonzinie0ad0b42014-08-20 10:08:23 +02005467 ctxt->exception.vector = -1;
Takuya Yoshikawa9d741912011-05-29 21:53:48 +09005468 ctxt->perm_ok = false;
Carsten Ottebbd9b642007-10-30 18:44:21 +01005469
Borislav Petkovb51e9742013-09-22 16:44:52 +02005470 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
Avi Kivity40059962011-02-01 16:32:04 +02005471
Takuya Yoshikawa9d741912011-05-29 21:53:48 +09005472 r = x86_decode_insn(ctxt, insn, insn_len);
Joerg Roedeld47f00a2010-09-10 17:30:56 +02005473
Avi Kivitye46479f2010-04-11 13:05:16 +03005474 trace_kvm_emulate_insn_start(vcpu);
Avi Kivityf2b57562007-11-18 15:17:51 +02005475 ++vcpu->stat.insn_emulation;
Takuya Yoshikawa1d2887e2011-07-30 18:03:34 +09005476 if (r != EMULATION_OK) {
Avi Kivity40059962011-02-01 16:32:04 +02005477 if (emulation_type & EMULTYPE_TRAP_UD)
5478 return EMULATE_FAIL;
Gleb Natapov991eebf2013-04-11 12:10:51 +03005479 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5480 emulation_type))
Carsten Ottebbd9b642007-10-30 18:44:21 +01005481 return EMULATE_DONE;
Gleb Natapov6d77dbf2010-05-10 11:16:56 +03005482 if (emulation_type & EMULTYPE_SKIP)
5483 return EMULATE_FAIL;
5484 return handle_emulation_failure(vcpu);
Carsten Ottebbd9b642007-10-30 18:44:21 +01005485 }
5486 }
5487
Gleb Natapovba8afb62009-04-12 13:36:57 +03005488 if (emulation_type & EMULTYPE_SKIP) {
Avi Kivity9dac77f2011-06-01 15:34:25 +03005489 kvm_rip_write(vcpu, ctxt->_eip);
Nadav Amitbb663c72014-07-21 14:37:26 +03005490 if (ctxt->eflags & X86_EFLAGS_RF)
5491 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
Gleb Natapovba8afb62009-04-12 13:36:57 +03005492 return EMULATE_DONE;
5493 }
5494
Xiao Guangrong1cb3f3a2011-09-22 17:02:48 +08005495 if (retry_instruction(ctxt, cr2, emulation_type))
5496 return EMULATE_DONE;
5497
Gleb Natapov7ae441e2011-03-31 12:06:41 +02005498 /* this is needed for vmware backdoor interface to work since it
Gleb Natapov4d2179e2010-04-28 19:15:42 +03005499 changes registers values during IO operation */
Gleb Natapov7ae441e2011-03-31 12:06:41 +02005500 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
5501 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
Avi Kivitydd856ef2012-08-27 23:46:17 +03005502 emulator_invalidate_register_cache(ctxt);
Gleb Natapov7ae441e2011-03-31 12:06:41 +02005503 }
Gleb Natapov4d2179e2010-04-28 19:15:42 +03005504
Gleb Natapov5cd21912010-03-18 15:20:26 +02005505restart:
Takuya Yoshikawa9d741912011-05-29 21:53:48 +09005506 r = x86_emulate_insn(ctxt);
Gleb Natapovc3cd7ff2010-04-28 19:15:35 +03005507
Joerg Roedel775fde82011-04-04 12:39:24 +02005508 if (r == EMULATION_INTERCEPTED)
5509 return EMULATE_DONE;
5510
Gleb Natapovd2ddd1c2010-08-25 12:47:43 +03005511 if (r == EMULATION_FAILED) {
Gleb Natapov991eebf2013-04-11 12:10:51 +03005512 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5513 emulation_type))
Gleb Natapovc3cd7ff2010-04-28 19:15:35 +03005514 return EMULATE_DONE;
5515
Gleb Natapov6d77dbf2010-05-10 11:16:56 +03005516 return handle_emulation_failure(vcpu);
Carsten Ottebbd9b642007-10-30 18:44:21 +01005517 }
5518
Takuya Yoshikawa9d741912011-05-29 21:53:48 +09005519 if (ctxt->have_exception) {
Gleb Natapovd2ddd1c2010-08-25 12:47:43 +03005520 r = EMULATE_DONE;
Paolo Bonzinief54bcf2014-09-04 19:46:15 +02005521 if (inject_emulated_exception(vcpu))
5522 return r;
Gleb Natapovd2ddd1c2010-08-25 12:47:43 +03005523 } else if (vcpu->arch.pio.count) {
Paolo Bonzini0912c972013-08-27 15:41:43 +02005524 if (!vcpu->arch.pio.in) {
5525 /* FIXME: return into emulator if single-stepping. */
Gleb Natapove85d28f2010-07-29 15:11:52 +03005526 vcpu->arch.pio.count = 0;
Paolo Bonzini0912c972013-08-27 15:41:43 +02005527 } else {
Gleb Natapov7ae441e2011-03-31 12:06:41 +02005528 writeback = false;
Gleb Natapov716d51a2012-09-03 15:24:26 +03005529 vcpu->arch.complete_userspace_io = complete_emulated_pio;
5530 }
Paolo Bonziniac0a48c2013-06-25 18:24:41 +02005531 r = EMULATE_USER_EXIT;
Gleb Natapov7ae441e2011-03-31 12:06:41 +02005532 } else if (vcpu->mmio_needed) {
5533 if (!vcpu->mmio_is_write)
5534 writeback = false;
Paolo Bonziniac0a48c2013-06-25 18:24:41 +02005535 r = EMULATE_USER_EXIT;
Gleb Natapov716d51a2012-09-03 15:24:26 +03005536 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
Gleb Natapov7ae441e2011-03-31 12:06:41 +02005537 } else if (r == EMULATION_RESTART)
Gleb Natapove85d28f2010-07-29 15:11:52 +03005538 goto restart;
Gleb Natapovd2ddd1c2010-08-25 12:47:43 +03005539 else
5540 r = EMULATE_DONE;
Gleb Natapove85d28f2010-07-29 15:11:52 +03005541
Gleb Natapov7ae441e2011-03-31 12:06:41 +02005542 if (writeback) {
Paolo Bonzini6addfc42014-03-27 11:29:28 +01005543 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
Takuya Yoshikawa9d741912011-05-29 21:53:48 +09005544 toggle_interruptibility(vcpu, ctxt->interruptibility);
Gleb Natapov7ae441e2011-03-31 12:06:41 +02005545 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
Paolo Bonzinia5845392015-04-01 18:18:53 +02005546 if (vcpu->arch.hflags != ctxt->emul_flags)
5547 kvm_set_hflags(vcpu, ctxt->emul_flags);
Takuya Yoshikawa9d741912011-05-29 21:53:48 +09005548 kvm_rip_write(vcpu, ctxt->eip);
Paolo Bonzini663f4c62013-06-25 18:32:07 +02005549 if (r == EMULATE_DONE)
Paolo Bonzini6addfc42014-03-27 11:29:28 +01005550 kvm_vcpu_check_singlestep(vcpu, rflags, &r);
Nadav Amit38827db2014-11-02 11:54:53 +02005551 if (!ctxt->have_exception ||
5552 exception_type(ctxt->exception.vector) == EXCPT_TRAP)
5553 __kvm_set_rflags(vcpu, ctxt->eflags);
Paolo Bonzini6addfc42014-03-27 11:29:28 +01005554
5555 /*
5556 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
5557 * do nothing, and it will be requested again as soon as
5558 * the shadow expires. But we still need to check here,
5559 * because POPF has no interrupt shadow.
5560 */
5561 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
5562 kvm_make_request(KVM_REQ_EVENT, vcpu);
Gleb Natapov7ae441e2011-03-31 12:06:41 +02005563 } else
5564 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
Gleb Natapov3457e412010-04-28 19:15:38 +03005565
Gleb Natapove85d28f2010-07-29 15:11:52 +03005566 return r;
Carsten Ottebbd9b642007-10-30 18:44:21 +01005567}
Andre Przywara51d8b662010-12-21 11:12:02 +01005568EXPORT_SYMBOL_GPL(x86_emulate_instruction);
Carsten Ottebbd9b642007-10-30 18:44:21 +01005569
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02005570int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
Carsten Ottede7d7892007-10-30 18:44:25 +01005571{
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02005572 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
Avi Kivityca1d4a92011-04-20 13:37:53 +03005573 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
5574 size, port, &val, 1);
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02005575 /* do not return to emulator after return from userspace */
Gleb Natapov79729952010-03-18 15:20:24 +02005576 vcpu->arch.pio.count = 0;
Izik Eidus0f346072008-12-29 01:42:20 +02005577 return ret;
Carsten Ottede7d7892007-10-30 18:44:25 +01005578}
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02005579EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
Carsten Ottede7d7892007-10-30 18:44:25 +01005580
Sebastian Andrzej Siewior251a5fd2016-07-13 17:16:33 +00005581static int kvmclock_cpu_down_prep(unsigned int cpu)
Gerd Hoffmannc8076602009-02-04 17:52:04 +01005582{
Tejun Heo0a3aee02010-12-18 16:28:55 +01005583 __this_cpu_write(cpu_tsc_khz, 0);
Sebastian Andrzej Siewior251a5fd2016-07-13 17:16:33 +00005584 return 0;
Zachary Amsden8cfdc002010-08-19 22:07:21 -10005585}
5586
5587static void tsc_khz_changed(void *data)
5588{
5589 struct cpufreq_freqs *freq = data;
5590 unsigned long khz = 0;
5591
5592 if (data)
5593 khz = freq->new;
5594 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5595 khz = cpufreq_quick_get(raw_smp_processor_id());
5596 if (!khz)
5597 khz = tsc_khz;
Tejun Heo0a3aee02010-12-18 16:28:55 +01005598 __this_cpu_write(cpu_tsc_khz, khz);
Gerd Hoffmannc8076602009-02-04 17:52:04 +01005599}
5600
Gerd Hoffmannc8076602009-02-04 17:52:04 +01005601static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
5602 void *data)
5603{
5604 struct cpufreq_freqs *freq = data;
5605 struct kvm *kvm;
5606 struct kvm_vcpu *vcpu;
5607 int i, send_ipi = 0;
5608
Zachary Amsden8cfdc002010-08-19 22:07:21 -10005609 /*
5610 * We allow guests to temporarily run on slowing clocks,
5611 * provided we notify them after, or to run on accelerating
5612 * clocks, provided we notify them before. Thus time never
5613 * goes backwards.
5614 *
5615 * However, we have a problem. We can't atomically update
5616 * the frequency of a given CPU from this function; it is
5617 * merely a notifier, which can be called from any CPU.
5618 * Changing the TSC frequency at arbitrary points in time
5619 * requires a recomputation of local variables related to
5620 * the TSC for each VCPU. We must flag these local variables
5621 * to be updated and be sure the update takes place with the
5622 * new frequency before any guests proceed.
5623 *
5624 * Unfortunately, the combination of hotplug CPU and frequency
5625 * change creates an intractable locking scenario; the order
5626 * of when these callouts happen is undefined with respect to
5627 * CPU hotplug, and they can race with each other. As such,
5628 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
5629 * undefined; you can actually have a CPU frequency change take
5630 * place in between the computation of X and the setting of the
5631 * variable. To protect against this problem, all updates of
5632 * the per_cpu tsc_khz variable are done in an interrupt
5633 * protected IPI, and all callers wishing to update the value
5634 * must wait for a synchronous IPI to complete (which is trivial
5635 * if the caller is on the CPU already). This establishes the
5636 * necessary total order on variable updates.
5637 *
5638 * Note that because a guest time update may take place
5639 * anytime after the setting of the VCPU's request bit, the
5640 * correct TSC value must be set before the request. However,
5641 * to ensure the update actually makes it to any guest which
5642 * starts running in hardware virtualization between the set
5643 * and the acquisition of the spinlock, we must also ping the
5644 * CPU after setting the request bit.
5645 *
5646 */
5647
Gerd Hoffmannc8076602009-02-04 17:52:04 +01005648 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
5649 return 0;
5650 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
5651 return 0;
Zachary Amsden8cfdc002010-08-19 22:07:21 -10005652
5653 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
Gerd Hoffmannc8076602009-02-04 17:52:04 +01005654
Paolo Bonzini2f303b72013-09-25 13:53:07 +02005655 spin_lock(&kvm_lock);
Gerd Hoffmannc8076602009-02-04 17:52:04 +01005656 list_for_each_entry(kvm, &vm_list, vm_list) {
Gleb Natapov988a2ca2009-06-09 15:56:29 +03005657 kvm_for_each_vcpu(i, vcpu, kvm) {
Gerd Hoffmannc8076602009-02-04 17:52:04 +01005658 if (vcpu->cpu != freq->cpu)
5659 continue;
Zachary Amsdenc2855452010-09-18 14:38:15 -10005660 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
Gerd Hoffmannc8076602009-02-04 17:52:04 +01005661 if (vcpu->cpu != smp_processor_id())
Zachary Amsden8cfdc002010-08-19 22:07:21 -10005662 send_ipi = 1;
Gerd Hoffmannc8076602009-02-04 17:52:04 +01005663 }
5664 }
Paolo Bonzini2f303b72013-09-25 13:53:07 +02005665 spin_unlock(&kvm_lock);
Gerd Hoffmannc8076602009-02-04 17:52:04 +01005666
5667 if (freq->old < freq->new && send_ipi) {
5668 /*
5669 * We upscale the frequency. Must make the guest
5670 * doesn't see old kvmclock values while running with
5671 * the new frequency, otherwise we risk the guest sees
5672 * time go backwards.
5673 *
5674 * In case we update the frequency for another cpu
5675 * (which might be in guest context) send an interrupt
5676 * to kick the cpu out of guest context. Next time
5677 * guest context is entered kvmclock will be updated,
5678 * so the guest will not see stale values.
5679 */
Zachary Amsden8cfdc002010-08-19 22:07:21 -10005680 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
Gerd Hoffmannc8076602009-02-04 17:52:04 +01005681 }
5682 return 0;
5683}
5684
5685static struct notifier_block kvmclock_cpufreq_notifier_block = {
Zachary Amsden8cfdc002010-08-19 22:07:21 -10005686 .notifier_call = kvmclock_cpufreq_notifier
5687};
5688
Sebastian Andrzej Siewior251a5fd2016-07-13 17:16:33 +00005689static int kvmclock_cpu_online(unsigned int cpu)
Zachary Amsden8cfdc002010-08-19 22:07:21 -10005690{
Sebastian Andrzej Siewior251a5fd2016-07-13 17:16:33 +00005691 tsc_khz_changed(NULL);
5692 return 0;
Zachary Amsden8cfdc002010-08-19 22:07:21 -10005693}
5694
Zachary Amsdenb820cc02009-09-29 11:38:34 -10005695static void kvm_timer_init(void)
5696{
5697 int cpu;
5698
Zachary Amsdenc2855452010-09-18 14:38:15 -10005699 max_tsc_khz = tsc_khz;
Srivatsa S. Bhat460dd422014-03-11 02:09:01 +05305700
Zachary Amsdenb820cc02009-09-29 11:38:34 -10005701 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
Zachary Amsdenc2855452010-09-18 14:38:15 -10005702#ifdef CONFIG_CPU_FREQ
5703 struct cpufreq_policy policy;
5704 memset(&policy, 0, sizeof(policy));
Avi Kivity3e26f232010-12-16 12:16:34 +02005705 cpu = get_cpu();
5706 cpufreq_get_policy(&policy, cpu);
Zachary Amsdenc2855452010-09-18 14:38:15 -10005707 if (policy.cpuinfo.max_freq)
5708 max_tsc_khz = policy.cpuinfo.max_freq;
Avi Kivity3e26f232010-12-16 12:16:34 +02005709 put_cpu();
Zachary Amsdenc2855452010-09-18 14:38:15 -10005710#endif
Zachary Amsdenb820cc02009-09-29 11:38:34 -10005711 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
5712 CPUFREQ_TRANSITION_NOTIFIER);
5713 }
Zachary Amsdenc2855452010-09-18 14:38:15 -10005714 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
Srivatsa S. Bhat460dd422014-03-11 02:09:01 +05305715
Sebastian Andrzej Siewior251a5fd2016-07-13 17:16:33 +00005716 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "AP_X86_KVM_CLK_ONLINE",
5717 kvmclock_cpu_online, kvmclock_cpu_down_prep);
Zachary Amsdenb820cc02009-09-29 11:38:34 -10005718}
5719
Zhang, Yanminff9d07a2010-04-19 13:32:45 +08005720static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5721
Gleb Natapovf5132b02011-11-10 14:57:22 +02005722int kvm_is_in_guest(void)
Zhang, Yanminff9d07a2010-04-19 13:32:45 +08005723{
Alex,Shi086c9852011-10-20 15:34:01 +08005724 return __this_cpu_read(current_vcpu) != NULL;
Zhang, Yanminff9d07a2010-04-19 13:32:45 +08005725}
5726
5727static int kvm_is_user_mode(void)
5728{
5729 int user_mode = 3;
Zhang, Yanmindcf46b92010-04-20 10:13:58 +08005730
Alex,Shi086c9852011-10-20 15:34:01 +08005731 if (__this_cpu_read(current_vcpu))
5732 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
Zhang, Yanmindcf46b92010-04-20 10:13:58 +08005733
Zhang, Yanminff9d07a2010-04-19 13:32:45 +08005734 return user_mode != 0;
5735}
5736
5737static unsigned long kvm_get_guest_ip(void)
5738{
5739 unsigned long ip = 0;
Zhang, Yanmindcf46b92010-04-20 10:13:58 +08005740
Alex,Shi086c9852011-10-20 15:34:01 +08005741 if (__this_cpu_read(current_vcpu))
5742 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
Zhang, Yanmindcf46b92010-04-20 10:13:58 +08005743
Zhang, Yanminff9d07a2010-04-19 13:32:45 +08005744 return ip;
5745}
5746
5747static struct perf_guest_info_callbacks kvm_guest_cbs = {
5748 .is_in_guest = kvm_is_in_guest,
5749 .is_user_mode = kvm_is_user_mode,
5750 .get_guest_ip = kvm_get_guest_ip,
5751};
5752
5753void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
5754{
Alex,Shi086c9852011-10-20 15:34:01 +08005755 __this_cpu_write(current_vcpu, vcpu);
Zhang, Yanminff9d07a2010-04-19 13:32:45 +08005756}
5757EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
5758
5759void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
5760{
Alex,Shi086c9852011-10-20 15:34:01 +08005761 __this_cpu_write(current_vcpu, NULL);
Zhang, Yanminff9d07a2010-04-19 13:32:45 +08005762}
5763EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
5764
Xiao Guangrongce88dec2011-07-12 03:33:44 +08005765static void kvm_set_mmio_spte_mask(void)
5766{
5767 u64 mask;
5768 int maxphyaddr = boot_cpu_data.x86_phys_bits;
5769
5770 /*
5771 * Set the reserved bits and the present bit of an paging-structure
5772 * entry to generate page fault with PFER.RSV = 1.
5773 */
Xiao Guangrong885032b2013-06-07 16:51:23 +08005774 /* Mask the reserved physical address bits. */
Tiejun Chend1431482014-09-01 18:44:04 +08005775 mask = rsvd_bits(maxphyaddr, 51);
Xiao Guangrong885032b2013-06-07 16:51:23 +08005776
5777 /* Bit 62 is always reserved for 32bit host. */
5778 mask |= 0x3ull << 62;
5779
5780 /* Set the present bit. */
Xiao Guangrongce88dec2011-07-12 03:33:44 +08005781 mask |= 1ull;
5782
5783#ifdef CONFIG_X86_64
5784 /*
5785 * If reserved bit is not supported, clear the present bit to disable
5786 * mmio page fault.
