blob: f71500af1f813245bb12092665ac7dea3ba5f24f [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
Sheng Yangaad82702010-05-12 16:40:42 +0800875 /* Update reserved bits */
876 if ((efer ^ old_efer) & EFER_NX)
877 kvm_mmu_reset_context(vcpu);
878
Roedel, Joergb69e8ca2010-05-06 11:38:43 +0200879 return 0;
Carsten Otte15c4a642007-10-30 18:44:17 +0100880}
881
Joerg Roedelf2b4b7d2008-01-31 14:57:37 +0100882void kvm_enable_efer_bits(u64 mask)
883{
884 efer_reserved_bits &= ~mask;
885}
886EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
887
888
Carsten Otte15c4a642007-10-30 18:44:17 +0100889/*
890 * Writes msr value into into the appropriate "register".
891 * Returns 0 on success, non-0 otherwise.
892 * Assumes vcpu_load() was already called.
893 */
Will Auld8fe8ab42012-11-29 12:42:12 -0800894int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
Carsten Otte15c4a642007-10-30 18:44:17 +0100895{
Will Auld8fe8ab42012-11-29 12:42:12 -0800896 return kvm_x86_ops->set_msr(vcpu, msr);
Carsten Otte15c4a642007-10-30 18:44:17 +0100897}
898
Carsten Otte313a3dc2007-10-11 19:16:52 +0200899/*
900 * Adapt set_msr() to msr_io()'s calling convention
901 */
902static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
903{
Will Auld8fe8ab42012-11-29 12:42:12 -0800904 struct msr_data msr;
905
906 msr.data = *data;
907 msr.index = index;
908 msr.host_initiated = true;
909 return kvm_set_msr(vcpu, &msr);
Carsten Otte313a3dc2007-10-11 19:16:52 +0200910}
911
Marcelo Tosatti16e8d742012-11-27 23:29:00 -0200912#ifdef CONFIG_X86_64
913struct pvclock_gtod_data {
914 seqcount_t seq;
915
916 struct { /* extract of a clocksource struct */
917 int vclock_mode;
918 cycle_t cycle_last;
919 cycle_t mask;
920 u32 mult;
921 u32 shift;
922 } clock;
923
924 /* open coded 'struct timespec' */
925 u64 monotonic_time_snsec;
926 time_t monotonic_time_sec;
927};
928
929static struct pvclock_gtod_data pvclock_gtod_data;
930
931static void update_pvclock_gtod(struct timekeeper *tk)
932{
933 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
934
935 write_seqcount_begin(&vdata->seq);
936
937 /* copy pvclock gtod data */
938 vdata->clock.vclock_mode = tk->clock->archdata.vclock_mode;
939 vdata->clock.cycle_last = tk->clock->cycle_last;
940 vdata->clock.mask = tk->clock->mask;
941 vdata->clock.mult = tk->mult;
942 vdata->clock.shift = tk->shift;
943
944 vdata->monotonic_time_sec = tk->xtime_sec
945 + tk->wall_to_monotonic.tv_sec;
946 vdata->monotonic_time_snsec = tk->xtime_nsec
947 + (tk->wall_to_monotonic.tv_nsec
948 << tk->shift);
949 while (vdata->monotonic_time_snsec >=
950 (((u64)NSEC_PER_SEC) << tk->shift)) {
951 vdata->monotonic_time_snsec -=
952 ((u64)NSEC_PER_SEC) << tk->shift;
953 vdata->monotonic_time_sec++;
954 }
955
956 write_seqcount_end(&vdata->seq);
957}
958#endif
959
960
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -0200961static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
962{
Avi Kivity9ed3c442010-05-04 15:00:37 +0300963 int version;
964 int r;
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +0200965 struct pvclock_wall_clock wc;
Jason Wang923de3c2010-01-27 19:13:49 +0800966 struct timespec boot;
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -0200967
968 if (!wall_clock)
969 return;
970
Avi Kivity9ed3c442010-05-04 15:00:37 +0300971 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
972 if (r)
973 return;
974
975 if (version & 1)
976 ++version; /* first time write, random junk */
977
978 ++version;
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -0200979
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -0200980 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
981
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +0200982 /*
983 * The guest calculates current wall clock time by adding
Zachary Amsden34c238a2010-09-18 14:38:14 -1000984 * system time (updated by kvm_guest_time_update below) to the
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +0200985 * wall clock specified here. guest system time equals host
986 * system time for us, thus we must fill in host boot time here.
987 */
Jason Wang923de3c2010-01-27 19:13:49 +0800988 getboottime(&boot);
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +0200989
Bruce Rogers4b648662012-07-20 10:44:24 -0600990 if (kvm->arch.kvmclock_offset) {
991 struct timespec ts = ns_to_timespec(kvm->arch.kvmclock_offset);
992 boot = timespec_sub(boot, ts);
993 }
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +0200994 wc.sec = boot.tv_sec;
995 wc.nsec = boot.tv_nsec;
996 wc.version = version;
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -0200997
998 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
999
1000 version++;
1001 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001002}
1003
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +02001004static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1005{
1006 uint32_t quotient, remainder;
1007
1008 /* Don't try to replace with do_div(), this one calculates
1009 * "(dividend << 32) / divisor" */
1010 __asm__ ( "divl %4"
1011 : "=a" (quotient), "=d" (remainder)
1012 : "0" (0), "1" (dividend), "r" (divisor) );
1013 return quotient;
1014}
1015
Zachary Amsden5f4e3f82010-09-18 14:38:13 -10001016static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz,
1017 s8 *pshift, u32 *pmultiplier)
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +02001018{
Zachary Amsden5f4e3f82010-09-18 14:38:13 -10001019 uint64_t scaled64;
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +02001020 int32_t shift = 0;
1021 uint64_t tps64;
1022 uint32_t tps32;
1023
Zachary Amsden5f4e3f82010-09-18 14:38:13 -10001024 tps64 = base_khz * 1000LL;
1025 scaled64 = scaled_khz * 1000LL;
Jan Kiszka50933622010-09-26 13:00:53 +02001026 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +02001027 tps64 >>= 1;
1028 shift--;
1029 }
1030
1031 tps32 = (uint32_t)tps64;
Jan Kiszka50933622010-09-26 13:00:53 +02001032 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1033 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
Zachary Amsden5f4e3f82010-09-18 14:38:13 -10001034 scaled64 >>= 1;
1035 else
1036 tps32 <<= 1;
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +02001037 shift++;
1038 }
1039
Zachary Amsden5f4e3f82010-09-18 14:38:13 -10001040 *pshift = shift;
1041 *pmultiplier = div_frac(scaled64, tps32);
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +02001042
Zachary Amsden5f4e3f82010-09-18 14:38:13 -10001043 pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
1044 __func__, base_khz, scaled_khz, shift, *pmultiplier);
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +02001045}
1046
Zachary Amsden759379d2010-08-19 22:07:25 -10001047static inline u64 get_kernel_ns(void)
1048{
1049 struct timespec ts;
1050
1051 WARN_ON(preemptible());
1052 ktime_get_ts(&ts);
1053 monotonic_to_bootbased(&ts);
1054 return timespec_to_ns(&ts);
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001055}
1056
Marcelo Tosattid8281992012-11-27 23:29:01 -02001057#ifdef CONFIG_X86_64
Marcelo Tosatti16e8d742012-11-27 23:29:00 -02001058static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
Marcelo Tosattid8281992012-11-27 23:29:01 -02001059#endif
Marcelo Tosatti16e8d742012-11-27 23:29:00 -02001060
Gerd Hoffmannc8076602009-02-04 17:52:04 +01001061static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
Zachary Amsdenc2855452010-09-18 14:38:15 -10001062unsigned long max_tsc_khz;
Gerd Hoffmannc8076602009-02-04 17:52:04 +01001063
Joerg Roedel857e4092011-03-25 09:44:50 +01001064static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
Zachary Amsden759379d2010-08-19 22:07:25 -10001065{
Zachary Amsdencc578282012-02-03 15:43:50 -02001066 return pvclock_scale_delta(nsec, vcpu->arch.virtual_tsc_mult,
1067 vcpu->arch.virtual_tsc_shift);
Zachary Amsden759379d2010-08-19 22:07:25 -10001068}
1069
Zachary Amsdencc578282012-02-03 15:43:50 -02001070static u32 adjust_tsc_khz(u32 khz, s32 ppm)
Zachary Amsdenc2855452010-09-18 14:38:15 -10001071{
Zachary Amsdencc578282012-02-03 15:43:50 -02001072 u64 v = (u64)khz * (1000000 + ppm);
1073 do_div(v, 1000000);
1074 return v;
1075}
1076
1077static void kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 this_tsc_khz)
1078{
1079 u32 thresh_lo, thresh_hi;
1080 int use_scaling = 0;
1081
Zachary Amsdenc2855452010-09-18 14:38:15 -10001082 /* Compute a scale to convert nanoseconds in TSC cycles */
1083 kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000,
Zachary Amsdencc578282012-02-03 15:43:50 -02001084 &vcpu->arch.virtual_tsc_shift,
1085 &vcpu->arch.virtual_tsc_mult);
1086 vcpu->arch.virtual_tsc_khz = this_tsc_khz;
1087
1088 /*
1089 * Compute the variation in TSC rate which is acceptable
1090 * within the range of tolerance and decide if the
1091 * rate being applied is within that bounds of the hardware
1092 * rate. If so, no scaling or compensation need be done.
1093 */
1094 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1095 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1096 if (this_tsc_khz < thresh_lo || this_tsc_khz > thresh_hi) {
1097 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", this_tsc_khz, thresh_lo, thresh_hi);
1098 use_scaling = 1;
1099 }
1100 kvm_x86_ops->set_tsc_khz(vcpu, this_tsc_khz, use_scaling);
Zachary Amsdenc2855452010-09-18 14:38:15 -10001101}
1102
1103static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1104{
Zachary Amsdene26101b2012-02-03 15:43:57 -02001105 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
Zachary Amsdencc578282012-02-03 15:43:50 -02001106 vcpu->arch.virtual_tsc_mult,
1107 vcpu->arch.virtual_tsc_shift);
Zachary Amsdene26101b2012-02-03 15:43:57 -02001108 tsc += vcpu->arch.this_tsc_write;
Zachary Amsdenc2855452010-09-18 14:38:15 -10001109 return tsc;
1110}
1111
Marcelo Tosattib48aa972012-11-27 23:29:03 -02001112void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1113{
1114#ifdef CONFIG_X86_64
1115 bool vcpus_matched;
1116 bool do_request = false;
1117 struct kvm_arch *ka = &vcpu->kvm->arch;
1118 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1119
1120 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1121 atomic_read(&vcpu->kvm->online_vcpus));
1122
1123 if (vcpus_matched && gtod->clock.vclock_mode == VCLOCK_TSC)
1124 if (!ka->use_master_clock)
1125 do_request = 1;
1126
1127 if (!vcpus_matched && ka->use_master_clock)
1128 do_request = 1;
1129
1130 if (do_request)
1131 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1132
1133 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1134 atomic_read(&vcpu->kvm->online_vcpus),
1135 ka->use_master_clock, gtod->clock.vclock_mode);
1136#endif
1137}
1138
Will Auldba904632012-11-29 12:42:50 -08001139static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1140{
1141 u64 curr_offset = kvm_x86_ops->read_tsc_offset(vcpu);
1142 vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1143}
1144
Will Auld8fe8ab42012-11-29 12:42:12 -08001145void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
Zachary Amsden99e3e302010-08-19 22:07:17 -10001146{
1147 struct kvm *kvm = vcpu->kvm;
Zachary Amsdenf38e0982010-08-19 22:07:20 -10001148 u64 offset, ns, elapsed;
Zachary Amsden99e3e302010-08-19 22:07:17 -10001149 unsigned long flags;
Marcelo Tosatti02626b62012-03-08 18:46:57 -03001150 s64 usdiff;
Marcelo Tosattib48aa972012-11-27 23:29:03 -02001151 bool matched;
Will Auld8fe8ab42012-11-29 12:42:12 -08001152 u64 data = msr->data;
Zachary Amsden99e3e302010-08-19 22:07:17 -10001153
Jan Kiszka038f8c12011-02-04 10:49:11 +01001154 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
Joerg Roedel857e4092011-03-25 09:44:50 +01001155 offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
Zachary Amsden759379d2010-08-19 22:07:25 -10001156 ns = get_kernel_ns();
Zachary Amsdenf38e0982010-08-19 22:07:20 -10001157 elapsed = ns - kvm->arch.last_tsc_nsec;
Zachary Amsden5d3cb0f62012-02-03 15:43:51 -02001158
1159 /* n.b - signed multiplication and division required */
Marcelo Tosatti02626b62012-03-08 18:46:57 -03001160 usdiff = data - kvm->arch.last_tsc_write;
Zachary Amsden5d3cb0f62012-02-03 15:43:51 -02001161#ifdef CONFIG_X86_64
Marcelo Tosatti02626b62012-03-08 18:46:57 -03001162 usdiff = (usdiff * 1000) / vcpu->arch.virtual_tsc_khz;
Zachary Amsden5d3cb0f62012-02-03 15:43:51 -02001163#else
1164 /* do_div() only does unsigned */
1165 asm("idivl %2; xor %%edx, %%edx"
Marcelo Tosatti02626b62012-03-08 18:46:57 -03001166 : "=A"(usdiff)
1167 : "A"(usdiff * 1000), "rm"(vcpu->arch.virtual_tsc_khz));
Zachary Amsden5d3cb0f62012-02-03 15:43:51 -02001168#endif
Marcelo Tosatti02626b62012-03-08 18:46:57 -03001169 do_div(elapsed, 1000);
1170 usdiff -= elapsed;
1171 if (usdiff < 0)
1172 usdiff = -usdiff;
Zachary Amsdenf38e0982010-08-19 22:07:20 -10001173
1174 /*
Zachary Amsden5d3cb0f62012-02-03 15:43:51 -02001175 * Special case: TSC write with a small delta (1 second) of virtual
1176 * cycle time against real time is interpreted as an attempt to
1177 * synchronize the CPU.
1178 *
1179 * For a reliable TSC, we can match TSC offsets, and for an unstable
1180 * TSC, we add elapsed time in this computation. We could let the
1181 * compensation code attempt to catch up if we fall behind, but
1182 * it's better to try to match offsets from the beginning.
1183 */
Marcelo Tosatti02626b62012-03-08 18:46:57 -03001184 if (usdiff < USEC_PER_SEC &&
Zachary Amsden5d3cb0f62012-02-03 15:43:51 -02001185 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
Zachary Amsdenf38e0982010-08-19 22:07:20 -10001186 if (!check_tsc_unstable()) {
Zachary Amsdene26101b2012-02-03 15:43:57 -02001187 offset = kvm->arch.cur_tsc_offset;
Zachary Amsdenf38e0982010-08-19 22:07:20 -10001188 pr_debug("kvm: matched tsc offset for %llu\n", data);
1189 } else {
Joerg Roedel857e4092011-03-25 09:44:50 +01001190 u64 delta = nsec_to_cycles(vcpu, elapsed);
Zachary Amsden5d3cb0f62012-02-03 15:43:51 -02001191 data += delta;
1192 offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
Zachary Amsden759379d2010-08-19 22:07:25 -10001193 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
Zachary Amsdenf38e0982010-08-19 22:07:20 -10001194 }
Marcelo Tosattib48aa972012-11-27 23:29:03 -02001195 matched = true;
Zachary Amsdene26101b2012-02-03 15:43:57 -02001196 } else {
1197 /*
1198 * We split periods of matched TSC writes into generations.
1199 * For each generation, we track the original measured
1200 * nanosecond time, offset, and write, so if TSCs are in
1201 * sync, we can match exact offset, and if not, we can match
Guo Chao4a969982012-06-28 15:17:27 +08001202 * exact software computation in compute_guest_tsc()
Zachary Amsdene26101b2012-02-03 15:43:57 -02001203 *
1204 * These values are tracked in kvm->arch.cur_xxx variables.
1205 */
1206 kvm->arch.cur_tsc_generation++;
1207 kvm->arch.cur_tsc_nsec = ns;
1208 kvm->arch.cur_tsc_write = data;
1209 kvm->arch.cur_tsc_offset = offset;
Marcelo Tosattib48aa972012-11-27 23:29:03 -02001210 matched = false;
Zachary Amsdene26101b2012-02-03 15:43:57 -02001211 pr_debug("kvm: new tsc generation %u, clock %llu\n",
1212 kvm->arch.cur_tsc_generation, data);
Zachary Amsdenf38e0982010-08-19 22:07:20 -10001213 }
Zachary Amsdene26101b2012-02-03 15:43:57 -02001214
1215 /*
1216 * We also track th most recent recorded KHZ, write and time to
1217 * allow the matching interval to be extended at each write.
1218 */
Zachary Amsdenf38e0982010-08-19 22:07:20 -10001219 kvm->arch.last_tsc_nsec = ns;
1220 kvm->arch.last_tsc_write = data;
Zachary Amsden5d3cb0f62012-02-03 15:43:51 -02001221 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
Zachary Amsden99e3e302010-08-19 22:07:17 -10001222
1223 /* Reset of TSC must disable overshoot protection below */
1224 vcpu->arch.hv_clock.tsc_timestamp = 0;
Zachary Amsdenb183aa52012-02-03 15:43:53 -02001225 vcpu->arch.last_guest_tsc = data;
Zachary Amsdene26101b2012-02-03 15:43:57 -02001226
1227 /* Keep track of which generation this VCPU has synchronized to */
1228 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1229 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1230 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1231
Will Auldba904632012-11-29 12:42:50 -08001232 if (guest_cpuid_has_tsc_adjust(vcpu) && !msr->host_initiated)
1233 update_ia32_tsc_adjust_msr(vcpu, offset);
Zachary Amsdene26101b2012-02-03 15:43:57 -02001234 kvm_x86_ops->write_tsc_offset(vcpu, offset);
1235 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
Marcelo Tosattib48aa972012-11-27 23:29:03 -02001236
1237 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1238 if (matched)
1239 kvm->arch.nr_vcpus_matched_tsc++;
1240 else
1241 kvm->arch.nr_vcpus_matched_tsc = 0;
1242
1243 kvm_track_tsc_matching(vcpu);
1244 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
Zachary Amsden99e3e302010-08-19 22:07:17 -10001245}
Zachary Amsdene26101b2012-02-03 15:43:57 -02001246
Zachary Amsden99e3e302010-08-19 22:07:17 -10001247EXPORT_SYMBOL_GPL(kvm_write_tsc);
1248
Marcelo Tosattid8281992012-11-27 23:29:01 -02001249#ifdef CONFIG_X86_64
1250
1251static cycle_t read_tsc(void)
1252{
1253 cycle_t ret;
1254 u64 last;
1255
1256 /*
1257 * Empirically, a fence (of type that depends on the CPU)
1258 * before rdtsc is enough to ensure that rdtsc is ordered
1259 * with respect to loads. The various CPU manuals are unclear
1260 * as to whether rdtsc can be reordered with later loads,
1261 * but no one has ever seen it happen.
1262 */
1263 rdtsc_barrier();
1264 ret = (cycle_t)vget_cycles();
1265
1266 last = pvclock_gtod_data.clock.cycle_last;
1267
1268 if (likely(ret >= last))
1269 return ret;
1270
1271 /*
1272 * GCC likes to generate cmov here, but this branch is extremely
1273 * predictable (it's just a funciton of time and the likely is
1274 * very likely) and there's a data dependence, so force GCC
1275 * to generate a branch instead. I don't barrier() because
1276 * we don't actually need a barrier, and if this function
1277 * ever gets inlined it will generate worse code.
1278 */
1279 asm volatile ("");
1280 return last;
1281}
1282
1283static inline u64 vgettsc(cycle_t *cycle_now)
1284{
1285 long v;
1286 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1287
1288 *cycle_now = read_tsc();
1289
1290 v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask;
1291 return v * gtod->clock.mult;
1292}
1293
1294static int do_monotonic(struct timespec *ts, cycle_t *cycle_now)
1295{
1296 unsigned long seq;
1297 u64 ns;
1298 int mode;
1299 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1300
1301 ts->tv_nsec = 0;
1302 do {
1303 seq = read_seqcount_begin(&gtod->seq);
1304 mode = gtod->clock.vclock_mode;
1305 ts->tv_sec = gtod->monotonic_time_sec;
1306 ns = gtod->monotonic_time_snsec;
1307 ns += vgettsc(cycle_now);
1308 ns >>= gtod->clock.shift;
1309 } while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
1310 timespec_add_ns(ts, ns);
1311
1312 return mode;
1313}
1314
1315/* returns true if host is using tsc clocksource */
1316static bool kvm_get_time_and_clockread(s64 *kernel_ns, cycle_t *cycle_now)
1317{
1318 struct timespec ts;
1319
1320 /* checked again under seqlock below */
1321 if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1322 return false;
1323
1324 if (do_monotonic(&ts, cycle_now) != VCLOCK_TSC)
1325 return false;
1326
1327 monotonic_to_bootbased(&ts);
1328 *kernel_ns = timespec_to_ns(&ts);
1329
1330 return true;
1331}
1332#endif
1333
1334/*
1335 *
Marcelo Tosattib48aa972012-11-27 23:29:03 -02001336 * Assuming a stable TSC across physical CPUS, and a stable TSC
1337 * across virtual CPUs, the following condition is possible.
1338 * Each numbered line represents an event visible to both
Marcelo Tosattid8281992012-11-27 23:29:01 -02001339 * CPUs at the next numbered event.
1340 *
1341 * "timespecX" represents host monotonic time. "tscX" represents
1342 * RDTSC value.
1343 *
1344 * VCPU0 on CPU0 | VCPU1 on CPU1
1345 *
1346 * 1. read timespec0,tsc0
1347 * 2. | timespec1 = timespec0 + N
1348 * | tsc1 = tsc0 + M
1349 * 3. transition to guest | transition to guest
1350 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1351 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
1352 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1353 *
1354 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1355 *
1356 * - ret0 < ret1
1357 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1358 * ...
1359 * - 0 < N - M => M < N
1360 *
1361 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1362 * always the case (the difference between two distinct xtime instances
1363 * might be smaller then the difference between corresponding TSC reads,
1364 * when updating guest vcpus pvclock areas).
1365 *
1366 * To avoid that problem, do not allow visibility of distinct
1367 * system_timestamp/tsc_timestamp values simultaneously: use a master
1368 * copy of host monotonic time values. Update that master copy
1369 * in lockstep.
1370 *
Marcelo Tosattib48aa972012-11-27 23:29:03 -02001371 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
Marcelo Tosattid8281992012-11-27 23:29:01 -02001372 *
1373 */
1374
1375static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1376{
1377#ifdef CONFIG_X86_64
1378 struct kvm_arch *ka = &kvm->arch;
1379 int vclock_mode;
Marcelo Tosattib48aa972012-11-27 23:29:03 -02001380 bool host_tsc_clocksource, vcpus_matched;
1381
1382 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1383 atomic_read(&kvm->online_vcpus));
Marcelo Tosattid8281992012-11-27 23:29:01 -02001384
1385 /*
1386 * If the host uses TSC clock, then passthrough TSC as stable
1387 * to the guest.
1388 */
Marcelo Tosattib48aa972012-11-27 23:29:03 -02001389 host_tsc_clocksource = kvm_get_time_and_clockread(
Marcelo Tosattid8281992012-11-27 23:29:01 -02001390 &ka->master_kernel_ns,
1391 &ka->master_cycle_now);
1392
Marcelo Tosattib48aa972012-11-27 23:29:03 -02001393 ka->use_master_clock = host_tsc_clocksource & vcpus_matched;
1394
Marcelo Tosattid8281992012-11-27 23:29:01 -02001395 if (ka->use_master_clock)
1396 atomic_set(&kvm_guest_has_master_clock, 1);
1397
1398 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
Marcelo Tosattib48aa972012-11-27 23:29:03 -02001399 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1400 vcpus_matched);
Marcelo Tosattid8281992012-11-27 23:29:01 -02001401#endif
1402}
1403
Zachary Amsden34c238a2010-09-18 14:38:14 -10001404static int kvm_guest_time_update(struct kvm_vcpu *v)
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001405{
Marcelo Tosattid8281992012-11-27 23:29:01 -02001406 unsigned long flags, this_tsc_khz;
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001407 struct kvm_vcpu_arch *vcpu = &v->arch;
Marcelo Tosattid8281992012-11-27 23:29:01 -02001408 struct kvm_arch *ka = &v->kvm->arch;
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001409 void *shared_kaddr;
Zachary Amsden1d5f0662010-08-19 22:07:30 -10001410 s64 kernel_ns, max_kernel_ns;
Marcelo Tosattid8281992012-11-27 23:29:01 -02001411 u64 tsc_timestamp, host_tsc;
Marcelo Tosatti78c03372012-11-27 23:28:47 -02001412 struct pvclock_vcpu_time_info *guest_hv_clock;
Marcelo Tosatti51d59c62012-08-03 15:57:49 -03001413 u8 pvclock_flags;
Marcelo Tosattid8281992012-11-27 23:29:01 -02001414 bool use_master_clock;
1415
1416 kernel_ns = 0;
1417 host_tsc = 0;
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +02001418
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001419 /* Keep irq disabled to prevent changes to the clock */
1420 local_irq_save(flags);
Zachary Amsdencc578282012-02-03 15:43:50 -02001421 this_tsc_khz = __get_cpu_var(cpu_tsc_khz);
Zachary Amsden8cfdc002010-08-19 22:07:21 -10001422 if (unlikely(this_tsc_khz == 0)) {
Zachary Amsdenc2855452010-09-18 14:38:15 -10001423 local_irq_restore(flags);
Zachary Amsden34c238a2010-09-18 14:38:14 -10001424 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
Zachary Amsden8cfdc002010-08-19 22:07:21 -10001425 return 1;
1426 }
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001427
Zachary Amsden1d5f0662010-08-19 22:07:30 -10001428 /*
Marcelo Tosattid8281992012-11-27 23:29:01 -02001429 * If the host uses TSC clock, then passthrough TSC as stable
1430 * to the guest.
1431 */
1432 spin_lock(&ka->pvclock_gtod_sync_lock);
1433 use_master_clock = ka->use_master_clock;
1434 if (use_master_clock) {
1435 host_tsc = ka->master_cycle_now;
1436 kernel_ns = ka->master_kernel_ns;
1437 }
1438 spin_unlock(&ka->pvclock_gtod_sync_lock);
1439 if (!use_master_clock) {
1440 host_tsc = native_read_tsc();
1441 kernel_ns = get_kernel_ns();
1442 }
1443
1444 tsc_timestamp = kvm_x86_ops->read_l1_tsc(v, host_tsc);
1445
1446 /*
Zachary Amsdenc2855452010-09-18 14:38:15 -10001447 * We may have to catch up the TSC to match elapsed wall clock
1448 * time for two reasons, even if kvmclock is used.
1449 * 1) CPU could have been running below the maximum TSC rate
1450 * 2) Broken TSC compensation resets the base at each VCPU
1451 * entry to avoid unknown leaps of TSC even when running
1452 * again on the same CPU. This may cause apparent elapsed
1453 * time to disappear, and the guest to stand still or run
1454 * very slowly.
1455 */
1456 if (vcpu->tsc_catchup) {
1457 u64 tsc = compute_guest_tsc(v, kernel_ns);
1458 if (tsc > tsc_timestamp) {
Marcelo Tosattif1e2b262012-02-03 15:43:55 -02001459 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
Zachary Amsdenc2855452010-09-18 14:38:15 -10001460 tsc_timestamp = tsc;
1461 }
1462 }
1463
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001464 local_irq_restore(flags);
1465
Zachary Amsdenc2855452010-09-18 14:38:15 -10001466 if (!vcpu->time_page)
1467 return 0;
1468
1469 /*
Zachary Amsden1d5f0662010-08-19 22:07:30 -10001470 * Time as measured by the TSC may go backwards when resetting the base
1471 * tsc_timestamp. The reason for this is that the TSC resolution is
1472 * higher than the resolution of the other clock scales. Thus, many
1473 * possible measurments of the TSC correspond to one measurement of any
1474 * other clock, and so a spread of values is possible. This is not a
1475 * problem for the computation of the nanosecond clock; with TSC rates
1476 * around 1GHZ, there can only be a few cycles which correspond to one
1477 * nanosecond value, and any path through this code will inevitably
1478 * take longer than that. However, with the kernel_ns value itself,
1479 * the precision may be much lower, down to HZ granularity. If the
1480 * first sampling of TSC against kernel_ns ends in the low part of the
1481 * range, and the second in the high end of the range, we can get:
1482 *
1483 * (TSC - offset_low) * S + kns_old > (TSC - offset_high) * S + kns_new
1484 *
1485 * As the sampling errors potentially range in the thousands of cycles,
1486 * it is possible such a time value has already been observed by the
1487 * guest. To protect against this, we must compute the system time as
1488 * observed by the guest and ensure the new system time is greater.
1489 */
1490 max_kernel_ns = 0;
Zachary Amsdenb183aa52012-02-03 15:43:53 -02001491 if (vcpu->hv_clock.tsc_timestamp) {
Zachary Amsden1d5f0662010-08-19 22:07:30 -10001492 max_kernel_ns = vcpu->last_guest_tsc -
1493 vcpu->hv_clock.tsc_timestamp;
1494 max_kernel_ns = pvclock_scale_delta(max_kernel_ns,
1495 vcpu->hv_clock.tsc_to_system_mul,
1496 vcpu->hv_clock.tsc_shift);
1497 max_kernel_ns += vcpu->last_kernel_ns;
1498 }
1499
Zachary Amsdene48672f2010-08-19 22:07:23 -10001500 if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
Zachary Amsden5f4e3f82010-09-18 14:38:13 -10001501 kvm_get_time_scale(NSEC_PER_SEC / 1000, this_tsc_khz,
1502 &vcpu->hv_clock.tsc_shift,
1503 &vcpu->hv_clock.tsc_to_system_mul);
Zachary Amsdene48672f2010-08-19 22:07:23 -10001504 vcpu->hw_tsc_khz = this_tsc_khz;
Zachary Amsden8cfdc002010-08-19 22:07:21 -10001505 }
1506
Marcelo Tosattid8281992012-11-27 23:29:01 -02001507 /* with a master <monotonic time, tsc value> tuple,
1508 * pvclock clock reads always increase at the (scaled) rate
1509 * of guest TSC - no need to deal with sampling errors.
1510 */
1511 if (!use_master_clock) {
1512 if (max_kernel_ns > kernel_ns)
1513 kernel_ns = max_kernel_ns;
1514 }
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001515 /* With all the info we got, fill in the values */
Zachary Amsden1d5f0662010-08-19 22:07:30 -10001516 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
Zachary Amsden759379d2010-08-19 22:07:25 -10001517 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
Zachary Amsden1d5f0662010-08-19 22:07:30 -10001518 vcpu->last_kernel_ns = kernel_ns;
Zachary Amsden28e46392010-09-18 14:38:12 -10001519 vcpu->last_guest_tsc = tsc_timestamp;
Marcelo Tosatti51d59c62012-08-03 15:57:49 -03001520
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001521 /*
1522 * The interface expects us to write an even number signaling that the
1523 * update is finished. Since the guest won't see the intermediate
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +02001524 * state, we just increase by 2 at the end.
