blob: bf040c4e02b332b7dd2126ae6ed2013261ecfdb4 [file] [log] [blame]
Avi Kivity6aa8b732006-12-10 02:21:36 -08001/*
2 * Kernel-based Virtual Machine driver for Linux
3 *
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
6 *
7 * Copyright (C) 2006 Qumranet, Inc.
Nicolas Kaiser9611c182010-10-06 14:23:22 +02008 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
Avi Kivity6aa8b732006-12-10 02:21:36 -08009 *
10 * Authors:
11 * Avi Kivity <avi@qumranet.com>
12 * Yaniv Kamay <yaniv@qumranet.com>
13 *
14 * This work is licensed under the terms of the GNU GPL, version 2. See
15 * the COPYING file in the top-level directory.
16 *
17 */
18
Hollis Blancharde2174022007-12-03 15:30:24 -060019#include "iodev.h"
Avi Kivity6aa8b732006-12-10 02:21:36 -080020
Avi Kivityedf88412007-12-16 11:02:48 +020021#include <linux/kvm_host.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080022#include <linux/kvm.h>
23#include <linux/module.h>
24#include <linux/errno.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080025#include <linux/percpu.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080026#include <linux/mm.h>
27#include <linux/miscdevice.h>
28#include <linux/vmalloc.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080029#include <linux/reboot.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080030#include <linux/debugfs.h>
31#include <linux/highmem.h>
32#include <linux/file.h>
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +010033#include <linux/syscore_ops.h>
Avi Kivity774c47f2007-02-12 00:54:47 -080034#include <linux/cpu.h>
Alexey Dobriyane8edc6e2007-05-21 01:22:52 +040035#include <linux/sched.h>
Avi Kivityd9e368d2007-06-07 19:18:30 +030036#include <linux/cpumask.h>
37#include <linux/smp.h>
Avi Kivityd6d28162007-06-28 08:38:16 -040038#include <linux/anon_inodes.h>
Avi Kivity04d2cc72007-09-10 18:10:54 +030039#include <linux/profile.h>
Anthony Liguori7aa81cc2007-09-17 14:57:50 -050040#include <linux/kvm_para.h>
Izik Eidus6fc138d2007-10-09 19:20:39 +020041#include <linux/pagemap.h>
Anthony Liguori8d4e1282007-10-18 09:59:34 -050042#include <linux/mman.h>
Anthony Liguori35149e22008-04-02 14:46:56 -050043#include <linux/swap.h>
Sheng Yange56d5322009-03-12 21:45:39 +080044#include <linux/bitops.h>
Marcelo Tosatti547de292009-05-07 17:55:13 -030045#include <linux/spinlock.h>
Arnd Bergmann6ff58942009-10-22 14:19:27 +020046#include <linux/compat.h>
Marcelo Tosattibc6678a2009-12-23 14:35:21 -020047#include <linux/srcu.h>
Joerg Roedel8f0b1ab2010-01-28 12:37:56 +010048#include <linux/hugetlb.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090049#include <linux/slab.h>
Sasha Levin743eeb02011-07-27 16:00:48 +030050#include <linux/sort.h>
51#include <linux/bsearch.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080052
Avi Kivitye4956062007-06-28 14:15:57 -040053#include <asm/processor.h>
Avi Kivitye4956062007-06-28 14:15:57 -040054#include <asm/io.h>
55#include <asm/uaccess.h>
Izik Eidus3e021bf2007-11-19 11:16:57 +020056#include <asm/pgtable.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080057
Laurent Vivier5f94c172008-05-30 16:05:54 +020058#include "coalesced_mmio.h"
Gleb Natapovaf585b92010-10-14 11:22:46 +020059#include "async_pf.h"
Laurent Vivier5f94c172008-05-30 16:05:54 +020060
Marcelo Tosatti229456f2009-06-17 09:22:14 -030061#define CREATE_TRACE_POINTS
62#include <trace/events/kvm.h>
63
Avi Kivity6aa8b732006-12-10 02:21:36 -080064MODULE_AUTHOR("Qumranet");
65MODULE_LICENSE("GPL");
66
Marcelo Tosattifa40a822009-06-04 15:08:24 -030067/*
68 * Ordering of locks:
69 *
Sheng Yangfae3a352009-12-15 10:28:07 +080070 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
Marcelo Tosattifa40a822009-06-04 15:08:24 -030071 */
72
Jan Kiszkae935b832011-02-08 12:55:33 +010073DEFINE_RAW_SPINLOCK(kvm_lock);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +080074LIST_HEAD(vm_list);
Avi Kivity133de902007-02-12 00:54:44 -080075
Rusty Russell7f59f492008-12-07 21:25:45 +103076static cpumask_var_t cpus_hardware_enabled;
Alexander Graf10474ae2009-09-15 11:37:46 +020077static int kvm_usage_count = 0;
78static atomic_t hardware_enable_failed;
Avi Kivity1b6c0162007-05-24 13:03:52 +030079
Rusty Russellc16f8622007-07-30 21:12:19 +100080struct kmem_cache *kvm_vcpu_cache;
81EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
Avi Kivity1165f5f2007-04-19 17:27:43 +030082
Avi Kivity15ad7142007-07-11 18:17:21 +030083static __read_mostly struct preempt_ops kvm_preempt_ops;
84
Hollis Blanchard76f7c872008-04-15 16:05:42 -050085struct dentry *kvm_debugfs_dir;
Avi Kivity6aa8b732006-12-10 02:21:36 -080086
Avi Kivitybccf2152007-02-21 18:04:26 +020087static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
88 unsigned long arg);
Alexander Graf1dda6062011-06-08 02:45:37 +020089#ifdef CONFIG_COMPAT
90static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
91 unsigned long arg);
92#endif
Alexander Graf10474ae2009-09-15 11:37:46 +020093static int hardware_enable_all(void);
94static void hardware_disable_all(void);
Avi Kivitybccf2152007-02-21 18:04:26 +020095
Marcelo Tosattie93f8a02009-12-23 14:35:24 -020096static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
97
Avi Kivityb7c41452010-12-02 17:52:50 +020098bool kvm_rebooting;
99EXPORT_SYMBOL_GPL(kvm_rebooting);
Avi Kivity4ecac3f2008-05-13 13:23:38 +0300100
Marcelo Tosatti54dee992009-06-11 12:07:44 -0300101static bool largepages_enabled = true;
102
Xiao Guangronga2766322012-07-26 11:58:59 +0800103bool kvm_is_mmio_pfn(pfn_t pfn)
Ben-Ami Yassourcbff90a2008-07-28 19:26:24 +0300104{
Andrea Arcangeli11feeb42013-07-25 03:04:38 +0200105 if (pfn_valid(pfn))
106 return PageReserved(pfn_to_page(pfn));
Ben-Ami Yassourcbff90a2008-07-28 19:26:24 +0300107
108 return true;
109}
110
Avi Kivity6aa8b732006-12-10 02:21:36 -0800111/*
112 * Switches to specified vcpu, until a matching vcpu_put()
113 */
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +0300114int vcpu_load(struct kvm_vcpu *vcpu)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800115{
Avi Kivity15ad7142007-07-11 18:17:21 +0300116 int cpu;
117
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +0300118 if (mutex_lock_killable(&vcpu->mutex))
119 return -EINTR;
Rik van Riel34bb10b2011-02-01 09:52:41 -0500120 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
121 /* The thread running this VCPU changed. */
122 struct pid *oldpid = vcpu->pid;
123 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
124 rcu_assign_pointer(vcpu->pid, newpid);
125 synchronize_rcu();
126 put_pid(oldpid);
127 }
Avi Kivity15ad7142007-07-11 18:17:21 +0300128 cpu = get_cpu();
129 preempt_notifier_register(&vcpu->preempt_notifier);
Carsten Otte313a3dc2007-10-11 19:16:52 +0200130 kvm_arch_vcpu_load(vcpu, cpu);
Avi Kivity15ad7142007-07-11 18:17:21 +0300131 put_cpu();
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +0300132 return 0;
Avi Kivitybccf2152007-02-21 18:04:26 +0200133}
134
Carsten Otte313a3dc2007-10-11 19:16:52 +0200135void vcpu_put(struct kvm_vcpu *vcpu)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800136{
Avi Kivity15ad7142007-07-11 18:17:21 +0300137 preempt_disable();
Carsten Otte313a3dc2007-10-11 19:16:52 +0200138 kvm_arch_vcpu_put(vcpu);
Avi Kivity15ad7142007-07-11 18:17:21 +0300139 preempt_notifier_unregister(&vcpu->preempt_notifier);
140 preempt_enable();
Avi Kivity6aa8b732006-12-10 02:21:36 -0800141 mutex_unlock(&vcpu->mutex);
142}
143
Avi Kivityd9e368d2007-06-07 19:18:30 +0300144static void ack_flush(void *_completed)
145{
Avi Kivityd9e368d2007-06-07 19:18:30 +0300146}
147
Rusty Russell49846892008-12-08 20:26:24 +1030148static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
Avi Kivityd9e368d2007-06-07 19:18:30 +0300149{
Avi Kivity597a5f52008-07-20 14:24:22 +0300150 int i, cpu, me;
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030151 cpumask_var_t cpus;
152 bool called = true;
Avi Kivityd9e368d2007-06-07 19:18:30 +0300153 struct kvm_vcpu *vcpu;
Avi Kivityd9e368d2007-06-07 19:18:30 +0300154
Li Zefan79f55992009-06-15 14:58:26 +0800155 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030156
Xiao Guangrong3cba4132011-01-12 15:41:22 +0800157 me = get_cpu();
Gleb Natapov988a2ca2009-06-09 15:56:29 +0300158 kvm_for_each_vcpu(i, vcpu, kvm) {
Xiao Guangrong3cba4132011-01-12 15:41:22 +0800159 kvm_make_request(req, vcpu);
Avi Kivityd9e368d2007-06-07 19:18:30 +0300160 cpu = vcpu->cpu;
Xiao Guangrong6b7e2d02011-01-12 15:40:31 +0800161
162 /* Set ->requests bit before we read ->mode */
163 smp_mb();
164
165 if (cpus != NULL && cpu != -1 && cpu != me &&
166 kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030167 cpumask_set_cpu(cpu, cpus);
Avi Kivityd9e368d2007-06-07 19:18:30 +0300168 }
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030169 if (unlikely(cpus == NULL))
170 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
171 else if (!cpumask_empty(cpus))
172 smp_call_function_many(cpus, ack_flush, NULL, 1);
173 else
174 called = false;
Xiao Guangrong3cba4132011-01-12 15:41:22 +0800175 put_cpu();
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030176 free_cpumask_var(cpus);
Rusty Russell49846892008-12-08 20:26:24 +1030177 return called;
178}
179
180void kvm_flush_remote_tlbs(struct kvm *kvm)
181{
Alex Shibec87d62012-03-04 13:13:38 +0800182 long dirty_count = kvm->tlbs_dirty;
Xiao Guangronga4ee1ca2010-11-23 11:13:00 +0800183
184 smp_mb();
Rusty Russell49846892008-12-08 20:26:24 +1030185 if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
186 ++kvm->stat.remote_tlb_flush;
Xiao Guangronga4ee1ca2010-11-23 11:13:00 +0800187 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
Avi Kivityd9e368d2007-06-07 19:18:30 +0300188}
189
Marcelo Tosatti2e53d632008-02-20 14:47:24 -0500190void kvm_reload_remote_mmus(struct kvm *kvm)
191{
Rusty Russell49846892008-12-08 20:26:24 +1030192 make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
Marcelo Tosatti2e53d632008-02-20 14:47:24 -0500193}
194
Marcelo Tosattid8281992012-11-27 23:29:01 -0200195void kvm_make_mclock_inprogress_request(struct kvm *kvm)
196{
197 make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
198}
199
Yang Zhang3d81bc72013-04-11 19:25:13 +0800200void kvm_make_scan_ioapic_request(struct kvm *kvm)
Yang Zhangc7c9c562013-01-25 10:18:51 +0800201{
Yang Zhang3d81bc72013-04-11 19:25:13 +0800202 make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
Yang Zhangc7c9c562013-01-25 10:18:51 +0800203}
204
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000205int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
206{
207 struct page *page;
208 int r;
209
210 mutex_init(&vcpu->mutex);
211 vcpu->cpu = -1;
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000212 vcpu->kvm = kvm;
213 vcpu->vcpu_id = id;
Rik van Riel34bb10b2011-02-01 09:52:41 -0500214 vcpu->pid = NULL;
Eddie Dongb6958ce2007-07-18 12:15:21 +0300215 init_waitqueue_head(&vcpu->wq);
Gleb Natapovaf585b92010-10-14 11:22:46 +0200216 kvm_async_pf_vcpu_init(vcpu);
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000217
218 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
219 if (!page) {
220 r = -ENOMEM;
221 goto fail;
222 }
223 vcpu->run = page_address(page);
224
Raghavendra K T4c088492012-07-18 19:07:46 +0530225 kvm_vcpu_set_in_spin_loop(vcpu, false);
226 kvm_vcpu_set_dy_eligible(vcpu, false);
Raghavendra K T3a08a8f2013-03-04 23:32:07 +0530227 vcpu->preempted = false;
Raghavendra K T4c088492012-07-18 19:07:46 +0530228
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +0800229 r = kvm_arch_vcpu_init(vcpu);
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000230 if (r < 0)
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +0800231 goto fail_free_run;
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000232 return 0;
233
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000234fail_free_run:
235 free_page((unsigned long)vcpu->run);
236fail:
Rusty Russell76fafa52007-10-08 10:50:48 +1000237 return r;
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000238}
239EXPORT_SYMBOL_GPL(kvm_vcpu_init);
240
241void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
242{
Rik van Riel34bb10b2011-02-01 09:52:41 -0500243 put_pid(vcpu->pid);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +0800244 kvm_arch_vcpu_uninit(vcpu);
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000245 free_page((unsigned long)vcpu->run);
246}
247EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
248
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200249#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
250static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
251{
252 return container_of(mn, struct kvm, mmu_notifier);
253}
254
255static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
256 struct mm_struct *mm,
257 unsigned long address)
258{
259 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200260 int need_tlb_flush, idx;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200261
262 /*
263 * When ->invalidate_page runs, the linux pte has been zapped
264 * already but the page is still allocated until
265 * ->invalidate_page returns. So if we increase the sequence
266 * here the kvm page fault will notice if the spte can't be
267 * established because the page is going to be freed. If
268 * instead the kvm page fault establishes the spte before
269 * ->invalidate_page runs, kvm_unmap_hva will release it
270 * before returning.