5787 */
5788 if (maxphyaddr == 52)
5789 mask &= ~1ull;
5790#endif
5791
5792 kvm_mmu_set_mmio_spte_mask(mask);
5793}
5794
Marcelo Tosatti16e8d742012-11-27 23:29:00 -02005795#ifdef CONFIG_X86_64
5796static void pvclock_gtod_update_fn(struct work_struct *work)
5797{
Marcelo Tosattid8281992012-11-27 23:29:01 -02005798 struct kvm *kvm;
5799
5800 struct kvm_vcpu *vcpu;
5801 int i;
5802
Paolo Bonzini2f303b72013-09-25 13:53:07 +02005803 spin_lock(&kvm_lock);
Marcelo Tosattid8281992012-11-27 23:29:01 -02005804 list_for_each_entry(kvm, &vm_list, vm_list)
5805 kvm_for_each_vcpu(i, vcpu, kvm)
Guo Hui Liu105b21b2014-09-12 13:43:19 +08005806 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
Marcelo Tosattid8281992012-11-27 23:29:01 -02005807 atomic_set(&kvm_guest_has_master_clock, 0);
Paolo Bonzini2f303b72013-09-25 13:53:07 +02005808 spin_unlock(&kvm_lock);
Marcelo Tosatti16e8d742012-11-27 23:29:00 -02005809}
5810
5811static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
5812
5813/*
5814 * Notification about pvclock gtod data update.
5815 */
5816static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
5817 void *priv)
5818{
5819 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
5820 struct timekeeper *tk = priv;
5821
5822 update_pvclock_gtod(tk);
5823
5824 /* disable master clock if host does not trust, or does not
5825 * use, TSC clocksource
5826 */
5827 if (gtod->clock.vclock_mode != VCLOCK_TSC &&
5828 atomic_read(&kvm_guest_has_master_clock) != 0)
5829 queue_work(system_long_wq, &pvclock_gtod_work);
5830
5831 return 0;
5832}
5833
5834static struct notifier_block pvclock_gtod_notifier = {
5835 .notifier_call = pvclock_gtod_notify,
5836};
5837#endif
5838
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08005839int kvm_arch_init(void *opaque)
Carsten Otte043405e2007-10-10 17:16:19 +02005840{
Zachary Amsdenb820cc02009-09-29 11:38:34 -10005841 int r;
Mathias Krause6b61edf2013-06-26 20:36:23 +02005842 struct kvm_x86_ops *ops = opaque;
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08005843
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08005844 if (kvm_x86_ops) {
5845 printk(KERN_ERR "kvm: already loaded the other module\n");
Zhang Xiantao56c6d282007-11-18 20:43:21 +08005846 r = -EEXIST;
5847 goto out;
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08005848 }
5849
5850 if (!ops->cpu_has_kvm_support()) {
5851 printk(KERN_ERR "kvm: no hardware support\n");
Zhang Xiantao56c6d282007-11-18 20:43:21 +08005852 r = -EOPNOTSUPP;
5853 goto out;
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08005854 }
5855 if (ops->disabled_by_bios()) {
5856 printk(KERN_ERR "kvm: disabled by bios\n");
Zhang Xiantao56c6d282007-11-18 20:43:21 +08005857 r = -EOPNOTSUPP;
5858 goto out;
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08005859 }
5860
Marcelo Tosatti013f6a52013-01-03 11:41:39 -02005861 r = -ENOMEM;
5862 shared_msrs = alloc_percpu(struct kvm_shared_msrs);
5863 if (!shared_msrs) {
5864 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
5865 goto out;
5866 }
5867
Avi Kivity97db56c2008-01-13 13:23:56 +02005868 r = kvm_mmu_module_init();
5869 if (r)
Marcelo Tosatti013f6a52013-01-03 11:41:39 -02005870 goto out_free_percpu;
Avi Kivity97db56c2008-01-13 13:23:56 +02005871
Xiao Guangrongce88dec2011-07-12 03:33:44 +08005872 kvm_set_mmio_spte_mask();
Avi Kivity97db56c2008-01-13 13:23:56 +02005873
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08005874 kvm_x86_ops = ops;
Paolo Bonzini920c8372014-03-26 15:54:00 +01005875
Sheng Yang7b523452008-04-25 21:13:50 +08005876 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
Bandan Dasffb128c2016-07-12 18:18:49 -04005877 PT_DIRTY_MASK, PT64_NX_MASK, 0,
5878 PT_PRESENT_MASK);
Zachary Amsdenb820cc02009-09-29 11:38:34 -10005879 kvm_timer_init();
Gerd Hoffmannc8076602009-02-04 17:52:04 +01005880
Zhang, Yanminff9d07a2010-04-19 13:32:45 +08005881 perf_register_guest_info_callbacks(&kvm_guest_cbs);
5882
Borislav Petkovd366bf72016-04-04 22:25:02 +02005883 if (boot_cpu_has(X86_FEATURE_XSAVE))
Dexuan Cui2acf9232010-06-10 11:27:12 +08005884 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
5885
Gleb Natapovc5cc4212012-08-05 15:58:30 +03005886 kvm_lapic_init();
Marcelo Tosatti16e8d742012-11-27 23:29:00 -02005887#ifdef CONFIG_X86_64
5888 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
5889#endif
5890
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08005891 return 0;
Zhang Xiantao56c6d282007-11-18 20:43:21 +08005892
Marcelo Tosatti013f6a52013-01-03 11:41:39 -02005893out_free_percpu:
5894 free_percpu(shared_msrs);
Zhang Xiantao56c6d282007-11-18 20:43:21 +08005895out:
Zhang Xiantao56c6d282007-11-18 20:43:21 +08005896 return r;
Carsten Otte043405e2007-10-10 17:16:19 +02005897}
Hollis Blanchard8776e512007-10-31 17:24:24 -05005898
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08005899void kvm_arch_exit(void)
5900{
Zhang, Yanminff9d07a2010-04-19 13:32:45 +08005901 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
5902
Jan Kiszka888d2562009-04-17 19:24:58 +02005903 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5904 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
5905 CPUFREQ_TRANSITION_NOTIFIER);
Sebastian Andrzej Siewior251a5fd2016-07-13 17:16:33 +00005906 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
Marcelo Tosatti16e8d742012-11-27 23:29:00 -02005907#ifdef CONFIG_X86_64
5908 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
5909#endif
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08005910 kvm_x86_ops = NULL;
Zhang Xiantao56c6d282007-11-18 20:43:21 +08005911 kvm_mmu_module_exit();
Marcelo Tosatti013f6a52013-01-03 11:41:39 -02005912 free_percpu(shared_msrs);
Zhang Xiantao56c6d282007-11-18 20:43:21 +08005913}
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08005914
Joel Schopp5cb56052015-03-02 13:43:31 -06005915int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
Hollis Blanchard8776e512007-10-31 17:24:24 -05005916{
5917 ++vcpu->stat.halt_exits;
Paolo Bonzini35754c92015-07-29 12:05:37 +02005918 if (lapic_in_kernel(vcpu)) {
Avi Kivitya4535292008-04-13 17:54:35 +03005919 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
Hollis Blanchard8776e512007-10-31 17:24:24 -05005920 return 1;
5921 } else {
5922 vcpu->run->exit_reason = KVM_EXIT_HLT;
5923 return 0;
5924 }
5925}
Joel Schopp5cb56052015-03-02 13:43:31 -06005926EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
5927
5928int kvm_emulate_halt(struct kvm_vcpu *vcpu)
5929{
5930 kvm_x86_ops->skip_emulated_instruction(vcpu);
5931 return kvm_vcpu_halt(vcpu);
5932}
Hollis Blanchard8776e512007-10-31 17:24:24 -05005933EXPORT_SYMBOL_GPL(kvm_emulate_halt);
5934
Srivatsa Vaddagiri6aef2662013-08-26 14:18:34 +05305935/*
5936 * kvm_pv_kick_cpu_op: Kick a vcpu.
5937 *
5938 * @apicid - apicid of vcpu to be kicked.
5939 */
5940static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
5941{
Raghavendra K T24d21662013-08-26 14:18:35 +05305942 struct kvm_lapic_irq lapic_irq;
Srivatsa Vaddagiri6aef2662013-08-26 14:18:34 +05305943
Raghavendra K T24d21662013-08-26 14:18:35 +05305944 lapic_irq.shorthand = 0;
5945 lapic_irq.dest_mode = 0;
5946 lapic_irq.dest_id = apicid;
James Sullivan93bbf0b2015-03-18 19:26:03 -06005947 lapic_irq.msi_redir_hint = false;
Srivatsa Vaddagiri6aef2662013-08-26 14:18:34 +05305948
Raghavendra K T24d21662013-08-26 14:18:35 +05305949 lapic_irq.delivery_mode = APIC_DM_REMRD;
Xiubo Li795a1492015-03-13 17:39:44 +08005950 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
Srivatsa Vaddagiri6aef2662013-08-26 14:18:34 +05305951}
5952
Andrey Smetanind62caab2015-11-10 15:36:33 +03005953void kvm_vcpu_deactivate_apicv(struct kvm_vcpu *vcpu)
5954{
5955 vcpu->arch.apicv_active = false;
5956 kvm_x86_ops->refresh_apicv_exec_ctrl(vcpu);
5957}
5958
Hollis Blanchard8776e512007-10-31 17:24:24 -05005959int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
5960{
5961 unsigned long nr, a0, a1, a2, a3, ret;
Nadav Amita449c7a2014-06-18 17:19:24 +03005962 int op_64_bit, r = 1;
Hollis Blanchard8776e512007-10-31 17:24:24 -05005963
Joel Schopp5cb56052015-03-02 13:43:31 -06005964 kvm_x86_ops->skip_emulated_instruction(vcpu);
5965
Gleb Natapov55cd8e52010-01-17 15:51:22 +02005966 if (kvm_hv_hypercall_enabled(vcpu->kvm))
5967 return kvm_hv_hypercall(vcpu);
5968
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03005969 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
5970 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
5971 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
5972 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
5973 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
Hollis Blanchard8776e512007-10-31 17:24:24 -05005974
Marcelo Tosatti229456f2009-06-17 09:22:14 -03005975 trace_kvm_hypercall(nr, a0, a1, a2, a3);
Feng (Eric) Liu2714d1d2008-04-10 15:31:10 -04005976
Nadav Amita449c7a2014-06-18 17:19:24 +03005977 op_64_bit = is_64_bit_mode(vcpu);
5978 if (!op_64_bit) {
Hollis Blanchard8776e512007-10-31 17:24:24 -05005979 nr &= 0xFFFFFFFF;
5980 a0 &= 0xFFFFFFFF;
5981 a1 &= 0xFFFFFFFF;
5982 a2 &= 0xFFFFFFFF;
5983 a3 &= 0xFFFFFFFF;
5984 }
5985
Jan Kiszka07708c42009-08-03 18:43:28 +02005986 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
5987 ret = -KVM_EPERM;
5988 goto out;
5989 }
5990
Hollis Blanchard8776e512007-10-31 17:24:24 -05005991 switch (nr) {
Avi Kivityb93463a2007-10-25 16:52:32 +02005992 case KVM_HC_VAPIC_POLL_IRQ:
5993 ret = 0;
5994 break;
Srivatsa Vaddagiri6aef2662013-08-26 14:18:34 +05305995 case KVM_HC_KICK_CPU:
5996 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
5997 ret = 0;
5998 break;
Hollis Blanchard8776e512007-10-31 17:24:24 -05005999 default:
6000 ret = -KVM_ENOSYS;
6001 break;
6002 }
Jan Kiszka07708c42009-08-03 18:43:28 +02006003out:
Nadav Amita449c7a2014-06-18 17:19:24 +03006004 if (!op_64_bit)
6005 ret = (u32)ret;
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03006006 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
Amit Shahf11c3a82008-02-21 01:00:30 +05306007 ++vcpu->stat.hypercalls;
Marcelo Tosatti2f333bc2008-02-22 12:21:37 -05006008 return r;
Hollis Blanchard8776e512007-10-31 17:24:24 -05006009}
6010EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
6011
Jan Kiszkab6785de2012-09-20 07:43:17 +02006012static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
Hollis Blanchard8776e512007-10-31 17:24:24 -05006013{
Avi Kivityd6aa1002011-04-20 15:47:13 +03006014 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
Hollis Blanchard8776e512007-10-31 17:24:24 -05006015 char instruction[3];
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03006016 unsigned long rip = kvm_rip_read(vcpu);
Hollis Blanchard8776e512007-10-31 17:24:24 -05006017
Hollis Blanchard8776e512007-10-31 17:24:24 -05006018 kvm_x86_ops->patch_hypercall(vcpu, instruction);
Hollis Blanchard8776e512007-10-31 17:24:24 -05006019
Takuya Yoshikawa9d741912011-05-29 21:53:48 +09006020 return emulator_write_emulated(ctxt, rip, instruction, 3, NULL);
Hollis Blanchard8776e512007-10-31 17:24:24 -05006021}
6022
Avi Kivity851ba692009-08-24 11:10:17 +03006023static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006024{
Matt Gingell782d4222015-11-16 15:26:00 -08006025 return vcpu->run->request_interrupt_window &&
6026 likely(!pic_in_kernel(vcpu->kvm));
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006027}
6028
Avi Kivity851ba692009-08-24 11:10:17 +03006029static void post_kvm_run_save(struct kvm_vcpu *vcpu)
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006030{
Avi Kivity851ba692009-08-24 11:10:17 +03006031 struct kvm_run *kvm_run = vcpu->run;
6032
Jan Kiszka91586a32009-10-05 13:07:21 +02006033 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
Paolo Bonzinif0778252015-04-01 15:06:40 +02006034 kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
Avi Kivity2d3ad1f2008-02-24 11:20:43 +02006035 kvm_run->cr8 = kvm_get_cr8(vcpu);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006036 kvm_run->apic_base = kvm_get_apic_base(vcpu);
Matt Gingell127a4572015-11-17 17:32:05 +01006037 kvm_run->ready_for_interrupt_injection =
6038 pic_in_kernel(vcpu->kvm) ||
Matt Gingell782d4222015-11-16 15:26:00 -08006039 kvm_vcpu_ready_for_interrupt_injection(vcpu);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006040}
6041
Gleb Natapov95ba8273132009-04-21 17:45:08 +03006042static void update_cr8_intercept(struct kvm_vcpu *vcpu)
6043{
6044 int max_irr, tpr;
6045
6046 if (!kvm_x86_ops->update_cr8_intercept)
6047 return;
6048
Paolo Bonzinibce87cc2016-01-08 13:48:51 +01006049 if (!lapic_in_kernel(vcpu))
Avi Kivity88c808f2009-08-17 22:49:40 +03006050 return;
6051
Andrey Smetanind62caab2015-11-10 15:36:33 +03006052 if (vcpu->arch.apicv_active)
6053 return;
6054
Gleb Natapov8db3baa2009-05-11 13:35:54 +03006055 if (!vcpu->arch.apic->vapic_addr)
6056 max_irr = kvm_lapic_find_highest_irr(vcpu);
6057 else
6058 max_irr = -1;
Gleb Natapov95ba8273132009-04-21 17:45:08 +03006059
6060 if (max_irr != -1)
6061 max_irr >>= 4;
6062
6063 tpr = kvm_lapic_get_cr8(vcpu);
6064
6065 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
6066}
6067
Jan Kiszkab6b8a142014-03-07 20:03:12 +01006068static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
Gleb Natapov95ba8273132009-04-21 17:45:08 +03006069{
Jan Kiszkab6b8a142014-03-07 20:03:12 +01006070 int r;
6071
Gleb Natapov95ba8273132009-04-21 17:45:08 +03006072 /* try to reinject previous events if any */
Gleb Natapovb59bb7b2009-07-09 15:33:51 +03006073 if (vcpu->arch.exception.pending) {
Avi Kivity5c1c85d2010-03-11 13:01:59 +02006074 trace_kvm_inj_exception(vcpu->arch.exception.nr,
6075 vcpu->arch.exception.has_error_code,
6076 vcpu->arch.exception.error_code);
Nadav Amitd6e8c852014-07-24 14:51:24 +03006077
6078 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
6079 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
6080 X86_EFLAGS_RF);
6081
Nadav Amit6bdf0662014-09-30 20:49:14 +03006082 if (vcpu->arch.exception.nr == DB_VECTOR &&
6083 (vcpu->arch.dr7 & DR7_GD)) {
6084 vcpu->arch.dr7 &= ~DR7_GD;
6085 kvm_update_dr7(vcpu);
6086 }
6087
Gleb Natapovb59bb7b2009-07-09 15:33:51 +03006088 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
6089 vcpu->arch.exception.has_error_code,
Joerg Roedelce7ddec2010-04-22 12:33:13 +02006090 vcpu->arch.exception.error_code,
6091 vcpu->arch.exception.reinject);
Jan Kiszkab6b8a142014-03-07 20:03:12 +01006092 return 0;
Gleb Natapovb59bb7b2009-07-09 15:33:51 +03006093 }
6094
Gleb Natapov95ba8273132009-04-21 17:45:08 +03006095 if (vcpu->arch.nmi_injected) {
6096 kvm_x86_ops->set_nmi(vcpu);
Jan Kiszkab6b8a142014-03-07 20:03:12 +01006097 return 0;
Gleb Natapov95ba8273132009-04-21 17:45:08 +03006098 }
6099
6100 if (vcpu->arch.interrupt.pending) {
Gleb Natapov66fd3f72009-05-11 13:35:50 +03006101 kvm_x86_ops->set_irq(vcpu);
Jan Kiszkab6b8a142014-03-07 20:03:12 +01006102 return 0;
6103 }
6104
6105 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6106 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6107 if (r != 0)
6108 return r;
Gleb Natapov95ba8273132009-04-21 17:45:08 +03006109 }
6110
6111 /* try to inject new event if pending */
Paolo Bonzinic43203c2016-06-01 22:26:00 +02006112 if (vcpu->arch.smi_pending && !is_smm(vcpu)) {
6113 vcpu->arch.smi_pending = false;
Paolo Bonziniee2cd4b2016-06-01 22:26:01 +02006114 enter_smm(vcpu);
Paolo Bonzinic43203c2016-06-01 22:26:00 +02006115 } else if (vcpu->arch.nmi_pending && kvm_x86_ops->nmi_allowed(vcpu)) {
Yuki Shibuya321c5652016-03-24 05:17:03 +00006116 --vcpu->arch.nmi_pending;
6117 vcpu->arch.nmi_injected = true;
6118 kvm_x86_ops->set_nmi(vcpu);
Yang Zhangc7c9c562013-01-25 10:18:51 +08006119 } else if (kvm_cpu_has_injectable_intr(vcpu)) {
Bandan Das9242b5b2014-07-08 00:30:23 -04006120 /*
6121 * Because interrupts can be injected asynchronously, we are
6122 * calling check_nested_events again here to avoid a race condition.
6123 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
6124 * proposal and current concerns. Perhaps we should be setting
6125 * KVM_REQ_EVENT only on certain events and not unconditionally?
6126 */
6127 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6128 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6129 if (r != 0)
6130 return r;
6131 }
Gleb Natapov95ba8273132009-04-21 17:45:08 +03006132 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
Gleb Natapov66fd3f72009-05-11 13:35:50 +03006133 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
6134 false);
6135 kvm_x86_ops->set_irq(vcpu);
Gleb Natapov95ba8273132009-04-21 17:45:08 +03006136 }
6137 }
Paolo Bonziniee2cd4b2016-06-01 22:26:01 +02006138
Jan Kiszkab6b8a142014-03-07 20:03:12 +01006139 return 0;
Gleb Natapov95ba8273132009-04-21 17:45:08 +03006140}
6141
Avi Kivity7460fb4a2011-09-20 13:43:14 +03006142static void process_nmi(struct kvm_vcpu *vcpu)
6143{
6144 unsigned limit = 2;
6145
6146 /*
6147 * x86 is limited to one NMI running, and one NMI pending after it.
6148 * If an NMI is already in progress, limit further NMIs to just one.
6149 * Otherwise, allow two (and we'll inject the first one immediately).