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001525 */
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +02001526 vcpu->hv_clock.version += 2;
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001527
Cong Wang8fd75e12011-11-25 23:14:17 +08001528 shared_kaddr = kmap_atomic(vcpu->time_page);
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001529
Marcelo Tosatti78c03372012-11-27 23:28:47 -02001530 guest_hv_clock = shared_kaddr + vcpu->time_offset;
1531
1532 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1533 pvclock_flags = (guest_hv_clock->flags & PVCLOCK_GUEST_STOPPED);
1534
1535 if (vcpu->pvclock_set_guest_stopped_request) {
1536 pvclock_flags |= PVCLOCK_GUEST_STOPPED;
1537 vcpu->pvclock_set_guest_stopped_request = false;
1538 }
1539
Marcelo Tosattid8281992012-11-27 23:29:01 -02001540 /* If the host uses TSC clocksource, then it is stable */
1541 if (use_master_clock)
1542 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
1543
Marcelo Tosatti78c03372012-11-27 23:28:47 -02001544 vcpu->hv_clock.flags = pvclock_flags;
1545
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001546 memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
Gerd Hoffmann50d0a0f2008-06-03 16:17:31 +02001547 sizeof(vcpu->hv_clock));
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001548
Cong Wang8fd75e12011-11-25 23:14:17 +08001549 kunmap_atomic(shared_kaddr);
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001550
1551 mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
Zachary Amsden8cfdc002010-08-19 22:07:21 -10001552 return 0;
Gerd Hoffmannc8076602009-02-04 17:52:04 +01001553}
1554
Avi Kivity9ba075a2008-05-26 20:06:35 +03001555static bool msr_mtrr_valid(unsigned msr)
1556{
1557 switch (msr) {
1558 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
1559 case MSR_MTRRfix64K_00000:
1560 case MSR_MTRRfix16K_80000:
1561 case MSR_MTRRfix16K_A0000:
1562 case MSR_MTRRfix4K_C0000:
1563 case MSR_MTRRfix4K_C8000:
1564 case MSR_MTRRfix4K_D0000:
1565 case MSR_MTRRfix4K_D8000:
1566 case MSR_MTRRfix4K_E0000:
1567 case MSR_MTRRfix4K_E8000:
1568 case MSR_MTRRfix4K_F0000:
1569 case MSR_MTRRfix4K_F8000:
1570 case MSR_MTRRdefType:
1571 case MSR_IA32_CR_PAT:
1572 return true;
1573 case 0x2f8:
1574 return true;
1575 }
1576 return false;
1577}
1578
Marcelo Tosattid6289b92009-06-22 15:27:56 -03001579static bool valid_pat_type(unsigned t)
1580{
1581 return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
1582}
1583
1584static bool valid_mtrr_type(unsigned t)
1585{
1586 return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
1587}
1588
1589static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1590{
1591 int i;
1592
1593 if (!msr_mtrr_valid(msr))
1594 return false;
1595
1596 if (msr == MSR_IA32_CR_PAT) {
1597 for (i = 0; i < 8; i++)
1598 if (!valid_pat_type((data >> (i * 8)) & 0xff))
1599 return false;
1600 return true;
1601 } else if (msr == MSR_MTRRdefType) {
1602 if (data & ~0xcff)
1603 return false;
1604 return valid_mtrr_type(data & 0xff);
1605 } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
1606 for (i = 0; i < 8 ; i++)
1607 if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
1608 return false;
1609 return true;
1610 }
1611
1612 /* variable MTRRs */
1613 return valid_mtrr_type(data & 0xff);
1614}
1615
Avi Kivity9ba075a2008-05-26 20:06:35 +03001616static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1617{
Sheng Yang0bed3b52008-10-09 16:01:54 +08001618 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1619
Marcelo Tosattid6289b92009-06-22 15:27:56 -03001620 if (!mtrr_valid(vcpu, msr, data))
Avi Kivity9ba075a2008-05-26 20:06:35 +03001621 return 1;
1622
Sheng Yang0bed3b52008-10-09 16:01:54 +08001623 if (msr == MSR_MTRRdefType) {
1624 vcpu->arch.mtrr_state.def_type = data;
1625 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
1626 } else if (msr == MSR_MTRRfix64K_00000)
1627 p[0] = data;
1628 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1629 p[1 + msr - MSR_MTRRfix16K_80000] = data;
1630 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1631 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
1632 else if (msr == MSR_IA32_CR_PAT)
1633 vcpu->arch.pat = data;
1634 else { /* Variable MTRRs */
1635 int idx, is_mtrr_mask;
1636 u64 *pt;
1637
1638 idx = (msr - 0x200) / 2;
1639 is_mtrr_mask = msr - 0x200 - 2 * idx;
1640 if (!is_mtrr_mask)
1641 pt =
1642 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1643 else
1644 pt =
1645 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1646 *pt = data;
1647 }
1648
1649 kvm_mmu_reset_context(vcpu);
Avi Kivity9ba075a2008-05-26 20:06:35 +03001650 return 0;
1651}
Carsten Otte15c4a642007-10-30 18:44:17 +01001652
Huang Ying890ca9a2009-05-11 16:48:15 +08001653static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1654{
1655 u64 mcg_cap = vcpu->arch.mcg_cap;
1656 unsigned bank_num = mcg_cap & 0xff;
1657
1658 switch (msr) {
1659 case MSR_IA32_MCG_STATUS:
1660 vcpu->arch.mcg_status = data;
1661 break;
1662 case MSR_IA32_MCG_CTL:
1663 if (!(mcg_cap & MCG_CTL_P))
1664 return 1;
1665 if (data != 0 && data != ~(u64)0)
1666 return -1;
1667 vcpu->arch.mcg_ctl = data;
1668 break;
1669 default:
1670 if (msr >= MSR_IA32_MC0_CTL &&
1671 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1672 u32 offset = msr - MSR_IA32_MC0_CTL;
Andre Przywara114be422010-03-24 17:46:42 +01001673 /* only 0 or all 1s can be written to IA32_MCi_CTL
1674 * some Linux kernels though clear bit 10 in bank 4 to
1675 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1676 * this to avoid an uncatched #GP in the guest
1677 */
Huang Ying890ca9a2009-05-11 16:48:15 +08001678 if ((offset & 0x3) == 0 &&
Andre Przywara114be422010-03-24 17:46:42 +01001679 data != 0 && (data | (1 << 10)) != ~(u64)0)
Huang Ying890ca9a2009-05-11 16:48:15 +08001680 return -1;
1681 vcpu->arch.mce_banks[offset] = data;
1682 break;
1683 }
1684 return 1;
1685 }
1686 return 0;
1687}
1688
Ed Swierkffde22a2009-10-15 15:21:43 -07001689static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1690{
1691 struct kvm *kvm = vcpu->kvm;
1692 int lm = is_long_mode(vcpu);
1693 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1694 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1695 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1696 : kvm->arch.xen_hvm_config.blob_size_32;
1697 u32 page_num = data & ~PAGE_MASK;
1698 u64 page_addr = data & PAGE_MASK;
1699 u8 *page;
1700 int r;
1701
1702 r = -E2BIG;
1703 if (page_num >= blob_size)
1704 goto out;
1705 r = -ENOMEM;
Sasha Levinff5c2c02011-12-04 19:36:29 +02001706 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
1707 if (IS_ERR(page)) {
1708 r = PTR_ERR(page);
Ed Swierkffde22a2009-10-15 15:21:43 -07001709 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02001710 }
Ed Swierkffde22a2009-10-15 15:21:43 -07001711 if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
1712 goto out_free;
1713 r = 0;
1714out_free:
1715 kfree(page);
1716out:
1717 return r;
1718}
1719
Gleb Natapov55cd8e52010-01-17 15:51:22 +02001720static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1721{
1722 return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
1723}
1724
1725static bool kvm_hv_msr_partition_wide(u32 msr)
1726{
1727 bool r = false;
1728 switch (msr) {
1729 case HV_X64_MSR_GUEST_OS_ID:
1730 case HV_X64_MSR_HYPERCALL:
1731 r = true;
1732 break;
1733 }
1734
1735 return r;
1736}
1737
1738static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1739{
1740 struct kvm *kvm = vcpu->kvm;
1741
1742 switch (msr) {
1743 case HV_X64_MSR_GUEST_OS_ID:
1744 kvm->arch.hv_guest_os_id = data;
1745 /* setting guest os id to zero disables hypercall page */
1746 if (!kvm->arch.hv_guest_os_id)
1747 kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1748 break;
1749 case HV_X64_MSR_HYPERCALL: {
1750 u64 gfn;
1751 unsigned long addr;
1752 u8 instructions[4];
1753
1754 /* if guest os id is not set hypercall should remain disabled */
1755 if (!kvm->arch.hv_guest_os_id)
1756 break;
1757 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1758 kvm->arch.hv_hypercall = data;
1759 break;
1760 }
1761 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1762 addr = gfn_to_hva(kvm, gfn);
1763 if (kvm_is_error_hva(addr))
1764 return 1;
1765 kvm_x86_ops->patch_hypercall(vcpu, instructions);
1766 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
Xiao Guangrong8b0cedf2011-05-15 23:22:04 +08001767 if (__copy_to_user((void __user *)addr, instructions, 4))
Gleb Natapov55cd8e52010-01-17 15:51:22 +02001768 return 1;
1769 kvm->arch.hv_hypercall = data;
1770 break;
1771 }
1772 default:
Christoffer Dalla737f252012-06-03 21:17:48 +03001773 vcpu_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1774 "data 0x%llx\n", msr, data);
Gleb Natapov55cd8e52010-01-17 15:51:22 +02001775 return 1;
1776 }
1777 return 0;
1778}
1779
1780static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1781{
Gleb Natapov10388a02010-01-17 15:51:23 +02001782 switch (msr) {
1783 case HV_X64_MSR_APIC_ASSIST_PAGE: {
1784 unsigned long addr;
Gleb Natapov55cd8e52010-01-17 15:51:22 +02001785
Gleb Natapov10388a02010-01-17 15:51:23 +02001786 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1787 vcpu->arch.hv_vapic = data;
1788 break;
1789 }
1790 addr = gfn_to_hva(vcpu->kvm, data >>
1791 HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT);
1792 if (kvm_is_error_hva(addr))
1793 return 1;
Xiao Guangrong8b0cedf2011-05-15 23:22:04 +08001794 if (__clear_user((void __user *)addr, PAGE_SIZE))
Gleb Natapov10388a02010-01-17 15:51:23 +02001795 return 1;
1796 vcpu->arch.hv_vapic = data;
1797 break;
1798 }
1799 case HV_X64_MSR_EOI:
1800 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1801 case HV_X64_MSR_ICR:
1802 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1803 case HV_X64_MSR_TPR:
1804 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1805 default:
Christoffer Dalla737f252012-06-03 21:17:48 +03001806 vcpu_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1807 "data 0x%llx\n", msr, data);
Gleb Natapov10388a02010-01-17 15:51:23 +02001808 return 1;
1809 }
1810
1811 return 0;
Gleb Natapov55cd8e52010-01-17 15:51:22 +02001812}
1813
Gleb Natapov344d9582010-10-14 11:22:50 +02001814static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
1815{
1816 gpa_t gpa = data & ~0x3f;
1817
Guo Chao4a969982012-06-28 15:17:27 +08001818 /* Bits 2:5 are reserved, Should be zero */
Gleb Natapov6adba522010-10-14 11:22:55 +02001819 if (data & 0x3c)
Gleb Natapov344d9582010-10-14 11:22:50 +02001820 return 1;
1821
1822 vcpu->arch.apf.msr_val = data;
1823
1824 if (!(data & KVM_ASYNC_PF_ENABLED)) {
1825 kvm_clear_async_pf_completion_queue(vcpu);
1826 kvm_async_pf_hash_reset(vcpu);
1827 return 0;
1828 }
1829
1830 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa))
1831 return 1;
1832
Gleb Natapov6adba522010-10-14 11:22:55 +02001833 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
Gleb Natapov344d9582010-10-14 11:22:50 +02001834 kvm_async_pf_wakeup_all(vcpu);
1835 return 0;
1836}
1837
Glauber Costa12f9a482011-02-01 14:16:40 -05001838static void kvmclock_reset(struct kvm_vcpu *vcpu)
1839{
1840 if (vcpu->arch.time_page) {
1841 kvm_release_page_dirty(vcpu->arch.time_page);
1842 vcpu->arch.time_page = NULL;
1843 }
1844}
1845
Glauber Costac9aaa892011-07-11 15:28:14 -04001846static void accumulate_steal_time(struct kvm_vcpu *vcpu)
1847{
1848 u64 delta;
1849
1850 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
1851 return;
1852
1853 delta = current->sched_info.run_delay - vcpu->arch.st.last_steal;
1854 vcpu->arch.st.last_steal = current->sched_info.run_delay;
1855 vcpu->arch.st.accum_steal = delta;
1856}
1857
1858static void record_steal_time(struct kvm_vcpu *vcpu)
1859{
1860 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
1861 return;
1862
1863 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
1864 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
1865 return;
1866
1867 vcpu->arch.st.steal.steal += vcpu->arch.st.accum_steal;
1868 vcpu->arch.st.steal.version += 2;
1869 vcpu->arch.st.accum_steal = 0;
1870
1871 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
1872 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
1873}
1874
Will Auld8fe8ab42012-11-29 12:42:12 -08001875int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
Carsten Otte15c4a642007-10-30 18:44:17 +01001876{
Gleb Natapov57537852012-01-15 14:17:22 +02001877 bool pr = false;
Will Auld8fe8ab42012-11-29 12:42:12 -08001878 u32 msr = msr_info->index;
1879 u64 data = msr_info->data;
Gleb Natapov57537852012-01-15 14:17:22 +02001880
Carsten Otte15c4a642007-10-30 18:44:17 +01001881 switch (msr) {
Borislav Petkov2e32b712013-02-19 19:33:13 +01001882 case MSR_AMD64_NB_CFG:
1883 case MSR_IA32_UCODE_REV:
1884 case MSR_IA32_UCODE_WRITE:
1885 case MSR_VM_HSAVE_PA:
1886 case MSR_AMD64_PATCH_LOADER:
1887 case MSR_AMD64_BU_CFG2:
1888 break;
1889
Carsten Otte15c4a642007-10-30 18:44:17 +01001890 case MSR_EFER:
Roedel, Joergb69e8ca2010-05-06 11:38:43 +02001891 return set_efer(vcpu, data);
Andre Przywara8f1589d2009-06-24 12:44:33 +02001892 case MSR_K7_HWCR:
1893 data &= ~(u64)0x40; /* ignore flush filter disable */
Joerg Roedel82494022010-02-24 18:59:16 +01001894 data &= ~(u64)0x100; /* ignore ignne emulation enable */
Nicolae Mogoreanua223c312012-02-21 13:44:21 -08001895 data &= ~(u64)0x8; /* ignore TLB cache disable */
Andre Przywara8f1589d2009-06-24 12:44:33 +02001896 if (data != 0) {
Christoffer Dalla737f252012-06-03 21:17:48 +03001897 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
1898 data);
Andre Przywara8f1589d2009-06-24 12:44:33 +02001899 return 1;
1900 }
Carsten Otte15c4a642007-10-30 18:44:17 +01001901 break;
Andre Przywaraf7c6d142009-07-02 15:04:14 +02001902 case MSR_FAM10H_MMIO_CONF_BASE:
1903 if (data != 0) {
Christoffer Dalla737f252012-06-03 21:17:48 +03001904 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
1905 "0x%llx\n", data);
Andre Przywaraf7c6d142009-07-02 15:04:14 +02001906 return 1;
1907 }
Carsten Otte15c4a642007-10-30 18:44:17 +01001908 break;
Alexander Grafb5e2fec2008-07-22 08:00:45 +02001909 case MSR_IA32_DEBUGCTLMSR:
1910 if (!data) {
1911 /* We support the non-activated case already */
1912 break;
1913 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
1914 /* Values other than LBR and BTF are vendor-specific,
1915 thus reserved and should throw a #GP */
1916 return 1;
1917 }
Christoffer Dalla737f252012-06-03 21:17:48 +03001918 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1919 __func__, data);
Alexander Grafb5e2fec2008-07-22 08:00:45 +02001920 break;
Avi Kivity9ba075a2008-05-26 20:06:35 +03001921 case 0x200 ... 0x2ff:
1922 return set_msr_mtrr(vcpu, msr, data);
Carsten Otte15c4a642007-10-30 18:44:17 +01001923 case MSR_IA32_APICBASE:
1924 kvm_set_apic_base(vcpu, data);
1925 break;
Gleb Natapov0105d1a2009-07-05 17:39:36 +03001926 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1927 return kvm_x2apic_msr_write(vcpu, msr, data);
Liu, Jinsonga3e06bb2011-09-22 16:55:52 +08001928 case MSR_IA32_TSCDEADLINE:
1929 kvm_set_lapic_tscdeadline_msr(vcpu, data);
1930 break;
Will Auldba904632012-11-29 12:42:50 -08001931 case MSR_IA32_TSC_ADJUST:
1932 if (guest_cpuid_has_tsc_adjust(vcpu)) {
1933 if (!msr_info->host_initiated) {
1934 u64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
1935 kvm_x86_ops->adjust_tsc_offset(vcpu, adj, true);
1936 }
1937 vcpu->arch.ia32_tsc_adjust_msr = data;
1938 }
1939 break;
Carsten Otte15c4a642007-10-30 18:44:17 +01001940 case MSR_IA32_MISC_ENABLE:
Zhang Xiantaoad312c72007-12-13 23:50:52 +08001941 vcpu->arch.ia32_misc_enable_msr = data;
Carsten Otte15c4a642007-10-30 18:44:17 +01001942 break;
Glauber Costa11c6bff2010-05-11 12:17:41 -04001943 case MSR_KVM_WALL_CLOCK_NEW:
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001944 case MSR_KVM_WALL_CLOCK:
1945 vcpu->kvm->arch.wall_clock = data;
1946 kvm_write_wall_clock(vcpu->kvm, data);
1947 break;
Glauber Costa11c6bff2010-05-11 12:17:41 -04001948 case MSR_KVM_SYSTEM_TIME_NEW:
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001949 case MSR_KVM_SYSTEM_TIME: {
Glauber Costa12f9a482011-02-01 14:16:40 -05001950 kvmclock_reset(vcpu);
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001951
1952 vcpu->arch.time = data;
Zachary Amsdenc2855452010-09-18 14:38:15 -10001953 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001954
1955 /* we verify if the enable bit is set... */
1956 if (!(data & 1))
1957 break;
1958
1959 /* ...but clean it before doing the actual write */
1960 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
1961
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001962 vcpu->arch.time_page =
1963 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001964
Xiao Guangrong32cad842012-08-03 15:42:52 +08001965 if (is_error_page(vcpu->arch.time_page))
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001966 vcpu->arch.time_page = NULL;
Xiao Guangrong32cad842012-08-03 15:42:52 +08001967
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02001968 break;
1969 }
Gleb Natapov344d9582010-10-14 11:22:50 +02001970 case MSR_KVM_ASYNC_PF_EN:
1971 if (kvm_pv_enable_async_pf(vcpu, data))
1972 return 1;
1973 break;
Glauber Costac9aaa892011-07-11 15:28:14 -04001974 case MSR_KVM_STEAL_TIME:
1975
1976 if (unlikely(!sched_info_on()))
1977 return 1;
1978
1979 if (data & KVM_STEAL_RESERVED_MASK)
1980 return 1;
1981
1982 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
1983 data & KVM_STEAL_VALID_BITS))
1984 return 1;
1985
1986 vcpu->arch.st.msr_val = data;
1987
1988 if (!(data & KVM_MSR_ENABLED))
1989 break;
1990
1991 vcpu->arch.st.last_steal = current->sched_info.run_delay;
1992
1993 preempt_disable();
1994 accumulate_steal_time(vcpu);
1995 preempt_enable();
1996
1997 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
1998
1999 break;
Michael S. Tsirkinae7a2a32012-06-24 19:25:07 +03002000 case MSR_KVM_PV_EOI_EN:
2001 if (kvm_lapic_enable_pv_eoi(vcpu, data))
2002 return 1;
2003 break;
Glauber Costac9aaa892011-07-11 15:28:14 -04002004
Huang Ying890ca9a2009-05-11 16:48:15 +08002005 case MSR_IA32_MCG_CTL:
2006 case MSR_IA32_MCG_STATUS:
2007 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
2008 return set_msr_mce(vcpu, msr, data);
Andre Przywara71db6022009-06-12 22:01:29 +02002009
2010 /* Performance counters are not protected by a CPUID bit,
2011 * so we should check all of them in the generic path for the sake of
2012 * cross vendor migration.
2013 * Writing a zero into the event select MSRs disables them,
2014 * which we perfectly emulate ;-). Any other value should be at least
2015 * reported, some guests depend on them.
2016 */
Andre Przywara71db6022009-06-12 22:01:29 +02002017 case MSR_K7_EVNTSEL0:
2018 case MSR_K7_EVNTSEL1:
2019 case MSR_K7_EVNTSEL2:
2020 case MSR_K7_EVNTSEL3:
2021 if (data != 0)
Christoffer Dalla737f252012-06-03 21:17:48 +03002022 vcpu_unimpl(vcpu, "unimplemented perfctr wrmsr: "
2023 "0x%x data 0x%llx\n", msr, data);
Andre Przywara71db6022009-06-12 22:01:29 +02002024 break;
2025 /* at least RHEL 4 unconditionally writes to the perfctr registers,
2026 * so we ignore writes to make it happy.
2027 */
Andre Przywara71db6022009-06-12 22:01:29 +02002028 case MSR_K7_PERFCTR0:
2029 case MSR_K7_PERFCTR1:
2030 case MSR_K7_PERFCTR2:
2031 case MSR_K7_PERFCTR3:
Christoffer Dalla737f252012-06-03 21:17:48 +03002032 vcpu_unimpl(vcpu, "unimplemented perfctr wrmsr: "
2033 "0x%x data 0x%llx\n", msr, data);
Andre Przywara71db6022009-06-12 22:01:29 +02002034 break;
Gleb Natapov57537852012-01-15 14:17:22 +02002035 case MSR_P6_PERFCTR0:
2036 case MSR_P6_PERFCTR1:
2037 pr = true;
2038 case MSR_P6_EVNTSEL0:
2039 case MSR_P6_EVNTSEL1:
2040 if (kvm_pmu_msr(vcpu, msr))
2041 return kvm_pmu_set_msr(vcpu, msr, data);
2042
2043 if (pr || data != 0)
Christoffer Dalla737f252012-06-03 21:17:48 +03002044 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2045 "0x%x data 0x%llx\n", msr, data);
Gleb Natapov57537852012-01-15 14:17:22 +02002046 break;
Jes Sorensen84e0cef2010-09-01 11:42:04 +02002047 case MSR_K7_CLK_CTL:
2048 /*
2049 * Ignore all writes to this no longer documented MSR.
2050 * Writes are only relevant for old K7 processors,
2051 * all pre-dating SVM, but a recommended workaround from
Guo Chao4a969982012-06-28 15:17:27 +08002052 * AMD for these chips. It is possible to specify the
Jes Sorensen84e0cef2010-09-01 11:42:04 +02002053 * affected processor models on the command line, hence
2054 * the need to ignore the workaround.
2055 */
2056 break;
Gleb Natapov55cd8e52010-01-17 15:51:22 +02002057 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2058 if (kvm_hv_msr_partition_wide(msr)) {
2059 int r;
2060 mutex_lock(&vcpu->kvm->lock);
2061 r = set_msr_hyperv_pw(vcpu, msr, data);
2062 mutex_unlock(&vcpu->kvm->lock);
2063 return r;
2064 } else
2065 return set_msr_hyperv(vcpu, msr, data);
2066 break;
john cooper91c9c3e2011-01-21 00:21:00 -05002067 case MSR_IA32_BBL_CR_CTL3:
2068 /* Drop writes to this legacy MSR -- see rdmsr
2069 * counterpart for further detail.
2070 */
Christoffer Dalla737f252012-06-03 21:17:48 +03002071 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
john cooper91c9c3e2011-01-21 00:21:00 -05002072 break;
Boris Ostrovsky2b036c62012-01-09 14:00:35 -05002073 case MSR_AMD64_OSVW_ID_LENGTH:
2074 if (!guest_cpuid_has_osvw(vcpu))
2075 return 1;
2076 vcpu->arch.osvw.length = data;
2077 break;
2078 case MSR_AMD64_OSVW_STATUS:
2079 if (!guest_cpuid_has_osvw(vcpu))
2080 return 1;
2081 vcpu->arch.osvw.status = data;
2082 break;
Carsten Otte15c4a642007-10-30 18:44:17 +01002083 default:
Ed Swierkffde22a2009-10-15 15:21:43 -07002084 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2085 return xen_hvm_config(vcpu, data);
Gleb Natapovf5132b02011-11-10 14:57:22 +02002086 if (kvm_pmu_msr(vcpu, msr))
2087 return kvm_pmu_set_msr(vcpu, msr, data);
Andre Przywaraed85c062009-06-25 12:36:49 +02002088 if (!ignore_msrs) {
Christoffer Dalla737f252012-06-03 21:17:48 +03002089 vcpu_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
2090 msr, data);
Andre Przywaraed85c062009-06-25 12:36:49 +02002091 return 1;
2092 } else {
Christoffer Dalla737f252012-06-03 21:17:48 +03002093 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
2094 msr, data);
Andre Przywaraed85c062009-06-25 12:36:49 +02002095 break;
2096 }
Carsten Otte15c4a642007-10-30 18:44:17 +01002097 }
2098 return 0;
2099}
2100EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2101
2102
2103/*
2104 * Reads an msr value (of 'msr_index') into 'pdata'.
2105 * Returns 0 on success, non-0 otherwise.
2106 * Assumes vcpu_load() was already called.
2107 */
2108int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
2109{
2110 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
2111}
2112
Avi Kivity9ba075a2008-05-26 20:06:35 +03002113static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2114{
Sheng Yang0bed3b52008-10-09 16:01:54 +08002115 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
2116
Avi Kivity9ba075a2008-05-26 20:06:35 +03002117 if (!msr_mtrr_valid(msr))
2118 return 1;
2119
Sheng Yang0bed3b52008-10-09 16:01:54 +08002120 if (msr == MSR_MTRRdefType)
2121 *pdata = vcpu->arch.mtrr_state.def_type +
2122 (vcpu->arch.mtrr_state.enabled << 10);
2123 else if (msr == MSR_MTRRfix64K_00000)
2124 *pdata = p[0];
2125 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
2126 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
2127 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
2128 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
2129 else if (msr == MSR_IA32_CR_PAT)
2130 *pdata = vcpu->arch.pat;
2131 else { /* Variable MTRRs */
2132 int idx, is_mtrr_mask;
2133 u64 *pt;
2134
2135 idx = (msr - 0x200) / 2;
2136 is_mtrr_mask = msr - 0x200 - 2 * idx;
2137 if (!is_mtrr_mask)
2138 pt =
2139 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
2140 else
2141 pt =
2142 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
2143 *pdata = *pt;
2144 }
2145
Avi Kivity9ba075a2008-05-26 20:06:35 +03002146 return 0;
2147}
2148
Huang Ying890ca9a2009-05-11 16:48:15 +08002149static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2150{
2151 u64 data;
2152 u64 mcg_cap = vcpu->arch.mcg_cap;
2153 unsigned bank_num = mcg_cap & 0xff;
2154
2155 switch (msr) {
2156 case MSR_IA32_P5_MC_ADDR:
2157 case MSR_IA32_P5_MC_TYPE:
2158 data = 0;
2159 break;
2160 case MSR_IA32_MCG_CAP:
2161 data = vcpu->arch.mcg_cap;
2162 break;
2163 case MSR_IA32_MCG_CTL:
2164 if (!(mcg_cap & MCG_CTL_P))
2165 return 1;
2166 data = vcpu->arch.mcg_ctl;
2167 break;
2168 case MSR_IA32_MCG_STATUS:
2169 data = vcpu->arch.mcg_status;
2170 break;
2171 default:
2172 if (msr >= MSR_IA32_MC0_CTL &&
2173 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
2174 u32 offset = msr - MSR_IA32_MC0_CTL;
2175 data = vcpu->arch.mce_banks[offset];
2176 break;
2177 }
2178 return 1;
2179 }
2180 *pdata = data;
2181 return 0;
2182}
2183
Gleb Natapov55cd8e52010-01-17 15:51:22 +02002184static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2185{
2186 u64 data = 0;
2187 struct kvm *kvm = vcpu->kvm;
2188
2189 switch (msr) {
2190 case HV_X64_MSR_GUEST_OS_ID:
2191 data = kvm->arch.hv_guest_os_id;
2192 break;
2193 case HV_X64_MSR_HYPERCALL:
2194 data = kvm->arch.hv_hypercall;
2195 break;
2196 default:
Christoffer Dalla737f252012-06-03 21:17:48 +03002197 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
Gleb Natapov55cd8e52010-01-17 15:51:22 +02002198 return 1;
2199 }
2200
2201 *pdata = data;
2202 return 0;
2203}
2204
2205static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2206{
2207 u64 data = 0;
2208
2209 switch (msr) {
2210 case HV_X64_MSR_VP_INDEX: {
2211 int r;
2212 struct kvm_vcpu *v;
2213 kvm_for_each_vcpu(r, v, vcpu->kvm)
2214 if (v == vcpu)
2215 data = r;
2216 break;
2217 }
Gleb Natapov10388a02010-01-17 15:51:23 +02002218 case HV_X64_MSR_EOI:
2219 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
2220 case HV_X64_MSR_ICR:
2221 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
2222 case HV_X64_MSR_TPR:
2223 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
Mike Waychison14fa67e2011-07-21 15:38:10 -07002224 case HV_X64_MSR_APIC_ASSIST_PAGE:
Mike Waychisond1613ad2011-07-23 00:31:45 -07002225 data = vcpu->arch.hv_vapic;
2226 break;
Gleb Natapov55cd8e52010-01-17 15:51:22 +02002227 default:
Christoffer Dalla737f252012-06-03 21:17:48 +03002228 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
Gleb Natapov55cd8e52010-01-17 15:51:22 +02002229 return 1;
2230 }
2231 *pdata = data;
2232 return 0;
2233}
2234
Carsten Otte15c4a642007-10-30 18:44:17 +01002235int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2236{
2237 u64 data;
2238
2239 switch (msr) {
Carsten Otte15c4a642007-10-30 18:44:17 +01002240 case MSR_IA32_PLATFORM_ID:
Carsten Otte15c4a642007-10-30 18:44:17 +01002241 case MSR_IA32_EBL_CR_POWERON:
Alexander Grafb5e2fec2008-07-22 08:00:45 +02002242 case MSR_IA32_DEBUGCTLMSR:
2243 case MSR_IA32_LASTBRANCHFROMIP:
2244 case MSR_IA32_LASTBRANCHTOIP:
2245 case MSR_IA32_LASTINTFROMIP:
2246 case MSR_IA32_LASTINTTOIP:
Jaswinder Singh Rajput60af2ec2009-05-14 11:00:10 +05302247 case MSR_K8_SYSCFG:
2248 case MSR_K7_HWCR:
Avi Kivity61a6bd62008-12-29 17:32:28 +02002249 case MSR_VM_HSAVE_PA:
Amit Shah9e699622009-06-15 13:25:34 +05302250 case MSR_K7_EVNTSEL0:
Amit Shah1f3ee612009-06-30 16:24:28 +05302251 case MSR_K7_PERFCTR0:
Andre Przywara1fdbd482009-06-24 12:44:34 +02002252 case MSR_K8_INT_PENDING_MSG:
Andre Przywarac323c0e2009-06-24 15:37:05 +02002253 case MSR_AMD64_NB_CFG:
Andre Przywaraf7c6d142009-07-02 15:04:14 +02002254 case MSR_FAM10H_MMIO_CONF_BASE:
Borislav Petkov2e32b712013-02-19 19:33:13 +01002255 case MSR_AMD64_BU_CFG2:
Carsten Otte15c4a642007-10-30 18:44:17 +01002256 data = 0;
2257 break;
Gleb Natapov57537852012-01-15 14:17:22 +02002258 case MSR_P6_PERFCTR0:
2259 case MSR_P6_PERFCTR1:
2260 case MSR_P6_EVNTSEL0:
2261 case MSR_P6_EVNTSEL1:
2262 if (kvm_pmu_msr(vcpu, msr))
2263 return kvm_pmu_get_msr(vcpu, msr, pdata);
2264 data = 0;
2265 break;
Marcelo Tosatti742bc672011-07-29 19:44:21 -03002266 case MSR_IA32_UCODE_REV:
2267 data = 0x100000000ULL;
2268 break;
Avi Kivity9ba075a2008-05-26 20:06:35 +03002269 case MSR_MTRRcap:
2270 data = 0x500 | KVM_NR_VAR_MTRR;
2271 break;
2272 case 0x200 ... 0x2ff:
2273 return get_msr_mtrr(vcpu, msr, pdata);
Carsten Otte15c4a642007-10-30 18:44:17 +01002274 case 0xcd: /* fsb frequency */
2275 data = 3;
2276 break;
Jes Sorensen7b914092010-09-09 12:06:46 +02002277 /*
2278 * MSR_EBC_FREQUENCY_ID
2279 * Conservative value valid for even the basic CPU models.
2280 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2281 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2282 * and 266MHz for model 3, or 4. Set Core Clock
2283 * Frequency to System Bus Frequency Ratio to 1 (bits
2284 * 31:24) even though these are only valid for CPU
2285 * models > 2, however guests may end up dividing or
2286 * multiplying by zero otherwise.
2287 */
2288 case MSR_EBC_FREQUENCY_ID:
2289 data = 1 << 24;
2290 break;
Carsten Otte15c4a642007-10-30 18:44:17 +01002291 case MSR_IA32_APICBASE:
2292 data = kvm_get_apic_base(vcpu);
2293 break;
Gleb Natapov0105d1a2009-07-05 17:39:36 +03002294 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2295 return kvm_x2apic_msr_read(vcpu, msr, pdata);
2296 break;
Liu, Jinsonga3e06bb2011-09-22 16:55:52 +08002297 case MSR_IA32_TSCDEADLINE:
2298 data = kvm_get_lapic_tscdeadline_msr(vcpu);
2299 break;
Will Auldba904632012-11-29 12:42:50 -08002300 case MSR_IA32_TSC_ADJUST:
2301 data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2302 break;
Carsten Otte15c4a642007-10-30 18:44:17 +01002303 case MSR_IA32_MISC_ENABLE:
Zhang Xiantaoad312c72007-12-13 23:50:52 +08002304 data = vcpu->arch.ia32_misc_enable_msr;
Carsten Otte15c4a642007-10-30 18:44:17 +01002305 break;
Alexander Graf847f0ad2008-02-21 12:11:01 +01002306 case MSR_IA32_PERF_STATUS:
2307 /* TSC increment by tick */
2308 data = 1000ULL;
2309 /* CPU multiplier */
2310 data |= (((uint64_t)4ULL) << 40);
2311 break;
Carsten Otte15c4a642007-10-30 18:44:17 +01002312 case MSR_EFER:
Avi Kivityf6801df2010-01-21 15:31:50 +02002313 data = vcpu->arch.efer;
Carsten Otte15c4a642007-10-30 18:44:17 +01002314 break;
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02002315 case MSR_KVM_WALL_CLOCK:
Glauber Costa11c6bff2010-05-11 12:17:41 -04002316 case MSR_KVM_WALL_CLOCK_NEW:
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02002317 data = vcpu->kvm->arch.wall_clock;
2318 break;
2319 case MSR_KVM_SYSTEM_TIME:
Glauber Costa11c6bff2010-05-11 12:17:41 -04002320 case MSR_KVM_SYSTEM_TIME_NEW:
Glauber de Oliveira Costa18068522008-02-15 17:52:47 -02002321 data = vcpu->arch.time;
2322 break;
Gleb Natapov344d9582010-10-14 11:22:50 +02002323 case MSR_KVM_ASYNC_PF_EN:
2324 data = vcpu->arch.apf.msr_val;
2325 break;
Glauber Costac9aaa892011-07-11 15:28:14 -04002326 case MSR_KVM_STEAL_TIME:
2327 data = vcpu->arch.st.msr_val;
2328 break;
Michael S. Tsirkin1d921282012-08-26 18:00:29 +03002329 case MSR_KVM_PV_EOI_EN:
2330 data = vcpu->arch.pv_eoi.msr_val;
2331 break;
Huang Ying890ca9a2009-05-11 16:48:15 +08002332 case MSR_IA32_P5_MC_ADDR:
2333 case MSR_IA32_P5_MC_TYPE:
2334 case MSR_IA32_MCG_CAP:
2335 case MSR_IA32_MCG_CTL:
2336 case MSR_IA32_MCG_STATUS:
2337 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
2338 return get_msr_mce(vcpu, msr, pdata);
Jes Sorensen84e0cef2010-09-01 11:42:04 +02002339 case MSR_K7_CLK_CTL:
2340 /*
2341 * Provide expected ramp-up count for K7. All other
2342 * are set to zero, indicating minimum divisors for
2343 * every field.
2344 *
2345 * This prevents guest kernels on AMD host with CPU
2346 * type 6, model 8 and higher from exploding due to
2347 * the rdmsr failing.
2348 */
2349 data = 0x20000000;
2350 break;
Gleb Natapov55cd8e52010-01-17 15:51:22 +02002351 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2352 if (kvm_hv_msr_partition_wide(msr)) {
2353 int r;
2354 mutex_lock(&vcpu->kvm->lock);
2355 r = get_msr_hyperv_pw(vcpu, msr, pdata);
2356 mutex_unlock(&vcpu->kvm->lock);
2357 return r;
2358 } else
2359 return get_msr_hyperv(vcpu, msr, pdata);
2360 break;
john cooper91c9c3e2011-01-21 00:21:00 -05002361 case MSR_IA32_BBL_CR_CTL3:
2362 /* This legacy MSR exists but isn't fully documented in current
2363 * silicon. It is however accessed by winxp in very narrow
2364 * scenarios where it sets bit #19, itself documented as
2365 * a "reserved" bit. Best effort attempt to source coherent
2366 * read data here should the balance of the register be
2367 * interpreted by the guest:
2368 *
2369 * L2 cache control register 3: 64GB range, 256KB size,
2370 * enabled, latency 0x1, configured
2371 */
2372 data = 0xbe702111;
2373 break;
Boris Ostrovsky2b036c62012-01-09 14:00:35 -05002374 case MSR_AMD64_OSVW_ID_LENGTH:
2375 if (!guest_cpuid_has_osvw(vcpu))
2376 return 1;
2377 data = vcpu->arch.osvw.length;
2378 break;
2379 case MSR_AMD64_OSVW_STATUS:
2380 if (!guest_cpuid_has_osvw(vcpu))
2381 return 1;
2382 data = vcpu->arch.osvw.status;
2383 break;
Carsten Otte15c4a642007-10-30 18:44:17 +01002384 default:
Gleb Natapovf5132b02011-11-10 14:57:22 +02002385 if (kvm_pmu_msr(vcpu, msr))
2386 return kvm_pmu_get_msr(vcpu, msr, pdata);
Andre Przywaraed85c062009-06-25 12:36:49 +02002387 if (!ignore_msrs) {
Christoffer Dalla737f252012-06-03 21:17:48 +03002388 vcpu_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
Andre Przywaraed85c062009-06-25 12:36:49 +02002389 return 1;
2390 } else {
Christoffer Dalla737f252012-06-03 21:17:48 +03002391 vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
Andre Przywaraed85c062009-06-25 12:36:49 +02002392 data = 0;
2393 }
2394 break;
Carsten Otte15c4a642007-10-30 18:44:17 +01002395 }
2396 *pdata = data;
2397 return 0;
2398}
2399EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2400
Carsten Otte313a3dc2007-10-11 19:16:52 +02002401/*
2402 * Read or write a bunch of msrs. All parameters are kernel addresses.
2403 *
2404 * @return number of msrs set successfully.
2405 */
2406static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2407 struct kvm_msr_entry *entries,
2408 int (*do_msr)(struct kvm_vcpu *vcpu,
2409 unsigned index, u64 *data))
2410{
Marcelo Tosattif656ce02009-12-23 14:35:25 -02002411 int i, idx;
Carsten Otte313a3dc2007-10-11 19:16:52 +02002412
Marcelo Tosattif656ce02009-12-23 14:35:25 -02002413 idx = srcu_read_lock(&vcpu->kvm->srcu);
Carsten Otte313a3dc2007-10-11 19:16:52 +02002414 for (i = 0; i < msrs->nmsrs; ++i)
2415 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2416 break;
Marcelo Tosattif656ce02009-12-23 14:35:25 -02002417 srcu_read_unlock(&vcpu->kvm->srcu, idx);
Carsten Otte313a3dc2007-10-11 19:16:52 +02002418
Carsten Otte313a3dc2007-10-11 19:16:52 +02002419 return i;
2420}
2421
2422/*
2423 * Read or write a bunch of msrs. Parameters are user addresses.
2424 *
2425 * @return number of msrs set successfully.
