blob: 37040079cd6bec44276a4161b84e120fe923293f [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"
Carsten Otte313a3dc2007-10-11 19:16:52 +020030
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -020031#include <linux/clocksource.h>
Ben-Ami Yassour4d5c5d02008-07-28 19:26:26 +030032#include <linux/interrupt.h>
Carsten Otte313a3dc2007-10-11 19:16:52 +020033#include <linux/kvm.h>
34#include <linux/fs.h>
35#include <linux/vmalloc.h>
Carsten Otte5fb76f92007-10-29 16:08:51 +010036#include <linux/module.h>
Zhang Xiantao0de10342007-11-20 16:25:04 +080037#include <linux/mman.h>
Marcelo Tosatti2bacc552007-12-12 10:46:12 -050038#include <linux/highmem.h>
Joerg Roedel19de40a2008-12-03 14:43:34 +010039#include <linux/iommu.h>
Ben-Ami Yassour62c476c2008-09-14 03:48:28 +030040#include <linux/intel-iommu.h>
Gerd Hoffmannc8076602009-02-04 17:52:04 +010041#include <linux/cpufreq.h>
Avi Kivity18863bd2009-09-07 11:12:18 +030042#include <linux/user-return-notifier.h>
Marcelo Tosattia983fb22009-12-23 14:35:23 -020043#include <linux/srcu.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090044#include <linux/slab.h>
Zhang, Yanminff9d07a2010-04-19 13:32:45 +080045#include <linux/perf_event.h>
Lai Jiangshan7bee3422010-06-02 17:06:03 +080046#include <linux/uaccess.h>
Gleb Natapovaf585b92010-10-14 11:22:46 +020047#include <linux/hash.h>
Joerg Roedela1b60c12011-09-06 18:46:34 +020048#include <linux/pci.h>
Marcelo Tosatti16e8d742012-11-27 23:29:00 -020049#include <linux/timekeeper_internal.h>
50#include <linux/pvclock_gtod.h>
Avi Kivityaec51dc2009-07-01 16:01:02 +030051#include <trace/events/kvm.h>
Xiao Guangrong2ed152a2010-03-10 19:00:43 +080052
Marcelo Tosatti229456f2009-06-17 09:22:14 -030053#define CREATE_TRACE_POINTS
54#include "trace.h"
Carsten Otte043405e2007-10-10 17:16:19 +020055
Frederic Weisbecker24f1e32c2009-09-09 19:22:48 +020056#include <asm/debugreg.h>
Zhang Xiantaod825ed02007-11-14 20:08:51 +080057#include <asm/msr.h>
Avi Kivitya5f61302008-02-20 17:57:21 +020058#include <asm/desc.h>
Sheng Yang0bed3b52008-10-09 16:01:54 +080059#include <asm/mtrr.h>
Huang Ying890ca9a2009-05-11 16:48:15 +080060#include <asm/mce.h>
Sheng Yang7cf30852010-05-17 17:08:27 +080061#include <asm/i387.h>
Linus Torvalds1361b832012-02-21 13:19:22 -080062#include <asm/fpu-internal.h> /* Ugh! */
Sheng Yang98918832010-05-17 17:08:28 +080063#include <asm/xcr.h>
Zachary Amsden1d5f0662010-08-19 22:07:30 -100064#include <asm/pvclock.h>
Avi Kivity217fc9c2010-08-26 13:38:03 +030065#include <asm/div64.h>
Carsten Otte043405e2007-10-10 17:16:19 +020066
Carsten Otte313a3dc2007-10-11 19:16:52 +020067#define MAX_IO_MSRS 256
Huang Ying890ca9a2009-05-11 16:48:15 +080068#define KVM_MAX_MCE_BANKS 32
Huang Ying5854dbc2010-10-08 16:24:14 +080069#define KVM_MCE_CAP_SUPPORTED (MCG_CTL_P | MCG_SER_P)
Huang Ying890ca9a2009-05-11 16:48:15 +080070
Avi Kivity0f65dd72011-04-20 13:37:53 +030071#define emul_to_vcpu(ctxt) \
72 container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
73
Joerg Roedel50a37eb2008-01-31 14:57:38 +010074/* EFER defaults:
75 * - enable syscall per default because its emulated by KVM
76 * - enable LME and LMA per default on 64 bit KVM
77 */
78#ifdef CONFIG_X86_64
Lai Jiangshan1260edbe2011-02-21 11:51:35 +080079static
80u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
Joerg Roedel50a37eb2008-01-31 14:57:38 +010081#else
Lai Jiangshan1260edbe2011-02-21 11:51:35 +080082static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
Joerg Roedel50a37eb2008-01-31 14:57:38 +010083#endif
Carsten Otte313a3dc2007-10-11 19:16:52 +020084
Avi Kivityba1389b2007-11-18 16:24:12 +020085#define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
86#define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
Hollis Blanchard417bc302007-10-31 17:24:23 -050087
Gleb Natapovcb142eb2009-08-09 15:17:40 +030088static void update_cr8_intercept(struct kvm_vcpu *vcpu);
Avi Kivity7460fb4a2011-09-20 13:43:14 +030089static void process_nmi(struct kvm_vcpu *vcpu);
Avi Kivity674eea02008-02-11 18:37:23 +020090
Zhang Xiantao97896d02007-11-14 20:09:30 +080091struct kvm_x86_ops *kvm_x86_ops;
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -030092EXPORT_SYMBOL_GPL(kvm_x86_ops);
Zhang Xiantao97896d02007-11-14 20:09:30 +080093
Rusty Russell476bc002012-01-13 09:32:18 +103094static bool ignore_msrs = 0;
95module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
Andre Przywaraed85c062009-06-25 12:36:49 +020096
Joerg Roedel92a1f122011-03-25 09:44:51 +010097bool kvm_has_tsc_control;
98EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
99u32 kvm_max_guest_tsc_khz;
100EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
101
Zachary Amsdencc578282012-02-03 15:43:50 -0200102/* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
103static u32 tsc_tolerance_ppm = 250;
104module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
105
Avi Kivity18863bd2009-09-07 11:12:18 +0300106#define KVM_NR_SHARED_MSRS 16
107
108struct kvm_shared_msrs_global {
109 int nr;
Sheng Yang2bf78fa2009-12-18 16:48:44 +0800110 u32 msrs[KVM_NR_SHARED_MSRS];
Avi Kivity18863bd2009-09-07 11:12:18 +0300111};
112
113struct kvm_shared_msrs {
114 struct user_return_notifier urn;
115 bool registered;
Sheng Yang2bf78fa2009-12-18 16:48:44 +0800116 struct kvm_shared_msr_values {
117 u64 host;
118 u64 curr;
119 } values[KVM_NR_SHARED_MSRS];
Avi Kivity18863bd2009-09-07 11:12:18 +0300120};
121
122static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
Marcelo Tosatti013f6a52013-01-03 11:41:39 -0200123static struct kvm_shared_msrs __percpu *shared_msrs;
Avi Kivity18863bd2009-09-07 11:12:18 +0300124
Hollis Blanchard417bc302007-10-31 17:24:23 -0500125struct kvm_stats_debugfs_item debugfs_entries[] = {
Avi Kivityba1389b2007-11-18 16:24:12 +0200126 { "pf_fixed", VCPU_STAT(pf_fixed) },
127 { "pf_guest", VCPU_STAT(pf_guest) },
128 { "tlb_flush", VCPU_STAT(tlb_flush) },
129 { "invlpg", VCPU_STAT(invlpg) },
130 { "exits", VCPU_STAT(exits) },
131 { "io_exits", VCPU_STAT(io_exits) },
132 { "mmio_exits", VCPU_STAT(mmio_exits) },
133 { "signal_exits", VCPU_STAT(signal_exits) },
134 { "irq_window", VCPU_STAT(irq_window_exits) },
Sheng Yangf08864b2008-05-15 18:23:25 +0800135 { "nmi_window", VCPU_STAT(nmi_window_exits) },
Avi Kivityba1389b2007-11-18 16:24:12 +0200136 { "halt_exits", VCPU_STAT(halt_exits) },
137 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
Amit Shahf11c3a82008-02-21 01:00:30 +0530138 { "hypercalls", VCPU_STAT(hypercalls) },
Avi Kivityba1389b2007-11-18 16:24:12 +0200139 { "request_irq", VCPU_STAT(request_irq_exits) },
140 { "irq_exits", VCPU_STAT(irq_exits) },
141 { "host_state_reload", VCPU_STAT(host_state_reload) },
142 { "efer_reload", VCPU_STAT(efer_reload) },
143 { "fpu_reload", VCPU_STAT(fpu_reload) },
144 { "insn_emulation", VCPU_STAT(insn_emulation) },
145 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
Avi Kivityfa89a812008-09-01 15:57:51 +0300146 { "irq_injections", VCPU_STAT(irq_injections) },
Jan Kiszkac4abb7c2008-09-26 09:30:55 +0200147 { "nmi_injections", VCPU_STAT(nmi_injections) },
Avi Kivity4cee5762007-11-18 16:37:07 +0200148 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
149 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
150 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
151 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
152 { "mmu_flooded", VM_STAT(mmu_flooded) },
153 { "mmu_recycled", VM_STAT(mmu_recycled) },
Avi Kivitydfc5aa02007-12-18 19:47:18 +0200154 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
Marcelo Tosatti4731d4c2008-09-23 13:18:39 -0300155 { "mmu_unsync", VM_STAT(mmu_unsync) },
Avi Kivity0f74a242007-11-20 23:01:14 +0200156 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
Marcelo Tosatti05da4552008-02-23 11:44:30 -0300157 { "largepages", VM_STAT(lpages) },
Hollis Blanchard417bc302007-10-31 17:24:23 -0500158 { NULL }
159};
160
Dexuan Cui2acf9232010-06-10 11:27:12 +0800161u64 __read_mostly host_xcr0;
162
Jan Kiszkab6785de2012-09-20 07:43:17 +0200163static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
Avi Kivityd6aa1002011-04-20 15:47:13 +0300164
Jan Kiszka8b6e4542012-09-20 07:43:08 +0200165static int kvm_vcpu_reset(struct kvm_vcpu *vcpu);
Dexuan Cui2acf9232010-06-10 11:27:12 +0800166
Gleb Natapovaf585b92010-10-14 11:22:46 +0200167static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
168{
169 int i;
170 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
171 vcpu->arch.apf.gfns[i] = ~0;
172}
173
Avi Kivity18863bd2009-09-07 11:12:18 +0300174static void kvm_on_user_return(struct user_return_notifier *urn)
175{
176 unsigned slot;
Avi Kivity18863bd2009-09-07 11:12:18 +0300177 struct kvm_shared_msrs *locals
178 = container_of(urn, struct kvm_shared_msrs, urn);
Sheng Yang2bf78fa2009-12-18 16:48:44 +0800179 struct kvm_shared_msr_values *values;
Avi Kivity18863bd2009-09-07 11:12:18 +0300180
181 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
Sheng Yang2bf78fa2009-12-18 16:48:44 +0800182 values = &locals->values[slot];
183 if (values->host != values->curr) {
184 wrmsrl(shared_msrs_global.msrs[slot], values->host);
185 values->curr = values->host;
Avi Kivity18863bd2009-09-07 11:12:18 +0300186 }
187 }
188 locals->registered = false;
189 user_return_notifier_unregister(urn);
190}
191
Sheng Yang2bf78fa2009-12-18 16:48:44 +0800192static void shared_msr_update(unsigned slot, u32 msr)
Avi Kivity18863bd2009-09-07 11:12:18 +0300193{
Avi Kivity18863bd2009-09-07 11:12:18 +0300194 u64 value;
Marcelo Tosatti013f6a52013-01-03 11:41:39 -0200195 unsigned int cpu = smp_processor_id();
196 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
Avi Kivity18863bd2009-09-07 11:12:18 +0300197
Sheng Yang2bf78fa2009-12-18 16:48:44 +0800198 /* only read, and nobody should modify it at this time,
199 * so don't need lock */
200 if (slot >= shared_msrs_global.nr) {
201 printk(KERN_ERR "kvm: invalid MSR slot!");
202 return;
203 }
204 rdmsrl_safe(msr, &value);
205 smsr->values[slot].host = value;
206 smsr->values[slot].curr = value;
207}
208
209void kvm_define_shared_msr(unsigned slot, u32 msr)
210{
Avi Kivity18863bd2009-09-07 11:12:18 +0300211 if (slot >= shared_msrs_global.nr)
212 shared_msrs_global.nr = slot + 1;
Sheng Yang2bf78fa2009-12-18 16:48:44 +0800213 shared_msrs_global.msrs[slot] = msr;
214 /* we need ensured the shared_msr_global have been updated */
215 smp_wmb();
Avi Kivity18863bd2009-09-07 11:12:18 +0300216}
217EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
218
219static void kvm_shared_msr_cpu_online(void)
220{
221 unsigned i;
Avi Kivity18863bd2009-09-07 11:12:18 +0300222
223 for (i = 0; i < shared_msrs_global.nr; ++i)
Sheng Yang2bf78fa2009-12-18 16:48:44 +0800224 shared_msr_update(i, shared_msrs_global.msrs[i]);
Avi Kivity18863bd2009-09-07 11:12:18 +0300225}
226
Avi Kivityd5696722009-12-02 12:28:47 +0200227void kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
Avi Kivity18863bd2009-09-07 11:12:18 +0300228{
Marcelo Tosatti013f6a52013-01-03 11:41:39 -0200229 unsigned int cpu = smp_processor_id();
230 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
Avi Kivity18863bd2009-09-07 11:12:18 +0300231
Sheng Yang2bf78fa2009-12-18 16:48:44 +0800232 if (((value ^ smsr->values[slot].curr) & mask) == 0)
Avi Kivity18863bd2009-09-07 11:12:18 +0300233 return;
Sheng Yang2bf78fa2009-12-18 16:48:44 +0800234 smsr->values[slot].curr = value;
235 wrmsrl(shared_msrs_global.msrs[slot], value);
Avi Kivity18863bd2009-09-07 11:12:18 +0300236 if (!smsr->registered) {
237 smsr->urn.on_user_return = kvm_on_user_return;
238 user_return_notifier_register(&smsr->urn);
239 smsr->registered = true;
240 }
241}
242EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
243
Avi Kivity3548bab2009-11-28 14:18:47 +0200244static void drop_user_return_notifiers(void *ignore)
245{
Marcelo Tosatti013f6a52013-01-03 11:41:39 -0200246 unsigned int cpu = smp_processor_id();
247 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
Avi Kivity3548bab2009-11-28 14:18:47 +0200248
249 if (smsr->registered)
250 kvm_on_user_return(&smsr->urn);
251}
252
Carsten Otte6866b832007-10-29 16:09:10 +0100253u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
254{
Gleb Natapov8a5a87d2012-08-05 15:58:26 +0300255 return vcpu->arch.apic_base;
Carsten Otte6866b832007-10-29 16:09:10 +0100256}
257EXPORT_SYMBOL_GPL(kvm_get_apic_base);
258
259void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
260{
261 /* TODO: reserve bits check */
Gleb Natapov8a5a87d2012-08-05 15:58:26 +0300262 kvm_lapic_set_base(vcpu, data);
Carsten Otte6866b832007-10-29 16:09:10 +0100263}
264EXPORT_SYMBOL_GPL(kvm_set_apic_base);
265
Eddie Dong3fd28fc2009-11-19 17:54:07 +0200266#define EXCPT_BENIGN 0
267#define EXCPT_CONTRIBUTORY 1
268#define EXCPT_PF 2
269
270static int exception_class(int vector)
271{
272 switch (vector) {
273 case PF_VECTOR:
274 return EXCPT_PF;
275 case DE_VECTOR:
276 case TS_VECTOR:
277 case NP_VECTOR:
278 case SS_VECTOR:
279 case GP_VECTOR:
280 return EXCPT_CONTRIBUTORY;
281 default:
282 break;
283 }
284 return EXCPT_BENIGN;
285}
286
287static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
Joerg Roedelce7ddec2010-04-22 12:33:13 +0200288 unsigned nr, bool has_error, u32 error_code,
289 bool reinject)
Eddie Dong3fd28fc2009-11-19 17:54:07 +0200290{
291 u32 prev_nr;
292 int class1, class2;
293
Avi Kivity3842d132010-07-27 12:30:24 +0300294 kvm_make_request(KVM_REQ_EVENT, vcpu);
295
Eddie Dong3fd28fc2009-11-19 17:54:07 +0200296 if (!vcpu->arch.exception.pending) {
297 queue:
298 vcpu->arch.exception.pending = true;
299 vcpu->arch.exception.has_error_code = has_error;
300 vcpu->arch.exception.nr = nr;
301 vcpu->arch.exception.error_code = error_code;
Joerg Roedel3f0fd292010-05-05 16:04:41 +0200302 vcpu->arch.exception.reinject = reinject;
Eddie Dong3fd28fc2009-11-19 17:54:07 +0200303 return;
304 }
305
306 /* to check exception */
307 prev_nr = vcpu->arch.exception.nr;
308 if (prev_nr == DF_VECTOR) {
309 /* triple fault -> shutdown */
Avi Kivitya8eeb042010-05-10 12:34:53 +0300310 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
Eddie Dong3fd28fc2009-11-19 17:54:07 +0200311 return;
312 }
313 class1 = exception_class(prev_nr);
314 class2 = exception_class(nr);
315 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
316 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
317 /* generate double fault per SDM Table 5-5 */
318 vcpu->arch.exception.pending = true;
319 vcpu->arch.exception.has_error_code = true;
320 vcpu->arch.exception.nr = DF_VECTOR;
321 vcpu->arch.exception.error_code = 0;
322 } else
323 /* replace previous exception with a new one in a hope
324 that instruction re-execution will regenerate lost
325 exception */
326 goto queue;
327}
328
Avi Kivity298101d2007-11-25 13:41:11 +0200329void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
330{
Joerg Roedelce7ddec2010-04-22 12:33:13 +0200331 kvm_multiple_exception(vcpu, nr, false, 0, false);
Avi Kivity298101d2007-11-25 13:41:11 +0200332}
333EXPORT_SYMBOL_GPL(kvm_queue_exception);
334
Joerg Roedelce7ddec2010-04-22 12:33:13 +0200335void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
336{
337 kvm_multiple_exception(vcpu, nr, false, 0, true);
338}
339EXPORT_SYMBOL_GPL(kvm_requeue_exception);
340
Andre Przywaradb8fcef2010-12-21 11:12:01 +0100341void kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
Avi Kivityc3c91fe2007-11-25 14:04:58 +0200342{
Andre Przywaradb8fcef2010-12-21 11:12:01 +0100343 if (err)
344 kvm_inject_gp(vcpu, 0);
345 else
346 kvm_x86_ops->skip_emulated_instruction(vcpu);
347}
348EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
Joerg Roedel8df25a32010-09-10 17:30:46 +0200349
Avi Kivity6389ee92010-11-29 16:12:30 +0200350void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
Avi Kivityc3c91fe2007-11-25 14:04:58 +0200351{
Avi Kivityc3c91fe2007-11-25 14:04:58 +0200352 ++vcpu->stat.pf_guest;
Avi Kivity6389ee92010-11-29 16:12:30 +0200353 vcpu->arch.cr2 = fault->address;
354 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
Avi Kivityc3c91fe2007-11-25 14:04:58 +0200355}
Nadav Har'El27d6c862011-05-25 23:06:59 +0300356EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
Avi Kivityc3c91fe2007-11-25 14:04:58 +0200357
Avi Kivity6389ee92010-11-29 16:12:30 +0200358void kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
Joerg Roedeld4f8cf62010-09-10 17:30:55 +0200359{
Avi Kivity6389ee92010-11-29 16:12:30 +0200360 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
361 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
Joerg Roedeld4f8cf62010-09-10 17:30:55 +0200362 else
Avi Kivity6389ee92010-11-29 16:12:30 +0200363 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
Joerg Roedeld4f8cf62010-09-10 17:30:55 +0200364}
365
Sheng Yang3419ffc2008-05-15 09:52:48 +0800366void kvm_inject_nmi(struct kvm_vcpu *vcpu)
367{
Avi Kivity7460fb4a2011-09-20 13:43:14 +0300368 atomic_inc(&vcpu->arch.nmi_queued);
369 kvm_make_request(KVM_REQ_NMI, vcpu);
Sheng Yang3419ffc2008-05-15 09:52:48 +0800370}
371EXPORT_SYMBOL_GPL(kvm_inject_nmi);
372
Avi Kivity298101d2007-11-25 13:41:11 +0200373void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
374{
Joerg Roedelce7ddec2010-04-22 12:33:13 +0200375 kvm_multiple_exception(vcpu, nr, true, error_code, false);
Avi Kivity298101d2007-11-25 13:41:11 +0200376}
377EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
378
Joerg Roedelce7ddec2010-04-22 12:33:13 +0200379void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
380{
381 kvm_multiple_exception(vcpu, nr, true, error_code, true);
382}
383EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
384
Carsten Ottea03490e2007-10-29 16:09:35 +0100385/*
Avi Kivity0a79b002009-09-01 12:03:25 +0300386 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
387 * a #GP and return false.
388 */
389bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
Carsten Otte043405e2007-10-10 17:16:19 +0200390{
Avi Kivity0a79b002009-09-01 12:03:25 +0300391 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
392 return true;
393 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
394 return false;
Carsten Ottea03490e2007-10-29 16:09:35 +0100395}
Avi Kivity0a79b002009-09-01 12:03:25 +0300396EXPORT_SYMBOL_GPL(kvm_require_cpl);
Carsten Ottea03490e2007-10-29 16:09:35 +0100397
398/*
Joerg Roedelec92fe42010-09-10 17:30:51 +0200399 * This function will be used to read from the physical memory of the currently
400 * running guest. The difference to kvm_read_guest_page is that this function
401 * can read from guest physical or from the guest's guest physical memory.
402 */
403int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
404 gfn_t ngfn, void *data, int offset, int len,
405 u32 access)
406{
407 gfn_t real_gfn;
408 gpa_t ngpa;
409
410 ngpa = gfn_to_gpa(ngfn);
411 real_gfn = mmu->translate_gpa(vcpu, ngpa, access);
412 if (real_gfn == UNMAPPED_GVA)
413 return -EFAULT;
414
415 real_gfn = gpa_to_gfn(real_gfn);
416
417 return kvm_read_guest_page(vcpu->kvm, real_gfn, data, offset, len);
418}
419EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
420
Joerg Roedel3d06b8b2010-09-10 17:30:53 +0200421int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
422 void *data, int offset, int len, u32 access)
423{
424 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
425 data, offset, len, access);
426}
427
Joerg Roedelec92fe42010-09-10 17:30:51 +0200428/*
Carsten Ottea03490e2007-10-29 16:09:35 +0100429 * Load the pae pdptrs. Return true is they are all valid.
430 */
Joerg Roedelff03a072010-09-10 17:30:57 +0200431int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
Carsten Ottea03490e2007-10-29 16:09:35 +0100432{
433 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
434 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
435 int i;
436 int ret;
Joerg Roedelff03a072010-09-10 17:30:57 +0200437 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
Carsten Ottea03490e2007-10-29 16:09:35 +0100438
Joerg Roedelff03a072010-09-10 17:30:57 +0200439 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
440 offset * sizeof(u64), sizeof(pdpte),
441 PFERR_USER_MASK|PFERR_WRITE_MASK);
Carsten Ottea03490e2007-10-29 16:09:35 +0100442 if (ret < 0) {
443 ret = 0;
444 goto out;
445 }
446 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
Avi Kivity43a37952009-06-10 14:12:05 +0300447 if (is_present_gpte(pdpte[i]) &&
Dong, Eddie20c466b2009-03-31 23:03:45 +0800448 (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
Carsten Ottea03490e2007-10-29 16:09:35 +0100449 ret = 0;
450 goto out;
451 }
452 }
453 ret = 1;
454
Joerg Roedelff03a072010-09-10 17:30:57 +0200455 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
Avi Kivity6de4f3a2009-05-31 22:58:47 +0300456 __set_bit(VCPU_EXREG_PDPTR,
457 (unsigned long *)&vcpu->arch.regs_avail);
458 __set_bit(VCPU_EXREG_PDPTR,
459 (unsigned long *)&vcpu->arch.regs_dirty);
Carsten Ottea03490e2007-10-29 16:09:35 +0100460out:
Carsten Ottea03490e2007-10-29 16:09:35 +0100461
462 return ret;
463}
Joerg Roedelcc4b6872008-02-07 13:47:43 +0100464EXPORT_SYMBOL_GPL(load_pdptrs);
Carsten Ottea03490e2007-10-29 16:09:35 +0100465
Avi Kivityd835dfe2007-11-21 02:57:59 +0200466static bool pdptrs_changed(struct kvm_vcpu *vcpu)
467{
Joerg Roedelff03a072010-09-10 17:30:57 +0200468 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
Avi Kivityd835dfe2007-11-21 02:57:59 +0200469 bool changed = true;
Joerg Roedel3d06b8b2010-09-10 17:30:53 +0200470 int offset;
471 gfn_t gfn;
Avi Kivityd835dfe2007-11-21 02:57:59 +0200472 int r;
473
474 if (is_long_mode(vcpu) || !is_pae(vcpu))
475 return false;
476
Avi Kivity6de4f3a2009-05-31 22:58:47 +0300477 if (!test_bit(VCPU_EXREG_PDPTR,
478 (unsigned long *)&vcpu->arch.regs_avail))
479 return true;
480
Avi Kivity9f8fe502010-12-05 17:30:00 +0200481 gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT;
482 offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1);
Joerg Roedel3d06b8b2010-09-10 17:30:53 +0200483 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
484 PFERR_USER_MASK | PFERR_WRITE_MASK);
Avi Kivityd835dfe2007-11-21 02:57:59 +0200485 if (r < 0)
486 goto out;
Joerg Roedelff03a072010-09-10 17:30:57 +0200487 changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
Avi Kivityd835dfe2007-11-21 02:57:59 +0200488out:
Avi Kivityd835dfe2007-11-21 02:57:59 +0200489
490 return changed;
491}
492
Avi Kivity49a9b072010-06-10 17:02:14 +0300493int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
Carsten Ottea03490e2007-10-29 16:09:35 +0100494{
Sheng Yangaad82702010-05-12 16:40:42 +0800495 unsigned long old_cr0 = kvm_read_cr0(vcpu);
496 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP |
497 X86_CR0_CD | X86_CR0_NW;
498
Avi Kivityf9a48e62010-01-06 19:10:22 +0200499 cr0 |= X86_CR0_ET;
500
Gleb Natapovab344822010-01-21 15:28:46 +0200501#ifdef CONFIG_X86_64
Gleb Natapov0f122442010-04-28 19:15:31 +0300502 if (cr0 & 0xffffffff00000000UL)
503 return 1;
Gleb Natapovab344822010-01-21 15:28:46 +0200504#endif
505
506 cr0 &= ~CR0_RESERVED_BITS;
Carsten Ottea03490e2007-10-29 16:09:35 +0100507
Gleb Natapov0f122442010-04-28 19:15:31 +0300508 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
509 return 1;
Carsten Ottea03490e2007-10-29 16:09:35 +0100510
Gleb Natapov0f122442010-04-28 19:15:31 +0300511 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
512 return 1;
Carsten Ottea03490e2007-10-29 16:09:35 +0100513
514 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
515#ifdef CONFIG_X86_64
Avi Kivityf6801df2010-01-21 15:31:50 +0200516 if ((vcpu->arch.efer & EFER_LME)) {
Carsten Ottea03490e2007-10-29 16:09:35 +0100517 int cs_db, cs_l;
518
Gleb Natapov0f122442010-04-28 19:15:31 +0300519 if (!is_pae(vcpu))
520 return 1;
Carsten Ottea03490e2007-10-29 16:09:35 +0100521 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
Gleb Natapov0f122442010-04-28 19:15:31 +0300522 if (cs_l)
523 return 1;
Carsten Ottea03490e2007-10-29 16:09:35 +0100524 } else
525#endif
Joerg Roedelff03a072010-09-10 17:30:57 +0200526 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
Avi Kivity9f8fe502010-12-05 17:30:00 +0200527 kvm_read_cr3(vcpu)))
Gleb Natapov0f122442010-04-28 19:15:31 +0300528 return 1;
Carsten Ottea03490e2007-10-29 16:09:35 +0100529 }
530
Mao, Junjiead756a12012-07-02 01:18:48 +0000531 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
532 return 1;
533
Carsten Ottea03490e2007-10-29 16:09:35 +0100534 kvm_x86_ops->set_cr0(vcpu, cr0);
Carsten Ottea03490e2007-10-29 16:09:35 +0100535
Lai Jiangshand170c412011-02-21 11:21:30 +0800536 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
Xiao Guangronge5f3f022010-11-12 14:47:01 +0800537 kvm_clear_async_pf_completion_queue(vcpu);
Lai Jiangshand170c412011-02-21 11:21:30 +0800538 kvm_async_pf_hash_reset(vcpu);
539 }
Xiao Guangronge5f3f022010-11-12 14:47:01 +0800540
Sheng Yangaad82702010-05-12 16:40:42 +0800541 if ((cr0 ^ old_cr0) & update_bits)
542 kvm_mmu_reset_context(vcpu);
Gleb Natapov0f122442010-04-28 19:15:31 +0300543 return 0;
544}
Avi Kivity2d3ad1f2008-02-24 11:20:43 +0200545EXPORT_SYMBOL_GPL(kvm_set_cr0);
Carsten Ottea03490e2007-10-29 16:09:35 +0100546
Avi Kivity2d3ad1f2008-02-24 11:20:43 +0200547void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
Carsten Ottea03490e2007-10-29 16:09:35 +0100548{
Avi Kivity49a9b072010-06-10 17:02:14 +0300549 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
Carsten Ottea03490e2007-10-29 16:09:35 +0100550}
Avi Kivity2d3ad1f2008-02-24 11:20:43 +0200551EXPORT_SYMBOL_GPL(kvm_lmsw);
Carsten Ottea03490e2007-10-29 16:09:35 +0100552
Dexuan Cui2acf9232010-06-10 11:27:12 +0800553int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
554{
555 u64 xcr0;
556
557 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
558 if (index != XCR_XFEATURE_ENABLED_MASK)
559 return 1;
560 xcr0 = xcr;
561 if (kvm_x86_ops->get_cpl(vcpu) != 0)
562 return 1;
563 if (!(xcr0 & XSTATE_FP))
564 return 1;
565 if ((xcr0 & XSTATE_YMM) && !(xcr0 & XSTATE_SSE))
566 return 1;
567 if (xcr0 & ~host_xcr0)
568 return 1;
569 vcpu->arch.xcr0 = xcr0;
570 vcpu->guest_xcr0_loaded = 0;
571 return 0;
572}
573
574int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
575{
576 if (__kvm_set_xcr(vcpu, index, xcr)) {
577 kvm_inject_gp(vcpu, 0);
578 return 1;
579 }
580 return 0;
581}
582EXPORT_SYMBOL_GPL(kvm_set_xcr);
583
Avi Kivitya83b29c2010-06-10 17:02:15 +0300584int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
Carsten Ottea03490e2007-10-29 16:09:35 +0100585{
Avi Kivityfc78f512009-12-07 12:16:48 +0200586 unsigned long old_cr4 = kvm_read_cr4(vcpu);
Yang, Wei Yc68b7342011-06-03 11:13:42 +0800587 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE |
588 X86_CR4_PAE | X86_CR4_SMEP;
Gleb Natapov0f122442010-04-28 19:15:31 +0300589 if (cr4 & CR4_RESERVED_BITS)
590 return 1;
Carsten Ottea03490e2007-10-29 16:09:35 +0100591
Dexuan Cui2acf9232010-06-10 11:27:12 +0800592 if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
593 return 1;
594
Yang, Wei Yc68b7342011-06-03 11:13:42 +0800595 if (!guest_cpuid_has_smep(vcpu) && (cr4 & X86_CR4_SMEP))
596 return 1;
597
Yang, Wei74dc2b42011-06-14 20:10:18 +0800598 if (!guest_cpuid_has_fsgsbase(vcpu) && (cr4 & X86_CR4_RDWRGSFS))
599 return 1;
600
Carsten Ottea03490e2007-10-29 16:09:35 +0100601 if (is_long_mode(vcpu)) {
Gleb Natapov0f122442010-04-28 19:15:31 +0300602 if (!(cr4 & X86_CR4_PAE))
603 return 1;
Avi Kivitya2edf572009-05-24 22:19:00 +0300604 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
605 && ((cr4 ^ old_cr4) & pdptr_bits)
Avi Kivity9f8fe502010-12-05 17:30:00 +0200606 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
607 kvm_read_cr3(vcpu)))
Gleb Natapov0f122442010-04-28 19:15:31 +0300608 return 1;
Carsten Ottea03490e2007-10-29 16:09:35 +0100609
Mao, Junjiead756a12012-07-02 01:18:48 +0000610 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
611 if (!guest_cpuid_has_pcid(vcpu))
612 return 1;
613
614 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
615 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
616 return 1;
617 }
618
Nadav Har'El5e1746d2011-05-25 23:03:24 +0300619 if (kvm_x86_ops->set_cr4(vcpu, cr4))
Gleb Natapov0f122442010-04-28 19:15:31 +0300620 return 1;
621
Mao, Junjiead756a12012-07-02 01:18:48 +0000622 if (((cr4 ^ old_cr4) & pdptr_bits) ||
623 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
Sheng Yangaad82702010-05-12 16:40:42 +0800624 kvm_mmu_reset_context(vcpu);
Gleb Natapov0f122442010-04-28 19:15:31 +0300625
Dexuan Cui2acf9232010-06-10 11:27:12 +0800626 if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE)
Avi Kivity00b27a32011-11-23 16:30:32 +0200627 kvm_update_cpuid(vcpu);
Dexuan Cui2acf9232010-06-10 11:27:12 +0800628
Gleb Natapov0f122442010-04-28 19:15:31 +0300629 return 0;
630}
Avi Kivity2d3ad1f2008-02-24 11:20:43 +0200631EXPORT_SYMBOL_GPL(kvm_set_cr4);
Carsten Ottea03490e2007-10-29 16:09:35 +0100632
Avi Kivity23902182010-06-10 17:02:16 +0300633int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
Carsten Ottea03490e2007-10-29 16:09:35 +0100634{
Avi Kivity9f8fe502010-12-05 17:30:00 +0200635 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
Marcelo Tosatti0ba73cd2008-09-23 13:18:34 -0300636 kvm_mmu_sync_roots(vcpu);
Avi Kivityd835dfe2007-11-21 02:57:59 +0200637 kvm_mmu_flush_tlb(vcpu);
Gleb Natapov0f122442010-04-28 19:15:31 +0300638 return 0;
Avi Kivityd835dfe2007-11-21 02:57:59 +0200639 }
640
Carsten Ottea03490e2007-10-29 16:09:35 +0100641 if (is_long_mode(vcpu)) {
Gleb Natapov471842e2012-10-15 09:49:08 +0200642 if (kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE)) {
Mao, Junjiead756a12012-07-02 01:18:48 +0000643 if (cr3 & CR3_PCID_ENABLED_RESERVED_BITS)
644 return 1;
645 } else
646 if (cr3 & CR3_L_MODE_RESERVED_BITS)
647 return 1;
Carsten Ottea03490e2007-10-29 16:09:35 +0100648 } else {
649 if (is_pae(vcpu)) {
Gleb Natapov0f122442010-04-28 19:15:31 +0300650 if (cr3 & CR3_PAE_RESERVED_BITS)
651 return 1;
Joerg Roedelff03a072010-09-10 17:30:57 +0200652 if (is_paging(vcpu) &&
653 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
Gleb Natapov0f122442010-04-28 19:15:31 +0300654 return 1;
Carsten Ottea03490e2007-10-29 16:09:35 +0100655 }
656 /*
657 * We don't check reserved bits in nonpae mode, because
658 * this isn't enforced, and VMware depends on this.
659 */
660 }
661
Carsten Ottea03490e2007-10-29 16:09:35 +0100662 /*
663 * Does the new cr3 value map to physical memory? (Note, we
664 * catch an invalid cr3 even in real-mode, because it would
665 * cause trouble later on when we turn on paging anyway.)
666 *
667 * A real CPU would silently accept an invalid cr3 and would
668 * attempt to use it - with largely undefined (and often hard
669 * to debug) behavior on the guest side.
670 */
671 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
Gleb Natapov0f122442010-04-28 19:15:31 +0300672 return 1;
673 vcpu->arch.cr3 = cr3;
Avi Kivityaff48ba2010-12-05 18:56:11 +0200674 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
Gleb Natapov0f122442010-04-28 19:15:31 +0300675 vcpu->arch.mmu.new_cr3(vcpu);
676 return 0;
677}
Avi Kivity2d3ad1f2008-02-24 11:20:43 +0200678EXPORT_SYMBOL_GPL(kvm_set_cr3);
Carsten Ottea03490e2007-10-29 16:09:35 +0100679
Andre Przywaraeea1cff2010-12-21 11:12:00 +0100680int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
Carsten Ottea03490e2007-10-29 16:09:35 +0100681{
Gleb Natapov0f122442010-04-28 19:15:31 +0300682 if (cr8 & CR8_RESERVED_BITS)
683 return 1;
Carsten Ottea03490e2007-10-29 16:09:35 +0100684 if (irqchip_in_kernel(vcpu->kvm))
685 kvm_lapic_set_tpr(vcpu, cr8);
686 else
Zhang Xiantaoad312c72007-12-13 23:50:52 +0800687 vcpu->arch.cr8 = cr8;
Gleb Natapov0f122442010-04-28 19:15:31 +0300688 return 0;
689}
Avi Kivity2d3ad1f2008-02-24 11:20:43 +0200690EXPORT_SYMBOL_GPL(kvm_set_cr8);
Carsten Ottea03490e2007-10-29 16:09:35 +0100691
Avi Kivity2d3ad1f2008-02-24 11:20:43 +0200692unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
Carsten Ottea03490e2007-10-29 16:09:35 +0100693{
694 if (irqchip_in_kernel(vcpu->kvm))
695 return kvm_lapic_get_cr8(vcpu);
696 else
Zhang Xiantaoad312c72007-12-13 23:50:52 +0800697 return vcpu->arch.cr8;
Carsten Ottea03490e2007-10-29 16:09:35 +0100698}
Avi Kivity2d3ad1f2008-02-24 11:20:43 +0200699EXPORT_SYMBOL_GPL(kvm_get_cr8);
Carsten Ottea03490e2007-10-29 16:09:35 +0100700
Jan Kiszkac8639012012-09-21 05:42:55 +0200701static void kvm_update_dr7(struct kvm_vcpu *vcpu)
702{
703 unsigned long dr7;
704
705 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
706 dr7 = vcpu->arch.guest_debug_dr7;
707 else
708 dr7 = vcpu->arch.dr7;
709 kvm_x86_ops->set_dr7(vcpu, dr7);
710 vcpu->arch.switch_db_regs = (dr7 & DR7_BP_EN_MASK);
711}
712
Gleb Natapov338dbc92010-04-28 19:15:32 +0300713static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
Gleb Natapov020df072010-04-13 10:05:23 +0300714{
715 switch (dr) {
716 case 0 ... 3:
717 vcpu->arch.db[dr] = val;
718 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
719 vcpu->arch.eff_db[dr] = val;
720 break;
721 case 4:
Gleb Natapov338dbc92010-04-28 19:15:32 +0300722 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
723 return 1; /* #UD */
Gleb Natapov020df072010-04-13 10:05:23 +0300724 /* fall through */
725 case 6:
Gleb Natapov338dbc92010-04-28 19:15:32 +0300726 if (val & 0xffffffff00000000ULL)
727 return -1; /* #GP */
Gleb Natapov020df072010-04-13 10:05:23 +0300728 vcpu->arch.dr6 = (val & DR6_VOLATILE) | DR6_FIXED_1;
729 break;
730 case 5:
Gleb Natapov338dbc92010-04-28 19:15:32 +0300731 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
732 return 1; /* #UD */
Gleb Natapov020df072010-04-13 10:05:23 +0300733 /* fall through */
734 default: /* 7 */
Gleb Natapov338dbc92010-04-28 19:15:32 +0300735 if (val & 0xffffffff00000000ULL)
736 return -1; /* #GP */
Gleb Natapov020df072010-04-13 10:05:23 +0300737 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
Jan Kiszkac8639012012-09-21 05:42:55 +0200738 kvm_update_dr7(vcpu);
Gleb Natapov020df072010-04-13 10:05:23 +0300739 break;
740 }
741
742 return 0;
743}
Gleb Natapov338dbc92010-04-28 19:15:32 +0300744
745int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
746{
747 int res;
748
749 res = __kvm_set_dr(vcpu, dr, val);
750 if (res > 0)
751 kvm_queue_exception(vcpu, UD_VECTOR);
752 else if (res < 0)
753 kvm_inject_gp(vcpu, 0);
754
755 return res;
756}
Gleb Natapov020df072010-04-13 10:05:23 +0300757EXPORT_SYMBOL_GPL(kvm_set_dr);
758
Gleb Natapov338dbc92010-04-28 19:15:32 +0300759static int _kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
Gleb Natapov020df072010-04-13 10:05:23 +0300760{
761 switch (dr) {
762 case 0 ... 3:
763 *val = vcpu->arch.db[dr];
764 break;
765 case 4:
Gleb Natapov338dbc92010-04-28 19:15:32 +0300766 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
Gleb Natapov020df072010-04-13 10:05:23 +0300767 return 1;
Gleb Natapov020df072010-04-13 10:05:23 +0300768 /* fall through */
769 case 6:
770 *val = vcpu->arch.dr6;
771 break;
772 case 5:
Gleb Natapov338dbc92010-04-28 19:15:32 +0300773 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
Gleb Natapov020df072010-04-13 10:05:23 +0300774 return 1;
Gleb Natapov020df072010-04-13 10:05:23 +0300775 /* fall through */
776 default: /* 7 */
777 *val = vcpu->arch.dr7;
778 break;
779 }
780
781 return 0;
782}
Gleb Natapov338dbc92010-04-28 19:15:32 +0300783
784int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
785{
786 if (_kvm_get_dr(vcpu, dr, val)) {
787 kvm_queue_exception(vcpu, UD_VECTOR);
788 return 1;
789 }
790 return 0;
791}
Gleb Natapov020df072010-04-13 10:05:23 +0300792EXPORT_SYMBOL_GPL(kvm_get_dr);
793
Avi Kivity022cd0e2011-11-10 14:57:23 +0200794bool kvm_rdpmc(struct kvm_vcpu *vcpu)
795{
796 u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
797 u64 data;
798 int err;
799
800 err = kvm_pmu_read_pmc(vcpu, ecx, &data);
801 if (err)
802 return err;
803 kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
804 kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
805 return err;
806}
807EXPORT_SYMBOL_GPL(kvm_rdpmc);
808
Carsten Otte043405e2007-10-10 17:16:19 +0200809/*
810 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
811 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
812 *
813 * This list is modified at module load time to reflect the
Glauber Costae3267cb2009-10-06 13:24:50 -0400814 * capabilities of the host cpu. This capabilities test skips MSRs that are
815 * kvm-specific. Those are put in the beginning of the list.