6150 */
6151 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
6152 limit = 1;
6153
6154 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
6155 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
6156 kvm_make_request(KVM_REQ_EVENT, vcpu);
6157}
6158
Paolo Bonzini660a5d52015-05-05 11:50:23 +02006159#define put_smstate(type, buf, offset, val) \
6160 *(type *)((buf) + (offset) - 0x7e00) = val
6161
Paolo Bonziniee2cd4b2016-06-01 22:26:01 +02006162static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
Paolo Bonzini660a5d52015-05-05 11:50:23 +02006163{
6164 u32 flags = 0;
6165 flags |= seg->g << 23;
6166 flags |= seg->db << 22;
6167 flags |= seg->l << 21;
6168 flags |= seg->avl << 20;
6169 flags |= seg->present << 15;
6170 flags |= seg->dpl << 13;
6171 flags |= seg->s << 12;
6172 flags |= seg->type << 8;
6173 return flags;
6174}
6175
Paolo Bonziniee2cd4b2016-06-01 22:26:01 +02006176static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
Paolo Bonzini660a5d52015-05-05 11:50:23 +02006177{
6178 struct kvm_segment seg;
6179 int offset;
6180
6181 kvm_get_segment(vcpu, &seg, n);
6182 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
6183
6184 if (n < 3)
6185 offset = 0x7f84 + n * 12;
6186 else
6187 offset = 0x7f2c + (n - 3) * 12;
6188
6189 put_smstate(u32, buf, offset + 8, seg.base);
6190 put_smstate(u32, buf, offset + 4, seg.limit);
Paolo Bonziniee2cd4b2016-06-01 22:26:01 +02006191 put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
Paolo Bonzini660a5d52015-05-05 11:50:23 +02006192}
6193
Alexander Kuleshovefbb2882015-09-06 19:35:41 +06006194#ifdef CONFIG_X86_64
Paolo Bonziniee2cd4b2016-06-01 22:26:01 +02006195static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
Paolo Bonzini660a5d52015-05-05 11:50:23 +02006196{
6197 struct kvm_segment seg;
6198 int offset;
6199 u16 flags;
6200
6201 kvm_get_segment(vcpu, &seg, n);
6202 offset = 0x7e00 + n * 16;
6203
Paolo Bonziniee2cd4b2016-06-01 22:26:01 +02006204 flags = enter_smm_get_segment_flags(&seg) >> 8;
Paolo Bonzini660a5d52015-05-05 11:50:23 +02006205 put_smstate(u16, buf, offset, seg.selector);
6206 put_smstate(u16, buf, offset + 2, flags);
6207 put_smstate(u32, buf, offset + 4, seg.limit);
6208 put_smstate(u64, buf, offset + 8, seg.base);
6209}
Alexander Kuleshovefbb2882015-09-06 19:35:41 +06006210#endif
Paolo Bonzini660a5d52015-05-05 11:50:23 +02006211
Paolo Bonziniee2cd4b2016-06-01 22:26:01 +02006212static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
Paolo Bonzini660a5d52015-05-05 11:50:23 +02006213{
6214 struct desc_ptr dt;
6215 struct kvm_segment seg;
6216 unsigned long val;
6217 int i;
6218
6219 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
6220 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
6221 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
6222 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
6223
6224 for (i = 0; i < 8; i++)
6225 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
6226
6227 kvm_get_dr(vcpu, 6, &val);
6228 put_smstate(u32, buf, 0x7fcc, (u32)val);
6229 kvm_get_dr(vcpu, 7, &val);
6230 put_smstate(u32, buf, 0x7fc8, (u32)val);
6231
6232 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6233 put_smstate(u32, buf, 0x7fc4, seg.selector);
6234 put_smstate(u32, buf, 0x7f64, seg.base);
6235 put_smstate(u32, buf, 0x7f60, seg.limit);
Paolo Bonziniee2cd4b2016-06-01 22:26:01 +02006236 put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
Paolo Bonzini660a5d52015-05-05 11:50:23 +02006237
6238 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6239 put_smstate(u32, buf, 0x7fc0, seg.selector);
6240 put_smstate(u32, buf, 0x7f80, seg.base);
6241 put_smstate(u32, buf, 0x7f7c, seg.limit);
Paolo Bonziniee2cd4b2016-06-01 22:26:01 +02006242 put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
Paolo Bonzini660a5d52015-05-05 11:50:23 +02006243
6244 kvm_x86_ops->get_gdt(vcpu, &dt);
6245 put_smstate(u32, buf, 0x7f74, dt.address);
6246 put_smstate(u32, buf, 0x7f70, dt.size);
6247
6248 kvm_x86_ops->get_idt(vcpu, &dt);
6249 put_smstate(u32, buf, 0x7f58, dt.address);
6250 put_smstate(u32, buf, 0x7f54, dt.size);
6251
6252 for (i = 0; i < 6; i++)
Paolo Bonziniee2cd4b2016-06-01 22:26:01 +02006253 enter_smm_save_seg_32(vcpu, buf, i);
Paolo Bonzini660a5d52015-05-05 11:50:23 +02006254
6255 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
6256
6257 /* revision id */
6258 put_smstate(u32, buf, 0x7efc, 0x00020000);
6259 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
6260}
6261
Paolo Bonziniee2cd4b2016-06-01 22:26:01 +02006262static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
Paolo Bonzini660a5d52015-05-05 11:50:23 +02006263{
6264#ifdef CONFIG_X86_64
6265 struct desc_ptr dt;
6266 struct kvm_segment seg;
6267 unsigned long val;
6268 int i;
6269
6270 for (i = 0; i < 16; i++)
6271 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
6272
6273 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
6274 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
6275
6276 kvm_get_dr(vcpu, 6, &val);
6277 put_smstate(u64, buf, 0x7f68, val);
6278 kvm_get_dr(vcpu, 7, &val);
6279 put_smstate(u64, buf, 0x7f60, val);
6280
6281 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
6282 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
6283 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
6284
6285 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
6286
6287 /* revision id */
6288 put_smstate(u32, buf, 0x7efc, 0x00020064);
6289
6290 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
6291
6292 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6293 put_smstate(u16, buf, 0x7e90, seg.selector);
Paolo Bonziniee2cd4b2016-06-01 22:26:01 +02006294 put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
Paolo Bonzini660a5d52015-05-05 11:50:23 +02006295 put_smstate(u32, buf, 0x7e94, seg.limit);
6296 put_smstate(u64, buf, 0x7e98, seg.base);
6297
6298 kvm_x86_ops->get_idt(vcpu, &dt);
6299 put_smstate(u32, buf, 0x7e84, dt.size);
6300 put_smstate(u64, buf, 0x7e88, dt.address);
6301
6302 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6303 put_smstate(u16, buf, 0x7e70, seg.selector);
Paolo Bonziniee2cd4b2016-06-01 22:26:01 +02006304 put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
Paolo Bonzini660a5d52015-05-05 11:50:23 +02006305 put_smstate(u32, buf, 0x7e74, seg.limit);
6306 put_smstate(u64, buf, 0x7e78, seg.base);
6307
6308 kvm_x86_ops->get_gdt(vcpu, &dt);
6309 put_smstate(u32, buf, 0x7e64, dt.size);
6310 put_smstate(u64, buf, 0x7e68, dt.address);
6311
6312 for (i = 0; i < 6; i++)
Paolo Bonziniee2cd4b2016-06-01 22:26:01 +02006313 enter_smm_save_seg_64(vcpu, buf, i);
Paolo Bonzini660a5d52015-05-05 11:50:23 +02006314#else
6315 WARN_ON_ONCE(1);
6316#endif
6317}
6318
Paolo Bonziniee2cd4b2016-06-01 22:26:01 +02006319static void enter_smm(struct kvm_vcpu *vcpu)
Paolo Bonzini64d60672015-05-07 11:36:11 +02006320{
Paolo Bonzini660a5d52015-05-05 11:50:23 +02006321 struct kvm_segment cs, ds;
Paolo Bonzini18c36262015-08-07 12:27:54 +02006322 struct desc_ptr dt;
Paolo Bonzini660a5d52015-05-05 11:50:23 +02006323 char buf[512];
6324 u32 cr0;
6325
Paolo Bonzini660a5d52015-05-05 11:50:23 +02006326 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
6327 vcpu->arch.hflags |= HF_SMM_MASK;
6328 memset(buf, 0, 512);
6329 if (guest_cpuid_has_longmode(vcpu))
Paolo Bonziniee2cd4b2016-06-01 22:26:01 +02006330 enter_smm_save_state_64(vcpu, buf);
Paolo Bonzini660a5d52015-05-05 11:50:23 +02006331 else
Paolo Bonziniee2cd4b2016-06-01 22:26:01 +02006332 enter_smm_save_state_32(vcpu, buf);
Paolo Bonzini660a5d52015-05-05 11:50:23 +02006333
Paolo Bonzini54bf36a2015-04-08 15:39:23 +02006334 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
Paolo Bonzini660a5d52015-05-05 11:50:23 +02006335
6336 if (kvm_x86_ops->get_nmi_mask(vcpu))
6337 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
6338 else
6339 kvm_x86_ops->set_nmi_mask(vcpu, true);
6340
6341 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
6342 kvm_rip_write(vcpu, 0x8000);
6343
6344 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
6345 kvm_x86_ops->set_cr0(vcpu, cr0);
6346 vcpu->arch.cr0 = cr0;
6347
6348 kvm_x86_ops->set_cr4(vcpu, 0);
6349
Paolo Bonzini18c36262015-08-07 12:27:54 +02006350 /* Undocumented: IDT limit is set to zero on entry to SMM. */
6351 dt.address = dt.size = 0;
6352 kvm_x86_ops->set_idt(vcpu, &dt);
6353
Paolo Bonzini660a5d52015-05-05 11:50:23 +02006354 __kvm_set_dr(vcpu, 7, DR7_FIXED_1);
6355
6356 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
6357 cs.base = vcpu->arch.smbase;
6358
6359 ds.selector = 0;
6360 ds.base = 0;
6361
6362 cs.limit = ds.limit = 0xffffffff;
6363 cs.type = ds.type = 0x3;
6364 cs.dpl = ds.dpl = 0;
6365 cs.db = ds.db = 0;
6366 cs.s = ds.s = 1;
6367 cs.l = ds.l = 0;
6368 cs.g = ds.g = 1;
6369 cs.avl = ds.avl = 0;
6370 cs.present = ds.present = 1;
6371 cs.unusable = ds.unusable = 0;
6372 cs.padding = ds.padding = 0;
6373
6374 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
6375 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
6376 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
6377 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
6378 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
6379 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
6380
6381 if (guest_cpuid_has_longmode(vcpu))
6382 kvm_x86_ops->set_efer(vcpu, 0);
6383
6384 kvm_update_cpuid(vcpu);
6385 kvm_mmu_reset_context(vcpu);
Paolo Bonzini64d60672015-05-07 11:36:11 +02006386}
6387
Paolo Bonziniee2cd4b2016-06-01 22:26:01 +02006388static void process_smi(struct kvm_vcpu *vcpu)
Paolo Bonzinic43203c2016-06-01 22:26:00 +02006389{
6390 vcpu->arch.smi_pending = true;
6391 kvm_make_request(KVM_REQ_EVENT, vcpu);
6392}
6393
Paolo Bonzini2860c4b2016-01-07 15:05:10 +01006394void kvm_make_scan_ioapic_request(struct kvm *kvm)
6395{
6396 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
6397}
6398
Yang Zhang3d81bc72013-04-11 19:25:13 +08006399static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
Yang Zhangc7c9c562013-01-25 10:18:51 +08006400{
Andrey Smetanin5c9194122015-11-10 15:36:34 +03006401 u64 eoi_exit_bitmap[4];
6402
Yang Zhang3d81bc72013-04-11 19:25:13 +08006403 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
6404 return;
Yang Zhangc7c9c562013-01-25 10:18:51 +08006405
Andrey Smetanin63086302015-11-10 15:36:32 +03006406 bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
Yang Zhangc7c9c562013-01-25 10:18:51 +08006407
Steve Rutherfordb053b2a2015-07-29 23:32:35 -07006408 if (irqchip_split(vcpu->kvm))
Andrey Smetanin63086302015-11-10 15:36:32 +03006409 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
Radim Krčmářdb2bdcb2015-10-08 20:23:34 +02006410 else {
Andrey Smetanind62caab2015-11-10 15:36:33 +03006411 if (vcpu->arch.apicv_active)
6412 kvm_x86_ops->sync_pir_to_irr(vcpu);
Andrey Smetanin63086302015-11-10 15:36:32 +03006413 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
Radim Krčmářdb2bdcb2015-10-08 20:23:34 +02006414 }
Andrey Smetanin5c9194122015-11-10 15:36:34 +03006415 bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
6416 vcpu_to_synic(vcpu)->vec_bitmap, 256);
6417 kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
Yang Zhangc7c9c562013-01-25 10:18:51 +08006418}
6419
Radim Krčmářa70656b2014-09-18 12:38:36 -04006420static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu)
6421{
6422 ++vcpu->stat.tlb_flush;
6423 kvm_x86_ops->tlb_flush(vcpu);
6424}
6425
Tang Chen4256f432014-09-24 15:57:54 +08006426void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
6427{
Tang Chenc24ae0d2014-09-24 15:57:58 +08006428 struct page *page = NULL;
6429
Paolo Bonzini35754c92015-07-29 12:05:37 +02006430 if (!lapic_in_kernel(vcpu))
Paolo Bonzinif439ed22014-10-02 13:53:24 +02006431 return;
6432
Tang Chen4256f432014-09-24 15:57:54 +08006433 if (!kvm_x86_ops->set_apic_access_page_addr)
6434 return;
6435
Tang Chenc24ae0d2014-09-24 15:57:58 +08006436 page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
Andrea Arcangelie8fd5e92015-05-08 14:32:56 +02006437 if (is_error_page(page))
6438 return;
Tang Chenc24ae0d2014-09-24 15:57:58 +08006439 kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
6440
6441 /*
6442 * Do not pin apic access page in memory, the MMU notifier
6443 * will call us again if it is migrated or swapped out.
6444 */
6445 put_page(page);
Tang Chen4256f432014-09-24 15:57:54 +08006446}
6447EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
6448
Tang Chenfe715572014-09-24 15:57:57 +08006449void kvm_arch_mmu_notifier_invalidate_page(struct kvm *kvm,
6450 unsigned long address)
6451{
Tang Chenc24ae0d2014-09-24 15:57:58 +08006452 /*
6453 * The physical address of apic access page is stored in the VMCS.
6454 * Update it when it becomes invalid.
6455 */
6456 if (address == gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT))
6457 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
Tang Chenfe715572014-09-24 15:57:57 +08006458}
6459
Takuya Yoshikawa9357d932013-12-13 15:08:38 +09006460/*
Paolo Bonzini362c6982015-02-06 12:48:04 +01006461 * Returns 1 to let vcpu_run() continue the guest execution loop without
Takuya Yoshikawa9357d932013-12-13 15:08:38 +09006462 * exiting to the userspace. Otherwise, the value will be returned to the
6463 * userspace.
6464 */
Avi Kivity851ba692009-08-24 11:10:17 +03006465static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006466{
6467 int r;
Matt Gingell62a193e2015-11-16 15:26:07 -08006468 bool req_int_win =
6469 dm_request_for_irq_injection(vcpu) &&
6470 kvm_cpu_accept_dm_intr(vcpu);
6471
Jan Kiszka730dca42013-04-28 10:50:52 +02006472 bool req_immediate_exit = false;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006473
Avi Kivity3e007502010-06-23 14:26:18 +03006474 if (vcpu->requests) {
Avi Kivitya8eeb042010-05-10 12:34:53 +03006475 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
Marcelo Tosatti2e53d632008-02-20 14:47:24 -05006476 kvm_mmu_unload(vcpu);
Avi Kivitya8eeb042010-05-10 12:34:53 +03006477 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
Marcelo Tosatti2f599712008-05-27 12:10:20 -03006478 __kvm_migrate_timers(vcpu);
Marcelo Tosattid8281992012-11-27 23:29:01 -02006479 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
6480 kvm_gen_update_masterclock(vcpu->kvm);
Marcelo Tosatti0061d532013-05-09 20:21:41 -03006481 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
6482 kvm_gen_kvmclock_update(vcpu);
Zachary Amsden34c238a2010-09-18 14:38:14 -10006483 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
6484 r = kvm_guest_time_update(vcpu);
Zachary Amsden8cfdc002010-08-19 22:07:21 -10006485 if (unlikely(r))
6486 goto out;
6487 }
Avi Kivitya8eeb042010-05-10 12:34:53 +03006488 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
Marcelo Tosatti4731d4c2008-09-23 13:18:39 -03006489 kvm_mmu_sync_roots(vcpu);
Avi Kivitya8eeb042010-05-10 12:34:53 +03006490 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
Radim Krčmářa70656b2014-09-18 12:38:36 -04006491 kvm_vcpu_flush_tlb(vcpu);
Avi Kivitya8eeb042010-05-10 12:34:53 +03006492 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
Avi Kivity851ba692009-08-24 11:10:17 +03006493 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
Avi Kivityb93463a2007-10-25 16:52:32 +02006494 r = 0;
6495 goto out;
6496 }
Avi Kivitya8eeb042010-05-10 12:34:53 +03006497 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
Avi Kivity851ba692009-08-24 11:10:17 +03006498 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
Joerg Roedel71c4dfa2008-02-26 16:49:16 +01006499 r = 0;
6500 goto out;
6501 }
Avi Kivitya8eeb042010-05-10 12:34:53 +03006502 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
Avi Kivity02daab22009-12-30 12:40:26 +02006503 vcpu->fpu_active = 0;
6504 kvm_x86_ops->fpu_deactivate(vcpu);
6505 }
Gleb Natapovaf585b92010-10-14 11:22:46 +02006506 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
6507 /* Page is swapped out. Do synthetic halt */
6508 vcpu->arch.apf.halted = true;
6509 r = 1;
6510 goto out;
6511 }
Glauber Costac9aaa892011-07-11 15:28:14 -04006512 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
6513 record_steal_time(vcpu);
Paolo Bonzini64d60672015-05-07 11:36:11 +02006514 if (kvm_check_request(KVM_REQ_SMI, vcpu))
6515 process_smi(vcpu);
Avi Kivity7460fb4a2011-09-20 13:43:14 +03006516 if (kvm_check_request(KVM_REQ_NMI, vcpu))
6517 process_nmi(vcpu);
Gleb Natapovf5132b02011-11-10 14:57:22 +02006518 if (kvm_check_request(KVM_REQ_PMU, vcpu))
Wei Huangc6702c92015-06-19 13:44:45 +02006519 kvm_pmu_handle_event(vcpu);
Gleb Natapovf5132b02011-11-10 14:57:22 +02006520 if (kvm_check_request(KVM_REQ_PMI, vcpu))
Wei Huangc6702c92015-06-19 13:44:45 +02006521 kvm_pmu_deliver_pmi(vcpu);
Steve Rutherford7543a632015-07-29 23:21:41 -07006522 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
6523 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
6524 if (test_bit(vcpu->arch.pending_ioapic_eoi,
Andrey Smetanin63086302015-11-10 15:36:32 +03006525 vcpu->arch.ioapic_handled_vectors)) {
Steve Rutherford7543a632015-07-29 23:21:41 -07006526 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
6527 vcpu->run->eoi.vector =
6528 vcpu->arch.pending_ioapic_eoi;
6529 r = 0;
6530 goto out;
6531 }
6532 }
Yang Zhang3d81bc72013-04-11 19:25:13 +08006533 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
6534 vcpu_scan_ioapic(vcpu);
Tang Chen4256f432014-09-24 15:57:54 +08006535 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
6536 kvm_vcpu_reload_apic_access_page(vcpu);
Andrey Smetanin2ce79182015-07-03 15:01:41 +03006537 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
6538 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
6539 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
6540 r = 0;
6541 goto out;
6542 }
Andrey Smetanine516ceb2015-09-16 12:29:48 +03006543 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
6544 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
6545 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
6546 r = 0;
6547 goto out;
6548 }
Andrey Smetanindb3975712015-11-10 15:36:35 +03006549 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
6550 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
6551 vcpu->run->hyperv = vcpu->arch.hyperv.exit;
6552 r = 0;
6553 goto out;
6554 }
Andrey Smetaninf3b138c2015-12-28 18:27:24 +03006555
6556 /*
6557 * KVM_REQ_HV_STIMER has to be processed after
6558 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
6559 * depend on the guest clock being up-to-date
6560 */
Andrey Smetanin1f4b34f2015-11-30 19:22:21 +03006561 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
6562 kvm_hv_process_stimers(vcpu);
Avi Kivity2f52d582008-01-16 12:49:30 +02006563 }
Avi Kivityb93463a2007-10-25 16:52:32 +02006564
Feng Wubf9f6ac2015-09-18 22:29:55 +08006565 /*
6566 * KVM_REQ_EVENT is not set when posted interrupts are set by
6567 * VT-d hardware, so we have to update RVI unconditionally.
