blob: 1cec2c62a0b08405d2bd7c8908d6b7f33de3b63c [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"
Wei Huang474a5bb2015-06-19 13:54:23 +020030#include "pmu.h"
Andrey Smetanine83d5882015-07-03 15:01:34 +030031#include "hyperv.h"
Carsten Otte313a3dc2007-10-11 19:16:52 +020032
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -020033#include <linux/clocksource.h>
Ben-Ami Yassour4d5c5d02008-07-28 19:26:26 +030034#include <linux/interrupt.h>
Carsten Otte313a3dc2007-10-11 19:16:52 +020035#include <linux/kvm.h>
36#include <linux/fs.h>
37#include <linux/vmalloc.h>
Paul Gortmaker1767e932016-07-13 20:19:00 -040038#include <linux/export.h>
39#include <linux/moduleparam.h>
Zhang Xiantao0de10342007-11-20 16:25:04 +080040#include <linux/mman.h>
Marcelo Tosatti2bacc552007-12-12 10:46:12 -050041#include <linux/highmem.h>
Joerg Roedel19de40a2008-12-03 14:43:34 +010042#include <linux/iommu.h>
Ben-Ami Yassour62c476c2008-09-14 03:48:28 +030043#include <linux/intel-iommu.h>
Gerd Hoffmannc8076602009-02-04 17:52:04 +010044#include <linux/cpufreq.h>
Avi Kivity18863bd2009-09-07 11:12:18 +030045#include <linux/user-return-notifier.h>
Marcelo Tosattia983fb22009-12-23 14:35:23 -020046#include <linux/srcu.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090047#include <linux/slab.h>
Zhang, Yanminff9d07a2010-04-19 13:32:45 +080048#include <linux/perf_event.h>
Lai Jiangshan7bee3422010-06-02 17:06:03 +080049#include <linux/uaccess.h>
Gleb Natapovaf585b92010-10-14 11:22:46 +020050#include <linux/hash.h>
Joerg Roedela1b60c12011-09-06 18:46:34 +020051#include <linux/pci.h>
Marcelo Tosatti16e8d742012-11-27 23:29:00 -020052#include <linux/timekeeper_internal.h>
53#include <linux/pvclock_gtod.h>
Feng Wu87276882015-09-18 22:29:40 +080054#include <linux/kvm_irqfd.h>
55#include <linux/irqbypass.h>
Ingo Molnar3905f9a2017-02-05 12:07:04 +010056#include <linux/sched/stat.h>
Tom Lendackyd0ec49d2017-07-17 16:10:27 -050057#include <linux/mem_encrypt.h>
Ingo Molnar3905f9a2017-02-05 12:07:04 +010058
Avi Kivityaec51dc2009-07-01 16:01:02 +030059#include <trace/events/kvm.h>
Xiao Guangrong2ed152a2010-03-10 19:00:43 +080060
Frederic Weisbecker24f1e32c2009-09-09 19:22:48 +020061#include <asm/debugreg.h>
Zhang Xiantaod825ed02007-11-14 20:08:51 +080062#include <asm/msr.h>
Avi Kivitya5f61302008-02-20 17:57:21 +020063#include <asm/desc.h>
Huang Ying890ca9a2009-05-11 16:48:15 +080064#include <asm/mce.h>
Ingo Molnarf89e32e2015-04-22 10:58:10 +020065#include <linux/kernel_stat.h>
Ingo Molnar78f7f1e2015-04-24 02:54:44 +020066#include <asm/fpu/internal.h> /* Ugh! */
Zachary Amsden1d5f0662010-08-19 22:07:30 -100067#include <asm/pvclock.h>
Avi Kivity217fc9c2010-08-26 13:38:03 +030068#include <asm/div64.h>
Feng Wuefc64402015-09-18 22:29:51 +080069#include <asm/irq_remapping.h>
Carsten Otte043405e2007-10-10 17:16:19 +020070
Dave Hansend1898b72016-06-01 10:42:20 -070071#define CREATE_TRACE_POINTS
72#include "trace.h"
73
Carsten Otte313a3dc2007-10-11 19:16:52 +020074#define MAX_IO_MSRS 256
Huang Ying890ca9a2009-05-11 16:48:15 +080075#define KVM_MAX_MCE_BANKS 32
Ashok Rajc45dcc72016-06-22 14:59:56 +080076u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
77EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
Huang Ying890ca9a2009-05-11 16:48:15 +080078
Avi Kivity0f65dd72011-04-20 13:37:53 +030079#define emul_to_vcpu(ctxt) \
80 container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
81
Joerg Roedel50a37eb2008-01-31 14:57:38 +010082/* EFER defaults:
83 * - enable syscall per default because its emulated by KVM
84 * - enable LME and LMA per default on 64 bit KVM
85 */
86#ifdef CONFIG_X86_64
Lai Jiangshan1260edbe2011-02-21 11:51:35 +080087static
88u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
Joerg Roedel50a37eb2008-01-31 14:57:38 +010089#else
Lai Jiangshan1260edbe2011-02-21 11:51:35 +080090static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
Joerg Roedel50a37eb2008-01-31 14:57:38 +010091#endif
Carsten Otte313a3dc2007-10-11 19:16:52 +020092
Avi Kivityba1389b2007-11-18 16:24:12 +020093#define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
94#define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
Hollis Blanchard417bc302007-10-31 17:24:23 -050095
Radim Krčmářc5192652016-07-12 22:09:28 +020096#define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
97 KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
Radim Krčmář371313132016-07-12 22:09:27 +020098
Gleb Natapovcb142eb2009-08-09 15:17:40 +030099static void update_cr8_intercept(struct kvm_vcpu *vcpu);
Avi Kivity7460fb4a2011-09-20 13:43:14 +0300100static void process_nmi(struct kvm_vcpu *vcpu);
Paolo Bonziniee2cd4b2016-06-01 22:26:01 +0200101static void enter_smm(struct kvm_vcpu *vcpu);
Paolo Bonzini6addfc42014-03-27 11:29:28 +0100102static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
Avi Kivity674eea02008-02-11 18:37:23 +0200103
Paolo Bonzini893590c2015-11-06 11:46:24 +0100104struct kvm_x86_ops *kvm_x86_ops __read_mostly;
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -0300105EXPORT_SYMBOL_GPL(kvm_x86_ops);
Zhang Xiantao97896d02007-11-14 20:09:30 +0800106
Paolo Bonzini893590c2015-11-06 11:46:24 +0100107static bool __read_mostly ignore_msrs = 0;
Rusty Russell476bc002012-01-13 09:32:18 +1030108module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
Andre Przywaraed85c062009-06-25 12:36:49 +0200109
Eyal Moscovicifab0aa32017-11-08 14:32:08 +0200110static bool __read_mostly report_ignored_msrs = true;
111module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR);
112
Marcelo Tosatti9ed96e82014-01-06 12:00:02 -0200113unsigned int min_timer_period_us = 500;
114module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
115
Marcelo Tosatti630994b2015-05-12 22:42:04 -0300116static bool __read_mostly kvmclock_periodic_sync = true;
117module_param(kvmclock_periodic_sync, bool, S_IRUGO);
118
Paolo Bonzini893590c2015-11-06 11:46:24 +0100119bool __read_mostly kvm_has_tsc_control;
Joerg Roedel92a1f122011-03-25 09:44:51 +0100120EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
Paolo Bonzini893590c2015-11-06 11:46:24 +0100121u32 __read_mostly kvm_max_guest_tsc_khz;
Joerg Roedel92a1f122011-03-25 09:44:51 +0100122EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
Haozhong Zhangbc9b9612015-10-20 15:39:01 +0800123u8 __read_mostly kvm_tsc_scaling_ratio_frac_bits;
124EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
125u64 __read_mostly kvm_max_tsc_scaling_ratio;
126EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
Yunhong Jiang64672c92016-06-13 14:19:59 -0700127u64 __read_mostly kvm_default_tsc_scaling_ratio;
128EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
Joerg Roedel92a1f122011-03-25 09:44:51 +0100129
Zachary Amsdencc578282012-02-03 15:43:50 -0200130/* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
Paolo Bonzini893590c2015-11-06 11:46:24 +0100131static u32 __read_mostly tsc_tolerance_ppm = 250;
Zachary Amsdencc578282012-02-03 15:43:50 -0200132module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
133
Marcelo Tosattid0659d92014-12-16 09:08:15 -0500134/* lapic timer advance (tscdeadline mode only) in nanoseconds */
Paolo Bonzini893590c2015-11-06 11:46:24 +0100135unsigned int __read_mostly lapic_timer_advance_ns = 0;
Marcelo Tosattid0659d92014-12-16 09:08:15 -0500136module_param(lapic_timer_advance_ns, uint, S_IRUGO | S_IWUSR);
137
Feng Wu520040142016-01-25 16:53:33 +0800138static bool __read_mostly vector_hashing = true;
139module_param(vector_hashing, bool, S_IRUGO);
140
Avi Kivity18863bd2009-09-07 11:12:18 +0300141#define KVM_NR_SHARED_MSRS 16
142
143struct kvm_shared_msrs_global {
144 int nr;
Sheng Yang2bf78fa2009-12-18 16:48:44 +0800145 u32 msrs[KVM_NR_SHARED_MSRS];
Avi Kivity18863bd2009-09-07 11:12:18 +0300146};
147
148struct kvm_shared_msrs {
149 struct user_return_notifier urn;
150 bool registered;
Sheng Yang2bf78fa2009-12-18 16:48:44 +0800151 struct kvm_shared_msr_values {
152 u64 host;
153 u64 curr;
154 } values[KVM_NR_SHARED_MSRS];
Avi Kivity18863bd2009-09-07 11:12:18 +0300155};
156
157static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
Marcelo Tosatti013f6a52013-01-03 11:41:39 -0200158static struct kvm_shared_msrs __percpu *shared_msrs;
Avi Kivity18863bd2009-09-07 11:12:18 +0300159
Hollis Blanchard417bc302007-10-31 17:24:23 -0500160struct kvm_stats_debugfs_item debugfs_entries[] = {
Avi Kivityba1389b2007-11-18 16:24:12 +0200161 { "pf_fixed", VCPU_STAT(pf_fixed) },
162 { "pf_guest", VCPU_STAT(pf_guest) },
163 { "tlb_flush", VCPU_STAT(tlb_flush) },
164 { "invlpg", VCPU_STAT(invlpg) },
165 { "exits", VCPU_STAT(exits) },
166 { "io_exits", VCPU_STAT(io_exits) },
167 { "mmio_exits", VCPU_STAT(mmio_exits) },
168 { "signal_exits", VCPU_STAT(signal_exits) },
169 { "irq_window", VCPU_STAT(irq_window_exits) },
Sheng Yangf08864b2008-05-15 18:23:25 +0800170 { "nmi_window", VCPU_STAT(nmi_window_exits) },
Avi Kivityba1389b2007-11-18 16:24:12 +0200171 { "halt_exits", VCPU_STAT(halt_exits) },
Paolo Bonzinif7819512015-02-04 18:20:58 +0100172 { "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
Paolo Bonzini62bea5b2015-09-15 18:27:57 +0200173 { "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
Christian Borntraeger3491caf2016-05-13 12:16:35 +0200174 { "halt_poll_invalid", VCPU_STAT(halt_poll_invalid) },
Avi Kivityba1389b2007-11-18 16:24:12 +0200175 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
Amit Shahf11c3a82008-02-21 01:00:30 +0530176 { "hypercalls", VCPU_STAT(hypercalls) },
Avi Kivityba1389b2007-11-18 16:24:12 +0200177 { "request_irq", VCPU_STAT(request_irq_exits) },
178 { "irq_exits", VCPU_STAT(irq_exits) },
179 { "host_state_reload", VCPU_STAT(host_state_reload) },
180 { "efer_reload", VCPU_STAT(efer_reload) },
181 { "fpu_reload", VCPU_STAT(fpu_reload) },
182 { "insn_emulation", VCPU_STAT(insn_emulation) },
183 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
Avi Kivityfa89a812008-09-01 15:57:51 +0300184 { "irq_injections", VCPU_STAT(irq_injections) },
Jan Kiszkac4abb7c2008-09-26 09:30:55 +0200185 { "nmi_injections", VCPU_STAT(nmi_injections) },
Paolo Bonzini0f1e2612016-12-17 16:05:19 +0100186 { "req_event", VCPU_STAT(req_event) },
Avi Kivity4cee5762007-11-18 16:37:07 +0200187 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
188 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
189 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
190 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
191 { "mmu_flooded", VM_STAT(mmu_flooded) },
192 { "mmu_recycled", VM_STAT(mmu_recycled) },
Avi Kivitydfc5aa02007-12-18 19:47:18 +0200193 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
Marcelo Tosatti4731d4c2008-09-23 13:18:39 -0300194 { "mmu_unsync", VM_STAT(mmu_unsync) },
Avi Kivity0f74a242007-11-20 23:01:14 +0200195 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
Marcelo Tosatti05da4552008-02-23 11:44:30 -0300196 { "largepages", VM_STAT(lpages) },
David Matlackf3414bc2016-12-20 15:25:57 -0800197 { "max_mmu_page_hash_collisions",
198 VM_STAT(max_mmu_page_hash_collisions) },
Hollis Blanchard417bc302007-10-31 17:24:23 -0500199 { NULL }
200};
201
Dexuan Cui2acf9232010-06-10 11:27:12 +0800202u64 __read_mostly host_xcr0;
203
Jan Kiszkab6785de2012-09-20 07:43:17 +0200204static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
Avi Kivityd6aa1002011-04-20 15:47:13 +0300205
Gleb Natapovaf585b92010-10-14 11:22:46 +0200206static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
207{
208 int i;
209 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
210 vcpu->arch.apf.gfns[i] = ~0;
211}
212
Avi Kivity18863bd2009-09-07 11:12:18 +0300213static void kvm_on_user_return(struct user_return_notifier *urn)
214{
215 unsigned slot;
Avi Kivity18863bd2009-09-07 11:12:18 +0300216 struct kvm_shared_msrs *locals
217 = container_of(urn, struct kvm_shared_msrs, urn);
Sheng Yang2bf78fa2009-12-18 16:48:44 +0800218 struct kvm_shared_msr_values *values;
Ignacio Alvarado1650b4e2016-11-04 12:15:55 -0700219 unsigned long flags;
Avi Kivity18863bd2009-09-07 11:12:18 +0300220
Ignacio Alvarado1650b4e2016-11-04 12:15:55 -0700221 /*
222 * Disabling irqs at this point since the following code could be
223 * interrupted and executed through kvm_arch_hardware_disable()
224 */
225 local_irq_save(flags);
226 if (locals->registered) {
227 locals->registered = false;
228 user_return_notifier_unregister(urn);
229 }
230 local_irq_restore(flags);
Avi Kivity18863bd2009-09-07 11:12:18 +0300231 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
Sheng Yang2bf78fa2009-12-18 16:48:44 +0800232 values = &locals->values[slot];
233 if (values->host != values->curr) {
234 wrmsrl(shared_msrs_global.msrs[slot], values->host);
235 values->curr = values->host;
Avi Kivity18863bd2009-09-07 11:12:18 +0300236 }
237 }
Avi Kivity18863bd2009-09-07 11:12:18 +0300238}
239
Sheng Yang2bf78fa2009-12-18 16:48:44 +0800240static void shared_msr_update(unsigned slot, u32 msr)
Avi Kivity18863bd2009-09-07 11:12:18 +0300241{
Avi Kivity18863bd2009-09-07 11:12:18 +0300242 u64 value;
Marcelo Tosatti013f6a52013-01-03 11:41:39 -0200243 unsigned int cpu = smp_processor_id();
244 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
Avi Kivity18863bd2009-09-07 11:12:18 +0300245
Sheng Yang2bf78fa2009-12-18 16:48:44 +0800246 /* only read, and nobody should modify it at this time,
247 * so don't need lock */
248 if (slot >= shared_msrs_global.nr) {
249 printk(KERN_ERR "kvm: invalid MSR slot!");
250 return;
251 }
252 rdmsrl_safe(msr, &value);
253 smsr->values[slot].host = value;
254 smsr->values[slot].curr = value;
255}
256
257void kvm_define_shared_msr(unsigned slot, u32 msr)
258{
Nadav Amit0123be42014-07-24 15:06:56 +0300259 BUG_ON(slot >= KVM_NR_SHARED_MSRS);
Paolo Bonzinic847fe82015-07-29 11:06:34 +0200260 shared_msrs_global.msrs[slot] = msr;
Avi Kivity18863bd2009-09-07 11:12:18 +0300261 if (slot >= shared_msrs_global.nr)
262 shared_msrs_global.nr = slot + 1;
Avi Kivity18863bd2009-09-07 11:12:18 +0300263}
264EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
265
266static void kvm_shared_msr_cpu_online(void)
267{
268 unsigned i;
Avi Kivity18863bd2009-09-07 11:12:18 +0300269
270 for (i = 0; i < shared_msrs_global.nr; ++i)
Sheng Yang2bf78fa2009-12-18 16:48:44 +0800271 shared_msr_update(i, shared_msrs_global.msrs[i]);
Avi Kivity18863bd2009-09-07 11:12:18 +0300272}
273
Andy Honig8b3c3102014-08-27 11:16:44 -0700274int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
Avi Kivity18863bd2009-09-07 11:12:18 +0300275{
Marcelo Tosatti013f6a52013-01-03 11:41:39 -0200276 unsigned int cpu = smp_processor_id();
277 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
Andy Honig8b3c3102014-08-27 11:16:44 -0700278 int err;
Avi Kivity18863bd2009-09-07 11:12:18 +0300279
Sheng Yang2bf78fa2009-12-18 16:48:44 +0800280 if (((value ^ smsr->values[slot].curr) & mask) == 0)
Andy Honig8b3c3102014-08-27 11:16:44 -0700281 return 0;
Sheng Yang2bf78fa2009-12-18 16:48:44 +0800282 smsr->values[slot].curr = value;
Andy Honig8b3c3102014-08-27 11:16:44 -0700283 err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
284 if (err)
285 return 1;
286
Avi Kivity18863bd2009-09-07 11:12:18 +0300287 if (!smsr->registered) {
288 smsr->urn.on_user_return = kvm_on_user_return;
289 user_return_notifier_register(&smsr->urn);
290 smsr->registered = true;
291 }
Andy Honig8b3c3102014-08-27 11:16:44 -0700292 return 0;
Avi Kivity18863bd2009-09-07 11:12:18 +0300293}
294EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
295
Radim Krčmář13a34e02014-08-28 15:13:03 +0200296static void drop_user_return_notifiers(void)
Avi Kivity3548bab2009-11-28 14:18:47 +0200297{
Marcelo Tosatti013f6a52013-01-03 11:41:39 -0200298 unsigned int cpu = smp_processor_id();
299 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
Avi Kivity3548bab2009-11-28 14:18:47 +0200300
301 if (smsr->registered)
302 kvm_on_user_return(&smsr->urn);
303}
304
Carsten Otte6866b832007-10-29 16:09:10 +0100305u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
306{
Gleb Natapov8a5a87d2012-08-05 15:58:26 +0300307 return vcpu->arch.apic_base;
Carsten Otte6866b832007-10-29 16:09:10 +0100308}
309EXPORT_SYMBOL_GPL(kvm_get_apic_base);
310
Jan Kiszka58cb6282014-01-24 16:48:44 +0100311int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
Carsten Otte6866b832007-10-29 16:09:10 +0100312{
Jan Kiszka58cb6282014-01-24 16:48:44 +0100313 u64 old_state = vcpu->arch.apic_base &
314 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
315 u64 new_state = msr_info->data &
316 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
Radim Krčmářd6321d42017-08-05 00:12:49 +0200317 u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) | 0x2ff |
318 (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
Jan Kiszka58cb6282014-01-24 16:48:44 +0100319
Jim Mattsond3802282017-08-10 10:14:13 -0700320 if ((msr_info->data & reserved_bits) || new_state == X2APIC_ENABLE)
321 return 1;
Jan Kiszka58cb6282014-01-24 16:48:44 +0100322 if (!msr_info->host_initiated &&
Jim Mattsond3802282017-08-10 10:14:13 -0700323 ((new_state == MSR_IA32_APICBASE_ENABLE &&
Jan Kiszka58cb6282014-01-24 16:48:44 +0100324 old_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE)) ||
325 (new_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE) &&
326 old_state == 0)))
327 return 1;
328
329 kvm_lapic_set_base(vcpu, msr_info->data);
330 return 0;
Carsten Otte6866b832007-10-29 16:09:10 +0100331}
332EXPORT_SYMBOL_GPL(kvm_set_apic_base);
333
Andi Kleen2605fc22014-05-02 00:44:37 +0200334asmlinkage __visible void kvm_spurious_fault(void)
Geoff Levande3ba45b2013-04-05 19:20:30 +0000335{
336 /* Fault while not rebooting. We want the trace. */
337 BUG();
338}
339EXPORT_SYMBOL_GPL(kvm_spurious_fault);
340
Eddie Dong3fd28fc2009-11-19 17:54:07 +0200341#define EXCPT_BENIGN 0
342#define EXCPT_CONTRIBUTORY 1
343#define EXCPT_PF 2
344
345static int exception_class(int vector)
346{
347 switch (vector) {
348 case PF_VECTOR:
349 return EXCPT_PF;
350 case DE_VECTOR:
351 case TS_VECTOR:
352 case NP_VECTOR:
353 case SS_VECTOR:
354 case GP_VECTOR:
355 return EXCPT_CONTRIBUTORY;
356 default:
357 break;
358 }
359 return EXCPT_BENIGN;
360}
361
Nadav Amitd6e8c852014-07-24 14:51:24 +0300362#define EXCPT_FAULT 0
363#define EXCPT_TRAP 1
364#define EXCPT_ABORT 2
365#define EXCPT_INTERRUPT 3
366
367static int exception_type(int vector)
368{
369 unsigned int mask;
370
371 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
372 return EXCPT_INTERRUPT;
373
374 mask = 1 << vector;
375
376 /* #DB is trap, as instruction watchpoints are handled elsewhere */
377 if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
378 return EXCPT_TRAP;
379
380 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
381 return EXCPT_ABORT;
382
383 /* Reserved exceptions will result in fault */
384 return EXCPT_FAULT;
385}
386
Eddie Dong3fd28fc2009-11-19 17:54:07 +0200387static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
Joerg Roedelce7ddec2010-04-22 12:33:13 +0200388 unsigned nr, bool has_error, u32 error_code,
389 bool reinject)
Eddie Dong3fd28fc2009-11-19 17:54:07 +0200390{
391 u32 prev_nr;
392 int class1, class2;
393
Avi Kivity3842d132010-07-27 12:30:24 +0300394 kvm_make_request(KVM_REQ_EVENT, vcpu);
395
Wanpeng Li664f8e22017-08-24 03:35:09 -0700396 if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
Eddie Dong3fd28fc2009-11-19 17:54:07 +0200397 queue:
Nadav Amit3ffb2462014-11-02 11:54:42 +0200398 if (has_error && !is_protmode(vcpu))
399 has_error = false;
Wanpeng Li664f8e22017-08-24 03:35:09 -0700400 if (reinject) {
401 /*
402 * On vmentry, vcpu->arch.exception.pending is only
403 * true if an event injection was blocked by
404 * nested_run_pending. In that case, however,
405 * vcpu_enter_guest requests an immediate exit,
406 * and the guest shouldn't proceed far enough to
407 * need reinjection.
408 */
409 WARN_ON_ONCE(vcpu->arch.exception.pending);
410 vcpu->arch.exception.injected = true;
411 } else {
412 vcpu->arch.exception.pending = true;
413 vcpu->arch.exception.injected = false;
414 }
Eddie Dong3fd28fc2009-11-19 17:54:07 +0200415 vcpu->arch.exception.has_error_code = has_error;
416 vcpu->arch.exception.nr = nr;
417 vcpu->arch.exception.error_code = error_code;
418 return;
419 }
420
421 /* to check exception */
422 prev_nr = vcpu->arch.exception.nr;
423 if (prev_nr == DF_VECTOR) {
424 /* triple fault -> shutdown */
Avi Kivitya8eeb042010-05-10 12:34:53 +0300425 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
Eddie Dong3fd28fc2009-11-19 17:54:07 +0200426 return;
427 }
428 class1 = exception_class(prev_nr);
429 class2 = exception_class(nr);
430 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
431 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
Wanpeng Li664f8e22017-08-24 03:35:09 -0700432 /*
433 * Generate double fault per SDM Table 5-5. Set
434 * exception.pending = true so that the double fault
435 * can trigger a nested vmexit.
436 */
Eddie Dong3fd28fc2009-11-19 17:54:07 +0200437 vcpu->arch.exception.pending = true;
Wanpeng Li664f8e22017-08-24 03:35:09 -0700438 vcpu->arch.exception.injected = false;
Eddie Dong3fd28fc2009-11-19 17:54:07 +0200439 vcpu->arch.exception.has_error_code = true;
440 vcpu->arch.exception.nr = DF_VECTOR;
441 vcpu->arch.exception.error_code = 0;
442 } else
443 /* replace previous exception with a new one in a hope
444 that instruction re-execution will regenerate lost
445 exception */
446 goto queue;
447}
448
Avi Kivity298101d2007-11-25 13:41:11 +0200449void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
450{
Joerg Roedelce7ddec2010-04-22 12:33:13 +0200451 kvm_multiple_exception(vcpu, nr, false, 0, false);
Avi Kivity298101d2007-11-25 13:41:11 +0200452}
453EXPORT_SYMBOL_GPL(kvm_queue_exception);
454
Joerg Roedelce7ddec2010-04-22 12:33:13 +0200455void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
456{
457 kvm_multiple_exception(vcpu, nr, false, 0, true);
458}
459EXPORT_SYMBOL_GPL(kvm_requeue_exception);
460
Kyle Huey6affcbe2016-11-29 12:40:40 -0800461int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
Avi Kivityc3c91fe2007-11-25 14:04:58 +0200462{
Andre Przywaradb8fcef2010-12-21 11:12:01 +0100463 if (err)
464 kvm_inject_gp(vcpu, 0);
465 else
Kyle Huey6affcbe2016-11-29 12:40:40 -0800466 return kvm_skip_emulated_instruction(vcpu);
467
468 return 1;
Andre Przywaradb8fcef2010-12-21 11:12:01 +0100469}
470EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
Joerg Roedel8df25a32010-09-10 17:30:46 +0200471
Avi Kivity6389ee92010-11-29 16:12:30 +0200472void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
Avi Kivityc3c91fe2007-11-25 14:04:58 +0200473{
Avi Kivityc3c91fe2007-11-25 14:04:58 +0200474 ++vcpu->stat.pf_guest;
Wanpeng Liadfe20f2017-07-13 18:30:41 -0700475 vcpu->arch.exception.nested_apf =
476 is_guest_mode(vcpu) && fault->async_page_fault;
477 if (vcpu->arch.exception.nested_apf)
478 vcpu->arch.apf.nested_apf_token = fault->address;
479 else
480 vcpu->arch.cr2 = fault->address;
Avi Kivity6389ee92010-11-29 16:12:30 +0200481 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
Avi Kivityc3c91fe2007-11-25 14:04:58 +0200482}
Nadav Har'El27d6c862011-05-25 23:06:59 +0300483EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
Avi Kivityc3c91fe2007-11-25 14:04:58 +0200484
Paolo Bonzinief54bcf2014-09-04 19:46:15 +0200485static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
Joerg Roedeld4f8cf62010-09-10 17:30:55 +0200486{
Avi Kivity6389ee92010-11-29 16:12:30 +0200487 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
488 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
Joerg Roedeld4f8cf62010-09-10 17:30:55 +0200489 else
Avi Kivity6389ee92010-11-29 16:12:30 +0200490 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
Paolo Bonzinief54bcf2014-09-04 19:46:15 +0200491
492 return fault->nested_page_fault;
Joerg Roedeld4f8cf62010-09-10 17:30:55 +0200493}
494
Sheng Yang3419ffc2008-05-15 09:52:48 +0800495void kvm_inject_nmi(struct kvm_vcpu *vcpu)
496{
Avi Kivity7460fb4a2011-09-20 13:43:14 +0300497 atomic_inc(&vcpu->arch.nmi_queued);
498 kvm_make_request(KVM_REQ_NMI, vcpu);
Sheng Yang3419ffc2008-05-15 09:52:48 +0800499}
500EXPORT_SYMBOL_GPL(kvm_inject_nmi);
501
Avi Kivity298101d2007-11-25 13:41:11 +0200502void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
503{
Joerg Roedelce7ddec2010-04-22 12:33:13 +0200504 kvm_multiple_exception(vcpu, nr, true, error_code, false);
Avi Kivity298101d2007-11-25 13:41:11 +0200505}
506EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
507
Joerg Roedelce7ddec2010-04-22 12:33:13 +0200508void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
509{
510 kvm_multiple_exception(vcpu, nr, true, error_code, true);
511}
512EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
513
Carsten Ottea03490e2007-10-29 16:09:35 +0100514/*
Avi Kivity0a79b002009-09-01 12:03:25 +0300515 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
516 * a #GP and return false.
517 */
518bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
Carsten Otte043405e2007-10-10 17:16:19 +0200519{
Avi Kivity0a79b002009-09-01 12:03:25 +0300520 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
521 return true;
522 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
523 return false;
Carsten Ottea03490e2007-10-29 16:09:35 +0100524}
Avi Kivity0a79b002009-09-01 12:03:25 +0300525EXPORT_SYMBOL_GPL(kvm_require_cpl);
Carsten Ottea03490e2007-10-29 16:09:35 +0100526
Nadav Amit16f8a6f2014-10-03 01:10:05 +0300527bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
528{
529 if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
530 return true;
531
532 kvm_queue_exception(vcpu, UD_VECTOR);
533 return false;
534}
535EXPORT_SYMBOL_GPL(kvm_require_dr);
536
Carsten Ottea03490e2007-10-29 16:09:35 +0100537/*
Joerg Roedelec92fe42010-09-10 17:30:51 +0200538 * This function will be used to read from the physical memory of the currently
Paolo Bonzini54bf36a2015-04-08 15:39:23 +0200539 * running guest. The difference to kvm_vcpu_read_guest_page is that this function
Joerg Roedelec92fe42010-09-10 17:30:51 +0200540 * can read from guest physical or from the guest's guest physical memory.
541 */
542int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
543 gfn_t ngfn, void *data, int offset, int len,
544 u32 access)
545{
Paolo Bonzini54987b72014-09-02 13:23:06 +0200546 struct x86_exception exception;
Joerg Roedelec92fe42010-09-10 17:30:51 +0200547 gfn_t real_gfn;
548 gpa_t ngpa;
549
550 ngpa = gfn_to_gpa(ngfn);
Paolo Bonzini54987b72014-09-02 13:23:06 +0200551 real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
Joerg Roedelec92fe42010-09-10 17:30:51 +0200552 if (real_gfn == UNMAPPED_GVA)
553 return -EFAULT;
554
555 real_gfn = gpa_to_gfn(real_gfn);
556
Paolo Bonzini54bf36a2015-04-08 15:39:23 +0200557 return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
Joerg Roedelec92fe42010-09-10 17:30:51 +0200558}
559EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
560
Fengguang Wu69b00492015-01-19 22:33:39 +0800561static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
Joerg Roedel3d06b8b2010-09-10 17:30:53 +0200562 void *data, int offset, int len, u32 access)
563{
564 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
565 data, offset, len, access);
566}
567
Joerg Roedelec92fe42010-09-10 17:30:51 +0200568/*
Carsten Ottea03490e2007-10-29 16:09:35 +0100569 * Load the pae pdptrs. Return true is they are all valid.
570 */
Joerg Roedelff03a072010-09-10 17:30:57 +0200571int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
Carsten Ottea03490e2007-10-29 16:09:35 +0100572{
573 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
574 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
575 int i;
576 int ret;
Joerg Roedelff03a072010-09-10 17:30:57 +0200577 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
Carsten Ottea03490e2007-10-29 16:09:35 +0100578
Joerg Roedelff03a072010-09-10 17:30:57 +0200579 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
580 offset * sizeof(u64), sizeof(pdpte),
581 PFERR_USER_MASK|PFERR_WRITE_MASK);
Carsten Ottea03490e2007-10-29 16:09:35 +0100582 if (ret < 0) {
583 ret = 0;
584 goto out;
585 }
586 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
Bandan Das812f30b2016-07-12 18:18:50 -0400587 if ((pdpte[i] & PT_PRESENT_MASK) &&
Xiao Guangronga0a64f52015-08-05 12:04:21 +0800588 (pdpte[i] &
589 vcpu->arch.mmu.guest_rsvd_check.rsvd_bits_mask[0][2])) {
Carsten Ottea03490e2007-10-29 16:09:35 +0100590 ret = 0;
591 goto out;
592 }
593 }
594 ret = 1;
595
Joerg Roedelff03a072010-09-10 17:30:57 +0200596 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
Avi Kivity6de4f3a2009-05-31 22:58:47 +0300597 __set_bit(VCPU_EXREG_PDPTR,
598 (unsigned long *)&vcpu->arch.regs_avail);
599 __set_bit(VCPU_EXREG_PDPTR,
600 (unsigned long *)&vcpu->arch.regs_dirty);
Carsten Ottea03490e2007-10-29 16:09:35 +0100601out:
Carsten Ottea03490e2007-10-29 16:09:35 +0100602
603 return ret;
604}
Joerg Roedelcc4b6872008-02-07 13:47:43 +0100605EXPORT_SYMBOL_GPL(load_pdptrs);
Carsten Ottea03490e2007-10-29 16:09:35 +0100606
Ladi Prosek9ed38ffa2016-11-30 16:03:10 +0100607bool pdptrs_changed(struct kvm_vcpu *vcpu)
Avi Kivityd835dfe2007-11-21 02:57:59 +0200608{
Joerg Roedelff03a072010-09-10 17:30:57 +0200609 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
Avi Kivityd835dfe2007-11-21 02:57:59 +0200610 bool changed = true;
Joerg Roedel3d06b8b2010-09-10 17:30:53 +0200611 int offset;
612 gfn_t gfn;
Avi Kivityd835dfe2007-11-21 02:57:59 +0200613 int r;
614
615 if (is_long_mode(vcpu) || !is_pae(vcpu))
616 return false;
617
Avi Kivity6de4f3a2009-05-31 22:58:47 +0300618 if (!test_bit(VCPU_EXREG_PDPTR,
619 (unsigned long *)&vcpu->arch.regs_avail))
620 return true;
621
Paolo Bonzinia5121772017-07-24 18:54:38 +0200622 gfn = (kvm_read_cr3(vcpu) & 0xffffffe0ul) >> PAGE_SHIFT;
623 offset = (kvm_read_cr3(vcpu) & 0xffffffe0ul) & (PAGE_SIZE - 1);
Joerg Roedel3d06b8b2010-09-10 17:30:53 +0200624 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
625 PFERR_USER_MASK | PFERR_WRITE_MASK);
Avi Kivityd835dfe2007-11-21 02:57:59 +0200626 if (r < 0)
627 goto out;
Joerg Roedelff03a072010-09-10 17:30:57 +0200628 changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
Avi Kivityd835dfe2007-11-21 02:57:59 +0200629out:
Avi Kivityd835dfe2007-11-21 02:57:59 +0200630
631 return changed;
632}
Ladi Prosek9ed38ffa2016-11-30 16:03:10 +0100633EXPORT_SYMBOL_GPL(pdptrs_changed);
Avi Kivityd835dfe2007-11-21 02:57:59 +0200634
Avi Kivity49a9b072010-06-10 17:02:14 +0300635int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
Carsten Ottea03490e2007-10-29 16:09:35 +0100636{
Sheng Yangaad82702010-05-12 16:40:42 +0800637 unsigned long old_cr0 = kvm_read_cr0(vcpu);
Xiao Guangrongd81135a2015-05-13 14:42:28 +0800638 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
Sheng Yangaad82702010-05-12 16:40:42 +0800639
Avi Kivityf9a48e62010-01-06 19:10:22 +0200640 cr0 |= X86_CR0_ET;
641
Gleb Natapovab344822010-01-21 15:28:46 +0200642#ifdef CONFIG_X86_64
Gleb Natapov0f122442010-04-28 19:15:31 +0300643 if (cr0 & 0xffffffff00000000UL)
644 return 1;
Gleb Natapovab344822010-01-21 15:28:46 +0200645#endif
646
647 cr0 &= ~CR0_RESERVED_BITS;
Carsten Ottea03490e2007-10-29 16:09:35 +0100648
Gleb Natapov0f122442010-04-28 19:15:31 +0300649 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
650 return 1;
Carsten Ottea03490e2007-10-29 16:09:35 +0100651
Gleb Natapov0f122442010-04-28 19:15:31 +0300652 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
653 return 1;
Carsten Ottea03490e2007-10-29 16:09:35 +0100654
655 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
656#ifdef CONFIG_X86_64
Avi Kivityf6801df2010-01-21 15:31:50 +0200657 if ((vcpu->arch.efer & EFER_LME)) {
Carsten Ottea03490e2007-10-29 16:09:35 +0100658 int cs_db, cs_l;
659
Gleb Natapov0f122442010-04-28 19:15:31 +0300660 if (!is_pae(vcpu))
661 return 1;
Carsten Ottea03490e2007-10-29 16:09:35 +0100662 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
Gleb Natapov0f122442010-04-28 19:15:31 +0300663 if (cs_l)
664 return 1;
Carsten Ottea03490e2007-10-29 16:09:35 +0100665 } else
666#endif
Joerg Roedelff03a072010-09-10 17:30:57 +0200667 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
Avi Kivity9f8fe502010-12-05 17:30:00 +0200668 kvm_read_cr3(vcpu)))
Gleb Natapov0f122442010-04-28 19:15:31 +0300669 return 1;
Carsten Ottea03490e2007-10-29 16:09:35 +0100670 }
671
Mao, Junjiead756a12012-07-02 01:18:48 +0000672 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
673 return 1;
674
Carsten Ottea03490e2007-10-29 16:09:35 +0100675 kvm_x86_ops->set_cr0(vcpu, cr0);
Carsten Ottea03490e2007-10-29 16:09:35 +0100676
Lai Jiangshand170c412011-02-21 11:21:30 +0800677 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
Xiao Guangronge5f3f022010-11-12 14:47:01 +0800678 kvm_clear_async_pf_completion_queue(vcpu);
Lai Jiangshand170c412011-02-21 11:21:30 +0800679 kvm_async_pf_hash_reset(vcpu);
680 }
Xiao Guangronge5f3f022010-11-12 14:47:01 +0800681
Sheng Yangaad82702010-05-12 16:40:42 +0800682 if ((cr0 ^ old_cr0) & update_bits)
683 kvm_mmu_reset_context(vcpu);
Xiao Guangrongb18d5432015-06-15 16:55:21 +0800684
Laszlo Ersek879ae182015-11-04 12:54:41 +0100685 if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
686 kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
687 !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
Xiao Guangrongb18d5432015-06-15 16:55:21 +0800688 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
689
Gleb Natapov0f122442010-04-28 19:15:31 +0300690 return 0;
691}
Avi Kivity2d3ad1f2008-02-24 11:20:43 +0200692EXPORT_SYMBOL_GPL(kvm_set_cr0);
Carsten Ottea03490e2007-10-29 16:09:35 +0100693
Avi Kivity2d3ad1f2008-02-24 11:20:43 +0200694void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
Carsten Ottea03490e2007-10-29 16:09:35 +0100695{
Avi Kivity49a9b072010-06-10 17:02:14 +0300696 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
Carsten Ottea03490e2007-10-29 16:09:35 +0100697}
Avi Kivity2d3ad1f2008-02-24 11:20:43 +0200698EXPORT_SYMBOL_GPL(kvm_lmsw);
Carsten Ottea03490e2007-10-29 16:09:35 +0100699
Marcelo Tosatti42bdf992013-04-15 23:30:13 -0300700static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
701{
702 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
703 !vcpu->guest_xcr0_loaded) {
704 /* kvm_set_xcr() also depends on this */
705 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
706 vcpu->guest_xcr0_loaded = 1;
707 }
708}
709
710static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
711{
712 if (vcpu->guest_xcr0_loaded) {
713 if (vcpu->arch.xcr0 != host_xcr0)
714 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
715 vcpu->guest_xcr0_loaded = 0;
716 }
717}
718
Fengguang Wu69b00492015-01-19 22:33:39 +0800719static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
Dexuan Cui2acf9232010-06-10 11:27:12 +0800720{
Liu, Jinsong56c103e2014-02-21 17:39:02 +0000721 u64 xcr0 = xcr;
722 u64 old_xcr0 = vcpu->arch.xcr0;
Paolo Bonzini46c34cb2013-10-17 16:50:46 +0200723 u64 valid_bits;
Dexuan Cui2acf9232010-06-10 11:27:12 +0800724
725 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
726 if (index != XCR_XFEATURE_ENABLED_MASK)
727 return 1;
Dave Hansend91cab72015-09-02 16:31:26 -0700728 if (!(xcr0 & XFEATURE_MASK_FP))
Dexuan Cui2acf9232010-06-10 11:27:12 +0800729 return 1;
Dave Hansend91cab72015-09-02 16:31:26 -0700730 if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
Dexuan Cui2acf9232010-06-10 11:27:12 +0800731 return 1;
Paolo Bonzini46c34cb2013-10-17 16:50:46 +0200732
733 /*
734 * Do not allow the guest to set bits that we do not support
735 * saving. However, xcr0 bit 0 is always set, even if the
736 * emulated CPU does not support XSAVE (see fx_init).
737 */
Dave Hansend91cab72015-09-02 16:31:26 -0700738 valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
Paolo Bonzini46c34cb2013-10-17 16:50:46 +0200739 if (xcr0 & ~valid_bits)
Dexuan Cui2acf9232010-06-10 11:27:12 +0800740 return 1;
Paolo Bonzini46c34cb2013-10-17 16:50:46 +0200741
Dave Hansend91cab72015-09-02 16:31:26 -0700742 if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
743 (!(xcr0 & XFEATURE_MASK_BNDCSR)))
Liu, Jinsong390bd522014-02-24 10:58:09 +0000744 return 1;
745
Dave Hansend91cab72015-09-02 16:31:26 -0700746 if (xcr0 & XFEATURE_MASK_AVX512) {
747 if (!(xcr0 & XFEATURE_MASK_YMM))
Chao Peng612263b2014-10-22 17:35:24 +0800748 return 1;
Dave Hansend91cab72015-09-02 16:31:26 -0700749 if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
Chao Peng612263b2014-10-22 17:35:24 +0800750 return 1;
751 }
Dexuan Cui2acf9232010-06-10 11:27:12 +0800752 vcpu->arch.xcr0 = xcr0;
Liu, Jinsong56c103e2014-02-21 17:39:02 +0000753
Dave Hansend91cab72015-09-02 16:31:26 -0700754 if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
Liu, Jinsong56c103e2014-02-21 17:39:02 +0000755 kvm_update_cpuid(vcpu);
Dexuan Cui2acf9232010-06-10 11:27:12 +0800756 return 0;
757}
758
759int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
760{
Zhanghaoyu (A)764bcbc2013-06-14 07:36:13 +0000761 if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
762 __kvm_set_xcr(vcpu, index, xcr)) {
Dexuan Cui2acf9232010-06-10 11:27:12 +0800763 kvm_inject_gp(vcpu, 0);
764 return 1;
765 }
766 return 0;
767}
768EXPORT_SYMBOL_GPL(kvm_set_xcr);
769
Avi Kivitya83b29c2010-06-10 17:02:15 +0300770int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
Carsten Ottea03490e2007-10-29 16:09:35 +0100771{
Avi Kivityfc78f512009-12-07 12:16:48 +0200772 unsigned long old_cr4 = kvm_read_cr4(vcpu);
Xiao Guangrong0be02262015-05-11 22:55:21 +0800773 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
Huaitong Hanb9baba82016-03-22 16:51:21 +0800774 X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
Xiao Guangrong0be02262015-05-11 22:55:21 +0800775
Gleb Natapov0f122442010-04-28 19:15:31 +0300776 if (cr4 & CR4_RESERVED_BITS)
777 return 1;
Carsten Ottea03490e2007-10-29 16:09:35 +0100778
Radim Krčmářd6321d42017-08-05 00:12:49 +0200779 if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) && (cr4 & X86_CR4_OSXSAVE))
Dexuan Cui2acf9232010-06-10 11:27:12 +0800780 return 1;
781
Radim Krčmářd6321d42017-08-05 00:12:49 +0200782 if (!guest_cpuid_has(vcpu, X86_FEATURE_SMEP) && (cr4 & X86_CR4_SMEP))
Yang, Wei Yc68b7342011-06-03 11:13:42 +0800783 return 1;
784
Radim Krčmářd6321d42017-08-05 00:12:49 +0200785 if (!guest_cpuid_has(vcpu, X86_FEATURE_SMAP) && (cr4 & X86_CR4_SMAP))
Feng Wu97ec8c02014-04-01 17:46:34 +0800786 return 1;
787
Radim Krčmářd6321d42017-08-05 00:12:49 +0200788 if (!guest_cpuid_has(vcpu, X86_FEATURE_FSGSBASE) && (cr4 & X86_CR4_FSGSBASE))
Yang, Wei74dc2b42011-06-14 20:10:18 +0800789 return 1;
790
Radim Krčmářd6321d42017-08-05 00:12:49 +0200791 if (!guest_cpuid_has(vcpu, X86_FEATURE_PKU) && (cr4 & X86_CR4_PKE))
Huaitong Hanb9baba82016-03-22 16:51:21 +0800792 return 1;
793
Yu Zhangfd8cb432017-08-24 20:27:56 +0800794 if (!guest_cpuid_has(vcpu, X86_FEATURE_LA57) && (cr4 & X86_CR4_LA57))
Nadav Har'El5e1746d2011-05-25 23:03:24 +0300795 return 1;
Gleb Natapov0f122442010-04-28 19:15:31 +0300796
797 if (is_long_mode(vcpu)) {
Sheng Yangaad82702010-05-12 16:40:42 +0800798 if (!(cr4 & X86_CR4_PAE))
799 return 1;
Gleb Natapov0f122442010-04-28 19:15:31 +0300800 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
Dexuan Cui2acf9232010-06-10 11:27:12 +0800801 && ((cr4 ^ old_cr4) & pdptr_bits)
802 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
803 kvm_read_cr3(vcpu)))
Gleb Natapov0f122442010-04-28 19:15:31 +0300804 return 1;
805
Mao, Junjiead756a12012-07-02 01:18:48 +0000806 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
Radim Krčmářd6321d42017-08-05 00:12:49 +0200807 if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
Mao, Junjiead756a12012-07-02 01:18:48 +0000808 return 1;
809
810 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
811 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
812 return 1;
813 }
814
Carsten Ottea03490e2007-10-29 16:09:35 +0100815 if (kvm_x86_ops->set_cr4(vcpu, cr4))
816 return 1;
817
Mao, Junjiead756a12012-07-02 01:18:48 +0000818 if (((cr4 ^ old_cr4) & pdptr_bits) ||
819 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
Carsten Ottea03490e2007-10-29 16:09:35 +0100820 kvm_mmu_reset_context(vcpu);
821
Huaitong Hanb9baba82016-03-22 16:51:21 +0800822 if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
Avi Kivity00b27a32011-11-23 16:30:32 +0200823 kvm_update_cpuid(vcpu);
Gleb Natapov0f122442010-04-28 19:15:31 +0300824
825 return 0;
826}
Avi Kivity2d3ad1f2008-02-24 11:20:43 +0200827EXPORT_SYMBOL_GPL(kvm_set_cr4);
Carsten Ottea03490e2007-10-29 16:09:35 +0100828
Avi Kivity23902182010-06-10 17:02:16 +0300829int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
Carsten Ottea03490e2007-10-29 16:09:35 +0100830{
Paolo Bonziniac146232014-11-10 13:53:25 +0100831#ifdef CONFIG_X86_64
Nadav Amit9d88fca2014-11-02 11:54:52 +0200832 cr3 &= ~CR3_PCID_INVD;
Paolo Bonziniac146232014-11-10 13:53:25 +0100833#endif
Nadav Amit9d88fca2014-11-02 11:54:52 +0200834
Avi Kivity9f8fe502010-12-05 17:30:00 +0200835 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
Marcelo Tosatti0ba73cd2008-09-23 13:18:34 -0300836 kvm_mmu_sync_roots(vcpu);
Liang Chen77c39132014-09-18 12:38:37 -0400837 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
Gleb Natapov0f122442010-04-28 19:15:31 +0300838 return 0;
Avi Kivityd835dfe2007-11-21 02:57:59 +0200839 }
840
Yu Zhangd1cd3ce2017-08-24 20:27:53 +0800841 if (is_long_mode(vcpu) &&
842 (cr3 & rsvd_bits(cpuid_maxphyaddr(vcpu), 62)))
843 return 1;
844 else if (is_pae(vcpu) && is_paging(vcpu) &&
Jan Kiszkad9f89b82014-05-10 09:24:34 +0200845 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
Nadav Amit346874c2014-04-18 03:35:09 +0300846 return 1;
Carsten Ottea03490e2007-10-29 16:09:35 +0100847
Gleb Natapov0f122442010-04-28 19:15:31 +0300848 vcpu->arch.cr3 = cr3;
Avi Kivityaff48ba2010-12-05 18:56:11 +0200849 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
Paolo Bonzinid8d173d2013-10-02 16:56:11 +0200850 kvm_mmu_new_cr3(vcpu);
Gleb Natapov0f122442010-04-28 19:15:31 +0300851 return 0;
852}
Avi Kivity2d3ad1f2008-02-24 11:20:43 +0200853EXPORT_SYMBOL_GPL(kvm_set_cr3);
Carsten Ottea03490e2007-10-29 16:09:35 +0100854
Andre Przywaraeea1cff2010-12-21 11:12:00 +0100855int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
Carsten Ottea03490e2007-10-29 16:09:35 +0100856{
Gleb Natapov0f122442010-04-28 19:15:31 +0300857 if (cr8 & CR8_RESERVED_BITS)
858 return 1;
Paolo Bonzini35754c92015-07-29 12:05:37 +0200859 if (lapic_in_kernel(vcpu))
Carsten Ottea03490e2007-10-29 16:09:35 +0100860 kvm_lapic_set_tpr(vcpu, cr8);
861 else
Zhang Xiantaoad312c72007-12-13 23:50:52 +0800862 vcpu->arch.cr8 = cr8;
Gleb Natapov0f122442010-04-28 19:15:31 +0300863 return 0;
864}
Avi Kivity2d3ad1f2008-02-24 11:20:43 +0200865EXPORT_SYMBOL_GPL(kvm_set_cr8);
Carsten Ottea03490e2007-10-29 16:09:35 +0100866
Avi Kivity2d3ad1f2008-02-24 11:20:43 +0200867unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
Carsten Ottea03490e2007-10-29 16:09:35 +0100868{
Paolo Bonzini35754c92015-07-29 12:05:37 +0200869 if (lapic_in_kernel(vcpu))
Carsten Ottea03490e2007-10-29 16:09:35 +0100870 return kvm_lapic_get_cr8(vcpu);
871 else
Zhang Xiantaoad312c72007-12-13 23:50:52 +0800872 return vcpu->arch.cr8;
Carsten Ottea03490e2007-10-29 16:09:35 +0100873}
Avi Kivity2d3ad1f2008-02-24 11:20:43 +0200874EXPORT_SYMBOL_GPL(kvm_get_cr8);
Carsten Ottea03490e2007-10-29 16:09:35 +0100875
Nadav Amitae561ed2015-04-02 03:10:37 +0300876static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
877{
878 int i;
879
880 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
881 for (i = 0; i < KVM_NR_DB_REGS; i++)
882 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
883 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
884 }
885}
886
Jan Kiszka73aaf249e2014-01-04 18:47:16 +0100887static void kvm_update_dr6(struct kvm_vcpu *vcpu)
888{
889 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
890 kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
891}
892
Jan Kiszkac8639012012-09-21 05:42:55 +0200893static void kvm_update_dr7(struct kvm_vcpu *vcpu)
894{
895 unsigned long dr7;
896
897 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
898 dr7 = vcpu->arch.guest_debug_dr7;
899 else
900 dr7 = vcpu->arch.dr7;
901 kvm_x86_ops->set_dr7(vcpu, dr7);
Paolo Bonzini360b9482014-02-21 09:55:56 +0100902 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
903 if (dr7 & DR7_BP_EN_MASK)
904 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
Jan Kiszkac8639012012-09-21 05:42:55 +0200905}
906
Nadav Amit6f43ed02014-07-15 17:37:46 +0300907static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
908{
909 u64 fixed = DR6_FIXED_1;
910
Radim Krčmářd6321d42017-08-05 00:12:49 +0200911 if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
Nadav Amit6f43ed02014-07-15 17:37:46 +0300912 fixed |= DR6_RTM;
913 return fixed;
914}
915
Gleb Natapov338dbc92010-04-28 19:15:32 +0300916static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
Gleb Natapov020df072010-04-13 10:05:23 +0300917{
918 switch (dr) {
919 case 0 ... 3:
920 vcpu->arch.db[dr] = val;
921 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
922 vcpu->arch.eff_db[dr] = val;
923 break;
924 case 4:
Gleb Natapov020df072010-04-13 10:05:23 +0300925 /* fall through */
926 case 6:
Gleb Natapov338dbc92010-04-28 19:15:32 +0300927 if (val & 0xffffffff00000000ULL)
928 return -1; /* #GP */
Nadav Amit6f43ed02014-07-15 17:37:46 +0300929 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
Jan Kiszka73aaf249e2014-01-04 18:47:16 +0100930 kvm_update_dr6(vcpu);
Gleb Natapov020df072010-04-13 10:05:23 +0300931 break;
932 case 5:
Gleb Natapov020df072010-04-13 10:05:23 +0300933 /* fall through */
934 default: /* 7 */
Gleb Natapov338dbc92010-04-28 19:15:32 +0300935 if (val & 0xffffffff00000000ULL)
936 return -1; /* #GP */
Gleb Natapov020df072010-04-13 10:05:23 +0300937 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
Jan Kiszkac8639012012-09-21 05:42:55 +0200938 kvm_update_dr7(vcpu);
Gleb Natapov020df072010-04-13 10:05:23 +0300939 break;
940 }
941
942 return 0;
943}
Gleb Natapov338dbc92010-04-28 19:15:32 +0300944
945int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
946{
Nadav Amit16f8a6f2014-10-03 01:10:05 +0300947 if (__kvm_set_dr(vcpu, dr, val)) {
Gleb Natapov338dbc92010-04-28 19:15:32 +0300948 kvm_inject_gp(vcpu, 0);
Nadav Amit16f8a6f2014-10-03 01:10:05 +0300949 return 1;
950 }
951 return 0;
Gleb Natapov338dbc92010-04-28 19:15:32 +0300952}
Gleb Natapov020df072010-04-13 10:05:23 +0300953EXPORT_SYMBOL_GPL(kvm_set_dr);
954
Nadav Amit16f8a6f2014-10-03 01:10:05 +0300955int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
Gleb Natapov020df072010-04-13 10:05:23 +0300956{
957 switch (dr) {
958 case 0 ... 3:
959 *val = vcpu->arch.db[dr];
960 break;
961 case 4:
Gleb Natapov020df072010-04-13 10:05:23 +0300962 /* fall through */
963 case 6:
Jan Kiszka73aaf249e2014-01-04 18:47:16 +0100964 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
965 *val = vcpu->arch.dr6;
966 else
967 *val = kvm_x86_ops->get_dr6(vcpu);
Gleb Natapov020df072010-04-13 10:05:23 +0300968 break;
969 case 5:
Gleb Natapov020df072010-04-13 10:05:23 +0300970 /* fall through */
971 default: /* 7 */
972 *val = vcpu->arch.dr7;
973 break;
974 }
Gleb Natapov338dbc92010-04-28 19:15:32 +0300975 return 0;
976}
Gleb Natapov020df072010-04-13 10:05:23 +0300977EXPORT_SYMBOL_GPL(kvm_get_dr);
978
Avi Kivity022cd0e2011-11-10 14:57:23 +0200979bool kvm_rdpmc(struct kvm_vcpu *vcpu)
980{
981 u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
982 u64 data;
983 int err;
984
Wei Huangc6702c92015-06-19 13:44:45 +0200985 err = kvm_pmu_rdpmc(vcpu, ecx, &data);
Avi Kivity022cd0e2011-11-10 14:57:23 +0200986 if (err)
987 return err;
988 kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
989 kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
990 return err;
991}
992EXPORT_SYMBOL_GPL(kvm_rdpmc);
993
Carsten Otte043405e2007-10-10 17:16:19 +0200994/*
995 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
996 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
997 *
998 * This list is modified at module load time to reflect the
Glauber Costae3267cb2009-10-06 13:24:50 -0400999 * capabilities of the host cpu. This capabilities test skips MSRs that are
Paolo Bonzini62ef68b2015-05-05 12:08:55 +02001000 * kvm-specific. Those are put in emulated_msrs; filtering of emulated_msrs
1001 * may depend on host virtualization features rather than host cpu features.