Carsten Otte043405e2007-10-10 17:16:19 +0200816 */
Glauber Costae3267cb2009-10-06 13:24:50 -0400817
Gleb Natapov439793d2012-08-01 17:01:42 +0300818#define KVM_SAVE_MSRS_BEGIN 10
Carsten Otte043405e2007-10-10 17:16:19 +0200819static u32 msrs_to_save[] = {
Glauber Costae3267cb2009-10-06 13:24:50 -0400820 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
Glauber Costa11c6bff2010-05-11 12:17:41 -0400821 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
Gleb Natapov55cd8e52010-01-17 15:51:22 +0200822 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
Glauber Costac9aaa892011-07-11 15:28:14 -0400823 HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
Michael S. Tsirkinae7a2a32012-06-24 19:25:07 +0300824 MSR_KVM_PV_EOI_EN,
Carsten Otte043405e2007-10-10 17:16:19 +0200825 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
Brian Gerst8c065852010-07-17 09:03:26 -0400826 MSR_STAR,
Carsten Otte043405e2007-10-10 17:16:19 +0200827#ifdef CONFIG_X86_64
828 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
829#endif
Avi Kivitye90aa412010-09-01 10:23:35 +0300830 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
Carsten Otte043405e2007-10-10 17:16:19 +0200831};
832
833static unsigned num_msrs_to_save;
834
Mathias Krausef1d24832012-08-30 01:30:18 +0200835static const u32 emulated_msrs[] = {
Will Auldba904632012-11-29 12:42:50 -0800836 MSR_IA32_TSC_ADJUST,
Liu, Jinsonga3e06bb2011-09-22 16:55:52 +0800837 MSR_IA32_TSCDEADLINE,
Carsten Otte043405e2007-10-10 17:16:19 +0200838 MSR_IA32_MISC_ENABLE,
Avi Kivity908e75f2010-07-07 14:09:38 +0300839 MSR_IA32_MCG_STATUS,
840 MSR_IA32_MCG_CTL,
Carsten Otte043405e2007-10-10 17:16:19 +0200841};
842
Roedel, Joergb69e8ca2010-05-06 11:38:43 +0200843static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
Carsten Otte15c4a642007-10-30 18:44:17 +0100844{
Sheng Yangaad82702010-05-12 16:40:42 +0800845 u64 old_efer = vcpu->arch.efer;
846
Roedel, Joergb69e8ca2010-05-06 11:38:43 +0200847 if (efer & efer_reserved_bits)
848 return 1;
Carsten Otte15c4a642007-10-30 18:44:17 +0100849
850 if (is_paging(vcpu)
Roedel, Joergb69e8ca2010-05-06 11:38:43 +0200851 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
852 return 1;
Carsten Otte15c4a642007-10-30 18:44:17 +0100853
Alexander Graf1b2fd702009-02-02 16:23:51 +0100854 if (efer & EFER_FFXSR) {
855 struct kvm_cpuid_entry2 *feat;
856
857 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
Roedel, Joergb69e8ca2010-05-06 11:38:43 +0200858 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
859 return 1;
Alexander Graf1b2fd702009-02-02 16:23:51 +0100860 }
861
Alexander Grafd8017472008-11-25 20:17:11 +0100862 if (efer & EFER_SVME) {
863 struct kvm_cpuid_entry2 *feat;
864
865 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
Roedel, Joergb69e8ca2010-05-06 11:38:43 +0200866 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
867 return 1;
Alexander Grafd8017472008-11-25 20:17:11 +0100868 }
869
Carsten Otte15c4a642007-10-30 18:44:17 +0100870 efer &= ~EFER_LMA;
Avi Kivityf6801df2010-01-21 15:31:50 +0200871 efer |= vcpu->arch.efer & EFER_LMA;
Carsten Otte15c4a642007-10-30 18:44:17 +0100872
Sheng Yanga3d204e2010-05-12 16:40:40 +0800873 kvm_x86_ops->set_efer(vcpu, efer);
874
Avi Kivity9645bb562009-03-31 11:31:54 +0300875 vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled;
Roedel, Joergb69e8ca2010-05-06 11:38:43 +0200876
Sheng Yangaad82702010-05-12 16:40:42 +0800877 /* Update reserved bits */
878 if ((efer ^ old_efer) & EFER_NX)
879 kvm_mmu_reset_context(vcpu);
880
Roedel, Joergb69e8ca2010-05-06 11:38:43 +0200881 return 0;
Carsten Otte15c4a642007-10-30 18:44:17 +0100882}
883
Joerg Roedelf2b4b7d2008-01-31 14:57:37 +0100884void kvm_enable_efer_bits(u64 mask)
885{
886 efer_reserved_bits &= ~mask;
887}
888EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
889
890
Carsten Otte15c4a642007-10-30 18:44:17 +0100891/*
892 * Writes msr value into into the appropriate "register".
893 * Returns 0 on success, non-0 otherwise.
894 * Assumes vcpu_load() was already called.
895 */
Will Auld8fe8ab42012-11-29 12:42:12 -0800896int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
Carsten Otte15c4a642007-10-30 18:44:17 +0100897{
Will Auld8fe8ab42012-11-29 12:42:12 -0800898 return kvm_x86_ops->set_msr(vcpu, msr);
Carsten Otte15c4a642007-10-30 18:44:17 +0100899}
900
Carsten Otte313a3dc2007-10-11 19:16:52 +0200901/*
902 * Adapt set_msr() to msr_io()'s calling convention
903 */
904static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
905{
Will Auld8fe8ab42012-11-29 12:42:12 -0800906 struct msr_data msr;
907
908 msr.data = *data;
909 msr.index = index;
910 msr.host_initiated = true;
911 return kvm_set_msr(vcpu, &msr);
Carsten Otte313a3dc2007-10-11 19:16:52 +0200912}
913
Marcelo Tosatti16e8d742012-11-27 23:29:00 -0200914#ifdef CONFIG_X86_64
915struct pvclock_gtod_data {
916 seqcount_t seq;
917
918 struct { /* extract of a clocksource struct */
919 int vclock_mode;
920 cycle_t cycle_last;
921 cycle_t mask;
922 u32 mult;
923 u32 shift;
924 } clock;
925
926 /* open coded 'struct timespec' */
927 u64 monotonic_time_snsec;
928 time_t monotonic_time_sec;
929};
930
931static struct pvclock_gtod_data pvclock_gtod_data;
932
933static void update_pvclock_gtod(struct timekeeper *tk)
934{
935 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
936
937 write_seqcount_begin(&vdata->seq);
938
939 /* copy pvclock gtod data */
940 vdata->clock.vclock_mode = tk->clock->archdata.vclock_mode;
941 vdata->clock.cycle_last = tk->clock->cycle_last;
942 vdata->clock.mask = tk->clock->mask;
943 vdata->clock.mult = tk->mult;
944 vdata->clock.shift = tk->shift;
945
946 vdata->monotonic_time_sec = tk->xtime_sec
947 + tk->wall_to_monotonic.tv_sec;
948 vdata->monotonic_time_snsec = tk->xtime_nsec
949 + (tk->wall_to_monotonic.tv_nsec
950 << tk->shift);
951 while (vdata->monotonic_time_snsec >=
952 (((u64)NSEC_PER_SEC) << tk->shift)) {
953 vdata->monotonic_time_snsec -=
954 ((u64)NSEC_PER_SEC) << tk->shift;
955 vdata->monotonic_time_sec++;
956 }
957
958 write_seqcount_end(&vdata->seq);
959}
960#endif
961
962
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -0200963static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
964{
Avi Kivity9ed3c442010-05-04 15:00:37 +0300965 int version;
966 int r;
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +0200967 struct pvclock_wall_clock wc;
Jason Wang923de3c2010-01-27 19:13:49 +0800968 struct timespec boot;
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -0200969
970 if (!wall_clock)
971 return;
972
Avi Kivity9ed3c442010-05-04 15:00:37 +0300973 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
974 if (r)
975 return;
976
977 if (version & 1)
978 ++version; /* first time write, random junk */
979
980 ++version;
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -0200981
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -0200982 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
983
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +0200984 /*
985 * The guest calculates current wall clock time by adding
Zachary Amsden34c238a2010-09-18 14:38:14 -1000986 * system time (updated by kvm_guest_time_update below) to the
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +0200987 * wall clock specified here. guest system time equals host
988 * system time for us, thus we must fill in host boot time here.
989 */
Jason Wang923de3c2010-01-27 19:13:49 +0800990 getboottime(&boot);
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +0200991
Bruce Rogers4b648662012-07-20 10:44:24 -0600992 if (kvm->arch.kvmclock_offset) {
993 struct timespec ts = ns_to_timespec(kvm->arch.kvmclock_offset);
994 boot = timespec_sub(boot, ts);
995 }
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +0200996 wc.sec = boot.tv_sec;
997 wc.nsec = boot.tv_nsec;
998 wc.version = version;
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -0200999
1000 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1001
1002 version++;
1003 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001004}
1005
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +02001006static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1007{
1008 uint32_t quotient, remainder;
1009
1010 /* Don't try to replace with do_div(), this one calculates
1011 * "(dividend << 32) / divisor" */
1012 __asm__ ( "divl %4"
1013 : "=a" (quotient), "=d" (remainder)
1014 : "0" (0), "1" (dividend), "r" (divisor) );
1015 return quotient;
1016}
1017
Zachary Amsden5f4e3f82010-09-18 14:38:13 -10001018static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz,
1019 s8 *pshift, u32 *pmultiplier)
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +02001020{
Zachary Amsden5f4e3f82010-09-18 14:38:13 -10001021 uint64_t scaled64;
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +02001022 int32_t shift = 0;
1023 uint64_t tps64;
1024 uint32_t tps32;
1025
Zachary Amsden5f4e3f82010-09-18 14:38:13 -10001026 tps64 = base_khz * 1000LL;
1027 scaled64 = scaled_khz * 1000LL;
Jan Kiszka50933622010-09-26 13:00:53 +02001028 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +02001029 tps64 >>= 1;
1030 shift--;
1031 }
1032
1033 tps32 = (uint32_t)tps64;
Jan Kiszka50933622010-09-26 13:00:53 +02001034 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1035 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
Zachary Amsden5f4e3f82010-09-18 14:38:13 -10001036 scaled64 >>= 1;
1037 else
1038 tps32 <<= 1;
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +02001039 shift++;
1040 }
1041
Zachary Amsden5f4e3f82010-09-18 14:38:13 -10001042 *pshift = shift;
1043 *pmultiplier = div_frac(scaled64, tps32);
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +02001044
Zachary Amsden5f4e3f82010-09-18 14:38:13 -10001045 pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
1046 __func__, base_khz, scaled_khz, shift, *pmultiplier);
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +02001047}
1048
Zachary Amsden759379d2010-08-19 22:07:25 -10001049static inline u64 get_kernel_ns(void)
1050{
1051 struct timespec ts;
1052
1053 WARN_ON(preemptible());
1054 ktime_get_ts(&ts);
1055 monotonic_to_bootbased(&ts);
1056 return timespec_to_ns(&ts);
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001057}
1058
Marcelo Tosattid8281992012-11-27 23:29:01 -02001059#ifdef CONFIG_X86_64
Marcelo Tosatti16e8d742012-11-27 23:29:00 -02001060static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
Marcelo Tosattid8281992012-11-27 23:29:01 -02001061#endif
Marcelo Tosatti16e8d742012-11-27 23:29:00 -02001062
Gerd Hoffmannc8076602009-02-04 17:52:04 +01001063static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
Zachary Amsdenc2855452010-09-18 14:38:15 -10001064unsigned long max_tsc_khz;
Gerd Hoffmannc8076602009-02-04 17:52:04 +01001065
Joerg Roedel857e4092011-03-25 09:44:50 +01001066static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
Zachary Amsden759379d2010-08-19 22:07:25 -10001067{
Zachary Amsdencc578282012-02-03 15:43:50 -02001068 return pvclock_scale_delta(nsec, vcpu->arch.virtual_tsc_mult,
1069 vcpu->arch.virtual_tsc_shift);
Zachary Amsden759379d2010-08-19 22:07:25 -10001070}
1071
Zachary Amsdencc578282012-02-03 15:43:50 -02001072static u32 adjust_tsc_khz(u32 khz, s32 ppm)
Zachary Amsdenc2855452010-09-18 14:38:15 -10001073{
Zachary Amsdencc578282012-02-03 15:43:50 -02001074 u64 v = (u64)khz * (1000000 + ppm);
1075 do_div(v, 1000000);
1076 return v;
1077}
1078
1079static void kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 this_tsc_khz)
1080{
1081 u32 thresh_lo, thresh_hi;
1082 int use_scaling = 0;
1083
Zachary Amsdenc2855452010-09-18 14:38:15 -10001084 /* Compute a scale to convert nanoseconds in TSC cycles */
1085 kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000,
Zachary Amsdencc578282012-02-03 15:43:50 -02001086 &vcpu->arch.virtual_tsc_shift,
1087 &vcpu->arch.virtual_tsc_mult);
1088 vcpu->arch.virtual_tsc_khz = this_tsc_khz;
1089
1090 /*
1091 * Compute the variation in TSC rate which is acceptable
1092 * within the range of tolerance and decide if the
1093 * rate being applied is within that bounds of the hardware
1094 * rate. If so, no scaling or compensation need be done.
1095 */
1096 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1097 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1098 if (this_tsc_khz < thresh_lo || this_tsc_khz > thresh_hi) {
1099 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", this_tsc_khz, thresh_lo, thresh_hi);
1100 use_scaling = 1;
1101 }
1102 kvm_x86_ops->set_tsc_khz(vcpu, this_tsc_khz, use_scaling);
Zachary Amsdenc2855452010-09-18 14:38:15 -10001103}
1104
1105static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1106{
Zachary Amsdene26101b2012-02-03 15:43:57 -02001107 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
Zachary Amsdencc578282012-02-03 15:43:50 -02001108 vcpu->arch.virtual_tsc_mult,
1109 vcpu->arch.virtual_tsc_shift);
Zachary Amsdene26101b2012-02-03 15:43:57 -02001110 tsc += vcpu->arch.this_tsc_write;
Zachary Amsdenc2855452010-09-18 14:38:15 -10001111 return tsc;
1112}
1113
Marcelo Tosattib48aa972012-11-27 23:29:03 -02001114void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1115{
1116#ifdef CONFIG_X86_64
1117 bool vcpus_matched;
1118 bool do_request = false;
1119 struct kvm_arch *ka = &vcpu->kvm->arch;
1120 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1121
1122 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1123 atomic_read(&vcpu->kvm->online_vcpus));
1124
1125 if (vcpus_matched && gtod->clock.vclock_mode == VCLOCK_TSC)
1126 if (!ka->use_master_clock)
1127 do_request = 1;
1128
1129 if (!vcpus_matched && ka->use_master_clock)
1130 do_request = 1;
1131
1132 if (do_request)
1133 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1134
1135 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1136 atomic_read(&vcpu->kvm->online_vcpus),
1137 ka->use_master_clock, gtod->clock.vclock_mode);
1138#endif
1139}
1140
Will Auldba904632012-11-29 12:42:50 -08001141static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1142{
1143 u64 curr_offset = kvm_x86_ops->read_tsc_offset(vcpu);
1144 vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1145}
1146
Will Auld8fe8ab42012-11-29 12:42:12 -08001147void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
Zachary Amsden99e3e302010-08-19 22:07:17 -10001148{
1149 struct kvm *kvm = vcpu->kvm;
Zachary Amsdenf38e0982010-08-19 22:07:20 -10001150 u64 offset, ns, elapsed;
Zachary Amsden99e3e302010-08-19 22:07:17 -10001151 unsigned long flags;
Marcelo Tosatti02626b62012-03-08 18:46:57 -03001152 s64 usdiff;
Marcelo Tosattib48aa972012-11-27 23:29:03 -02001153 bool matched;
Will Auld8fe8ab42012-11-29 12:42:12 -08001154 u64 data = msr->data;
Zachary Amsden99e3e302010-08-19 22:07:17 -10001155
Jan Kiszka038f8c12011-02-04 10:49:11 +01001156 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
Joerg Roedel857e4092011-03-25 09:44:50 +01001157 offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
Zachary Amsden759379d2010-08-19 22:07:25 -10001158 ns = get_kernel_ns();
Zachary Amsdenf38e0982010-08-19 22:07:20 -10001159 elapsed = ns - kvm->arch.last_tsc_nsec;
Zachary Amsden5d3cb0f62012-02-03 15:43:51 -02001160
1161 /* n.b - signed multiplication and division required */
Marcelo Tosatti02626b62012-03-08 18:46:57 -03001162 usdiff = data - kvm->arch.last_tsc_write;
Zachary Amsden5d3cb0f62012-02-03 15:43:51 -02001163#ifdef CONFIG_X86_64
Marcelo Tosatti02626b62012-03-08 18:46:57 -03001164 usdiff = (usdiff * 1000) / vcpu->arch.virtual_tsc_khz;
Zachary Amsden5d3cb0f62012-02-03 15:43:51 -02001165#else
1166 /* do_div() only does unsigned */
1167 asm("idivl %2; xor %%edx, %%edx"
Marcelo Tosatti02626b62012-03-08 18:46:57 -03001168 : "=A"(usdiff)
1169 : "A"(usdiff * 1000), "rm"(vcpu->arch.virtual_tsc_khz));
Zachary Amsden5d3cb0f62012-02-03 15:43:51 -02001170#endif
Marcelo Tosatti02626b62012-03-08 18:46:57 -03001171 do_div(elapsed, 1000);
1172 usdiff -= elapsed;
1173 if (usdiff < 0)
1174 usdiff = -usdiff;
Zachary Amsdenf38e0982010-08-19 22:07:20 -10001175
1176 /*
Zachary Amsden5d3cb0f62012-02-03 15:43:51 -02001177 * Special case: TSC write with a small delta (1 second) of virtual
1178 * cycle time against real time is interpreted as an attempt to
1179 * synchronize the CPU.
1180 *
1181 * For a reliable TSC, we can match TSC offsets, and for an unstable
1182 * TSC, we add elapsed time in this computation. We could let the
1183 * compensation code attempt to catch up if we fall behind, but
1184 * it's better to try to match offsets from the beginning.
1185 */
Marcelo Tosatti02626b62012-03-08 18:46:57 -03001186 if (usdiff < USEC_PER_SEC &&
Zachary Amsden5d3cb0f62012-02-03 15:43:51 -02001187 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
Zachary Amsdenf38e0982010-08-19 22:07:20 -10001188 if (!check_tsc_unstable()) {
Zachary Amsdene26101b2012-02-03 15:43:57 -02001189 offset = kvm->arch.cur_tsc_offset;
Zachary Amsdenf38e0982010-08-19 22:07:20 -10001190 pr_debug("kvm: matched tsc offset for %llu\n", data);
1191 } else {
Joerg Roedel857e4092011-03-25 09:44:50 +01001192 u64 delta = nsec_to_cycles(vcpu, elapsed);
Zachary Amsden5d3cb0f62012-02-03 15:43:51 -02001193 data += delta;
1194 offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
Zachary Amsden759379d2010-08-19 22:07:25 -10001195 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
Zachary Amsdenf38e0982010-08-19 22:07:20 -10001196 }
Marcelo Tosattib48aa972012-11-27 23:29:03 -02001197 matched = true;
Zachary Amsdene26101b2012-02-03 15:43:57 -02001198 } else {
1199 /*
1200 * We split periods of matched TSC writes into generations.
1201 * For each generation, we track the original measured
1202 * nanosecond time, offset, and write, so if TSCs are in
1203 * sync, we can match exact offset, and if not, we can match
Guo Chao4a969982012-06-28 15:17:27 +08001204 * exact software computation in compute_guest_tsc()
Zachary Amsdene26101b2012-02-03 15:43:57 -02001205 *
1206 * These values are tracked in kvm->arch.cur_xxx variables.
1207 */
1208 kvm->arch.cur_tsc_generation++;
1209 kvm->arch.cur_tsc_nsec = ns;
1210 kvm->arch.cur_tsc_write = data;
1211 kvm->arch.cur_tsc_offset = offset;
Marcelo Tosattib48aa972012-11-27 23:29:03 -02001212 matched = false;
Zachary Amsdene26101b2012-02-03 15:43:57 -02001213 pr_debug("kvm: new tsc generation %u, clock %llu\n",
1214 kvm->arch.cur_tsc_generation, data);
Zachary Amsdenf38e0982010-08-19 22:07:20 -10001215 }
Zachary Amsdene26101b2012-02-03 15:43:57 -02001216
1217 /*
1218 * We also track th most recent recorded KHZ, write and time to
1219 * allow the matching interval to be extended at each write.
1220 */
Zachary Amsdenf38e0982010-08-19 22:07:20 -10001221 kvm->arch.last_tsc_nsec = ns;
1222 kvm->arch.last_tsc_write = data;
Zachary Amsden5d3cb0f62012-02-03 15:43:51 -02001223 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
Zachary Amsden99e3e302010-08-19 22:07:17 -10001224
1225 /* Reset of TSC must disable overshoot protection below */
1226 vcpu->arch.hv_clock.tsc_timestamp = 0;
Zachary Amsdenb183aa52012-02-03 15:43:53 -02001227 vcpu->arch.last_guest_tsc = data;
Zachary Amsdene26101b2012-02-03 15:43:57 -02001228
1229 /* Keep track of which generation this VCPU has synchronized to */
1230 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1231 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1232 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1233
Will Auldba904632012-11-29 12:42:50 -08001234 if (guest_cpuid_has_tsc_adjust(vcpu) && !msr->host_initiated)
1235 update_ia32_tsc_adjust_msr(vcpu, offset);
Zachary Amsdene26101b2012-02-03 15:43:57 -02001236 kvm_x86_ops->write_tsc_offset(vcpu, offset);
1237 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
Marcelo Tosattib48aa972012-11-27 23:29:03 -02001238
1239 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1240 if (matched)
1241 kvm->arch.nr_vcpus_matched_tsc++;
1242 else
1243 kvm->arch.nr_vcpus_matched_tsc = 0;
1244
1245 kvm_track_tsc_matching(vcpu);
1246 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
Zachary Amsden99e3e302010-08-19 22:07:17 -10001247}
Zachary Amsdene26101b2012-02-03 15:43:57 -02001248
Zachary Amsden99e3e302010-08-19 22:07:17 -10001249EXPORT_SYMBOL_GPL(kvm_write_tsc);
1250
Marcelo Tosattid8281992012-11-27 23:29:01 -02001251#ifdef CONFIG_X86_64
1252
1253static cycle_t read_tsc(void)
1254{
1255 cycle_t ret;
1256 u64 last;
1257
1258 /*
1259 * Empirically, a fence (of type that depends on the CPU)
1260 * before rdtsc is enough to ensure that rdtsc is ordered
1261 * with respect to loads. The various CPU manuals are unclear
1262 * as to whether rdtsc can be reordered with later loads,
1263 * but no one has ever seen it happen.
1264 */
1265 rdtsc_barrier();
1266 ret = (cycle_t)vget_cycles();
1267
1268 last = pvclock_gtod_data.clock.cycle_last;
1269
1270 if (likely(ret >= last))
1271 return ret;
1272
1273 /*
1274 * GCC likes to generate cmov here, but this branch is extremely
1275 * predictable (it's just a funciton of time and the likely is
1276 * very likely) and there's a data dependence, so force GCC
1277 * to generate a branch instead. I don't barrier() because
1278 * we don't actually need a barrier, and if this function
1279 * ever gets inlined it will generate worse code.
1280 */
1281 asm volatile ("");
1282 return last;
1283}
1284
1285static inline u64 vgettsc(cycle_t *cycle_now)
1286{
1287 long v;
1288 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1289
1290 *cycle_now = read_tsc();
1291
1292 v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask;
1293 return v * gtod->clock.mult;
1294}
1295
1296static int do_monotonic(struct timespec *ts, cycle_t *cycle_now)
1297{
1298 unsigned long seq;
1299 u64 ns;
1300 int mode;
1301 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1302
1303 ts->tv_nsec = 0;
1304 do {
1305 seq = read_seqcount_begin(&gtod->seq);
1306 mode = gtod->clock.vclock_mode;
1307 ts->tv_sec = gtod->monotonic_time_sec;
1308 ns = gtod->monotonic_time_snsec;
1309 ns += vgettsc(cycle_now);
1310 ns >>= gtod->clock.shift;
1311 } while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
1312 timespec_add_ns(ts, ns);
1313
1314 return mode;
1315}
1316
1317/* returns true if host is using tsc clocksource */
1318static bool kvm_get_time_and_clockread(s64 *kernel_ns, cycle_t *cycle_now)
1319{
1320 struct timespec ts;
1321
1322 /* checked again under seqlock below */
1323 if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1324 return false;
1325
1326 if (do_monotonic(&ts, cycle_now) != VCLOCK_TSC)
1327 return false;
1328
1329 monotonic_to_bootbased(&ts);
1330 *kernel_ns = timespec_to_ns(&ts);
1331
1332 return true;
1333}
1334#endif
1335
1336/*
1337 *
Marcelo Tosattib48aa972012-11-27 23:29:03 -02001338 * Assuming a stable TSC across physical CPUS, and a stable TSC
1339 * across virtual CPUs, the following condition is possible.
1340 * Each numbered line represents an event visible to both
Marcelo Tosattid8281992012-11-27 23:29:01 -02001341 * CPUs at the next numbered event.
1342 *
1343 * "timespecX" represents host monotonic time. "tscX" represents
1344 * RDTSC value.
1345 *
1346 * VCPU0 on CPU0 | VCPU1 on CPU1
1347 *
1348 * 1. read timespec0,tsc0
1349 * 2. | timespec1 = timespec0 + N
1350 * | tsc1 = tsc0 + M
1351 * 3. transition to guest | transition to guest
1352 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1353 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
1354 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1355 *
1356 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1357 *
1358 * - ret0 < ret1
1359 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1360 * ...
1361 * - 0 < N - M => M < N
1362 *
1363 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1364 * always the case (the difference between two distinct xtime instances
1365 * might be smaller then the difference between corresponding TSC reads,
1366 * when updating guest vcpus pvclock areas).
1367 *
1368 * To avoid that problem, do not allow visibility of distinct
1369 * system_timestamp/tsc_timestamp values simultaneously: use a master
1370 * copy of host monotonic time values. Update that master copy
1371 * in lockstep.
1372 *
Marcelo Tosattib48aa972012-11-27 23:29:03 -02001373 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
Marcelo Tosattid8281992012-11-27 23:29:01 -02001374 *
1375 */
1376
1377static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1378{
1379#ifdef CONFIG_X86_64
1380 struct kvm_arch *ka = &kvm->arch;
1381 int vclock_mode;
Marcelo Tosattib48aa972012-11-27 23:29:03 -02001382 bool host_tsc_clocksource, vcpus_matched;
1383
1384 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1385 atomic_read(&kvm->online_vcpus));
Marcelo Tosattid8281992012-11-27 23:29:01 -02001386
1387 /*
1388 * If the host uses TSC clock, then passthrough TSC as stable
1389 * to the guest.
1390 */
Marcelo Tosattib48aa972012-11-27 23:29:03 -02001391 host_tsc_clocksource = kvm_get_time_and_clockread(
Marcelo Tosattid8281992012-11-27 23:29:01 -02001392 &ka->master_kernel_ns,
1393 &ka->master_cycle_now);
1394
Marcelo Tosattib48aa972012-11-27 23:29:03 -02001395 ka->use_master_clock = host_tsc_clocksource & vcpus_matched;
1396
Marcelo Tosattid8281992012-11-27 23:29:01 -02001397 if (ka->use_master_clock)
1398 atomic_set(&kvm_guest_has_master_clock, 1);
1399
1400 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
Marcelo Tosattib48aa972012-11-27 23:29:03 -02001401 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1402 vcpus_matched);
Marcelo Tosattid8281992012-11-27 23:29:01 -02001403#endif
1404}
1405
Zachary Amsden34c238a2010-09-18 14:38:14 -10001406static int kvm_guest_time_update(struct kvm_vcpu *v)
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001407{
Marcelo Tosattid8281992012-11-27 23:29:01 -02001408 unsigned long flags, this_tsc_khz;
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001409 struct kvm_vcpu_arch *vcpu = &v->arch;
Marcelo Tosattid8281992012-11-27 23:29:01 -02001410 struct kvm_arch *ka = &v->kvm->arch;
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001411 void *shared_kaddr;
Zachary Amsden1d5f0662010-08-19 22:07:30 -10001412 s64 kernel_ns, max_kernel_ns;
Marcelo Tosattid8281992012-11-27 23:29:01 -02001413 u64 tsc_timestamp, host_tsc;
Marcelo Tosatti78c03372012-11-27 23:28:47 -02001414 struct pvclock_vcpu_time_info *guest_hv_clock;
Marcelo Tosatti51d59c62012-08-03 15:57:49 -03001415 u8 pvclock_flags;
Marcelo Tosattid8281992012-11-27 23:29:01 -02001416 bool use_master_clock;
1417
1418 kernel_ns = 0;
1419 host_tsc = 0;
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +02001420
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001421 /* Keep irq disabled to prevent changes to the clock */
1422 local_irq_save(flags);
Zachary Amsdencc578282012-02-03 15:43:50 -02001423 this_tsc_khz = __get_cpu_var(cpu_tsc_khz);
Zachary Amsden8cfdc002010-08-19 22:07:21 -10001424 if (unlikely(this_tsc_khz == 0)) {
Zachary Amsdenc2855452010-09-18 14:38:15 -10001425 local_irq_restore(flags);
Zachary Amsden34c238a2010-09-18 14:38:14 -10001426 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
Zachary Amsden8cfdc002010-08-19 22:07:21 -10001427 return 1;
1428 }
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001429
Zachary Amsden1d5f0662010-08-19 22:07:30 -10001430 /*
Marcelo Tosattid8281992012-11-27 23:29:01 -02001431 * If the host uses TSC clock, then passthrough TSC as stable
1432 * to the guest.
1433 */
1434 spin_lock(&ka->pvclock_gtod_sync_lock);
1435 use_master_clock = ka->use_master_clock;
1436 if (use_master_clock) {
1437 host_tsc = ka->master_cycle_now;
1438 kernel_ns = ka->master_kernel_ns;
1439 }
1440 spin_unlock(&ka->pvclock_gtod_sync_lock);
1441 if (!use_master_clock) {
1442 host_tsc = native_read_tsc();
1443 kernel_ns = get_kernel_ns();
1444 }
1445
1446 tsc_timestamp = kvm_x86_ops->read_l1_tsc(v, host_tsc);
1447
1448 /*
Zachary Amsdenc2855452010-09-18 14:38:15 -10001449 * We may have to catch up the TSC to match elapsed wall clock
1450 * time for two reasons, even if kvmclock is used.
1451 * 1) CPU could have been running below the maximum TSC rate
1452 * 2) Broken TSC compensation resets the base at each VCPU
1453 * entry to avoid unknown leaps of TSC even when running
1454 * again on the same CPU. This may cause apparent elapsed
1455 * time to disappear, and the guest to stand still or run
1456 * very slowly.
1457 */
1458 if (vcpu->tsc_catchup) {
1459 u64 tsc = compute_guest_tsc(v, kernel_ns);
1460 if (tsc > tsc_timestamp) {
Marcelo Tosattif1e2b262012-02-03 15:43:55 -02001461 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
Zachary Amsdenc2855452010-09-18 14:38:15 -10001462 tsc_timestamp = tsc;
1463 }
1464 }
1465
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001466 local_irq_restore(flags);
1467
Zachary Amsdenc2855452010-09-18 14:38:15 -10001468 if (!vcpu->time_page)
1469 return 0;
1470
1471 /*
Zachary Amsden1d5f0662010-08-19 22:07:30 -10001472 * Time as measured by the TSC may go backwards when resetting the base
1473 * tsc_timestamp. The reason for this is that the TSC resolution is
1474 * higher than the resolution of the other clock scales. Thus, many
1475 * possible measurments of the TSC correspond to one measurement of any
1476 * other clock, and so a spread of values is possible. This is not a
1477 * problem for the computation of the nanosecond clock; with TSC rates
1478 * around 1GHZ, there can only be a few cycles which correspond to one
1479 * nanosecond value, and any path through this code will inevitably
1480 * take longer than that. However, with the kernel_ns value itself,
1481 * the precision may be much lower, down to HZ granularity. If the
1482 * first sampling of TSC against kernel_ns ends in the low part of the
1483 * range, and the second in the high end of the range, we can get:
1484 *
1485 * (TSC - offset_low) * S + kns_old > (TSC - offset_high) * S + kns_new
1486 *
1487 * As the sampling errors potentially range in the thousands of cycles,
1488 * it is possible such a time value has already been observed by the
1489 * guest. To protect against this, we must compute the system time as
1490 * observed by the guest and ensure the new system time is greater.
1491 */
1492 max_kernel_ns = 0;
Zachary Amsdenb183aa52012-02-03 15:43:53 -02001493 if (vcpu->hv_clock.tsc_timestamp) {
Zachary Amsden1d5f0662010-08-19 22:07:30 -10001494 max_kernel_ns = vcpu->last_guest_tsc -
1495 vcpu->hv_clock.tsc_timestamp;
1496 max_kernel_ns = pvclock_scale_delta(max_kernel_ns,
1497 vcpu->hv_clock.tsc_to_system_mul,
1498 vcpu->hv_clock.tsc_shift);
1499 max_kernel_ns += vcpu->last_kernel_ns;
1500 }
1501
Zachary Amsdene48672f2010-08-19 22:07:23 -10001502 if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
Zachary Amsden5f4e3f82010-09-18 14:38:13 -10001503 kvm_get_time_scale(NSEC_PER_SEC / 1000, this_tsc_khz,
1504 &vcpu->hv_clock.tsc_shift,
1505 &vcpu->hv_clock.tsc_to_system_mul);
Zachary Amsdene48672f2010-08-19 22:07:23 -10001506 vcpu->hw_tsc_khz = this_tsc_khz;
Zachary Amsden8cfdc002010-08-19 22:07:21 -10001507 }
1508
Marcelo Tosattid8281992012-11-27 23:29:01 -02001509 /* with a master <monotonic time, tsc value> tuple,
1510 * pvclock clock reads always increase at the (scaled) rate
1511 * of guest TSC - no need to deal with sampling errors.
1512 */
1513 if (!use_master_clock) {
1514 if (max_kernel_ns > kernel_ns)
1515 kernel_ns = max_kernel_ns;
1516 }
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001517 /* With all the info we got, fill in the values */
Zachary Amsden1d5f0662010-08-19 22:07:30 -10001518 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
Zachary Amsden759379d2010-08-19 22:07:25 -10001519 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
Zachary Amsden1d5f0662010-08-19 22:07:30 -10001520 vcpu->last_kernel_ns = kernel_ns;
Zachary Amsden28e46392010-09-18 14:38:12 -10001521 vcpu->last_guest_tsc = tsc_timestamp;
Marcelo Tosatti51d59c62012-08-03 15:57:49 -03001522
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001523 /*
1524 * The interface expects us to write an even number signaling that the
1525 * update is finished. Since the guest won't see the intermediate
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +02001526 * state, we just increase by 2 at the end.
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001527 */
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +02001528 vcpu->hv_clock.version += 2;
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001529
Cong Wang8fd75e12011-11-25 23:14:17 +08001530 shared_kaddr = kmap_atomic(vcpu->time_page);
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001531
Marcelo Tosatti78c03372012-11-27 23:28:47 -02001532 guest_hv_clock = shared_kaddr + vcpu->time_offset;
1533
1534 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1535 pvclock_flags = (guest_hv_clock->flags & PVCLOCK_GUEST_STOPPED);
1536
1537 if (vcpu->pvclock_set_guest_stopped_request) {
1538 pvclock_flags |= PVCLOCK_GUEST_STOPPED;
1539 vcpu->pvclock_set_guest_stopped_request = false;
1540 }
1541
Marcelo Tosattid8281992012-11-27 23:29:01 -02001542 /* If the host uses TSC clocksource, then it is stable */
1543 if (use_master_clock)
1544 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
1545
Marcelo Tosatti78c03372012-11-27 23:28:47 -02001546 vcpu->hv_clock.flags = pvclock_flags;
1547
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001548 memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +02001549 sizeof(vcpu->hv_clock));
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001550
Cong Wang8fd75e12011-11-25 23:14:17 +08001551 kunmap_atomic(shared_kaddr);
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001552
1553 mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
Zachary Amsden8cfdc002010-08-19 22:07:21 -10001554 return 0;
Gerd Hoffmannc8076602009-02-04 17:52:04 +01001555}
1556
Avi Kivity9ba075a2008-05-26 20:06:35 +03001557static bool msr_mtrr_valid(unsigned msr)
1558{
1559 switch (msr) {
1560 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
1561 case MSR_MTRRfix64K_00000:
1562 case MSR_MTRRfix16K_80000:
1563 case MSR_MTRRfix16K_A0000:
1564 case MSR_MTRRfix4K_C0000:
1565 case MSR_MTRRfix4K_C8000:
1566 case MSR_MTRRfix4K_D0000:
1567 case MSR_MTRRfix4K_D8000:
1568 case MSR_MTRRfix4K_E0000:
1569 case MSR_MTRRfix4K_E8000:
1570 case MSR_MTRRfix4K_F0000:
1571 case MSR_MTRRfix4K_F8000:
1572 case MSR_MTRRdefType:
1573 case MSR_IA32_CR_PAT:
1574 return true;
1575 case 0x2f8:
1576 return true;
1577 }
1578 return false;
1579}
1580
Marcelo Tosattid6289b92009-06-22 15:27:56 -03001581static bool valid_pat_type(unsigned t)
1582{
1583 return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
1584}
1585
1586static bool valid_mtrr_type(unsigned t)
1587{
1588 return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
1589}
1590
1591static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1592{
1593 int i;
1594
1595 if (!msr_mtrr_valid(msr))
1596 return false;
1597
1598 if (msr == MSR_IA32_CR_PAT) {
1599 for (i = 0; i < 8; i++)
1600 if (!valid_pat_type((data >> (i * 8)) & 0xff))
1601 return false;
1602 return true;
1603 } else if (msr == MSR_MTRRdefType) {
1604 if (data & ~0xcff)
1605 return false;
1606 return valid_mtrr_type(data & 0xff);
1607 } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
1608 for (i = 0; i < 8 ; i++)
1609 if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
1610 return false;
1611 return true;
1612 }
1613
1614 /* variable MTRRs */
1615 return valid_mtrr_type(data & 0xff);
1616}
1617
Avi Kivity9ba075a2008-05-26 20:06:35 +03001618static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1619{
Sheng Yang0bed3b52008-10-09 16:01:54 +08001620 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1621
Marcelo Tosattid6289b92009-06-22 15:27:56 -03001622 if (!mtrr_valid(vcpu, msr, data))
Avi Kivity9ba075a2008-05-26 20:06:35 +03001623 return 1;
1624
Sheng Yang0bed3b52008-10-09 16:01:54 +08001625 if (msr == MSR_MTRRdefType) {
1626 vcpu->arch.mtrr_state.def_type = data;
1627 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
1628 } else if (msr == MSR_MTRRfix64K_00000)
1629 p[0] = data;
1630 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1631 p[1 + msr - MSR_MTRRfix16K_80000] = data;
1632 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1633 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
1634 else if (msr == MSR_IA32_CR_PAT)
1635 vcpu->arch.pat = data;
1636 else { /* Variable MTRRs */
1637 int idx, is_mtrr_mask;
1638 u64 *pt;
1639
1640 idx = (msr - 0x200) / 2;
1641 is_mtrr_mask = msr - 0x200 - 2 * idx;
1642 if (!is_mtrr_mask)
1643 pt =
1644 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1645 else
1646 pt =
1647 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1648 *pt = data;
1649 }
1650
1651 kvm_mmu_reset_context(vcpu);
Avi Kivity9ba075a2008-05-26 20:06:35 +03001652 return 0;
1653}
Carsten Otte15c4a642007-10-30 18:44:17 +01001654
Huang Ying890ca9a2009-05-11 16:48:15 +08001655static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1656{
1657 u64 mcg_cap = vcpu->arch.mcg_cap;
1658 unsigned bank_num = mcg_cap & 0xff;
1659
1660 switch (msr) {
1661 case MSR_IA32_MCG_STATUS:
1662 vcpu->arch.mcg_status = data;
1663 break;
1664 case MSR_IA32_MCG_CTL:
1665 if (!(mcg_cap & MCG_CTL_P))
1666 return 1;
1667 if (data != 0 && data != ~(u64)0)
1668 return -1;
1669 vcpu->arch.mcg_ctl = data;
1670 break;
1671 default:
1672 if (msr >= MSR_IA32_MC0_CTL &&
1673 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1674 u32 offset = msr - MSR_IA32_MC0_CTL;
Andre Przywara114be422010-03-24 17:46:42 +01001675 /* only 0 or all 1s can be written to IA32_MCi_CTL
1676 * some Linux kernels though clear bit 10 in bank 4 to
1677 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1678 * this to avoid an uncatched #GP in the guest
1679 */
Huang Ying890ca9a2009-05-11 16:48:15 +08001680 if ((offset & 0x3) == 0 &&
Andre Przywara114be422010-03-24 17:46:42 +01001681 data != 0 && (data | (1 << 10)) != ~(u64)0)
Huang Ying890ca9a2009-05-11 16:48:15 +08001682 return -1;
1683 vcpu->arch.mce_banks[offset] = data;
1684 break;
1685 }
1686 return 1;
1687 }
1688 return 0;
1689}
1690
Ed Swierkffde22a2009-10-15 15:21:43 -07001691static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1692{
1693 struct kvm *kvm = vcpu->kvm;
1694 int lm = is_long_mode(vcpu);
1695 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1696 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1697 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1698 : kvm->arch.xen_hvm_config.blob_size_32;
1699 u32 page_num = data & ~PAGE_MASK;
1700 u64 page_addr = data & PAGE_MASK;
1701 u8 *page;
1702 int r;
1703
1704 r = -E2BIG;
1705 if (page_num >= blob_size)
1706 goto out;
1707 r = -ENOMEM;
Sasha Levinff5c2c02011-12-04 19:36:29 +02001708 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
1709 if (IS_ERR(page)) {
1710 r = PTR_ERR(page);
Ed Swierkffde22a2009-10-15 15:21:43 -07001711 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02001712 }
Ed Swierkffde22a2009-10-15 15:21:43 -07001713 if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
1714 goto out_free;
1715 r = 0;
1716out_free:
1717 kfree(page);
1718out:
1719 return r;
1720}
1721
Gleb Natapov55cd8e52010-01-17 15:51:22 +02001722static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1723{
1724 return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
1725}
1726
1727static bool kvm_hv_msr_partition_wide(u32 msr)
1728{
1729 bool r = false;
1730 switch (msr) {
1731 case HV_X64_MSR_GUEST_OS_ID:
1732 case HV_X64_MSR_HYPERCALL:
1733 r = true;
1734 break;
1735 }
1736
1737 return r;
1738}
1739
1740static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1741{
1742 struct kvm *kvm = vcpu->kvm;
1743
1744 switch (msr) {
1745 case HV_X64_MSR_GUEST_OS_ID:
1746 kvm->arch.hv_guest_os_id = data;
1747 /* setting guest os id to zero disables hypercall page */
1748 if (!kvm->arch.hv_guest_os_id)
1749 kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1750 break;
1751 case HV_X64_MSR_HYPERCALL: {
1752 u64 gfn;
1753 unsigned long addr;
1754 u8 instructions[4];
1755
1756 /* if guest os id is not set hypercall should remain disabled */
1757 if (!kvm->arch.hv_guest_os_id)
1758 break;
1759 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1760 kvm->arch.hv_hypercall = data;
1761 break;
1762 }
1763 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1764 addr = gfn_to_hva(kvm, gfn);
1765 if (kvm_is_error_hva(addr))
1766 return 1;
1767 kvm_x86_ops->patch_hypercall(vcpu, instructions);
1768 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
Xiao Guangrong8b0cedf2011-05-15 23:22:04 +08001769 if (__copy_to_user((void __user *)addr, instructions, 4))
Gleb Natapov55cd8e52010-01-17 15:51:22 +02001770 return 1;
1771 kvm->arch.hv_hypercall = data;
1772 break;
1773 }
1774 default:
Christoffer Dalla737f252012-06-03 21:17:48 +03001775 vcpu_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1776 "data 0x%llx\n", msr, data);
Gleb Natapov55cd8e52010-01-17 15:51:22 +02001777 return 1;
1778 }
1779 return 0;
1780}
1781
1782static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1783{
Gleb Natapov10388a02010-01-17 15:51:23 +02001784 switch (msr) {
1785 case HV_X64_MSR_APIC_ASSIST_PAGE: {
1786 unsigned long addr;
Gleb Natapov55cd8e52010-01-17 15:51:22 +02001787
Gleb Natapov10388a02010-01-17 15:51:23 +02001788 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1789 vcpu->arch.hv_vapic = data;
1790 break;
1791 }
1792 addr = gfn_to_hva(vcpu->kvm, data >>
1793 HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT);
1794 if (kvm_is_error_hva(addr))
1795 return 1;
Xiao Guangrong8b0cedf2011-05-15 23:22:04 +08001796 if (__clear_user((void __user *)addr, PAGE_SIZE))
Gleb Natapov10388a02010-01-17 15:51:23 +02001797 return 1;
1798 vcpu->arch.hv_vapic = data;
1799 break;
1800 }
1801 case HV_X64_MSR_EOI:
1802 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1803 case HV_X64_MSR_ICR:
1804 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1805 case HV_X64_MSR_TPR:
1806 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1807 default:
Christoffer Dalla737f252012-06-03 21:17:48 +03001808 vcpu_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1809 "data 0x%llx\n", msr, data);
Gleb Natapov10388a02010-01-17 15:51:23 +02001810 return 1;
1811 }
1812
1813 return 0;
Gleb Natapov55cd8e52010-01-17 15:51:22 +02001814}
1815
Gleb Natapov344d9582010-10-14 11:22:50 +02001816static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
1817{
1818 gpa_t gpa = data & ~0x3f;
1819
Guo Chao4a969982012-06-28 15:17:27 +08001820 /* Bits 2:5 are reserved, Should be zero */
Gleb Natapov6adba522010-10-14 11:22:55 +02001821 if (data & 0x3c)
Gleb Natapov344d9582010-10-14 11:22:50 +02001822 return 1;
1823
1824 vcpu->arch.apf.msr_val = data;
1825
1826 if (!(data & KVM_ASYNC_PF_ENABLED)) {
1827 kvm_clear_async_pf_completion_queue(vcpu);
1828 kvm_async_pf_hash_reset(vcpu);
1829 return 0;
1830 }
1831
1832 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa))
1833 return 1;
1834
Gleb Natapov6adba522010-10-14 11:22:55 +02001835 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
Gleb Natapov344d9582010-10-14 11:22:50 +02001836 kvm_async_pf_wakeup_all(vcpu);
1837 return 0;
1838}
1839
Glauber Costa12f9a482011-02-01 14:16:40 -05001840static void kvmclock_reset(struct kvm_vcpu *vcpu)
1841{
1842 if (vcpu->arch.time_page) {
1843 kvm_release_page_dirty(vcpu->arch.time_page);
1844 vcpu->arch.time_page = NULL;
1845 }
1846}
1847
Glauber Costac9aaa892011-07-11 15:28:14 -04001848static void accumulate_steal_time(struct kvm_vcpu *vcpu)
1849{
1850 u64 delta;
1851
1852 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
1853 return;
1854
1855 delta = current->sched_info.run_delay - vcpu->arch.st.last_steal;
1856 vcpu->arch.st.last_steal = current->sched_info.run_delay;
1857 vcpu->arch.st.accum_steal = delta;
1858}
1859
1860static void record_steal_time(struct kvm_vcpu *vcpu)
1861{
1862 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
1863 return;
1864
1865 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
1866 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
1867 return;
1868
1869 vcpu->arch.st.steal.steal += vcpu->arch.st.accum_steal;
1870 vcpu->arch.st.steal.version += 2;
1871 vcpu->arch.st.accum_steal = 0;
1872
1873 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
1874 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
1875}
1876
Will Auld8fe8ab42012-11-29 12:42:12 -08001877int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
Carsten Otte15c4a642007-10-30 18:44:17 +01001878{
Gleb Natapov57537852012-01-15 14:17:22 +02001879 bool pr = false;
Will Auld8fe8ab42012-11-29 12:42:12 -08001880 u32 msr = msr_info->index;
1881 u64 data = msr_info->data;
Gleb Natapov57537852012-01-15 14:17:22 +02001882
Carsten Otte15c4a642007-10-30 18:44:17 +01001883 switch (msr) {
Borislav Petkov2e32b712013-02-19 19:33:13 +01001884 case MSR_AMD64_NB_CFG:
1885 case MSR_IA32_UCODE_REV:
1886 case MSR_IA32_UCODE_WRITE:
1887 case MSR_VM_HSAVE_PA:
1888 case MSR_AMD64_PATCH_LOADER:
1889 case MSR_AMD64_BU_CFG2:
1890 break;
1891
Carsten Otte15c4a642007-10-30 18:44:17 +01001892 case MSR_EFER:
Roedel, Joergb69e8ca2010-05-06 11:38:43 +02001893 return set_efer(vcpu, data);
Andre Przywara8f1589d2009-06-24 12:44:33 +02001894 case MSR_K7_HWCR:
1895 data &= ~(u64)0x40; /* ignore flush filter disable */
Joerg Roedel82494022010-02-24 18:59:16 +01001896 data &= ~(u64)0x100; /* ignore ignne emulation enable */
Nicolae Mogoreanua223c312012-02-21 13:44:21 -08001897 data &= ~(u64)0x8; /* ignore TLB cache disable */
Andre Przywara8f1589d2009-06-24 12:44:33 +02001898 if (data != 0) {
Christoffer Dalla737f252012-06-03 21:17:48 +03001899 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
1900 data);
Andre Przywara8f1589d2009-06-24 12:44:33 +02001901 return 1;
1902 }
Carsten Otte15c4a642007-10-30 18:44:17 +01001903 break;
Andre Przywaraf7c6d142009-07-02 15:04:14 +02001904 case MSR_FAM10H_MMIO_CONF_BASE:
1905 if (data != 0) {
Christoffer Dalla737f252012-06-03 21:17:48 +03001906 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
1907 "0x%llx\n", data);
Andre Przywaraf7c6d142009-07-02 15:04:14 +02001908 return 1;
1909 }
Carsten Otte15c4a642007-10-30 18:44:17 +01001910 break;
Alexander Grafb5e2fec2008-07-22 08:00:45 +02001911 case MSR_IA32_DEBUGCTLMSR:
1912 if (!data) {
1913 /* We support the non-activated case already */
1914 break;
1915 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
1916 /* Values other than LBR and BTF are vendor-specific,
1917 thus reserved and should throw a #GP */
1918 return 1;
1919 }
Christoffer Dalla737f252012-06-03 21:17:48 +03001920 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1921 __func__, data);
Alexander Grafb5e2fec2008-07-22 08:00:45 +02001922 break;
Avi Kivity9ba075a2008-05-26 20:06:35 +03001923 case 0x200 ... 0x2ff:
1924 return set_msr_mtrr(vcpu, msr, data);
Carsten Otte15c4a642007-10-30 18:44:17 +01001925 case MSR_IA32_APICBASE:
1926 kvm_set_apic_base(vcpu, data);
1927 break;
Gleb Natapov0105d1a2009-07-05 17:39:36 +03001928 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1929 return kvm_x2apic_msr_write(vcpu, msr, data);
Liu, Jinsonga3e06bb2011-09-22 16:55:52 +08001930 case MSR_IA32_TSCDEADLINE:
1931 kvm_set_lapic_tscdeadline_msr(vcpu, data);
1932 break;
Will Auldba904632012-11-29 12:42:50 -08001933 case MSR_IA32_TSC_ADJUST:
1934 if (guest_cpuid_has_tsc_adjust(vcpu)) {
1935 if (!msr_info->host_initiated) {
1936 u64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
1937 kvm_x86_ops->adjust_tsc_offset(vcpu, adj, true);
1938 }
1939 vcpu->arch.ia32_tsc_adjust_msr = data;
1940 }
1941 break;
Carsten Otte15c4a642007-10-30 18:44:17 +01001942 case MSR_IA32_MISC_ENABLE:
Zhang Xiantaoad312c72007-12-13 23:50:52 +08001943 vcpu->arch.ia32_misc_enable_msr = data;
Carsten Otte15c4a642007-10-30 18:44:17 +01001944 break;
Glauber Costa11c6bff2010-05-11 12:17:41 -04001945 case MSR_KVM_WALL_CLOCK_NEW:
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001946 case MSR_KVM_WALL_CLOCK:
1947 vcpu->kvm->arch.wall_clock = data;
1948 kvm_write_wall_clock(vcpu->kvm, data);
1949 break;
Glauber Costa11c6bff2010-05-11 12:17:41 -04001950 case MSR_KVM_SYSTEM_TIME_NEW:
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001951 case MSR_KVM_SYSTEM_TIME: {
Glauber Costa12f9a482011-02-01 14:16:40 -05001952 kvmclock_reset(vcpu);
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001953
1954 vcpu->arch.time = data;
Zachary Amsdenc2855452010-09-18 14:38:15 -10001955 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001956
1957 /* we verify if the enable bit is set... */
1958 if (!(data & 1))
1959 break;
1960
1961 /* ...but clean it before doing the actual write */
1962 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
1963
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001964 vcpu->arch.time_page =
1965 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001966
Xiao Guangrong32cad842012-08-03 15:42:52 +08001967 if (is_error_page(vcpu->arch.time_page))
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001968 vcpu->arch.time_page = NULL;
Xiao Guangrong32cad842012-08-03 15:42:52 +08001969
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001970 break;
1971 }
Gleb Natapov344d9582010-10-14 11:22:50 +02001972 case MSR_KVM_ASYNC_PF_EN:
1973 if (kvm_pv_enable_async_pf(vcpu, data))
1974 return 1;
1975 break;
Glauber Costac9aaa892011-07-11 15:28:14 -04001976 case MSR_KVM_STEAL_TIME:
1977
1978 if (unlikely(!sched_info_on()))
1979 return 1;
1980
1981 if (data & KVM_STEAL_RESERVED_MASK)
1982 return 1;
1983
1984 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
1985 data & KVM_STEAL_VALID_BITS))
1986 return 1;
1987
1988 vcpu->arch.st.msr_val = data;
1989
1990 if (!(data & KVM_MSR_ENABLED))
1991 break;
1992
1993 vcpu->arch.st.last_steal = current->sched_info.run_delay;
1994
1995 preempt_disable();
1996 accumulate_steal_time(vcpu);
1997 preempt_enable();
1998
1999 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2000
2001 break;
Michael S. Tsirkinae7a2a32012-06-24 19:25:07 +03002002 case MSR_KVM_PV_EOI_EN:
2003 if (kvm_lapic_enable_pv_eoi(vcpu, data))
2004 return 1;
2005 break;
Glauber Costac9aaa892011-07-11 15:28:14 -04002006
Huang Ying890ca9a2009-05-11 16:48:15 +08002007 case MSR_IA32_MCG_CTL:
2008 case MSR_IA32_MCG_STATUS:
2009 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
2010 return set_msr_mce(vcpu, msr, data);
Andre Przywara71db6022009-06-12 22:01:29 +02002011
2012 /* Performance counters are not protected by a CPUID bit,
2013 * so we should check all of them in the generic path for the sake of
2014 * cross vendor migration.