6568 */
6569 if (kvm_lapic_enabled(vcpu)) {
6570 /*
6571 * Update architecture specific hints for APIC
6572 * virtual interrupt delivery.
6573 */
Andrey Smetanind62caab2015-11-10 15:36:33 +03006574 if (vcpu->arch.apicv_active)
Feng Wubf9f6ac2015-09-18 22:29:55 +08006575 kvm_x86_ops->hwapic_irr_update(vcpu,
6576 kvm_lapic_find_highest_irr(vcpu));
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006577 }
6578
Avi Kivityb463a6f2010-07-20 15:06:17 +03006579 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
Jan Kiszka66450a22013-03-13 12:42:34 +01006580 kvm_apic_accept_events(vcpu);
6581 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
6582 r = 1;
6583 goto out;
6584 }
6585
Jan Kiszkab6b8a142014-03-07 20:03:12 +01006586 if (inject_pending_event(vcpu, req_int_win) != 0)
6587 req_immediate_exit = true;
Yuki Shibuya321c5652016-03-24 05:17:03 +00006588 else {
Paolo Bonzinic43203c2016-06-01 22:26:00 +02006589 /* Enable NMI/IRQ window open exits if needed.
6590 *
6591 * SMIs have two cases: 1) they can be nested, and
6592 * then there is nothing to do here because RSM will
6593 * cause a vmexit anyway; 2) or the SMI can be pending
6594 * because inject_pending_event has completed the
6595 * injection of an IRQ or NMI from the previous vmexit,
6596 * and then we request an immediate exit to inject the SMI.
6597 */
6598 if (vcpu->arch.smi_pending && !is_smm(vcpu))
6599 req_immediate_exit = true;
Yuki Shibuya321c5652016-03-24 05:17:03 +00006600 if (vcpu->arch.nmi_pending)
6601 kvm_x86_ops->enable_nmi_window(vcpu);
6602 if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
6603 kvm_x86_ops->enable_irq_window(vcpu);
6604 }
Gleb Natapov6a8b1d12009-05-11 13:35:51 +03006605
Avi Kivity3842d132010-07-27 12:30:24 +03006606 if (kvm_lapic_enabled(vcpu)) {
6607 update_cr8_intercept(vcpu);
6608 kvm_lapic_sync_to_vapic(vcpu);
6609 }
Gleb Natapov95ba8273132009-04-21 17:45:08 +03006610 }
Avi Kivityb93463a2007-10-25 16:52:32 +02006611
Avi Kivityd8368af2012-05-14 18:07:56 +03006612 r = kvm_mmu_reload(vcpu);
6613 if (unlikely(r)) {
Michael S. Tsirkind905c062012-06-24 19:25:00 +03006614 goto cancel_injection;
Avi Kivityd8368af2012-05-14 18:07:56 +03006615 }
6616
Jan Kiszka42dbaa52008-12-15 13:52:10 +01006617 preempt_disable();
6618
6619 kvm_x86_ops->prepare_guest_switch(vcpu);
6620 if (vcpu->fpu_active)
6621 kvm_load_guest_fpu(vcpu);
Xiao Guangrong6b7e2d02011-01-12 15:40:31 +08006622 vcpu->mode = IN_GUEST_MODE;
6623
Michael S. Tsirkin01b71912013-11-04 22:36:25 +02006624 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6625
Lan Tianyu0f127d12016-03-13 11:10:29 +08006626 /*
6627 * We should set ->mode before check ->requests,
6628 * Please see the comment in kvm_make_all_cpus_request.
6629 * This also orders the write to mode from any reads
6630 * to the page tables done while the VCPU is running.
6631 * Please see the comment in kvm_flush_remote_tlbs.
Xiao Guangrong6b7e2d02011-01-12 15:40:31 +08006632 */
Michael S. Tsirkin01b71912013-11-04 22:36:25 +02006633 smp_mb__after_srcu_read_unlock();
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006634
6635 local_irq_disable();
6636
Xiao Guangrong6b7e2d02011-01-12 15:40:31 +08006637 if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006638 || need_resched() || signal_pending(current)) {
Xiao Guangrong6b7e2d02011-01-12 15:40:31 +08006639 vcpu->mode = OUTSIDE_GUEST_MODE;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006640 smp_wmb();
6641 local_irq_enable();
6642 preempt_enable();
Michael S. Tsirkin01b71912013-11-04 22:36:25 +02006643 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006644 r = 1;
Michael S. Tsirkind905c062012-06-24 19:25:00 +03006645 goto cancel_injection;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006646 }
6647
David Matlackfc5b7f32016-03-30 12:24:47 -07006648 kvm_load_guest_xcr0(vcpu);
6649
Paolo Bonzinic43203c2016-06-01 22:26:00 +02006650 if (req_immediate_exit) {
6651 kvm_make_request(KVM_REQ_EVENT, vcpu);
Nadav Har'Eld6185f22011-09-22 13:52:56 +03006652 smp_send_reschedule(vcpu->cpu);
Paolo Bonzinic43203c2016-06-01 22:26:00 +02006653 }
Nadav Har'Eld6185f22011-09-22 13:52:56 +03006654
Paolo Bonzini8b89fe12015-12-10 18:37:32 +01006655 trace_kvm_entry(vcpu->vcpu_id);
6656 wait_lapic_expire(vcpu);
Paolo Bonzini6edaa532016-06-15 15:18:26 +02006657 guest_enter_irqoff();
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006658
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006659 if (unlikely(vcpu->arch.switch_db_regs)) {
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006660 set_debugreg(0, 7);
6661 set_debugreg(vcpu->arch.eff_db[0], 0);
6662 set_debugreg(vcpu->arch.eff_db[1], 1);
6663 set_debugreg(vcpu->arch.eff_db[2], 2);
6664 set_debugreg(vcpu->arch.eff_db[3], 3);
Paolo Bonzinic77fb5f2014-02-21 10:17:24 +01006665 set_debugreg(vcpu->arch.dr6, 6);
Nadav Amitae561ed2015-04-02 03:10:37 +03006666 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006667 }
6668
Avi Kivity851ba692009-08-24 11:10:17 +03006669 kvm_x86_ops->run(vcpu);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006670
Frederic Weisbecker24f1e32c2009-09-09 19:22:48 +02006671 /*
Paolo Bonzinic77fb5f2014-02-21 10:17:24 +01006672 * Do this here before restoring debug registers on the host. And
6673 * since we do this before handling the vmexit, a DR access vmexit
6674 * can (a) read the correct value of the debug registers, (b) set
6675 * KVM_DEBUGREG_WONT_EXIT again.
6676 */
6677 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
Paolo Bonzinic77fb5f2014-02-21 10:17:24 +01006678 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
6679 kvm_x86_ops->sync_dirty_debug_regs(vcpu);
Paolo Bonzini70e4da72016-02-26 12:28:40 +01006680 kvm_update_dr0123(vcpu);
6681 kvm_update_dr6(vcpu);
6682 kvm_update_dr7(vcpu);
6683 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
Paolo Bonzinic77fb5f2014-02-21 10:17:24 +01006684 }
6685
6686 /*
Frederic Weisbecker24f1e32c2009-09-09 19:22:48 +02006687 * If the guest has used debug registers, at least dr7
6688 * will be disabled while returning to the host.
6689 * If we don't have active breakpoints in the host, we don't
6690 * care about the messed up debug address registers. But if
6691 * we have some of them active, restore the old state.
6692 */
Frederic Weisbecker59d8eb52009-11-10 11:03:12 +01006693 if (hw_breakpoint_active())
Frederic Weisbecker24f1e32c2009-09-09 19:22:48 +02006694 hw_breakpoint_restore();
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006695
Haozhong Zhang4ba76532015-10-20 15:39:07 +08006696 vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
Zachary Amsden1d5f0662010-08-19 22:07:30 -10006697
Xiao Guangrong6b7e2d02011-01-12 15:40:31 +08006698 vcpu->mode = OUTSIDE_GUEST_MODE;
Avi Kivityd94e1dc2010-05-03 16:54:48 +03006699 smp_wmb();
Yang Zhanga547c6d2013-04-11 19:25:10 +08006700
David Matlackfc5b7f32016-03-30 12:24:47 -07006701 kvm_put_guest_xcr0(vcpu);
6702
Yang Zhanga547c6d2013-04-11 19:25:10 +08006703 /* Interrupt is enabled by handle_external_intr() */
6704 kvm_x86_ops->handle_external_intr(vcpu);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006705
6706 ++vcpu->stat.exits;
6707
Paolo Bonzinif2485b32016-06-15 15:23:11 +02006708 guest_exit_irqoff();
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006709
Paolo Bonzinif2485b32016-06-15 15:23:11 +02006710 local_irq_enable();
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006711 preempt_enable();
6712
Marcelo Tosattif656ce02009-12-23 14:35:25 -02006713 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
Marcelo Tosatti3200f402008-03-29 20:17:59 -03006714
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006715 /*
6716 * Profile KVM exit RIPs:
6717 */
6718 if (unlikely(prof_on == KVM_PROFILING)) {
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03006719 unsigned long rip = kvm_rip_read(vcpu);
6720 profile_hit(KVM_PROFILING, (void *)rip);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006721 }
6722
Zachary Amsdencc578282012-02-03 15:43:50 -02006723 if (unlikely(vcpu->arch.tsc_always_catchup))
6724 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
Avi Kivity298101d2007-11-25 13:41:11 +02006725
Michael S. Tsirkin5cfb1d52012-06-24 19:24:54 +03006726 if (vcpu->arch.apic_attention)
6727 kvm_lapic_sync_from_vapic(vcpu);
Avi Kivityb93463a2007-10-25 16:52:32 +02006728
Avi Kivity851ba692009-08-24 11:10:17 +03006729 r = kvm_x86_ops->handle_exit(vcpu);
Michael S. Tsirkind905c062012-06-24 19:25:00 +03006730 return r;
6731
6732cancel_injection:
6733 kvm_x86_ops->cancel_injection(vcpu);
Michael S. Tsirkinae7a2a32012-06-24 19:25:07 +03006734 if (unlikely(vcpu->arch.apic_attention))
6735 kvm_lapic_sync_from_vapic(vcpu);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006736out:
Marcelo Tosattid7690172008-09-08 15:23:48 -03006737 return r;
6738}
6739
Paolo Bonzini362c6982015-02-06 12:48:04 +01006740static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
6741{
Feng Wubf9f6ac2015-09-18 22:29:55 +08006742 if (!kvm_arch_vcpu_runnable(vcpu) &&
6743 (!kvm_x86_ops->pre_block || kvm_x86_ops->pre_block(vcpu) == 0)) {
Paolo Bonzini9c8fd1b2015-02-06 12:58:42 +01006744 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6745 kvm_vcpu_block(vcpu);
6746 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
Feng Wubf9f6ac2015-09-18 22:29:55 +08006747
6748 if (kvm_x86_ops->post_block)
6749 kvm_x86_ops->post_block(vcpu);
6750
Paolo Bonzini9c8fd1b2015-02-06 12:58:42 +01006751 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
6752 return 1;
6753 }
Gleb Natapov09cec752009-03-23 15:11:44 +02006754
Paolo Bonzini362c6982015-02-06 12:48:04 +01006755 kvm_apic_accept_events(vcpu);
6756 switch(vcpu->arch.mp_state) {
6757 case KVM_MP_STATE_HALTED:
6758 vcpu->arch.pv.pv_unhalted = false;
6759 vcpu->arch.mp_state =
6760 KVM_MP_STATE_RUNNABLE;
6761 case KVM_MP_STATE_RUNNABLE:
6762 vcpu->arch.apf.halted = false;
6763 break;
6764 case KVM_MP_STATE_INIT_RECEIVED:
6765 break;
6766 default:
6767 return -EINTR;
6768 break;
6769 }
6770 return 1;
6771}
6772
Paolo Bonzini5d9bc642015-10-13 10:18:53 +02006773static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
6774{
6775 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6776 !vcpu->arch.apf.halted);
6777}
6778
Paolo Bonzini362c6982015-02-06 12:48:04 +01006779static int vcpu_run(struct kvm_vcpu *vcpu)
Marcelo Tosattid7690172008-09-08 15:23:48 -03006780{
6781 int r;
Marcelo Tosattif656ce02009-12-23 14:35:25 -02006782 struct kvm *kvm = vcpu->kvm;
Marcelo Tosattid7690172008-09-08 15:23:48 -03006783
Marcelo Tosattif656ce02009-12-23 14:35:25 -02006784 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
Marcelo Tosattid7690172008-09-08 15:23:48 -03006785
Paolo Bonzini362c6982015-02-06 12:48:04 +01006786 for (;;) {
Paolo Bonzini58f800d2015-10-13 21:32:50 +02006787 if (kvm_vcpu_running(vcpu)) {
Avi Kivity851ba692009-08-24 11:10:17 +03006788 r = vcpu_enter_guest(vcpu);
Feng Wubf9f6ac2015-09-18 22:29:55 +08006789 } else {
Paolo Bonzini362c6982015-02-06 12:48:04 +01006790 r = vcpu_block(kvm, vcpu);
Feng Wubf9f6ac2015-09-18 22:29:55 +08006791 }
6792
Gleb Natapov09cec752009-03-23 15:11:44 +02006793 if (r <= 0)
6794 break;
6795
6796 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
6797 if (kvm_cpu_has_pending_timer(vcpu))
6798 kvm_inject_pending_timer_irqs(vcpu);
6799
Matt Gingell782d4222015-11-16 15:26:00 -08006800 if (dm_request_for_irq_injection(vcpu) &&
6801 kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
Paolo Bonzini4ca7dd82015-07-30 10:32:16 +02006802 r = 0;
6803 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
Gleb Natapov09cec752009-03-23 15:11:44 +02006804 ++vcpu->stat.request_irq_exits;
Paolo Bonzini362c6982015-02-06 12:48:04 +01006805 break;
Gleb Natapov09cec752009-03-23 15:11:44 +02006806 }
Gleb Natapovaf585b92010-10-14 11:22:46 +02006807
6808 kvm_check_async_pf_completion(vcpu);
6809
Gleb Natapov09cec752009-03-23 15:11:44 +02006810 if (signal_pending(current)) {
6811 r = -EINTR;
Avi Kivity851ba692009-08-24 11:10:17 +03006812 vcpu->run->exit_reason = KVM_EXIT_INTR;
Gleb Natapov09cec752009-03-23 15:11:44 +02006813 ++vcpu->stat.signal_exits;
Paolo Bonzini362c6982015-02-06 12:48:04 +01006814 break;
Gleb Natapov09cec752009-03-23 15:11:44 +02006815 }
6816 if (need_resched()) {
Marcelo Tosattif656ce02009-12-23 14:35:25 -02006817 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
Takuya Yoshikawac08ac062013-12-13 15:07:21 +09006818 cond_resched();
Marcelo Tosattif656ce02009-12-23 14:35:25 -02006819 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
Marcelo Tosattid7690172008-09-08 15:23:48 -03006820 }
6821 }
6822
Marcelo Tosattif656ce02009-12-23 14:35:25 -02006823 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006824
6825 return r;
6826}
6827
Gleb Natapov716d51a2012-09-03 15:24:26 +03006828static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
6829{
6830 int r;
6831 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6832 r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
6833 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6834 if (r != EMULATE_DONE)
6835 return 0;
6836 return 1;
6837}
6838
6839static int complete_emulated_pio(struct kvm_vcpu *vcpu)
6840{
6841 BUG_ON(!vcpu->arch.pio.count);
6842
6843 return complete_emulated_io(vcpu);
6844}
6845
Avi Kivityf78146b2012-04-18 19:22:47 +03006846/*
6847 * Implements the following, as a state machine:
6848 *
6849 * read:
6850 * for each fragment
Xiao Guangrong87da7e62012-10-24 14:07:59 +08006851 * for each mmio piece in the fragment
6852 * write gpa, len
6853 * exit
6854 * copy data
Avi Kivityf78146b2012-04-18 19:22:47 +03006855 * execute insn
6856 *
6857 * write:
6858 * for each fragment
Xiao Guangrong87da7e62012-10-24 14:07:59 +08006859 * for each mmio piece in the fragment
6860 * write gpa, len
6861 * copy data
6862 * exit
Avi Kivityf78146b2012-04-18 19:22:47 +03006863 */
Gleb Natapov716d51a2012-09-03 15:24:26 +03006864static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
Avi Kivity5287f192010-01-19 14:20:10 +02006865{
6866 struct kvm_run *run = vcpu->run;
Avi Kivityf78146b2012-04-18 19:22:47 +03006867 struct kvm_mmio_fragment *frag;
Xiao Guangrong87da7e62012-10-24 14:07:59 +08006868 unsigned len;
Avi Kivity5287f192010-01-19 14:20:10 +02006869
Gleb Natapov716d51a2012-09-03 15:24:26 +03006870 BUG_ON(!vcpu->mmio_needed);
Avi Kivity5287f192010-01-19 14:20:10 +02006871
Gleb Natapov716d51a2012-09-03 15:24:26 +03006872 /* Complete previous fragment */
Xiao Guangrong87da7e62012-10-24 14:07:59 +08006873 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
6874 len = min(8u, frag->len);
Gleb Natapov716d51a2012-09-03 15:24:26 +03006875 if (!vcpu->mmio_is_write)
Xiao Guangrong87da7e62012-10-24 14:07:59 +08006876 memcpy(frag->data, run->mmio.data, len);
6877
6878 if (frag->len <= 8) {
6879 /* Switch to the next fragment. */
6880 frag++;
6881 vcpu->mmio_cur_fragment++;
6882 } else {
6883 /* Go forward to the next mmio piece. */
6884 frag->data += len;
6885 frag->gpa += len;
6886 frag->len -= len;
6887 }
6888
Andrew Honiga08d3b32014-02-27 19:35:14 +01006889 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
Gleb Natapov716d51a2012-09-03 15:24:26 +03006890 vcpu->mmio_needed = 0;
Paolo Bonzini0912c972013-08-27 15:41:43 +02006891
6892 /* FIXME: return into emulator if single-stepping. */
Avi Kivitycef4dea2010-01-20 12:01:20 +02006893 if (vcpu->mmio_is_write)
Gleb Natapov716d51a2012-09-03 15:24:26 +03006894 return 1;
6895 vcpu->mmio_read_completed = 1;
6896 return complete_emulated_io(vcpu);
Avi Kivity5287f192010-01-19 14:20:10 +02006897 }
Xiao Guangrong87da7e62012-10-24 14:07:59 +08006898
Gleb Natapov716d51a2012-09-03 15:24:26 +03006899 run->exit_reason = KVM_EXIT_MMIO;
6900 run->mmio.phys_addr = frag->gpa;
6901 if (vcpu->mmio_is_write)
Xiao Guangrong87da7e62012-10-24 14:07:59 +08006902 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
6903 run->mmio.len = min(8u, frag->len);
Gleb Natapov716d51a2012-09-03 15:24:26 +03006904 run->mmio.is_write = vcpu->mmio_is_write;
6905 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6906 return 0;
Avi Kivity5287f192010-01-19 14:20:10 +02006907}
6908
Gleb Natapov716d51a2012-09-03 15:24:26 +03006909
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006910int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
6911{
Ingo Molnarc5bedc62015-04-23 12:49:20 +02006912 struct fpu *fpu = &current->thread.fpu;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006913 int r;
6914 sigset_t sigsaved;
6915
Ingo Molnarc4d72e22015-04-27 07:18:17 +02006916 fpu__activate_curr(fpu);
Avi Kivitye5c30142011-01-11 12:15:54 +02006917
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006918 if (vcpu->sigset_active)
6919 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
6920
Avi Kivityac9f6dc2008-07-06 15:48:31 +03006921 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
6922 kvm_vcpu_block(vcpu);
Jan Kiszka66450a22013-03-13 12:42:34 +01006923 kvm_apic_accept_events(vcpu);
Marcelo Tosattid7690172008-09-08 15:23:48 -03006924 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
Avi Kivityac9f6dc2008-07-06 15:48:31 +03006925 r = -EAGAIN;
6926 goto out;
6927 }
6928
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006929 /* re-sync apic's tpr */
Paolo Bonzini35754c92015-07-29 12:05:37 +02006930 if (!lapic_in_kernel(vcpu)) {
Andre Przywaraeea1cff2010-12-21 11:12:00 +01006931 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
6932 r = -EINVAL;
6933 goto out;
6934 }
6935 }
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006936
Gleb Natapov716d51a2012-09-03 15:24:26 +03006937 if (unlikely(vcpu->arch.complete_userspace_io)) {
6938 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
6939 vcpu->arch.complete_userspace_io = NULL;
6940 r = cui(vcpu);
6941 if (r <= 0)
6942 goto out;
6943 } else
6944 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
Avi Kivity5287f192010-01-19 14:20:10 +02006945
Paolo Bonzini362c6982015-02-06 12:48:04 +01006946 r = vcpu_run(vcpu);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006947
6948out:
Marcelo Tosattif1d86e42010-05-03 23:04:27 -03006949 post_kvm_run_save(vcpu);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006950 if (vcpu->sigset_active)
6951 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
6952
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006953 return r;
6954}
6955
6956int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6957{
Gleb Natapov7ae441e2011-03-31 12:06:41 +02006958 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
6959 /*
6960 * We are here if userspace calls get_regs() in the middle of
6961 * instruction emulation. Registers state needs to be copied
Guo Chao4a969982012-06-28 15:17:27 +08006962 * back from emulation context to vcpu. Userspace shouldn't do
Gleb Natapov7ae441e2011-03-31 12:06:41 +02006963 * that usually, but some bad designed PV devices (vmware
6964 * backdoor interface) need this to work
6965 */
Avi Kivitydd856ef2012-08-27 23:46:17 +03006966 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
Gleb Natapov7ae441e2011-03-31 12:06:41 +02006967 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6968 }
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03006969 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
6970 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
6971 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
6972 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
6973 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
6974 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
6975 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