2426 */
2427static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2428 int (*do_msr)(struct kvm_vcpu *vcpu,
2429 unsigned index, u64 *data),
2430 int writeback)
2431{
2432 struct kvm_msrs msrs;
2433 struct kvm_msr_entry *entries;
2434 int r, n;
2435 unsigned size;
2436
2437 r = -EFAULT;
2438 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2439 goto out;
2440
2441 r = -E2BIG;
2442 if (msrs.nmsrs >= MAX_IO_MSRS)
2443 goto out;
2444
Carsten Otte313a3dc2007-10-11 19:16:52 +02002445 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002446 entries = memdup_user(user_msrs->entries, size);
2447 if (IS_ERR(entries)) {
2448 r = PTR_ERR(entries);
Carsten Otte313a3dc2007-10-11 19:16:52 +02002449 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002450 }
Carsten Otte313a3dc2007-10-11 19:16:52 +02002451
2452 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2453 if (r < 0)
2454 goto out_free;
2455
2456 r = -EFAULT;
2457 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2458 goto out_free;
2459
2460 r = n;
2461
2462out_free:
Avi Kivity7a73c022010-07-22 23:24:52 +03002463 kfree(entries);
Carsten Otte313a3dc2007-10-11 19:16:52 +02002464out:
2465 return r;
2466}
2467
Zhang Xiantao018d00d2007-11-15 23:07:47 +08002468int kvm_dev_ioctl_check_extension(long ext)
2469{
2470 int r;
2471
2472 switch (ext) {
2473 case KVM_CAP_IRQCHIP:
2474 case KVM_CAP_HLT:
2475 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
Zhang Xiantao018d00d2007-11-15 23:07:47 +08002476 case KVM_CAP_SET_TSS_ADDR:
Dan Kenigsberg07716712007-11-21 17:10:04 +02002477 case KVM_CAP_EXT_CPUID:
Gerd Hoffmannc8076602009-02-04 17:52:04 +01002478 case KVM_CAP_CLOCKSOURCE:
Sheng Yang78376992008-01-28 05:10:22 +08002479 case KVM_CAP_PIT:
Marcelo Tosattia28e4f52008-02-22 12:21:36 -05002480 case KVM_CAP_NOP_IO_DELAY:
Marcelo Tosatti62d9f0d2008-04-11 13:24:45 -03002481 case KVM_CAP_MP_STATE:
Avi Kivityed848622008-07-29 11:30:57 +03002482 case KVM_CAP_SYNC_MMU:
Lai Jiangshana355c852010-12-14 17:57:47 +08002483 case KVM_CAP_USER_NMI:
Marcelo Tosatti52d939a2008-12-30 15:55:06 -02002484 case KVM_CAP_REINJECT_CONTROL:
Gleb Natapov49256632009-02-04 17:28:14 +02002485 case KVM_CAP_IRQ_INJECT_STATUS:
Sheng Yange56d5322009-03-12 21:45:39 +08002486 case KVM_CAP_ASSIGN_DEV_IRQ:
Gregory Haskins721eecb2009-05-20 10:30:49 -04002487 case KVM_CAP_IRQFD:
Gregory Haskinsd34e6b12009-07-07 17:08:49 -04002488 case KVM_CAP_IOEVENTFD:
Jan Kiszkac5ff41c2009-05-14 22:42:53 +02002489 case KVM_CAP_PIT2:
Beth Kone9f42752009-07-07 11:50:38 -04002490 case KVM_CAP_PIT_STATE2:
Sheng Yangb927a3c2009-07-21 10:42:48 +08002491 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
Ed Swierkffde22a2009-10-15 15:21:43 -07002492 case KVM_CAP_XEN_HVM:
Glauber Costaafbcf7a2009-10-16 15:28:36 -04002493 case KVM_CAP_ADJUST_CLOCK:
Jan Kiszka3cfc3092009-11-12 01:04:25 +01002494 case KVM_CAP_VCPU_EVENTS:
Gleb Natapov55cd8e52010-01-17 15:51:22 +02002495 case KVM_CAP_HYPERV:
Gleb Natapov10388a02010-01-17 15:51:23 +02002496 case KVM_CAP_HYPERV_VAPIC:
Gleb Natapovc25bc162010-01-17 15:51:24 +02002497 case KVM_CAP_HYPERV_SPIN:
Zhai, Edwinab9f4ec2010-01-29 14:38:44 +08002498 case KVM_CAP_PCI_SEGMENT:
Jan Kiszkaa1efbe72010-02-15 10:45:43 +01002499 case KVM_CAP_DEBUGREGS:
Jan Kiszkad2be1652010-02-23 17:47:57 +01002500 case KVM_CAP_X86_ROBUST_SINGLESTEP:
Sheng Yang2d5b5a62010-06-13 17:29:39 +08002501 case KVM_CAP_XSAVE:
Gleb Natapov344d9582010-10-14 11:22:50 +02002502 case KVM_CAP_ASYNC_PF:
Joerg Roedel92a1f122011-03-25 09:44:51 +01002503 case KVM_CAP_GET_TSC_KHZ:
Jan Kiszka07700a92012-02-28 14:19:54 +01002504 case KVM_CAP_PCI_2_3:
Eric B Munson1c0b28c2012-03-10 14:37:27 -05002505 case KVM_CAP_KVMCLOCK_CTRL:
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08002506 case KVM_CAP_READONLY_MEM:
Alex Williamson7a844282012-09-21 11:58:03 -06002507 case KVM_CAP_IRQFD_RESAMPLE:
Zhang Xiantao018d00d2007-11-15 23:07:47 +08002508 r = 1;
2509 break;
Laurent Vivier542472b2008-05-30 16:05:55 +02002510 case KVM_CAP_COALESCED_MMIO:
2511 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2512 break;
Avi Kivity774ead32007-12-26 13:57:04 +02002513 case KVM_CAP_VAPIC:
2514 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2515 break;
Avi Kivityf7252302008-02-20 11:53:16 +02002516 case KVM_CAP_NR_VCPUS:
Sasha Levin8c3ba332011-07-18 17:17:15 +03002517 r = KVM_SOFT_MAX_VCPUS;
2518 break;
2519 case KVM_CAP_MAX_VCPUS:
Avi Kivityf7252302008-02-20 11:53:16 +02002520 r = KVM_MAX_VCPUS;
2521 break;
Avi Kivitya988b912008-02-20 11:59:20 +02002522 case KVM_CAP_NR_MEMSLOTS:
Alex Williamsonbbacc0c2012-12-10 10:33:09 -07002523 r = KVM_USER_MEM_SLOTS;
Avi Kivitya988b912008-02-20 11:59:20 +02002524 break;
Marcelo Tosattia68a6a72009-10-01 19:28:39 -03002525 case KVM_CAP_PV_MMU: /* obsolete */
2526 r = 0;
Marcelo Tosatti2f333bc2008-02-22 12:21:37 -05002527 break;
Ben-Ami Yassour62c476c2008-09-14 03:48:28 +03002528 case KVM_CAP_IOMMU:
Joerg Roedela1b60c12011-09-06 18:46:34 +02002529 r = iommu_present(&pci_bus_type);
Ben-Ami Yassour62c476c2008-09-14 03:48:28 +03002530 break;
Huang Ying890ca9a2009-05-11 16:48:15 +08002531 case KVM_CAP_MCE:
2532 r = KVM_MAX_MCE_BANKS;
2533 break;
Sheng Yang2d5b5a62010-06-13 17:29:39 +08002534 case KVM_CAP_XCRS:
2535 r = cpu_has_xsave;
2536 break;
Joerg Roedel92a1f122011-03-25 09:44:51 +01002537 case KVM_CAP_TSC_CONTROL:
2538 r = kvm_has_tsc_control;
2539 break;
Jan Kiszka4d25a0662011-12-21 12:28:29 +01002540 case KVM_CAP_TSC_DEADLINE_TIMER:
2541 r = boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER);
2542 break;
Zhang Xiantao018d00d2007-11-15 23:07:47 +08002543 default:
2544 r = 0;
2545 break;
2546 }
2547 return r;
2548
2549}
2550
Carsten Otte043405e2007-10-10 17:16:19 +02002551long kvm_arch_dev_ioctl(struct file *filp,
2552 unsigned int ioctl, unsigned long arg)
2553{
2554 void __user *argp = (void __user *)arg;
2555 long r;
2556
2557 switch (ioctl) {
2558 case KVM_GET_MSR_INDEX_LIST: {
2559 struct kvm_msr_list __user *user_msr_list = argp;
2560 struct kvm_msr_list msr_list;
2561 unsigned n;
2562
2563 r = -EFAULT;
2564 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2565 goto out;
2566 n = msr_list.nmsrs;
2567 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2568 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2569 goto out;
2570 r = -E2BIG;
Jan Kiszkae125e7b2009-07-02 21:45:47 +02002571 if (n < msr_list.nmsrs)
Carsten Otte043405e2007-10-10 17:16:19 +02002572 goto out;
2573 r = -EFAULT;
2574 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2575 num_msrs_to_save * sizeof(u32)))
2576 goto out;
Jan Kiszkae125e7b2009-07-02 21:45:47 +02002577 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
Carsten Otte043405e2007-10-10 17:16:19 +02002578 &emulated_msrs,
2579 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2580 goto out;
2581 r = 0;
2582 break;
2583 }
Avi Kivity674eea02008-02-11 18:37:23 +02002584 case KVM_GET_SUPPORTED_CPUID: {
2585 struct kvm_cpuid2 __user *cpuid_arg = argp;
2586 struct kvm_cpuid2 cpuid;
2587
2588 r = -EFAULT;
2589 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2590 goto out;
2591 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
Amit Shah19355472009-01-14 16:56:00 +00002592 cpuid_arg->entries);
Avi Kivity674eea02008-02-11 18:37:23 +02002593 if (r)
2594 goto out;
2595
2596 r = -EFAULT;
2597 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2598 goto out;
2599 r = 0;
2600 break;
2601 }
Huang Ying890ca9a2009-05-11 16:48:15 +08002602 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2603 u64 mce_cap;
2604
2605 mce_cap = KVM_MCE_CAP_SUPPORTED;
2606 r = -EFAULT;
2607 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2608 goto out;
2609 r = 0;
2610 break;
2611 }
Carsten Otte043405e2007-10-10 17:16:19 +02002612 default:
2613 r = -EINVAL;
2614 }
2615out:
2616 return r;
2617}
2618
Sheng Yangf5f48ee2010-06-30 12:25:15 +08002619static void wbinvd_ipi(void *garbage)
2620{
2621 wbinvd();
2622}
2623
2624static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2625{
2626 return vcpu->kvm->arch.iommu_domain &&
2627 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY);
2628}
2629
Carsten Otte313a3dc2007-10-11 19:16:52 +02002630void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2631{
Sheng Yangf5f48ee2010-06-30 12:25:15 +08002632 /* Address WBINVD may be executed by guest */
2633 if (need_emulate_wbinvd(vcpu)) {
2634 if (kvm_x86_ops->has_wbinvd_exit())
2635 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2636 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2637 smp_call_function_single(vcpu->cpu,
2638 wbinvd_ipi, NULL, 1);
2639 }
2640
Carsten Otte313a3dc2007-10-11 19:16:52 +02002641 kvm_x86_ops->vcpu_load(vcpu, cpu);
Zachary Amsden0dd6a6e2012-02-03 15:43:56 -02002642
2643 /* Apply any externally detected TSC adjustments (due to suspend) */
2644 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
2645 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
2646 vcpu->arch.tsc_offset_adjustment = 0;
2647 set_bit(KVM_REQ_CLOCK_UPDATE, &vcpu->requests);
2648 }
2649
Zachary Amsden48434c202010-08-19 22:07:24 -10002650 if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
Zachary Amsden6f526ec2012-02-03 15:43:54 -02002651 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2652 native_read_tsc() - vcpu->arch.last_host_tsc;
Zachary Amsdene48672f2010-08-19 22:07:23 -10002653 if (tsc_delta < 0)
2654 mark_tsc_unstable("KVM discovered backwards TSC");
Zachary Amsdenc2855452010-09-18 14:38:15 -10002655 if (check_tsc_unstable()) {
Zachary Amsdenb183aa52012-02-03 15:43:53 -02002656 u64 offset = kvm_x86_ops->compute_tsc_offset(vcpu,
2657 vcpu->arch.last_guest_tsc);
2658 kvm_x86_ops->write_tsc_offset(vcpu, offset);
Zachary Amsdenc2855452010-09-18 14:38:15 -10002659 vcpu->arch.tsc_catchup = 1;
Zachary Amsdenc2855452010-09-18 14:38:15 -10002660 }
Marcelo Tosattid98d07c2012-11-27 23:29:04 -02002661 /*
2662 * On a host with synchronized TSC, there is no need to update
2663 * kvmclock on vcpu->cpu migration
2664 */
2665 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
2666 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
Zachary Amsdenc2855452010-09-18 14:38:15 -10002667 if (vcpu->cpu != cpu)
2668 kvm_migrate_timers(vcpu);
Zachary Amsdene48672f2010-08-19 22:07:23 -10002669 vcpu->cpu = cpu;
Zachary Amsden6b7d7e72009-10-09 16:26:08 -10002670 }
Glauber Costac9aaa892011-07-11 15:28:14 -04002671
2672 accumulate_steal_time(vcpu);
2673 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
Carsten Otte313a3dc2007-10-11 19:16:52 +02002674}
2675
2676void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2677{
Avi Kivity02daab22009-12-30 12:40:26 +02002678 kvm_x86_ops->vcpu_put(vcpu);
Avi Kivity1c11e712010-05-03 16:05:44 +03002679 kvm_put_guest_fpu(vcpu);
Zachary Amsden6f526ec2012-02-03 15:43:54 -02002680 vcpu->arch.last_host_tsc = native_read_tsc();
Carsten Otte313a3dc2007-10-11 19:16:52 +02002681}
2682
Carsten Otte313a3dc2007-10-11 19:16:52 +02002683static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2684 struct kvm_lapic_state *s)
2685{
Zhang Xiantaoad312c72007-12-13 23:50:52 +08002686 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
Carsten Otte313a3dc2007-10-11 19:16:52 +02002687
2688 return 0;
2689}
2690
2691static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2692 struct kvm_lapic_state *s)
2693{
Gleb Natapov64eb0622012-08-08 15:24:36 +03002694 kvm_apic_post_state_restore(vcpu, s);
Gleb Natapovcb142eb2009-08-09 15:17:40 +03002695 update_cr8_intercept(vcpu);
Carsten Otte313a3dc2007-10-11 19:16:52 +02002696
2697 return 0;
2698}
2699
Zhang Xiantaof77bc6a2007-11-21 04:36:41 +08002700static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2701 struct kvm_interrupt *irq)
2702{
Michael S. Tsirkina50abc32012-09-05 20:00:52 +03002703 if (irq->irq < 0 || irq->irq >= KVM_NR_INTERRUPTS)
Zhang Xiantaof77bc6a2007-11-21 04:36:41 +08002704 return -EINVAL;
2705 if (irqchip_in_kernel(vcpu->kvm))
2706 return -ENXIO;
Zhang Xiantaof77bc6a2007-11-21 04:36:41 +08002707
Gleb Natapov66fd3f72009-05-11 13:35:50 +03002708 kvm_queue_interrupt(vcpu, irq->irq, false);
Avi Kivity3842d132010-07-27 12:30:24 +03002709 kvm_make_request(KVM_REQ_EVENT, vcpu);
Zhang Xiantaof77bc6a2007-11-21 04:36:41 +08002710
Zhang Xiantaof77bc6a2007-11-21 04:36:41 +08002711 return 0;
2712}
2713
Jan Kiszkac4abb7c2008-09-26 09:30:55 +02002714static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2715{
Jan Kiszkac4abb7c2008-09-26 09:30:55 +02002716 kvm_inject_nmi(vcpu);
Jan Kiszkac4abb7c2008-09-26 09:30:55 +02002717
2718 return 0;
2719}
2720
Avi Kivityb209749f2007-10-22 16:50:39 +02002721static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2722 struct kvm_tpr_access_ctl *tac)
2723{
2724 if (tac->flags)
2725 return -EINVAL;
2726 vcpu->arch.tpr_access_reporting = !!tac->enabled;
2727 return 0;
2728}
2729
Huang Ying890ca9a2009-05-11 16:48:15 +08002730static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2731 u64 mcg_cap)
2732{
2733 int r;
2734 unsigned bank_num = mcg_cap & 0xff, bank;
2735
2736 r = -EINVAL;
Jan Kiszkaa9e38c3e2009-10-23 09:37:00 +02002737 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
Huang Ying890ca9a2009-05-11 16:48:15 +08002738 goto out;
2739 if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2740 goto out;
2741 r = 0;
2742 vcpu->arch.mcg_cap = mcg_cap;
2743 /* Init IA32_MCG_CTL to all 1s */
2744 if (mcg_cap & MCG_CTL_P)
2745 vcpu->arch.mcg_ctl = ~(u64)0;
2746 /* Init IA32_MCi_CTL to all 1s */
2747 for (bank = 0; bank < bank_num; bank++)
2748 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2749out:
2750 return r;
2751}
2752
2753static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2754 struct kvm_x86_mce *mce)
2755{
2756 u64 mcg_cap = vcpu->arch.mcg_cap;
2757 unsigned bank_num = mcg_cap & 0xff;
2758 u64 *banks = vcpu->arch.mce_banks;
2759
2760 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2761 return -EINVAL;
2762 /*
2763 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2764 * reporting is disabled
2765 */
2766 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2767 vcpu->arch.mcg_ctl != ~(u64)0)
2768 return 0;
2769 banks += 4 * mce->bank;
2770 /*
2771 * if IA32_MCi_CTL is not all 1s, the uncorrected error
2772 * reporting is disabled for the bank
2773 */
2774 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2775 return 0;
2776 if (mce->status & MCI_STATUS_UC) {
2777 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
Avi Kivityfc78f512009-12-07 12:16:48 +02002778 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
Avi Kivitya8eeb042010-05-10 12:34:53 +03002779 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
Huang Ying890ca9a2009-05-11 16:48:15 +08002780 return 0;
2781 }
2782 if (banks[1] & MCI_STATUS_VAL)
2783 mce->status |= MCI_STATUS_OVER;
2784 banks[2] = mce->addr;
2785 banks[3] = mce->misc;
2786 vcpu->arch.mcg_status = mce->mcg_status;
2787 banks[1] = mce->status;
2788 kvm_queue_exception(vcpu, MC_VECTOR);
2789 } else if (!(banks[1] & MCI_STATUS_VAL)
2790 || !(banks[1] & MCI_STATUS_UC)) {
2791 if (banks[1] & MCI_STATUS_VAL)
2792 mce->status |= MCI_STATUS_OVER;
2793 banks[2] = mce->addr;
2794 banks[3] = mce->misc;
2795 banks[1] = mce->status;
2796 } else
2797 banks[1] |= MCI_STATUS_OVER;
2798 return 0;
2799}
2800
Jan Kiszka3cfc3092009-11-12 01:04:25 +01002801static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2802 struct kvm_vcpu_events *events)
2803{
Avi Kivity7460fb4a2011-09-20 13:43:14 +03002804 process_nmi(vcpu);
Jan Kiszka03b82a32010-02-15 10:45:41 +01002805 events->exception.injected =
2806 vcpu->arch.exception.pending &&
2807 !kvm_exception_is_soft(vcpu->arch.exception.nr);
Jan Kiszka3cfc3092009-11-12 01:04:25 +01002808 events->exception.nr = vcpu->arch.exception.nr;
2809 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
Vasiliy Kulikov97e69aa2010-10-30 22:54:47 +04002810 events->exception.pad = 0;
Jan Kiszka3cfc3092009-11-12 01:04:25 +01002811 events->exception.error_code = vcpu->arch.exception.error_code;
2812
Jan Kiszka03b82a32010-02-15 10:45:41 +01002813 events->interrupt.injected =
2814 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
Jan Kiszka3cfc3092009-11-12 01:04:25 +01002815 events->interrupt.nr = vcpu->arch.interrupt.nr;
Jan Kiszka03b82a32010-02-15 10:45:41 +01002816 events->interrupt.soft = 0;
Jan Kiszka48005f62010-02-19 19:38:07 +01002817 events->interrupt.shadow =
2818 kvm_x86_ops->get_interrupt_shadow(vcpu,
2819 KVM_X86_SHADOW_INT_MOV_SS | KVM_X86_SHADOW_INT_STI);
Jan Kiszka3cfc3092009-11-12 01:04:25 +01002820
2821 events->nmi.injected = vcpu->arch.nmi_injected;
Avi Kivity7460fb4a2011-09-20 13:43:14 +03002822 events->nmi.pending = vcpu->arch.nmi_pending != 0;
Jan Kiszka3cfc3092009-11-12 01:04:25 +01002823 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
Vasiliy Kulikov97e69aa2010-10-30 22:54:47 +04002824 events->nmi.pad = 0;
Jan Kiszka3cfc3092009-11-12 01:04:25 +01002825
2826 events->sipi_vector = vcpu->arch.sipi_vector;
2827
Jan Kiszkadab4b912009-12-06 18:24:15 +01002828 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
Jan Kiszka48005f62010-02-19 19:38:07 +01002829 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2830 | KVM_VCPUEVENT_VALID_SHADOW);
Vasiliy Kulikov97e69aa2010-10-30 22:54:47 +04002831 memset(&events->reserved, 0, sizeof(events->reserved));
Jan Kiszka3cfc3092009-11-12 01:04:25 +01002832}
2833
2834static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2835 struct kvm_vcpu_events *events)
2836{
Jan Kiszkadab4b912009-12-06 18:24:15 +01002837 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
Jan Kiszka48005f62010-02-19 19:38:07 +01002838 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2839 | KVM_VCPUEVENT_VALID_SHADOW))
Jan Kiszka3cfc3092009-11-12 01:04:25 +01002840 return -EINVAL;
2841
Avi Kivity7460fb4a2011-09-20 13:43:14 +03002842 process_nmi(vcpu);
Jan Kiszka3cfc3092009-11-12 01:04:25 +01002843 vcpu->arch.exception.pending = events->exception.injected;
2844 vcpu->arch.exception.nr = events->exception.nr;
2845 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2846 vcpu->arch.exception.error_code = events->exception.error_code;
2847
2848 vcpu->arch.interrupt.pending = events->interrupt.injected;
2849 vcpu->arch.interrupt.nr = events->interrupt.nr;
2850 vcpu->arch.interrupt.soft = events->interrupt.soft;
Jan Kiszka48005f62010-02-19 19:38:07 +01002851 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2852 kvm_x86_ops->set_interrupt_shadow(vcpu,
2853 events->interrupt.shadow);
Jan Kiszka3cfc3092009-11-12 01:04:25 +01002854
2855 vcpu->arch.nmi_injected = events->nmi.injected;
Jan Kiszkadab4b912009-12-06 18:24:15 +01002856 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2857 vcpu->arch.nmi_pending = events->nmi.pending;
Jan Kiszka3cfc3092009-11-12 01:04:25 +01002858 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2859
Jan Kiszkadab4b912009-12-06 18:24:15 +01002860 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR)
2861 vcpu->arch.sipi_vector = events->sipi_vector;
Jan Kiszka3cfc3092009-11-12 01:04:25 +01002862
Avi Kivity3842d132010-07-27 12:30:24 +03002863 kvm_make_request(KVM_REQ_EVENT, vcpu);
2864
Jan Kiszka3cfc3092009-11-12 01:04:25 +01002865 return 0;
2866}
2867
Jan Kiszkaa1efbe72010-02-15 10:45:43 +01002868static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
2869 struct kvm_debugregs *dbgregs)
2870{
Jan Kiszkaa1efbe72010-02-15 10:45:43 +01002871 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
2872 dbgregs->dr6 = vcpu->arch.dr6;
2873 dbgregs->dr7 = vcpu->arch.dr7;
2874 dbgregs->flags = 0;
Vasiliy Kulikov97e69aa2010-10-30 22:54:47 +04002875 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
Jan Kiszkaa1efbe72010-02-15 10:45:43 +01002876}
2877
2878static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
2879 struct kvm_debugregs *dbgregs)
2880{
2881 if (dbgregs->flags)
2882 return -EINVAL;
2883
Jan Kiszkaa1efbe72010-02-15 10:45:43 +01002884 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
2885 vcpu->arch.dr6 = dbgregs->dr6;
2886 vcpu->arch.dr7 = dbgregs->dr7;
2887
Jan Kiszkaa1efbe72010-02-15 10:45:43 +01002888 return 0;
2889}
2890
Sheng Yang2d5b5a62010-06-13 17:29:39 +08002891static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
2892 struct kvm_xsave *guest_xsave)
2893{
2894 if (cpu_has_xsave)
2895 memcpy(guest_xsave->region,
2896 &vcpu->arch.guest_fpu.state->xsave,
Xiaotian Fengf45755b2010-08-13 15:19:11 +08002897 xstate_size);
Sheng Yang2d5b5a62010-06-13 17:29:39 +08002898 else {
2899 memcpy(guest_xsave->region,
2900 &vcpu->arch.guest_fpu.state->fxsave,
2901 sizeof(struct i387_fxsave_struct));
2902 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
2903 XSTATE_FPSSE;
2904 }
2905}
2906
2907static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
2908 struct kvm_xsave *guest_xsave)
2909{
2910 u64 xstate_bv =
2911 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
2912
2913 if (cpu_has_xsave)
2914 memcpy(&vcpu->arch.guest_fpu.state->xsave,
Xiaotian Fengf45755b2010-08-13 15:19:11 +08002915 guest_xsave->region, xstate_size);
Sheng Yang2d5b5a62010-06-13 17:29:39 +08002916 else {
2917 if (xstate_bv & ~XSTATE_FPSSE)
2918 return -EINVAL;
2919 memcpy(&vcpu->arch.guest_fpu.state->fxsave,
2920 guest_xsave->region, sizeof(struct i387_fxsave_struct));
2921 }
2922 return 0;
2923}
2924
2925static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
2926 struct kvm_xcrs *guest_xcrs)
2927{
2928 if (!cpu_has_xsave) {
2929 guest_xcrs->nr_xcrs = 0;
2930 return;
2931 }
2932
2933 guest_xcrs->nr_xcrs = 1;
2934 guest_xcrs->flags = 0;
2935 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
2936 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
2937}
2938
2939static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
2940 struct kvm_xcrs *guest_xcrs)
2941{
2942 int i, r = 0;
2943
2944 if (!cpu_has_xsave)
2945 return -EINVAL;
2946
2947 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
2948 return -EINVAL;
2949
2950 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
2951 /* Only support XCR0 currently */
2952 if (guest_xcrs->xcrs[0].xcr == XCR_XFEATURE_ENABLED_MASK) {
2953 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
2954 guest_xcrs->xcrs[0].value);
2955 break;
2956 }
2957 if (r)
2958 r = -EINVAL;
2959 return r;
2960}
2961
Eric B Munson1c0b28c2012-03-10 14:37:27 -05002962/*
2963 * kvm_set_guest_paused() indicates to the guest kernel that it has been
2964 * stopped by the hypervisor. This function will be called from the host only.
2965 * EINVAL is returned when the host attempts to set the flag for a guest that
2966 * does not support pv clocks.
2967 */
2968static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
2969{
Eric B Munson1c0b28c2012-03-10 14:37:27 -05002970 if (!vcpu->arch.time_page)
2971 return -EINVAL;
Marcelo Tosatti51d59c62012-08-03 15:57:49 -03002972 vcpu->arch.pvclock_set_guest_stopped_request = true;
Eric B Munson1c0b28c2012-03-10 14:37:27 -05002973 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2974 return 0;
2975}
2976
Carsten Otte313a3dc2007-10-11 19:16:52 +02002977long kvm_arch_vcpu_ioctl(struct file *filp,
2978 unsigned int ioctl, unsigned long arg)
2979{
2980 struct kvm_vcpu *vcpu = filp->private_data;
2981 void __user *argp = (void __user *)arg;
2982 int r;
Avi Kivityd1ac91d2010-06-20 15:54:43 +03002983 union {
2984 struct kvm_lapic_state *lapic;
2985 struct kvm_xsave *xsave;
2986 struct kvm_xcrs *xcrs;
2987 void *buffer;
2988 } u;
Carsten Otte313a3dc2007-10-11 19:16:52 +02002989
Avi Kivityd1ac91d2010-06-20 15:54:43 +03002990 u.buffer = NULL;
Carsten Otte313a3dc2007-10-11 19:16:52 +02002991 switch (ioctl) {
2992 case KVM_GET_LAPIC: {
Marcelo Tosatti2204ae32009-10-29 13:44:16 -02002993 r = -EINVAL;
2994 if (!vcpu->arch.apic)
2995 goto out;
Avi Kivityd1ac91d2010-06-20 15:54:43 +03002996 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
Carsten Otte313a3dc2007-10-11 19:16:52 +02002997
Dave Hansenb772ff32008-08-11 10:01:47 -07002998 r = -ENOMEM;
Avi Kivityd1ac91d2010-06-20 15:54:43 +03002999 if (!u.lapic)
Dave Hansenb772ff32008-08-11 10:01:47 -07003000 goto out;
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003001 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
Carsten Otte313a3dc2007-10-11 19:16:52 +02003002 if (r)
3003 goto out;
3004 r = -EFAULT;
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003005 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
Carsten Otte313a3dc2007-10-11 19:16:52 +02003006 goto out;
3007 r = 0;
3008 break;
3009 }
3010 case KVM_SET_LAPIC: {
Marcelo Tosatti2204ae32009-10-29 13:44:16 -02003011 r = -EINVAL;
3012 if (!vcpu->arch.apic)
3013 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02003014 u.lapic = memdup_user(argp, sizeof(*u.lapic));
Guo Chao18595412012-11-02 18:33:21 +08003015 if (IS_ERR(u.lapic))
3016 return PTR_ERR(u.lapic);
Sasha Levinff5c2c02011-12-04 19:36:29 +02003017
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003018 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
Carsten Otte313a3dc2007-10-11 19:16:52 +02003019 break;
3020 }
Zhang Xiantaof77bc6a2007-11-21 04:36:41 +08003021 case KVM_INTERRUPT: {
3022 struct kvm_interrupt irq;
3023
3024 r = -EFAULT;
3025 if (copy_from_user(&irq, argp, sizeof irq))
3026 goto out;
3027 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
Zhang Xiantaof77bc6a2007-11-21 04:36:41 +08003028 break;
3029 }
Jan Kiszkac4abb7c2008-09-26 09:30:55 +02003030 case KVM_NMI: {
3031 r = kvm_vcpu_ioctl_nmi(vcpu);
Jan Kiszkac4abb7c2008-09-26 09:30:55 +02003032 break;
3033 }
Carsten Otte313a3dc2007-10-11 19:16:52 +02003034 case KVM_SET_CPUID: {
3035 struct kvm_cpuid __user *cpuid_arg = argp;
3036 struct kvm_cpuid cpuid;
3037
3038 r = -EFAULT;
3039 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3040 goto out;
3041 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
Carsten Otte313a3dc2007-10-11 19:16:52 +02003042 break;
3043 }
Dan Kenigsberg07716712007-11-21 17:10:04 +02003044 case KVM_SET_CPUID2: {
3045 struct kvm_cpuid2 __user *cpuid_arg = argp;
3046 struct kvm_cpuid2 cpuid;
3047
3048 r = -EFAULT;
3049 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3050 goto out;
3051 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
Amit Shah19355472009-01-14 16:56:00 +00003052 cpuid_arg->entries);
Dan Kenigsberg07716712007-11-21 17:10:04 +02003053 break;
3054 }
3055 case KVM_GET_CPUID2: {
3056 struct kvm_cpuid2 __user *cpuid_arg = argp;
3057 struct kvm_cpuid2 cpuid;
3058
3059 r = -EFAULT;
3060 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3061 goto out;
3062 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
Amit Shah19355472009-01-14 16:56:00 +00003063 cpuid_arg->entries);
Dan Kenigsberg07716712007-11-21 17:10:04 +02003064 if (r)
3065 goto out;
3066 r = -EFAULT;
3067 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3068 goto out;
3069 r = 0;
3070 break;
3071 }
Carsten Otte313a3dc2007-10-11 19:16:52 +02003072 case KVM_GET_MSRS:
3073 r = msr_io(vcpu, argp, kvm_get_msr, 1);
3074 break;
3075 case KVM_SET_MSRS:
3076 r = msr_io(vcpu, argp, do_set_msr, 0);
3077 break;
Avi Kivityb209749f2007-10-22 16:50:39 +02003078 case KVM_TPR_ACCESS_REPORTING: {
3079 struct kvm_tpr_access_ctl tac;
3080
3081 r = -EFAULT;
3082 if (copy_from_user(&tac, argp, sizeof tac))
3083 goto out;
3084 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3085 if (r)
3086 goto out;
3087 r = -EFAULT;
3088 if (copy_to_user(argp, &tac, sizeof tac))
3089 goto out;
3090 r = 0;
3091 break;
3092 };
Avi Kivityb93463a2007-10-25 16:52:32 +02003093 case KVM_SET_VAPIC_ADDR: {
3094 struct kvm_vapic_addr va;
3095
3096 r = -EINVAL;
3097 if (!irqchip_in_kernel(vcpu->kvm))
3098 goto out;
3099 r = -EFAULT;
3100 if (copy_from_user(&va, argp, sizeof va))
3101 goto out;
3102 r = 0;
3103 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3104 break;
3105 }
Huang Ying890ca9a2009-05-11 16:48:15 +08003106 case KVM_X86_SETUP_MCE: {
3107 u64 mcg_cap;
3108
3109 r = -EFAULT;
3110 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3111 goto out;
3112 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3113 break;
3114 }
3115 case KVM_X86_SET_MCE: {
3116 struct kvm_x86_mce mce;
3117
3118 r = -EFAULT;
3119 if (copy_from_user(&mce, argp, sizeof mce))
3120 goto out;
3121 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3122 break;
3123 }
Jan Kiszka3cfc3092009-11-12 01:04:25 +01003124 case KVM_GET_VCPU_EVENTS: {
3125 struct kvm_vcpu_events events;
3126
3127 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3128
3129 r = -EFAULT;
3130 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3131 break;
3132 r = 0;
3133 break;
3134 }
3135 case KVM_SET_VCPU_EVENTS: {
3136 struct kvm_vcpu_events events;
3137
3138 r = -EFAULT;
3139 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3140 break;
3141
3142 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3143 break;
3144 }
Jan Kiszkaa1efbe72010-02-15 10:45:43 +01003145 case KVM_GET_DEBUGREGS: {
3146 struct kvm_debugregs dbgregs;
3147
3148 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3149
3150 r = -EFAULT;
3151 if (copy_to_user(argp, &dbgregs,
3152 sizeof(struct kvm_debugregs)))
3153 break;
3154 r = 0;
3155 break;
3156 }
3157 case KVM_SET_DEBUGREGS: {
3158 struct kvm_debugregs dbgregs;
3159
3160 r = -EFAULT;
3161 if (copy_from_user(&dbgregs, argp,
3162 sizeof(struct kvm_debugregs)))
3163 break;
3164
3165 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3166 break;
3167 }
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003168 case KVM_GET_XSAVE: {
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003169 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003170 r = -ENOMEM;
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003171 if (!u.xsave)
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003172 break;
3173
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003174 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003175
3176 r = -EFAULT;
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003177 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003178 break;
3179 r = 0;
3180 break;
3181 }
3182 case KVM_SET_XSAVE: {
Sasha Levinff5c2c02011-12-04 19:36:29 +02003183 u.xsave = memdup_user(argp, sizeof(*u.xsave));
Guo Chao18595412012-11-02 18:33:21 +08003184 if (IS_ERR(u.xsave))
3185 return PTR_ERR(u.xsave);
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003186
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003187 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003188 break;
3189 }
3190 case KVM_GET_XCRS: {
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003191 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003192 r = -ENOMEM;
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003193 if (!u.xcrs)
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003194 break;
3195
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003196 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003197
3198 r = -EFAULT;
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003199 if (copy_to_user(argp, u.xcrs,
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003200 sizeof(struct kvm_xcrs)))
3201 break;
3202 r = 0;
3203 break;
3204 }
3205 case KVM_SET_XCRS: {
Sasha Levinff5c2c02011-12-04 19:36:29 +02003206 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
Guo Chao18595412012-11-02 18:33:21 +08003207 if (IS_ERR(u.xcrs))
3208 return PTR_ERR(u.xcrs);
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003209
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003210 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
Sheng Yang2d5b5a62010-06-13 17:29:39 +08003211 break;
3212 }
Joerg Roedel92a1f122011-03-25 09:44:51 +01003213 case KVM_SET_TSC_KHZ: {
3214 u32 user_tsc_khz;
3215
3216 r = -EINVAL;
Joerg Roedel92a1f122011-03-25 09:44:51 +01003217 user_tsc_khz = (u32)arg;
3218
3219 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3220 goto out;
3221
Zachary Amsdencc578282012-02-03 15:43:50 -02003222 if (user_tsc_khz == 0)
3223 user_tsc_khz = tsc_khz;
3224
3225 kvm_set_tsc_khz(vcpu, user_tsc_khz);
Joerg Roedel92a1f122011-03-25 09:44:51 +01003226
3227 r = 0;
3228 goto out;
3229 }
3230 case KVM_GET_TSC_KHZ: {
Zachary Amsdencc578282012-02-03 15:43:50 -02003231 r = vcpu->arch.virtual_tsc_khz;
Joerg Roedel92a1f122011-03-25 09:44:51 +01003232 goto out;
3233 }
Eric B Munson1c0b28c2012-03-10 14:37:27 -05003234 case KVM_KVMCLOCK_CTRL: {
3235 r = kvm_set_guest_paused(vcpu);
3236 goto out;
3237 }
Carsten Otte313a3dc2007-10-11 19:16:52 +02003238 default:
3239 r = -EINVAL;
3240 }
3241out:
Avi Kivityd1ac91d2010-06-20 15:54:43 +03003242 kfree(u.buffer);
Carsten Otte313a3dc2007-10-11 19:16:52 +02003243 return r;
3244}
3245
Carsten Otte5b1c1492012-01-04 10:25:23 +01003246int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3247{
3248 return VM_FAULT_SIGBUS;
3249}
3250
Carsten Otte1fe779f2007-10-29 16:08:35 +01003251static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3252{
3253 int ret;
3254
3255 if (addr > (unsigned int)(-3 * PAGE_SIZE))
Guo Chao951179c2012-11-02 18:33:22 +08003256 return -EINVAL;
Carsten Otte1fe779f2007-10-29 16:08:35 +01003257 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3258 return ret;
3259}
3260
Sheng Yangb927a3c2009-07-21 10:42:48 +08003261static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3262 u64 ident_addr)
3263{
3264 kvm->arch.ept_identity_map_addr = ident_addr;
3265 return 0;
3266}
3267
Carsten Otte1fe779f2007-10-29 16:08:35 +01003268static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3269 u32 kvm_nr_mmu_pages)
3270{
3271 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3272 return -EINVAL;
3273
Marcelo Tosatti79fac952009-12-23 14:35:26 -02003274 mutex_lock(&kvm->slots_lock);
Carsten Otte1fe779f2007-10-29 16:08:35 +01003275
3276 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
Zhang Xiantaof05e70a2007-12-14 10:01:48 +08003277 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
Carsten Otte1fe779f2007-10-29 16:08:35 +01003278
Marcelo Tosatti79fac952009-12-23 14:35:26 -02003279 mutex_unlock(&kvm->slots_lock);
Carsten Otte1fe779f2007-10-29 16:08:35 +01003280 return 0;
3281}
3282
3283static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3284{
Dave Hansen39de71e2010-08-19 18:11:14 -07003285 return kvm->arch.n_max_mmu_pages;
Carsten Otte1fe779f2007-10-29 16:08:35 +01003286}
3287
Carsten Otte1fe779f2007-10-29 16:08:35 +01003288static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3289{
3290 int r;
3291
3292 r = 0;
3293 switch (chip->chip_id) {
3294 case KVM_IRQCHIP_PIC_MASTER:
3295 memcpy(&chip->chip.pic,
3296 &pic_irqchip(kvm)->pics[0],
3297 sizeof(struct kvm_pic_state));
3298 break;
3299 case KVM_IRQCHIP_PIC_SLAVE:
3300 memcpy(&chip->chip.pic,
3301 &pic_irqchip(kvm)->pics[1],
3302 sizeof(struct kvm_pic_state));
3303 break;
3304 case KVM_IRQCHIP_IOAPIC:
Gleb Natapoveba02262009-08-24 11:54:25 +03003305 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
Carsten Otte1fe779f2007-10-29 16:08:35 +01003306 break;
3307 default:
3308 r = -EINVAL;
3309 break;
3310 }
3311 return r;
3312}
3313
3314static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3315{
3316 int r;
3317
3318 r = 0;
3319 switch (chip->chip_id) {
3320 case KVM_IRQCHIP_PIC_MASTER:
Avi Kivityf4f51052010-09-19 18:44:07 +02003321 spin_lock(&pic_irqchip(kvm)->lock);
Carsten Otte1fe779f2007-10-29 16:08:35 +01003322 memcpy(&pic_irqchip(kvm)->pics[0],
3323 &chip->chip.pic,
3324 sizeof(struct kvm_pic_state));
Avi Kivityf4f51052010-09-19 18:44:07 +02003325 spin_unlock(&pic_irqchip(kvm)->lock);
Carsten Otte1fe779f2007-10-29 16:08:35 +01003326 break;
3327 case KVM_IRQCHIP_PIC_SLAVE:
Avi Kivityf4f51052010-09-19 18:44:07 +02003328 spin_lock(&pic_irqchip(kvm)->lock);
Carsten Otte1fe779f2007-10-29 16:08:35 +01003329 memcpy(&pic_irqchip(kvm)->pics[1],
3330 &chip->chip.pic,
3331 sizeof(struct kvm_pic_state));
Avi Kivityf4f51052010-09-19 18:44:07 +02003332 spin_unlock(&pic_irqchip(kvm)->lock);
Carsten Otte1fe779f2007-10-29 16:08:35 +01003333 break;
3334 case KVM_IRQCHIP_IOAPIC:
Gleb Natapoveba02262009-08-24 11:54:25 +03003335 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
Carsten Otte1fe779f2007-10-29 16:08:35 +01003336 break;
3337 default:
3338 r = -EINVAL;
3339 break;
3340 }
3341 kvm_pic_update_irq(pic_irqchip(kvm));
3342 return r;
3343}
3344
Sheng Yange0f63cb2008-03-04 00:50:59 +08003345static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3346{
3347 int r = 0;
3348
Marcelo Tosatti894a9c52009-06-23 15:05:14 -03003349 mutex_lock(&kvm->arch.vpit->pit_state.lock);
Sheng Yange0f63cb2008-03-04 00:50:59 +08003350 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
Marcelo Tosatti894a9c52009-06-23 15:05:14 -03003351 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
Sheng Yange0f63cb2008-03-04 00:50:59 +08003352 return r;
3353}
3354
3355static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3356{
3357 int r = 0;
3358
Marcelo Tosatti894a9c52009-06-23 15:05:14 -03003359 mutex_lock(&kvm->arch.vpit->pit_state.lock);
Sheng Yange0f63cb2008-03-04 00:50:59 +08003360 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
Beth Kone9f42752009-07-07 11:50:38 -04003361 kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
3362 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3363 return r;
3364}
3365
3366static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3367{
3368 int r = 0;
3369
3370 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3371 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3372 sizeof(ps->channels));
3373 ps->flags = kvm->arch.vpit->pit_state.flags;
3374 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
Vasiliy Kulikov97e69aa2010-10-30 22:54:47 +04003375 memset(&ps->reserved, 0, sizeof(ps->reserved));
Beth Kone9f42752009-07-07 11:50:38 -04003376 return r;
3377}
3378
3379static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3380{
3381 int r = 0, start = 0;
3382 u32 prev_legacy, cur_legacy;
3383 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3384 prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3385 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3386 if (!prev_legacy && cur_legacy)
3387 start = 1;
3388 memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3389 sizeof(kvm->arch.vpit->pit_state.channels));
3390 kvm->arch.vpit->pit_state.flags = ps->flags;
3391 kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
Marcelo Tosatti894a9c52009-06-23 15:05:14 -03003392 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
Sheng Yange0f63cb2008-03-04 00:50:59 +08003393 return r;
3394}
3395
Marcelo Tosatti52d939a2008-12-30 15:55:06 -02003396static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3397 struct kvm_reinject_control *control)
3398{
3399 if (!kvm->arch.vpit)
3400 return -ENXIO;
Marcelo Tosatti894a9c52009-06-23 15:05:14 -03003401 mutex_lock(&kvm->arch.vpit->pit_state.lock);
Avi Kivity26ef1922012-07-26 18:01:53 +03003402 kvm->arch.vpit->pit_state.reinject = control->pit_reinject;
Marcelo Tosatti894a9c52009-06-23 15:05:14 -03003403 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
Marcelo Tosatti52d939a2008-12-30 15:55:06 -02003404 return 0;
3405}
3406
Takuya Yoshikawa95d4c162011-11-14 18:24:50 +09003407/**
Takuya Yoshikawa60c34612012-03-03 14:21:48 +09003408 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3409 * @kvm: kvm instance
3410 * @log: slot id and address to which we copy the log
Takuya Yoshikawa95d4c162011-11-14 18:24:50 +09003411 *
Takuya Yoshikawa60c34612012-03-03 14:21:48 +09003412 * We need to keep it in mind that VCPU threads can write to the bitmap
3413 * concurrently. So, to avoid losing data, we keep the following order for
3414 * each bit:
Takuya Yoshikawa95d4c162011-11-14 18:24:50 +09003415 *
Takuya Yoshikawa60c34612012-03-03 14:21:48 +09003416 * 1. Take a snapshot of the bit and clear it if needed.