2015 * Writing a zero into the event select MSRs disables them,
2016 * which we perfectly emulate ;-). Any other value should be at least
2017 * reported, some guests depend on them.
2018 */
Andre Przywara71db6022009-06-12 22:01:29 +02002019 case MSR_K7_EVNTSEL0:
2020 case MSR_K7_EVNTSEL1:
2021 case MSR_K7_EVNTSEL2:
2022 case MSR_K7_EVNTSEL3:
2023 if (data != 0)
Christoffer Dalla737f252012-06-03 21:17:48 +03002024 vcpu_unimpl(vcpu, "unimplemented perfctr wrmsr: "
2025 "0x%x data 0x%llx\n", msr, data);
Andre Przywara71db6022009-06-12 22:01:29 +02002026 break;
2027 /* at least RHEL 4 unconditionally writes to the perfctr registers,
2028 * so we ignore writes to make it happy.
2029 */
Andre Przywara71db6022009-06-12 22:01:29 +02002030 case MSR_K7_PERFCTR0:
2031 case MSR_K7_PERFCTR1:
2032 case MSR_K7_PERFCTR2:
2033 case MSR_K7_PERFCTR3:
Christoffer Dalla737f252012-06-03 21:17:48 +03002034 vcpu_unimpl(vcpu, "unimplemented perfctr wrmsr: "
2035 "0x%x data 0x%llx\n", msr, data);
Andre Przywara71db6022009-06-12 22:01:29 +02002036 break;
Gleb Natapov57537852012-01-15 14:17:22 +02002037 case MSR_P6_PERFCTR0:
2038 case MSR_P6_PERFCTR1:
2039 pr = true;
2040 case MSR_P6_EVNTSEL0:
2041 case MSR_P6_EVNTSEL1:
2042 if (kvm_pmu_msr(vcpu, msr))
2043 return kvm_pmu_set_msr(vcpu, msr, data);
2044
2045 if (pr || data != 0)
Christoffer Dalla737f252012-06-03 21:17:48 +03002046 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2047 "0x%x data 0x%llx\n", msr, data);
Gleb Natapov57537852012-01-15 14:17:22 +02002048 break;
Jes Sorensen84e0cef2010-09-01 11:42:04 +02002049 case MSR_K7_CLK_CTL:
2050 /*
2051 * Ignore all writes to this no longer documented MSR.
2052 * Writes are only relevant for old K7 processors,
2053 * all pre-dating SVM, but a recommended workaround from
Guo Chao4a969982012-06-28 15:17:27 +08002054 * AMD for these chips. It is possible to specify the
Jes Sorensen84e0cef2010-09-01 11:42:04 +02002055 * affected processor models on the command line, hence
2056 * the need to ignore the workaround.
2057 */
2058 break;
Gleb Natapov55cd8e52010-01-17 15:51:22 +02002059 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2060 if (kvm_hv_msr_partition_wide(msr)) {
2061 int r;
2062 mutex_lock(&vcpu->kvm->lock);
2063 r = set_msr_hyperv_pw(vcpu, msr, data);
2064 mutex_unlock(&vcpu->kvm->lock);
2065 return r;
2066 } else
2067 return set_msr_hyperv(vcpu, msr, data);
2068 break;
john cooper91c9c3e2011-01-21 00:21:00 -05002069 case MSR_IA32_BBL_CR_CTL3:
2070 /* Drop writes to this legacy MSR -- see rdmsr
2071 * counterpart for further detail.
2072 */
Christoffer Dalla737f252012-06-03 21:17:48 +03002073 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
john cooper91c9c3e2011-01-21 00:21:00 -05002074 break;
Boris Ostrovsky2b036c62012-01-09 14:00:35 -05002075 case MSR_AMD64_OSVW_ID_LENGTH:
2076 if (!guest_cpuid_has_osvw(vcpu))
2077 return 1;
2078 vcpu->arch.osvw.length = data;
2079 break;
2080 case MSR_AMD64_OSVW_STATUS:
2081 if (!guest_cpuid_has_osvw(vcpu))
2082 return 1;
2083 vcpu->arch.osvw.status = data;
2084 break;
Carsten Otte15c4a642007-10-30 18:44:17 +01002085 default:
Ed Swierkffde22a2009-10-15 15:21:43 -07002086 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2087 return xen_hvm_config(vcpu, data);
Gleb Natapovf5132b02011-11-10 14:57:22 +02002088 if (kvm_pmu_msr(vcpu, msr))
2089 return kvm_pmu_set_msr(vcpu, msr, data);
Andre Przywaraed85c062009-06-25 12:36:49 +02002090 if (!ignore_msrs) {
Christoffer Dalla737f252012-06-03 21:17:48 +03002091 vcpu_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
2092 msr, data);
Andre Przywaraed85c062009-06-25 12:36:49 +02002093 return 1;
2094 } else {
Christoffer Dalla737f252012-06-03 21:17:48 +03002095 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
2096 msr, data);
Andre Przywaraed85c062009-06-25 12:36:49 +02002097 break;
2098 }
Carsten Otte15c4a642007-10-30 18:44:17 +01002099 }
2100 return 0;
2101}
2102EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2103
2104
2105/*
2106 * Reads an msr value (of 'msr_index') into 'pdata'.
2107 * Returns 0 on success, non-0 otherwise.
2108 * Assumes vcpu_load() was already called.
2109 */
2110int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
2111{
2112 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
2113}
2114
Avi Kivity9ba075a2008-05-26 20:06:35 +03002115static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2116{
Sheng Yang0bed3b52008-10-09 16:01:54 +08002117 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
2118
Avi Kivity9ba075a2008-05-26 20:06:35 +03002119 if (!msr_mtrr_valid(msr))
2120 return 1;
2121
Sheng Yang0bed3b52008-10-09 16:01:54 +08002122 if (msr == MSR_MTRRdefType)
2123 *pdata = vcpu->arch.mtrr_state.def_type +
2124 (vcpu->arch.mtrr_state.enabled << 10);
2125 else if (msr == MSR_MTRRfix64K_00000)
2126 *pdata = p[0];
2127 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
2128 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
2129 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
2130 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
2131 else if (msr == MSR_IA32_CR_PAT)
2132 *pdata = vcpu->arch.pat;
2133 else { /* Variable MTRRs */
2134 int idx, is_mtrr_mask;
2135 u64 *pt;
2136
2137 idx = (msr - 0x200) / 2;
2138 is_mtrr_mask = msr - 0x200 - 2 * idx;
2139 if (!is_mtrr_mask)
2140 pt =
2141 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
2142 else
2143 pt =
2144 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
2145 *pdata = *pt;
2146 }
2147
Avi Kivity9ba075a2008-05-26 20:06:35 +03002148 return 0;
2149}
2150
Huang Ying890ca9a2009-05-11 16:48:15 +08002151static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2152{
2153 u64 data;
2154 u64 mcg_cap = vcpu->arch.mcg_cap;
2155 unsigned bank_num = mcg_cap & 0xff;
2156
2157 switch (msr) {
2158 case MSR_IA32_P5_MC_ADDR:
2159 case MSR_IA32_P5_MC_TYPE:
2160 data = 0;
2161 break;
2162 case MSR_IA32_MCG_CAP:
2163 data = vcpu->arch.mcg_cap;
2164 break;
2165 case MSR_IA32_MCG_CTL:
2166 if (!(mcg_cap & MCG_CTL_P))
2167 return 1;
2168 data = vcpu->arch.mcg_ctl;
2169 break;
2170 case MSR_IA32_MCG_STATUS:
2171 data = vcpu->arch.mcg_status;
2172 break;
2173 default:
2174 if (msr >= MSR_IA32_MC0_CTL &&
2175 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
2176 u32 offset = msr - MSR_IA32_MC0_CTL;
2177 data = vcpu->arch.mce_banks[offset];
2178 break;
2179 }
2180 return 1;
2181 }
2182 *pdata = data;
2183 return 0;
2184}
2185
Gleb Natapov55cd8e52010-01-17 15:51:22 +02002186static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2187{
2188 u64 data = 0;
2189 struct kvm *kvm = vcpu->kvm;
2190
2191 switch (msr) {
2192 case HV_X64_MSR_GUEST_OS_ID:
2193 data = kvm->arch.hv_guest_os_id;
2194 break;
2195 case HV_X64_MSR_HYPERCALL:
2196 data = kvm->arch.hv_hypercall;
2197 break;
2198 default:
Christoffer Dalla737f252012-06-03 21:17:48 +03002199 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
Gleb Natapov55cd8e52010-01-17 15:51:22 +02002200 return 1;
2201 }
2202
2203 *pdata = data;
2204 return 0;
2205}
2206
2207static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2208{
2209 u64 data = 0;
2210
2211 switch (msr) {
2212 case HV_X64_MSR_VP_INDEX: {
2213 int r;
2214 struct kvm_vcpu *v;
2215 kvm_for_each_vcpu(r, v, vcpu->kvm)
2216 if (v == vcpu)
2217 data = r;
2218 break;
2219 }
Gleb Natapov10388a02010-01-17 15:51:23 +02002220 case HV_X64_MSR_EOI:
2221 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
2222 case HV_X64_MSR_ICR:
2223 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
2224 case HV_X64_MSR_TPR:
2225 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
Mike Waychison14fa67e2011-07-21 15:38:10 -07002226 case HV_X64_MSR_APIC_ASSIST_PAGE:
Mike Waychisond1613ad2011-07-23 00:31:45 -07002227 data = vcpu->arch.hv_vapic;
2228 break;
Gleb Natapov55cd8e52010-01-17 15:51:22 +02002229 default:
Christoffer Dalla737f252012-06-03 21:17:48 +03002230 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
Gleb Natapov55cd8e52010-01-17 15:51:22 +02002231 return 1;
2232 }
2233 *pdata = data;
2234 return 0;
2235}
2236
Carsten Otte15c4a642007-10-30 18:44:17 +01002237int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2238{
2239 u64 data;
2240
2241 switch (msr) {
Carsten Otte15c4a642007-10-30 18:44:17 +01002242 case MSR_IA32_PLATFORM_ID:
Carsten Otte15c4a642007-10-30 18:44:17 +01002243 case MSR_IA32_EBL_CR_POWERON:
Alexander Grafb5e2fec2008-07-22 08:00:45 +02002244 case MSR_IA32_DEBUGCTLMSR:
2245 case MSR_IA32_LASTBRANCHFROMIP:
2246 case MSR_IA32_LASTBRANCHTOIP:
2247 case MSR_IA32_LASTINTFROMIP:
2248 case MSR_IA32_LASTINTTOIP:
Jaswinder Singh Rajput60af2ec2009-05-14 11:00:10 +05302249 case MSR_K8_SYSCFG:
2250 case MSR_K7_HWCR:
Avi Kivity61a6bd62008-12-29 17:32:28 +02002251 case MSR_VM_HSAVE_PA:
Amit Shah9e699622009-06-15 13:25:34 +05302252 case MSR_K7_EVNTSEL0:
Amit Shah1f3ee612009-06-30 16:24:28 +05302253 case MSR_K7_PERFCTR0:
Andre Przywara1fdbd482009-06-24 12:44:34 +02002254 case MSR_K8_INT_PENDING_MSG:
Andre Przywarac323c0e2009-06-24 15:37:05 +02002255 case MSR_AMD64_NB_CFG:
Andre Przywaraf7c6d142009-07-02 15:04:14 +02002256 case MSR_FAM10H_MMIO_CONF_BASE:
Borislav Petkov2e32b712013-02-19 19:33:13 +01002257 case MSR_AMD64_BU_CFG2:
Carsten Otte15c4a642007-10-30 18:44:17 +01002258 data = 0;
2259 break;
Gleb Natapov57537852012-01-15 14:17:22 +02002260 case MSR_P6_PERFCTR0:
2261 case MSR_P6_PERFCTR1:
2262 case MSR_P6_EVNTSEL0:
2263 case MSR_P6_EVNTSEL1:
2264 if (kvm_pmu_msr(vcpu, msr))
2265 return kvm_pmu_get_msr(vcpu, msr, pdata);
2266 data = 0;
2267 break;
Marcelo Tosatti742bc672011-07-29 19:44:21 -03002268 case MSR_IA32_UCODE_REV:
2269 data = 0x100000000ULL;
2270 break;
Avi Kivity9ba075a2008-05-26 20:06:35 +03002271 case MSR_MTRRcap:
2272 data = 0x500 | KVM_NR_VAR_MTRR;
2273 break;
2274 case 0x200 ... 0x2ff:
2275 return get_msr_mtrr(vcpu, msr, pdata);
Carsten Otte15c4a642007-10-30 18:44:17 +01002276 case 0xcd: /* fsb frequency */
2277 data = 3;
2278 break;
Jes Sorensen7b914092010-09-09 12:06:46 +02002279 /*
2280 * MSR_EBC_FREQUENCY_ID
2281 * Conservative value valid for even the basic CPU models.
2282 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2283 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2284 * and 266MHz for model 3, or 4. Set Core Clock
2285 * Frequency to System Bus Frequency Ratio to 1 (bits
2286 * 31:24) even though these are only valid for CPU
2287 * models > 2, however guests may end up dividing or
2288 * multiplying by zero otherwise.
2289 */
2290 case MSR_EBC_FREQUENCY_ID:
2291 data = 1 << 24;
2292 break;
Carsten Otte15c4a642007-10-30 18:44:17 +01002293 case MSR_IA32_APICBASE:
2294 data = kvm_get_apic_base(vcpu);
2295 break;
Gleb Natapov0105d1a2009-07-05 17:39:36 +03002296 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2297 return kvm_x2apic_msr_read(vcpu, msr, pdata);
2298 break;
Liu, Jinsonga3e06bb2011-09-22 16:55:52 +08002299 case MSR_IA32_TSCDEADLINE:
2300 data = kvm_get_lapic_tscdeadline_msr(vcpu);
2301 break;
Will Auldba904632012-11-29 12:42:50 -08002302 case MSR_IA32_TSC_ADJUST:
2303 data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2304 break;
Carsten Otte15c4a642007-10-30 18:44:17 +01002305 case MSR_IA32_MISC_ENABLE:
Zhang Xiantaoad312c72007-12-13 23:50:52 +08002306 data = vcpu->arch.ia32_misc_enable_msr;
Carsten Otte15c4a642007-10-30 18:44:17 +01002307 break;
Alexander Graf847f0ad2008-02-21 12:11:01 +01002308 case MSR_IA32_PERF_STATUS:
2309 /* TSC increment by tick */
2310 data = 1000ULL;
2311 /* CPU multiplier */
2312 data |= (((uint64_t)4ULL) << 40);
2313 break;
Carsten Otte15c4a642007-10-30 18:44:17 +01002314 case MSR_EFER:
Avi Kivityf6801df2010-01-21 15:31:50 +02002315 data = vcpu->arch.efer;
Carsten Otte15c4a642007-10-30 18:44:17 +01002316 break;
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02002317 case MSR_KVM_WALL_CLOCK:
Glauber Costa11c6bff2010-05-11 12:17:41 -04002318 case MSR_KVM_WALL_CLOCK_NEW:
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02002319 data = vcpu->kvm->arch.wall_clock;
2320 break;
2321 case MSR_KVM_SYSTEM_TIME:
Glauber Costa11c6bff2010-05-11 12:17:41 -04002322 case MSR_KVM_SYSTEM_TIME_NEW:
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02002323 data = vcpu->arch.time;
2324 break;
Gleb Natapov344d9582010-10-14 11:22:50 +02002325 case MSR_KVM_ASYNC_PF_EN:
2326 data = vcpu->arch.apf.msr_val;
2327 break;
Glauber Costac9aaa892011-07-11 15:28:14 -04002328 case MSR_KVM_STEAL_TIME:
2329 data = vcpu->arch.st.msr_val;
2330 break;
Michael S. Tsirkin1d921282012-08-26 18:00:29 +03002331 case MSR_KVM_PV_EOI_EN:
2332 data = vcpu->arch.pv_eoi.msr_val;
2333 break;
Huang Ying890ca9a2009-05-11 16:48:15 +08002334 case MSR_IA32_P5_MC_ADDR:
2335 case MSR_IA32_P5_MC_TYPE:
2336 case MSR_IA32_MCG_CAP:
2337 case MSR_IA32_MCG_CTL:
2338 case MSR_IA32_MCG_STATUS:
2339 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
2340 return get_msr_mce(vcpu, msr, pdata);
Jes Sorensen84e0cef2010-09-01 11:42:04 +02002341 case MSR_K7_CLK_CTL:
2342 /*
2343 * Provide expected ramp-up count for K7. All other
2344 * are set to zero, indicating minimum divisors for
2345 * every field.
2346 *
2347 * This prevents guest kernels on AMD host with CPU
2348 * type 6, model 8 and higher from exploding due to
2349 * the rdmsr failing.
2350 */
2351 data = 0x20000000;
2352 break;
Gleb Natapov55cd8e52010-01-17 15:51:22 +02002353 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2354 if (kvm_hv_msr_partition_wide(msr)) {
2355 int r;
2356 mutex_lock(&vcpu->kvm->lock);
2357 r = get_msr_hyperv_pw(vcpu, msr, pdata);
2358 mutex_unlock(&vcpu->kvm->lock);
2359 return r;
2360 } else
2361 return get_msr_hyperv(vcpu, msr, pdata);
2362 break;
john cooper91c9c3e2011-01-21 00:21:00 -05002363 case MSR_IA32_BBL_CR_CTL3:
2364 /* This legacy MSR exists but isn't fully documented in current
2365 * silicon. It is however accessed by winxp in very narrow
2366 * scenarios where it sets bit #19, itself documented as
2367 * a "reserved" bit. Best effort attempt to source coherent
2368 * read data here should the balance of the register be
2369 * interpreted by the guest:
2370 *
2371 * L2 cache control register 3: 64GB range, 256KB size,
2372 * enabled, latency 0x1, configured
2373 */
2374 data = 0xbe702111;
2375 break;
Boris Ostrovsky2b036c62012-01-09 14:00:35 -05002376 case MSR_AMD64_OSVW_ID_LENGTH:
2377 if (!guest_cpuid_has_osvw(vcpu))
2378 return 1;
2379 data = vcpu->arch.osvw.length;
2380 break;
2381 case MSR_AMD64_OSVW_STATUS:
2382 if (!guest_cpuid_has_osvw(vcpu))
2383 return 1;
2384 data = vcpu->arch.osvw.status;
2385 break;
Carsten Otte15c4a642007-10-30 18:44:17 +01002386 default:
Gleb Natapovf5132b02011-11-10 14:57:22 +02002387 if (kvm_pmu_msr(vcpu, msr))
2388 return kvm_pmu_get_msr(vcpu, msr, pdata);
Andre Przywaraed85c062009-06-25 12:36:49 +02002389 if (!ignore_msrs) {
Christoffer Dalla737f252012-06-03 21:17:48 +03002390 vcpu_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
Andre Przywaraed85c062009-06-25 12:36:49 +02002391 return 1;
2392 } else {
Christoffer Dalla737f252012-06-03 21:17:48 +03002393 vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
Andre Przywaraed85c062009-06-25 12:36:49 +02002394 data = 0;
2395 }
2396 break;
Carsten Otte15c4a642007-10-30 18:44:17 +01002397 }
2398 *pdata = data;
2399 return 0;
2400}
2401EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2402
Carsten Otte313a3dc2007-10-11 19:16:52 +02002403/*
2404 * Read or write a bunch of msrs. All parameters are kernel addresses.
2405 *
2406 * @return number of msrs set successfully.
2407 */
2408static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2409 struct kvm_msr_entry *entries,
2410 int (*do_msr)(struct kvm_vcpu *vcpu,
2411 unsigned index, u64 *data))
2412{
Marcelo Tosattif656ce02009-12-23 14:35:25 -02002413 int i, idx;
Carsten Otte313a3dc2007-10-11 19:16:52 +02002414
Marcelo Tosattif656ce02009-12-23 14:35:25 -02002415 idx = srcu_read_lock(&vcpu->kvm->srcu);
Carsten Otte313a3dc2007-10-11 19:16:52 +02002416 for (i = 0; i < msrs->nmsrs; ++i)
2417 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2418 break;
Marcelo Tosattif656ce02009-12-23 14:35:25 -02002419 srcu_read_unlock(&vcpu->kvm->srcu, idx);
Carsten Otte313a3dc2007-10-11 19:16:52 +02002420
Carsten Otte313a3dc2007-10-11 19:16:52 +02002421 return i;
2422}
2423
2424/*
2425 * Read or write a bunch of msrs. Parameters are user addresses.
2426 *
2427 * @return number of msrs set successfully.
2428 */
2429static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2430 int (*do_msr)(struct kvm_vcpu *vcpu,
2431 unsigned index, u64 *data),
2432 int writeback)
2433{
2434 struct kvm_msrs msrs;
2435 struct kvm_msr_entry *entries;
2436 int r, n;
2437 unsigned size;
2438
2439 r = -EFAULT;
2440 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2441 goto out;
2442
2443 r = -E2BIG;
2444 if (msrs.nmsrs >= MAX_IO_MSRS)
2445 goto out;
2446
Carsten Otte313a3dc2007-10-11 19:16:52 +02002447 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002448 entries = memdup_user(user_msrs->entries, size);
2449 if (IS_ERR(entries)) {
2450 r = PTR_ERR(entries);
Carsten Otte313a3dc2007-10-11 19:16:52 +02002451 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002452 }
Carsten Otte313a3dc2007-10-11 19:16:52 +02002453
2454 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2455 if (r < 0)
2456 goto out_free;
2457
2458 r = -EFAULT;
2459 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2460 goto out_free;
2461
2462 r = n;
2463
2464out_free:
Avi Kivity7a73c022010-07-22 23:24:52 +03002465 kfree(entries);
Carsten Otte313a3dc2007-10-11 19:16:52 +02002466out:
2467 return r;
2468}
2469
Zhang Xiantao018d00d2007-11-15 23:07:47 +08002470int kvm_dev_ioctl_check_extension(long ext)
2471{
2472 int r;
2473
2474 switch (ext) {
2475 case KVM_CAP_IRQCHIP:
2476 case KVM_CAP_HLT:
2477 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
Zhang Xiantao018d00d2007-11-15 23:07:47 +08002478 case KVM_CAP_SET_TSS_ADDR:
Dan Kenigsberg07716712007-11-21 17:10:04 +02002479 case KVM_CAP_EXT_CPUID:
Gerd Hoffmannc8076602009-02-04 17:52:04 +01002480 case KVM_CAP_CLOCKSOURCE:
Sheng Yang78376992008-01-28 05:10:22 +08002481 case KVM_CAP_PIT:
Marcelo Tosattia28e4f52008-02-22 12:21:36 -05002482 case KVM_CAP_NOP_IO_DELAY:
Marcelo Tosatti62d9f0d2008-04-11 13:24:45 -03002483 case KVM_CAP_MP_STATE:
Avi Kivityed848622008-07-29 11:30:57 +03002484 case KVM_CAP_SYNC_MMU:
Lai Jiangshana355c852010-12-14 17:57:47 +08002485 case KVM_CAP_USER_NMI:
Marcelo Tosatti52d939a2008-12-30 15:55:06 -02002486 case KVM_CAP_REINJECT_CONTROL:
Gleb Natapov49256632009-02-04 17:28:14 +02002487 case KVM_CAP_IRQ_INJECT_STATUS:
Sheng Yange56d5322009-03-12 21:45:39 +08002488 case KVM_CAP_ASSIGN_DEV_IRQ:
Gregory Haskins721eecbf2009-05-20 10:30:49 -04002489 case KVM_CAP_IRQFD:
Gregory Haskinsd34e6b12009-07-07 17:08:49 -04002490 case KVM_CAP_IOEVENTFD:
Jan Kiszkac5ff41c2009-05-14 22:42:53 +02002491 case KVM_CAP_PIT2:
Beth Kone9f42752009-07-07 11:50:38 -04002492 case KVM_CAP_PIT_STATE2:
Sheng Yangb927a3c2009-07-21 10:42:48 +08002493 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
Ed Swierkffde22a2009-10-15 15:21:43 -07002494 case KVM_CAP_XEN_HVM:
Glauber Costaafbcf7a2009-10-16 15:28:36 -04002495 case KVM_CAP_ADJUST_CLOCK:
Jan Kiszka3cfc3092009-11-12 01:04:25 +01002496 case KVM_CAP_VCPU_EVENTS:
Gleb Natapov55cd8e52010-01-17 15:51:22 +02002497 case KVM_CAP_HYPERV:
Gleb Natapov10388a02010-01-17 15:51:23 +02002498 case KVM_CAP_HYPERV_VAPIC:
Gleb Natapovc25bc162010-01-17 15:51:24 +02002499 case KVM_CAP_HYPERV_SPIN:
Zhai, Edwinab9f4ec2010-01-29 14:38:44 +08002500 case KVM_CAP_PCI_SEGMENT:
Jan Kiszkaa1efbe72010-02-15 10:45:43 +01002501 case KVM_CAP_DEBUGREGS:
Jan Kiszkad2be1652010-02-23 17:47:57 +01002502 case KVM_CAP_X86_ROBUST_SINGLESTEP:
Sheng Yang2d5b5a62010-06-13 17:29:39 +08002503 case KVM_CAP_XSAVE:
Gleb Natapov344d9582010-10-14 11:22:50 +02002504 case KVM_CAP_ASYNC_PF:
Joerg Roedel92a1f122011-03-25 09:44:51 +01002505 case KVM_CAP_GET_TSC_KHZ:
Jan Kiszka07700a92012-02-28 14:19:54 +01002506 case KVM_CAP_PCI_2_3:
Eric B Munson1c0b28c2012-03-10 14:37:27 -05002507 case KVM_CAP_KVMCLOCK_CTRL:
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08002508 case KVM_CAP_READONLY_MEM:
Alex Williamson7a844282012-09-21 11:58:03 -06002509 case KVM_CAP_IRQFD_RESAMPLE:
Zhang Xiantao018d00d2007-11-15 23:07:47 +08002510 r = 1;
2511 break;
Laurent Vivier542472b2008-05-30 16:05:55 +02002512 case KVM_CAP_COALESCED_MMIO:
2513 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2514 break;
Avi Kivity774ead32007-12-26 13:57:04 +02002515 case KVM_CAP_VAPIC:
2516 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2517 break;
Avi Kivityf7252302008-02-20 11:53:16 +02002518 case KVM_CAP_NR_VCPUS:
Sasha Levin8c3ba332011-07-18 17:17:15 +03002519 r = KVM_SOFT_MAX_VCPUS;
2520 break;
2521 case KVM_CAP_MAX_VCPUS:
Avi Kivityf7252302008-02-20 11:53:16 +02002522 r = KVM_MAX_VCPUS;
2523 break;
Avi Kivitya988b912008-02-20 11:59:20 +02002524 case KVM_CAP_NR_MEMSLOTS:
2525 r = KVM_MEMORY_SLOTS;
2526 break;
Marcelo Tosattia68a6a72009-10-01 19:28:39 -03002527 case KVM_CAP_PV_MMU: /* obsolete */
2528 r = 0;
Marcelo Tosatti2f333bc2008-02-22 12:21:37 -05002529 break;
Ben-Ami Yassour62c476c2008-09-14 03:48:28 +03002530 case KVM_CAP_IOMMU:
Joerg Roedela1b60c12011-09-06 18:46:34 +02002531 r = iommu_present(&pci_bus_type);
Ben-Ami Yassour62c476c2008-09-14 03:48:28 +03002532 break;
Huang Ying890ca9a2009-05-11 16:48:15 +08002533 case KVM_CAP_MCE:
2534 r = KVM_MAX_MCE_BANKS;
2535 break;
Sheng Yang2d5b5a62010-06-13 17:29:39 +08002536 case KVM_CAP_XCRS:
2537 r = cpu_has_xsave;
2538 break;
Joerg Roedel92a1f122011-03-25 09:44:51 +01002539 case KVM_CAP_TSC_CONTROL:
2540 r = kvm_has_tsc_control;
2541 break;
Jan Kiszka4d25a0662011-12-21 12:28:29 +01002542 case KVM_CAP_TSC_DEADLINE_TIMER:
2543 r = boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER);
2544 break;
Zhang Xiantao018d00d2007-11-15 23:07:47 +08002545 default:
2546 r = 0;
2547 break;
2548 }
2549 return r;
2550
2551}
2552
Carsten Otte043405e2007-10-10 17:16:19 +02002553long kvm_arch_dev_ioctl(struct file *filp,
2554 unsigned int ioctl, unsigned long arg)
2555{
2556 void __user *argp = (void __user *)arg;
2557 long r;
2558
2559 switch (ioctl) {
2560 case KVM_GET_MSR_INDEX_LIST: {
2561 struct kvm_msr_list __user *user_msr_list = argp;
2562 struct kvm_msr_list msr_list;
2563 unsigned n;
2564
2565 r = -EFAULT;
2566 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2567 goto out;
2568 n = msr_list.nmsrs;
2569 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2570 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2571 goto out;
2572 r = -E2BIG;
Jan Kiszkae125e7b2009-07-02 21:45:47 +02002573 if (n < msr_list.nmsrs)
Carsten Otte043405e2007-10-10 17:16:19 +02002574 goto out;
2575 r = -EFAULT;
2576 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2577 num_msrs_to_save * sizeof(u32)))
2578 goto out;
Jan Kiszkae125e7b2009-07-02 21:45:47 +02002579 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
Carsten Otte043405e2007-10-10 17:16:19 +02002580 &emulated_msrs,
2581 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2582 goto out;
2583 r = 0;
2584 break;
2585 }
Avi Kivity674eea02008-02-11 18:37:23 +02002586 case KVM_GET_SUPPORTED_CPUID: {
2587 struct kvm_cpuid2 __user *cpuid_arg = argp;
2588 struct kvm_cpuid2 cpuid;
2589
2590 r = -EFAULT;
2591 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2592 goto out;
2593 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
Amit Shah19355472009-01-14 16:56:00 +00002594 cpuid_arg->entries);
Avi Kivity674eea02008-02-11 18:37:23 +02002595 if (r)
2596 goto out;
2597
2598 r = -EFAULT;
2599 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2600 goto out;
2601 r = 0;
2602 break;
2603 }
Huang Ying890ca9a2009-05-11 16:48:15 +08002604 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2605 u64 mce_cap;
2606
2607 mce_cap = KVM_MCE_CAP_SUPPORTED;
2608 r = -EFAULT;
2609 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2610 goto out;
2611 r = 0;
2612 break;
2613 }
Carsten Otte043405e2007-10-10 17:16:19 +02002614 default:
2615 r = -EINVAL;
2616 }
2617out:
2618 return r;
2619}
2620
Sheng Yangf5f48ee2010-06-30 12:25:15 +08002621static void wbinvd_ipi(void *garbage)
2622{
2623 wbinvd();
2624}
2625
2626static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2627{
2628 return vcpu->kvm->arch.iommu_domain &&
2629 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY);
2630}
2631
Carsten Otte313a3dc2007-10-11 19:16:52 +02002632void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2633{
Sheng Yangf5f48ee2010-06-30 12:25:15 +08002634 /* Address WBINVD may be executed by guest */
2635 if (need_emulate_wbinvd(vcpu)) {
2636 if (kvm_x86_ops->has_wbinvd_exit())
2637 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2638 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2639 smp_call_function_single(vcpu->cpu,
2640 wbinvd_ipi, NULL, 1);
2641 }
2642
Carsten Otte313a3dc2007-10-11 19:16:52 +02002643 kvm_x86_ops->vcpu_load(vcpu, cpu);
Zachary Amsden0dd6a6e2012-02-03 15:43:56 -02002644
2645 /* Apply any externally detected TSC adjustments (due to suspend) */
2646 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
2647 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
2648 vcpu->arch.tsc_offset_adjustment = 0;
2649 set_bit(KVM_REQ_CLOCK_UPDATE, &vcpu->requests);
2650 }
2651
Zachary Amsden48434c202010-08-19 22:07:24 -10002652 if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
Zachary Amsden6f526ec2012-02-03 15:43:54 -02002653 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2654 native_read_tsc() - vcpu->arch.last_host_tsc;
Zachary Amsdene48672f2010-08-19 22:07:23 -10002655 if (tsc_delta < 0)
2656 mark_tsc_unstable("KVM discovered backwards TSC");
Zachary Amsdenc2855452010-09-18 14:38:15 -10002657 if (check_tsc_unstable()) {
Zachary Amsdenb183aa52012-02-03 15:43:53 -02002658 u64 offset = kvm_x86_ops->compute_tsc_offset(vcpu,
2659 vcpu->arch.last_guest_tsc);
2660 kvm_x86_ops->write_tsc_offset(vcpu, offset);
Zachary Amsdenc2855452010-09-18 14:38:15 -10002661 vcpu->arch.tsc_catchup = 1;
Zachary Amsdenc2855452010-09-18 14:38:15 -10002662 }
Marcelo Tosattid98d07c2012-11-27 23:29:04 -02002663 /*
2664 * On a host with synchronized TSC, there is no need to update
2665 * kvmclock on vcpu->cpu migration
2666 */
2667 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
2668 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
Zachary Amsdenc2855452010-09-18 14:38:15 -10002669 if (vcpu->cpu != cpu)
2670 kvm_migrate_timers(vcpu);
Zachary Amsdene48672f2010-08-19 22:07:23 -10002671 vcpu->cpu = cpu;
Zachary Amsden6b7d7e72009-10-09 16:26:08 -10002672 }
Glauber Costac9aaa892011-07-11 15:28:14 -04002673
2674 accumulate_steal_time(vcpu);
2675 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
Carsten Otte313a3dc2007-10-11 19:16:52 +02002676}
2677
2678void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2679{
Avi Kivity02daab22009-12-30 12:40:26 +02002680 kvm_x86_ops->vcpu_put(vcpu);
Avi Kivity1c11e712010-05-03 16:05:44 +03002681 kvm_put_guest_fpu(vcpu);
Zachary Amsden6f526ec2012-02-03 15:43:54 -02002682 vcpu->arch.last_host_tsc = native_read_tsc();
Carsten Otte313a3dc2007-10-11 19:16:52 +02002683}
2684
Carsten Otte313a3dc2007-10-11 19:16:52 +02002685static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2686 struct kvm_lapic_state *s)
2687{
Zhang Xiantaoad312c72007-12-13 23:50:52 +08002688 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
Carsten Otte313a3dc2007-10-11 19:16:52 +02002689
2690 return 0;
2691}
2692
2693static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2694 struct kvm_lapic_state *s)
2695{
Gleb Natapov64eb0622012-08-08 15:24:36 +03002696 kvm_apic_post_state_restore(vcpu, s);
Gleb Natapovcb142eb2009-08-09 15:17:40 +03002697 update_cr8_intercept(vcpu);
Carsten Otte313a3dc2007-10-11 19:16:52 +02002698
2699 return 0;
2700}
2701
Zhang Xiantaof77bc6a2007-11-21 04:36:41 +08002702static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2703 struct kvm_interrupt *irq)
2704{
Michael S. Tsirkina50abc32012-09-05 20:00:52 +03002705 if (irq->irq < 0 || irq->irq >= KVM_NR_INTERRUPTS)
Zhang Xiantaof77bc6a2007-11-21 04:36:41 +08002706 return -EINVAL;
2707 if (irqchip_in_kernel(vcpu->kvm))
2708 return -ENXIO;
Zhang Xiantaof77bc6a2007-11-21 04:36:41 +08002709
Gleb Natapov66fd3f72009-05-11 13:35:50 +03002710 kvm_queue_interrupt(vcpu, irq->irq, false);
Avi Kivity3842d132010-07-27 12:30:24 +03002711 kvm_make_request(KVM_REQ_EVENT, vcpu);
Zhang Xiantaof77bc6a2007-11-21 04:36:41 +08002712
Zhang Xiantaof77bc6a2007-11-21 04:36:41 +08002713 return 0;
2714}
2715
Jan Kiszkac4abb7c2008-09-26 09:30:55 +02002716static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2717{
Jan Kiszkac4abb7c2008-09-26 09:30:55 +02002718 kvm_inject_nmi(vcpu);
Jan Kiszkac4abb7c2008-09-26 09:30:55 +02002719
2720 return 0;
2721}
2722
Avi Kivityb209749f2007-10-22 16:50:39 +02002723static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2724 struct kvm_tpr_access_ctl *tac)
2725{
2726 if (tac->flags)
2727 return -EINVAL;
2728 vcpu->arch.tpr_access_reporting = !!tac->enabled;
2729 return 0;
2730}
2731
Huang Ying890ca9a2009-05-11 16:48:15 +08002732static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2733 u64 mcg_cap)
2734{
2735 int r;
2736 unsigned bank_num = mcg_cap & 0xff, bank;
2737
2738 r = -EINVAL;
Jan Kiszkaa9e38c3e2009-10-23 09:37:00 +02002739 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
Huang Ying890ca9a2009-05-11 16:48:15 +08002740 goto out;
2741 if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2742 goto out;
2743 r = 0;
2744 vcpu->arch.mcg_cap = mcg_cap;
2745 /* Init IA32_MCG_CTL to all 1s */
2746 if (mcg_cap & MCG_CTL_P)
2747 vcpu->arch.mcg_ctl = ~(u64)0;
2748 /* Init IA32_MCi_CTL to all 1s */
2749 for (bank = 0; bank < bank_num; bank++)
2750 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2751out:
2752 return r;
2753}
2754
2755static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2756 struct kvm_x86_mce *mce)
2757{
2758 u64 mcg_cap = vcpu->arch.mcg_cap;
2759 unsigned bank_num = mcg_cap & 0xff;
2760 u64 *banks = vcpu->arch.mce_banks;
2761
2762 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2763 return -EINVAL;
2764 /*
2765 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2766 * reporting is disabled
2767 */
2768 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2769 vcpu->arch.mcg_ctl != ~(u64)0)
2770 return 0;
2771 banks += 4 * mce->bank;
2772 /*
2773 * if IA32_MCi_CTL is not all 1s, the uncorrected error
2774 * reporting is disabled for the bank
2775 */
2776 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2777 return 0;
2778 if (mce->status & MCI_STATUS_UC) {
2779 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
Avi Kivityfc78f512009-12-07 12:16:48 +02002780 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
Avi Kivitya8eeb042010-05-10 12:34:53 +03002781 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
Huang Ying890ca9a2009-05-11 16:48:15 +08002782 return 0;
2783 }
2784 if (banks[1] & MCI_STATUS_VAL)
2785 mce->status |= MCI_STATUS_OVER;
2786 banks[2] = mce->addr;
2787 banks[3] = mce->misc;
2788 vcpu->arch.mcg_status = mce->mcg_status;
2789 banks[1] = mce->status;
2790 kvm_queue_exception(vcpu, MC_VECTOR);
2791 } else if (!(banks[1] & MCI_STATUS_VAL)
2792 || !(banks[1] & MCI_STATUS_UC)) {
2793 if (banks[1] & MCI_STATUS_VAL)
2794 mce->status |= MCI_STATUS_OVER;
2795 banks[2] = mce->addr;
2796 banks[3] = mce->misc;
2797 banks[1] = mce->status;
2798 } else
2799 banks[1] |= MCI_STATUS_OVER;
2800 return 0;
2801}
2802
Jan Kiszka3cfc3092009-11-12 01:04:25 +01002803static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2804 struct kvm_vcpu_events *events)
2805{
Avi Kivity7460fb4a2011-09-20 13:43:14 +03002806 process_nmi(vcpu);
Jan Kiszka03b82a32010-02-15 10:45:41 +01002807 events->exception.injected =
2808 vcpu->arch.exception.pending &&
2809 !kvm_exception_is_soft(vcpu->arch.exception.nr);
Jan Kiszka3cfc3092009-11-12 01:04:25 +01002810 events->exception.nr = vcpu->arch.exception.nr;
2811 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
Vasiliy Kulikov97e69aa2010-10-30 22:54:47 +04002812 events->exception.pad = 0;
Jan Kiszka3cfc3092009-11-12 01:04:25 +01002813 events->exception.error_code = vcpu->arch.exception.error_code;
2814
Jan Kiszka03b82a32010-02-15 10:45:41 +01002815 events->interrupt.injected =
2816 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
Jan Kiszka3cfc3092009-11-12 01:04:25 +01002817 events->interrupt.nr = vcpu->arch.interrupt.nr;
Jan Kiszka03b82a32010-02-15 10:45:41 +01002818 events->interrupt.soft = 0;
Jan Kiszka48005f62010-02-19 19:38:07 +01002819 events->interrupt.shadow =
2820 kvm_x86_ops->get_interrupt_shadow(vcpu,
2821 KVM_X86_SHADOW_INT_MOV_SS | KVM_X86_SHADOW_INT_STI);
Jan Kiszka3cfc3092009-11-12 01:04:25 +01002822
2823 events->nmi.injected = vcpu->arch.nmi_injected;
Avi Kivity7460fb4a2011-09-20 13:43:14 +03002824 events->nmi.pending = vcpu->arch.nmi_pending != 0;
Jan Kiszka3cfc3092009-11-12 01:04:25 +01002825 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
Vasiliy Kulikov97e69aa2010-10-30 22:54:47 +04002826 events->nmi.pad = 0;
Jan Kiszka3cfc3092009-11-12 01:04:25 +01002827
2828 events->sipi_vector = vcpu->arch.sipi_vector;
2829
Jan Kiszkadab4b912009-12-06 18:24:15 +01002830 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
Jan Kiszka48005f62010-02-19 19:38:07 +01002831 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2832 | KVM_VCPUEVENT_VALID_SHADOW);
Vasiliy Kulikov97e69aa2010-10-30 22:54:47 +04002833 memset(&events->reserved, 0, sizeof(events->reserved));
Jan Kiszka3cfc3092009-11-12 01:04:25 +01002834}
2835
2836static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2837 struct kvm_vcpu_events *events)
2838{
Jan Kiszkadab4b912009-12-06 18:24:15 +01002839 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
Jan Kiszka48005f62010-02-19 19:38:07 +01002840 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2841 | KVM_VCPUEVENT_VALID_SHADOW))
Jan Kiszka3cfc3092009-11-12 01:04:25 +01002842 return -EINVAL;
2843
Avi Kivity7460fb4a2011-09-20 13:43:14 +03002844 process_nmi(vcpu);
Jan Kiszka3cfc3092009-11-12 01:04:25 +01002845 vcpu->arch.exception.pending = events->exception.injected;
2846 vcpu->arch.exception.nr = events->exception.nr;
2847 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2848 vcpu->arch.exception.error_code = events->exception.error_code;
2849
2850 vcpu->arch.interrupt.pending = events->interrupt.injected;
2851 vcpu->arch.interrupt.nr = events->interrupt.nr;
2852 vcpu->arch.interrupt.soft = events->interrupt.soft;
Jan Kiszka48005f62010-02-19 19:38:07 +01002853 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2854 kvm_x86_ops->set_interrupt_shadow(vcpu,
2855 events->interrupt.shadow);
Jan Kiszka3cfc3092009-11-12 01:04:25 +01002856
2857 vcpu->arch.nmi_injected = events->nmi.injected;
Jan Kiszkadab4b912009-12-06 18:24:15 +01002858 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2859 vcpu->arch.nmi_pending = events->nmi.pending;
Jan Kiszka3cfc3092009-11-12 01:04:25 +01002860 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2861
Jan Kiszkadab4b912009-12-06 18:24:15 +01002862 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR)
2863 vcpu->arch.sipi_vector = events->sipi_vector;
Jan Kiszka3cfc3092009-11-12 01:04:25 +01002864
Avi Kivity3842d132010-07-27 12:30:24 +03002865 kvm_make_request(KVM_REQ_EVENT, vcpu);
2866
Jan Kiszka3cfc3092009-11-12 01:04:25 +01002867 return 0;
2868}
2869
Jan Kiszkaa1efbe72010-02-15 10:45:43 +01002870static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
2871 struct kvm_debugregs *dbgregs)
2872{
Jan Kiszkaa1efbe72010-02-15 10:45:43 +01002873 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
2874 dbgregs->dr6 = vcpu->arch.dr6;
2875 dbgregs->dr7 = vcpu->arch.dr7;
2876 dbgregs->flags = 0;
Vasiliy Kulikov97e69aa2010-10-30 22:54:47 +04002877 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
Jan Kiszkaa1efbe72010-02-15 10:45:43 +01002878}
2879
2880static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
2881 struct kvm_debugregs *dbgregs)
2882{
2883 if (dbgregs->flags)
2884 return -EINVAL;
2885
Jan Kiszkaa1efbe72010-02-15 10:45:43 +01002886 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
2887 vcpu->arch.dr6 = dbgregs->dr6;
2888 vcpu->arch.dr7 = dbgregs->dr7;
2889
Jan Kiszkaa1efbe72010-02-15 10:45:43 +01002890 return 0;
2891}
2892
Sheng Yang2d5b5a62010-06-13 17:29:39 +08002893static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
2894 struct kvm_xsave *guest_xsave)
2895{
2896 if (cpu_has_xsave)
2897 memcpy(guest_xsave->region,
2898 &vcpu->arch.guest_fpu.state->xsave,
Xiaotian Fengf45755b2010-08-13 15:19:11 +08002899 xstate_size);
Sheng Yang2d5b5a62010-06-13 17:29:39 +08002900 else {
2901 memcpy(guest_xsave->region,
2902 &vcpu->arch.guest_fpu.state->fxsave,
2903 sizeof(struct i387_fxsave_struct));
2904 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
2905 XSTATE_FPSSE;
2906 }
2907}
2908
2909static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
2910 struct kvm_xsave *guest_xsave)
2911{
2912 u64 xstate_bv =
2913 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
2914
2915 if (cpu_has_xsave)
2916 memcpy(&vcpu->arch.guest_fpu.state->xsave,
Xiaotian Fengf45755b2010-08-13 15:19:11 +08002917 guest_xsave->region, xstate_size);
Sheng Yang2d5b5a62010-06-13 17:29:39 +08002918 else {
2919 if (xstate_bv & ~XSTATE_FPSSE)
2920 return -EINVAL;
2921 memcpy(&vcpu->arch.guest_fpu.state->fxsave,
2922 guest_xsave->region, sizeof(struct i387_fxsave_struct));
2923 }
2924 return 0;
2925}
2926
2927static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
2928 struct kvm_xcrs *guest_xcrs)
2929{
2930 if (!cpu_has_xsave) {
2931 guest_xcrs->nr_xcrs = 0;
2932 return;
2933 }
2934
2935 guest_xcrs->nr_xcrs = 1;
2936 guest_xcrs->flags = 0;
2937 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
2938 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
2939}
2940
2941static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
2942 struct kvm_xcrs *guest_xcrs)
2943{
2944 int i, r = 0;
2945
2946 if (!cpu_has_xsave)
2947 return -EINVAL;
2948
2949 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
2950 return -EINVAL;
2951
2952 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
2953 /* Only support XCR0 currently */
2954 if (guest_xcrs->xcrs[0].xcr == XCR_XFEATURE_ENABLED_MASK) {
2955 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
2956 guest_xcrs->xcrs[0].value);
2957 break;
2958 }
2959 if (r)
2960 r = -EINVAL;
2961 return r;
2962}
2963
Eric B Munson1c0b28c2012-03-10 14:37:27 -05002964/*
2965 * kvm_set_guest_paused() indicates to the guest kernel that it has been
2966 * stopped by the hypervisor. This function will be called from the host only.