6976 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006977#ifdef CONFIG_X86_64
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03006978 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
6979 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
6980 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
6981 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
6982 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
6983 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
6984 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
6985 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006986#endif
6987
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03006988 regs->rip = kvm_rip_read(vcpu);
Jan Kiszka91586a32009-10-05 13:07:21 +02006989 regs->rflags = kvm_get_rflags(vcpu);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006990
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006991 return 0;
6992}
6993
6994int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6995{
Gleb Natapov7ae441e2011-03-31 12:06:41 +02006996 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
6997 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6998
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03006999 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
7000 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
7001 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
7002 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
7003 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
7004 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
7005 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
7006 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007007#ifdef CONFIG_X86_64
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03007008 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
7009 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
7010 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
7011 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
7012 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
7013 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
7014 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
7015 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007016#endif
7017
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03007018 kvm_rip_write(vcpu, regs->rip);
Jan Kiszka91586a32009-10-05 13:07:21 +02007019 kvm_set_rflags(vcpu, regs->rflags);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007020
Jan Kiszkab4f14ab2008-04-30 17:59:04 +02007021 vcpu->arch.exception.pending = false;
7022
Avi Kivity3842d132010-07-27 12:30:24 +03007023 kvm_make_request(KVM_REQ_EVENT, vcpu);
7024
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007025 return 0;
7026}
7027
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007028void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
7029{
7030 struct kvm_segment cs;
7031
Guillaume Thouvenin3e6e0aa2008-05-27 10:18:46 +02007032 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007033 *db = cs.db;
7034 *l = cs.l;
7035}
7036EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
7037
7038int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
7039 struct kvm_sregs *sregs)
7040{
Gleb Natapov89a27f42010-02-16 10:51:48 +02007041 struct desc_ptr dt;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007042
Guillaume Thouvenin3e6e0aa2008-05-27 10:18:46 +02007043 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7044 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7045 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7046 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7047 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7048 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007049
Guillaume Thouvenin3e6e0aa2008-05-27 10:18:46 +02007050 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7051 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007052
7053 kvm_x86_ops->get_idt(vcpu, &dt);
Gleb Natapov89a27f42010-02-16 10:51:48 +02007054 sregs->idt.limit = dt.size;
7055 sregs->idt.base = dt.address;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007056 kvm_x86_ops->get_gdt(vcpu, &dt);
Gleb Natapov89a27f42010-02-16 10:51:48 +02007057 sregs->gdt.limit = dt.size;
7058 sregs->gdt.base = dt.address;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007059
Avi Kivity4d4ec082009-12-29 18:07:30 +02007060 sregs->cr0 = kvm_read_cr0(vcpu);
Zhang Xiantaoad312c72007-12-13 23:50:52 +08007061 sregs->cr2 = vcpu->arch.cr2;
Avi Kivity9f8fe502010-12-05 17:30:00 +02007062 sregs->cr3 = kvm_read_cr3(vcpu);
Avi Kivityfc78f512009-12-07 12:16:48 +02007063 sregs->cr4 = kvm_read_cr4(vcpu);
Avi Kivity2d3ad1f2008-02-24 11:20:43 +02007064 sregs->cr8 = kvm_get_cr8(vcpu);
Avi Kivityf6801df2010-01-21 15:31:50 +02007065 sregs->efer = vcpu->arch.efer;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007066 sregs->apic_base = kvm_get_apic_base(vcpu);
7067
Gleb Natapov923c61b2009-05-11 13:35:48 +03007068 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007069
Gleb Natapov36752c92009-05-11 13:35:53 +03007070 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
Gleb Natapov14d0bc12009-04-21 17:45:11 +03007071 set_bit(vcpu->arch.interrupt.nr,
7072 (unsigned long *)sregs->interrupt_bitmap);
Gleb Natapov16d7a192009-04-21 17:45:10 +03007073
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007074 return 0;
7075}
7076
Marcelo Tosatti62d9f0d2008-04-11 13:24:45 -03007077int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
7078 struct kvm_mp_state *mp_state)
7079{
Jan Kiszka66450a22013-03-13 12:42:34 +01007080 kvm_apic_accept_events(vcpu);
Srivatsa Vaddagiri6aef2662013-08-26 14:18:34 +05307081 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
7082 vcpu->arch.pv.pv_unhalted)
7083 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
7084 else
7085 mp_state->mp_state = vcpu->arch.mp_state;
7086
Marcelo Tosatti62d9f0d2008-04-11 13:24:45 -03007087 return 0;
7088}
7089
7090int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
7091 struct kvm_mp_state *mp_state)
7092{
Paolo Bonzinibce87cc2016-01-08 13:48:51 +01007093 if (!lapic_in_kernel(vcpu) &&
Jan Kiszka66450a22013-03-13 12:42:34 +01007094 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
7095 return -EINVAL;
7096
7097 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
7098 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
7099 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
7100 } else
7101 vcpu->arch.mp_state = mp_state->mp_state;
Avi Kivity3842d132010-07-27 12:30:24 +03007102 kvm_make_request(KVM_REQ_EVENT, vcpu);
Marcelo Tosatti62d9f0d2008-04-11 13:24:45 -03007103 return 0;
7104}
7105
Kevin Wolf7f3d35f2012-02-08 14:34:38 +01007106int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
7107 int reason, bool has_error_code, u32 error_code)
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007108{
Takuya Yoshikawa9d741912011-05-29 21:53:48 +09007109 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
Mohammed Gamal8ec47222010-08-16 00:47:01 +03007110 int ret;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007111
Mohammed Gamal8ec47222010-08-16 00:47:01 +03007112 init_emulate_ctxt(vcpu);
Izik Eidus37817f22008-03-24 23:14:53 +02007113
Kevin Wolf7f3d35f2012-02-08 14:34:38 +01007114 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
Takuya Yoshikawa9d741912011-05-29 21:53:48 +09007115 has_error_code, error_code);
Izik Eidus37817f22008-03-24 23:14:53 +02007116
Gleb Natapovc6975182010-02-18 12:15:01 +02007117 if (ret)
Gleb Natapov19d04432010-04-15 12:29:50 +03007118 return EMULATE_FAIL;
Gleb Natapovc6975182010-02-18 12:15:01 +02007119
Takuya Yoshikawa9d741912011-05-29 21:53:48 +09007120 kvm_rip_write(vcpu, ctxt->eip);
7121 kvm_set_rflags(vcpu, ctxt->eflags);
Avi Kivity3842d132010-07-27 12:30:24 +03007122 kvm_make_request(KVM_REQ_EVENT, vcpu);
Gleb Natapov19d04432010-04-15 12:29:50 +03007123 return EMULATE_DONE;
Izik Eidus37817f22008-03-24 23:14:53 +02007124}
7125EXPORT_SYMBOL_GPL(kvm_task_switch);
7126
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007127int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
7128 struct kvm_sregs *sregs)
7129{
Jan Kiszka58cb6282014-01-24 16:48:44 +01007130 struct msr_data apic_base_msr;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007131 int mmu_reset_needed = 0;
Xiao Guangrong63f42e02011-01-12 15:39:18 +08007132 int pending_vec, max_bits, idx;
Gleb Natapov89a27f42010-02-16 10:51:48 +02007133 struct desc_ptr dt;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007134
Petr Matousek6d1068b2012-11-06 19:24:07 +01007135 if (!guest_cpuid_has_xsave(vcpu) && (sregs->cr4 & X86_CR4_OSXSAVE))
7136 return -EINVAL;
7137
Gleb Natapov89a27f42010-02-16 10:51:48 +02007138 dt.size = sregs->idt.limit;
7139 dt.address = sregs->idt.base;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007140 kvm_x86_ops->set_idt(vcpu, &dt);
Gleb Natapov89a27f42010-02-16 10:51:48 +02007141 dt.size = sregs->gdt.limit;
7142 dt.address = sregs->gdt.base;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007143 kvm_x86_ops->set_gdt(vcpu, &dt);
7144
Zhang Xiantaoad312c72007-12-13 23:50:52 +08007145 vcpu->arch.cr2 = sregs->cr2;
Avi Kivity9f8fe502010-12-05 17:30:00 +02007146 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
Jan Kiszkadc7e7952009-07-01 20:52:03 +02007147 vcpu->arch.cr3 = sregs->cr3;
Avi Kivityaff48ba2010-12-05 18:56:11 +02007148 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007149
Avi Kivity2d3ad1f2008-02-24 11:20:43 +02007150 kvm_set_cr8(vcpu, sregs->cr8);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007151
Avi Kivityf6801df2010-01-21 15:31:50 +02007152 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007153 kvm_x86_ops->set_efer(vcpu, sregs->efer);
Jan Kiszka58cb6282014-01-24 16:48:44 +01007154 apic_base_msr.data = sregs->apic_base;
7155 apic_base_msr.host_initiated = true;
7156 kvm_set_apic_base(vcpu, &apic_base_msr);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007157
Avi Kivity4d4ec082009-12-29 18:07:30 +02007158 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007159 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
Paul Knowlesd7306162008-02-06 11:02:35 +00007160 vcpu->arch.cr0 = sregs->cr0;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007161
Avi Kivityfc78f512009-12-07 12:16:48 +02007162 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007163 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
Huaitong Hanb9baba82016-03-22 16:51:21 +08007164 if (sregs->cr4 & (X86_CR4_OSXSAVE | X86_CR4_PKE))
Avi Kivity00b27a32011-11-23 16:30:32 +02007165 kvm_update_cpuid(vcpu);
Xiao Guangrong63f42e02011-01-12 15:39:18 +08007166
7167 idx = srcu_read_lock(&vcpu->kvm->srcu);
Marcelo Tosatti7c93be42009-10-26 16:48:33 -02007168 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
Avi Kivity9f8fe502010-12-05 17:30:00 +02007169 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
Marcelo Tosatti7c93be42009-10-26 16:48:33 -02007170 mmu_reset_needed = 1;
7171 }
Xiao Guangrong63f42e02011-01-12 15:39:18 +08007172 srcu_read_unlock(&vcpu->kvm->srcu, idx);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007173
7174 if (mmu_reset_needed)
7175 kvm_mmu_reset_context(vcpu);
7176
Michael S. Tsirkina50abc32012-09-05 20:00:52 +03007177 max_bits = KVM_NR_INTERRUPTS;
Gleb Natapov923c61b2009-05-11 13:35:48 +03007178 pending_vec = find_first_bit(
7179 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
7180 if (pending_vec < max_bits) {
Gleb Natapov66fd3f72009-05-11 13:35:50 +03007181 kvm_queue_interrupt(vcpu, pending_vec, false);
Gleb Natapov923c61b2009-05-11 13:35:48 +03007182 pr_debug("Set back pending irq %d\n", pending_vec);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007183 }
7184
Guillaume Thouvenin3e6e0aa2008-05-27 10:18:46 +02007185 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7186 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7187 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7188 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7189 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7190 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007191
Guillaume Thouvenin3e6e0aa2008-05-27 10:18:46 +02007192 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7193 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007194
Mikhail Ershov5f0269f2009-08-03 14:58:25 +03007195 update_cr8_intercept(vcpu);
7196
Marcelo Tosatti9c3e4aa2008-09-10 16:40:55 -03007197 /* Older userspace won't unhalt the vcpu on reset. */
Gleb Natapovc5af89b2009-06-09 15:56:26 +03007198 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
Marcelo Tosatti9c3e4aa2008-09-10 16:40:55 -03007199 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
Avi Kivity3eeb3282010-01-21 15:31:48 +02007200 !is_protmode(vcpu))
Marcelo Tosatti9c3e4aa2008-09-10 16:40:55 -03007201 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7202
Avi Kivity3842d132010-07-27 12:30:24 +03007203 kvm_make_request(KVM_REQ_EVENT, vcpu);
7204
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007205 return 0;
7206}
7207
Jan Kiszkad0bfb942008-12-15 13:52:10 +01007208int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
7209 struct kvm_guest_debug *dbg)
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007210{
Jan Kiszka355be0b2009-10-03 00:31:21 +02007211 unsigned long rflags;
Jan Kiszkaae675ef2008-12-15 13:52:10 +01007212 int i, r;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007213
Jan Kiszka4f926bf22009-10-30 12:46:59 +01007214 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
7215 r = -EBUSY;
7216 if (vcpu->arch.exception.pending)
Avi Kivity2122ff52010-05-13 11:25:04 +03007217 goto out;
Jan Kiszka4f926bf22009-10-30 12:46:59 +01007218 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
7219 kvm_queue_exception(vcpu, DB_VECTOR);
7220 else
7221 kvm_queue_exception(vcpu, BP_VECTOR);
7222 }
7223
Jan Kiszka91586a32009-10-05 13:07:21 +02007224 /*
7225 * Read rflags as long as potentially injected trace flags are still
7226 * filtered out.
7227 */
7228 rflags = kvm_get_rflags(vcpu);
Jan Kiszka355be0b2009-10-03 00:31:21 +02007229
7230 vcpu->guest_debug = dbg->control;
7231 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
7232 vcpu->guest_debug = 0;
7233
7234 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
Jan Kiszkaae675ef2008-12-15 13:52:10 +01007235 for (i = 0; i < KVM_NR_DB_REGS; ++i)
7236 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
Jan Kiszkac8639012012-09-21 05:42:55 +02007237 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
Jan Kiszkaae675ef2008-12-15 13:52:10 +01007238 } else {
7239 for (i = 0; i < KVM_NR_DB_REGS; i++)
7240 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
Jan Kiszkaae675ef2008-12-15 13:52:10 +01007241 }
Jan Kiszkac8639012012-09-21 05:42:55 +02007242 kvm_update_dr7(vcpu);
Jan Kiszkaae675ef2008-12-15 13:52:10 +01007243
Jan Kiszkaf92653e2010-02-23 17:47:55 +01007244 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7245 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
7246 get_segment_base(vcpu, VCPU_SREG_CS);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007247
Jan Kiszka91586a32009-10-05 13:07:21 +02007248 /*
7249 * Trigger an rflags update that will inject or remove the trace
7250 * flags.
7251 */
7252 kvm_set_rflags(vcpu, rflags);
Jan Kiszkad0bfb942008-12-15 13:52:10 +01007253
Paolo Bonzinia96036b2015-11-10 11:55:36 +01007254 kvm_x86_ops->update_bp_intercept(vcpu);
Jan Kiszka355be0b2009-10-03 00:31:21 +02007255
Jan Kiszka4f926bf22009-10-30 12:46:59 +01007256 r = 0;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007257
Avi Kivity2122ff52010-05-13 11:25:04 +03007258out:
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007259
7260 return r;
7261}
7262
7263/*
Zhang Xiantao8b006792007-11-16 13:05:55 +08007264 * Translate a guest virtual address to a guest physical address.
7265 */
7266int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
7267 struct kvm_translation *tr)
7268{
7269 unsigned long vaddr = tr->linear_address;
7270 gpa_t gpa;
Marcelo Tosattif656ce02009-12-23 14:35:25 -02007271 int idx;
Zhang Xiantao8b006792007-11-16 13:05:55 +08007272
Marcelo Tosattif656ce02009-12-23 14:35:25 -02007273 idx = srcu_read_lock(&vcpu->kvm->srcu);
Gleb Natapov1871c602010-02-10 14:21:32 +02007274 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
Marcelo Tosattif656ce02009-12-23 14:35:25 -02007275 srcu_read_unlock(&vcpu->kvm->srcu, idx);
Zhang Xiantao8b006792007-11-16 13:05:55 +08007276 tr->physical_address = gpa;
7277 tr->valid = gpa != UNMAPPED_GVA;
7278 tr->writeable = 1;
7279 tr->usermode = 0;
Zhang Xiantao8b006792007-11-16 13:05:55 +08007280
7281 return 0;
7282}
7283
Hollis Blanchardd0752062007-10-31 17:24:25 -05007284int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7285{
Ingo Molnarc47ada32015-04-30 17:15:32 +02007286 struct fxregs_state *fxsave =
Ingo Molnar7366ed72015-04-27 04:19:39 +02007287 &vcpu->arch.guest_fpu.state.fxsave;
Hollis Blanchardd0752062007-10-31 17:24:25 -05007288
Hollis Blanchardd0752062007-10-31 17:24:25 -05007289 memcpy(fpu->fpr, fxsave->st_space, 128);
7290 fpu->fcw = fxsave->cwd;
7291 fpu->fsw = fxsave->swd;
7292 fpu->ftwx = fxsave->twd;
7293 fpu->last_opcode = fxsave->fop;
7294 fpu->last_ip = fxsave->rip;
7295 fpu->last_dp = fxsave->rdp;
7296 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
7297
Hollis Blanchardd0752062007-10-31 17:24:25 -05007298 return 0;
7299}
7300
7301int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7302{
Ingo Molnarc47ada32015-04-30 17:15:32 +02007303 struct fxregs_state *fxsave =
Ingo Molnar7366ed72015-04-27 04:19:39 +02007304 &vcpu->arch.guest_fpu.state.fxsave;
Hollis Blanchardd0752062007-10-31 17:24:25 -05007305
Hollis Blanchardd0752062007-10-31 17:24:25 -05007306 memcpy(fxsave->st_space, fpu->fpr, 128);
7307 fxsave->cwd = fpu->fcw;
7308 fxsave->swd = fpu->fsw;
7309 fxsave->twd = fpu->ftwx;
7310 fxsave->fop = fpu->last_opcode;
7311 fxsave->rip = fpu->last_ip;
7312 fxsave->rdp = fpu->last_dp;
7313 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
7314
Hollis Blanchardd0752062007-10-31 17:24:25 -05007315 return 0;
7316}
7317
Ingo Molnar0ee6a512015-04-27 06:58:22 +02007318static void fx_init(struct kvm_vcpu *vcpu)
Hollis Blanchardd0752062007-10-31 17:24:25 -05007319{
Ingo Molnarbf935b02015-04-30 10:23:42 +02007320 fpstate_init(&vcpu->arch.guest_fpu.state);
Borislav Petkov782511b2016-04-04 22:25:03 +02007321 if (boot_cpu_has(X86_FEATURE_XSAVES))
Ingo Molnar7366ed72015-04-27 04:19:39 +02007322 vcpu->arch.guest_fpu.state.xsave.header.xcomp_bv =
Paolo Bonzinidf1daba2014-11-21 19:05:07 +01007323 host_xcr0 | XSTATE_COMPACTION_ENABLED;
Hollis Blanchardd0752062007-10-31 17:24:25 -05007324
Dexuan Cui2acf9232010-06-10 11:27:12 +08007325 /*
7326 * Ensure guest xcr0 is valid for loading
7327 */
Dave Hansend91cab72015-09-02 16:31:26 -07007328 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
Dexuan Cui2acf9232010-06-10 11:27:12 +08007329
Zhang Xiantaoad312c72007-12-13 23:50:52 +08007330 vcpu->arch.cr0 |= X86_CR0_ET;
Hollis Blanchardd0752062007-10-31 17:24:25 -05007331}
Hollis Blanchardd0752062007-10-31 17:24:25 -05007332
7333void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
7334{
Avi Kivity2608d7a2010-01-21 15:31:45 +02007335 if (vcpu->guest_fpu_loaded)
Hollis Blanchardd0752062007-10-31 17:24:25 -05007336 return;
7337
Dexuan Cui2acf9232010-06-10 11:27:12 +08007338 /*
7339 * Restore all possible states in the guest,
7340 * and assume host would use all available bits.