Carsten Otte043405e2007-10-10 17:16:19 +02001002 */
Glauber Costae3267cb2009-10-06 13:24:50 -04001003
Carsten Otte043405e2007-10-10 17:16:19 +02001004static u32 msrs_to_save[] = {
1005 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
Brian Gerst8c065852010-07-17 09:03:26 -04001006 MSR_STAR,
Carsten Otte043405e2007-10-10 17:16:19 +02001007#ifdef CONFIG_X86_64
1008 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1009#endif
Nadav Har'Elb3897a42013-07-08 19:12:35 +08001010 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
Paolo Bonzini9dbe6cf2015-11-12 14:49:17 +01001011 MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
Carsten Otte043405e2007-10-10 17:16:19 +02001012};
1013
1014static unsigned num_msrs_to_save;
1015
Paolo Bonzini62ef68b2015-05-05 12:08:55 +02001016static u32 emulated_msrs[] = {
1017 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1018 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1019 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1020 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
Ladi Prosek72c139b2017-07-26 13:32:59 +02001021 HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
Andrey Smetanine7d95132015-07-03 15:01:37 +03001022 HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1023 HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
Andrey Smetanine516ceb2015-09-16 12:29:48 +03001024 HV_X64_MSR_RESET,
Andrey Smetanin11c4b1c2015-09-16 12:29:49 +03001025 HV_X64_MSR_VP_INDEX,
Andrey Smetanin9eec50b2015-09-16 12:29:50 +03001026 HV_X64_MSR_VP_RUNTIME,
Andrey Smetanin5c9194122015-11-10 15:36:34 +03001027 HV_X64_MSR_SCONTROL,
Andrey Smetanin1f4b34f2015-11-30 19:22:21 +03001028 HV_X64_MSR_STIMER0_CONFIG,
Paolo Bonzini62ef68b2015-05-05 12:08:55 +02001029 HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1030 MSR_KVM_PV_EOI_EN,
1031
Will Auldba904632012-11-29 12:42:50 -08001032 MSR_IA32_TSC_ADJUST,
Liu, Jinsonga3e06bb2011-09-22 16:55:52 +08001033 MSR_IA32_TSCDEADLINE,
Carsten Otte043405e2007-10-10 17:16:19 +02001034 MSR_IA32_MISC_ENABLE,
Avi Kivity908e75f2010-07-07 14:09:38 +03001035 MSR_IA32_MCG_STATUS,
1036 MSR_IA32_MCG_CTL,
Ashok Rajc45dcc72016-06-22 14:59:56 +08001037 MSR_IA32_MCG_EXT_CTL,
Paolo Bonzini64d60672015-05-07 11:36:11 +02001038 MSR_IA32_SMBASE,
Kyle Hueydb2336a2017-03-20 01:16:28 -07001039 MSR_PLATFORM_INFO,
1040 MSR_MISC_FEATURES_ENABLES,
Carsten Otte043405e2007-10-10 17:16:19 +02001041};
1042
Paolo Bonzini62ef68b2015-05-05 12:08:55 +02001043static unsigned num_emulated_msrs;
1044
Jan Kiszka384bb782013-04-20 10:52:36 +02001045bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
Carsten Otte15c4a642007-10-30 18:44:17 +01001046{
Roedel, Joergb69e8ca2010-05-06 11:38:43 +02001047 if (efer & efer_reserved_bits)
Jan Kiszka384bb782013-04-20 10:52:36 +02001048 return false;
Carsten Otte15c4a642007-10-30 18:44:17 +01001049
Radim Krčmář1b4d56b2017-08-05 00:12:50 +02001050 if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
Jan Kiszka384bb782013-04-20 10:52:36 +02001051 return false;
Alexander Graf1b2fd702009-02-02 16:23:51 +01001052
Radim Krčmář1b4d56b2017-08-05 00:12:50 +02001053 if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
Jan Kiszka384bb782013-04-20 10:52:36 +02001054 return false;
Alexander Grafd8017472008-11-25 20:17:11 +01001055
Jan Kiszka384bb782013-04-20 10:52:36 +02001056 return true;
1057}
1058EXPORT_SYMBOL_GPL(kvm_valid_efer);
1059
1060static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
1061{
1062 u64 old_efer = vcpu->arch.efer;
1063
1064 if (!kvm_valid_efer(vcpu, efer))
1065 return 1;
1066
1067 if (is_paging(vcpu)
1068 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1069 return 1;
1070
Carsten Otte15c4a642007-10-30 18:44:17 +01001071 efer &= ~EFER_LMA;
Avi Kivityf6801df2010-01-21 15:31:50 +02001072 efer |= vcpu->arch.efer & EFER_LMA;
Carsten Otte15c4a642007-10-30 18:44:17 +01001073
Sheng Yanga3d204e2010-05-12 16:40:40 +08001074 kvm_x86_ops->set_efer(vcpu, efer);
1075
Sheng Yangaad82702010-05-12 16:40:42 +08001076 /* Update reserved bits */
1077 if ((efer ^ old_efer) & EFER_NX)
1078 kvm_mmu_reset_context(vcpu);
1079
Roedel, Joergb69e8ca2010-05-06 11:38:43 +02001080 return 0;
Carsten Otte15c4a642007-10-30 18:44:17 +01001081}
1082
Joerg Roedelf2b4b7d2008-01-31 14:57:37 +01001083void kvm_enable_efer_bits(u64 mask)
1084{
1085 efer_reserved_bits &= ~mask;
1086}
1087EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1088
Carsten Otte15c4a642007-10-30 18:44:17 +01001089/*
1090 * Writes msr value into into the appropriate "register".
1091 * Returns 0 on success, non-0 otherwise.
1092 * Assumes vcpu_load() was already called.
1093 */
Will Auld8fe8ab42012-11-29 12:42:12 -08001094int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
Carsten Otte15c4a642007-10-30 18:44:17 +01001095{
Nadav Amit854e8bb2014-09-16 03:24:05 +03001096 switch (msr->index) {
1097 case MSR_FS_BASE:
1098 case MSR_GS_BASE:
1099 case MSR_KERNEL_GS_BASE:
1100 case MSR_CSTAR:
1101 case MSR_LSTAR:
Yu Zhangfd8cb432017-08-24 20:27:56 +08001102 if (is_noncanonical_address(msr->data, vcpu))
Nadav Amit854e8bb2014-09-16 03:24:05 +03001103 return 1;
1104 break;
1105 case MSR_IA32_SYSENTER_EIP:
1106 case MSR_IA32_SYSENTER_ESP:
1107 /*
1108 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1109 * non-canonical address is written on Intel but not on
1110 * AMD (which ignores the top 32-bits, because it does
1111 * not implement 64-bit SYSENTER).
1112 *
1113 * 64-bit code should hence be able to write a non-canonical
1114 * value on AMD. Making the address canonical ensures that
1115 * vmentry does not fail on Intel after writing a non-canonical
1116 * value, and that something deterministic happens if the guest
1117 * invokes 64-bit SYSENTER.
1118 */
Yu Zhangfd8cb432017-08-24 20:27:56 +08001119 msr->data = get_canonical(msr->data, vcpu_virt_addr_bits(vcpu));
Nadav Amit854e8bb2014-09-16 03:24:05 +03001120 }
Will Auld8fe8ab42012-11-29 12:42:12 -08001121 return kvm_x86_ops->set_msr(vcpu, msr);
Carsten Otte15c4a642007-10-30 18:44:17 +01001122}
Nadav Amit854e8bb2014-09-16 03:24:05 +03001123EXPORT_SYMBOL_GPL(kvm_set_msr);
Carsten Otte15c4a642007-10-30 18:44:17 +01001124
Carsten Otte313a3dc2007-10-11 19:16:52 +02001125/*
1126 * Adapt set_msr() to msr_io()'s calling convention
1127 */
Paolo Bonzini609e36d2015-04-08 15:30:38 +02001128static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1129{
1130 struct msr_data msr;
1131 int r;
1132
1133 msr.index = index;
1134 msr.host_initiated = true;
1135 r = kvm_get_msr(vcpu, &msr);
1136 if (r)
1137 return r;
1138
1139 *data = msr.data;
1140 return 0;
1141}
1142
Carsten Otte313a3dc2007-10-11 19:16:52 +02001143static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1144{
Will Auld8fe8ab42012-11-29 12:42:12 -08001145 struct msr_data msr;
1146
1147 msr.data = *data;
1148 msr.index = index;
1149 msr.host_initiated = true;
1150 return kvm_set_msr(vcpu, &msr);
Carsten Otte313a3dc2007-10-11 19:16:52 +02001151}
1152
Marcelo Tosatti16e8d742012-11-27 23:29:00 -02001153#ifdef CONFIG_X86_64
1154struct pvclock_gtod_data {
1155 seqcount_t seq;
1156
1157 struct { /* extract of a clocksource struct */
1158 int vclock_mode;
Thomas Gleixnera5a1d1c2016-12-21 20:32:01 +01001159 u64 cycle_last;
1160 u64 mask;
Marcelo Tosatti16e8d742012-11-27 23:29:00 -02001161 u32 mult;
1162 u32 shift;
1163 } clock;
1164
Thomas Gleixnercbcf2dd2014-07-16 21:04:54 +00001165 u64 boot_ns;
1166 u64 nsec_base;
Marcelo Tosatti55dd00a2017-01-24 15:09:39 -02001167 u64 wall_time_sec;
Marcelo Tosatti16e8d742012-11-27 23:29:00 -02001168};
1169
1170static struct pvclock_gtod_data pvclock_gtod_data;
1171
1172static void update_pvclock_gtod(struct timekeeper *tk)
1173{
1174 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
Thomas Gleixnercbcf2dd2014-07-16 21:04:54 +00001175 u64 boot_ns;
1176
Peter Zijlstra876e7882015-03-19 10:09:06 +01001177 boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
Marcelo Tosatti16e8d742012-11-27 23:29:00 -02001178
1179 write_seqcount_begin(&vdata->seq);
1180
1181 /* copy pvclock gtod data */
Peter Zijlstra876e7882015-03-19 10:09:06 +01001182 vdata->clock.vclock_mode = tk->tkr_mono.clock->archdata.vclock_mode;
1183 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
1184 vdata->clock.mask = tk->tkr_mono.mask;
1185 vdata->clock.mult = tk->tkr_mono.mult;
1186 vdata->clock.shift = tk->tkr_mono.shift;
Marcelo Tosatti16e8d742012-11-27 23:29:00 -02001187
Thomas Gleixnercbcf2dd2014-07-16 21:04:54 +00001188 vdata->boot_ns = boot_ns;
Peter Zijlstra876e7882015-03-19 10:09:06 +01001189 vdata->nsec_base = tk->tkr_mono.xtime_nsec;
Marcelo Tosatti16e8d742012-11-27 23:29:00 -02001190
Marcelo Tosatti55dd00a2017-01-24 15:09:39 -02001191 vdata->wall_time_sec = tk->xtime_sec;
1192
Marcelo Tosatti16e8d742012-11-27 23:29:00 -02001193 write_seqcount_end(&vdata->seq);
1194}
1195#endif
1196
Nicholas Krausebab5bb32015-01-01 22:05:18 -05001197void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1198{
1199 /*
1200 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1201 * vcpu_enter_guest. This function is only called from
1202 * the physical CPU that is running vcpu.
1203 */
1204 kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1205}
Marcelo Tosatti16e8d742012-11-27 23:29:00 -02001206
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001207static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1208{
Avi Kivity9ed3c442010-05-04 15:00:37 +03001209 int version;
1210 int r;
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +02001211 struct pvclock_wall_clock wc;
Arnd Bergmann87aeb542016-06-17 17:48:56 +02001212 struct timespec64 boot;
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001213
1214 if (!wall_clock)
1215 return;
1216
Avi Kivity9ed3c442010-05-04 15:00:37 +03001217 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1218 if (r)
1219 return;
1220
1221 if (version & 1)
1222 ++version; /* first time write, random junk */
1223
1224 ++version;
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001225
Nicholas Krause1dab1342015-12-30 13:08:46 -05001226 if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1227 return;
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001228
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +02001229 /*
1230 * The guest calculates current wall clock time by adding
Zachary Amsden34c238a2010-09-18 14:38:14 -10001231 * system time (updated by kvm_guest_time_update below) to the
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +02001232 * wall clock specified here. guest system time equals host
1233 * system time for us, thus we must fill in host boot time here.
1234 */
Arnd Bergmann87aeb542016-06-17 17:48:56 +02001235 getboottime64(&boot);
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +02001236
Bruce Rogers4b648662012-07-20 10:44:24 -06001237 if (kvm->arch.kvmclock_offset) {
Arnd Bergmann87aeb542016-06-17 17:48:56 +02001238 struct timespec64 ts = ns_to_timespec64(kvm->arch.kvmclock_offset);
1239 boot = timespec64_sub(boot, ts);
Bruce Rogers4b648662012-07-20 10:44:24 -06001240 }
Arnd Bergmann87aeb542016-06-17 17:48:56 +02001241 wc.sec = (u32)boot.tv_sec; /* overflow in 2106 guest time */
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +02001242 wc.nsec = boot.tv_nsec;
1243 wc.version = version;
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001244
1245 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1246
1247 version++;
1248 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001249}
1250
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +02001251static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1252{
Paolo Bonzinib51012d2016-01-22 11:39:22 +01001253 do_shl32_div32(dividend, divisor);
1254 return dividend;
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +02001255}
1256
Paolo Bonzini3ae13fa2016-02-08 15:11:15 +01001257static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
Zachary Amsden5f4e3f82010-09-18 14:38:13 -10001258 s8 *pshift, u32 *pmultiplier)
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +02001259{
Zachary Amsden5f4e3f82010-09-18 14:38:13 -10001260 uint64_t scaled64;
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +02001261 int32_t shift = 0;
1262 uint64_t tps64;
1263 uint32_t tps32;
1264
Paolo Bonzini3ae13fa2016-02-08 15:11:15 +01001265 tps64 = base_hz;
1266 scaled64 = scaled_hz;
Jan Kiszka50933622010-09-26 13:00:53 +02001267 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +02001268 tps64 >>= 1;
1269 shift--;
1270 }
1271
1272 tps32 = (uint32_t)tps64;
Jan Kiszka50933622010-09-26 13:00:53 +02001273 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1274 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
Zachary Amsden5f4e3f82010-09-18 14:38:13 -10001275 scaled64 >>= 1;
1276 else
1277 tps32 <<= 1;
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +02001278 shift++;
1279 }
1280
Zachary Amsden5f4e3f82010-09-18 14:38:13 -10001281 *pshift = shift;
1282 *pmultiplier = div_frac(scaled64, tps32);
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +02001283
Paolo Bonzini3ae13fa2016-02-08 15:11:15 +01001284 pr_debug("%s: base_hz %llu => %llu, shift %d, mul %u\n",
1285 __func__, base_hz, scaled_hz, shift, *pmultiplier);
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +02001286}
1287
Marcelo Tosattid8281992012-11-27 23:29:01 -02001288#ifdef CONFIG_X86_64
Marcelo Tosatti16e8d742012-11-27 23:29:00 -02001289static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
Marcelo Tosattid8281992012-11-27 23:29:01 -02001290#endif
Marcelo Tosatti16e8d742012-11-27 23:29:00 -02001291
Gerd Hoffmannc8076602009-02-04 17:52:04 +01001292static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
Fengguang Wu69b00492015-01-19 22:33:39 +08001293static unsigned long max_tsc_khz;
Gerd Hoffmannc8076602009-02-04 17:52:04 +01001294
Zachary Amsdencc578282012-02-03 15:43:50 -02001295static u32 adjust_tsc_khz(u32 khz, s32 ppm)
Zachary Amsdenc2855452010-09-18 14:38:15 -10001296{
Zachary Amsdencc578282012-02-03 15:43:50 -02001297 u64 v = (u64)khz * (1000000 + ppm);
1298 do_div(v, 1000000);
1299 return v;
1300}
1301
Haozhong Zhang381d5852015-10-20 15:39:04 +08001302static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1303{
1304 u64 ratio;
1305
1306 /* Guest TSC same frequency as host TSC? */
1307 if (!scale) {
1308 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1309 return 0;
1310 }
1311
1312 /* TSC scaling supported? */
1313 if (!kvm_has_tsc_control) {
1314 if (user_tsc_khz > tsc_khz) {
1315 vcpu->arch.tsc_catchup = 1;
1316 vcpu->arch.tsc_always_catchup = 1;
1317 return 0;
1318 } else {
1319 WARN(1, "user requested TSC rate below hardware speed\n");
1320 return -1;
1321 }
1322 }
1323
1324 /* TSC scaling required - calculate ratio */
1325 ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1326 user_tsc_khz, tsc_khz);
1327
1328 if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1329 WARN_ONCE(1, "Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1330 user_tsc_khz);
1331 return -1;
1332 }
1333
1334 vcpu->arch.tsc_scaling_ratio = ratio;
1335 return 0;
1336}
1337
Paolo Bonzini4941b8c2016-02-08 14:51:12 +01001338static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
Zachary Amsdencc578282012-02-03 15:43:50 -02001339{
1340 u32 thresh_lo, thresh_hi;
1341 int use_scaling = 0;
1342
Marcelo Tosatti03ba32c2013-03-11 23:10:24 -03001343 /* tsc_khz can be zero if TSC calibration fails */
Paolo Bonzini4941b8c2016-02-08 14:51:12 +01001344 if (user_tsc_khz == 0) {
Haozhong Zhangad721882015-10-20 15:39:02 +08001345 /* set tsc_scaling_ratio to a safe value */
1346 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
Haozhong Zhang381d5852015-10-20 15:39:04 +08001347 return -1;
Haozhong Zhangad721882015-10-20 15:39:02 +08001348 }
Marcelo Tosatti03ba32c2013-03-11 23:10:24 -03001349
Zachary Amsdenc2855452010-09-18 14:38:15 -10001350 /* Compute a scale to convert nanoseconds in TSC cycles */
Paolo Bonzini3ae13fa2016-02-08 15:11:15 +01001351 kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
Zachary Amsdencc578282012-02-03 15:43:50 -02001352 &vcpu->arch.virtual_tsc_shift,
1353 &vcpu->arch.virtual_tsc_mult);
Paolo Bonzini4941b8c2016-02-08 14:51:12 +01001354 vcpu->arch.virtual_tsc_khz = user_tsc_khz;
Zachary Amsdencc578282012-02-03 15:43:50 -02001355
1356 /*
1357 * Compute the variation in TSC rate which is acceptable
1358 * within the range of tolerance and decide if the
1359 * rate being applied is within that bounds of the hardware
1360 * rate. If so, no scaling or compensation need be done.
1361 */
1362 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1363 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
Paolo Bonzini4941b8c2016-02-08 14:51:12 +01001364 if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
1365 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 -02001366 use_scaling = 1;
1367 }
Paolo Bonzini4941b8c2016-02-08 14:51:12 +01001368 return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
Zachary Amsdenc2855452010-09-18 14:38:15 -10001369}
1370
1371static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1372{
Zachary Amsdene26101b2012-02-03 15:43:57 -02001373 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
Zachary Amsdencc578282012-02-03 15:43:50 -02001374 vcpu->arch.virtual_tsc_mult,
1375 vcpu->arch.virtual_tsc_shift);
Zachary Amsdene26101b2012-02-03 15:43:57 -02001376 tsc += vcpu->arch.this_tsc_write;
Zachary Amsdenc2855452010-09-18 14:38:15 -10001377 return tsc;
1378}
1379
Fengguang Wu69b00492015-01-19 22:33:39 +08001380static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
Marcelo Tosattib48aa972012-11-27 23:29:03 -02001381{
1382#ifdef CONFIG_X86_64
1383 bool vcpus_matched;
Marcelo Tosattib48aa972012-11-27 23:29:03 -02001384 struct kvm_arch *ka = &vcpu->kvm->arch;
1385 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1386
1387 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1388 atomic_read(&vcpu->kvm->online_vcpus));
1389
Marcelo Tosatti7f187922014-11-04 21:30:44 -02001390 /*
1391 * Once the masterclock is enabled, always perform request in
1392 * order to update it.
1393 *
1394 * In order to enable masterclock, the host clocksource must be TSC
1395 * and the vcpus need to have matched TSCs. When that happens,
1396 * perform request to enable masterclock.
1397 */
1398 if (ka->use_master_clock ||
1399 (gtod->clock.vclock_mode == VCLOCK_TSC && vcpus_matched))
Marcelo Tosattib48aa972012-11-27 23:29:03 -02001400 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1401
1402 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1403 atomic_read(&vcpu->kvm->online_vcpus),
1404 ka->use_master_clock, gtod->clock.vclock_mode);
1405#endif
1406}
1407
Will Auldba904632012-11-29 12:42:50 -08001408static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1409{
Luiz Capitulino3e3f5022016-09-07 14:47:20 -04001410 u64 curr_offset = vcpu->arch.tsc_offset;
Will Auldba904632012-11-29 12:42:50 -08001411 vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1412}
1413
Haozhong Zhang35181e82015-10-20 15:39:03 +08001414/*
1415 * Multiply tsc by a fixed point number represented by ratio.
1416 *
1417 * The most significant 64-N bits (mult) of ratio represent the
1418 * integral part of the fixed point number; the remaining N bits
1419 * (frac) represent the fractional part, ie. ratio represents a fixed
1420 * point number (mult + frac * 2^(-N)).
1421 *
1422 * N equals to kvm_tsc_scaling_ratio_frac_bits.
1423 */
1424static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1425{
1426 return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1427}
1428
1429u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1430{
1431 u64 _tsc = tsc;
1432 u64 ratio = vcpu->arch.tsc_scaling_ratio;
1433
1434 if (ratio != kvm_default_tsc_scaling_ratio)
1435 _tsc = __scale_tsc(ratio, tsc);
1436
1437 return _tsc;
1438}
1439EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1440
Haozhong Zhang07c14192015-10-20 15:39:05 +08001441static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1442{
1443 u64 tsc;
1444
1445 tsc = kvm_scale_tsc(vcpu, rdtsc());
1446
1447 return target_tsc - tsc;
1448}
1449
Haozhong Zhang4ba76532015-10-20 15:39:07 +08001450u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1451{
Paolo Bonziniea26e4e2016-11-01 00:39:48 +01001452 return vcpu->arch.tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
Haozhong Zhang4ba76532015-10-20 15:39:07 +08001453}
1454EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1455
Luiz Capitulinoa545ab62016-09-07 14:47:19 -04001456static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1457{
1458 kvm_x86_ops->write_tsc_offset(vcpu, offset);
1459 vcpu->arch.tsc_offset = offset;
1460}
1461
Will Auld8fe8ab42012-11-29 12:42:12 -08001462void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
Zachary Amsden99e3e302010-08-19 22:07:17 -10001463{
1464 struct kvm *kvm = vcpu->kvm;
Zachary Amsdenf38e0982010-08-19 22:07:20 -10001465 u64 offset, ns, elapsed;
Zachary Amsden99e3e302010-08-19 22:07:17 -10001466 unsigned long flags;
Marcelo Tosattib48aa972012-11-27 23:29:03 -02001467 bool matched;
Tomasz Grabiec0d3da0d2014-06-24 09:42:43 +02001468 bool already_matched;
Will Auld8fe8ab42012-11-29 12:42:12 -08001469 u64 data = msr->data;
Denis Plotnikovc5e8ec82017-04-07 12:09:52 +03001470 bool synchronizing = false;
Zachary Amsden99e3e302010-08-19 22:07:17 -10001471
Jan Kiszka038f8c12011-02-04 10:49:11 +01001472 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
Haozhong Zhang07c14192015-10-20 15:39:05 +08001473 offset = kvm_compute_tsc_offset(vcpu, data);
Paolo Bonzini108b2492016-09-01 14:21:03 +02001474 ns = ktime_get_boot_ns();
Zachary Amsdenf38e0982010-08-19 22:07:20 -10001475 elapsed = ns - kvm->arch.last_tsc_nsec;
Zachary Amsden5d3cb0f62012-02-03 15:43:51 -02001476
Marcelo Tosatti03ba32c2013-03-11 23:10:24 -03001477 if (vcpu->arch.virtual_tsc_khz) {
Denis Plotnikovbd8fab32017-04-07 12:09:53 +03001478 if (data == 0 && msr->host_initiated) {
1479 /*
1480 * detection of vcpu initialization -- need to sync
1481 * with other vCPUs. This particularly helps to keep
1482 * kvm_clock stable after CPU hotplug
1483 */
1484 synchronizing = true;
1485 } else {
1486 u64 tsc_exp = kvm->arch.last_tsc_write +
1487 nsec_to_cycles(vcpu, elapsed);
1488 u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
1489 /*
1490 * Special case: TSC write with a small delta (1 second)
1491 * of virtual cycle time against real time is
1492 * interpreted as an attempt to synchronize the CPU.
1493 */
1494 synchronizing = data < tsc_exp + tsc_hz &&
1495 data + tsc_hz > tsc_exp;
1496 }
Denis Plotnikovc5e8ec82017-04-07 12:09:52 +03001497 }
Zachary Amsdenf38e0982010-08-19 22:07:20 -10001498
1499 /*
Zachary Amsden5d3cb0f62012-02-03 15:43:51 -02001500 * For a reliable TSC, we can match TSC offsets, and for an unstable
1501 * TSC, we add elapsed time in this computation. We could let the
1502 * compensation code attempt to catch up if we fall behind, but
1503 * it's better to try to match offsets from the beginning.
1504 */
Denis Plotnikovc5e8ec82017-04-07 12:09:52 +03001505 if (synchronizing &&
Zachary Amsden5d3cb0f62012-02-03 15:43:51 -02001506 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
Zachary Amsdenf38e0982010-08-19 22:07:20 -10001507 if (!check_tsc_unstable()) {
Zachary Amsdene26101b2012-02-03 15:43:57 -02001508 offset = kvm->arch.cur_tsc_offset;
Zachary Amsdenf38e0982010-08-19 22:07:20 -10001509 pr_debug("kvm: matched tsc offset for %llu\n", data);
1510 } else {
Joerg Roedel857e4092011-03-25 09:44:50 +01001511 u64 delta = nsec_to_cycles(vcpu, elapsed);
Zachary Amsden5d3cb0f62012-02-03 15:43:51 -02001512 data += delta;
Haozhong Zhang07c14192015-10-20 15:39:05 +08001513 offset = kvm_compute_tsc_offset(vcpu, data);
Zachary Amsden759379d2010-08-19 22:07:25 -10001514 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
Zachary Amsdenf38e0982010-08-19 22:07:20 -10001515 }
Marcelo Tosattib48aa972012-11-27 23:29:03 -02001516 matched = true;
Tomasz Grabiec0d3da0d2014-06-24 09:42:43 +02001517 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
Zachary Amsdene26101b2012-02-03 15:43:57 -02001518 } else {
1519 /*
1520 * We split periods of matched TSC writes into generations.
1521 * For each generation, we track the original measured
1522 * nanosecond time, offset, and write, so if TSCs are in
1523 * sync, we can match exact offset, and if not, we can match
Guo Chao4a969982012-06-28 15:17:27 +08001524 * exact software computation in compute_guest_tsc()
Zachary Amsdene26101b2012-02-03 15:43:57 -02001525 *
1526 * These values are tracked in kvm->arch.cur_xxx variables.
1527 */
1528 kvm->arch.cur_tsc_generation++;
1529 kvm->arch.cur_tsc_nsec = ns;
1530 kvm->arch.cur_tsc_write = data;
1531 kvm->arch.cur_tsc_offset = offset;
Marcelo Tosattib48aa972012-11-27 23:29:03 -02001532 matched = false;
Tomasz Grabiec0d3da0d2014-06-24 09:42:43 +02001533 pr_debug("kvm: new tsc generation %llu, clock %llu\n",
Zachary Amsdene26101b2012-02-03 15:43:57 -02001534 kvm->arch.cur_tsc_generation, data);
Zachary Amsdenf38e0982010-08-19 22:07:20 -10001535 }
Zachary Amsdene26101b2012-02-03 15:43:57 -02001536
1537 /*
1538 * We also track th most recent recorded KHZ, write and time to
1539 * allow the matching interval to be extended at each write.
1540 */
Zachary Amsdenf38e0982010-08-19 22:07:20 -10001541 kvm->arch.last_tsc_nsec = ns;
1542 kvm->arch.last_tsc_write = data;
Zachary Amsden5d3cb0f62012-02-03 15:43:51 -02001543 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
Zachary Amsden99e3e302010-08-19 22:07:17 -10001544
Zachary Amsdenb183aa52012-02-03 15:43:53 -02001545 vcpu->arch.last_guest_tsc = data;
Zachary Amsdene26101b2012-02-03 15:43:57 -02001546
1547 /* Keep track of which generation this VCPU has synchronized to */
1548 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1549 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1550 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1551
Radim Krčmářd6321d42017-08-05 00:12:49 +02001552 if (!msr->host_initiated && guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST))
Will Auldba904632012-11-29 12:42:50 -08001553 update_ia32_tsc_adjust_msr(vcpu, offset);
Radim Krčmářd6321d42017-08-05 00:12:49 +02001554
Luiz Capitulinoa545ab62016-09-07 14:47:19 -04001555 kvm_vcpu_write_tsc_offset(vcpu, offset);
Zachary Amsdene26101b2012-02-03 15:43:57 -02001556 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
Marcelo Tosattib48aa972012-11-27 23:29:03 -02001557
1558 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
Tomasz Grabiec0d3da0d2014-06-24 09:42:43 +02001559 if (!matched) {
Marcelo Tosattib48aa972012-11-27 23:29:03 -02001560 kvm->arch.nr_vcpus_matched_tsc = 0;
Tomasz Grabiec0d3da0d2014-06-24 09:42:43 +02001561 } else if (!already_matched) {
1562 kvm->arch.nr_vcpus_matched_tsc++;
1563 }
Marcelo Tosattib48aa972012-11-27 23:29:03 -02001564
1565 kvm_track_tsc_matching(vcpu);
1566 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
Zachary Amsden99e3e302010-08-19 22:07:17 -10001567}
Zachary Amsdene26101b2012-02-03 15:43:57 -02001568
Zachary Amsden99e3e302010-08-19 22:07:17 -10001569EXPORT_SYMBOL_GPL(kvm_write_tsc);
1570
Haozhong Zhang58ea6762015-10-20 15:39:06 +08001571static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
1572 s64 adjustment)
1573{
Paolo Bonziniea26e4e2016-11-01 00:39:48 +01001574 kvm_vcpu_write_tsc_offset(vcpu, vcpu->arch.tsc_offset + adjustment);
Haozhong Zhang58ea6762015-10-20 15:39:06 +08001575}
1576
1577static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
1578{
1579 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
1580 WARN_ON(adjustment < 0);
1581 adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
Paolo Bonziniea26e4e2016-11-01 00:39:48 +01001582 adjust_tsc_offset_guest(vcpu, adjustment);
Haozhong Zhang58ea6762015-10-20 15:39:06 +08001583}
1584
Marcelo Tosattid8281992012-11-27 23:29:01 -02001585#ifdef CONFIG_X86_64
1586
Thomas Gleixnera5a1d1c2016-12-21 20:32:01 +01001587static u64 read_tsc(void)
Marcelo Tosattid8281992012-11-27 23:29:01 -02001588{
Thomas Gleixnera5a1d1c2016-12-21 20:32:01 +01001589 u64 ret = (u64)rdtsc_ordered();
Andy Lutomirski03b97302015-06-25 18:44:08 +02001590 u64 last = pvclock_gtod_data.clock.cycle_last;
Marcelo Tosattid8281992012-11-27 23:29:01 -02001591
1592 if (likely(ret >= last))
1593 return ret;
1594
1595 /*
1596 * GCC likes to generate cmov here, but this branch is extremely
Adam Buchbinder6a6256f2016-02-23 15:34:30 -08001597 * predictable (it's just a function of time and the likely is
Marcelo Tosattid8281992012-11-27 23:29:01 -02001598 * very likely) and there's a data dependence, so force GCC
1599 * to generate a branch instead. I don't barrier() because
1600 * we don't actually need a barrier, and if this function
1601 * ever gets inlined it will generate worse code.
1602 */
1603 asm volatile ("");
1604 return last;
1605}
1606
Thomas Gleixnera5a1d1c2016-12-21 20:32:01 +01001607static inline u64 vgettsc(u64 *cycle_now)
Marcelo Tosattid8281992012-11-27 23:29:01 -02001608{
1609 long v;
1610 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1611
1612 *cycle_now = read_tsc();
1613
1614 v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask;
1615 return v * gtod->clock.mult;
1616}
1617
Thomas Gleixnera5a1d1c2016-12-21 20:32:01 +01001618static int do_monotonic_boot(s64 *t, u64 *cycle_now)
Marcelo Tosattid8281992012-11-27 23:29:01 -02001619{
Marcelo Tosattid8281992012-11-27 23:29:01 -02001620 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
Thomas Gleixnercbcf2dd2014-07-16 21:04:54 +00001621 unsigned long seq;
1622 int mode;
1623 u64 ns;
Marcelo Tosattid8281992012-11-27 23:29:01 -02001624
Marcelo Tosattid8281992012-11-27 23:29:01 -02001625 do {
1626 seq = read_seqcount_begin(&gtod->seq);
1627 mode = gtod->clock.vclock_mode;
Thomas Gleixnercbcf2dd2014-07-16 21:04:54 +00001628 ns = gtod->nsec_base;
Marcelo Tosattid8281992012-11-27 23:29:01 -02001629 ns += vgettsc(cycle_now);
1630 ns >>= gtod->clock.shift;
Thomas Gleixnercbcf2dd2014-07-16 21:04:54 +00001631 ns += gtod->boot_ns;
Marcelo Tosattid8281992012-11-27 23:29:01 -02001632 } while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
Thomas Gleixnercbcf2dd2014-07-16 21:04:54 +00001633 *t = ns;
Marcelo Tosattid8281992012-11-27 23:29:01 -02001634
1635 return mode;
1636}
1637
Marcelo Tosatti55dd00a2017-01-24 15:09:39 -02001638static int do_realtime(struct timespec *ts, u64 *cycle_now)
1639{
1640 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1641 unsigned long seq;
1642 int mode;
1643 u64 ns;
1644
1645 do {
1646 seq = read_seqcount_begin(&gtod->seq);
1647 mode = gtod->clock.vclock_mode;
1648 ts->tv_sec = gtod->wall_time_sec;
1649 ns = gtod->nsec_base;
1650 ns += vgettsc(cycle_now);
1651 ns >>= gtod->clock.shift;
1652 } while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
1653
1654 ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
1655 ts->tv_nsec = ns;
1656
1657 return mode;
1658}
1659
Marcelo Tosattid8281992012-11-27 23:29:01 -02001660/* returns true if host is using tsc clocksource */
Thomas Gleixnera5a1d1c2016-12-21 20:32:01 +01001661static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *cycle_now)
Marcelo Tosattid8281992012-11-27 23:29:01 -02001662{
Marcelo Tosattid8281992012-11-27 23:29:01 -02001663 /* checked again under seqlock below */
1664 if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1665 return false;
1666
Thomas Gleixnercbcf2dd2014-07-16 21:04:54 +00001667 return do_monotonic_boot(kernel_ns, cycle_now) == VCLOCK_TSC;
Marcelo Tosattid8281992012-11-27 23:29:01 -02001668}
Marcelo Tosatti55dd00a2017-01-24 15:09:39 -02001669
1670/* returns true if host is using tsc clocksource */
1671static bool kvm_get_walltime_and_clockread(struct timespec *ts,
1672 u64 *cycle_now)
1673{
1674 /* checked again under seqlock below */
1675 if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1676 return false;
1677
1678 return do_realtime(ts, cycle_now) == VCLOCK_TSC;
1679}
Marcelo Tosattid8281992012-11-27 23:29:01 -02001680#endif
1681
1682/*
1683 *
Marcelo Tosattib48aa972012-11-27 23:29:03 -02001684 * Assuming a stable TSC across physical CPUS, and a stable TSC
1685 * across virtual CPUs, the following condition is possible.
1686 * Each numbered line represents an event visible to both
Marcelo Tosattid8281992012-11-27 23:29:01 -02001687 * CPUs at the next numbered event.
1688 *
1689 * "timespecX" represents host monotonic time. "tscX" represents
1690 * RDTSC value.
1691 *
1692 * VCPU0 on CPU0 | VCPU1 on CPU1
1693 *
1694 * 1. read timespec0,tsc0
1695 * 2. | timespec1 = timespec0 + N
1696 * | tsc1 = tsc0 + M
1697 * 3. transition to guest | transition to guest
1698 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1699 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
1700 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1701 *
1702 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1703 *
1704 * - ret0 < ret1
1705 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1706 * ...
1707 * - 0 < N - M => M < N
1708 *
1709 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1710 * always the case (the difference between two distinct xtime instances
1711 * might be smaller then the difference between corresponding TSC reads,
1712 * when updating guest vcpus pvclock areas).
1713 *
1714 * To avoid that problem, do not allow visibility of distinct
1715 * system_timestamp/tsc_timestamp values simultaneously: use a master
1716 * copy of host monotonic time values. Update that master copy
1717 * in lockstep.
1718 *
Marcelo Tosattib48aa972012-11-27 23:29:03 -02001719 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
Marcelo Tosattid8281992012-11-27 23:29:01 -02001720 *
1721 */
1722
1723static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1724{
1725#ifdef CONFIG_X86_64
1726 struct kvm_arch *ka = &kvm->arch;
1727 int vclock_mode;
Marcelo Tosattib48aa972012-11-27 23:29:03 -02001728 bool host_tsc_clocksource, vcpus_matched;
1729
1730 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1731 atomic_read(&kvm->online_vcpus));
Marcelo Tosattid8281992012-11-27 23:29:01 -02001732
1733 /*
1734 * If the host uses TSC clock, then passthrough TSC as stable
1735 * to the guest.
1736 */
Marcelo Tosattib48aa972012-11-27 23:29:03 -02001737 host_tsc_clocksource = kvm_get_time_and_clockread(
Marcelo Tosattid8281992012-11-27 23:29:01 -02001738 &ka->master_kernel_ns,
1739 &ka->master_cycle_now);
1740
Marcelo Tosatti16a96022014-05-14 12:43:24 -03001741 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
Ladi Proseka826faf2017-06-26 09:56:43 +02001742 && !ka->backwards_tsc_observed
Marcelo Tosatti54750f22015-01-20 15:54:52 -02001743 && !ka->boot_vcpu_runs_old_kvmclock;
Marcelo Tosattib48aa972012-11-27 23:29:03 -02001744
Marcelo Tosattid8281992012-11-27 23:29:01 -02001745 if (ka->use_master_clock)
1746 atomic_set(&kvm_guest_has_master_clock, 1);
1747
1748 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
Marcelo Tosattib48aa972012-11-27 23:29:03 -02001749 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1750 vcpus_matched);
Marcelo Tosattid8281992012-11-27 23:29:01 -02001751#endif
1752}
1753
Paolo Bonzini2860c4b2016-01-07 15:05:10 +01001754void kvm_make_mclock_inprogress_request(struct kvm *kvm)
1755{
1756 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
1757}
1758
Marcelo Tosatti2e762ff2013-08-27 23:55:29 -03001759static void kvm_gen_update_masterclock(struct kvm *kvm)
1760{
1761#ifdef CONFIG_X86_64
1762 int i;
1763 struct kvm_vcpu *vcpu;
1764 struct kvm_arch *ka = &kvm->arch;
1765
1766 spin_lock(&ka->pvclock_gtod_sync_lock);
1767 kvm_make_mclock_inprogress_request(kvm);
1768 /* no guest entries from this point */
1769 pvclock_update_vm_gtod_copy(kvm);
1770
1771 kvm_for_each_vcpu(i, vcpu, kvm)
Guo Hui Liu105b21b2014-09-12 13:43:19 +08001772 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
Marcelo Tosatti2e762ff2013-08-27 23:55:29 -03001773
1774 /* guest entries allowed */
1775 kvm_for_each_vcpu(i, vcpu, kvm)
Radim Krčmář72875d8a2017-04-26 22:32:19 +02001776 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
Marcelo Tosatti2e762ff2013-08-27 23:55:29 -03001777
1778 spin_unlock(&ka->pvclock_gtod_sync_lock);
1779#endif
1780}
1781
Marcelo Tosattie891a322017-04-17 12:51:37 -03001782u64 get_kvmclock_ns(struct kvm *kvm)
Paolo Bonzini108b2492016-09-01 14:21:03 +02001783{
Paolo Bonzini108b2492016-09-01 14:21:03 +02001784 struct kvm_arch *ka = &kvm->arch;
Paolo Bonzini8b953442016-11-16 18:31:30 +01001785 struct pvclock_vcpu_time_info hv_clock;
Wanpeng Lie2c22062017-05-11 18:12:05 -07001786 u64 ret;
Paolo Bonzini108b2492016-09-01 14:21:03 +02001787
Paolo Bonzini8b953442016-11-16 18:31:30 +01001788 spin_lock(&ka->pvclock_gtod_sync_lock);
1789 if (!ka->use_master_clock) {
1790 spin_unlock(&ka->pvclock_gtod_sync_lock);
1791 return ktime_get_boot_ns() + ka->kvmclock_offset;
Paolo Bonzini108b2492016-09-01 14:21:03 +02001792 }
1793
Paolo Bonzini8b953442016-11-16 18:31:30 +01001794 hv_clock.tsc_timestamp = ka->master_cycle_now;
1795 hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
1796 spin_unlock(&ka->pvclock_gtod_sync_lock);
1797
Wanpeng Lie2c22062017-05-11 18:12:05 -07001798 /* both __this_cpu_read() and rdtsc() should be on the same cpu */
1799 get_cpu();
1800
Wanpeng Lie70b57a2017-11-20 14:55:05 -08001801 if (__this_cpu_read(cpu_tsc_khz)) {
1802 kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
1803 &hv_clock.tsc_shift,
1804 &hv_clock.tsc_to_system_mul);
1805 ret = __pvclock_read_cycles(&hv_clock, rdtsc());
1806 } else
1807 ret = ktime_get_boot_ns() + ka->kvmclock_offset;
Wanpeng Lie2c22062017-05-11 18:12:05 -07001808
1809 put_cpu();
1810
1811 return ret;
Paolo Bonzini108b2492016-09-01 14:21:03 +02001812}
1813
Paolo Bonzini0d6dd2f2016-09-01 14:20:09 +02001814static void kvm_setup_pvclock_page(struct kvm_vcpu *v)
1815{
1816 struct kvm_vcpu_arch *vcpu = &v->arch;
1817 struct pvclock_vcpu_time_info guest_hv_clock;
1818
Paolo Bonzini4e335d92017-05-02 16:20:18 +02001819 if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
Paolo Bonzini0d6dd2f2016-09-01 14:20:09 +02001820 &guest_hv_clock, sizeof(guest_hv_clock))))
1821 return;
1822
1823 /* This VCPU is paused, but it's legal for a guest to read another
1824 * VCPU's kvmclock, so we really have to follow the specification where
1825 * it says that version is odd if data is being modified, and even after
1826 * it is consistent.
1827 *
1828 * Version field updates must be kept separate. This is because
1829 * kvm_write_guest_cached might use a "rep movs" instruction, and
1830 * writes within a string instruction are weakly ordered. So there
1831 * are three writes overall.
1832 *
1833 * As a small optimization, only write the version field in the first
1834 * and third write. The vcpu->pv_time cache is still valid, because the
1835 * version field is the first in the struct.
1836 */
1837 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
1838
Liran Alon51c4b8b2017-11-05 16:11:30 +02001839 if (guest_hv_clock.version & 1)
1840 ++guest_hv_clock.version; /* first time write, random junk */
1841
Paolo Bonzini0d6dd2f2016-09-01 14:20:09 +02001842 vcpu->hv_clock.version = guest_hv_clock.version + 1;
Paolo Bonzini4e335d92017-05-02 16:20:18 +02001843 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1844 &vcpu->hv_clock,
1845 sizeof(vcpu->hv_clock.version));
Paolo Bonzini0d6dd2f2016-09-01 14:20:09 +02001846
1847 smp_wmb();
1848
1849 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1850 vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
1851
1852 if (vcpu->pvclock_set_guest_stopped_request) {
1853 vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
1854 vcpu->pvclock_set_guest_stopped_request = false;
1855 }
1856
1857 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
1858
Paolo Bonzini4e335d92017-05-02 16:20:18 +02001859 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1860 &vcpu->hv_clock,
1861 sizeof(vcpu->hv_clock));
Paolo Bonzini0d6dd2f2016-09-01 14:20:09 +02001862
1863 smp_wmb();
1864
1865 vcpu->hv_clock.version++;
Paolo Bonzini4e335d92017-05-02 16:20:18 +02001866 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1867 &vcpu->hv_clock,
1868 sizeof(vcpu->hv_clock.version));
Paolo Bonzini0d6dd2f2016-09-01 14:20:09 +02001869}
1870
Zachary Amsden34c238a2010-09-18 14:38:14 -10001871static int kvm_guest_time_update(struct kvm_vcpu *v)
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001872{
Paolo Bonzini78db6a52016-02-08 14:51:40 +01001873 unsigned long flags, tgt_tsc_khz;
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001874 struct kvm_vcpu_arch *vcpu = &v->arch;
Marcelo Tosattid8281992012-11-27 23:29:01 -02001875 struct kvm_arch *ka = &v->kvm->arch;
Marcelo Tosattif25e6562014-01-06 12:18:59 -02001876 s64 kernel_ns;
Marcelo Tosattid8281992012-11-27 23:29:01 -02001877 u64 tsc_timestamp, host_tsc;
Marcelo Tosatti51d59c62012-08-03 15:57:49 -03001878 u8 pvclock_flags;
Marcelo Tosattid8281992012-11-27 23:29:01 -02001879 bool use_master_clock;
1880
1881 kernel_ns = 0;
1882 host_tsc = 0;
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +02001883
Zachary Amsden1d5f0662010-08-19 22:07:30 -10001884 /*
Marcelo Tosattid8281992012-11-27 23:29:01 -02001885 * If the host uses TSC clock, then passthrough TSC as stable
1886 * to the guest.
1887 */
1888 spin_lock(&ka->pvclock_gtod_sync_lock);
1889 use_master_clock = ka->use_master_clock;
1890 if (use_master_clock) {
1891 host_tsc = ka->master_cycle_now;
1892 kernel_ns = ka->master_kernel_ns;
1893 }
1894 spin_unlock(&ka->pvclock_gtod_sync_lock);
Marcelo Tosattic09664b2013-03-18 13:54:32 -03001895
1896 /* Keep irq disabled to prevent changes to the clock */
1897 local_irq_save(flags);
Paolo Bonzini78db6a52016-02-08 14:51:40 +01001898 tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
1899 if (unlikely(tgt_tsc_khz == 0)) {
Marcelo Tosattic09664b2013-03-18 13:54:32 -03001900 local_irq_restore(flags);
1901 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1902 return 1;
1903 }
Marcelo Tosattid8281992012-11-27 23:29:01 -02001904 if (!use_master_clock) {
Andy Lutomirski4ea16362015-06-25 18:44:07 +02001905 host_tsc = rdtsc();
Paolo Bonzini108b2492016-09-01 14:21:03 +02001906 kernel_ns = ktime_get_boot_ns();
Marcelo Tosattid8281992012-11-27 23:29:01 -02001907 }
1908
Haozhong Zhang4ba76532015-10-20 15:39:07 +08001909 tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
Marcelo Tosattid8281992012-11-27 23:29:01 -02001910
1911 /*
Zachary Amsdenc2855452010-09-18 14:38:15 -10001912 * We may have to catch up the TSC to match elapsed wall clock
1913 * time for two reasons, even if kvmclock is used.
1914 * 1) CPU could have been running below the maximum TSC rate
1915 * 2) Broken TSC compensation resets the base at each VCPU
1916 * entry to avoid unknown leaps of TSC even when running
1917 * again on the same CPU. This may cause apparent elapsed
1918 * time to disappear, and the guest to stand still or run
1919 * very slowly.
1920 */
1921 if (vcpu->tsc_catchup) {
1922 u64 tsc = compute_guest_tsc(v, kernel_ns);
1923 if (tsc > tsc_timestamp) {
Marcelo Tosattif1e2b262012-02-03 15:43:55 -02001924 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
Zachary Amsdenc2855452010-09-18 14:38:15 -10001925 tsc_timestamp = tsc;
1926 }
1927 }
1928
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001929 local_irq_restore(flags);
1930
Paolo Bonzini0d6dd2f2016-09-01 14:20:09 +02001931 /* With all the info we got, fill in the values */
Zachary Amsdenc2855452010-09-18 14:38:15 -10001932
Paolo Bonzini78db6a52016-02-08 14:51:40 +01001933 if (kvm_has_tsc_control)
1934 tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
1935
1936 if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
Paolo Bonzini3ae13fa2016-02-08 15:11:15 +01001937 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
Zachary Amsden5f4e3f82010-09-18 14:38:13 -10001938 &vcpu->hv_clock.tsc_shift,
1939 &vcpu->hv_clock.tsc_to_system_mul);
Paolo Bonzini78db6a52016-02-08 14:51:40 +01001940 vcpu->hw_tsc_khz = tgt_tsc_khz;
Zachary Amsden8cfdc002010-08-19 22:07:21 -10001941 }
1942
Zachary Amsden1d5f0662010-08-19 22:07:30 -10001943 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
Zachary Amsden759379d2010-08-19 22:07:25 -10001944 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
Zachary Amsden28e46392010-09-18 14:38:12 -10001945 vcpu->last_guest_tsc = tsc_timestamp;
Marcelo Tosatti51d59c62012-08-03 15:57:49 -03001946
Marcelo Tosattid8281992012-11-27 23:29:01 -02001947 /* If the host uses TSC clocksource, then it is stable */
Paolo Bonzini0d6dd2f2016-09-01 14:20:09 +02001948 pvclock_flags = 0;
Marcelo Tosattid8281992012-11-27 23:29:01 -02001949 if (use_master_clock)
1950 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
1951
Marcelo Tosatti78c03372012-11-27 23:28:47 -02001952 vcpu->hv_clock.flags = pvclock_flags;
1953
Paolo Bonzini095cf552016-02-08 12:54:12 +01001954 if (vcpu->pv_time_enabled)
1955 kvm_setup_pvclock_page(v);
1956 if (v == kvm_get_vcpu(v->kvm, 0))
1957 kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
Zachary Amsden8cfdc002010-08-19 22:07:21 -10001958 return 0;
Gerd Hoffmannc8076602009-02-04 17:52:04 +01001959}
1960
Marcelo Tosatti0061d532013-05-09 20:21:41 -03001961/*
1962 * kvmclock updates which are isolated to a given vcpu, such as
1963 * vcpu->cpu migration, should not allow system_timestamp from
1964 * the rest of the vcpus to remain static. Otherwise ntp frequency
1965 * correction applies to one vcpu's system_timestamp but not
1966 * the others.
1967 *
1968 * So in those cases, request a kvmclock update for all vcpus.
Andrew Jones7e44e442014-02-28 12:52:54 +01001969 * We need to rate-limit these requests though, as they can
1970 * considerably slow guests that have a large number of vcpus.
1971 * The time for a remote vcpu to update its kvmclock is bound
1972 * by the delay we use to rate-limit the updates.
Marcelo Tosatti0061d532013-05-09 20:21:41 -03001973 */
1974
Andrew Jones7e44e442014-02-28 12:52:54 +01001975#define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
1976
1977static void kvmclock_update_fn(struct work_struct *work)
Marcelo Tosatti0061d532013-05-09 20:21:41 -03001978{
1979 int i;
Andrew Jones7e44e442014-02-28 12:52:54 +01001980 struct delayed_work *dwork = to_delayed_work(work);
1981 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1982 kvmclock_update_work);
1983 struct kvm *kvm = container_of(ka, struct kvm, arch);
Marcelo Tosatti0061d532013-05-09 20:21:41 -03001984 struct kvm_vcpu *vcpu;
1985
1986 kvm_for_each_vcpu(i, vcpu, kvm) {
Guo Hui Liu105b21b2014-09-12 13:43:19 +08001987 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
Marcelo Tosatti0061d532013-05-09 20:21:41 -03001988 kvm_vcpu_kick(vcpu);
1989 }
1990}
1991
Andrew Jones7e44e442014-02-28 12:52:54 +01001992static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
1993{
1994 struct kvm *kvm = v->kvm;
1995
Guo Hui Liu105b21b2014-09-12 13:43:19 +08001996 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
Andrew Jones7e44e442014-02-28 12:52:54 +01001997 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
1998 KVMCLOCK_UPDATE_DELAY);
1999}
2000
Andrew Jones332967a2014-02-28 12:52:55 +01002001#define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2002
2003static void kvmclock_sync_fn(struct work_struct *work)
2004{
2005 struct delayed_work *dwork = to_delayed_work(work);
2006 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2007 kvmclock_sync_work);
2008 struct kvm *kvm = container_of(ka, struct kvm, arch);
2009
Marcelo Tosatti630994b2015-05-12 22:42:04 -03002010 if (!kvmclock_periodic_sync)
2011 return;
2012
Andrew Jones332967a2014-02-28 12:52:55 +01002013 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
2014 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
2015 KVMCLOCK_SYNC_PERIOD);
2016}
2017
Wanpeng Li9ffd9862017-10-19 06:47:56 -07002018static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
Huang Ying890ca9a2009-05-11 16:48:15 +08002019{
2020 u64 mcg_cap = vcpu->arch.mcg_cap;
2021 unsigned bank_num = mcg_cap & 0xff;
Wanpeng Li9ffd9862017-10-19 06:47:56 -07002022 u32 msr = msr_info->index;
2023 u64 data = msr_info->data;
Huang Ying890ca9a2009-05-11 16:48:15 +08002024
2025 switch (msr) {
2026 case MSR_IA32_MCG_STATUS:
2027 vcpu->arch.mcg_status = data;
2028 break;
2029 case MSR_IA32_MCG_CTL:
2030 if (!(mcg_cap & MCG_CTL_P))
2031 return 1;
2032 if (data != 0 && data != ~(u64)0)
2033 return -1;
2034 vcpu->arch.mcg_ctl = data;
2035 break;
2036 default:
2037 if (msr >= MSR_IA32_MC0_CTL &&
Chen Yucong81760dc2014-09-23 10:44:35 +08002038 msr < MSR_IA32_MCx_CTL(bank_num)) {
Huang Ying890ca9a2009-05-11 16:48:15 +08002039 u32 offset = msr - MSR_IA32_MC0_CTL;
Andre Przywara114be422010-03-24 17:46:42 +01002040 /* only 0 or all 1s can be written to IA32_MCi_CTL
2041 * some Linux kernels though clear bit 10 in bank 4 to
2042 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2043 * this to avoid an uncatched #GP in the guest
2044 */
Huang Ying890ca9a2009-05-11 16:48:15 +08002045 if ((offset & 0x3) == 0 &&
Andre Przywara114be422010-03-24 17:46:42 +01002046 data != 0 && (data | (1 << 10)) != ~(u64)0)
Huang Ying890ca9a2009-05-11 16:48:15 +08002047 return -1;
Wanpeng Li9ffd9862017-10-19 06:47:56 -07002048 if (!msr_info->host_initiated &&
2049 (offset & 0x3) == 1 && data != 0)
2050 return -1;
Huang Ying890ca9a2009-05-11 16:48:15 +08002051 vcpu->arch.mce_banks[offset] = data;
2052 break;
2053 }
2054 return 1;
2055 }
2056 return 0;
2057}
2058
Ed Swierkffde22a2009-10-15 15:21:43 -07002059static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
2060{
2061 struct kvm *kvm = vcpu->kvm;
2062 int lm = is_long_mode(vcpu);
2063 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
2064 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
2065 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
2066 : kvm->arch.xen_hvm_config.blob_size_32;
2067 u32 page_num = data & ~PAGE_MASK;
2068 u64 page_addr = data & PAGE_MASK;
2069 u8 *page;
2070 int r;
2071
2072 r = -E2BIG;
2073 if (page_num >= blob_size)
2074 goto out;
2075 r = -ENOMEM;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002076 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
2077 if (IS_ERR(page)) {
2078 r = PTR_ERR(page);
Ed Swierkffde22a2009-10-15 15:21:43 -07002079 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002080 }
Paolo Bonzini54bf36a2015-04-08 15:39:23 +02002081 if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
Ed Swierkffde22a2009-10-15 15:21:43 -07002082 goto out_free;
2083 r = 0;
2084out_free:
2085 kfree(page);
2086out:
2087 return r;
2088}
2089
Gleb Natapov344d9582010-10-14 11:22:50 +02002090static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2091{
2092 gpa_t gpa = data & ~0x3f;
2093
Wanpeng Li52a5c152017-07-13 18:30:42 -07002094 /* Bits 3:5 are reserved, Should be zero */
2095 if (data & 0x38)
Gleb Natapov344d9582010-10-14 11:22:50 +02002096 return 1;
2097
2098 vcpu->arch.apf.msr_val = data;
2099
2100 if (!(data & KVM_ASYNC_PF_ENABLED)) {
2101 kvm_clear_async_pf_completion_queue(vcpu);
2102 kvm_async_pf_hash_reset(vcpu);
2103 return 0;
2104 }
2105
Paolo Bonzini4e335d92017-05-02 16:20:18 +02002106 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
Andrew Honig8f964522013-03-29 09:35:21 -07002107 sizeof(u32)))
Gleb Natapov344d9582010-10-14 11:22:50 +02002108 return 1;
2109
Gleb Natapov6adba522010-10-14 11:22:55 +02002110 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
Wanpeng Li52a5c152017-07-13 18:30:42 -07002111 vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
Gleb Natapov344d9582010-10-14 11:22:50 +02002112 kvm_async_pf_wakeup_all(vcpu);
2113 return 0;
2114}
2115
Glauber Costa12f9a482011-02-01 14:16:40 -05002116static void kvmclock_reset(struct kvm_vcpu *vcpu)
2117{
Andy Honig0b794592013-02-20 14:48:10 -08002118 vcpu->arch.pv_time_enabled = false;
Glauber Costa12f9a482011-02-01 14:16:40 -05002119}
2120
Glauber Costac9aaa892011-07-11 15:28:14 -04002121static void record_steal_time(struct kvm_vcpu *vcpu)
2122{
2123 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2124 return;
2125
Paolo Bonzini4e335d92017-05-02 16:20:18 +02002126 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
Glauber Costac9aaa892011-07-11 15:28:14 -04002127 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2128 return;
2129
Pan Xinhui0b9f6c42016-11-02 05:08:35 -04002130 vcpu->arch.st.steal.preempted = 0;
2131
Wanpeng Li35f3fae12016-05-03 11:43:10 +08002132 if (vcpu->arch.st.steal.version & 1)
2133 vcpu->arch.st.steal.version += 1; /* first time write, random junk */
2134
2135 vcpu->arch.st.steal.version += 1;
2136
Paolo Bonzini4e335d92017-05-02 16:20:18 +02002137 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
Wanpeng Li35f3fae12016-05-03 11:43:10 +08002138 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2139
2140 smp_wmb();
2141
Liang Chenc54cdf12016-03-16 19:33:16 +08002142 vcpu->arch.st.steal.steal += current->sched_info.run_delay -
2143 vcpu->arch.st.last_steal;
2144 vcpu->arch.st.last_steal = current->sched_info.run_delay;
Wanpeng Li35f3fae12016-05-03 11:43:10 +08002145
Paolo Bonzini4e335d92017-05-02 16:20:18 +02002146 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
Wanpeng Li35f3fae12016-05-03 11:43:10 +08002147 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2148
2149 smp_wmb();
2150
2151 vcpu->arch.st.steal.version += 1;
Glauber Costac9aaa892011-07-11 15:28:14 -04002152
Paolo Bonzini4e335d92017-05-02 16:20:18 +02002153 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
Glauber Costac9aaa892011-07-11 15:28:14 -04002154 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2155}
2156
Will Auld8fe8ab42012-11-29 12:42:12 -08002157int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
Carsten Otte15c4a642007-10-30 18:44:17 +01002158{
Gleb Natapov57537852012-01-15 14:17:22 +02002159 bool pr = false;
Will Auld8fe8ab42012-11-29 12:42:12 -08002160 u32 msr = msr_info->index;
2161 u64 data = msr_info->data;
Gleb Natapov57537852012-01-15 14:17:22 +02002162
Carsten Otte15c4a642007-10-30 18:44:17 +01002163 switch (msr) {
Borislav Petkov2e32b712013-02-19 19:33:13 +01002164 case MSR_AMD64_NB_CFG:
2165 case MSR_IA32_UCODE_REV:
2166 case MSR_IA32_UCODE_WRITE:
2167 case MSR_VM_HSAVE_PA:
2168 case MSR_AMD64_PATCH_LOADER:
2169 case MSR_AMD64_BU_CFG2:
Ladi Prosek405a3532017-04-06 15:22:20 +02002170 case MSR_AMD64_DC_CFG:
Borislav Petkov2e32b712013-02-19 19:33:13 +01002171 break;
2172
Carsten Otte15c4a642007-10-30 18:44:17 +01002173 case MSR_EFER:
Roedel, Joergb69e8ca2010-05-06 11:38:43 +02002174 return set_efer(vcpu, data);
Andre Przywara8f1589d2009-06-24 12:44:33 +02002175 case MSR_K7_HWCR:
2176 data &= ~(u64)0x40; /* ignore flush filter disable */
Joerg Roedel82494022010-02-24 18:59:16 +01002177 data &= ~(u64)0x100; /* ignore ignne emulation enable */
Nicolae Mogoreanua223c312012-02-21 13:44:21 -08002178 data &= ~(u64)0x8; /* ignore TLB cache disable */
Matthias Lange22d48b2d2014-06-26 13:50:15 +02002179 data &= ~(u64)0x40000; /* ignore Mc status write enable */
Andre Przywara8f1589d2009-06-24 12:44:33 +02002180 if (data != 0) {
Christoffer Dalla737f252012-06-03 21:17:48 +03002181 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2182 data);
Andre Przywara8f1589d2009-06-24 12:44:33 +02002183 return 1;
2184 }
Carsten Otte15c4a642007-10-30 18:44:17 +01002185 break;
Andre Przywaraf7c6d142009-07-02 15:04:14 +02002186 case MSR_FAM10H_MMIO_CONF_BASE:
2187 if (data != 0) {
Christoffer Dalla737f252012-06-03 21:17:48 +03002188 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2189 "0x%llx\n", data);
Andre Przywaraf7c6d142009-07-02 15:04:14 +02002190 return 1;
2191 }
Carsten Otte15c4a642007-10-30 18:44:17 +01002192 break;
Alexander Grafb5e2fec2008-07-22 08:00:45 +02002193 case MSR_IA32_DEBUGCTLMSR:
2194 if (!data) {
2195 /* We support the non-activated case already */
2196 break;
2197 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2198 /* Values other than LBR and BTF are vendor-specific,
2199 thus reserved and should throw a #GP */
2200 return 1;
2201 }
Christoffer Dalla737f252012-06-03 21:17:48 +03002202 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2203 __func__, data);
Alexander Grafb5e2fec2008-07-22 08:00:45 +02002204 break;
Avi Kivity9ba075a2008-05-26 20:06:35 +03002205 case 0x200 ... 0x2ff:
Xiao Guangrongff536042015-06-15 16:55:22 +08002206 return kvm_mtrr_set_msr(vcpu, msr, data);
Carsten Otte15c4a642007-10-30 18:44:17 +01002207 case MSR_IA32_APICBASE:
Jan Kiszka58cb6282014-01-24 16:48:44 +01002208 return kvm_set_apic_base(vcpu, msr_info);
Gleb Natapov0105d1a2009-07-05 17:39:36 +03002209 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2210 return kvm_x2apic_msr_write(vcpu, msr, data);
Liu, Jinsonga3e06bb2011-09-22 16:55:52 +08002211 case MSR_IA32_TSCDEADLINE:
2212 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2213 break;
Will Auldba904632012-11-29 12:42:50 -08002214 case MSR_IA32_TSC_ADJUST:
Radim Krčmářd6321d42017-08-05 00:12:49 +02002215 if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
Will Auldba904632012-11-29 12:42:50 -08002216 if (!msr_info->host_initiated) {
Chris J Argesd913b902014-11-12 21:00:39 -06002217 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
Haozhong Zhangd7add052015-08-07 11:24:32 +08002218 adjust_tsc_offset_guest(vcpu, adj);
Will Auldba904632012-11-29 12:42:50 -08002219 }
2220 vcpu->arch.ia32_tsc_adjust_msr = data;
2221 }
2222 break;
Carsten Otte15c4a642007-10-30 18:44:17 +01002223 case MSR_IA32_MISC_ENABLE:
Zhang Xiantaoad312c72007-12-13 23:50:52 +08002224 vcpu->arch.ia32_misc_enable_msr = data;
Carsten Otte15c4a642007-10-30 18:44:17 +01002225 break;
Paolo Bonzini64d60672015-05-07 11:36:11 +02002226 case MSR_IA32_SMBASE:
2227 if (!msr_info->host_initiated)
2228 return 1;
2229 vcpu->arch.smbase = data;
2230 break;
Glauber Costa11c6bff2010-05-11 12:17:41 -04002231 case MSR_KVM_WALL_CLOCK_NEW:
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02002232 case MSR_KVM_WALL_CLOCK:
2233 vcpu->kvm->arch.wall_clock = data;
2234 kvm_write_wall_clock(vcpu->kvm, data);
2235 break;
Glauber Costa11c6bff2010-05-11 12:17:41 -04002236 case MSR_KVM_SYSTEM_TIME_NEW:
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02002237 case MSR_KVM_SYSTEM_TIME: {
Marcelo Tosatti54750f22015-01-20 15:54:52 -02002238 struct kvm_arch *ka = &vcpu->kvm->arch;
2239
Glauber Costa12f9a482011-02-01 14:16:40 -05002240 kvmclock_reset(vcpu);
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02002241
Marcelo Tosatti54750f22015-01-20 15:54:52 -02002242 if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2243 bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2244
2245 if (ka->boot_vcpu_runs_old_kvmclock != tmp)
Radim Krčmář1bd20092017-04-26 22:32:20 +02002246 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
Marcelo Tosatti54750f22015-01-20 15:54:52 -02002247
2248 ka->boot_vcpu_runs_old_kvmclock = tmp;
2249 }
2250
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02002251 vcpu->arch.time = data;
Marcelo Tosatti0061d532013-05-09 20:21:41 -03002252 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02002253
2254 /* we verify if the enable bit is set... */
2255 if (!(data & 1))
2256 break;
2257
Paolo Bonzini4e335d92017-05-02 16:20:18 +02002258 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
Andrew Honig8f964522013-03-29 09:35:21 -07002259 &vcpu->arch.pv_time, data & ~1ULL,
2260 sizeof(struct pvclock_vcpu_time_info)))
Andy Honig0b794592013-02-20 14:48:10 -08002261 vcpu->arch.pv_time_enabled = false;
2262 else
2263 vcpu->arch.pv_time_enabled = true;
Xiao Guangrong32cad842012-08-03 15:42:52 +08002264
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02002265 break;
2266 }
Gleb Natapov344d9582010-10-14 11:22:50 +02002267 case MSR_KVM_ASYNC_PF_EN:
2268 if (kvm_pv_enable_async_pf(vcpu, data))
2269 return 1;
2270 break;
Glauber Costac9aaa892011-07-11 15:28:14 -04002271 case MSR_KVM_STEAL_TIME:
2272
2273 if (unlikely(!sched_info_on()))
2274 return 1;
2275
2276 if (data & KVM_STEAL_RESERVED_MASK)
2277 return 1;
2278
Paolo Bonzini4e335d92017-05-02 16:20:18 +02002279 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
Andrew Honig8f964522013-03-29 09:35:21 -07002280 data & KVM_STEAL_VALID_BITS,
2281 sizeof(struct kvm_steal_time)))
Glauber Costac9aaa892011-07-11 15:28:14 -04002282 return 1;
2283
2284 vcpu->arch.st.msr_val = data;
2285
2286 if (!(data & KVM_MSR_ENABLED))
2287 break;
2288
Glauber Costac9aaa892011-07-11 15:28:14 -04002289 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2290
2291 break;
Michael S. Tsirkinae7a2a32012-06-24 19:25:07 +03002292 case MSR_KVM_PV_EOI_EN:
2293 if (kvm_lapic_enable_pv_eoi(vcpu, data))
2294 return 1;
2295 break;
Glauber Costac9aaa892011-07-11 15:28:14 -04002296
Huang Ying890ca9a2009-05-11 16:48:15 +08002297 case MSR_IA32_MCG_CTL:
2298 case MSR_IA32_MCG_STATUS:
Chen Yucong81760dc2014-09-23 10:44:35 +08002299 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
Wanpeng Li9ffd9862017-10-19 06:47:56 -07002300 return set_msr_mce(vcpu, msr_info);
Andre Przywara71db6022009-06-12 22:01:29 +02002301
Wei Huang6912ac32015-06-12 01:34:56 -04002302 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2303 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2304 pr = true; /* fall through */
2305 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2306 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
Wei Huangc6702c92015-06-19 13:44:45 +02002307 if (kvm_pmu_is_valid_msr(vcpu, msr))
Paolo Bonziniafd80d82013-03-28 17:18:35 +01002308 return kvm_pmu_set_msr(vcpu, msr_info);
Gleb Natapov57537852012-01-15 14:17:22 +02002309
2310 if (pr || data != 0)
Christoffer Dalla737f252012-06-03 21:17:48 +03002311 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2312 "0x%x data 0x%llx\n", msr, data);
Gleb Natapov57537852012-01-15 14:17:22 +02002313 break;
Jes Sorensen84e0cef2010-09-01 11:42:04 +02002314 case MSR_K7_CLK_CTL:
2315 /*
2316 * Ignore all writes to this no longer documented MSR.