2967 * EINVAL is returned when the host attempts to set the flag for a guest that
2968 * does not support pv clocks.
2969 */
2970static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
2971{
Eric B Munson1c0b28c2012-03-10 14:37:27 -05002972 if (!vcpu->arch.time_page)
2973 return -EINVAL;
Marcelo Tosatti51d59c62012-08-03 15:57:49 -03002974 vcpu->arch.pvclock_set_guest_stopped_request = true;
Eric B Munson1c0b28c2012-03-10 14:37:27 -05002975 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2976 return 0;
2977}
2978
Carsten Otte313a3dc2007-10-11 19:16:52 +02002979long kvm_arch_vcpu_ioctl(struct file *filp,
2980 unsigned int ioctl, unsigned long arg)
2981{
2982 struct kvm_vcpu *vcpu = filp->private_data;
2983 void __user *argp = (void __user *)arg;
2984 int r;
Avi Kivityd1ac91d2010-06-20 15:54:43 +03002985 union {
2986 struct kvm_lapic_state *lapic;
2987 struct kvm_xsave *xsave;
2988 struct kvm_xcrs *xcrs;
2989 void *buffer;
2990 } u;
Carsten Otte313a3dc2007-10-11 19:16:52 +02002991
Avi Kivityd1ac91d2010-06-20 15:54:43 +03002992 u.buffer = NULL;
Carsten Otte313a3dc2007-10-11 19:16:52 +02002993 switch (ioctl) {
2994 case KVM_GET_LAPIC: {
Marcelo Tosatti2204ae32009-10-29 13:44:16 -02002995 r = -EINVAL;
2996 if (!vcpu->arch.apic)
2997 goto out;
Avi Kivityd1ac91d2010-06-20 15:54:43 +03002998 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
Carsten Otte313a3dc2007-10-11 19:16:52 +02002999
Dave Hansenb772ff32008-08-11 10:01:47 -07003000 r = -ENOMEM;
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003001 if (!u.lapic)
Dave Hansenb772ff32008-08-11 10:01:47 -07003002 goto out;
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003003 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
Carsten Otte313a3dc2007-10-11 19:16:52 +02003004 if (r)
3005 goto out;
3006 r = -EFAULT;
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003007 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
Carsten Otte313a3dc2007-10-11 19:16:52 +02003008 goto out;
3009 r = 0;
3010 break;
3011 }
3012 case KVM_SET_LAPIC: {
Marcelo Tosatti2204ae32009-10-29 13:44:16 -02003013 r = -EINVAL;
3014 if (!vcpu->arch.apic)
3015 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02003016 u.lapic = memdup_user(argp, sizeof(*u.lapic));
Guo Chao18595412012-11-02 18:33:21 +08003017 if (IS_ERR(u.lapic))
3018 return PTR_ERR(u.lapic);
Sasha Levinff5c2c02011-12-04 19:36:29 +02003019
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003020 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
Carsten Otte313a3dc2007-10-11 19:16:52 +02003021 break;
3022 }
Zhang Xiantaof77bc6a2007-11-21 04:36:41 +08003023 case KVM_INTERRUPT: {
3024 struct kvm_interrupt irq;
3025
3026 r = -EFAULT;
3027 if (copy_from_user(&irq, argp, sizeof irq))
3028 goto out;
3029 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
Zhang Xiantaof77bc6a2007-11-21 04:36:41 +08003030 break;
3031 }
Jan Kiszkac4abb7c2008-09-26 09:30:55 +02003032 case KVM_NMI: {
3033 r = kvm_vcpu_ioctl_nmi(vcpu);
Jan Kiszkac4abb7c2008-09-26 09:30:55 +02003034 break;
3035 }
Carsten Otte313a3dc2007-10-11 19:16:52 +02003036 case KVM_SET_CPUID: {
3037 struct kvm_cpuid __user *cpuid_arg = argp;
3038 struct kvm_cpuid cpuid;
3039
3040 r = -EFAULT;
3041 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3042 goto out;
3043 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
Carsten Otte313a3dc2007-10-11 19:16:52 +02003044 break;
3045 }
Dan Kenigsberg07716712007-11-21 17:10:04 +02003046 case KVM_SET_CPUID2: {
3047 struct kvm_cpuid2 __user *cpuid_arg = argp;
3048 struct kvm_cpuid2 cpuid;
3049
3050 r = -EFAULT;
3051 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3052 goto out;
3053 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
Amit Shah19355472009-01-14 16:56:00 +00003054 cpuid_arg->entries);
Dan Kenigsberg07716712007-11-21 17:10:04 +02003055 break;
3056 }
3057 case KVM_GET_CPUID2: {
3058 struct kvm_cpuid2 __user *cpuid_arg = argp;
3059 struct kvm_cpuid2 cpuid;
3060
3061 r = -EFAULT;
3062 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3063 goto out;
3064 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
Amit Shah19355472009-01-14 16:56:00 +00003065 cpuid_arg->entries);
Dan Kenigsberg07716712007-11-21 17:10:04 +02003066 if (r)
3067 goto out;
3068 r = -EFAULT;
3069 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3070 goto out;
3071 r = 0;
3072 break;
3073 }
Carsten Otte313a3dc2007-10-11 19:16:52 +02003074 case KVM_GET_MSRS:
3075 r = msr_io(vcpu, argp, kvm_get_msr, 1);
3076 break;
3077 case KVM_SET_MSRS:
3078 r = msr_io(vcpu, argp, do_set_msr, 0);
3079 break;
Avi Kivityb209749f2007-10-22 16:50:39 +02003080 case KVM_TPR_ACCESS_REPORTING: {
3081 struct kvm_tpr_access_ctl tac;
3082
3083 r = -EFAULT;
3084 if (copy_from_user(&tac, argp, sizeof tac))
3085 goto out;
3086 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3087 if (r)
3088 goto out;
3089 r = -EFAULT;
3090 if (copy_to_user(argp, &tac, sizeof tac))
3091 goto out;
3092 r = 0;
3093 break;
3094 };
Avi Kivityb93463a2007-10-25 16:52:32 +02003095 case KVM_SET_VAPIC_ADDR: {
3096 struct kvm_vapic_addr va;
3097
3098 r = -EINVAL;
3099 if (!irqchip_in_kernel(vcpu->kvm))
3100 goto out;
3101 r = -EFAULT;
3102 if (copy_from_user(&va, argp, sizeof va))
3103 goto out;
3104 r = 0;
3105 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3106 break;
3107 }
Huang Ying890ca9a2009-05-11 16:48:15 +08003108 case KVM_X86_SETUP_MCE: {
3109 u64 mcg_cap;
3110
3111 r = -EFAULT;
3112 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3113 goto out;
3114 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3115 break;
3116 }
3117 case KVM_X86_SET_MCE: {
3118 struct kvm_x86_mce mce;
3119
3120 r = -EFAULT;
3121 if (copy_from_user(&mce, argp, sizeof mce))
3122 goto out;
3123 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3124 break;
3125 }
Jan Kiszka3cfc3092009-11-12 01:04:25 +01003126 case KVM_GET_VCPU_EVENTS: {
3127 struct kvm_vcpu_events events;
3128
3129 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3130
3131 r = -EFAULT;
3132 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3133 break;
3134 r = 0;
3135 break;
3136 }
3137 case KVM_SET_VCPU_EVENTS: {
3138 struct kvm_vcpu_events events;
3139
3140 r = -EFAULT;
3141 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3142 break;
3143
3144 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3145 break;
3146 }
Jan Kiszkaa1efbe72010-02-15 10:45:43 +01003147 case KVM_GET_DEBUGREGS: {
3148 struct kvm_debugregs dbgregs;
3149
3150 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3151
3152 r = -EFAULT;
3153 if (copy_to_user(argp, &dbgregs,
3154 sizeof(struct kvm_debugregs)))
3155 break;
3156 r = 0;
3157 break;
3158 }
3159 case KVM_SET_DEBUGREGS: {
3160 struct kvm_debugregs dbgregs;
3161
3162 r = -EFAULT;
3163 if (copy_from_user(&dbgregs, argp,
3164 sizeof(struct kvm_debugregs)))
3165 break;
3166
3167 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3168 break;
3169 }
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003170 case KVM_GET_XSAVE: {
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003171 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003172 r = -ENOMEM;
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003173 if (!u.xsave)
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003174 break;
3175
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003176 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003177
3178 r = -EFAULT;
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003179 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003180 break;
3181 r = 0;
3182 break;
3183 }
3184 case KVM_SET_XSAVE: {
Sasha Levinff5c2c02011-12-04 19:36:29 +02003185 u.xsave = memdup_user(argp, sizeof(*u.xsave));
Guo Chao18595412012-11-02 18:33:21 +08003186 if (IS_ERR(u.xsave))
3187 return PTR_ERR(u.xsave);
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003188
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003189 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003190 break;
3191 }
3192 case KVM_GET_XCRS: {
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003193 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003194 r = -ENOMEM;
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003195 if (!u.xcrs)
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003196 break;
3197
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003198 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003199
3200 r = -EFAULT;
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003201 if (copy_to_user(argp, u.xcrs,
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003202 sizeof(struct kvm_xcrs)))
3203 break;
3204 r = 0;
3205 break;
3206 }
3207 case KVM_SET_XCRS: {
Sasha Levinff5c2c02011-12-04 19:36:29 +02003208 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
Guo Chao18595412012-11-02 18:33:21 +08003209 if (IS_ERR(u.xcrs))
3210 return PTR_ERR(u.xcrs);
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003211
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003212 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003213 break;
3214 }
Joerg Roedel92a1f122011-03-25 09:44:51 +01003215 case KVM_SET_TSC_KHZ: {
3216 u32 user_tsc_khz;
3217
3218 r = -EINVAL;
Joerg Roedel92a1f122011-03-25 09:44:51 +01003219 user_tsc_khz = (u32)arg;
3220
3221 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3222 goto out;
3223
Zachary Amsdencc578282012-02-03 15:43:50 -02003224 if (user_tsc_khz == 0)
3225 user_tsc_khz = tsc_khz;
3226
3227 kvm_set_tsc_khz(vcpu, user_tsc_khz);
Joerg Roedel92a1f122011-03-25 09:44:51 +01003228
3229 r = 0;
3230 goto out;
3231 }
3232 case KVM_GET_TSC_KHZ: {
Zachary Amsdencc578282012-02-03 15:43:50 -02003233 r = vcpu->arch.virtual_tsc_khz;
Joerg Roedel92a1f122011-03-25 09:44:51 +01003234 goto out;
3235 }
Eric B Munson1c0b28c2012-03-10 14:37:27 -05003236 case KVM_KVMCLOCK_CTRL: {
3237 r = kvm_set_guest_paused(vcpu);
3238 goto out;
3239 }
Carsten Otte313a3dc2007-10-11 19:16:52 +02003240 default:
3241 r = -EINVAL;
3242 }
3243out:
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003244 kfree(u.buffer);
Carsten Otte313a3dc2007-10-11 19:16:52 +02003245 return r;
3246}
3247
Carsten Otte5b1c1492012-01-04 10:25:23 +01003248int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3249{
3250 return VM_FAULT_SIGBUS;
3251}
3252
Carsten Otte1fe779f2007-10-29 16:08:35 +01003253static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3254{
3255 int ret;
3256
3257 if (addr > (unsigned int)(-3 * PAGE_SIZE))
Guo Chao951179c2012-11-02 18:33:22 +08003258 return -EINVAL;
Carsten Otte1fe779f2007-10-29 16:08:35 +01003259 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3260 return ret;
3261}
3262
Sheng Yangb927a3c2009-07-21 10:42:48 +08003263static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3264 u64 ident_addr)
3265{
3266 kvm->arch.ept_identity_map_addr = ident_addr;
3267 return 0;
3268}
3269
Carsten Otte1fe779f2007-10-29 16:08:35 +01003270static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3271 u32 kvm_nr_mmu_pages)
3272{
3273 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3274 return -EINVAL;
3275
Marcelo Tosatti79fac952009-12-23 14:35:26 -02003276 mutex_lock(&kvm->slots_lock);
Marcelo Tosatti7c8a83b2009-05-12 18:55:43 -03003277 spin_lock(&kvm->mmu_lock);
Carsten Otte1fe779f2007-10-29 16:08:35 +01003278
3279 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
Zhang Xiantaof05e70a2007-12-14 10:01:48 +08003280 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
Carsten Otte1fe779f2007-10-29 16:08:35 +01003281
Marcelo Tosatti7c8a83b2009-05-12 18:55:43 -03003282 spin_unlock(&kvm->mmu_lock);
Marcelo Tosatti79fac952009-12-23 14:35:26 -02003283 mutex_unlock(&kvm->slots_lock);
Carsten Otte1fe779f2007-10-29 16:08:35 +01003284 return 0;
3285}
3286
3287static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3288{
Dave Hansen39de71e2010-08-19 18:11:14 -07003289 return kvm->arch.n_max_mmu_pages;
Carsten Otte1fe779f2007-10-29 16:08:35 +01003290}
3291
Carsten Otte1fe779f2007-10-29 16:08:35 +01003292static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3293{
3294 int r;
3295
3296 r = 0;
3297 switch (chip->chip_id) {
3298 case KVM_IRQCHIP_PIC_MASTER:
3299 memcpy(&chip->chip.pic,
3300 &pic_irqchip(kvm)->pics[0],
3301 sizeof(struct kvm_pic_state));
3302 break;
3303 case KVM_IRQCHIP_PIC_SLAVE:
3304 memcpy(&chip->chip.pic,
3305 &pic_irqchip(kvm)->pics[1],
3306 sizeof(struct kvm_pic_state));
3307 break;
3308 case KVM_IRQCHIP_IOAPIC:
Gleb Natapoveba02262009-08-24 11:54:25 +03003309 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
Carsten Otte1fe779f2007-10-29 16:08:35 +01003310 break;
3311 default:
3312 r = -EINVAL;
3313 break;
3314 }
3315 return r;
3316}
3317
3318static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3319{
3320 int r;
3321
3322 r = 0;
3323 switch (chip->chip_id) {
3324 case KVM_IRQCHIP_PIC_MASTER:
Avi Kivityf4f51052010-09-19 18:44:07 +02003325 spin_lock(&pic_irqchip(kvm)->lock);
Carsten Otte1fe779f2007-10-29 16:08:35 +01003326 memcpy(&pic_irqchip(kvm)->pics[0],
3327 &chip->chip.pic,
3328 sizeof(struct kvm_pic_state));
Avi Kivityf4f51052010-09-19 18:44:07 +02003329 spin_unlock(&pic_irqchip(kvm)->lock);
Carsten Otte1fe779f2007-10-29 16:08:35 +01003330 break;
3331 case KVM_IRQCHIP_PIC_SLAVE:
Avi Kivityf4f51052010-09-19 18:44:07 +02003332 spin_lock(&pic_irqchip(kvm)->lock);
Carsten Otte1fe779f2007-10-29 16:08:35 +01003333 memcpy(&pic_irqchip(kvm)->pics[1],
3334 &chip->chip.pic,
3335 sizeof(struct kvm_pic_state));
Avi Kivityf4f51052010-09-19 18:44:07 +02003336 spin_unlock(&pic_irqchip(kvm)->lock);
Carsten Otte1fe779f2007-10-29 16:08:35 +01003337 break;
3338 case KVM_IRQCHIP_IOAPIC:
Gleb Natapoveba02262009-08-24 11:54:25 +03003339 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
Carsten Otte1fe779f2007-10-29 16:08:35 +01003340 break;
3341 default:
3342 r = -EINVAL;
3343 break;
3344 }
3345 kvm_pic_update_irq(pic_irqchip(kvm));
3346 return r;
3347}
3348
Sheng Yange0f63cb2008-03-04 00:50:59 +08003349static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3350{
3351 int r = 0;
3352
Marcelo Tosatti894a9c52009-06-23 15:05:14 -03003353 mutex_lock(&kvm->arch.vpit->pit_state.lock);
Sheng Yange0f63cb2008-03-04 00:50:59 +08003354 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
Marcelo Tosatti894a9c52009-06-23 15:05:14 -03003355 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
Sheng Yange0f63cb2008-03-04 00:50:59 +08003356 return r;
3357}
3358
3359static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3360{
3361 int r = 0;
3362
Marcelo Tosatti894a9c52009-06-23 15:05:14 -03003363 mutex_lock(&kvm->arch.vpit->pit_state.lock);
Sheng Yange0f63cb2008-03-04 00:50:59 +08003364 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
Beth Kone9f42752009-07-07 11:50:38 -04003365 kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
3366 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3367 return r;
3368}
3369
3370static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3371{
3372 int r = 0;
3373
3374 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3375 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3376 sizeof(ps->channels));
3377 ps->flags = kvm->arch.vpit->pit_state.flags;
3378 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
Vasiliy Kulikov97e69aa2010-10-30 22:54:47 +04003379 memset(&ps->reserved, 0, sizeof(ps->reserved));
Beth Kone9f42752009-07-07 11:50:38 -04003380 return r;
3381}
3382
3383static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3384{
3385 int r = 0, start = 0;
3386 u32 prev_legacy, cur_legacy;
3387 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3388 prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3389 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3390 if (!prev_legacy && cur_legacy)
3391 start = 1;
3392 memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3393 sizeof(kvm->arch.vpit->pit_state.channels));
3394 kvm->arch.vpit->pit_state.flags = ps->flags;
3395 kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
Marcelo Tosatti894a9c52009-06-23 15:05:14 -03003396 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
Sheng Yange0f63cb2008-03-04 00:50:59 +08003397 return r;
3398}
3399
Marcelo Tosatti52d939a2008-12-30 15:55:06 -02003400static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3401 struct kvm_reinject_control *control)
3402{
3403 if (!kvm->arch.vpit)
3404 return -ENXIO;
Marcelo Tosatti894a9c52009-06-23 15:05:14 -03003405 mutex_lock(&kvm->arch.vpit->pit_state.lock);
Avi Kivity26ef1922012-07-26 18:01:53 +03003406 kvm->arch.vpit->pit_state.reinject = control->pit_reinject;
Marcelo Tosatti894a9c52009-06-23 15:05:14 -03003407 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
Marcelo Tosatti52d939a2008-12-30 15:55:06 -02003408 return 0;
3409}
3410
Takuya Yoshikawa95d4c162011-11-14 18:24:50 +09003411/**
Takuya Yoshikawa60c34612012-03-03 14:21:48 +09003412 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3413 * @kvm: kvm instance
3414 * @log: slot id and address to which we copy the log
Takuya Yoshikawa95d4c162011-11-14 18:24:50 +09003415 *
Takuya Yoshikawa60c34612012-03-03 14:21:48 +09003416 * We need to keep it in mind that VCPU threads can write to the bitmap
3417 * concurrently. So, to avoid losing data, we keep the following order for
3418 * each bit:
Takuya Yoshikawa95d4c162011-11-14 18:24:50 +09003419 *
Takuya Yoshikawa60c34612012-03-03 14:21:48 +09003420 * 1. Take a snapshot of the bit and clear it if needed.
3421 * 2. Write protect the corresponding page.
3422 * 3. Flush TLB's if needed.
3423 * 4. Copy the snapshot to the userspace.
Takuya Yoshikawa95d4c162011-11-14 18:24:50 +09003424 *
Takuya Yoshikawa60c34612012-03-03 14:21:48 +09003425 * Between 2 and 3, the guest may write to the page using the remaining TLB
3426 * entry. This is not a problem because the page will be reported dirty at
3427 * step 4 using the snapshot taken before and step 3 ensures that successive
3428 * writes will be logged for the next call.
Takuya Yoshikawa95d4c162011-11-14 18:24:50 +09003429 */
Takuya Yoshikawa60c34612012-03-03 14:21:48 +09003430int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
Zhang Xiantao5bb064d2007-11-18 20:29:43 +08003431{
Takuya Yoshikawa7850ac52011-11-14 18:23:34 +09003432 int r;
Zhang Xiantao5bb064d2007-11-18 20:29:43 +08003433 struct kvm_memory_slot *memslot;
Takuya Yoshikawa60c34612012-03-03 14:21:48 +09003434 unsigned long n, i;
3435 unsigned long *dirty_bitmap;
3436 unsigned long *dirty_bitmap_buffer;
3437 bool is_dirty = false;
Zhang Xiantao5bb064d2007-11-18 20:29:43 +08003438
Marcelo Tosatti79fac952009-12-23 14:35:26 -02003439 mutex_lock(&kvm->slots_lock);
Zhang Xiantao5bb064d2007-11-18 20:29:43 +08003440
Marcelo Tosattib050b012009-12-23 14:35:22 -02003441 r = -EINVAL;
3442 if (log->slot >= KVM_MEMORY_SLOTS)
Zhang Xiantao5bb064d2007-11-18 20:29:43 +08003443 goto out;
3444
Xiao Guangrong28a37542011-11-24 19:04:35 +08003445 memslot = id_to_memslot(kvm->memslots, log->slot);
Takuya Yoshikawa60c34612012-03-03 14:21:48 +09003446
3447 dirty_bitmap = memslot->dirty_bitmap;
Marcelo Tosattib050b012009-12-23 14:35:22 -02003448 r = -ENOENT;
Takuya Yoshikawa60c34612012-03-03 14:21:48 +09003449 if (!dirty_bitmap)
Marcelo Tosattib050b012009-12-23 14:35:22 -02003450 goto out;
3451
Takuya Yoshikawa87bf6e72010-04-12 19:35:35 +09003452 n = kvm_dirty_bitmap_bytes(memslot);
Marcelo Tosattib050b012009-12-23 14:35:22 -02003453
Takuya Yoshikawa60c34612012-03-03 14:21:48 +09003454 dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
3455 memset(dirty_bitmap_buffer, 0, n);
Marcelo Tosattib050b012009-12-23 14:35:22 -02003456
Takuya Yoshikawa60c34612012-03-03 14:21:48 +09003457 spin_lock(&kvm->mmu_lock);
Marcelo Tosattib050b012009-12-23 14:35:22 -02003458
Takuya Yoshikawa60c34612012-03-03 14:21:48 +09003459 for (i = 0; i < n / sizeof(long); i++) {
3460 unsigned long mask;
3461 gfn_t offset;
Sasha Levincdfca7b2011-12-04 19:36:28 +02003462
Takuya Yoshikawa60c34612012-03-03 14:21:48 +09003463 if (!dirty_bitmap[i])
3464 continue;
Marcelo Tosattib050b012009-12-23 14:35:22 -02003465
Takuya Yoshikawa60c34612012-03-03 14:21:48 +09003466 is_dirty = true;
Takuya Yoshikawa914ebcc2010-04-28 18:50:36 +09003467
Takuya Yoshikawa60c34612012-03-03 14:21:48 +09003468 mask = xchg(&dirty_bitmap[i], 0);
3469 dirty_bitmap_buffer[i] = mask;
Michael S. Tsirkinedde99c2010-10-25 03:21:24 +02003470
Takuya Yoshikawa60c34612012-03-03 14:21:48 +09003471 offset = i * BITS_PER_LONG;
3472 kvm_mmu_write_protect_pt_masked(kvm, memslot, offset, mask);
Zhang Xiantao5bb064d2007-11-18 20:29:43 +08003473 }
Takuya Yoshikawa60c34612012-03-03 14:21:48 +09003474 if (is_dirty)
3475 kvm_flush_remote_tlbs(kvm);
3476
3477 spin_unlock(&kvm->mmu_lock);
3478
3479 r = -EFAULT;
3480 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
3481 goto out;
Marcelo Tosattib050b012009-12-23 14:35:22 -02003482
Zhang Xiantao5bb064d2007-11-18 20:29:43 +08003483 r = 0;
3484out:
Marcelo Tosatti79fac952009-12-23 14:35:26 -02003485 mutex_unlock(&kvm->slots_lock);
Zhang Xiantao5bb064d2007-11-18 20:29:43 +08003486 return r;
3487}
3488
Christoffer Dall23d43cf2012-07-24 08:51:20 -04003489int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event)
3490{
3491 if (!irqchip_in_kernel(kvm))
3492 return -ENXIO;
3493
3494 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3495 irq_event->irq, irq_event->level);
3496 return 0;
3497}
3498
Carsten Otte1fe779f2007-10-29 16:08:35 +01003499long kvm_arch_vm_ioctl(struct file *filp,
3500 unsigned int ioctl, unsigned long arg)
3501{
3502 struct kvm *kvm = filp->private_data;
3503 void __user *argp = (void __user *)arg;
Avi Kivity367e1312009-08-26 14:57:07 +03003504 int r = -ENOTTY;
Dave Hansenf0d66272008-08-11 10:01:45 -07003505 /*
3506 * This union makes it completely explicit to gcc-3.x
3507 * that these two variables' stack usage should be
3508 * combined, not added together.