7341 * Guest xcr0 would be loaded later.
7342 */
Hollis Blanchardd0752062007-10-31 17:24:25 -05007343 vcpu->guest_fpu_loaded = 1;
Suresh Siddhab1a74bf2012-09-20 11:01:49 -07007344 __kernel_fpu_begin();
Ingo Molnar003e2e82015-05-25 11:59:35 +02007345 __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu.state);
Avi Kivity0c048512010-01-21 15:31:52 +02007346 trace_kvm_fpu(1);
Hollis Blanchardd0752062007-10-31 17:24:25 -05007347}
Hollis Blanchardd0752062007-10-31 17:24:25 -05007348
7349void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
7350{
Rik van Riel653f52c2015-04-23 11:52:37 -04007351 if (!vcpu->guest_fpu_loaded) {
7352 vcpu->fpu_counter = 0;
Hollis Blanchardd0752062007-10-31 17:24:25 -05007353 return;
Rik van Riel653f52c2015-04-23 11:52:37 -04007354 }
Hollis Blanchardd0752062007-10-31 17:24:25 -05007355
7356 vcpu->guest_fpu_loaded = 0;
Ingo Molnar4f836342015-04-27 02:53:16 +02007357 copy_fpregs_to_fpstate(&vcpu->arch.guest_fpu);
Suresh Siddhab1a74bf2012-09-20 11:01:49 -07007358 __kernel_fpu_end();
Avi Kivityf096ed82007-11-18 13:54:33 +02007359 ++vcpu->stat.fpu_reload;
Rik van Riel653f52c2015-04-23 11:52:37 -04007360 /*
7361 * If using eager FPU mode, or if the guest is a frequent user
7362 * of the FPU, just leave the FPU active for next time.
7363 * Every 255 times fpu_counter rolls over to 0; a guest that uses
7364 * the FPU in bursts will revert to loading it on demand.
7365 */
Paolo Bonzini5a5fbdc2016-03-08 10:00:11 +01007366 if (!use_eager_fpu()) {
Rik van Riel653f52c2015-04-23 11:52:37 -04007367 if (++vcpu->fpu_counter < 5)
7368 kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
7369 }
Avi Kivity0c048512010-01-21 15:31:52 +02007370 trace_kvm_fpu(0);
Hollis Blanchardd0752062007-10-31 17:24:25 -05007371}
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08007372
7373void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
7374{
Glauber Costa12f9a482011-02-01 14:16:40 -05007375 kvmclock_reset(vcpu);
Joerg Roedel7f1ea202009-02-25 16:08:31 +01007376
Sheng Yangf5f48ee2010-06-30 12:25:15 +08007377 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08007378 kvm_x86_ops->vcpu_free(vcpu);
7379}
7380
7381struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
7382 unsigned int id)
7383{
Liang Lic447e762015-05-21 04:41:25 +08007384 struct kvm_vcpu *vcpu;
7385
Zachary Amsden6755bae2010-08-19 22:07:22 -10007386 if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
7387 printk_once(KERN_WARNING
7388 "kvm: SMP vm created on host with unstable TSC; "
7389 "guest TSC will not be reliable\n");
Liang Lic447e762015-05-21 04:41:25 +08007390
7391 vcpu = kvm_x86_ops->vcpu_create(kvm, id);
7392
Liang Lic447e762015-05-21 04:41:25 +08007393 return vcpu;
Avi Kivity26e52152007-11-20 15:30:24 +02007394}
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08007395
Avi Kivity26e52152007-11-20 15:30:24 +02007396int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
7397{
7398 int r;
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08007399
Xiao Guangrong19efffa2015-06-15 16:55:31 +08007400 kvm_vcpu_mtrr_init(vcpu);
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +03007401 r = vcpu_load(vcpu);
7402 if (r)
7403 return r;
Nadav Amitd28bc9d2015-04-13 14:34:08 +03007404 kvm_vcpu_reset(vcpu, false);
Paolo Bonzini8a3c1a332013-10-02 16:56:13 +02007405 kvm_mmu_setup(vcpu);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08007406 vcpu_put(vcpu);
Avi Kivity26e52152007-11-20 15:30:24 +02007407 return r;
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08007408}
7409
Dominik Dingel31928aa2014-12-04 15:47:07 +01007410void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
Marcelo Tosatti42897d82012-11-27 23:29:02 -02007411{
Will Auld8fe8ab42012-11-29 12:42:12 -08007412 struct msr_data msr;
Andrew Jones332967a2014-02-28 12:52:55 +01007413 struct kvm *kvm = vcpu->kvm;
Marcelo Tosatti42897d82012-11-27 23:29:02 -02007414
Dominik Dingel31928aa2014-12-04 15:47:07 +01007415 if (vcpu_load(vcpu))
7416 return;
Will Auld8fe8ab42012-11-29 12:42:12 -08007417 msr.data = 0x0;
7418 msr.index = MSR_IA32_TSC;
7419 msr.host_initiated = true;
7420 kvm_write_tsc(vcpu, &msr);
Marcelo Tosatti42897d82012-11-27 23:29:02 -02007421 vcpu_put(vcpu);
7422
Marcelo Tosatti630994b2015-05-12 22:42:04 -03007423 if (!kvmclock_periodic_sync)
7424 return;
7425
Andrew Jones332967a2014-02-28 12:52:55 +01007426 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
7427 KVMCLOCK_SYNC_PERIOD);
Marcelo Tosatti42897d82012-11-27 23:29:02 -02007428}
7429
Hollis Blanchardd40ccc62007-11-19 14:04:43 -06007430void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08007431{
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +03007432 int r;
Gleb Natapov344d9582010-10-14 11:22:50 +02007433 vcpu->arch.apf.msr_val = 0;
7434
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +03007435 r = vcpu_load(vcpu);
7436 BUG_ON(r);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08007437 kvm_mmu_unload(vcpu);
7438 vcpu_put(vcpu);
7439
7440 kvm_x86_ops->vcpu_free(vcpu);
7441}
7442
Nadav Amitd28bc9d2015-04-13 14:34:08 +03007443void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08007444{
Paolo Bonzinie69fab52015-06-04 10:44:44 +02007445 vcpu->arch.hflags = 0;
7446
Paolo Bonzinic43203c2016-06-01 22:26:00 +02007447 vcpu->arch.smi_pending = 0;
Avi Kivity7460fb4a2011-09-20 13:43:14 +03007448 atomic_set(&vcpu->arch.nmi_queued, 0);
7449 vcpu->arch.nmi_pending = 0;
Jan Kiszka448fa4a2008-09-26 09:30:48 +02007450 vcpu->arch.nmi_injected = false;
Nadav Amit5f7552d2014-06-30 12:03:02 +03007451 kvm_clear_interrupt_queue(vcpu);
7452 kvm_clear_exception_queue(vcpu);
Jan Kiszka448fa4a2008-09-26 09:30:48 +02007453
Jan Kiszka42dbaa52008-12-15 13:52:10 +01007454 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
Nadav Amitae561ed2015-04-02 03:10:37 +03007455 kvm_update_dr0123(vcpu);
Nadav Amit6f43ed02014-07-15 17:37:46 +03007456 vcpu->arch.dr6 = DR6_INIT;
Jan Kiszka73aaf249e2014-01-04 18:47:16 +01007457 kvm_update_dr6(vcpu);
Jan Kiszka42dbaa52008-12-15 13:52:10 +01007458 vcpu->arch.dr7 = DR7_FIXED_1;
Jan Kiszkac8639012012-09-21 05:42:55 +02007459 kvm_update_dr7(vcpu);
Jan Kiszka42dbaa52008-12-15 13:52:10 +01007460
Nadav Amit11190222015-04-02 03:10:38 +03007461 vcpu->arch.cr2 = 0;
7462
Avi Kivity3842d132010-07-27 12:30:24 +03007463 kvm_make_request(KVM_REQ_EVENT, vcpu);
Gleb Natapov344d9582010-10-14 11:22:50 +02007464 vcpu->arch.apf.msr_val = 0;
Glauber Costac9aaa892011-07-11 15:28:14 -04007465 vcpu->arch.st.msr_val = 0;
Avi Kivity3842d132010-07-27 12:30:24 +03007466
Glauber Costa12f9a482011-02-01 14:16:40 -05007467 kvmclock_reset(vcpu);
7468
Gleb Natapovaf585b92010-10-14 11:22:46 +02007469 kvm_clear_async_pf_completion_queue(vcpu);
7470 kvm_async_pf_hash_reset(vcpu);
7471 vcpu->arch.apf.halted = false;
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08007472
Paolo Bonzini64d60672015-05-07 11:36:11 +02007473 if (!init_event) {
Nadav Amitd28bc9d2015-04-13 14:34:08 +03007474 kvm_pmu_reset(vcpu);
Paolo Bonzini64d60672015-05-07 11:36:11 +02007475 vcpu->arch.smbase = 0x30000;
7476 }
Gleb Natapovf5132b02011-11-10 14:57:22 +02007477
Julian Stecklina66f7b722012-12-05 15:26:19 +01007478 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
7479 vcpu->arch.regs_avail = ~0;
7480 vcpu->arch.regs_dirty = ~0;
7481
Nadav Amitd28bc9d2015-04-13 14:34:08 +03007482 kvm_x86_ops->vcpu_reset(vcpu, init_event);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08007483}
7484
Paolo Bonzini2b4a2732014-11-24 14:35:24 +01007485void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
Jan Kiszka66450a22013-03-13 12:42:34 +01007486{
7487 struct kvm_segment cs;
7488
7489 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7490 cs.selector = vector << 8;
7491 cs.base = vector << 12;
7492 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7493 kvm_rip_write(vcpu, 0);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08007494}
7495
Radim Krčmář13a34e02014-08-28 15:13:03 +02007496int kvm_arch_hardware_enable(void)
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08007497{
Zachary Amsdenca84d1a2010-08-19 22:07:28 -10007498 struct kvm *kvm;
7499 struct kvm_vcpu *vcpu;
7500 int i;
Zachary Amsden0dd6a6e2012-02-03 15:43:56 -02007501 int ret;
7502 u64 local_tsc;
7503 u64 max_tsc = 0;
7504 bool stable, backwards_tsc = false;
Avi Kivity18863bd2009-09-07 11:12:18 +03007505
7506 kvm_shared_msr_cpu_online();
Radim Krčmář13a34e02014-08-28 15:13:03 +02007507 ret = kvm_x86_ops->hardware_enable();
Zachary Amsden0dd6a6e2012-02-03 15:43:56 -02007508 if (ret != 0)
7509 return ret;
7510
Andy Lutomirski4ea16362015-06-25 18:44:07 +02007511 local_tsc = rdtsc();
Zachary Amsden0dd6a6e2012-02-03 15:43:56 -02007512 stable = !check_tsc_unstable();
7513 list_for_each_entry(kvm, &vm_list, vm_list) {
7514 kvm_for_each_vcpu(i, vcpu, kvm) {
7515 if (!stable && vcpu->cpu == smp_processor_id())
Guo Hui Liu105b21b2014-09-12 13:43:19 +08007516 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
Zachary Amsden0dd6a6e2012-02-03 15:43:56 -02007517 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
7518 backwards_tsc = true;
7519 if (vcpu->arch.last_host_tsc > max_tsc)
7520 max_tsc = vcpu->arch.last_host_tsc;
7521 }
7522 }
7523 }
7524
7525 /*
7526 * Sometimes, even reliable TSCs go backwards. This happens on
7527 * platforms that reset TSC during suspend or hibernate actions, but
7528 * maintain synchronization. We must compensate. Fortunately, we can
7529 * detect that condition here, which happens early in CPU bringup,
7530 * before any KVM threads can be running. Unfortunately, we can't
7531 * bring the TSCs fully up to date with real time, as we aren't yet far
7532 * enough into CPU bringup that we know how much real time has actually
7533 * elapsed; our helper function, get_kernel_ns() will be using boot
7534 * variables that haven't been updated yet.
7535 *
7536 * So we simply find the maximum observed TSC above, then record the
7537 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
7538 * the adjustment will be applied. Note that we accumulate
7539 * adjustments, in case multiple suspend cycles happen before some VCPU
7540 * gets a chance to run again. In the event that no KVM threads get a
7541 * chance to run, we will miss the entire elapsed period, as we'll have
7542 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
7543 * loose cycle time. This isn't too big a deal, since the loss will be
7544 * uniform across all VCPUs (not to mention the scenario is extremely
7545 * unlikely). It is possible that a second hibernate recovery happens
7546 * much faster than a first, causing the observed TSC here to be
7547 * smaller; this would require additional padding adjustment, which is
7548 * why we set last_host_tsc to the local tsc observed here.
7549 *
7550 * N.B. - this code below runs only on platforms with reliable TSC,
7551 * as that is the only way backwards_tsc is set above. Also note
7552 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
7553 * have the same delta_cyc adjustment applied if backwards_tsc
7554 * is detected. Note further, this adjustment is only done once,
7555 * as we reset last_host_tsc on all VCPUs to stop this from being
7556 * called multiple times (one for each physical CPU bringup).
7557 *
Guo Chao4a969982012-06-28 15:17:27 +08007558 * Platforms with unreliable TSCs don't have to deal with this, they
Zachary Amsden0dd6a6e2012-02-03 15:43:56 -02007559 * will be compensated by the logic in vcpu_load, which sets the TSC to
7560 * catchup mode. This will catchup all VCPUs to real time, but cannot
7561 * guarantee that they stay in perfect synchronization.
7562 */
7563 if (backwards_tsc) {
7564 u64 delta_cyc = max_tsc - local_tsc;
Marcelo Tosatti16a96022014-05-14 12:43:24 -03007565 backwards_tsc_observed = true;
Zachary Amsden0dd6a6e2012-02-03 15:43:56 -02007566 list_for_each_entry(kvm, &vm_list, vm_list) {
7567 kvm_for_each_vcpu(i, vcpu, kvm) {
7568 vcpu->arch.tsc_offset_adjustment += delta_cyc;
7569 vcpu->arch.last_host_tsc = local_tsc;
Guo Hui Liu105b21b2014-09-12 13:43:19 +08007570 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
Zachary Amsden0dd6a6e2012-02-03 15:43:56 -02007571 }
7572
7573 /*
7574 * We have to disable TSC offset matching.. if you were
7575 * booting a VM while issuing an S4 host suspend....
7576 * you may have some problem. Solving this issue is
7577 * left as an exercise to the reader.
7578 */
7579 kvm->arch.last_tsc_nsec = 0;
7580 kvm->arch.last_tsc_write = 0;
7581 }
7582
7583 }
7584 return 0;
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08007585}
7586
Radim Krčmář13a34e02014-08-28 15:13:03 +02007587void kvm_arch_hardware_disable(void)
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08007588{
Radim Krčmář13a34e02014-08-28 15:13:03 +02007589 kvm_x86_ops->hardware_disable();
7590 drop_user_return_notifiers();
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08007591}
7592
7593int kvm_arch_hardware_setup(void)
7594{
Nadav Amit9e9c3fe2015-04-12 21:47:15 +03007595 int r;
7596
7597 r = kvm_x86_ops->hardware_setup();
7598 if (r != 0)
7599 return r;
7600
Haozhong Zhang35181e82015-10-20 15:39:03 +08007601 if (kvm_has_tsc_control) {
7602 /*
7603 * Make sure the user can only configure tsc_khz values that
7604 * fit into a signed integer.
7605 * A min value is not calculated needed because it will always
7606 * be 1 on all machines.