271 *
272 * The sequence increase only need to be seen at spin_unlock
273 * time, and not at spin_lock time.
274 *
275 * Increasing the sequence after the spin_unlock would be
276 * unsafe because the kvm page fault could then establish the
277 * pte after kvm_unmap_hva returned, without noticing the page
278 * is going to be freed.
279 */
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200280 idx = srcu_read_lock(&kvm->srcu);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200281 spin_lock(&kvm->mmu_lock);
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900282
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200283 kvm->mmu_notifier_seq++;
Xiao Guangronga4ee1ca2010-11-23 11:13:00 +0800284 need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200285 /* we've to flush the tlb before the pages can be freed */
286 if (need_tlb_flush)
287 kvm_flush_remote_tlbs(kvm);
288
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900289 spin_unlock(&kvm->mmu_lock);
290 srcu_read_unlock(&kvm->srcu, idx);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200291}
292
Izik Eidus3da0dd42009-09-23 21:47:18 +0300293static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
294 struct mm_struct *mm,
295 unsigned long address,
296 pte_t pte)
297{
298 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200299 int idx;
Izik Eidus3da0dd42009-09-23 21:47:18 +0300300
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200301 idx = srcu_read_lock(&kvm->srcu);
Izik Eidus3da0dd42009-09-23 21:47:18 +0300302 spin_lock(&kvm->mmu_lock);
303 kvm->mmu_notifier_seq++;
304 kvm_set_spte_hva(kvm, address, pte);
305 spin_unlock(&kvm->mmu_lock);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200306 srcu_read_unlock(&kvm->srcu, idx);
Izik Eidus3da0dd42009-09-23 21:47:18 +0300307}
308
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200309static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
310 struct mm_struct *mm,
311 unsigned long start,
312 unsigned long end)
313{
314 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200315 int need_tlb_flush = 0, idx;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200316
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200317 idx = srcu_read_lock(&kvm->srcu);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200318 spin_lock(&kvm->mmu_lock);
319 /*
320 * The count increase must become visible at unlock time as no
321 * spte can be established without taking the mmu_lock and
322 * count is also read inside the mmu_lock critical section.
323 */
324 kvm->mmu_notifier_count++;
Takuya Yoshikawab3ae2092012-07-02 17:56:33 +0900325 need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
Xiao Guangronga4ee1ca2010-11-23 11:13:00 +0800326 need_tlb_flush |= kvm->tlbs_dirty;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200327 /* we've to flush the tlb before the pages can be freed */
328 if (need_tlb_flush)
329 kvm_flush_remote_tlbs(kvm);
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900330
331 spin_unlock(&kvm->mmu_lock);
332 srcu_read_unlock(&kvm->srcu, idx);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200333}
334
335static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
336 struct mm_struct *mm,
337 unsigned long start,
338 unsigned long end)
339{
340 struct kvm *kvm = mmu_notifier_to_kvm(mn);
341
342 spin_lock(&kvm->mmu_lock);
343 /*
344 * This sequence increase will notify the kvm page fault that
345 * the page that is going to be mapped in the spte could have
346 * been freed.
347 */
348 kvm->mmu_notifier_seq++;
Paul Mackerrasa355aa52011-12-12 12:37:21 +0000349 smp_wmb();
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200350 /*
351 * The above sequence increase must be visible before the
Paul Mackerrasa355aa52011-12-12 12:37:21 +0000352 * below count decrease, which is ensured by the smp_wmb above
353 * in conjunction with the smp_rmb in mmu_notifier_retry().
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200354 */
355 kvm->mmu_notifier_count--;
356 spin_unlock(&kvm->mmu_lock);
357
358 BUG_ON(kvm->mmu_notifier_count < 0);
359}
360
361static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
362 struct mm_struct *mm,
363 unsigned long address)
364{
365 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200366 int young, idx;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200367
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200368 idx = srcu_read_lock(&kvm->srcu);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200369 spin_lock(&kvm->mmu_lock);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200370
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900371 young = kvm_age_hva(kvm, address);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200372 if (young)
373 kvm_flush_remote_tlbs(kvm);
374
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900375 spin_unlock(&kvm->mmu_lock);
376 srcu_read_unlock(&kvm->srcu, idx);
377
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200378 return young;
379}
380
Andrea Arcangeli8ee53822011-01-13 15:47:10 -0800381static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
382 struct mm_struct *mm,
383 unsigned long address)
384{
385 struct kvm *kvm = mmu_notifier_to_kvm(mn);
386 int young, idx;
387
388 idx = srcu_read_lock(&kvm->srcu);
389 spin_lock(&kvm->mmu_lock);
390 young = kvm_test_age_hva(kvm, address);
391 spin_unlock(&kvm->mmu_lock);
392 srcu_read_unlock(&kvm->srcu, idx);
393
394 return young;
395}
396
Marcelo Tosatti85db06e2008-12-10 21:23:26 +0100397static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
398 struct mm_struct *mm)
399{
400 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Lai Jiangshaneda2bed2010-04-20 14:29:29 +0800401 int idx;
402
403 idx = srcu_read_lock(&kvm->srcu);
Marcelo Tosatti2df72e92012-08-24 15:54:57 -0300404 kvm_arch_flush_shadow_all(kvm);
Lai Jiangshaneda2bed2010-04-20 14:29:29 +0800405 srcu_read_unlock(&kvm->srcu, idx);
Marcelo Tosatti85db06e2008-12-10 21:23:26 +0100406}
407
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200408static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
409 .invalidate_page = kvm_mmu_notifier_invalidate_page,
410 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
411 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
412 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
Andrea Arcangeli8ee53822011-01-13 15:47:10 -0800413 .test_young = kvm_mmu_notifier_test_young,
Izik Eidus3da0dd42009-09-23 21:47:18 +0300414 .change_pte = kvm_mmu_notifier_change_pte,
Marcelo Tosatti85db06e2008-12-10 21:23:26 +0100415 .release = kvm_mmu_notifier_release,
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200416};
Avi Kivity4c07b0a2009-12-20 14:54:04 +0200417
418static int kvm_init_mmu_notifier(struct kvm *kvm)
419{
420 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
421 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
422}
423
424#else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
425
426static int kvm_init_mmu_notifier(struct kvm *kvm)
427{
428 return 0;
429}
430
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200431#endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
432
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800433static void kvm_init_memslots_id(struct kvm *kvm)
434{
435 int i;
436 struct kvm_memslots *slots = kvm->memslots;
437
438 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
Xiao Guangrongf85e2cb2011-11-24 17:41:54 +0800439 slots->id_to_index[i] = slots->memslots[i].id = i;
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800440}
441
Carsten Ottee08b9632012-01-04 10:25:20 +0100442static struct kvm *kvm_create_vm(unsigned long type)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800443{
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100444 int r, i;
445 struct kvm *kvm = kvm_arch_alloc_vm();
Avi Kivity6aa8b732006-12-10 02:21:36 -0800446
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100447 if (!kvm)
448 return ERR_PTR(-ENOMEM);
449
Carsten Ottee08b9632012-01-04 10:25:20 +0100450 r = kvm_arch_init_vm(kvm, type);
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100451 if (r)
452 goto out_err_nodisable;
Alexander Graf10474ae2009-09-15 11:37:46 +0200453
454 r = hardware_enable_all();
455 if (r)
456 goto out_err_nodisable;
457
Avi Kivity75858a82009-01-04 17:10:50 +0200458#ifdef CONFIG_HAVE_KVM_IRQCHIP
459 INIT_HLIST_HEAD(&kvm->mask_notifier_list);
Gleb Natapov136bdfe2009-08-24 11:54:23 +0300460 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
Avi Kivity75858a82009-01-04 17:10:50 +0200461#endif
Avi Kivity6aa8b732006-12-10 02:21:36 -0800462
Alex Williamson1e702d92012-12-10 10:33:32 -0700463 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
464
Marcelo Tosatti46a26bf2009-12-23 14:35:16 -0200465 r = -ENOMEM;
466 kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
467 if (!kvm->memslots)
Jan Kiszka57e7fbe2010-11-09 12:42:12 +0100468 goto out_err_nosrcu;
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800469 kvm_init_memslots_id(kvm);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200470 if (init_srcu_struct(&kvm->srcu))
Jan Kiszka57e7fbe2010-11-09 12:42:12 +0100471 goto out_err_nosrcu;
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200472 for (i = 0; i < KVM_NR_BUSES; i++) {
473 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
474 GFP_KERNEL);
Jan Kiszka57e7fbe2010-11-09 12:42:12 +0100475 if (!kvm->buses[i])
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200476 goto out_err;
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200477 }
Mike Waychison74b5c5b2011-06-03 13:04:53 -0700478
OGAWA Hirofumi85722cd2011-05-11 09:28:28 +0900479 spin_lock_init(&kvm->mmu_lock);
Avi Kivity6d4e4c42007-11-21 16:41:05 +0200480 kvm->mm = current->mm;
481 atomic_inc(&kvm->mm->mm_count);
Gregory Haskinsd34e6b12009-07-07 17:08:49 -0400482 kvm_eventfd_init(kvm);
Shaohua Li11ec2802007-07-23 14:51:37 +0800483 mutex_init(&kvm->lock);
Marcelo Tosatti60eead72009-06-04 15:08:23 -0300484 mutex_init(&kvm->irq_lock);
Marcelo Tosatti79fac952009-12-23 14:35:26 -0200485 mutex_init(&kvm->slots_lock);
Izik Eidusd39f13b2008-03-30 16:01:25 +0300486 atomic_set(&kvm->users_count, 1);
Scott Wood07f0a7b2013-04-25 14:11:23 +0000487 INIT_LIST_HEAD(&kvm->devices);
Mike Waychison74b5c5b2011-06-03 13:04:53 -0700488
489 r = kvm_init_mmu_notifier(kvm);
490 if (r)
491 goto out_err;
492
Jan Kiszkae935b832011-02-08 12:55:33 +0100493 raw_spin_lock(&kvm_lock);
Rusty Russell5e58cfe2007-07-23 17:08:21 +1000494 list_add(&kvm->vm_list, &vm_list);
Jan Kiszkae935b832011-02-08 12:55:33 +0100495 raw_spin_unlock(&kvm_lock);
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100496
Avi Kivityf17abe92007-02-21 19:28:04 +0200497 return kvm;
Alexander Graf10474ae2009-09-15 11:37:46 +0200498
499out_err:
Jan Kiszka57e7fbe2010-11-09 12:42:12 +0100500 cleanup_srcu_struct(&kvm->srcu);
501out_err_nosrcu:
Alexander Graf10474ae2009-09-15 11:37:46 +0200502 hardware_disable_all();
503out_err_nodisable:
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200504 for (i = 0; i < KVM_NR_BUSES; i++)
505 kfree(kvm->buses[i]);
Marcelo Tosatti46a26bf2009-12-23 14:35:16 -0200506 kfree(kvm->memslots);
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100507 kvm_arch_free_vm(kvm);
Alexander Graf10474ae2009-09-15 11:37:46 +0200508 return ERR_PTR(r);
Avi Kivityf17abe92007-02-21 19:28:04 +0200509}
510
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900511/*
512 * Avoid using vmalloc for a small buffer.