3509 */
3510 union {
3511 struct kvm_pit_state ps;
Beth Kone9f42752009-07-07 11:50:38 -04003512 struct kvm_pit_state2 ps2;
Jan Kiszkac5ff41c2009-05-14 22:42:53 +02003513 struct kvm_pit_config pit_config;
Dave Hansenf0d66272008-08-11 10:01:45 -07003514 } u;
Carsten Otte1fe779f2007-10-29 16:08:35 +01003515
3516 switch (ioctl) {
3517 case KVM_SET_TSS_ADDR:
3518 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
Carsten Otte1fe779f2007-10-29 16:08:35 +01003519 break;
Sheng Yangb927a3c2009-07-21 10:42:48 +08003520 case KVM_SET_IDENTITY_MAP_ADDR: {
3521 u64 ident_addr;
3522
3523 r = -EFAULT;
3524 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3525 goto out;
3526 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
Sheng Yangb927a3c2009-07-21 10:42:48 +08003527 break;
3528 }
Carsten Otte1fe779f2007-10-29 16:08:35 +01003529 case KVM_SET_NR_MMU_PAGES:
3530 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
Carsten Otte1fe779f2007-10-29 16:08:35 +01003531 break;
3532 case KVM_GET_NR_MMU_PAGES:
3533 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3534 break;
Marcelo Tosatti3ddea122009-10-29 13:44:15 -02003535 case KVM_CREATE_IRQCHIP: {
3536 struct kvm_pic *vpic;
3537
3538 mutex_lock(&kvm->lock);
3539 r = -EEXIST;
3540 if (kvm->arch.vpic)
3541 goto create_irqchip_unlock;
Avi Kivity3e515702012-03-05 14:23:29 +02003542 r = -EINVAL;
3543 if (atomic_read(&kvm->online_vcpus))
3544 goto create_irqchip_unlock;
Carsten Otte1fe779f2007-10-29 16:08:35 +01003545 r = -ENOMEM;
Marcelo Tosatti3ddea122009-10-29 13:44:15 -02003546 vpic = kvm_create_pic(kvm);
3547 if (vpic) {
Carsten Otte1fe779f2007-10-29 16:08:35 +01003548 r = kvm_ioapic_init(kvm);
3549 if (r) {
Takuya Yoshikawa175504c2010-12-16 01:41:37 +09003550 mutex_lock(&kvm->slots_lock);
Wei Yongjun72bb2fc2010-02-09 10:33:03 +08003551 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
Sasha Levin743eeb02011-07-27 16:00:48 +03003552 &vpic->dev_master);
3553 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3554 &vpic->dev_slave);
3555 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3556 &vpic->dev_eclr);
Takuya Yoshikawa175504c2010-12-16 01:41:37 +09003557 mutex_unlock(&kvm->slots_lock);
Marcelo Tosatti3ddea122009-10-29 13:44:15 -02003558 kfree(vpic);
3559 goto create_irqchip_unlock;
Carsten Otte1fe779f2007-10-29 16:08:35 +01003560 }
3561 } else
Marcelo Tosatti3ddea122009-10-29 13:44:15 -02003562 goto create_irqchip_unlock;
3563 smp_wmb();
3564 kvm->arch.vpic = vpic;
3565 smp_wmb();
Avi Kivity399ec802008-11-19 13:58:46 +02003566 r = kvm_setup_default_irq_routing(kvm);
3567 if (r) {
Takuya Yoshikawa175504c2010-12-16 01:41:37 +09003568 mutex_lock(&kvm->slots_lock);
Marcelo Tosatti3ddea122009-10-29 13:44:15 -02003569 mutex_lock(&kvm->irq_lock);
Wei Yongjun72bb2fc2010-02-09 10:33:03 +08003570 kvm_ioapic_destroy(kvm);
3571 kvm_destroy_pic(kvm);
Marcelo Tosatti3ddea122009-10-29 13:44:15 -02003572 mutex_unlock(&kvm->irq_lock);
Takuya Yoshikawa175504c2010-12-16 01:41:37 +09003573 mutex_unlock(&kvm->slots_lock);
Avi Kivity399ec802008-11-19 13:58:46 +02003574 }
Marcelo Tosatti3ddea122009-10-29 13:44:15 -02003575 create_irqchip_unlock:
3576 mutex_unlock(&kvm->lock);
Carsten Otte1fe779f2007-10-29 16:08:35 +01003577 break;
Marcelo Tosatti3ddea122009-10-29 13:44:15 -02003578 }
Sheng Yang78376992008-01-28 05:10:22 +08003579 case KVM_CREATE_PIT:
Jan Kiszkac5ff41c2009-05-14 22:42:53 +02003580 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3581 goto create_pit;
3582 case KVM_CREATE_PIT2:
3583 r = -EFAULT;
3584 if (copy_from_user(&u.pit_config, argp,
3585 sizeof(struct kvm_pit_config)))
3586 goto out;
3587 create_pit:
Marcelo Tosatti79fac952009-12-23 14:35:26 -02003588 mutex_lock(&kvm->slots_lock);
Avi Kivity269e05e2009-01-05 15:21:42 +02003589 r = -EEXIST;
3590 if (kvm->arch.vpit)
3591 goto create_pit_unlock;
Sheng Yang78376992008-01-28 05:10:22 +08003592 r = -ENOMEM;
Jan Kiszkac5ff41c2009-05-14 22:42:53 +02003593 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
Sheng Yang78376992008-01-28 05:10:22 +08003594 if (kvm->arch.vpit)
3595 r = 0;
Avi Kivity269e05e2009-01-05 15:21:42 +02003596 create_pit_unlock:
Marcelo Tosatti79fac952009-12-23 14:35:26 -02003597 mutex_unlock(&kvm->slots_lock);
Sheng Yang78376992008-01-28 05:10:22 +08003598 break;
Carsten Otte1fe779f2007-10-29 16:08:35 +01003599 case KVM_GET_IRQCHIP: {
3600 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
Sasha Levinff5c2c02011-12-04 19:36:29 +02003601 struct kvm_irqchip *chip;
Carsten Otte1fe779f2007-10-29 16:08:35 +01003602
Sasha Levinff5c2c02011-12-04 19:36:29 +02003603 chip = memdup_user(argp, sizeof(*chip));
3604 if (IS_ERR(chip)) {
3605 r = PTR_ERR(chip);
Carsten Otte1fe779f2007-10-29 16:08:35 +01003606 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02003607 }
3608
Carsten Otte1fe779f2007-10-29 16:08:35 +01003609 r = -ENXIO;
3610 if (!irqchip_in_kernel(kvm))
Dave Hansenf0d66272008-08-11 10:01:45 -07003611 goto get_irqchip_out;
3612 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3613 if (r)
3614 goto get_irqchip_out;
3615 r = -EFAULT;
3616 if (copy_to_user(argp, chip, sizeof *chip))
3617 goto get_irqchip_out;
3618 r = 0;
3619 get_irqchip_out:
3620 kfree(chip);
Carsten Otte1fe779f2007-10-29 16:08:35 +01003621 break;
3622 }
3623 case KVM_SET_IRQCHIP: {
3624 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
Sasha Levinff5c2c02011-12-04 19:36:29 +02003625 struct kvm_irqchip *chip;
Carsten Otte1fe779f2007-10-29 16:08:35 +01003626
Sasha Levinff5c2c02011-12-04 19:36:29 +02003627 chip = memdup_user(argp, sizeof(*chip));
3628 if (IS_ERR(chip)) {
3629 r = PTR_ERR(chip);
Carsten Otte1fe779f2007-10-29 16:08:35 +01003630 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02003631 }
3632
Carsten Otte1fe779f2007-10-29 16:08:35 +01003633 r = -ENXIO;
3634 if (!irqchip_in_kernel(kvm))
Dave Hansenf0d66272008-08-11 10:01:45 -07003635 goto set_irqchip_out;
3636 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3637 if (r)
3638 goto set_irqchip_out;
3639 r = 0;
3640 set_irqchip_out:
3641 kfree(chip);
Carsten Otte1fe779f2007-10-29 16:08:35 +01003642 break;
3643 }
Sheng Yange0f63cb2008-03-04 00:50:59 +08003644 case KVM_GET_PIT: {
Sheng Yange0f63cb2008-03-04 00:50:59 +08003645 r = -EFAULT;
Dave Hansenf0d66272008-08-11 10:01:45 -07003646 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
Sheng Yange0f63cb2008-03-04 00:50:59 +08003647 goto out;
3648 r = -ENXIO;
3649 if (!kvm->arch.vpit)
3650 goto out;
Dave Hansenf0d66272008-08-11 10:01:45 -07003651 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
Sheng Yange0f63cb2008-03-04 00:50:59 +08003652 if (r)
3653 goto out;
3654 r = -EFAULT;
Dave Hansenf0d66272008-08-11 10:01:45 -07003655 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
Sheng Yange0f63cb2008-03-04 00:50:59 +08003656 goto out;
3657 r = 0;
3658 break;
3659 }
3660 case KVM_SET_PIT: {
Sheng Yange0f63cb2008-03-04 00:50:59 +08003661 r = -EFAULT;
Dave Hansenf0d66272008-08-11 10:01:45 -07003662 if (copy_from_user(&u.ps, argp, sizeof u.ps))
Sheng Yange0f63cb2008-03-04 00:50:59 +08003663 goto out;
3664 r = -ENXIO;
3665 if (!kvm->arch.vpit)
3666 goto out;
Dave Hansenf0d66272008-08-11 10:01:45 -07003667 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
Sheng Yange0f63cb2008-03-04 00:50:59 +08003668 break;
3669 }
Beth Kone9f42752009-07-07 11:50:38 -04003670 case KVM_GET_PIT2: {
3671 r = -ENXIO;
3672 if (!kvm->arch.vpit)
3673 goto out;
3674 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3675 if (r)
3676 goto out;
3677 r = -EFAULT;
3678 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3679 goto out;
3680 r = 0;
3681 break;
3682 }
3683 case KVM_SET_PIT2: {
3684 r = -EFAULT;
3685 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3686 goto out;
3687 r = -ENXIO;
3688 if (!kvm->arch.vpit)
3689 goto out;
3690 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
Beth Kone9f42752009-07-07 11:50:38 -04003691 break;
3692 }
Marcelo Tosatti52d939a2008-12-30 15:55:06 -02003693 case KVM_REINJECT_CONTROL: {
3694 struct kvm_reinject_control control;
3695 r = -EFAULT;
3696 if (copy_from_user(&control, argp, sizeof(control)))
3697 goto out;
3698 r = kvm_vm_ioctl_reinject(kvm, &control);
Marcelo Tosatti52d939a2008-12-30 15:55:06 -02003699 break;
3700 }
Ed Swierkffde22a2009-10-15 15:21:43 -07003701 case KVM_XEN_HVM_CONFIG: {
3702 r = -EFAULT;
3703 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3704 sizeof(struct kvm_xen_hvm_config)))
3705 goto out;
3706 r = -EINVAL;
3707 if (kvm->arch.xen_hvm_config.flags)
3708 goto out;
3709 r = 0;
3710 break;
3711 }
Glauber Costaafbcf7a2009-10-16 15:28:36 -04003712 case KVM_SET_CLOCK: {
Glauber Costaafbcf7a2009-10-16 15:28:36 -04003713 struct kvm_clock_data user_ns;
3714 u64 now_ns;
3715 s64 delta;
3716
3717 r = -EFAULT;
3718 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3719 goto out;
3720
3721 r = -EINVAL;
3722 if (user_ns.flags)
3723 goto out;
3724
3725 r = 0;
Avi Kivity395c6b02010-10-04 12:55:49 +02003726 local_irq_disable();
Zachary Amsden759379d2010-08-19 22:07:25 -10003727 now_ns = get_kernel_ns();
Glauber Costaafbcf7a2009-10-16 15:28:36 -04003728 delta = user_ns.clock - now_ns;
Avi Kivity395c6b02010-10-04 12:55:49 +02003729 local_irq_enable();
Glauber Costaafbcf7a2009-10-16 15:28:36 -04003730 kvm->arch.kvmclock_offset = delta;
3731 break;
3732 }
3733 case KVM_GET_CLOCK: {
Glauber Costaafbcf7a2009-10-16 15:28:36 -04003734 struct kvm_clock_data user_ns;
3735 u64 now_ns;
3736
Avi Kivity395c6b02010-10-04 12:55:49 +02003737 local_irq_disable();
Zachary Amsden759379d2010-08-19 22:07:25 -10003738 now_ns = get_kernel_ns();
Glauber Costaafbcf7a2009-10-16 15:28:36 -04003739 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
Avi Kivity395c6b02010-10-04 12:55:49 +02003740 local_irq_enable();
Glauber Costaafbcf7a2009-10-16 15:28:36 -04003741 user_ns.flags = 0;
Vasiliy Kulikov97e69aa2010-10-30 22:54:47 +04003742 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
Glauber Costaafbcf7a2009-10-16 15:28:36 -04003743
3744 r = -EFAULT;
3745 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
3746 goto out;
3747 r = 0;
3748 break;
3749 }
3750
Carsten Otte1fe779f2007-10-29 16:08:35 +01003751 default:
3752 ;
3753 }
3754out:
3755 return r;
3756}
3757
Zhang Xiantaoa16b0432007-11-16 14:38:21 +08003758static void kvm_init_msr_list(void)
Carsten Otte043405e2007-10-10 17:16:19 +02003759{
3760 u32 dummy[2];
3761 unsigned i, j;
3762
Glauber Costae3267cb2009-10-06 13:24:50 -04003763 /* skip the first msrs in the list. KVM-specific */
3764 for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
Carsten Otte043405e2007-10-10 17:16:19 +02003765 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
3766 continue;
3767 if (j < i)
3768 msrs_to_save[j] = msrs_to_save[i];
3769 j++;
3770 }
3771 num_msrs_to_save = j;
3772}
3773
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03003774static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3775 const void *v)
Carsten Ottebbd9b642007-10-30 18:44:21 +01003776{
Avi Kivity70252a12010-01-19 12:51:22 +02003777 int handled = 0;
3778 int n;
Carsten Ottebbd9b642007-10-30 18:44:21 +01003779
Avi Kivity70252a12010-01-19 12:51:22 +02003780 do {
3781 n = min(len, 8);
3782 if (!(vcpu->arch.apic &&
3783 !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, n, v))
3784 && kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3785 break;
3786 handled += n;
3787 addr += n;
3788 len -= n;
3789 v += n;
3790 } while (len);
3791
3792 return handled;
Carsten Ottebbd9b642007-10-30 18:44:21 +01003793}
3794
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03003795static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
Carsten Ottebbd9b642007-10-30 18:44:21 +01003796{
Avi Kivity70252a12010-01-19 12:51:22 +02003797 int handled = 0;
3798 int n;
Carsten Ottebbd9b642007-10-30 18:44:21 +01003799
Avi Kivity70252a12010-01-19 12:51:22 +02003800 do {
3801 n = min(len, 8);
3802 if (!(vcpu->arch.apic &&
3803 !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, n, v))
3804 && kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3805 break;
3806 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
3807 handled += n;
3808 addr += n;
3809 len -= n;
3810 v += n;
3811 } while (len);
3812
3813 return handled;
Carsten Ottebbd9b642007-10-30 18:44:21 +01003814}
3815
Gleb Natapov2dafc6c2010-03-18 15:20:16 +02003816static void kvm_set_segment(struct kvm_vcpu *vcpu,
3817 struct kvm_segment *var, int seg)
3818{
3819 kvm_x86_ops->set_segment(vcpu, var, seg);
3820}
3821
3822void kvm_get_segment(struct kvm_vcpu *vcpu,
3823 struct kvm_segment *var, int seg)
3824{
3825 kvm_x86_ops->get_segment(vcpu, var, seg);
3826}
3827
Xiao Guangronge459e322011-11-28 20:42:16 +08003828gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
Joerg Roedel02f59dc2010-09-10 17:30:54 +02003829{
3830 gpa_t t_gpa;
Avi Kivityab9ae312010-11-22 17:53:26 +02003831 struct x86_exception exception;
Joerg Roedel02f59dc2010-09-10 17:30:54 +02003832
3833 BUG_ON(!mmu_is_nested(vcpu));
3834
3835 /* NPT walks are always user-walks */
3836 access |= PFERR_USER_MASK;
Avi Kivityab9ae312010-11-22 17:53:26 +02003837 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, &exception);
Joerg Roedel02f59dc2010-09-10 17:30:54 +02003838
3839 return t_gpa;
3840}
3841
Avi Kivityab9ae312010-11-22 17:53:26 +02003842gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
3843 struct x86_exception *exception)
Gleb Natapov1871c602010-02-10 14:21:32 +02003844{
3845 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
Avi Kivityab9ae312010-11-22 17:53:26 +02003846 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
Gleb Natapov1871c602010-02-10 14:21:32 +02003847}
3848
Avi Kivityab9ae312010-11-22 17:53:26 +02003849 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
3850 struct x86_exception *exception)
Gleb Natapov1871c602010-02-10 14:21:32 +02003851{
3852 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3853 access |= PFERR_FETCH_MASK;
Avi Kivityab9ae312010-11-22 17:53:26 +02003854 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
Gleb Natapov1871c602010-02-10 14:21:32 +02003855}
3856
Avi Kivityab9ae312010-11-22 17:53:26 +02003857gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
3858 struct x86_exception *exception)
Gleb Natapov1871c602010-02-10 14:21:32 +02003859{
3860 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3861 access |= PFERR_WRITE_MASK;
Avi Kivityab9ae312010-11-22 17:53:26 +02003862 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
Gleb Natapov1871c602010-02-10 14:21:32 +02003863}
3864
3865/* uses this to access any guest's mapped memory without checking CPL */
Avi Kivityab9ae312010-11-22 17:53:26 +02003866gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
3867 struct x86_exception *exception)
Gleb Natapov1871c602010-02-10 14:21:32 +02003868{
Avi Kivityab9ae312010-11-22 17:53:26 +02003869 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
Gleb Natapov1871c602010-02-10 14:21:32 +02003870}
3871
3872static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
3873 struct kvm_vcpu *vcpu, u32 access,
Avi Kivitybcc55cb2010-11-22 17:53:22 +02003874 struct x86_exception *exception)
Carsten Ottebbd9b642007-10-30 18:44:21 +01003875{
3876 void *data = val;
Marcelo Tosatti10589a42007-12-20 19:18:22 -05003877 int r = X86EMUL_CONTINUE;
Carsten Ottebbd9b642007-10-30 18:44:21 +01003878
3879 while (bytes) {
Joerg Roedel14dfe852010-09-10 17:30:49 +02003880 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
Avi Kivityab9ae312010-11-22 17:53:26 +02003881 exception);
Carsten Ottebbd9b642007-10-30 18:44:21 +01003882 unsigned offset = addr & (PAGE_SIZE-1);
Izik Eidus77c20022008-12-29 01:42:19 +02003883 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
Carsten Ottebbd9b642007-10-30 18:44:21 +01003884 int ret;
3885
Avi Kivitybcc55cb2010-11-22 17:53:22 +02003886 if (gpa == UNMAPPED_GVA)
Avi Kivityab9ae312010-11-22 17:53:26 +02003887 return X86EMUL_PROPAGATE_FAULT;
Izik Eidus77c20022008-12-29 01:42:19 +02003888 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
Marcelo Tosatti10589a42007-12-20 19:18:22 -05003889 if (ret < 0) {
Gleb Natapovc3cd7ff2010-04-28 19:15:35 +03003890 r = X86EMUL_IO_NEEDED;
Marcelo Tosatti10589a42007-12-20 19:18:22 -05003891 goto out;
3892 }
Carsten Ottebbd9b642007-10-30 18:44:21 +01003893
Izik Eidus77c20022008-12-29 01:42:19 +02003894 bytes -= toread;
3895 data += toread;
3896 addr += toread;
Carsten Ottebbd9b642007-10-30 18:44:21 +01003897 }
Marcelo Tosatti10589a42007-12-20 19:18:22 -05003898out:
Marcelo Tosatti10589a42007-12-20 19:18:22 -05003899 return r;
Carsten Ottebbd9b642007-10-30 18:44:21 +01003900}
Izik Eidus77c20022008-12-29 01:42:19 +02003901
Gleb Natapov1871c602010-02-10 14:21:32 +02003902/* used for instruction fetching */
Avi Kivity0f65dd72011-04-20 13:37:53 +03003903static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
3904 gva_t addr, void *val, unsigned int bytes,
Avi Kivitybcc55cb2010-11-22 17:53:22 +02003905 struct x86_exception *exception)
Gleb Natapov1871c602010-02-10 14:21:32 +02003906{
Avi Kivity0f65dd72011-04-20 13:37:53 +03003907 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
Gleb Natapov1871c602010-02-10 14:21:32 +02003908 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
Avi Kivity0f65dd72011-04-20 13:37:53 +03003909
Gleb Natapov1871c602010-02-10 14:21:32 +02003910 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu,
Avi Kivitybcc55cb2010-11-22 17:53:22 +02003911 access | PFERR_FETCH_MASK,
3912 exception);
Gleb Natapov1871c602010-02-10 14:21:32 +02003913}
3914
Nadav Har'El064aea72011-05-25 23:04:56 +03003915int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
Avi Kivity0f65dd72011-04-20 13:37:53 +03003916 gva_t addr, void *val, unsigned int bytes,
Avi Kivitybcc55cb2010-11-22 17:53:22 +02003917 struct x86_exception *exception)
Gleb Natapov1871c602010-02-10 14:21:32 +02003918{
Avi Kivity0f65dd72011-04-20 13:37:53 +03003919 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
Gleb Natapov1871c602010-02-10 14:21:32 +02003920 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
Avi Kivity0f65dd72011-04-20 13:37:53 +03003921
Gleb Natapov1871c602010-02-10 14:21:32 +02003922 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
Avi Kivitybcc55cb2010-11-22 17:53:22 +02003923 exception);
Gleb Natapov1871c602010-02-10 14:21:32 +02003924}
Nadav Har'El064aea72011-05-25 23:04:56 +03003925EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
Gleb Natapov1871c602010-02-10 14:21:32 +02003926
Avi Kivity0f65dd72011-04-20 13:37:53 +03003927static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
3928 gva_t addr, void *val, unsigned int bytes,
Avi Kivitybcc55cb2010-11-22 17:53:22 +02003929 struct x86_exception *exception)
Gleb Natapov1871c602010-02-10 14:21:32 +02003930{
Avi Kivity0f65dd72011-04-20 13:37:53 +03003931 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
Avi Kivitybcc55cb2010-11-22 17:53:22 +02003932 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
Gleb Natapov1871c602010-02-10 14:21:32 +02003933}
3934
Nadav Har'El6a4d7552011-05-25 23:08:00 +03003935int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
Avi Kivity0f65dd72011-04-20 13:37:53 +03003936 gva_t addr, void *val,
Gleb Natapov2dafc6c2010-03-18 15:20:16 +02003937 unsigned int bytes,
Avi Kivitybcc55cb2010-11-22 17:53:22 +02003938 struct x86_exception *exception)
Izik Eidus77c20022008-12-29 01:42:19 +02003939{
Avi Kivity0f65dd72011-04-20 13:37:53 +03003940 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
Izik Eidus77c20022008-12-29 01:42:19 +02003941 void *data = val;
3942 int r = X86EMUL_CONTINUE;
3943
3944 while (bytes) {
Joerg Roedel14dfe852010-09-10 17:30:49 +02003945 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
3946 PFERR_WRITE_MASK,
Avi Kivityab9ae312010-11-22 17:53:26 +02003947 exception);
Izik Eidus77c20022008-12-29 01:42:19 +02003948 unsigned offset = addr & (PAGE_SIZE-1);
3949 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
3950 int ret;
3951
Avi Kivitybcc55cb2010-11-22 17:53:22 +02003952 if (gpa == UNMAPPED_GVA)
Avi Kivityab9ae312010-11-22 17:53:26 +02003953 return X86EMUL_PROPAGATE_FAULT;
Izik Eidus77c20022008-12-29 01:42:19 +02003954 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
3955 if (ret < 0) {
Gleb Natapovc3cd7ff2010-04-28 19:15:35 +03003956 r = X86EMUL_IO_NEEDED;
Izik Eidus77c20022008-12-29 01:42:19 +02003957 goto out;
3958 }
3959
3960 bytes -= towrite;
3961 data += towrite;
3962 addr += towrite;
3963 }
3964out:
3965 return r;
3966}
Nadav Har'El6a4d7552011-05-25 23:08:00 +03003967EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
Izik Eidus77c20022008-12-29 01:42:19 +02003968
Xiao Guangrongaf7cc7d2011-07-12 03:22:46 +08003969static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
3970 gpa_t *gpa, struct x86_exception *exception,
3971 bool write)
3972{
Avi Kivity97d64b72012-09-12 14:52:00 +03003973 u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
3974 | (write ? PFERR_WRITE_MASK : 0);
Xiao Guangrongaf7cc7d2011-07-12 03:22:46 +08003975
Avi Kivity97d64b72012-09-12 14:52:00 +03003976 if (vcpu_match_mmio_gva(vcpu, gva)
3977 && !permission_fault(vcpu->arch.walk_mmu, vcpu->arch.access, access)) {
Xiao Guangrongbebb1062011-07-12 03:23:20 +08003978 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
3979 (gva & (PAGE_SIZE - 1));
Xiao Guangrong4f022642011-07-12 03:34:24 +08003980 trace_vcpu_match_mmio(gva, *gpa, write, false);
Xiao Guangrongbebb1062011-07-12 03:23:20 +08003981 return 1;
3982 }
3983
Xiao Guangrongaf7cc7d2011-07-12 03:22:46 +08003984 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3985
3986 if (*gpa == UNMAPPED_GVA)
3987 return -1;
3988
3989 /* For APIC access vmexit */
3990 if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3991 return 1;
3992
Xiao Guangrong4f022642011-07-12 03:34:24 +08003993 if (vcpu_match_mmio_gpa(vcpu, *gpa)) {
3994 trace_vcpu_match_mmio(gva, *gpa, write, true);
Xiao Guangrongbebb1062011-07-12 03:23:20 +08003995 return 1;
Xiao Guangrong4f022642011-07-12 03:34:24 +08003996 }
Xiao Guangrongbebb1062011-07-12 03:23:20 +08003997
Xiao Guangrongaf7cc7d2011-07-12 03:22:46 +08003998 return 0;
3999}
4000
Marcelo Tosatti3200f402008-03-29 20:17:59 -03004001int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
Avi Kivitybcc55cb2010-11-22 17:53:22 +02004002 const void *val, int bytes)
Avi Kivity9f811282008-03-02 14:06:05 +02004003{
4004 int ret;
4005
4006 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
4007 if (ret < 0)
4008 return 0;
Xiao Guangrongf57f2ef2011-09-22 16:56:39 +08004009 kvm_mmu_pte_write(vcpu, gpa, val, bytes);
Avi Kivity9f811282008-03-02 14:06:05 +02004010 return 1;
4011}
4012
Xiao Guangrong77d197b2011-07-13 14:31:50 +08004013struct read_write_emulator_ops {
4014 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4015 int bytes);
4016 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4017 void *val, int bytes);
4018 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4019 int bytes, void *val);
4020 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4021 void *val, int bytes);
4022 bool write;
4023};
4024
4025static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4026{
4027 if (vcpu->mmio_read_completed) {
Xiao Guangrong77d197b2011-07-13 14:31:50 +08004028 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
Avi Kivityf78146b2012-04-18 19:22:47 +03004029 vcpu->mmio_fragments[0].gpa, *(u64 *)val);
Xiao Guangrong77d197b2011-07-13 14:31:50 +08004030 vcpu->mmio_read_completed = 0;
4031 return 1;
4032 }
4033
4034 return 0;
4035}
4036
4037static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4038 void *val, int bytes)
4039{
4040 return !kvm_read_guest(vcpu->kvm, gpa, val, bytes);
4041}
4042
4043static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4044 void *val, int bytes)
4045{
4046 return emulator_write_phys(vcpu, gpa, val, bytes);
4047}
4048
4049static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4050{
4051 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4052 return vcpu_mmio_write(vcpu, gpa, bytes, val);
4053}
4054
4055static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4056 void *val, int bytes)
4057{
4058 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4059 return X86EMUL_IO_NEEDED;
4060}
4061
4062static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4063 void *val, int bytes)
4064{
Avi Kivityf78146b2012-04-18 19:22:47 +03004065 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4066
Xiao Guangrong87da7e62012-10-24 14:07:59 +08004067 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
Xiao Guangrong77d197b2011-07-13 14:31:50 +08004068 return X86EMUL_CONTINUE;
4069}
4070
Mathias Krause0fbe9b02012-08-30 01:30:17 +02004071static const struct read_write_emulator_ops read_emultor = {
Xiao Guangrong77d197b2011-07-13 14:31:50 +08004072 .read_write_prepare = read_prepare,
4073 .read_write_emulate = read_emulate,
4074 .read_write_mmio = vcpu_mmio_read,
4075 .read_write_exit_mmio = read_exit_mmio,
4076};
4077
Mathias Krause0fbe9b02012-08-30 01:30:17 +02004078static const struct read_write_emulator_ops write_emultor = {
Xiao Guangrong77d197b2011-07-13 14:31:50 +08004079 .read_write_emulate = write_emulate,
4080 .read_write_mmio = write_mmio,
4081 .read_write_exit_mmio = write_exit_mmio,
4082 .write = true,
4083};
4084
Xiao Guangrong22388a32011-07-13 14:32:31 +08004085static int emulator_read_write_onepage(unsigned long addr, void *val,
4086 unsigned int bytes,
4087 struct x86_exception *exception,
4088 struct kvm_vcpu *vcpu,
Mathias Krause0fbe9b02012-08-30 01:30:17 +02004089 const struct read_write_emulator_ops *ops)
Carsten Ottebbd9b642007-10-30 18:44:21 +01004090{
Xiao Guangrongaf7cc7d2011-07-12 03:22:46 +08004091 gpa_t gpa;
4092 int handled, ret;
Xiao Guangrong22388a32011-07-13 14:32:31 +08004093 bool write = ops->write;
Avi Kivityf78146b2012-04-18 19:22:47 +03004094 struct kvm_mmio_fragment *frag;
Xiao Guangrong22388a32011-07-13 14:32:31 +08004095
4096 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
Carsten Ottebbd9b642007-10-30 18:44:21 +01004097
Xiao Guangrongaf7cc7d2011-07-12 03:22:46 +08004098 if (ret < 0)
Carsten Ottebbd9b642007-10-30 18:44:21 +01004099 return X86EMUL_PROPAGATE_FAULT;
Carsten Ottebbd9b642007-10-30 18:44:21 +01004100
4101 /* For APIC access vmexit */
Xiao Guangrongaf7cc7d2011-07-12 03:22:46 +08004102 if (ret)
Carsten Ottebbd9b642007-10-30 18:44:21 +01004103 goto mmio;
4104
Xiao Guangrong22388a32011-07-13 14:32:31 +08004105 if (ops->read_write_emulate(vcpu, gpa, val, bytes))
Carsten Ottebbd9b642007-10-30 18:44:21 +01004106 return X86EMUL_CONTINUE;
4107
4108mmio:
4109 /*
4110 * Is this MMIO handled locally?
4111 */
Xiao Guangrong22388a32011-07-13 14:32:31 +08004112 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
Avi Kivity70252a12010-01-19 12:51:22 +02004113 if (handled == bytes)
Carsten Ottebbd9b642007-10-30 18:44:21 +01004114 return X86EMUL_CONTINUE;
Carsten Ottebbd9b642007-10-30 18:44:21 +01004115
Avi Kivity70252a12010-01-19 12:51:22 +02004116 gpa += handled;
4117 bytes -= handled;
4118 val += handled;
4119
Xiao Guangrong87da7e62012-10-24 14:07:59 +08004120 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
4121 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
4122 frag->gpa = gpa;
4123 frag->data = val;
4124 frag->len = bytes;
Avi Kivityf78146b2012-04-18 19:22:47 +03004125 return X86EMUL_CONTINUE;
Carsten Ottebbd9b642007-10-30 18:44:21 +01004126}
4127
Xiao Guangrong22388a32011-07-13 14:32:31 +08004128int emulator_read_write(struct x86_emulate_ctxt *ctxt, unsigned long addr,
4129 void *val, unsigned int bytes,
4130 struct x86_exception *exception,
Mathias Krause0fbe9b02012-08-30 01:30:17 +02004131 const struct read_write_emulator_ops *ops)
Carsten Ottebbd9b642007-10-30 18:44:21 +01004132{
Avi Kivity0f65dd72011-04-20 13:37:53 +03004133 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
Avi Kivityf78146b2012-04-18 19:22:47 +03004134 gpa_t gpa;
4135 int rc;
4136
4137 if (ops->read_write_prepare &&
4138 ops->read_write_prepare(vcpu, val, bytes))
4139 return X86EMUL_CONTINUE;
4140
4141 vcpu->mmio_nr_fragments = 0;
Avi Kivity0f65dd72011-04-20 13:37:53 +03004142
Carsten Ottebbd9b642007-10-30 18:44:21 +01004143 /* Crossing a page boundary? */
4144 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
Avi Kivityf78146b2012-04-18 19:22:47 +03004145 int now;
Carsten Ottebbd9b642007-10-30 18:44:21 +01004146
4147 now = -addr & ~PAGE_MASK;
Xiao Guangrong22388a32011-07-13 14:32:31 +08004148 rc = emulator_read_write_onepage(addr, val, now, exception,
4149 vcpu, ops);
4150
Carsten Ottebbd9b642007-10-30 18:44:21 +01004151 if (rc != X86EMUL_CONTINUE)
4152 return rc;
4153 addr += now;
4154 val += now;
4155 bytes -= now;
4156 }
Xiao Guangrong22388a32011-07-13 14:32:31 +08004157
Avi Kivityf78146b2012-04-18 19:22:47 +03004158 rc = emulator_read_write_onepage(addr, val, bytes, exception,
4159 vcpu, ops);
4160 if (rc != X86EMUL_CONTINUE)
4161 return rc;
4162
4163 if (!vcpu->mmio_nr_fragments)
4164 return rc;
4165
4166 gpa = vcpu->mmio_fragments[0].gpa;
4167
4168 vcpu->mmio_needed = 1;
4169 vcpu->mmio_cur_fragment = 0;
4170
Xiao Guangrong87da7e62012-10-24 14:07:59 +08004171 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
Avi Kivityf78146b2012-04-18 19:22:47 +03004172 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
4173 vcpu->run->exit_reason = KVM_EXIT_MMIO;
4174 vcpu->run->mmio.phys_addr = gpa;
4175
4176 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
Xiao Guangrong22388a32011-07-13 14:32:31 +08004177}
4178
4179static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4180 unsigned long addr,
4181 void *val,
4182 unsigned int bytes,
4183 struct x86_exception *exception)
4184{
4185 return emulator_read_write(ctxt, addr, val, bytes,
4186 exception, &read_emultor);
4187}
4188
4189int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
4190 unsigned long addr,
4191 const void *val,
4192 unsigned int bytes,
4193 struct x86_exception *exception)
4194{
4195 return emulator_read_write(ctxt, addr, (void *)val, bytes,
4196 exception, &write_emultor);
Carsten Ottebbd9b642007-10-30 18:44:21 +01004197}
Carsten Ottebbd9b642007-10-30 18:44:21 +01004198
Avi Kivitydaea3e72010-03-15 13:59:54 +02004199#define CMPXCHG_TYPE(t, ptr, old, new) \
4200 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4201
4202#ifdef CONFIG_X86_64
4203# define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4204#else
4205# define CMPXCHG64(ptr, old, new) \
Jan Kiszka9749a6c2010-03-20 10:14:13 +01004206 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
Avi Kivitydaea3e72010-03-15 13:59:54 +02004207#endif
4208
Avi Kivity0f65dd72011-04-20 13:37:53 +03004209static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4210 unsigned long addr,
Carsten Ottebbd9b642007-10-30 18:44:21 +01004211 const void *old,
4212 const void *new,
4213 unsigned int bytes,
Avi Kivity0f65dd72011-04-20 13:37:53 +03004214 struct x86_exception *exception)
Carsten Ottebbd9b642007-10-30 18:44:21 +01004215{
Avi Kivity0f65dd72011-04-20 13:37:53 +03004216 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
Avi Kivitydaea3e72010-03-15 13:59:54 +02004217 gpa_t gpa;
4218 struct page *page;
4219 char *kaddr;
4220 bool exchanged;
4221
Marcelo Tosatti2bacc552007-12-12 10:46:12 -05004222 /* guests cmpxchg8b have to be emulated atomically */
Avi Kivitydaea3e72010-03-15 13:59:54 +02004223 if (bytes > 8 || (bytes & (bytes - 1)))
4224 goto emul_write;
Marcelo Tosatti2bacc552007-12-12 10:46:12 -05004225
Avi Kivitydaea3e72010-03-15 13:59:54 +02004226 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
Marcelo Tosatti10589a42007-12-20 19:18:22 -05004227
Avi Kivitydaea3e72010-03-15 13:59:54 +02004228 if (gpa == UNMAPPED_GVA ||
4229 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4230 goto emul_write;
Marcelo Tosatti2bacc552007-12-12 10:46:12 -05004231
Avi Kivitydaea3e72010-03-15 13:59:54 +02004232 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4233 goto emul_write;
Marcelo Tosatti2bacc552007-12-12 10:46:12 -05004234
Avi Kivitydaea3e72010-03-15 13:59:54 +02004235 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
Xiao Guangrong32cad842012-08-03 15:42:52 +08004236 if (is_error_page(page))
Wei Yongjunc19b8bd2010-07-15 08:51:58 +08004237 goto emul_write;
Izik Eidus72dc67a2008-02-10 18:04:15 +02004238
Cong Wang8fd75e12011-11-25 23:14:17 +08004239 kaddr = kmap_atomic(page);
Avi Kivitydaea3e72010-03-15 13:59:54 +02004240 kaddr += offset_in_page(gpa);
4241 switch (bytes) {
4242 case 1:
4243 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4244 break;
4245 case 2:
4246 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4247 break;
4248 case 4:
4249 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4250 break;
4251 case 8:
4252 exchanged = CMPXCHG64(kaddr, old, new);
4253 break;
4254 default:
4255 BUG();
Marcelo Tosatti2bacc552007-12-12 10:46:12 -05004256 }
Cong Wang8fd75e12011-11-25 23:14:17 +08004257 kunmap_atomic(kaddr);
Avi Kivitydaea3e72010-03-15 13:59:54 +02004258 kvm_release_page_dirty(page);
4259
4260 if (!exchanged)
4261 return X86EMUL_CMPXCHG_FAILED;
4262
Xiao Guangrongf57f2ef2011-09-22 16:56:39 +08004263 kvm_mmu_pte_write(vcpu, gpa, new, bytes);
Gleb Natapov8f6abd02010-04-13 10:21:56 +03004264
4265 return X86EMUL_CONTINUE;
Avi Kivity4a5f48f2010-03-15 13:59:55 +02004266
Marcelo Tosatti3200f402008-03-29 20:17:59 -03004267emul_write:
Avi Kivitydaea3e72010-03-15 13:59:54 +02004268 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
Marcelo Tosatti2bacc552007-12-12 10:46:12 -05004269
Avi Kivity0f65dd72011-04-20 13:37:53 +03004270 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
Carsten Ottebbd9b642007-10-30 18:44:21 +01004271}
4272
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02004273static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4274{
4275 /* TODO: String I/O for in kernel device */
4276 int r;
4277
4278 if (vcpu->arch.pio.in)
4279 r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
4280 vcpu->arch.pio.size, pd);
4281 else
4282 r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
4283 vcpu->arch.pio.port, vcpu->arch.pio.size,
4284 pd);
4285 return r;
4286}
4287
Xiao Guangrong6f6fbe92011-09-22 16:55:10 +08004288static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
4289 unsigned short port, void *val,
4290 unsigned int count, bool in)
4291{
4292 trace_kvm_pio(!in, port, size, count);
4293
4294 vcpu->arch.pio.port = port;
4295 vcpu->arch.pio.in = in;
4296 vcpu->arch.pio.count = count;
4297 vcpu->arch.pio.size = size;
4298
4299 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4300 vcpu->arch.pio.count = 0;
4301 return 1;
4302 }
4303
4304 vcpu->run->exit_reason = KVM_EXIT_IO;
4305 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
4306 vcpu->run->io.size = size;
4307 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4308 vcpu->run->io.count = count;
4309 vcpu->run->io.port = port;
4310
4311 return 0;
4312}
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02004313
Avi Kivityca1d4a92011-04-20 13:37:53 +03004314static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4315 int size, unsigned short port, void *val,
4316 unsigned int count)
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02004317{
Avi Kivityca1d4a92011-04-20 13:37:53 +03004318 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
Xiao Guangrong6f6fbe92011-09-22 16:55:10 +08004319 int ret;
Avi Kivityca1d4a92011-04-20 13:37:53 +03004320
Gleb Natapov79729952010-03-18 15:20:24 +02004321 if (vcpu->arch.pio.count)
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02004322 goto data_avail;
4323
Xiao Guangrong6f6fbe92011-09-22 16:55:10 +08004324 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
4325 if (ret) {
4326data_avail:
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02004327 memcpy(val, vcpu->arch.pio_data, size * count);
Gleb Natapov79729952010-03-18 15:20:24 +02004328 vcpu->arch.pio.count = 0;
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02004329 return 1;
4330 }
4331
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02004332 return 0;
4333}
4334
Avi Kivityca1d4a92011-04-20 13:37:53 +03004335static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4336 int size, unsigned short port,
4337 const void *val, unsigned int count)
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02004338{
Avi Kivityca1d4a92011-04-20 13:37:53 +03004339 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4340
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02004341 memcpy(vcpu->arch.pio_data, val, size * count);
Xiao Guangrong6f6fbe92011-09-22 16:55:10 +08004342 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02004343}
4344
Carsten Ottebbd9b642007-10-30 18:44:21 +01004345static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4346{
4347 return kvm_x86_ops->get_segment_base(vcpu, seg);
4348}
4349
Avi Kivity3cb16fe2011-04-20 15:38:44 +03004350static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
Carsten Ottebbd9b642007-10-30 18:44:21 +01004351{
Avi Kivity3cb16fe2011-04-20 15:38:44 +03004352 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
Carsten Ottebbd9b642007-10-30 18:44:21 +01004353}
4354
Sheng Yangf5f48ee2010-06-30 12:25:15 +08004355int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4356{
4357 if (!need_emulate_wbinvd(vcpu))
4358 return X86EMUL_CONTINUE;
4359
4360 if (kvm_x86_ops->has_wbinvd_exit()) {
Jan Kiszka2eec7342010-11-01 14:01:29 +01004361 int cpu = get_cpu();
4362
4363 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
Sheng Yangf5f48ee2010-06-30 12:25:15 +08004364 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4365 wbinvd_ipi, NULL, 1);
Jan Kiszka2eec7342010-11-01 14:01:29 +01004366 put_cpu();
Sheng Yangf5f48ee2010-06-30 12:25:15 +08004367 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
Jan Kiszka2eec7342010-11-01 14:01:29 +01004368 } else
4369 wbinvd();
Sheng Yangf5f48ee2010-06-30 12:25:15 +08004370 return X86EMUL_CONTINUE;
4371}
4372EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4373
Avi Kivitybcaf5cc2011-04-20 15:53:23 +03004374static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4375{
4376 kvm_emulate_wbinvd(emul_to_vcpu(ctxt));
4377}
4378
Avi Kivity717746e2011-04-20 13:37:53 +03004379int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
Carsten Ottebbd9b642007-10-30 18:44:21 +01004380{
Avi Kivity717746e2011-04-20 13:37:53 +03004381 return _kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
Carsten Ottebbd9b642007-10-30 18:44:21 +01004382}
4383
Avi Kivity717746e2011-04-20 13:37:53 +03004384int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
Carsten Ottebbd9b642007-10-30 18:44:21 +01004385{
Gleb Natapov338dbc92010-04-28 19:15:32 +03004386
Avi Kivity717746e2011-04-20 13:37:53 +03004387 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
Carsten Ottebbd9b642007-10-30 18:44:21 +01004388}
4389
Gleb Natapov52a46612010-03-18 15:20:03 +02004390static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4391{
4392 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4393}
4394
Avi Kivity717746e2011-04-20 13:37:53 +03004395static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
Gleb Natapov52a46612010-03-18 15:20:03 +02004396{
Avi Kivity717746e2011-04-20 13:37:53 +03004397 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
Gleb Natapov52a46612010-03-18 15:20:03 +02004398 unsigned long value;
4399
4400 switch (cr) {
4401 case 0:
4402 value = kvm_read_cr0(vcpu);
4403 break;
4404 case 2:
4405 value = vcpu->arch.cr2;
4406 break;
4407 case 3:
Avi Kivity9f8fe502010-12-05 17:30:00 +02004408 value = kvm_read_cr3(vcpu);
Gleb Natapov52a46612010-03-18 15:20:03 +02004409 break;
4410 case 4:
4411 value = kvm_read_cr4(vcpu);
4412 break;
4413 case 8:
4414 value = kvm_get_cr8(vcpu);
4415 break;
4416 default:
Christoffer Dalla737f252012-06-03 21:17:48 +03004417 kvm_err("%s: unexpected cr %u\n", __func__, cr);
Gleb Natapov52a46612010-03-18 15:20:03 +02004418 return 0;
4419 }
4420
4421 return value;
4422}
4423
Avi Kivity717746e2011-04-20 13:37:53 +03004424static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
Gleb Natapov52a46612010-03-18 15:20:03 +02004425{
Avi Kivity717746e2011-04-20 13:37:53 +03004426 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
Gleb Natapov0f122442010-04-28 19:15:31 +03004427 int res = 0;
4428
Gleb Natapov52a46612010-03-18 15:20:03 +02004429 switch (cr) {
4430 case 0:
Avi Kivity49a9b072010-06-10 17:02:14 +03004431 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
Gleb Natapov52a46612010-03-18 15:20:03 +02004432 break;
4433 case 2:
4434 vcpu->arch.cr2 = val;
4435 break;
4436 case 3:
Avi Kivity23902182010-06-10 17:02:16 +03004437 res = kvm_set_cr3(vcpu, val);
Gleb Natapov52a46612010-03-18 15:20:03 +02004438 break;
4439 case 4:
Avi Kivitya83b29c2010-06-10 17:02:15 +03004440 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
Gleb Natapov52a46612010-03-18 15:20:03 +02004441 break;
4442 case 8:
Andre Przywaraeea1cff2010-12-21 11:12:00 +01004443 res = kvm_set_cr8(vcpu, val);
Gleb Natapov52a46612010-03-18 15:20:03 +02004444 break;
4445 default:
Christoffer Dalla737f252012-06-03 21:17:48 +03004446 kvm_err("%s: unexpected cr %u\n", __func__, cr);
Gleb Natapov0f122442010-04-28 19:15:31 +03004447 res = -1;
Gleb Natapov52a46612010-03-18 15:20:03 +02004448 }
Gleb Natapov0f122442010-04-28 19:15:31 +03004449
4450 return res;
Gleb Natapov52a46612010-03-18 15:20:03 +02004451}
4452
Kevin Wolf4cee4792012-02-08 14:34:41 +01004453static void emulator_set_rflags(struct x86_emulate_ctxt *ctxt, ulong val)
4454{
4455 kvm_set_rflags(emul_to_vcpu(ctxt), val);
4456}
4457
Avi Kivity717746e2011-04-20 13:37:53 +03004458static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
Gleb Natapov9c537242010-03-18 15:20:05 +02004459{
Avi Kivity717746e2011-04-20 13:37:53 +03004460 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
Gleb Natapov9c537242010-03-18 15:20:05 +02004461}
4462
Avi Kivity4bff1e862011-04-20 13:37:53 +03004463static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
Gleb Natapov2dafc6c2010-03-18 15:20:16 +02004464{
Avi Kivity4bff1e862011-04-20 13:37:53 +03004465 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
Gleb Natapov2dafc6c2010-03-18 15:20:16 +02004466}
4467
Avi Kivity4bff1e862011-04-20 13:37:53 +03004468static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
Mohammed Gamal160ce1f2010-08-04 05:44:24 +03004469{
Avi Kivity4bff1e862011-04-20 13:37:53 +03004470 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
Mohammed Gamal160ce1f2010-08-04 05:44:24 +03004471}
4472
Avi Kivity1ac9d0c2011-04-20 15:12:00 +03004473static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4474{
4475 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4476}
4477
4478static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4479{
4480 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4481}
4482
Avi Kivity4bff1e862011-04-20 13:37:53 +03004483static unsigned long emulator_get_cached_segment_base(
4484 struct x86_emulate_ctxt *ctxt, int seg)
Gleb Natapov5951c442010-04-28 19:15:29 +03004485{
Avi Kivity4bff1e862011-04-20 13:37:53 +03004486 return get_segment_base(emul_to_vcpu(ctxt), seg);
Gleb Natapov5951c442010-04-28 19:15:29 +03004487}
4488
Avi Kivity1aa36612011-04-27 13:20:30 +03004489static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
4490 struct desc_struct *desc, u32 *base3,
4491 int seg)
Gleb Natapov2dafc6c2010-03-18 15:20:16 +02004492{
4493 struct kvm_segment var;
4494
Avi Kivity4bff1e862011-04-20 13:37:53 +03004495 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
Avi Kivity1aa36612011-04-27 13:20:30 +03004496 *selector = var.selector;
Gleb Natapov2dafc6c2010-03-18 15:20:16 +02004497
4498 if (var.unusable)
4499 return false;
4500
4501 if (var.g)
4502 var.limit >>= 12;
4503 set_desc_limit(desc, var.limit);
4504 set_desc_base(desc, (unsigned long)var.base);
Gleb Natapov5601d052011-03-07 14:55:06 +02004505#ifdef CONFIG_X86_64
4506 if (base3)
4507 *base3 = var.base >> 32;
4508#endif
Gleb Natapov2dafc6c2010-03-18 15:20:16 +02004509 desc->type = var.type;
4510 desc->s = var.s;
4511 desc->dpl = var.dpl;
4512 desc->p = var.present;
4513 desc->avl = var.avl;
4514 desc->l = var.l;
4515 desc->d = var.db;
4516 desc->g = var.g;
4517
4518 return true;
4519}
4520
Avi Kivity1aa36612011-04-27 13:20:30 +03004521static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
4522 struct desc_struct *desc, u32 base3,
4523 int seg)
Gleb Natapov2dafc6c2010-03-18 15:20:16 +02004524{
Avi Kivity4bff1e862011-04-20 13:37:53 +03004525 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
Gleb Natapov2dafc6c2010-03-18 15:20:16 +02004526 struct kvm_segment var;
4527
Avi Kivity1aa36612011-04-27 13:20:30 +03004528 var.selector = selector;
Gleb Natapov2dafc6c2010-03-18 15:20:16 +02004529 var.base = get_desc_base(desc);
Gleb Natapov5601d052011-03-07 14:55:06 +02004530#ifdef CONFIG_X86_64
4531 var.base |= ((u64)base3) << 32;
4532#endif
Gleb Natapov2dafc6c2010-03-18 15:20:16 +02004533 var.limit = get_desc_limit(desc);
4534 if (desc->g)
4535 var.limit = (var.limit << 12) | 0xfff;
4536 var.type = desc->type;
4537 var.present = desc->p;
4538 var.dpl = desc->dpl;
4539 var.db = desc->d;
4540 var.s = desc->s;
4541 var.l = desc->l;
4542 var.g = desc->g;
4543 var.avl = desc->avl;
4544 var.present = desc->p;
4545 var.unusable = !var.present;
4546 var.padding = 0;
4547
4548 kvm_set_segment(vcpu, &var, seg);
4549 return;
4550}
4551
Avi Kivity717746e2011-04-20 13:37:53 +03004552static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
4553 u32 msr_index, u64 *pdata)
4554{
4555 return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
4556}
4557
4558static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
4559 u32 msr_index, u64 data)
4560{
Will Auld8fe8ab42012-11-29 12:42:12 -08004561 struct msr_data msr;
4562
4563 msr.data = data;
4564 msr.index = msr_index;
4565 msr.host_initiated = false;
4566 return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
Avi Kivity717746e2011-04-20 13:37:53 +03004567}
4568
Avi Kivity222d21a2011-11-10 14:57:30 +02004569static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
4570 u32 pmc, u64 *pdata)
4571{
4572 return kvm_pmu_read_pmc(emul_to_vcpu(ctxt), pmc, pdata);
4573}
4574
Avi Kivity6c3287f2011-04-20 15:43:05 +03004575static void emulator_halt(struct x86_emulate_ctxt *ctxt)
4576{
4577 emul_to_vcpu(ctxt)->arch.halt_request = 1;
4578}
4579
Avi Kivity5037f6f2011-03-28 16:53:59 +02004580static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
4581{
4582 preempt_disable();
Avi Kivity5197b802011-04-20 15:55:40 +03004583 kvm_load_guest_fpu(emul_to_vcpu(ctxt));
Avi Kivity5037f6f2011-03-28 16:53:59 +02004584 /*
4585 * CR0.TS may reference the host fpu state, not the guest fpu state,
4586 * so it may be clear at this point.