7607 */
7608 u64 max = min(0x7fffffffULL,
7609 __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
7610 kvm_max_guest_tsc_khz = max;
7611
Haozhong Zhangad721882015-10-20 15:39:02 +08007612 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
Haozhong Zhang35181e82015-10-20 15:39:03 +08007613 }
Haozhong Zhangad721882015-10-20 15:39:02 +08007614
Nadav Amit9e9c3fe2015-04-12 21:47:15 +03007615 kvm_init_msr_list();
7616 return 0;
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08007617}
7618
7619void kvm_arch_hardware_unsetup(void)
7620{
7621 kvm_x86_ops->hardware_unsetup();
7622}
7623
7624void kvm_arch_check_processor_compat(void *rtn)
7625{
7626 kvm_x86_ops->check_processor_compatibility(rtn);
7627}
7628
Paolo Bonzinid71ba782015-07-29 11:56:48 +02007629bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
7630{
7631 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
7632}
7633EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
7634
7635bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
7636{
7637 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
7638}
7639
Gleb Natapov54e98182012-08-05 15:58:32 +03007640struct static_key kvm_no_apic_vcpu __read_mostly;
Paolo Bonzinibce87cc2016-01-08 13:48:51 +01007641EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
Gleb Natapov54e98182012-08-05 15:58:32 +03007642
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08007643int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
7644{
7645 struct page *page;
7646 struct kvm *kvm;
7647 int r;
7648
7649 BUG_ON(vcpu->kvm == NULL);
7650 kvm = vcpu->kvm;
7651
Andrey Smetanind62caab2015-11-10 15:36:33 +03007652 vcpu->arch.apicv_active = kvm_x86_ops->get_enable_apicv();
Srivatsa Vaddagiri6aef2662013-08-26 14:18:34 +05307653 vcpu->arch.pv.pv_unhalted = false;
Avi Kivity9aabc88f2010-07-29 15:11:50 +03007654 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
Nadav Amit58d269d2015-04-02 03:10:36 +03007655 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_reset_bsp(vcpu))
Avi Kivitya4535292008-04-13 17:54:35 +03007656 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08007657 else
Avi Kivitya4535292008-04-13 17:54:35 +03007658 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08007659
7660 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
7661 if (!page) {
7662 r = -ENOMEM;
7663 goto fail;
7664 }
Zhang Xiantaoad312c72007-12-13 23:50:52 +08007665 vcpu->arch.pio_data = page_address(page);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08007666
Zachary Amsdencc578282012-02-03 15:43:50 -02007667 kvm_set_tsc_khz(vcpu, max_tsc_khz);
Zachary Amsdenc2855452010-09-18 14:38:15 -10007668
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08007669 r = kvm_mmu_create(vcpu);
7670 if (r < 0)
7671 goto fail_free_pio_data;
7672
7673 if (irqchip_in_kernel(kvm)) {
7674 r = kvm_create_lapic(vcpu);
7675 if (r < 0)
7676 goto fail_mmu_destroy;
Gleb Natapov54e98182012-08-05 15:58:32 +03007677 } else
7678 static_key_slow_inc(&kvm_no_apic_vcpu);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08007679
Huang Ying890ca9a2009-05-11 16:48:15 +08007680 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
7681 GFP_KERNEL);
7682 if (!vcpu->arch.mce_banks) {
7683 r = -ENOMEM;
Wei Yongjun443c39b2010-01-22 14:21:29 +08007684 goto fail_free_lapic;
Huang Ying890ca9a2009-05-11 16:48:15 +08007685 }
7686 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
7687
Wei Yongjunf1797352013-04-18 07:41:00 +08007688 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
7689 r = -ENOMEM;
Sheng Yangf5f48ee2010-06-30 12:25:15 +08007690 goto fail_free_mce_banks;
Wei Yongjunf1797352013-04-18 07:41:00 +08007691 }
Sheng Yangf5f48ee2010-06-30 12:25:15 +08007692
Ingo Molnar0ee6a512015-04-27 06:58:22 +02007693 fx_init(vcpu);
Julian Stecklina66f7b722012-12-05 15:26:19 +01007694
Will Auldba904632012-11-29 12:42:50 -08007695 vcpu->arch.ia32_tsc_adjust_msr = 0x0;
Andy Honig0b794592013-02-20 14:48:10 -08007696 vcpu->arch.pv_time_enabled = false;
Paolo Bonzinid7876f12013-10-02 16:06:15 +02007697
7698 vcpu->arch.guest_supported_xcr0 = 0;
Paolo Bonzini4344ee92013-10-02 16:06:16 +02007699 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
Paolo Bonzinid7876f12013-10-02 16:06:15 +02007700
Eugene Korenevsky5a4f55c2015-03-29 23:56:12 +03007701 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
7702
Radim Krčmář74545702015-04-27 15:11:25 +02007703 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
7704
Gleb Natapovaf585b92010-10-14 11:22:46 +02007705 kvm_async_pf_hash_reset(vcpu);
Gleb Natapovf5132b02011-11-10 14:57:22 +02007706 kvm_pmu_init(vcpu);
Gleb Natapovaf585b92010-10-14 11:22:46 +02007707
Steve Rutherford1c1a9ce2015-07-30 11:27:16 +02007708 vcpu->arch.pending_external_vector = -1;
7709
Andrey Smetanin5c9194122015-11-10 15:36:34 +03007710 kvm_hv_vcpu_init(vcpu);
7711
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08007712 return 0;
Ingo Molnar0ee6a512015-04-27 06:58:22 +02007713
Sheng Yangf5f48ee2010-06-30 12:25:15 +08007714fail_free_mce_banks:
7715 kfree(vcpu->arch.mce_banks);
Wei Yongjun443c39b2010-01-22 14:21:29 +08007716fail_free_lapic:
7717 kvm_free_lapic(vcpu);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08007718fail_mmu_destroy:
7719 kvm_mmu_destroy(vcpu);
7720fail_free_pio_data:
Zhang Xiantaoad312c72007-12-13 23:50:52 +08007721 free_page((unsigned long)vcpu->arch.pio_data);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08007722fail:
7723 return r;
7724}
7725
7726void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
7727{
Marcelo Tosattif656ce02009-12-23 14:35:25 -02007728 int idx;
7729
Andrey Smetanin1f4b34f2015-11-30 19:22:21 +03007730 kvm_hv_vcpu_uninit(vcpu);
Gleb Natapovf5132b02011-11-10 14:57:22 +02007731 kvm_pmu_destroy(vcpu);
Wei Yongjun36cb93f2010-01-22 14:18:47 +08007732 kfree(vcpu->arch.mce_banks);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08007733 kvm_free_lapic(vcpu);
Marcelo Tosattif656ce02009-12-23 14:35:25 -02007734 idx = srcu_read_lock(&vcpu->kvm->srcu);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08007735 kvm_mmu_destroy(vcpu);
Marcelo Tosattif656ce02009-12-23 14:35:25 -02007736 srcu_read_unlock(&vcpu->kvm->srcu, idx);
Zhang Xiantaoad312c72007-12-13 23:50:52 +08007737 free_page((unsigned long)vcpu->arch.pio_data);
Paolo Bonzini35754c92015-07-29 12:05:37 +02007738 if (!lapic_in_kernel(vcpu))
Gleb Natapov54e98182012-08-05 15:58:32 +03007739 static_key_slow_dec(&kvm_no_apic_vcpu);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08007740}
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +08007741
Radim Krčmáře790d9e2014-08-21 18:08:05 +02007742void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
7743{
Radim Krčmářae97a3b2014-08-21 18:08:06 +02007744 kvm_x86_ops->sched_in(vcpu, cpu);
Radim Krčmáře790d9e2014-08-21 18:08:05 +02007745}
7746
Carsten Ottee08b9632012-01-04 10:25:20 +01007747int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +08007748{
Carsten Ottee08b9632012-01-04 10:25:20 +01007749 if (type)
7750 return -EINVAL;
7751
Paolo Bonzini6ef768f2014-11-20 13:45:31 +01007752 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
Zhang Xiantaof05e70a2007-12-14 10:01:48 +08007753 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
Xiao Guangrong365c8862013-05-31 08:36:29 +08007754 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
Ben-Ami Yassour4d5c5d02008-07-28 19:26:26 +03007755 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
Alex Williamsone0f0bbc2013-10-30 11:02:30 -06007756 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +08007757
Sheng Yang5550af42008-10-15 20:15:06 +08007758 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
7759 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
Alex Williamson7a844282012-09-21 11:58:03 -06007760 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
7761 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
7762 &kvm->arch.irq_sources_bitmap);
Sheng Yang5550af42008-10-15 20:15:06 +08007763
Jan Kiszka038f8c12011-02-04 10:49:11 +01007764 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
Gleb Natapov1e08ec42012-09-13 17:19:24 +03007765 mutex_init(&kvm->arch.apic_map_lock);
Marcelo Tosattid8281992012-11-27 23:29:01 -02007766 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
7767
7768 pvclock_update_vm_gtod_copy(kvm);
Marcelo Tosatti53f658b32008-12-11 20:45:05 +01007769
Andrew Jones7e44e442014-02-28 12:52:54 +01007770 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
Andrew Jones332967a2014-02-28 12:52:55 +01007771 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
Andrew Jones7e44e442014-02-28 12:52:54 +01007772
Xiao Guangrong0eb05bf2016-02-24 17:51:13 +08007773 kvm_page_track_init(kvm);
Xiao Guangrong13d268c2016-02-24 17:51:16 +08007774 kvm_mmu_init_vm(kvm);
Xiao Guangrong0eb05bf2016-02-24 17:51:13 +08007775
Suravee Suthikulpanit03543132016-05-04 14:09:42 -05007776 if (kvm_x86_ops->vm_init)
7777 return kvm_x86_ops->vm_init(kvm);
7778
Jan Kiszkad89f5ef2010-11-09 17:02:49 +01007779 return 0;
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +08007780}
7781
7782static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
7783{
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +03007784 int r;
7785 r = vcpu_load(vcpu);
7786 BUG_ON(r);
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +08007787 kvm_mmu_unload(vcpu);
7788 vcpu_put(vcpu);
7789}
7790
7791static void kvm_free_vcpus(struct kvm *kvm)
7792{
7793 unsigned int i;
Gleb Natapov988a2ca2009-06-09 15:56:29 +03007794 struct kvm_vcpu *vcpu;
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +08007795
7796 /*
7797 * Unpin any mmu pages first.
7798 */
Gleb Natapovaf585b92010-10-14 11:22:46 +02007799 kvm_for_each_vcpu(i, vcpu, kvm) {
7800 kvm_clear_async_pf_completion_queue(vcpu);
Gleb Natapov988a2ca2009-06-09 15:56:29 +03007801 kvm_unload_vcpu_mmu(vcpu);
Gleb Natapovaf585b92010-10-14 11:22:46 +02007802 }
Gleb Natapov988a2ca2009-06-09 15:56:29 +03007803 kvm_for_each_vcpu(i, vcpu, kvm)
7804 kvm_arch_vcpu_free(vcpu);
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +08007805
Gleb Natapov988a2ca2009-06-09 15:56:29 +03007806 mutex_lock(&kvm->lock);
7807 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
7808 kvm->vcpus[i] = NULL;
7809
7810 atomic_set(&kvm->online_vcpus, 0);
7811 mutex_unlock(&kvm->lock);
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +08007812}
7813
Sheng Yangad8ba2c2009-01-06 10:03:02 +08007814void kvm_arch_sync_events(struct kvm *kvm)
7815{
Andrew Jones332967a2014-02-28 12:52:55 +01007816 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
Andrew Jones7e44e442014-02-28 12:52:54 +01007817 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
Sheng Yangba4cef32009-01-06 10:03:03 +08007818 kvm_free_all_assigned_devices(kvm);
Xiao Guangrongaea924f2010-07-10 17:37:56 +08007819 kvm_free_pit(kvm);
Sheng Yangad8ba2c2009-01-06 10:03:02 +08007820}
7821
Paolo Bonzini1d8007b2015-10-12 13:38:32 +02007822int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
Paolo Bonzini9da0e4d2015-05-18 13:33:16 +02007823{
7824 int i, r;
Paolo Bonzini25188b92015-10-14 15:51:08 +02007825 unsigned long hva;
Paolo Bonzinif0d648b2015-10-12 13:56:27 +02007826 struct kvm_memslots *slots = kvm_memslots(kvm);
7827 struct kvm_memory_slot *slot, old;
Paolo Bonzini9da0e4d2015-05-18 13:33:16 +02007828
7829 /* Called with kvm->slots_lock held. */
Paolo Bonzini1d8007b2015-10-12 13:38:32 +02007830 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
7831 return -EINVAL;
Paolo Bonzini9da0e4d2015-05-18 13:33:16 +02007832
Paolo Bonzinif0d648b2015-10-12 13:56:27 +02007833 slot = id_to_memslot(slots, id);
7834 if (size) {
Paolo Bonzinib21629d2016-06-01 14:09:18 +02007835 if (slot->npages)
Paolo Bonzinif0d648b2015-10-12 13:56:27 +02007836 return -EEXIST;
7837
7838 /*
7839 * MAP_SHARED to prevent internal slot pages from being moved
7840 * by fork()/COW.
7841 */
7842 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
7843 MAP_SHARED | MAP_ANONYMOUS, 0);
7844 if (IS_ERR((void *)hva))
7845 return PTR_ERR((void *)hva);
7846 } else {
7847 if (!slot->npages)
7848 return 0;
7849
7850 hva = 0;
7851 }
7852
7853 old = *slot;
Paolo Bonzini9da0e4d2015-05-18 13:33:16 +02007854 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
Paolo Bonzini1d8007b2015-10-12 13:38:32 +02007855 struct kvm_userspace_memory_region m;
Paolo Bonzini9da0e4d2015-05-18 13:33:16 +02007856
Paolo Bonzini1d8007b2015-10-12 13:38:32 +02007857 m.slot = id | (i << 16);
7858 m.flags = 0;
7859 m.guest_phys_addr = gpa;
Paolo Bonzinif0d648b2015-10-12 13:56:27 +02007860 m.userspace_addr = hva;
Paolo Bonzini1d8007b2015-10-12 13:38:32 +02007861 m.memory_size = size;
Paolo Bonzini9da0e4d2015-05-18 13:33:16 +02007862 r = __kvm_set_memory_region(kvm, &m);
7863 if (r < 0)
7864 return r;
7865 }
7866
Paolo Bonzinif0d648b2015-10-12 13:56:27 +02007867 if (!size) {
7868 r = vm_munmap(old.userspace_addr, old.npages * PAGE_SIZE);
7869 WARN_ON(r < 0);
7870 }
7871
Paolo Bonzini9da0e4d2015-05-18 13:33:16 +02007872 return 0;
7873}
7874EXPORT_SYMBOL_GPL(__x86_set_memory_region);
7875
Paolo Bonzini1d8007b2015-10-12 13:38:32 +02007876int x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
Paolo Bonzini9da0e4d2015-05-18 13:33:16 +02007877{
7878 int r;
7879
7880 mutex_lock(&kvm->slots_lock);
Paolo Bonzini1d8007b2015-10-12 13:38:32 +02007881 r = __x86_set_memory_region(kvm, id, gpa, size);
Paolo Bonzini9da0e4d2015-05-18 13:33:16 +02007882 mutex_unlock(&kvm->slots_lock);
7883
7884 return r;
7885}
7886EXPORT_SYMBOL_GPL(x86_set_memory_region);
7887
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +08007888void kvm_arch_destroy_vm(struct kvm *kvm)
7889{
Andrew Honig27469d22013-04-18 09:38:14 -07007890 if (current->mm == kvm->mm) {
7891 /*
7892 * Free memory regions allocated on behalf of userspace,
7893 * unless the the memory map has changed due to process exit
7894 * or fd copying.
7895 */
Paolo Bonzini1d8007b2015-10-12 13:38:32 +02007896 x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0);
7897 x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, 0, 0);
7898 x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
Andrew Honig27469d22013-04-18 09:38:14 -07007899 }
Suravee Suthikulpanit03543132016-05-04 14:09:42 -05007900 if (kvm_x86_ops->vm_destroy)
7901 kvm_x86_ops->vm_destroy(kvm);
Sheng Yang6eb55812008-10-31 12:37:41 +08007902 kvm_iommu_unmap_guest(kvm);
Zhang Xiantaod7deeeb02007-12-14 10:17:34 +08007903 kfree(kvm->arch.vpic);
7904 kfree(kvm->arch.vioapic);
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +08007905 kvm_free_vcpus(kvm);
Radim Krčmářaf1bae52016-07-12 22:09:30 +02007906 kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
Xiao Guangrong13d268c2016-02-24 17:51:16 +08007907 kvm_mmu_uninit_vm(kvm);
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +08007908}
Zhang Xiantao0de10342007-11-20 16:25:04 +08007909
Aneesh Kumar K.V55870272013-10-07 22:18:00 +05307910void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +09007911 struct kvm_memory_slot *dont)
7912{
7913 int i;
7914
Takuya Yoshikawad89cc612012-08-01 18:03:28 +09007915 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7916 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
Thomas Huth548ef282015-02-24 21:29:25 +01007917 kvfree(free->arch.rmap[i]);
Takuya Yoshikawad89cc612012-08-01 18:03:28 +09007918 free->arch.rmap[i] = NULL;
Takuya Yoshikawa77d11302012-07-02 17:57:17 +09007919 }
Takuya Yoshikawad89cc612012-08-01 18:03:28 +09007920 if (i == 0)
7921 continue;
7922
7923 if (!dont || free->arch.lpage_info[i - 1] !=
7924 dont->arch.lpage_info[i - 1]) {
Thomas Huth548ef282015-02-24 21:29:25 +01007925 kvfree(free->arch.lpage_info[i - 1]);
Takuya Yoshikawad89cc612012-08-01 18:03:28 +09007926 free->arch.lpage_info[i - 1] = NULL;
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +09007927 }
7928 }
Xiao Guangrong21ebbed2016-02-24 17:51:09 +08007929
7930 kvm_page_track_free_memslot(free, dont);
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +09007931}
7932
Aneesh Kumar K.V55870272013-10-07 22:18:00 +05307933int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
7934 unsigned long npages)
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +09007935{
7936 int i;
7937
Takuya Yoshikawad89cc612012-08-01 18:03:28 +09007938 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
Xiao Guangrong92f94f12016-02-24 17:51:06 +08007939 struct kvm_lpage_info *linfo;
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +09007940 unsigned long ugfn;
7941 int lpages;
Takuya Yoshikawad89cc612012-08-01 18:03:28 +09007942 int level = i + 1;
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +09007943
7944 lpages = gfn_to_index(slot->base_gfn + npages - 1,
7945 slot->base_gfn, level) + 1;
7946
Takuya Yoshikawad89cc612012-08-01 18:03:28 +09007947 slot->arch.rmap[i] =
7948 kvm_kvzalloc(lpages * sizeof(*slot->arch.rmap[i]));
7949 if (!slot->arch.rmap[i])
Takuya Yoshikawa77d11302012-07-02 17:57:17 +09007950 goto out_free;
Takuya Yoshikawad89cc612012-08-01 18:03:28 +09007951 if (i == 0)
7952 continue;
Takuya Yoshikawa77d11302012-07-02 17:57:17 +09007953
Xiao Guangrong92f94f12016-02-24 17:51:06 +08007954 linfo = kvm_kvzalloc(lpages * sizeof(*linfo));
7955 if (!linfo)
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +09007956 goto out_free;
7957
Xiao Guangrong92f94f12016-02-24 17:51:06 +08007958 slot->arch.lpage_info[i - 1] = linfo;
7959
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +09007960 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
Xiao Guangrong92f94f12016-02-24 17:51:06 +08007961 linfo[0].disallow_lpage = 1;
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +09007962 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
Xiao Guangrong92f94f12016-02-24 17:51:06 +08007963 linfo[lpages - 1].disallow_lpage = 1;
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +09007964 ugfn = slot->userspace_addr >> PAGE_SHIFT;
7965 /*
7966 * If the gfn and userspace address are not aligned wrt each
7967 * other, or if explicitly asked to, disable large page
7968 * support for this slot
7969 */
7970 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
7971 !kvm_largepages_enabled()) {
7972 unsigned long j;
7973
7974 for (j = 0; j < lpages; ++j)
Xiao Guangrong92f94f12016-02-24 17:51:06 +08007975 linfo[j].disallow_lpage = 1;
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +09007976 }
7977 }
7978
Xiao Guangrong21ebbed2016-02-24 17:51:09 +08007979 if (kvm_page_track_create_memslot(slot, npages))
7980 goto out_free;
7981
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +09007982 return 0;
7983
7984out_free:
Takuya Yoshikawad89cc612012-08-01 18:03:28 +09007985 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
Thomas Huth548ef282015-02-24 21:29:25 +01007986 kvfree(slot->arch.rmap[i]);
Takuya Yoshikawad89cc612012-08-01 18:03:28 +09007987 slot->arch.rmap[i] = NULL;
7988 if (i == 0)
7989 continue;
7990
Thomas Huth548ef282015-02-24 21:29:25 +01007991 kvfree(slot->arch.lpage_info[i - 1]);
Takuya Yoshikawad89cc612012-08-01 18:03:28 +09007992 slot->arch.lpage_info[i - 1] = NULL;
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +09007993 }
7994 return -ENOMEM;
7995}
7996
Paolo Bonzini15f46012015-05-17 21:26:08 +02007997void kvm_arch_memslots_updated(struct kvm *kvm, struct kvm_memslots *slots)
Takuya Yoshikawae59dbe02013-07-04 13:40:29 +09007998{
Takuya Yoshikawae6dff7d2013-07-04 13:41:26 +09007999 /*
8000 * memslots->generation has been incremented.
8001 * mmio generation may have reached its maximum value.
8002 */
Paolo Bonzini54bf36a2015-04-08 15:39:23 +02008003 kvm_mmu_invalidate_mmio_sptes(kvm, slots);
Takuya Yoshikawae59dbe02013-07-04 13:40:29 +09008004}
8005
Marcelo Tosattif7784b82009-12-23 14:35:18 -02008006int kvm_arch_prepare_memory_region(struct kvm *kvm,
8007 struct kvm_memory_slot *memslot,
Paolo Bonzini09170a42015-05-18 13:59:39 +02008008 const struct kvm_userspace_memory_region *mem,
Takuya Yoshikawa7b6195a2013-02-27 19:44:34 +09008009 enum kvm_mr_change change)
Zhang Xiantao0de10342007-11-20 16:25:04 +08008010{
Marcelo Tosattif7784b82009-12-23 14:35:18 -02008011 return 0;
8012}
8013
Kai Huang88178fd2015-01-28 10:54:27 +08008014static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
8015 struct kvm_memory_slot *new)
8016{
8017 /* Still write protect RO slot */
8018 if (new->flags & KVM_MEM_READONLY) {
8019 kvm_mmu_slot_remove_write_access(kvm, new);
8020 return;
8021 }
8022
8023 /*
8024 * Call kvm_x86_ops dirty logging hooks when they are valid.