513 * Should not be used when the size is statically known.
514 */
Takuya Yoshikawac1a7b322012-05-20 13:15:07 +0900515void *kvm_kvzalloc(unsigned long size)
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900516{
517 if (size > PAGE_SIZE)
518 return vzalloc(size);
519 else
520 return kzalloc(size, GFP_KERNEL);
521}
522
Takuya Yoshikawac1a7b322012-05-20 13:15:07 +0900523void kvm_kvfree(const void *addr)
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900524{
525 if (is_vmalloc_addr(addr))
526 vfree(addr);
527 else
528 kfree(addr);
529}
530
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900531static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
532{
533 if (!memslot->dirty_bitmap)
534 return;
535
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900536 kvm_kvfree(memslot->dirty_bitmap);
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900537 memslot->dirty_bitmap = NULL;
538}
539
Avi Kivity6aa8b732006-12-10 02:21:36 -0800540/*
541 * Free any memory in @free but not in @dont.
542 */
543static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
544 struct kvm_memory_slot *dont)
545{
Avi Kivity6aa8b732006-12-10 02:21:36 -0800546 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900547 kvm_destroy_dirty_bitmap(free);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800548
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +0900549 kvm_arch_free_memslot(free, dont);
Marcelo Tosatti05da4552008-02-23 11:44:30 -0300550
Avi Kivity6aa8b732006-12-10 02:21:36 -0800551 free->npages = 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800552}
553
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +0800554void kvm_free_physmem(struct kvm *kvm)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800555{
Marcelo Tosatti46a26bf2009-12-23 14:35:16 -0200556 struct kvm_memslots *slots = kvm->memslots;
Xiao Guangrongbe6ba0f2011-11-24 17:39:18 +0800557 struct kvm_memory_slot *memslot;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800558
Xiao Guangrongbe6ba0f2011-11-24 17:39:18 +0800559 kvm_for_each_memslot(memslot, slots)
560 kvm_free_physmem_slot(memslot, NULL);
Marcelo Tosatti46a26bf2009-12-23 14:35:16 -0200561
562 kfree(kvm->memslots);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800563}
564
Scott Wood07f0a7b2013-04-25 14:11:23 +0000565static void kvm_destroy_devices(struct kvm *kvm)
566{
567 struct list_head *node, *tmp;
568
569 list_for_each_safe(node, tmp, &kvm->devices) {
570 struct kvm_device *dev =
571 list_entry(node, struct kvm_device, vm_node);
572
573 list_del(node);
574 dev->ops->destroy(dev);
575 }
576}
577
Avi Kivityf17abe92007-02-21 19:28:04 +0200578static void kvm_destroy_vm(struct kvm *kvm)
579{
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200580 int i;
Avi Kivity6d4e4c42007-11-21 16:41:05 +0200581 struct mm_struct *mm = kvm->mm;
582
Sheng Yangad8ba2c2009-01-06 10:03:02 +0800583 kvm_arch_sync_events(kvm);
Jan Kiszkae935b832011-02-08 12:55:33 +0100584 raw_spin_lock(&kvm_lock);
Avi Kivity133de902007-02-12 00:54:44 -0800585 list_del(&kvm->vm_list);
Jan Kiszkae935b832011-02-08 12:55:33 +0100586 raw_spin_unlock(&kvm_lock);
Avi Kivity399ec802008-11-19 13:58:46 +0200587 kvm_free_irq_routing(kvm);
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200588 for (i = 0; i < KVM_NR_BUSES; i++)
589 kvm_io_bus_destroy(kvm->buses[i]);
Avi Kivity980da6c2009-12-20 15:13:43 +0200590 kvm_coalesced_mmio_free(kvm);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200591#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
592 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
Gleb Natapovf00be0c2009-03-19 12:20:36 +0200593#else
Marcelo Tosatti2df72e92012-08-24 15:54:57 -0300594 kvm_arch_flush_shadow_all(kvm);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200595#endif
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +0800596 kvm_arch_destroy_vm(kvm);
Scott Wood07f0a7b2013-04-25 14:11:23 +0000597 kvm_destroy_devices(kvm);
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100598 kvm_free_physmem(kvm);
599 cleanup_srcu_struct(&kvm->srcu);
600 kvm_arch_free_vm(kvm);
Alexander Graf10474ae2009-09-15 11:37:46 +0200601 hardware_disable_all();
Avi Kivity6d4e4c42007-11-21 16:41:05 +0200602 mmdrop(mm);
Avi Kivityf17abe92007-02-21 19:28:04 +0200603}
604
Izik Eidusd39f13b2008-03-30 16:01:25 +0300605void kvm_get_kvm(struct kvm *kvm)
606{
607 atomic_inc(&kvm->users_count);
608}
609EXPORT_SYMBOL_GPL(kvm_get_kvm);
610
611void kvm_put_kvm(struct kvm *kvm)
612{
613 if (atomic_dec_and_test(&kvm->users_count))
614 kvm_destroy_vm(kvm);
615}
616EXPORT_SYMBOL_GPL(kvm_put_kvm);
617
618
Avi Kivityf17abe92007-02-21 19:28:04 +0200619static int kvm_vm_release(struct inode *inode, struct file *filp)
620{
621 struct kvm *kvm = filp->private_data;
622
Gregory Haskins721eecbf2009-05-20 10:30:49 -0400623 kvm_irqfd_release(kvm);
624
Izik Eidusd39f13b2008-03-30 16:01:25 +0300625 kvm_put_kvm(kvm);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800626 return 0;
627}
628
Takuya Yoshikawa515a0122010-10-27 18:23:54 +0900629/*
630 * Allocation size is twice as large as the actual dirty bitmap size.
Takuya Yoshikawa93474b22012-03-01 19:34:45 +0900631 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
Takuya Yoshikawa515a0122010-10-27 18:23:54 +0900632 */
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900633static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
634{
Takuya Yoshikawa189a2f72012-02-08 13:01:09 +0900635#ifndef CONFIG_S390
Takuya Yoshikawa515a0122010-10-27 18:23:54 +0900636 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900637
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900638 memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900639 if (!memslot->dirty_bitmap)
640 return -ENOMEM;
641
Takuya Yoshikawa189a2f72012-02-08 13:01:09 +0900642#endif /* !CONFIG_S390 */
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900643 return 0;
644}
645
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800646static int cmp_memslot(const void *slot1, const void *slot2)
647{
648 struct kvm_memory_slot *s1, *s2;
649
650 s1 = (struct kvm_memory_slot *)slot1;
651 s2 = (struct kvm_memory_slot *)slot2;
652
653 if (s1->npages < s2->npages)
654 return 1;
655 if (s1->npages > s2->npages)
656 return -1;
657
658 return 0;
659}
660
661/*
662 * Sort the memslots base on its size, so the larger slots
663 * will get better fit.
664 */
665static void sort_memslots(struct kvm_memslots *slots)
666{
Xiao Guangrongf85e2cb2011-11-24 17:41:54 +0800667 int i;
668
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800669 sort(slots->memslots, KVM_MEM_SLOTS_NUM,
670 sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
Xiao Guangrongf85e2cb2011-11-24 17:41:54 +0800671
672 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
673 slots->id_to_index[slots->memslots[i].id] = i;
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800674}
675
Alex Williamson116c14c2012-12-21 08:20:16 -0700676void update_memslots(struct kvm_memslots *slots, struct kvm_memory_slot *new,
677 u64 last_generation)
Xiao Guangrongbe593d62011-11-24 17:38:24 +0800678{
679 if (new) {
680 int id = new->id;
Xiao Guangrong28a37542011-11-24 19:04:35 +0800681 struct kvm_memory_slot *old = id_to_memslot(slots, id);
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800682 unsigned long npages = old->npages;
Xiao Guangrongbe593d62011-11-24 17:38:24 +0800683
Xiao Guangrong28a37542011-11-24 19:04:35 +0800684 *old = *new;
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800685 if (new->npages != npages)
686 sort_memslots(slots);
Xiao Guangrongbe593d62011-11-24 17:38:24 +0800687 }
688
Alex Williamson116c14c2012-12-21 08:20:16 -0700689 slots->generation = last_generation + 1;
Xiao Guangrongbe593d62011-11-24 17:38:24 +0800690}
691
Xiao Guangronga50d64d2012-08-21 10:58:13 +0800692static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
693{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +0800694 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
695
696#ifdef KVM_CAP_READONLY_MEM
697 valid_flags |= KVM_MEM_READONLY;
698#endif
699
700 if (mem->flags & ~valid_flags)
Xiao Guangronga50d64d2012-08-21 10:58:13 +0800701 return -EINVAL;
702
703 return 0;
704}
705
Gleb Natapov7ec4fb42012-12-24 17:49:30 +0200706static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
707 struct kvm_memslots *slots, struct kvm_memory_slot *new)
708{
709 struct kvm_memslots *old_memslots = kvm->memslots;
710
711 update_memslots(slots, new, kvm->memslots->generation);
712 rcu_assign_pointer(kvm->memslots, slots);
713 synchronize_srcu_expedited(&kvm->srcu);
Takuya Yoshikawae59dbe02013-07-04 13:40:29 +0900714
715 kvm_arch_memslots_updated(kvm);
716
717 return old_memslots;
Gleb Natapov7ec4fb42012-12-24 17:49:30 +0200718}
719
Avi Kivity6aa8b732006-12-10 02:21:36 -0800720/*
Avi Kivity6aa8b732006-12-10 02:21:36 -0800721 * Allocate some memory and give it an address in the guest physical address
722 * space.
723 *
724 * Discontiguous memory is allowed, mostly for framebuffers.
Sheng Yangf78e0e22007-10-29 09:40:42 +0800725 *
Marcelo Tosatti10589a42007-12-20 19:18:22 -0500726 * Must be called holding mmap_sem for write.