2317 * Writes are only relevant for old K7 processors,
2318 * all pre-dating SVM, but a recommended workaround from
Guo Chao4a969982012-06-28 15:17:27 +08002319 * AMD for these chips. It is possible to specify the
Jes Sorensen84e0cef2010-09-01 11:42:04 +02002320 * affected processor models on the command line, hence
2321 * the need to ignore the workaround.
2322 */
2323 break;
Gleb Natapov55cd8e52010-01-17 15:51:22 +02002324 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
Andrey Smetanine7d95132015-07-03 15:01:37 +03002325 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2326 case HV_X64_MSR_CRASH_CTL:
Andrey Smetanin1f4b34f2015-11-30 19:22:21 +03002327 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
Andrey Smetanine7d95132015-07-03 15:01:37 +03002328 return kvm_hv_set_msr_common(vcpu, msr, data,
2329 msr_info->host_initiated);
john cooper91c9c3e2011-01-21 00:21:00 -05002330 case MSR_IA32_BBL_CR_CTL3:
2331 /* Drop writes to this legacy MSR -- see rdmsr
2332 * counterpart for further detail.
2333 */
Eyal Moscovicifab0aa32017-11-08 14:32:08 +02002334 if (report_ignored_msrs)
2335 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
2336 msr, data);
john cooper91c9c3e2011-01-21 00:21:00 -05002337 break;
Boris Ostrovsky2b036c62012-01-09 14:00:35 -05002338 case MSR_AMD64_OSVW_ID_LENGTH:
Radim Krčmářd6321d42017-08-05 00:12:49 +02002339 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
Boris Ostrovsky2b036c62012-01-09 14:00:35 -05002340 return 1;
2341 vcpu->arch.osvw.length = data;
2342 break;
2343 case MSR_AMD64_OSVW_STATUS:
Radim Krčmářd6321d42017-08-05 00:12:49 +02002344 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
Boris Ostrovsky2b036c62012-01-09 14:00:35 -05002345 return 1;
2346 vcpu->arch.osvw.status = data;
2347 break;
Kyle Hueydb2336a2017-03-20 01:16:28 -07002348 case MSR_PLATFORM_INFO:
2349 if (!msr_info->host_initiated ||
2350 data & ~MSR_PLATFORM_INFO_CPUID_FAULT ||
2351 (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
2352 cpuid_fault_enabled(vcpu)))
2353 return 1;
2354 vcpu->arch.msr_platform_info = data;
2355 break;
2356 case MSR_MISC_FEATURES_ENABLES:
2357 if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
2358 (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
2359 !supports_cpuid_fault(vcpu)))
2360 return 1;
2361 vcpu->arch.msr_misc_features_enables = data;
2362 break;
Carsten Otte15c4a642007-10-30 18:44:17 +01002363 default:
Ed Swierkffde22a2009-10-15 15:21:43 -07002364 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2365 return xen_hvm_config(vcpu, data);
Wei Huangc6702c92015-06-19 13:44:45 +02002366 if (kvm_pmu_is_valid_msr(vcpu, msr))
Paolo Bonziniafd80d82013-03-28 17:18:35 +01002367 return kvm_pmu_set_msr(vcpu, msr_info);
Andre Przywaraed85c062009-06-25 12:36:49 +02002368 if (!ignore_msrs) {
Bandan Dasae0f5492016-11-15 01:36:18 -05002369 vcpu_debug_ratelimited(vcpu, "unhandled wrmsr: 0x%x data 0x%llx\n",
Christoffer Dalla737f252012-06-03 21:17:48 +03002370 msr, data);
Andre Przywaraed85c062009-06-25 12:36:49 +02002371 return 1;
2372 } else {
Eyal Moscovicifab0aa32017-11-08 14:32:08 +02002373 if (report_ignored_msrs)
2374 vcpu_unimpl(vcpu,
2375 "ignored wrmsr: 0x%x data 0x%llx\n",
2376 msr, data);
Andre Przywaraed85c062009-06-25 12:36:49 +02002377 break;
2378 }
Carsten Otte15c4a642007-10-30 18:44:17 +01002379 }
2380 return 0;
2381}
2382EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2383
2384
2385/*
2386 * Reads an msr value (of 'msr_index') into 'pdata'.
2387 * Returns 0 on success, non-0 otherwise.
2388 * Assumes vcpu_load() was already called.
2389 */
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002390int kvm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
Carsten Otte15c4a642007-10-30 18:44:17 +01002391{
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002392 return kvm_x86_ops->get_msr(vcpu, msr);
Carsten Otte15c4a642007-10-30 18:44:17 +01002393}
Wincy Vanff651cb2014-12-11 08:52:58 +03002394EXPORT_SYMBOL_GPL(kvm_get_msr);
Carsten Otte15c4a642007-10-30 18:44:17 +01002395
Huang Ying890ca9a2009-05-11 16:48:15 +08002396static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2397{
2398 u64 data;
2399 u64 mcg_cap = vcpu->arch.mcg_cap;
2400 unsigned bank_num = mcg_cap & 0xff;
2401
2402 switch (msr) {
2403 case MSR_IA32_P5_MC_ADDR:
2404 case MSR_IA32_P5_MC_TYPE:
2405 data = 0;
2406 break;
2407 case MSR_IA32_MCG_CAP:
2408 data = vcpu->arch.mcg_cap;
2409 break;
2410 case MSR_IA32_MCG_CTL:
2411 if (!(mcg_cap & MCG_CTL_P))
2412 return 1;
2413 data = vcpu->arch.mcg_ctl;
2414 break;
2415 case MSR_IA32_MCG_STATUS:
2416 data = vcpu->arch.mcg_status;
2417 break;
2418 default:
2419 if (msr >= MSR_IA32_MC0_CTL &&
Chen Yucong81760dc2014-09-23 10:44:35 +08002420 msr < MSR_IA32_MCx_CTL(bank_num)) {
Huang Ying890ca9a2009-05-11 16:48:15 +08002421 u32 offset = msr - MSR_IA32_MC0_CTL;
2422 data = vcpu->arch.mce_banks[offset];
2423 break;
2424 }
2425 return 1;
2426 }
2427 *pdata = data;
2428 return 0;
2429}
2430
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002431int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
Carsten Otte15c4a642007-10-30 18:44:17 +01002432{
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002433 switch (msr_info->index) {
Carsten Otte15c4a642007-10-30 18:44:17 +01002434 case MSR_IA32_PLATFORM_ID:
Carsten Otte15c4a642007-10-30 18:44:17 +01002435 case MSR_IA32_EBL_CR_POWERON:
Alexander Grafb5e2fec2008-07-22 08:00:45 +02002436 case MSR_IA32_DEBUGCTLMSR:
2437 case MSR_IA32_LASTBRANCHFROMIP:
2438 case MSR_IA32_LASTBRANCHTOIP:
2439 case MSR_IA32_LASTINTFROMIP:
2440 case MSR_IA32_LASTINTTOIP:
Jaswinder Singh Rajput60af2ec2009-05-14 11:00:10 +05302441 case MSR_K8_SYSCFG:
Paolo Bonzini3afb1122015-09-18 17:33:04 +02002442 case MSR_K8_TSEG_ADDR:
2443 case MSR_K8_TSEG_MASK:
Jaswinder Singh Rajput60af2ec2009-05-14 11:00:10 +05302444 case MSR_K7_HWCR:
Avi Kivity61a6bd62008-12-29 17:32:28 +02002445 case MSR_VM_HSAVE_PA:
Andre Przywara1fdbd482009-06-24 12:44:34 +02002446 case MSR_K8_INT_PENDING_MSG:
Andre Przywarac323c0e2009-06-24 15:37:05 +02002447 case MSR_AMD64_NB_CFG:
Andre Przywaraf7c6d142009-07-02 15:04:14 +02002448 case MSR_FAM10H_MMIO_CONF_BASE:
Borislav Petkov2e32b712013-02-19 19:33:13 +01002449 case MSR_AMD64_BU_CFG2:
Dmitry Bilunov0c2df2a2016-05-31 17:38:24 +03002450 case MSR_IA32_PERF_CTL:
Ladi Prosek405a3532017-04-06 15:22:20 +02002451 case MSR_AMD64_DC_CFG:
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002452 msr_info->data = 0;
Carsten Otte15c4a642007-10-30 18:44:17 +01002453 break;
Wei Huang6912ac32015-06-12 01:34:56 -04002454 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2455 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2456 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2457 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
Wei Huangc6702c92015-06-19 13:44:45 +02002458 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002459 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2460 msr_info->data = 0;
Gleb Natapov57537852012-01-15 14:17:22 +02002461 break;
Marcelo Tosatti742bc672011-07-29 19:44:21 -03002462 case MSR_IA32_UCODE_REV:
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002463 msr_info->data = 0x100000000ULL;
Marcelo Tosatti742bc672011-07-29 19:44:21 -03002464 break;
Avi Kivity9ba075a2008-05-26 20:06:35 +03002465 case MSR_MTRRcap:
Avi Kivity9ba075a2008-05-26 20:06:35 +03002466 case 0x200 ... 0x2ff:
Xiao Guangrongff536042015-06-15 16:55:22 +08002467 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
Carsten Otte15c4a642007-10-30 18:44:17 +01002468 case 0xcd: /* fsb frequency */
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002469 msr_info->data = 3;
Carsten Otte15c4a642007-10-30 18:44:17 +01002470 break;
Jes Sorensen7b914092010-09-09 12:06:46 +02002471 /*
2472 * MSR_EBC_FREQUENCY_ID
2473 * Conservative value valid for even the basic CPU models.
2474 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2475 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2476 * and 266MHz for model 3, or 4. Set Core Clock
2477 * Frequency to System Bus Frequency Ratio to 1 (bits
2478 * 31:24) even though these are only valid for CPU
2479 * models > 2, however guests may end up dividing or
2480 * multiplying by zero otherwise.
2481 */
2482 case MSR_EBC_FREQUENCY_ID:
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002483 msr_info->data = 1 << 24;
Jes Sorensen7b914092010-09-09 12:06:46 +02002484 break;
Carsten Otte15c4a642007-10-30 18:44:17 +01002485 case MSR_IA32_APICBASE:
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002486 msr_info->data = kvm_get_apic_base(vcpu);
Carsten Otte15c4a642007-10-30 18:44:17 +01002487 break;
Gleb Natapov0105d1a2009-07-05 17:39:36 +03002488 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002489 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
Gleb Natapov0105d1a2009-07-05 17:39:36 +03002490 break;
Liu, Jinsonga3e06bb2011-09-22 16:55:52 +08002491 case MSR_IA32_TSCDEADLINE:
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002492 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
Liu, Jinsonga3e06bb2011-09-22 16:55:52 +08002493 break;
Will Auldba904632012-11-29 12:42:50 -08002494 case MSR_IA32_TSC_ADJUST:
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002495 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
Will Auldba904632012-11-29 12:42:50 -08002496 break;
Carsten Otte15c4a642007-10-30 18:44:17 +01002497 case MSR_IA32_MISC_ENABLE:
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002498 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
Carsten Otte15c4a642007-10-30 18:44:17 +01002499 break;
Paolo Bonzini64d60672015-05-07 11:36:11 +02002500 case MSR_IA32_SMBASE:
2501 if (!msr_info->host_initiated)
2502 return 1;
2503 msr_info->data = vcpu->arch.smbase;
Carsten Otte15c4a642007-10-30 18:44:17 +01002504 break;
Alexander Graf847f0ad2008-02-21 12:11:01 +01002505 case MSR_IA32_PERF_STATUS:
2506 /* TSC increment by tick */
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002507 msr_info->data = 1000ULL;
Alexander Graf847f0ad2008-02-21 12:11:01 +01002508 /* CPU multiplier */
Nicolas Ioossb0996ae2015-06-29 18:39:23 +08002509 msr_info->data |= (((uint64_t)4ULL) << 40);
Alexander Graf847f0ad2008-02-21 12:11:01 +01002510 break;
Carsten Otte15c4a642007-10-30 18:44:17 +01002511 case MSR_EFER:
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002512 msr_info->data = vcpu->arch.efer;
Carsten Otte15c4a642007-10-30 18:44:17 +01002513 break;
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02002514 case MSR_KVM_WALL_CLOCK:
Glauber Costa11c6bff2010-05-11 12:17:41 -04002515 case MSR_KVM_WALL_CLOCK_NEW:
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002516 msr_info->data = vcpu->kvm->arch.wall_clock;
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02002517 break;
2518 case MSR_KVM_SYSTEM_TIME:
Glauber Costa11c6bff2010-05-11 12:17:41 -04002519 case MSR_KVM_SYSTEM_TIME_NEW:
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002520 msr_info->data = vcpu->arch.time;
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02002521 break;
Gleb Natapov344d9582010-10-14 11:22:50 +02002522 case MSR_KVM_ASYNC_PF_EN:
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002523 msr_info->data = vcpu->arch.apf.msr_val;
Gleb Natapov344d9582010-10-14 11:22:50 +02002524 break;
Glauber Costac9aaa892011-07-11 15:28:14 -04002525 case MSR_KVM_STEAL_TIME:
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002526 msr_info->data = vcpu->arch.st.msr_val;
Glauber Costac9aaa892011-07-11 15:28:14 -04002527 break;
Michael S. Tsirkin1d921282012-08-26 18:00:29 +03002528 case MSR_KVM_PV_EOI_EN:
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002529 msr_info->data = vcpu->arch.pv_eoi.msr_val;
Michael S. Tsirkin1d921282012-08-26 18:00:29 +03002530 break;
Huang Ying890ca9a2009-05-11 16:48:15 +08002531 case MSR_IA32_P5_MC_ADDR:
2532 case MSR_IA32_P5_MC_TYPE:
2533 case MSR_IA32_MCG_CAP:
2534 case MSR_IA32_MCG_CTL:
2535 case MSR_IA32_MCG_STATUS:
Chen Yucong81760dc2014-09-23 10:44:35 +08002536 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002537 return get_msr_mce(vcpu, msr_info->index, &msr_info->data);
Jes Sorensen84e0cef2010-09-01 11:42:04 +02002538 case MSR_K7_CLK_CTL:
2539 /*
2540 * Provide expected ramp-up count for K7. All other
2541 * are set to zero, indicating minimum divisors for
2542 * every field.
2543 *
2544 * This prevents guest kernels on AMD host with CPU
2545 * type 6, model 8 and higher from exploding due to
2546 * the rdmsr failing.
2547 */
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002548 msr_info->data = 0x20000000;
Jes Sorensen84e0cef2010-09-01 11:42:04 +02002549 break;
Gleb Natapov55cd8e52010-01-17 15:51:22 +02002550 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
Andrey Smetanine7d95132015-07-03 15:01:37 +03002551 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2552 case HV_X64_MSR_CRASH_CTL:
Andrey Smetanin1f4b34f2015-11-30 19:22:21 +03002553 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
Andrey Smetanine83d5882015-07-03 15:01:34 +03002554 return kvm_hv_get_msr_common(vcpu,
2555 msr_info->index, &msr_info->data);
Gleb Natapov55cd8e52010-01-17 15:51:22 +02002556 break;
john cooper91c9c3e2011-01-21 00:21:00 -05002557 case MSR_IA32_BBL_CR_CTL3:
2558 /* This legacy MSR exists but isn't fully documented in current
2559 * silicon. It is however accessed by winxp in very narrow
2560 * scenarios where it sets bit #19, itself documented as
2561 * a "reserved" bit. Best effort attempt to source coherent
2562 * read data here should the balance of the register be
2563 * interpreted by the guest:
2564 *
2565 * L2 cache control register 3: 64GB range, 256KB size,
2566 * enabled, latency 0x1, configured
2567 */
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002568 msr_info->data = 0xbe702111;
john cooper91c9c3e2011-01-21 00:21:00 -05002569 break;
Boris Ostrovsky2b036c62012-01-09 14:00:35 -05002570 case MSR_AMD64_OSVW_ID_LENGTH:
Radim Krčmářd6321d42017-08-05 00:12:49 +02002571 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
Boris Ostrovsky2b036c62012-01-09 14:00:35 -05002572 return 1;
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002573 msr_info->data = vcpu->arch.osvw.length;
Boris Ostrovsky2b036c62012-01-09 14:00:35 -05002574 break;
2575 case MSR_AMD64_OSVW_STATUS:
Radim Krčmářd6321d42017-08-05 00:12:49 +02002576 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
Boris Ostrovsky2b036c62012-01-09 14:00:35 -05002577 return 1;
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002578 msr_info->data = vcpu->arch.osvw.status;
Boris Ostrovsky2b036c62012-01-09 14:00:35 -05002579 break;
Kyle Hueydb2336a2017-03-20 01:16:28 -07002580 case MSR_PLATFORM_INFO:
2581 msr_info->data = vcpu->arch.msr_platform_info;
2582 break;
2583 case MSR_MISC_FEATURES_ENABLES:
2584 msr_info->data = vcpu->arch.msr_misc_features_enables;
2585 break;
Carsten Otte15c4a642007-10-30 18:44:17 +01002586 default:
Wei Huangc6702c92015-06-19 13:44:45 +02002587 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002588 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
Andre Przywaraed85c062009-06-25 12:36:49 +02002589 if (!ignore_msrs) {
Bandan Dasae0f5492016-11-15 01:36:18 -05002590 vcpu_debug_ratelimited(vcpu, "unhandled rdmsr: 0x%x\n",
2591 msr_info->index);
Andre Przywaraed85c062009-06-25 12:36:49 +02002592 return 1;
2593 } else {
Eyal Moscovicifab0aa32017-11-08 14:32:08 +02002594 if (report_ignored_msrs)
2595 vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n",
2596 msr_info->index);
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002597 msr_info->data = 0;
Andre Przywaraed85c062009-06-25 12:36:49 +02002598 }
2599 break;
Carsten Otte15c4a642007-10-30 18:44:17 +01002600 }
Carsten Otte15c4a642007-10-30 18:44:17 +01002601 return 0;
2602}
2603EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2604
Carsten Otte313a3dc2007-10-11 19:16:52 +02002605/*
2606 * Read or write a bunch of msrs. All parameters are kernel addresses.
2607 *
2608 * @return number of msrs set successfully.
2609 */
2610static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2611 struct kvm_msr_entry *entries,
2612 int (*do_msr)(struct kvm_vcpu *vcpu,
2613 unsigned index, u64 *data))
2614{
Marcelo Tosattif656ce02009-12-23 14:35:25 -02002615 int i, idx;
Carsten Otte313a3dc2007-10-11 19:16:52 +02002616
Marcelo Tosattif656ce02009-12-23 14:35:25 -02002617 idx = srcu_read_lock(&vcpu->kvm->srcu);
Carsten Otte313a3dc2007-10-11 19:16:52 +02002618 for (i = 0; i < msrs->nmsrs; ++i)
2619 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2620 break;
Marcelo Tosattif656ce02009-12-23 14:35:25 -02002621 srcu_read_unlock(&vcpu->kvm->srcu, idx);
Carsten Otte313a3dc2007-10-11 19:16:52 +02002622
Carsten Otte313a3dc2007-10-11 19:16:52 +02002623 return i;
2624}
2625
2626/*
2627 * Read or write a bunch of msrs. Parameters are user addresses.
2628 *
2629 * @return number of msrs set successfully.
2630 */
2631static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2632 int (*do_msr)(struct kvm_vcpu *vcpu,
2633 unsigned index, u64 *data),
2634 int writeback)
2635{
2636 struct kvm_msrs msrs;
2637 struct kvm_msr_entry *entries;
2638 int r, n;
2639 unsigned size;
2640
2641 r = -EFAULT;
2642 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2643 goto out;
2644
2645 r = -E2BIG;
2646 if (msrs.nmsrs >= MAX_IO_MSRS)
2647 goto out;
2648
Carsten Otte313a3dc2007-10-11 19:16:52 +02002649 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002650 entries = memdup_user(user_msrs->entries, size);
2651 if (IS_ERR(entries)) {
2652 r = PTR_ERR(entries);
Carsten Otte313a3dc2007-10-11 19:16:52 +02002653 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002654 }
Carsten Otte313a3dc2007-10-11 19:16:52 +02002655
2656 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2657 if (r < 0)
2658 goto out_free;
2659
2660 r = -EFAULT;
2661 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2662 goto out_free;
2663
2664 r = n;
2665
2666out_free:
Avi Kivity7a73c022010-07-22 23:24:52 +03002667 kfree(entries);
Carsten Otte313a3dc2007-10-11 19:16:52 +02002668out:
2669 return r;
2670}
2671
Alexander Graf784aa3d2014-07-14 18:27:35 +02002672int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
Zhang Xiantao018d00d2007-11-15 23:07:47 +08002673{
2674 int r;
2675
2676 switch (ext) {
2677 case KVM_CAP_IRQCHIP:
2678 case KVM_CAP_HLT:
2679 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
Zhang Xiantao018d00d2007-11-15 23:07:47 +08002680 case KVM_CAP_SET_TSS_ADDR:
Dan Kenigsberg07716712007-11-21 17:10:04 +02002681 case KVM_CAP_EXT_CPUID:
Borislav Petkov9c15bb12013-09-22 16:44:50 +02002682 case KVM_CAP_EXT_EMUL_CPUID:
Gerd Hoffmannc8076602009-02-04 17:52:04 +01002683 case KVM_CAP_CLOCKSOURCE:
Sheng Yang78376992008-01-28 05:10:22 +08002684 case KVM_CAP_PIT:
Marcelo Tosattia28e4f52008-02-22 12:21:36 -05002685 case KVM_CAP_NOP_IO_DELAY:
Marcelo Tosatti62d9f0d2008-04-11 13:24:45 -03002686 case KVM_CAP_MP_STATE:
Avi Kivityed848622008-07-29 11:30:57 +03002687 case KVM_CAP_SYNC_MMU:
Lai Jiangshana355c852010-12-14 17:57:47 +08002688 case KVM_CAP_USER_NMI:
Marcelo Tosatti52d939a2008-12-30 15:55:06 -02002689 case KVM_CAP_REINJECT_CONTROL:
Gleb Natapov49256632009-02-04 17:28:14 +02002690 case KVM_CAP_IRQ_INJECT_STATUS:
Gregory Haskinsd34e6b12009-07-07 17:08:49 -04002691 case KVM_CAP_IOEVENTFD:
Michael S. Tsirkinf848a5a2014-03-31 21:50:38 +03002692 case KVM_CAP_IOEVENTFD_NO_LENGTH:
Jan Kiszkac5ff41c2009-05-14 22:42:53 +02002693 case KVM_CAP_PIT2:
Beth Kone9f42752009-07-07 11:50:38 -04002694 case KVM_CAP_PIT_STATE2:
Sheng Yangb927a3c2009-07-21 10:42:48 +08002695 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
Ed Swierkffde22a2009-10-15 15:21:43 -07002696 case KVM_CAP_XEN_HVM:
Jan Kiszka3cfc3092009-11-12 01:04:25 +01002697 case KVM_CAP_VCPU_EVENTS:
Gleb Natapov55cd8e52010-01-17 15:51:22 +02002698 case KVM_CAP_HYPERV:
Gleb Natapov10388a02010-01-17 15:51:23 +02002699 case KVM_CAP_HYPERV_VAPIC:
Gleb Natapovc25bc162010-01-17 15:51:24 +02002700 case KVM_CAP_HYPERV_SPIN:
Andrey Smetanin5c9194122015-11-10 15:36:34 +03002701 case KVM_CAP_HYPERV_SYNIC:
Roman Kaganefc479e2017-06-22 16:51:01 +03002702 case KVM_CAP_HYPERV_SYNIC2:
Roman Kagand3457c82017-07-14 17:13:20 +03002703 case KVM_CAP_HYPERV_VP_INDEX:
Zhai, Edwinab9f4ec2010-01-29 14:38:44 +08002704 case KVM_CAP_PCI_SEGMENT:
Jan Kiszkaa1efbe72010-02-15 10:45:43 +01002705 case KVM_CAP_DEBUGREGS:
Jan Kiszkad2be1652010-02-23 17:47:57 +01002706 case KVM_CAP_X86_ROBUST_SINGLESTEP:
Sheng Yang2d5b5a62010-06-13 17:29:39 +08002707 case KVM_CAP_XSAVE:
Gleb Natapov344d9582010-10-14 11:22:50 +02002708 case KVM_CAP_ASYNC_PF:
Joerg Roedel92a1f122011-03-25 09:44:51 +01002709 case KVM_CAP_GET_TSC_KHZ:
Eric B Munson1c0b28c2012-03-10 14:37:27 -05002710 case KVM_CAP_KVMCLOCK_CTRL:
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08002711 case KVM_CAP_READONLY_MEM:
Paolo Bonzini5f66b622014-01-29 18:10:45 +01002712 case KVM_CAP_HYPERV_TIME:
Gabriel L. Somlo100943c2014-02-27 23:06:17 -05002713 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
Radim Krčmářdefcf512015-01-08 15:59:30 +01002714 case KVM_CAP_TSC_DEADLINE_TIMER:
Nadav Amit90de4a12015-04-13 01:53:41 +03002715 case KVM_CAP_ENABLE_CAP_VM:
2716 case KVM_CAP_DISABLE_QUIRKS:
Paolo Bonzinid71ba782015-07-29 11:56:48 +02002717 case KVM_CAP_SET_BOOT_CPU_ID:
Steve Rutherford49df6392015-07-29 23:21:40 -07002718 case KVM_CAP_SPLIT_IRQCHIP:
Paolo Bonzini460df4c2017-02-08 11:50:15 +01002719 case KVM_CAP_IMMEDIATE_EXIT:
Zhang Xiantao018d00d2007-11-15 23:07:47 +08002720 r = 1;
2721 break;
Paolo Bonzinie3fd9a92016-11-09 17:48:15 +01002722 case KVM_CAP_ADJUST_CLOCK:
2723 r = KVM_CLOCK_TSC_STABLE;
2724 break;
Michael S. Tsirkin668fffa2017-04-21 12:27:17 +02002725 case KVM_CAP_X86_GUEST_MWAIT:
2726 r = kvm_mwait_in_guest();
2727 break;
Paolo Bonzini6d396b52015-04-01 14:25:33 +02002728 case KVM_CAP_X86_SMM:
2729 /* SMBASE is usually relocated above 1M on modern chipsets,
2730 * and SMM handlers might indeed rely on 4G segment limits,
2731 * so do not report SMM to be available if real mode is
2732 * emulated via vm86 mode. Still, do not go to great lengths
2733 * to avoid userspace's usage of the feature, because it is a
2734 * fringe case that is not enabled except via specific settings
2735 * of the module parameters.
2736 */
2737 r = kvm_x86_ops->cpu_has_high_real_mode_segbase();
2738 break;
Avi Kivity774ead32007-12-26 13:57:04 +02002739 case KVM_CAP_VAPIC:
2740 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2741 break;
Avi Kivityf7252302008-02-20 11:53:16 +02002742 case KVM_CAP_NR_VCPUS:
Sasha Levin8c3ba332011-07-18 17:17:15 +03002743 r = KVM_SOFT_MAX_VCPUS;
2744 break;
2745 case KVM_CAP_MAX_VCPUS:
Avi Kivityf7252302008-02-20 11:53:16 +02002746 r = KVM_MAX_VCPUS;
2747 break;
Avi Kivitya988b912008-02-20 11:59:20 +02002748 case KVM_CAP_NR_MEMSLOTS:
Alex Williamsonbbacc0c2012-12-10 10:33:09 -07002749 r = KVM_USER_MEM_SLOTS;
Avi Kivitya988b912008-02-20 11:59:20 +02002750 break;
Marcelo Tosattia68a6a72009-10-01 19:28:39 -03002751 case KVM_CAP_PV_MMU: /* obsolete */
2752 r = 0;
Marcelo Tosatti2f333bc2008-02-22 12:21:37 -05002753 break;
Huang Ying890ca9a2009-05-11 16:48:15 +08002754 case KVM_CAP_MCE:
2755 r = KVM_MAX_MCE_BANKS;
2756 break;
Sheng Yang2d5b5a62010-06-13 17:29:39 +08002757 case KVM_CAP_XCRS:
Borislav Petkovd366bf72016-04-04 22:25:02 +02002758 r = boot_cpu_has(X86_FEATURE_XSAVE);
Sheng Yang2d5b5a62010-06-13 17:29:39 +08002759 break;
Joerg Roedel92a1f122011-03-25 09:44:51 +01002760 case KVM_CAP_TSC_CONTROL:
2761 r = kvm_has_tsc_control;
2762 break;
Radim Krčmář371313132016-07-12 22:09:27 +02002763 case KVM_CAP_X2APIC_API:
2764 r = KVM_X2APIC_API_VALID_FLAGS;
2765 break;
Zhang Xiantao018d00d2007-11-15 23:07:47 +08002766 default:
2767 r = 0;
2768 break;
2769 }
2770 return r;
2771
2772}
2773
Carsten Otte043405e2007-10-10 17:16:19 +02002774long kvm_arch_dev_ioctl(struct file *filp,
2775 unsigned int ioctl, unsigned long arg)
2776{
2777 void __user *argp = (void __user *)arg;
2778 long r;
2779
2780 switch (ioctl) {
2781 case KVM_GET_MSR_INDEX_LIST: {
2782 struct kvm_msr_list __user *user_msr_list = argp;
2783 struct kvm_msr_list msr_list;
2784 unsigned n;
2785
2786 r = -EFAULT;
2787 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2788 goto out;
2789 n = msr_list.nmsrs;
Paolo Bonzini62ef68b2015-05-05 12:08:55 +02002790 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
Carsten Otte043405e2007-10-10 17:16:19 +02002791 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2792 goto out;
2793 r = -E2BIG;
Jan Kiszkae125e7b2009-07-02 21:45:47 +02002794 if (n < msr_list.nmsrs)
Carsten Otte043405e2007-10-10 17:16:19 +02002795 goto out;
2796 r = -EFAULT;
2797 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2798 num_msrs_to_save * sizeof(u32)))
2799 goto out;
Jan Kiszkae125e7b2009-07-02 21:45:47 +02002800 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
Carsten Otte043405e2007-10-10 17:16:19 +02002801 &emulated_msrs,
Paolo Bonzini62ef68b2015-05-05 12:08:55 +02002802 num_emulated_msrs * sizeof(u32)))
Carsten Otte043405e2007-10-10 17:16:19 +02002803 goto out;
2804 r = 0;
2805 break;
2806 }
Borislav Petkov9c15bb12013-09-22 16:44:50 +02002807 case KVM_GET_SUPPORTED_CPUID:
2808 case KVM_GET_EMULATED_CPUID: {
Avi Kivity674eea02008-02-11 18:37:23 +02002809 struct kvm_cpuid2 __user *cpuid_arg = argp;
2810 struct kvm_cpuid2 cpuid;
2811
2812 r = -EFAULT;
2813 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2814 goto out;
Borislav Petkov9c15bb12013-09-22 16:44:50 +02002815
2816 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
2817 ioctl);
Avi Kivity674eea02008-02-11 18:37:23 +02002818 if (r)
2819 goto out;
2820
2821 r = -EFAULT;
2822 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2823 goto out;
2824 r = 0;
2825 break;
2826 }
Huang Ying890ca9a2009-05-11 16:48:15 +08002827 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
Huang Ying890ca9a2009-05-11 16:48:15 +08002828 r = -EFAULT;
Ashok Rajc45dcc72016-06-22 14:59:56 +08002829 if (copy_to_user(argp, &kvm_mce_cap_supported,
2830 sizeof(kvm_mce_cap_supported)))
Huang Ying890ca9a2009-05-11 16:48:15 +08002831 goto out;
2832 r = 0;
2833 break;
2834 }
Carsten Otte043405e2007-10-10 17:16:19 +02002835 default:
2836 r = -EINVAL;
2837 }
2838out:
2839 return r;
2840}
2841
Sheng Yangf5f48ee2010-06-30 12:25:15 +08002842static void wbinvd_ipi(void *garbage)
2843{
2844 wbinvd();
2845}
2846
2847static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2848{
Alex Williamsone0f0bbc2013-10-30 11:02:30 -06002849 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
Sheng Yangf5f48ee2010-06-30 12:25:15 +08002850}
2851
Carsten Otte313a3dc2007-10-11 19:16:52 +02002852void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2853{
Sheng Yangf5f48ee2010-06-30 12:25:15 +08002854 /* Address WBINVD may be executed by guest */
2855 if (need_emulate_wbinvd(vcpu)) {
2856 if (kvm_x86_ops->has_wbinvd_exit())
2857 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2858 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2859 smp_call_function_single(vcpu->cpu,
2860 wbinvd_ipi, NULL, 1);
2861 }
2862
Carsten Otte313a3dc2007-10-11 19:16:52 +02002863 kvm_x86_ops->vcpu_load(vcpu, cpu);
Zachary Amsden0dd6a6e2012-02-03 15:43:56 -02002864
2865 /* Apply any externally detected TSC adjustments (due to suspend) */
2866 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
2867 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
2868 vcpu->arch.tsc_offset_adjustment = 0;
Guo Hui Liu105b21b2014-09-12 13:43:19 +08002869 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
Zachary Amsden0dd6a6e2012-02-03 15:43:56 -02002870 }
2871
Zachary Amsden48434c202010-08-19 22:07:24 -10002872 if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
Zachary Amsden6f526ec2012-02-03 15:43:54 -02002873 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
Andy Lutomirski4ea16362015-06-25 18:44:07 +02002874 rdtsc() - vcpu->arch.last_host_tsc;
Zachary Amsdene48672f2010-08-19 22:07:23 -10002875 if (tsc_delta < 0)
2876 mark_tsc_unstable("KVM discovered backwards TSC");
Yunhong Jiangce7a0582016-06-13 14:20:01 -07002877
Zachary Amsdenc2855452010-09-18 14:38:15 -10002878 if (check_tsc_unstable()) {
Haozhong Zhang07c14192015-10-20 15:39:05 +08002879 u64 offset = kvm_compute_tsc_offset(vcpu,
Zachary Amsdenb183aa52012-02-03 15:43:53 -02002880 vcpu->arch.last_guest_tsc);
Luiz Capitulinoa545ab62016-09-07 14:47:19 -04002881 kvm_vcpu_write_tsc_offset(vcpu, offset);
Zachary Amsdenc2855452010-09-18 14:38:15 -10002882 vcpu->arch.tsc_catchup = 1;
Zachary Amsdenc2855452010-09-18 14:38:15 -10002883 }
Paolo Bonzinia749e242017-06-29 17:14:50 +02002884
2885 if (kvm_lapic_hv_timer_in_use(vcpu))
2886 kvm_lapic_restart_hv_timer(vcpu);
2887
Marcelo Tosattid98d07c2012-11-27 23:29:04 -02002888 /*
2889 * On a host with synchronized TSC, there is no need to update
2890 * kvmclock on vcpu->cpu migration
2891 */
2892 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
Marcelo Tosatti0061d532013-05-09 20:21:41 -03002893 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
Zachary Amsdenc2855452010-09-18 14:38:15 -10002894 if (vcpu->cpu != cpu)
Radim Krčmář1bd20092017-04-26 22:32:20 +02002895 kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
Zachary Amsdene48672f2010-08-19 22:07:23 -10002896 vcpu->cpu = cpu;
Zachary Amsden6b7d7e72009-10-09 16:26:08 -10002897 }
Glauber Costac9aaa892011-07-11 15:28:14 -04002898
Glauber Costac9aaa892011-07-11 15:28:14 -04002899 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
Carsten Otte313a3dc2007-10-11 19:16:52 +02002900}
2901
Pan Xinhui0b9f6c42016-11-02 05:08:35 -04002902static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
2903{
2904 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2905 return;
2906
2907 vcpu->arch.st.steal.preempted = 1;
2908
Paolo Bonzini4e335d92017-05-02 16:20:18 +02002909 kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.st.stime,
Pan Xinhui0b9f6c42016-11-02 05:08:35 -04002910 &vcpu->arch.st.steal.preempted,
2911 offsetof(struct kvm_steal_time, preempted),
2912 sizeof(vcpu->arch.st.steal.preempted));
2913}
2914
Carsten Otte313a3dc2007-10-11 19:16:52 +02002915void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2916{
Andrea Arcangelicc0d9072016-12-17 19:13:32 +01002917 int idx;
Longpeng(Mike)de63ad42017-08-08 12:05:33 +08002918
2919 if (vcpu->preempted)
2920 vcpu->arch.preempted_in_kernel = !kvm_x86_ops->get_cpl(vcpu);
2921
Andrea Arcangeli931f2612016-12-17 18:43:52 +01002922 /*
2923 * Disable page faults because we're in atomic context here.
2924 * kvm_write_guest_offset_cached() would call might_fault()
2925 * that relies on pagefault_disable() to tell if there's a
2926 * bug. NOTE: the write to guest memory may not go through if
2927 * during postcopy live migration or if there's heavy guest
2928 * paging.
2929 */
2930 pagefault_disable();
Andrea Arcangelicc0d9072016-12-17 19:13:32 +01002931 /*
2932 * kvm_memslots() will be called by
2933 * kvm_write_guest_offset_cached() so take the srcu lock.
2934 */
2935 idx = srcu_read_lock(&vcpu->kvm->srcu);
Pan Xinhui0b9f6c42016-11-02 05:08:35 -04002936 kvm_steal_time_set_preempted(vcpu);
Andrea Arcangelicc0d9072016-12-17 19:13:32 +01002937 srcu_read_unlock(&vcpu->kvm->srcu, idx);
Andrea Arcangeli931f2612016-12-17 18:43:52 +01002938 pagefault_enable();
Avi Kivity02daab22009-12-30 12:40:26 +02002939 kvm_x86_ops->vcpu_put(vcpu);
Andy Lutomirski4ea16362015-06-25 18:44:07 +02002940 vcpu->arch.last_host_tsc = rdtsc();
Carsten Otte313a3dc2007-10-11 19:16:52 +02002941}
2942
Carsten Otte313a3dc2007-10-11 19:16:52 +02002943static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2944 struct kvm_lapic_state *s)
2945{
Paolo Bonzini76dfafd52016-12-19 17:17:11 +01002946 if (kvm_x86_ops->sync_pir_to_irr && vcpu->arch.apicv_active)
Andrey Smetanind62caab2015-11-10 15:36:33 +03002947 kvm_x86_ops->sync_pir_to_irr(vcpu);
2948
Radim Krčmářa92e2542016-07-12 22:09:22 +02002949 return kvm_apic_get_state(vcpu, s);
Carsten Otte313a3dc2007-10-11 19:16:52 +02002950}
2951
2952static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2953 struct kvm_lapic_state *s)
2954{
Radim Krčmářa92e2542016-07-12 22:09:22 +02002955 int r;
2956
2957 r = kvm_apic_set_state(vcpu, s);
2958 if (r)
2959 return r;
Gleb Natapovcb142eb2009-08-09 15:17:40 +03002960 update_cr8_intercept(vcpu);
Carsten Otte313a3dc2007-10-11 19:16:52 +02002961
2962 return 0;
2963}
2964
Matt Gingell127a4572015-11-17 17:32:05 +01002965static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
2966{
2967 return (!lapic_in_kernel(vcpu) ||
2968 kvm_apic_accept_pic_intr(vcpu));
2969}
2970
Matt Gingell782d4222015-11-16 15:26:00 -08002971/*
2972 * if userspace requested an interrupt window, check that the
2973 * interrupt window is open.
2974 *
2975 * No need to exit to userspace if we already have an interrupt queued.
2976 */
2977static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
2978{
2979 return kvm_arch_interrupt_allowed(vcpu) &&
2980 !kvm_cpu_has_interrupt(vcpu) &&
2981 !kvm_event_needs_reinjection(vcpu) &&
2982 kvm_cpu_accept_dm_intr(vcpu);
2983}
2984
Zhang Xiantaof77bc6a2007-11-21 04:36:41 +08002985static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2986 struct kvm_interrupt *irq)
2987{
Chen Gang02cdb502013-02-27 11:33:25 +08002988 if (irq->irq >= KVM_NR_INTERRUPTS)
Zhang Xiantaof77bc6a2007-11-21 04:36:41 +08002989 return -EINVAL;
Steve Rutherford1c1a9ce2015-07-30 11:27:16 +02002990
2991 if (!irqchip_in_kernel(vcpu->kvm)) {
2992 kvm_queue_interrupt(vcpu, irq->irq, false);
2993 kvm_make_request(KVM_REQ_EVENT, vcpu);
2994 return 0;
2995 }
2996
2997 /*
2998 * With in-kernel LAPIC, we only use this to inject EXTINT, so
2999 * fail for in-kernel 8259.
3000 */
3001 if (pic_in_kernel(vcpu->kvm))
Zhang Xiantaof77bc6a2007-11-21 04:36:41 +08003002 return -ENXIO;
Zhang Xiantaof77bc6a2007-11-21 04:36:41 +08003003
Steve Rutherford1c1a9ce2015-07-30 11:27:16 +02003004 if (vcpu->arch.pending_external_vector != -1)
3005 return -EEXIST;
Zhang Xiantaof77bc6a2007-11-21 04:36:41 +08003006
Steve Rutherford1c1a9ce2015-07-30 11:27:16 +02003007 vcpu->arch.pending_external_vector = irq->irq;
Matt Gingell934bf652015-11-16 15:26:05 -08003008 kvm_make_request(KVM_REQ_EVENT, vcpu);
Zhang Xiantaof77bc6a2007-11-21 04:36:41 +08003009 return 0;
3010}
3011
Jan Kiszkac4abb7c2008-09-26 09:30:55 +02003012static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
3013{
Jan Kiszkac4abb7c2008-09-26 09:30:55 +02003014 kvm_inject_nmi(vcpu);
Jan Kiszkac4abb7c2008-09-26 09:30:55 +02003015
3016 return 0;
3017}
3018
Paolo Bonzinif0778252015-04-01 15:06:40 +02003019static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
3020{
Paolo Bonzini64d60672015-05-07 11:36:11 +02003021 kvm_make_request(KVM_REQ_SMI, vcpu);
3022
Paolo Bonzinif0778252015-04-01 15:06:40 +02003023 return 0;
3024}
3025
Avi Kivityb209749f2007-10-22 16:50:39 +02003026static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
3027 struct kvm_tpr_access_ctl *tac)
3028{
3029 if (tac->flags)
3030 return -EINVAL;
3031 vcpu->arch.tpr_access_reporting = !!tac->enabled;
3032 return 0;
3033}
3034
Huang Ying890ca9a2009-05-11 16:48:15 +08003035static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
3036 u64 mcg_cap)
3037{
3038 int r;
3039 unsigned bank_num = mcg_cap & 0xff, bank;
3040
3041 r = -EINVAL;
Jan Kiszkaa9e38c3e2009-10-23 09:37:00 +02003042 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
Huang Ying890ca9a2009-05-11 16:48:15 +08003043 goto out;
Ashok Rajc45dcc72016-06-22 14:59:56 +08003044 if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
Huang Ying890ca9a2009-05-11 16:48:15 +08003045 goto out;
3046 r = 0;
3047 vcpu->arch.mcg_cap = mcg_cap;
3048 /* Init IA32_MCG_CTL to all 1s */
3049 if (mcg_cap & MCG_CTL_P)
3050 vcpu->arch.mcg_ctl = ~(u64)0;
3051 /* Init IA32_MCi_CTL to all 1s */
3052 for (bank = 0; bank < bank_num; bank++)
3053 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
Ashok Rajc45dcc72016-06-22 14:59:56 +08003054
3055 if (kvm_x86_ops->setup_mce)
3056 kvm_x86_ops->setup_mce(vcpu);
Huang Ying890ca9a2009-05-11 16:48:15 +08003057out:
3058 return r;
3059}
3060
3061static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
3062 struct kvm_x86_mce *mce)
3063{
3064 u64 mcg_cap = vcpu->arch.mcg_cap;
3065 unsigned bank_num = mcg_cap & 0xff;
3066 u64 *banks = vcpu->arch.mce_banks;
3067
3068 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
3069 return -EINVAL;
3070 /*
3071 * if IA32_MCG_CTL is not all 1s, the uncorrected error
3072 * reporting is disabled
3073 */
3074 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
3075 vcpu->arch.mcg_ctl != ~(u64)0)
3076 return 0;
3077 banks += 4 * mce->bank;
3078 /*
3079 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3080 * reporting is disabled for the bank
3081 */
3082 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
3083 return 0;
3084 if (mce->status & MCI_STATUS_UC) {
3085 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
Avi Kivityfc78f512009-12-07 12:16:48 +02003086 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
Avi Kivitya8eeb042010-05-10 12:34:53 +03003087 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
Huang Ying890ca9a2009-05-11 16:48:15 +08003088 return 0;
3089 }
3090 if (banks[1] & MCI_STATUS_VAL)
3091 mce->status |= MCI_STATUS_OVER;
3092 banks[2] = mce->addr;
3093 banks[3] = mce->misc;
3094 vcpu->arch.mcg_status = mce->mcg_status;
3095 banks[1] = mce->status;
3096 kvm_queue_exception(vcpu, MC_VECTOR);
3097 } else if (!(banks[1] & MCI_STATUS_VAL)
3098 || !(banks[1] & MCI_STATUS_UC)) {
3099 if (banks[1] & MCI_STATUS_VAL)
3100 mce->status |= MCI_STATUS_OVER;
3101 banks[2] = mce->addr;
3102 banks[3] = mce->misc;
3103 banks[1] = mce->status;
3104 } else
3105 banks[1] |= MCI_STATUS_OVER;
3106 return 0;
3107}
3108
Jan Kiszka3cfc3092009-11-12 01:04:25 +01003109static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3110 struct kvm_vcpu_events *events)
3111{
Avi Kivity7460fb4a2011-09-20 13:43:14 +03003112 process_nmi(vcpu);
Wanpeng Li664f8e22017-08-24 03:35:09 -07003113 /*
3114 * FIXME: pass injected and pending separately. This is only
3115 * needed for nested virtualization, whose state cannot be
3116 * migrated yet. For now we can combine them.
3117 */
Jan Kiszka03b82a32010-02-15 10:45:41 +01003118 events->exception.injected =
Wanpeng Li664f8e22017-08-24 03:35:09 -07003119 (vcpu->arch.exception.pending ||
3120 vcpu->arch.exception.injected) &&
Jan Kiszka03b82a32010-02-15 10:45:41 +01003121 !kvm_exception_is_soft(vcpu->arch.exception.nr);
Jan Kiszka3cfc3092009-11-12 01:04:25 +01003122 events->exception.nr = vcpu->arch.exception.nr;
3123 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
Vasiliy Kulikov97e69aa2010-10-30 22:54:47 +04003124 events->exception.pad = 0;
Jan Kiszka3cfc3092009-11-12 01:04:25 +01003125 events->exception.error_code = vcpu->arch.exception.error_code;
3126
Jan Kiszka03b82a32010-02-15 10:45:41 +01003127 events->interrupt.injected =
3128 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
Jan Kiszka3cfc3092009-11-12 01:04:25 +01003129 events->interrupt.nr = vcpu->arch.interrupt.nr;
Jan Kiszka03b82a32010-02-15 10:45:41 +01003130 events->interrupt.soft = 0;
Paolo Bonzini37ccdcb2014-05-20 14:29:47 +02003131 events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
Jan Kiszka3cfc3092009-11-12 01:04:25 +01003132
3133 events->nmi.injected = vcpu->arch.nmi_injected;
Avi Kivity7460fb4a2011-09-20 13:43:14 +03003134 events->nmi.pending = vcpu->arch.nmi_pending != 0;
Jan Kiszka3cfc3092009-11-12 01:04:25 +01003135 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
Vasiliy Kulikov97e69aa2010-10-30 22:54:47 +04003136 events->nmi.pad = 0;
Jan Kiszka3cfc3092009-11-12 01:04:25 +01003137
Jan Kiszka66450a22013-03-13 12:42:34 +01003138 events->sipi_vector = 0; /* never valid when reporting to user space */
Jan Kiszka3cfc3092009-11-12 01:04:25 +01003139
Paolo Bonzinif0778252015-04-01 15:06:40 +02003140 events->smi.smm = is_smm(vcpu);
3141 events->smi.pending = vcpu->arch.smi_pending;
3142 events->smi.smm_inside_nmi =
3143 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
3144 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
3145
Jan Kiszkadab4b912009-12-06 18:24:15 +01003146 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
Paolo Bonzinif0778252015-04-01 15:06:40 +02003147 | KVM_VCPUEVENT_VALID_SHADOW
3148 | KVM_VCPUEVENT_VALID_SMM);
Vasiliy Kulikov97e69aa2010-10-30 22:54:47 +04003149 memset(&events->reserved, 0, sizeof(events->reserved));
Jan Kiszka3cfc3092009-11-12 01:04:25 +01003150}
3151
Xiao Guangrong6ef4e072016-12-24 10:00:42 +01003152static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags);
3153
Jan Kiszka3cfc3092009-11-12 01:04:25 +01003154static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3155 struct kvm_vcpu_events *events)
3156{
Jan Kiszkadab4b912009-12-06 18:24:15 +01003157 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
Jan Kiszka48005f62010-02-19 19:38:07 +01003158 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
Paolo Bonzinif0778252015-04-01 15:06:40 +02003159 | KVM_VCPUEVENT_VALID_SHADOW
3160 | KVM_VCPUEVENT_VALID_SMM))
Jan Kiszka3cfc3092009-11-12 01:04:25 +01003161 return -EINVAL;
3162
Paolo Bonzini78e546c2016-06-01 14:09:20 +02003163 if (events->exception.injected &&
Jim Mattson28d06352017-04-05 09:14:40 -07003164 (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR ||
3165 is_guest_mode(vcpu)))
Paolo Bonzini78e546c2016-06-01 14:09:20 +02003166 return -EINVAL;
3167
David Hildenbrand28bf2882017-03-23 11:46:03 +01003168 /* INITs are latched while in SMM */
3169 if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
3170 (events->smi.smm || events->smi.pending) &&
3171 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
3172 return -EINVAL;
3173
Avi Kivity7460fb4a2011-09-20 13:43:14 +03003174 process_nmi(vcpu);
Wanpeng Li664f8e22017-08-24 03:35:09 -07003175 vcpu->arch.exception.injected = false;
Jan Kiszka3cfc3092009-11-12 01:04:25 +01003176 vcpu->arch.exception.pending = events->exception.injected;
3177 vcpu->arch.exception.nr = events->exception.nr;
3178 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3179 vcpu->arch.exception.error_code = events->exception.error_code;
3180
3181 vcpu->arch.interrupt.pending = events->interrupt.injected;
3182 vcpu->arch.interrupt.nr = events->interrupt.nr;
3183 vcpu->arch.interrupt.soft = events->interrupt.soft;
Jan Kiszka48005f62010-02-19 19:38:07 +01003184 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3185 kvm_x86_ops->set_interrupt_shadow(vcpu,
3186 events->interrupt.shadow);
Jan Kiszka3cfc3092009-11-12 01:04:25 +01003187
3188 vcpu->arch.nmi_injected = events->nmi.injected;
Jan Kiszkadab4b912009-12-06 18:24:15 +01003189 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3190 vcpu->arch.nmi_pending = events->nmi.pending;
Jan Kiszka3cfc3092009-11-12 01:04:25 +01003191 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3192
Jan Kiszka66450a22013-03-13 12:42:34 +01003193 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
Paolo Bonzinibce87cc2016-01-08 13:48:51 +01003194 lapic_in_kernel(vcpu))
Jan Kiszka66450a22013-03-13 12:42:34 +01003195 vcpu->arch.apic->sipi_vector = events->sipi_vector;
Jan Kiszka3cfc3092009-11-12 01:04:25 +01003196
Paolo Bonzinif0778252015-04-01 15:06:40 +02003197 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
Xiao Guangrong6ef4e072016-12-24 10:00:42 +01003198 u32 hflags = vcpu->arch.hflags;
Paolo Bonzinif0778252015-04-01 15:06:40 +02003199 if (events->smi.smm)
Xiao Guangrong6ef4e072016-12-24 10:00:42 +01003200 hflags |= HF_SMM_MASK;
Paolo Bonzinif0778252015-04-01 15:06:40 +02003201 else
Xiao Guangrong6ef4e072016-12-24 10:00:42 +01003202 hflags &= ~HF_SMM_MASK;
3203 kvm_set_hflags(vcpu, hflags);
3204
Paolo Bonzinif0778252015-04-01 15:06:40 +02003205 vcpu->arch.smi_pending = events->smi.pending;
Wanpeng Lif4ef1912017-08-01 16:05:25 -07003206
3207 if (events->smi.smm) {
3208 if (events->smi.smm_inside_nmi)
3209 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
Paolo Bonzinif0778252015-04-01 15:06:40 +02003210 else
Wanpeng Lif4ef1912017-08-01 16:05:25 -07003211 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
3212 if (lapic_in_kernel(vcpu)) {
3213 if (events->smi.latched_init)
3214 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3215 else
3216 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3217 }
Paolo Bonzinif0778252015-04-01 15:06:40 +02003218 }
3219 }
3220
Avi Kivity3842d132010-07-27 12:30:24 +03003221 kvm_make_request(KVM_REQ_EVENT, vcpu);
3222
Jan Kiszka3cfc3092009-11-12 01:04:25 +01003223 return 0;
3224}
3225
Jan Kiszkaa1efbe72010-02-15 10:45:43 +01003226static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3227 struct kvm_debugregs *dbgregs)
3228{
Jan Kiszka73aaf249e2014-01-04 18:47:16 +01003229 unsigned long val;
3230
Jan Kiszkaa1efbe72010-02-15 10:45:43 +01003231 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
Nadav Amit16f8a6f2014-10-03 01:10:05 +03003232 kvm_get_dr(vcpu, 6, &val);
Jan Kiszka73aaf249e2014-01-04 18:47:16 +01003233 dbgregs->dr6 = val;
Jan Kiszkaa1efbe72010-02-15 10:45:43 +01003234 dbgregs->dr7 = vcpu->arch.dr7;
3235 dbgregs->flags = 0;
Vasiliy Kulikov97e69aa2010-10-30 22:54:47 +04003236 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
Jan Kiszkaa1efbe72010-02-15 10:45:43 +01003237}
3238
3239static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3240 struct kvm_debugregs *dbgregs)
3241{
3242 if (dbgregs->flags)
3243 return -EINVAL;
3244
Paolo Bonzinid14bdb52016-06-01 14:09:23 +02003245 if (dbgregs->dr6 & ~0xffffffffull)
3246 return -EINVAL;
3247 if (dbgregs->dr7 & ~0xffffffffull)
3248 return -EINVAL;
3249
Jan Kiszkaa1efbe72010-02-15 10:45:43 +01003250 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
Nadav Amitae561ed2015-04-02 03:10:37 +03003251 kvm_update_dr0123(vcpu);
Jan Kiszkaa1efbe72010-02-15 10:45:43 +01003252 vcpu->arch.dr6 = dbgregs->dr6;
Jan Kiszka73aaf249e2014-01-04 18:47:16 +01003253 kvm_update_dr6(vcpu);
Jan Kiszkaa1efbe72010-02-15 10:45:43 +01003254 vcpu->arch.dr7 = dbgregs->dr7;
Jan Kiszka9926c9f2014-01-04 18:47:15 +01003255 kvm_update_dr7(vcpu);
Jan Kiszkaa1efbe72010-02-15 10:45:43 +01003256
Jan Kiszkaa1efbe72010-02-15 10:45:43 +01003257 return 0;
3258}
3259
Paolo Bonzinidf1daba2014-11-21 19:05:07 +01003260#define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3261
3262static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3263{
Ingo Molnarc47ada32015-04-30 17:15:32 +02003264 struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
Ingo Molnar400e4b22015-04-24 10:19:47 +02003265 u64 xstate_bv = xsave->header.xfeatures;
Paolo Bonzinidf1daba2014-11-21 19:05:07 +01003266 u64 valid;
3267
3268 /*
3269 * Copy legacy XSAVE area, to avoid complications with CPUID
3270 * leaves 0 and 1 in the loop below.