4587 */
4588 clts();
4589}
4590
4591static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
4592{
4593 preempt_enable();
4594}
4595
Avi Kivity29535382011-04-20 13:37:53 +03004596static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
Joerg Roedel8a76d7f2011-04-04 12:39:27 +02004597 struct x86_instruction_info *info,
Avi Kivityc4f035c2011-04-04 12:39:22 +02004598 enum x86_intercept_stage stage)
4599{
Avi Kivity29535382011-04-20 13:37:53 +03004600 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
Avi Kivityc4f035c2011-04-04 12:39:22 +02004601}
4602
Avi Kivity0017f932012-06-07 14:10:16 +03004603static void emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
Stephan Bärwolfbdb42f52012-01-12 16:43:03 +01004604 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
4605{
Avi Kivity0017f932012-06-07 14:10:16 +03004606 kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx);
Stephan Bärwolfbdb42f52012-01-12 16:43:03 +01004607}
4608
Avi Kivitydd856ef2012-08-27 23:46:17 +03004609static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
4610{
4611 return kvm_register_read(emul_to_vcpu(ctxt), reg);
4612}
4613
4614static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
4615{
4616 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
4617}
4618
Mathias Krause0225fb52012-08-30 01:30:16 +02004619static const struct x86_emulate_ops emulate_ops = {
Avi Kivitydd856ef2012-08-27 23:46:17 +03004620 .read_gpr = emulator_read_gpr,
4621 .write_gpr = emulator_write_gpr,
Gleb Natapov1871c602010-02-10 14:21:32 +02004622 .read_std = kvm_read_guest_virt_system,
Gleb Natapov2dafc6c2010-03-18 15:20:16 +02004623 .write_std = kvm_write_guest_virt_system,
Gleb Natapov1871c602010-02-10 14:21:32 +02004624 .fetch = kvm_fetch_guest_virt,
Carsten Ottebbd9b642007-10-30 18:44:21 +01004625 .read_emulated = emulator_read_emulated,
4626 .write_emulated = emulator_write_emulated,
4627 .cmpxchg_emulated = emulator_cmpxchg_emulated,
Avi Kivity3cb16fe2011-04-20 15:38:44 +03004628 .invlpg = emulator_invlpg,
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02004629 .pio_in_emulated = emulator_pio_in_emulated,
4630 .pio_out_emulated = emulator_pio_out_emulated,
Avi Kivity1aa36612011-04-27 13:20:30 +03004631 .get_segment = emulator_get_segment,
4632 .set_segment = emulator_set_segment,
Gleb Natapov5951c442010-04-28 19:15:29 +03004633 .get_cached_segment_base = emulator_get_cached_segment_base,
Gleb Natapov2dafc6c2010-03-18 15:20:16 +02004634 .get_gdt = emulator_get_gdt,
Mohammed Gamal160ce1f2010-08-04 05:44:24 +03004635 .get_idt = emulator_get_idt,
Avi Kivity1ac9d0c2011-04-20 15:12:00 +03004636 .set_gdt = emulator_set_gdt,
4637 .set_idt = emulator_set_idt,
Gleb Natapov52a46612010-03-18 15:20:03 +02004638 .get_cr = emulator_get_cr,
4639 .set_cr = emulator_set_cr,
Kevin Wolf4cee4792012-02-08 14:34:41 +01004640 .set_rflags = emulator_set_rflags,
Gleb Natapov9c537242010-03-18 15:20:05 +02004641 .cpl = emulator_get_cpl,
Gleb Natapov35aa5372010-04-28 19:15:27 +03004642 .get_dr = emulator_get_dr,
4643 .set_dr = emulator_set_dr,
Avi Kivity717746e2011-04-20 13:37:53 +03004644 .set_msr = emulator_set_msr,
4645 .get_msr = emulator_get_msr,
Avi Kivity222d21a2011-11-10 14:57:30 +02004646 .read_pmc = emulator_read_pmc,
Avi Kivity6c3287f2011-04-20 15:43:05 +03004647 .halt = emulator_halt,
Avi Kivitybcaf5cc2011-04-20 15:53:23 +03004648 .wbinvd = emulator_wbinvd,
Avi Kivityd6aa1002011-04-20 15:47:13 +03004649 .fix_hypercall = emulator_fix_hypercall,
Avi Kivity5037f6f2011-03-28 16:53:59 +02004650 .get_fpu = emulator_get_fpu,
4651 .put_fpu = emulator_put_fpu,
Avi Kivityc4f035c2011-04-04 12:39:22 +02004652 .intercept = emulator_intercept,
Stephan Bärwolfbdb42f52012-01-12 16:43:03 +01004653 .get_cpuid = emulator_get_cpuid,
Carsten Ottebbd9b642007-10-30 18:44:21 +01004654};
4655
Gleb Natapov95cb2292010-04-28 19:15:43 +03004656static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
4657{
4658 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu, mask);
4659 /*
4660 * an sti; sti; sequence only disable interrupts for the first
4661 * instruction. So, if the last instruction, be it emulated or
4662 * not, left the system with the INT_STI flag enabled, it
4663 * means that the last instruction is an sti. We should not
4664 * leave the flag on in this case. The same goes for mov ss
4665 */
4666 if (!(int_shadow & mask))
4667 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
4668}
4669
Gleb Natapov54b84862010-04-28 19:15:44 +03004670static void inject_emulated_exception(struct kvm_vcpu *vcpu)
4671{
4672 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
Avi Kivityda9cb572010-11-22 17:53:21 +02004673 if (ctxt->exception.vector == PF_VECTOR)
Avi Kivity6389ee92010-11-29 16:12:30 +02004674 kvm_propagate_fault(vcpu, &ctxt->exception);
Avi Kivityda9cb572010-11-22 17:53:21 +02004675 else if (ctxt->exception.error_code_valid)
4676 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
4677 ctxt->exception.error_code);
Gleb Natapov54b84862010-04-28 19:15:44 +03004678 else
Avi Kivityda9cb572010-11-22 17:53:21 +02004679 kvm_queue_exception(vcpu, ctxt->exception.vector);
Gleb Natapov54b84862010-04-28 19:15:44 +03004680}
4681
Avi Kivitydd856ef2012-08-27 23:46:17 +03004682static void init_decode_cache(struct x86_emulate_ctxt *ctxt)
Takuya Yoshikawab5c9ff72011-05-25 11:09:38 +09004683{
Avi Kivity9dac77f2011-06-01 15:34:25 +03004684 memset(&ctxt->twobyte, 0,
Avi Kivitydd856ef2012-08-27 23:46:17 +03004685 (void *)&ctxt->_regs - (void *)&ctxt->twobyte);
Takuya Yoshikawab5c9ff72011-05-25 11:09:38 +09004686
Avi Kivity9dac77f2011-06-01 15:34:25 +03004687 ctxt->fetch.start = 0;
4688 ctxt->fetch.end = 0;
4689 ctxt->io_read.pos = 0;
4690 ctxt->io_read.end = 0;
4691 ctxt->mem_read.pos = 0;
4692 ctxt->mem_read.end = 0;
Takuya Yoshikawab5c9ff72011-05-25 11:09:38 +09004693}
4694
Mohammed Gamal8ec47222010-08-16 00:47:01 +03004695static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
4696{
Takuya Yoshikawaadf52232011-05-25 11:06:16 +09004697 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
Mohammed Gamal8ec47222010-08-16 00:47:01 +03004698 int cs_db, cs_l;
4699
Mohammed Gamal8ec47222010-08-16 00:47:01 +03004700 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4701
Takuya Yoshikawaadf52232011-05-25 11:06:16 +09004702 ctxt->eflags = kvm_get_rflags(vcpu);
4703 ctxt->eip = kvm_rip_read(vcpu);
4704 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
4705 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
4706 cs_l ? X86EMUL_MODE_PROT64 :
4707 cs_db ? X86EMUL_MODE_PROT32 :
4708 X86EMUL_MODE_PROT16;
4709 ctxt->guest_mode = is_guest_mode(vcpu);
4710
Avi Kivitydd856ef2012-08-27 23:46:17 +03004711 init_decode_cache(ctxt);
Gleb Natapov7ae441e2011-03-31 12:06:41 +02004712 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
Mohammed Gamal8ec47222010-08-16 00:47:01 +03004713}
4714
Serge E. Hallyn71f98332011-04-13 09:12:54 -05004715int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
Mohammed Gamal63995652010-09-19 14:34:06 +02004716{
Takuya Yoshikawa9d741912011-05-29 21:53:48 +09004717 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
Mohammed Gamal63995652010-09-19 14:34:06 +02004718 int ret;
4719
4720 init_emulate_ctxt(vcpu);
4721
Avi Kivity9dac77f2011-06-01 15:34:25 +03004722 ctxt->op_bytes = 2;
4723 ctxt->ad_bytes = 2;
4724 ctxt->_eip = ctxt->eip + inc_eip;
Takuya Yoshikawa9d741912011-05-29 21:53:48 +09004725 ret = emulate_int_real(ctxt, irq);
Mohammed Gamal63995652010-09-19 14:34:06 +02004726
4727 if (ret != X86EMUL_CONTINUE)
4728 return EMULATE_FAIL;
4729
Avi Kivity9dac77f2011-06-01 15:34:25 +03004730 ctxt->eip = ctxt->_eip;
Takuya Yoshikawa9d741912011-05-29 21:53:48 +09004731 kvm_rip_write(vcpu, ctxt->eip);
4732 kvm_set_rflags(vcpu, ctxt->eflags);
Mohammed Gamal63995652010-09-19 14:34:06 +02004733
4734 if (irq == NMI_VECTOR)
Avi Kivity7460fb4a2011-09-20 13:43:14 +03004735 vcpu->arch.nmi_pending = 0;
Mohammed Gamal63995652010-09-19 14:34:06 +02004736 else
4737 vcpu->arch.interrupt.pending = false;
4738
4739 return EMULATE_DONE;
4740}
4741EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
4742
Gleb Natapov6d77dbf2010-05-10 11:16:56 +03004743static int handle_emulation_failure(struct kvm_vcpu *vcpu)
4744{
Joerg Roedelfc3a9152010-11-29 17:51:49 +01004745 int r = EMULATE_DONE;
4746
Gleb Natapov6d77dbf2010-05-10 11:16:56 +03004747 ++vcpu->stat.insn_emulation_fail;
4748 trace_kvm_emulate_insn_failed(vcpu);
Joerg Roedelfc3a9152010-11-29 17:51:49 +01004749 if (!is_guest_mode(vcpu)) {
4750 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4751 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4752 vcpu->run->internal.ndata = 0;
4753 r = EMULATE_FAIL;
4754 }
Gleb Natapov6d77dbf2010-05-10 11:16:56 +03004755 kvm_queue_exception(vcpu, UD_VECTOR);
Joerg Roedelfc3a9152010-11-29 17:51:49 +01004756
4757 return r;
Gleb Natapov6d77dbf2010-05-10 11:16:56 +03004758}
4759
Gleb Natapova6f177e2010-07-08 12:41:12 +03004760static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t gva)
4761{
4762 gpa_t gpa;
Xiao Guangrong8e3d9d02012-08-21 10:57:42 +08004763 pfn_t pfn;
Gleb Natapova6f177e2010-07-08 12:41:12 +03004764
Gleb Natapov68be0802010-07-14 19:05:45 +03004765 if (tdp_enabled)
4766 return false;
4767
Gleb Natapova6f177e2010-07-08 12:41:12 +03004768 /*
4769 * if emulation was due to access to shadowed page table
Guo Chao4a969982012-06-28 15:17:27 +08004770 * and it failed try to unshadow page and re-enter the
Gleb Natapova6f177e2010-07-08 12:41:12 +03004771 * guest to let CPU execute the instruction.
4772 */
4773 if (kvm_mmu_unprotect_page_virt(vcpu, gva))
4774 return true;
4775
4776 gpa = kvm_mmu_gva_to_gpa_system(vcpu, gva, NULL);
4777
4778 if (gpa == UNMAPPED_GVA)
4779 return true; /* let cpu generate fault */
4780
Xiao Guangrong8e3d9d02012-08-21 10:57:42 +08004781 /*
4782 * Do not retry the unhandleable instruction if it faults on the
4783 * readonly host memory, otherwise it will goto a infinite loop:
4784 * retry instruction -> write #PF -> emulation fail -> retry
4785 * instruction -> ...
4786 */
4787 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
Xiao Guangrong81c52c52012-10-16 20:10:59 +08004788 if (!is_error_noslot_pfn(pfn)) {
Xiao Guangrong8e3d9d02012-08-21 10:57:42 +08004789 kvm_release_pfn_clean(pfn);
Gleb Natapova6f177e2010-07-08 12:41:12 +03004790 return true;
Xiao Guangrong8e3d9d02012-08-21 10:57:42 +08004791 }
Gleb Natapova6f177e2010-07-08 12:41:12 +03004792
4793 return false;
4794}
4795
Xiao Guangrong1cb3f3a2011-09-22 17:02:48 +08004796static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
4797 unsigned long cr2, int emulation_type)
4798{
4799 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4800 unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
4801
4802 last_retry_eip = vcpu->arch.last_retry_eip;
4803 last_retry_addr = vcpu->arch.last_retry_addr;
4804
4805 /*
4806 * If the emulation is caused by #PF and it is non-page_table
4807 * writing instruction, it means the VM-EXIT is caused by shadow
4808 * page protected, we can zap the shadow page and retry this
4809 * instruction directly.
4810 *
4811 * Note: if the guest uses a non-page-table modifying instruction
4812 * on the PDE that points to the instruction, then we will unmap
4813 * the instruction and go to an infinite loop. So, we cache the
4814 * last retried eip and the last fault address, if we meet the eip
4815 * and the address again, we can break out of the potential infinite
4816 * loop.
4817 */
4818 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
4819
4820 if (!(emulation_type & EMULTYPE_RETRY))
4821 return false;
4822
4823 if (x86_page_table_writing_insn(ctxt))
4824 return false;
4825
4826 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
4827 return false;
4828
4829 vcpu->arch.last_retry_eip = ctxt->eip;
4830 vcpu->arch.last_retry_addr = cr2;
4831
4832 if (!vcpu->arch.mmu.direct_map)
4833 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
4834
4835 kvm_mmu_unprotect_page(vcpu->kvm, gpa >> PAGE_SHIFT);
4836
4837 return true;
4838}
4839
Gleb Natapov716d51a2012-09-03 15:24:26 +03004840static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
4841static int complete_emulated_pio(struct kvm_vcpu *vcpu);
4842
Andre Przywara51d8b662010-12-21 11:12:02 +01004843int x86_emulate_instruction(struct kvm_vcpu *vcpu,
4844 unsigned long cr2,
Andre Przywaradc25e892010-12-21 11:12:07 +01004845 int emulation_type,
4846 void *insn,
4847 int insn_len)
Carsten Ottebbd9b642007-10-30 18:44:21 +01004848{
Gleb Natapov95cb2292010-04-28 19:15:43 +03004849 int r;
Takuya Yoshikawa9d741912011-05-29 21:53:48 +09004850 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
Gleb Natapov7ae441e2011-03-31 12:06:41 +02004851 bool writeback = true;
Carsten Ottebbd9b642007-10-30 18:44:21 +01004852
Avi Kivity26eef702008-07-03 14:59:22 +03004853 kvm_clear_exception_queue(vcpu);
Gleb Natapov8d7d81022011-04-12 12:36:21 +03004854
Sheng Yang571008d2008-01-02 14:49:22 +08004855 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
Mohammed Gamal8ec47222010-08-16 00:47:01 +03004856 init_emulate_ctxt(vcpu);
Takuya Yoshikawa9d741912011-05-29 21:53:48 +09004857 ctxt->interruptibility = 0;
4858 ctxt->have_exception = false;
4859 ctxt->perm_ok = false;
Carsten Ottebbd9b642007-10-30 18:44:21 +01004860
Takuya Yoshikawa9d741912011-05-29 21:53:48 +09004861 ctxt->only_vendor_specific_insn
Avi Kivity40059962011-02-01 16:32:04 +02004862 = emulation_type & EMULTYPE_TRAP_UD;
4863
Takuya Yoshikawa9d741912011-05-29 21:53:48 +09004864 r = x86_decode_insn(ctxt, insn, insn_len);
Joerg Roedeld47f00a2010-09-10 17:30:56 +02004865
Avi Kivitye46479f2010-04-11 13:05:16 +03004866 trace_kvm_emulate_insn_start(vcpu);
Avi Kivityf2b57562007-11-18 15:17:51 +02004867 ++vcpu->stat.insn_emulation;
Takuya Yoshikawa1d2887e2011-07-30 18:03:34 +09004868 if (r != EMULATION_OK) {
Avi Kivity40059962011-02-01 16:32:04 +02004869 if (emulation_type & EMULTYPE_TRAP_UD)
4870 return EMULATE_FAIL;
Gleb Natapova6f177e2010-07-08 12:41:12 +03004871 if (reexecute_instruction(vcpu, cr2))
Carsten Ottebbd9b642007-10-30 18:44:21 +01004872 return EMULATE_DONE;
Gleb Natapov6d77dbf2010-05-10 11:16:56 +03004873 if (emulation_type & EMULTYPE_SKIP)
4874 return EMULATE_FAIL;
4875 return handle_emulation_failure(vcpu);
Carsten Ottebbd9b642007-10-30 18:44:21 +01004876 }
4877 }
4878
Gleb Natapovba8afb62009-04-12 13:36:57 +03004879 if (emulation_type & EMULTYPE_SKIP) {
Avi Kivity9dac77f2011-06-01 15:34:25 +03004880 kvm_rip_write(vcpu, ctxt->_eip);
Gleb Natapovba8afb62009-04-12 13:36:57 +03004881 return EMULATE_DONE;
4882 }
4883
Xiao Guangrong1cb3f3a2011-09-22 17:02:48 +08004884 if (retry_instruction(ctxt, cr2, emulation_type))
4885 return EMULATE_DONE;
4886
Gleb Natapov7ae441e2011-03-31 12:06:41 +02004887 /* this is needed for vmware backdoor interface to work since it
Gleb Natapov4d2179e2010-04-28 19:15:42 +03004888 changes registers values during IO operation */
Gleb Natapov7ae441e2011-03-31 12:06:41 +02004889 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
4890 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
Avi Kivitydd856ef2012-08-27 23:46:17 +03004891 emulator_invalidate_register_cache(ctxt);
Gleb Natapov7ae441e2011-03-31 12:06:41 +02004892 }
Gleb Natapov4d2179e2010-04-28 19:15:42 +03004893
Gleb Natapov5cd21912010-03-18 15:20:26 +02004894restart:
Takuya Yoshikawa9d741912011-05-29 21:53:48 +09004895 r = x86_emulate_insn(ctxt);
Gleb Natapovc3cd7ff2010-04-28 19:15:35 +03004896
Joerg Roedel775fde82011-04-04 12:39:24 +02004897 if (r == EMULATION_INTERCEPTED)
4898 return EMULATE_DONE;
4899
Gleb Natapovd2ddd1c2010-08-25 12:47:43 +03004900 if (r == EMULATION_FAILED) {
Gleb Natapova6f177e2010-07-08 12:41:12 +03004901 if (reexecute_instruction(vcpu, cr2))
Gleb Natapovc3cd7ff2010-04-28 19:15:35 +03004902 return EMULATE_DONE;
4903
Gleb Natapov6d77dbf2010-05-10 11:16:56 +03004904 return handle_emulation_failure(vcpu);
Carsten Ottebbd9b642007-10-30 18:44:21 +01004905 }
4906
Takuya Yoshikawa9d741912011-05-29 21:53:48 +09004907 if (ctxt->have_exception) {
Gleb Natapove85d28f2010-07-29 15:11:52 +03004908 inject_emulated_exception(vcpu);
Gleb Natapovd2ddd1c2010-08-25 12:47:43 +03004909 r = EMULATE_DONE;
4910 } else if (vcpu->arch.pio.count) {
Gleb Natapove85d28f2010-07-29 15:11:52 +03004911 if (!vcpu->arch.pio.in)
4912 vcpu->arch.pio.count = 0;
Gleb Natapov716d51a2012-09-03 15:24:26 +03004913 else {
Gleb Natapov7ae441e2011-03-31 12:06:41 +02004914 writeback = false;
Gleb Natapov716d51a2012-09-03 15:24:26 +03004915 vcpu->arch.complete_userspace_io = complete_emulated_pio;
4916 }
Gleb Natapove85d28f2010-07-29 15:11:52 +03004917 r = EMULATE_DO_MMIO;
Gleb Natapov7ae441e2011-03-31 12:06:41 +02004918 } else if (vcpu->mmio_needed) {
4919 if (!vcpu->mmio_is_write)
4920 writeback = false;
Gleb Natapove85d28f2010-07-29 15:11:52 +03004921 r = EMULATE_DO_MMIO;
Gleb Natapov716d51a2012-09-03 15:24:26 +03004922 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
Gleb Natapov7ae441e2011-03-31 12:06:41 +02004923 } else if (r == EMULATION_RESTART)
Gleb Natapove85d28f2010-07-29 15:11:52 +03004924 goto restart;
Gleb Natapovd2ddd1c2010-08-25 12:47:43 +03004925 else
4926 r = EMULATE_DONE;
Gleb Natapove85d28f2010-07-29 15:11:52 +03004927
Gleb Natapov7ae441e2011-03-31 12:06:41 +02004928 if (writeback) {
Takuya Yoshikawa9d741912011-05-29 21:53:48 +09004929 toggle_interruptibility(vcpu, ctxt->interruptibility);
4930 kvm_set_rflags(vcpu, ctxt->eflags);
Gleb Natapov7ae441e2011-03-31 12:06:41 +02004931 kvm_make_request(KVM_REQ_EVENT, vcpu);
Gleb Natapov7ae441e2011-03-31 12:06:41 +02004932 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
Takuya Yoshikawa9d741912011-05-29 21:53:48 +09004933 kvm_rip_write(vcpu, ctxt->eip);
Gleb Natapov7ae441e2011-03-31 12:06:41 +02004934 } else
4935 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
Gleb Natapov3457e412010-04-28 19:15:38 +03004936
Gleb Natapove85d28f2010-07-29 15:11:52 +03004937 return r;
Carsten Ottebbd9b642007-10-30 18:44:21 +01004938}
Andre Przywara51d8b662010-12-21 11:12:02 +01004939EXPORT_SYMBOL_GPL(x86_emulate_instruction);
Carsten Ottebbd9b642007-10-30 18:44:21 +01004940
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02004941int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
Carsten Ottede7d7892007-10-30 18:44:25 +01004942{
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02004943 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
Avi Kivityca1d4a92011-04-20 13:37:53 +03004944 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
4945 size, port, &val, 1);
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02004946 /* do not return to emulator after return from userspace */
Gleb Natapov79729952010-03-18 15:20:24 +02004947 vcpu->arch.pio.count = 0;
Izik Eidus0f346072008-12-29 01:42:20 +02004948 return ret;
Carsten Ottede7d7892007-10-30 18:44:25 +01004949}
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02004950EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
Carsten Ottede7d7892007-10-30 18:44:25 +01004951
Zachary Amsden8cfdc002010-08-19 22:07:21 -10004952static void tsc_bad(void *info)
Gerd Hoffmannc8076602009-02-04 17:52:04 +01004953{
Tejun Heo0a3aee02010-12-18 16:28:55 +01004954 __this_cpu_write(cpu_tsc_khz, 0);
Zachary Amsden8cfdc002010-08-19 22:07:21 -10004955}
4956
4957static void tsc_khz_changed(void *data)
4958{
4959 struct cpufreq_freqs *freq = data;
4960 unsigned long khz = 0;
4961
4962 if (data)
4963 khz = freq->new;
4964 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
4965 khz = cpufreq_quick_get(raw_smp_processor_id());
4966 if (!khz)
4967 khz = tsc_khz;
Tejun Heo0a3aee02010-12-18 16:28:55 +01004968 __this_cpu_write(cpu_tsc_khz, khz);
Gerd Hoffmannc8076602009-02-04 17:52:04 +01004969}
4970
Gerd Hoffmannc8076602009-02-04 17:52:04 +01004971static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
4972 void *data)
4973{
4974 struct cpufreq_freqs *freq = data;
4975 struct kvm *kvm;
4976 struct kvm_vcpu *vcpu;
4977 int i, send_ipi = 0;
4978
Zachary Amsden8cfdc002010-08-19 22:07:21 -10004979 /*
4980 * We allow guests to temporarily run on slowing clocks,
4981 * provided we notify them after, or to run on accelerating
4982 * clocks, provided we notify them before. Thus time never
4983 * goes backwards.
4984 *
4985 * However, we have a problem. We can't atomically update
4986 * the frequency of a given CPU from this function; it is
4987 * merely a notifier, which can be called from any CPU.
4988 * Changing the TSC frequency at arbitrary points in time
4989 * requires a recomputation of local variables related to
4990 * the TSC for each VCPU. We must flag these local variables
4991 * to be updated and be sure the update takes place with the
4992 * new frequency before any guests proceed.
4993 *
4994 * Unfortunately, the combination of hotplug CPU and frequency
4995 * change creates an intractable locking scenario; the order
4996 * of when these callouts happen is undefined with respect to
4997 * CPU hotplug, and they can race with each other. As such,
4998 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
4999 * undefined; you can actually have a CPU frequency change take
5000 * place in between the computation of X and the setting of the
5001 * variable. To protect against this problem, all updates of
5002 * the per_cpu tsc_khz variable are done in an interrupt
5003 * protected IPI, and all callers wishing to update the value
5004 * must wait for a synchronous IPI to complete (which is trivial
5005 * if the caller is on the CPU already). This establishes the
5006 * necessary total order on variable updates.
5007 *
5008 * Note that because a guest time update may take place
5009 * anytime after the setting of the VCPU's request bit, the
5010 * correct TSC value must be set before the request. However,
5011 * to ensure the update actually makes it to any guest which
5012 * starts running in hardware virtualization between the set
5013 * and the acquisition of the spinlock, we must also ping the
5014 * CPU after setting the request bit.
5015 *
5016 */
5017
Gerd Hoffmannc8076602009-02-04 17:52:04 +01005018 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
5019 return 0;
5020 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
5021 return 0;
Zachary Amsden8cfdc002010-08-19 22:07:21 -10005022
5023 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
Gerd Hoffmannc8076602009-02-04 17:52:04 +01005024
Jan Kiszkae935b832011-02-08 12:55:33 +01005025 raw_spin_lock(&kvm_lock);
Gerd Hoffmannc8076602009-02-04 17:52:04 +01005026 list_for_each_entry(kvm, &vm_list, vm_list) {
Gleb Natapov988a2ca2009-06-09 15:56:29 +03005027 kvm_for_each_vcpu(i, vcpu, kvm) {
Gerd Hoffmannc8076602009-02-04 17:52:04 +01005028 if (vcpu->cpu != freq->cpu)
5029 continue;
Zachary Amsdenc2855452010-09-18 14:38:15 -10005030 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
Gerd Hoffmannc8076602009-02-04 17:52:04 +01005031 if (vcpu->cpu != smp_processor_id())
Zachary Amsden8cfdc002010-08-19 22:07:21 -10005032 send_ipi = 1;
Gerd Hoffmannc8076602009-02-04 17:52:04 +01005033 }
5034 }
Jan Kiszkae935b832011-02-08 12:55:33 +01005035 raw_spin_unlock(&kvm_lock);
Gerd Hoffmannc8076602009-02-04 17:52:04 +01005036
5037 if (freq->old < freq->new && send_ipi) {
5038 /*
5039 * We upscale the frequency. Must make the guest
5040 * doesn't see old kvmclock values while running with
5041 * the new frequency, otherwise we risk the guest sees
5042 * time go backwards.
5043 *
5044 * In case we update the frequency for another cpu
5045 * (which might be in guest context) send an interrupt
5046 * to kick the cpu out of guest context. Next time
5047 * guest context is entered kvmclock will be updated,
5048 * so the guest will not see stale values.
5049 */
Zachary Amsden8cfdc002010-08-19 22:07:21 -10005050 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
Gerd Hoffmannc8076602009-02-04 17:52:04 +01005051 }
5052 return 0;
5053}
5054
5055static struct notifier_block kvmclock_cpufreq_notifier_block = {
Zachary Amsden8cfdc002010-08-19 22:07:21 -10005056 .notifier_call = kvmclock_cpufreq_notifier
5057};
5058
5059static int kvmclock_cpu_notifier(struct notifier_block *nfb,
5060 unsigned long action, void *hcpu)
5061{
5062 unsigned int cpu = (unsigned long)hcpu;
5063
5064 switch (action) {
5065 case CPU_ONLINE:
5066 case CPU_DOWN_FAILED:
5067 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5068 break;
5069 case CPU_DOWN_PREPARE:
5070 smp_call_function_single(cpu, tsc_bad, NULL, 1);
5071 break;
5072 }
5073 return NOTIFY_OK;
5074}
5075
5076static struct notifier_block kvmclock_cpu_notifier_block = {
5077 .notifier_call = kvmclock_cpu_notifier,
5078 .priority = -INT_MAX
Gerd Hoffmannc8076602009-02-04 17:52:04 +01005079};
5080
Zachary Amsdenb820cc02009-09-29 11:38:34 -10005081static void kvm_timer_init(void)
5082{
5083 int cpu;
5084
Zachary Amsdenc2855452010-09-18 14:38:15 -10005085 max_tsc_khz = tsc_khz;
Zachary Amsden8cfdc002010-08-19 22:07:21 -10005086 register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
Zachary Amsdenb820cc02009-09-29 11:38:34 -10005087 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
Zachary Amsdenc2855452010-09-18 14:38:15 -10005088#ifdef CONFIG_CPU_FREQ
5089 struct cpufreq_policy policy;
5090 memset(&policy, 0, sizeof(policy));
Avi Kivity3e26f232010-12-16 12:16:34 +02005091 cpu = get_cpu();
5092 cpufreq_get_policy(&policy, cpu);
Zachary Amsdenc2855452010-09-18 14:38:15 -10005093 if (policy.cpuinfo.max_freq)
5094 max_tsc_khz = policy.cpuinfo.max_freq;
Avi Kivity3e26f232010-12-16 12:16:34 +02005095 put_cpu();
Zachary Amsdenc2855452010-09-18 14:38:15 -10005096#endif
Zachary Amsdenb820cc02009-09-29 11:38:34 -10005097 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
5098 CPUFREQ_TRANSITION_NOTIFIER);
5099 }
Zachary Amsdenc2855452010-09-18 14:38:15 -10005100 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
Zachary Amsden8cfdc002010-08-19 22:07:21 -10005101 for_each_online_cpu(cpu)
5102 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
Zachary Amsdenb820cc02009-09-29 11:38:34 -10005103}
5104
Zhang, Yanminff9d07a2010-04-19 13:32:45 +08005105static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5106
Gleb Natapovf5132b02011-11-10 14:57:22 +02005107int kvm_is_in_guest(void)
Zhang, Yanminff9d07a2010-04-19 13:32:45 +08005108{
Alex,Shi086c9852011-10-20 15:34:01 +08005109 return __this_cpu_read(current_vcpu) != NULL;
Zhang, Yanminff9d07a2010-04-19 13:32:45 +08005110}
5111
5112static int kvm_is_user_mode(void)
5113{
5114 int user_mode = 3;
Zhang, Yanmindcf46b92010-04-20 10:13:58 +08005115
Alex,Shi086c9852011-10-20 15:34:01 +08005116 if (__this_cpu_read(current_vcpu))
5117 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
Zhang, Yanmindcf46b92010-04-20 10:13:58 +08005118
Zhang, Yanminff9d07a2010-04-19 13:32:45 +08005119 return user_mode != 0;
5120}
5121
5122static unsigned long kvm_get_guest_ip(void)
5123{
5124 unsigned long ip = 0;
Zhang, Yanmindcf46b92010-04-20 10:13:58 +08005125
Alex,Shi086c9852011-10-20 15:34:01 +08005126 if (__this_cpu_read(current_vcpu))
5127 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
Zhang, Yanmindcf46b92010-04-20 10:13:58 +08005128
Zhang, Yanminff9d07a2010-04-19 13:32:45 +08005129 return ip;
5130}
5131
5132static struct perf_guest_info_callbacks kvm_guest_cbs = {
5133 .is_in_guest = kvm_is_in_guest,
5134 .is_user_mode = kvm_is_user_mode,
5135 .get_guest_ip = kvm_get_guest_ip,
5136};
5137
5138void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
5139{
Alex,Shi086c9852011-10-20 15:34:01 +08005140 __this_cpu_write(current_vcpu, vcpu);
Zhang, Yanminff9d07a2010-04-19 13:32:45 +08005141}
5142EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
5143
5144void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
5145{
Alex,Shi086c9852011-10-20 15:34:01 +08005146 __this_cpu_write(current_vcpu, NULL);
Zhang, Yanminff9d07a2010-04-19 13:32:45 +08005147}
5148EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
5149
Xiao Guangrongce88dec2011-07-12 03:33:44 +08005150static void kvm_set_mmio_spte_mask(void)
5151{
5152 u64 mask;
5153 int maxphyaddr = boot_cpu_data.x86_phys_bits;
5154
5155 /*
5156 * Set the reserved bits and the present bit of an paging-structure
5157 * entry to generate page fault with PFER.RSV = 1.
5158 */
5159 mask = ((1ull << (62 - maxphyaddr + 1)) - 1) << maxphyaddr;
5160 mask |= 1ull;
5161
5162#ifdef CONFIG_X86_64
5163 /*
5164 * If reserved bit is not supported, clear the present bit to disable
5165 * mmio page fault.
5166 */
5167 if (maxphyaddr == 52)
5168 mask &= ~1ull;
5169#endif
5170
5171 kvm_mmu_set_mmio_spte_mask(mask);
5172}
5173
Marcelo Tosatti16e8d742012-11-27 23:29:00 -02005174#ifdef CONFIG_X86_64
5175static void pvclock_gtod_update_fn(struct work_struct *work)
5176{
Marcelo Tosattid8281992012-11-27 23:29:01 -02005177 struct kvm *kvm;
5178
5179 struct kvm_vcpu *vcpu;
5180 int i;
5181
5182 raw_spin_lock(&kvm_lock);
5183 list_for_each_entry(kvm, &vm_list, vm_list)
5184 kvm_for_each_vcpu(i, vcpu, kvm)
5185 set_bit(KVM_REQ_MASTERCLOCK_UPDATE, &vcpu->requests);
5186 atomic_set(&kvm_guest_has_master_clock, 0);
5187 raw_spin_unlock(&kvm_lock);
Marcelo Tosatti16e8d742012-11-27 23:29:00 -02005188}
5189
5190static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
5191
5192/*
5193 * Notification about pvclock gtod data update.
5194 */
5195static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
5196 void *priv)
5197{
5198 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
5199 struct timekeeper *tk = priv;
5200
5201 update_pvclock_gtod(tk);
5202
5203 /* disable master clock if host does not trust, or does not
5204 * use, TSC clocksource
5205 */
5206 if (gtod->clock.vclock_mode != VCLOCK_TSC &&
5207 atomic_read(&kvm_guest_has_master_clock) != 0)
5208 queue_work(system_long_wq, &pvclock_gtod_work);
5209
5210 return 0;
5211}
5212
5213static struct notifier_block pvclock_gtod_notifier = {
5214 .notifier_call = pvclock_gtod_notify,
5215};
5216#endif
5217
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08005218int kvm_arch_init(void *opaque)
Carsten Otte043405e2007-10-10 17:16:19 +02005219{
Zachary Amsdenb820cc02009-09-29 11:38:34 -10005220 int r;
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08005221 struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
5222
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08005223 if (kvm_x86_ops) {
5224 printk(KERN_ERR "kvm: already loaded the other module\n");
Zhang Xiantao56c6d282007-11-18 20:43:21 +08005225 r = -EEXIST;
5226 goto out;
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08005227 }
5228
5229 if (!ops->cpu_has_kvm_support()) {
5230 printk(KERN_ERR "kvm: no hardware support\n");
Zhang Xiantao56c6d282007-11-18 20:43:21 +08005231 r = -EOPNOTSUPP;
5232 goto out;
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08005233 }
5234 if (ops->disabled_by_bios()) {
5235 printk(KERN_ERR "kvm: disabled by bios\n");
Zhang Xiantao56c6d282007-11-18 20:43:21 +08005236 r = -EOPNOTSUPP;
5237 goto out;
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08005238 }
5239
Marcelo Tosatti013f6a52013-01-03 11:41:39 -02005240 r = -ENOMEM;
5241 shared_msrs = alloc_percpu(struct kvm_shared_msrs);
5242 if (!shared_msrs) {
5243 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
5244 goto out;
5245 }
5246
Avi Kivity97db56c2008-01-13 13:23:56 +02005247 r = kvm_mmu_module_init();
5248 if (r)
Marcelo Tosatti013f6a52013-01-03 11:41:39 -02005249 goto out_free_percpu;
Avi Kivity97db56c2008-01-13 13:23:56 +02005250
Xiao Guangrongce88dec2011-07-12 03:33:44 +08005251 kvm_set_mmio_spte_mask();
Avi Kivity97db56c2008-01-13 13:23:56 +02005252 kvm_init_msr_list();
5253
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08005254 kvm_x86_ops = ops;
Sheng Yang7b523452008-04-25 21:13:50 +08005255 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
Sheng Yang4b12f0d2009-04-27 20:35:42 +08005256 PT_DIRTY_MASK, PT64_NX_MASK, 0);
Gerd Hoffmannc8076602009-02-04 17:52:04 +01005257
Zachary Amsdenb820cc02009-09-29 11:38:34 -10005258 kvm_timer_init();
Gerd Hoffmannc8076602009-02-04 17:52:04 +01005259
Zhang, Yanminff9d07a2010-04-19 13:32:45 +08005260 perf_register_guest_info_callbacks(&kvm_guest_cbs);
5261
Dexuan Cui2acf9232010-06-10 11:27:12 +08005262 if (cpu_has_xsave)
5263 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
5264
Gleb Natapovc5cc4212012-08-05 15:58:30 +03005265 kvm_lapic_init();
Marcelo Tosatti16e8d742012-11-27 23:29:00 -02005266#ifdef CONFIG_X86_64
5267 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
5268#endif
5269
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08005270 return 0;
Zhang Xiantao56c6d282007-11-18 20:43:21 +08005271
Marcelo Tosatti013f6a52013-01-03 11:41:39 -02005272out_free_percpu:
5273 free_percpu(shared_msrs);
Zhang Xiantao56c6d282007-11-18 20:43:21 +08005274out:
Zhang Xiantao56c6d282007-11-18 20:43:21 +08005275 return r;
Carsten Otte043405e2007-10-10 17:16:19 +02005276}
Hollis Blanchard8776e512007-10-31 17:24:24 -05005277
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08005278void kvm_arch_exit(void)
5279{
Zhang, Yanminff9d07a2010-04-19 13:32:45 +08005280 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
5281
Jan Kiszka888d2562009-04-17 19:24:58 +02005282 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5283 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
5284 CPUFREQ_TRANSITION_NOTIFIER);
Zachary Amsden8cfdc002010-08-19 22:07:21 -10005285 unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
Marcelo Tosatti16e8d742012-11-27 23:29:00 -02005286#ifdef CONFIG_X86_64
5287 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
5288#endif
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08005289 kvm_x86_ops = NULL;
Zhang Xiantao56c6d282007-11-18 20:43:21 +08005290 kvm_mmu_module_exit();
Marcelo Tosatti013f6a52013-01-03 11:41:39 -02005291 free_percpu(shared_msrs);
Zhang Xiantao56c6d282007-11-18 20:43:21 +08005292}
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08005293
Hollis Blanchard8776e512007-10-31 17:24:24 -05005294int kvm_emulate_halt(struct kvm_vcpu *vcpu)
5295{
5296 ++vcpu->stat.halt_exits;
5297 if (irqchip_in_kernel(vcpu->kvm)) {
Avi Kivitya4535292008-04-13 17:54:35 +03005298 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
Hollis Blanchard8776e512007-10-31 17:24:24 -05005299 return 1;
5300 } else {
5301 vcpu->run->exit_reason = KVM_EXIT_HLT;
5302 return 0;
5303 }
5304}
5305EXPORT_SYMBOL_GPL(kvm_emulate_halt);
5306
Gleb Natapov55cd8e52010-01-17 15:51:22 +02005307int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
5308{
5309 u64 param, ingpa, outgpa, ret;
5310 uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
5311 bool fast, longmode;
5312 int cs_db, cs_l;
5313
5314 /*
5315 * hypercall generates UD from non zero cpl and real mode
5316 * per HYPER-V spec
5317 */
Avi Kivity3eeb3282010-01-21 15:31:48 +02005318 if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
Gleb Natapov55cd8e52010-01-17 15:51:22 +02005319 kvm_queue_exception(vcpu, UD_VECTOR);
5320 return 0;
5321 }
5322
5323 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5324 longmode = is_long_mode(vcpu) && cs_l == 1;
5325
5326 if (!longmode) {
Gleb Natapovccd46932010-01-19 15:06:38 +02005327 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
5328 (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
5329 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
5330 (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
5331 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
5332 (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
Gleb Natapov55cd8e52010-01-17 15:51:22 +02005333 }
5334#ifdef CONFIG_X86_64
5335 else {
5336 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
5337 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
5338 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
5339 }
5340#endif
5341
5342 code = param & 0xffff;
5343 fast = (param >> 16) & 0x1;
5344 rep_cnt = (param >> 32) & 0xfff;
5345 rep_idx = (param >> 48) & 0xfff;
5346
5347 trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
5348
Gleb Natapovc25bc162010-01-17 15:51:24 +02005349 switch (code) {
5350 case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
5351 kvm_vcpu_on_spin(vcpu);
5352 break;
5353 default:
5354 res = HV_STATUS_INVALID_HYPERCALL_CODE;
5355 break;
5356 }
Gleb Natapov55cd8e52010-01-17 15:51:22 +02005357
5358 ret = res | (((u64)rep_done & 0xfff) << 32);
5359 if (longmode) {
5360 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5361 } else {
5362 kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
5363 kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
5364 }
5365
5366 return 1;
5367}
5368
Hollis Blanchard8776e512007-10-31 17:24:24 -05005369int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
5370{
5371 unsigned long nr, a0, a1, a2, a3, ret;
Marcelo Tosatti2f333bc2008-02-22 12:21:37 -05005372 int r = 1;
Hollis Blanchard8776e512007-10-31 17:24:24 -05005373
Gleb Natapov55cd8e52010-01-17 15:51:22 +02005374 if (kvm_hv_hypercall_enabled(vcpu->kvm))
5375 return kvm_hv_hypercall(vcpu);
5376
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03005377 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
5378 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
5379 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
5380 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
5381 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
Hollis Blanchard8776e512007-10-31 17:24:24 -05005382
Marcelo Tosatti229456f2009-06-17 09:22:14 -03005383 trace_kvm_hypercall(nr, a0, a1, a2, a3);
Feng (Eric) Liu2714d1d2008-04-10 15:31:10 -04005384
Hollis Blanchard8776e512007-10-31 17:24:24 -05005385 if (!is_long_mode(vcpu)) {
5386 nr &= 0xFFFFFFFF;
5387 a0 &= 0xFFFFFFFF;
5388 a1 &= 0xFFFFFFFF;
5389 a2 &= 0xFFFFFFFF;
5390 a3 &= 0xFFFFFFFF;
5391 }
5392
Jan Kiszka07708c42009-08-03 18:43:28 +02005393 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
5394 ret = -KVM_EPERM;
5395 goto out;
5396 }
5397
Hollis Blanchard8776e512007-10-31 17:24:24 -05005398 switch (nr) {
Avi Kivityb93463a2007-10-25 16:52:32 +02005399 case KVM_HC_VAPIC_POLL_IRQ:
5400 ret = 0;
5401 break;
Hollis Blanchard8776e512007-10-31 17:24:24 -05005402 default:
5403 ret = -KVM_ENOSYS;
5404 break;
5405 }
Jan Kiszka07708c42009-08-03 18:43:28 +02005406out:
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03005407 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
Amit Shahf11c3a82008-02-21 01:00:30 +05305408 ++vcpu->stat.hypercalls;
Marcelo Tosatti2f333bc2008-02-22 12:21:37 -05005409 return r;
Hollis Blanchard8776e512007-10-31 17:24:24 -05005410}
5411EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
5412
Jan Kiszkab6785de2012-09-20 07:43:17 +02005413static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
Hollis Blanchard8776e512007-10-31 17:24:24 -05005414{
Avi Kivityd6aa1002011-04-20 15:47:13 +03005415 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
Hollis Blanchard8776e512007-10-31 17:24:24 -05005416 char instruction[3];
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03005417 unsigned long rip = kvm_rip_read(vcpu);
Hollis Blanchard8776e512007-10-31 17:24:24 -05005418
Hollis Blanchard8776e512007-10-31 17:24:24 -05005419 /*
5420 * Blow out the MMU to ensure that no other VCPU has an active mapping
5421 * to ensure that the updated hypercall appears atomically across all
5422 * VCPUs.