3417 * 2. Write protect the corresponding page.
3418 * 3. Flush TLB's if needed.
3419 * 4. Copy the snapshot to the userspace.
Takuya Yoshikawa95d4c162011-11-14 18:24:50 +09003420 *
Takuya Yoshikawa60c34612012-03-03 14:21:48 +09003421 * Between 2 and 3, the guest may write to the page using the remaining TLB
3422 * entry. This is not a problem because the page will be reported dirty at
3423 * step 4 using the snapshot taken before and step 3 ensures that successive
3424 * writes will be logged for the next call.
Takuya Yoshikawa95d4c162011-11-14 18:24:50 +09003425 */
Takuya Yoshikawa60c34612012-03-03 14:21:48 +09003426int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
Zhang Xiantao5bb064d2007-11-18 20:29:43 +08003427{
Takuya Yoshikawa7850ac52011-11-14 18:23:34 +09003428 int r;
Zhang Xiantao5bb064d2007-11-18 20:29:43 +08003429 struct kvm_memory_slot *memslot;
Takuya Yoshikawa60c34612012-03-03 14:21:48 +09003430 unsigned long n, i;
3431 unsigned long *dirty_bitmap;
3432 unsigned long *dirty_bitmap_buffer;
3433 bool is_dirty = false;
Zhang Xiantao5bb064d2007-11-18 20:29:43 +08003434
Marcelo Tosatti79fac952009-12-23 14:35:26 -02003435 mutex_lock(&kvm->slots_lock);
Zhang Xiantao5bb064d2007-11-18 20:29:43 +08003436
Marcelo Tosattib050b012009-12-23 14:35:22 -02003437 r = -EINVAL;
Alex Williamsonbbacc0c2012-12-10 10:33:09 -07003438 if (log->slot >= KVM_USER_MEM_SLOTS)
Zhang Xiantao5bb064d2007-11-18 20:29:43 +08003439 goto out;
3440
Xiao Guangrong28a37542011-11-24 19:04:35 +08003441 memslot = id_to_memslot(kvm->memslots, log->slot);
Takuya Yoshikawa60c34612012-03-03 14:21:48 +09003442
3443 dirty_bitmap = memslot->dirty_bitmap;
Marcelo Tosattib050b012009-12-23 14:35:22 -02003444 r = -ENOENT;
Takuya Yoshikawa60c34612012-03-03 14:21:48 +09003445 if (!dirty_bitmap)
Marcelo Tosattib050b012009-12-23 14:35:22 -02003446 goto out;
3447
Takuya Yoshikawa87bf6e72010-04-12 19:35:35 +09003448 n = kvm_dirty_bitmap_bytes(memslot);
Marcelo Tosattib050b012009-12-23 14:35:22 -02003449
Takuya Yoshikawa60c34612012-03-03 14:21:48 +09003450 dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
3451 memset(dirty_bitmap_buffer, 0, n);
Marcelo Tosattib050b012009-12-23 14:35:22 -02003452
Takuya Yoshikawa60c34612012-03-03 14:21:48 +09003453 spin_lock(&kvm->mmu_lock);
Marcelo Tosattib050b012009-12-23 14:35:22 -02003454
Takuya Yoshikawa60c34612012-03-03 14:21:48 +09003455 for (i = 0; i < n / sizeof(long); i++) {
3456 unsigned long mask;
3457 gfn_t offset;
Sasha Levincdfca7b2011-12-04 19:36:28 +02003458
Takuya Yoshikawa60c34612012-03-03 14:21:48 +09003459 if (!dirty_bitmap[i])
3460 continue;
Marcelo Tosattib050b012009-12-23 14:35:22 -02003461
Takuya Yoshikawa60c34612012-03-03 14:21:48 +09003462 is_dirty = true;
Takuya Yoshikawa914ebcc2010-04-28 18:50:36 +09003463
Takuya Yoshikawa60c34612012-03-03 14:21:48 +09003464 mask = xchg(&dirty_bitmap[i], 0);
3465 dirty_bitmap_buffer[i] = mask;
Michael S. Tsirkinedde99c2010-10-25 03:21:24 +02003466
Takuya Yoshikawa60c34612012-03-03 14:21:48 +09003467 offset = i * BITS_PER_LONG;
3468 kvm_mmu_write_protect_pt_masked(kvm, memslot, offset, mask);
Zhang Xiantao5bb064d2007-11-18 20:29:43 +08003469 }
Takuya Yoshikawa60c34612012-03-03 14:21:48 +09003470 if (is_dirty)
3471 kvm_flush_remote_tlbs(kvm);
3472
3473 spin_unlock(&kvm->mmu_lock);
3474
3475 r = -EFAULT;
3476 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
3477 goto out;
Marcelo Tosattib050b012009-12-23 14:35:22 -02003478
Zhang Xiantao5bb064d2007-11-18 20:29:43 +08003479 r = 0;
3480out:
Marcelo Tosatti79fac952009-12-23 14:35:26 -02003481 mutex_unlock(&kvm->slots_lock);
Zhang Xiantao5bb064d2007-11-18 20:29:43 +08003482 return r;
3483}
3484
Christoffer Dall23d43cf2012-07-24 08:51:20 -04003485int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event)
3486{
3487 if (!irqchip_in_kernel(kvm))
3488 return -ENXIO;
3489
3490 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3491 irq_event->irq, irq_event->level);
3492 return 0;
3493}
3494
Carsten Otte1fe779f2007-10-29 16:08:35 +01003495long kvm_arch_vm_ioctl(struct file *filp,
3496 unsigned int ioctl, unsigned long arg)
3497{
3498 struct kvm *kvm = filp->private_data;
3499 void __user *argp = (void __user *)arg;
Avi Kivity367e1312009-08-26 14:57:07 +03003500 int r = -ENOTTY;
Dave Hansenf0d66272008-08-11 10:01:45 -07003501 /*
3502 * This union makes it completely explicit to gcc-3.x
3503 * that these two variables' stack usage should be
3504 * combined, not added together.
3505 */
3506 union {
3507 struct kvm_pit_state ps;
Beth Kone9f42752009-07-07 11:50:38 -04003508 struct kvm_pit_state2 ps2;
Jan Kiszkac5ff41c2009-05-14 22:42:53 +02003509 struct kvm_pit_config pit_config;
Dave Hansenf0d66272008-08-11 10:01:45 -07003510 } u;
Carsten Otte1fe779f2007-10-29 16:08:35 +01003511
3512 switch (ioctl) {
3513 case KVM_SET_TSS_ADDR:
3514 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
Carsten Otte1fe779f2007-10-29 16:08:35 +01003515 break;
Sheng Yangb927a3c2009-07-21 10:42:48 +08003516 case KVM_SET_IDENTITY_MAP_ADDR: {
3517 u64 ident_addr;
3518
3519 r = -EFAULT;
3520 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3521 goto out;
3522 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
Sheng Yangb927a3c2009-07-21 10:42:48 +08003523 break;
3524 }
Carsten Otte1fe779f2007-10-29 16:08:35 +01003525 case KVM_SET_NR_MMU_PAGES:
3526 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
Carsten Otte1fe779f2007-10-29 16:08:35 +01003527 break;
3528 case KVM_GET_NR_MMU_PAGES:
3529 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3530 break;
Marcelo Tosatti3ddea122009-10-29 13:44:15 -02003531 case KVM_CREATE_IRQCHIP: {
3532 struct kvm_pic *vpic;
3533
3534 mutex_lock(&kvm->lock);
3535 r = -EEXIST;
3536 if (kvm->arch.vpic)
3537 goto create_irqchip_unlock;
Avi Kivity3e515702012-03-05 14:23:29 +02003538 r = -EINVAL;
3539 if (atomic_read(&kvm->online_vcpus))
3540 goto create_irqchip_unlock;
Carsten Otte1fe779f2007-10-29 16:08:35 +01003541 r = -ENOMEM;
Marcelo Tosatti3ddea122009-10-29 13:44:15 -02003542 vpic = kvm_create_pic(kvm);
3543 if (vpic) {
Carsten Otte1fe779f2007-10-29 16:08:35 +01003544 r = kvm_ioapic_init(kvm);
3545 if (r) {
Takuya Yoshikawa175504c2010-12-16 01:41:37 +09003546 mutex_lock(&kvm->slots_lock);
Wei Yongjun72bb2fc2010-02-09 10:33:03 +08003547 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
Sasha Levin743eeb02011-07-27 16:00:48 +03003548 &vpic->dev_master);
3549 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3550 &vpic->dev_slave);
3551 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3552 &vpic->dev_eclr);
Takuya Yoshikawa175504c2010-12-16 01:41:37 +09003553 mutex_unlock(&kvm->slots_lock);
Marcelo Tosatti3ddea122009-10-29 13:44:15 -02003554 kfree(vpic);
3555 goto create_irqchip_unlock;
Carsten Otte1fe779f2007-10-29 16:08:35 +01003556 }
3557 } else
Marcelo Tosatti3ddea122009-10-29 13:44:15 -02003558 goto create_irqchip_unlock;
3559 smp_wmb();
3560 kvm->arch.vpic = vpic;
3561 smp_wmb();
Avi Kivity399ec802008-11-19 13:58:46 +02003562 r = kvm_setup_default_irq_routing(kvm);
3563 if (r) {
Takuya Yoshikawa175504c2010-12-16 01:41:37 +09003564 mutex_lock(&kvm->slots_lock);
Marcelo Tosatti3ddea122009-10-29 13:44:15 -02003565 mutex_lock(&kvm->irq_lock);
Wei Yongjun72bb2fc2010-02-09 10:33:03 +08003566 kvm_ioapic_destroy(kvm);
3567 kvm_destroy_pic(kvm);
Marcelo Tosatti3ddea122009-10-29 13:44:15 -02003568 mutex_unlock(&kvm->irq_lock);
Takuya Yoshikawa175504c2010-12-16 01:41:37 +09003569 mutex_unlock(&kvm->slots_lock);
Avi Kivity399ec802008-11-19 13:58:46 +02003570 }
Marcelo Tosatti3ddea122009-10-29 13:44:15 -02003571 create_irqchip_unlock:
3572 mutex_unlock(&kvm->lock);
Carsten Otte1fe779f2007-10-29 16:08:35 +01003573 break;
Marcelo Tosatti3ddea122009-10-29 13:44:15 -02003574 }
Sheng Yang78376992008-01-28 05:10:22 +08003575 case KVM_CREATE_PIT:
Jan Kiszkac5ff41c2009-05-14 22:42:53 +02003576 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3577 goto create_pit;
3578 case KVM_CREATE_PIT2:
3579 r = -EFAULT;
3580 if (copy_from_user(&u.pit_config, argp,
3581 sizeof(struct kvm_pit_config)))
3582 goto out;
3583 create_pit:
Marcelo Tosatti79fac952009-12-23 14:35:26 -02003584 mutex_lock(&kvm->slots_lock);
Avi Kivity269e05e2009-01-05 15:21:42 +02003585 r = -EEXIST;
3586 if (kvm->arch.vpit)
3587 goto create_pit_unlock;
Sheng Yang78376992008-01-28 05:10:22 +08003588 r = -ENOMEM;
Jan Kiszkac5ff41c2009-05-14 22:42:53 +02003589 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
Sheng Yang78376992008-01-28 05:10:22 +08003590 if (kvm->arch.vpit)
3591 r = 0;
Avi Kivity269e05e2009-01-05 15:21:42 +02003592 create_pit_unlock:
Marcelo Tosatti79fac952009-12-23 14:35:26 -02003593 mutex_unlock(&kvm->slots_lock);
Sheng Yang78376992008-01-28 05:10:22 +08003594 break;
Carsten Otte1fe779f2007-10-29 16:08:35 +01003595 case KVM_GET_IRQCHIP: {
3596 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
Sasha Levinff5c2c02011-12-04 19:36:29 +02003597 struct kvm_irqchip *chip;
Carsten Otte1fe779f2007-10-29 16:08:35 +01003598
Sasha Levinff5c2c02011-12-04 19:36:29 +02003599 chip = memdup_user(argp, sizeof(*chip));
3600 if (IS_ERR(chip)) {
3601 r = PTR_ERR(chip);
Carsten Otte1fe779f2007-10-29 16:08:35 +01003602 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02003603 }
3604
Carsten Otte1fe779f2007-10-29 16:08:35 +01003605 r = -ENXIO;
3606 if (!irqchip_in_kernel(kvm))
Dave Hansenf0d66272008-08-11 10:01:45 -07003607 goto get_irqchip_out;
3608 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3609 if (r)
3610 goto get_irqchip_out;
3611 r = -EFAULT;
3612 if (copy_to_user(argp, chip, sizeof *chip))
3613 goto get_irqchip_out;
3614 r = 0;
3615 get_irqchip_out:
3616 kfree(chip);
Carsten Otte1fe779f2007-10-29 16:08:35 +01003617 break;
3618 }
3619 case KVM_SET_IRQCHIP: {
3620 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
Sasha Levinff5c2c02011-12-04 19:36:29 +02003621 struct kvm_irqchip *chip;
Carsten Otte1fe779f2007-10-29 16:08:35 +01003622
Sasha Levinff5c2c02011-12-04 19:36:29 +02003623 chip = memdup_user(argp, sizeof(*chip));
3624 if (IS_ERR(chip)) {
3625 r = PTR_ERR(chip);
Carsten Otte1fe779f2007-10-29 16:08:35 +01003626 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02003627 }
3628
Carsten Otte1fe779f2007-10-29 16:08:35 +01003629 r = -ENXIO;
3630 if (!irqchip_in_kernel(kvm))
Dave Hansenf0d66272008-08-11 10:01:45 -07003631 goto set_irqchip_out;
3632 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3633 if (r)
3634 goto set_irqchip_out;
3635 r = 0;
3636 set_irqchip_out:
3637 kfree(chip);
Carsten Otte1fe779f2007-10-29 16:08:35 +01003638 break;
3639 }
Sheng Yange0f63cb2008-03-04 00:50:59 +08003640 case KVM_GET_PIT: {
Sheng Yange0f63cb2008-03-04 00:50:59 +08003641 r = -EFAULT;
Dave Hansenf0d66272008-08-11 10:01:45 -07003642 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
Sheng Yange0f63cb2008-03-04 00:50:59 +08003643 goto out;
3644 r = -ENXIO;
3645 if (!kvm->arch.vpit)
3646 goto out;
Dave Hansenf0d66272008-08-11 10:01:45 -07003647 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
Sheng Yange0f63cb2008-03-04 00:50:59 +08003648 if (r)
3649 goto out;
3650 r = -EFAULT;
Dave Hansenf0d66272008-08-11 10:01:45 -07003651 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
Sheng Yange0f63cb2008-03-04 00:50:59 +08003652 goto out;
3653 r = 0;
3654 break;
3655 }
3656 case KVM_SET_PIT: {
Sheng Yange0f63cb2008-03-04 00:50:59 +08003657 r = -EFAULT;
Dave Hansenf0d66272008-08-11 10:01:45 -07003658 if (copy_from_user(&u.ps, argp, sizeof u.ps))
Sheng Yange0f63cb2008-03-04 00:50:59 +08003659 goto out;
3660 r = -ENXIO;
3661 if (!kvm->arch.vpit)
3662 goto out;
Dave Hansenf0d66272008-08-11 10:01:45 -07003663 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
Sheng Yange0f63cb2008-03-04 00:50:59 +08003664 break;
3665 }
Beth Kone9f42752009-07-07 11:50:38 -04003666 case KVM_GET_PIT2: {
3667 r = -ENXIO;
3668 if (!kvm->arch.vpit)
3669 goto out;
3670 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3671 if (r)
3672 goto out;
3673 r = -EFAULT;
3674 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3675 goto out;
3676 r = 0;
3677 break;
3678 }
3679 case KVM_SET_PIT2: {
3680 r = -EFAULT;
3681 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3682 goto out;
3683 r = -ENXIO;
3684 if (!kvm->arch.vpit)
3685 goto out;
3686 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
Beth Kone9f42752009-07-07 11:50:38 -04003687 break;
3688 }
Marcelo Tosatti52d939a2008-12-30 15:55:06 -02003689 case KVM_REINJECT_CONTROL: {
3690 struct kvm_reinject_control control;
3691 r = -EFAULT;
3692 if (copy_from_user(&control, argp, sizeof(control)))
3693 goto out;
3694 r = kvm_vm_ioctl_reinject(kvm, &control);
Marcelo Tosatti52d939a2008-12-30 15:55:06 -02003695 break;
3696 }
Ed Swierkffde22a2009-10-15 15:21:43 -07003697 case KVM_XEN_HVM_CONFIG: {
3698 r = -EFAULT;
3699 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3700 sizeof(struct kvm_xen_hvm_config)))
3701 goto out;
3702 r = -EINVAL;
3703 if (kvm->arch.xen_hvm_config.flags)
3704 goto out;
3705 r = 0;
3706 break;
3707 }
Glauber Costaafbcf7a2009-10-16 15:28:36 -04003708 case KVM_SET_CLOCK: {
Glauber Costaafbcf7a2009-10-16 15:28:36 -04003709 struct kvm_clock_data user_ns;
3710 u64 now_ns;
3711 s64 delta;
3712
3713 r = -EFAULT;
3714 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3715 goto out;
3716
3717 r = -EINVAL;
3718 if (user_ns.flags)
3719 goto out;
3720
3721 r = 0;
Avi Kivity395c6b02010-10-04 12:55:49 +02003722 local_irq_disable();
Zachary Amsden759379d2010-08-19 22:07:25 -10003723 now_ns = get_kernel_ns();
Glauber Costaafbcf7a2009-10-16 15:28:36 -04003724 delta = user_ns.clock - now_ns;
Avi Kivity395c6b02010-10-04 12:55:49 +02003725 local_irq_enable();
Glauber Costaafbcf7a2009-10-16 15:28:36 -04003726 kvm->arch.kvmclock_offset = delta;
3727 break;
3728 }
3729 case KVM_GET_CLOCK: {
Glauber Costaafbcf7a2009-10-16 15:28:36 -04003730 struct kvm_clock_data user_ns;
3731 u64 now_ns;
3732
Avi Kivity395c6b02010-10-04 12:55:49 +02003733 local_irq_disable();
Zachary Amsden759379d2010-08-19 22:07:25 -10003734 now_ns = get_kernel_ns();
Glauber Costaafbcf7a2009-10-16 15:28:36 -04003735 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
Avi Kivity395c6b02010-10-04 12:55:49 +02003736 local_irq_enable();
Glauber Costaafbcf7a2009-10-16 15:28:36 -04003737 user_ns.flags = 0;
Vasiliy Kulikov97e69aa2010-10-30 22:54:47 +04003738 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
Glauber Costaafbcf7a2009-10-16 15:28:36 -04003739
3740 r = -EFAULT;
3741 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
3742 goto out;
3743 r = 0;
3744 break;
3745 }
3746
Carsten Otte1fe779f2007-10-29 16:08:35 +01003747 default:
3748 ;
3749 }
3750out:
3751 return r;
3752}
3753
Zhang Xiantaoa16b0432007-11-16 14:38:21 +08003754static void kvm_init_msr_list(void)
Carsten Otte043405e2007-10-10 17:16:19 +02003755{
3756 u32 dummy[2];
3757 unsigned i, j;
3758
Glauber Costae3267cb2009-10-06 13:24:50 -04003759 /* skip the first msrs in the list. KVM-specific */
3760 for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
Carsten Otte043405e2007-10-10 17:16:19 +02003761 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
3762 continue;
3763 if (j < i)
3764 msrs_to_save[j] = msrs_to_save[i];
3765 j++;
3766 }
3767 num_msrs_to_save = j;
3768}
3769
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03003770static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3771 const void *v)
Carsten Ottebbd9b642007-10-30 18:44:21 +01003772{
Avi Kivity70252a12010-01-19 12:51:22 +02003773 int handled = 0;
3774 int n;
Carsten Ottebbd9b642007-10-30 18:44:21 +01003775
Avi Kivity70252a12010-01-19 12:51:22 +02003776 do {
3777 n = min(len, 8);
3778 if (!(vcpu->arch.apic &&
3779 !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, n, v))
3780 && kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3781 break;
3782 handled += n;
3783 addr += n;
3784 len -= n;
3785 v += n;
3786 } while (len);
3787
3788 return handled;
Carsten Ottebbd9b642007-10-30 18:44:21 +01003789}
3790
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03003791static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
Carsten Ottebbd9b642007-10-30 18:44:21 +01003792{
Avi Kivity70252a12010-01-19 12:51:22 +02003793 int handled = 0;
3794 int n;
Carsten Ottebbd9b642007-10-30 18:44:21 +01003795
Avi Kivity70252a12010-01-19 12:51:22 +02003796 do {
3797 n = min(len, 8);
3798 if (!(vcpu->arch.apic &&
3799 !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, n, v))
3800 && kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3801 break;
3802 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
3803 handled += n;
3804 addr += n;
3805 len -= n;
3806 v += n;
3807 } while (len);
3808
3809 return handled;
Carsten Ottebbd9b642007-10-30 18:44:21 +01003810}
3811
Gleb Natapov2dafc6c2010-03-18 15:20:16 +02003812static void kvm_set_segment(struct kvm_vcpu *vcpu,
3813 struct kvm_segment *var, int seg)
3814{
3815 kvm_x86_ops->set_segment(vcpu, var, seg);
3816}
3817
3818void kvm_get_segment(struct kvm_vcpu *vcpu,
3819 struct kvm_segment *var, int seg)
3820{
3821 kvm_x86_ops->get_segment(vcpu, var, seg);
3822}
3823
Xiao Guangronge459e322011-11-28 20:42:16 +08003824gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
Joerg Roedel02f59dc2010-09-10 17:30:54 +02003825{
3826 gpa_t t_gpa;
Avi Kivityab9ae312010-11-22 17:53:26 +02003827 struct x86_exception exception;
Joerg Roedel02f59dc2010-09-10 17:30:54 +02003828
3829 BUG_ON(!mmu_is_nested(vcpu));
3830
3831 /* NPT walks are always user-walks */
3832 access |= PFERR_USER_MASK;
Avi Kivityab9ae312010-11-22 17:53:26 +02003833 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, &exception);
Joerg Roedel02f59dc2010-09-10 17:30:54 +02003834
3835 return t_gpa;
3836}
3837
Avi Kivityab9ae312010-11-22 17:53:26 +02003838gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
3839 struct x86_exception *exception)
Gleb Natapov1871c602010-02-10 14:21:32 +02003840{
3841 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
Avi Kivityab9ae312010-11-22 17:53:26 +02003842 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
Gleb Natapov1871c602010-02-10 14:21:32 +02003843}
3844
Avi Kivityab9ae312010-11-22 17:53:26 +02003845 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
3846 struct x86_exception *exception)
Gleb Natapov1871c602010-02-10 14:21:32 +02003847{
3848 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3849 access |= PFERR_FETCH_MASK;
Avi Kivityab9ae312010-11-22 17:53:26 +02003850 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
Gleb Natapov1871c602010-02-10 14:21:32 +02003851}
3852
Avi Kivityab9ae312010-11-22 17:53:26 +02003853gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
3854 struct x86_exception *exception)
Gleb Natapov1871c602010-02-10 14:21:32 +02003855{
3856 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3857 access |= PFERR_WRITE_MASK;
Avi Kivityab9ae312010-11-22 17:53:26 +02003858 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
Gleb Natapov1871c602010-02-10 14:21:32 +02003859}
3860
3861/* uses this to access any guest's mapped memory without checking CPL */
Avi Kivityab9ae312010-11-22 17:53:26 +02003862gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
3863 struct x86_exception *exception)
Gleb Natapov1871c602010-02-10 14:21:32 +02003864{
Avi Kivityab9ae312010-11-22 17:53:26 +02003865 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
Gleb Natapov1871c602010-02-10 14:21:32 +02003866}
3867
3868static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
3869 struct kvm_vcpu *vcpu, u32 access,
Avi Kivitybcc55cb2010-11-22 17:53:22 +02003870 struct x86_exception *exception)
Carsten Ottebbd9b642007-10-30 18:44:21 +01003871{
3872 void *data = val;
Marcelo Tosatti10589a42007-12-20 19:18:22 -05003873 int r = X86EMUL_CONTINUE;
Carsten Ottebbd9b642007-10-30 18:44:21 +01003874
3875 while (bytes) {
Joerg Roedel14dfe852010-09-10 17:30:49 +02003876 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
Avi Kivityab9ae312010-11-22 17:53:26 +02003877 exception);
Carsten Ottebbd9b642007-10-30 18:44:21 +01003878 unsigned offset = addr & (PAGE_SIZE-1);
Izik Eidus77c20022008-12-29 01:42:19 +02003879 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
Carsten Ottebbd9b642007-10-30 18:44:21 +01003880 int ret;
3881
Avi Kivitybcc55cb2010-11-22 17:53:22 +02003882 if (gpa == UNMAPPED_GVA)
Avi Kivityab9ae312010-11-22 17:53:26 +02003883 return X86EMUL_PROPAGATE_FAULT;
Izik Eidus77c20022008-12-29 01:42:19 +02003884 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
Marcelo Tosatti10589a42007-12-20 19:18:22 -05003885 if (ret < 0) {
Gleb Natapovc3cd7ff2010-04-28 19:15:35 +03003886 r = X86EMUL_IO_NEEDED;
Marcelo Tosatti10589a42007-12-20 19:18:22 -05003887 goto out;
3888 }
Carsten Ottebbd9b642007-10-30 18:44:21 +01003889
Izik Eidus77c20022008-12-29 01:42:19 +02003890 bytes -= toread;
3891 data += toread;
3892 addr += toread;
Carsten Ottebbd9b642007-10-30 18:44:21 +01003893 }
Marcelo Tosatti10589a42007-12-20 19:18:22 -05003894out:
Marcelo Tosatti10589a42007-12-20 19:18:22 -05003895 return r;
Carsten Ottebbd9b642007-10-30 18:44:21 +01003896}
Izik Eidus77c20022008-12-29 01:42:19 +02003897
Gleb Natapov1871c602010-02-10 14:21:32 +02003898/* used for instruction fetching */
Avi Kivity0f65dd72011-04-20 13:37:53 +03003899static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
3900 gva_t addr, void *val, unsigned int bytes,
Avi Kivitybcc55cb2010-11-22 17:53:22 +02003901 struct x86_exception *exception)
Gleb Natapov1871c602010-02-10 14:21:32 +02003902{
Avi Kivity0f65dd72011-04-20 13:37:53 +03003903 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
Gleb Natapov1871c602010-02-10 14:21:32 +02003904 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
Avi Kivity0f65dd72011-04-20 13:37:53 +03003905
Gleb Natapov1871c602010-02-10 14:21:32 +02003906 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu,
Avi Kivitybcc55cb2010-11-22 17:53:22 +02003907 access | PFERR_FETCH_MASK,
3908 exception);
Gleb Natapov1871c602010-02-10 14:21:32 +02003909}
3910
Nadav Har'El064aea72011-05-25 23:04:56 +03003911int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
Avi Kivity0f65dd72011-04-20 13:37:53 +03003912 gva_t addr, void *val, unsigned int bytes,
Avi Kivitybcc55cb2010-11-22 17:53:22 +02003913 struct x86_exception *exception)
Gleb Natapov1871c602010-02-10 14:21:32 +02003914{
Avi Kivity0f65dd72011-04-20 13:37:53 +03003915 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
Gleb Natapov1871c602010-02-10 14:21:32 +02003916 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
Avi Kivity0f65dd72011-04-20 13:37:53 +03003917
Gleb Natapov1871c602010-02-10 14:21:32 +02003918 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
Avi Kivitybcc55cb2010-11-22 17:53:22 +02003919 exception);
Gleb Natapov1871c602010-02-10 14:21:32 +02003920}
Nadav Har'El064aea72011-05-25 23:04:56 +03003921EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
Gleb Natapov1871c602010-02-10 14:21:32 +02003922
Avi Kivity0f65dd72011-04-20 13:37:53 +03003923static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
3924 gva_t addr, void *val, unsigned int bytes,
Avi Kivitybcc55cb2010-11-22 17:53:22 +02003925 struct x86_exception *exception)
Gleb Natapov1871c602010-02-10 14:21:32 +02003926{
Avi Kivity0f65dd72011-04-20 13:37:53 +03003927 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
Avi Kivitybcc55cb2010-11-22 17:53:22 +02003928 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
Gleb Natapov1871c602010-02-10 14:21:32 +02003929}
3930
Nadav Har'El6a4d7552011-05-25 23:08:00 +03003931int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
Avi Kivity0f65dd72011-04-20 13:37:53 +03003932 gva_t addr, void *val,
Gleb Natapov2dafc6c2010-03-18 15:20:16 +02003933 unsigned int bytes,
Avi Kivitybcc55cb2010-11-22 17:53:22 +02003934 struct x86_exception *exception)
Izik Eidus77c20022008-12-29 01:42:19 +02003935{
Avi Kivity0f65dd72011-04-20 13:37:53 +03003936 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
Izik Eidus77c20022008-12-29 01:42:19 +02003937 void *data = val;
3938 int r = X86EMUL_CONTINUE;
3939
3940 while (bytes) {
Joerg Roedel14dfe852010-09-10 17:30:49 +02003941 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
3942 PFERR_WRITE_MASK,
Avi Kivityab9ae312010-11-22 17:53:26 +02003943 exception);
Izik Eidus77c20022008-12-29 01:42:19 +02003944 unsigned offset = addr & (PAGE_SIZE-1);
3945 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
3946 int ret;
3947
Avi Kivitybcc55cb2010-11-22 17:53:22 +02003948 if (gpa == UNMAPPED_GVA)
Avi Kivityab9ae312010-11-22 17:53:26 +02003949 return X86EMUL_PROPAGATE_FAULT;
Izik Eidus77c20022008-12-29 01:42:19 +02003950 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
3951 if (ret < 0) {
Gleb Natapovc3cd7ff2010-04-28 19:15:35 +03003952 r = X86EMUL_IO_NEEDED;
Izik Eidus77c20022008-12-29 01:42:19 +02003953 goto out;
3954 }
3955
3956 bytes -= towrite;
3957 data += towrite;
3958 addr += towrite;
3959 }
3960out:
3961 return r;
3962}
Nadav Har'El6a4d7552011-05-25 23:08:00 +03003963EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
Izik Eidus77c20022008-12-29 01:42:19 +02003964
Xiao Guangrongaf7cc7d2011-07-12 03:22:46 +08003965static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
3966 gpa_t *gpa, struct x86_exception *exception,
3967 bool write)
3968{
Avi Kivity97d64b72012-09-12 14:52:00 +03003969 u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
3970 | (write ? PFERR_WRITE_MASK : 0);
Xiao Guangrongaf7cc7d2011-07-12 03:22:46 +08003971
Avi Kivity97d64b72012-09-12 14:52:00 +03003972 if (vcpu_match_mmio_gva(vcpu, gva)
3973 && !permission_fault(vcpu->arch.walk_mmu, vcpu->arch.access, access)) {
Xiao Guangrongbebb1062011-07-12 03:23:20 +08003974 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
3975 (gva & (PAGE_SIZE - 1));
Xiao Guangrong4f022642011-07-12 03:34:24 +08003976 trace_vcpu_match_mmio(gva, *gpa, write, false);
Xiao Guangrongbebb1062011-07-12 03:23:20 +08003977 return 1;
3978 }
3979
Xiao Guangrongaf7cc7d2011-07-12 03:22:46 +08003980 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3981
3982 if (*gpa == UNMAPPED_GVA)
3983 return -1;
3984
3985 /* For APIC access vmexit */
3986 if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3987 return 1;
3988
Xiao Guangrong4f022642011-07-12 03:34:24 +08003989 if (vcpu_match_mmio_gpa(vcpu, *gpa)) {
3990 trace_vcpu_match_mmio(gva, *gpa, write, true);
Xiao Guangrongbebb1062011-07-12 03:23:20 +08003991 return 1;
Xiao Guangrong4f022642011-07-12 03:34:24 +08003992 }
Xiao Guangrongbebb1062011-07-12 03:23:20 +08003993
Xiao Guangrongaf7cc7d2011-07-12 03:22:46 +08003994 return 0;
3995}
3996
Marcelo Tosatti3200f402008-03-29 20:17:59 -03003997int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
Avi Kivitybcc55cb2010-11-22 17:53:22 +02003998 const void *val, int bytes)
Avi Kivity9f811282008-03-02 14:06:05 +02003999{
4000 int ret;
4001
4002 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
4003 if (ret < 0)
4004 return 0;
Xiao Guangrongf57f2ef2011-09-22 16:56:39 +08004005 kvm_mmu_pte_write(vcpu, gpa, val, bytes);
Avi Kivity9f811282008-03-02 14:06:05 +02004006 return 1;
4007}
4008
Xiao Guangrong77d197b2011-07-13 14:31:50 +08004009struct read_write_emulator_ops {
4010 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4011 int bytes);
4012 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4013 void *val, int bytes);
4014 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4015 int bytes, void *val);
4016 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4017 void *val, int bytes);
4018 bool write;
4019};
4020
4021static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4022{
4023 if (vcpu->mmio_read_completed) {
Xiao Guangrong77d197b2011-07-13 14:31:50 +08004024 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
Avi Kivityf78146b2012-04-18 19:22:47 +03004025 vcpu->mmio_fragments[0].gpa, *(u64 *)val);
Xiao Guangrong77d197b2011-07-13 14:31:50 +08004026 vcpu->mmio_read_completed = 0;
4027 return 1;
4028 }
4029
4030 return 0;
4031}
4032
4033static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4034 void *val, int bytes)
4035{
4036 return !kvm_read_guest(vcpu->kvm, gpa, val, bytes);
4037}
4038
4039static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4040 void *val, int bytes)
4041{
4042 return emulator_write_phys(vcpu, gpa, val, bytes);
4043}
4044
4045static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4046{
4047 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4048 return vcpu_mmio_write(vcpu, gpa, bytes, val);
4049}
4050
4051static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4052 void *val, int bytes)
4053{
4054 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4055 return X86EMUL_IO_NEEDED;
4056}
4057
4058static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4059 void *val, int bytes)
4060{
Avi Kivityf78146b2012-04-18 19:22:47 +03004061 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4062
Xiao Guangrong87da7e62012-10-24 14:07:59 +08004063 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
Xiao Guangrong77d197b2011-07-13 14:31:50 +08004064 return X86EMUL_CONTINUE;
4065}
4066
Mathias Krause0fbe9b02012-08-30 01:30:17 +02004067static const struct read_write_emulator_ops read_emultor = {
Xiao Guangrong77d197b2011-07-13 14:31:50 +08004068 .read_write_prepare = read_prepare,
4069 .read_write_emulate = read_emulate,
4070 .read_write_mmio = vcpu_mmio_read,
4071 .read_write_exit_mmio = read_exit_mmio,
4072};
4073
Mathias Krause0fbe9b02012-08-30 01:30:17 +02004074static const struct read_write_emulator_ops write_emultor = {
Xiao Guangrong77d197b2011-07-13 14:31:50 +08004075 .read_write_emulate = write_emulate,
4076 .read_write_mmio = write_mmio,
4077 .read_write_exit_mmio = write_exit_mmio,
4078 .write = true,
4079};
4080
Xiao Guangrong22388a32011-07-13 14:32:31 +08004081static int emulator_read_write_onepage(unsigned long addr, void *val,
4082 unsigned int bytes,
4083 struct x86_exception *exception,
4084 struct kvm_vcpu *vcpu,
Mathias Krause0fbe9b02012-08-30 01:30:17 +02004085 const struct read_write_emulator_ops *ops)
Carsten Ottebbd9b642007-10-30 18:44:21 +01004086{
Xiao Guangrongaf7cc7d2011-07-12 03:22:46 +08004087 gpa_t gpa;
4088 int handled, ret;
Xiao Guangrong22388a32011-07-13 14:32:31 +08004089 bool write = ops->write;
Avi Kivityf78146b2012-04-18 19:22:47 +03004090 struct kvm_mmio_fragment *frag;
Xiao Guangrong22388a32011-07-13 14:32:31 +08004091
4092 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
Carsten Ottebbd9b642007-10-30 18:44:21 +01004093
Xiao Guangrongaf7cc7d2011-07-12 03:22:46 +08004094 if (ret < 0)
Carsten Ottebbd9b642007-10-30 18:44:21 +01004095 return X86EMUL_PROPAGATE_FAULT;
Carsten Ottebbd9b642007-10-30 18:44:21 +01004096
4097 /* For APIC access vmexit */
Xiao Guangrongaf7cc7d2011-07-12 03:22:46 +08004098 if (ret)
Carsten Ottebbd9b642007-10-30 18:44:21 +01004099 goto mmio;
4100
Xiao Guangrong22388a32011-07-13 14:32:31 +08004101 if (ops->read_write_emulate(vcpu, gpa, val, bytes))
Carsten Ottebbd9b642007-10-30 18:44:21 +01004102 return X86EMUL_CONTINUE;
4103
4104mmio:
4105 /*
4106 * Is this MMIO handled locally?