Avi Kivity6aa8b732006-12-10 02:21:36 -0800727 */
Sheng Yangf78e0e22007-10-29 09:40:42 +0800728int __kvm_set_memory_region(struct kvm *kvm,
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +0900729 struct kvm_userspace_memory_region *mem)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800730{
Avi Kivity8234b222010-12-27 12:08:45 +0200731 int r;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800732 gfn_t base_gfn;
Heiko Carstens28bcb112009-09-03 17:35:35 +0200733 unsigned long npages;
Takuya Yoshikawaa843fac2013-01-11 18:27:43 +0900734 struct kvm_memory_slot *slot;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800735 struct kvm_memory_slot old, new;
Alex Williamsonb7f69c52012-12-10 10:33:03 -0700736 struct kvm_memslots *slots = NULL, *old_memslots;
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900737 enum kvm_mr_change change;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800738
Xiao Guangronga50d64d2012-08-21 10:58:13 +0800739 r = check_memory_region_flags(mem);
740 if (r)
741 goto out;
742
Avi Kivity6aa8b732006-12-10 02:21:36 -0800743 r = -EINVAL;
744 /* General sanity checks */
745 if (mem->memory_size & (PAGE_SIZE - 1))
746 goto out;
747 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
748 goto out;
Takuya Yoshikawafa3d3152011-05-07 16:35:38 +0900749 /* We can read the guest memory with __xxx_user() later on. */
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +0900750 if ((mem->slot < KVM_USER_MEM_SLOTS) &&
Takuya Yoshikawafa3d3152011-05-07 16:35:38 +0900751 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
Heiko Carstens9e3bb6b2011-05-24 07:51:27 +0200752 !access_ok(VERIFY_WRITE,
753 (void __user *)(unsigned long)mem->userspace_addr,
754 mem->memory_size)))
Hollis Blanchard78749802008-11-07 13:32:12 -0600755 goto out;
Xiao Guangrong93a5cef2011-11-24 17:37:48 +0800756 if (mem->slot >= KVM_MEM_SLOTS_NUM)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800757 goto out;
758 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
759 goto out;
760
Takuya Yoshikawaa843fac2013-01-11 18:27:43 +0900761 slot = id_to_memslot(kvm->memslots, mem->slot);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800762 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
763 npages = mem->memory_size >> PAGE_SHIFT;
764
Takuya Yoshikawa660c22c2010-04-13 22:47:24 +0900765 r = -EINVAL;
766 if (npages > KVM_MEM_MAX_NR_PAGES)
767 goto out;
768
Avi Kivity6aa8b732006-12-10 02:21:36 -0800769 if (!npages)
770 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
771
Takuya Yoshikawaa843fac2013-01-11 18:27:43 +0900772 new = old = *slot;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800773
Avi Kivitye36d96f2010-06-21 10:56:36 +0300774 new.id = mem->slot;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800775 new.base_gfn = base_gfn;
776 new.npages = npages;
777 new.flags = mem->flags;
778
Avi Kivity6aa8b732006-12-10 02:21:36 -0800779 r = -EINVAL;
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900780 if (npages) {
781 if (!old.npages)
782 change = KVM_MR_CREATE;
783 else { /* Modify an existing slot. */
784 if ((mem->userspace_addr != old.userspace_addr) ||
Takuya Yoshikawa75d61fb2013-01-30 19:40:41 +0900785 (npages != old.npages) ||
786 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900787 goto out;
788
789 if (base_gfn != old.base_gfn)
790 change = KVM_MR_MOVE;
791 else if (new.flags != old.flags)
792 change = KVM_MR_FLAGS_ONLY;
793 else { /* Nothing to change. */
794 r = 0;
795 goto out;
796 }
797 }
798 } else if (old.npages) {
799 change = KVM_MR_DELETE;
800 } else /* Modify a non-existent slot: disallowed. */
Takuya Yoshikawa0ea75e12013-01-11 18:26:10 +0900801 goto out;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800802
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900803 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
Takuya Yoshikawa0a706be2013-01-11 18:26:55 +0900804 /* Check for overlaps */
805 r = -EEXIST;
806 kvm_for_each_memslot(slot, kvm->memslots) {
Takuya Yoshikawaa843fac2013-01-11 18:27:43 +0900807 if ((slot->id >= KVM_USER_MEM_SLOTS) ||
808 (slot->id == mem->slot))
Takuya Yoshikawa0a706be2013-01-11 18:26:55 +0900809 continue;
810 if (!((base_gfn + npages <= slot->base_gfn) ||
811 (base_gfn >= slot->base_gfn + slot->npages)))
812 goto out;
813 }
Avi Kivity6aa8b732006-12-10 02:21:36 -0800814 }
Avi Kivity6aa8b732006-12-10 02:21:36 -0800815
Avi Kivity6aa8b732006-12-10 02:21:36 -0800816 /* Free page dirty bitmap if unneeded */
817 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
Al Viro8b6d44c2007-02-09 16:38:40 +0000818 new.dirty_bitmap = NULL;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800819
820 r = -ENOMEM;
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900821 if (change == KVM_MR_CREATE) {
Takuya Yoshikawa189a2f72012-02-08 13:01:09 +0900822 new.userspace_addr = mem->userspace_addr;
Takuya Yoshikawad89cc612012-08-01 18:03:28 +0900823
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +0900824 if (kvm_arch_create_memslot(&new, npages))
825 goto out_free;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800826 }
Joerg Roedelec04b262009-06-19 15:16:23 +0200827
Avi Kivity6aa8b732006-12-10 02:21:36 -0800828 /* Allocate page dirty bitmap if needed */
829 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900830 if (kvm_create_dirty_bitmap(&new) < 0)
Sheng Yangf78e0e22007-10-29 09:40:42 +0800831 goto out_free;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800832 }
833
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900834 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200835 r = -ENOMEM;
Thomas Meyer6da64fd2011-11-08 20:32:19 +0100836 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
837 GFP_KERNEL);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200838 if (!slots)
839 goto out_free;
Xiao Guangrong28a37542011-11-24 19:04:35 +0800840 slot = id_to_memslot(slots, mem->slot);
841 slot->flags |= KVM_MEMSLOT_INVALID;
842
Gleb Natapov7ec4fb42012-12-24 17:49:30 +0200843 old_memslots = install_new_memslots(kvm, slots, NULL);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200844
Alex Williamsone40f1932012-12-10 10:32:57 -0700845 /* slot was deleted or moved, clear iommu mapping */
846 kvm_iommu_unmap_pages(kvm, &old);
Marcelo Tosatti12d6e752012-08-24 15:54:58 -0300847 /* From this point no new shadow pages pointing to a deleted,
848 * or moved, memslot will be created.
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200849 *
850 * validation of sp->gfn happens in:
851 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
852 * - kvm_is_visible_gfn (mmu_check_roots)
853 */
Marcelo Tosatti2df72e92012-08-24 15:54:57 -0300854 kvm_arch_flush_shadow_memslot(kvm, slot);
Alex Williamsonb7f69c52012-12-10 10:33:03 -0700855 slots = old_memslots;
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200856 }
Marcelo Tosatti34d4cb82008-07-10 20:49:31 -0300857
Takuya Yoshikawa7b6195a2013-02-27 19:44:34 +0900858 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
Marcelo Tosattif7784b82009-12-23 14:35:18 -0200859 if (r)
Alex Williamsonb7f69c52012-12-10 10:33:03 -0700860 goto out_slots;
Marcelo Tosattif7784b82009-12-23 14:35:18 -0200861
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200862 r = -ENOMEM;
Alex Williamsonb7f69c52012-12-10 10:33:03 -0700863 /*
864 * We can re-use the old_memslots from above, the only difference
865 * from the currently installed memslots is the invalid flag. This
866 * will get overwritten by update_memslots anyway.
867 */
868 if (!slots) {
869 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
870 GFP_KERNEL);
871 if (!slots)
872 goto out_free;
873 }
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200874
Alex Williamson261874b2013-01-24 15:04:03 -0700875 /*
876 * IOMMU mapping: New slots need to be mapped. Old slots need to be
Takuya Yoshikawa75d61fb2013-01-30 19:40:41 +0900877 * un-mapped and re-mapped if their base changes. Since base change
878 * unmapping is handled above with slot deletion, mapping alone is
879 * needed here. Anything else the iommu might care about for existing
880 * slots (size changes, userspace addr changes and read-only flag
881 * changes) is disallowed above, so any other attribute changes getting
882 * here can be skipped.
Alex Williamson261874b2013-01-24 15:04:03 -0700883 */
Takuya Yoshikawa75d61fb2013-01-30 19:40:41 +0900884 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
885 r = kvm_iommu_map_pages(kvm, &new);
886 if (r)
887 goto out_slots;
Alex Williamsone40f1932012-12-10 10:32:57 -0700888 }
889
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200890 /* actual memory is freed via old in kvm_free_physmem_slot below */
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900891 if (change == KVM_MR_DELETE) {
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200892 new.dirty_bitmap = NULL;
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +0900893 memset(&new.arch, 0, sizeof(new.arch));
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200894 }
895
Gleb Natapov7ec4fb42012-12-24 17:49:30 +0200896 old_memslots = install_new_memslots(kvm, slots, &new);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800897
Takuya Yoshikawa84826442013-02-27 19:45:25 +0900898 kvm_arch_commit_memory_region(kvm, mem, &old, change);
Zhang Xiantao3ad82a72007-11-20 13:11:38 +0800899
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200900 kvm_free_physmem_slot(&old, &new);
901 kfree(old_memslots);
902
Avi Kivity6aa8b732006-12-10 02:21:36 -0800903 return 0;
904
Alex Williamsone40f1932012-12-10 10:32:57 -0700905out_slots:
906 kfree(slots);
Sheng Yangf78e0e22007-10-29 09:40:42 +0800907out_free:
Avi Kivity6aa8b732006-12-10 02:21:36 -0800908 kvm_free_physmem_slot(&new, &old);
909out:
910 return r;
Izik Eidus210c7c42007-10-24 23:52:57 +0200911}
Sheng Yangf78e0e22007-10-29 09:40:42 +0800912EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
913
914int kvm_set_memory_region(struct kvm *kvm,
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +0900915 struct kvm_userspace_memory_region *mem)
Sheng Yangf78e0e22007-10-29 09:40:42 +0800916{
917 int r;
918
Marcelo Tosatti79fac952009-12-23 14:35:26 -0200919 mutex_lock(&kvm->slots_lock);
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +0900920 r = __kvm_set_memory_region(kvm, mem);
Marcelo Tosatti79fac952009-12-23 14:35:26 -0200921 mutex_unlock(&kvm->slots_lock);
Sheng Yangf78e0e22007-10-29 09:40:42 +0800922 return r;
923}
Izik Eidus210c7c42007-10-24 23:52:57 +0200924EXPORT_SYMBOL_GPL(kvm_set_memory_region);
925