3271 */
3272 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3273
3274 /* Set XSTATE_BV */
Radim Krčmář00c87e92017-02-01 14:19:53 +01003275 xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
Paolo Bonzinidf1daba2014-11-21 19:05:07 +01003276 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3277
3278 /*
3279 * Copy each region from the possibly compacted offset to the
3280 * non-compacted offset.
3281 */
Dave Hansend91cab72015-09-02 16:31:26 -07003282 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
Paolo Bonzinidf1daba2014-11-21 19:05:07 +01003283 while (valid) {
3284 u64 feature = valid & -valid;
3285 int index = fls64(feature) - 1;
3286 void *src = get_xsave_addr(xsave, feature);
3287
3288 if (src) {
3289 u32 size, offset, ecx, edx;
3290 cpuid_count(XSTATE_CPUID, index,
3291 &size, &offset, &ecx, &edx);
Paolo Bonzini38cfd5e2017-08-23 23:16:29 +02003292 if (feature == XFEATURE_MASK_PKRU)
3293 memcpy(dest + offset, &vcpu->arch.pkru,
3294 sizeof(vcpu->arch.pkru));
3295 else
3296 memcpy(dest + offset, src, size);
3297
Paolo Bonzinidf1daba2014-11-21 19:05:07 +01003298 }
3299
3300 valid -= feature;
3301 }
3302}
3303
3304static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3305{
Ingo Molnarc47ada32015-04-30 17:15:32 +02003306 struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
Paolo Bonzinidf1daba2014-11-21 19:05:07 +01003307 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3308 u64 valid;
3309
3310 /*
3311 * Copy legacy XSAVE area, to avoid complications with CPUID
3312 * leaves 0 and 1 in the loop below.
3313 */
3314 memcpy(xsave, src, XSAVE_HDR_OFFSET);
3315
3316 /* Set XSTATE_BV and possibly XCOMP_BV. */
Ingo Molnar400e4b22015-04-24 10:19:47 +02003317 xsave->header.xfeatures = xstate_bv;
Borislav Petkov782511b2016-04-04 22:25:03 +02003318 if (boot_cpu_has(X86_FEATURE_XSAVES))
Ingo Molnar3a544502015-04-24 10:14:36 +02003319 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
Paolo Bonzinidf1daba2014-11-21 19:05:07 +01003320
3321 /*
3322 * Copy each region from the non-compacted offset to the
3323 * possibly compacted offset.
3324 */
Dave Hansend91cab72015-09-02 16:31:26 -07003325 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
Paolo Bonzinidf1daba2014-11-21 19:05:07 +01003326 while (valid) {
3327 u64 feature = valid & -valid;
3328 int index = fls64(feature) - 1;
3329 void *dest = get_xsave_addr(xsave, feature);
3330
3331 if (dest) {
3332 u32 size, offset, ecx, edx;
3333 cpuid_count(XSTATE_CPUID, index,
3334 &size, &offset, &ecx, &edx);
Paolo Bonzini38cfd5e2017-08-23 23:16:29 +02003335 if (feature == XFEATURE_MASK_PKRU)
3336 memcpy(&vcpu->arch.pkru, src + offset,
3337 sizeof(vcpu->arch.pkru));
3338 else
3339 memcpy(dest, src + offset, size);
Wanpeng Liee4100d2015-07-09 15:44:52 +08003340 }
Paolo Bonzinidf1daba2014-11-21 19:05:07 +01003341
3342 valid -= feature;
3343 }
3344}
3345
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003346static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3347 struct kvm_xsave *guest_xsave)
3348{
Borislav Petkovd366bf72016-04-04 22:25:02 +02003349 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
Paolo Bonzinidf1daba2014-11-21 19:05:07 +01003350 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3351 fill_xsave((u8 *) guest_xsave->region, vcpu);
Paolo Bonzini4344ee92013-10-02 16:06:16 +02003352 } else {
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003353 memcpy(guest_xsave->region,
Ingo Molnar7366ed72015-04-27 04:19:39 +02003354 &vcpu->arch.guest_fpu.state.fxsave,
Ingo Molnarc47ada32015-04-30 17:15:32 +02003355 sizeof(struct fxregs_state));
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003356 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
Dave Hansend91cab72015-09-02 16:31:26 -07003357 XFEATURE_MASK_FPSSE;
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003358 }
3359}
3360
Wanpeng Lia5758132017-05-11 02:58:55 -07003361#define XSAVE_MXCSR_OFFSET 24
3362
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003363static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3364 struct kvm_xsave *guest_xsave)
3365{
3366 u64 xstate_bv =
3367 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
Wanpeng Lia5758132017-05-11 02:58:55 -07003368 u32 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003369
Borislav Petkovd366bf72016-04-04 22:25:02 +02003370 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
Paolo Bonzinid7876f12013-10-02 16:06:15 +02003371 /*
3372 * Here we allow setting states that are not present in
3373 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
3374 * with old userspace.
3375 */
Wanpeng Lia5758132017-05-11 02:58:55 -07003376 if (xstate_bv & ~kvm_supported_xcr0() ||
3377 mxcsr & ~mxcsr_feature_mask)
Paolo Bonzinid7876f12013-10-02 16:06:15 +02003378 return -EINVAL;
Paolo Bonzinidf1daba2014-11-21 19:05:07 +01003379 load_xsave(vcpu, (u8 *)guest_xsave->region);
Paolo Bonzinid7876f12013-10-02 16:06:15 +02003380 } else {
Wanpeng Lia5758132017-05-11 02:58:55 -07003381 if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
3382 mxcsr & ~mxcsr_feature_mask)
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003383 return -EINVAL;
Ingo Molnar7366ed72015-04-27 04:19:39 +02003384 memcpy(&vcpu->arch.guest_fpu.state.fxsave,
Ingo Molnarc47ada32015-04-30 17:15:32 +02003385 guest_xsave->region, sizeof(struct fxregs_state));
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003386 }
3387 return 0;
3388}
3389
3390static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3391 struct kvm_xcrs *guest_xcrs)
3392{
Borislav Petkovd366bf72016-04-04 22:25:02 +02003393 if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003394 guest_xcrs->nr_xcrs = 0;
3395 return;
3396 }
3397
3398 guest_xcrs->nr_xcrs = 1;
3399 guest_xcrs->flags = 0;
3400 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3401 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3402}
3403
3404static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3405 struct kvm_xcrs *guest_xcrs)
3406{
3407 int i, r = 0;
3408
Borislav Petkovd366bf72016-04-04 22:25:02 +02003409 if (!boot_cpu_has(X86_FEATURE_XSAVE))
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003410 return -EINVAL;
3411
3412 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3413 return -EINVAL;
3414
3415 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3416 /* Only support XCR0 currently */
Paolo Bonzinic67a04c2013-10-17 16:50:47 +02003417 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003418 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
Paolo Bonzinic67a04c2013-10-17 16:50:47 +02003419 guest_xcrs->xcrs[i].value);
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003420 break;
3421 }
3422 if (r)
3423 r = -EINVAL;
3424 return r;
3425}
3426
Eric B Munson1c0b28c2012-03-10 14:37:27 -05003427/*
3428 * kvm_set_guest_paused() indicates to the guest kernel that it has been
3429 * stopped by the hypervisor. This function will be called from the host only.
3430 * EINVAL is returned when the host attempts to set the flag for a guest that
3431 * does not support pv clocks.
3432 */
3433static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3434{
Andy Honig0b794592013-02-20 14:48:10 -08003435 if (!vcpu->arch.pv_time_enabled)
Eric B Munson1c0b28c2012-03-10 14:37:27 -05003436 return -EINVAL;
Marcelo Tosatti51d59c62012-08-03 15:57:49 -03003437 vcpu->arch.pvclock_set_guest_stopped_request = true;
Eric B Munson1c0b28c2012-03-10 14:37:27 -05003438 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3439 return 0;
3440}
3441
Andrey Smetanin5c9194122015-11-10 15:36:34 +03003442static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
3443 struct kvm_enable_cap *cap)
3444{
3445 if (cap->flags)
3446 return -EINVAL;
3447
3448 switch (cap->cap) {
Roman Kaganefc479e2017-06-22 16:51:01 +03003449 case KVM_CAP_HYPERV_SYNIC2:
3450 if (cap->args[0])
3451 return -EINVAL;
Andrey Smetanin5c9194122015-11-10 15:36:34 +03003452 case KVM_CAP_HYPERV_SYNIC:
Wanpeng Li546d87e2017-01-03 18:56:19 -08003453 if (!irqchip_in_kernel(vcpu->kvm))
3454 return -EINVAL;
Roman Kaganefc479e2017-06-22 16:51:01 +03003455 return kvm_hv_activate_synic(vcpu, cap->cap ==
3456 KVM_CAP_HYPERV_SYNIC2);
Andrey Smetanin5c9194122015-11-10 15:36:34 +03003457 default:
3458 return -EINVAL;
3459 }
3460}
3461
Carsten Otte313a3dc2007-10-11 19:16:52 +02003462long kvm_arch_vcpu_ioctl(struct file *filp,
3463 unsigned int ioctl, unsigned long arg)
3464{
3465 struct kvm_vcpu *vcpu = filp->private_data;
3466 void __user *argp = (void __user *)arg;
3467 int r;
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003468 union {
3469 struct kvm_lapic_state *lapic;
3470 struct kvm_xsave *xsave;
3471 struct kvm_xcrs *xcrs;
3472 void *buffer;
3473 } u;
Carsten Otte313a3dc2007-10-11 19:16:52 +02003474
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003475 u.buffer = NULL;
Carsten Otte313a3dc2007-10-11 19:16:52 +02003476 switch (ioctl) {
3477 case KVM_GET_LAPIC: {
Marcelo Tosatti2204ae32009-10-29 13:44:16 -02003478 r = -EINVAL;
Paolo Bonzinibce87cc2016-01-08 13:48:51 +01003479 if (!lapic_in_kernel(vcpu))
Marcelo Tosatti2204ae32009-10-29 13:44:16 -02003480 goto out;
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003481 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
Carsten Otte313a3dc2007-10-11 19:16:52 +02003482
Dave Hansenb772ff32008-08-11 10:01:47 -07003483 r = -ENOMEM;
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003484 if (!u.lapic)
Dave Hansenb772ff32008-08-11 10:01:47 -07003485 goto out;
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003486 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
Carsten Otte313a3dc2007-10-11 19:16:52 +02003487 if (r)
3488 goto out;
3489 r = -EFAULT;
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003490 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
Carsten Otte313a3dc2007-10-11 19:16:52 +02003491 goto out;
3492 r = 0;
3493 break;
3494 }
3495 case KVM_SET_LAPIC: {
Marcelo Tosatti2204ae32009-10-29 13:44:16 -02003496 r = -EINVAL;
Paolo Bonzinibce87cc2016-01-08 13:48:51 +01003497 if (!lapic_in_kernel(vcpu))
Marcelo Tosatti2204ae32009-10-29 13:44:16 -02003498 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02003499 u.lapic = memdup_user(argp, sizeof(*u.lapic));
Guo Chao18595412012-11-02 18:33:21 +08003500 if (IS_ERR(u.lapic))
3501 return PTR_ERR(u.lapic);
Sasha Levinff5c2c02011-12-04 19:36:29 +02003502
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003503 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
Carsten Otte313a3dc2007-10-11 19:16:52 +02003504 break;
3505 }
Zhang Xiantaof77bc6a2007-11-21 04:36:41 +08003506 case KVM_INTERRUPT: {
3507 struct kvm_interrupt irq;
3508
3509 r = -EFAULT;
3510 if (copy_from_user(&irq, argp, sizeof irq))
3511 goto out;
3512 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
Zhang Xiantaof77bc6a2007-11-21 04:36:41 +08003513 break;
3514 }
Jan Kiszkac4abb7c2008-09-26 09:30:55 +02003515 case KVM_NMI: {
3516 r = kvm_vcpu_ioctl_nmi(vcpu);
Jan Kiszkac4abb7c2008-09-26 09:30:55 +02003517 break;
3518 }
Paolo Bonzinif0778252015-04-01 15:06:40 +02003519 case KVM_SMI: {
3520 r = kvm_vcpu_ioctl_smi(vcpu);
3521 break;
3522 }
Carsten Otte313a3dc2007-10-11 19:16:52 +02003523 case KVM_SET_CPUID: {
3524 struct kvm_cpuid __user *cpuid_arg = argp;
3525 struct kvm_cpuid cpuid;
3526
3527 r = -EFAULT;
3528 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3529 goto out;
3530 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
Carsten Otte313a3dc2007-10-11 19:16:52 +02003531 break;
3532 }
Dan Kenigsberg07716712007-11-21 17:10:04 +02003533 case KVM_SET_CPUID2: {
3534 struct kvm_cpuid2 __user *cpuid_arg = argp;
3535 struct kvm_cpuid2 cpuid;
3536
3537 r = -EFAULT;
3538 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3539 goto out;
3540 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
Amit Shah19355472009-01-14 16:56:00 +00003541 cpuid_arg->entries);
Dan Kenigsberg07716712007-11-21 17:10:04 +02003542 break;
3543 }
3544 case KVM_GET_CPUID2: {
3545 struct kvm_cpuid2 __user *cpuid_arg = argp;
3546 struct kvm_cpuid2 cpuid;
3547
3548 r = -EFAULT;
3549 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3550 goto out;
3551 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
Amit Shah19355472009-01-14 16:56:00 +00003552 cpuid_arg->entries);
Dan Kenigsberg07716712007-11-21 17:10:04 +02003553 if (r)
3554 goto out;
3555 r = -EFAULT;
3556 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3557 goto out;
3558 r = 0;
3559 break;
3560 }
Carsten Otte313a3dc2007-10-11 19:16:52 +02003561 case KVM_GET_MSRS:
Paolo Bonzini609e36d2015-04-08 15:30:38 +02003562 r = msr_io(vcpu, argp, do_get_msr, 1);
Carsten Otte313a3dc2007-10-11 19:16:52 +02003563 break;
3564 case KVM_SET_MSRS:
3565 r = msr_io(vcpu, argp, do_set_msr, 0);
3566 break;
Avi Kivityb209749f2007-10-22 16:50:39 +02003567 case KVM_TPR_ACCESS_REPORTING: {
3568 struct kvm_tpr_access_ctl tac;
3569
3570 r = -EFAULT;
3571 if (copy_from_user(&tac, argp, sizeof tac))
3572 goto out;
3573 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3574 if (r)
3575 goto out;
3576 r = -EFAULT;
3577 if (copy_to_user(argp, &tac, sizeof tac))
3578 goto out;
3579 r = 0;
3580 break;
3581 };
Avi Kivityb93463a2007-10-25 16:52:32 +02003582 case KVM_SET_VAPIC_ADDR: {
3583 struct kvm_vapic_addr va;
Paolo Bonzini7301d6a2016-11-17 15:55:46 +01003584 int idx;
Avi Kivityb93463a2007-10-25 16:52:32 +02003585
3586 r = -EINVAL;
Paolo Bonzini35754c92015-07-29 12:05:37 +02003587 if (!lapic_in_kernel(vcpu))
Avi Kivityb93463a2007-10-25 16:52:32 +02003588 goto out;
3589 r = -EFAULT;
3590 if (copy_from_user(&va, argp, sizeof va))
3591 goto out;
Paolo Bonzini7301d6a2016-11-17 15:55:46 +01003592 idx = srcu_read_lock(&vcpu->kvm->srcu);
Andy Honigfda4e2e2013-11-20 10:23:22 -08003593 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
Paolo Bonzini7301d6a2016-11-17 15:55:46 +01003594 srcu_read_unlock(&vcpu->kvm->srcu, idx);
Avi Kivityb93463a2007-10-25 16:52:32 +02003595 break;
3596 }
Huang Ying890ca9a2009-05-11 16:48:15 +08003597 case KVM_X86_SETUP_MCE: {
3598 u64 mcg_cap;
3599
3600 r = -EFAULT;
3601 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3602 goto out;
3603 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3604 break;
3605 }
3606 case KVM_X86_SET_MCE: {
3607 struct kvm_x86_mce mce;
3608
3609 r = -EFAULT;
3610 if (copy_from_user(&mce, argp, sizeof mce))
3611 goto out;
3612 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3613 break;
3614 }
Jan Kiszka3cfc3092009-11-12 01:04:25 +01003615 case KVM_GET_VCPU_EVENTS: {
3616 struct kvm_vcpu_events events;
3617
3618 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3619
3620 r = -EFAULT;
3621 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3622 break;
3623 r = 0;
3624 break;
3625 }
3626 case KVM_SET_VCPU_EVENTS: {
3627 struct kvm_vcpu_events events;
3628
3629 r = -EFAULT;
3630 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3631 break;
3632
3633 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3634 break;
3635 }
Jan Kiszkaa1efbe72010-02-15 10:45:43 +01003636 case KVM_GET_DEBUGREGS: {
3637 struct kvm_debugregs dbgregs;
3638
3639 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3640
3641 r = -EFAULT;
3642 if (copy_to_user(argp, &dbgregs,
3643 sizeof(struct kvm_debugregs)))
3644 break;
3645 r = 0;
3646 break;
3647 }
3648 case KVM_SET_DEBUGREGS: {
3649 struct kvm_debugregs dbgregs;
3650
3651 r = -EFAULT;
3652 if (copy_from_user(&dbgregs, argp,
3653 sizeof(struct kvm_debugregs)))
3654 break;
3655
3656 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3657 break;
3658 }
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003659 case KVM_GET_XSAVE: {
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003660 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003661 r = -ENOMEM;
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003662 if (!u.xsave)
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003663 break;
3664
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003665 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003666
3667 r = -EFAULT;
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003668 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003669 break;
3670 r = 0;
3671 break;
3672 }
3673 case KVM_SET_XSAVE: {
Sasha Levinff5c2c02011-12-04 19:36:29 +02003674 u.xsave = memdup_user(argp, sizeof(*u.xsave));
Guo Chao18595412012-11-02 18:33:21 +08003675 if (IS_ERR(u.xsave))
3676 return PTR_ERR(u.xsave);
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003677
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003678 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003679 break;
3680 }
3681 case KVM_GET_XCRS: {
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003682 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003683 r = -ENOMEM;
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003684 if (!u.xcrs)
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003685 break;
3686
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003687 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003688
3689 r = -EFAULT;
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003690 if (copy_to_user(argp, u.xcrs,
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003691 sizeof(struct kvm_xcrs)))
3692 break;
3693 r = 0;
3694 break;
3695 }
3696 case KVM_SET_XCRS: {
Sasha Levinff5c2c02011-12-04 19:36:29 +02003697 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
Guo Chao18595412012-11-02 18:33:21 +08003698 if (IS_ERR(u.xcrs))
3699 return PTR_ERR(u.xcrs);
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003700
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003701 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003702 break;
3703 }
Joerg Roedel92a1f122011-03-25 09:44:51 +01003704 case KVM_SET_TSC_KHZ: {
3705 u32 user_tsc_khz;
3706
3707 r = -EINVAL;
Joerg Roedel92a1f122011-03-25 09:44:51 +01003708 user_tsc_khz = (u32)arg;
3709
3710 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3711 goto out;
3712
Zachary Amsdencc578282012-02-03 15:43:50 -02003713 if (user_tsc_khz == 0)
3714 user_tsc_khz = tsc_khz;
3715
Haozhong Zhang381d5852015-10-20 15:39:04 +08003716 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
3717 r = 0;
Joerg Roedel92a1f122011-03-25 09:44:51 +01003718
Joerg Roedel92a1f122011-03-25 09:44:51 +01003719 goto out;
3720 }
3721 case KVM_GET_TSC_KHZ: {
Zachary Amsdencc578282012-02-03 15:43:50 -02003722 r = vcpu->arch.virtual_tsc_khz;
Joerg Roedel92a1f122011-03-25 09:44:51 +01003723 goto out;
3724 }
Eric B Munson1c0b28c2012-03-10 14:37:27 -05003725 case KVM_KVMCLOCK_CTRL: {
3726 r = kvm_set_guest_paused(vcpu);
3727 goto out;
3728 }
Andrey Smetanin5c9194122015-11-10 15:36:34 +03003729 case KVM_ENABLE_CAP: {
3730 struct kvm_enable_cap cap;
3731
3732 r = -EFAULT;
3733 if (copy_from_user(&cap, argp, sizeof(cap)))
3734 goto out;
3735 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
3736 break;
3737 }
Carsten Otte313a3dc2007-10-11 19:16:52 +02003738 default:
3739 r = -EINVAL;
3740 }
3741out:
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003742 kfree(u.buffer);
Carsten Otte313a3dc2007-10-11 19:16:52 +02003743 return r;
3744}
3745
Carsten Otte5b1c1492012-01-04 10:25:23 +01003746int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3747{
3748 return VM_FAULT_SIGBUS;
3749}
3750
Carsten Otte1fe779f2007-10-29 16:08:35 +01003751static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3752{
3753 int ret;
3754
3755 if (addr > (unsigned int)(-3 * PAGE_SIZE))
Guo Chao951179c2012-11-02 18:33:22 +08003756 return -EINVAL;
Carsten Otte1fe779f2007-10-29 16:08:35 +01003757 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3758 return ret;
3759}
3760
Sheng Yangb927a3c2009-07-21 10:42:48 +08003761static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3762 u64 ident_addr)
3763{
3764 kvm->arch.ept_identity_map_addr = ident_addr;
3765 return 0;
3766}
3767
Carsten Otte1fe779f2007-10-29 16:08:35 +01003768static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3769 u32 kvm_nr_mmu_pages)
3770{
3771 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3772 return -EINVAL;
3773
Marcelo Tosatti79fac952009-12-23 14:35:26 -02003774 mutex_lock(&kvm->slots_lock);
Carsten Otte1fe779f2007-10-29 16:08:35 +01003775
3776 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
Zhang Xiantaof05e70a2007-12-14 10:01:48 +08003777 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
Carsten Otte1fe779f2007-10-29 16:08:35 +01003778
Marcelo Tosatti79fac952009-12-23 14:35:26 -02003779 mutex_unlock(&kvm->slots_lock);
Carsten Otte1fe779f2007-10-29 16:08:35 +01003780 return 0;
3781}
3782
3783static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3784{
Dave Hansen39de71e2010-08-19 18:11:14 -07003785 return kvm->arch.n_max_mmu_pages;
Carsten Otte1fe779f2007-10-29 16:08:35 +01003786}
3787
Carsten Otte1fe779f2007-10-29 16:08:35 +01003788static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3789{
David Hildenbrand90bca052017-04-07 10:50:23 +02003790 struct kvm_pic *pic = kvm->arch.vpic;
Carsten Otte1fe779f2007-10-29 16:08:35 +01003791 int r;
3792
3793 r = 0;
3794 switch (chip->chip_id) {
3795 case KVM_IRQCHIP_PIC_MASTER:
David Hildenbrand90bca052017-04-07 10:50:23 +02003796 memcpy(&chip->chip.pic, &pic->pics[0],
Carsten Otte1fe779f2007-10-29 16:08:35 +01003797 sizeof(struct kvm_pic_state));
3798 break;
3799 case KVM_IRQCHIP_PIC_SLAVE:
David Hildenbrand90bca052017-04-07 10:50:23 +02003800 memcpy(&chip->chip.pic, &pic->pics[1],
Carsten Otte1fe779f2007-10-29 16:08:35 +01003801 sizeof(struct kvm_pic_state));
3802 break;
3803 case KVM_IRQCHIP_IOAPIC:
David Hildenbrand33392b42017-04-07 10:50:27 +02003804 kvm_get_ioapic(kvm, &chip->chip.ioapic);
Carsten Otte1fe779f2007-10-29 16:08:35 +01003805 break;
3806 default:
3807 r = -EINVAL;
3808 break;
3809 }
3810 return r;
3811}
3812
3813static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3814{
David Hildenbrand90bca052017-04-07 10:50:23 +02003815 struct kvm_pic *pic = kvm->arch.vpic;
Carsten Otte1fe779f2007-10-29 16:08:35 +01003816 int r;
3817
3818 r = 0;
3819 switch (chip->chip_id) {
3820 case KVM_IRQCHIP_PIC_MASTER:
David Hildenbrand90bca052017-04-07 10:50:23 +02003821 spin_lock(&pic->lock);
3822 memcpy(&pic->pics[0], &chip->chip.pic,
Carsten Otte1fe779f2007-10-29 16:08:35 +01003823 sizeof(struct kvm_pic_state));
David Hildenbrand90bca052017-04-07 10:50:23 +02003824 spin_unlock(&pic->lock);
Carsten Otte1fe779f2007-10-29 16:08:35 +01003825 break;
3826 case KVM_IRQCHIP_PIC_SLAVE:
David Hildenbrand90bca052017-04-07 10:50:23 +02003827 spin_lock(&pic->lock);
3828 memcpy(&pic->pics[1], &chip->chip.pic,
Carsten Otte1fe779f2007-10-29 16:08:35 +01003829 sizeof(struct kvm_pic_state));
David Hildenbrand90bca052017-04-07 10:50:23 +02003830 spin_unlock(&pic->lock);
Carsten Otte1fe779f2007-10-29 16:08:35 +01003831 break;
3832 case KVM_IRQCHIP_IOAPIC:
David Hildenbrand33392b42017-04-07 10:50:27 +02003833 kvm_set_ioapic(kvm, &chip->chip.ioapic);
Carsten Otte1fe779f2007-10-29 16:08:35 +01003834 break;
3835 default:
3836 r = -EINVAL;
3837 break;
3838 }
David Hildenbrand90bca052017-04-07 10:50:23 +02003839 kvm_pic_update_irq(pic);
Carsten Otte1fe779f2007-10-29 16:08:35 +01003840 return r;
3841}
3842
Sheng Yange0f63cb2008-03-04 00:50:59 +08003843static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3844{
Radim Krčmář34f39412016-03-02 22:56:50 +01003845 struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
3846
3847 BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
3848
3849 mutex_lock(&kps->lock);
3850 memcpy(ps, &kps->channels, sizeof(*ps));
3851 mutex_unlock(&kps->lock);
Saurabh Sengar2da29bc2015-10-30 12:56:11 +05303852 return 0;
Sheng Yange0f63cb2008-03-04 00:50:59 +08003853}
3854
3855static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3856{
Andrew Honig01856042015-11-18 14:50:23 -08003857 int i;
Radim Krčmář09edea72016-03-02 22:56:43 +01003858 struct kvm_pit *pit = kvm->arch.vpit;
3859
3860 mutex_lock(&pit->pit_state.lock);
Radim Krčmář34f39412016-03-02 22:56:50 +01003861 memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
Andrew Honig01856042015-11-18 14:50:23 -08003862 for (i = 0; i < 3; i++)
Radim Krčmář09edea72016-03-02 22:56:43 +01003863 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
3864 mutex_unlock(&pit->pit_state.lock);
Saurabh Sengar2da29bc2015-10-30 12:56:11 +05303865 return 0;
Beth Kone9f42752009-07-07 11:50:38 -04003866}
3867
3868static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3869{
Beth Kone9f42752009-07-07 11:50:38 -04003870 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3871 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3872 sizeof(ps->channels));
3873 ps->flags = kvm->arch.vpit->pit_state.flags;
3874 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
Vasiliy Kulikov97e69aa2010-10-30 22:54:47 +04003875 memset(&ps->reserved, 0, sizeof(ps->reserved));
Saurabh Sengar2da29bc2015-10-30 12:56:11 +05303876 return 0;
Beth Kone9f42752009-07-07 11:50:38 -04003877}
3878
3879static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3880{
Saurabh Sengar2da29bc2015-10-30 12:56:11 +05303881 int start = 0;
Andrew Honig01856042015-11-18 14:50:23 -08003882 int i;
Beth Kone9f42752009-07-07 11:50:38 -04003883 u32 prev_legacy, cur_legacy;
Radim Krčmář09edea72016-03-02 22:56:43 +01003884 struct kvm_pit *pit = kvm->arch.vpit;
3885
3886 mutex_lock(&pit->pit_state.lock);
3887 prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
Beth Kone9f42752009-07-07 11:50:38 -04003888 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3889 if (!prev_legacy && cur_legacy)
3890 start = 1;
Radim Krčmář09edea72016-03-02 22:56:43 +01003891 memcpy(&pit->pit_state.channels, &ps->channels,
3892 sizeof(pit->pit_state.channels));
3893 pit->pit_state.flags = ps->flags;
Andrew Honig01856042015-11-18 14:50:23 -08003894 for (i = 0; i < 3; i++)
Radim Krčmář09edea72016-03-02 22:56:43 +01003895 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
Paolo Bonzinie5e57e72016-01-07 13:50:38 +01003896 start && i == 0);
Radim Krčmář09edea72016-03-02 22:56:43 +01003897 mutex_unlock(&pit->pit_state.lock);
Saurabh Sengar2da29bc2015-10-30 12:56:11 +05303898 return 0;
Sheng Yange0f63cb2008-03-04 00:50:59 +08003899}
3900
Marcelo Tosatti52d939a2008-12-30 15:55:06 -02003901static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3902 struct kvm_reinject_control *control)
3903{
Radim Krčmář71474e22016-03-02 22:56:45 +01003904 struct kvm_pit *pit = kvm->arch.vpit;
3905
3906 if (!pit)
Marcelo Tosatti52d939a2008-12-30 15:55:06 -02003907 return -ENXIO;
Radim Krčmářb39c90b2016-03-02 22:56:44 +01003908
Radim Krčmář71474e22016-03-02 22:56:45 +01003909 /* pit->pit_state.lock was overloaded to prevent userspace from getting
3910 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
3911 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
3912 */
3913 mutex_lock(&pit->pit_state.lock);
3914 kvm_pit_set_reinject(pit, control->pit_reinject);
3915 mutex_unlock(&pit->pit_state.lock);
Radim Krčmářb39c90b2016-03-02 22:56:44 +01003916
Marcelo Tosatti52d939a2008-12-30 15:55:06 -02003917 return 0;
3918}
3919
Takuya Yoshikawa95d4c162011-11-14 18:24:50 +09003920/**
Takuya Yoshikawa60c34612012-03-03 14:21:48 +09003921 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3922 * @kvm: kvm instance
3923 * @log: slot id and address to which we copy the log
Takuya Yoshikawa95d4c162011-11-14 18:24:50 +09003924 *
Paolo Bonzinie108ff22015-01-15 15:58:54 -08003925 * Steps 1-4 below provide general overview of dirty page logging. See
3926 * kvm_get_dirty_log_protect() function description for additional details.
3927 *
3928 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
3929 * always flush the TLB (step 4) even if previous step failed and the dirty
3930 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
3931 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
3932 * writes will be marked dirty for next log read.
Takuya Yoshikawa95d4c162011-11-14 18:24:50 +09003933 *
Takuya Yoshikawa60c34612012-03-03 14:21:48 +09003934 * 1. Take a snapshot of the bit and clear it if needed.
3935 * 2. Write protect the corresponding page.
Paolo Bonzinie108ff22015-01-15 15:58:54 -08003936 * 3. Copy the snapshot to the userspace.
3937 * 4. Flush TLB's if needed.
Takuya Yoshikawa95d4c162011-11-14 18:24:50 +09003938 */
Takuya Yoshikawa60c34612012-03-03 14:21:48 +09003939int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
Zhang Xiantao5bb064d2007-11-18 20:29:43 +08003940{
Takuya Yoshikawa60c34612012-03-03 14:21:48 +09003941 bool is_dirty = false;
Paolo Bonzinie108ff22015-01-15 15:58:54 -08003942 int r;
Zhang Xiantao5bb064d2007-11-18 20:29:43 +08003943
Marcelo Tosatti79fac952009-12-23 14:35:26 -02003944 mutex_lock(&kvm->slots_lock);
Zhang Xiantao5bb064d2007-11-18 20:29:43 +08003945
Kai Huang88178fd2015-01-28 10:54:27 +08003946 /*
3947 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
3948 */
3949 if (kvm_x86_ops->flush_log_dirty)
3950 kvm_x86_ops->flush_log_dirty(kvm);
3951
Paolo Bonzinie108ff22015-01-15 15:58:54 -08003952 r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
Xiao Guangrong198c74f2014-04-17 17:06:16 +08003953
3954 /*
3955 * All the TLBs can be flushed out of mmu lock, see the comments in
3956 * kvm_mmu_slot_remove_write_access().
3957 */
Paolo Bonzinie108ff22015-01-15 15:58:54 -08003958 lockdep_assert_held(&kvm->slots_lock);
Xiao Guangrong198c74f2014-04-17 17:06:16 +08003959 if (is_dirty)
3960 kvm_flush_remote_tlbs(kvm);
3961
Marcelo Tosatti79fac952009-12-23 14:35:26 -02003962 mutex_unlock(&kvm->slots_lock);
Zhang Xiantao5bb064d2007-11-18 20:29:43 +08003963 return r;
3964}
3965
Yang Zhangaa2fbe62013-04-11 19:21:40 +08003966int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
3967 bool line_status)
Christoffer Dall23d43cf2012-07-24 08:51:20 -04003968{
3969 if (!irqchip_in_kernel(kvm))
3970 return -ENXIO;
3971
3972 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
Yang Zhangaa2fbe62013-04-11 19:21:40 +08003973 irq_event->irq, irq_event->level,
3974 line_status);
Christoffer Dall23d43cf2012-07-24 08:51:20 -04003975 return 0;
3976}
3977
Nadav Amit90de4a12015-04-13 01:53:41 +03003978static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3979 struct kvm_enable_cap *cap)
3980{
3981 int r;
3982
3983 if (cap->flags)
3984 return -EINVAL;
3985
3986 switch (cap->cap) {
3987 case KVM_CAP_DISABLE_QUIRKS:
3988 kvm->arch.disabled_quirks = cap->args[0];
3989 r = 0;
3990 break;
Steve Rutherford49df6392015-07-29 23:21:40 -07003991 case KVM_CAP_SPLIT_IRQCHIP: {
3992 mutex_lock(&kvm->lock);
Steve Rutherfordb053b2a2015-07-29 23:32:35 -07003993 r = -EINVAL;
3994 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
3995 goto split_irqchip_unlock;
Steve Rutherford49df6392015-07-29 23:21:40 -07003996 r = -EEXIST;
3997 if (irqchip_in_kernel(kvm))
3998 goto split_irqchip_unlock;
Paolo Bonzini557abc42016-06-13 14:50:04 +02003999 if (kvm->created_vcpus)
Steve Rutherford49df6392015-07-29 23:21:40 -07004000 goto split_irqchip_unlock;
4001 r = kvm_setup_empty_irq_routing(kvm);
David Hildenbrand5c0aea02017-04-28 17:06:20 +02004002 if (r)
Steve Rutherford49df6392015-07-29 23:21:40 -07004003 goto split_irqchip_unlock;
4004 /* Pairs with irqchip_in_kernel. */
4005 smp_wmb();
Radim Krčmář49776fa2016-12-16 16:10:02 +01004006 kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
Steve Rutherfordb053b2a2015-07-29 23:32:35 -07004007 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
Steve Rutherford49df6392015-07-29 23:21:40 -07004008 r = 0;
4009split_irqchip_unlock:
4010 mutex_unlock(&kvm->lock);
4011 break;
4012 }
Radim Krčmář371313132016-07-12 22:09:27 +02004013 case KVM_CAP_X2APIC_API:
4014 r = -EINVAL;
4015 if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
4016 break;
4017
4018 if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
4019 kvm->arch.x2apic_format = true;
Radim Krčmářc5192652016-07-12 22:09:28 +02004020 if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
4021 kvm->arch.x2apic_broadcast_quirk_disabled = true;
Radim Krčmář371313132016-07-12 22:09:27 +02004022
4023 r = 0;
4024 break;
Nadav Amit90de4a12015-04-13 01:53:41 +03004025 default:
4026 r = -EINVAL;
4027 break;
4028 }
4029 return r;
4030}
4031
Carsten Otte1fe779f2007-10-29 16:08:35 +01004032long kvm_arch_vm_ioctl(struct file *filp,
4033 unsigned int ioctl, unsigned long arg)
4034{
4035 struct kvm *kvm = filp->private_data;
4036 void __user *argp = (void __user *)arg;
Avi Kivity367e1312009-08-26 14:57:07 +03004037 int r = -ENOTTY;
Dave Hansenf0d66272008-08-11 10:01:45 -07004038 /*
4039 * This union makes it completely explicit to gcc-3.x
4040 * that these two variables' stack usage should be
4041 * combined, not added together.
4042 */
4043 union {
4044 struct kvm_pit_state ps;
Beth Kone9f42752009-07-07 11:50:38 -04004045 struct kvm_pit_state2 ps2;
Jan Kiszkac5ff41c2009-05-14 22:42:53 +02004046 struct kvm_pit_config pit_config;
Dave Hansenf0d66272008-08-11 10:01:45 -07004047 } u;
Carsten Otte1fe779f2007-10-29 16:08:35 +01004048
4049 switch (ioctl) {
4050 case KVM_SET_TSS_ADDR:
4051 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
Carsten Otte1fe779f2007-10-29 16:08:35 +01004052 break;
Sheng Yangb927a3c2009-07-21 10:42:48 +08004053 case KVM_SET_IDENTITY_MAP_ADDR: {
4054 u64 ident_addr;
4055
David Hildenbrand1af1ac92017-08-24 20:51:36 +02004056 mutex_lock(&kvm->lock);
4057 r = -EINVAL;
4058 if (kvm->created_vcpus)
4059 goto set_identity_unlock;
Sheng Yangb927a3c2009-07-21 10:42:48 +08004060 r = -EFAULT;
4061 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
David Hildenbrand1af1ac92017-08-24 20:51:36 +02004062 goto set_identity_unlock;
Sheng Yangb927a3c2009-07-21 10:42:48 +08004063 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
David Hildenbrand1af1ac92017-08-24 20:51:36 +02004064set_identity_unlock:
4065 mutex_unlock(&kvm->lock);
Sheng Yangb927a3c2009-07-21 10:42:48 +08004066 break;
4067 }
Carsten Otte1fe779f2007-10-29 16:08:35 +01004068 case KVM_SET_NR_MMU_PAGES:
4069 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
Carsten Otte1fe779f2007-10-29 16:08:35 +01004070 break;
4071 case KVM_GET_NR_MMU_PAGES:
4072 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
4073 break;
Marcelo Tosatti3ddea122009-10-29 13:44:15 -02004074 case KVM_CREATE_IRQCHIP: {
Marcelo Tosatti3ddea122009-10-29 13:44:15 -02004075 mutex_lock(&kvm->lock);
Radim Krčmář09941362016-12-16 16:10:03 +01004076
Marcelo Tosatti3ddea122009-10-29 13:44:15 -02004077 r = -EEXIST;
Radim Krčmář35e6eaa2016-12-16 16:10:01 +01004078 if (irqchip_in_kernel(kvm))
Marcelo Tosatti3ddea122009-10-29 13:44:15 -02004079 goto create_irqchip_unlock;
Radim Krčmář09941362016-12-16 16:10:03 +01004080
Avi Kivity3e515702012-03-05 14:23:29 +02004081 r = -EINVAL;
Paolo Bonzini557abc42016-06-13 14:50:04 +02004082 if (kvm->created_vcpus)
Avi Kivity3e515702012-03-05 14:23:29 +02004083 goto create_irqchip_unlock;
Radim Krčmář09941362016-12-16 16:10:03 +01004084
4085 r = kvm_pic_init(kvm);
4086 if (r)
Marcelo Tosatti3ddea122009-10-29 13:44:15 -02004087 goto create_irqchip_unlock;
Radim Krčmář09941362016-12-16 16:10:03 +01004088
4089 r = kvm_ioapic_init(kvm);
4090 if (r) {
Radim Krčmář09941362016-12-16 16:10:03 +01004091 kvm_pic_destroy(kvm);
Radim Krčmář09941362016-12-16 16:10:03 +01004092 goto create_irqchip_unlock;
4093 }
4094
Avi Kivity399ec802008-11-19 13:58:46 +02004095 r = kvm_setup_default_irq_routing(kvm);
4096 if (r) {
Wei Yongjun72bb2fc2010-02-09 10:33:03 +08004097 kvm_ioapic_destroy(kvm);
Radim Krčmář09941362016-12-16 16:10:03 +01004098 kvm_pic_destroy(kvm);
Paolo Bonzini71ba9942015-07-29 12:31:15 +02004099 goto create_irqchip_unlock;
Avi Kivity399ec802008-11-19 13:58:46 +02004100 }
Radim Krčmář49776fa2016-12-16 16:10:02 +01004101 /* Write kvm->irq_routing before enabling irqchip_in_kernel. */
Paolo Bonzini71ba9942015-07-29 12:31:15 +02004102 smp_wmb();
Radim Krčmář49776fa2016-12-16 16:10:02 +01004103 kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
Marcelo Tosatti3ddea122009-10-29 13:44:15 -02004104 create_irqchip_unlock:
4105 mutex_unlock(&kvm->lock);
Carsten Otte1fe779f2007-10-29 16:08:35 +01004106 break;
Marcelo Tosatti3ddea122009-10-29 13:44:15 -02004107 }
Sheng Yang78376992008-01-28 05:10:22 +08004108 case KVM_CREATE_PIT:
Jan Kiszkac5ff41c2009-05-14 22:42:53 +02004109 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
4110 goto create_pit;
4111 case KVM_CREATE_PIT2:
4112 r = -EFAULT;
4113 if (copy_from_user(&u.pit_config, argp,
4114 sizeof(struct kvm_pit_config)))
4115 goto out;
4116 create_pit:
Paolo Bonzini250715a2016-06-01 14:09:24 +02004117 mutex_lock(&kvm->lock);
Avi Kivity269e05e2009-01-05 15:21:42 +02004118 r = -EEXIST;
4119 if (kvm->arch.vpit)
4120 goto create_pit_unlock;
Sheng Yang78376992008-01-28 05:10:22 +08004121 r = -ENOMEM;
Jan Kiszkac5ff41c2009-05-14 22:42:53 +02004122 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
Sheng Yang78376992008-01-28 05:10:22 +08004123 if (kvm->arch.vpit)
4124 r = 0;
Avi Kivity269e05e2009-01-05 15:21:42 +02004125 create_pit_unlock:
Paolo Bonzini250715a2016-06-01 14:09:24 +02004126 mutex_unlock(&kvm->lock);
Sheng Yang78376992008-01-28 05:10:22 +08004127 break;
Carsten Otte1fe779f2007-10-29 16:08:35 +01004128 case KVM_GET_IRQCHIP: {
4129 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
Sasha Levinff5c2c02011-12-04 19:36:29 +02004130 struct kvm_irqchip *chip;
Carsten Otte1fe779f2007-10-29 16:08:35 +01004131
Sasha Levinff5c2c02011-12-04 19:36:29 +02004132 chip = memdup_user(argp, sizeof(*chip));
4133 if (IS_ERR(chip)) {
4134 r = PTR_ERR(chip);
Carsten Otte1fe779f2007-10-29 16:08:35 +01004135 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02004136 }
4137
Carsten Otte1fe779f2007-10-29 16:08:35 +01004138 r = -ENXIO;
Radim Krčmář826da322016-12-16 16:10:06 +01004139 if (!irqchip_kernel(kvm))
Dave Hansenf0d66272008-08-11 10:01:45 -07004140 goto get_irqchip_out;
4141 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
4142 if (r)
4143 goto get_irqchip_out;
4144 r = -EFAULT;
4145 if (copy_to_user(argp, chip, sizeof *chip))
4146 goto get_irqchip_out;
4147 r = 0;
4148 get_irqchip_out:
4149 kfree(chip);
Carsten Otte1fe779f2007-10-29 16:08:35 +01004150 break;
4151 }
4152 case KVM_SET_IRQCHIP: {
4153 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
Sasha Levinff5c2c02011-12-04 19:36:29 +02004154 struct kvm_irqchip *chip;
Carsten Otte1fe779f2007-10-29 16:08:35 +01004155
Sasha Levinff5c2c02011-12-04 19:36:29 +02004156 chip = memdup_user(argp, sizeof(*chip));
4157 if (IS_ERR(chip)) {
4158 r = PTR_ERR(chip);
Carsten Otte1fe779f2007-10-29 16:08:35 +01004159 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02004160 }
4161
Carsten Otte1fe779f2007-10-29 16:08:35 +01004162 r = -ENXIO;
Radim Krčmář826da322016-12-16 16:10:06 +01004163 if (!irqchip_kernel(kvm))
Dave Hansenf0d66272008-08-11 10:01:45 -07004164 goto set_irqchip_out;
4165 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
4166 if (r)
4167 goto set_irqchip_out;
4168 r = 0;
4169 set_irqchip_out:
4170 kfree(chip);
Carsten Otte1fe779f2007-10-29 16:08:35 +01004171 break;
4172 }
Sheng Yange0f63cb2008-03-04 00:50:59 +08004173 case KVM_GET_PIT: {
Sheng Yange0f63cb2008-03-04 00:50:59 +08004174 r = -EFAULT;
Dave Hansenf0d66272008-08-11 10:01:45 -07004175 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
Sheng Yange0f63cb2008-03-04 00:50:59 +08004176 goto out;
4177 r = -ENXIO;
4178 if (!kvm->arch.vpit)
4179 goto out;
Dave Hansenf0d66272008-08-11 10:01:45 -07004180 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
Sheng Yange0f63cb2008-03-04 00:50:59 +08004181 if (r)
4182 goto out;
4183 r = -EFAULT;
Dave Hansenf0d66272008-08-11 10:01:45 -07004184 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
Sheng Yange0f63cb2008-03-04 00:50:59 +08004185 goto out;
4186 r = 0;
4187 break;
4188 }
4189 case KVM_SET_PIT: {
Sheng Yange0f63cb2008-03-04 00:50:59 +08004190 r = -EFAULT;
Dave Hansenf0d66272008-08-11 10:01:45 -07004191 if (copy_from_user(&u.ps, argp, sizeof u.ps))
Sheng Yange0f63cb2008-03-04 00:50:59 +08004192 goto out;
4193 r = -ENXIO;
4194 if (!kvm->arch.vpit)
4195 goto out;
Dave Hansenf0d66272008-08-11 10:01:45 -07004196 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
Sheng Yange0f63cb2008-03-04 00:50:59 +08004197 break;
4198 }
Beth Kone9f42752009-07-07 11:50:38 -04004199 case KVM_GET_PIT2: {
4200 r = -ENXIO;
4201 if (!kvm->arch.vpit)
4202 goto out;
4203 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
4204 if (r)
4205 goto out;
4206 r = -EFAULT;
4207 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
4208 goto out;
4209 r = 0;
4210 break;
4211 }
4212 case KVM_SET_PIT2: {
4213 r = -EFAULT;
4214 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
4215 goto out;
4216 r = -ENXIO;
4217 if (!kvm->arch.vpit)
4218 goto out;
4219 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
Beth Kone9f42752009-07-07 11:50:38 -04004220 break;
4221 }
Marcelo Tosatti52d939a2008-12-30 15:55:06 -02004222 case KVM_REINJECT_CONTROL: {
4223 struct kvm_reinject_control control;
4224 r = -EFAULT;
4225 if (copy_from_user(&control, argp, sizeof(control)))
4226 goto out;
4227 r = kvm_vm_ioctl_reinject(kvm, &control);
Marcelo Tosatti52d939a2008-12-30 15:55:06 -02004228 break;
4229 }
Paolo Bonzinid71ba782015-07-29 11:56:48 +02004230 case KVM_SET_BOOT_CPU_ID:
4231 r = 0;
4232 mutex_lock(&kvm->lock);
Paolo Bonzini557abc42016-06-13 14:50:04 +02004233 if (kvm->created_vcpus)
Paolo Bonzinid71ba782015-07-29 11:56:48 +02004234 r = -EBUSY;
4235 else
4236 kvm->arch.bsp_vcpu_id = arg;
4237 mutex_unlock(&kvm->lock);
4238 break;
Ed Swierkffde22a2009-10-15 15:21:43 -07004239 case KVM_XEN_HVM_CONFIG: {
4240 r = -EFAULT;
4241 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
4242 sizeof(struct kvm_xen_hvm_config)))
4243 goto out;
4244 r = -EINVAL;
4245 if (kvm->arch.xen_hvm_config.flags)
4246 goto out;
4247 r = 0;
4248 break;
4249 }
Glauber Costaafbcf7a2009-10-16 15:28:36 -04004250 case KVM_SET_CLOCK: {
Glauber Costaafbcf7a2009-10-16 15:28:36 -04004251 struct kvm_clock_data user_ns;
4252 u64 now_ns;
Glauber Costaafbcf7a2009-10-16 15:28:36 -04004253
4254 r = -EFAULT;
4255 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
4256 goto out;
4257
4258 r = -EINVAL;
4259 if (user_ns.flags)
4260 goto out;
4261
4262 r = 0;
Radim Krčmář0bc48be2017-05-16 22:50:00 +02004263 /*
4264 * TODO: userspace has to take care of races with VCPU_RUN, so
4265 * kvm_gen_update_masterclock() can be cut down to locked
4266 * pvclock_update_vm_gtod_copy().
4267 */
4268 kvm_gen_update_masterclock(kvm);
Marcelo Tosattie891a322017-04-17 12:51:37 -03004269 now_ns = get_kvmclock_ns(kvm);
Paolo Bonzini108b2492016-09-01 14:21:03 +02004270 kvm->arch.kvmclock_offset += user_ns.clock - now_ns;
Radim Krčmář0bc48be2017-05-16 22:50:00 +02004271 kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
Glauber Costaafbcf7a2009-10-16 15:28:36 -04004272 break;
4273 }
4274 case KVM_GET_CLOCK: {
Glauber Costaafbcf7a2009-10-16 15:28:36 -04004275 struct kvm_clock_data user_ns;
4276 u64 now_ns;
4277
Marcelo Tosattie891a322017-04-17 12:51:37 -03004278 now_ns = get_kvmclock_ns(kvm);
Paolo Bonzini108b2492016-09-01 14:21:03 +02004279 user_ns.clock = now_ns;
Paolo Bonzinie3fd9a92016-11-09 17:48:15 +01004280 user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
Vasiliy Kulikov97e69aa2010-10-30 22:54:47 +04004281 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
Glauber Costaafbcf7a2009-10-16 15:28:36 -04004282
4283 r = -EFAULT;
4284 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
4285 goto out;
4286 r = 0;
4287 break;
4288 }
Nadav Amit90de4a12015-04-13 01:53:41 +03004289 case KVM_ENABLE_CAP: {
4290 struct kvm_enable_cap cap;
Glauber Costaafbcf7a2009-10-16 15:28:36 -04004291
Nadav Amit90de4a12015-04-13 01:53:41 +03004292 r = -EFAULT;
4293 if (copy_from_user(&cap, argp, sizeof(cap)))
4294 goto out;
4295 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
4296 break;
4297 }
Carsten Otte1fe779f2007-10-29 16:08:35 +01004298 default:
Paolo Bonziniad6260d2017-03-27 14:30:40 +02004299 r = -ENOTTY;
Carsten Otte1fe779f2007-10-29 16:08:35 +01004300 }
4301out:
4302 return r;
4303}
4304
Zhang Xiantaoa16b0432007-11-16 14:38:21 +08004305static void kvm_init_msr_list(void)
Carsten Otte043405e2007-10-10 17:16:19 +02004306{
4307 u32 dummy[2];
4308 unsigned i, j;
4309
Paolo Bonzini62ef68b2015-05-05 12:08:55 +02004310 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
Carsten Otte043405e2007-10-10 17:16:19 +02004311 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
4312 continue;
Paolo Bonzini93c4adc2014-03-05 23:19:52 +01004313
4314 /*
4315 * Even MSRs that are valid in the host may not be exposed
Paolo Bonzini9dbe6cf2015-11-12 14:49:17 +01004316 * to the guests in some cases.