5423 */
5424 kvm_mmu_zap_all(vcpu->kvm);
5425
Hollis Blanchard8776e512007-10-31 17:24:24 -05005426 kvm_x86_ops->patch_hypercall(vcpu, instruction);
Hollis Blanchard8776e512007-10-31 17:24:24 -05005427
Takuya Yoshikawa9d741912011-05-29 21:53:48 +09005428 return emulator_write_emulated(ctxt, rip, instruction, 3, NULL);
Hollis Blanchard8776e512007-10-31 17:24:24 -05005429}
5430
Hollis Blanchardd0752062007-10-31 17:24:25 -05005431/*
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005432 * Check if userspace requested an interrupt window, and that the
5433 * interrupt window is open.
5434 *
5435 * No need to exit to userspace if we already have an interrupt queued.
5436 */
Avi Kivity851ba692009-08-24 11:10:17 +03005437static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005438{
Gleb Natapov80618232009-04-21 17:44:56 +03005439 return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
Avi Kivity851ba692009-08-24 11:10:17 +03005440 vcpu->run->request_interrupt_window &&
Gleb Natapov5df56642009-04-21 17:44:59 +03005441 kvm_arch_interrupt_allowed(vcpu));
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005442}
5443
Avi Kivity851ba692009-08-24 11:10:17 +03005444static void post_kvm_run_save(struct kvm_vcpu *vcpu)
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005445{
Avi Kivity851ba692009-08-24 11:10:17 +03005446 struct kvm_run *kvm_run = vcpu->run;
5447
Jan Kiszka91586a32009-10-05 13:07:21 +02005448 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
Avi Kivity2d3ad1f2008-02-24 11:20:43 +02005449 kvm_run->cr8 = kvm_get_cr8(vcpu);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005450 kvm_run->apic_base = kvm_get_apic_base(vcpu);
Jan Kiszka45312202008-12-11 16:54:54 +01005451 if (irqchip_in_kernel(vcpu->kvm))
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005452 kvm_run->ready_for_interrupt_injection = 1;
Jan Kiszka45312202008-12-11 16:54:54 +01005453 else
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005454 kvm_run->ready_for_interrupt_injection =
Gleb Natapovfa9726b2009-05-11 13:35:47 +03005455 kvm_arch_interrupt_allowed(vcpu) &&
5456 !kvm_cpu_has_interrupt(vcpu) &&
5457 !kvm_event_needs_reinjection(vcpu);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005458}
5459
Xiao Guangrong44841412012-09-07 14:14:20 +08005460static int vapic_enter(struct kvm_vcpu *vcpu)
Avi Kivityb93463a2007-10-25 16:52:32 +02005461{
5462 struct kvm_lapic *apic = vcpu->arch.apic;
5463 struct page *page;
5464
5465 if (!apic || !apic->vapic_addr)
Xiao Guangrong44841412012-09-07 14:14:20 +08005466 return 0;
Avi Kivityb93463a2007-10-25 16:52:32 +02005467
5468 page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
Xiao Guangrong44841412012-09-07 14:14:20 +08005469 if (is_error_page(page))
5470 return -EFAULT;
Izik Eidus72dc67a2008-02-10 18:04:15 +02005471
5472 vcpu->arch.apic->vapic_page = page;
Xiao Guangrong44841412012-09-07 14:14:20 +08005473 return 0;
Avi Kivityb93463a2007-10-25 16:52:32 +02005474}
5475
5476static void vapic_exit(struct kvm_vcpu *vcpu)
5477{
5478 struct kvm_lapic *apic = vcpu->arch.apic;
Marcelo Tosattif656ce02009-12-23 14:35:25 -02005479 int idx;
Avi Kivityb93463a2007-10-25 16:52:32 +02005480
5481 if (!apic || !apic->vapic_addr)
5482 return;
5483
Marcelo Tosattif656ce02009-12-23 14:35:25 -02005484 idx = srcu_read_lock(&vcpu->kvm->srcu);
Avi Kivityb93463a2007-10-25 16:52:32 +02005485 kvm_release_page_dirty(apic->vapic_page);
5486 mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
Marcelo Tosattif656ce02009-12-23 14:35:25 -02005487 srcu_read_unlock(&vcpu->kvm->srcu, idx);
Avi Kivityb93463a2007-10-25 16:52:32 +02005488}
5489
Gleb Natapov95ba8273132009-04-21 17:45:08 +03005490static void update_cr8_intercept(struct kvm_vcpu *vcpu)
5491{
5492 int max_irr, tpr;
5493
5494 if (!kvm_x86_ops->update_cr8_intercept)
5495 return;
5496
Avi Kivity88c808f2009-08-17 22:49:40 +03005497 if (!vcpu->arch.apic)
5498 return;
5499
Gleb Natapov8db3baa2009-05-11 13:35:54 +03005500 if (!vcpu->arch.apic->vapic_addr)
5501 max_irr = kvm_lapic_find_highest_irr(vcpu);
5502 else
5503 max_irr = -1;
Gleb Natapov95ba8273132009-04-21 17:45:08 +03005504
5505 if (max_irr != -1)
5506 max_irr >>= 4;
5507
5508 tpr = kvm_lapic_get_cr8(vcpu);
5509
5510 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
5511}
5512
Avi Kivity851ba692009-08-24 11:10:17 +03005513static void inject_pending_event(struct kvm_vcpu *vcpu)
Gleb Natapov95ba8273132009-04-21 17:45:08 +03005514{
5515 /* try to reinject previous events if any */
Gleb Natapovb59bb7b2009-07-09 15:33:51 +03005516 if (vcpu->arch.exception.pending) {
Avi Kivity5c1c85d02010-03-11 13:01:59 +02005517 trace_kvm_inj_exception(vcpu->arch.exception.nr,
5518 vcpu->arch.exception.has_error_code,
5519 vcpu->arch.exception.error_code);
Gleb Natapovb59bb7b2009-07-09 15:33:51 +03005520 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
5521 vcpu->arch.exception.has_error_code,
Joerg Roedelce7ddec2010-04-22 12:33:13 +02005522 vcpu->arch.exception.error_code,
5523 vcpu->arch.exception.reinject);
Gleb Natapovb59bb7b2009-07-09 15:33:51 +03005524 return;
5525 }
5526
Gleb Natapov95ba8273132009-04-21 17:45:08 +03005527 if (vcpu->arch.nmi_injected) {
5528 kvm_x86_ops->set_nmi(vcpu);
5529 return;
5530 }
5531
5532 if (vcpu->arch.interrupt.pending) {
Gleb Natapov66fd3f72009-05-11 13:35:50 +03005533 kvm_x86_ops->set_irq(vcpu);
Gleb Natapov95ba8273132009-04-21 17:45:08 +03005534 return;
5535 }
5536
5537 /* try to inject new event if pending */
5538 if (vcpu->arch.nmi_pending) {
5539 if (kvm_x86_ops->nmi_allowed(vcpu)) {
Avi Kivity7460fb4a2011-09-20 13:43:14 +03005540 --vcpu->arch.nmi_pending;
Gleb Natapov95ba8273132009-04-21 17:45:08 +03005541 vcpu->arch.nmi_injected = true;
5542 kvm_x86_ops->set_nmi(vcpu);
5543 }
5544 } else if (kvm_cpu_has_interrupt(vcpu)) {
5545 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
Gleb Natapov66fd3f72009-05-11 13:35:50 +03005546 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
5547 false);
5548 kvm_x86_ops->set_irq(vcpu);
Gleb Natapov95ba8273132009-04-21 17:45:08 +03005549 }
5550 }
5551}
5552
Dexuan Cui2acf9232010-06-10 11:27:12 +08005553static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
5554{
5555 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
5556 !vcpu->guest_xcr0_loaded) {
5557 /* kvm_set_xcr() also depends on this */
5558 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
5559 vcpu->guest_xcr0_loaded = 1;
5560 }
5561}
5562
5563static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
5564{
5565 if (vcpu->guest_xcr0_loaded) {
5566 if (vcpu->arch.xcr0 != host_xcr0)
5567 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
5568 vcpu->guest_xcr0_loaded = 0;
5569 }
5570}
5571
Avi Kivity7460fb4a2011-09-20 13:43:14 +03005572static void process_nmi(struct kvm_vcpu *vcpu)
5573{
5574 unsigned limit = 2;
5575
5576 /*
5577 * x86 is limited to one NMI running, and one NMI pending after it.
5578 * If an NMI is already in progress, limit further NMIs to just one.
5579 * Otherwise, allow two (and we'll inject the first one immediately).
5580 */
5581 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
5582 limit = 1;
5583
5584 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
5585 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
5586 kvm_make_request(KVM_REQ_EVENT, vcpu);
5587}
5588
Marcelo Tosattid8281992012-11-27 23:29:01 -02005589static void kvm_gen_update_masterclock(struct kvm *kvm)
5590{
5591#ifdef CONFIG_X86_64
5592 int i;
5593 struct kvm_vcpu *vcpu;
5594 struct kvm_arch *ka = &kvm->arch;
5595
5596 spin_lock(&ka->pvclock_gtod_sync_lock);
5597 kvm_make_mclock_inprogress_request(kvm);
5598 /* no guest entries from this point */
5599 pvclock_update_vm_gtod_copy(kvm);
5600
5601 kvm_for_each_vcpu(i, vcpu, kvm)
5602 set_bit(KVM_REQ_CLOCK_UPDATE, &vcpu->requests);
5603
5604 /* guest entries allowed */
5605 kvm_for_each_vcpu(i, vcpu, kvm)
5606 clear_bit(KVM_REQ_MCLOCK_INPROGRESS, &vcpu->requests);
5607
5608 spin_unlock(&ka->pvclock_gtod_sync_lock);
5609#endif
5610}
5611
Avi Kivity851ba692009-08-24 11:10:17 +03005612static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005613{
5614 int r;
Gleb Natapov6a8b1d12009-05-11 13:35:51 +03005615 bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
Avi Kivity851ba692009-08-24 11:10:17 +03005616 vcpu->run->request_interrupt_window;
Nadav Har'Eld6185f22011-09-22 13:52:56 +03005617 bool req_immediate_exit = 0;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005618
Avi Kivity3e007502010-06-23 14:26:18 +03005619 if (vcpu->requests) {
Avi Kivitya8eeb042010-05-10 12:34:53 +03005620 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
Marcelo Tosatti2e53d632008-02-20 14:47:24 -05005621 kvm_mmu_unload(vcpu);
Avi Kivitya8eeb042010-05-10 12:34:53 +03005622 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
Marcelo Tosatti2f599712008-05-27 12:10:20 -03005623 __kvm_migrate_timers(vcpu);
Marcelo Tosattid8281992012-11-27 23:29:01 -02005624 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
5625 kvm_gen_update_masterclock(vcpu->kvm);
Zachary Amsden34c238a2010-09-18 14:38:14 -10005626 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
5627 r = kvm_guest_time_update(vcpu);
Zachary Amsden8cfdc002010-08-19 22:07:21 -10005628 if (unlikely(r))
5629 goto out;
5630 }
Avi Kivitya8eeb042010-05-10 12:34:53 +03005631 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
Marcelo Tosatti4731d4c2008-09-23 13:18:39 -03005632 kvm_mmu_sync_roots(vcpu);
Avi Kivitya8eeb042010-05-10 12:34:53 +03005633 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
Marcelo Tosattid4acf7e2008-06-06 16:37:35 -03005634 kvm_x86_ops->tlb_flush(vcpu);
Avi Kivitya8eeb042010-05-10 12:34:53 +03005635 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
Avi Kivity851ba692009-08-24 11:10:17 +03005636 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
Avi Kivityb93463a2007-10-25 16:52:32 +02005637 r = 0;
5638 goto out;
5639 }
Avi Kivitya8eeb042010-05-10 12:34:53 +03005640 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
Avi Kivity851ba692009-08-24 11:10:17 +03005641 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
Joerg Roedel71c4dfa2008-02-26 16:49:16 +01005642 r = 0;
5643 goto out;
5644 }
Avi Kivitya8eeb042010-05-10 12:34:53 +03005645 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
Avi Kivity02daab22009-12-30 12:40:26 +02005646 vcpu->fpu_active = 0;
5647 kvm_x86_ops->fpu_deactivate(vcpu);
5648 }
Gleb Natapovaf585b92010-10-14 11:22:46 +02005649 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
5650 /* Page is swapped out. Do synthetic halt */
5651 vcpu->arch.apf.halted = true;
5652 r = 1;
5653 goto out;
5654 }
Glauber Costac9aaa892011-07-11 15:28:14 -04005655 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
5656 record_steal_time(vcpu);
Avi Kivity7460fb4a2011-09-20 13:43:14 +03005657 if (kvm_check_request(KVM_REQ_NMI, vcpu))
5658 process_nmi(vcpu);
Nadav Har'Eld6185f22011-09-22 13:52:56 +03005659 req_immediate_exit =
5660 kvm_check_request(KVM_REQ_IMMEDIATE_EXIT, vcpu);
Gleb Natapovf5132b02011-11-10 14:57:22 +02005661 if (kvm_check_request(KVM_REQ_PMU, vcpu))
5662 kvm_handle_pmu_event(vcpu);
5663 if (kvm_check_request(KVM_REQ_PMI, vcpu))
5664 kvm_deliver_pmi(vcpu);
Avi Kivity2f52d582008-01-16 12:49:30 +02005665 }
Avi Kivityb93463a2007-10-25 16:52:32 +02005666
Avi Kivityb463a6f2010-07-20 15:06:17 +03005667 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
Avi Kivity3842d132010-07-27 12:30:24 +03005668 inject_pending_event(vcpu);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005669
Avi Kivity3842d132010-07-27 12:30:24 +03005670 /* enable NMI/IRQ window open exits if needed */
Avi Kivity7460fb4a2011-09-20 13:43:14 +03005671 if (vcpu->arch.nmi_pending)
Avi Kivity3842d132010-07-27 12:30:24 +03005672 kvm_x86_ops->enable_nmi_window(vcpu);
5673 else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
5674 kvm_x86_ops->enable_irq_window(vcpu);
Gleb Natapov6a8b1d12009-05-11 13:35:51 +03005675
Avi Kivity3842d132010-07-27 12:30:24 +03005676 if (kvm_lapic_enabled(vcpu)) {
5677 update_cr8_intercept(vcpu);
5678 kvm_lapic_sync_to_vapic(vcpu);
5679 }
Gleb Natapov95ba8273132009-04-21 17:45:08 +03005680 }
Avi Kivityb93463a2007-10-25 16:52:32 +02005681
Avi Kivityd8368af2012-05-14 18:07:56 +03005682 r = kvm_mmu_reload(vcpu);
5683 if (unlikely(r)) {
Michael S. Tsirkind905c062012-06-24 19:25:00 +03005684 goto cancel_injection;
Avi Kivityd8368af2012-05-14 18:07:56 +03005685 }
5686
Jan Kiszka42dbaa52008-12-15 13:52:10 +01005687 preempt_disable();
5688
5689 kvm_x86_ops->prepare_guest_switch(vcpu);
5690 if (vcpu->fpu_active)
5691 kvm_load_guest_fpu(vcpu);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005692 kvm_load_guest_xcr0(vcpu);
5693
Xiao Guangrong6b7e2d02011-01-12 15:40:31 +08005694 vcpu->mode = IN_GUEST_MODE;
5695
5696 /* We should set ->mode before check ->requests,
5697 * see the comment in make_all_cpus_request.
5698 */
5699 smp_mb();
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005700
5701 local_irq_disable();
5702
Xiao Guangrong6b7e2d02011-01-12 15:40:31 +08005703 if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005704 || need_resched() || signal_pending(current)) {
Xiao Guangrong6b7e2d02011-01-12 15:40:31 +08005705 vcpu->mode = OUTSIDE_GUEST_MODE;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005706 smp_wmb();
5707 local_irq_enable();
5708 preempt_enable();
5709 r = 1;
Michael S. Tsirkind905c062012-06-24 19:25:00 +03005710 goto cancel_injection;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005711 }
5712
Marcelo Tosattif656ce02009-12-23 14:35:25 -02005713 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005714
Nadav Har'Eld6185f22011-09-22 13:52:56 +03005715 if (req_immediate_exit)
5716 smp_send_reschedule(vcpu->cpu);
5717
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005718 kvm_guest_enter();
5719
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005720 if (unlikely(vcpu->arch.switch_db_regs)) {
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005721 set_debugreg(0, 7);
5722 set_debugreg(vcpu->arch.eff_db[0], 0);
5723 set_debugreg(vcpu->arch.eff_db[1], 1);
5724 set_debugreg(vcpu->arch.eff_db[2], 2);
5725 set_debugreg(vcpu->arch.eff_db[3], 3);
5726 }
5727
Marcelo Tosatti229456f2009-06-17 09:22:14 -03005728 trace_kvm_entry(vcpu->vcpu_id);
Avi Kivity851ba692009-08-24 11:10:17 +03005729 kvm_x86_ops->run(vcpu);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005730
Frederic Weisbecker24f1e32c2009-09-09 19:22:48 +02005731 /*
5732 * If the guest has used debug registers, at least dr7
5733 * will be disabled while returning to the host.
5734 * If we don't have active breakpoints in the host, we don't
5735 * care about the messed up debug address registers. But if
5736 * we have some of them active, restore the old state.
5737 */
Frederic Weisbecker59d8eb52009-11-10 11:03:12 +01005738 if (hw_breakpoint_active())
Frederic Weisbecker24f1e32c2009-09-09 19:22:48 +02005739 hw_breakpoint_restore();
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005740
Marcelo Tosatti886b4702012-11-27 23:28:58 -02005741 vcpu->arch.last_guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu,
5742 native_read_tsc());
Zachary Amsden1d5f0662010-08-19 22:07:30 -10005743
Xiao Guangrong6b7e2d02011-01-12 15:40:31 +08005744 vcpu->mode = OUTSIDE_GUEST_MODE;
Avi Kivityd94e1dc2010-05-03 16:54:48 +03005745 smp_wmb();
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005746 local_irq_enable();
5747
5748 ++vcpu->stat.exits;
5749
5750 /*
5751 * We must have an instruction between local_irq_enable() and
5752 * kvm_guest_exit(), so the timer interrupt isn't delayed by
5753 * the interrupt shadow. The stat.exits increment will do nicely.
5754 * But we need to prevent reordering, hence this barrier():
5755 */
5756 barrier();
5757
5758 kvm_guest_exit();
5759
5760 preempt_enable();
5761
Marcelo Tosattif656ce02009-12-23 14:35:25 -02005762 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
Marcelo Tosatti3200f402008-03-29 20:17:59 -03005763
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005764 /*
5765 * Profile KVM exit RIPs:
5766 */
5767 if (unlikely(prof_on == KVM_PROFILING)) {
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03005768 unsigned long rip = kvm_rip_read(vcpu);
5769 profile_hit(KVM_PROFILING, (void *)rip);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005770 }
5771
Zachary Amsdencc578282012-02-03 15:43:50 -02005772 if (unlikely(vcpu->arch.tsc_always_catchup))
5773 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
Avi Kivity298101d2007-11-25 13:41:11 +02005774
Michael S. Tsirkin5cfb1d52012-06-24 19:24:54 +03005775 if (vcpu->arch.apic_attention)
5776 kvm_lapic_sync_from_vapic(vcpu);
Avi Kivityb93463a2007-10-25 16:52:32 +02005777
Avi Kivity851ba692009-08-24 11:10:17 +03005778 r = kvm_x86_ops->handle_exit(vcpu);
Michael S. Tsirkind905c062012-06-24 19:25:00 +03005779 return r;
5780
5781cancel_injection:
5782 kvm_x86_ops->cancel_injection(vcpu);
Michael S. Tsirkinae7a2a32012-06-24 19:25:07 +03005783 if (unlikely(vcpu->arch.apic_attention))
5784 kvm_lapic_sync_from_vapic(vcpu);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005785out:
Marcelo Tosattid7690172008-09-08 15:23:48 -03005786 return r;
5787}
5788
Gleb Natapov09cec752009-03-23 15:11:44 +02005789
Avi Kivity851ba692009-08-24 11:10:17 +03005790static int __vcpu_run(struct kvm_vcpu *vcpu)
Marcelo Tosattid7690172008-09-08 15:23:48 -03005791{
5792 int r;
Marcelo Tosattif656ce02009-12-23 14:35:25 -02005793 struct kvm *kvm = vcpu->kvm;
Marcelo Tosattid7690172008-09-08 15:23:48 -03005794
5795 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
Jan Kiszka1b10bf32008-09-30 10:41:06 +02005796 pr_debug("vcpu %d received sipi with vector # %x\n",
5797 vcpu->vcpu_id, vcpu->arch.sipi_vector);
Marcelo Tosattid7690172008-09-08 15:23:48 -03005798 kvm_lapic_reset(vcpu);
Jan Kiszka8b6e4542012-09-20 07:43:08 +02005799 r = kvm_vcpu_reset(vcpu);
Marcelo Tosattid7690172008-09-08 15:23:48 -03005800 if (r)
5801 return r;
5802 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005803 }
5804
Marcelo Tosattif656ce02009-12-23 14:35:25 -02005805 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
Xiao Guangrong44841412012-09-07 14:14:20 +08005806 r = vapic_enter(vcpu);
5807 if (r) {
5808 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5809 return r;
5810 }
Marcelo Tosattid7690172008-09-08 15:23:48 -03005811
5812 r = 1;
5813 while (r > 0) {
Gleb Natapovaf585b92010-10-14 11:22:46 +02005814 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
5815 !vcpu->arch.apf.halted)
Avi Kivity851ba692009-08-24 11:10:17 +03005816 r = vcpu_enter_guest(vcpu);
Marcelo Tosattid7690172008-09-08 15:23:48 -03005817 else {
Marcelo Tosattif656ce02009-12-23 14:35:25 -02005818 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
Marcelo Tosattid7690172008-09-08 15:23:48 -03005819 kvm_vcpu_block(vcpu);
Marcelo Tosattif656ce02009-12-23 14:35:25 -02005820 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
Avi Kivitya8eeb042010-05-10 12:34:53 +03005821 if (kvm_check_request(KVM_REQ_UNHALT, vcpu))
Gleb Natapov09cec752009-03-23 15:11:44 +02005822 {
5823 switch(vcpu->arch.mp_state) {
5824 case KVM_MP_STATE_HALTED:
Marcelo Tosattid7690172008-09-08 15:23:48 -03005825 vcpu->arch.mp_state =
Gleb Natapov09cec752009-03-23 15:11:44 +02005826 KVM_MP_STATE_RUNNABLE;
5827 case KVM_MP_STATE_RUNNABLE:
Gleb Natapovaf585b92010-10-14 11:22:46 +02005828 vcpu->arch.apf.halted = false;
Gleb Natapov09cec752009-03-23 15:11:44 +02005829 break;
5830 case KVM_MP_STATE_SIPI_RECEIVED:
5831 default:
5832 r = -EINTR;
5833 break;
5834 }
5835 }
Marcelo Tosattid7690172008-09-08 15:23:48 -03005836 }
5837
Gleb Natapov09cec752009-03-23 15:11:44 +02005838 if (r <= 0)
5839 break;
5840
5841 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
5842 if (kvm_cpu_has_pending_timer(vcpu))
5843 kvm_inject_pending_timer_irqs(vcpu);
5844
Avi Kivity851ba692009-08-24 11:10:17 +03005845 if (dm_request_for_irq_injection(vcpu)) {
Gleb Natapov09cec752009-03-23 15:11:44 +02005846 r = -EINTR;
Avi Kivity851ba692009-08-24 11:10:17 +03005847 vcpu->run->exit_reason = KVM_EXIT_INTR;
Gleb Natapov09cec752009-03-23 15:11:44 +02005848 ++vcpu->stat.request_irq_exits;
5849 }
Gleb Natapovaf585b92010-10-14 11:22:46 +02005850
5851 kvm_check_async_pf_completion(vcpu);
5852
Gleb Natapov09cec752009-03-23 15:11:44 +02005853 if (signal_pending(current)) {
5854 r = -EINTR;
Avi Kivity851ba692009-08-24 11:10:17 +03005855 vcpu->run->exit_reason = KVM_EXIT_INTR;
Gleb Natapov09cec752009-03-23 15:11:44 +02005856 ++vcpu->stat.signal_exits;
5857 }
5858 if (need_resched()) {
Marcelo Tosattif656ce02009-12-23 14:35:25 -02005859 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
Gleb Natapov09cec752009-03-23 15:11:44 +02005860 kvm_resched(vcpu);
Marcelo Tosattif656ce02009-12-23 14:35:25 -02005861 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
Marcelo Tosattid7690172008-09-08 15:23:48 -03005862 }
5863 }
5864
Marcelo Tosattif656ce02009-12-23 14:35:25 -02005865 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005866
Avi Kivityb93463a2007-10-25 16:52:32 +02005867 vapic_exit(vcpu);
5868
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005869 return r;
5870}
5871
Gleb Natapov716d51a2012-09-03 15:24:26 +03005872static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
5873{
5874 int r;
5875 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5876 r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
5877 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5878 if (r != EMULATE_DONE)
5879 return 0;
5880 return 1;
5881}
5882
5883static int complete_emulated_pio(struct kvm_vcpu *vcpu)
5884{
5885 BUG_ON(!vcpu->arch.pio.count);
5886
5887 return complete_emulated_io(vcpu);
5888}
5889
Avi Kivityf78146b2012-04-18 19:22:47 +03005890/*
5891 * Implements the following, as a state machine:
5892 *
5893 * read:
5894 * for each fragment
Xiao Guangrong87da7e62012-10-24 14:07:59 +08005895 * for each mmio piece in the fragment
5896 * write gpa, len
5897 * exit
5898 * copy data
Avi Kivityf78146b2012-04-18 19:22:47 +03005899 * execute insn
5900 *
5901 * write:
5902 * for each fragment
Xiao Guangrong87da7e62012-10-24 14:07:59 +08005903 * for each mmio piece in the fragment
5904 * write gpa, len
5905 * copy data
5906 * exit
Avi Kivityf78146b2012-04-18 19:22:47 +03005907 */
Gleb Natapov716d51a2012-09-03 15:24:26 +03005908static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
Avi Kivity5287f192010-01-19 14:20:10 +02005909{
5910 struct kvm_run *run = vcpu->run;
Avi Kivityf78146b2012-04-18 19:22:47 +03005911 struct kvm_mmio_fragment *frag;
Xiao Guangrong87da7e62012-10-24 14:07:59 +08005912 unsigned len;
Avi Kivity5287f192010-01-19 14:20:10 +02005913
Gleb Natapov716d51a2012-09-03 15:24:26 +03005914 BUG_ON(!vcpu->mmio_needed);
Avi Kivity5287f192010-01-19 14:20:10 +02005915
Gleb Natapov716d51a2012-09-03 15:24:26 +03005916 /* Complete previous fragment */
Xiao Guangrong87da7e62012-10-24 14:07:59 +08005917 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
5918 len = min(8u, frag->len);
Gleb Natapov716d51a2012-09-03 15:24:26 +03005919 if (!vcpu->mmio_is_write)
Xiao Guangrong87da7e62012-10-24 14:07:59 +08005920 memcpy(frag->data, run->mmio.data, len);
5921
5922 if (frag->len <= 8) {
5923 /* Switch to the next fragment. */
5924 frag++;
5925 vcpu->mmio_cur_fragment++;
5926 } else {
5927 /* Go forward to the next mmio piece. */
5928 frag->data += len;
5929 frag->gpa += len;
5930 frag->len -= len;
5931 }
5932
Gleb Natapov716d51a2012-09-03 15:24:26 +03005933 if (vcpu->mmio_cur_fragment == vcpu->mmio_nr_fragments) {
5934 vcpu->mmio_needed = 0;
Avi Kivitycef4dea2010-01-20 12:01:20 +02005935 if (vcpu->mmio_is_write)
Gleb Natapov716d51a2012-09-03 15:24:26 +03005936 return 1;
5937 vcpu->mmio_read_completed = 1;
5938 return complete_emulated_io(vcpu);
Avi Kivity5287f192010-01-19 14:20:10 +02005939 }
Xiao Guangrong87da7e62012-10-24 14:07:59 +08005940
Gleb Natapov716d51a2012-09-03 15:24:26 +03005941 run->exit_reason = KVM_EXIT_MMIO;
5942 run->mmio.phys_addr = frag->gpa;
5943 if (vcpu->mmio_is_write)
Xiao Guangrong87da7e62012-10-24 14:07:59 +08005944 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
5945 run->mmio.len = min(8u, frag->len);
Gleb Natapov716d51a2012-09-03 15:24:26 +03005946 run->mmio.is_write = vcpu->mmio_is_write;
5947 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
5948 return 0;
Avi Kivity5287f192010-01-19 14:20:10 +02005949}
5950
Gleb Natapov716d51a2012-09-03 15:24:26 +03005951
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005952int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
5953{
5954 int r;
5955 sigset_t sigsaved;
5956
Avi Kivitye5c30142011-01-11 12:15:54 +02005957 if (!tsk_used_math(current) && init_fpu(current))
5958 return -ENOMEM;
5959
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005960 if (vcpu->sigset_active)
5961 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
5962
Avi Kivityac9f6dc2008-07-06 15:48:31 +03005963 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
5964 kvm_vcpu_block(vcpu);
Marcelo Tosattid7690172008-09-08 15:23:48 -03005965 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
Avi Kivityac9f6dc2008-07-06 15:48:31 +03005966 r = -EAGAIN;
5967 goto out;
5968 }
5969
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005970 /* re-sync apic's tpr */
Andre Przywaraeea1cff2010-12-21 11:12:00 +01005971 if (!irqchip_in_kernel(vcpu->kvm)) {
5972 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
5973 r = -EINVAL;
5974 goto out;
5975 }
5976 }
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005977
Gleb Natapov716d51a2012-09-03 15:24:26 +03005978 if (unlikely(vcpu->arch.complete_userspace_io)) {
5979 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
5980 vcpu->arch.complete_userspace_io = NULL;
5981 r = cui(vcpu);
5982 if (r <= 0)
5983 goto out;
5984 } else
5985 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
Avi Kivity5287f192010-01-19 14:20:10 +02005986
Avi Kivity851ba692009-08-24 11:10:17 +03005987 r = __vcpu_run(vcpu);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005988
5989out:
Marcelo Tosattif1d86e42010-05-03 23:04:27 -03005990 post_kvm_run_save(vcpu);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005991 if (vcpu->sigset_active)
5992 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
5993
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005994 return r;
5995}
5996
5997int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
5998{
Gleb Natapov7ae441e2011-03-31 12:06:41 +02005999 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
6000 /*
6001 * We are here if userspace calls get_regs() in the middle of
6002 * instruction emulation. Registers state needs to be copied
Guo Chao4a969982012-06-28 15:17:27 +08006003 * back from emulation context to vcpu. Userspace shouldn't do
Gleb Natapov7ae441e2011-03-31 12:06:41 +02006004 * that usually, but some bad designed PV devices (vmware
6005 * backdoor interface) need this to work
6006 */
Avi Kivitydd856ef2012-08-27 23:46:17 +03006007 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
Gleb Natapov7ae441e2011-03-31 12:06:41 +02006008 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6009 }
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03006010 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
6011 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
6012 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
6013 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
6014 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
6015 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
6016 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
6017 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006018#ifdef CONFIG_X86_64
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03006019 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
6020 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
6021 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
6022 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
6023 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
6024 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
6025 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
6026 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006027#endif
6028
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03006029 regs->rip = kvm_rip_read(vcpu);
Jan Kiszka91586a32009-10-05 13:07:21 +02006030 regs->rflags = kvm_get_rflags(vcpu);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006031
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006032 return 0;
6033}
6034
6035int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6036{
Gleb Natapov7ae441e2011-03-31 12:06:41 +02006037 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
6038 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6039
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03006040 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
6041 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
6042 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
6043 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
6044 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
6045 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
6046 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
6047 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006048#ifdef CONFIG_X86_64
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03006049 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
6050 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
6051 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
6052 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
6053 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
6054 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
6055 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
6056 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006057#endif
6058
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03006059 kvm_rip_write(vcpu, regs->rip);
Jan Kiszka91586a32009-10-05 13:07:21 +02006060 kvm_set_rflags(vcpu, regs->rflags);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006061
Jan Kiszkab4f14ab2008-04-30 17:59:04 +02006062 vcpu->arch.exception.pending = false;
6063
Avi Kivity3842d132010-07-27 12:30:24 +03006064 kvm_make_request(KVM_REQ_EVENT, vcpu);
6065
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006066 return 0;
6067}
6068
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006069void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
6070{
6071 struct kvm_segment cs;
6072
Guillaume Thouvenin3e6e0aa2008-05-27 10:18:46 +02006073 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006074 *db = cs.db;
6075 *l = cs.l;
6076}
6077EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
6078
6079int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
6080 struct kvm_sregs *sregs)
6081{
Gleb Natapov89a27f42010-02-16 10:51:48 +02006082 struct desc_ptr dt;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006083
Guillaume Thouvenin3e6e0aa2008-05-27 10:18:46 +02006084 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6085 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6086 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6087 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6088 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6089 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006090
Guillaume Thouvenin3e6e0aa2008-05-27 10:18:46 +02006091 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6092 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006093
6094 kvm_x86_ops->get_idt(vcpu, &dt);
Gleb Natapov89a27f42010-02-16 10:51:48 +02006095 sregs->idt.limit = dt.size;
6096 sregs->idt.base = dt.address;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006097 kvm_x86_ops->get_gdt(vcpu, &dt);
Gleb Natapov89a27f42010-02-16 10:51:48 +02006098 sregs->gdt.limit = dt.size;
6099 sregs->gdt.base = dt.address;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006100
Avi Kivity4d4ec082009-12-29 18:07:30 +02006101 sregs->cr0 = kvm_read_cr0(vcpu);
Zhang Xiantaoad312c72007-12-13 23:50:52 +08006102 sregs->cr2 = vcpu->arch.cr2;
Avi Kivity9f8fe502010-12-05 17:30:00 +02006103 sregs->cr3 = kvm_read_cr3(vcpu);
Avi Kivityfc78f512009-12-07 12:16:48 +02006104 sregs->cr4 = kvm_read_cr4(vcpu);
Avi Kivity2d3ad1f2008-02-24 11:20:43 +02006105 sregs->cr8 = kvm_get_cr8(vcpu);
Avi Kivityf6801df2010-01-21 15:31:50 +02006106 sregs->efer = vcpu->arch.efer;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006107 sregs->apic_base = kvm_get_apic_base(vcpu);
6108
Gleb Natapov923c61b2009-05-11 13:35:48 +03006109 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006110
Gleb Natapov36752c92009-05-11 13:35:53 +03006111 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
Gleb Natapov14d0bc12009-04-21 17:45:11 +03006112 set_bit(vcpu->arch.interrupt.nr,
6113 (unsigned long *)sregs->interrupt_bitmap);
Gleb Natapov16d7a192009-04-21 17:45:10 +03006114
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006115 return 0;
6116}
6117
Marcelo Tosatti62d9f0d2008-04-11 13:24:45 -03006118int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
6119 struct kvm_mp_state *mp_state)
6120{
Marcelo Tosatti62d9f0d2008-04-11 13:24:45 -03006121 mp_state->mp_state = vcpu->arch.mp_state;
Marcelo Tosatti62d9f0d2008-04-11 13:24:45 -03006122 return 0;
6123}
6124
6125int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
6126 struct kvm_mp_state *mp_state)
6127{
Marcelo Tosatti62d9f0d2008-04-11 13:24:45 -03006128 vcpu->arch.mp_state = mp_state->mp_state;
Avi Kivity3842d132010-07-27 12:30:24 +03006129 kvm_make_request(KVM_REQ_EVENT, vcpu);
Marcelo Tosatti62d9f0d2008-04-11 13:24:45 -03006130 return 0;
6131}
6132
Kevin Wolf7f3d35f2012-02-08 14:34:38 +01006133int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
6134 int reason, bool has_error_code, u32 error_code)
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006135{
Takuya Yoshikawa9d741912011-05-29 21:53:48 +09006136 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
Mohammed Gamal8ec47222010-08-16 00:47:01 +03006137 int ret;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006138
Mohammed Gamal8ec47222010-08-16 00:47:01 +03006139 init_emulate_ctxt(vcpu);
Izik Eidus37817f22008-03-24 23:14:53 +02006140
Kevin Wolf7f3d35f2012-02-08 14:34:38 +01006141 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
Takuya Yoshikawa9d741912011-05-29 21:53:48 +09006142 has_error_code, error_code);
Izik Eidus37817f22008-03-24 23:14:53 +02006143
Gleb Natapovc6975182010-02-18 12:15:01 +02006144 if (ret)
Gleb Natapov19d04432010-04-15 12:29:50 +03006145 return EMULATE_FAIL;
Gleb Natapovc6975182010-02-18 12:15:01 +02006146
Takuya Yoshikawa9d741912011-05-29 21:53:48 +09006147 kvm_rip_write(vcpu, ctxt->eip);
6148 kvm_set_rflags(vcpu, ctxt->eflags);
Avi Kivity3842d132010-07-27 12:30:24 +03006149 kvm_make_request(KVM_REQ_EVENT, vcpu);
Gleb Natapov19d04432010-04-15 12:29:50 +03006150 return EMULATE_DONE;
Izik Eidus37817f22008-03-24 23:14:53 +02006151}
6152EXPORT_SYMBOL_GPL(kvm_task_switch);
6153
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006154int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
6155 struct kvm_sregs *sregs)
6156{
6157 int mmu_reset_needed = 0;
Xiao Guangrong63f42e02011-01-12 15:39:18 +08006158 int pending_vec, max_bits, idx;
Gleb Natapov89a27f42010-02-16 10:51:48 +02006159 struct desc_ptr dt;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006160
Petr Matousek6d1068b2012-11-06 19:24:07 +01006161 if (!guest_cpuid_has_xsave(vcpu) && (sregs->cr4 & X86_CR4_OSXSAVE))
6162 return -EINVAL;
6163
Gleb Natapov89a27f42010-02-16 10:51:48 +02006164 dt.size = sregs->idt.limit;
6165 dt.address = sregs->idt.base;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006166 kvm_x86_ops->set_idt(vcpu, &dt);
Gleb Natapov89a27f42010-02-16 10:51:48 +02006167 dt.size = sregs->gdt.limit;
6168 dt.address = sregs->gdt.base;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006169 kvm_x86_ops->set_gdt(vcpu, &dt);
6170
Zhang Xiantaoad312c72007-12-13 23:50:52 +08006171 vcpu->arch.cr2 = sregs->cr2;
Avi Kivity9f8fe502010-12-05 17:30:00 +02006172 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
Jan Kiszkadc7e7952009-07-01 20:52:03 +02006173 vcpu->arch.cr3 = sregs->cr3;
Avi Kivityaff48ba2010-12-05 18:56:11 +02006174 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006175
Avi Kivity2d3ad1f2008-02-24 11:20:43 +02006176 kvm_set_cr8(vcpu, sregs->cr8);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006177
Avi Kivityf6801df2010-01-21 15:31:50 +02006178 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006179 kvm_x86_ops->set_efer(vcpu, sregs->efer);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006180 kvm_set_apic_base(vcpu, sregs->apic_base);
6181
Avi Kivity4d4ec082009-12-29 18:07:30 +02006182 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006183 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
Paul Knowlesd7306162008-02-06 11:02:35 +00006184 vcpu->arch.cr0 = sregs->cr0;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006185
Avi Kivityfc78f512009-12-07 12:16:48 +02006186 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006187 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
Sheng Yang3ea3aa82010-12-08 10:49:43 +08006188 if (sregs->cr4 & X86_CR4_OSXSAVE)
Avi Kivity00b27a32011-11-23 16:30:32 +02006189 kvm_update_cpuid(vcpu);
Xiao Guangrong63f42e02011-01-12 15:39:18 +08006190
6191 idx = srcu_read_lock(&vcpu->kvm->srcu);
Marcelo Tosatti7c93be442009-10-26 16:48:33 -02006192 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
Avi Kivity9f8fe502010-12-05 17:30:00 +02006193 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
Marcelo Tosatti7c93be442009-10-26 16:48:33 -02006194 mmu_reset_needed = 1;
6195 }
Xiao Guangrong63f42e02011-01-12 15:39:18 +08006196 srcu_read_unlock(&vcpu->kvm->srcu, idx);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006197
6198 if (mmu_reset_needed)
6199 kvm_mmu_reset_context(vcpu);
6200
Michael S. Tsirkina50abc32012-09-05 20:00:52 +03006201 max_bits = KVM_NR_INTERRUPTS;
Gleb Natapov923c61b2009-05-11 13:35:48 +03006202 pending_vec = find_first_bit(
6203 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
6204 if (pending_vec < max_bits) {
Gleb Natapov66fd3f72009-05-11 13:35:50 +03006205 kvm_queue_interrupt(vcpu, pending_vec, false);
Gleb Natapov923c61b2009-05-11 13:35:48 +03006206 pr_debug("Set back pending irq %d\n", pending_vec);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006207 }
6208
Guillaume Thouvenin3e6e0aa2008-05-27 10:18:46 +02006209 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6210 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6211 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6212 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6213 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6214 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006215
Guillaume Thouvenin3e6e0aa2008-05-27 10:18:46 +02006216 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6217 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006218
Mikhail Ershov5f0269f2009-08-03 14:58:25 +03006219 update_cr8_intercept(vcpu);
6220
Marcelo Tosatti9c3e4aa2008-09-10 16:40:55 -03006221 /* Older userspace won't unhalt the vcpu on reset. */
Gleb Natapovc5af89b2009-06-09 15:56:26 +03006222 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
Marcelo Tosatti9c3e4aa2008-09-10 16:40:55 -03006223 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
Avi Kivity3eeb3282010-01-21 15:31:48 +02006224 !is_protmode(vcpu))
Marcelo Tosatti9c3e4aa2008-09-10 16:40:55 -03006225 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6226
Avi Kivity3842d132010-07-27 12:30:24 +03006227 kvm_make_request(KVM_REQ_EVENT, vcpu);
6228
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006229 return 0;
6230}
6231
Jan Kiszkad0bfb942008-12-15 13:52:10 +01006232int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
6233 struct kvm_guest_debug *dbg)
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006234{
Jan Kiszka355be0b2009-10-03 00:31:21 +02006235 unsigned long rflags;
Jan Kiszkaae675ef2008-12-15 13:52:10 +01006236 int i, r;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006237
Jan Kiszka4f926bf22009-10-30 12:46:59 +01006238 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
6239 r = -EBUSY;
6240 if (vcpu->arch.exception.pending)
Avi Kivity2122ff52010-05-13 11:25:04 +03006241 goto out;
Jan Kiszka4f926bf22009-10-30 12:46:59 +01006242 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
6243 kvm_queue_exception(vcpu, DB_VECTOR);
6244 else
6245 kvm_queue_exception(vcpu, BP_VECTOR);
6246 }
6247
Jan Kiszka91586a32009-10-05 13:07:21 +02006248 /*
6249 * Read rflags as long as potentially injected trace flags are still
6250 * filtered out.
6251 */
6252 rflags = kvm_get_rflags(vcpu);
Jan Kiszka355be0b2009-10-03 00:31:21 +02006253
6254 vcpu->guest_debug = dbg->control;
6255 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
6256 vcpu->guest_debug = 0;
6257
6258 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
Jan Kiszkaae675ef2008-12-15 13:52:10 +01006259 for (i = 0; i < KVM_NR_DB_REGS; ++i)
6260 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
Jan Kiszkac8639012012-09-21 05:42:55 +02006261 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
Jan Kiszkaae675ef2008-12-15 13:52:10 +01006262 } else {
6263 for (i = 0; i < KVM_NR_DB_REGS; i++)
6264 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
Jan Kiszkaae675ef2008-12-15 13:52:10 +01006265 }
Jan Kiszkac8639012012-09-21 05:42:55 +02006266 kvm_update_dr7(vcpu);
Jan Kiszkaae675ef2008-12-15 13:52:10 +01006267
Jan Kiszkaf92653e2010-02-23 17:47:55 +01006268 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6269 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
6270 get_segment_base(vcpu, VCPU_SREG_CS);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006271
Jan Kiszka91586a32009-10-05 13:07:21 +02006272 /*
6273 * Trigger an rflags update that will inject or remove the trace
6274 * flags.