4107 */
Xiao Guangrong22388a32011-07-13 14:32:31 +08004108 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
Avi Kivity70252a12010-01-19 12:51:22 +02004109 if (handled == bytes)
Carsten Ottebbd9b642007-10-30 18:44:21 +01004110 return X86EMUL_CONTINUE;
Carsten Ottebbd9b642007-10-30 18:44:21 +01004111
Avi Kivity70252a12010-01-19 12:51:22 +02004112 gpa += handled;
4113 bytes -= handled;
4114 val += handled;
4115
Xiao Guangrong87da7e62012-10-24 14:07:59 +08004116 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
4117 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
4118 frag->gpa = gpa;
4119 frag->data = val;
4120 frag->len = bytes;
Avi Kivityf78146b2012-04-18 19:22:47 +03004121 return X86EMUL_CONTINUE;
Carsten Ottebbd9b642007-10-30 18:44:21 +01004122}
4123
Xiao Guangrong22388a32011-07-13 14:32:31 +08004124int emulator_read_write(struct x86_emulate_ctxt *ctxt, unsigned long addr,
4125 void *val, unsigned int bytes,
4126 struct x86_exception *exception,
Mathias Krause0fbe9b02012-08-30 01:30:17 +02004127 const struct read_write_emulator_ops *ops)
Carsten Ottebbd9b642007-10-30 18:44:21 +01004128{
Avi Kivity0f65dd72011-04-20 13:37:53 +03004129 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
Avi Kivityf78146b2012-04-18 19:22:47 +03004130 gpa_t gpa;
4131 int rc;
4132
4133 if (ops->read_write_prepare &&
4134 ops->read_write_prepare(vcpu, val, bytes))
4135 return X86EMUL_CONTINUE;
4136
4137 vcpu->mmio_nr_fragments = 0;
Avi Kivity0f65dd72011-04-20 13:37:53 +03004138
Carsten Ottebbd9b642007-10-30 18:44:21 +01004139 /* Crossing a page boundary? */
4140 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
Avi Kivityf78146b2012-04-18 19:22:47 +03004141 int now;
Carsten Ottebbd9b642007-10-30 18:44:21 +01004142
4143 now = -addr & ~PAGE_MASK;
Xiao Guangrong22388a32011-07-13 14:32:31 +08004144 rc = emulator_read_write_onepage(addr, val, now, exception,
4145 vcpu, ops);
4146
Carsten Ottebbd9b642007-10-30 18:44:21 +01004147 if (rc != X86EMUL_CONTINUE)
4148 return rc;
4149 addr += now;
4150 val += now;
4151 bytes -= now;
4152 }
Xiao Guangrong22388a32011-07-13 14:32:31 +08004153
Avi Kivityf78146b2012-04-18 19:22:47 +03004154 rc = emulator_read_write_onepage(addr, val, bytes, exception,
4155 vcpu, ops);
4156 if (rc != X86EMUL_CONTINUE)
4157 return rc;
4158
4159 if (!vcpu->mmio_nr_fragments)
4160 return rc;
4161
4162 gpa = vcpu->mmio_fragments[0].gpa;
4163
4164 vcpu->mmio_needed = 1;
4165 vcpu->mmio_cur_fragment = 0;
4166
Xiao Guangrong87da7e62012-10-24 14:07:59 +08004167 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
Avi Kivityf78146b2012-04-18 19:22:47 +03004168 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
4169 vcpu->run->exit_reason = KVM_EXIT_MMIO;
4170 vcpu->run->mmio.phys_addr = gpa;
4171
4172 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
Xiao Guangrong22388a32011-07-13 14:32:31 +08004173}
4174
4175static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4176 unsigned long addr,
4177 void *val,
4178 unsigned int bytes,
4179 struct x86_exception *exception)
4180{
4181 return emulator_read_write(ctxt, addr, val, bytes,
4182 exception, &read_emultor);
4183}
4184
4185int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
4186 unsigned long addr,
4187 const void *val,
4188 unsigned int bytes,
4189 struct x86_exception *exception)
4190{
4191 return emulator_read_write(ctxt, addr, (void *)val, bytes,
4192 exception, &write_emultor);
Carsten Ottebbd9b642007-10-30 18:44:21 +01004193}
Carsten Ottebbd9b642007-10-30 18:44:21 +01004194
Avi Kivitydaea3e72010-03-15 13:59:54 +02004195#define CMPXCHG_TYPE(t, ptr, old, new) \
4196 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4197
4198#ifdef CONFIG_X86_64
4199# define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4200#else
4201# define CMPXCHG64(ptr, old, new) \
Jan Kiszka9749a6c2010-03-20 10:14:13 +01004202 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
Avi Kivitydaea3e72010-03-15 13:59:54 +02004203#endif
4204
Avi Kivity0f65dd72011-04-20 13:37:53 +03004205static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4206 unsigned long addr,
Carsten Ottebbd9b642007-10-30 18:44:21 +01004207 const void *old,
4208 const void *new,
4209 unsigned int bytes,
Avi Kivity0f65dd72011-04-20 13:37:53 +03004210 struct x86_exception *exception)
Carsten Ottebbd9b642007-10-30 18:44:21 +01004211{
Avi Kivity0f65dd72011-04-20 13:37:53 +03004212 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
Avi Kivitydaea3e72010-03-15 13:59:54 +02004213 gpa_t gpa;
4214 struct page *page;
4215 char *kaddr;
4216 bool exchanged;
4217
Marcelo Tosatti2bacc552007-12-12 10:46:12 -05004218 /* guests cmpxchg8b have to be emulated atomically */
Avi Kivitydaea3e72010-03-15 13:59:54 +02004219 if (bytes > 8 || (bytes & (bytes - 1)))
4220 goto emul_write;
Marcelo Tosatti2bacc552007-12-12 10:46:12 -05004221
Avi Kivitydaea3e72010-03-15 13:59:54 +02004222 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
Marcelo Tosatti10589a42007-12-20 19:18:22 -05004223
Avi Kivitydaea3e72010-03-15 13:59:54 +02004224 if (gpa == UNMAPPED_GVA ||
4225 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4226 goto emul_write;
Marcelo Tosatti2bacc552007-12-12 10:46:12 -05004227
Avi Kivitydaea3e72010-03-15 13:59:54 +02004228 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4229 goto emul_write;
Marcelo Tosatti2bacc552007-12-12 10:46:12 -05004230
Avi Kivitydaea3e72010-03-15 13:59:54 +02004231 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
Xiao Guangrong32cad842012-08-03 15:42:52 +08004232 if (is_error_page(page))
Wei Yongjunc19b8bd62010-07-15 08:51:58 +08004233 goto emul_write;
Izik Eidus72dc67a2008-02-10 18:04:15 +02004234
Cong Wang8fd75e12011-11-25 23:14:17 +08004235 kaddr = kmap_atomic(page);
Avi Kivitydaea3e72010-03-15 13:59:54 +02004236 kaddr += offset_in_page(gpa);
4237 switch (bytes) {
4238 case 1:
4239 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4240 break;
4241 case 2:
4242 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4243 break;
4244 case 4:
4245 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4246 break;
4247 case 8:
4248 exchanged = CMPXCHG64(kaddr, old, new);
4249 break;
4250 default:
4251 BUG();
Marcelo Tosatti2bacc552007-12-12 10:46:12 -05004252 }
Cong Wang8fd75e12011-11-25 23:14:17 +08004253 kunmap_atomic(kaddr);
Avi Kivitydaea3e72010-03-15 13:59:54 +02004254 kvm_release_page_dirty(page);
4255
4256 if (!exchanged)
4257 return X86EMUL_CMPXCHG_FAILED;
4258
Xiao Guangrongf57f2ef2011-09-22 16:56:39 +08004259 kvm_mmu_pte_write(vcpu, gpa, new, bytes);
Gleb Natapov8f6abd02010-04-13 10:21:56 +03004260
4261 return X86EMUL_CONTINUE;
Avi Kivity4a5f48f2010-03-15 13:59:55 +02004262
Marcelo Tosatti3200f402008-03-29 20:17:59 -03004263emul_write:
Avi Kivitydaea3e72010-03-15 13:59:54 +02004264 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
Marcelo Tosatti2bacc552007-12-12 10:46:12 -05004265
Avi Kivity0f65dd72011-04-20 13:37:53 +03004266 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
Carsten Ottebbd9b642007-10-30 18:44:21 +01004267}
4268
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02004269static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4270{
4271 /* TODO: String I/O for in kernel device */
4272 int r;
4273
4274 if (vcpu->arch.pio.in)
4275 r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
4276 vcpu->arch.pio.size, pd);
4277 else
4278 r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
4279 vcpu->arch.pio.port, vcpu->arch.pio.size,
4280 pd);
4281 return r;
4282}
4283
Xiao Guangrong6f6fbe92011-09-22 16:55:10 +08004284static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
4285 unsigned short port, void *val,
4286 unsigned int count, bool in)
4287{
4288 trace_kvm_pio(!in, port, size, count);
4289
4290 vcpu->arch.pio.port = port;
4291 vcpu->arch.pio.in = in;
4292 vcpu->arch.pio.count = count;
4293 vcpu->arch.pio.size = size;
4294
4295 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4296 vcpu->arch.pio.count = 0;
4297 return 1;
4298 }
4299
4300 vcpu->run->exit_reason = KVM_EXIT_IO;
4301 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
4302 vcpu->run->io.size = size;
4303 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4304 vcpu->run->io.count = count;
4305 vcpu->run->io.port = port;
4306
4307 return 0;
4308}
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02004309
Avi Kivityca1d4a92011-04-20 13:37:53 +03004310static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4311 int size, unsigned short port, void *val,
4312 unsigned int count)
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02004313{
Avi Kivityca1d4a92011-04-20 13:37:53 +03004314 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
Xiao Guangrong6f6fbe92011-09-22 16:55:10 +08004315 int ret;
Avi Kivityca1d4a92011-04-20 13:37:53 +03004316
Gleb Natapov79729952010-03-18 15:20:24 +02004317 if (vcpu->arch.pio.count)
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02004318 goto data_avail;
4319
Xiao Guangrong6f6fbe92011-09-22 16:55:10 +08004320 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
4321 if (ret) {
4322data_avail:
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02004323 memcpy(val, vcpu->arch.pio_data, size * count);
Gleb Natapov79729952010-03-18 15:20:24 +02004324 vcpu->arch.pio.count = 0;
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02004325 return 1;
4326 }
4327
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02004328 return 0;
4329}
4330
Avi Kivityca1d4a92011-04-20 13:37:53 +03004331static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4332 int size, unsigned short port,
4333 const void *val, unsigned int count)
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02004334{
Avi Kivityca1d4a92011-04-20 13:37:53 +03004335 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4336
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02004337 memcpy(vcpu->arch.pio_data, val, size * count);
Xiao Guangrong6f6fbe92011-09-22 16:55:10 +08004338 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02004339}
4340
Carsten Ottebbd9b642007-10-30 18:44:21 +01004341static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4342{
4343 return kvm_x86_ops->get_segment_base(vcpu, seg);
4344}
4345
Avi Kivity3cb16fe2011-04-20 15:38:44 +03004346static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
Carsten Ottebbd9b642007-10-30 18:44:21 +01004347{
Avi Kivity3cb16fe2011-04-20 15:38:44 +03004348 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
Carsten Ottebbd9b642007-10-30 18:44:21 +01004349}
4350
Sheng Yangf5f48ee2010-06-30 12:25:15 +08004351int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4352{
4353 if (!need_emulate_wbinvd(vcpu))
4354 return X86EMUL_CONTINUE;
4355
4356 if (kvm_x86_ops->has_wbinvd_exit()) {
Jan Kiszka2eec7342010-11-01 14:01:29 +01004357 int cpu = get_cpu();
4358
4359 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
Sheng Yangf5f48ee2010-06-30 12:25:15 +08004360 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4361 wbinvd_ipi, NULL, 1);
Jan Kiszka2eec7342010-11-01 14:01:29 +01004362 put_cpu();
Sheng Yangf5f48ee2010-06-30 12:25:15 +08004363 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
Jan Kiszka2eec7342010-11-01 14:01:29 +01004364 } else
4365 wbinvd();
Sheng Yangf5f48ee2010-06-30 12:25:15 +08004366 return X86EMUL_CONTINUE;
4367}
4368EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4369
Avi Kivitybcaf5cc2011-04-20 15:53:23 +03004370static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4371{
4372 kvm_emulate_wbinvd(emul_to_vcpu(ctxt));
4373}
4374
Avi Kivity717746e2011-04-20 13:37:53 +03004375int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
Carsten Ottebbd9b642007-10-30 18:44:21 +01004376{
Avi Kivity717746e2011-04-20 13:37:53 +03004377 return _kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
Carsten Ottebbd9b642007-10-30 18:44:21 +01004378}
4379
Avi Kivity717746e2011-04-20 13:37:53 +03004380int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
Carsten Ottebbd9b642007-10-30 18:44:21 +01004381{
Gleb Natapov338dbc92010-04-28 19:15:32 +03004382
Avi Kivity717746e2011-04-20 13:37:53 +03004383 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
Carsten Ottebbd9b642007-10-30 18:44:21 +01004384}
4385
Gleb Natapov52a46612010-03-18 15:20:03 +02004386static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4387{
4388 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4389}
4390
Avi Kivity717746e2011-04-20 13:37:53 +03004391static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
Gleb Natapov52a46612010-03-18 15:20:03 +02004392{
Avi Kivity717746e2011-04-20 13:37:53 +03004393 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
Gleb Natapov52a46612010-03-18 15:20:03 +02004394 unsigned long value;
4395
4396 switch (cr) {
4397 case 0:
4398 value = kvm_read_cr0(vcpu);
4399 break;
4400 case 2:
4401 value = vcpu->arch.cr2;
4402 break;
4403 case 3:
Avi Kivity9f8fe502010-12-05 17:30:00 +02004404 value = kvm_read_cr3(vcpu);
Gleb Natapov52a46612010-03-18 15:20:03 +02004405 break;
4406 case 4:
4407 value = kvm_read_cr4(vcpu);
4408 break;
4409 case 8:
4410 value = kvm_get_cr8(vcpu);
4411 break;
4412 default:
Christoffer Dalla737f252012-06-03 21:17:48 +03004413 kvm_err("%s: unexpected cr %u\n", __func__, cr);
Gleb Natapov52a46612010-03-18 15:20:03 +02004414 return 0;
4415 }
4416
4417 return value;
4418}
4419
Avi Kivity717746e2011-04-20 13:37:53 +03004420static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
Gleb Natapov52a46612010-03-18 15:20:03 +02004421{
Avi Kivity717746e2011-04-20 13:37:53 +03004422 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
Gleb Natapov0f122442010-04-28 19:15:31 +03004423 int res = 0;
4424
Gleb Natapov52a46612010-03-18 15:20:03 +02004425 switch (cr) {
4426 case 0:
Avi Kivity49a9b072010-06-10 17:02:14 +03004427 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
Gleb Natapov52a46612010-03-18 15:20:03 +02004428 break;
4429 case 2:
4430 vcpu->arch.cr2 = val;
4431 break;
4432 case 3:
Avi Kivity23902182010-06-10 17:02:16 +03004433 res = kvm_set_cr3(vcpu, val);
Gleb Natapov52a46612010-03-18 15:20:03 +02004434 break;
4435 case 4:
Avi Kivitya83b29c2010-06-10 17:02:15 +03004436 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
Gleb Natapov52a46612010-03-18 15:20:03 +02004437 break;
4438 case 8:
Andre Przywaraeea1cff2010-12-21 11:12:00 +01004439 res = kvm_set_cr8(vcpu, val);
Gleb Natapov52a46612010-03-18 15:20:03 +02004440 break;
4441 default:
Christoffer Dalla737f252012-06-03 21:17:48 +03004442 kvm_err("%s: unexpected cr %u\n", __func__, cr);
Gleb Natapov0f122442010-04-28 19:15:31 +03004443 res = -1;
Gleb Natapov52a46612010-03-18 15:20:03 +02004444 }
Gleb Natapov0f122442010-04-28 19:15:31 +03004445
4446 return res;
Gleb Natapov52a46612010-03-18 15:20:03 +02004447}
4448
Kevin Wolf4cee4792012-02-08 14:34:41 +01004449static void emulator_set_rflags(struct x86_emulate_ctxt *ctxt, ulong val)
4450{
4451 kvm_set_rflags(emul_to_vcpu(ctxt), val);
4452}
4453
Avi Kivity717746e2011-04-20 13:37:53 +03004454static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
Gleb Natapov9c537242010-03-18 15:20:05 +02004455{
Avi Kivity717746e2011-04-20 13:37:53 +03004456 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
Gleb Natapov9c537242010-03-18 15:20:05 +02004457}
4458
Avi Kivity4bff1e862011-04-20 13:37:53 +03004459static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
Gleb Natapov2dafc6c2010-03-18 15:20:16 +02004460{
Avi Kivity4bff1e862011-04-20 13:37:53 +03004461 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
Gleb Natapov2dafc6c2010-03-18 15:20:16 +02004462}
4463
Avi Kivity4bff1e862011-04-20 13:37:53 +03004464static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
Mohammed Gamal160ce1f2010-08-04 05:44:24 +03004465{
Avi Kivity4bff1e862011-04-20 13:37:53 +03004466 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
Mohammed Gamal160ce1f2010-08-04 05:44:24 +03004467}
4468
Avi Kivity1ac9d0c2011-04-20 15:12:00 +03004469static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4470{
4471 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4472}
4473
4474static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4475{
4476 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4477}
4478
Avi Kivity4bff1e862011-04-20 13:37:53 +03004479static unsigned long emulator_get_cached_segment_base(
4480 struct x86_emulate_ctxt *ctxt, int seg)
Gleb Natapov5951c442010-04-28 19:15:29 +03004481{
Avi Kivity4bff1e862011-04-20 13:37:53 +03004482 return get_segment_base(emul_to_vcpu(ctxt), seg);
Gleb Natapov5951c442010-04-28 19:15:29 +03004483}
4484
Avi Kivity1aa36612011-04-27 13:20:30 +03004485static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
4486 struct desc_struct *desc, u32 *base3,
4487 int seg)
Gleb Natapov2dafc6c2010-03-18 15:20:16 +02004488{
4489 struct kvm_segment var;
4490
Avi Kivity4bff1e862011-04-20 13:37:53 +03004491 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
Avi Kivity1aa36612011-04-27 13:20:30 +03004492 *selector = var.selector;
Gleb Natapov2dafc6c2010-03-18 15:20:16 +02004493
Gleb Natapov378a8b02013-01-21 15:36:48 +02004494 if (var.unusable) {
4495 memset(desc, 0, sizeof(*desc));
Gleb Natapov2dafc6c2010-03-18 15:20:16 +02004496 return false;
Gleb Natapov378a8b02013-01-21 15:36:48 +02004497 }
Gleb Natapov2dafc6c2010-03-18 15:20:16 +02004498
4499 if (var.g)
4500 var.limit >>= 12;
4501 set_desc_limit(desc, var.limit);
4502 set_desc_base(desc, (unsigned long)var.base);
Gleb Natapov5601d052011-03-07 14:55:06 +02004503#ifdef CONFIG_X86_64
4504 if (base3)
4505 *base3 = var.base >> 32;
4506#endif
Gleb Natapov2dafc6c2010-03-18 15:20:16 +02004507 desc->type = var.type;
4508 desc->s = var.s;
4509 desc->dpl = var.dpl;
4510 desc->p = var.present;
4511 desc->avl = var.avl;
4512 desc->l = var.l;
4513 desc->d = var.db;
4514 desc->g = var.g;
4515
4516 return true;
4517}
4518
Avi Kivity1aa36612011-04-27 13:20:30 +03004519static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
4520 struct desc_struct *desc, u32 base3,
4521 int seg)
Gleb Natapov2dafc6c2010-03-18 15:20:16 +02004522{
Avi Kivity4bff1e862011-04-20 13:37:53 +03004523 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
Gleb Natapov2dafc6c2010-03-18 15:20:16 +02004524 struct kvm_segment var;
4525
Avi Kivity1aa36612011-04-27 13:20:30 +03004526 var.selector = selector;
Gleb Natapov2dafc6c2010-03-18 15:20:16 +02004527 var.base = get_desc_base(desc);
Gleb Natapov5601d052011-03-07 14:55:06 +02004528#ifdef CONFIG_X86_64
4529 var.base |= ((u64)base3) << 32;
4530#endif
Gleb Natapov2dafc6c2010-03-18 15:20:16 +02004531 var.limit = get_desc_limit(desc);
4532 if (desc->g)
4533 var.limit = (var.limit << 12) | 0xfff;
4534 var.type = desc->type;
4535 var.present = desc->p;
4536 var.dpl = desc->dpl;
4537 var.db = desc->d;
4538 var.s = desc->s;
4539 var.l = desc->l;
4540 var.g = desc->g;
4541 var.avl = desc->avl;
4542 var.present = desc->p;
4543 var.unusable = !var.present;
4544 var.padding = 0;
4545
4546 kvm_set_segment(vcpu, &var, seg);
4547 return;
4548}
4549
Avi Kivity717746e2011-04-20 13:37:53 +03004550static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
4551 u32 msr_index, u64 *pdata)
4552{
4553 return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
4554}
4555
4556static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
4557 u32 msr_index, u64 data)
4558{
Will Auld8fe8ab42012-11-29 12:42:12 -08004559 struct msr_data msr;
4560
4561 msr.data = data;
4562 msr.index = msr_index;
4563 msr.host_initiated = false;
4564 return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
Avi Kivity717746e2011-04-20 13:37:53 +03004565}
4566
Avi Kivity222d21a2011-11-10 14:57:30 +02004567static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
4568 u32 pmc, u64 *pdata)
4569{
4570 return kvm_pmu_read_pmc(emul_to_vcpu(ctxt), pmc, pdata);
4571}
4572
Avi Kivity6c3287f2011-04-20 15:43:05 +03004573static void emulator_halt(struct x86_emulate_ctxt *ctxt)
4574{
4575 emul_to_vcpu(ctxt)->arch.halt_request = 1;
4576}
4577
Avi Kivity5037f6f2011-03-28 16:53:59 +02004578static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
4579{
4580 preempt_disable();
Avi Kivity5197b802011-04-20 15:55:40 +03004581 kvm_load_guest_fpu(emul_to_vcpu(ctxt));
Avi Kivity5037f6f2011-03-28 16:53:59 +02004582 /*
4583 * CR0.TS may reference the host fpu state, not the guest fpu state,
4584 * so it may be clear at this point.
4585 */
4586 clts();
4587}
4588
4589static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
4590{
4591 preempt_enable();
4592}
4593
Avi Kivity29535382011-04-20 13:37:53 +03004594static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
Joerg Roedel8a76d7f2011-04-04 12:39:27 +02004595 struct x86_instruction_info *info,
Avi Kivityc4f035c2011-04-04 12:39:22 +02004596 enum x86_intercept_stage stage)
4597{
Avi Kivity29535382011-04-20 13:37:53 +03004598 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
Avi Kivityc4f035c2011-04-04 12:39:22 +02004599}
4600
Avi Kivity0017f932012-06-07 14:10:16 +03004601static void emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
Stephan Bärwolfbdb42f52012-01-12 16:43:03 +01004602 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
4603{
Avi Kivity0017f932012-06-07 14:10:16 +03004604 kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx);
Stephan Bärwolfbdb42f52012-01-12 16:43:03 +01004605}
4606
Avi Kivitydd856ef2012-08-27 23:46:17 +03004607static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
4608{
4609 return kvm_register_read(emul_to_vcpu(ctxt), reg);
4610}
4611
4612static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
4613{
4614 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
4615}
4616
Mathias Krause0225fb52012-08-30 01:30:16 +02004617static const struct x86_emulate_ops emulate_ops = {
Avi Kivitydd856ef2012-08-27 23:46:17 +03004618 .read_gpr = emulator_read_gpr,
4619 .write_gpr = emulator_write_gpr,
Gleb Natapov1871c602010-02-10 14:21:32 +02004620 .read_std = kvm_read_guest_virt_system,
Gleb Natapov2dafc6c2010-03-18 15:20:16 +02004621 .write_std = kvm_write_guest_virt_system,
Gleb Natapov1871c602010-02-10 14:21:32 +02004622 .fetch = kvm_fetch_guest_virt,
Carsten Ottebbd9b642007-10-30 18:44:21 +01004623 .read_emulated = emulator_read_emulated,
4624 .write_emulated = emulator_write_emulated,
4625 .cmpxchg_emulated = emulator_cmpxchg_emulated,
Avi Kivity3cb16fe2011-04-20 15:38:44 +03004626 .invlpg = emulator_invlpg,
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02004627 .pio_in_emulated = emulator_pio_in_emulated,
4628 .pio_out_emulated = emulator_pio_out_emulated,
Avi Kivity1aa36612011-04-27 13:20:30 +03004629 .get_segment = emulator_get_segment,
4630 .set_segment = emulator_set_segment,
Gleb Natapov5951c442010-04-28 19:15:29 +03004631 .get_cached_segment_base = emulator_get_cached_segment_base,
Gleb Natapov2dafc6c2010-03-18 15:20:16 +02004632 .get_gdt = emulator_get_gdt,
Mohammed Gamal160ce1f2010-08-04 05:44:24 +03004633 .get_idt = emulator_get_idt,
Avi Kivity1ac9d0c2011-04-20 15:12:00 +03004634 .set_gdt = emulator_set_gdt,
4635 .set_idt = emulator_set_idt,
Gleb Natapov52a46612010-03-18 15:20:03 +02004636 .get_cr = emulator_get_cr,
4637 .set_cr = emulator_set_cr,
Kevin Wolf4cee4792012-02-08 14:34:41 +01004638 .set_rflags = emulator_set_rflags,
Gleb Natapov9c537242010-03-18 15:20:05 +02004639 .cpl = emulator_get_cpl,
Gleb Natapov35aa5372010-04-28 19:15:27 +03004640 .get_dr = emulator_get_dr,
4641 .set_dr = emulator_set_dr,
Avi Kivity717746e2011-04-20 13:37:53 +03004642 .set_msr = emulator_set_msr,
4643 .get_msr = emulator_get_msr,
Avi Kivity222d21a2011-11-10 14:57:30 +02004644 .read_pmc = emulator_read_pmc,
Avi Kivity6c3287f2011-04-20 15:43:05 +03004645 .halt = emulator_halt,
Avi Kivitybcaf5cc2011-04-20 15:53:23 +03004646 .wbinvd = emulator_wbinvd,
Avi Kivityd6aa1002011-04-20 15:47:13 +03004647 .fix_hypercall = emulator_fix_hypercall,
Avi Kivity5037f6f2011-03-28 16:53:59 +02004648 .get_fpu = emulator_get_fpu,
4649 .put_fpu = emulator_put_fpu,
Avi Kivityc4f035c2011-04-04 12:39:22 +02004650 .intercept = emulator_intercept,
Stephan Bärwolfbdb42f52012-01-12 16:43:03 +01004651 .get_cpuid = emulator_get_cpuid,
Carsten Ottebbd9b642007-10-30 18:44:21 +01004652};
4653
Gleb Natapov95cb2292010-04-28 19:15:43 +03004654static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
4655{
4656 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu, mask);
4657 /*
4658 * an sti; sti; sequence only disable interrupts for the first
4659 * instruction. So, if the last instruction, be it emulated or
4660 * not, left the system with the INT_STI flag enabled, it
4661 * means that the last instruction is an sti. We should not
4662 * leave the flag on in this case. The same goes for mov ss
4663 */
4664 if (!(int_shadow & mask))
4665 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
4666}
4667
Gleb Natapov54b84862010-04-28 19:15:44 +03004668static void inject_emulated_exception(struct kvm_vcpu *vcpu)
4669{
4670 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
Avi Kivityda9cb572010-11-22 17:53:21 +02004671 if (ctxt->exception.vector == PF_VECTOR)
Avi Kivity6389ee92010-11-29 16:12:30 +02004672 kvm_propagate_fault(vcpu, &ctxt->exception);
Avi Kivityda9cb572010-11-22 17:53:21 +02004673 else if (ctxt->exception.error_code_valid)
4674 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
4675 ctxt->exception.error_code);
Gleb Natapov54b84862010-04-28 19:15:44 +03004676 else
Avi Kivityda9cb572010-11-22 17:53:21 +02004677 kvm_queue_exception(vcpu, ctxt->exception.vector);
Gleb Natapov54b84862010-04-28 19:15:44 +03004678}
4679
Avi Kivitydd856ef2012-08-27 23:46:17 +03004680static void init_decode_cache(struct x86_emulate_ctxt *ctxt)
Takuya Yoshikawab5c9ff72011-05-25 11:09:38 +09004681{
Avi Kivity9dac77f2011-06-01 15:34:25 +03004682 memset(&ctxt->twobyte, 0,
Avi Kivitydd856ef2012-08-27 23:46:17 +03004683 (void *)&ctxt->_regs - (void *)&ctxt->twobyte);
Takuya Yoshikawab5c9ff72011-05-25 11:09:38 +09004684
Avi Kivity9dac77f2011-06-01 15:34:25 +03004685 ctxt->fetch.start = 0;
4686 ctxt->fetch.end = 0;
4687 ctxt->io_read.pos = 0;
4688 ctxt->io_read.end = 0;
4689 ctxt->mem_read.pos = 0;
4690 ctxt->mem_read.end = 0;
Takuya Yoshikawab5c9ff72011-05-25 11:09:38 +09004691}
4692
Mohammed Gamal8ec47222010-08-16 00:47:01 +03004693static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
4694{
Takuya Yoshikawaadf52232011-05-25 11:06:16 +09004695 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
Mohammed Gamal8ec47222010-08-16 00:47:01 +03004696 int cs_db, cs_l;
4697
Mohammed Gamal8ec47222010-08-16 00:47:01 +03004698 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4699
Takuya Yoshikawaadf52232011-05-25 11:06:16 +09004700 ctxt->eflags = kvm_get_rflags(vcpu);
4701 ctxt->eip = kvm_rip_read(vcpu);
4702 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
4703 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
4704 cs_l ? X86EMUL_MODE_PROT64 :
4705 cs_db ? X86EMUL_MODE_PROT32 :
4706 X86EMUL_MODE_PROT16;
4707 ctxt->guest_mode = is_guest_mode(vcpu);
4708
Avi Kivitydd856ef2012-08-27 23:46:17 +03004709 init_decode_cache(ctxt);
Gleb Natapov7ae441e2011-03-31 12:06:41 +02004710 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
Mohammed Gamal8ec47222010-08-16 00:47:01 +03004711}
4712
Serge E. Hallyn71f98332011-04-13 09:12:54 -05004713int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
Mohammed Gamal63995652010-09-19 14:34:06 +02004714{
Takuya Yoshikawa9d741912011-05-29 21:53:48 +09004715 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
Mohammed Gamal63995652010-09-19 14:34:06 +02004716 int ret;
4717
4718 init_emulate_ctxt(vcpu);
4719
Avi Kivity9dac77f2011-06-01 15:34:25 +03004720 ctxt->op_bytes = 2;
4721 ctxt->ad_bytes = 2;
4722 ctxt->_eip = ctxt->eip + inc_eip;
Takuya Yoshikawa9d741912011-05-29 21:53:48 +09004723 ret = emulate_int_real(ctxt, irq);
Mohammed Gamal63995652010-09-19 14:34:06 +02004724
4725 if (ret != X86EMUL_CONTINUE)
4726 return EMULATE_FAIL;
4727
Avi Kivity9dac77f2011-06-01 15:34:25 +03004728 ctxt->eip = ctxt->_eip;
Takuya Yoshikawa9d741912011-05-29 21:53:48 +09004729 kvm_rip_write(vcpu, ctxt->eip);
4730 kvm_set_rflags(vcpu, ctxt->eflags);
Mohammed Gamal63995652010-09-19 14:34:06 +02004731
4732 if (irq == NMI_VECTOR)
Avi Kivity7460fb4a2011-09-20 13:43:14 +03004733 vcpu->arch.nmi_pending = 0;
Mohammed Gamal63995652010-09-19 14:34:06 +02004734 else
4735 vcpu->arch.interrupt.pending = false;
4736
4737 return EMULATE_DONE;
4738}
4739EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
4740
Gleb Natapov6d77dbf2010-05-10 11:16:56 +03004741static int handle_emulation_failure(struct kvm_vcpu *vcpu)
4742{
Joerg Roedelfc3a9152010-11-29 17:51:49 +01004743 int r = EMULATE_DONE;
4744
Gleb Natapov6d77dbf2010-05-10 11:16:56 +03004745 ++vcpu->stat.insn_emulation_fail;
4746 trace_kvm_emulate_insn_failed(vcpu);
Joerg Roedelfc3a9152010-11-29 17:51:49 +01004747 if (!is_guest_mode(vcpu)) {
4748 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4749 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4750 vcpu->run->internal.ndata = 0;
4751 r = EMULATE_FAIL;
4752 }
Gleb Natapov6d77dbf2010-05-10 11:16:56 +03004753 kvm_queue_exception(vcpu, UD_VECTOR);
Joerg Roedelfc3a9152010-11-29 17:51:49 +01004754
4755 return r;
Gleb Natapov6d77dbf2010-05-10 11:16:56 +03004756}
4757
Xiao Guangrong93c05d32013-01-13 23:49:07 +08004758static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
4759 bool write_fault_to_shadow_pgtable)
Gleb Natapova6f177e2010-07-08 12:41:12 +03004760{
Xiao Guangrong95b3cf62013-01-13 23:46:52 +08004761 gpa_t gpa = cr2;
Xiao Guangrong8e3d9d02012-08-21 10:57:42 +08004762 pfn_t pfn;
Gleb Natapova6f177e2010-07-08 12:41:12 +03004763
Xiao Guangrong95b3cf62013-01-13 23:46:52 +08004764 if (!vcpu->arch.mmu.direct_map) {
4765 /*
4766 * Write permission should be allowed since only
4767 * write access need to be emulated.
4768 */
4769 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
Gleb Natapov68be0802010-07-14 19:05:45 +03004770
Xiao Guangrong95b3cf62013-01-13 23:46:52 +08004771 /*
4772 * If the mapping is invalid in guest, let cpu retry
4773 * it to generate fault.
4774 */
4775 if (gpa == UNMAPPED_GVA)
4776 return true;
4777 }
Gleb Natapova6f177e2010-07-08 12:41:12 +03004778
Xiao Guangrong8e3d9d02012-08-21 10:57:42 +08004779 /*
4780 * Do not retry the unhandleable instruction if it faults on the
4781 * readonly host memory, otherwise it will goto a infinite loop:
4782 * retry instruction -> write #PF -> emulation fail -> retry
4783 * instruction -> ...
4784 */
4785 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
Xiao Guangrong95b3cf62013-01-13 23:46:52 +08004786
4787 /*
4788 * If the instruction failed on the error pfn, it can not be fixed,
4789 * report the error to userspace.
4790 */
4791 if (is_error_noslot_pfn(pfn))
4792 return false;
4793
4794 kvm_release_pfn_clean(pfn);
4795
4796 /* The instructions are well-emulated on direct mmu. */
4797 if (vcpu->arch.mmu.direct_map) {
4798 unsigned int indirect_shadow_pages;
4799
4800 spin_lock(&vcpu->kvm->mmu_lock);
4801 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
4802 spin_unlock(&vcpu->kvm->mmu_lock);
4803
4804 if (indirect_shadow_pages)
4805 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
4806
Gleb Natapova6f177e2010-07-08 12:41:12 +03004807 return true;
Xiao Guangrong8e3d9d02012-08-21 10:57:42 +08004808 }
Gleb Natapova6f177e2010-07-08 12:41:12 +03004809
Xiao Guangrong95b3cf62013-01-13 23:46:52 +08004810 /*
4811 * if emulation was due to access to shadowed page table
4812 * and it failed try to unshadow page and re-enter the
4813 * guest to let CPU execute the instruction.
4814 */
4815 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
Xiao Guangrong93c05d32013-01-13 23:49:07 +08004816
4817 /*
4818 * If the access faults on its page table, it can not
4819 * be fixed by unprotecting shadow page and it should
4820 * be reported to userspace.
4821 */
4822 return !write_fault_to_shadow_pgtable;
Gleb Natapova6f177e2010-07-08 12:41:12 +03004823}
4824
Xiao Guangrong1cb3f3a2011-09-22 17:02:48 +08004825static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
4826 unsigned long cr2, int emulation_type)
4827{
4828 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4829 unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
4830
4831 last_retry_eip = vcpu->arch.last_retry_eip;
4832 last_retry_addr = vcpu->arch.last_retry_addr;
4833
4834 /*
4835 * If the emulation is caused by #PF and it is non-page_table
4836 * writing instruction, it means the VM-EXIT is caused by shadow
4837 * page protected, we can zap the shadow page and retry this
4838 * instruction directly.
4839 *
4840 * Note: if the guest uses a non-page-table modifying instruction
4841 * on the PDE that points to the instruction, then we will unmap
4842 * the instruction and go to an infinite loop. So, we cache the
4843 * last retried eip and the last fault address, if we meet the eip
4844 * and the address again, we can break out of the potential infinite
4845 * loop.
4846 */
4847 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
4848
4849 if (!(emulation_type & EMULTYPE_RETRY))
4850 return false;
4851
4852 if (x86_page_table_writing_insn(ctxt))
4853 return false;
4854
4855 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
4856 return false;
4857
4858 vcpu->arch.last_retry_eip = ctxt->eip;
4859 vcpu->arch.last_retry_addr = cr2;
4860
4861 if (!vcpu->arch.mmu.direct_map)
4862 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
4863
Xiao Guangrong22368022013-01-13 23:44:12 +08004864 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
Xiao Guangrong1cb3f3a2011-09-22 17:02:48 +08004865
4866 return true;
4867}
4868
Gleb Natapov716d51a2012-09-03 15:24:26 +03004869static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
4870static int complete_emulated_pio(struct kvm_vcpu *vcpu);
4871
Andre Przywara51d8b662010-12-21 11:12:02 +01004872int x86_emulate_instruction(struct kvm_vcpu *vcpu,
4873 unsigned long cr2,
Andre Przywaradc25e892010-12-21 11:12:07 +01004874 int emulation_type,
4875 void *insn,
4876 int insn_len)
Carsten Ottebbd9b642007-10-30 18:44:21 +01004877{
Gleb Natapov95cb2292010-04-28 19:15:43 +03004878 int r;
Takuya Yoshikawa9d741912011-05-29 21:53:48 +09004879 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
Gleb Natapov7ae441e2011-03-31 12:06:41 +02004880 bool writeback = true;
Xiao Guangrong93c05d32013-01-13 23:49:07 +08004881 bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
Carsten Ottebbd9b642007-10-30 18:44:21 +01004882
Xiao Guangrong93c05d32013-01-13 23:49:07 +08004883 /*
4884 * Clear write_fault_to_shadow_pgtable here to ensure it is
4885 * never reused.