Paolo Bonzini93c4adc2014-03-05 23:19:52 +01004317 */
4318 switch (msrs_to_save[i]) {
4319 case MSR_IA32_BNDCFGS:
4320 if (!kvm_x86_ops->mpx_supported())
4321 continue;
4322 break;
Paolo Bonzini9dbe6cf2015-11-12 14:49:17 +01004323 case MSR_TSC_AUX:
4324 if (!kvm_x86_ops->rdtscp_supported())
4325 continue;
4326 break;
Paolo Bonzini93c4adc2014-03-05 23:19:52 +01004327 default:
4328 break;
4329 }
4330
Carsten Otte043405e2007-10-10 17:16:19 +02004331 if (j < i)
4332 msrs_to_save[j] = msrs_to_save[i];
4333 j++;
4334 }
4335 num_msrs_to_save = j;
Paolo Bonzini62ef68b2015-05-05 12:08:55 +02004336
4337 for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
4338 switch (emulated_msrs[i]) {
Paolo Bonzini6d396b52015-04-01 14:25:33 +02004339 case MSR_IA32_SMBASE:
4340 if (!kvm_x86_ops->cpu_has_high_real_mode_segbase())
4341 continue;
4342 break;
Paolo Bonzini62ef68b2015-05-05 12:08:55 +02004343 default:
4344 break;
4345 }
4346
4347 if (j < i)
4348 emulated_msrs[j] = emulated_msrs[i];
4349 j++;
4350 }
4351 num_emulated_msrs = j;
Carsten Otte043405e2007-10-10 17:16:19 +02004352}
4353
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03004354static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
4355 const void *v)
Carsten Ottebbd9b642007-10-30 18:44:21 +01004356{
Avi Kivity70252a12010-01-19 12:51:22 +02004357 int handled = 0;
4358 int n;
Carsten Ottebbd9b642007-10-30 18:44:21 +01004359
Avi Kivity70252a12010-01-19 12:51:22 +02004360 do {
4361 n = min(len, 8);
Paolo Bonzinibce87cc2016-01-08 13:48:51 +01004362 if (!(lapic_in_kernel(vcpu) &&
Nikolay Nikolaeve32edf42015-03-26 14:39:28 +00004363 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
4364 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
Avi Kivity70252a12010-01-19 12:51:22 +02004365 break;
4366 handled += n;
4367 addr += n;
4368 len -= n;
4369 v += n;
4370 } while (len);
4371
4372 return handled;
Carsten Ottebbd9b642007-10-30 18:44:21 +01004373}
4374
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03004375static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
Carsten Ottebbd9b642007-10-30 18:44:21 +01004376{
Avi Kivity70252a12010-01-19 12:51:22 +02004377 int handled = 0;
4378 int n;
Carsten Ottebbd9b642007-10-30 18:44:21 +01004379
Avi Kivity70252a12010-01-19 12:51:22 +02004380 do {
4381 n = min(len, 8);
Paolo Bonzinibce87cc2016-01-08 13:48:51 +01004382 if (!(lapic_in_kernel(vcpu) &&
Nikolay Nikolaeve32edf42015-03-26 14:39:28 +00004383 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
4384 addr, n, v))
4385 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
Avi Kivity70252a12010-01-19 12:51:22 +02004386 break;
Wanpeng Lie39d200f2017-12-14 17:40:50 -08004387 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
Avi Kivity70252a12010-01-19 12:51:22 +02004388 handled += n;
4389 addr += n;
4390 len -= n;
4391 v += n;
4392 } while (len);
4393
4394 return handled;
Carsten Ottebbd9b642007-10-30 18:44:21 +01004395}
4396
Gleb Natapov2dafc6c2010-03-18 15:20:16 +02004397static void kvm_set_segment(struct kvm_vcpu *vcpu,
4398 struct kvm_segment *var, int seg)
4399{
4400 kvm_x86_ops->set_segment(vcpu, var, seg);
4401}
4402
4403void kvm_get_segment(struct kvm_vcpu *vcpu,
4404 struct kvm_segment *var, int seg)
4405{
4406 kvm_x86_ops->get_segment(vcpu, var, seg);
4407}
4408
Paolo Bonzini54987b72014-09-02 13:23:06 +02004409gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
4410 struct x86_exception *exception)
Joerg Roedel02f59dc2010-09-10 17:30:54 +02004411{
4412 gpa_t t_gpa;
Joerg Roedel02f59dc2010-09-10 17:30:54 +02004413
4414 BUG_ON(!mmu_is_nested(vcpu));
4415
4416 /* NPT walks are always user-walks */
4417 access |= PFERR_USER_MASK;
Paolo Bonzini54987b72014-09-02 13:23:06 +02004418 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, exception);
Joerg Roedel02f59dc2010-09-10 17:30:54 +02004419
4420 return t_gpa;
4421}
4422
Avi Kivityab9ae312010-11-22 17:53:26 +02004423gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
4424 struct x86_exception *exception)
Gleb Natapov1871c602010-02-10 14:21:32 +02004425{
4426 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
Avi Kivityab9ae312010-11-22 17:53:26 +02004427 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
Gleb Natapov1871c602010-02-10 14:21:32 +02004428}
4429
Avi Kivityab9ae312010-11-22 17:53:26 +02004430 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
4431 struct x86_exception *exception)
Gleb Natapov1871c602010-02-10 14:21:32 +02004432{
4433 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4434 access |= PFERR_FETCH_MASK;
Avi Kivityab9ae312010-11-22 17:53:26 +02004435 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
Gleb Natapov1871c602010-02-10 14:21:32 +02004436}
4437
Avi Kivityab9ae312010-11-22 17:53:26 +02004438gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
4439 struct x86_exception *exception)
Gleb Natapov1871c602010-02-10 14:21:32 +02004440{
4441 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4442 access |= PFERR_WRITE_MASK;
Avi Kivityab9ae312010-11-22 17:53:26 +02004443 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
Gleb Natapov1871c602010-02-10 14:21:32 +02004444}
4445
4446/* uses this to access any guest's mapped memory without checking CPL */
Avi Kivityab9ae312010-11-22 17:53:26 +02004447gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
4448 struct x86_exception *exception)
Gleb Natapov1871c602010-02-10 14:21:32 +02004449{
Avi Kivityab9ae312010-11-22 17:53:26 +02004450 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
Gleb Natapov1871c602010-02-10 14:21:32 +02004451}
4452
4453static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
4454 struct kvm_vcpu *vcpu, u32 access,
Avi Kivitybcc55cb2010-11-22 17:53:22 +02004455 struct x86_exception *exception)
Carsten Ottebbd9b642007-10-30 18:44:21 +01004456{
4457 void *data = val;
Marcelo Tosatti10589a42007-12-20 19:18:22 -05004458 int r = X86EMUL_CONTINUE;
Carsten Ottebbd9b642007-10-30 18:44:21 +01004459
4460 while (bytes) {
Joerg Roedel14dfe852010-09-10 17:30:49 +02004461 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
Avi Kivityab9ae312010-11-22 17:53:26 +02004462 exception);
Carsten Ottebbd9b642007-10-30 18:44:21 +01004463 unsigned offset = addr & (PAGE_SIZE-1);
Izik Eidus77c20022008-12-29 01:42:19 +02004464 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
Carsten Ottebbd9b642007-10-30 18:44:21 +01004465 int ret;
4466
Avi Kivitybcc55cb2010-11-22 17:53:22 +02004467 if (gpa == UNMAPPED_GVA)
Avi Kivityab9ae312010-11-22 17:53:26 +02004468 return X86EMUL_PROPAGATE_FAULT;
Paolo Bonzini54bf36a2015-04-08 15:39:23 +02004469 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
4470 offset, toread);
Marcelo Tosatti10589a42007-12-20 19:18:22 -05004471 if (ret < 0) {
Gleb Natapovc3cd7ff2010-04-28 19:15:35 +03004472 r = X86EMUL_IO_NEEDED;
Marcelo Tosatti10589a42007-12-20 19:18:22 -05004473 goto out;
4474 }
Carsten Ottebbd9b642007-10-30 18:44:21 +01004475
Izik Eidus77c20022008-12-29 01:42:19 +02004476 bytes -= toread;
4477 data += toread;
4478 addr += toread;
Carsten Ottebbd9b642007-10-30 18:44:21 +01004479 }
Marcelo Tosatti10589a42007-12-20 19:18:22 -05004480out:
Marcelo Tosatti10589a42007-12-20 19:18:22 -05004481 return r;
Carsten Ottebbd9b642007-10-30 18:44:21 +01004482}
Izik Eidus77c20022008-12-29 01:42:19 +02004483
Gleb Natapov1871c602010-02-10 14:21:32 +02004484/* used for instruction fetching */
Avi Kivity0f65dd72011-04-20 13:37:53 +03004485static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
4486 gva_t addr, void *val, unsigned int bytes,
Avi Kivitybcc55cb2010-11-22 17:53:22 +02004487 struct x86_exception *exception)
Gleb Natapov1871c602010-02-10 14:21:32 +02004488{
Avi Kivity0f65dd72011-04-20 13:37:53 +03004489 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
Gleb Natapov1871c602010-02-10 14:21:32 +02004490 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
Paolo Bonzini44583cb2014-05-13 14:02:13 +02004491 unsigned offset;
4492 int ret;
Avi Kivity0f65dd72011-04-20 13:37:53 +03004493
Paolo Bonzini44583cb2014-05-13 14:02:13 +02004494 /* Inline kvm_read_guest_virt_helper for speed. */
4495 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
4496 exception);
4497 if (unlikely(gpa == UNMAPPED_GVA))
4498 return X86EMUL_PROPAGATE_FAULT;
4499
4500 offset = addr & (PAGE_SIZE-1);
4501 if (WARN_ON(offset + bytes > PAGE_SIZE))
4502 bytes = (unsigned)PAGE_SIZE - offset;
Paolo Bonzini54bf36a2015-04-08 15:39:23 +02004503 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
4504 offset, bytes);
Paolo Bonzini44583cb2014-05-13 14:02:13 +02004505 if (unlikely(ret < 0))
4506 return X86EMUL_IO_NEEDED;
4507
4508 return X86EMUL_CONTINUE;
Gleb Natapov1871c602010-02-10 14:21:32 +02004509}
4510
Nadav Har'El064aea72011-05-25 23:04:56 +03004511int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
Avi Kivity0f65dd72011-04-20 13:37:53 +03004512 gva_t addr, void *val, unsigned int bytes,
Avi Kivitybcc55cb2010-11-22 17:53:22 +02004513 struct x86_exception *exception)
Gleb Natapov1871c602010-02-10 14:21:32 +02004514{
Avi Kivity0f65dd72011-04-20 13:37:53 +03004515 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
Gleb Natapov1871c602010-02-10 14:21:32 +02004516 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
Avi Kivity0f65dd72011-04-20 13:37:53 +03004517
Gleb Natapov1871c602010-02-10 14:21:32 +02004518 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
Avi Kivitybcc55cb2010-11-22 17:53:22 +02004519 exception);
Gleb Natapov1871c602010-02-10 14:21:32 +02004520}
Nadav Har'El064aea72011-05-25 23:04:56 +03004521EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
Gleb Natapov1871c602010-02-10 14:21:32 +02004522
Avi Kivity0f65dd72011-04-20 13:37:53 +03004523static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4524 gva_t addr, void *val, unsigned int bytes,
Avi Kivitybcc55cb2010-11-22 17:53:22 +02004525 struct x86_exception *exception)
Gleb Natapov1871c602010-02-10 14:21:32 +02004526{
Avi Kivity0f65dd72011-04-20 13:37:53 +03004527 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
Avi Kivitybcc55cb2010-11-22 17:53:22 +02004528 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
Gleb Natapov1871c602010-02-10 14:21:32 +02004529}
4530
Radim Krčmář7a036a62015-10-30 16:36:24 +01004531static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
4532 unsigned long addr, void *val, unsigned int bytes)
4533{
4534 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4535 int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
4536
4537 return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
4538}
4539
Nadav Har'El6a4d7552011-05-25 23:08:00 +03004540int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
Avi Kivity0f65dd72011-04-20 13:37:53 +03004541 gva_t addr, void *val,
Gleb Natapov2dafc6c2010-03-18 15:20:16 +02004542 unsigned int bytes,
Avi Kivitybcc55cb2010-11-22 17:53:22 +02004543 struct x86_exception *exception)
Izik Eidus77c20022008-12-29 01:42:19 +02004544{
Avi Kivity0f65dd72011-04-20 13:37:53 +03004545 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
Izik Eidus77c20022008-12-29 01:42:19 +02004546 void *data = val;
4547 int r = X86EMUL_CONTINUE;
4548
4549 while (bytes) {
Joerg Roedel14dfe852010-09-10 17:30:49 +02004550 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4551 PFERR_WRITE_MASK,
Avi Kivityab9ae312010-11-22 17:53:26 +02004552 exception);
Izik Eidus77c20022008-12-29 01:42:19 +02004553 unsigned offset = addr & (PAGE_SIZE-1);
4554 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4555 int ret;
4556
Avi Kivitybcc55cb2010-11-22 17:53:22 +02004557 if (gpa == UNMAPPED_GVA)
Avi Kivityab9ae312010-11-22 17:53:26 +02004558 return X86EMUL_PROPAGATE_FAULT;
Paolo Bonzini54bf36a2015-04-08 15:39:23 +02004559 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
Izik Eidus77c20022008-12-29 01:42:19 +02004560 if (ret < 0) {
Gleb Natapovc3cd7ff2010-04-28 19:15:35 +03004561 r = X86EMUL_IO_NEEDED;
Izik Eidus77c20022008-12-29 01:42:19 +02004562 goto out;
4563 }
4564
4565 bytes -= towrite;
4566 data += towrite;
4567 addr += towrite;
4568 }
4569out:
4570 return r;
4571}
Nadav Har'El6a4d7552011-05-25 23:08:00 +03004572EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
Izik Eidus77c20022008-12-29 01:42:19 +02004573
Tom Lendacky0f89b202016-12-14 14:59:23 -05004574static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4575 gpa_t gpa, bool write)
4576{
4577 /* For APIC access vmexit */
4578 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4579 return 1;
4580
4581 if (vcpu_match_mmio_gpa(vcpu, gpa)) {
4582 trace_vcpu_match_mmio(gva, gpa, write, true);
4583 return 1;
4584 }
4585
4586 return 0;
4587}
4588
Xiao Guangrongaf7cc7d2011-07-12 03:22:46 +08004589static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4590 gpa_t *gpa, struct x86_exception *exception,
4591 bool write)
4592{
Avi Kivity97d64b72012-09-12 14:52:00 +03004593 u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
4594 | (write ? PFERR_WRITE_MASK : 0);
Xiao Guangrongaf7cc7d2011-07-12 03:22:46 +08004595
Huaitong Hanbe94f6b2016-03-22 16:51:20 +08004596 /*
4597 * currently PKRU is only applied to ept enabled guest so
4598 * there is no pkey in EPT page table for L1 guest or EPT
4599 * shadow page table for L2 guest.
4600 */
Avi Kivity97d64b72012-09-12 14:52:00 +03004601 if (vcpu_match_mmio_gva(vcpu, gva)
Feng Wu97ec8c02014-04-01 17:46:34 +08004602 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
Huaitong Hanbe94f6b2016-03-22 16:51:20 +08004603 vcpu->arch.access, 0, access)) {
Xiao Guangrongbebb1062011-07-12 03:23:20 +08004604 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4605 (gva & (PAGE_SIZE - 1));
Xiao Guangrong4f022642011-07-12 03:34:24 +08004606 trace_vcpu_match_mmio(gva, *gpa, write, false);
Xiao Guangrongbebb1062011-07-12 03:23:20 +08004607 return 1;
4608 }
4609
Xiao Guangrongaf7cc7d2011-07-12 03:22:46 +08004610 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4611
4612 if (*gpa == UNMAPPED_GVA)
4613 return -1;
4614
Tom Lendacky0f89b202016-12-14 14:59:23 -05004615 return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
Xiao Guangrongaf7cc7d2011-07-12 03:22:46 +08004616}
4617
Marcelo Tosatti3200f402008-03-29 20:17:59 -03004618int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
Avi Kivitybcc55cb2010-11-22 17:53:22 +02004619 const void *val, int bytes)
Avi Kivity9f811282008-03-02 14:06:05 +02004620{
4621 int ret;
4622
Paolo Bonzini54bf36a2015-04-08 15:39:23 +02004623 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
Avi Kivity9f811282008-03-02 14:06:05 +02004624 if (ret < 0)
4625 return 0;
Xiao Guangrong0eb05bf2016-02-24 17:51:13 +08004626 kvm_page_track_write(vcpu, gpa, val, bytes);
Avi Kivity9f811282008-03-02 14:06:05 +02004627 return 1;
4628}
4629
Xiao Guangrong77d197b2011-07-13 14:31:50 +08004630struct read_write_emulator_ops {
4631 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4632 int bytes);
4633 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4634 void *val, int bytes);
4635 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4636 int bytes, void *val);
4637 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4638 void *val, int bytes);
4639 bool write;
4640};
4641
4642static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4643{
4644 if (vcpu->mmio_read_completed) {
Xiao Guangrong77d197b2011-07-13 14:31:50 +08004645 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
Wanpeng Lie39d200f2017-12-14 17:40:50 -08004646 vcpu->mmio_fragments[0].gpa, val);
Xiao Guangrong77d197b2011-07-13 14:31:50 +08004647 vcpu->mmio_read_completed = 0;
4648 return 1;
4649 }
4650
4651 return 0;
4652}
4653
4654static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4655 void *val, int bytes)
4656{
Paolo Bonzini54bf36a2015-04-08 15:39:23 +02004657 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
Xiao Guangrong77d197b2011-07-13 14:31:50 +08004658}
4659
4660static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4661 void *val, int bytes)
4662{
4663 return emulator_write_phys(vcpu, gpa, val, bytes);
4664}
4665
4666static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4667{
Wanpeng Lie39d200f2017-12-14 17:40:50 -08004668 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
Xiao Guangrong77d197b2011-07-13 14:31:50 +08004669 return vcpu_mmio_write(vcpu, gpa, bytes, val);
4670}
4671
4672static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4673 void *val, int bytes)
4674{
Wanpeng Lie39d200f2017-12-14 17:40:50 -08004675 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
Xiao Guangrong77d197b2011-07-13 14:31:50 +08004676 return X86EMUL_IO_NEEDED;
4677}
4678
4679static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4680 void *val, int bytes)
4681{
Avi Kivityf78146b2012-04-18 19:22:47 +03004682 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4683
Xiao Guangrong87da7e62012-10-24 14:07:59 +08004684 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
Xiao Guangrong77d197b2011-07-13 14:31:50 +08004685 return X86EMUL_CONTINUE;
4686}
4687
Mathias Krause0fbe9b02012-08-30 01:30:17 +02004688static const struct read_write_emulator_ops read_emultor = {
Xiao Guangrong77d197b2011-07-13 14:31:50 +08004689 .read_write_prepare = read_prepare,
4690 .read_write_emulate = read_emulate,
4691 .read_write_mmio = vcpu_mmio_read,
4692 .read_write_exit_mmio = read_exit_mmio,
4693};
4694
Mathias Krause0fbe9b02012-08-30 01:30:17 +02004695static const struct read_write_emulator_ops write_emultor = {
Xiao Guangrong77d197b2011-07-13 14:31:50 +08004696 .read_write_emulate = write_emulate,
4697 .read_write_mmio = write_mmio,
4698 .read_write_exit_mmio = write_exit_mmio,
4699 .write = true,
4700};
4701
Xiao Guangrong22388a32011-07-13 14:32:31 +08004702static int emulator_read_write_onepage(unsigned long addr, void *val,
4703 unsigned int bytes,
4704 struct x86_exception *exception,
4705 struct kvm_vcpu *vcpu,
Mathias Krause0fbe9b02012-08-30 01:30:17 +02004706 const struct read_write_emulator_ops *ops)
Carsten Ottebbd9b642007-10-30 18:44:21 +01004707{
Xiao Guangrongaf7cc7d2011-07-12 03:22:46 +08004708 gpa_t gpa;
4709 int handled, ret;
Xiao Guangrong22388a32011-07-13 14:32:31 +08004710 bool write = ops->write;
Avi Kivityf78146b2012-04-18 19:22:47 +03004711 struct kvm_mmio_fragment *frag;
Tom Lendacky0f89b202016-12-14 14:59:23 -05004712 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4713
4714 /*
4715 * If the exit was due to a NPF we may already have a GPA.
4716 * If the GPA is present, use it to avoid the GVA to GPA table walk.
4717 * Note, this cannot be used on string operations since string
4718 * operation using rep will only have the initial GPA from the NPF
4719 * occurred.
4720 */
4721 if (vcpu->arch.gpa_available &&
4722 emulator_can_use_gpa(ctxt) &&
Brijesh Singh618232e2017-08-17 18:36:57 +02004723 (addr & ~PAGE_MASK) == (vcpu->arch.gpa_val & ~PAGE_MASK)) {
4724 gpa = vcpu->arch.gpa_val;
4725 ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
4726 } else {
4727 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
4728 if (ret < 0)
4729 return X86EMUL_PROPAGATE_FAULT;
Tom Lendacky0f89b202016-12-14 14:59:23 -05004730 }
Xiao Guangrong22388a32011-07-13 14:32:31 +08004731
Brijesh Singh618232e2017-08-17 18:36:57 +02004732 if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
Carsten Ottebbd9b642007-10-30 18:44:21 +01004733 return X86EMUL_CONTINUE;
4734
Carsten Ottebbd9b642007-10-30 18:44:21 +01004735 /*
4736 * Is this MMIO handled locally?
4737 */
Xiao Guangrong22388a32011-07-13 14:32:31 +08004738 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
Avi Kivity70252a12010-01-19 12:51:22 +02004739 if (handled == bytes)
Carsten Ottebbd9b642007-10-30 18:44:21 +01004740 return X86EMUL_CONTINUE;
Carsten Ottebbd9b642007-10-30 18:44:21 +01004741
Avi Kivity70252a12010-01-19 12:51:22 +02004742 gpa += handled;
4743 bytes -= handled;
4744 val += handled;
4745
Xiao Guangrong87da7e62012-10-24 14:07:59 +08004746 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
4747 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
4748 frag->gpa = gpa;
4749 frag->data = val;
4750 frag->len = bytes;
Avi Kivityf78146b2012-04-18 19:22:47 +03004751 return X86EMUL_CONTINUE;
Carsten Ottebbd9b642007-10-30 18:44:21 +01004752}
4753
Xiubo Li52eb5a62015-03-13 17:39:45 +08004754static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
4755 unsigned long addr,
Xiao Guangrong22388a32011-07-13 14:32:31 +08004756 void *val, unsigned int bytes,
4757 struct x86_exception *exception,
Mathias Krause0fbe9b02012-08-30 01:30:17 +02004758 const struct read_write_emulator_ops *ops)
Carsten Ottebbd9b642007-10-30 18:44:21 +01004759{
Avi Kivity0f65dd72011-04-20 13:37:53 +03004760 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
Avi Kivityf78146b2012-04-18 19:22:47 +03004761 gpa_t gpa;
4762 int rc;
4763
4764 if (ops->read_write_prepare &&
4765 ops->read_write_prepare(vcpu, val, bytes))
4766 return X86EMUL_CONTINUE;
4767
4768 vcpu->mmio_nr_fragments = 0;
Avi Kivity0f65dd72011-04-20 13:37:53 +03004769
Carsten Ottebbd9b642007-10-30 18:44:21 +01004770 /* Crossing a page boundary? */
4771 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
Avi Kivityf78146b2012-04-18 19:22:47 +03004772 int now;
Carsten Ottebbd9b642007-10-30 18:44:21 +01004773
4774 now = -addr & ~PAGE_MASK;
Xiao Guangrong22388a32011-07-13 14:32:31 +08004775 rc = emulator_read_write_onepage(addr, val, now, exception,
4776 vcpu, ops);
4777
Carsten Ottebbd9b642007-10-30 18:44:21 +01004778 if (rc != X86EMUL_CONTINUE)
4779 return rc;
4780 addr += now;
Nadav Amitbac155312015-01-26 09:32:26 +02004781 if (ctxt->mode != X86EMUL_MODE_PROT64)
4782 addr = (u32)addr;
Carsten Ottebbd9b642007-10-30 18:44:21 +01004783 val += now;
4784 bytes -= now;
4785 }
Xiao Guangrong22388a32011-07-13 14:32:31 +08004786
Avi Kivityf78146b2012-04-18 19:22:47 +03004787 rc = emulator_read_write_onepage(addr, val, bytes, exception,
4788 vcpu, ops);
4789 if (rc != X86EMUL_CONTINUE)
4790 return rc;
4791
4792 if (!vcpu->mmio_nr_fragments)
4793 return rc;
4794
4795 gpa = vcpu->mmio_fragments[0].gpa;
4796
4797 vcpu->mmio_needed = 1;
4798 vcpu->mmio_cur_fragment = 0;
4799
Xiao Guangrong87da7e62012-10-24 14:07:59 +08004800 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
Avi Kivityf78146b2012-04-18 19:22:47 +03004801 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
4802 vcpu->run->exit_reason = KVM_EXIT_MMIO;
4803 vcpu->run->mmio.phys_addr = gpa;
4804
4805 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
Xiao Guangrong22388a32011-07-13 14:32:31 +08004806}
4807
4808static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4809 unsigned long addr,
4810 void *val,
4811 unsigned int bytes,
4812 struct x86_exception *exception)
4813{
4814 return emulator_read_write(ctxt, addr, val, bytes,
4815 exception, &read_emultor);
4816}
4817
Xiubo Li52eb5a62015-03-13 17:39:45 +08004818static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
Xiao Guangrong22388a32011-07-13 14:32:31 +08004819 unsigned long addr,
4820 const void *val,
4821 unsigned int bytes,
4822 struct x86_exception *exception)
4823{
4824 return emulator_read_write(ctxt, addr, (void *)val, bytes,
4825 exception, &write_emultor);
Carsten Ottebbd9b642007-10-30 18:44:21 +01004826}
Carsten Ottebbd9b642007-10-30 18:44:21 +01004827
Avi Kivitydaea3e72010-03-15 13:59:54 +02004828#define CMPXCHG_TYPE(t, ptr, old, new) \
4829 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4830
4831#ifdef CONFIG_X86_64
4832# define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4833#else
4834# define CMPXCHG64(ptr, old, new) \
Jan Kiszka9749a6c2010-03-20 10:14:13 +01004835 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
Avi Kivitydaea3e72010-03-15 13:59:54 +02004836#endif
4837
Avi Kivity0f65dd72011-04-20 13:37:53 +03004838static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4839 unsigned long addr,
Carsten Ottebbd9b642007-10-30 18:44:21 +01004840 const void *old,
4841 const void *new,
4842 unsigned int bytes,
Avi Kivity0f65dd72011-04-20 13:37:53 +03004843 struct x86_exception *exception)
Carsten Ottebbd9b642007-10-30 18:44:21 +01004844{
Avi Kivity0f65dd72011-04-20 13:37:53 +03004845 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
Avi Kivitydaea3e72010-03-15 13:59:54 +02004846 gpa_t gpa;
4847 struct page *page;
4848 char *kaddr;
4849 bool exchanged;
4850
Marcelo Tosatti2bacc552007-12-12 10:46:12 -05004851 /* guests cmpxchg8b have to be emulated atomically */
Avi Kivitydaea3e72010-03-15 13:59:54 +02004852 if (bytes > 8 || (bytes & (bytes - 1)))
4853 goto emul_write;
Marcelo Tosatti2bacc552007-12-12 10:46:12 -05004854
Avi Kivitydaea3e72010-03-15 13:59:54 +02004855 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
Marcelo Tosatti10589a42007-12-20 19:18:22 -05004856
Avi Kivitydaea3e72010-03-15 13:59:54 +02004857 if (gpa == UNMAPPED_GVA ||
4858 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4859 goto emul_write;
Marcelo Tosatti2bacc552007-12-12 10:46:12 -05004860
Avi Kivitydaea3e72010-03-15 13:59:54 +02004861 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4862 goto emul_write;
Marcelo Tosatti2bacc552007-12-12 10:46:12 -05004863
Paolo Bonzini54bf36a2015-04-08 15:39:23 +02004864 page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
Xiao Guangrong32cad842012-08-03 15:42:52 +08004865 if (is_error_page(page))
Wei Yongjunc19b8bd2010-07-15 08:51:58 +08004866 goto emul_write;
Izik Eidus72dc67a2008-02-10 18:04:15 +02004867
Cong Wang8fd75e12011-11-25 23:14:17 +08004868 kaddr = kmap_atomic(page);
Avi Kivitydaea3e72010-03-15 13:59:54 +02004869 kaddr += offset_in_page(gpa);
4870 switch (bytes) {
4871 case 1:
4872 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4873 break;
4874 case 2:
4875 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4876 break;
4877 case 4:
4878 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4879 break;
4880 case 8:
4881 exchanged = CMPXCHG64(kaddr, old, new);
4882 break;
4883 default:
4884 BUG();
Marcelo Tosatti2bacc552007-12-12 10:46:12 -05004885 }
Cong Wang8fd75e12011-11-25 23:14:17 +08004886 kunmap_atomic(kaddr);
Avi Kivitydaea3e72010-03-15 13:59:54 +02004887 kvm_release_page_dirty(page);
4888
4889 if (!exchanged)
4890 return X86EMUL_CMPXCHG_FAILED;
4891
Paolo Bonzini54bf36a2015-04-08 15:39:23 +02004892 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
Xiao Guangrong0eb05bf2016-02-24 17:51:13 +08004893 kvm_page_track_write(vcpu, gpa, new, bytes);
Gleb Natapov8f6abd02010-04-13 10:21:56 +03004894
4895 return X86EMUL_CONTINUE;
Avi Kivity4a5f48f2010-03-15 13:59:55 +02004896
Marcelo Tosatti3200f402008-03-29 20:17:59 -03004897emul_write:
Avi Kivitydaea3e72010-03-15 13:59:54 +02004898 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
Marcelo Tosatti2bacc552007-12-12 10:46:12 -05004899
Avi Kivity0f65dd72011-04-20 13:37:53 +03004900 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
Carsten Ottebbd9b642007-10-30 18:44:21 +01004901}
4902
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02004903static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4904{
Wanpeng Licbfc6c92017-05-19 02:46:56 -07004905 int r = 0, i;
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02004906
Wanpeng Licbfc6c92017-05-19 02:46:56 -07004907 for (i = 0; i < vcpu->arch.pio.count; i++) {
4908 if (vcpu->arch.pio.in)
4909 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
4910 vcpu->arch.pio.size, pd);
4911 else
4912 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
4913 vcpu->arch.pio.port, vcpu->arch.pio.size,
4914 pd);
4915 if (r)
4916 break;
4917 pd += vcpu->arch.pio.size;
4918 }
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02004919 return r;
4920}
4921
Xiao Guangrong6f6fbe92011-09-22 16:55:10 +08004922static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
4923 unsigned short port, void *val,
4924 unsigned int count, bool in)
4925{
Xiao Guangrong6f6fbe92011-09-22 16:55:10 +08004926 vcpu->arch.pio.port = port;
4927 vcpu->arch.pio.in = in;
4928 vcpu->arch.pio.count = count;
4929 vcpu->arch.pio.size = size;
4930
4931 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4932 vcpu->arch.pio.count = 0;
4933 return 1;
4934 }
4935
4936 vcpu->run->exit_reason = KVM_EXIT_IO;
4937 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
4938 vcpu->run->io.size = size;
4939 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4940 vcpu->run->io.count = count;
4941 vcpu->run->io.port = port;
4942
4943 return 0;
4944}
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02004945
Avi Kivityca1d4a92011-04-20 13:37:53 +03004946static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4947 int size, unsigned short port, void *val,
4948 unsigned int count)
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02004949{
Avi Kivityca1d4a92011-04-20 13:37:53 +03004950 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
Xiao Guangrong6f6fbe92011-09-22 16:55:10 +08004951 int ret;
Avi Kivityca1d4a92011-04-20 13:37:53 +03004952
Gleb Natapov79729952010-03-18 15:20:24 +02004953 if (vcpu->arch.pio.count)
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02004954 goto data_avail;
4955
Wanpeng Licbfc6c92017-05-19 02:46:56 -07004956 memset(vcpu->arch.pio_data, 0, size * count);
4957
Xiao Guangrong6f6fbe92011-09-22 16:55:10 +08004958 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
4959 if (ret) {
4960data_avail:
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02004961 memcpy(val, vcpu->arch.pio_data, size * count);
Ulrich Obergfell11719032014-05-02 17:57:47 +02004962 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
Gleb Natapov79729952010-03-18 15:20:24 +02004963 vcpu->arch.pio.count = 0;
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02004964 return 1;
4965 }
4966
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02004967 return 0;
4968}
4969
Avi Kivityca1d4a92011-04-20 13:37:53 +03004970static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4971 int size, unsigned short port,
4972 const void *val, unsigned int count)
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02004973{
Avi Kivityca1d4a92011-04-20 13:37:53 +03004974 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4975
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02004976 memcpy(vcpu->arch.pio_data, val, size * count);
Ulrich Obergfell11719032014-05-02 17:57:47 +02004977 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
Xiao Guangrong6f6fbe92011-09-22 16:55:10 +08004978 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02004979}
4980
Carsten Ottebbd9b642007-10-30 18:44:21 +01004981static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4982{
4983 return kvm_x86_ops->get_segment_base(vcpu, seg);
4984}
4985
Avi Kivity3cb16fe2011-04-20 15:38:44 +03004986static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
Carsten Ottebbd9b642007-10-30 18:44:21 +01004987{
Avi Kivity3cb16fe2011-04-20 15:38:44 +03004988 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
Carsten Ottebbd9b642007-10-30 18:44:21 +01004989}
4990
Jiang Biaoae6a2372016-11-07 08:54:51 +08004991static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
Sheng Yangf5f48ee2010-06-30 12:25:15 +08004992{
4993 if (!need_emulate_wbinvd(vcpu))
4994 return X86EMUL_CONTINUE;
4995
4996 if (kvm_x86_ops->has_wbinvd_exit()) {
Jan Kiszka2eec7342010-11-01 14:01:29 +01004997 int cpu = get_cpu();
4998
4999 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
Sheng Yangf5f48ee2010-06-30 12:25:15 +08005000 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
5001 wbinvd_ipi, NULL, 1);
Jan Kiszka2eec7342010-11-01 14:01:29 +01005002 put_cpu();
Sheng Yangf5f48ee2010-06-30 12:25:15 +08005003 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
Jan Kiszka2eec7342010-11-01 14:01:29 +01005004 } else
5005 wbinvd();
Sheng Yangf5f48ee2010-06-30 12:25:15 +08005006 return X86EMUL_CONTINUE;
5007}
Joel Schopp5cb56052015-03-02 13:43:31 -06005008
5009int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
5010{
Kyle Huey6affcbe2016-11-29 12:40:40 -08005011 kvm_emulate_wbinvd_noskip(vcpu);
5012 return kvm_skip_emulated_instruction(vcpu);
Joel Schopp5cb56052015-03-02 13:43:31 -06005013}
Sheng Yangf5f48ee2010-06-30 12:25:15 +08005014EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
5015
Joel Schopp5cb56052015-03-02 13:43:31 -06005016
5017
Avi Kivitybcaf5cc2011-04-20 15:53:23 +03005018static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
5019{
Joel Schopp5cb56052015-03-02 13:43:31 -06005020 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
Avi Kivitybcaf5cc2011-04-20 15:53:23 +03005021}
5022
Xiubo Li52eb5a62015-03-13 17:39:45 +08005023static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
5024 unsigned long *dest)
Carsten Ottebbd9b642007-10-30 18:44:21 +01005025{
Nadav Amit16f8a6f2014-10-03 01:10:05 +03005026 return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
Carsten Ottebbd9b642007-10-30 18:44:21 +01005027}
5028
Xiubo Li52eb5a62015-03-13 17:39:45 +08005029static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
5030 unsigned long value)
Carsten Ottebbd9b642007-10-30 18:44:21 +01005031{
Gleb Natapov338dbc92010-04-28 19:15:32 +03005032
Avi Kivity717746e2011-04-20 13:37:53 +03005033 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
Carsten Ottebbd9b642007-10-30 18:44:21 +01005034}
5035
Gleb Natapov52a46612010-03-18 15:20:03 +02005036static u64 mk_cr_64(u64 curr_cr, u32 new_val)
5037{
5038 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
5039}
5040
Avi Kivity717746e2011-04-20 13:37:53 +03005041static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
Gleb Natapov52a46612010-03-18 15:20:03 +02005042{
Avi Kivity717746e2011-04-20 13:37:53 +03005043 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
Gleb Natapov52a46612010-03-18 15:20:03 +02005044 unsigned long value;
5045
5046 switch (cr) {
5047 case 0:
5048 value = kvm_read_cr0(vcpu);
5049 break;
5050 case 2:
5051 value = vcpu->arch.cr2;
5052 break;
5053 case 3:
Avi Kivity9f8fe502010-12-05 17:30:00 +02005054 value = kvm_read_cr3(vcpu);
Gleb Natapov52a46612010-03-18 15:20:03 +02005055 break;
5056 case 4:
5057 value = kvm_read_cr4(vcpu);
5058 break;
5059 case 8:
5060 value = kvm_get_cr8(vcpu);
5061 break;
5062 default:
Christoffer Dalla737f252012-06-03 21:17:48 +03005063 kvm_err("%s: unexpected cr %u\n", __func__, cr);
Gleb Natapov52a46612010-03-18 15:20:03 +02005064 return 0;
5065 }
5066
5067 return value;
5068}
5069
Avi Kivity717746e2011-04-20 13:37:53 +03005070static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
Gleb Natapov52a46612010-03-18 15:20:03 +02005071{
Avi Kivity717746e2011-04-20 13:37:53 +03005072 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
Gleb Natapov0f122442010-04-28 19:15:31 +03005073 int res = 0;
5074
Gleb Natapov52a46612010-03-18 15:20:03 +02005075 switch (cr) {
5076 case 0:
Avi Kivity49a9b072010-06-10 17:02:14 +03005077 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
Gleb Natapov52a46612010-03-18 15:20:03 +02005078 break;
5079 case 2:
5080 vcpu->arch.cr2 = val;
5081 break;
5082 case 3:
Avi Kivity23902182010-06-10 17:02:16 +03005083 res = kvm_set_cr3(vcpu, val);
Gleb Natapov52a46612010-03-18 15:20:03 +02005084 break;
5085 case 4:
Avi Kivitya83b29c2010-06-10 17:02:15 +03005086 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
Gleb Natapov52a46612010-03-18 15:20:03 +02005087 break;
5088 case 8:
Andre Przywaraeea1cff2010-12-21 11:12:00 +01005089 res = kvm_set_cr8(vcpu, val);
Gleb Natapov52a46612010-03-18 15:20:03 +02005090 break;
5091 default:
Christoffer Dalla737f252012-06-03 21:17:48 +03005092 kvm_err("%s: unexpected cr %u\n", __func__, cr);
Gleb Natapov0f122442010-04-28 19:15:31 +03005093 res = -1;
Gleb Natapov52a46612010-03-18 15:20:03 +02005094 }
Gleb Natapov0f122442010-04-28 19:15:31 +03005095
5096 return res;
Gleb Natapov52a46612010-03-18 15:20:03 +02005097}
5098
Avi Kivity717746e2011-04-20 13:37:53 +03005099static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
Gleb Natapov9c537242010-03-18 15:20:05 +02005100{
Avi Kivity717746e2011-04-20 13:37:53 +03005101 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
Gleb Natapov9c537242010-03-18 15:20:05 +02005102}
5103
Avi Kivity4bff1e862011-04-20 13:37:53 +03005104static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
Gleb Natapov2dafc6c2010-03-18 15:20:16 +02005105{
Avi Kivity4bff1e862011-04-20 13:37:53 +03005106 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
Gleb Natapov2dafc6c2010-03-18 15:20:16 +02005107}
5108
Avi Kivity4bff1e862011-04-20 13:37:53 +03005109static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
Mohammed Gamal160ce1f2010-08-04 05:44:24 +03005110{
Avi Kivity4bff1e862011-04-20 13:37:53 +03005111 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
Mohammed Gamal160ce1f2010-08-04 05:44:24 +03005112}
5113
Avi Kivity1ac9d0c2011-04-20 15:12:00 +03005114static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5115{
5116 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
5117}
5118
5119static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5120{
5121 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
5122}
5123
Avi Kivity4bff1e862011-04-20 13:37:53 +03005124static unsigned long emulator_get_cached_segment_base(
5125 struct x86_emulate_ctxt *ctxt, int seg)
Gleb Natapov5951c442010-04-28 19:15:29 +03005126{
Avi Kivity4bff1e862011-04-20 13:37:53 +03005127 return get_segment_base(emul_to_vcpu(ctxt), seg);
Gleb Natapov5951c442010-04-28 19:15:29 +03005128}
5129
Avi Kivity1aa36612011-04-27 13:20:30 +03005130static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
5131 struct desc_struct *desc, u32 *base3,
5132 int seg)
Gleb Natapov2dafc6c2010-03-18 15:20:16 +02005133{
5134 struct kvm_segment var;
5135
Avi Kivity4bff1e862011-04-20 13:37:53 +03005136 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
Avi Kivity1aa36612011-04-27 13:20:30 +03005137 *selector = var.selector;
Gleb Natapov2dafc6c2010-03-18 15:20:16 +02005138
Gleb Natapov378a8b02013-01-21 15:36:48 +02005139 if (var.unusable) {
5140 memset(desc, 0, sizeof(*desc));
Radim Krčmářf0367ee2017-05-18 19:37:30 +02005141 if (base3)
5142 *base3 = 0;
Gleb Natapov2dafc6c2010-03-18 15:20:16 +02005143 return false;
Gleb Natapov378a8b02013-01-21 15:36:48 +02005144 }
Gleb Natapov2dafc6c2010-03-18 15:20:16 +02005145
5146 if (var.g)
5147 var.limit >>= 12;
5148 set_desc_limit(desc, var.limit);
5149 set_desc_base(desc, (unsigned long)var.base);
Gleb Natapov5601d052011-03-07 14:55:06 +02005150#ifdef CONFIG_X86_64
5151 if (base3)
5152 *base3 = var.base >> 32;
5153#endif
Gleb Natapov2dafc6c2010-03-18 15:20:16 +02005154 desc->type = var.type;
5155 desc->s = var.s;
5156 desc->dpl = var.dpl;
5157 desc->p = var.present;
5158 desc->avl = var.avl;
5159 desc->l = var.l;
5160 desc->d = var.db;
5161 desc->g = var.g;
5162
5163 return true;
5164}
5165
Avi Kivity1aa36612011-04-27 13:20:30 +03005166static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
5167 struct desc_struct *desc, u32 base3,
5168 int seg)
Gleb Natapov2dafc6c2010-03-18 15:20:16 +02005169{
Avi Kivity4bff1e862011-04-20 13:37:53 +03005170 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
Gleb Natapov2dafc6c2010-03-18 15:20:16 +02005171 struct kvm_segment var;
5172
Avi Kivity1aa36612011-04-27 13:20:30 +03005173 var.selector = selector;
Gleb Natapov2dafc6c2010-03-18 15:20:16 +02005174 var.base = get_desc_base(desc);
Gleb Natapov5601d052011-03-07 14:55:06 +02005175#ifdef CONFIG_X86_64
5176 var.base |= ((u64)base3) << 32;
5177#endif
Gleb Natapov2dafc6c2010-03-18 15:20:16 +02005178 var.limit = get_desc_limit(desc);
5179 if (desc->g)
5180 var.limit = (var.limit << 12) | 0xfff;
5181 var.type = desc->type;
Gleb Natapov2dafc6c2010-03-18 15:20:16 +02005182 var.dpl = desc->dpl;
5183 var.db = desc->d;
5184 var.s = desc->s;
5185 var.l = desc->l;
5186 var.g = desc->g;
5187 var.avl = desc->avl;
5188 var.present = desc->p;
5189 var.unusable = !var.present;
5190 var.padding = 0;
5191
5192 kvm_set_segment(vcpu, &var, seg);
5193 return;
5194}
5195
Avi Kivity717746e2011-04-20 13:37:53 +03005196static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
5197 u32 msr_index, u64 *pdata)
5198{
Paolo Bonzini609e36d2015-04-08 15:30:38 +02005199 struct msr_data msr;
5200 int r;
5201
5202 msr.index = msr_index;
5203 msr.host_initiated = false;
5204 r = kvm_get_msr(emul_to_vcpu(ctxt), &msr);
5205 if (r)
5206 return r;
5207
5208 *pdata = msr.data;
5209 return 0;
Avi Kivity717746e2011-04-20 13:37:53 +03005210}
5211
5212static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
5213 u32 msr_index, u64 data)
5214{
Will Auld8fe8ab42012-11-29 12:42:12 -08005215 struct msr_data msr;
5216
5217 msr.data = data;
5218 msr.index = msr_index;
5219 msr.host_initiated = false;
5220 return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
Avi Kivity717746e2011-04-20 13:37:53 +03005221}
5222
Paolo Bonzini64d60672015-05-07 11:36:11 +02005223static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
5224{
5225 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5226
5227 return vcpu->arch.smbase;
5228}
5229
5230static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
5231{
5232 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5233
5234 vcpu->arch.smbase = smbase;
5235}
5236
Nadav Amit67f4d422014-06-02 18:34:09 +03005237static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
5238 u32 pmc)
5239{
Wei Huangc6702c92015-06-19 13:44:45 +02005240 return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc);
Nadav Amit67f4d422014-06-02 18:34:09 +03005241}
5242
Avi Kivity222d21a2011-11-10 14:57:30 +02005243static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
5244 u32 pmc, u64 *pdata)
5245{
Wei Huangc6702c92015-06-19 13:44:45 +02005246 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
Avi Kivity222d21a2011-11-10 14:57:30 +02005247}
5248
Avi Kivity6c3287f2011-04-20 15:43:05 +03005249static void emulator_halt(struct x86_emulate_ctxt *ctxt)
5250{
5251 emul_to_vcpu(ctxt)->arch.halt_request = 1;
5252}
5253
Avi Kivity29535382011-04-20 13:37:53 +03005254static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
Joerg Roedel8a76d7f2011-04-04 12:39:27 +02005255 struct x86_instruction_info *info,
Avi Kivityc4f035c2011-04-04 12:39:22 +02005256 enum x86_intercept_stage stage)
5257{
Avi Kivity29535382011-04-20 13:37:53 +03005258 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
Avi Kivityc4f035c2011-04-04 12:39:22 +02005259}
5260
Yu Zhange911eb32017-08-24 20:27:52 +08005261static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
5262 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx, bool check_limit)
Stephan Bärwolfbdb42f52012-01-12 16:43:03 +01005263{
Yu Zhange911eb32017-08-24 20:27:52 +08005264 return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, check_limit);
Stephan Bärwolfbdb42f52012-01-12 16:43:03 +01005265}
5266
Avi Kivitydd856ef2012-08-27 23:46:17 +03005267static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
5268{
5269 return kvm_register_read(emul_to_vcpu(ctxt), reg);
5270}
5271
5272static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
5273{
5274 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
5275}
5276
Nadav Amit801806d2015-01-26 09:32:23 +02005277static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
5278{
5279 kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
5280}
5281
Ladi Prosek6ed071f2017-04-25 16:42:44 +02005282static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
5283{
5284 return emul_to_vcpu(ctxt)->arch.hflags;
5285}
5286
5287static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
5288{
5289 kvm_set_hflags(emul_to_vcpu(ctxt), emul_flags);
5290}
5291
Ladi Prosek0234bf82017-10-11 16:54:40 +02005292static int emulator_pre_leave_smm(struct x86_emulate_ctxt *ctxt, u64 smbase)
5293{
5294 return kvm_x86_ops->pre_leave_smm(emul_to_vcpu(ctxt), smbase);
5295}
5296
Mathias Krause0225fb52012-08-30 01:30:16 +02005297static const struct x86_emulate_ops emulate_ops = {
Avi Kivitydd856ef2012-08-27 23:46:17 +03005298 .read_gpr = emulator_read_gpr,
5299 .write_gpr = emulator_write_gpr,
Gleb Natapov1871c602010-02-10 14:21:32 +02005300 .read_std = kvm_read_guest_virt_system,
Gleb Natapov2dafc6c2010-03-18 15:20:16 +02005301 .write_std = kvm_write_guest_virt_system,
Radim Krčmář7a036a62015-10-30 16:36:24 +01005302 .read_phys = kvm_read_guest_phys_system,
Gleb Natapov1871c602010-02-10 14:21:32 +02005303 .fetch = kvm_fetch_guest_virt,
Carsten Ottebbd9b642007-10-30 18:44:21 +01005304 .read_emulated = emulator_read_emulated,
5305 .write_emulated = emulator_write_emulated,
5306 .cmpxchg_emulated = emulator_cmpxchg_emulated,
Avi Kivity3cb16fe2011-04-20 15:38:44 +03005307 .invlpg = emulator_invlpg,
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02005308 .pio_in_emulated = emulator_pio_in_emulated,
5309 .pio_out_emulated = emulator_pio_out_emulated,
Avi Kivity1aa36612011-04-27 13:20:30 +03005310 .get_segment = emulator_get_segment,
5311 .set_segment = emulator_set_segment,
Gleb Natapov5951c442010-04-28 19:15:29 +03005312 .get_cached_segment_base = emulator_get_cached_segment_base,
Gleb Natapov2dafc6c2010-03-18 15:20:16 +02005313 .get_gdt = emulator_get_gdt,
Mohammed Gamal160ce1f2010-08-04 05:44:24 +03005314 .get_idt = emulator_get_idt,
Avi Kivity1ac9d0c2011-04-20 15:12:00 +03005315 .set_gdt = emulator_set_gdt,
5316 .set_idt = emulator_set_idt,
Gleb Natapov52a46612010-03-18 15:20:03 +02005317 .get_cr = emulator_get_cr,
5318 .set_cr = emulator_set_cr,
Gleb Natapov9c537242010-03-18 15:20:05 +02005319 .cpl = emulator_get_cpl,
Gleb Natapov35aa5372010-04-28 19:15:27 +03005320 .get_dr = emulator_get_dr,
5321 .set_dr = emulator_set_dr,
Paolo Bonzini64d60672015-05-07 11:36:11 +02005322 .get_smbase = emulator_get_smbase,
5323 .set_smbase = emulator_set_smbase,
Avi Kivity717746e2011-04-20 13:37:53 +03005324 .set_msr = emulator_set_msr,
5325 .get_msr = emulator_get_msr,
Nadav Amit67f4d422014-06-02 18:34:09 +03005326 .check_pmc = emulator_check_pmc,
Avi Kivity222d21a2011-11-10 14:57:30 +02005327 .read_pmc = emulator_read_pmc,
Avi Kivity6c3287f2011-04-20 15:43:05 +03005328 .halt = emulator_halt,
Avi Kivitybcaf5cc2011-04-20 15:53:23 +03005329 .wbinvd = emulator_wbinvd,
Avi Kivityd6aa1002011-04-20 15:47:13 +03005330 .fix_hypercall = emulator_fix_hypercall,
Avi Kivityc4f035c2011-04-04 12:39:22 +02005331 .intercept = emulator_intercept,
Stephan Bärwolfbdb42f52012-01-12 16:43:03 +01005332 .get_cpuid = emulator_get_cpuid,
Nadav Amit801806d2015-01-26 09:32:23 +02005333 .set_nmi_mask = emulator_set_nmi_mask,
Ladi Prosek6ed071f2017-04-25 16:42:44 +02005334 .get_hflags = emulator_get_hflags,
5335 .set_hflags = emulator_set_hflags,
Ladi Prosek0234bf82017-10-11 16:54:40 +02005336 .pre_leave_smm = emulator_pre_leave_smm,
Carsten Ottebbd9b642007-10-30 18:44:21 +01005337};
5338
Gleb Natapov95cb2292010-04-28 19:15:43 +03005339static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
5340{
Paolo Bonzini37ccdcb2014-05-20 14:29:47 +02005341 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
Gleb Natapov95cb2292010-04-28 19:15:43 +03005342 /*
5343 * an sti; sti; sequence only disable interrupts for the first
5344 * instruction. So, if the last instruction, be it emulated or
5345 * not, left the system with the INT_STI flag enabled, it
5346 * means that the last instruction is an sti. We should not
5347 * leave the flag on in this case. The same goes for mov ss
5348 */
Paolo Bonzini37ccdcb2014-05-20 14:29:47 +02005349 if (int_shadow & mask)
5350 mask = 0;
Paolo Bonzini6addfc42014-03-27 11:29:28 +01005351 if (unlikely(int_shadow || mask)) {
Gleb Natapov95cb2292010-04-28 19:15:43 +03005352 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
Paolo Bonzini6addfc42014-03-27 11:29:28 +01005353 if (!mask)
5354 kvm_make_request(KVM_REQ_EVENT, vcpu);
5355 }
Gleb Natapov95cb2292010-04-28 19:15:43 +03005356}
5357
Paolo Bonzinief54bcf2014-09-04 19:46:15 +02005358static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
Gleb Natapov54b84862010-04-28 19:15:44 +03005359{
5360 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
Avi Kivityda9cb572010-11-22 17:53:21 +02005361 if (ctxt->exception.vector == PF_VECTOR)
Paolo Bonzinief54bcf2014-09-04 19:46:15 +02005362 return kvm_propagate_fault(vcpu, &ctxt->exception);
5363
5364 if (ctxt->exception.error_code_valid)
Avi Kivityda9cb572010-11-22 17:53:21 +02005365 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
5366 ctxt->exception.error_code);
Gleb Natapov54b84862010-04-28 19:15:44 +03005367 else
Avi Kivityda9cb572010-11-22 17:53:21 +02005368 kvm_queue_exception(vcpu, ctxt->exception.vector);
Paolo Bonzinief54bcf2014-09-04 19:46:15 +02005369 return false;
Gleb Natapov54b84862010-04-28 19:15:44 +03005370}
5371
Mohammed Gamal8ec47222010-08-16 00:47:01 +03005372static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
5373{
Takuya Yoshikawaadf52232011-05-25 11:06:16 +09005374 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
Mohammed Gamal8ec47222010-08-16 00:47:01 +03005375 int cs_db, cs_l;
5376
Mohammed Gamal8ec47222010-08-16 00:47:01 +03005377 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5378
Takuya Yoshikawaadf52232011-05-25 11:06:16 +09005379 ctxt->eflags = kvm_get_rflags(vcpu);
Paolo Bonzinic8401dd2017-06-07 15:13:14 +02005380 ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
5381
Takuya Yoshikawaadf52232011-05-25 11:06:16 +09005382 ctxt->eip = kvm_rip_read(vcpu);
5383 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
5384 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
Nadav Amit42bf5492014-04-18 07:11:34 +03005385 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
Takuya Yoshikawaadf52232011-05-25 11:06:16 +09005386 cs_db ? X86EMUL_MODE_PROT32 :
5387 X86EMUL_MODE_PROT16;
Paolo Bonzinia5845392015-04-01 18:18:53 +02005388 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
Paolo Bonzini64d60672015-05-07 11:36:11 +02005389 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
5390 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
Takuya Yoshikawaadf52232011-05-25 11:06:16 +09005391
Avi Kivitydd856ef2012-08-27 23:46:17 +03005392 init_decode_cache(ctxt);
Gleb Natapov7ae441e2011-03-31 12:06:41 +02005393 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
Mohammed Gamal8ec47222010-08-16 00:47:01 +03005394}
5395
Serge E. Hallyn71f98332011-04-13 09:12:54 -05005396int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
Mohammed Gamal63995652010-09-19 14:34:06 +02005397{
Takuya Yoshikawa9d741912011-05-29 21:53:48 +09005398 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
Mohammed Gamal63995652010-09-19 14:34:06 +02005399 int ret;
5400
5401 init_emulate_ctxt(vcpu);
5402
Avi Kivity9dac77f2011-06-01 15:34:25 +03005403 ctxt->op_bytes = 2;
5404 ctxt->ad_bytes = 2;
5405 ctxt->_eip = ctxt->eip + inc_eip;
Takuya Yoshikawa9d741912011-05-29 21:53:48 +09005406 ret = emulate_int_real(ctxt, irq);
Mohammed Gamal63995652010-09-19 14:34:06 +02005407
5408 if (ret != X86EMUL_CONTINUE)
5409 return EMULATE_FAIL;
5410
Avi Kivity9dac77f2011-06-01 15:34:25 +03005411 ctxt->eip = ctxt->_eip;
Takuya Yoshikawa9d741912011-05-29 21:53:48 +09005412 kvm_rip_write(vcpu, ctxt->eip);
5413 kvm_set_rflags(vcpu, ctxt->eflags);
Mohammed Gamal63995652010-09-19 14:34:06 +02005414
5415 if (irq == NMI_VECTOR)
Avi Kivity7460fb4a2011-09-20 13:43:14 +03005416 vcpu->arch.nmi_pending = 0;
Mohammed Gamal63995652010-09-19 14:34:06 +02005417 else
5418 vcpu->arch.interrupt.pending = false;
5419
5420 return EMULATE_DONE;
5421}
5422EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
5423
Gleb Natapov6d77dbf2010-05-10 11:16:56 +03005424static int handle_emulation_failure(struct kvm_vcpu *vcpu)
5425{
Joerg Roedelfc3a9152010-11-29 17:51:49 +01005426 int r = EMULATE_DONE;
5427
Gleb Natapov6d77dbf2010-05-10 11:16:56 +03005428 ++vcpu->stat.insn_emulation_fail;
5429 trace_kvm_emulate_insn_failed(vcpu);
Nadav Amita2b9e6c2014-09-17 02:50:50 +03005430 if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
Joerg Roedelfc3a9152010-11-29 17:51:49 +01005431 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5432 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5433 vcpu->run->internal.ndata = 0;
Liran Alon1f4dcb32017-11-05 16:56:33 +02005434 r = EMULATE_USER_EXIT;
Joerg Roedelfc3a9152010-11-29 17:51:49 +01005435 }
Gleb Natapov6d77dbf2010-05-10 11:16:56 +03005436 kvm_queue_exception(vcpu, UD_VECTOR);
Joerg Roedelfc3a9152010-11-29 17:51:49 +01005437
5438 return r;
Gleb Natapov6d77dbf2010-05-10 11:16:56 +03005439}
5440
Xiao Guangrong93c05d32013-01-13 23:49:07 +08005441static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
Gleb Natapov991eebf2013-04-11 12:10:51 +03005442 bool write_fault_to_shadow_pgtable,
5443 int emulation_type)
Gleb Natapova6f177e2010-07-08 12:41:12 +03005444{
Xiao Guangrong95b3cf62013-01-13 23:46:52 +08005445 gpa_t gpa = cr2;
Dan Williamsba049e92016-01-15 16:56:11 -08005446 kvm_pfn_t pfn;
Gleb Natapova6f177e2010-07-08 12:41:12 +03005447
Gleb Natapov991eebf2013-04-11 12:10:51 +03005448 if (emulation_type & EMULTYPE_NO_REEXECUTE)
5449 return false;
5450
Xiao Guangrong95b3cf62013-01-13 23:46:52 +08005451 if (!vcpu->arch.mmu.direct_map) {
5452 /*
5453 * Write permission should be allowed since only
5454 * write access need to be emulated.
5455 */
5456 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
Gleb Natapov68be0802010-07-14 19:05:45 +03005457
Xiao Guangrong95b3cf62013-01-13 23:46:52 +08005458 /*
5459 * If the mapping is invalid in guest, let cpu retry
5460 * it to generate fault.
5461 */
5462 if (gpa == UNMAPPED_GVA)
5463 return true;
5464 }
Gleb Natapova6f177e2010-07-08 12:41:12 +03005465
Xiao Guangrong8e3d9d02012-08-21 10:57:42 +08005466 /*
5467 * Do not retry the unhandleable instruction if it faults on the
5468 * readonly host memory, otherwise it will goto a infinite loop:
5469 * retry instruction -> write #PF -> emulation fail -> retry
5470 * instruction -> ...
5471 */
5472 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
Xiao Guangrong95b3cf62013-01-13 23:46:52 +08005473
5474 /*
5475 * If the instruction failed on the error pfn, it can not be fixed,
5476 * report the error to userspace.
5477 */
5478 if (is_error_noslot_pfn(pfn))
5479 return false;
5480
5481 kvm_release_pfn_clean(pfn);
5482
5483 /* The instructions are well-emulated on direct mmu. */
5484 if (vcpu->arch.mmu.direct_map) {
5485 unsigned int indirect_shadow_pages;
5486
5487 spin_lock(&vcpu->kvm->mmu_lock);
5488 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
5489 spin_unlock(&vcpu->kvm->mmu_lock);
5490
5491 if (indirect_shadow_pages)
5492 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5493
Gleb Natapova6f177e2010-07-08 12:41:12 +03005494 return true;
Xiao Guangrong8e3d9d02012-08-21 10:57:42 +08005495 }
Gleb Natapova6f177e2010-07-08 12:41:12 +03005496
Xiao Guangrong95b3cf62013-01-13 23:46:52 +08005497 /*
5498 * if emulation was due to access to shadowed page table
5499 * and it failed try to unshadow page and re-enter the
5500 * guest to let CPU execute the instruction.
5501 */
5502 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
Xiao Guangrong93c05d32013-01-13 23:49:07 +08005503
5504 /*
5505 * If the access faults on its page table, it can not
5506 * be fixed by unprotecting shadow page and it should
5507 * be reported to userspace.
5508 */
5509 return !write_fault_to_shadow_pgtable;
Gleb Natapova6f177e2010-07-08 12:41:12 +03005510}
5511
Xiao Guangrong1cb3f3a2011-09-22 17:02:48 +08005512static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
5513 unsigned long cr2, int emulation_type)
5514{
5515 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5516 unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
5517
5518 last_retry_eip = vcpu->arch.last_retry_eip;
5519 last_retry_addr = vcpu->arch.last_retry_addr;
5520
5521 /*
5522 * If the emulation is caused by #PF and it is non-page_table
5523 * writing instruction, it means the VM-EXIT is caused by shadow
5524 * page protected, we can zap the shadow page and retry this
5525 * instruction directly.
5526 *
5527 * Note: if the guest uses a non-page-table modifying instruction
5528 * on the PDE that points to the instruction, then we will unmap
5529 * the instruction and go to an infinite loop. So, we cache the
5530 * last retried eip and the last fault address, if we meet the eip
5531 * and the address again, we can break out of the potential infinite
5532 * loop.