6275 */
6276 kvm_set_rflags(vcpu, rflags);
Jan Kiszkad0bfb942008-12-15 13:52:10 +01006277
Jan Kiszkac8639012012-09-21 05:42:55 +02006278 kvm_x86_ops->update_db_bp_intercept(vcpu);
Jan Kiszka355be0b2009-10-03 00:31:21 +02006279
Jan Kiszka4f926bf22009-10-30 12:46:59 +01006280 r = 0;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006281
Avi Kivity2122ff52010-05-13 11:25:04 +03006282out:
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006283
6284 return r;
6285}
6286
6287/*
Zhang Xiantao8b006792007-11-16 13:05:55 +08006288 * Translate a guest virtual address to a guest physical address.
6289 */
6290int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
6291 struct kvm_translation *tr)
6292{
6293 unsigned long vaddr = tr->linear_address;
6294 gpa_t gpa;
Marcelo Tosattif656ce02009-12-23 14:35:25 -02006295 int idx;
Zhang Xiantao8b006792007-11-16 13:05:55 +08006296
Marcelo Tosattif656ce02009-12-23 14:35:25 -02006297 idx = srcu_read_lock(&vcpu->kvm->srcu);
Gleb Natapov1871c602010-02-10 14:21:32 +02006298 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
Marcelo Tosattif656ce02009-12-23 14:35:25 -02006299 srcu_read_unlock(&vcpu->kvm->srcu, idx);
Zhang Xiantao8b006792007-11-16 13:05:55 +08006300 tr->physical_address = gpa;
6301 tr->valid = gpa != UNMAPPED_GVA;
6302 tr->writeable = 1;
6303 tr->usermode = 0;
Zhang Xiantao8b006792007-11-16 13:05:55 +08006304
6305 return 0;
6306}
6307
Hollis Blanchardd0752062007-10-31 17:24:25 -05006308int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6309{
Sheng Yang98918832010-05-17 17:08:28 +08006310 struct i387_fxsave_struct *fxsave =
6311 &vcpu->arch.guest_fpu.state->fxsave;
Hollis Blanchardd0752062007-10-31 17:24:25 -05006312
Hollis Blanchardd0752062007-10-31 17:24:25 -05006313 memcpy(fpu->fpr, fxsave->st_space, 128);
6314 fpu->fcw = fxsave->cwd;
6315 fpu->fsw = fxsave->swd;
6316 fpu->ftwx = fxsave->twd;
6317 fpu->last_opcode = fxsave->fop;
6318 fpu->last_ip = fxsave->rip;
6319 fpu->last_dp = fxsave->rdp;
6320 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
6321
Hollis Blanchardd0752062007-10-31 17:24:25 -05006322 return 0;
6323}
6324
6325int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6326{
Sheng Yang98918832010-05-17 17:08:28 +08006327 struct i387_fxsave_struct *fxsave =
6328 &vcpu->arch.guest_fpu.state->fxsave;
Hollis Blanchardd0752062007-10-31 17:24:25 -05006329
Hollis Blanchardd0752062007-10-31 17:24:25 -05006330 memcpy(fxsave->st_space, fpu->fpr, 128);
6331 fxsave->cwd = fpu->fcw;
6332 fxsave->swd = fpu->fsw;
6333 fxsave->twd = fpu->ftwx;
6334 fxsave->fop = fpu->last_opcode;
6335 fxsave->rip = fpu->last_ip;
6336 fxsave->rdp = fpu->last_dp;
6337 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
6338
Hollis Blanchardd0752062007-10-31 17:24:25 -05006339 return 0;
6340}
6341
Jan Kiszka10ab25c2010-05-25 16:01:50 +02006342int fx_init(struct kvm_vcpu *vcpu)
Hollis Blanchardd0752062007-10-31 17:24:25 -05006343{
Jan Kiszka10ab25c2010-05-25 16:01:50 +02006344 int err;
6345
6346 err = fpu_alloc(&vcpu->arch.guest_fpu);
6347 if (err)
6348 return err;
6349
Sheng Yang98918832010-05-17 17:08:28 +08006350 fpu_finit(&vcpu->arch.guest_fpu);
Hollis Blanchardd0752062007-10-31 17:24:25 -05006351
Dexuan Cui2acf9232010-06-10 11:27:12 +08006352 /*
6353 * Ensure guest xcr0 is valid for loading
6354 */
6355 vcpu->arch.xcr0 = XSTATE_FP;
6356
Zhang Xiantaoad312c72007-12-13 23:50:52 +08006357 vcpu->arch.cr0 |= X86_CR0_ET;
Jan Kiszka10ab25c2010-05-25 16:01:50 +02006358
6359 return 0;
Hollis Blanchardd0752062007-10-31 17:24:25 -05006360}
6361EXPORT_SYMBOL_GPL(fx_init);
6362
Sheng Yang98918832010-05-17 17:08:28 +08006363static void fx_free(struct kvm_vcpu *vcpu)
6364{
6365 fpu_free(&vcpu->arch.guest_fpu);
6366}
6367
Hollis Blanchardd0752062007-10-31 17:24:25 -05006368void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
6369{
Avi Kivity2608d7a2010-01-21 15:31:45 +02006370 if (vcpu->guest_fpu_loaded)
Hollis Blanchardd0752062007-10-31 17:24:25 -05006371 return;
6372
Dexuan Cui2acf9232010-06-10 11:27:12 +08006373 /*
6374 * Restore all possible states in the guest,
6375 * and assume host would use all available bits.
6376 * Guest xcr0 would be loaded later.
6377 */
6378 kvm_put_guest_xcr0(vcpu);
Hollis Blanchardd0752062007-10-31 17:24:25 -05006379 vcpu->guest_fpu_loaded = 1;
Suresh Siddhab1a74bf2012-09-20 11:01:49 -07006380 __kernel_fpu_begin();
Sheng Yang98918832010-05-17 17:08:28 +08006381 fpu_restore_checking(&vcpu->arch.guest_fpu);
Avi Kivity0c048512010-01-21 15:31:52 +02006382 trace_kvm_fpu(1);
Hollis Blanchardd0752062007-10-31 17:24:25 -05006383}
Hollis Blanchardd0752062007-10-31 17:24:25 -05006384
6385void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
6386{
Dexuan Cui2acf9232010-06-10 11:27:12 +08006387 kvm_put_guest_xcr0(vcpu);
6388
Hollis Blanchardd0752062007-10-31 17:24:25 -05006389 if (!vcpu->guest_fpu_loaded)
6390 return;
6391
6392 vcpu->guest_fpu_loaded = 0;
Sheng Yang98918832010-05-17 17:08:28 +08006393 fpu_save_init(&vcpu->arch.guest_fpu);
Suresh Siddhab1a74bf2012-09-20 11:01:49 -07006394 __kernel_fpu_end();
Avi Kivityf096ed82007-11-18 13:54:33 +02006395 ++vcpu->stat.fpu_reload;
Avi Kivitya8eeb042010-05-10 12:34:53 +03006396 kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
Avi Kivity0c048512010-01-21 15:31:52 +02006397 trace_kvm_fpu(0);
Hollis Blanchardd0752062007-10-31 17:24:25 -05006398}
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08006399
6400void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
6401{
Glauber Costa12f9a482011-02-01 14:16:40 -05006402 kvmclock_reset(vcpu);
Joerg Roedel7f1ea202009-02-25 16:08:31 +01006403
Sheng Yangf5f48ee2010-06-30 12:25:15 +08006404 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
Sheng Yang98918832010-05-17 17:08:28 +08006405 fx_free(vcpu);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08006406 kvm_x86_ops->vcpu_free(vcpu);
6407}
6408
6409struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
6410 unsigned int id)
6411{
Zachary Amsden6755bae2010-08-19 22:07:22 -10006412 if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
6413 printk_once(KERN_WARNING
6414 "kvm: SMP vm created on host with unstable TSC; "
6415 "guest TSC will not be reliable\n");
Avi Kivity26e52152007-11-20 15:30:24 +02006416 return kvm_x86_ops->vcpu_create(kvm, id);
6417}
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08006418
Avi Kivity26e52152007-11-20 15:30:24 +02006419int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
6420{
6421 int r;
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08006422
Sheng Yang0bed3b52008-10-09 16:01:54 +08006423 vcpu->arch.mtrr_state.have_fixed = 1;
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +03006424 r = vcpu_load(vcpu);
6425 if (r)
6426 return r;
Jan Kiszka8b6e4542012-09-20 07:43:08 +02006427 r = kvm_vcpu_reset(vcpu);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08006428 if (r == 0)
6429 r = kvm_mmu_setup(vcpu);
6430 vcpu_put(vcpu);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08006431
Avi Kivity26e52152007-11-20 15:30:24 +02006432 return r;
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08006433}
6434
Marcelo Tosatti42897d82012-11-27 23:29:02 -02006435int kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
6436{
6437 int r;
Will Auld8fe8ab42012-11-29 12:42:12 -08006438 struct msr_data msr;
Marcelo Tosatti42897d82012-11-27 23:29:02 -02006439
6440 r = vcpu_load(vcpu);
6441 if (r)
6442 return r;
Will Auld8fe8ab42012-11-29 12:42:12 -08006443 msr.data = 0x0;
6444 msr.index = MSR_IA32_TSC;
6445 msr.host_initiated = true;
6446 kvm_write_tsc(vcpu, &msr);
Marcelo Tosatti42897d82012-11-27 23:29:02 -02006447 vcpu_put(vcpu);
6448
6449 return r;
6450}
6451
Hollis Blanchardd40ccc62007-11-19 14:04:43 -06006452void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08006453{
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +03006454 int r;
Gleb Natapov344d9582010-10-14 11:22:50 +02006455 vcpu->arch.apf.msr_val = 0;
6456
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +03006457 r = vcpu_load(vcpu);
6458 BUG_ON(r);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08006459 kvm_mmu_unload(vcpu);
6460 vcpu_put(vcpu);
6461
Sheng Yang98918832010-05-17 17:08:28 +08006462 fx_free(vcpu);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08006463 kvm_x86_ops->vcpu_free(vcpu);
6464}
6465
Jan Kiszka8b6e4542012-09-20 07:43:08 +02006466static int kvm_vcpu_reset(struct kvm_vcpu *vcpu)
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08006467{
Avi Kivity7460fb4a2011-09-20 13:43:14 +03006468 atomic_set(&vcpu->arch.nmi_queued, 0);
6469 vcpu->arch.nmi_pending = 0;
Jan Kiszka448fa4a2008-09-26 09:30:48 +02006470 vcpu->arch.nmi_injected = false;
6471
Jan Kiszka42dbaa52008-12-15 13:52:10 +01006472 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
6473 vcpu->arch.dr6 = DR6_FIXED_1;
6474 vcpu->arch.dr7 = DR7_FIXED_1;
Jan Kiszkac8639012012-09-21 05:42:55 +02006475 kvm_update_dr7(vcpu);
Jan Kiszka42dbaa52008-12-15 13:52:10 +01006476
Avi Kivity3842d132010-07-27 12:30:24 +03006477 kvm_make_request(KVM_REQ_EVENT, vcpu);
Gleb Natapov344d9582010-10-14 11:22:50 +02006478 vcpu->arch.apf.msr_val = 0;
Glauber Costac9aaa892011-07-11 15:28:14 -04006479 vcpu->arch.st.msr_val = 0;
Avi Kivity3842d132010-07-27 12:30:24 +03006480
Glauber Costa12f9a482011-02-01 14:16:40 -05006481 kvmclock_reset(vcpu);
6482
Gleb Natapovaf585b92010-10-14 11:22:46 +02006483 kvm_clear_async_pf_completion_queue(vcpu);
6484 kvm_async_pf_hash_reset(vcpu);
6485 vcpu->arch.apf.halted = false;
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08006486
Gleb Natapovf5132b02011-11-10 14:57:22 +02006487 kvm_pmu_reset(vcpu);
6488
Julian Stecklina66f7b722012-12-05 15:26:19 +01006489 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
6490 vcpu->arch.regs_avail = ~0;
6491 vcpu->arch.regs_dirty = ~0;
6492
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08006493 return kvm_x86_ops->vcpu_reset(vcpu);
6494}
6495
Alexander Graf10474ae2009-09-15 11:37:46 +02006496int kvm_arch_hardware_enable(void *garbage)
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08006497{
Zachary Amsdenca84d1a2010-08-19 22:07:28 -10006498 struct kvm *kvm;
6499 struct kvm_vcpu *vcpu;
6500 int i;
Zachary Amsden0dd6a6e2012-02-03 15:43:56 -02006501 int ret;
6502 u64 local_tsc;
6503 u64 max_tsc = 0;
6504 bool stable, backwards_tsc = false;
Avi Kivity18863bd2009-09-07 11:12:18 +03006505
6506 kvm_shared_msr_cpu_online();
Zachary Amsden0dd6a6e2012-02-03 15:43:56 -02006507 ret = kvm_x86_ops->hardware_enable(garbage);
6508 if (ret != 0)
6509 return ret;
6510
6511 local_tsc = native_read_tsc();
6512 stable = !check_tsc_unstable();
6513 list_for_each_entry(kvm, &vm_list, vm_list) {
6514 kvm_for_each_vcpu(i, vcpu, kvm) {
6515 if (!stable && vcpu->cpu == smp_processor_id())
6516 set_bit(KVM_REQ_CLOCK_UPDATE, &vcpu->requests);
6517 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
6518 backwards_tsc = true;
6519 if (vcpu->arch.last_host_tsc > max_tsc)
6520 max_tsc = vcpu->arch.last_host_tsc;
6521 }
6522 }
6523 }
6524
6525 /*
6526 * Sometimes, even reliable TSCs go backwards. This happens on
6527 * platforms that reset TSC during suspend or hibernate actions, but
6528 * maintain synchronization. We must compensate. Fortunately, we can
6529 * detect that condition here, which happens early in CPU bringup,
6530 * before any KVM threads can be running. Unfortunately, we can't
6531 * bring the TSCs fully up to date with real time, as we aren't yet far
6532 * enough into CPU bringup that we know how much real time has actually
6533 * elapsed; our helper function, get_kernel_ns() will be using boot
6534 * variables that haven't been updated yet.
6535 *
6536 * So we simply find the maximum observed TSC above, then record the
6537 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
6538 * the adjustment will be applied. Note that we accumulate
6539 * adjustments, in case multiple suspend cycles happen before some VCPU
6540 * gets a chance to run again. In the event that no KVM threads get a
6541 * chance to run, we will miss the entire elapsed period, as we'll have
6542 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
6543 * loose cycle time. This isn't too big a deal, since the loss will be
6544 * uniform across all VCPUs (not to mention the scenario is extremely
6545 * unlikely). It is possible that a second hibernate recovery happens
6546 * much faster than a first, causing the observed TSC here to be
6547 * smaller; this would require additional padding adjustment, which is
6548 * why we set last_host_tsc to the local tsc observed here.
6549 *
6550 * N.B. - this code below runs only on platforms with reliable TSC,
6551 * as that is the only way backwards_tsc is set above. Also note
6552 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
6553 * have the same delta_cyc adjustment applied if backwards_tsc
6554 * is detected. Note further, this adjustment is only done once,
6555 * as we reset last_host_tsc on all VCPUs to stop this from being
6556 * called multiple times (one for each physical CPU bringup).
6557 *
Guo Chao4a969982012-06-28 15:17:27 +08006558 * Platforms with unreliable TSCs don't have to deal with this, they
Zachary Amsden0dd6a6e2012-02-03 15:43:56 -02006559 * will be compensated by the logic in vcpu_load, which sets the TSC to
6560 * catchup mode. This will catchup all VCPUs to real time, but cannot
6561 * guarantee that they stay in perfect synchronization.
6562 */
6563 if (backwards_tsc) {
6564 u64 delta_cyc = max_tsc - local_tsc;
6565 list_for_each_entry(kvm, &vm_list, vm_list) {
6566 kvm_for_each_vcpu(i, vcpu, kvm) {
6567 vcpu->arch.tsc_offset_adjustment += delta_cyc;
6568 vcpu->arch.last_host_tsc = local_tsc;
Marcelo Tosattid8281992012-11-27 23:29:01 -02006569 set_bit(KVM_REQ_MASTERCLOCK_UPDATE,
6570 &vcpu->requests);
Zachary Amsden0dd6a6e2012-02-03 15:43:56 -02006571 }
6572
6573 /*
6574 * We have to disable TSC offset matching.. if you were
6575 * booting a VM while issuing an S4 host suspend....
6576 * you may have some problem. Solving this issue is
6577 * left as an exercise to the reader.
6578 */
6579 kvm->arch.last_tsc_nsec = 0;
6580 kvm->arch.last_tsc_write = 0;
6581 }
6582
6583 }
6584 return 0;
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08006585}
6586
6587void kvm_arch_hardware_disable(void *garbage)
6588{
6589 kvm_x86_ops->hardware_disable(garbage);
Avi Kivity3548bab2009-11-28 14:18:47 +02006590 drop_user_return_notifiers(garbage);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08006591}
6592
6593int kvm_arch_hardware_setup(void)
6594{
6595 return kvm_x86_ops->hardware_setup();
6596}
6597
6598void kvm_arch_hardware_unsetup(void)
6599{
6600 kvm_x86_ops->hardware_unsetup();
6601}
6602
6603void kvm_arch_check_processor_compat(void *rtn)
6604{
6605 kvm_x86_ops->check_processor_compatibility(rtn);
6606}
6607
Avi Kivity3e515702012-03-05 14:23:29 +02006608bool kvm_vcpu_compatible(struct kvm_vcpu *vcpu)
6609{
6610 return irqchip_in_kernel(vcpu->kvm) == (vcpu->arch.apic != NULL);
6611}
6612
Gleb Natapov54e98182012-08-05 15:58:32 +03006613struct static_key kvm_no_apic_vcpu __read_mostly;
6614
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08006615int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
6616{
6617 struct page *page;
6618 struct kvm *kvm;
6619 int r;
6620
6621 BUG_ON(vcpu->kvm == NULL);
6622 kvm = vcpu->kvm;
6623
Avi Kivity9aabc88f2010-07-29 15:11:50 +03006624 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
Gleb Natapovc5af89b2009-06-09 15:56:26 +03006625 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
Avi Kivitya4535292008-04-13 17:54:35 +03006626 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08006627 else
Avi Kivitya4535292008-04-13 17:54:35 +03006628 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08006629
6630 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
6631 if (!page) {
6632 r = -ENOMEM;
6633 goto fail;
6634 }
Zhang Xiantaoad312c72007-12-13 23:50:52 +08006635 vcpu->arch.pio_data = page_address(page);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08006636
Zachary Amsdencc578282012-02-03 15:43:50 -02006637 kvm_set_tsc_khz(vcpu, max_tsc_khz);
Zachary Amsdenc2855452010-09-18 14:38:15 -10006638
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08006639 r = kvm_mmu_create(vcpu);
6640 if (r < 0)
6641 goto fail_free_pio_data;
6642
6643 if (irqchip_in_kernel(kvm)) {
6644 r = kvm_create_lapic(vcpu);
6645 if (r < 0)
6646 goto fail_mmu_destroy;
Gleb Natapov54e98182012-08-05 15:58:32 +03006647 } else
6648 static_key_slow_inc(&kvm_no_apic_vcpu);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08006649
Huang Ying890ca9a2009-05-11 16:48:15 +08006650 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
6651 GFP_KERNEL);
6652 if (!vcpu->arch.mce_banks) {
6653 r = -ENOMEM;
Wei Yongjun443c39b2010-01-22 14:21:29 +08006654 goto fail_free_lapic;
Huang Ying890ca9a2009-05-11 16:48:15 +08006655 }
6656 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
6657
Sheng Yangf5f48ee2010-06-30 12:25:15 +08006658 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL))
6659 goto fail_free_mce_banks;
6660
Julian Stecklina66f7b722012-12-05 15:26:19 +01006661 r = fx_init(vcpu);
6662 if (r)
6663 goto fail_free_wbinvd_dirty_mask;
6664
Will Auldba904632012-11-29 12:42:50 -08006665 vcpu->arch.ia32_tsc_adjust_msr = 0x0;
Gleb Natapovaf585b92010-10-14 11:22:46 +02006666 kvm_async_pf_hash_reset(vcpu);
Gleb Natapovf5132b02011-11-10 14:57:22 +02006667 kvm_pmu_init(vcpu);
Gleb Natapovaf585b92010-10-14 11:22:46 +02006668
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08006669 return 0;
Julian Stecklina66f7b722012-12-05 15:26:19 +01006670fail_free_wbinvd_dirty_mask:
6671 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
Sheng Yangf5f48ee2010-06-30 12:25:15 +08006672fail_free_mce_banks:
6673 kfree(vcpu->arch.mce_banks);
Wei Yongjun443c39b2010-01-22 14:21:29 +08006674fail_free_lapic:
6675 kvm_free_lapic(vcpu);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08006676fail_mmu_destroy:
6677 kvm_mmu_destroy(vcpu);
6678fail_free_pio_data:
Zhang Xiantaoad312c72007-12-13 23:50:52 +08006679 free_page((unsigned long)vcpu->arch.pio_data);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08006680fail:
6681 return r;
6682}
6683
6684void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
6685{
Marcelo Tosattif656ce02009-12-23 14:35:25 -02006686 int idx;
6687
Gleb Natapovf5132b02011-11-10 14:57:22 +02006688 kvm_pmu_destroy(vcpu);
Wei Yongjun36cb93f2010-01-22 14:18:47 +08006689 kfree(vcpu->arch.mce_banks);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08006690 kvm_free_lapic(vcpu);
Marcelo Tosattif656ce02009-12-23 14:35:25 -02006691 idx = srcu_read_lock(&vcpu->kvm->srcu);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08006692 kvm_mmu_destroy(vcpu);
Marcelo Tosattif656ce02009-12-23 14:35:25 -02006693 srcu_read_unlock(&vcpu->kvm->srcu, idx);
Zhang Xiantaoad312c72007-12-13 23:50:52 +08006694 free_page((unsigned long)vcpu->arch.pio_data);
Gleb Natapov54e98182012-08-05 15:58:32 +03006695 if (!irqchip_in_kernel(vcpu->kvm))
6696 static_key_slow_dec(&kvm_no_apic_vcpu);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08006697}
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +08006698
Carsten Ottee08b9632012-01-04 10:25:20 +01006699int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +08006700{
Carsten Ottee08b9632012-01-04 10:25:20 +01006701 if (type)
6702 return -EINVAL;
6703
Zhang Xiantaof05e70a2007-12-14 10:01:48 +08006704 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
Ben-Ami Yassour4d5c5d02008-07-28 19:26:26 +03006705 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +08006706
Sheng Yang5550af42008-10-15 20:15:06 +08006707 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
6708 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
Alex Williamson7a844282012-09-21 11:58:03 -06006709 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
6710 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
6711 &kvm->arch.irq_sources_bitmap);
Sheng Yang5550af42008-10-15 20:15:06 +08006712
Jan Kiszka038f8c12011-02-04 10:49:11 +01006713 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
Gleb Natapov1e08ec42012-09-13 17:19:24 +03006714 mutex_init(&kvm->arch.apic_map_lock);
Marcelo Tosattid8281992012-11-27 23:29:01 -02006715 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
6716
6717 pvclock_update_vm_gtod_copy(kvm);
Marcelo Tosatti53f658b32008-12-11 20:45:05 +01006718
Jan Kiszkad89f5ef2010-11-09 17:02:49 +01006719 return 0;
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +08006720}
6721
6722static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
6723{
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +03006724 int r;
6725 r = vcpu_load(vcpu);
6726 BUG_ON(r);
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +08006727 kvm_mmu_unload(vcpu);
6728 vcpu_put(vcpu);
6729}
6730
6731static void kvm_free_vcpus(struct kvm *kvm)
6732{
6733 unsigned int i;
Gleb Natapov988a2ca2009-06-09 15:56:29 +03006734 struct kvm_vcpu *vcpu;
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +08006735
6736 /*
6737 * Unpin any mmu pages first.
6738 */
Gleb Natapovaf585b92010-10-14 11:22:46 +02006739 kvm_for_each_vcpu(i, vcpu, kvm) {
6740 kvm_clear_async_pf_completion_queue(vcpu);
Gleb Natapov988a2ca2009-06-09 15:56:29 +03006741 kvm_unload_vcpu_mmu(vcpu);
Gleb Natapovaf585b92010-10-14 11:22:46 +02006742 }
Gleb Natapov988a2ca2009-06-09 15:56:29 +03006743 kvm_for_each_vcpu(i, vcpu, kvm)
6744 kvm_arch_vcpu_free(vcpu);
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +08006745
Gleb Natapov988a2ca2009-06-09 15:56:29 +03006746 mutex_lock(&kvm->lock);
6747 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
6748 kvm->vcpus[i] = NULL;
6749
6750 atomic_set(&kvm->online_vcpus, 0);
6751 mutex_unlock(&kvm->lock);
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +08006752}
6753
Sheng Yangad8ba2c2009-01-06 10:03:02 +08006754void kvm_arch_sync_events(struct kvm *kvm)
6755{
Sheng Yangba4cef32009-01-06 10:03:03 +08006756 kvm_free_all_assigned_devices(kvm);
Xiao Guangrongaea924f2010-07-10 17:37:56 +08006757 kvm_free_pit(kvm);
Sheng Yangad8ba2c2009-01-06 10:03:02 +08006758}
6759
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +08006760void kvm_arch_destroy_vm(struct kvm *kvm)
6761{
Sheng Yang6eb55812008-10-31 12:37:41 +08006762 kvm_iommu_unmap_guest(kvm);
Zhang Xiantaod7deeeb02007-12-14 10:17:34 +08006763 kfree(kvm->arch.vpic);
6764 kfree(kvm->arch.vioapic);
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +08006765 kvm_free_vcpus(kvm);
Avi Kivity3d458302008-03-25 11:26:13 +02006766 if (kvm->arch.apic_access_page)
6767 put_page(kvm->arch.apic_access_page);
Sheng Yangb7ebfb02008-04-25 21:44:52 +08006768 if (kvm->arch.ept_identity_pagetable)
6769 put_page(kvm->arch.ept_identity_pagetable);
Gleb Natapov1e08ec42012-09-13 17:19:24 +03006770 kfree(rcu_dereference_check(kvm->arch.apic_map, 1));
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +08006771}
Zhang Xiantao0de10342007-11-20 16:25:04 +08006772
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +09006773void kvm_arch_free_memslot(struct kvm_memory_slot *free,
6774 struct kvm_memory_slot *dont)
6775{
6776 int i;
6777
Takuya Yoshikawad89cc612012-08-01 18:03:28 +09006778 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
6779 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
6780 kvm_kvfree(free->arch.rmap[i]);
6781 free->arch.rmap[i] = NULL;
Takuya Yoshikawa77d11302012-07-02 17:57:17 +09006782 }
Takuya Yoshikawad89cc612012-08-01 18:03:28 +09006783 if (i == 0)
6784 continue;
6785
6786 if (!dont || free->arch.lpage_info[i - 1] !=
6787 dont->arch.lpage_info[i - 1]) {
6788 kvm_kvfree(free->arch.lpage_info[i - 1]);
6789 free->arch.lpage_info[i - 1] = NULL;
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +09006790 }
6791 }
6792}
6793
6794int kvm_arch_create_memslot(struct kvm_memory_slot *slot, unsigned long npages)
6795{
6796 int i;
6797
Takuya Yoshikawad89cc612012-08-01 18:03:28 +09006798 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +09006799 unsigned long ugfn;
6800 int lpages;
Takuya Yoshikawad89cc612012-08-01 18:03:28 +09006801 int level = i + 1;
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +09006802
6803 lpages = gfn_to_index(slot->base_gfn + npages - 1,
6804 slot->base_gfn, level) + 1;
6805
Takuya Yoshikawad89cc612012-08-01 18:03:28 +09006806 slot->arch.rmap[i] =
6807 kvm_kvzalloc(lpages * sizeof(*slot->arch.rmap[i]));
6808 if (!slot->arch.rmap[i])
Takuya Yoshikawa77d11302012-07-02 17:57:17 +09006809 goto out_free;
Takuya Yoshikawad89cc612012-08-01 18:03:28 +09006810 if (i == 0)
6811 continue;
Takuya Yoshikawa77d11302012-07-02 17:57:17 +09006812
Takuya Yoshikawad89cc612012-08-01 18:03:28 +09006813 slot->arch.lpage_info[i - 1] = kvm_kvzalloc(lpages *
6814 sizeof(*slot->arch.lpage_info[i - 1]));
6815 if (!slot->arch.lpage_info[i - 1])
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +09006816 goto out_free;
6817
6818 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
Takuya Yoshikawad89cc612012-08-01 18:03:28 +09006819 slot->arch.lpage_info[i - 1][0].write_count = 1;
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +09006820 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
Takuya Yoshikawad89cc612012-08-01 18:03:28 +09006821 slot->arch.lpage_info[i - 1][lpages - 1].write_count = 1;
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +09006822 ugfn = slot->userspace_addr >> PAGE_SHIFT;
6823 /*
6824 * If the gfn and userspace address are not aligned wrt each
6825 * other, or if explicitly asked to, disable large page
6826 * support for this slot
6827 */
6828 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
6829 !kvm_largepages_enabled()) {
6830 unsigned long j;
6831
6832 for (j = 0; j < lpages; ++j)
Takuya Yoshikawad89cc612012-08-01 18:03:28 +09006833 slot->arch.lpage_info[i - 1][j].write_count = 1;
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +09006834 }
6835 }
6836
6837 return 0;
6838
6839out_free:
Takuya Yoshikawad89cc612012-08-01 18:03:28 +09006840 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
6841 kvm_kvfree(slot->arch.rmap[i]);
6842 slot->arch.rmap[i] = NULL;
6843 if (i == 0)
6844 continue;
6845
6846 kvm_kvfree(slot->arch.lpage_info[i - 1]);
6847 slot->arch.lpage_info[i - 1] = NULL;
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +09006848 }
6849 return -ENOMEM;
6850}
6851
Marcelo Tosattif7784b82009-12-23 14:35:18 -02006852int kvm_arch_prepare_memory_region(struct kvm *kvm,
6853 struct kvm_memory_slot *memslot,
Zhang Xiantao0de10342007-11-20 16:25:04 +08006854 struct kvm_memory_slot old,
Marcelo Tosattif7784b82009-12-23 14:35:18 -02006855 struct kvm_userspace_memory_region *mem,
Zhang Xiantao0de10342007-11-20 16:25:04 +08006856 int user_alloc)
6857{
Marcelo Tosattif7784b82009-12-23 14:35:18 -02006858 int npages = memslot->npages;
Avi Kivity7ac77092010-06-21 10:57:45 +03006859 int map_flags = MAP_PRIVATE | MAP_ANONYMOUS;
6860
6861 /* Prevent internal slot pages from being moved by fork()/COW. */
6862 if (memslot->id >= KVM_MEMORY_SLOTS)
6863 map_flags = MAP_SHARED | MAP_ANONYMOUS;
Zhang Xiantao0de10342007-11-20 16:25:04 +08006864
6865 /*To keep backward compatibility with older userspace,
Guo Chao4a969982012-06-28 15:17:27 +08006866 *x86 needs to handle !user_alloc case.
Zhang Xiantao0de10342007-11-20 16:25:04 +08006867 */
6868 if (!user_alloc) {
Takuya Yoshikawaaab2eb72012-08-01 18:01:10 +09006869 if (npages && !old.npages) {
Andrea Arcangeli604b38a2008-07-25 16:32:03 +02006870 unsigned long userspace_addr;
6871
Linus Torvalds6be5ceb2012-04-20 17:13:58 -07006872 userspace_addr = vm_mmap(NULL, 0,
Andrea Arcangeli604b38a2008-07-25 16:32:03 +02006873 npages * PAGE_SIZE,
6874 PROT_READ | PROT_WRITE,
Avi Kivity7ac77092010-06-21 10:57:45 +03006875 map_flags,
Andrea Arcangeli604b38a2008-07-25 16:32:03 +02006876 0);
Zhang Xiantao0de10342007-11-20 16:25:04 +08006877
Andrea Arcangeli604b38a2008-07-25 16:32:03 +02006878 if (IS_ERR((void *)userspace_addr))
6879 return PTR_ERR((void *)userspace_addr);
6880
Andrea Arcangeli604b38a2008-07-25 16:32:03 +02006881 memslot->userspace_addr = userspace_addr;
Zhang Xiantao0de10342007-11-20 16:25:04 +08006882 }
6883 }
6884
Marcelo Tosattif7784b82009-12-23 14:35:18 -02006885
6886 return 0;
6887}
6888
6889void kvm_arch_commit_memory_region(struct kvm *kvm,
6890 struct kvm_userspace_memory_region *mem,
6891 struct kvm_memory_slot old,
6892 int user_alloc)
6893{
6894
Xiao Guangrong48c0e4e2011-03-04 18:59:21 +08006895 int nr_mmu_pages = 0, npages = mem->memory_size >> PAGE_SHIFT;
Marcelo Tosattif7784b82009-12-23 14:35:18 -02006896
Takuya Yoshikawaaab2eb72012-08-01 18:01:10 +09006897 if (!user_alloc && !old.user_alloc && old.npages && !npages) {
Marcelo Tosattif7784b82009-12-23 14:35:18 -02006898 int ret;
6899
Al Virobfce2812012-04-20 21:57:04 -04006900 ret = vm_munmap(old.userspace_addr,
Marcelo Tosattif7784b82009-12-23 14:35:18 -02006901 old.npages * PAGE_SIZE);
Marcelo Tosattif7784b82009-12-23 14:35:18 -02006902 if (ret < 0)
6903 printk(KERN_WARNING
6904 "kvm_vm_ioctl_set_memory_region: "
6905 "failed to munmap memory\n");
6906 }
6907
Xiao Guangrong48c0e4e2011-03-04 18:59:21 +08006908 if (!kvm->arch.n_requested_mmu_pages)
6909 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
Zhang Xiantao0de10342007-11-20 16:25:04 +08006910
Xiao Guangrong48c0e4e2011-03-04 18:59:21 +08006911 spin_lock(&kvm->mmu_lock);
6912 if (nr_mmu_pages)
6913 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
Zhang Xiantao0de10342007-11-20 16:25:04 +08006914 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
Marcelo Tosatti7c8a83b2009-05-12 18:55:43 -03006915 spin_unlock(&kvm->mmu_lock);
Marcelo Tosatti3b4dc3a2012-08-28 17:43:26 -03006916 /*
6917 * If memory slot is created, or moved, we need to clear all
6918 * mmio sptes.
6919 */
6920 if (npages && old.base_gfn != mem->guest_phys_addr >> PAGE_SHIFT) {
6921 kvm_mmu_zap_all(kvm);
6922 kvm_reload_remote_mmus(kvm);
6923 }
Zhang Xiantao0de10342007-11-20 16:25:04 +08006924}
Zhang Xiantao1d737c82007-12-14 09:35:10 +08006925
Marcelo Tosatti2df72e92012-08-24 15:54:57 -03006926void kvm_arch_flush_shadow_all(struct kvm *kvm)
Marcelo Tosatti34d4cb82008-07-10 20:49:31 -03006927{
6928 kvm_mmu_zap_all(kvm);
Marcelo Tosatti8986ecc2009-05-12 18:55:45 -03006929 kvm_reload_remote_mmus(kvm);
Marcelo Tosatti34d4cb82008-07-10 20:49:31 -03006930}
6931
Marcelo Tosatti2df72e92012-08-24 15:54:57 -03006932void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
6933 struct kvm_memory_slot *slot)
6934{
6935 kvm_arch_flush_shadow_all(kvm);
6936}
6937
Zhang Xiantao1d737c82007-12-14 09:35:10 +08006938int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
6939{
Gleb Natapovaf585b92010-10-14 11:22:46 +02006940 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6941 !vcpu->arch.apf.halted)
6942 || !list_empty_careful(&vcpu->async_pf.done)
Gleb Natapova1b37102009-07-09 15:33:52 +03006943 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
Avi Kivity7460fb4a2011-09-20 13:43:14 +03006944 || atomic_read(&vcpu->arch.nmi_queued) ||
Gleb Natapova1b37102009-07-09 15:33:52 +03006945 (kvm_arch_interrupt_allowed(vcpu) &&
6946 kvm_cpu_has_interrupt(vcpu));
Zhang Xiantao1d737c82007-12-14 09:35:10 +08006947}
Zhang Xiantao57361992007-12-17 14:21:40 +08006948
Christoffer Dallb6d33832012-03-08 16:44:24 -05006949int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
Zhang Xiantao57361992007-12-17 14:21:40 +08006950{
Christoffer Dallb6d33832012-03-08 16:44:24 -05006951 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
Zhang Xiantao57361992007-12-17 14:21:40 +08006952}
Gleb Natapov78646122009-03-23 12:12:11 +02006953
6954int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
6955{
6956 return kvm_x86_ops->interrupt_allowed(vcpu);
6957}
Marcelo Tosatti229456f2009-06-17 09:22:14 -03006958
Jan Kiszkaf92653e2010-02-23 17:47:55 +01006959bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
6960{
6961 unsigned long current_rip = kvm_rip_read(vcpu) +
6962 get_segment_base(vcpu, VCPU_SREG_CS);
6963
6964 return current_rip == linear_rip;
6965}
6966EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
6967
Jan Kiszka94fe45d2009-10-18 13:24:44 +02006968unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
6969{
6970 unsigned long rflags;
6971
6972 rflags = kvm_x86_ops->get_rflags(vcpu);
6973 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
Jan Kiszkac310bac2010-02-23 17:47:58 +01006974 rflags &= ~X86_EFLAGS_TF;
Jan Kiszka94fe45d2009-10-18 13:24:44 +02006975 return rflags;
6976}
6977EXPORT_SYMBOL_GPL(kvm_get_rflags);
6978
6979void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
6980{
6981 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
Jan Kiszkaf92653e2010-02-23 17:47:55 +01006982 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
Jan Kiszkac310bac2010-02-23 17:47:58 +01006983 rflags |= X86_EFLAGS_TF;
Jan Kiszka94fe45d2009-10-18 13:24:44 +02006984 kvm_x86_ops->set_rflags(vcpu, rflags);
Avi Kivity3842d132010-07-27 12:30:24 +03006985 kvm_make_request(KVM_REQ_EVENT, vcpu);
Jan Kiszka94fe45d2009-10-18 13:24:44 +02006986}
6987EXPORT_SYMBOL_GPL(kvm_set_rflags);
6988
Gleb Natapov56028d02010-10-17 18:13:42 +02006989void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
6990{
6991 int r;
6992
Xiao Guangrongfb67e142010-12-07 10:35:25 +08006993 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
Xiao Guangrongc4806ac2010-11-12 14:49:55 +08006994 is_error_page(work->page))
Gleb Natapov56028d02010-10-17 18:13:42 +02006995 return;
6996
6997 r = kvm_mmu_reload(vcpu);
6998 if (unlikely(r))
6999 return;
7000
Xiao Guangrongfb67e142010-12-07 10:35:25 +08007001 if (!vcpu->arch.mmu.direct_map &&
7002 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
7003 return;
7004
Gleb Natapov56028d02010-10-17 18:13:42 +02007005 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
7006}
7007
Gleb Natapovaf585b92010-10-14 11:22:46 +02007008static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
7009{
7010 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
7011}
7012
7013static inline u32 kvm_async_pf_next_probe(u32 key)
7014{
7015 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
7016}
7017
7018static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7019{
7020 u32 key = kvm_async_pf_hash_fn(gfn);
7021
7022 while (vcpu->arch.apf.gfns[key] != ~0)
7023 key = kvm_async_pf_next_probe(key);
7024
7025 vcpu->arch.apf.gfns[key] = gfn;
7026}
7027
7028static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
7029{
7030 int i;
7031 u32 key = kvm_async_pf_hash_fn(gfn);
7032
7033 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
Xiao Guangrongc7d28c22010-11-01 17:00:30 +08007034 (vcpu->arch.apf.gfns[key] != gfn &&
7035 vcpu->arch.apf.gfns[key] != ~0); i++)
Gleb Natapovaf585b92010-10-14 11:22:46 +02007036 key = kvm_async_pf_next_probe(key);
7037
7038 return key;
7039}
7040
7041bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7042{
7043 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
7044}
7045
7046static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7047{
7048 u32 i, j, k;
7049
7050 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
7051 while (true) {
7052 vcpu->arch.apf.gfns[i] = ~0;
7053 do {
7054 j = kvm_async_pf_next_probe(j);
7055 if (vcpu->arch.apf.gfns[j] == ~0)
7056 return;
7057 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
7058 /*
7059 * k lies cyclically in ]i,j]
7060 * | i.k.j |
7061 * |....j i.k.| or |.k..j i...|
7062 */
7063 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
7064 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
7065 i = j;
7066 }
7067}
7068
Gleb Natapov7c907052010-10-14 11:22:53 +02007069static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
7070{
7071
7072 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
7073 sizeof(val));
7074}
7075
Gleb Natapovaf585b92010-10-14 11:22:46 +02007076void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
7077 struct kvm_async_pf *work)
7078{
Avi Kivity6389ee92010-11-29 16:12:30 +02007079 struct x86_exception fault;
7080
Gleb Natapov7c907052010-10-14 11:22:53 +02007081 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
Gleb Natapovaf585b92010-10-14 11:22:46 +02007082 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
Gleb Natapov7c907052010-10-14 11:22:53 +02007083
7084 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
Gleb Natapovfc5f06f2010-10-14 11:22:56 +02007085 (vcpu->arch.apf.send_user_only &&
7086 kvm_x86_ops->get_cpl(vcpu) == 0))
Gleb Natapov7c907052010-10-14 11:22:53 +02007087 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
7088 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
Avi Kivity6389ee92010-11-29 16:12:30 +02007089 fault.vector = PF_VECTOR;
7090 fault.error_code_valid = true;
7091 fault.error_code = 0;
7092 fault.nested_page_fault = false;
7093 fault.address = work->arch.token;
7094 kvm_inject_page_fault(vcpu, &fault);
Gleb Natapov7c907052010-10-14 11:22:53 +02007095 }
Gleb Natapovaf585b92010-10-14 11:22:46 +02007096}
7097
7098void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
7099 struct kvm_async_pf *work)
7100{
Avi Kivity6389ee92010-11-29 16:12:30 +02007101 struct x86_exception fault;
7102
Gleb Natapov7c907052010-10-14 11:22:53 +02007103 trace_kvm_async_pf_ready(work->arch.token, work->gva);
7104 if (is_error_page(work->page))
7105 work->arch.token = ~0; /* broadcast wakeup */
7106 else
7107 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
7108
7109 if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
7110 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
Avi Kivity6389ee92010-11-29 16:12:30 +02007111 fault.vector = PF_VECTOR;
7112 fault.error_code_valid = true;
7113 fault.error_code = 0;
7114 fault.nested_page_fault = false;
7115 fault.address = work->arch.token;
7116 kvm_inject_page_fault(vcpu, &fault);
Gleb Natapov7c907052010-10-14 11:22:53 +02007117 }
Xiao Guangronge6d53e32010-11-01 17:01:28 +08007118 vcpu->arch.apf.halted = false;
Gleb Natapova4fa1632012-05-03 11:36:39 +03007119 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
Gleb Natapov7c907052010-10-14 11:22:53 +02007120}
7121
7122bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
7123{
7124 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
7125 return true;
7126 else
7127 return !kvm_event_needs_reinjection(vcpu) &&
7128 kvm_x86_ops->interrupt_allowed(vcpu);
Gleb Natapovaf585b92010-10-14 11:22:46 +02007129}
7130
Marcelo Tosatti229456f2009-06-17 09:22:14 -03007131EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
7132EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
7133EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
7134EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
7135EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
Joerg Roedel0ac406d2009-10-09 16:08:27 +02007136EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
Joerg Roedeld8cabdd2009-10-09 16:08:28 +02007137EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
Joerg Roedel17897f32009-10-09 16:08:29 +02007138EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
Joerg Roedel236649d2009-10-09 16:08:30 +02007139EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
Joerg Roedelec1ff792009-10-09 16:08:31 +02007140EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
Joerg Roedel532a46b2009-10-09 16:08:32 +02007141EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
Joerg Roedel2e554e82010-02-24 18:59:14 +01007142EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);