Carsten Otte1fe779f2007-10-29 16:08:35 +0100926int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +0900927 struct kvm_userspace_memory_region *mem)
Izik Eidus210c7c42007-10-24 23:52:57 +0200928{
Alex Williamsonbbacc0c2012-12-10 10:33:09 -0700929 if (mem->slot >= KVM_USER_MEM_SLOTS)
Izik Eiduse0d62c72007-10-24 23:57:46 +0200930 return -EINVAL;
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +0900931 return kvm_set_memory_region(kvm, mem);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800932}
933
Zhang Xiantao5bb064d2007-11-18 20:29:43 +0800934int kvm_get_dirty_log(struct kvm *kvm,
935 struct kvm_dirty_log *log, int *is_dirty)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800936{
937 struct kvm_memory_slot *memslot;
938 int r, i;
Takuya Yoshikawa87bf6e72010-04-12 19:35:35 +0900939 unsigned long n;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800940 unsigned long any = 0;
941
Avi Kivity6aa8b732006-12-10 02:21:36 -0800942 r = -EINVAL;
Alex Williamsonbbacc0c2012-12-10 10:33:09 -0700943 if (log->slot >= KVM_USER_MEM_SLOTS)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800944 goto out;
945
Xiao Guangrong28a37542011-11-24 19:04:35 +0800946 memslot = id_to_memslot(kvm->memslots, log->slot);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800947 r = -ENOENT;
948 if (!memslot->dirty_bitmap)
949 goto out;
950
Takuya Yoshikawa87bf6e72010-04-12 19:35:35 +0900951 n = kvm_dirty_bitmap_bytes(memslot);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800952
Uri Lublincd1a4a92007-02-22 16:43:09 +0200953 for (i = 0; !any && i < n/sizeof(long); ++i)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800954 any = memslot->dirty_bitmap[i];
955
956 r = -EFAULT;
957 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
958 goto out;
959
Zhang Xiantao5bb064d2007-11-18 20:29:43 +0800960 if (any)
961 *is_dirty = 1;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800962
963 r = 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800964out:
Avi Kivity6aa8b732006-12-10 02:21:36 -0800965 return r;
966}
967
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +0900968bool kvm_largepages_enabled(void)
969{
970 return largepages_enabled;
971}
972
Marcelo Tosatti54dee992009-06-11 12:07:44 -0300973void kvm_disable_largepages(void)
974{
975 largepages_enabled = false;
976}
977EXPORT_SYMBOL_GPL(kvm_disable_largepages);
978
Gleb Natapov49c77542010-10-18 15:22:23 +0200979struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
980{
981 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
982}
Avi Kivitya1f4d3952010-06-21 11:44:20 +0300983EXPORT_SYMBOL_GPL(gfn_to_memslot);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800984
Izik Eiduse0d62c72007-10-24 23:57:46 +0200985int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
986{
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800987 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
Izik Eiduse0d62c72007-10-24 23:57:46 +0200988
Alex Williamsonbbacc0c2012-12-10 10:33:09 -0700989 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800990 memslot->flags & KVM_MEMSLOT_INVALID)
991 return 0;
Izik Eiduse0d62c72007-10-24 23:57:46 +0200992
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800993 return 1;
Izik Eiduse0d62c72007-10-24 23:57:46 +0200994}
995EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
996
Joerg Roedel8f0b1ab2010-01-28 12:37:56 +0100997unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
998{
999 struct vm_area_struct *vma;
1000 unsigned long addr, size;
1001
1002 size = PAGE_SIZE;
1003
1004 addr = gfn_to_hva(kvm, gfn);
1005 if (kvm_is_error_hva(addr))
1006 return PAGE_SIZE;
1007
1008 down_read(&current->mm->mmap_sem);
1009 vma = find_vma(current->mm, addr);
1010 if (!vma)
1011 goto out;
1012
1013 size = vma_kernel_pagesize(vma);
1014
1015out:
1016 up_read(&current->mm->mmap_sem);
1017
1018 return size;
1019}
1020
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001021static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1022{
1023 return slot->flags & KVM_MEM_READONLY;
1024}
1025
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001026static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1027 gfn_t *nr_pages, bool write)
Izik Eidus539cb662007-11-11 22:05:04 +02001028{
Marcelo Tosattibc6678a2009-12-23 14:35:21 -02001029 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
Xiao Guangrongca3a4902012-08-21 11:01:50 +08001030 return KVM_HVA_ERR_BAD;
Xiao Guangrong48987782010-08-22 19:11:43 +08001031
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001032 if (memslot_is_readonly(slot) && write)
1033 return KVM_HVA_ERR_RO_BAD;
Xiao Guangrong48987782010-08-22 19:11:43 +08001034
1035 if (nr_pages)
1036 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1037
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001038 return __gfn_to_hva_memslot(slot, gfn);
Izik Eidus539cb662007-11-11 22:05:04 +02001039}
Xiao Guangrong48987782010-08-22 19:11:43 +08001040
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001041static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1042 gfn_t *nr_pages)
1043{
1044 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1045}
1046
1047unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1048 gfn_t gfn)
1049{
1050 return gfn_to_hva_many(slot, gfn, NULL);
1051}
1052EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1053
Xiao Guangrong48987782010-08-22 19:11:43 +08001054unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1055{
Gleb Natapov49c77542010-10-18 15:22:23 +02001056 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
Xiao Guangrong48987782010-08-22 19:11:43 +08001057}
Sheng Yang0d150292008-04-25 21:44:50 +08001058EXPORT_SYMBOL_GPL(gfn_to_hva);
Izik Eidus539cb662007-11-11 22:05:04 +02001059
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001060/*
1061 * The hva returned by this function is only allowed to be read.
1062 * It should pair with kvm_read_hva() or kvm_read_hva_atomic().
1063 */
1064static unsigned long gfn_to_hva_read(struct kvm *kvm, gfn_t gfn)
Gleb Natapov80300892010-10-19 18:13:41 +02001065{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001066 return __gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL, false);
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001067}
1068
1069static int kvm_read_hva(void *data, void __user *hva, int len)
1070{
1071 return __copy_from_user(data, hva, len);
1072}
1073
1074static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1075{
1076 return __copy_from_user_inatomic(data, hva, len);
Gleb Natapov80300892010-10-19 18:13:41 +02001077}
1078
Geoff Levand39369f72013-04-05 19:20:30 +00001079static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
Gleb Natapov0857b9e2011-02-01 13:21:47 +02001080 unsigned long start, int write, struct page **page)
1081{
1082 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1083
1084 if (write)
1085 flags |= FOLL_WRITE;
1086
1087 return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1088}
1089
Huang Yingfafc3db2011-01-30 11:15:49 +08001090static inline int check_user_page_hwpoison(unsigned long addr)
1091{
1092 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1093
1094 rc = __get_user_pages(current, current->mm, addr, 1,
1095 flags, NULL, NULL, NULL);
1096 return rc == -EHWPOISON;
1097}
1098
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001099/*
1100 * The atomic path to get the writable pfn which will be stored in @pfn,
1101 * true indicates success, otherwise false is returned.
1102 */
1103static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1104 bool write_fault, bool *writable, pfn_t *pfn)
1105{
1106 struct page *page[1];
1107 int npages;
1108
1109 if (!(async || atomic))
1110 return false;
1111
Xiao Guangrong12ce13f2012-08-21 11:00:49 +08001112 /*
1113 * Fast pin a writable pfn only if it is a write fault request
1114 * or the caller allows to map a writable pfn for a read fault
1115 * request.
1116 */
1117 if (!(write_fault || writable))
1118 return false;
1119
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001120 npages = __get_user_pages_fast(addr, 1, 1, page);
1121 if (npages == 1) {
1122 *pfn = page_to_pfn(page[0]);
1123
1124 if (writable)
1125 *writable = true;
1126 return true;
1127 }
1128
1129 return false;
1130}
1131
1132/*
1133 * The slow path to get the pfn of the specified host virtual address,
1134 * 1 indicates success, -errno is returned if error is detected.
1135 */
1136static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1137 bool *writable, pfn_t *pfn)
Avi Kivity954bbbc2007-03-30 14:02:32 +03001138{
Anthony Liguori8d4e1282007-10-18 09:59:34 -05001139 struct page *page[1];
Gleb Natapovaf585b92010-10-14 11:22:46 +02001140 int npages = 0;
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001141
1142 might_sleep();
1143
1144 if (writable)
1145 *writable = write_fault;
1146
1147 if (async) {
1148 down_read(&current->mm->mmap_sem);
1149 npages = get_user_page_nowait(current, current->mm,
1150 addr, write_fault, page);
1151 up_read(&current->mm->mmap_sem);
1152 } else
1153 npages = get_user_pages_fast(addr, 1, write_fault,
1154 page);
1155 if (npages != 1)
1156 return npages;
1157
1158 /* map read fault as writable if possible */
Xiao Guangrong12ce13f2012-08-21 11:00:49 +08001159 if (unlikely(!write_fault) && writable) {
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001160 struct page *wpage[1];
1161
1162 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1163 if (npages == 1) {
1164 *writable = true;
1165 put_page(page[0]);
1166 page[0] = wpage[0];
1167 }
1168
1169 npages = 1;
1170 }
1171 *pfn = page_to_pfn(page[0]);
1172 return npages;
1173}
1174
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001175static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1176{
1177 if (unlikely(!(vma->vm_flags & VM_READ)))
1178 return false;
1179
1180 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1181 return false;
1182
1183 return true;
1184}
1185
Xiao Guangrong12ce13f2012-08-21 11:00:49 +08001186/*
1187 * Pin guest page in memory and return its pfn.
1188 * @addr: host virtual address which maps memory to the guest
1189 * @atomic: whether this function can sleep
1190 * @async: whether this function need to wait IO complete if the
1191 * host page is not in the memory
1192 * @write_fault: whether we should get a writable host page
1193 * @writable: whether it allows to map a writable host page for !@write_fault
1194 *
1195 * The function will map a writable host page for these two cases:
1196 * 1): @write_fault = true
1197 * 2): @write_fault = false && @writable, @writable will tell the caller
1198 * whether the mapping is writable.