5533 */
5534 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
5535
5536 if (!(emulation_type & EMULTYPE_RETRY))
5537 return false;
5538
5539 if (x86_page_table_writing_insn(ctxt))
5540 return false;
5541
5542 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
5543 return false;
5544
5545 vcpu->arch.last_retry_eip = ctxt->eip;
5546 vcpu->arch.last_retry_addr = cr2;
5547
5548 if (!vcpu->arch.mmu.direct_map)
5549 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5550
Xiao Guangrong22368022013-01-13 23:44:12 +08005551 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
Xiao Guangrong1cb3f3a2011-09-22 17:02:48 +08005552
5553 return true;
5554}
5555
Gleb Natapov716d51a2012-09-03 15:24:26 +03005556static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
5557static int complete_emulated_pio(struct kvm_vcpu *vcpu);
5558
Paolo Bonzini64d60672015-05-07 11:36:11 +02005559static void kvm_smm_changed(struct kvm_vcpu *vcpu)
Paolo Bonzinia5845392015-04-01 18:18:53 +02005560{
Paolo Bonzini64d60672015-05-07 11:36:11 +02005561 if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
Paolo Bonzini660a5d52015-05-05 11:50:23 +02005562 /* This is a good place to trace that we are exiting SMM. */
5563 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
5564
Paolo Bonzinic43203c2016-06-01 22:26:00 +02005565 /* Process a latched INIT or SMI, if any. */
5566 kvm_make_request(KVM_REQ_EVENT, vcpu);
Paolo Bonzini64d60672015-05-07 11:36:11 +02005567 }
Paolo Bonzini699023e2015-05-18 15:03:39 +02005568
5569 kvm_mmu_reset_context(vcpu);
Paolo Bonzini64d60672015-05-07 11:36:11 +02005570}
5571
5572static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags)
5573{
5574 unsigned changed = vcpu->arch.hflags ^ emul_flags;
5575
Paolo Bonzinia5845392015-04-01 18:18:53 +02005576 vcpu->arch.hflags = emul_flags;
Paolo Bonzini64d60672015-05-07 11:36:11 +02005577
5578 if (changed & HF_SMM_MASK)
5579 kvm_smm_changed(vcpu);
Paolo Bonzinia5845392015-04-01 18:18:53 +02005580}
5581
Paolo Bonzini4a1e10d2013-05-30 11:48:30 +02005582static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
5583 unsigned long *db)
5584{
5585 u32 dr6 = 0;
5586 int i;
5587 u32 enable, rwlen;
5588
5589 enable = dr7;
5590 rwlen = dr7 >> 16;
5591 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
5592 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
5593 dr6 |= (1 << i);
5594 return dr6;
5595}
5596
Paolo Bonzinic8401dd2017-06-07 15:13:14 +02005597static void kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu, int *r)
Paolo Bonzini663f4c62013-06-25 18:32:07 +02005598{
5599 struct kvm_run *kvm_run = vcpu->run;
5600
Paolo Bonzinic8401dd2017-06-07 15:13:14 +02005601 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
5602 kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 | DR6_RTM;
5603 kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
5604 kvm_run->debug.arch.exception = DB_VECTOR;
5605 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5606 *r = EMULATE_USER_EXIT;
5607 } else {
5608 /*
5609 * "Certain debug exceptions may clear bit 0-3. The
5610 * remaining contents of the DR6 register are never
5611 * cleared by the processor".
5612 */
5613 vcpu->arch.dr6 &= ~15;
5614 vcpu->arch.dr6 |= DR6_BS | DR6_RTM;
5615 kvm_queue_exception(vcpu, DB_VECTOR);
Paolo Bonzini663f4c62013-06-25 18:32:07 +02005616 }
5617}
5618
Kyle Huey6affcbe2016-11-29 12:40:40 -08005619int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
5620{
5621 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5622 int r = EMULATE_DONE;
5623
5624 kvm_x86_ops->skip_emulated_instruction(vcpu);
Paolo Bonzinic8401dd2017-06-07 15:13:14 +02005625
5626 /*
5627 * rflags is the old, "raw" value of the flags. The new value has
5628 * not been saved yet.
5629 *
5630 * This is correct even for TF set by the guest, because "the
5631 * processor will not generate this exception after the instruction
5632 * that sets the TF flag".
5633 */
5634 if (unlikely(rflags & X86_EFLAGS_TF))
5635 kvm_vcpu_do_singlestep(vcpu, &r);
Kyle Huey6affcbe2016-11-29 12:40:40 -08005636 return r == EMULATE_DONE;
5637}
5638EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
5639
Paolo Bonzini4a1e10d2013-05-30 11:48:30 +02005640static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
5641{
Paolo Bonzini4a1e10d2013-05-30 11:48:30 +02005642 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
5643 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
Nadav Amit82b32772014-11-02 11:54:45 +02005644 struct kvm_run *kvm_run = vcpu->run;
5645 unsigned long eip = kvm_get_linear_rip(vcpu);
5646 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
Paolo Bonzini4a1e10d2013-05-30 11:48:30 +02005647 vcpu->arch.guest_debug_dr7,
5648 vcpu->arch.eff_db);
5649
5650 if (dr6 != 0) {
Nadav Amit6f43ed02014-07-15 17:37:46 +03005651 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
Nadav Amit82b32772014-11-02 11:54:45 +02005652 kvm_run->debug.arch.pc = eip;
Paolo Bonzini4a1e10d2013-05-30 11:48:30 +02005653 kvm_run->debug.arch.exception = DB_VECTOR;
5654 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5655 *r = EMULATE_USER_EXIT;
5656 return true;
5657 }
5658 }
5659
Nadav Amit4161a562014-07-17 01:19:31 +03005660 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
5661 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
Nadav Amit82b32772014-11-02 11:54:45 +02005662 unsigned long eip = kvm_get_linear_rip(vcpu);
5663 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
Paolo Bonzini4a1e10d2013-05-30 11:48:30 +02005664 vcpu->arch.dr7,
5665 vcpu->arch.db);
5666
5667 if (dr6 != 0) {
5668 vcpu->arch.dr6 &= ~15;
Nadav Amit6f43ed02014-07-15 17:37:46 +03005669 vcpu->arch.dr6 |= dr6 | DR6_RTM;
Paolo Bonzini4a1e10d2013-05-30 11:48:30 +02005670 kvm_queue_exception(vcpu, DB_VECTOR);
5671 *r = EMULATE_DONE;
5672 return true;
5673 }
5674 }
5675
5676 return false;
5677}
5678
Andre Przywara51d8b662010-12-21 11:12:02 +01005679int x86_emulate_instruction(struct kvm_vcpu *vcpu,
5680 unsigned long cr2,
Andre Przywaradc25e892010-12-21 11:12:07 +01005681 int emulation_type,
5682 void *insn,
5683 int insn_len)
Carsten Ottebbd9b642007-10-30 18:44:21 +01005684{
Gleb Natapov95cb2292010-04-28 19:15:43 +03005685 int r;
Takuya Yoshikawa9d741912011-05-29 21:53:48 +09005686 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
Gleb Natapov7ae441e2011-03-31 12:06:41 +02005687 bool writeback = true;
Xiao Guangrong93c05d32013-01-13 23:49:07 +08005688 bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
Carsten Ottebbd9b642007-10-30 18:44:21 +01005689
Xiao Guangrong93c05d32013-01-13 23:49:07 +08005690 /*
5691 * Clear write_fault_to_shadow_pgtable here to ensure it is
5692 * never reused.
5693 */
5694 vcpu->arch.write_fault_to_shadow_pgtable = false;
Avi Kivity26eef702008-07-03 14:59:22 +03005695 kvm_clear_exception_queue(vcpu);
Gleb Natapov8d7d81022011-04-12 12:36:21 +03005696
Sheng Yang571008d2008-01-02 14:49:22 +08005697 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
Mohammed Gamal8ec47222010-08-16 00:47:01 +03005698 init_emulate_ctxt(vcpu);
Paolo Bonzini4a1e10d2013-05-30 11:48:30 +02005699
5700 /*
5701 * We will reenter on the same instruction since
5702 * we do not set complete_userspace_io. This does not
5703 * handle watchpoints yet, those would be handled in
5704 * the emulate_ops.
5705 */
5706 if (kvm_vcpu_check_breakpoint(vcpu, &r))
5707 return r;
5708
Takuya Yoshikawa9d741912011-05-29 21:53:48 +09005709 ctxt->interruptibility = 0;
5710 ctxt->have_exception = false;
Paolo Bonzinie0ad0b42014-08-20 10:08:23 +02005711 ctxt->exception.vector = -1;
Takuya Yoshikawa9d741912011-05-29 21:53:48 +09005712 ctxt->perm_ok = false;
Carsten Ottebbd9b642007-10-30 18:44:21 +01005713
Borislav Petkovb51e9742013-09-22 16:44:52 +02005714 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
Avi Kivity40059962011-02-01 16:32:04 +02005715
Takuya Yoshikawa9d741912011-05-29 21:53:48 +09005716 r = x86_decode_insn(ctxt, insn, insn_len);
Joerg Roedeld47f00a2010-09-10 17:30:56 +02005717
Avi Kivitye46479f2010-04-11 13:05:16 +03005718 trace_kvm_emulate_insn_start(vcpu);
Avi Kivityf2b57562007-11-18 15:17:51 +02005719 ++vcpu->stat.insn_emulation;
Takuya Yoshikawa1d2887e2011-07-30 18:03:34 +09005720 if (r != EMULATION_OK) {
Avi Kivity40059962011-02-01 16:32:04 +02005721 if (emulation_type & EMULTYPE_TRAP_UD)
5722 return EMULATE_FAIL;
Gleb Natapov991eebf2013-04-11 12:10:51 +03005723 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5724 emulation_type))
Carsten Ottebbd9b642007-10-30 18:44:21 +01005725 return EMULATE_DONE;
Paolo Bonzini6ea6e842017-11-10 10:49:38 +01005726 if (ctxt->have_exception && inject_emulated_exception(vcpu))
5727 return EMULATE_DONE;
Gleb Natapov6d77dbf2010-05-10 11:16:56 +03005728 if (emulation_type & EMULTYPE_SKIP)
5729 return EMULATE_FAIL;
5730 return handle_emulation_failure(vcpu);
Carsten Ottebbd9b642007-10-30 18:44:21 +01005731 }
5732 }
5733
Gleb Natapovba8afb62009-04-12 13:36:57 +03005734 if (emulation_type & EMULTYPE_SKIP) {
Avi Kivity9dac77f2011-06-01 15:34:25 +03005735 kvm_rip_write(vcpu, ctxt->_eip);
Nadav Amitbb663c72014-07-21 14:37:26 +03005736 if (ctxt->eflags & X86_EFLAGS_RF)
5737 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
Gleb Natapovba8afb62009-04-12 13:36:57 +03005738 return EMULATE_DONE;
5739 }
5740
Xiao Guangrong1cb3f3a2011-09-22 17:02:48 +08005741 if (retry_instruction(ctxt, cr2, emulation_type))
5742 return EMULATE_DONE;
5743
Gleb Natapov7ae441e2011-03-31 12:06:41 +02005744 /* this is needed for vmware backdoor interface to work since it
Gleb Natapov4d2179e2010-04-28 19:15:42 +03005745 changes registers values during IO operation */
Gleb Natapov7ae441e2011-03-31 12:06:41 +02005746 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
5747 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
Avi Kivitydd856ef2012-08-27 23:46:17 +03005748 emulator_invalidate_register_cache(ctxt);
Gleb Natapov7ae441e2011-03-31 12:06:41 +02005749 }
Gleb Natapov4d2179e2010-04-28 19:15:42 +03005750
Gleb Natapov5cd21912010-03-18 15:20:26 +02005751restart:
Tom Lendacky0f89b202016-12-14 14:59:23 -05005752 /* Save the faulting GPA (cr2) in the address field */
5753 ctxt->exception.address = cr2;
5754
Takuya Yoshikawa9d741912011-05-29 21:53:48 +09005755 r = x86_emulate_insn(ctxt);
Gleb Natapovc3cd7ff2010-04-28 19:15:35 +03005756
Joerg Roedel775fde82011-04-04 12:39:24 +02005757 if (r == EMULATION_INTERCEPTED)
5758 return EMULATE_DONE;
5759
Gleb Natapovd2ddd1c2010-08-25 12:47:43 +03005760 if (r == EMULATION_FAILED) {
Gleb Natapov991eebf2013-04-11 12:10:51 +03005761 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5762 emulation_type))
Gleb Natapovc3cd7ff2010-04-28 19:15:35 +03005763 return EMULATE_DONE;
5764
Gleb Natapov6d77dbf2010-05-10 11:16:56 +03005765 return handle_emulation_failure(vcpu);
Carsten Ottebbd9b642007-10-30 18:44:21 +01005766 }
5767
Takuya Yoshikawa9d741912011-05-29 21:53:48 +09005768 if (ctxt->have_exception) {
Gleb Natapovd2ddd1c2010-08-25 12:47:43 +03005769 r = EMULATE_DONE;
Paolo Bonzinief54bcf2014-09-04 19:46:15 +02005770 if (inject_emulated_exception(vcpu))
5771 return r;
Gleb Natapovd2ddd1c2010-08-25 12:47:43 +03005772 } else if (vcpu->arch.pio.count) {
Paolo Bonzini0912c972013-08-27 15:41:43 +02005773 if (!vcpu->arch.pio.in) {
5774 /* FIXME: return into emulator if single-stepping. */
Gleb Natapove85d28f2010-07-29 15:11:52 +03005775 vcpu->arch.pio.count = 0;
Paolo Bonzini0912c972013-08-27 15:41:43 +02005776 } else {
Gleb Natapov7ae441e2011-03-31 12:06:41 +02005777 writeback = false;
Gleb Natapov716d51a2012-09-03 15:24:26 +03005778 vcpu->arch.complete_userspace_io = complete_emulated_pio;
5779 }
Paolo Bonziniac0a48c2013-06-25 18:24:41 +02005780 r = EMULATE_USER_EXIT;
Gleb Natapov7ae441e2011-03-31 12:06:41 +02005781 } else if (vcpu->mmio_needed) {
5782 if (!vcpu->mmio_is_write)
5783 writeback = false;
Paolo Bonziniac0a48c2013-06-25 18:24:41 +02005784 r = EMULATE_USER_EXIT;
Gleb Natapov716d51a2012-09-03 15:24:26 +03005785 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
Gleb Natapov7ae441e2011-03-31 12:06:41 +02005786 } else if (r == EMULATION_RESTART)
Gleb Natapove85d28f2010-07-29 15:11:52 +03005787 goto restart;
Gleb Natapovd2ddd1c2010-08-25 12:47:43 +03005788 else
5789 r = EMULATE_DONE;
Gleb Natapove85d28f2010-07-29 15:11:52 +03005790
Gleb Natapov7ae441e2011-03-31 12:06:41 +02005791 if (writeback) {
Paolo Bonzini6addfc42014-03-27 11:29:28 +01005792 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
Takuya Yoshikawa9d741912011-05-29 21:53:48 +09005793 toggle_interruptibility(vcpu, ctxt->interruptibility);
Gleb Natapov7ae441e2011-03-31 12:06:41 +02005794 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
Takuya Yoshikawa9d741912011-05-29 21:53:48 +09005795 kvm_rip_write(vcpu, ctxt->eip);
Paolo Bonzinic8401dd2017-06-07 15:13:14 +02005796 if (r == EMULATE_DONE &&
5797 (ctxt->tf || (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)))
5798 kvm_vcpu_do_singlestep(vcpu, &r);
Nadav Amit38827db2014-11-02 11:54:53 +02005799 if (!ctxt->have_exception ||
5800 exception_type(ctxt->exception.vector) == EXCPT_TRAP)
5801 __kvm_set_rflags(vcpu, ctxt->eflags);
Paolo Bonzini6addfc42014-03-27 11:29:28 +01005802
5803 /*
5804 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
5805 * do nothing, and it will be requested again as soon as
5806 * the shadow expires. But we still need to check here,
5807 * because POPF has no interrupt shadow.
5808 */
5809 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
5810 kvm_make_request(KVM_REQ_EVENT, vcpu);
Gleb Natapov7ae441e2011-03-31 12:06:41 +02005811 } else
5812 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
Gleb Natapov3457e412010-04-28 19:15:38 +03005813
Gleb Natapove85d28f2010-07-29 15:11:52 +03005814 return r;
Carsten Ottebbd9b642007-10-30 18:44:21 +01005815}
Andre Przywara51d8b662010-12-21 11:12:02 +01005816EXPORT_SYMBOL_GPL(x86_emulate_instruction);
Carsten Ottebbd9b642007-10-30 18:44:21 +01005817
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02005818int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
Carsten Ottede7d7892007-10-30 18:44:25 +01005819{
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02005820 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
Avi Kivityca1d4a92011-04-20 13:37:53 +03005821 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
5822 size, port, &val, 1);
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02005823 /* do not return to emulator after return from userspace */
Gleb Natapov79729952010-03-18 15:20:24 +02005824 vcpu->arch.pio.count = 0;
Izik Eidus0f346072008-12-29 01:42:20 +02005825 return ret;
Carsten Ottede7d7892007-10-30 18:44:25 +01005826}
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02005827EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
Carsten Ottede7d7892007-10-30 18:44:25 +01005828
Tom Lendacky8370c3d2016-11-23 12:01:50 -05005829static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
5830{
5831 unsigned long val;
5832
5833 /* We should only ever be called with arch.pio.count equal to 1 */
5834 BUG_ON(vcpu->arch.pio.count != 1);
5835
5836 /* For size less than 4 we merge, else we zero extend */
5837 val = (vcpu->arch.pio.size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX)
5838 : 0;
5839
5840 /*
5841 * Since vcpu->arch.pio.count == 1 let emulator_pio_in_emulated perform
5842 * the copy and tracing
5843 */
5844 emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, vcpu->arch.pio.size,
5845 vcpu->arch.pio.port, &val, 1);
5846 kvm_register_write(vcpu, VCPU_REGS_RAX, val);
5847
5848 return 1;
5849}
5850
5851int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size, unsigned short port)
5852{
5853 unsigned long val;
5854 int ret;
5855
5856 /* For size less than 4 we merge, else we zero extend */
5857 val = (size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX) : 0;
5858
5859 ret = emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, size, port,
5860 &val, 1);
5861 if (ret) {
5862 kvm_register_write(vcpu, VCPU_REGS_RAX, val);
5863 return ret;
5864 }
5865
5866 vcpu->arch.complete_userspace_io = complete_fast_pio_in;
5867
5868 return 0;
5869}
5870EXPORT_SYMBOL_GPL(kvm_fast_pio_in);
5871
Sebastian Andrzej Siewior251a5fd2016-07-13 17:16:33 +00005872static int kvmclock_cpu_down_prep(unsigned int cpu)
Gerd Hoffmannc8076602009-02-04 17:52:04 +01005873{
Tejun Heo0a3aee02010-12-18 16:28:55 +01005874 __this_cpu_write(cpu_tsc_khz, 0);
Sebastian Andrzej Siewior251a5fd2016-07-13 17:16:33 +00005875 return 0;
Zachary Amsden8cfdc002010-08-19 22:07:21 -10005876}
5877
5878static void tsc_khz_changed(void *data)
5879{
5880 struct cpufreq_freqs *freq = data;
5881 unsigned long khz = 0;
5882
5883 if (data)
5884 khz = freq->new;
5885 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5886 khz = cpufreq_quick_get(raw_smp_processor_id());
5887 if (!khz)
5888 khz = tsc_khz;
Tejun Heo0a3aee02010-12-18 16:28:55 +01005889 __this_cpu_write(cpu_tsc_khz, khz);
Gerd Hoffmannc8076602009-02-04 17:52:04 +01005890}
5891
Gerd Hoffmannc8076602009-02-04 17:52:04 +01005892static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
5893 void *data)
5894{
5895 struct cpufreq_freqs *freq = data;
5896 struct kvm *kvm;
5897 struct kvm_vcpu *vcpu;
5898 int i, send_ipi = 0;
5899
Zachary Amsden8cfdc002010-08-19 22:07:21 -10005900 /*
5901 * We allow guests to temporarily run on slowing clocks,
5902 * provided we notify them after, or to run on accelerating
5903 * clocks, provided we notify them before. Thus time never
5904 * goes backwards.
5905 *
5906 * However, we have a problem. We can't atomically update
5907 * the frequency of a given CPU from this function; it is
5908 * merely a notifier, which can be called from any CPU.
5909 * Changing the TSC frequency at arbitrary points in time
5910 * requires a recomputation of local variables related to
5911 * the TSC for each VCPU. We must flag these local variables
5912 * to be updated and be sure the update takes place with the
5913 * new frequency before any guests proceed.
5914 *
5915 * Unfortunately, the combination of hotplug CPU and frequency
5916 * change creates an intractable locking scenario; the order
5917 * of when these callouts happen is undefined with respect to
5918 * CPU hotplug, and they can race with each other. As such,
5919 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
5920 * undefined; you can actually have a CPU frequency change take
5921 * place in between the computation of X and the setting of the
5922 * variable. To protect against this problem, all updates of
5923 * the per_cpu tsc_khz variable are done in an interrupt
5924 * protected IPI, and all callers wishing to update the value
5925 * must wait for a synchronous IPI to complete (which is trivial
5926 * if the caller is on the CPU already). This establishes the
5927 * necessary total order on variable updates.
5928 *
5929 * Note that because a guest time update may take place
5930 * anytime after the setting of the VCPU's request bit, the
5931 * correct TSC value must be set before the request. However,
5932 * to ensure the update actually makes it to any guest which
5933 * starts running in hardware virtualization between the set
5934 * and the acquisition of the spinlock, we must also ping the
5935 * CPU after setting the request bit.
5936 *
5937 */
5938
Gerd Hoffmannc8076602009-02-04 17:52:04 +01005939 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
5940 return 0;
5941 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
5942 return 0;
Zachary Amsden8cfdc002010-08-19 22:07:21 -10005943
5944 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
Gerd Hoffmannc8076602009-02-04 17:52:04 +01005945
Paolo Bonzini2f303b72013-09-25 13:53:07 +02005946 spin_lock(&kvm_lock);
Gerd Hoffmannc8076602009-02-04 17:52:04 +01005947 list_for_each_entry(kvm, &vm_list, vm_list) {
Gleb Natapov988a2ca2009-06-09 15:56:29 +03005948 kvm_for_each_vcpu(i, vcpu, kvm) {
Gerd Hoffmannc8076602009-02-04 17:52:04 +01005949 if (vcpu->cpu != freq->cpu)
5950 continue;
Zachary Amsdenc2855452010-09-18 14:38:15 -10005951 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
Gerd Hoffmannc8076602009-02-04 17:52:04 +01005952 if (vcpu->cpu != smp_processor_id())
Zachary Amsden8cfdc002010-08-19 22:07:21 -10005953 send_ipi = 1;
Gerd Hoffmannc8076602009-02-04 17:52:04 +01005954 }
5955 }
Paolo Bonzini2f303b72013-09-25 13:53:07 +02005956 spin_unlock(&kvm_lock);
Gerd Hoffmannc8076602009-02-04 17:52:04 +01005957
5958 if (freq->old < freq->new && send_ipi) {
5959 /*
5960 * We upscale the frequency. Must make the guest
5961 * doesn't see old kvmclock values while running with
5962 * the new frequency, otherwise we risk the guest sees
5963 * time go backwards.
5964 *
5965 * In case we update the frequency for another cpu
5966 * (which might be in guest context) send an interrupt
5967 * to kick the cpu out of guest context. Next time
5968 * guest context is entered kvmclock will be updated,
5969 * so the guest will not see stale values.
5970 */
Zachary Amsden8cfdc002010-08-19 22:07:21 -10005971 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
Gerd Hoffmannc8076602009-02-04 17:52:04 +01005972 }
5973 return 0;
5974}
5975
5976static struct notifier_block kvmclock_cpufreq_notifier_block = {
Zachary Amsden8cfdc002010-08-19 22:07:21 -10005977 .notifier_call = kvmclock_cpufreq_notifier
5978};
5979
Sebastian Andrzej Siewior251a5fd2016-07-13 17:16:33 +00005980static int kvmclock_cpu_online(unsigned int cpu)
Zachary Amsden8cfdc002010-08-19 22:07:21 -10005981{
Sebastian Andrzej Siewior251a5fd2016-07-13 17:16:33 +00005982 tsc_khz_changed(NULL);
5983 return 0;
Zachary Amsden8cfdc002010-08-19 22:07:21 -10005984}
5985
Zachary Amsdenb820cc02009-09-29 11:38:34 -10005986static void kvm_timer_init(void)
5987{
Zachary Amsdenc2855452010-09-18 14:38:15 -10005988 max_tsc_khz = tsc_khz;
Srivatsa S. Bhat460dd422014-03-11 02:09:01 +05305989
Zachary Amsdenb820cc02009-09-29 11:38:34 -10005990 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
Zachary Amsdenc2855452010-09-18 14:38:15 -10005991#ifdef CONFIG_CPU_FREQ
5992 struct cpufreq_policy policy;
Borislav Petkov758f5882016-09-04 19:13:57 +02005993 int cpu;
5994
Zachary Amsdenc2855452010-09-18 14:38:15 -10005995 memset(&policy, 0, sizeof(policy));
Avi Kivity3e26f232010-12-16 12:16:34 +02005996 cpu = get_cpu();
5997 cpufreq_get_policy(&policy, cpu);
Zachary Amsdenc2855452010-09-18 14:38:15 -10005998 if (policy.cpuinfo.max_freq)
5999 max_tsc_khz = policy.cpuinfo.max_freq;
Avi Kivity3e26f232010-12-16 12:16:34 +02006000 put_cpu();
Zachary Amsdenc2855452010-09-18 14:38:15 -10006001#endif
Zachary Amsdenb820cc02009-09-29 11:38:34 -10006002 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
6003 CPUFREQ_TRANSITION_NOTIFIER);
6004 }
Zachary Amsdenc2855452010-09-18 14:38:15 -10006005 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
Srivatsa S. Bhat460dd422014-03-11 02:09:01 +05306006
Thomas Gleixner73c1b412016-12-21 20:19:54 +01006007 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
Sebastian Andrzej Siewior251a5fd2016-07-13 17:16:33 +00006008 kvmclock_cpu_online, kvmclock_cpu_down_prep);
Zachary Amsdenb820cc02009-09-29 11:38:34 -10006009}
6010
Zhang, Yanminff9d07a2010-04-19 13:32:45 +08006011static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
6012
Gleb Natapovf5132b02011-11-10 14:57:22 +02006013int kvm_is_in_guest(void)
Zhang, Yanminff9d07a2010-04-19 13:32:45 +08006014{
Alex,Shi086c9852011-10-20 15:34:01 +08006015 return __this_cpu_read(current_vcpu) != NULL;
Zhang, Yanminff9d07a2010-04-19 13:32:45 +08006016}
6017
6018static int kvm_is_user_mode(void)
6019{
6020 int user_mode = 3;
Zhang, Yanmindcf46b92010-04-20 10:13:58 +08006021
Alex,Shi086c9852011-10-20 15:34:01 +08006022 if (__this_cpu_read(current_vcpu))
6023 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
Zhang, Yanmindcf46b92010-04-20 10:13:58 +08006024
Zhang, Yanminff9d07a2010-04-19 13:32:45 +08006025 return user_mode != 0;
6026}
6027
6028static unsigned long kvm_get_guest_ip(void)
6029{
6030 unsigned long ip = 0;
Zhang, Yanmindcf46b92010-04-20 10:13:58 +08006031
Alex,Shi086c9852011-10-20 15:34:01 +08006032 if (__this_cpu_read(current_vcpu))
6033 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
Zhang, Yanmindcf46b92010-04-20 10:13:58 +08006034
Zhang, Yanminff9d07a2010-04-19 13:32:45 +08006035 return ip;
6036}
6037
6038static struct perf_guest_info_callbacks kvm_guest_cbs = {
6039 .is_in_guest = kvm_is_in_guest,
6040 .is_user_mode = kvm_is_user_mode,
6041 .get_guest_ip = kvm_get_guest_ip,
6042};
6043
6044void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
6045{
Alex,Shi086c9852011-10-20 15:34:01 +08006046 __this_cpu_write(current_vcpu, vcpu);
Zhang, Yanminff9d07a2010-04-19 13:32:45 +08006047}
6048EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
6049
6050void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
6051{
Alex,Shi086c9852011-10-20 15:34:01 +08006052 __this_cpu_write(current_vcpu, NULL);
Zhang, Yanminff9d07a2010-04-19 13:32:45 +08006053}
6054EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
6055
Xiao Guangrongce88dec2011-07-12 03:33:44 +08006056static void kvm_set_mmio_spte_mask(void)
6057{
6058 u64 mask;
6059 int maxphyaddr = boot_cpu_data.x86_phys_bits;
6060
6061 /*
6062 * Set the reserved bits and the present bit of an paging-structure
6063 * entry to generate page fault with PFER.RSV = 1.
6064 */
Xiao Guangrong885032b2013-06-07 16:51:23 +08006065 /* Mask the reserved physical address bits. */
Tiejun Chend1431482014-09-01 18:44:04 +08006066 mask = rsvd_bits(maxphyaddr, 51);
Xiao Guangrong885032b2013-06-07 16:51:23 +08006067
Xiao Guangrong885032b2013-06-07 16:51:23 +08006068 /* Set the present bit. */
Xiao Guangrongce88dec2011-07-12 03:33:44 +08006069 mask |= 1ull;
6070
6071#ifdef CONFIG_X86_64
6072 /*
6073 * If reserved bit is not supported, clear the present bit to disable
6074 * mmio page fault.
6075 */
6076 if (maxphyaddr == 52)
6077 mask &= ~1ull;
6078#endif
6079
Peter Feinerdcdca5f2017-06-30 17:26:30 -07006080 kvm_mmu_set_mmio_spte_mask(mask, mask);
Xiao Guangrongce88dec2011-07-12 03:33:44 +08006081}
6082
Marcelo Tosatti16e8d742012-11-27 23:29:00 -02006083#ifdef CONFIG_X86_64
6084static void pvclock_gtod_update_fn(struct work_struct *work)
6085{
Marcelo Tosattid8281992012-11-27 23:29:01 -02006086 struct kvm *kvm;
6087
6088 struct kvm_vcpu *vcpu;
6089 int i;
6090
Paolo Bonzini2f303b72013-09-25 13:53:07 +02006091 spin_lock(&kvm_lock);
Marcelo Tosattid8281992012-11-27 23:29:01 -02006092 list_for_each_entry(kvm, &vm_list, vm_list)
6093 kvm_for_each_vcpu(i, vcpu, kvm)
Guo Hui Liu105b21b2014-09-12 13:43:19 +08006094 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
Marcelo Tosattid8281992012-11-27 23:29:01 -02006095 atomic_set(&kvm_guest_has_master_clock, 0);
Paolo Bonzini2f303b72013-09-25 13:53:07 +02006096 spin_unlock(&kvm_lock);
Marcelo Tosatti16e8d742012-11-27 23:29:00 -02006097}
6098
6099static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
6100
6101/*
6102 * Notification about pvclock gtod data update.
6103 */
6104static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
6105 void *priv)
6106{
6107 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
6108 struct timekeeper *tk = priv;
6109
6110 update_pvclock_gtod(tk);
6111
6112 /* disable master clock if host does not trust, or does not
6113 * use, TSC clocksource
6114 */
6115 if (gtod->clock.vclock_mode != VCLOCK_TSC &&
6116 atomic_read(&kvm_guest_has_master_clock) != 0)
6117 queue_work(system_long_wq, &pvclock_gtod_work);
6118
6119 return 0;
6120}
6121
6122static struct notifier_block pvclock_gtod_notifier = {
6123 .notifier_call = pvclock_gtod_notify,
6124};
6125#endif
6126
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08006127int kvm_arch_init(void *opaque)
Carsten Otte043405e2007-10-10 17:16:19 +02006128{
Zachary Amsdenb820cc02009-09-29 11:38:34 -10006129 int r;
Mathias Krause6b61edf2013-06-26 20:36:23 +02006130 struct kvm_x86_ops *ops = opaque;
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08006131
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08006132 if (kvm_x86_ops) {
6133 printk(KERN_ERR "kvm: already loaded the other module\n");
Zhang Xiantao56c6d282007-11-18 20:43:21 +08006134 r = -EEXIST;
6135 goto out;
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08006136 }
6137
6138 if (!ops->cpu_has_kvm_support()) {
6139 printk(KERN_ERR "kvm: no hardware support\n");
Zhang Xiantao56c6d282007-11-18 20:43:21 +08006140 r = -EOPNOTSUPP;
6141 goto out;
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08006142 }
6143 if (ops->disabled_by_bios()) {
6144 printk(KERN_ERR "kvm: disabled by bios\n");
Zhang Xiantao56c6d282007-11-18 20:43:21 +08006145 r = -EOPNOTSUPP;
6146 goto out;
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08006147 }
6148
Marcelo Tosatti013f6a52013-01-03 11:41:39 -02006149 r = -ENOMEM;
6150 shared_msrs = alloc_percpu(struct kvm_shared_msrs);
6151 if (!shared_msrs) {
6152 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
6153 goto out;
6154 }
6155
Avi Kivity97db56c2008-01-13 13:23:56 +02006156 r = kvm_mmu_module_init();
6157 if (r)
Marcelo Tosatti013f6a52013-01-03 11:41:39 -02006158 goto out_free_percpu;
Avi Kivity97db56c2008-01-13 13:23:56 +02006159
Xiao Guangrongce88dec2011-07-12 03:33:44 +08006160 kvm_set_mmio_spte_mask();
Avi Kivity97db56c2008-01-13 13:23:56 +02006161
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08006162 kvm_x86_ops = ops;
Paolo Bonzini920c8372014-03-26 15:54:00 +01006163
Sheng Yang7b523452008-04-25 21:13:50 +08006164 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
Bandan Dasffb128c2016-07-12 18:18:49 -04006165 PT_DIRTY_MASK, PT64_NX_MASK, 0,
Tom Lendackyd0ec49d2017-07-17 16:10:27 -05006166 PT_PRESENT_MASK, 0, sme_me_mask);
Zachary Amsdenb820cc02009-09-29 11:38:34 -10006167 kvm_timer_init();
Gerd Hoffmannc8076602009-02-04 17:52:04 +01006168
Zhang, Yanminff9d07a2010-04-19 13:32:45 +08006169 perf_register_guest_info_callbacks(&kvm_guest_cbs);
6170
Borislav Petkovd366bf72016-04-04 22:25:02 +02006171 if (boot_cpu_has(X86_FEATURE_XSAVE))
Dexuan Cui2acf9232010-06-10 11:27:12 +08006172 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
6173
Gleb Natapovc5cc4212012-08-05 15:58:30 +03006174 kvm_lapic_init();
Marcelo Tosatti16e8d742012-11-27 23:29:00 -02006175#ifdef CONFIG_X86_64
6176 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
6177#endif
6178
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08006179 return 0;
Zhang Xiantao56c6d282007-11-18 20:43:21 +08006180
Marcelo Tosatti013f6a52013-01-03 11:41:39 -02006181out_free_percpu:
6182 free_percpu(shared_msrs);
Zhang Xiantao56c6d282007-11-18 20:43:21 +08006183out:
Zhang Xiantao56c6d282007-11-18 20:43:21 +08006184 return r;
Carsten Otte043405e2007-10-10 17:16:19 +02006185}
Hollis Blanchard8776e512007-10-31 17:24:24 -05006186
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08006187void kvm_arch_exit(void)
6188{
David Matlackcef84c32016-12-16 14:30:36 -08006189 kvm_lapic_exit();
Zhang, Yanminff9d07a2010-04-19 13:32:45 +08006190 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
6191
Jan Kiszka888d2562009-04-17 19:24:58 +02006192 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
6193 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
6194 CPUFREQ_TRANSITION_NOTIFIER);
Sebastian Andrzej Siewior251a5fd2016-07-13 17:16:33 +00006195 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
Marcelo Tosatti16e8d742012-11-27 23:29:00 -02006196#ifdef CONFIG_X86_64
6197 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
6198#endif
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08006199 kvm_x86_ops = NULL;
Zhang Xiantao56c6d282007-11-18 20:43:21 +08006200 kvm_mmu_module_exit();
Marcelo Tosatti013f6a52013-01-03 11:41:39 -02006201 free_percpu(shared_msrs);
Zhang Xiantao56c6d282007-11-18 20:43:21 +08006202}
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08006203
Joel Schopp5cb56052015-03-02 13:43:31 -06006204int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
Hollis Blanchard8776e512007-10-31 17:24:24 -05006205{
6206 ++vcpu->stat.halt_exits;
Paolo Bonzini35754c92015-07-29 12:05:37 +02006207 if (lapic_in_kernel(vcpu)) {
Avi Kivitya4535292008-04-13 17:54:35 +03006208 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
Hollis Blanchard8776e512007-10-31 17:24:24 -05006209 return 1;
6210 } else {
6211 vcpu->run->exit_reason = KVM_EXIT_HLT;
6212 return 0;
6213 }
6214}
Joel Schopp5cb56052015-03-02 13:43:31 -06006215EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
6216
6217int kvm_emulate_halt(struct kvm_vcpu *vcpu)
6218{
Kyle Huey6affcbe2016-11-29 12:40:40 -08006219 int ret = kvm_skip_emulated_instruction(vcpu);
6220 /*
6221 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
6222 * KVM_EXIT_DEBUG here.
6223 */
6224 return kvm_vcpu_halt(vcpu) && ret;
Joel Schopp5cb56052015-03-02 13:43:31 -06006225}
Hollis Blanchard8776e512007-10-31 17:24:24 -05006226EXPORT_SYMBOL_GPL(kvm_emulate_halt);
6227
Arnd Bergmann8ef81a92017-02-09 16:10:42 +01006228#ifdef CONFIG_X86_64
Marcelo Tosatti55dd00a2017-01-24 15:09:39 -02006229static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
6230 unsigned long clock_type)
6231{
6232 struct kvm_clock_pairing clock_pairing;
6233 struct timespec ts;
Paolo Bonzini80fbd892017-02-08 10:57:24 +01006234 u64 cycle;
Marcelo Tosatti55dd00a2017-01-24 15:09:39 -02006235 int ret;
6236
6237 if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
6238 return -KVM_EOPNOTSUPP;
6239
6240 if (kvm_get_walltime_and_clockread(&ts, &cycle) == false)
6241 return -KVM_EOPNOTSUPP;
6242
6243 clock_pairing.sec = ts.tv_sec;
6244 clock_pairing.nsec = ts.tv_nsec;
6245 clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
6246 clock_pairing.flags = 0;
6247
6248 ret = 0;
6249 if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
6250 sizeof(struct kvm_clock_pairing)))
6251 ret = -KVM_EFAULT;
6252
6253 return ret;
6254}
Arnd Bergmann8ef81a92017-02-09 16:10:42 +01006255#endif
Marcelo Tosatti55dd00a2017-01-24 15:09:39 -02006256
Srivatsa Vaddagiri6aef2662013-08-26 14:18:34 +05306257/*
6258 * kvm_pv_kick_cpu_op: Kick a vcpu.
6259 *
6260 * @apicid - apicid of vcpu to be kicked.
6261 */
6262static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
6263{
Raghavendra K T24d21662013-08-26 14:18:35 +05306264 struct kvm_lapic_irq lapic_irq;
Srivatsa Vaddagiri6aef2662013-08-26 14:18:34 +05306265
Raghavendra K T24d21662013-08-26 14:18:35 +05306266 lapic_irq.shorthand = 0;
6267 lapic_irq.dest_mode = 0;
Longpeng(Mike)ebd28fc2017-08-02 11:20:51 +08006268 lapic_irq.level = 0;
Raghavendra K T24d21662013-08-26 14:18:35 +05306269 lapic_irq.dest_id = apicid;
James Sullivan93bbf0b2015-03-18 19:26:03 -06006270 lapic_irq.msi_redir_hint = false;
Srivatsa Vaddagiri6aef2662013-08-26 14:18:34 +05306271
Raghavendra K T24d21662013-08-26 14:18:35 +05306272 lapic_irq.delivery_mode = APIC_DM_REMRD;
Xiubo Li795a1492015-03-13 17:39:44 +08006273 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
Srivatsa Vaddagiri6aef2662013-08-26 14:18:34 +05306274}
6275
Andrey Smetanind62caab2015-11-10 15:36:33 +03006276void kvm_vcpu_deactivate_apicv(struct kvm_vcpu *vcpu)
6277{
6278 vcpu->arch.apicv_active = false;
6279 kvm_x86_ops->refresh_apicv_exec_ctrl(vcpu);
6280}
6281
Hollis Blanchard8776e512007-10-31 17:24:24 -05006282int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
6283{
6284 unsigned long nr, a0, a1, a2, a3, ret;
Kyle Huey6affcbe2016-11-29 12:40:40 -08006285 int op_64_bit, r;
Hollis Blanchard8776e512007-10-31 17:24:24 -05006286
Kyle Huey6affcbe2016-11-29 12:40:40 -08006287 r = kvm_skip_emulated_instruction(vcpu);
Joel Schopp5cb56052015-03-02 13:43:31 -06006288
Gleb Natapov55cd8e52010-01-17 15:51:22 +02006289 if (kvm_hv_hypercall_enabled(vcpu->kvm))
6290 return kvm_hv_hypercall(vcpu);
6291
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03006292 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
6293 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
6294 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
6295 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
6296 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
Hollis Blanchard8776e512007-10-31 17:24:24 -05006297
Marcelo Tosatti229456f2009-06-17 09:22:14 -03006298 trace_kvm_hypercall(nr, a0, a1, a2, a3);
Feng (Eric) Liu2714d1d2008-04-10 15:31:10 -04006299
Nadav Amita449c7a2014-06-18 17:19:24 +03006300 op_64_bit = is_64_bit_mode(vcpu);
6301 if (!op_64_bit) {
Hollis Blanchard8776e512007-10-31 17:24:24 -05006302 nr &= 0xFFFFFFFF;
6303 a0 &= 0xFFFFFFFF;
6304 a1 &= 0xFFFFFFFF;
6305 a2 &= 0xFFFFFFFF;
6306 a3 &= 0xFFFFFFFF;
6307 }
6308
Jan Kiszka07708c42009-08-03 18:43:28 +02006309 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
6310 ret = -KVM_EPERM;
6311 goto out;
6312 }
6313
Hollis Blanchard8776e512007-10-31 17:24:24 -05006314 switch (nr) {
Avi Kivityb93463a2007-10-25 16:52:32 +02006315 case KVM_HC_VAPIC_POLL_IRQ:
6316 ret = 0;
6317 break;
Srivatsa Vaddagiri6aef2662013-08-26 14:18:34 +05306318 case KVM_HC_KICK_CPU:
6319 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
6320 ret = 0;
6321 break;
Arnd Bergmann8ef81a92017-02-09 16:10:42 +01006322#ifdef CONFIG_X86_64
Marcelo Tosatti55dd00a2017-01-24 15:09:39 -02006323 case KVM_HC_CLOCK_PAIRING:
6324 ret = kvm_pv_clock_pairing(vcpu, a0, a1);
6325 break;
Arnd Bergmann8ef81a92017-02-09 16:10:42 +01006326#endif
Hollis Blanchard8776e512007-10-31 17:24:24 -05006327 default:
6328 ret = -KVM_ENOSYS;
6329 break;
6330 }
Jan Kiszka07708c42009-08-03 18:43:28 +02006331out:
Nadav Amita449c7a2014-06-18 17:19:24 +03006332 if (!op_64_bit)
6333 ret = (u32)ret;
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03006334 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
Amit Shahf11c3a82008-02-21 01:00:30 +05306335 ++vcpu->stat.hypercalls;
Marcelo Tosatti2f333bc2008-02-22 12:21:37 -05006336 return r;
Hollis Blanchard8776e512007-10-31 17:24:24 -05006337}
6338EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
6339
Jan Kiszkab6785de2012-09-20 07:43:17 +02006340static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
Hollis Blanchard8776e512007-10-31 17:24:24 -05006341{
Avi Kivityd6aa1002011-04-20 15:47:13 +03006342 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
Hollis Blanchard8776e512007-10-31 17:24:24 -05006343 char instruction[3];
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03006344 unsigned long rip = kvm_rip_read(vcpu);
Hollis Blanchard8776e512007-10-31 17:24:24 -05006345
Hollis Blanchard8776e512007-10-31 17:24:24 -05006346 kvm_x86_ops->patch_hypercall(vcpu, instruction);
Hollis Blanchard8776e512007-10-31 17:24:24 -05006347
Dmitry Vyukovce2e8522017-01-17 14:51:04 +01006348 return emulator_write_emulated(ctxt, rip, instruction, 3,
6349 &ctxt->exception);
Hollis Blanchard8776e512007-10-31 17:24:24 -05006350}
6351
Avi Kivity851ba692009-08-24 11:10:17 +03006352static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006353{
Matt Gingell782d4222015-11-16 15:26:00 -08006354 return vcpu->run->request_interrupt_window &&
6355 likely(!pic_in_kernel(vcpu->kvm));
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006356}
6357
Avi Kivity851ba692009-08-24 11:10:17 +03006358static void post_kvm_run_save(struct kvm_vcpu *vcpu)
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006359{
Avi Kivity851ba692009-08-24 11:10:17 +03006360 struct kvm_run *kvm_run = vcpu->run;
6361
Jan Kiszka91586a32009-10-05 13:07:21 +02006362 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
Paolo Bonzinif0778252015-04-01 15:06:40 +02006363 kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
Avi Kivity2d3ad1f2008-02-24 11:20:43 +02006364 kvm_run->cr8 = kvm_get_cr8(vcpu);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006365 kvm_run->apic_base = kvm_get_apic_base(vcpu);
Matt Gingell127a4572015-11-17 17:32:05 +01006366 kvm_run->ready_for_interrupt_injection =
6367 pic_in_kernel(vcpu->kvm) ||
Matt Gingell782d4222015-11-16 15:26:00 -08006368 kvm_vcpu_ready_for_interrupt_injection(vcpu);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006369}
6370
Gleb Natapov95ba8273132009-04-21 17:45:08 +03006371static void update_cr8_intercept(struct kvm_vcpu *vcpu)
6372{
6373 int max_irr, tpr;
6374
6375 if (!kvm_x86_ops->update_cr8_intercept)
6376 return;
6377
Paolo Bonzinibce87cc2016-01-08 13:48:51 +01006378 if (!lapic_in_kernel(vcpu))
Avi Kivity88c808f2009-08-17 22:49:40 +03006379 return;
6380
Andrey Smetanind62caab2015-11-10 15:36:33 +03006381 if (vcpu->arch.apicv_active)
6382 return;
6383
Gleb Natapov8db3baa2009-05-11 13:35:54 +03006384 if (!vcpu->arch.apic->vapic_addr)
6385 max_irr = kvm_lapic_find_highest_irr(vcpu);
6386 else
6387 max_irr = -1;
Gleb Natapov95ba8273132009-04-21 17:45:08 +03006388
6389 if (max_irr != -1)
6390 max_irr >>= 4;
6391
6392 tpr = kvm_lapic_get_cr8(vcpu);
6393
6394 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
6395}
6396
Jan Kiszkab6b8a142014-03-07 20:03:12 +01006397static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
Gleb Natapov95ba8273132009-04-21 17:45:08 +03006398{
Jan Kiszkab6b8a142014-03-07 20:03:12 +01006399 int r;
6400
Gleb Natapov95ba8273132009-04-21 17:45:08 +03006401 /* try to reinject previous events if any */
Wanpeng Li664f8e22017-08-24 03:35:09 -07006402 if (vcpu->arch.exception.injected) {
6403 kvm_x86_ops->queue_exception(vcpu);
6404 return 0;
6405 }
6406
6407 /*
6408 * Exceptions must be injected immediately, or the exception
6409 * frame will have the address of the NMI or interrupt handler.
6410 */
6411 if (!vcpu->arch.exception.pending) {
6412 if (vcpu->arch.nmi_injected) {
6413 kvm_x86_ops->set_nmi(vcpu);
6414 return 0;
6415 }
6416
6417 if (vcpu->arch.interrupt.pending) {
6418 kvm_x86_ops->set_irq(vcpu);
6419 return 0;
6420 }
6421 }
6422
6423 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6424 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6425 if (r != 0)
6426 return r;
6427 }
6428
6429 /* try to inject new event if pending */
Gleb Natapovb59bb7b2009-07-09 15:33:51 +03006430 if (vcpu->arch.exception.pending) {
Avi Kivity5c1c85d02010-03-11 13:01:59 +02006431 trace_kvm_inj_exception(vcpu->arch.exception.nr,
6432 vcpu->arch.exception.has_error_code,
6433 vcpu->arch.exception.error_code);
Nadav Amitd6e8c852014-07-24 14:51:24 +03006434
Wanpeng Li664f8e22017-08-24 03:35:09 -07006435 vcpu->arch.exception.pending = false;
6436 vcpu->arch.exception.injected = true;
6437
Nadav Amitd6e8c852014-07-24 14:51:24 +03006438 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
6439 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
6440 X86_EFLAGS_RF);
6441
Nadav Amit6bdf0662014-09-30 20:49:14 +03006442 if (vcpu->arch.exception.nr == DB_VECTOR &&
6443 (vcpu->arch.dr7 & DR7_GD)) {
6444 vcpu->arch.dr7 &= ~DR7_GD;
6445 kvm_update_dr7(vcpu);
6446 }
6447
Wanpeng Licfcd20e2017-07-13 18:30:39 -07006448 kvm_x86_ops->queue_exception(vcpu);
Ladi Prosek72d7b372017-10-11 16:54:41 +02006449 } else if (vcpu->arch.smi_pending && !is_smm(vcpu) && kvm_x86_ops->smi_allowed(vcpu)) {
Paolo Bonzinic43203c2016-06-01 22:26:00 +02006450 vcpu->arch.smi_pending = false;
Paolo Bonziniee2cd4b2016-06-01 22:26:01 +02006451 enter_smm(vcpu);
Paolo Bonzinic43203c2016-06-01 22:26:00 +02006452 } else if (vcpu->arch.nmi_pending && kvm_x86_ops->nmi_allowed(vcpu)) {
Yuki Shibuya321c5652016-03-24 05:17:03 +00006453 --vcpu->arch.nmi_pending;
6454 vcpu->arch.nmi_injected = true;
6455 kvm_x86_ops->set_nmi(vcpu);
Yang Zhangc7c9c562013-01-25 10:18:51 +08006456 } else if (kvm_cpu_has_injectable_intr(vcpu)) {
Bandan Das9242b5b2014-07-08 00:30:23 -04006457 /*
6458 * Because interrupts can be injected asynchronously, we are
6459 * calling check_nested_events again here to avoid a race condition.
6460 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
6461 * proposal and current concerns. Perhaps we should be setting
6462 * KVM_REQ_EVENT only on certain events and not unconditionally?
6463 */
6464 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6465 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6466 if (r != 0)
6467 return r;
6468 }
Gleb Natapov95ba8273132009-04-21 17:45:08 +03006469 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
Gleb Natapov66fd3f72009-05-11 13:35:50 +03006470 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
6471 false);
6472 kvm_x86_ops->set_irq(vcpu);
Gleb Natapov95ba8273132009-04-21 17:45:08 +03006473 }
6474 }
Paolo Bonziniee2cd4b2016-06-01 22:26:01 +02006475
Jan Kiszkab6b8a142014-03-07 20:03:12 +01006476 return 0;
Gleb Natapov95ba8273132009-04-21 17:45:08 +03006477}
6478
Avi Kivity7460fb4a2011-09-20 13:43:14 +03006479static void process_nmi(struct kvm_vcpu *vcpu)
6480{
6481 unsigned limit = 2;
6482
6483 /*
6484 * x86 is limited to one NMI running, and one NMI pending after it.
6485 * If an NMI is already in progress, limit further NMIs to just one.
6486 * Otherwise, allow two (and we'll inject the first one immediately).
6487 */
6488 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
6489 limit = 1;
6490
6491 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
6492 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
6493 kvm_make_request(KVM_REQ_EVENT, vcpu);
6494}
6495
Paolo Bonziniee2cd4b2016-06-01 22:26:01 +02006496static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
Paolo Bonzini660a5d52015-05-05 11:50:23 +02006497{
6498 u32 flags = 0;
6499 flags |= seg->g << 23;
6500 flags |= seg->db << 22;
6501 flags |= seg->l << 21;
6502 flags |= seg->avl << 20;
6503 flags |= seg->present << 15;
6504 flags |= seg->dpl << 13;
6505 flags |= seg->s << 12;
6506 flags |= seg->type << 8;
6507 return flags;
6508}
6509
Paolo Bonziniee2cd4b2016-06-01 22:26:01 +02006510static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
Paolo Bonzini660a5d52015-05-05 11:50:23 +02006511{
6512 struct kvm_segment seg;
6513 int offset;
6514
6515 kvm_get_segment(vcpu, &seg, n);
6516 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
6517
6518 if (n < 3)
6519 offset = 0x7f84 + n * 12;
6520 else
6521 offset = 0x7f2c + (n - 3) * 12;
6522
6523 put_smstate(u32, buf, offset + 8, seg.base);
6524 put_smstate(u32, buf, offset + 4, seg.limit);
Paolo Bonziniee2cd4b2016-06-01 22:26:01 +02006525 put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
Paolo Bonzini660a5d52015-05-05 11:50:23 +02006526}
6527
Alexander Kuleshovefbb2882015-09-06 19:35:41 +06006528#ifdef CONFIG_X86_64
Paolo Bonziniee2cd4b2016-06-01 22:26:01 +02006529static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
Paolo Bonzini660a5d52015-05-05 11:50:23 +02006530{
6531 struct kvm_segment seg;
6532 int offset;
6533 u16 flags;
6534
6535 kvm_get_segment(vcpu, &seg, n);
6536 offset = 0x7e00 + n * 16;
6537
Paolo Bonziniee2cd4b2016-06-01 22:26:01 +02006538 flags = enter_smm_get_segment_flags(&seg) >> 8;
Paolo Bonzini660a5d52015-05-05 11:50:23 +02006539 put_smstate(u16, buf, offset, seg.selector);
6540 put_smstate(u16, buf, offset + 2, flags);
6541 put_smstate(u32, buf, offset + 4, seg.limit);
6542 put_smstate(u64, buf, offset + 8, seg.base);
6543}
Alexander Kuleshovefbb2882015-09-06 19:35:41 +06006544#endif
Paolo Bonzini660a5d52015-05-05 11:50:23 +02006545
Paolo Bonziniee2cd4b2016-06-01 22:26:01 +02006546static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
Paolo Bonzini660a5d52015-05-05 11:50:23 +02006547{
6548 struct desc_ptr dt;
6549 struct kvm_segment seg;
6550 unsigned long val;
6551 int i;
6552
6553 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
6554 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
6555 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
6556 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
6557
6558 for (i = 0; i < 8; i++)
6559 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
6560
6561 kvm_get_dr(vcpu, 6, &val);
6562 put_smstate(u32, buf, 0x7fcc, (u32)val);
6563 kvm_get_dr(vcpu, 7, &val);
6564 put_smstate(u32, buf, 0x7fc8, (u32)val);
6565
6566 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6567 put_smstate(u32, buf, 0x7fc4, seg.selector);
6568 put_smstate(u32, buf, 0x7f64, seg.base);
6569 put_smstate(u32, buf, 0x7f60, seg.limit);
Paolo Bonziniee2cd4b2016-06-01 22:26:01 +02006570 put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
Paolo Bonzini660a5d52015-05-05 11:50:23 +02006571
6572 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6573 put_smstate(u32, buf, 0x7fc0, seg.selector);
6574 put_smstate(u32, buf, 0x7f80, seg.base);
6575 put_smstate(u32, buf, 0x7f7c, seg.limit);
Paolo Bonziniee2cd4b2016-06-01 22:26:01 +02006576 put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
Paolo Bonzini660a5d52015-05-05 11:50:23 +02006577
6578 kvm_x86_ops->get_gdt(vcpu, &dt);
6579 put_smstate(u32, buf, 0x7f74, dt.address);
6580 put_smstate(u32, buf, 0x7f70, dt.size);
6581
6582 kvm_x86_ops->get_idt(vcpu, &dt);
6583 put_smstate(u32, buf, 0x7f58, dt.address);
6584 put_smstate(u32, buf, 0x7f54, dt.size);
6585
6586 for (i = 0; i < 6; i++)
Paolo Bonziniee2cd4b2016-06-01 22:26:01 +02006587 enter_smm_save_seg_32(vcpu, buf, i);
Paolo Bonzini660a5d52015-05-05 11:50:23 +02006588
6589 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
6590
6591 /* revision id */
6592 put_smstate(u32, buf, 0x7efc, 0x00020000);
6593 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
6594}
6595
Paolo Bonziniee2cd4b2016-06-01 22:26:01 +02006596static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
Paolo Bonzini660a5d52015-05-05 11:50:23 +02006597{
6598#ifdef CONFIG_X86_64
6599 struct desc_ptr dt;
6600 struct kvm_segment seg;
6601 unsigned long val;
6602 int i;
6603
6604 for (i = 0; i < 16; i++)
6605 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
6606
6607 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
6608 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
6609
6610 kvm_get_dr(vcpu, 6, &val);
6611 put_smstate(u64, buf, 0x7f68, val);
6612 kvm_get_dr(vcpu, 7, &val);
6613 put_smstate(u64, buf, 0x7f60, val);
6614
6615 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
6616 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
6617 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
6618
6619 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
6620
6621 /* revision id */
6622 put_smstate(u32, buf, 0x7efc, 0x00020064);
6623
6624 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
6625
6626 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6627 put_smstate(u16, buf, 0x7e90, seg.selector);
Paolo Bonziniee2cd4b2016-06-01 22:26:01 +02006628 put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
Paolo Bonzini660a5d52015-05-05 11:50:23 +02006629 put_smstate(u32, buf, 0x7e94, seg.limit);
6630 put_smstate(u64, buf, 0x7e98, seg.base);
6631
6632 kvm_x86_ops->get_idt(vcpu, &dt);
6633 put_smstate(u32, buf, 0x7e84, dt.size);
6634 put_smstate(u64, buf, 0x7e88, dt.address);
6635
6636 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6637 put_smstate(u16, buf, 0x7e70, seg.selector);
Paolo Bonziniee2cd4b2016-06-01 22:26:01 +02006638 put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
Paolo Bonzini660a5d52015-05-05 11:50:23 +02006639 put_smstate(u32, buf, 0x7e74, seg.limit);
6640 put_smstate(u64, buf, 0x7e78, seg.base);
6641
6642 kvm_x86_ops->get_gdt(vcpu, &dt);
6643 put_smstate(u32, buf, 0x7e64, dt.size);
6644 put_smstate(u64, buf, 0x7e68, dt.address);
6645
6646 for (i = 0; i < 6; i++)
Paolo Bonziniee2cd4b2016-06-01 22:26:01 +02006647 enter_smm_save_seg_64(vcpu, buf, i);
Paolo Bonzini660a5d52015-05-05 11:50:23 +02006648#else
6649 WARN_ON_ONCE(1);
6650#endif
6651}
6652
Paolo Bonziniee2cd4b2016-06-01 22:26:01 +02006653static void enter_smm(struct kvm_vcpu *vcpu)
Paolo Bonzini64d60672015-05-07 11:36:11 +02006654{
Paolo Bonzini660a5d52015-05-05 11:50:23 +02006655 struct kvm_segment cs, ds;
Paolo Bonzini18c36262015-08-07 12:27:54 +02006656 struct desc_ptr dt;
Paolo Bonzini660a5d52015-05-05 11:50:23 +02006657 char buf[512];
6658 u32 cr0;
6659
Paolo Bonzini660a5d52015-05-05 11:50:23 +02006660 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
Paolo Bonzini660a5d52015-05-05 11:50:23 +02006661 memset(buf, 0, 512);
Radim Krčmářd6321d42017-08-05 00:12:49 +02006662 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
Paolo Bonziniee2cd4b2016-06-01 22:26:01 +02006663 enter_smm_save_state_64(vcpu, buf);
Paolo Bonzini660a5d52015-05-05 11:50:23 +02006664 else
Paolo Bonziniee2cd4b2016-06-01 22:26:01 +02006665 enter_smm_save_state_32(vcpu, buf);
Paolo Bonzini660a5d52015-05-05 11:50:23 +02006666
Ladi Prosek0234bf82017-10-11 16:54:40 +02006667 /*
6668 * Give pre_enter_smm() a chance to make ISA-specific changes to the
6669 * vCPU state (e.g. leave guest mode) after we've saved the state into
6670 * the SMM state-save area.