8025 *
8026 * kvm_x86_ops->slot_disable_log_dirty is called when:
8027 *
8028 * - KVM_MR_CREATE with dirty logging is disabled
8029 * - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
8030 *
8031 * The reason is, in case of PML, we need to set D-bit for any slots
8032 * with dirty logging disabled in order to eliminate unnecessary GPA
8033 * logging in PML buffer (and potential PML buffer full VMEXT). This
8034 * guarantees leaving PML enabled during guest's lifetime won't have
8035 * any additonal overhead from PML when guest is running with dirty
8036 * logging disabled for memory slots.
8037 *
8038 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
8039 * to dirty logging mode.
8040 *
8041 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
8042 *
8043 * In case of write protect:
8044 *
8045 * Write protect all pages for dirty logging.
8046 *
8047 * All the sptes including the large sptes which point to this
8048 * slot are set to readonly. We can not create any new large
8049 * spte on this slot until the end of the logging.
8050 *
8051 * See the comments in fast_page_fault().
8052 */
8053 if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
8054 if (kvm_x86_ops->slot_enable_log_dirty)
8055 kvm_x86_ops->slot_enable_log_dirty(kvm, new);
8056 else
8057 kvm_mmu_slot_remove_write_access(kvm, new);
8058 } else {
8059 if (kvm_x86_ops->slot_disable_log_dirty)
8060 kvm_x86_ops->slot_disable_log_dirty(kvm, new);
8061 }
8062}
8063
Marcelo Tosattif7784b82009-12-23 14:35:18 -02008064void kvm_arch_commit_memory_region(struct kvm *kvm,
Paolo Bonzini09170a42015-05-18 13:59:39 +02008065 const struct kvm_userspace_memory_region *mem,
Takuya Yoshikawa84826442013-02-27 19:45:25 +09008066 const struct kvm_memory_slot *old,
Paolo Bonzinif36f3f22015-05-18 13:20:23 +02008067 const struct kvm_memory_slot *new,
Takuya Yoshikawa84826442013-02-27 19:45:25 +09008068 enum kvm_mr_change change)
Marcelo Tosattif7784b82009-12-23 14:35:18 -02008069{
Takuya Yoshikawa84826442013-02-27 19:45:25 +09008070 int nr_mmu_pages = 0;
Marcelo Tosattif7784b82009-12-23 14:35:18 -02008071
Xiao Guangrong48c0e4e2011-03-04 18:59:21 +08008072 if (!kvm->arch.n_requested_mmu_pages)
8073 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
Zhang Xiantao0de10342007-11-20 16:25:04 +08008074
Xiao Guangrong48c0e4e2011-03-04 18:59:21 +08008075 if (nr_mmu_pages)
8076 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
Kai Huang1c91cad42015-01-28 10:54:26 +08008077
Takuya Yoshikawac972f3b2013-01-08 19:43:28 +09008078 /*
Wanpeng Li3ea3b7f2015-04-03 15:40:25 +08008079 * Dirty logging tracks sptes in 4k granularity, meaning that large
8080 * sptes have to be split. If live migration is successful, the guest
8081 * in the source machine will be destroyed and large sptes will be
8082 * created in the destination. However, if the guest continues to run
8083 * in the source machine (for example if live migration fails), small
8084 * sptes will remain around and cause bad performance.
8085 *
8086 * Scan sptes if dirty logging has been stopped, dropping those
8087 * which can be collapsed into a single large-page spte. Later
8088 * page faults will create the large-page sptes.
8089 */
8090 if ((change != KVM_MR_DELETE) &&
8091 (old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
8092 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
8093 kvm_mmu_zap_collapsible_sptes(kvm, new);
8094
8095 /*
Kai Huang88178fd2015-01-28 10:54:27 +08008096 * Set up write protection and/or dirty logging for the new slot.
Xiao Guangrongc126d942014-04-17 17:06:14 +08008097 *
Kai Huang88178fd2015-01-28 10:54:27 +08008098 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
8099 * been zapped so no dirty logging staff is needed for old slot. For
8100 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
8101 * new and it's also covered when dealing with the new slot.
Paolo Bonzinif36f3f22015-05-18 13:20:23 +02008102 *
8103 * FIXME: const-ify all uses of struct kvm_memory_slot.
Takuya Yoshikawac972f3b2013-01-08 19:43:28 +09008104 */
Kai Huang88178fd2015-01-28 10:54:27 +08008105 if (change != KVM_MR_DELETE)
Paolo Bonzinif36f3f22015-05-18 13:20:23 +02008106 kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
Zhang Xiantao0de10342007-11-20 16:25:04 +08008107}
Zhang Xiantao1d737c82007-12-14 09:35:10 +08008108
Marcelo Tosatti2df72e92012-08-24 15:54:57 -03008109void kvm_arch_flush_shadow_all(struct kvm *kvm)
Marcelo Tosatti34d4cb82008-07-10 20:49:31 -03008110{
Xiao Guangrong6ca18b62013-05-31 08:36:23 +08008111 kvm_mmu_invalidate_zap_all_pages(kvm);
Marcelo Tosatti34d4cb82008-07-10 20:49:31 -03008112}
8113
Marcelo Tosatti2df72e92012-08-24 15:54:57 -03008114void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
8115 struct kvm_memory_slot *slot)
8116{
Xiao Guangrong6ca18b62013-05-31 08:36:23 +08008117 kvm_mmu_invalidate_zap_all_pages(kvm);
Marcelo Tosatti2df72e92012-08-24 15:54:57 -03008118}
8119
Paolo Bonzini5d9bc642015-10-13 10:18:53 +02008120static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
8121{
8122 if (!list_empty_careful(&vcpu->async_pf.done))
8123 return true;
8124
8125 if (kvm_apic_has_events(vcpu))
8126 return true;
8127
8128 if (vcpu->arch.pv.pv_unhalted)
8129 return true;
8130
8131 if (atomic_read(&vcpu->arch.nmi_queued))
8132 return true;
8133
Paolo Bonzini73917732015-10-13 10:19:35 +02008134 if (test_bit(KVM_REQ_SMI, &vcpu->requests))
8135 return true;
8136
Paolo Bonzini5d9bc642015-10-13 10:18:53 +02008137 if (kvm_arch_interrupt_allowed(vcpu) &&
8138 kvm_cpu_has_interrupt(vcpu))
8139 return true;
8140
Andrey Smetanin1f4b34f2015-11-30 19:22:21 +03008141 if (kvm_hv_has_stimer_pending(vcpu))
8142 return true;
8143
Paolo Bonzini5d9bc642015-10-13 10:18:53 +02008144 return false;
8145}
8146
Zhang Xiantao1d737c82007-12-14 09:35:10 +08008147int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
8148{
Jan Kiszkab6b8a142014-03-07 20:03:12 +01008149 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
8150 kvm_x86_ops->check_nested_events(vcpu, false);
8151
Paolo Bonzini5d9bc642015-10-13 10:18:53 +02008152 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
Zhang Xiantao1d737c82007-12-14 09:35:10 +08008153}
Zhang Xiantao57361992007-12-17 14:21:40 +08008154
Christoffer Dallb6d33832012-03-08 16:44:24 -05008155int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
Zhang Xiantao57361992007-12-17 14:21:40 +08008156{
Christoffer Dallb6d33832012-03-08 16:44:24 -05008157 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
Zhang Xiantao57361992007-12-17 14:21:40 +08008158}
Gleb Natapov78646122009-03-23 12:12:11 +02008159
8160int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
8161{
8162 return kvm_x86_ops->interrupt_allowed(vcpu);
8163}
Marcelo Tosatti229456f2009-06-17 09:22:14 -03008164
Nadav Amit82b32772014-11-02 11:54:45 +02008165unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
8166{
8167 if (is_64_bit_mode(vcpu))
8168 return kvm_rip_read(vcpu);
8169 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
8170 kvm_rip_read(vcpu));
8171}
8172EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
8173
Jan Kiszkaf92653e2010-02-23 17:47:55 +01008174bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
8175{
Nadav Amit82b32772014-11-02 11:54:45 +02008176 return kvm_get_linear_rip(vcpu) == linear_rip;
Jan Kiszkaf92653e2010-02-23 17:47:55 +01008177}
8178EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
8179
Jan Kiszka94fe45d2009-10-18 13:24:44 +02008180unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
8181{
8182 unsigned long rflags;
8183
8184 rflags = kvm_x86_ops->get_rflags(vcpu);
8185 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
Jan Kiszkac310bac2010-02-23 17:47:58 +01008186 rflags &= ~X86_EFLAGS_TF;
Jan Kiszka94fe45d2009-10-18 13:24:44 +02008187 return rflags;
8188}
8189EXPORT_SYMBOL_GPL(kvm_get_rflags);
8190
Paolo Bonzini6addfc42014-03-27 11:29:28 +01008191static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
Jan Kiszka94fe45d2009-10-18 13:24:44 +02008192{
8193 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
Jan Kiszkaf92653e2010-02-23 17:47:55 +01008194 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
Jan Kiszkac310bac2010-02-23 17:47:58 +01008195 rflags |= X86_EFLAGS_TF;
Jan Kiszka94fe45d2009-10-18 13:24:44 +02008196 kvm_x86_ops->set_rflags(vcpu, rflags);
Paolo Bonzini6addfc42014-03-27 11:29:28 +01008197}
8198
8199void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
8200{
8201 __kvm_set_rflags(vcpu, rflags);
Avi Kivity3842d132010-07-27 12:30:24 +03008202 kvm_make_request(KVM_REQ_EVENT, vcpu);
Jan Kiszka94fe45d2009-10-18 13:24:44 +02008203}
8204EXPORT_SYMBOL_GPL(kvm_set_rflags);
8205
Gleb Natapov56028d02010-10-17 18:13:42 +02008206void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
8207{
8208 int r;
8209
Xiao Guangrongfb67e142010-12-07 10:35:25 +08008210 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
chai wenf2e10662013-10-14 22:22:33 +08008211 work->wakeup_all)
Gleb Natapov56028d02010-10-17 18:13:42 +02008212 return;
8213
8214 r = kvm_mmu_reload(vcpu);
8215 if (unlikely(r))
8216 return;
8217
Xiao Guangrongfb67e142010-12-07 10:35:25 +08008218 if (!vcpu->arch.mmu.direct_map &&
8219 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
8220 return;
8221
Gleb Natapov56028d02010-10-17 18:13:42 +02008222 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
8223}
8224
Gleb Natapovaf585b92010-10-14 11:22:46 +02008225static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
8226{
8227 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
8228}
8229
8230static inline u32 kvm_async_pf_next_probe(u32 key)
8231{
8232 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
8233}
8234
8235static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8236{
8237 u32 key = kvm_async_pf_hash_fn(gfn);
8238
8239 while (vcpu->arch.apf.gfns[key] != ~0)
8240 key = kvm_async_pf_next_probe(key);
8241
8242 vcpu->arch.apf.gfns[key] = gfn;
8243}
8244
8245static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
8246{
8247 int i;
8248 u32 key = kvm_async_pf_hash_fn(gfn);
8249
8250 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
Xiao Guangrongc7d28c22010-11-01 17:00:30 +08008251 (vcpu->arch.apf.gfns[key] != gfn &&
8252 vcpu->arch.apf.gfns[key] != ~0); i++)
Gleb Natapovaf585b92010-10-14 11:22:46 +02008253 key = kvm_async_pf_next_probe(key);
8254
8255 return key;
8256}
8257
8258bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8259{
8260 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
8261}
8262
8263static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8264{
8265 u32 i, j, k;
8266
8267 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
8268 while (true) {
8269 vcpu->arch.apf.gfns[i] = ~0;
8270 do {
8271 j = kvm_async_pf_next_probe(j);
8272 if (vcpu->arch.apf.gfns[j] == ~0)
8273 return;
8274 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
8275 /*
8276 * k lies cyclically in ]i,j]
8277 * | i.k.j |
8278 * |....j i.k.| or |.k..j i...|
8279 */
8280 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
8281 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
8282 i = j;
8283 }
8284}
8285
Gleb Natapov7c907052010-10-14 11:22:53 +02008286static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
8287{
8288
8289 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
8290 sizeof(val));
8291}
8292
Gleb Natapovaf585b92010-10-14 11:22:46 +02008293void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
8294 struct kvm_async_pf *work)
8295{
Avi Kivity6389ee92010-11-29 16:12:30 +02008296 struct x86_exception fault;
8297
Gleb Natapov7c907052010-10-14 11:22:53 +02008298 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
Gleb Natapovaf585b92010-10-14 11:22:46 +02008299 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
Gleb Natapov7c907052010-10-14 11:22:53 +02008300
8301 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
Gleb Natapovfc5f06f2010-10-14 11:22:56 +02008302 (vcpu->arch.apf.send_user_only &&
8303 kvm_x86_ops->get_cpl(vcpu) == 0))
Gleb Natapov7c907052010-10-14 11:22:53 +02008304 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
8305 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
Avi Kivity6389ee92010-11-29 16:12:30 +02008306 fault.vector = PF_VECTOR;
8307 fault.error_code_valid = true;
8308 fault.error_code = 0;
8309 fault.nested_page_fault = false;
8310 fault.address = work->arch.token;
8311 kvm_inject_page_fault(vcpu, &fault);
Gleb Natapov7c907052010-10-14 11:22:53 +02008312 }
Gleb Natapovaf585b92010-10-14 11:22:46 +02008313}
8314
8315void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
8316 struct kvm_async_pf *work)
8317{
Avi Kivity6389ee92010-11-29 16:12:30 +02008318 struct x86_exception fault;
8319
Gleb Natapov7c907052010-10-14 11:22:53 +02008320 trace_kvm_async_pf_ready(work->arch.token, work->gva);
chai wenf2e10662013-10-14 22:22:33 +08008321 if (work->wakeup_all)
Gleb Natapov7c907052010-10-14 11:22:53 +02008322 work->arch.token = ~0; /* broadcast wakeup */
8323 else
8324 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
8325
8326 if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
8327 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
Avi Kivity6389ee92010-11-29 16:12:30 +02008328 fault.vector = PF_VECTOR;
8329 fault.error_code_valid = true;
8330 fault.error_code = 0;
8331 fault.nested_page_fault = false;
8332 fault.address = work->arch.token;
8333 kvm_inject_page_fault(vcpu, &fault);
Gleb Natapov7c907052010-10-14 11:22:53 +02008334 }
Xiao Guangronge6d53e32010-11-01 17:01:28 +08008335 vcpu->arch.apf.halted = false;
Gleb Natapova4fa1632012-05-03 11:36:39 +03008336 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
Gleb Natapov7c907052010-10-14 11:22:53 +02008337}
8338
8339bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
8340{
8341 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
8342 return true;
8343 else
8344 return !kvm_event_needs_reinjection(vcpu) &&
8345 kvm_x86_ops->interrupt_allowed(vcpu);
Gleb Natapovaf585b92010-10-14 11:22:46 +02008346}
8347
Paolo Bonzini5544eb92015-07-07 15:41:58 +02008348void kvm_arch_start_assignment(struct kvm *kvm)
8349{
8350 atomic_inc(&kvm->arch.assigned_device_count);
8351}
8352EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
8353
8354void kvm_arch_end_assignment(struct kvm *kvm)
8355{
8356 atomic_dec(&kvm->arch.assigned_device_count);
8357}
8358EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
8359
8360bool kvm_arch_has_assigned_device(struct kvm *kvm)
8361{
8362 return atomic_read(&kvm->arch.assigned_device_count);
8363}
8364EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
8365
Alex Williamsone0f0bbc2013-10-30 11:02:30 -06008366void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
8367{
8368 atomic_inc(&kvm->arch.noncoherent_dma_count);
8369}
8370EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
8371
8372void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
8373{
8374 atomic_dec(&kvm->arch.noncoherent_dma_count);
8375}
8376EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
8377
8378bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
8379{
8380 return atomic_read(&kvm->arch.noncoherent_dma_count);
8381}
8382EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
8383
Alex Williamson14717e22016-05-05 11:58:35 -06008384bool kvm_arch_has_irq_bypass(void)
8385{
8386 return kvm_x86_ops->update_pi_irte != NULL;
8387}
8388
Feng Wu87276882015-09-18 22:29:40 +08008389int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
8390 struct irq_bypass_producer *prod)
8391{
8392 struct kvm_kernel_irqfd *irqfd =
8393 container_of(cons, struct kvm_kernel_irqfd, consumer);
8394
Alex Williamson14717e22016-05-05 11:58:35 -06008395 irqfd->producer = prod;
Feng Wu87276882015-09-18 22:29:40 +08008396
Alex Williamson14717e22016-05-05 11:58:35 -06008397 return kvm_x86_ops->update_pi_irte(irqfd->kvm,
8398 prod->irq, irqfd->gsi, 1);
Feng Wu87276882015-09-18 22:29:40 +08008399}
8400
8401void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
8402 struct irq_bypass_producer *prod)
8403{
8404 int ret;
8405 struct kvm_kernel_irqfd *irqfd =
8406 container_of(cons, struct kvm_kernel_irqfd, consumer);
8407
Feng Wu87276882015-09-18 22:29:40 +08008408 WARN_ON(irqfd->producer != prod);
8409 irqfd->producer = NULL;
8410
8411 /*
8412 * When producer of consumer is unregistered, we change back to
8413 * remapped mode, so we can re-use the current implementation
Andrea Gelminibb3541f2016-05-21 14:14:44 +02008414 * when the irq is masked/disabled or the consumer side (KVM
Feng Wu87276882015-09-18 22:29:40 +08008415 * int this case doesn't want to receive the interrupts.
8416 */
8417 ret = kvm_x86_ops->update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
8418 if (ret)
8419 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
8420 " fails: %d\n", irqfd->consumer.token, ret);
8421}
8422
8423int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
8424 uint32_t guest_irq, bool set)
8425{
8426 if (!kvm_x86_ops->update_pi_irte)
8427 return -EINVAL;
8428
8429 return kvm_x86_ops->update_pi_irte(kvm, host_irq, guest_irq, set);
8430}
8431
Feng Wu520040142016-01-25 16:53:33 +08008432bool kvm_vector_hashing_enabled(void)
8433{
8434 return vector_hashing;
8435}
8436EXPORT_SYMBOL_GPL(kvm_vector_hashing_enabled);
8437
Marcelo Tosatti229456f2009-06-17 09:22:14 -03008438EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
Jason Wang931c33b2015-09-15 14:41:58 +08008439EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
Marcelo Tosatti229456f2009-06-17 09:22:14 -03008440EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
8441EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
8442EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
8443EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
Joerg Roedel0ac406d2009-10-09 16:08:27 +02008444EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
Joerg Roedeld8cabdd2009-10-09 16:08:28 +02008445EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
Joerg Roedel17897f32009-10-09 16:08:29 +02008446EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
Joerg Roedel236649d2009-10-09 16:08:30 +02008447EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
Joerg Roedelec1ff792009-10-09 16:08:31 +02008448EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
Joerg Roedel532a46b2009-10-09 16:08:32 +02008449EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
Joerg Roedel2e554e82010-02-24 18:59:14 +01008450EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
Yoshihiro YUNOMAE489223e2013-06-12 16:43:44 +09008451EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
Radim Krčmář7b462682014-08-21 18:08:09 +02008452EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window);
Kai Huang843e4332015-01-28 10:54:28 +08008453EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
Feng Wuefc64402015-09-18 22:29:51 +08008454EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
Suravee Suthikulpanit18f40c52016-05-04 14:09:48 -05008455EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
8456EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);