1199 */
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001200static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1201 bool write_fault, bool *writable)
1202{
1203 struct vm_area_struct *vma;
1204 pfn_t pfn = 0;
1205 int npages;
Avi Kivity954bbbc2007-03-30 14:02:32 +03001206
Gleb Natapovaf585b92010-10-14 11:22:46 +02001207 /* we can do it either atomically or asynchronously, not both */
1208 BUG_ON(atomic && async);
1209
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001210 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1211 return pfn;
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001212
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001213 if (atomic)
1214 return KVM_PFN_ERR_FAULT;
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001215
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001216 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1217 if (npages == 1)
1218 return pfn;
Gleb Natapovaf585b92010-10-14 11:22:46 +02001219
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001220 down_read(&current->mm->mmap_sem);
1221 if (npages == -EHWPOISON ||
1222 (!async && check_user_page_hwpoison(addr))) {
1223 pfn = KVM_PFN_ERR_HWPOISON;
1224 goto exit;
Xiao Guangrong887c08a2010-08-22 19:10:28 +08001225 }
Izik Eidus539cb662007-11-11 22:05:04 +02001226
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001227 vma = find_vma_intersection(current->mm, addr, addr + 1);
Anthony Liguori8d4e1282007-10-18 09:59:34 -05001228
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001229 if (vma == NULL)
1230 pfn = KVM_PFN_ERR_FAULT;
1231 else if ((vma->vm_flags & VM_PFNMAP)) {
1232 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1233 vma->vm_pgoff;
1234 BUG_ON(!kvm_is_mmio_pfn(pfn));
1235 } else {
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001236 if (async && vma_is_valid(vma, write_fault))
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001237 *async = true;
1238 pfn = KVM_PFN_ERR_FAULT;
1239 }
1240exit:
1241 up_read(&current->mm->mmap_sem);
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001242 return pfn;
Anthony Liguori35149e22008-04-02 14:46:56 -05001243}
1244
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001245static pfn_t
1246__gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1247 bool *async, bool write_fault, bool *writable)
Xiao Guangrong887c08a2010-08-22 19:10:28 +08001248{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001249 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1250
1251 if (addr == KVM_HVA_ERR_RO_BAD)
1252 return KVM_PFN_ERR_RO_FAULT;
1253
1254 if (kvm_is_error_hva(addr))
Xiao Guangrong81c52c52012-10-16 20:10:59 +08001255 return KVM_PFN_NOSLOT;
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001256
1257 /* Do not map writable pfn in the readonly memslot. */
1258 if (writable && memslot_is_readonly(slot)) {
1259 *writable = false;
1260 writable = NULL;
1261 }
1262
1263 return hva_to_pfn(addr, atomic, async, write_fault,
1264 writable);
Xiao Guangrong887c08a2010-08-22 19:10:28 +08001265}
Xiao Guangrong887c08a2010-08-22 19:10:28 +08001266
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001267static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1268 bool write_fault, bool *writable)
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001269{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001270 struct kvm_memory_slot *slot;
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001271
Gleb Natapovaf585b92010-10-14 11:22:46 +02001272 if (async)
1273 *async = false;
1274
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001275 slot = gfn_to_memslot(kvm, gfn);
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001276
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001277 return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1278 writable);
Xiao Guangrong365fb3f2010-08-28 19:24:13 +08001279}
1280
1281pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1282{
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001283 return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
Xiao Guangrong365fb3f2010-08-28 19:24:13 +08001284}
1285EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1286
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001287pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1288 bool write_fault, bool *writable)
Gleb Natapovaf585b92010-10-14 11:22:46 +02001289{
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001290 return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
Gleb Natapovaf585b92010-10-14 11:22:46 +02001291}
1292EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1293
Xiao Guangrong365fb3f2010-08-28 19:24:13 +08001294pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1295{
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001296 return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001297}
Anthony Liguori35149e22008-04-02 14:46:56 -05001298EXPORT_SYMBOL_GPL(gfn_to_pfn);
1299
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001300pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1301 bool *writable)
1302{
1303 return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1304}
1305EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1306
Xiao Guangrongd5661042012-07-17 21:56:16 +08001307pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001308{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001309 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001310}
1311
Xiao Guangrong037d92d2012-08-21 10:59:12 +08001312pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1313{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001314 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
Xiao Guangrong037d92d2012-08-21 10:59:12 +08001315}
1316EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1317
Xiao Guangrong48987782010-08-22 19:11:43 +08001318int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1319 int nr_pages)
1320{
1321 unsigned long addr;
1322 gfn_t entry;
1323
Gleb Natapov49c77542010-10-18 15:22:23 +02001324 addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
Xiao Guangrong48987782010-08-22 19:11:43 +08001325 if (kvm_is_error_hva(addr))
1326 return -1;
1327
1328 if (entry < nr_pages)
1329 return 0;
1330
1331 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1332}
1333EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1334
Xiao Guangronga2766322012-07-26 11:58:59 +08001335static struct page *kvm_pfn_to_page(pfn_t pfn)
1336{
Xiao Guangrong81c52c52012-10-16 20:10:59 +08001337 if (is_error_noslot_pfn(pfn))
Xiao Guangrong6cede2e2012-08-03 15:41:22 +08001338 return KVM_ERR_PTR_BAD_PAGE;
Xiao Guangronga2766322012-07-26 11:58:59 +08001339
Xiao Guangrongcb9aaa32012-08-03 15:42:10 +08001340 if (kvm_is_mmio_pfn(pfn)) {
1341 WARN_ON(1);
1342 return KVM_ERR_PTR_BAD_PAGE;
1343 }
1344
Xiao Guangronga2766322012-07-26 11:58:59 +08001345 return pfn_to_page(pfn);
1346}
1347
Anthony Liguori35149e22008-04-02 14:46:56 -05001348struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1349{
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001350 pfn_t pfn;
1351
1352 pfn = gfn_to_pfn(kvm, gfn);
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001353
Xiao Guangronga2766322012-07-26 11:58:59 +08001354 return kvm_pfn_to_page(pfn);
Avi Kivity954bbbc2007-03-30 14:02:32 +03001355}
Anthony Liguoriaab61cc2007-10-29 15:15:20 -05001356
Avi Kivity954bbbc2007-03-30 14:02:32 +03001357EXPORT_SYMBOL_GPL(gfn_to_page);
1358
Izik Eidusb4231d62007-11-20 11:49:33 +02001359void kvm_release_page_clean(struct page *page)
1360{
Xiao Guangrong32cad842012-08-03 15:42:52 +08001361 WARN_ON(is_error_page(page));
1362
Anthony Liguori35149e22008-04-02 14:46:56 -05001363 kvm_release_pfn_clean(page_to_pfn(page));
Izik Eidusb4231d62007-11-20 11:49:33 +02001364}
1365EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1366
Anthony Liguori35149e22008-04-02 14:46:56 -05001367void kvm_release_pfn_clean(pfn_t pfn)
1368{
Xiao Guangrong81c52c52012-10-16 20:10:59 +08001369 if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001370 put_page(pfn_to_page(pfn));
Anthony Liguori35149e22008-04-02 14:46:56 -05001371}
1372EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1373
Izik Eidusb4231d62007-11-20 11:49:33 +02001374void kvm_release_page_dirty(struct page *page)
Izik Eidus8a7ae052007-10-18 11:09:33 +02001375{
Xiao Guangronga2766322012-07-26 11:58:59 +08001376 WARN_ON(is_error_page(page));
1377
Anthony Liguori35149e22008-04-02 14:46:56 -05001378 kvm_release_pfn_dirty(page_to_pfn(page));
Izik Eidus8a7ae052007-10-18 11:09:33 +02001379}
Izik Eidusb4231d62007-11-20 11:49:33 +02001380EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
Izik Eidus8a7ae052007-10-18 11:09:33 +02001381
Anthony Liguori35149e22008-04-02 14:46:56 -05001382void kvm_release_pfn_dirty(pfn_t pfn)
1383{
1384 kvm_set_pfn_dirty(pfn);
1385 kvm_release_pfn_clean(pfn);
1386}
1387EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1388
1389void kvm_set_page_dirty(struct page *page)
1390{
1391 kvm_set_pfn_dirty(page_to_pfn(page));
1392}
1393EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1394
1395void kvm_set_pfn_dirty(pfn_t pfn)
1396{
Xiantao Zhangc77fb9d2008-09-27 10:55:40 +08001397 if (!kvm_is_mmio_pfn(pfn)) {
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001398 struct page *page = pfn_to_page(pfn);
1399 if (!PageReserved(page))
1400 SetPageDirty(page);
1401 }
Anthony Liguori35149e22008-04-02 14:46:56 -05001402}
1403EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1404
1405void kvm_set_pfn_accessed(pfn_t pfn)
1406{
Xiantao Zhangc77fb9d2008-09-27 10:55:40 +08001407 if (!kvm_is_mmio_pfn(pfn))
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001408 mark_page_accessed(pfn_to_page(pfn));
Anthony Liguori35149e22008-04-02 14:46:56 -05001409}
1410EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1411
1412void kvm_get_pfn(pfn_t pfn)
1413{
Xiantao Zhangc77fb9d2008-09-27 10:55:40 +08001414 if (!kvm_is_mmio_pfn(pfn))
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001415 get_page(pfn_to_page(pfn));
Anthony Liguori35149e22008-04-02 14:46:56 -05001416}
1417EXPORT_SYMBOL_GPL(kvm_get_pfn);
1418
Izik Eidus195aefd2007-10-01 22:14:18 +02001419static int next_segment(unsigned long len, int offset)
1420{
1421 if (len > PAGE_SIZE - offset)
1422 return PAGE_SIZE - offset;
1423 else
1424 return len;
1425}
1426
1427int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1428 int len)
1429{
Izik Eiduse0506bc2007-11-11 22:10:22 +02001430 int r;
1431 unsigned long addr;
Izik Eidus195aefd2007-10-01 22:14:18 +02001432
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001433 addr = gfn_to_hva_read(kvm, gfn);
Izik Eiduse0506bc2007-11-11 22:10:22 +02001434 if (kvm_is_error_hva(addr))
Izik Eidus195aefd2007-10-01 22:14:18 +02001435 return -EFAULT;
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001436 r = kvm_read_hva(data, (void __user *)addr + offset, len);
Izik Eiduse0506bc2007-11-11 22:10:22 +02001437 if (r)
1438 return -EFAULT;
Izik Eidus195aefd2007-10-01 22:14:18 +02001439 return 0;
1440}
1441EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1442
1443int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1444{
1445 gfn_t gfn = gpa >> PAGE_SHIFT;
1446 int seg;
1447 int offset = offset_in_page(gpa);
1448 int ret;
1449
1450 while ((seg = next_segment(len, offset)) != 0) {
1451 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1452 if (ret < 0)
1453 return ret;
1454 offset = 0;
1455 len -= seg;
1456 data += seg;
1457 ++gfn;
1458 }
1459 return 0;
1460}
1461EXPORT_SYMBOL_GPL(kvm_read_guest);
1462
Marcelo Tosatti7ec54582007-12-20 19:18:23 -05001463int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1464 unsigned long len)
1465{
1466 int r;
1467 unsigned long addr;
1468 gfn_t gfn = gpa >> PAGE_SHIFT;
1469 int offset = offset_in_page(gpa);
1470
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001471 addr = gfn_to_hva_read(kvm, gfn);
Marcelo Tosatti7ec54582007-12-20 19:18:23 -05001472 if (kvm_is_error_hva(addr))
1473 return -EFAULT;
Andrea Arcangeli0aac03f2008-01-30 19:57:35 +01001474 pagefault_disable();
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001475 r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
Andrea Arcangeli0aac03f2008-01-30 19:57:35 +01001476 pagefault_enable();
Marcelo Tosatti7ec54582007-12-20 19:18:23 -05001477 if (r)
1478 return -EFAULT;
1479 return 0;
1480}
1481EXPORT_SYMBOL(kvm_read_guest_atomic);
1482
Izik Eidus195aefd2007-10-01 22:14:18 +02001483int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1484 int offset, int len)
1485{
Izik Eiduse0506bc2007-11-11 22:10:22 +02001486 int r;
1487 unsigned long addr;
Izik Eidus195aefd2007-10-01 22:14:18 +02001488
Izik Eiduse0506bc2007-11-11 22:10:22 +02001489 addr = gfn_to_hva(kvm, gfn);
1490 if (kvm_is_error_hva(addr))
Izik Eidus195aefd2007-10-01 22:14:18 +02001491 return -EFAULT;
Xiao Guangrong8b0cedf2011-05-15 23:22:04 +08001492 r = __copy_to_user((void __user *)addr + offset, data, len);
Izik Eiduse0506bc2007-11-11 22:10:22 +02001493 if (r)
1494 return -EFAULT;
Izik Eidus195aefd2007-10-01 22:14:18 +02001495 mark_page_dirty(kvm, gfn);
1496 return 0;
1497}
1498EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1499
1500int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1501 unsigned long len)
1502{
1503 gfn_t gfn = gpa >> PAGE_SHIFT;
1504 int seg;
1505 int offset = offset_in_page(gpa);
1506 int ret;
1507
1508 while ((seg = next_segment(len, offset)) != 0) {
1509 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1510 if (ret < 0)
1511 return ret;
1512 offset = 0;
1513 len -= seg;
1514 data += seg;
1515 ++gfn;
1516 }
1517 return 0;
1518}
1519
Gleb Natapov49c77542010-10-18 15:22:23 +02001520int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
Andrew Honig8f964522013-03-29 09:35:21 -07001521 gpa_t gpa, unsigned long len)
Gleb Natapov49c77542010-10-18 15:22:23 +02001522{
1523 struct kvm_memslots *slots = kvm_memslots(kvm);
1524 int offset = offset_in_page(gpa);
Andrew Honig8f964522013-03-29 09:35:21 -07001525 gfn_t start_gfn = gpa >> PAGE_SHIFT;
1526 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1527 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1528 gfn_t nr_pages_avail;
Gleb Natapov49c77542010-10-18 15:22:23 +02001529
1530 ghc->gpa = gpa;
1531 ghc->generation = slots->generation;
Andrew Honig8f964522013-03-29 09:35:21 -07001532 ghc->len = len;
1533 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1534 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, &nr_pages_avail);
1535 if (!kvm_is_error_hva(ghc->hva) && nr_pages_avail >= nr_pages_needed) {
Gleb Natapov49c77542010-10-18 15:22:23 +02001536 ghc->hva += offset;
Andrew Honig8f964522013-03-29 09:35:21 -07001537 } else {
1538 /*
1539 * If the requested region crosses two memslots, we still
1540 * verify that the entire region is valid here.