6671 */
6672 kvm_x86_ops->pre_enter_smm(vcpu, buf);
6673
6674 vcpu->arch.hflags |= HF_SMM_MASK;
Paolo Bonzini54bf36a2015-04-08 15:39:23 +02006675 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
Paolo Bonzini660a5d52015-05-05 11:50:23 +02006676
6677 if (kvm_x86_ops->get_nmi_mask(vcpu))
6678 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
6679 else
6680 kvm_x86_ops->set_nmi_mask(vcpu, true);
6681
6682 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
6683 kvm_rip_write(vcpu, 0x8000);
6684
6685 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
6686 kvm_x86_ops->set_cr0(vcpu, cr0);
6687 vcpu->arch.cr0 = cr0;
6688
6689 kvm_x86_ops->set_cr4(vcpu, 0);
6690
Paolo Bonzini18c36262015-08-07 12:27:54 +02006691 /* Undocumented: IDT limit is set to zero on entry to SMM. */
6692 dt.address = dt.size = 0;
6693 kvm_x86_ops->set_idt(vcpu, &dt);
6694
Paolo Bonzini660a5d52015-05-05 11:50:23 +02006695 __kvm_set_dr(vcpu, 7, DR7_FIXED_1);
6696
6697 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
6698 cs.base = vcpu->arch.smbase;
6699
6700 ds.selector = 0;
6701 ds.base = 0;
6702
6703 cs.limit = ds.limit = 0xffffffff;
6704 cs.type = ds.type = 0x3;
6705 cs.dpl = ds.dpl = 0;
6706 cs.db = ds.db = 0;
6707 cs.s = ds.s = 1;
6708 cs.l = ds.l = 0;
6709 cs.g = ds.g = 1;
6710 cs.avl = ds.avl = 0;
6711 cs.present = ds.present = 1;
6712 cs.unusable = ds.unusable = 0;
6713 cs.padding = ds.padding = 0;
6714
6715 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
6716 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
6717 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
6718 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
6719 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
6720 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
6721
Radim Krčmářd6321d42017-08-05 00:12:49 +02006722 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
Paolo Bonzini660a5d52015-05-05 11:50:23 +02006723 kvm_x86_ops->set_efer(vcpu, 0);
6724
6725 kvm_update_cpuid(vcpu);
6726 kvm_mmu_reset_context(vcpu);
Paolo Bonzini64d60672015-05-07 11:36:11 +02006727}
6728
Paolo Bonziniee2cd4b2016-06-01 22:26:01 +02006729static void process_smi(struct kvm_vcpu *vcpu)
Paolo Bonzinic43203c2016-06-01 22:26:00 +02006730{
6731 vcpu->arch.smi_pending = true;
6732 kvm_make_request(KVM_REQ_EVENT, vcpu);
6733}
6734
Paolo Bonzini2860c4b2016-01-07 15:05:10 +01006735void kvm_make_scan_ioapic_request(struct kvm *kvm)
6736{
6737 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
6738}
6739
Yang Zhang3d81bc72013-04-11 19:25:13 +08006740static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
Yang Zhangc7c9c562013-01-25 10:18:51 +08006741{
Andrey Smetanin5c9194122015-11-10 15:36:34 +03006742 u64 eoi_exit_bitmap[4];
6743
Yang Zhang3d81bc72013-04-11 19:25:13 +08006744 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
6745 return;
Yang Zhangc7c9c562013-01-25 10:18:51 +08006746
Andrey Smetanin63086302015-11-10 15:36:32 +03006747 bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
Yang Zhangc7c9c562013-01-25 10:18:51 +08006748
Steve Rutherfordb053b2a2015-07-29 23:32:35 -07006749 if (irqchip_split(vcpu->kvm))
Andrey Smetanin63086302015-11-10 15:36:32 +03006750 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
Radim Krčmářdb2bdcb2015-10-08 20:23:34 +02006751 else {
Paolo Bonzini76dfafd52016-12-19 17:17:11 +01006752 if (kvm_x86_ops->sync_pir_to_irr && vcpu->arch.apicv_active)
Andrey Smetanind62caab2015-11-10 15:36:33 +03006753 kvm_x86_ops->sync_pir_to_irr(vcpu);
Andrey Smetanin63086302015-11-10 15:36:32 +03006754 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
Radim Krčmářdb2bdcb2015-10-08 20:23:34 +02006755 }
Andrey Smetanin5c9194122015-11-10 15:36:34 +03006756 bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
6757 vcpu_to_synic(vcpu)->vec_bitmap, 256);
6758 kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
Yang Zhangc7c9c562013-01-25 10:18:51 +08006759}
6760
Radim Krčmářa70656b2014-09-18 12:38:36 -04006761static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu)
6762{
6763 ++vcpu->stat.tlb_flush;
6764 kvm_x86_ops->tlb_flush(vcpu);
6765}
6766
Radim Krčmářb1394e72017-11-30 19:05:45 +01006767void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
6768 unsigned long start, unsigned long end)
6769{
6770 unsigned long apic_address;
6771
6772 /*
6773 * The physical address of apic access page is stored in the VMCS.
6774 * Update it when it becomes invalid.
6775 */
6776 apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
6777 if (start <= apic_address && apic_address < end)
6778 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
6779}
6780
Tang Chen4256f432014-09-24 15:57:54 +08006781void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
6782{
Tang Chenc24ae0d2014-09-24 15:57:58 +08006783 struct page *page = NULL;
6784
Paolo Bonzini35754c92015-07-29 12:05:37 +02006785 if (!lapic_in_kernel(vcpu))
Paolo Bonzinif439ed22014-10-02 13:53:24 +02006786 return;
6787
Tang Chen4256f432014-09-24 15:57:54 +08006788 if (!kvm_x86_ops->set_apic_access_page_addr)
6789 return;
6790
Tang Chenc24ae0d2014-09-24 15:57:58 +08006791 page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
Andrea Arcangelie8fd5e92015-05-08 14:32:56 +02006792 if (is_error_page(page))
6793 return;
Tang Chenc24ae0d2014-09-24 15:57:58 +08006794 kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
6795
6796 /*
6797 * Do not pin apic access page in memory, the MMU notifier
6798 * will call us again if it is migrated or swapped out.
6799 */
6800 put_page(page);
Tang Chen4256f432014-09-24 15:57:54 +08006801}
6802EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
6803
Takuya Yoshikawa9357d932013-12-13 15:08:38 +09006804/*
Paolo Bonzini362c6982015-02-06 12:48:04 +01006805 * Returns 1 to let vcpu_run() continue the guest execution loop without
Takuya Yoshikawa9357d932013-12-13 15:08:38 +09006806 * exiting to the userspace. Otherwise, the value will be returned to the
6807 * userspace.
6808 */
Avi Kivity851ba692009-08-24 11:10:17 +03006809static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006810{
6811 int r;
Matt Gingell62a193e2015-11-16 15:26:07 -08006812 bool req_int_win =
6813 dm_request_for_irq_injection(vcpu) &&
6814 kvm_cpu_accept_dm_intr(vcpu);
6815
Jan Kiszka730dca42013-04-28 10:50:52 +02006816 bool req_immediate_exit = false;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006817
Radim Krčmář2fa6e1e2017-06-04 14:43:52 +02006818 if (kvm_request_pending(vcpu)) {
Avi Kivitya8eeb042010-05-10 12:34:53 +03006819 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
Marcelo Tosatti2e53d632008-02-20 14:47:24 -05006820 kvm_mmu_unload(vcpu);
Avi Kivitya8eeb042010-05-10 12:34:53 +03006821 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
Marcelo Tosatti2f599712008-05-27 12:10:20 -03006822 __kvm_migrate_timers(vcpu);
Marcelo Tosattid8281992012-11-27 23:29:01 -02006823 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
6824 kvm_gen_update_masterclock(vcpu->kvm);
Marcelo Tosatti0061d532013-05-09 20:21:41 -03006825 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
6826 kvm_gen_kvmclock_update(vcpu);
Zachary Amsden34c238a2010-09-18 14:38:14 -10006827 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
6828 r = kvm_guest_time_update(vcpu);
Zachary Amsden8cfdc002010-08-19 22:07:21 -10006829 if (unlikely(r))
6830 goto out;
6831 }
Avi Kivitya8eeb042010-05-10 12:34:53 +03006832 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
Marcelo Tosatti4731d4c2008-09-23 13:18:39 -03006833 kvm_mmu_sync_roots(vcpu);
Avi Kivitya8eeb042010-05-10 12:34:53 +03006834 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
Radim Krčmářa70656b2014-09-18 12:38:36 -04006835 kvm_vcpu_flush_tlb(vcpu);
Avi Kivitya8eeb042010-05-10 12:34:53 +03006836 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
Avi Kivity851ba692009-08-24 11:10:17 +03006837 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
Avi Kivityb93463a2007-10-25 16:52:32 +02006838 r = 0;
6839 goto out;
6840 }
Avi Kivitya8eeb042010-05-10 12:34:53 +03006841 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
Avi Kivity851ba692009-08-24 11:10:17 +03006842 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
Wanpeng Libbeac282017-08-09 22:33:12 -07006843 vcpu->mmio_needed = 0;
Joerg Roedel71c4dfa2008-02-26 16:49:16 +01006844 r = 0;
6845 goto out;
6846 }
Gleb Natapovaf585b92010-10-14 11:22:46 +02006847 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
6848 /* Page is swapped out. Do synthetic halt */
6849 vcpu->arch.apf.halted = true;
6850 r = 1;
6851 goto out;
6852 }
Glauber Costac9aaa892011-07-11 15:28:14 -04006853 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
6854 record_steal_time(vcpu);
Paolo Bonzini64d60672015-05-07 11:36:11 +02006855 if (kvm_check_request(KVM_REQ_SMI, vcpu))
6856 process_smi(vcpu);
Avi Kivity7460fb4a2011-09-20 13:43:14 +03006857 if (kvm_check_request(KVM_REQ_NMI, vcpu))
6858 process_nmi(vcpu);
Gleb Natapovf5132b02011-11-10 14:57:22 +02006859 if (kvm_check_request(KVM_REQ_PMU, vcpu))
Wei Huangc6702c92015-06-19 13:44:45 +02006860 kvm_pmu_handle_event(vcpu);
Gleb Natapovf5132b02011-11-10 14:57:22 +02006861 if (kvm_check_request(KVM_REQ_PMI, vcpu))
Wei Huangc6702c92015-06-19 13:44:45 +02006862 kvm_pmu_deliver_pmi(vcpu);
Steve Rutherford7543a632015-07-29 23:21:41 -07006863 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
6864 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
6865 if (test_bit(vcpu->arch.pending_ioapic_eoi,
Andrey Smetanin63086302015-11-10 15:36:32 +03006866 vcpu->arch.ioapic_handled_vectors)) {
Steve Rutherford7543a632015-07-29 23:21:41 -07006867 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
6868 vcpu->run->eoi.vector =
6869 vcpu->arch.pending_ioapic_eoi;
6870 r = 0;
6871 goto out;
6872 }
6873 }
Yang Zhang3d81bc72013-04-11 19:25:13 +08006874 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
6875 vcpu_scan_ioapic(vcpu);
Tang Chen4256f432014-09-24 15:57:54 +08006876 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
6877 kvm_vcpu_reload_apic_access_page(vcpu);
Andrey Smetanin2ce79182015-07-03 15:01:41 +03006878 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
6879 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
6880 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
6881 r = 0;
6882 goto out;
6883 }
Andrey Smetanine516ceb2015-09-16 12:29:48 +03006884 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
6885 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
6886 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
6887 r = 0;
6888 goto out;
6889 }
Andrey Smetanindb3975712015-11-10 15:36:35 +03006890 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
6891 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
6892 vcpu->run->hyperv = vcpu->arch.hyperv.exit;
6893 r = 0;
6894 goto out;
6895 }
Andrey Smetaninf3b138c2015-12-28 18:27:24 +03006896
6897 /*
6898 * KVM_REQ_HV_STIMER has to be processed after
6899 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
6900 * depend on the guest clock being up-to-date
6901 */
Andrey Smetanin1f4b34f2015-11-30 19:22:21 +03006902 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
6903 kvm_hv_process_stimers(vcpu);
Avi Kivity2f52d582008-01-16 12:49:30 +02006904 }
Avi Kivityb93463a2007-10-25 16:52:32 +02006905
Avi Kivityb463a6f2010-07-20 15:06:17 +03006906 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
Paolo Bonzini0f1e2612016-12-17 16:05:19 +01006907 ++vcpu->stat.req_event;
Jan Kiszka66450a22013-03-13 12:42:34 +01006908 kvm_apic_accept_events(vcpu);
6909 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
6910 r = 1;
6911 goto out;
6912 }
6913
Jan Kiszkab6b8a142014-03-07 20:03:12 +01006914 if (inject_pending_event(vcpu, req_int_win) != 0)
6915 req_immediate_exit = true;
Yuki Shibuya321c5652016-03-24 05:17:03 +00006916 else {
Ladi Prosekcc3d9672017-10-17 16:02:39 +02006917 /* Enable SMI/NMI/IRQ window open exits if needed.
Paolo Bonzinic43203c2016-06-01 22:26:00 +02006918 *
Ladi Prosekcc3d9672017-10-17 16:02:39 +02006919 * SMIs have three cases:
6920 * 1) They can be nested, and then there is nothing to
6921 * do here because RSM will cause a vmexit anyway.
6922 * 2) There is an ISA-specific reason why SMI cannot be
6923 * injected, and the moment when this changes can be
6924 * intercepted.
6925 * 3) Or the SMI can be pending because
6926 * inject_pending_event has completed the injection
6927 * of an IRQ or NMI from the previous vmexit, and
6928 * then we request an immediate exit to inject the
6929 * SMI.
Paolo Bonzinic43203c2016-06-01 22:26:00 +02006930 */
6931 if (vcpu->arch.smi_pending && !is_smm(vcpu))
Ladi Prosekcc3d9672017-10-17 16:02:39 +02006932 if (!kvm_x86_ops->enable_smi_window(vcpu))
6933 req_immediate_exit = true;
Yuki Shibuya321c5652016-03-24 05:17:03 +00006934 if (vcpu->arch.nmi_pending)
6935 kvm_x86_ops->enable_nmi_window(vcpu);
6936 if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
6937 kvm_x86_ops->enable_irq_window(vcpu);
Wanpeng Li664f8e22017-08-24 03:35:09 -07006938 WARN_ON(vcpu->arch.exception.pending);
Yuki Shibuya321c5652016-03-24 05:17:03 +00006939 }
Gleb Natapov6a8b1d12009-05-11 13:35:51 +03006940
Avi Kivity3842d132010-07-27 12:30:24 +03006941 if (kvm_lapic_enabled(vcpu)) {
6942 update_cr8_intercept(vcpu);
6943 kvm_lapic_sync_to_vapic(vcpu);
6944 }
Gleb Natapov95ba8273132009-04-21 17:45:08 +03006945 }
Avi Kivityb93463a2007-10-25 16:52:32 +02006946
Avi Kivityd8368af2012-05-14 18:07:56 +03006947 r = kvm_mmu_reload(vcpu);
6948 if (unlikely(r)) {
Michael S. Tsirkind905c062012-06-24 19:25:00 +03006949 goto cancel_injection;
Avi Kivityd8368af2012-05-14 18:07:56 +03006950 }
6951
Jan Kiszka42dbaa52008-12-15 13:52:10 +01006952 preempt_disable();
6953
6954 kvm_x86_ops->prepare_guest_switch(vcpu);
Paolo Bonzinib95234c2016-12-19 13:57:33 +01006955
6956 /*
6957 * Disable IRQs before setting IN_GUEST_MODE. Posted interrupt
6958 * IPI are then delayed after guest entry, which ensures that they
6959 * result in virtual interrupt delivery.
6960 */
6961 local_irq_disable();
Xiao Guangrong6b7e2d02011-01-12 15:40:31 +08006962 vcpu->mode = IN_GUEST_MODE;
6963
Michael S. Tsirkin01b71912013-11-04 22:36:25 +02006964 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6965
Lan Tianyu0f127d12016-03-13 11:10:29 +08006966 /*
Paolo Bonzinib95234c2016-12-19 13:57:33 +01006967 * 1) We should set ->mode before checking ->requests. Please see
Andrew Jonescde9af62017-04-26 22:32:24 +02006968 * the comment in kvm_vcpu_exiting_guest_mode().
Paolo Bonzinib95234c2016-12-19 13:57:33 +01006969 *
6970 * 2) For APICv, we should set ->mode before checking PIR.ON. This
6971 * pairs with the memory barrier implicit in pi_test_and_set_on
6972 * (see vmx_deliver_posted_interrupt).
6973 *
6974 * 3) This also orders the write to mode from any reads to the page
6975 * tables done while the VCPU is running. Please see the comment
6976 * in kvm_flush_remote_tlbs.
Xiao Guangrong6b7e2d02011-01-12 15:40:31 +08006977 */
Michael S. Tsirkin01b71912013-11-04 22:36:25 +02006978 smp_mb__after_srcu_read_unlock();
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006979
Paolo Bonzinib95234c2016-12-19 13:57:33 +01006980 /*
6981 * This handles the case where a posted interrupt was
6982 * notified with kvm_vcpu_kick.
6983 */
6984 if (kvm_lapic_enabled(vcpu)) {
6985 if (kvm_x86_ops->sync_pir_to_irr && vcpu->arch.apicv_active)
6986 kvm_x86_ops->sync_pir_to_irr(vcpu);
6987 }
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006988
Radim Krčmář2fa6e1e2017-06-04 14:43:52 +02006989 if (vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu)
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006990 || need_resched() || signal_pending(current)) {
Xiao Guangrong6b7e2d02011-01-12 15:40:31 +08006991 vcpu->mode = OUTSIDE_GUEST_MODE;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006992 smp_wmb();
6993 local_irq_enable();
6994 preempt_enable();
Michael S. Tsirkin01b71912013-11-04 22:36:25 +02006995 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006996 r = 1;
Michael S. Tsirkind905c062012-06-24 19:25:00 +03006997 goto cancel_injection;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006998 }
6999
David Matlackfc5b7f32016-03-30 12:24:47 -07007000 kvm_load_guest_xcr0(vcpu);
7001
Paolo Bonzinic43203c2016-06-01 22:26:00 +02007002 if (req_immediate_exit) {
7003 kvm_make_request(KVM_REQ_EVENT, vcpu);
Nadav Har'Eld6185f22011-09-22 13:52:56 +03007004 smp_send_reschedule(vcpu->cpu);
Paolo Bonzinic43203c2016-06-01 22:26:00 +02007005 }
Nadav Har'Eld6185f22011-09-22 13:52:56 +03007006
Paolo Bonzini8b89fe12015-12-10 18:37:32 +01007007 trace_kvm_entry(vcpu->vcpu_id);
7008 wait_lapic_expire(vcpu);
Paolo Bonzini6edaa532016-06-15 15:18:26 +02007009 guest_enter_irqoff();
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007010
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007011 if (unlikely(vcpu->arch.switch_db_regs)) {
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007012 set_debugreg(0, 7);
7013 set_debugreg(vcpu->arch.eff_db[0], 0);
7014 set_debugreg(vcpu->arch.eff_db[1], 1);
7015 set_debugreg(vcpu->arch.eff_db[2], 2);
7016 set_debugreg(vcpu->arch.eff_db[3], 3);
Paolo Bonzinic77fb5f2014-02-21 10:17:24 +01007017 set_debugreg(vcpu->arch.dr6, 6);
Nadav Amitae561ed2015-04-02 03:10:37 +03007018 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007019 }
7020
Avi Kivity851ba692009-08-24 11:10:17 +03007021 kvm_x86_ops->run(vcpu);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007022
Frederic Weisbecker24f1e32c2009-09-09 19:22:48 +02007023 /*
Paolo Bonzinic77fb5f2014-02-21 10:17:24 +01007024 * Do this here before restoring debug registers on the host. And
7025 * since we do this before handling the vmexit, a DR access vmexit
7026 * can (a) read the correct value of the debug registers, (b) set
7027 * KVM_DEBUGREG_WONT_EXIT again.
7028 */
7029 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
Paolo Bonzinic77fb5f2014-02-21 10:17:24 +01007030 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
7031 kvm_x86_ops->sync_dirty_debug_regs(vcpu);
Paolo Bonzini70e4da72016-02-26 12:28:40 +01007032 kvm_update_dr0123(vcpu);
7033 kvm_update_dr6(vcpu);
7034 kvm_update_dr7(vcpu);
7035 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
Paolo Bonzinic77fb5f2014-02-21 10:17:24 +01007036 }
7037
7038 /*
Frederic Weisbecker24f1e32c2009-09-09 19:22:48 +02007039 * If the guest has used debug registers, at least dr7
7040 * will be disabled while returning to the host.
7041 * If we don't have active breakpoints in the host, we don't
7042 * care about the messed up debug address registers. But if
7043 * we have some of them active, restore the old state.
7044 */
Frederic Weisbecker59d8eb52009-11-10 11:03:12 +01007045 if (hw_breakpoint_active())
Frederic Weisbecker24f1e32c2009-09-09 19:22:48 +02007046 hw_breakpoint_restore();
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007047
Haozhong Zhang4ba76532015-10-20 15:39:07 +08007048 vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
Zachary Amsden1d5f0662010-08-19 22:07:30 -10007049
Xiao Guangrong6b7e2d02011-01-12 15:40:31 +08007050 vcpu->mode = OUTSIDE_GUEST_MODE;
Avi Kivityd94e1dc2010-05-03 16:54:48 +03007051 smp_wmb();
Yang Zhanga547c6d2013-04-11 19:25:10 +08007052
David Matlackfc5b7f32016-03-30 12:24:47 -07007053 kvm_put_guest_xcr0(vcpu);
7054
Yang Zhanga547c6d2013-04-11 19:25:10 +08007055 kvm_x86_ops->handle_external_intr(vcpu);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007056
7057 ++vcpu->stat.exits;
7058
Paolo Bonzinif2485b32016-06-15 15:23:11 +02007059 guest_exit_irqoff();
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007060
Paolo Bonzinif2485b32016-06-15 15:23:11 +02007061 local_irq_enable();
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007062 preempt_enable();
7063
Marcelo Tosattif656ce02009-12-23 14:35:25 -02007064 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
Marcelo Tosatti3200f402008-03-29 20:17:59 -03007065
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007066 /*
7067 * Profile KVM exit RIPs:
7068 */
7069 if (unlikely(prof_on == KVM_PROFILING)) {
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03007070 unsigned long rip = kvm_rip_read(vcpu);
7071 profile_hit(KVM_PROFILING, (void *)rip);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007072 }
7073
Zachary Amsdencc578282012-02-03 15:43:50 -02007074 if (unlikely(vcpu->arch.tsc_always_catchup))
7075 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
Avi Kivity298101d2007-11-25 13:41:11 +02007076
Michael S. Tsirkin5cfb1d52012-06-24 19:24:54 +03007077 if (vcpu->arch.apic_attention)
7078 kvm_lapic_sync_from_vapic(vcpu);
Avi Kivityb93463a2007-10-25 16:52:32 +02007079
Brijesh Singh618232e2017-08-17 18:36:57 +02007080 vcpu->arch.gpa_available = false;
Avi Kivity851ba692009-08-24 11:10:17 +03007081 r = kvm_x86_ops->handle_exit(vcpu);
Michael S. Tsirkind905c062012-06-24 19:25:00 +03007082 return r;
7083
7084cancel_injection:
7085 kvm_x86_ops->cancel_injection(vcpu);
Michael S. Tsirkinae7a2a32012-06-24 19:25:07 +03007086 if (unlikely(vcpu->arch.apic_attention))
7087 kvm_lapic_sync_from_vapic(vcpu);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007088out:
Marcelo Tosattid7690172008-09-08 15:23:48 -03007089 return r;
7090}
7091
Paolo Bonzini362c6982015-02-06 12:48:04 +01007092static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
7093{
Feng Wubf9f6ac2015-09-18 22:29:55 +08007094 if (!kvm_arch_vcpu_runnable(vcpu) &&
7095 (!kvm_x86_ops->pre_block || kvm_x86_ops->pre_block(vcpu) == 0)) {
Paolo Bonzini9c8fd1b2015-02-06 12:58:42 +01007096 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7097 kvm_vcpu_block(vcpu);
7098 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
Feng Wubf9f6ac2015-09-18 22:29:55 +08007099
7100 if (kvm_x86_ops->post_block)
7101 kvm_x86_ops->post_block(vcpu);
7102
Paolo Bonzini9c8fd1b2015-02-06 12:58:42 +01007103 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
7104 return 1;
7105 }
Gleb Natapov09cec752009-03-23 15:11:44 +02007106
Paolo Bonzini362c6982015-02-06 12:48:04 +01007107 kvm_apic_accept_events(vcpu);
7108 switch(vcpu->arch.mp_state) {
7109 case KVM_MP_STATE_HALTED:
7110 vcpu->arch.pv.pv_unhalted = false;
7111 vcpu->arch.mp_state =
7112 KVM_MP_STATE_RUNNABLE;
7113 case KVM_MP_STATE_RUNNABLE:
7114 vcpu->arch.apf.halted = false;
7115 break;
7116 case KVM_MP_STATE_INIT_RECEIVED:
7117 break;
7118 default:
7119 return -EINTR;
7120 break;
7121 }
7122 return 1;
7123}
7124
Paolo Bonzini5d9bc642015-10-13 10:18:53 +02007125static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
7126{
Paolo Bonzini0ad3bed2016-12-19 15:23:54 +01007127 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
7128 kvm_x86_ops->check_nested_events(vcpu, false);
7129
Paolo Bonzini5d9bc642015-10-13 10:18:53 +02007130 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
7131 !vcpu->arch.apf.halted);
7132}
7133
Paolo Bonzini362c6982015-02-06 12:48:04 +01007134static int vcpu_run(struct kvm_vcpu *vcpu)
Marcelo Tosattid7690172008-09-08 15:23:48 -03007135{
7136 int r;
Marcelo Tosattif656ce02009-12-23 14:35:25 -02007137 struct kvm *kvm = vcpu->kvm;
Marcelo Tosattid7690172008-09-08 15:23:48 -03007138
Marcelo Tosattif656ce02009-12-23 14:35:25 -02007139 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
Marcelo Tosattid7690172008-09-08 15:23:48 -03007140
Paolo Bonzini362c6982015-02-06 12:48:04 +01007141 for (;;) {
Paolo Bonzini58f800d2015-10-13 21:32:50 +02007142 if (kvm_vcpu_running(vcpu)) {
Avi Kivity851ba692009-08-24 11:10:17 +03007143 r = vcpu_enter_guest(vcpu);
Feng Wubf9f6ac2015-09-18 22:29:55 +08007144 } else {
Paolo Bonzini362c6982015-02-06 12:48:04 +01007145 r = vcpu_block(kvm, vcpu);
Feng Wubf9f6ac2015-09-18 22:29:55 +08007146 }
7147
Gleb Natapov09cec752009-03-23 15:11:44 +02007148 if (r <= 0)
7149 break;
7150
Radim Krčmář72875d8a2017-04-26 22:32:19 +02007151 kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
Gleb Natapov09cec752009-03-23 15:11:44 +02007152 if (kvm_cpu_has_pending_timer(vcpu))
7153 kvm_inject_pending_timer_irqs(vcpu);
7154
Matt Gingell782d4222015-11-16 15:26:00 -08007155 if (dm_request_for_irq_injection(vcpu) &&
7156 kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
Paolo Bonzini4ca7dd82015-07-30 10:32:16 +02007157 r = 0;
7158 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
Gleb Natapov09cec752009-03-23 15:11:44 +02007159 ++vcpu->stat.request_irq_exits;
Paolo Bonzini362c6982015-02-06 12:48:04 +01007160 break;
Gleb Natapov09cec752009-03-23 15:11:44 +02007161 }
Gleb Natapovaf585b92010-10-14 11:22:46 +02007162
7163 kvm_check_async_pf_completion(vcpu);
7164
Gleb Natapov09cec752009-03-23 15:11:44 +02007165 if (signal_pending(current)) {
7166 r = -EINTR;
Avi Kivity851ba692009-08-24 11:10:17 +03007167 vcpu->run->exit_reason = KVM_EXIT_INTR;
Gleb Natapov09cec752009-03-23 15:11:44 +02007168 ++vcpu->stat.signal_exits;
Paolo Bonzini362c6982015-02-06 12:48:04 +01007169 break;
Gleb Natapov09cec752009-03-23 15:11:44 +02007170 }
7171 if (need_resched()) {
Marcelo Tosattif656ce02009-12-23 14:35:25 -02007172 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
Takuya Yoshikawac08ac062013-12-13 15:07:21 +09007173 cond_resched();
Marcelo Tosattif656ce02009-12-23 14:35:25 -02007174 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
Marcelo Tosattid7690172008-09-08 15:23:48 -03007175 }
7176 }
7177
Marcelo Tosattif656ce02009-12-23 14:35:25 -02007178 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007179
7180 return r;
7181}
7182
Gleb Natapov716d51a2012-09-03 15:24:26 +03007183static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
7184{
7185 int r;
7186 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
7187 r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
7188 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
7189 if (r != EMULATE_DONE)
7190 return 0;
7191 return 1;
7192}
7193
7194static int complete_emulated_pio(struct kvm_vcpu *vcpu)
7195{
7196 BUG_ON(!vcpu->arch.pio.count);
7197
7198 return complete_emulated_io(vcpu);
7199}
7200
Avi Kivityf78146b2012-04-18 19:22:47 +03007201/*
7202 * Implements the following, as a state machine:
7203 *
7204 * read:
7205 * for each fragment
Xiao Guangrong87da7e62012-10-24 14:07:59 +08007206 * for each mmio piece in the fragment
7207 * write gpa, len
7208 * exit
7209 * copy data
Avi Kivityf78146b2012-04-18 19:22:47 +03007210 * execute insn
7211 *
7212 * write:
7213 * for each fragment
Xiao Guangrong87da7e62012-10-24 14:07:59 +08007214 * for each mmio piece in the fragment
7215 * write gpa, len
7216 * copy data
7217 * exit
Avi Kivityf78146b2012-04-18 19:22:47 +03007218 */
Gleb Natapov716d51a2012-09-03 15:24:26 +03007219static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
Avi Kivity5287f192010-01-19 14:20:10 +02007220{
7221 struct kvm_run *run = vcpu->run;
Avi Kivityf78146b2012-04-18 19:22:47 +03007222 struct kvm_mmio_fragment *frag;
Xiao Guangrong87da7e62012-10-24 14:07:59 +08007223 unsigned len;
Avi Kivity5287f192010-01-19 14:20:10 +02007224
Gleb Natapov716d51a2012-09-03 15:24:26 +03007225 BUG_ON(!vcpu->mmio_needed);
Avi Kivity5287f192010-01-19 14:20:10 +02007226
Gleb Natapov716d51a2012-09-03 15:24:26 +03007227 /* Complete previous fragment */
Xiao Guangrong87da7e62012-10-24 14:07:59 +08007228 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
7229 len = min(8u, frag->len);
Gleb Natapov716d51a2012-09-03 15:24:26 +03007230 if (!vcpu->mmio_is_write)
Xiao Guangrong87da7e62012-10-24 14:07:59 +08007231 memcpy(frag->data, run->mmio.data, len);
7232
7233 if (frag->len <= 8) {
7234 /* Switch to the next fragment. */
7235 frag++;
7236 vcpu->mmio_cur_fragment++;
7237 } else {
7238 /* Go forward to the next mmio piece. */
7239 frag->data += len;
7240 frag->gpa += len;
7241 frag->len -= len;
7242 }
7243
Andrew Honiga08d3b32014-02-27 19:35:14 +01007244 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
Gleb Natapov716d51a2012-09-03 15:24:26 +03007245 vcpu->mmio_needed = 0;
Paolo Bonzini0912c972013-08-27 15:41:43 +02007246
7247 /* FIXME: return into emulator if single-stepping. */
Avi Kivitycef4dea2010-01-20 12:01:20 +02007248 if (vcpu->mmio_is_write)
Gleb Natapov716d51a2012-09-03 15:24:26 +03007249 return 1;
7250 vcpu->mmio_read_completed = 1;
7251 return complete_emulated_io(vcpu);
Avi Kivity5287f192010-01-19 14:20:10 +02007252 }
Xiao Guangrong87da7e62012-10-24 14:07:59 +08007253
Gleb Natapov716d51a2012-09-03 15:24:26 +03007254 run->exit_reason = KVM_EXIT_MMIO;
7255 run->mmio.phys_addr = frag->gpa;
7256 if (vcpu->mmio_is_write)
Xiao Guangrong87da7e62012-10-24 14:07:59 +08007257 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
7258 run->mmio.len = min(8u, frag->len);
Gleb Natapov716d51a2012-09-03 15:24:26 +03007259 run->mmio.is_write = vcpu->mmio_is_write;
7260 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
7261 return 0;
Avi Kivity5287f192010-01-19 14:20:10 +02007262}
7263
Gleb Natapov716d51a2012-09-03 15:24:26 +03007264
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007265int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
7266{
7267 int r;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007268
Jan H. Schönherr20b70352017-11-24 22:39:01 +01007269 kvm_sigset_activate(vcpu);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007270
Peter Xu5663d8f2017-12-12 17:15:02 +01007271 kvm_load_guest_fpu(vcpu);
7272
Avi Kivityac9f6dc2008-07-06 15:48:31 +03007273 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
Jan H. Schönherr2f173d22017-09-06 18:34:06 +02007274 if (kvm_run->immediate_exit) {
7275 r = -EINTR;
7276 goto out;
7277 }
Avi Kivityac9f6dc2008-07-06 15:48:31 +03007278 kvm_vcpu_block(vcpu);
Jan Kiszka66450a22013-03-13 12:42:34 +01007279 kvm_apic_accept_events(vcpu);
Radim Krčmář72875d8a2017-04-26 22:32:19 +02007280 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
Avi Kivityac9f6dc2008-07-06 15:48:31 +03007281 r = -EAGAIN;
Jan H. Schönherra0595002017-09-06 00:27:19 +02007282 if (signal_pending(current)) {
7283 r = -EINTR;
7284 vcpu->run->exit_reason = KVM_EXIT_INTR;
7285 ++vcpu->stat.signal_exits;
7286 }
Avi Kivityac9f6dc2008-07-06 15:48:31 +03007287 goto out;
7288 }
7289
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007290 /* re-sync apic's tpr */
Paolo Bonzini35754c92015-07-29 12:05:37 +02007291 if (!lapic_in_kernel(vcpu)) {
Andre Przywaraeea1cff2010-12-21 11:12:00 +01007292 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
7293 r = -EINVAL;
7294 goto out;
7295 }
7296 }
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007297
Gleb Natapov716d51a2012-09-03 15:24:26 +03007298 if (unlikely(vcpu->arch.complete_userspace_io)) {
7299 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
7300 vcpu->arch.complete_userspace_io = NULL;
7301 r = cui(vcpu);
7302 if (r <= 0)
Peter Xu5663d8f2017-12-12 17:15:02 +01007303 goto out;
Gleb Natapov716d51a2012-09-03 15:24:26 +03007304 } else
7305 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
Avi Kivity5287f192010-01-19 14:20:10 +02007306
Paolo Bonzini460df4c2017-02-08 11:50:15 +01007307 if (kvm_run->immediate_exit)
7308 r = -EINTR;
7309 else
7310 r = vcpu_run(vcpu);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007311
7312out:
Peter Xu5663d8f2017-12-12 17:15:02 +01007313 kvm_put_guest_fpu(vcpu);
Marcelo Tosattif1d86e42010-05-03 23:04:27 -03007314 post_kvm_run_save(vcpu);
Jan H. Schönherr20b70352017-11-24 22:39:01 +01007315 kvm_sigset_deactivate(vcpu);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007316
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007317 return r;
7318}
7319
7320int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
7321{
Gleb Natapov7ae441e2011-03-31 12:06:41 +02007322 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
7323 /*
7324 * We are here if userspace calls get_regs() in the middle of
7325 * instruction emulation. Registers state needs to be copied
Guo Chao4a969982012-06-28 15:17:27 +08007326 * back from emulation context to vcpu. Userspace shouldn't do
Gleb Natapov7ae441e2011-03-31 12:06:41 +02007327 * that usually, but some bad designed PV devices (vmware
7328 * backdoor interface) need this to work
7329 */
Avi Kivitydd856ef2012-08-27 23:46:17 +03007330 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
Gleb Natapov7ae441e2011-03-31 12:06:41 +02007331 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7332 }
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03007333 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
7334 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
7335 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
7336 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
7337 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
7338 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
7339 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
7340 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007341#ifdef CONFIG_X86_64
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03007342 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
7343 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
7344 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
7345 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
7346 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
7347 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
7348 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
7349 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007350#endif
7351
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03007352 regs->rip = kvm_rip_read(vcpu);
Jan Kiszka91586a32009-10-05 13:07:21 +02007353 regs->rflags = kvm_get_rflags(vcpu);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007354
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007355 return 0;
7356}
7357
7358int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
7359{
Gleb Natapov7ae441e2011-03-31 12:06:41 +02007360 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
7361 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7362
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03007363 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
7364 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
7365 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
7366 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
7367 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
7368 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
7369 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
7370 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007371#ifdef CONFIG_X86_64
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03007372 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
7373 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
7374 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
7375 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
7376 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
7377 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
7378 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
7379 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007380#endif
7381
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03007382 kvm_rip_write(vcpu, regs->rip);
Wanpeng Lid73235d2017-12-07 00:30:08 -08007383 kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007384
Jan Kiszkab4f14ab2008-04-30 17:59:04 +02007385 vcpu->arch.exception.pending = false;
7386
Avi Kivity3842d132010-07-27 12:30:24 +03007387 kvm_make_request(KVM_REQ_EVENT, vcpu);
7388
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007389 return 0;
7390}
7391
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007392void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
7393{
7394 struct kvm_segment cs;
7395
Guillaume Thouvenin3e6e0aa2008-05-27 10:18:46 +02007396 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007397 *db = cs.db;
7398 *l = cs.l;
7399}
7400EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
7401
7402int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
7403 struct kvm_sregs *sregs)
7404{
Gleb Natapov89a27f42010-02-16 10:51:48 +02007405 struct desc_ptr dt;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007406
Guillaume Thouvenin3e6e0aa2008-05-27 10:18:46 +02007407 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7408 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7409 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7410 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7411 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7412 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007413
Guillaume Thouvenin3e6e0aa2008-05-27 10:18:46 +02007414 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7415 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007416
7417 kvm_x86_ops->get_idt(vcpu, &dt);
Gleb Natapov89a27f42010-02-16 10:51:48 +02007418 sregs->idt.limit = dt.size;
7419 sregs->idt.base = dt.address;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007420 kvm_x86_ops->get_gdt(vcpu, &dt);
Gleb Natapov89a27f42010-02-16 10:51:48 +02007421 sregs->gdt.limit = dt.size;
7422 sregs->gdt.base = dt.address;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007423
Avi Kivity4d4ec082009-12-29 18:07:30 +02007424 sregs->cr0 = kvm_read_cr0(vcpu);
Zhang Xiantaoad312c72007-12-13 23:50:52 +08007425 sregs->cr2 = vcpu->arch.cr2;
Avi Kivity9f8fe502010-12-05 17:30:00 +02007426 sregs->cr3 = kvm_read_cr3(vcpu);
Avi Kivityfc78f512009-12-07 12:16:48 +02007427 sregs->cr4 = kvm_read_cr4(vcpu);
Avi Kivity2d3ad1f2008-02-24 11:20:43 +02007428 sregs->cr8 = kvm_get_cr8(vcpu);
Avi Kivityf6801df2010-01-21 15:31:50 +02007429 sregs->efer = vcpu->arch.efer;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007430 sregs->apic_base = kvm_get_apic_base(vcpu);
7431
Gleb Natapov923c61b2009-05-11 13:35:48 +03007432 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007433
Gleb Natapov36752c92009-05-11 13:35:53 +03007434 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
Gleb Natapov14d0bc12009-04-21 17:45:11 +03007435 set_bit(vcpu->arch.interrupt.nr,
7436 (unsigned long *)sregs->interrupt_bitmap);
Gleb Natapov16d7a192009-04-21 17:45:10 +03007437
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007438 return 0;
7439}
7440
Marcelo Tosatti62d9f0d2008-04-11 13:24:45 -03007441int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
7442 struct kvm_mp_state *mp_state)
7443{
Jan Kiszka66450a22013-03-13 12:42:34 +01007444 kvm_apic_accept_events(vcpu);
Srivatsa Vaddagiri6aef2662013-08-26 14:18:34 +05307445 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
7446 vcpu->arch.pv.pv_unhalted)
7447 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
7448 else
7449 mp_state->mp_state = vcpu->arch.mp_state;
7450
Marcelo Tosatti62d9f0d2008-04-11 13:24:45 -03007451 return 0;
7452}
7453
7454int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
7455 struct kvm_mp_state *mp_state)
7456{
Paolo Bonzinibce87cc2016-01-08 13:48:51 +01007457 if (!lapic_in_kernel(vcpu) &&
Jan Kiszka66450a22013-03-13 12:42:34 +01007458 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
7459 return -EINVAL;
7460
David Hildenbrand28bf2882017-03-23 11:46:03 +01007461 /* INITs are latched while in SMM */
7462 if ((is_smm(vcpu) || vcpu->arch.smi_pending) &&
7463 (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
7464 mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
7465 return -EINVAL;
7466
Jan Kiszka66450a22013-03-13 12:42:34 +01007467 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
7468 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
7469 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
7470 } else
7471 vcpu->arch.mp_state = mp_state->mp_state;
Avi Kivity3842d132010-07-27 12:30:24 +03007472 kvm_make_request(KVM_REQ_EVENT, vcpu);
Marcelo Tosatti62d9f0d2008-04-11 13:24:45 -03007473 return 0;
7474}
7475
Kevin Wolf7f3d35f2012-02-08 14:34:38 +01007476int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
7477 int reason, bool has_error_code, u32 error_code)
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007478{
Takuya Yoshikawa9d741912011-05-29 21:53:48 +09007479 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
Mohammed Gamal8ec47222010-08-16 00:47:01 +03007480 int ret;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007481
Mohammed Gamal8ec47222010-08-16 00:47:01 +03007482 init_emulate_ctxt(vcpu);
Izik Eidus37817f22008-03-24 23:14:53 +02007483
Kevin Wolf7f3d35f2012-02-08 14:34:38 +01007484 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
Takuya Yoshikawa9d741912011-05-29 21:53:48 +09007485 has_error_code, error_code);
Izik Eidus37817f22008-03-24 23:14:53 +02007486
Gleb Natapovc6975182010-02-18 12:15:01 +02007487 if (ret)
Gleb Natapov19d04432010-04-15 12:29:50 +03007488 return EMULATE_FAIL;
Gleb Natapovc6975182010-02-18 12:15:01 +02007489
Takuya Yoshikawa9d741912011-05-29 21:53:48 +09007490 kvm_rip_write(vcpu, ctxt->eip);
7491 kvm_set_rflags(vcpu, ctxt->eflags);
Avi Kivity3842d132010-07-27 12:30:24 +03007492 kvm_make_request(KVM_REQ_EVENT, vcpu);
Gleb Natapov19d04432010-04-15 12:29:50 +03007493 return EMULATE_DONE;
Izik Eidus37817f22008-03-24 23:14:53 +02007494}
7495EXPORT_SYMBOL_GPL(kvm_task_switch);
7496
Lan Tianyuf2981032017-12-14 03:01:52 -05007497int kvm_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
7498{
7499 if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG_BIT)) {
7500 /*
7501 * When EFER.LME and CR0.PG are set, the processor is in
7502 * 64-bit mode (though maybe in a 32-bit code segment).
7503 * CR4.PAE and EFER.LMA must be set.
7504 */
7505 if (!(sregs->cr4 & X86_CR4_PAE_BIT)
7506 || !(sregs->efer & EFER_LMA))
7507 return -EINVAL;
7508 } else {
7509 /*
7510 * Not in 64-bit mode: EFER.LMA is clear and the code
7511 * segment cannot be 64-bit.
7512 */
7513 if (sregs->efer & EFER_LMA || sregs->cs.l)
7514 return -EINVAL;
7515 }
7516
7517 return 0;
7518}
7519
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007520int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
7521 struct kvm_sregs *sregs)
7522{
Jan Kiszka58cb6282014-01-24 16:48:44 +01007523 struct msr_data apic_base_msr;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007524 int mmu_reset_needed = 0;
Xiao Guangrong63f42e02011-01-12 15:39:18 +08007525 int pending_vec, max_bits, idx;
Gleb Natapov89a27f42010-02-16 10:51:48 +02007526 struct desc_ptr dt;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007527
Radim Krčmářd6321d42017-08-05 00:12:49 +02007528 if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
7529 (sregs->cr4 & X86_CR4_OSXSAVE))
Petr Matousek6d1068b2012-11-06 19:24:07 +01007530 return -EINVAL;
7531
Lan Tianyuf2981032017-12-14 03:01:52 -05007532 if (kvm_valid_sregs(vcpu, sregs))
7533 return -EINVAL;
7534
Jim Mattsond3802282017-08-10 10:14:13 -07007535 apic_base_msr.data = sregs->apic_base;
7536 apic_base_msr.host_initiated = true;
7537 if (kvm_set_apic_base(vcpu, &apic_base_msr))
Gleb Natapov89a27f42010-02-16 10:51:48 +02007538 return -EINVAL;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007539
Gleb Natapov89a27f42010-02-16 10:51:48 +02007540 dt.size = sregs->idt.limit;
7541 dt.address = sregs->idt.base;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007542 kvm_x86_ops->set_idt(vcpu, &dt);
7543 dt.size = sregs->gdt.limit;
Zhang Xiantaoad312c72007-12-13 23:50:52 +08007544 dt.address = sregs->gdt.base;
Avi Kivity9f8fe502010-12-05 17:30:00 +02007545 kvm_x86_ops->set_gdt(vcpu, &dt);
Jan Kiszkadc7e7952009-07-01 20:52:03 +02007546
Avi Kivityaff48ba2010-12-05 18:56:11 +02007547 vcpu->arch.cr2 = sregs->cr2;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007548 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
Avi Kivity2d3ad1f2008-02-24 11:20:43 +02007549 vcpu->arch.cr3 = sregs->cr3;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007550 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
Avi Kivityf6801df2010-01-21 15:31:50 +02007551
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007552 kvm_set_cr8(vcpu, sregs->cr8);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007553
7554 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
Avi Kivity4d4ec082009-12-29 18:07:30 +02007555 kvm_x86_ops->set_efer(vcpu, sregs->efer);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007556
Zhang Xiantaoad312c72007-12-13 23:50:52 +08007557 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007558 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
Paul Knowlesd7306162008-02-06 11:02:35 +00007559 vcpu->arch.cr0 = sregs->cr0;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007560
Avi Kivityfc78f512009-12-07 12:16:48 +02007561 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007562 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
Huaitong Hanb9baba82016-03-22 16:51:21 +08007563 if (sregs->cr4 & (X86_CR4_OSXSAVE | X86_CR4_PKE))
Avi Kivity00b27a32011-11-23 16:30:32 +02007564 kvm_update_cpuid(vcpu);
Xiao Guangrong63f42e02011-01-12 15:39:18 +08007565
7566 idx = srcu_read_lock(&vcpu->kvm->srcu);
Marcelo Tosatti7c93be42009-10-26 16:48:33 -02007567 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
Avi Kivity9f8fe502010-12-05 17:30:00 +02007568 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
Marcelo Tosatti7c93be42009-10-26 16:48:33 -02007569 mmu_reset_needed = 1;
7570 }
Xiao Guangrong63f42e02011-01-12 15:39:18 +08007571 srcu_read_unlock(&vcpu->kvm->srcu, idx);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007572
7573 if (mmu_reset_needed)
7574 kvm_mmu_reset_context(vcpu);
7575
Michael S. Tsirkina50abc32012-09-05 20:00:52 +03007576 max_bits = KVM_NR_INTERRUPTS;
Gleb Natapov923c61b2009-05-11 13:35:48 +03007577 pending_vec = find_first_bit(
7578 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
7579 if (pending_vec < max_bits) {
Gleb Natapov66fd3f72009-05-11 13:35:50 +03007580 kvm_queue_interrupt(vcpu, pending_vec, false);
Gleb Natapov923c61b2009-05-11 13:35:48 +03007581 pr_debug("Set back pending irq %d\n", pending_vec);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007582 }
7583
Guillaume Thouvenin3e6e0aa2008-05-27 10:18:46 +02007584 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7585 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7586 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7587 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7588 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7589 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007590
Guillaume Thouvenin3e6e0aa2008-05-27 10:18:46 +02007591 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7592 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007593
Mikhail Ershov5f0269f2009-08-03 14:58:25 +03007594 update_cr8_intercept(vcpu);
7595
Marcelo Tosatti9c3e4aa2008-09-10 16:40:55 -03007596 /* Older userspace won't unhalt the vcpu on reset. */
Gleb Natapovc5af89b2009-06-09 15:56:26 +03007597 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
Marcelo Tosatti9c3e4aa2008-09-10 16:40:55 -03007598 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
Avi Kivity3eeb3282010-01-21 15:31:48 +02007599 !is_protmode(vcpu))
Marcelo Tosatti9c3e4aa2008-09-10 16:40:55 -03007600 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7601
Avi Kivity3842d132010-07-27 12:30:24 +03007602 kvm_make_request(KVM_REQ_EVENT, vcpu);
7603
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007604 return 0;
7605}
7606
Jan Kiszkad0bfb942008-12-15 13:52:10 +01007607int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
7608 struct kvm_guest_debug *dbg)
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007609{
Jan Kiszka355be0b2009-10-03 00:31:21 +02007610 unsigned long rflags;
Jan Kiszkaae675ef2008-12-15 13:52:10 +01007611 int i, r;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007612
Jan Kiszka4f926bf22009-10-30 12:46:59 +01007613 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
7614 r = -EBUSY;
7615 if (vcpu->arch.exception.pending)
Avi Kivity2122ff52010-05-13 11:25:04 +03007616 goto out;
Jan Kiszka4f926bf22009-10-30 12:46:59 +01007617 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
7618 kvm_queue_exception(vcpu, DB_VECTOR);
7619 else
7620 kvm_queue_exception(vcpu, BP_VECTOR);
7621 }
7622
Jan Kiszka91586a32009-10-05 13:07:21 +02007623 /*
7624 * Read rflags as long as potentially injected trace flags are still
7625 * filtered out.
7626 */
7627 rflags = kvm_get_rflags(vcpu);
Jan Kiszka355be0b2009-10-03 00:31:21 +02007628
7629 vcpu->guest_debug = dbg->control;
7630 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
7631 vcpu->guest_debug = 0;
7632
7633 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
Jan Kiszkaae675ef2008-12-15 13:52:10 +01007634 for (i = 0; i < KVM_NR_DB_REGS; ++i)
7635 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
Jan Kiszkac8639012012-09-21 05:42:55 +02007636 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
Jan Kiszkaae675ef2008-12-15 13:52:10 +01007637 } else {
7638 for (i = 0; i < KVM_NR_DB_REGS; i++)
7639 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
Jan Kiszkaae675ef2008-12-15 13:52:10 +01007640 }
Jan Kiszkac8639012012-09-21 05:42:55 +02007641 kvm_update_dr7(vcpu);
Jan Kiszkaae675ef2008-12-15 13:52:10 +01007642
Jan Kiszkaf92653e2010-02-23 17:47:55 +01007643 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7644 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
7645 get_segment_base(vcpu, VCPU_SREG_CS);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007646
Jan Kiszka91586a32009-10-05 13:07:21 +02007647 /*
7648 * Trigger an rflags update that will inject or remove the trace
7649 * flags.
7650 */
7651 kvm_set_rflags(vcpu, rflags);
Jan Kiszkad0bfb942008-12-15 13:52:10 +01007652
Paolo Bonzinia96036b2015-11-10 11:55:36 +01007653 kvm_x86_ops->update_bp_intercept(vcpu);
Jan Kiszka355be0b2009-10-03 00:31:21 +02007654
Jan Kiszka4f926bf22009-10-30 12:46:59 +01007655 r = 0;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007656
Avi Kivity2122ff52010-05-13 11:25:04 +03007657out:
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05007658
7659 return r;
7660}
7661
7662/*
Zhang Xiantao8b006792007-11-16 13:05:55 +08007663 * Translate a guest virtual address to a guest physical address.