4886 */
4887 vcpu->arch.write_fault_to_shadow_pgtable = false;
Avi Kivity26eef702008-07-03 14:59:22 +03004888 kvm_clear_exception_queue(vcpu);
Gleb Natapov8d7d81022011-04-12 12:36:21 +03004889
Sheng Yang571008d2008-01-02 14:49:22 +08004890 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
Mohammed Gamal8ec47222010-08-16 00:47:01 +03004891 init_emulate_ctxt(vcpu);
Takuya Yoshikawa9d741912011-05-29 21:53:48 +09004892 ctxt->interruptibility = 0;
4893 ctxt->have_exception = false;
4894 ctxt->perm_ok = false;
Carsten Ottebbd9b642007-10-30 18:44:21 +01004895
Takuya Yoshikawa9d741912011-05-29 21:53:48 +09004896 ctxt->only_vendor_specific_insn
Avi Kivity40059962011-02-01 16:32:04 +02004897 = emulation_type & EMULTYPE_TRAP_UD;
4898
Takuya Yoshikawa9d741912011-05-29 21:53:48 +09004899 r = x86_decode_insn(ctxt, insn, insn_len);
Joerg Roedeld47f00a2010-09-10 17:30:56 +02004900
Avi Kivitye46479f2010-04-11 13:05:16 +03004901 trace_kvm_emulate_insn_start(vcpu);
Avi Kivityf2b57562007-11-18 15:17:51 +02004902 ++vcpu->stat.insn_emulation;
Takuya Yoshikawa1d2887e2011-07-30 18:03:34 +09004903 if (r != EMULATION_OK) {
Avi Kivity40059962011-02-01 16:32:04 +02004904 if (emulation_type & EMULTYPE_TRAP_UD)
4905 return EMULATE_FAIL;
Xiao Guangrong93c05d32013-01-13 23:49:07 +08004906 if (reexecute_instruction(vcpu, cr2,
4907 write_fault_to_spt))
Carsten Ottebbd9b642007-10-30 18:44:21 +01004908 return EMULATE_DONE;
Gleb Natapov6d77dbf2010-05-10 11:16:56 +03004909 if (emulation_type & EMULTYPE_SKIP)
4910 return EMULATE_FAIL;
4911 return handle_emulation_failure(vcpu);
Carsten Ottebbd9b642007-10-30 18:44:21 +01004912 }
4913 }
4914
Gleb Natapovba8afb62009-04-12 13:36:57 +03004915 if (emulation_type & EMULTYPE_SKIP) {
Avi Kivity9dac77f2011-06-01 15:34:25 +03004916 kvm_rip_write(vcpu, ctxt->_eip);
Gleb Natapovba8afb62009-04-12 13:36:57 +03004917 return EMULATE_DONE;
4918 }
4919
Xiao Guangrong1cb3f3a2011-09-22 17:02:48 +08004920 if (retry_instruction(ctxt, cr2, emulation_type))
4921 return EMULATE_DONE;
4922
Gleb Natapov7ae441e2011-03-31 12:06:41 +02004923 /* this is needed for vmware backdoor interface to work since it
Gleb Natapov4d2179e2010-04-28 19:15:42 +03004924 changes registers values during IO operation */
Gleb Natapov7ae441e2011-03-31 12:06:41 +02004925 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
4926 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
Avi Kivitydd856ef2012-08-27 23:46:17 +03004927 emulator_invalidate_register_cache(ctxt);
Gleb Natapov7ae441e2011-03-31 12:06:41 +02004928 }
Gleb Natapov4d2179e2010-04-28 19:15:42 +03004929
Gleb Natapov5cd21912010-03-18 15:20:26 +02004930restart:
Takuya Yoshikawa9d741912011-05-29 21:53:48 +09004931 r = x86_emulate_insn(ctxt);
Gleb Natapovc3cd7ff2010-04-28 19:15:35 +03004932
Joerg Roedel775fde82011-04-04 12:39:24 +02004933 if (r == EMULATION_INTERCEPTED)
4934 return EMULATE_DONE;
4935
Gleb Natapovd2ddd1c2010-08-25 12:47:43 +03004936 if (r == EMULATION_FAILED) {
Xiao Guangrong93c05d32013-01-13 23:49:07 +08004937 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt))
Gleb Natapovc3cd7ff2010-04-28 19:15:35 +03004938 return EMULATE_DONE;
4939
Gleb Natapov6d77dbf2010-05-10 11:16:56 +03004940 return handle_emulation_failure(vcpu);
Carsten Ottebbd9b642007-10-30 18:44:21 +01004941 }
4942
Takuya Yoshikawa9d741912011-05-29 21:53:48 +09004943 if (ctxt->have_exception) {
Gleb Natapove85d28f2010-07-29 15:11:52 +03004944 inject_emulated_exception(vcpu);
Gleb Natapovd2ddd1c2010-08-25 12:47:43 +03004945 r = EMULATE_DONE;
4946 } else if (vcpu->arch.pio.count) {
Gleb Natapove85d28f2010-07-29 15:11:52 +03004947 if (!vcpu->arch.pio.in)
4948 vcpu->arch.pio.count = 0;
Gleb Natapov716d51a2012-09-03 15:24:26 +03004949 else {
Gleb Natapov7ae441e2011-03-31 12:06:41 +02004950 writeback = false;
Gleb Natapov716d51a2012-09-03 15:24:26 +03004951 vcpu->arch.complete_userspace_io = complete_emulated_pio;
4952 }
Gleb Natapove85d28f2010-07-29 15:11:52 +03004953 r = EMULATE_DO_MMIO;
Gleb Natapov7ae441e2011-03-31 12:06:41 +02004954 } else if (vcpu->mmio_needed) {
4955 if (!vcpu->mmio_is_write)
4956 writeback = false;
Gleb Natapove85d28f2010-07-29 15:11:52 +03004957 r = EMULATE_DO_MMIO;
Gleb Natapov716d51a2012-09-03 15:24:26 +03004958 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
Gleb Natapov7ae441e2011-03-31 12:06:41 +02004959 } else if (r == EMULATION_RESTART)
Gleb Natapove85d28f2010-07-29 15:11:52 +03004960 goto restart;
Gleb Natapovd2ddd1c2010-08-25 12:47:43 +03004961 else
4962 r = EMULATE_DONE;
Gleb Natapove85d28f2010-07-29 15:11:52 +03004963
Gleb Natapov7ae441e2011-03-31 12:06:41 +02004964 if (writeback) {
Takuya Yoshikawa9d741912011-05-29 21:53:48 +09004965 toggle_interruptibility(vcpu, ctxt->interruptibility);
4966 kvm_set_rflags(vcpu, ctxt->eflags);
Gleb Natapov7ae441e2011-03-31 12:06:41 +02004967 kvm_make_request(KVM_REQ_EVENT, vcpu);
Gleb Natapov7ae441e2011-03-31 12:06:41 +02004968 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
Takuya Yoshikawa9d741912011-05-29 21:53:48 +09004969 kvm_rip_write(vcpu, ctxt->eip);
Gleb Natapov7ae441e2011-03-31 12:06:41 +02004970 } else
4971 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
Gleb Natapov3457e412010-04-28 19:15:38 +03004972
Gleb Natapove85d28f2010-07-29 15:11:52 +03004973 return r;
Carsten Ottebbd9b642007-10-30 18:44:21 +01004974}
Andre Przywara51d8b662010-12-21 11:12:02 +01004975EXPORT_SYMBOL_GPL(x86_emulate_instruction);
Carsten Ottebbd9b642007-10-30 18:44:21 +01004976
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02004977int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
Carsten Ottede7d7892007-10-30 18:44:25 +01004978{
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02004979 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
Avi Kivityca1d4a92011-04-20 13:37:53 +03004980 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
4981 size, port, &val, 1);
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02004982 /* do not return to emulator after return from userspace */
Gleb Natapov79729952010-03-18 15:20:24 +02004983 vcpu->arch.pio.count = 0;
Izik Eidus0f346072008-12-29 01:42:20 +02004984 return ret;
Carsten Ottede7d7892007-10-30 18:44:25 +01004985}
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02004986EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
Carsten Ottede7d7892007-10-30 18:44:25 +01004987
Zachary Amsden8cfdc002010-08-19 22:07:21 -10004988static void tsc_bad(void *info)
Gerd Hoffmannc8076602009-02-04 17:52:04 +01004989{
Tejun Heo0a3aee02010-12-18 16:28:55 +01004990 __this_cpu_write(cpu_tsc_khz, 0);
Zachary Amsden8cfdc002010-08-19 22:07:21 -10004991}
4992
4993static void tsc_khz_changed(void *data)
4994{
4995 struct cpufreq_freqs *freq = data;
4996 unsigned long khz = 0;
4997
4998 if (data)
4999 khz = freq->new;
5000 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5001 khz = cpufreq_quick_get(raw_smp_processor_id());
5002 if (!khz)
5003 khz = tsc_khz;
Tejun Heo0a3aee02010-12-18 16:28:55 +01005004 __this_cpu_write(cpu_tsc_khz, khz);
Gerd Hoffmannc8076602009-02-04 17:52:04 +01005005}
5006
Gerd Hoffmannc8076602009-02-04 17:52:04 +01005007static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
5008 void *data)
5009{
5010 struct cpufreq_freqs *freq = data;
5011 struct kvm *kvm;
5012 struct kvm_vcpu *vcpu;
5013 int i, send_ipi = 0;
5014
Zachary Amsden8cfdc002010-08-19 22:07:21 -10005015 /*
5016 * We allow guests to temporarily run on slowing clocks,
5017 * provided we notify them after, or to run on accelerating
5018 * clocks, provided we notify them before. Thus time never
5019 * goes backwards.
5020 *
5021 * However, we have a problem. We can't atomically update
5022 * the frequency of a given CPU from this function; it is
5023 * merely a notifier, which can be called from any CPU.
5024 * Changing the TSC frequency at arbitrary points in time
5025 * requires a recomputation of local variables related to
5026 * the TSC for each VCPU. We must flag these local variables
5027 * to be updated and be sure the update takes place with the
5028 * new frequency before any guests proceed.
5029 *
5030 * Unfortunately, the combination of hotplug CPU and frequency
5031 * change creates an intractable locking scenario; the order
5032 * of when these callouts happen is undefined with respect to
5033 * CPU hotplug, and they can race with each other. As such,
5034 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
5035 * undefined; you can actually have a CPU frequency change take
5036 * place in between the computation of X and the setting of the
5037 * variable. To protect against this problem, all updates of
5038 * the per_cpu tsc_khz variable are done in an interrupt
5039 * protected IPI, and all callers wishing to update the value
5040 * must wait for a synchronous IPI to complete (which is trivial
5041 * if the caller is on the CPU already). This establishes the
5042 * necessary total order on variable updates.
5043 *
5044 * Note that because a guest time update may take place
5045 * anytime after the setting of the VCPU's request bit, the
5046 * correct TSC value must be set before the request. However,
5047 * to ensure the update actually makes it to any guest which
5048 * starts running in hardware virtualization between the set
5049 * and the acquisition of the spinlock, we must also ping the
5050 * CPU after setting the request bit.
5051 *
5052 */
5053
Gerd Hoffmannc8076602009-02-04 17:52:04 +01005054 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
5055 return 0;
5056 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
5057 return 0;
Zachary Amsden8cfdc002010-08-19 22:07:21 -10005058
5059 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
Gerd Hoffmannc8076602009-02-04 17:52:04 +01005060
Jan Kiszkae935b832011-02-08 12:55:33 +01005061 raw_spin_lock(&kvm_lock);
Gerd Hoffmannc8076602009-02-04 17:52:04 +01005062 list_for_each_entry(kvm, &vm_list, vm_list) {
Gleb Natapov988a2ca2009-06-09 15:56:29 +03005063 kvm_for_each_vcpu(i, vcpu, kvm) {
Gerd Hoffmannc8076602009-02-04 17:52:04 +01005064 if (vcpu->cpu != freq->cpu)
5065 continue;
Zachary Amsdenc2855452010-09-18 14:38:15 -10005066 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
Gerd Hoffmannc8076602009-02-04 17:52:04 +01005067 if (vcpu->cpu != smp_processor_id())
Zachary Amsden8cfdc002010-08-19 22:07:21 -10005068 send_ipi = 1;
Gerd Hoffmannc8076602009-02-04 17:52:04 +01005069 }
5070 }
Jan Kiszkae935b832011-02-08 12:55:33 +01005071 raw_spin_unlock(&kvm_lock);
Gerd Hoffmannc8076602009-02-04 17:52:04 +01005072
5073 if (freq->old < freq->new && send_ipi) {
5074 /*
5075 * We upscale the frequency. Must make the guest
5076 * doesn't see old kvmclock values while running with
5077 * the new frequency, otherwise we risk the guest sees
5078 * time go backwards.
5079 *
5080 * In case we update the frequency for another cpu
5081 * (which might be in guest context) send an interrupt
5082 * to kick the cpu out of guest context. Next time
5083 * guest context is entered kvmclock will be updated,
5084 * so the guest will not see stale values.
5085 */
Zachary Amsden8cfdc002010-08-19 22:07:21 -10005086 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
Gerd Hoffmannc8076602009-02-04 17:52:04 +01005087 }
5088 return 0;
5089}
5090
5091static struct notifier_block kvmclock_cpufreq_notifier_block = {
Zachary Amsden8cfdc002010-08-19 22:07:21 -10005092 .notifier_call = kvmclock_cpufreq_notifier
5093};
5094
5095static int kvmclock_cpu_notifier(struct notifier_block *nfb,
5096 unsigned long action, void *hcpu)
5097{
5098 unsigned int cpu = (unsigned long)hcpu;
5099
5100 switch (action) {
5101 case CPU_ONLINE:
5102 case CPU_DOWN_FAILED:
5103 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5104 break;
5105 case CPU_DOWN_PREPARE:
5106 smp_call_function_single(cpu, tsc_bad, NULL, 1);
5107 break;
5108 }
5109 return NOTIFY_OK;
5110}
5111
5112static struct notifier_block kvmclock_cpu_notifier_block = {
5113 .notifier_call = kvmclock_cpu_notifier,
5114 .priority = -INT_MAX
Gerd Hoffmannc8076602009-02-04 17:52:04 +01005115};
5116
Zachary Amsdenb820cc02009-09-29 11:38:34 -10005117static void kvm_timer_init(void)
5118{
5119 int cpu;
5120
Zachary Amsdenc2855452010-09-18 14:38:15 -10005121 max_tsc_khz = tsc_khz;
Zachary Amsden8cfdc002010-08-19 22:07:21 -10005122 register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
Zachary Amsdenb820cc02009-09-29 11:38:34 -10005123 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
Zachary Amsdenc2855452010-09-18 14:38:15 -10005124#ifdef CONFIG_CPU_FREQ
5125 struct cpufreq_policy policy;
5126 memset(&policy, 0, sizeof(policy));
Avi Kivity3e26f232010-12-16 12:16:34 +02005127 cpu = get_cpu();
5128 cpufreq_get_policy(&policy, cpu);
Zachary Amsdenc2855452010-09-18 14:38:15 -10005129 if (policy.cpuinfo.max_freq)
5130 max_tsc_khz = policy.cpuinfo.max_freq;
Avi Kivity3e26f232010-12-16 12:16:34 +02005131 put_cpu();
Zachary Amsdenc2855452010-09-18 14:38:15 -10005132#endif
Zachary Amsdenb820cc02009-09-29 11:38:34 -10005133 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
5134 CPUFREQ_TRANSITION_NOTIFIER);
5135 }
Zachary Amsdenc2855452010-09-18 14:38:15 -10005136 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
Zachary Amsden8cfdc002010-08-19 22:07:21 -10005137 for_each_online_cpu(cpu)
5138 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
Zachary Amsdenb820cc02009-09-29 11:38:34 -10005139}
5140
Zhang, Yanminff9d07a2010-04-19 13:32:45 +08005141static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5142
Gleb Natapovf5132b02011-11-10 14:57:22 +02005143int kvm_is_in_guest(void)
Zhang, Yanminff9d07a2010-04-19 13:32:45 +08005144{
Alex,Shi086c9852011-10-20 15:34:01 +08005145 return __this_cpu_read(current_vcpu) != NULL;
Zhang, Yanminff9d07a2010-04-19 13:32:45 +08005146}
5147
5148static int kvm_is_user_mode(void)
5149{
5150 int user_mode = 3;
Zhang, Yanmindcf46b92010-04-20 10:13:58 +08005151
Alex,Shi086c9852011-10-20 15:34:01 +08005152 if (__this_cpu_read(current_vcpu))
5153 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
Zhang, Yanmindcf46b92010-04-20 10:13:58 +08005154
Zhang, Yanminff9d07a2010-04-19 13:32:45 +08005155 return user_mode != 0;
5156}
5157
5158static unsigned long kvm_get_guest_ip(void)
5159{
5160 unsigned long ip = 0;
Zhang, Yanmindcf46b92010-04-20 10:13:58 +08005161
Alex,Shi086c9852011-10-20 15:34:01 +08005162 if (__this_cpu_read(current_vcpu))
5163 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
Zhang, Yanmindcf46b92010-04-20 10:13:58 +08005164
Zhang, Yanminff9d07a2010-04-19 13:32:45 +08005165 return ip;
5166}
5167
5168static struct perf_guest_info_callbacks kvm_guest_cbs = {
5169 .is_in_guest = kvm_is_in_guest,
5170 .is_user_mode = kvm_is_user_mode,
5171 .get_guest_ip = kvm_get_guest_ip,
5172};
5173
5174void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
5175{
Alex,Shi086c9852011-10-20 15:34:01 +08005176 __this_cpu_write(current_vcpu, vcpu);
Zhang, Yanminff9d07a2010-04-19 13:32:45 +08005177}
5178EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
5179
5180void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
5181{
Alex,Shi086c9852011-10-20 15:34:01 +08005182 __this_cpu_write(current_vcpu, NULL);
Zhang, Yanminff9d07a2010-04-19 13:32:45 +08005183}
5184EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
5185
Xiao Guangrongce88dec2011-07-12 03:33:44 +08005186static void kvm_set_mmio_spte_mask(void)
5187{
5188 u64 mask;
5189 int maxphyaddr = boot_cpu_data.x86_phys_bits;
5190
5191 /*
5192 * Set the reserved bits and the present bit of an paging-structure
5193 * entry to generate page fault with PFER.RSV = 1.
5194 */
5195 mask = ((1ull << (62 - maxphyaddr + 1)) - 1) << maxphyaddr;
5196 mask |= 1ull;
5197
5198#ifdef CONFIG_X86_64
5199 /*
5200 * If reserved bit is not supported, clear the present bit to disable
5201 * mmio page fault.
5202 */
5203 if (maxphyaddr == 52)
5204 mask &= ~1ull;
5205#endif
5206
5207 kvm_mmu_set_mmio_spte_mask(mask);
5208}
5209
Marcelo Tosatti16e8d742012-11-27 23:29:00 -02005210#ifdef CONFIG_X86_64
5211static void pvclock_gtod_update_fn(struct work_struct *work)
5212{
Marcelo Tosattid8281992012-11-27 23:29:01 -02005213 struct kvm *kvm;
5214
5215 struct kvm_vcpu *vcpu;
5216 int i;
5217
5218 raw_spin_lock(&kvm_lock);
5219 list_for_each_entry(kvm, &vm_list, vm_list)
5220 kvm_for_each_vcpu(i, vcpu, kvm)
5221 set_bit(KVM_REQ_MASTERCLOCK_UPDATE, &vcpu->requests);
5222 atomic_set(&kvm_guest_has_master_clock, 0);
5223 raw_spin_unlock(&kvm_lock);
Marcelo Tosatti16e8d742012-11-27 23:29:00 -02005224}
5225
5226static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
5227
5228/*
5229 * Notification about pvclock gtod data update.
5230 */
5231static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
5232 void *priv)
5233{
5234 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
5235 struct timekeeper *tk = priv;
5236
5237 update_pvclock_gtod(tk);
5238
5239 /* disable master clock if host does not trust, or does not
5240 * use, TSC clocksource
5241 */
5242 if (gtod->clock.vclock_mode != VCLOCK_TSC &&
5243 atomic_read(&kvm_guest_has_master_clock) != 0)
5244 queue_work(system_long_wq, &pvclock_gtod_work);
5245
5246 return 0;
5247}
5248
5249static struct notifier_block pvclock_gtod_notifier = {
5250 .notifier_call = pvclock_gtod_notify,
5251};
5252#endif
5253
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08005254int kvm_arch_init(void *opaque)
Carsten Otte043405e2007-10-10 17:16:19 +02005255{
Zachary Amsdenb820cc02009-09-29 11:38:34 -10005256 int r;
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08005257 struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
5258
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08005259 if (kvm_x86_ops) {
5260 printk(KERN_ERR "kvm: already loaded the other module\n");
Zhang Xiantao56c6d282007-11-18 20:43:21 +08005261 r = -EEXIST;
5262 goto out;
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08005263 }
5264
5265 if (!ops->cpu_has_kvm_support()) {
5266 printk(KERN_ERR "kvm: no hardware support\n");
Zhang Xiantao56c6d282007-11-18 20:43:21 +08005267 r = -EOPNOTSUPP;
5268 goto out;
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08005269 }
5270 if (ops->disabled_by_bios()) {
5271 printk(KERN_ERR "kvm: disabled by bios\n");
Zhang Xiantao56c6d282007-11-18 20:43:21 +08005272 r = -EOPNOTSUPP;
5273 goto out;
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08005274 }
5275
Marcelo Tosatti013f6a52013-01-03 11:41:39 -02005276 r = -ENOMEM;
5277 shared_msrs = alloc_percpu(struct kvm_shared_msrs);
5278 if (!shared_msrs) {
5279 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
5280 goto out;
5281 }
5282
Avi Kivity97db56c2008-01-13 13:23:56 +02005283 r = kvm_mmu_module_init();
5284 if (r)
Marcelo Tosatti013f6a52013-01-03 11:41:39 -02005285 goto out_free_percpu;
Avi Kivity97db56c2008-01-13 13:23:56 +02005286
Xiao Guangrongce88dec2011-07-12 03:33:44 +08005287 kvm_set_mmio_spte_mask();
Avi Kivity97db56c2008-01-13 13:23:56 +02005288 kvm_init_msr_list();
5289
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08005290 kvm_x86_ops = ops;
Sheng Yang7b523452008-04-25 21:13:50 +08005291 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
Sheng Yang4b12f0d2009-04-27 20:35:42 +08005292 PT_DIRTY_MASK, PT64_NX_MASK, 0);
Gerd Hoffmannc8076602009-02-04 17:52:04 +01005293
Zachary Amsdenb820cc02009-09-29 11:38:34 -10005294 kvm_timer_init();
Gerd Hoffmannc8076602009-02-04 17:52:04 +01005295
Zhang, Yanminff9d07a2010-04-19 13:32:45 +08005296 perf_register_guest_info_callbacks(&kvm_guest_cbs);
5297
Dexuan Cui2acf9232010-06-10 11:27:12 +08005298 if (cpu_has_xsave)
5299 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
5300
Gleb Natapovc5cc4212012-08-05 15:58:30 +03005301 kvm_lapic_init();
Marcelo Tosatti16e8d742012-11-27 23:29:00 -02005302#ifdef CONFIG_X86_64
5303 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
5304#endif
5305
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08005306 return 0;
Zhang Xiantao56c6d282007-11-18 20:43:21 +08005307
Marcelo Tosatti013f6a52013-01-03 11:41:39 -02005308out_free_percpu:
5309 free_percpu(shared_msrs);
Zhang Xiantao56c6d282007-11-18 20:43:21 +08005310out:
Zhang Xiantao56c6d282007-11-18 20:43:21 +08005311 return r;
Carsten Otte043405e2007-10-10 17:16:19 +02005312}
Hollis Blanchard8776e512007-10-31 17:24:24 -05005313
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08005314void kvm_arch_exit(void)
5315{
Zhang, Yanminff9d07a2010-04-19 13:32:45 +08005316 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
5317
Jan Kiszka888d2562009-04-17 19:24:58 +02005318 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5319 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
5320 CPUFREQ_TRANSITION_NOTIFIER);
Zachary Amsden8cfdc002010-08-19 22:07:21 -10005321 unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
Marcelo Tosatti16e8d742012-11-27 23:29:00 -02005322#ifdef CONFIG_X86_64
5323 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
5324#endif
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08005325 kvm_x86_ops = NULL;
Zhang Xiantao56c6d282007-11-18 20:43:21 +08005326 kvm_mmu_module_exit();
Marcelo Tosatti013f6a52013-01-03 11:41:39 -02005327 free_percpu(shared_msrs);
Zhang Xiantao56c6d282007-11-18 20:43:21 +08005328}
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08005329
Hollis Blanchard8776e512007-10-31 17:24:24 -05005330int kvm_emulate_halt(struct kvm_vcpu *vcpu)
5331{
5332 ++vcpu->stat.halt_exits;
5333 if (irqchip_in_kernel(vcpu->kvm)) {
Avi Kivitya4535292008-04-13 17:54:35 +03005334 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
Hollis Blanchard8776e512007-10-31 17:24:24 -05005335 return 1;
5336 } else {
5337 vcpu->run->exit_reason = KVM_EXIT_HLT;
5338 return 0;
5339 }
5340}
5341EXPORT_SYMBOL_GPL(kvm_emulate_halt);
5342
Gleb Natapov55cd8e52010-01-17 15:51:22 +02005343int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
5344{
5345 u64 param, ingpa, outgpa, ret;
5346 uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
5347 bool fast, longmode;
5348 int cs_db, cs_l;
5349
5350 /*
5351 * hypercall generates UD from non zero cpl and real mode
5352 * per HYPER-V spec
5353 */
Avi Kivity3eeb3282010-01-21 15:31:48 +02005354 if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
Gleb Natapov55cd8e52010-01-17 15:51:22 +02005355 kvm_queue_exception(vcpu, UD_VECTOR);
5356 return 0;
5357 }
5358
5359 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5360 longmode = is_long_mode(vcpu) && cs_l == 1;
5361
5362 if (!longmode) {
Gleb Natapovccd46932010-01-19 15:06:38 +02005363 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
5364 (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
5365 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
5366 (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
5367 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
5368 (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
Gleb Natapov55cd8e52010-01-17 15:51:22 +02005369 }
5370#ifdef CONFIG_X86_64
5371 else {
5372 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
5373 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
5374 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
5375 }
5376#endif
5377
5378 code = param & 0xffff;
5379 fast = (param >> 16) & 0x1;
5380 rep_cnt = (param >> 32) & 0xfff;
5381 rep_idx = (param >> 48) & 0xfff;
5382
5383 trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
5384
Gleb Natapovc25bc162010-01-17 15:51:24 +02005385 switch (code) {
5386 case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
5387 kvm_vcpu_on_spin(vcpu);
5388 break;
5389 default:
5390 res = HV_STATUS_INVALID_HYPERCALL_CODE;
5391 break;
5392 }
Gleb Natapov55cd8e52010-01-17 15:51:22 +02005393
5394 ret = res | (((u64)rep_done & 0xfff) << 32);
5395 if (longmode) {
5396 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5397 } else {
5398 kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
5399 kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
5400 }
5401
5402 return 1;
5403}
5404
Hollis Blanchard8776e512007-10-31 17:24:24 -05005405int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
5406{
5407 unsigned long nr, a0, a1, a2, a3, ret;
Marcelo Tosatti2f333bc2008-02-22 12:21:37 -05005408 int r = 1;
Hollis Blanchard8776e512007-10-31 17:24:24 -05005409
Gleb Natapov55cd8e52010-01-17 15:51:22 +02005410 if (kvm_hv_hypercall_enabled(vcpu->kvm))
5411 return kvm_hv_hypercall(vcpu);
5412
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03005413 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
5414 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
5415 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
5416 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
5417 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
Hollis Blanchard8776e512007-10-31 17:24:24 -05005418
Marcelo Tosatti229456f2009-06-17 09:22:14 -03005419 trace_kvm_hypercall(nr, a0, a1, a2, a3);
Feng (Eric) Liu2714d1d2008-04-10 15:31:10 -04005420
Hollis Blanchard8776e512007-10-31 17:24:24 -05005421 if (!is_long_mode(vcpu)) {
5422 nr &= 0xFFFFFFFF;
5423 a0 &= 0xFFFFFFFF;
5424 a1 &= 0xFFFFFFFF;
5425 a2 &= 0xFFFFFFFF;
5426 a3 &= 0xFFFFFFFF;
5427 }
5428
Jan Kiszka07708c42009-08-03 18:43:28 +02005429 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
5430 ret = -KVM_EPERM;
5431 goto out;
5432 }
5433
Hollis Blanchard8776e512007-10-31 17:24:24 -05005434 switch (nr) {
Avi Kivityb93463a2007-10-25 16:52:32 +02005435 case KVM_HC_VAPIC_POLL_IRQ:
5436 ret = 0;
5437 break;
Hollis Blanchard8776e512007-10-31 17:24:24 -05005438 default:
5439 ret = -KVM_ENOSYS;
5440 break;
5441 }
Jan Kiszka07708c42009-08-03 18:43:28 +02005442out:
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03005443 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
Amit Shahf11c3a82008-02-21 01:00:30 +05305444 ++vcpu->stat.hypercalls;
Marcelo Tosatti2f333bc2008-02-22 12:21:37 -05005445 return r;
Hollis Blanchard8776e512007-10-31 17:24:24 -05005446}
5447EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
5448
Jan Kiszkab6785de2012-09-20 07:43:17 +02005449static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
Hollis Blanchard8776e512007-10-31 17:24:24 -05005450{
Avi Kivityd6aa1002011-04-20 15:47:13 +03005451 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
Hollis Blanchard8776e512007-10-31 17:24:24 -05005452 char instruction[3];
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03005453 unsigned long rip = kvm_rip_read(vcpu);
Hollis Blanchard8776e512007-10-31 17:24:24 -05005454
Hollis Blanchard8776e512007-10-31 17:24:24 -05005455 /*
5456 * Blow out the MMU to ensure that no other VCPU has an active mapping
5457 * to ensure that the updated hypercall appears atomically across all
5458 * VCPUs.
5459 */
5460 kvm_mmu_zap_all(vcpu->kvm);
5461
Hollis Blanchard8776e512007-10-31 17:24:24 -05005462 kvm_x86_ops->patch_hypercall(vcpu, instruction);
Hollis Blanchard8776e512007-10-31 17:24:24 -05005463
Takuya Yoshikawa9d741912011-05-29 21:53:48 +09005464 return emulator_write_emulated(ctxt, rip, instruction, 3, NULL);
Hollis Blanchard8776e512007-10-31 17:24:24 -05005465}
5466
Hollis Blanchardd0752062007-10-31 17:24:25 -05005467/*
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005468 * Check if userspace requested an interrupt window, and that the
5469 * interrupt window is open.
5470 *
5471 * No need to exit to userspace if we already have an interrupt queued.
5472 */
Avi Kivity851ba692009-08-24 11:10:17 +03005473static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005474{
Gleb Natapov80618232009-04-21 17:44:56 +03005475 return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
Avi Kivity851ba692009-08-24 11:10:17 +03005476 vcpu->run->request_interrupt_window &&
Gleb Natapov5df56642009-04-21 17:44:59 +03005477 kvm_arch_interrupt_allowed(vcpu));
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005478}
5479
Avi Kivity851ba692009-08-24 11:10:17 +03005480static void post_kvm_run_save(struct kvm_vcpu *vcpu)
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005481{
Avi Kivity851ba692009-08-24 11:10:17 +03005482 struct kvm_run *kvm_run = vcpu->run;
5483
Jan Kiszka91586a32009-10-05 13:07:21 +02005484 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
Avi Kivity2d3ad1f2008-02-24 11:20:43 +02005485 kvm_run->cr8 = kvm_get_cr8(vcpu);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005486 kvm_run->apic_base = kvm_get_apic_base(vcpu);
Jan Kiszka45312202008-12-11 16:54:54 +01005487 if (irqchip_in_kernel(vcpu->kvm))
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005488 kvm_run->ready_for_interrupt_injection = 1;
Jan Kiszka45312202008-12-11 16:54:54 +01005489 else
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005490 kvm_run->ready_for_interrupt_injection =
Gleb Natapovfa9726b2009-05-11 13:35:47 +03005491 kvm_arch_interrupt_allowed(vcpu) &&
5492 !kvm_cpu_has_interrupt(vcpu) &&
5493 !kvm_event_needs_reinjection(vcpu);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005494}
5495
Xiao Guangrong44841412012-09-07 14:14:20 +08005496static int vapic_enter(struct kvm_vcpu *vcpu)
Avi Kivityb93463a2007-10-25 16:52:32 +02005497{
5498 struct kvm_lapic *apic = vcpu->arch.apic;
5499 struct page *page;
5500
5501 if (!apic || !apic->vapic_addr)
Xiao Guangrong44841412012-09-07 14:14:20 +08005502 return 0;
Avi Kivityb93463a2007-10-25 16:52:32 +02005503
5504 page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
Xiao Guangrong44841412012-09-07 14:14:20 +08005505 if (is_error_page(page))
5506 return -EFAULT;
Izik Eidus72dc67a2008-02-10 18:04:15 +02005507
5508 vcpu->arch.apic->vapic_page = page;
Xiao Guangrong44841412012-09-07 14:14:20 +08005509 return 0;
Avi Kivityb93463a2007-10-25 16:52:32 +02005510}
5511
5512static void vapic_exit(struct kvm_vcpu *vcpu)
5513{
5514 struct kvm_lapic *apic = vcpu->arch.apic;
Marcelo Tosattif656ce02009-12-23 14:35:25 -02005515 int idx;
Avi Kivityb93463a2007-10-25 16:52:32 +02005516
5517 if (!apic || !apic->vapic_addr)
5518 return;
5519
Marcelo Tosattif656ce02009-12-23 14:35:25 -02005520 idx = srcu_read_lock(&vcpu->kvm->srcu);
Avi Kivityb93463a2007-10-25 16:52:32 +02005521 kvm_release_page_dirty(apic->vapic_page);
5522 mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
Marcelo Tosattif656ce02009-12-23 14:35:25 -02005523 srcu_read_unlock(&vcpu->kvm->srcu, idx);
Avi Kivityb93463a2007-10-25 16:52:32 +02005524}
5525
Gleb Natapov95ba8273132009-04-21 17:45:08 +03005526static void update_cr8_intercept(struct kvm_vcpu *vcpu)
5527{
5528 int max_irr, tpr;
5529
5530 if (!kvm_x86_ops->update_cr8_intercept)
5531 return;
5532
Avi Kivity88c808f2009-08-17 22:49:40 +03005533 if (!vcpu->arch.apic)
5534 return;
5535
Gleb Natapov8db3baa2009-05-11 13:35:54 +03005536 if (!vcpu->arch.apic->vapic_addr)
5537 max_irr = kvm_lapic_find_highest_irr(vcpu);
5538 else
5539 max_irr = -1;
Gleb Natapov95ba8273132009-04-21 17:45:08 +03005540
5541 if (max_irr != -1)
5542 max_irr >>= 4;
5543
5544 tpr = kvm_lapic_get_cr8(vcpu);
5545
5546 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
5547}
5548
Avi Kivity851ba692009-08-24 11:10:17 +03005549static void inject_pending_event(struct kvm_vcpu *vcpu)
Gleb Natapov95ba8273132009-04-21 17:45:08 +03005550{
5551 /* try to reinject previous events if any */
Gleb Natapovb59bb7b2009-07-09 15:33:51 +03005552 if (vcpu->arch.exception.pending) {
Avi Kivity5c1c85d2010-03-11 13:01:59 +02005553 trace_kvm_inj_exception(vcpu->arch.exception.nr,
5554 vcpu->arch.exception.has_error_code,
5555 vcpu->arch.exception.error_code);
Gleb Natapovb59bb7b2009-07-09 15:33:51 +03005556 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
5557 vcpu->arch.exception.has_error_code,
Joerg Roedelce7ddec2010-04-22 12:33:13 +02005558 vcpu->arch.exception.error_code,
5559 vcpu->arch.exception.reinject);
Gleb Natapovb59bb7b2009-07-09 15:33:51 +03005560 return;
5561 }
5562
Gleb Natapov95ba8273132009-04-21 17:45:08 +03005563 if (vcpu->arch.nmi_injected) {
5564 kvm_x86_ops->set_nmi(vcpu);
5565 return;
5566 }
5567
5568 if (vcpu->arch.interrupt.pending) {
Gleb Natapov66fd3f72009-05-11 13:35:50 +03005569 kvm_x86_ops->set_irq(vcpu);
Gleb Natapov95ba8273132009-04-21 17:45:08 +03005570 return;
5571 }
5572
5573 /* try to inject new event if pending */
5574 if (vcpu->arch.nmi_pending) {
5575 if (kvm_x86_ops->nmi_allowed(vcpu)) {
Avi Kivity7460fb4a2011-09-20 13:43:14 +03005576 --vcpu->arch.nmi_pending;
Gleb Natapov95ba8273132009-04-21 17:45:08 +03005577 vcpu->arch.nmi_injected = true;
5578 kvm_x86_ops->set_nmi(vcpu);
5579 }
Yang Zhangc7c9c562013-01-25 10:18:51 +08005580 } else if (kvm_cpu_has_injectable_intr(vcpu)) {
Gleb Natapov95ba8273132009-04-21 17:45:08 +03005581 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
Gleb Natapov66fd3f72009-05-11 13:35:50 +03005582 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
5583 false);
5584 kvm_x86_ops->set_irq(vcpu);
Gleb Natapov95ba8273132009-04-21 17:45:08 +03005585 }
5586 }
5587}
5588
Dexuan Cui2acf9232010-06-10 11:27:12 +08005589static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
5590{
5591 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
5592 !vcpu->guest_xcr0_loaded) {
5593 /* kvm_set_xcr() also depends on this */
5594 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
5595 vcpu->guest_xcr0_loaded = 1;
5596 }
5597}
5598
5599static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
5600{
5601 if (vcpu->guest_xcr0_loaded) {
5602 if (vcpu->arch.xcr0 != host_xcr0)
5603 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
5604 vcpu->guest_xcr0_loaded = 0;
5605 }
5606}
5607
Avi Kivity7460fb4a2011-09-20 13:43:14 +03005608static void process_nmi(struct kvm_vcpu *vcpu)
5609{
5610 unsigned limit = 2;
5611
5612 /*
5613 * x86 is limited to one NMI running, and one NMI pending after it.
5614 * If an NMI is already in progress, limit further NMIs to just one.
5615 * Otherwise, allow two (and we'll inject the first one immediately).