1541 */
1542 while (start_gfn <= end_gfn) {
1543 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1544 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1545 &nr_pages_avail);
1546 if (kvm_is_error_hva(ghc->hva))
1547 return -EFAULT;
1548 start_gfn += nr_pages_avail;
1549 }
1550 /* Use the slow path for cross page reads and writes. */
1551 ghc->memslot = NULL;
1552 }
Gleb Natapov49c77542010-10-18 15:22:23 +02001553 return 0;
1554}
1555EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1556
1557int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1558 void *data, unsigned long len)
1559{
1560 struct kvm_memslots *slots = kvm_memslots(kvm);
1561 int r;
1562
Andrew Honig8f964522013-03-29 09:35:21 -07001563 BUG_ON(len > ghc->len);
1564
Gleb Natapov49c77542010-10-18 15:22:23 +02001565 if (slots->generation != ghc->generation)
Andrew Honig8f964522013-03-29 09:35:21 -07001566 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1567
1568 if (unlikely(!ghc->memslot))
1569 return kvm_write_guest(kvm, ghc->gpa, data, len);
Gleb Natapov49c77542010-10-18 15:22:23 +02001570
1571 if (kvm_is_error_hva(ghc->hva))
1572 return -EFAULT;
1573
Xiao Guangrong8b0cedf2011-05-15 23:22:04 +08001574 r = __copy_to_user((void __user *)ghc->hva, data, len);
Gleb Natapov49c77542010-10-18 15:22:23 +02001575 if (r)
1576 return -EFAULT;
1577 mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1578
1579 return 0;
1580}
1581EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1582
Gleb Natapove03b6442011-07-11 15:28:11 -04001583int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1584 void *data, unsigned long len)
1585{
1586 struct kvm_memslots *slots = kvm_memslots(kvm);
1587 int r;
1588
Andrew Honig8f964522013-03-29 09:35:21 -07001589 BUG_ON(len > ghc->len);
1590
Gleb Natapove03b6442011-07-11 15:28:11 -04001591 if (slots->generation != ghc->generation)
Andrew Honig8f964522013-03-29 09:35:21 -07001592 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1593
1594 if (unlikely(!ghc->memslot))
1595 return kvm_read_guest(kvm, ghc->gpa, data, len);
Gleb Natapove03b6442011-07-11 15:28:11 -04001596
1597 if (kvm_is_error_hva(ghc->hva))
1598 return -EFAULT;
1599
1600 r = __copy_from_user(data, (void __user *)ghc->hva, len);
1601 if (r)
1602 return -EFAULT;
1603
1604 return 0;
1605}
1606EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1607
Izik Eidus195aefd2007-10-01 22:14:18 +02001608int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1609{
Heiko Carstens3bcc8a82010-10-27 17:21:21 +02001610 return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1611 offset, len);
Izik Eidus195aefd2007-10-01 22:14:18 +02001612}
1613EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1614
1615int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1616{
1617 gfn_t gfn = gpa >> PAGE_SHIFT;
1618 int seg;
1619 int offset = offset_in_page(gpa);
1620 int ret;
1621
1622 while ((seg = next_segment(len, offset)) != 0) {
1623 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1624 if (ret < 0)
1625 return ret;
1626 offset = 0;
1627 len -= seg;
1628 ++gfn;
1629 }
1630 return 0;
1631}
1632EXPORT_SYMBOL_GPL(kvm_clear_guest);
1633
Gleb Natapov49c77542010-10-18 15:22:23 +02001634void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1635 gfn_t gfn)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001636{
Rusty Russell7e9d6192007-07-31 20:41:14 +10001637 if (memslot && memslot->dirty_bitmap) {
1638 unsigned long rel_gfn = gfn - memslot->base_gfn;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001639
Takuya Yoshikawab74ca3b2012-10-04 17:13:12 -07001640 set_bit_le(rel_gfn, memslot->dirty_bitmap);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001641 }
1642}
1643
Gleb Natapov49c77542010-10-18 15:22:23 +02001644void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1645{
1646 struct kvm_memory_slot *memslot;
1647
1648 memslot = gfn_to_memslot(kvm, gfn);
1649 mark_page_dirty_in_slot(kvm, memslot, gfn);
1650}
1651
Eddie Dongb6958ce2007-07-18 12:15:21 +03001652/*
1653 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1654 */
Hollis Blanchard8776e512007-10-31 17:24:24 -05001655void kvm_vcpu_block(struct kvm_vcpu *vcpu)
Eddie Dongb6958ce2007-07-18 12:15:21 +03001656{
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001657 DEFINE_WAIT(wait);
Eddie Dongb6958ce2007-07-18 12:15:21 +03001658
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001659 for (;;) {
1660 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
Eddie Dongb6958ce2007-07-18 12:15:21 +03001661
Gleb Natapova1b37102009-07-09 15:33:52 +03001662 if (kvm_arch_vcpu_runnable(vcpu)) {
Avi Kivitya8eeb042010-05-10 12:34:53 +03001663 kvm_make_request(KVM_REQ_UNHALT, vcpu);
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001664 break;
Marcelo Tosattid7690172008-09-08 15:23:48 -03001665 }
Gleb Natapov09cec752009-03-23 15:11:44 +02001666 if (kvm_cpu_has_pending_timer(vcpu))
1667 break;
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001668 if (signal_pending(current))
1669 break;
1670
Eddie Dongb6958ce2007-07-18 12:15:21 +03001671 schedule();
Eddie Dongb6958ce2007-07-18 12:15:21 +03001672 }
1673
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001674 finish_wait(&vcpu->wq, &wait);
Eddie Dongb6958ce2007-07-18 12:15:21 +03001675}
1676
Marcelo Tosatti8c847802012-03-14 17:58:48 -03001677#ifndef CONFIG_S390
Christoffer Dallb6d33832012-03-08 16:44:24 -05001678/*
1679 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1680 */
1681void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1682{
1683 int me;
1684 int cpu = vcpu->cpu;
1685 wait_queue_head_t *wqp;
1686
1687 wqp = kvm_arch_vcpu_wq(vcpu);
1688 if (waitqueue_active(wqp)) {
1689 wake_up_interruptible(wqp);
1690 ++vcpu->stat.halt_wakeup;
1691 }
1692
1693 me = get_cpu();
1694 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1695 if (kvm_arch_vcpu_should_kick(vcpu))
1696 smp_send_reschedule(cpu);
1697 put_cpu();
1698}
Yang Zhanga20ed542013-04-11 19:25:15 +08001699EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
Marcelo Tosatti8c847802012-03-14 17:58:48 -03001700#endif /* !CONFIG_S390 */
Christoffer Dallb6d33832012-03-08 16:44:24 -05001701
Avi Kivity6aa8b732006-12-10 02:21:36 -08001702void kvm_resched(struct kvm_vcpu *vcpu)
1703{
Yaozu Dong3fca0362007-04-25 16:49:19 +03001704 if (!need_resched())
1705 return;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001706 cond_resched();
Avi Kivity6aa8b732006-12-10 02:21:36 -08001707}
1708EXPORT_SYMBOL_GPL(kvm_resched);
1709
Konstantin Weitz41628d32012-04-25 15:30:38 +02001710bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1711{
1712 struct pid *pid;
1713 struct task_struct *task = NULL;
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301714 bool ret = false;
Konstantin Weitz41628d32012-04-25 15:30:38 +02001715
1716 rcu_read_lock();
1717 pid = rcu_dereference(target->pid);
1718 if (pid)
1719 task = get_pid_task(target->pid, PIDTYPE_PID);
1720 rcu_read_unlock();
1721 if (!task)
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301722 return ret;
Konstantin Weitz41628d32012-04-25 15:30:38 +02001723 if (task->flags & PF_VCPU) {
1724 put_task_struct(task);
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301725 return ret;
Konstantin Weitz41628d32012-04-25 15:30:38 +02001726 }
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301727 ret = yield_to(task, 1);
Konstantin Weitz41628d32012-04-25 15:30:38 +02001728 put_task_struct(task);
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301729
1730 return ret;
Konstantin Weitz41628d32012-04-25 15:30:38 +02001731}
1732EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1733
Raghavendra K T06e48c52012-07-19 15:17:52 +05301734#ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1735/*
1736 * Helper that checks whether a VCPU is eligible for directed yield.
1737 * Most eligible candidate to yield is decided by following heuristics:
1738 *
1739 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1740 * (preempted lock holder), indicated by @in_spin_loop.
1741 * Set at the beiginning and cleared at the end of interception/PLE handler.
1742 *
1743 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1744 * chance last time (mostly it has become eligible now since we have probably
1745 * yielded to lockholder in last iteration. This is done by toggling
1746 * @dy_eligible each time a VCPU checked for eligibility.)
1747 *
1748 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1749 * to preempted lock-holder could result in wrong VCPU selection and CPU
1750 * burning. Giving priority for a potential lock-holder increases lock
1751 * progress.
1752 *
1753 * Since algorithm is based on heuristics, accessing another VCPU data without
1754 * locking does not harm. It may result in trying to yield to same VCPU, fail
1755 * and continue with next VCPU and so on.
1756 */
1757bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1758{
1759 bool eligible;
1760
1761 eligible = !vcpu->spin_loop.in_spin_loop ||
1762 (vcpu->spin_loop.in_spin_loop &&
1763 vcpu->spin_loop.dy_eligible);
1764
1765 if (vcpu->spin_loop.in_spin_loop)
1766 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1767
1768 return eligible;
1769}
1770#endif
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301771
Rik van Riel217ece62011-02-01 09:53:28 -05001772void kvm_vcpu_on_spin(struct kvm_vcpu *me)
Zhai, Edwind255f4f2009-10-09 18:03:20 +08001773{
Rik van Riel217ece62011-02-01 09:53:28 -05001774 struct kvm *kvm = me->kvm;
1775 struct kvm_vcpu *vcpu;
1776 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1777 int yielded = 0;
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301778 int try = 3;
Rik van Riel217ece62011-02-01 09:53:28 -05001779 int pass;
1780 int i;
Zhai, Edwind255f4f2009-10-09 18:03:20 +08001781
Raghavendra K T4c088492012-07-18 19:07:46 +05301782 kvm_vcpu_set_in_spin_loop(me, true);
Rik van Riel217ece62011-02-01 09:53:28 -05001783 /*
1784 * We boost the priority of a VCPU that is runnable but not
1785 * currently running, because it got preempted by something
1786 * else and called schedule in __vcpu_run. Hopefully that
1787 * VCPU is holding the lock that we need and will release it.
1788 * We approximate round-robin by starting at the last boosted VCPU.