7664 */
7665int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
7666 struct kvm_translation *tr)
7667{
7668 unsigned long vaddr = tr->linear_address;
7669 gpa_t gpa;
Marcelo Tosattif656ce02009-12-23 14:35:25 -02007670 int idx;
Zhang Xiantao8b006792007-11-16 13:05:55 +08007671
Marcelo Tosattif656ce02009-12-23 14:35:25 -02007672 idx = srcu_read_lock(&vcpu->kvm->srcu);
Gleb Natapov1871c602010-02-10 14:21:32 +02007673 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
Marcelo Tosattif656ce02009-12-23 14:35:25 -02007674 srcu_read_unlock(&vcpu->kvm->srcu, idx);
Zhang Xiantao8b006792007-11-16 13:05:55 +08007675 tr->physical_address = gpa;
7676 tr->valid = gpa != UNMAPPED_GVA;
7677 tr->writeable = 1;
7678 tr->usermode = 0;
Zhang Xiantao8b006792007-11-16 13:05:55 +08007679
7680 return 0;
7681}
7682
Hollis Blanchardd0752062007-10-31 17:24:25 -05007683int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7684{
Ingo Molnarc47ada32015-04-30 17:15:32 +02007685 struct fxregs_state *fxsave =
Ingo Molnar7366ed72015-04-27 04:19:39 +02007686 &vcpu->arch.guest_fpu.state.fxsave;
Hollis Blanchardd0752062007-10-31 17:24:25 -05007687
Hollis Blanchardd0752062007-10-31 17:24:25 -05007688 memcpy(fpu->fpr, fxsave->st_space, 128);
7689 fpu->fcw = fxsave->cwd;
7690 fpu->fsw = fxsave->swd;
7691 fpu->ftwx = fxsave->twd;
7692 fpu->last_opcode = fxsave->fop;
7693 fpu->last_ip = fxsave->rip;
7694 fpu->last_dp = fxsave->rdp;
7695 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
7696
Hollis Blanchardd0752062007-10-31 17:24:25 -05007697 return 0;
7698}
7699
7700int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7701{
Ingo Molnarc47ada32015-04-30 17:15:32 +02007702 struct fxregs_state *fxsave =
Ingo Molnar7366ed72015-04-27 04:19:39 +02007703 &vcpu->arch.guest_fpu.state.fxsave;
Hollis Blanchardd0752062007-10-31 17:24:25 -05007704
Hollis Blanchardd0752062007-10-31 17:24:25 -05007705 memcpy(fxsave->st_space, fpu->fpr, 128);
7706 fxsave->cwd = fpu->fcw;
7707 fxsave->swd = fpu->fsw;
7708 fxsave->twd = fpu->ftwx;
7709 fxsave->fop = fpu->last_opcode;
7710 fxsave->rip = fpu->last_ip;
7711 fxsave->rdp = fpu->last_dp;
7712 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
7713
Hollis Blanchardd0752062007-10-31 17:24:25 -05007714 return 0;
7715}
7716
Ingo Molnar0ee6a512015-04-27 06:58:22 +02007717static void fx_init(struct kvm_vcpu *vcpu)
Hollis Blanchardd0752062007-10-31 17:24:25 -05007718{
Ingo Molnarbf935b02015-04-30 10:23:42 +02007719 fpstate_init(&vcpu->arch.guest_fpu.state);
Borislav Petkov782511b2016-04-04 22:25:03 +02007720 if (boot_cpu_has(X86_FEATURE_XSAVES))
Ingo Molnar7366ed72015-04-27 04:19:39 +02007721 vcpu->arch.guest_fpu.state.xsave.header.xcomp_bv =
Paolo Bonzinidf1daba2014-11-21 19:05:07 +01007722 host_xcr0 | XSTATE_COMPACTION_ENABLED;
Hollis Blanchardd0752062007-10-31 17:24:25 -05007723
Dexuan Cui2acf9232010-06-10 11:27:12 +08007724 /*
7725 * Ensure guest xcr0 is valid for loading
7726 */
Dave Hansend91cab72015-09-02 16:31:26 -07007727 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
Dexuan Cui2acf9232010-06-10 11:27:12 +08007728
Zhang Xiantaoad312c72007-12-13 23:50:52 +08007729 vcpu->arch.cr0 |= X86_CR0_ET;
Hollis Blanchardd0752062007-10-31 17:24:25 -05007730}
Hollis Blanchardd0752062007-10-31 17:24:25 -05007731
Rik van Rielf775b132017-11-14 16:54:23 -05007732/* Swap (qemu) user FPU context for the guest FPU context. */
Hollis Blanchardd0752062007-10-31 17:24:25 -05007733void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
7734{
Rik van Rielf775b132017-11-14 16:54:23 -05007735 preempt_disable();
7736 copy_fpregs_to_fpstate(&vcpu->arch.user_fpu);
Paolo Bonzini38cfd5e2017-08-23 23:16:29 +02007737 /* PKRU is separately restored in kvm_x86_ops->run. */
7738 __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu.state,
7739 ~XFEATURE_MASK_PKRU);
Rik van Rielf775b132017-11-14 16:54:23 -05007740 preempt_enable();
Avi Kivity0c048512010-01-21 15:31:52 +02007741 trace_kvm_fpu(1);
Hollis Blanchardd0752062007-10-31 17:24:25 -05007742}
Hollis Blanchardd0752062007-10-31 17:24:25 -05007743
Rik van Rielf775b132017-11-14 16:54:23 -05007744/* When vcpu_run ends, restore user space FPU context. */
Hollis Blanchardd0752062007-10-31 17:24:25 -05007745void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
7746{
Rik van Rielf775b132017-11-14 16:54:23 -05007747 preempt_disable();
Ingo Molnar4f836342015-04-27 02:53:16 +02007748 copy_fpregs_to_fpstate(&vcpu->arch.guest_fpu);
Rik van Rielf775b132017-11-14 16:54:23 -05007749 copy_kernel_to_fpregs(&vcpu->arch.user_fpu.state);
7750 preempt_enable();
Avi Kivityf096ed82007-11-18 13:54:33 +02007751 ++vcpu->stat.fpu_reload;
Avi Kivity0c048512010-01-21 15:31:52 +02007752 trace_kvm_fpu(0);
Hollis Blanchardd0752062007-10-31 17:24:25 -05007753}
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08007754
7755void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
7756{
Ido Yarivbd768e12016-10-21 12:39:57 -04007757 void *wbinvd_dirty_mask = vcpu->arch.wbinvd_dirty_mask;
7758
Glauber Costa12f9a482011-02-01 14:16:40 -05007759 kvmclock_reset(vcpu);
Joerg Roedel7f1ea202009-02-25 16:08:31 +01007760
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08007761 kvm_x86_ops->vcpu_free(vcpu);
Ido Yarivbd768e12016-10-21 12:39:57 -04007762 free_cpumask_var(wbinvd_dirty_mask);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08007763}
7764
7765struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
7766 unsigned int id)
7767{
Liang Lic447e762015-05-21 04:41:25 +08007768 struct kvm_vcpu *vcpu;
7769
Zachary Amsden6755bae2010-08-19 22:07:22 -10007770 if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
7771 printk_once(KERN_WARNING
7772 "kvm: SMP vm created on host with unstable TSC; "
7773 "guest TSC will not be reliable\n");
Liang Lic447e762015-05-21 04:41:25 +08007774
7775 vcpu = kvm_x86_ops->vcpu_create(kvm, id);
7776
Liang Lic447e762015-05-21 04:41:25 +08007777 return vcpu;
Avi Kivity26e52152007-11-20 15:30:24 +02007778}
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08007779
Avi Kivity26e52152007-11-20 15:30:24 +02007780int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
7781{
7782 int r;
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08007783
Xiao Guangrong19efffa2015-06-15 16:55:31 +08007784 kvm_vcpu_mtrr_init(vcpu);
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +03007785 r = vcpu_load(vcpu);
7786 if (r)
7787 return r;
Nadav Amitd28bc9d2015-04-13 14:34:08 +03007788 kvm_vcpu_reset(vcpu, false);
Paolo Bonzini8a3c1a332013-10-02 16:56:13 +02007789 kvm_mmu_setup(vcpu);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08007790 vcpu_put(vcpu);
Avi Kivity26e52152007-11-20 15:30:24 +02007791 return r;
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08007792}
7793
Dominik Dingel31928aa2014-12-04 15:47:07 +01007794void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
Marcelo Tosatti42897d82012-11-27 23:29:02 -02007795{
Will Auld8fe8ab42012-11-29 12:42:12 -08007796 struct msr_data msr;
Andrew Jones332967a2014-02-28 12:52:55 +01007797 struct kvm *kvm = vcpu->kvm;
Marcelo Tosatti42897d82012-11-27 23:29:02 -02007798
Roman Kagand3457c82017-07-14 17:13:20 +03007799 kvm_hv_vcpu_postcreate(vcpu);
7800
Dominik Dingel31928aa2014-12-04 15:47:07 +01007801 if (vcpu_load(vcpu))
7802 return;
Will Auld8fe8ab42012-11-29 12:42:12 -08007803 msr.data = 0x0;
7804 msr.index = MSR_IA32_TSC;
7805 msr.host_initiated = true;
7806 kvm_write_tsc(vcpu, &msr);
Marcelo Tosatti42897d82012-11-27 23:29:02 -02007807 vcpu_put(vcpu);
7808
Marcelo Tosatti630994b2015-05-12 22:42:04 -03007809 if (!kvmclock_periodic_sync)
7810 return;
7811
Andrew Jones332967a2014-02-28 12:52:55 +01007812 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
7813 KVMCLOCK_SYNC_PERIOD);
Marcelo Tosatti42897d82012-11-27 23:29:02 -02007814}
7815
Hollis Blanchardd40ccc62007-11-19 14:04:43 -06007816void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08007817{
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +03007818 int r;
Gleb Natapov344d9582010-10-14 11:22:50 +02007819 vcpu->arch.apf.msr_val = 0;
7820
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +03007821 r = vcpu_load(vcpu);
7822 BUG_ON(r);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08007823 kvm_mmu_unload(vcpu);
7824 vcpu_put(vcpu);
7825
7826 kvm_x86_ops->vcpu_free(vcpu);
7827}
7828
Nadav Amitd28bc9d2015-04-13 14:34:08 +03007829void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08007830{
Paolo Bonzinie69fab52015-06-04 10:44:44 +02007831 vcpu->arch.hflags = 0;
7832
Paolo Bonzinic43203c2016-06-01 22:26:00 +02007833 vcpu->arch.smi_pending = 0;
Avi Kivity7460fb4a2011-09-20 13:43:14 +03007834 atomic_set(&vcpu->arch.nmi_queued, 0);
7835 vcpu->arch.nmi_pending = 0;
Jan Kiszka448fa4a2008-09-26 09:30:48 +02007836 vcpu->arch.nmi_injected = false;
Nadav Amit5f7552d2014-06-30 12:03:02 +03007837 kvm_clear_interrupt_queue(vcpu);
7838 kvm_clear_exception_queue(vcpu);
Wanpeng Li664f8e22017-08-24 03:35:09 -07007839 vcpu->arch.exception.pending = false;
Jan Kiszka448fa4a2008-09-26 09:30:48 +02007840
Jan Kiszka42dbaa52008-12-15 13:52:10 +01007841 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
Nadav Amitae561ed2015-04-02 03:10:37 +03007842 kvm_update_dr0123(vcpu);
Nadav Amit6f43ed02014-07-15 17:37:46 +03007843 vcpu->arch.dr6 = DR6_INIT;
Jan Kiszka73aaf249e2014-01-04 18:47:16 +01007844 kvm_update_dr6(vcpu);
Jan Kiszka42dbaa52008-12-15 13:52:10 +01007845 vcpu->arch.dr7 = DR7_FIXED_1;
Jan Kiszkac8639012012-09-21 05:42:55 +02007846 kvm_update_dr7(vcpu);
Jan Kiszka42dbaa52008-12-15 13:52:10 +01007847
Nadav Amit11190222015-04-02 03:10:38 +03007848 vcpu->arch.cr2 = 0;
7849
Avi Kivity3842d132010-07-27 12:30:24 +03007850 kvm_make_request(KVM_REQ_EVENT, vcpu);
Gleb Natapov344d9582010-10-14 11:22:50 +02007851 vcpu->arch.apf.msr_val = 0;
Glauber Costac9aaa892011-07-11 15:28:14 -04007852 vcpu->arch.st.msr_val = 0;
Avi Kivity3842d132010-07-27 12:30:24 +03007853
Glauber Costa12f9a482011-02-01 14:16:40 -05007854 kvmclock_reset(vcpu);
7855
Gleb Natapovaf585b92010-10-14 11:22:46 +02007856 kvm_clear_async_pf_completion_queue(vcpu);
7857 kvm_async_pf_hash_reset(vcpu);
7858 vcpu->arch.apf.halted = false;
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08007859
Wanpeng Lia554d202017-10-11 05:10:19 -07007860 if (kvm_mpx_supported()) {
7861 void *mpx_state_buffer;
7862
7863 /*
7864 * To avoid have the INIT path from kvm_apic_has_events() that be
7865 * called with loaded FPU and does not let userspace fix the state.
7866 */
Rik van Rielf775b132017-11-14 16:54:23 -05007867 if (init_event)
7868 kvm_put_guest_fpu(vcpu);
Wanpeng Lia554d202017-10-11 05:10:19 -07007869 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu.state.xsave,
7870 XFEATURE_MASK_BNDREGS);
7871 if (mpx_state_buffer)
7872 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
7873 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu.state.xsave,
7874 XFEATURE_MASK_BNDCSR);
7875 if (mpx_state_buffer)
7876 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
Rik van Rielf775b132017-11-14 16:54:23 -05007877 if (init_event)
7878 kvm_load_guest_fpu(vcpu);
Wanpeng Lia554d202017-10-11 05:10:19 -07007879 }
7880
Paolo Bonzini64d60672015-05-07 11:36:11 +02007881 if (!init_event) {
Nadav Amitd28bc9d2015-04-13 14:34:08 +03007882 kvm_pmu_reset(vcpu);
Paolo Bonzini64d60672015-05-07 11:36:11 +02007883 vcpu->arch.smbase = 0x30000;
Kyle Hueydb2336a2017-03-20 01:16:28 -07007884
7885 vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
7886 vcpu->arch.msr_misc_features_enables = 0;
Wanpeng Lia554d202017-10-11 05:10:19 -07007887
7888 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
Paolo Bonzini64d60672015-05-07 11:36:11 +02007889 }
Gleb Natapovf5132b02011-11-10 14:57:22 +02007890
Julian Stecklina66f7b722012-12-05 15:26:19 +01007891 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
7892 vcpu->arch.regs_avail = ~0;
7893 vcpu->arch.regs_dirty = ~0;
7894
Wanpeng Lia554d202017-10-11 05:10:19 -07007895 vcpu->arch.ia32_xss = 0;
7896
Nadav Amitd28bc9d2015-04-13 14:34:08 +03007897 kvm_x86_ops->vcpu_reset(vcpu, init_event);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08007898}
7899
Paolo Bonzini2b4a2732014-11-24 14:35:24 +01007900void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
Jan Kiszka66450a22013-03-13 12:42:34 +01007901{
7902 struct kvm_segment cs;
7903
7904 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7905 cs.selector = vector << 8;
7906 cs.base = vector << 12;
7907 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7908 kvm_rip_write(vcpu, 0);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08007909}
7910
Radim Krčmář13a34e02014-08-28 15:13:03 +02007911int kvm_arch_hardware_enable(void)
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08007912{
Zachary Amsdenca84d1a2010-08-19 22:07:28 -10007913 struct kvm *kvm;
7914 struct kvm_vcpu *vcpu;
7915 int i;
Zachary Amsden0dd6a6e2012-02-03 15:43:56 -02007916 int ret;
7917 u64 local_tsc;
7918 u64 max_tsc = 0;
7919 bool stable, backwards_tsc = false;
Avi Kivity18863bd2009-09-07 11:12:18 +03007920
7921 kvm_shared_msr_cpu_online();
Radim Krčmář13a34e02014-08-28 15:13:03 +02007922 ret = kvm_x86_ops->hardware_enable();
Zachary Amsden0dd6a6e2012-02-03 15:43:56 -02007923 if (ret != 0)
7924 return ret;
7925
Andy Lutomirski4ea16362015-06-25 18:44:07 +02007926 local_tsc = rdtsc();
Zachary Amsden0dd6a6e2012-02-03 15:43:56 -02007927 stable = !check_tsc_unstable();
7928 list_for_each_entry(kvm, &vm_list, vm_list) {
7929 kvm_for_each_vcpu(i, vcpu, kvm) {
7930 if (!stable && vcpu->cpu == smp_processor_id())
Guo Hui Liu105b21b2014-09-12 13:43:19 +08007931 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
Zachary Amsden0dd6a6e2012-02-03 15:43:56 -02007932 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
7933 backwards_tsc = true;
7934 if (vcpu->arch.last_host_tsc > max_tsc)
7935 max_tsc = vcpu->arch.last_host_tsc;
7936 }
7937 }
7938 }
7939
7940 /*
7941 * Sometimes, even reliable TSCs go backwards. This happens on
7942 * platforms that reset TSC during suspend or hibernate actions, but
7943 * maintain synchronization. We must compensate. Fortunately, we can
7944 * detect that condition here, which happens early in CPU bringup,
7945 * before any KVM threads can be running. Unfortunately, we can't
7946 * bring the TSCs fully up to date with real time, as we aren't yet far
7947 * enough into CPU bringup that we know how much real time has actually
Paolo Bonzini108b2492016-09-01 14:21:03 +02007948 * elapsed; our helper function, ktime_get_boot_ns() will be using boot
Zachary Amsden0dd6a6e2012-02-03 15:43:56 -02007949 * variables that haven't been updated yet.
7950 *
7951 * So we simply find the maximum observed TSC above, then record the
7952 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
7953 * the adjustment will be applied. Note that we accumulate
7954 * adjustments, in case multiple suspend cycles happen before some VCPU
7955 * gets a chance to run again. In the event that no KVM threads get a
7956 * chance to run, we will miss the entire elapsed period, as we'll have
7957 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
7958 * loose cycle time. This isn't too big a deal, since the loss will be
7959 * uniform across all VCPUs (not to mention the scenario is extremely
7960 * unlikely). It is possible that a second hibernate recovery happens
7961 * much faster than a first, causing the observed TSC here to be
7962 * smaller; this would require additional padding adjustment, which is
7963 * why we set last_host_tsc to the local tsc observed here.
7964 *
7965 * N.B. - this code below runs only on platforms with reliable TSC,
7966 * as that is the only way backwards_tsc is set above. Also note
7967 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
7968 * have the same delta_cyc adjustment applied if backwards_tsc
7969 * is detected. Note further, this adjustment is only done once,
7970 * as we reset last_host_tsc on all VCPUs to stop this from being
7971 * called multiple times (one for each physical CPU bringup).
7972 *
Guo Chao4a969982012-06-28 15:17:27 +08007973 * Platforms with unreliable TSCs don't have to deal with this, they
Zachary Amsden0dd6a6e2012-02-03 15:43:56 -02007974 * will be compensated by the logic in vcpu_load, which sets the TSC to
7975 * catchup mode. This will catchup all VCPUs to real time, but cannot
7976 * guarantee that they stay in perfect synchronization.
7977 */
7978 if (backwards_tsc) {
7979 u64 delta_cyc = max_tsc - local_tsc;
7980 list_for_each_entry(kvm, &vm_list, vm_list) {
Ladi Proseka826faf2017-06-26 09:56:43 +02007981 kvm->arch.backwards_tsc_observed = true;
Zachary Amsden0dd6a6e2012-02-03 15:43:56 -02007982 kvm_for_each_vcpu(i, vcpu, kvm) {
7983 vcpu->arch.tsc_offset_adjustment += delta_cyc;
7984 vcpu->arch.last_host_tsc = local_tsc;
Guo Hui Liu105b21b2014-09-12 13:43:19 +08007985 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
Zachary Amsden0dd6a6e2012-02-03 15:43:56 -02007986 }
7987
7988 /*
7989 * We have to disable TSC offset matching.. if you were
7990 * booting a VM while issuing an S4 host suspend....
7991 * you may have some problem. Solving this issue is
7992 * left as an exercise to the reader.
7993 */
7994 kvm->arch.last_tsc_nsec = 0;
7995 kvm->arch.last_tsc_write = 0;
7996 }
7997
7998 }
7999 return 0;
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08008000}
8001
Radim Krčmář13a34e02014-08-28 15:13:03 +02008002void kvm_arch_hardware_disable(void)
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08008003{
Radim Krčmář13a34e02014-08-28 15:13:03 +02008004 kvm_x86_ops->hardware_disable();
8005 drop_user_return_notifiers();
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08008006}
8007
8008int kvm_arch_hardware_setup(void)
8009{
Nadav Amit9e9c3fe2015-04-12 21:47:15 +03008010 int r;
8011
8012 r = kvm_x86_ops->hardware_setup();
8013 if (r != 0)
8014 return r;
8015
Haozhong Zhang35181e82015-10-20 15:39:03 +08008016 if (kvm_has_tsc_control) {
8017 /*
8018 * Make sure the user can only configure tsc_khz values that
8019 * fit into a signed integer.
8020 * A min value is not calculated needed because it will always
8021 * be 1 on all machines.
8022 */
8023 u64 max = min(0x7fffffffULL,
8024 __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
8025 kvm_max_guest_tsc_khz = max;
8026
Haozhong Zhangad721882015-10-20 15:39:02 +08008027 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
Haozhong Zhang35181e82015-10-20 15:39:03 +08008028 }
Haozhong Zhangad721882015-10-20 15:39:02 +08008029
Nadav Amit9e9c3fe2015-04-12 21:47:15 +03008030 kvm_init_msr_list();
8031 return 0;
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08008032}
8033
8034void kvm_arch_hardware_unsetup(void)
8035{
8036 kvm_x86_ops->hardware_unsetup();
8037}
8038
8039void kvm_arch_check_processor_compat(void *rtn)
8040{
8041 kvm_x86_ops->check_processor_compatibility(rtn);
8042}
8043
Paolo Bonzinid71ba782015-07-29 11:56:48 +02008044bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
8045{
8046 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
8047}
8048EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
8049
8050bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
8051{
8052 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
8053}
8054
Gleb Natapov54e98182012-08-05 15:58:32 +03008055struct static_key kvm_no_apic_vcpu __read_mostly;
Paolo Bonzinibce87cc2016-01-08 13:48:51 +01008056EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
Gleb Natapov54e98182012-08-05 15:58:32 +03008057
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08008058int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
8059{
8060 struct page *page;
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08008061 int r;
8062
Suravee Suthikulpanitb2a05fe2017-09-12 10:42:41 -05008063 vcpu->arch.apicv_active = kvm_x86_ops->get_enable_apicv(vcpu);
Avi Kivity9aabc88f2010-07-29 15:11:50 +03008064 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
David Hildenbrand26de7982017-08-24 20:51:25 +02008065 if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
Avi Kivitya4535292008-04-13 17:54:35 +03008066 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08008067 else
Avi Kivitya4535292008-04-13 17:54:35 +03008068 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08008069
8070 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
8071 if (!page) {
8072 r = -ENOMEM;
8073 goto fail;
8074 }
Zhang Xiantaoad312c72007-12-13 23:50:52 +08008075 vcpu->arch.pio_data = page_address(page);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08008076
Zachary Amsdencc578282012-02-03 15:43:50 -02008077 kvm_set_tsc_khz(vcpu, max_tsc_khz);
Zachary Amsdenc2855452010-09-18 14:38:15 -10008078
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08008079 r = kvm_mmu_create(vcpu);
8080 if (r < 0)
8081 goto fail_free_pio_data;
8082
David Hildenbrand26de7982017-08-24 20:51:25 +02008083 if (irqchip_in_kernel(vcpu->kvm)) {
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08008084 r = kvm_create_lapic(vcpu);
8085 if (r < 0)
8086 goto fail_mmu_destroy;
Gleb Natapov54e98182012-08-05 15:58:32 +03008087 } else
8088 static_key_slow_inc(&kvm_no_apic_vcpu);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08008089
Huang Ying890ca9a2009-05-11 16:48:15 +08008090 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
8091 GFP_KERNEL);
8092 if (!vcpu->arch.mce_banks) {
8093 r = -ENOMEM;
Wei Yongjun443c39b2010-01-22 14:21:29 +08008094 goto fail_free_lapic;
Huang Ying890ca9a2009-05-11 16:48:15 +08008095 }
8096 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
8097
Wei Yongjunf1797352013-04-18 07:41:00 +08008098 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
8099 r = -ENOMEM;
Sheng Yangf5f48ee2010-06-30 12:25:15 +08008100 goto fail_free_mce_banks;
Wei Yongjunf1797352013-04-18 07:41:00 +08008101 }
Sheng Yangf5f48ee2010-06-30 12:25:15 +08008102
Ingo Molnar0ee6a512015-04-27 06:58:22 +02008103 fx_init(vcpu);
Julian Stecklina66f7b722012-12-05 15:26:19 +01008104
Paolo Bonzini4344ee92013-10-02 16:06:16 +02008105 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
Paolo Bonzinid7876f12013-10-02 16:06:15 +02008106
Eugene Korenevsky5a4f55c2015-03-29 23:56:12 +03008107 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
8108
Radim Krčmář74545702015-04-27 15:11:25 +02008109 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
8110
Gleb Natapovaf585b92010-10-14 11:22:46 +02008111 kvm_async_pf_hash_reset(vcpu);
Gleb Natapovf5132b02011-11-10 14:57:22 +02008112 kvm_pmu_init(vcpu);
Gleb Natapovaf585b92010-10-14 11:22:46 +02008113
Steve Rutherford1c1a9ce2015-07-30 11:27:16 +02008114 vcpu->arch.pending_external_vector = -1;
Longpeng(Mike)de63ad42017-08-08 12:05:33 +08008115 vcpu->arch.preempted_in_kernel = false;
Steve Rutherford1c1a9ce2015-07-30 11:27:16 +02008116
Andrey Smetanin5c9194122015-11-10 15:36:34 +03008117 kvm_hv_vcpu_init(vcpu);
8118
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08008119 return 0;
Ingo Molnar0ee6a512015-04-27 06:58:22 +02008120
Sheng Yangf5f48ee2010-06-30 12:25:15 +08008121fail_free_mce_banks:
8122 kfree(vcpu->arch.mce_banks);
Wei Yongjun443c39b2010-01-22 14:21:29 +08008123fail_free_lapic:
8124 kvm_free_lapic(vcpu);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08008125fail_mmu_destroy:
8126 kvm_mmu_destroy(vcpu);
8127fail_free_pio_data:
Zhang Xiantaoad312c72007-12-13 23:50:52 +08008128 free_page((unsigned long)vcpu->arch.pio_data);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08008129fail:
8130 return r;
8131}
8132
8133void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
8134{
Marcelo Tosattif656ce02009-12-23 14:35:25 -02008135 int idx;
8136
Andrey Smetanin1f4b34f2015-11-30 19:22:21 +03008137 kvm_hv_vcpu_uninit(vcpu);
Gleb Natapovf5132b02011-11-10 14:57:22 +02008138 kvm_pmu_destroy(vcpu);
Wei Yongjun36cb93f2010-01-22 14:18:47 +08008139 kfree(vcpu->arch.mce_banks);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08008140 kvm_free_lapic(vcpu);
Marcelo Tosattif656ce02009-12-23 14:35:25 -02008141 idx = srcu_read_lock(&vcpu->kvm->srcu);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08008142 kvm_mmu_destroy(vcpu);
Marcelo Tosattif656ce02009-12-23 14:35:25 -02008143 srcu_read_unlock(&vcpu->kvm->srcu, idx);
Zhang Xiantaoad312c72007-12-13 23:50:52 +08008144 free_page((unsigned long)vcpu->arch.pio_data);
Paolo Bonzini35754c92015-07-29 12:05:37 +02008145 if (!lapic_in_kernel(vcpu))
Gleb Natapov54e98182012-08-05 15:58:32 +03008146 static_key_slow_dec(&kvm_no_apic_vcpu);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08008147}
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +08008148
Radim Krčmáře790d9e2014-08-21 18:08:05 +02008149void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
8150{
Radim Krčmářae97a3b2014-08-21 18:08:06 +02008151 kvm_x86_ops->sched_in(vcpu, cpu);
Radim Krčmáře790d9e2014-08-21 18:08:05 +02008152}
8153
Carsten Ottee08b9632012-01-04 10:25:20 +01008154int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +08008155{
Carsten Ottee08b9632012-01-04 10:25:20 +01008156 if (type)
8157 return -EINVAL;
8158
Paolo Bonzini6ef768f2014-11-20 13:45:31 +01008159 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
Zhang Xiantaof05e70a2007-12-14 10:01:48 +08008160 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
Xiao Guangrong365c8862013-05-31 08:36:29 +08008161 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
Ben-Ami Yassour4d5c5d02008-07-28 19:26:26 +03008162 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
Alex Williamsone0f0bbc2013-10-30 11:02:30 -06008163 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +08008164
Sheng Yang5550af42008-10-15 20:15:06 +08008165 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
8166 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
Alex Williamson7a844282012-09-21 11:58:03 -06008167 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
8168 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
8169 &kvm->arch.irq_sources_bitmap);
Sheng Yang5550af42008-10-15 20:15:06 +08008170
Jan Kiszka038f8c12011-02-04 10:49:11 +01008171 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
Gleb Natapov1e08ec42012-09-13 17:19:24 +03008172 mutex_init(&kvm->arch.apic_map_lock);
Paolo Bonzini3f5ad8b2016-12-12 10:12:53 +01008173 mutex_init(&kvm->arch.hyperv.hv_lock);
Marcelo Tosattid8281992012-11-27 23:29:01 -02008174 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
8175
Paolo Bonzini108b2492016-09-01 14:21:03 +02008176 kvm->arch.kvmclock_offset = -ktime_get_boot_ns();
Marcelo Tosattid8281992012-11-27 23:29:01 -02008177 pvclock_update_vm_gtod_copy(kvm);
Marcelo Tosatti53f658b32008-12-11 20:45:05 +01008178
Andrew Jones7e44e442014-02-28 12:52:54 +01008179 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
Andrew Jones332967a2014-02-28 12:52:55 +01008180 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
Andrew Jones7e44e442014-02-28 12:52:54 +01008181
Xiao Guangrong0eb05bf2016-02-24 17:51:13 +08008182 kvm_page_track_init(kvm);
Xiao Guangrong13d268c2016-02-24 17:51:16 +08008183 kvm_mmu_init_vm(kvm);
Xiao Guangrong0eb05bf2016-02-24 17:51:13 +08008184
Suravee Suthikulpanit03543132016-05-04 14:09:42 -05008185 if (kvm_x86_ops->vm_init)
8186 return kvm_x86_ops->vm_init(kvm);
8187
Jan Kiszkad89f5ef2010-11-09 17:02:49 +01008188 return 0;
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +08008189}
8190
8191static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
8192{
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +03008193 int r;
8194 r = vcpu_load(vcpu);
8195 BUG_ON(r);
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +08008196 kvm_mmu_unload(vcpu);
8197 vcpu_put(vcpu);
8198}
8199
8200static void kvm_free_vcpus(struct kvm *kvm)
8201{
8202 unsigned int i;
Gleb Natapov988a2ca2009-06-09 15:56:29 +03008203 struct kvm_vcpu *vcpu;
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +08008204
8205 /*
8206 * Unpin any mmu pages first.
8207 */
Gleb Natapovaf585b92010-10-14 11:22:46 +02008208 kvm_for_each_vcpu(i, vcpu, kvm) {
8209 kvm_clear_async_pf_completion_queue(vcpu);
Gleb Natapov988a2ca2009-06-09 15:56:29 +03008210 kvm_unload_vcpu_mmu(vcpu);
Gleb Natapovaf585b92010-10-14 11:22:46 +02008211 }
Gleb Natapov988a2ca2009-06-09 15:56:29 +03008212 kvm_for_each_vcpu(i, vcpu, kvm)
8213 kvm_arch_vcpu_free(vcpu);
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +08008214
Gleb Natapov988a2ca2009-06-09 15:56:29 +03008215 mutex_lock(&kvm->lock);
8216 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
8217 kvm->vcpus[i] = NULL;
8218
8219 atomic_set(&kvm->online_vcpus, 0);
8220 mutex_unlock(&kvm->lock);
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +08008221}
8222
Sheng Yangad8ba2c2009-01-06 10:03:02 +08008223void kvm_arch_sync_events(struct kvm *kvm)
8224{
Andrew Jones332967a2014-02-28 12:52:55 +01008225 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
Andrew Jones7e44e442014-02-28 12:52:54 +01008226 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
Xiao Guangrongaea924f2010-07-10 17:37:56 +08008227 kvm_free_pit(kvm);
Sheng Yangad8ba2c2009-01-06 10:03:02 +08008228}
8229
Paolo Bonzini1d8007b2015-10-12 13:38:32 +02008230int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
Paolo Bonzini9da0e4d2015-05-18 13:33:16 +02008231{
8232 int i, r;
Paolo Bonzini25188b92015-10-14 15:51:08 +02008233 unsigned long hva;
Paolo Bonzinif0d648b2015-10-12 13:56:27 +02008234 struct kvm_memslots *slots = kvm_memslots(kvm);
8235 struct kvm_memory_slot *slot, old;
Paolo Bonzini9da0e4d2015-05-18 13:33:16 +02008236
8237 /* Called with kvm->slots_lock held. */
Paolo Bonzini1d8007b2015-10-12 13:38:32 +02008238 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
8239 return -EINVAL;
Paolo Bonzini9da0e4d2015-05-18 13:33:16 +02008240
Paolo Bonzinif0d648b2015-10-12 13:56:27 +02008241 slot = id_to_memslot(slots, id);
8242 if (size) {
Paolo Bonzinib21629d2016-06-01 14:09:18 +02008243 if (slot->npages)
Paolo Bonzinif0d648b2015-10-12 13:56:27 +02008244 return -EEXIST;
8245
8246 /*
8247 * MAP_SHARED to prevent internal slot pages from being moved
8248 * by fork()/COW.
8249 */
8250 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
8251 MAP_SHARED | MAP_ANONYMOUS, 0);
8252 if (IS_ERR((void *)hva))
8253 return PTR_ERR((void *)hva);
8254 } else {
8255 if (!slot->npages)
8256 return 0;
8257
8258 hva = 0;
8259 }
8260
8261 old = *slot;
Paolo Bonzini9da0e4d2015-05-18 13:33:16 +02008262 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
Paolo Bonzini1d8007b2015-10-12 13:38:32 +02008263 struct kvm_userspace_memory_region m;
Paolo Bonzini9da0e4d2015-05-18 13:33:16 +02008264
Paolo Bonzini1d8007b2015-10-12 13:38:32 +02008265 m.slot = id | (i << 16);
8266 m.flags = 0;
8267 m.guest_phys_addr = gpa;
Paolo Bonzinif0d648b2015-10-12 13:56:27 +02008268 m.userspace_addr = hva;
Paolo Bonzini1d8007b2015-10-12 13:38:32 +02008269 m.memory_size = size;
Paolo Bonzini9da0e4d2015-05-18 13:33:16 +02008270 r = __kvm_set_memory_region(kvm, &m);
8271 if (r < 0)
8272 return r;
8273 }
8274
Paolo Bonzinif0d648b2015-10-12 13:56:27 +02008275 if (!size) {
8276 r = vm_munmap(old.userspace_addr, old.npages * PAGE_SIZE);
8277 WARN_ON(r < 0);
8278 }
8279
Paolo Bonzini9da0e4d2015-05-18 13:33:16 +02008280 return 0;
8281}
8282EXPORT_SYMBOL_GPL(__x86_set_memory_region);
8283
Paolo Bonzini1d8007b2015-10-12 13:38:32 +02008284int x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
Paolo Bonzini9da0e4d2015-05-18 13:33:16 +02008285{
8286 int r;
8287
8288 mutex_lock(&kvm->slots_lock);
Paolo Bonzini1d8007b2015-10-12 13:38:32 +02008289 r = __x86_set_memory_region(kvm, id, gpa, size);
Paolo Bonzini9da0e4d2015-05-18 13:33:16 +02008290 mutex_unlock(&kvm->slots_lock);
8291
8292 return r;
8293}
8294EXPORT_SYMBOL_GPL(x86_set_memory_region);
8295
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +08008296void kvm_arch_destroy_vm(struct kvm *kvm)
8297{
Andrew Honig27469d22013-04-18 09:38:14 -07008298 if (current->mm == kvm->mm) {
8299 /*
8300 * Free memory regions allocated on behalf of userspace,
8301 * unless the the memory map has changed due to process exit
8302 * or fd copying.
8303 */
Paolo Bonzini1d8007b2015-10-12 13:38:32 +02008304 x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0);
8305 x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, 0, 0);
8306 x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
Andrew Honig27469d22013-04-18 09:38:14 -07008307 }
Suravee Suthikulpanit03543132016-05-04 14:09:42 -05008308 if (kvm_x86_ops->vm_destroy)
8309 kvm_x86_ops->vm_destroy(kvm);
Peter Xuc7611592017-03-15 16:01:19 +08008310 kvm_pic_destroy(kvm);
8311 kvm_ioapic_destroy(kvm);
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +08008312 kvm_free_vcpus(kvm);
Radim Krčmářaf1bae52016-07-12 22:09:30 +02008313 kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
Xiao Guangrong13d268c2016-02-24 17:51:16 +08008314 kvm_mmu_uninit_vm(kvm);
Paolo Bonzini2beb6dad2017-03-27 17:53:50 +02008315 kvm_page_track_cleanup(kvm);
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +08008316}
Zhang Xiantao0de10342007-11-20 16:25:04 +08008317
Aneesh Kumar K.V55870272013-10-07 22:18:00 +05308318void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +09008319 struct kvm_memory_slot *dont)
8320{
8321 int i;
8322
Takuya Yoshikawad89cc612012-08-01 18:03:28 +09008323 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
8324 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
Thomas Huth548ef282015-02-24 21:29:25 +01008325 kvfree(free->arch.rmap[i]);
Takuya Yoshikawad89cc612012-08-01 18:03:28 +09008326 free->arch.rmap[i] = NULL;
Takuya Yoshikawa77d11302012-07-02 17:57:17 +09008327 }
Takuya Yoshikawad89cc612012-08-01 18:03:28 +09008328 if (i == 0)
8329 continue;
8330
8331 if (!dont || free->arch.lpage_info[i - 1] !=
8332 dont->arch.lpage_info[i - 1]) {
Thomas Huth548ef282015-02-24 21:29:25 +01008333 kvfree(free->arch.lpage_info[i - 1]);
Takuya Yoshikawad89cc612012-08-01 18:03:28 +09008334 free->arch.lpage_info[i - 1] = NULL;
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +09008335 }
8336 }
Xiao Guangrong21ebbed2016-02-24 17:51:09 +08008337
8338 kvm_page_track_free_memslot(free, dont);
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +09008339}
8340
Aneesh Kumar K.V55870272013-10-07 22:18:00 +05308341int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
8342 unsigned long npages)
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +09008343{
8344 int i;
8345
Takuya Yoshikawad89cc612012-08-01 18:03:28 +09008346 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
Xiao Guangrong92f94f12016-02-24 17:51:06 +08008347 struct kvm_lpage_info *linfo;
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +09008348 unsigned long ugfn;
8349 int lpages;
Takuya Yoshikawad89cc612012-08-01 18:03:28 +09008350 int level = i + 1;
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +09008351
8352 lpages = gfn_to_index(slot->base_gfn + npages - 1,
8353 slot->base_gfn, level) + 1;
8354
Takuya Yoshikawad89cc612012-08-01 18:03:28 +09008355 slot->arch.rmap[i] =
Michal Hockoa7c3e902017-05-08 15:57:09 -07008356 kvzalloc(lpages * sizeof(*slot->arch.rmap[i]), GFP_KERNEL);
Takuya Yoshikawad89cc612012-08-01 18:03:28 +09008357 if (!slot->arch.rmap[i])
Takuya Yoshikawa77d11302012-07-02 17:57:17 +09008358 goto out_free;
Takuya Yoshikawad89cc612012-08-01 18:03:28 +09008359 if (i == 0)
8360 continue;
Takuya Yoshikawa77d11302012-07-02 17:57:17 +09008361
Michal Hockoa7c3e902017-05-08 15:57:09 -07008362 linfo = kvzalloc(lpages * sizeof(*linfo), GFP_KERNEL);
Xiao Guangrong92f94f12016-02-24 17:51:06 +08008363 if (!linfo)
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +09008364 goto out_free;
8365
Xiao Guangrong92f94f12016-02-24 17:51:06 +08008366 slot->arch.lpage_info[i - 1] = linfo;
8367
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +09008368 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
Xiao Guangrong92f94f12016-02-24 17:51:06 +08008369 linfo[0].disallow_lpage = 1;
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +09008370 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
Xiao Guangrong92f94f12016-02-24 17:51:06 +08008371 linfo[lpages - 1].disallow_lpage = 1;
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +09008372 ugfn = slot->userspace_addr >> PAGE_SHIFT;
8373 /*
8374 * If the gfn and userspace address are not aligned wrt each
8375 * other, or if explicitly asked to, disable large page
8376 * support for this slot
8377 */
8378 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
8379 !kvm_largepages_enabled()) {
8380 unsigned long j;
8381
8382 for (j = 0; j < lpages; ++j)
Xiao Guangrong92f94f12016-02-24 17:51:06 +08008383 linfo[j].disallow_lpage = 1;
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +09008384 }
8385 }
8386
Xiao Guangrong21ebbed2016-02-24 17:51:09 +08008387 if (kvm_page_track_create_memslot(slot, npages))
8388 goto out_free;
8389
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +09008390 return 0;
8391
8392out_free:
Takuya Yoshikawad89cc612012-08-01 18:03:28 +09008393 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
Thomas Huth548ef282015-02-24 21:29:25 +01008394 kvfree(slot->arch.rmap[i]);
Takuya Yoshikawad89cc612012-08-01 18:03:28 +09008395 slot->arch.rmap[i] = NULL;
8396 if (i == 0)
8397 continue;
8398
Thomas Huth548ef282015-02-24 21:29:25 +01008399 kvfree(slot->arch.lpage_info[i - 1]);
Takuya Yoshikawad89cc612012-08-01 18:03:28 +09008400 slot->arch.lpage_info[i - 1] = NULL;
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +09008401 }
8402 return -ENOMEM;
8403}
8404
Paolo Bonzini15f46012015-05-17 21:26:08 +02008405void kvm_arch_memslots_updated(struct kvm *kvm, struct kvm_memslots *slots)
Takuya Yoshikawae59dbe02013-07-04 13:40:29 +09008406{
Takuya Yoshikawae6dff7d2013-07-04 13:41:26 +09008407 /*
8408 * memslots->generation has been incremented.
8409 * mmio generation may have reached its maximum value.
8410 */
Paolo Bonzini54bf36a2015-04-08 15:39:23 +02008411 kvm_mmu_invalidate_mmio_sptes(kvm, slots);
Takuya Yoshikawae59dbe02013-07-04 13:40:29 +09008412}
8413
Marcelo Tosattif7784b82009-12-23 14:35:18 -02008414int kvm_arch_prepare_memory_region(struct kvm *kvm,
8415 struct kvm_memory_slot *memslot,
Paolo Bonzini09170a42015-05-18 13:59:39 +02008416 const struct kvm_userspace_memory_region *mem,
Takuya Yoshikawa7b6195a2013-02-27 19:44:34 +09008417 enum kvm_mr_change change)
Zhang Xiantao0de10342007-11-20 16:25:04 +08008418{
Marcelo Tosattif7784b82009-12-23 14:35:18 -02008419 return 0;
8420}
8421
Kai Huang88178fd2015-01-28 10:54:27 +08008422static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
8423 struct kvm_memory_slot *new)
8424{
8425 /* Still write protect RO slot */
8426 if (new->flags & KVM_MEM_READONLY) {
8427 kvm_mmu_slot_remove_write_access(kvm, new);
8428 return;
8429 }
8430
8431 /*
8432 * Call kvm_x86_ops dirty logging hooks when they are valid.
8433 *
8434 * kvm_x86_ops->slot_disable_log_dirty is called when:
8435 *
8436 * - KVM_MR_CREATE with dirty logging is disabled
8437 * - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
8438 *
8439 * The reason is, in case of PML, we need to set D-bit for any slots
8440 * with dirty logging disabled in order to eliminate unnecessary GPA
8441 * logging in PML buffer (and potential PML buffer full VMEXT). This
8442 * guarantees leaving PML enabled during guest's lifetime won't have
8443 * any additonal overhead from PML when guest is running with dirty
8444 * logging disabled for memory slots.
8445 *
8446 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
8447 * to dirty logging mode.
8448 *
8449 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
8450 *
8451 * In case of write protect:
8452 *
8453 * Write protect all pages for dirty logging.
8454 *
8455 * All the sptes including the large sptes which point to this
8456 * slot are set to readonly. We can not create any new large
8457 * spte on this slot until the end of the logging.
8458 *
8459 * See the comments in fast_page_fault().
8460 */
8461 if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
8462 if (kvm_x86_ops->slot_enable_log_dirty)
8463 kvm_x86_ops->slot_enable_log_dirty(kvm, new);
8464 else
8465 kvm_mmu_slot_remove_write_access(kvm, new);
8466 } else {
8467 if (kvm_x86_ops->slot_disable_log_dirty)
8468 kvm_x86_ops->slot_disable_log_dirty(kvm, new);
8469 }
8470}
8471
Marcelo Tosattif7784b82009-12-23 14:35:18 -02008472void kvm_arch_commit_memory_region(struct kvm *kvm,
Paolo Bonzini09170a42015-05-18 13:59:39 +02008473 const struct kvm_userspace_memory_region *mem,
Takuya Yoshikawa84826442013-02-27 19:45:25 +09008474 const struct kvm_memory_slot *old,
Paolo Bonzinif36f3f22015-05-18 13:20:23 +02008475 const struct kvm_memory_slot *new,
Takuya Yoshikawa84826442013-02-27 19:45:25 +09008476 enum kvm_mr_change change)
Marcelo Tosattif7784b82009-12-23 14:35:18 -02008477{
Takuya Yoshikawa84826442013-02-27 19:45:25 +09008478 int nr_mmu_pages = 0;
Marcelo Tosattif7784b82009-12-23 14:35:18 -02008479
Xiao Guangrong48c0e4e2011-03-04 18:59:21 +08008480 if (!kvm->arch.n_requested_mmu_pages)
8481 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
Zhang Xiantao0de10342007-11-20 16:25:04 +08008482
Xiao Guangrong48c0e4e2011-03-04 18:59:21 +08008483 if (nr_mmu_pages)
8484 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
Kai Huang1c91cad42015-01-28 10:54:26 +08008485
Takuya Yoshikawac972f3b2013-01-08 19:43:28 +09008486 /*
Wanpeng Li3ea3b7f2015-04-03 15:40:25 +08008487 * Dirty logging tracks sptes in 4k granularity, meaning that large
8488 * sptes have to be split. If live migration is successful, the guest
8489 * in the source machine will be destroyed and large sptes will be
8490 * created in the destination. However, if the guest continues to run
8491 * in the source machine (for example if live migration fails), small
8492 * sptes will remain around and cause bad performance.
8493 *
8494 * Scan sptes if dirty logging has been stopped, dropping those
8495 * which can be collapsed into a single large-page spte. Later
8496 * page faults will create the large-page sptes.
8497 */
8498 if ((change != KVM_MR_DELETE) &&
8499 (old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
8500 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
8501 kvm_mmu_zap_collapsible_sptes(kvm, new);
8502
8503 /*
Kai Huang88178fd2015-01-28 10:54:27 +08008504 * Set up write protection and/or dirty logging for the new slot.
Xiao Guangrongc126d942014-04-17 17:06:14 +08008505 *
Kai Huang88178fd2015-01-28 10:54:27 +08008506 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
8507 * been zapped so no dirty logging staff is needed for old slot. For
8508 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
8509 * new and it's also covered when dealing with the new slot.
Paolo Bonzinif36f3f22015-05-18 13:20:23 +02008510 *
8511 * FIXME: const-ify all uses of struct kvm_memory_slot.
Takuya Yoshikawac972f3b2013-01-08 19:43:28 +09008512 */
Kai Huang88178fd2015-01-28 10:54:27 +08008513 if (change != KVM_MR_DELETE)
Paolo Bonzinif36f3f22015-05-18 13:20:23 +02008514 kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
Zhang Xiantao0de10342007-11-20 16:25:04 +08008515}
Zhang Xiantao1d737c82007-12-14 09:35:10 +08008516
Marcelo Tosatti2df72e92012-08-24 15:54:57 -03008517void kvm_arch_flush_shadow_all(struct kvm *kvm)
Marcelo Tosatti34d4cb82008-07-10 20:49:31 -03008518{
Xiao Guangrong6ca18b62013-05-31 08:36:23 +08008519 kvm_mmu_invalidate_zap_all_pages(kvm);
Marcelo Tosatti34d4cb82008-07-10 20:49:31 -03008520}
8521
Marcelo Tosatti2df72e92012-08-24 15:54:57 -03008522void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
8523 struct kvm_memory_slot *slot)
8524{
Xiaoguang Chenae7cd872016-10-09 15:41:44 +08008525 kvm_page_track_flush_slot(kvm, slot);
Marcelo Tosatti2df72e92012-08-24 15:54:57 -03008526}
8527
Paolo Bonzini5d9bc642015-10-13 10:18:53 +02008528static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
8529{
8530 if (!list_empty_careful(&vcpu->async_pf.done))
8531 return true;
8532
8533 if (kvm_apic_has_events(vcpu))
8534 return true;
8535
8536 if (vcpu->arch.pv.pv_unhalted)
8537 return true;
8538
Wanpeng Lia5f01f82017-09-13 04:04:01 -07008539 if (vcpu->arch.exception.pending)
8540 return true;
8541
ZhuangYanying47a66ee2017-05-26 13:16:48 +08008542 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
8543 (vcpu->arch.nmi_pending &&
8544 kvm_x86_ops->nmi_allowed(vcpu)))
Paolo Bonzini5d9bc642015-10-13 10:18:53 +02008545 return true;
8546
ZhuangYanying47a66ee2017-05-26 13:16:48 +08008547 if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
8548 (vcpu->arch.smi_pending && !is_smm(vcpu)))
Paolo Bonzini73917732015-10-13 10:19:35 +02008549 return true;
8550
Paolo Bonzini5d9bc642015-10-13 10:18:53 +02008551 if (kvm_arch_interrupt_allowed(vcpu) &&
8552 kvm_cpu_has_interrupt(vcpu))
8553 return true;
8554
Andrey Smetanin1f4b34f2015-11-30 19:22:21 +03008555 if (kvm_hv_has_stimer_pending(vcpu))
8556 return true;
8557
Paolo Bonzini5d9bc642015-10-13 10:18:53 +02008558 return false;
8559}
8560
Zhang Xiantao1d737c82007-12-14 09:35:10 +08008561int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
8562{
Paolo Bonzini5d9bc642015-10-13 10:18:53 +02008563 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
Zhang Xiantao1d737c82007-12-14 09:35:10 +08008564}
Zhang Xiantao57361992007-12-17 14:21:40 +08008565
Longpeng(Mike)199b5762017-08-08 12:05:32 +08008566bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
8567{
Longpeng(Mike)de63ad42017-08-08 12:05:33 +08008568 return vcpu->arch.preempted_in_kernel;
Longpeng(Mike)199b5762017-08-08 12:05:32 +08008569}
8570
Christoffer Dallb6d33832012-03-08 16:44:24 -05008571int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
Zhang Xiantao57361992007-12-17 14:21:40 +08008572{
Christoffer Dallb6d33832012-03-08 16:44:24 -05008573 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
Zhang Xiantao57361992007-12-17 14:21:40 +08008574}
Gleb Natapov78646122009-03-23 12:12:11 +02008575
8576int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
8577{
8578 return kvm_x86_ops->interrupt_allowed(vcpu);
8579}
Marcelo Tosatti229456f2009-06-17 09:22:14 -03008580
Nadav Amit82b32772014-11-02 11:54:45 +02008581unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
8582{
8583 if (is_64_bit_mode(vcpu))
8584 return kvm_rip_read(vcpu);
8585 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
8586 kvm_rip_read(vcpu));
8587}
8588EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
8589
Jan Kiszkaf92653e2010-02-23 17:47:55 +01008590bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
8591{
Nadav Amit82b32772014-11-02 11:54:45 +02008592 return kvm_get_linear_rip(vcpu) == linear_rip;
Jan Kiszkaf92653e2010-02-23 17:47:55 +01008593}
8594EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
8595
Jan Kiszka94fe45d2009-10-18 13:24:44 +02008596unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
8597{
8598 unsigned long rflags;
8599
8600 rflags = kvm_x86_ops->get_rflags(vcpu);
8601 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
Jan Kiszkac310bac2010-02-23 17:47:58 +01008602 rflags &= ~X86_EFLAGS_TF;
Jan Kiszka94fe45d2009-10-18 13:24:44 +02008603 return rflags;
8604}
8605EXPORT_SYMBOL_GPL(kvm_get_rflags);
8606
Paolo Bonzini6addfc42014-03-27 11:29:28 +01008607static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
Jan Kiszka94fe45d2009-10-18 13:24:44 +02008608{
8609 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
Jan Kiszkaf92653e2010-02-23 17:47:55 +01008610 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
Jan Kiszkac310bac2010-02-23 17:47:58 +01008611 rflags |= X86_EFLAGS_TF;
Jan Kiszka94fe45d2009-10-18 13:24:44 +02008612 kvm_x86_ops->set_rflags(vcpu, rflags);
Paolo Bonzini6addfc42014-03-27 11:29:28 +01008613}
8614
8615void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
8616{
8617 __kvm_set_rflags(vcpu, rflags);
Avi Kivity3842d132010-07-27 12:30:24 +03008618 kvm_make_request(KVM_REQ_EVENT, vcpu);
Jan Kiszka94fe45d2009-10-18 13:24:44 +02008619}
8620EXPORT_SYMBOL_GPL(kvm_set_rflags);
8621
Gleb Natapov56028d02010-10-17 18:13:42 +02008622void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
8623{
8624 int r;
8625
Xiao Guangrongfb67e142010-12-07 10:35:25 +08008626 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
chai wenf2e10662013-10-14 22:22:33 +08008627 work->wakeup_all)
Gleb Natapov56028d02010-10-17 18:13:42 +02008628 return;
8629
8630 r = kvm_mmu_reload(vcpu);
8631 if (unlikely(r))
8632 return;
8633
Xiao Guangrongfb67e142010-12-07 10:35:25 +08008634 if (!vcpu->arch.mmu.direct_map &&
8635 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
8636 return;
8637
Gleb Natapov56028d02010-10-17 18:13:42 +02008638 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
8639}
8640
Gleb Natapovaf585b92010-10-14 11:22:46 +02008641static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
8642{
8643 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
8644}
8645
8646static inline u32 kvm_async_pf_next_probe(u32 key)
8647{
8648 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
8649}
8650
8651static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8652{
8653 u32 key = kvm_async_pf_hash_fn(gfn);
8654
8655 while (vcpu->arch.apf.gfns[key] != ~0)
8656 key = kvm_async_pf_next_probe(key);
8657
8658 vcpu->arch.apf.gfns[key] = gfn;
8659}
8660
8661static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
8662{
8663 int i;
8664 u32 key = kvm_async_pf_hash_fn(gfn);
8665
8666 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
Xiao Guangrongc7d28c22010-11-01 17:00:30 +08008667 (vcpu->arch.apf.gfns[key] != gfn &&
8668 vcpu->arch.apf.gfns[key] != ~0); i++)
Gleb Natapovaf585b92010-10-14 11:22:46 +02008669 key = kvm_async_pf_next_probe(key);
8670
8671 return key;
8672}
8673
8674bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8675{
8676 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
8677}
8678
8679static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8680{
8681 u32 i, j, k;
8682
8683 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
8684 while (true) {
8685 vcpu->arch.apf.gfns[i] = ~0;
8686 do {
8687 j = kvm_async_pf_next_probe(j);
8688 if (vcpu->arch.apf.gfns[j] == ~0)
8689 return;
8690 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
8691 /*
8692 * k lies cyclically in ]i,j]
8693 * | i.k.j |
8694 * |....j i.k.| or |.k..j i...|
8695 */
8696 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
8697 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
8698 i = j;
8699 }
8700}
8701
Gleb Natapov7c907052010-10-14 11:22:53 +02008702static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
8703{
Paolo Bonzini4e335d92017-05-02 16:20:18 +02008704
8705 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
8706 sizeof(val));
Gleb Natapov7c907052010-10-14 11:22:53 +02008707}
8708
Wanpeng Li9a6e7c32017-09-14 03:54:16 -07008709static int apf_get_user(struct kvm_vcpu *vcpu, u32 *val)
8710{
8711
8712 return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, val,
8713 sizeof(u32));
8714}
8715
Gleb Natapovaf585b92010-10-14 11:22:46 +02008716void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
8717 struct kvm_async_pf *work)
8718{
Avi Kivity6389ee92010-11-29 16:12:30 +02008719 struct x86_exception fault;
8720
Gleb Natapov7c907052010-10-14 11:22:53 +02008721 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
Gleb Natapovaf585b92010-10-14 11:22:46 +02008722 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
Gleb Natapov7c907052010-10-14 11:22:53 +02008723
8724 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
Gleb Natapovfc5f06f2010-10-14 11:22:56 +02008725 (vcpu->arch.apf.send_user_only &&
8726 kvm_x86_ops->get_cpl(vcpu) == 0))
Gleb Natapov7c907052010-10-14 11:22:53 +02008727 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
8728 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
Avi Kivity6389ee92010-11-29 16:12:30 +02008729 fault.vector = PF_VECTOR;
8730 fault.error_code_valid = true;
8731 fault.error_code = 0;
8732 fault.nested_page_fault = false;
8733 fault.address = work->arch.token;
Wanpeng Liadfe20f2017-07-13 18:30:41 -07008734 fault.async_page_fault = true;
Avi Kivity6389ee92010-11-29 16:12:30 +02008735 kvm_inject_page_fault(vcpu, &fault);
Gleb Natapov7c907052010-10-14 11:22:53 +02008736 }
Gleb Natapovaf585b92010-10-14 11:22:46 +02008737}
8738
8739void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
8740 struct kvm_async_pf *work)
8741{
Avi Kivity6389ee92010-11-29 16:12:30 +02008742 struct x86_exception fault;
Wanpeng Li9a6e7c32017-09-14 03:54:16 -07008743 u32 val;
Avi Kivity6389ee92010-11-29 16:12:30 +02008744
chai wenf2e10662013-10-14 22:22:33 +08008745 if (work->wakeup_all)
Gleb Natapov7c907052010-10-14 11:22:53 +02008746 work->arch.token = ~0; /* broadcast wakeup */
8747 else
8748 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
Wanpeng Li24dccf82017-03-20 21:18:55 -07008749 trace_kvm_async_pf_ready(work->arch.token, work->gva);
Gleb Natapov7c907052010-10-14 11:22:53 +02008750
Wanpeng Li9a6e7c32017-09-14 03:54:16 -07008751 if (vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED &&
8752 !apf_get_user(vcpu, &val)) {
8753 if (val == KVM_PV_REASON_PAGE_NOT_PRESENT &&
8754 vcpu->arch.exception.pending &&
8755 vcpu->arch.exception.nr == PF_VECTOR &&
8756 !apf_put_user(vcpu, 0)) {
8757 vcpu->arch.exception.injected = false;
8758 vcpu->arch.exception.pending = false;
8759 vcpu->arch.exception.nr = 0;
8760 vcpu->arch.exception.has_error_code = false;
8761 vcpu->arch.exception.error_code = 0;
8762 } else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
8763 fault.vector = PF_VECTOR;
8764 fault.error_code_valid = true;
8765 fault.error_code = 0;
8766 fault.nested_page_fault = false;
8767 fault.address = work->arch.token;
8768 fault.async_page_fault = true;
8769 kvm_inject_page_fault(vcpu, &fault);
8770 }
Gleb Natapov7c907052010-10-14 11:22:53 +02008771 }
Xiao Guangronge6d53e32010-11-01 17:01:28 +08008772 vcpu->arch.apf.halted = false;
Gleb Natapova4fa1632012-05-03 11:36:39 +03008773 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
Gleb Natapov7c907052010-10-14 11:22:53 +02008774}
8775
8776bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
8777{
8778 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
8779 return true;
8780 else
Wanpeng Li9bc1f092017-06-08 20:13:40 -07008781 return kvm_can_do_async_pf(vcpu);
Gleb Natapovaf585b92010-10-14 11:22:46 +02008782}
8783
Paolo Bonzini5544eb92015-07-07 15:41:58 +02008784void kvm_arch_start_assignment(struct kvm *kvm)
8785{
8786 atomic_inc(&kvm->arch.assigned_device_count);
8787}
8788EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
8789
8790void kvm_arch_end_assignment(struct kvm *kvm)
8791{
8792 atomic_dec(&kvm->arch.assigned_device_count);
8793}
8794EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
8795
8796bool kvm_arch_has_assigned_device(struct kvm *kvm)
8797{
8798 return atomic_read(&kvm->arch.assigned_device_count);
8799}
8800EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
8801
Alex Williamsone0f0bbc2013-10-30 11:02:30 -06008802void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
8803{
8804 atomic_inc(&kvm->arch.noncoherent_dma_count);
8805}
8806EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
8807
8808void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
8809{
8810 atomic_dec(&kvm->arch.noncoherent_dma_count);
8811}
8812EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
8813
8814bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
8815{
8816 return atomic_read(&kvm->arch.noncoherent_dma_count);
8817}
8818EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
8819
Alex Williamson14717e22016-05-05 11:58:35 -06008820bool kvm_arch_has_irq_bypass(void)
8821{
8822 return kvm_x86_ops->update_pi_irte != NULL;
8823}
8824
Feng Wu87276882015-09-18 22:29:40 +08008825int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
8826 struct irq_bypass_producer *prod)
8827{
8828 struct kvm_kernel_irqfd *irqfd =
8829 container_of(cons, struct kvm_kernel_irqfd, consumer);
8830
Alex Williamson14717e22016-05-05 11:58:35 -06008831 irqfd->producer = prod;
Feng Wu87276882015-09-18 22:29:40 +08008832
Alex Williamson14717e22016-05-05 11:58:35 -06008833 return kvm_x86_ops->update_pi_irte(irqfd->kvm,
8834 prod->irq, irqfd->gsi, 1);
Feng Wu87276882015-09-18 22:29:40 +08008835}
8836
8837void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
8838 struct irq_bypass_producer *prod)
8839{
8840 int ret;
8841 struct kvm_kernel_irqfd *irqfd =
8842 container_of(cons, struct kvm_kernel_irqfd, consumer);
8843
Feng Wu87276882015-09-18 22:29:40 +08008844 WARN_ON(irqfd->producer != prod);
8845 irqfd->producer = NULL;
8846
8847 /*
8848 * When producer of consumer is unregistered, we change back to
8849 * remapped mode, so we can re-use the current implementation
Andrea Gelminibb3541f2016-05-21 14:14:44 +02008850 * when the irq is masked/disabled or the consumer side (KVM
Feng Wu87276882015-09-18 22:29:40 +08008851 * int this case doesn't want to receive the interrupts.
8852 */
8853 ret = kvm_x86_ops->update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
8854 if (ret)
8855 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
8856 " fails: %d\n", irqfd->consumer.token, ret);
8857}
8858
8859int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
8860 uint32_t guest_irq, bool set)
8861{
8862 if (!kvm_x86_ops->update_pi_irte)
8863 return -EINVAL;
8864
8865 return kvm_x86_ops->update_pi_irte(kvm, host_irq, guest_irq, set);
8866}
8867
Feng Wu520040142016-01-25 16:53:33 +08008868bool kvm_vector_hashing_enabled(void)
8869{
8870 return vector_hashing;
8871}
8872EXPORT_SYMBOL_GPL(kvm_vector_hashing_enabled);
8873
Marcelo Tosatti229456f2009-06-17 09:22:14 -03008874EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
Jason Wang931c33b2015-09-15 14:41:58 +08008875EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
Marcelo Tosatti229456f2009-06-17 09:22:14 -03008876EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
8877EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
8878EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
8879EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
Joerg Roedel0ac406d2009-10-09 16:08:27 +02008880EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
Joerg Roedeld8cabdd2009-10-09 16:08:28 +02008881EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
Joerg Roedel17897f32009-10-09 16:08:29 +02008882EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
Joerg Roedel236649d2009-10-09 16:08:30 +02008883EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
Joerg Roedelec1ff792009-10-09 16:08:31 +02008884EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
Joerg Roedel532a46b2009-10-09 16:08:32 +02008885EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
Joerg Roedel2e554e82010-02-24 18:59:14 +01008886EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
Yoshihiro YUNOMAE489223e2013-06-12 16:43:44 +09008887EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
Radim Krčmář7b462682014-08-21 18:08:09 +02008888EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window);
Kai Huang843e4332015-01-28 10:54:28 +08008889EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
Feng Wuefc64402015-09-18 22:29:51 +08008890EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
Suravee Suthikulpanit18f40c52016-05-04 14:09:48 -05008891EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
8892EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);