5616 */
5617 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
5618 limit = 1;
5619
5620 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
5621 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
5622 kvm_make_request(KVM_REQ_EVENT, vcpu);
5623}
5624
Marcelo Tosattid8281992012-11-27 23:29:01 -02005625static void kvm_gen_update_masterclock(struct kvm *kvm)
5626{
5627#ifdef CONFIG_X86_64
5628 int i;
5629 struct kvm_vcpu *vcpu;
5630 struct kvm_arch *ka = &kvm->arch;
5631
5632 spin_lock(&ka->pvclock_gtod_sync_lock);
5633 kvm_make_mclock_inprogress_request(kvm);
5634 /* no guest entries from this point */
5635 pvclock_update_vm_gtod_copy(kvm);
5636
5637 kvm_for_each_vcpu(i, vcpu, kvm)
5638 set_bit(KVM_REQ_CLOCK_UPDATE, &vcpu->requests);
5639
5640 /* guest entries allowed */
5641 kvm_for_each_vcpu(i, vcpu, kvm)
5642 clear_bit(KVM_REQ_MCLOCK_INPROGRESS, &vcpu->requests);
5643
5644 spin_unlock(&ka->pvclock_gtod_sync_lock);
5645#endif
5646}
5647
Yang Zhangc7c9c562013-01-25 10:18:51 +08005648static void update_eoi_exitmap(struct kvm_vcpu *vcpu)
5649{
5650 u64 eoi_exit_bitmap[4];
5651
5652 memset(eoi_exit_bitmap, 0, 32);
5653
5654 kvm_ioapic_calculate_eoi_exitmap(vcpu, eoi_exit_bitmap);
5655 kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
5656}
5657
Avi Kivity851ba692009-08-24 11:10:17 +03005658static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005659{
5660 int r;
Gleb Natapov6a8b1d12009-05-11 13:35:51 +03005661 bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
Avi Kivity851ba692009-08-24 11:10:17 +03005662 vcpu->run->request_interrupt_window;
Nadav Har'Eld6185f22011-09-22 13:52:56 +03005663 bool req_immediate_exit = 0;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005664
Avi Kivity3e007502010-06-23 14:26:18 +03005665 if (vcpu->requests) {
Avi Kivitya8eeb042010-05-10 12:34:53 +03005666 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
Marcelo Tosatti2e53d632008-02-20 14:47:24 -05005667 kvm_mmu_unload(vcpu);
Avi Kivitya8eeb042010-05-10 12:34:53 +03005668 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
Marcelo Tosatti2f599712008-05-27 12:10:20 -03005669 __kvm_migrate_timers(vcpu);
Marcelo Tosattid8281992012-11-27 23:29:01 -02005670 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
5671 kvm_gen_update_masterclock(vcpu->kvm);
Zachary Amsden34c238a2010-09-18 14:38:14 -10005672 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
5673 r = kvm_guest_time_update(vcpu);
Zachary Amsden8cfdc002010-08-19 22:07:21 -10005674 if (unlikely(r))
5675 goto out;
5676 }
Avi Kivitya8eeb042010-05-10 12:34:53 +03005677 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
Marcelo Tosatti4731d4c2008-09-23 13:18:39 -03005678 kvm_mmu_sync_roots(vcpu);
Avi Kivitya8eeb042010-05-10 12:34:53 +03005679 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
Marcelo Tosattid4acf7e2008-06-06 16:37:35 -03005680 kvm_x86_ops->tlb_flush(vcpu);
Avi Kivitya8eeb042010-05-10 12:34:53 +03005681 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
Avi Kivity851ba692009-08-24 11:10:17 +03005682 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
Avi Kivityb93463a2007-10-25 16:52:32 +02005683 r = 0;
5684 goto out;
5685 }
Avi Kivitya8eeb042010-05-10 12:34:53 +03005686 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
Avi Kivity851ba692009-08-24 11:10:17 +03005687 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
Joerg Roedel71c4dfa2008-02-26 16:49:16 +01005688 r = 0;
5689 goto out;
5690 }
Avi Kivitya8eeb042010-05-10 12:34:53 +03005691 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
Avi Kivity02daab22009-12-30 12:40:26 +02005692 vcpu->fpu_active = 0;
5693 kvm_x86_ops->fpu_deactivate(vcpu);
5694 }
Gleb Natapovaf585b92010-10-14 11:22:46 +02005695 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
5696 /* Page is swapped out. Do synthetic halt */
5697 vcpu->arch.apf.halted = true;
5698 r = 1;
5699 goto out;
5700 }
Glauber Costac9aaa892011-07-11 15:28:14 -04005701 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
5702 record_steal_time(vcpu);
Avi Kivity7460fb4a2011-09-20 13:43:14 +03005703 if (kvm_check_request(KVM_REQ_NMI, vcpu))
5704 process_nmi(vcpu);
Nadav Har'Eld6185f22011-09-22 13:52:56 +03005705 req_immediate_exit =
5706 kvm_check_request(KVM_REQ_IMMEDIATE_EXIT, vcpu);
Gleb Natapovf5132b02011-11-10 14:57:22 +02005707 if (kvm_check_request(KVM_REQ_PMU, vcpu))
5708 kvm_handle_pmu_event(vcpu);
5709 if (kvm_check_request(KVM_REQ_PMI, vcpu))
5710 kvm_deliver_pmi(vcpu);
Yang Zhangc7c9c562013-01-25 10:18:51 +08005711 if (kvm_check_request(KVM_REQ_EOIBITMAP, vcpu))
5712 update_eoi_exitmap(vcpu);
Avi Kivity2f52d582008-01-16 12:49:30 +02005713 }
Avi Kivityb93463a2007-10-25 16:52:32 +02005714
Avi Kivityb463a6f2010-07-20 15:06:17 +03005715 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
Avi Kivity3842d132010-07-27 12:30:24 +03005716 inject_pending_event(vcpu);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005717
Avi Kivity3842d132010-07-27 12:30:24 +03005718 /* enable NMI/IRQ window open exits if needed */
Avi Kivity7460fb4a2011-09-20 13:43:14 +03005719 if (vcpu->arch.nmi_pending)
Avi Kivity3842d132010-07-27 12:30:24 +03005720 kvm_x86_ops->enable_nmi_window(vcpu);
Yang Zhangc7c9c562013-01-25 10:18:51 +08005721 else if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
Avi Kivity3842d132010-07-27 12:30:24 +03005722 kvm_x86_ops->enable_irq_window(vcpu);
Gleb Natapov6a8b1d12009-05-11 13:35:51 +03005723
Avi Kivity3842d132010-07-27 12:30:24 +03005724 if (kvm_lapic_enabled(vcpu)) {
Yang Zhangc7c9c562013-01-25 10:18:51 +08005725 /*
5726 * Update architecture specific hints for APIC
5727 * virtual interrupt delivery.
5728 */
5729 if (kvm_x86_ops->hwapic_irr_update)
5730 kvm_x86_ops->hwapic_irr_update(vcpu,
5731 kvm_lapic_find_highest_irr(vcpu));
Avi Kivity3842d132010-07-27 12:30:24 +03005732 update_cr8_intercept(vcpu);
5733 kvm_lapic_sync_to_vapic(vcpu);
5734 }
Gleb Natapov95ba8273132009-04-21 17:45:08 +03005735 }
Avi Kivityb93463a2007-10-25 16:52:32 +02005736
Avi Kivityd8368af2012-05-14 18:07:56 +03005737 r = kvm_mmu_reload(vcpu);
5738 if (unlikely(r)) {
Michael S. Tsirkind905c062012-06-24 19:25:00 +03005739 goto cancel_injection;
Avi Kivityd8368af2012-05-14 18:07:56 +03005740 }
5741
Jan Kiszka42dbaa52008-12-15 13:52:10 +01005742 preempt_disable();
5743
5744 kvm_x86_ops->prepare_guest_switch(vcpu);
5745 if (vcpu->fpu_active)
5746 kvm_load_guest_fpu(vcpu);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005747 kvm_load_guest_xcr0(vcpu);
5748
Xiao Guangrong6b7e2d02011-01-12 15:40:31 +08005749 vcpu->mode = IN_GUEST_MODE;
5750
5751 /* We should set ->mode before check ->requests,
5752 * see the comment in make_all_cpus_request.
5753 */
5754 smp_mb();
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005755
5756 local_irq_disable();
5757
Xiao Guangrong6b7e2d02011-01-12 15:40:31 +08005758 if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005759 || need_resched() || signal_pending(current)) {
Xiao Guangrong6b7e2d02011-01-12 15:40:31 +08005760 vcpu->mode = OUTSIDE_GUEST_MODE;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005761 smp_wmb();
5762 local_irq_enable();
5763 preempt_enable();
5764 r = 1;
Michael S. Tsirkind905c062012-06-24 19:25:00 +03005765 goto cancel_injection;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005766 }
5767
Marcelo Tosattif656ce02009-12-23 14:35:25 -02005768 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005769
Nadav Har'Eld6185f22011-09-22 13:52:56 +03005770 if (req_immediate_exit)
5771 smp_send_reschedule(vcpu->cpu);
5772
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005773 kvm_guest_enter();
5774
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005775 if (unlikely(vcpu->arch.switch_db_regs)) {
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005776 set_debugreg(0, 7);
5777 set_debugreg(vcpu->arch.eff_db[0], 0);
5778 set_debugreg(vcpu->arch.eff_db[1], 1);
5779 set_debugreg(vcpu->arch.eff_db[2], 2);
5780 set_debugreg(vcpu->arch.eff_db[3], 3);
5781 }
5782
Marcelo Tosatti229456f2009-06-17 09:22:14 -03005783 trace_kvm_entry(vcpu->vcpu_id);
Avi Kivity851ba692009-08-24 11:10:17 +03005784 kvm_x86_ops->run(vcpu);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005785
Frederic Weisbecker24f1e32c2009-09-09 19:22:48 +02005786 /*
5787 * If the guest has used debug registers, at least dr7
5788 * will be disabled while returning to the host.
5789 * If we don't have active breakpoints in the host, we don't
5790 * care about the messed up debug address registers. But if
5791 * we have some of them active, restore the old state.
5792 */
Frederic Weisbecker59d8eb52009-11-10 11:03:12 +01005793 if (hw_breakpoint_active())
Frederic Weisbecker24f1e32c2009-09-09 19:22:48 +02005794 hw_breakpoint_restore();
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005795
Marcelo Tosatti886b4702012-11-27 23:28:58 -02005796 vcpu->arch.last_guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu,
5797 native_read_tsc());
Zachary Amsden1d5f0662010-08-19 22:07:30 -10005798
Xiao Guangrong6b7e2d02011-01-12 15:40:31 +08005799 vcpu->mode = OUTSIDE_GUEST_MODE;
Avi Kivityd94e1dc2010-05-03 16:54:48 +03005800 smp_wmb();
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005801 local_irq_enable();
5802
5803 ++vcpu->stat.exits;
5804
5805 /*
5806 * We must have an instruction between local_irq_enable() and
5807 * kvm_guest_exit(), so the timer interrupt isn't delayed by
5808 * the interrupt shadow. The stat.exits increment will do nicely.
5809 * But we need to prevent reordering, hence this barrier():
5810 */
5811 barrier();
5812
5813 kvm_guest_exit();
5814
5815 preempt_enable();
5816
Marcelo Tosattif656ce02009-12-23 14:35:25 -02005817 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
Marcelo Tosatti3200f402008-03-29 20:17:59 -03005818
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005819 /*
5820 * Profile KVM exit RIPs:
5821 */
5822 if (unlikely(prof_on == KVM_PROFILING)) {
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03005823 unsigned long rip = kvm_rip_read(vcpu);
5824 profile_hit(KVM_PROFILING, (void *)rip);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005825 }
5826
Zachary Amsdencc578282012-02-03 15:43:50 -02005827 if (unlikely(vcpu->arch.tsc_always_catchup))
5828 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
Avi Kivity298101d2007-11-25 13:41:11 +02005829
Michael S. Tsirkin5cfb1d52012-06-24 19:24:54 +03005830 if (vcpu->arch.apic_attention)
5831 kvm_lapic_sync_from_vapic(vcpu);
Avi Kivityb93463a2007-10-25 16:52:32 +02005832
Avi Kivity851ba692009-08-24 11:10:17 +03005833 r = kvm_x86_ops->handle_exit(vcpu);
Michael S. Tsirkind905c062012-06-24 19:25:00 +03005834 return r;
5835
5836cancel_injection:
5837 kvm_x86_ops->cancel_injection(vcpu);
Michael S. Tsirkinae7a2a32012-06-24 19:25:07 +03005838 if (unlikely(vcpu->arch.apic_attention))
5839 kvm_lapic_sync_from_vapic(vcpu);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005840out:
Marcelo Tosattid7690172008-09-08 15:23:48 -03005841 return r;
5842}
5843
Gleb Natapov09cec752009-03-23 15:11:44 +02005844
Avi Kivity851ba692009-08-24 11:10:17 +03005845static int __vcpu_run(struct kvm_vcpu *vcpu)
Marcelo Tosattid7690172008-09-08 15:23:48 -03005846{
5847 int r;
Marcelo Tosattif656ce02009-12-23 14:35:25 -02005848 struct kvm *kvm = vcpu->kvm;
Marcelo Tosattid7690172008-09-08 15:23:48 -03005849
5850 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
Jan Kiszka1b10bf32008-09-30 10:41:06 +02005851 pr_debug("vcpu %d received sipi with vector # %x\n",
5852 vcpu->vcpu_id, vcpu->arch.sipi_vector);
Marcelo Tosattid7690172008-09-08 15:23:48 -03005853 kvm_lapic_reset(vcpu);
Jan Kiszka8b6e4542012-09-20 07:43:08 +02005854 r = kvm_vcpu_reset(vcpu);
Marcelo Tosattid7690172008-09-08 15:23:48 -03005855 if (r)
5856 return r;
5857 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005858 }
5859
Marcelo Tosattif656ce02009-12-23 14:35:25 -02005860 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
Xiao Guangrong44841412012-09-07 14:14:20 +08005861 r = vapic_enter(vcpu);
5862 if (r) {
5863 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5864 return r;
5865 }
Marcelo Tosattid7690172008-09-08 15:23:48 -03005866
5867 r = 1;
5868 while (r > 0) {
Gleb Natapovaf585b92010-10-14 11:22:46 +02005869 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
5870 !vcpu->arch.apf.halted)
Avi Kivity851ba692009-08-24 11:10:17 +03005871 r = vcpu_enter_guest(vcpu);
Marcelo Tosattid7690172008-09-08 15:23:48 -03005872 else {
Marcelo Tosattif656ce02009-12-23 14:35:25 -02005873 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
Marcelo Tosattid7690172008-09-08 15:23:48 -03005874 kvm_vcpu_block(vcpu);
Marcelo Tosattif656ce02009-12-23 14:35:25 -02005875 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
Avi Kivitya8eeb042010-05-10 12:34:53 +03005876 if (kvm_check_request(KVM_REQ_UNHALT, vcpu))
Gleb Natapov09cec752009-03-23 15:11:44 +02005877 {
5878 switch(vcpu->arch.mp_state) {
5879 case KVM_MP_STATE_HALTED:
Marcelo Tosattid7690172008-09-08 15:23:48 -03005880 vcpu->arch.mp_state =
Gleb Natapov09cec752009-03-23 15:11:44 +02005881 KVM_MP_STATE_RUNNABLE;
5882 case KVM_MP_STATE_RUNNABLE:
Gleb Natapovaf585b92010-10-14 11:22:46 +02005883 vcpu->arch.apf.halted = false;
Gleb Natapov09cec752009-03-23 15:11:44 +02005884 break;
5885 case KVM_MP_STATE_SIPI_RECEIVED:
5886 default:
5887 r = -EINTR;
5888 break;
5889 }
5890 }
Marcelo Tosattid7690172008-09-08 15:23:48 -03005891 }
5892
Gleb Natapov09cec752009-03-23 15:11:44 +02005893 if (r <= 0)
5894 break;
5895
5896 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
5897 if (kvm_cpu_has_pending_timer(vcpu))
5898 kvm_inject_pending_timer_irqs(vcpu);
5899
Avi Kivity851ba692009-08-24 11:10:17 +03005900 if (dm_request_for_irq_injection(vcpu)) {
Gleb Natapov09cec752009-03-23 15:11:44 +02005901 r = -EINTR;
Avi Kivity851ba692009-08-24 11:10:17 +03005902 vcpu->run->exit_reason = KVM_EXIT_INTR;
Gleb Natapov09cec752009-03-23 15:11:44 +02005903 ++vcpu->stat.request_irq_exits;
5904 }
Gleb Natapovaf585b92010-10-14 11:22:46 +02005905
5906 kvm_check_async_pf_completion(vcpu);
5907
Gleb Natapov09cec752009-03-23 15:11:44 +02005908 if (signal_pending(current)) {
5909 r = -EINTR;
Avi Kivity851ba692009-08-24 11:10:17 +03005910 vcpu->run->exit_reason = KVM_EXIT_INTR;
Gleb Natapov09cec752009-03-23 15:11:44 +02005911 ++vcpu->stat.signal_exits;
5912 }
5913 if (need_resched()) {
Marcelo Tosattif656ce02009-12-23 14:35:25 -02005914 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
Gleb Natapov09cec752009-03-23 15:11:44 +02005915 kvm_resched(vcpu);
Marcelo Tosattif656ce02009-12-23 14:35:25 -02005916 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
Marcelo Tosattid7690172008-09-08 15:23:48 -03005917 }
5918 }
5919
Marcelo Tosattif656ce02009-12-23 14:35:25 -02005920 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005921
Avi Kivityb93463a2007-10-25 16:52:32 +02005922 vapic_exit(vcpu);
5923
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05005924 return r;
5925}
5926
Gleb Natapov716d51a2012-09-03 15:24:26 +03005927static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
5928{
5929 int r;
5930 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5931 r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
5932 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5933 if (r != EMULATE_DONE)
5934 return 0;
5935 return 1;
5936}
5937
5938static int complete_emulated_pio(struct kvm_vcpu *vcpu)
5939{
5940 BUG_ON(!vcpu->arch.pio.count);
5941
5942 return complete_emulated_io(vcpu);
5943}
5944
Avi Kivityf78146b2012-04-18 19:22:47 +03005945/*
5946 * Implements the following, as a state machine:
5947 *
5948 * read:
5949 * for each fragment
Xiao Guangrong87da7e62012-10-24 14:07:59 +08005950 * for each mmio piece in the fragment
5951 * write gpa, len
5952 * exit
5953 * copy data
Avi Kivityf78146b2012-04-18 19:22:47 +03005954 * execute insn
5955 *
5956 * write:
5957 * for each fragment
Xiao Guangrong87da7e62012-10-24 14:07:59 +08005958 * for each mmio piece in the fragment
5959 * write gpa, len
5960 * copy data
5961 * exit
Avi Kivityf78146b2012-04-18 19:22:47 +03005962 */
Gleb Natapov716d51a2012-09-03 15:24:26 +03005963static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
Avi Kivity5287f192010-01-19 14:20:10 +02005964{
5965 struct kvm_run *run = vcpu->run;
Avi Kivityf78146b2012-04-18 19:22:47 +03005966 struct kvm_mmio_fragment *frag;
Xiao Guangrong87da7e62012-10-24 14:07:59 +08005967 unsigned len;
Avi Kivity5287f192010-01-19 14:20:10 +02005968
Gleb Natapov716d51a2012-09-03 15:24:26 +03005969 BUG_ON(!vcpu->mmio_needed);
Avi Kivity5287f192010-01-19 14:20:10 +02005970
Gleb Natapov716d51a2012-09-03 15:24:26 +03005971 /* Complete previous fragment */
Xiao Guangrong87da7e62012-10-24 14:07:59 +08005972 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
5973 len = min(8u, frag->len);
Gleb Natapov716d51a2012-09-03 15:24:26 +03005974 if (!vcpu->mmio_is_write)
Xiao Guangrong87da7e62012-10-24 14:07:59 +08005975 memcpy(frag->data, run->mmio.data, len);
5976
5977 if (frag->len <= 8) {
5978 /* Switch to the next fragment. */
5979 frag++;
5980 vcpu->mmio_cur_fragment++;
5981 } else {
5982 /* Go forward to the next mmio piece. */
5983 frag->data += len;
5984 frag->gpa += len;
5985 frag->len -= len;
5986 }
5987
Gleb Natapov716d51a2012-09-03 15:24:26 +03005988 if (vcpu->mmio_cur_fragment == vcpu->mmio_nr_fragments) {
5989 vcpu->mmio_needed = 0;
Avi Kivitycef4dea2010-01-20 12:01:20 +02005990 if (vcpu->mmio_is_write)
Gleb Natapov716d51a2012-09-03 15:24:26 +03005991 return 1;
5992 vcpu->mmio_read_completed = 1;
5993 return complete_emulated_io(vcpu);
Avi Kivity5287f192010-01-19 14:20:10 +02005994 }
Xiao Guangrong87da7e62012-10-24 14:07:59 +08005995
Gleb Natapov716d51a2012-09-03 15:24:26 +03005996 run->exit_reason = KVM_EXIT_MMIO;
5997 run->mmio.phys_addr = frag->gpa;
5998 if (vcpu->mmio_is_write)
Xiao Guangrong87da7e62012-10-24 14:07:59 +08005999 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
6000 run->mmio.len = min(8u, frag->len);
Gleb Natapov716d51a2012-09-03 15:24:26 +03006001 run->mmio.is_write = vcpu->mmio_is_write;
6002 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6003 return 0;
Avi Kivity5287f192010-01-19 14:20:10 +02006004}
6005
Gleb Natapov716d51a2012-09-03 15:24:26 +03006006
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006007int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
6008{
6009 int r;
6010 sigset_t sigsaved;
6011
Avi Kivitye5c30142011-01-11 12:15:54 +02006012 if (!tsk_used_math(current) && init_fpu(current))
6013 return -ENOMEM;
6014
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006015 if (vcpu->sigset_active)
6016 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
6017
Avi Kivityac9f6dc2008-07-06 15:48:31 +03006018 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
6019 kvm_vcpu_block(vcpu);
Marcelo Tosattid7690172008-09-08 15:23:48 -03006020 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
Avi Kivityac9f6dc2008-07-06 15:48:31 +03006021 r = -EAGAIN;
6022 goto out;
6023 }
6024
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006025 /* re-sync apic's tpr */
Andre Przywaraeea1cff2010-12-21 11:12:00 +01006026 if (!irqchip_in_kernel(vcpu->kvm)) {
6027 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
6028 r = -EINVAL;
6029 goto out;
6030 }
6031 }
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006032
Gleb Natapov716d51a2012-09-03 15:24:26 +03006033 if (unlikely(vcpu->arch.complete_userspace_io)) {
6034 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
6035 vcpu->arch.complete_userspace_io = NULL;
6036 r = cui(vcpu);
6037 if (r <= 0)
6038 goto out;
6039 } else
6040 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
Avi Kivity5287f192010-01-19 14:20:10 +02006041
Avi Kivity851ba692009-08-24 11:10:17 +03006042 r = __vcpu_run(vcpu);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006043
6044out:
Marcelo Tosattif1d86e42010-05-03 23:04:27 -03006045 post_kvm_run_save(vcpu);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006046 if (vcpu->sigset_active)
6047 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
6048
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006049 return r;
6050}
6051
6052int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6053{
Gleb Natapov7ae441e2011-03-31 12:06:41 +02006054 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
6055 /*
6056 * We are here if userspace calls get_regs() in the middle of
6057 * instruction emulation. Registers state needs to be copied
Guo Chao4a969982012-06-28 15:17:27 +08006058 * back from emulation context to vcpu. Userspace shouldn't do
Gleb Natapov7ae441e2011-03-31 12:06:41 +02006059 * that usually, but some bad designed PV devices (vmware
6060 * backdoor interface) need this to work
6061 */
Avi Kivitydd856ef2012-08-27 23:46:17 +03006062 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
Gleb Natapov7ae441e2011-03-31 12:06:41 +02006063 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6064 }
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03006065 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
6066 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
6067 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
6068 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
6069 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
6070 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
6071 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
6072 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006073#ifdef CONFIG_X86_64
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03006074 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
6075 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
6076 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
6077 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
6078 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
6079 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
6080 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
6081 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006082#endif
6083
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03006084 regs->rip = kvm_rip_read(vcpu);
Jan Kiszka91586a32009-10-05 13:07:21 +02006085 regs->rflags = kvm_get_rflags(vcpu);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006086
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006087 return 0;
6088}
6089
6090int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6091{
Gleb Natapov7ae441e2011-03-31 12:06:41 +02006092 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
6093 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6094
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03006095 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
6096 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
6097 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
6098 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
6099 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
6100 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
6101 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
6102 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006103#ifdef CONFIG_X86_64
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03006104 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
6105 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
6106 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
6107 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
6108 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
6109 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
6110 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
6111 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006112#endif
6113
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03006114 kvm_rip_write(vcpu, regs->rip);
Jan Kiszka91586a32009-10-05 13:07:21 +02006115 kvm_set_rflags(vcpu, regs->rflags);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006116
Jan Kiszkab4f14ab2008-04-30 17:59:04 +02006117 vcpu->arch.exception.pending = false;
6118
Avi Kivity3842d132010-07-27 12:30:24 +03006119 kvm_make_request(KVM_REQ_EVENT, vcpu);
6120
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006121 return 0;
6122}
6123
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006124void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
6125{
6126 struct kvm_segment cs;
6127
Guillaume Thouvenin3e6e0aa2008-05-27 10:18:46 +02006128 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006129 *db = cs.db;
6130 *l = cs.l;
6131}
6132EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
6133
6134int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
6135 struct kvm_sregs *sregs)
6136{
Gleb Natapov89a27f42010-02-16 10:51:48 +02006137 struct desc_ptr dt;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006138
Guillaume Thouvenin3e6e0aa2008-05-27 10:18:46 +02006139 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6140 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6141 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6142 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6143 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6144 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006145
Guillaume Thouvenin3e6e0aa2008-05-27 10:18:46 +02006146 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6147 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006148
6149 kvm_x86_ops->get_idt(vcpu, &dt);
Gleb Natapov89a27f42010-02-16 10:51:48 +02006150 sregs->idt.limit = dt.size;
6151 sregs->idt.base = dt.address;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006152 kvm_x86_ops->get_gdt(vcpu, &dt);
Gleb Natapov89a27f42010-02-16 10:51:48 +02006153 sregs->gdt.limit = dt.size;
6154 sregs->gdt.base = dt.address;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006155
Avi Kivity4d4ec082009-12-29 18:07:30 +02006156 sregs->cr0 = kvm_read_cr0(vcpu);
Zhang Xiantaoad312c72007-12-13 23:50:52 +08006157 sregs->cr2 = vcpu->arch.cr2;
Avi Kivity9f8fe502010-12-05 17:30:00 +02006158 sregs->cr3 = kvm_read_cr3(vcpu);
Avi Kivityfc78f512009-12-07 12:16:48 +02006159 sregs->cr4 = kvm_read_cr4(vcpu);
Avi Kivity2d3ad1f2008-02-24 11:20:43 +02006160 sregs->cr8 = kvm_get_cr8(vcpu);
Avi Kivityf6801df2010-01-21 15:31:50 +02006161 sregs->efer = vcpu->arch.efer;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006162 sregs->apic_base = kvm_get_apic_base(vcpu);
6163
Gleb Natapov923c61b2009-05-11 13:35:48 +03006164 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006165
Gleb Natapov36752c92009-05-11 13:35:53 +03006166 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
Gleb Natapov14d0bc12009-04-21 17:45:11 +03006167 set_bit(vcpu->arch.interrupt.nr,
6168 (unsigned long *)sregs->interrupt_bitmap);
Gleb Natapov16d7a192009-04-21 17:45:10 +03006169
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006170 return 0;
6171}
6172
Marcelo Tosatti62d9f0d2008-04-11 13:24:45 -03006173int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
6174 struct kvm_mp_state *mp_state)
6175{
Marcelo Tosatti62d9f0d2008-04-11 13:24:45 -03006176 mp_state->mp_state = vcpu->arch.mp_state;
Marcelo Tosatti62d9f0d2008-04-11 13:24:45 -03006177 return 0;
6178}
6179
6180int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
6181 struct kvm_mp_state *mp_state)
6182{
Marcelo Tosatti62d9f0d2008-04-11 13:24:45 -03006183 vcpu->arch.mp_state = mp_state->mp_state;
Avi Kivity3842d132010-07-27 12:30:24 +03006184 kvm_make_request(KVM_REQ_EVENT, vcpu);
Marcelo Tosatti62d9f0d2008-04-11 13:24:45 -03006185 return 0;
6186}
6187
Kevin Wolf7f3d35f2012-02-08 14:34:38 +01006188int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
6189 int reason, bool has_error_code, u32 error_code)
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006190{
Takuya Yoshikawa9d741912011-05-29 21:53:48 +09006191 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
Mohammed Gamal8ec47222010-08-16 00:47:01 +03006192 int ret;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006193
Mohammed Gamal8ec47222010-08-16 00:47:01 +03006194 init_emulate_ctxt(vcpu);
Izik Eidus37817f22008-03-24 23:14:53 +02006195
Kevin Wolf7f3d35f2012-02-08 14:34:38 +01006196 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
Takuya Yoshikawa9d741912011-05-29 21:53:48 +09006197 has_error_code, error_code);
Izik Eidus37817f22008-03-24 23:14:53 +02006198
Gleb Natapovc6975182010-02-18 12:15:01 +02006199 if (ret)
Gleb Natapov19d04432010-04-15 12:29:50 +03006200 return EMULATE_FAIL;
Gleb Natapovc6975182010-02-18 12:15:01 +02006201
Takuya Yoshikawa9d741912011-05-29 21:53:48 +09006202 kvm_rip_write(vcpu, ctxt->eip);
6203 kvm_set_rflags(vcpu, ctxt->eflags);
Avi Kivity3842d132010-07-27 12:30:24 +03006204 kvm_make_request(KVM_REQ_EVENT, vcpu);
Gleb Natapov19d04432010-04-15 12:29:50 +03006205 return EMULATE_DONE;
Izik Eidus37817f22008-03-24 23:14:53 +02006206}
6207EXPORT_SYMBOL_GPL(kvm_task_switch);
6208
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006209int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
6210 struct kvm_sregs *sregs)
6211{
6212 int mmu_reset_needed = 0;
Xiao Guangrong63f42e02011-01-12 15:39:18 +08006213 int pending_vec, max_bits, idx;
Gleb Natapov89a27f42010-02-16 10:51:48 +02006214 struct desc_ptr dt;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006215
Petr Matousek6d1068b2012-11-06 19:24:07 +01006216 if (!guest_cpuid_has_xsave(vcpu) && (sregs->cr4 & X86_CR4_OSXSAVE))
6217 return -EINVAL;
6218
Gleb Natapov89a27f42010-02-16 10:51:48 +02006219 dt.size = sregs->idt.limit;
6220 dt.address = sregs->idt.base;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006221 kvm_x86_ops->set_idt(vcpu, &dt);
Gleb Natapov89a27f42010-02-16 10:51:48 +02006222 dt.size = sregs->gdt.limit;
6223 dt.address = sregs->gdt.base;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006224 kvm_x86_ops->set_gdt(vcpu, &dt);
6225
Zhang Xiantaoad312c72007-12-13 23:50:52 +08006226 vcpu->arch.cr2 = sregs->cr2;
Avi Kivity9f8fe502010-12-05 17:30:00 +02006227 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
Jan Kiszkadc7e7952009-07-01 20:52:03 +02006228 vcpu->arch.cr3 = sregs->cr3;
Avi Kivityaff48ba2010-12-05 18:56:11 +02006229 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006230
Avi Kivity2d3ad1f2008-02-24 11:20:43 +02006231 kvm_set_cr8(vcpu, sregs->cr8);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006232
Avi Kivityf6801df2010-01-21 15:31:50 +02006233 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006234 kvm_x86_ops->set_efer(vcpu, sregs->efer);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006235 kvm_set_apic_base(vcpu, sregs->apic_base);
6236
Avi Kivity4d4ec082009-12-29 18:07:30 +02006237 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006238 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
Paul Knowlesd7306162008-02-06 11:02:35 +00006239 vcpu->arch.cr0 = sregs->cr0;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006240
Avi Kivityfc78f512009-12-07 12:16:48 +02006241 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006242 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
Sheng Yang3ea3aa82010-12-08 10:49:43 +08006243 if (sregs->cr4 & X86_CR4_OSXSAVE)
Avi Kivity00b27a32011-11-23 16:30:32 +02006244 kvm_update_cpuid(vcpu);
Xiao Guangrong63f42e02011-01-12 15:39:18 +08006245
6246 idx = srcu_read_lock(&vcpu->kvm->srcu);
Marcelo Tosatti7c93be42009-10-26 16:48:33 -02006247 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
Avi Kivity9f8fe502010-12-05 17:30:00 +02006248 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
Marcelo Tosatti7c93be42009-10-26 16:48:33 -02006249 mmu_reset_needed = 1;
6250 }
Xiao Guangrong63f42e02011-01-12 15:39:18 +08006251 srcu_read_unlock(&vcpu->kvm->srcu, idx);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006252
6253 if (mmu_reset_needed)
6254 kvm_mmu_reset_context(vcpu);
6255
Michael S. Tsirkina50abc32012-09-05 20:00:52 +03006256 max_bits = KVM_NR_INTERRUPTS;
Gleb Natapov923c61b2009-05-11 13:35:48 +03006257 pending_vec = find_first_bit(
6258 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
6259 if (pending_vec < max_bits) {
Gleb Natapov66fd3f72009-05-11 13:35:50 +03006260 kvm_queue_interrupt(vcpu, pending_vec, false);
Gleb Natapov923c61b2009-05-11 13:35:48 +03006261 pr_debug("Set back pending irq %d\n", pending_vec);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006262 }
6263
Guillaume Thouvenin3e6e0aa2008-05-27 10:18:46 +02006264 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6265 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6266 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6267 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6268 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6269 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006270
Guillaume Thouvenin3e6e0aa2008-05-27 10:18:46 +02006271 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6272 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006273
Mikhail Ershov5f0269f2009-08-03 14:58:25 +03006274 update_cr8_intercept(vcpu);
6275
Marcelo Tosatti9c3e4aa2008-09-10 16:40:55 -03006276 /* Older userspace won't unhalt the vcpu on reset. */
Gleb Natapovc5af89b2009-06-09 15:56:26 +03006277 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
Marcelo Tosatti9c3e4aa2008-09-10 16:40:55 -03006278 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
Avi Kivity3eeb3282010-01-21 15:31:48 +02006279 !is_protmode(vcpu))
Marcelo Tosatti9c3e4aa2008-09-10 16:40:55 -03006280 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6281
Avi Kivity3842d132010-07-27 12:30:24 +03006282 kvm_make_request(KVM_REQ_EVENT, vcpu);
6283
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006284 return 0;
6285}
6286
Jan Kiszkad0bfb942008-12-15 13:52:10 +01006287int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
6288 struct kvm_guest_debug *dbg)
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006289{
Jan Kiszka355be0b2009-10-03 00:31:21 +02006290 unsigned long rflags;
Jan Kiszkaae675ef2008-12-15 13:52:10 +01006291 int i, r;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006292
Jan Kiszka4f926bf22009-10-30 12:46:59 +01006293 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
6294 r = -EBUSY;
6295 if (vcpu->arch.exception.pending)
Avi Kivity2122ff52010-05-13 11:25:04 +03006296 goto out;
Jan Kiszka4f926bf22009-10-30 12:46:59 +01006297 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
6298 kvm_queue_exception(vcpu, DB_VECTOR);
6299 else
6300 kvm_queue_exception(vcpu, BP_VECTOR);
6301 }
6302
Jan Kiszka91586a32009-10-05 13:07:21 +02006303 /*
6304 * Read rflags as long as potentially injected trace flags are still
6305 * filtered out.
6306 */
6307 rflags = kvm_get_rflags(vcpu);
Jan Kiszka355be0b2009-10-03 00:31:21 +02006308
6309 vcpu->guest_debug = dbg->control;
6310 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
6311 vcpu->guest_debug = 0;
6312
6313 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
Jan Kiszkaae675ef2008-12-15 13:52:10 +01006314 for (i = 0; i < KVM_NR_DB_REGS; ++i)
6315 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
Jan Kiszkac8639012012-09-21 05:42:55 +02006316 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
Jan Kiszkaae675ef2008-12-15 13:52:10 +01006317 } else {
6318 for (i = 0; i < KVM_NR_DB_REGS; i++)
6319 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
Jan Kiszkaae675ef2008-12-15 13:52:10 +01006320 }
Jan Kiszkac8639012012-09-21 05:42:55 +02006321 kvm_update_dr7(vcpu);
Jan Kiszkaae675ef2008-12-15 13:52:10 +01006322
Jan Kiszkaf92653e2010-02-23 17:47:55 +01006323 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6324 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
6325 get_segment_base(vcpu, VCPU_SREG_CS);
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006326
Jan Kiszka91586a32009-10-05 13:07:21 +02006327 /*
6328 * Trigger an rflags update that will inject or remove the trace
6329 * flags.
6330 */
6331 kvm_set_rflags(vcpu, rflags);
Jan Kiszkad0bfb942008-12-15 13:52:10 +01006332
Jan Kiszkac8639012012-09-21 05:42:55 +02006333 kvm_x86_ops->update_db_bp_intercept(vcpu);
Jan Kiszka355be0b2009-10-03 00:31:21 +02006334
Jan Kiszka4f926bf22009-10-30 12:46:59 +01006335 r = 0;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006336
Avi Kivity2122ff52010-05-13 11:25:04 +03006337out:
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05006338
6339 return r;
6340}
6341
6342/*
Zhang Xiantao8b006792007-11-16 13:05:55 +08006343 * Translate a guest virtual address to a guest physical address.
6344 */
6345int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
6346 struct kvm_translation *tr)
6347{
6348 unsigned long vaddr = tr->linear_address;
6349 gpa_t gpa;
Marcelo Tosattif656ce02009-12-23 14:35:25 -02006350 int idx;
Zhang Xiantao8b006792007-11-16 13:05:55 +08006351
Marcelo Tosattif656ce02009-12-23 14:35:25 -02006352 idx = srcu_read_lock(&vcpu->kvm->srcu);
Gleb Natapov1871c602010-02-10 14:21:32 +02006353 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
Marcelo Tosattif656ce02009-12-23 14:35:25 -02006354 srcu_read_unlock(&vcpu->kvm->srcu, idx);
Zhang Xiantao8b006792007-11-16 13:05:55 +08006355 tr->physical_address = gpa;
6356 tr->valid = gpa != UNMAPPED_GVA;
6357 tr->writeable = 1;
6358 tr->usermode = 0;
Zhang Xiantao8b006792007-11-16 13:05:55 +08006359
6360 return 0;
6361}
6362
Hollis Blanchardd0752062007-10-31 17:24:25 -05006363int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6364{
Sheng Yang98918832010-05-17 17:08:28 +08006365 struct i387_fxsave_struct *fxsave =
6366 &vcpu->arch.guest_fpu.state->fxsave;
Hollis Blanchardd0752062007-10-31 17:24:25 -05006367
Hollis Blanchardd0752062007-10-31 17:24:25 -05006368 memcpy(fpu->fpr, fxsave->st_space, 128);
6369 fpu->fcw = fxsave->cwd;
6370 fpu->fsw = fxsave->swd;
6371 fpu->ftwx = fxsave->twd;
6372 fpu->last_opcode = fxsave->fop;
6373 fpu->last_ip = fxsave->rip;
6374 fpu->last_dp = fxsave->rdp;
6375 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
6376
Hollis Blanchardd0752062007-10-31 17:24:25 -05006377 return 0;
6378}
6379
6380int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6381{
Sheng Yang98918832010-05-17 17:08:28 +08006382 struct i387_fxsave_struct *fxsave =
6383 &vcpu->arch.guest_fpu.state->fxsave;
Hollis Blanchardd0752062007-10-31 17:24:25 -05006384
Hollis Blanchardd0752062007-10-31 17:24:25 -05006385 memcpy(fxsave->st_space, fpu->fpr, 128);
6386 fxsave->cwd = fpu->fcw;
6387 fxsave->swd = fpu->fsw;
6388 fxsave->twd = fpu->ftwx;
6389 fxsave->fop = fpu->last_opcode;
6390 fxsave->rip = fpu->last_ip;
6391 fxsave->rdp = fpu->last_dp;
6392 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
6393
Hollis Blanchardd0752062007-10-31 17:24:25 -05006394 return 0;
6395}
6396
Jan Kiszka10ab25c2010-05-25 16:01:50 +02006397int fx_init(struct kvm_vcpu *vcpu)
Hollis Blanchardd0752062007-10-31 17:24:25 -05006398{
Jan Kiszka10ab25c2010-05-25 16:01:50 +02006399 int err;
6400
6401 err = fpu_alloc(&vcpu->arch.guest_fpu);
6402 if (err)
6403 return err;
6404
Sheng Yang98918832010-05-17 17:08:28 +08006405 fpu_finit(&vcpu->arch.guest_fpu);
Hollis Blanchardd0752062007-10-31 17:24:25 -05006406
Dexuan Cui2acf9232010-06-10 11:27:12 +08006407 /*
6408 * Ensure guest xcr0 is valid for loading
6409 */
6410 vcpu->arch.xcr0 = XSTATE_FP;
6411
Zhang Xiantaoad312c72007-12-13 23:50:52 +08006412 vcpu->arch.cr0 |= X86_CR0_ET;
Jan Kiszka10ab25c2010-05-25 16:01:50 +02006413
6414 return 0;
Hollis Blanchardd0752062007-10-31 17:24:25 -05006415}
6416EXPORT_SYMBOL_GPL(fx_init);
6417
Sheng Yang98918832010-05-17 17:08:28 +08006418static void fx_free(struct kvm_vcpu *vcpu)
6419{
6420 fpu_free(&vcpu->arch.guest_fpu);
6421}
6422
Hollis Blanchardd0752062007-10-31 17:24:25 -05006423void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
6424{
Avi Kivity2608d7a2010-01-21 15:31:45 +02006425 if (vcpu->guest_fpu_loaded)
Hollis Blanchardd0752062007-10-31 17:24:25 -05006426 return;
6427
Dexuan Cui2acf9232010-06-10 11:27:12 +08006428 /*
6429 * Restore all possible states in the guest,
6430 * and assume host would use all available bits.
6431 * Guest xcr0 would be loaded later.
6432 */
6433 kvm_put_guest_xcr0(vcpu);
Hollis Blanchardd0752062007-10-31 17:24:25 -05006434 vcpu->guest_fpu_loaded = 1;
Suresh Siddhab1a74bf2012-09-20 11:01:49 -07006435 __kernel_fpu_begin();
Sheng Yang98918832010-05-17 17:08:28 +08006436 fpu_restore_checking(&vcpu->arch.guest_fpu);
Avi Kivity0c048512010-01-21 15:31:52 +02006437 trace_kvm_fpu(1);
Hollis Blanchardd0752062007-10-31 17:24:25 -05006438}
Hollis Blanchardd0752062007-10-31 17:24:25 -05006439
6440void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
6441{
Dexuan Cui2acf9232010-06-10 11:27:12 +08006442 kvm_put_guest_xcr0(vcpu);
6443
Hollis Blanchardd0752062007-10-31 17:24:25 -05006444 if (!vcpu->guest_fpu_loaded)
6445 return;
6446
6447 vcpu->guest_fpu_loaded = 0;
Sheng Yang98918832010-05-17 17:08:28 +08006448 fpu_save_init(&vcpu->arch.guest_fpu);
Suresh Siddhab1a74bf2012-09-20 11:01:49 -07006449 __kernel_fpu_end();
Avi Kivityf096ed82007-11-18 13:54:33 +02006450 ++vcpu->stat.fpu_reload;
Avi Kivitya8eeb042010-05-10 12:34:53 +03006451 kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
Avi Kivity0c048512010-01-21 15:31:52 +02006452 trace_kvm_fpu(0);
Hollis Blanchardd0752062007-10-31 17:24:25 -05006453}
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08006454
6455void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
6456{
Glauber Costa12f9a482011-02-01 14:16:40 -05006457 kvmclock_reset(vcpu);
Joerg Roedel7f1ea202009-02-25 16:08:31 +01006458
Sheng Yangf5f48ee2010-06-30 12:25:15 +08006459 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
Sheng Yang98918832010-05-17 17:08:28 +08006460 fx_free(vcpu);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08006461 kvm_x86_ops->vcpu_free(vcpu);
6462}
6463
6464struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
6465 unsigned int id)
6466{
Zachary Amsden6755bae2010-08-19 22:07:22 -10006467 if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
6468 printk_once(KERN_WARNING
6469 "kvm: SMP vm created on host with unstable TSC; "
6470 "guest TSC will not be reliable\n");
Avi Kivity26e52152007-11-20 15:30:24 +02006471 return kvm_x86_ops->vcpu_create(kvm, id);
6472}
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08006473
Avi Kivity26e52152007-11-20 15:30:24 +02006474int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
6475{
6476 int r;
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08006477
Sheng Yang0bed3b52008-10-09 16:01:54 +08006478 vcpu->arch.mtrr_state.have_fixed = 1;
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +03006479 r = vcpu_load(vcpu);
6480 if (r)
6481 return r;
Jan Kiszka8b6e4542012-09-20 07:43:08 +02006482 r = kvm_vcpu_reset(vcpu);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08006483 if (r == 0)
6484 r = kvm_mmu_setup(vcpu);
6485 vcpu_put(vcpu);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08006486
Avi Kivity26e52152007-11-20 15:30:24 +02006487 return r;
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08006488}
6489
Marcelo Tosatti42897d82012-11-27 23:29:02 -02006490int kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
6491{
6492 int r;
Will Auld8fe8ab42012-11-29 12:42:12 -08006493 struct msr_data msr;
Marcelo Tosatti42897d82012-11-27 23:29:02 -02006494
6495 r = vcpu_load(vcpu);
6496 if (r)
6497 return r;
Will Auld8fe8ab42012-11-29 12:42:12 -08006498 msr.data = 0x0;
6499 msr.index = MSR_IA32_TSC;
6500 msr.host_initiated = true;
6501 kvm_write_tsc(vcpu, &msr);
Marcelo Tosatti42897d82012-11-27 23:29:02 -02006502 vcpu_put(vcpu);
6503
6504 return r;
6505}
6506
Hollis Blanchardd40ccc62007-11-19 14:04:43 -06006507void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08006508{
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +03006509 int r;
Gleb Natapov344d9582010-10-14 11:22:50 +02006510 vcpu->arch.apf.msr_val = 0;
6511
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +03006512 r = vcpu_load(vcpu);
6513 BUG_ON(r);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08006514 kvm_mmu_unload(vcpu);
6515 vcpu_put(vcpu);
6516
Sheng Yang98918832010-05-17 17:08:28 +08006517 fx_free(vcpu);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08006518 kvm_x86_ops->vcpu_free(vcpu);
6519}
6520
Jan Kiszka8b6e4542012-09-20 07:43:08 +02006521static int kvm_vcpu_reset(struct kvm_vcpu *vcpu)
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08006522{
Avi Kivity7460fb4a2011-09-20 13:43:14 +03006523 atomic_set(&vcpu->arch.nmi_queued, 0);
6524 vcpu->arch.nmi_pending = 0;
Jan Kiszka448fa4a2008-09-26 09:30:48 +02006525 vcpu->arch.nmi_injected = false;
6526
Jan Kiszka42dbaa52008-12-15 13:52:10 +01006527 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
6528 vcpu->arch.dr6 = DR6_FIXED_1;
6529 vcpu->arch.dr7 = DR7_FIXED_1;
Jan Kiszkac8639012012-09-21 05:42:55 +02006530 kvm_update_dr7(vcpu);
Jan Kiszka42dbaa52008-12-15 13:52:10 +01006531
Avi Kivity3842d132010-07-27 12:30:24 +03006532 kvm_make_request(KVM_REQ_EVENT, vcpu);
Gleb Natapov344d9582010-10-14 11:22:50 +02006533 vcpu->arch.apf.msr_val = 0;
Glauber Costac9aaa892011-07-11 15:28:14 -04006534 vcpu->arch.st.msr_val = 0;
Avi Kivity3842d132010-07-27 12:30:24 +03006535
Glauber Costa12f9a482011-02-01 14:16:40 -05006536 kvmclock_reset(vcpu);
6537
Gleb Natapovaf585b92010-10-14 11:22:46 +02006538 kvm_clear_async_pf_completion_queue(vcpu);
6539 kvm_async_pf_hash_reset(vcpu);
6540 vcpu->arch.apf.halted = false;
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08006541
Gleb Natapovf5132b02011-11-10 14:57:22 +02006542 kvm_pmu_reset(vcpu);
6543
Julian Stecklina66f7b722012-12-05 15:26:19 +01006544 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
6545 vcpu->arch.regs_avail = ~0;
6546 vcpu->arch.regs_dirty = ~0;
6547
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08006548 return kvm_x86_ops->vcpu_reset(vcpu);
6549}
6550
Alexander Graf10474ae2009-09-15 11:37:46 +02006551int kvm_arch_hardware_enable(void *garbage)
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08006552{
Zachary Amsdenca84d1a2010-08-19 22:07:28 -10006553 struct kvm *kvm;
6554 struct kvm_vcpu *vcpu;
6555 int i;
Zachary Amsden0dd6a6e2012-02-03 15:43:56 -02006556 int ret;
6557 u64 local_tsc;
6558 u64 max_tsc = 0;
6559 bool stable, backwards_tsc = false;
Avi Kivity18863bd2009-09-07 11:12:18 +03006560
6561 kvm_shared_msr_cpu_online();
Zachary Amsden0dd6a6e2012-02-03 15:43:56 -02006562 ret = kvm_x86_ops->hardware_enable(garbage);
6563 if (ret != 0)
6564 return ret;
6565
6566 local_tsc = native_read_tsc();
6567 stable = !check_tsc_unstable();
6568 list_for_each_entry(kvm, &vm_list, vm_list) {
6569 kvm_for_each_vcpu(i, vcpu, kvm) {
6570 if (!stable && vcpu->cpu == smp_processor_id())
6571 set_bit(KVM_REQ_CLOCK_UPDATE, &vcpu->requests);
6572 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
6573 backwards_tsc = true;
6574 if (vcpu->arch.last_host_tsc > max_tsc)
6575 max_tsc = vcpu->arch.last_host_tsc;
6576 }
6577 }
6578 }
6579
6580 /*
6581 * Sometimes, even reliable TSCs go backwards. This happens on
6582 * platforms that reset TSC during suspend or hibernate actions, but
6583 * maintain synchronization. We must compensate. Fortunately, we can
6584 * detect that condition here, which happens early in CPU bringup,
6585 * before any KVM threads can be running. Unfortunately, we can't
6586 * bring the TSCs fully up to date with real time, as we aren't yet far
6587 * enough into CPU bringup that we know how much real time has actually
6588 * elapsed; our helper function, get_kernel_ns() will be using boot
6589 * variables that haven't been updated yet.
6590 *
6591 * So we simply find the maximum observed TSC above, then record the
6592 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
6593 * the adjustment will be applied. Note that we accumulate
6594 * adjustments, in case multiple suspend cycles happen before some VCPU
6595 * gets a chance to run again. In the event that no KVM threads get a
6596 * chance to run, we will miss the entire elapsed period, as we'll have
6597 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
6598 * loose cycle time. This isn't too big a deal, since the loss will be
6599 * uniform across all VCPUs (not to mention the scenario is extremely
6600 * unlikely). It is possible that a second hibernate recovery happens
6601 * much faster than a first, causing the observed TSC here to be
6602 * smaller; this would require additional padding adjustment, which is
6603 * why we set last_host_tsc to the local tsc observed here.
6604 *
6605 * N.B. - this code below runs only on platforms with reliable TSC,
6606 * as that is the only way backwards_tsc is set above. Also note
6607 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
6608 * have the same delta_cyc adjustment applied if backwards_tsc
6609 * is detected. Note further, this adjustment is only done once,
6610 * as we reset last_host_tsc on all VCPUs to stop this from being
6611 * called multiple times (one for each physical CPU bringup).
6612 *
Guo Chao4a969982012-06-28 15:17:27 +08006613 * Platforms with unreliable TSCs don't have to deal with this, they
Zachary Amsden0dd6a6e2012-02-03 15:43:56 -02006614 * will be compensated by the logic in vcpu_load, which sets the TSC to
6615 * catchup mode. This will catchup all VCPUs to real time, but cannot
6616 * guarantee that they stay in perfect synchronization.
6617 */
6618 if (backwards_tsc) {
6619 u64 delta_cyc = max_tsc - local_tsc;
6620 list_for_each_entry(kvm, &vm_list, vm_list) {
6621 kvm_for_each_vcpu(i, vcpu, kvm) {
6622 vcpu->arch.tsc_offset_adjustment += delta_cyc;
6623 vcpu->arch.last_host_tsc = local_tsc;
Marcelo Tosattid8281992012-11-27 23:29:01 -02006624 set_bit(KVM_REQ_MASTERCLOCK_UPDATE,
6625 &vcpu->requests);
Zachary Amsden0dd6a6e2012-02-03 15:43:56 -02006626 }
6627
6628 /*
6629 * We have to disable TSC offset matching.. if you were
6630 * booting a VM while issuing an S4 host suspend....
6631 * you may have some problem. Solving this issue is
6632 * left as an exercise to the reader.
6633 */
6634 kvm->arch.last_tsc_nsec = 0;
6635 kvm->arch.last_tsc_write = 0;
6636 }
6637
6638 }
6639 return 0;
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08006640}
6641
6642void kvm_arch_hardware_disable(void *garbage)
6643{
6644 kvm_x86_ops->hardware_disable(garbage);
Avi Kivity3548bab2009-11-28 14:18:47 +02006645 drop_user_return_notifiers(garbage);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08006646}
6647
6648int kvm_arch_hardware_setup(void)
6649{
6650 return kvm_x86_ops->hardware_setup();
6651}
6652
6653void kvm_arch_hardware_unsetup(void)
6654{
6655 kvm_x86_ops->hardware_unsetup();
6656}
6657
6658void kvm_arch_check_processor_compat(void *rtn)
6659{
6660 kvm_x86_ops->check_processor_compatibility(rtn);
6661}
6662
Avi Kivity3e515702012-03-05 14:23:29 +02006663bool kvm_vcpu_compatible(struct kvm_vcpu *vcpu)
6664{
6665 return irqchip_in_kernel(vcpu->kvm) == (vcpu->arch.apic != NULL);
6666}
6667
Gleb Natapov54e98182012-08-05 15:58:32 +03006668struct static_key kvm_no_apic_vcpu __read_mostly;
6669
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08006670int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
6671{
6672 struct page *page;
6673 struct kvm *kvm;
6674 int r;
6675
6676 BUG_ON(vcpu->kvm == NULL);
6677 kvm = vcpu->kvm;
6678
Avi Kivity9aabc88f2010-07-29 15:11:50 +03006679 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
Gleb Natapovc5af89b2009-06-09 15:56:26 +03006680 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
Avi Kivitya4535292008-04-13 17:54:35 +03006681 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08006682 else
Avi Kivitya4535292008-04-13 17:54:35 +03006683 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08006684
6685 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
6686 if (!page) {
6687 r = -ENOMEM;
6688 goto fail;
6689 }
Zhang Xiantaoad312c72007-12-13 23:50:52 +08006690 vcpu->arch.pio_data = page_address(page);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08006691
Zachary Amsdencc578282012-02-03 15:43:50 -02006692 kvm_set_tsc_khz(vcpu, max_tsc_khz);
Zachary Amsdenc2855452010-09-18 14:38:15 -10006693
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08006694 r = kvm_mmu_create(vcpu);
6695 if (r < 0)
6696 goto fail_free_pio_data;
6697
6698 if (irqchip_in_kernel(kvm)) {
6699 r = kvm_create_lapic(vcpu);
6700 if (r < 0)
6701 goto fail_mmu_destroy;
Gleb Natapov54e98182012-08-05 15:58:32 +03006702 } else
6703 static_key_slow_inc(&kvm_no_apic_vcpu);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08006704
Huang Ying890ca9a2009-05-11 16:48:15 +08006705 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
6706 GFP_KERNEL);
6707 if (!vcpu->arch.mce_banks) {
6708 r = -ENOMEM;
Wei Yongjun443c39b2010-01-22 14:21:29 +08006709 goto fail_free_lapic;
Huang Ying890ca9a2009-05-11 16:48:15 +08006710 }
6711 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
6712
Sheng Yangf5f48ee2010-06-30 12:25:15 +08006713 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL))
6714 goto fail_free_mce_banks;
6715
Julian Stecklina66f7b722012-12-05 15:26:19 +01006716 r = fx_init(vcpu);
6717 if (r)
6718 goto fail_free_wbinvd_dirty_mask;
6719
Will Auldba904632012-11-29 12:42:50 -08006720 vcpu->arch.ia32_tsc_adjust_msr = 0x0;
Gleb Natapovaf585b92010-10-14 11:22:46 +02006721 kvm_async_pf_hash_reset(vcpu);
Gleb Natapovf5132b02011-11-10 14:57:22 +02006722 kvm_pmu_init(vcpu);
Gleb Natapovaf585b92010-10-14 11:22:46 +02006723
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08006724 return 0;
Julian Stecklina66f7b722012-12-05 15:26:19 +01006725fail_free_wbinvd_dirty_mask:
6726 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
Sheng Yangf5f48ee2010-06-30 12:25:15 +08006727fail_free_mce_banks:
6728 kfree(vcpu->arch.mce_banks);
Wei Yongjun443c39b2010-01-22 14:21:29 +08006729fail_free_lapic:
6730 kvm_free_lapic(vcpu);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08006731fail_mmu_destroy:
6732 kvm_mmu_destroy(vcpu);
6733fail_free_pio_data:
Zhang Xiantaoad312c72007-12-13 23:50:52 +08006734 free_page((unsigned long)vcpu->arch.pio_data);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08006735fail:
6736 return r;
6737}
6738
6739void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
6740{
Marcelo Tosattif656ce02009-12-23 14:35:25 -02006741 int idx;
6742
Gleb Natapovf5132b02011-11-10 14:57:22 +02006743 kvm_pmu_destroy(vcpu);
Wei Yongjun36cb93f2010-01-22 14:18:47 +08006744 kfree(vcpu->arch.mce_banks);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08006745 kvm_free_lapic(vcpu);
Marcelo Tosattif656ce02009-12-23 14:35:25 -02006746 idx = srcu_read_lock(&vcpu->kvm->srcu);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08006747 kvm_mmu_destroy(vcpu);
Marcelo Tosattif656ce02009-12-23 14:35:25 -02006748 srcu_read_unlock(&vcpu->kvm->srcu, idx);
Zhang Xiantaoad312c72007-12-13 23:50:52 +08006749 free_page((unsigned long)vcpu->arch.pio_data);
Gleb Natapov54e98182012-08-05 15:58:32 +03006750 if (!irqchip_in_kernel(vcpu->kvm))
6751 static_key_slow_dec(&kvm_no_apic_vcpu);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08006752}
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +08006753
Carsten Ottee08b9632012-01-04 10:25:20 +01006754int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +08006755{
Carsten Ottee08b9632012-01-04 10:25:20 +01006756 if (type)
6757 return -EINVAL;
6758
Zhang Xiantaof05e70a2007-12-14 10:01:48 +08006759 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
Ben-Ami Yassour4d5c5d02008-07-28 19:26:26 +03006760 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +08006761
Sheng Yang5550af42008-10-15 20:15:06 +08006762 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
6763 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
Alex Williamson7a844282012-09-21 11:58:03 -06006764 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
6765 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
6766 &kvm->arch.irq_sources_bitmap);
Sheng Yang5550af42008-10-15 20:15:06 +08006767
Jan Kiszka038f8c12011-02-04 10:49:11 +01006768 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
Gleb Natapov1e08ec42012-09-13 17:19:24 +03006769 mutex_init(&kvm->arch.apic_map_lock);
Marcelo Tosattid8281992012-11-27 23:29:01 -02006770 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
6771
6772 pvclock_update_vm_gtod_copy(kvm);
Marcelo Tosatti53f658b32008-12-11 20:45:05 +01006773
Jan Kiszkad89f5ef2010-11-09 17:02:49 +01006774 return 0;
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +08006775}
6776
6777static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
6778{
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +03006779 int r;
6780 r = vcpu_load(vcpu);
6781 BUG_ON(r);
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +08006782 kvm_mmu_unload(vcpu);
6783 vcpu_put(vcpu);
6784}
6785
6786static void kvm_free_vcpus(struct kvm *kvm)
6787{
6788 unsigned int i;
Gleb Natapov988a2ca2009-06-09 15:56:29 +03006789 struct kvm_vcpu *vcpu;
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +08006790
6791 /*
6792 * Unpin any mmu pages first.
6793 */
Gleb Natapovaf585b92010-10-14 11:22:46 +02006794 kvm_for_each_vcpu(i, vcpu, kvm) {
6795 kvm_clear_async_pf_completion_queue(vcpu);
Gleb Natapov988a2ca2009-06-09 15:56:29 +03006796 kvm_unload_vcpu_mmu(vcpu);
Gleb Natapovaf585b92010-10-14 11:22:46 +02006797 }
Gleb Natapov988a2ca2009-06-09 15:56:29 +03006798 kvm_for_each_vcpu(i, vcpu, kvm)
6799 kvm_arch_vcpu_free(vcpu);
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +08006800
Gleb Natapov988a2ca2009-06-09 15:56:29 +03006801 mutex_lock(&kvm->lock);
6802 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
6803 kvm->vcpus[i] = NULL;
6804
6805 atomic_set(&kvm->online_vcpus, 0);
6806 mutex_unlock(&kvm->lock);
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +08006807}
6808
Sheng Yangad8ba2c2009-01-06 10:03:02 +08006809void kvm_arch_sync_events(struct kvm *kvm)
6810{
Sheng Yangba4cef32009-01-06 10:03:03 +08006811 kvm_free_all_assigned_devices(kvm);
Xiao Guangrongaea924f2010-07-10 17:37:56 +08006812 kvm_free_pit(kvm);
Sheng Yangad8ba2c2009-01-06 10:03:02 +08006813}
6814
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +08006815void kvm_arch_destroy_vm(struct kvm *kvm)
6816{
Sheng Yang6eb55812008-10-31 12:37:41 +08006817 kvm_iommu_unmap_guest(kvm);
Zhang Xiantaod7deeeb02007-12-14 10:17:34 +08006818 kfree(kvm->arch.vpic);
6819 kfree(kvm->arch.vioapic);
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +08006820 kvm_free_vcpus(kvm);
Avi Kivity3d458302008-03-25 11:26:13 +02006821 if (kvm->arch.apic_access_page)
6822 put_page(kvm->arch.apic_access_page);
Sheng Yangb7ebfb02008-04-25 21:44:52 +08006823 if (kvm->arch.ept_identity_pagetable)
6824 put_page(kvm->arch.ept_identity_pagetable);
Gleb Natapov1e08ec42012-09-13 17:19:24 +03006825 kfree(rcu_dereference_check(kvm->arch.apic_map, 1));
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +08006826}
Zhang Xiantao0de10342007-11-20 16:25:04 +08006827
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +09006828void kvm_arch_free_memslot(struct kvm_memory_slot *free,
6829 struct kvm_memory_slot *dont)
6830{
6831 int i;
6832
Takuya Yoshikawad89cc612012-08-01 18:03:28 +09006833 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
6834 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
6835 kvm_kvfree(free->arch.rmap[i]);
6836 free->arch.rmap[i] = NULL;
Takuya Yoshikawa77d11302012-07-02 17:57:17 +09006837 }
Takuya Yoshikawad89cc612012-08-01 18:03:28 +09006838 if (i == 0)
6839 continue;
6840
6841 if (!dont || free->arch.lpage_info[i - 1] !=
6842 dont->arch.lpage_info[i - 1]) {
6843 kvm_kvfree(free->arch.lpage_info[i - 1]);
6844 free->arch.lpage_info[i - 1] = NULL;
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +09006845 }
6846 }
6847}
6848
6849int kvm_arch_create_memslot(struct kvm_memory_slot *slot, unsigned long npages)
6850{
6851 int i;
6852
Takuya Yoshikawad89cc612012-08-01 18:03:28 +09006853 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +09006854 unsigned long ugfn;
6855 int lpages;
Takuya Yoshikawad89cc612012-08-01 18:03:28 +09006856 int level = i + 1;
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +09006857
6858 lpages = gfn_to_index(slot->base_gfn + npages - 1,
6859 slot->base_gfn, level) + 1;
6860
Takuya Yoshikawad89cc612012-08-01 18:03:28 +09006861 slot->arch.rmap[i] =
6862 kvm_kvzalloc(lpages * sizeof(*slot->arch.rmap[i]));
6863 if (!slot->arch.rmap[i])
Takuya Yoshikawa77d11302012-07-02 17:57:17 +09006864 goto out_free;
Takuya Yoshikawad89cc612012-08-01 18:03:28 +09006865 if (i == 0)
6866 continue;
Takuya Yoshikawa77d11302012-07-02 17:57:17 +09006867
Takuya Yoshikawad89cc612012-08-01 18:03:28 +09006868 slot->arch.lpage_info[i - 1] = kvm_kvzalloc(lpages *
6869 sizeof(*slot->arch.lpage_info[i - 1]));
6870 if (!slot->arch.lpage_info[i - 1])
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +09006871 goto out_free;
6872
6873 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
Takuya Yoshikawad89cc612012-08-01 18:03:28 +09006874 slot->arch.lpage_info[i - 1][0].write_count = 1;
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +09006875 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
Takuya Yoshikawad89cc612012-08-01 18:03:28 +09006876 slot->arch.lpage_info[i - 1][lpages - 1].write_count = 1;
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +09006877 ugfn = slot->userspace_addr >> PAGE_SHIFT;
6878 /*
6879 * If the gfn and userspace address are not aligned wrt each
6880 * other, or if explicitly asked to, disable large page
6881 * support for this slot
6882 */
6883 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
6884 !kvm_largepages_enabled()) {
6885 unsigned long j;
6886
6887 for (j = 0; j < lpages; ++j)
Takuya Yoshikawad89cc612012-08-01 18:03:28 +09006888 slot->arch.lpage_info[i - 1][j].write_count = 1;
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +09006889 }
6890 }
6891
6892 return 0;
6893
6894out_free:
Takuya Yoshikawad89cc612012-08-01 18:03:28 +09006895 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
6896 kvm_kvfree(slot->arch.rmap[i]);
6897 slot->arch.rmap[i] = NULL;
6898 if (i == 0)
6899 continue;
6900
6901 kvm_kvfree(slot->arch.lpage_info[i - 1]);
6902 slot->arch.lpage_info[i - 1] = NULL;
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +09006903 }
6904 return -ENOMEM;
6905}
6906
Marcelo Tosattif7784b82009-12-23 14:35:18 -02006907int kvm_arch_prepare_memory_region(struct kvm *kvm,
6908 struct kvm_memory_slot *memslot,
Zhang Xiantao0de10342007-11-20 16:25:04 +08006909 struct kvm_memory_slot old,
Marcelo Tosattif7784b82009-12-23 14:35:18 -02006910 struct kvm_userspace_memory_region *mem,
Alex Williamsonf82a8cf2012-12-10 10:33:21 -07006911 bool user_alloc)
Zhang Xiantao0de10342007-11-20 16:25:04 +08006912{
Marcelo Tosattif7784b82009-12-23 14:35:18 -02006913 int npages = memslot->npages;
Avi Kivity7ac77092010-06-21 10:57:45 +03006914
Takuya Yoshikawa7a905b12013-02-07 18:55:57 +09006915 /*
6916 * Only private memory slots need to be mapped here since
6917 * KVM_SET_MEMORY_REGION ioctl is no longer supported.
Zhang Xiantao0de10342007-11-20 16:25:04 +08006918 */
Takuya Yoshikawa7a905b12013-02-07 18:55:57 +09006919 if ((memslot->id >= KVM_USER_MEM_SLOTS) && npages && !old.npages) {
6920 unsigned long userspace_addr;
Andrea Arcangeli604b38a2008-07-25 16:32:03 +02006921
Takuya Yoshikawa7a905b12013-02-07 18:55:57 +09006922 /*
6923 * MAP_SHARED to prevent internal slot pages from being moved
6924 * by fork()/COW.
6925 */
6926 userspace_addr = vm_mmap(NULL, 0, npages * PAGE_SIZE,
6927 PROT_READ | PROT_WRITE,
6928 MAP_SHARED | MAP_ANONYMOUS, 0);
Zhang Xiantao0de10342007-11-20 16:25:04 +08006929
Takuya Yoshikawa7a905b12013-02-07 18:55:57 +09006930 if (IS_ERR((void *)userspace_addr))
6931 return PTR_ERR((void *)userspace_addr);
Andrea Arcangeli604b38a2008-07-25 16:32:03 +02006932
Takuya Yoshikawa7a905b12013-02-07 18:55:57 +09006933 memslot->userspace_addr = userspace_addr;
Zhang Xiantao0de10342007-11-20 16:25:04 +08006934 }
6935
Marcelo Tosattif7784b82009-12-23 14:35:18 -02006936 return 0;
6937}
6938
6939void kvm_arch_commit_memory_region(struct kvm *kvm,
6940 struct kvm_userspace_memory_region *mem,
6941 struct kvm_memory_slot old,
Alex Williamsonf82a8cf2012-12-10 10:33:21 -07006942 bool user_alloc)
Marcelo Tosattif7784b82009-12-23 14:35:18 -02006943{
6944
Xiao Guangrong48c0e4e2011-03-04 18:59:21 +08006945 int nr_mmu_pages = 0, npages = mem->memory_size >> PAGE_SHIFT;
Marcelo Tosattif7784b82009-12-23 14:35:18 -02006946
Takuya Yoshikawa7a905b12013-02-07 18:55:57 +09006947 if ((mem->slot >= KVM_USER_MEM_SLOTS) && old.npages && !npages) {
Marcelo Tosattif7784b82009-12-23 14:35:18 -02006948 int ret;
6949
Al Virobfce2812012-04-20 21:57:04 -04006950 ret = vm_munmap(old.userspace_addr,
Marcelo Tosattif7784b82009-12-23 14:35:18 -02006951 old.npages * PAGE_SIZE);
Marcelo Tosattif7784b82009-12-23 14:35:18 -02006952 if (ret < 0)
6953 printk(KERN_WARNING
6954 "kvm_vm_ioctl_set_memory_region: "
6955 "failed to munmap memory\n");
6956 }
6957
Xiao Guangrong48c0e4e2011-03-04 18:59:21 +08006958 if (!kvm->arch.n_requested_mmu_pages)
6959 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
Zhang Xiantao0de10342007-11-20 16:25:04 +08006960
Xiao Guangrong48c0e4e2011-03-04 18:59:21 +08006961 if (nr_mmu_pages)
6962 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
Takuya Yoshikawac972f3b2013-01-08 19:43:28 +09006963 /*
6964 * Write protect all pages for dirty logging.
6965 * Existing largepage mappings are destroyed here and new ones will
6966 * not be created until the end of the logging.
6967 */
Takuya Yoshikawa9d1beef2013-01-08 19:46:48 +09006968 if (npages && (mem->flags & KVM_MEM_LOG_DIRTY_PAGES))
Takuya Yoshikawac972f3b2013-01-08 19:43:28 +09006969 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
Marcelo Tosatti3b4dc3a2012-08-28 17:43:26 -03006970 /*
6971 * If memory slot is created, or moved, we need to clear all
6972 * mmio sptes.
6973 */
6974 if (npages && old.base_gfn != mem->guest_phys_addr >> PAGE_SHIFT) {
6975 kvm_mmu_zap_all(kvm);
6976 kvm_reload_remote_mmus(kvm);
6977 }
Zhang Xiantao0de10342007-11-20 16:25:04 +08006978}
Zhang Xiantao1d737c82007-12-14 09:35:10 +08006979
Marcelo Tosatti2df72e92012-08-24 15:54:57 -03006980void kvm_arch_flush_shadow_all(struct kvm *kvm)
Marcelo Tosatti34d4cb82008-07-10 20:49:31 -03006981{
6982 kvm_mmu_zap_all(kvm);
Marcelo Tosatti8986ecc2009-05-12 18:55:45 -03006983 kvm_reload_remote_mmus(kvm);
Marcelo Tosatti34d4cb82008-07-10 20:49:31 -03006984}
6985
Marcelo Tosatti2df72e92012-08-24 15:54:57 -03006986void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
6987 struct kvm_memory_slot *slot)
6988{
6989 kvm_arch_flush_shadow_all(kvm);
6990}
6991
Zhang Xiantao1d737c82007-12-14 09:35:10 +08006992int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
6993{
Gleb Natapovaf585b92010-10-14 11:22:46 +02006994 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6995 !vcpu->arch.apf.halted)
6996 || !list_empty_careful(&vcpu->async_pf.done)
Gleb Natapova1b37102009-07-09 15:33:52 +03006997 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
Avi Kivity7460fb4a2011-09-20 13:43:14 +03006998 || atomic_read(&vcpu->arch.nmi_queued) ||
Gleb Natapova1b37102009-07-09 15:33:52 +03006999 (kvm_arch_interrupt_allowed(vcpu) &&
7000 kvm_cpu_has_interrupt(vcpu));
Zhang Xiantao1d737c82007-12-14 09:35:10 +08007001}
Zhang Xiantao57361992007-12-17 14:21:40 +08007002
Christoffer Dallb6d33832012-03-08 16:44:24 -05007003int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
Zhang Xiantao57361992007-12-17 14:21:40 +08007004{
Christoffer Dallb6d33832012-03-08 16:44:24 -05007005 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
Zhang Xiantao57361992007-12-17 14:21:40 +08007006}
Gleb Natapov78646122009-03-23 12:12:11 +02007007
7008int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
7009{
7010 return kvm_x86_ops->interrupt_allowed(vcpu);
7011}
Marcelo Tosatti229456f2009-06-17 09:22:14 -03007012
Jan Kiszkaf92653e2010-02-23 17:47:55 +01007013bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
7014{
7015 unsigned long current_rip = kvm_rip_read(vcpu) +
7016 get_segment_base(vcpu, VCPU_SREG_CS);
7017
7018 return current_rip == linear_rip;
7019}
7020EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
7021
Jan Kiszka94fe45d2009-10-18 13:24:44 +02007022unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
7023{
7024 unsigned long rflags;
7025
7026 rflags = kvm_x86_ops->get_rflags(vcpu);
7027 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
Jan Kiszkac310bac2010-02-23 17:47:58 +01007028 rflags &= ~X86_EFLAGS_TF;
Jan Kiszka94fe45d2009-10-18 13:24:44 +02007029 return rflags;
7030}
7031EXPORT_SYMBOL_GPL(kvm_get_rflags);
7032
7033void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
7034{
7035 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
Jan Kiszkaf92653e2010-02-23 17:47:55 +01007036 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
Jan Kiszkac310bac2010-02-23 17:47:58 +01007037 rflags |= X86_EFLAGS_TF;
Jan Kiszka94fe45d2009-10-18 13:24:44 +02007038 kvm_x86_ops->set_rflags(vcpu, rflags);
Avi Kivity3842d132010-07-27 12:30:24 +03007039 kvm_make_request(KVM_REQ_EVENT, vcpu);
Jan Kiszka94fe45d2009-10-18 13:24:44 +02007040}
7041EXPORT_SYMBOL_GPL(kvm_set_rflags);
7042
Gleb Natapov56028d02010-10-17 18:13:42 +02007043void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
7044{
7045 int r;
7046
Xiao Guangrongfb67e142010-12-07 10:35:25 +08007047 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
Xiao Guangrongc4806ac2010-11-12 14:49:55 +08007048 is_error_page(work->page))
Gleb Natapov56028d02010-10-17 18:13:42 +02007049 return;
7050
7051 r = kvm_mmu_reload(vcpu);
7052 if (unlikely(r))
7053 return;
7054
Xiao Guangrongfb67e142010-12-07 10:35:25 +08007055 if (!vcpu->arch.mmu.direct_map &&
7056 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
7057 return;
7058
Gleb Natapov56028d02010-10-17 18:13:42 +02007059 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
7060}
7061
Gleb Natapovaf585b92010-10-14 11:22:46 +02007062static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
7063{
7064 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
7065}
7066
7067static inline u32 kvm_async_pf_next_probe(u32 key)
7068{
7069 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
7070}
7071
7072static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7073{
7074 u32 key = kvm_async_pf_hash_fn(gfn);
7075
7076 while (vcpu->arch.apf.gfns[key] != ~0)
7077 key = kvm_async_pf_next_probe(key);
7078
7079 vcpu->arch.apf.gfns[key] = gfn;
7080}
7081
7082static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
7083{
7084 int i;
7085 u32 key = kvm_async_pf_hash_fn(gfn);
7086
7087 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
Xiao Guangrongc7d28c22010-11-01 17:00:30 +08007088 (vcpu->arch.apf.gfns[key] != gfn &&
7089 vcpu->arch.apf.gfns[key] != ~0); i++)
Gleb Natapovaf585b92010-10-14 11:22:46 +02007090 key = kvm_async_pf_next_probe(key);
7091
7092 return key;
7093}
7094
7095bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7096{
7097 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
7098}
7099
7100static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7101{
7102 u32 i, j, k;
7103
7104 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
7105 while (true) {
7106 vcpu->arch.apf.gfns[i] = ~0;
7107 do {
7108 j = kvm_async_pf_next_probe(j);
7109 if (vcpu->arch.apf.gfns[j] == ~0)
7110 return;
7111 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
7112 /*
7113 * k lies cyclically in ]i,j]
7114 * | i.k.j |
7115 * |....j i.k.| or |.k..j i...|
7116 */
7117 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
7118 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
7119 i = j;
7120 }
7121}
7122
Gleb Natapov7c907052010-10-14 11:22:53 +02007123static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
7124{
7125
7126 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
7127 sizeof(val));
7128}
7129
Gleb Natapovaf585b92010-10-14 11:22:46 +02007130void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
7131 struct kvm_async_pf *work)
7132{
Avi Kivity6389ee92010-11-29 16:12:30 +02007133 struct x86_exception fault;
7134
Gleb Natapov7c907052010-10-14 11:22:53 +02007135 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
Gleb Natapovaf585b92010-10-14 11:22:46 +02007136 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
Gleb Natapov7c907052010-10-14 11:22:53 +02007137
7138 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
Gleb Natapovfc5f06f2010-10-14 11:22:56 +02007139 (vcpu->arch.apf.send_user_only &&
7140 kvm_x86_ops->get_cpl(vcpu) == 0))
Gleb Natapov7c907052010-10-14 11:22:53 +02007141 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
7142 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
Avi Kivity6389ee92010-11-29 16:12:30 +02007143 fault.vector = PF_VECTOR;
7144 fault.error_code_valid = true;
7145 fault.error_code = 0;
7146 fault.nested_page_fault = false;
7147 fault.address = work->arch.token;
7148 kvm_inject_page_fault(vcpu, &fault);
Gleb Natapov7c907052010-10-14 11:22:53 +02007149 }
Gleb Natapovaf585b92010-10-14 11:22:46 +02007150}
7151
7152void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
7153 struct kvm_async_pf *work)
7154{
Avi Kivity6389ee92010-11-29 16:12:30 +02007155 struct x86_exception fault;
7156
Gleb Natapov7c907052010-10-14 11:22:53 +02007157 trace_kvm_async_pf_ready(work->arch.token, work->gva);
7158 if (is_error_page(work->page))
7159 work->arch.token = ~0; /* broadcast wakeup */
7160 else
7161 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
7162
7163 if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
7164 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
Avi Kivity6389ee92010-11-29 16:12:30 +02007165 fault.vector = PF_VECTOR;
7166 fault.error_code_valid = true;
7167 fault.error_code = 0;
7168 fault.nested_page_fault = false;
7169 fault.address = work->arch.token;
7170 kvm_inject_page_fault(vcpu, &fault);
Gleb Natapov7c907052010-10-14 11:22:53 +02007171 }
Xiao Guangronge6d53e32010-11-01 17:01:28 +08007172 vcpu->arch.apf.halted = false;
Gleb Natapova4fa1632012-05-03 11:36:39 +03007173 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
Gleb Natapov7c907052010-10-14 11:22:53 +02007174}
7175
7176bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
7177{
7178 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
7179 return true;
7180 else
7181 return !kvm_event_needs_reinjection(vcpu) &&
7182 kvm_x86_ops->interrupt_allowed(vcpu);
Gleb Natapovaf585b92010-10-14 11:22:46 +02007183}
7184
Marcelo Tosatti229456f2009-06-17 09:22:14 -03007185EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
7186EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
7187EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
7188EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
7189EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
Joerg Roedel0ac406d2009-10-09 16:08:27 +02007190EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
Joerg Roedeld8cabdd2009-10-09 16:08:28 +02007191EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
Joerg Roedel17897f32009-10-09 16:08:29 +02007192EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
Joerg Roedel236649d2009-10-09 16:08:30 +02007193EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
Joerg Roedelec1ff792009-10-09 16:08:31 +02007194EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
Joerg Roedel532a46b2009-10-09 16:08:32 +02007195EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
Joerg Roedel2e554e82010-02-24 18:59:14 +01007196EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);