1789 */
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301790 for (pass = 0; pass < 2 && !yielded && try; pass++) {
Rik van Riel217ece62011-02-01 09:53:28 -05001791 kvm_for_each_vcpu(i, vcpu, kvm) {
Rik van Riel5cfc2aa2012-06-19 16:51:04 -04001792 if (!pass && i <= last_boosted_vcpu) {
Rik van Riel217ece62011-02-01 09:53:28 -05001793 i = last_boosted_vcpu;
1794 continue;
1795 } else if (pass && i > last_boosted_vcpu)
1796 break;
Raghavendra K T7bc7ae22013-03-04 23:32:27 +05301797 if (!ACCESS_ONCE(vcpu->preempted))
1798 continue;
Rik van Riel217ece62011-02-01 09:53:28 -05001799 if (vcpu == me)
1800 continue;
1801 if (waitqueue_active(&vcpu->wq))
1802 continue;
Raghavendra K T06e48c52012-07-19 15:17:52 +05301803 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1804 continue;
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301805
1806 yielded = kvm_vcpu_yield_to(vcpu);
1807 if (yielded > 0) {
Rik van Riel217ece62011-02-01 09:53:28 -05001808 kvm->last_boosted_vcpu = i;
Rik van Riel217ece62011-02-01 09:53:28 -05001809 break;
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301810 } else if (yielded < 0) {
1811 try--;
1812 if (!try)
1813 break;
Rik van Riel217ece62011-02-01 09:53:28 -05001814 }
Rik van Riel217ece62011-02-01 09:53:28 -05001815 }
1816 }
Raghavendra K T4c088492012-07-18 19:07:46 +05301817 kvm_vcpu_set_in_spin_loop(me, false);
Raghavendra K T06e48c52012-07-19 15:17:52 +05301818
1819 /* Ensure vcpu is not eligible during next spinloop */
1820 kvm_vcpu_set_dy_eligible(me, false);
Zhai, Edwind255f4f2009-10-09 18:03:20 +08001821}
1822EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1823
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001824static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001825{
1826 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001827 struct page *page;
1828
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001829 if (vmf->pgoff == 0)
Avi Kivity039576c2007-03-20 12:46:50 +02001830 page = virt_to_page(vcpu->run);
Avi Kivity09566762008-01-23 18:14:23 +02001831#ifdef CONFIG_X86
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001832 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
Zhang Xiantaoad312c72007-12-13 23:50:52 +08001833 page = virt_to_page(vcpu->arch.pio_data);
Avi Kivity09566762008-01-23 18:14:23 +02001834#endif
Laurent Vivier5f94c172008-05-30 16:05:54 +02001835#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1836 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1837 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1838#endif
Avi Kivity039576c2007-03-20 12:46:50 +02001839 else
Carsten Otte5b1c1492012-01-04 10:25:23 +01001840 return kvm_arch_vcpu_fault(vcpu, vmf);
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001841 get_page(page);
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001842 vmf->page = page;
1843 return 0;
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001844}
1845
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04001846static const struct vm_operations_struct kvm_vcpu_vm_ops = {
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001847 .fault = kvm_vcpu_fault,
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001848};
1849
1850static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1851{
1852 vma->vm_ops = &kvm_vcpu_vm_ops;
1853 return 0;
1854}
1855
Avi Kivitybccf2152007-02-21 18:04:26 +02001856static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1857{
1858 struct kvm_vcpu *vcpu = filp->private_data;
1859
Al Viro66c0b392008-04-19 20:33:56 +01001860 kvm_put_kvm(vcpu->kvm);
Avi Kivitybccf2152007-02-21 18:04:26 +02001861 return 0;
1862}
1863
Christian Borntraeger3d3aab12008-12-02 11:17:32 +01001864static struct file_operations kvm_vcpu_fops = {
Avi Kivitybccf2152007-02-21 18:04:26 +02001865 .release = kvm_vcpu_release,
1866 .unlocked_ioctl = kvm_vcpu_ioctl,
Alexander Graf1dda6062011-06-08 02:45:37 +02001867#ifdef CONFIG_COMPAT
1868 .compat_ioctl = kvm_vcpu_compat_ioctl,
1869#endif
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001870 .mmap = kvm_vcpu_mmap,
Arnd Bergmann6038f372010-08-15 18:52:59 +02001871 .llseek = noop_llseek,
Avi Kivitybccf2152007-02-21 18:04:26 +02001872};
1873
1874/*
1875 * Allocates an inode for the vcpu.
1876 */
1877static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1878{
Yann Droneaud24009b02013-08-24 22:14:07 +02001879 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
Avi Kivitybccf2152007-02-21 18:04:26 +02001880}
1881
Avi Kivityc5ea7662007-02-20 18:41:05 +02001882/*
1883 * Creates some virtual cpus. Good luck creating more than one.
1884 */
Gleb Natapov73880c82009-06-09 15:56:28 +03001885static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
Avi Kivityc5ea7662007-02-20 18:41:05 +02001886{
1887 int r;
Gleb Natapov988a2ca2009-06-09 15:56:29 +03001888 struct kvm_vcpu *vcpu, *v;
Avi Kivityc5ea7662007-02-20 18:41:05 +02001889
Gleb Natapov73880c82009-06-09 15:56:28 +03001890 vcpu = kvm_arch_vcpu_create(kvm, id);
Rusty Russellfb3f0f52007-07-27 17:16:56 +10001891 if (IS_ERR(vcpu))
1892 return PTR_ERR(vcpu);
Avi Kivityc5ea7662007-02-20 18:41:05 +02001893
Avi Kivity15ad7142007-07-11 18:17:21 +03001894 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1895
Avi Kivity26e52152007-11-20 15:30:24 +02001896 r = kvm_arch_vcpu_setup(vcpu);
1897 if (r)
Jan Kiszkad7805922011-05-23 10:33:05 +02001898 goto vcpu_destroy;
Avi Kivity26e52152007-11-20 15:30:24 +02001899
Shaohua Li11ec2802007-07-23 14:51:37 +08001900 mutex_lock(&kvm->lock);
Avi Kivity3e515702012-03-05 14:23:29 +02001901 if (!kvm_vcpu_compatible(vcpu)) {
1902 r = -EINVAL;
1903 goto unlock_vcpu_destroy;
1904 }
Gleb Natapov73880c82009-06-09 15:56:28 +03001905 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1906 r = -EINVAL;
Jan Kiszkad7805922011-05-23 10:33:05 +02001907 goto unlock_vcpu_destroy;
Rusty Russellfb3f0f52007-07-27 17:16:56 +10001908 }
Gleb Natapov73880c82009-06-09 15:56:28 +03001909
Gleb Natapov988a2ca2009-06-09 15:56:29 +03001910 kvm_for_each_vcpu(r, v, kvm)
1911 if (v->vcpu_id == id) {
Gleb Natapov73880c82009-06-09 15:56:28 +03001912 r = -EEXIST;
Jan Kiszkad7805922011-05-23 10:33:05 +02001913 goto unlock_vcpu_destroy;
Gleb Natapov73880c82009-06-09 15:56:28 +03001914 }
1915
1916 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
Rusty Russellfb3f0f52007-07-27 17:16:56 +10001917
1918 /* Now it's all set up, let userspace reach it */
Al Viro66c0b392008-04-19 20:33:56 +01001919 kvm_get_kvm(kvm);
Avi Kivitybccf2152007-02-21 18:04:26 +02001920 r = create_vcpu_fd(vcpu);
Gleb Natapov73880c82009-06-09 15:56:28 +03001921 if (r < 0) {
1922 kvm_put_kvm(kvm);
Jan Kiszkad7805922011-05-23 10:33:05 +02001923 goto unlock_vcpu_destroy;
Gleb Natapov73880c82009-06-09 15:56:28 +03001924 }
1925
1926 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1927 smp_wmb();
1928 atomic_inc(&kvm->online_vcpus);
1929
Gleb Natapov73880c82009-06-09 15:56:28 +03001930 mutex_unlock(&kvm->lock);
Marcelo Tosatti42897d82012-11-27 23:29:02 -02001931 kvm_arch_vcpu_postcreate(vcpu);
Avi Kivitybccf2152007-02-21 18:04:26 +02001932 return r;
Avi Kivityc5ea7662007-02-20 18:41:05 +02001933
Jan Kiszkad7805922011-05-23 10:33:05 +02001934unlock_vcpu_destroy:
Glauber Costa7d8fece2008-09-17 23:16:59 -03001935 mutex_unlock(&kvm->lock);
Jan Kiszkad7805922011-05-23 10:33:05 +02001936vcpu_destroy:
Hollis Blanchardd40ccc62007-11-19 14:04:43 -06001937 kvm_arch_vcpu_destroy(vcpu);
Avi Kivityc5ea7662007-02-20 18:41:05 +02001938 return r;
1939}
1940
Avi Kivity1961d272007-03-05 19:46:05 +02001941static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1942{
1943 if (sigset) {
1944 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1945 vcpu->sigset_active = 1;
1946 vcpu->sigset = *sigset;
1947 } else
1948 vcpu->sigset_active = 0;
1949 return 0;
1950}
1951
Avi Kivitybccf2152007-02-21 18:04:26 +02001952static long kvm_vcpu_ioctl(struct file *filp,
1953 unsigned int ioctl, unsigned long arg)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001954{
Avi Kivitybccf2152007-02-21 18:04:26 +02001955 struct kvm_vcpu *vcpu = filp->private_data;
Al Viro2f3669872007-02-09 16:38:35 +00001956 void __user *argp = (void __user *)arg;
Carsten Otte313a3dc2007-10-11 19:16:52 +02001957 int r;
Dave Hansenfa3795a2008-08-11 10:01:46 -07001958 struct kvm_fpu *fpu = NULL;
1959 struct kvm_sregs *kvm_sregs = NULL;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001960
Avi Kivity6d4e4c42007-11-21 16:41:05 +02001961 if (vcpu->kvm->mm != current->mm)
1962 return -EIO;
Avi Kivity2122ff52010-05-13 11:25:04 +03001963
Sanjay Lal2f4d9b52012-11-21 18:34:15 -08001964#if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
Avi Kivity2122ff52010-05-13 11:25:04 +03001965 /*
1966 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1967 * so vcpu_load() would break it.
1968 */
1969 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1970 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1971#endif
1972
1973
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +03001974 r = vcpu_load(vcpu);
1975 if (r)
1976 return r;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001977 switch (ioctl) {
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001978 case KVM_RUN:
Avi Kivityf0fe5102007-03-07 13:11:17 +02001979 r = -EINVAL;
1980 if (arg)
1981 goto out;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05001982 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
Gleb Natapov64be5002010-10-24 16:49:08 +02001983 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001984 break;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001985 case KVM_GET_REGS: {
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08001986 struct kvm_regs *kvm_regs;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001987
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08001988 r = -ENOMEM;
1989 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1990 if (!kvm_regs)
1991 goto out;
1992 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001993 if (r)
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08001994 goto out_free1;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001995 r = -EFAULT;
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08001996 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1997 goto out_free1;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001998 r = 0;
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08001999out_free1:
2000 kfree(kvm_regs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002001 break;
2002 }
2003 case KVM_SET_REGS: {
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002004 struct kvm_regs *kvm_regs;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002005
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002006 r = -ENOMEM;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002007 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2008 if (IS_ERR(kvm_regs)) {
2009 r = PTR_ERR(kvm_regs);
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002010 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002011 }
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002012 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002013 kfree(kvm_regs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002014 break;
2015 }
2016 case KVM_GET_SREGS: {
Dave Hansenfa3795a2008-08-11 10:01:46 -07002017 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2018 r = -ENOMEM;
2019 if (!kvm_sregs)
2020 goto out;
2021 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002022 if (r)
2023 goto out;
2024 r = -EFAULT;
Dave Hansenfa3795a2008-08-11 10:01:46 -07002025 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002026 goto out;
2027 r = 0;
2028 break;
2029 }
2030 case KVM_SET_SREGS: {
Sasha Levinff5c2c02011-12-04 19:36:29 +02002031 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2032 if (IS_ERR(kvm_sregs)) {
2033 r = PTR_ERR(kvm_sregs);
Guo Chao18595412012-11-02 18:33:21 +08002034 kvm_sregs = NULL;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002035 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002036 }
Dave Hansenfa3795a2008-08-11 10:01:46 -07002037 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002038 break;
2039 }
Marcelo Tosatti62d9f0d2008-04-11 13:24:45 -03002040 case KVM_GET_MP_STATE: {
2041 struct kvm_mp_state mp_state;
2042
2043 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2044 if (r)
2045 goto out;
2046 r = -EFAULT;
2047 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2048 goto out;
2049 r = 0;
2050 break;
2051 }
2052 case KVM_SET_MP_STATE: {
2053 struct kvm_mp_state mp_state;
2054
2055 r = -EFAULT;
2056 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2057 goto out;
2058 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
Marcelo Tosatti62d9f0d2008-04-11 13:24:45 -03002059 break;
2060 }
Avi Kivity6aa8b732006-12-10 02:21:36 -08002061 case KVM_TRANSLATE: {
2062 struct kvm_translation tr;
2063
2064 r = -EFAULT;
Al Viro2f3669872007-02-09 16:38:35 +00002065 if (copy_from_user(&tr, argp, sizeof tr))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002066 goto out;
Zhang Xiantao8b006792007-11-16 13:05:55 +08002067 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002068 if (r)
2069 goto out;
2070 r = -EFAULT;
Al Viro2f3669872007-02-09 16:38:35 +00002071 if (copy_to_user(argp, &tr, sizeof tr))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002072 goto out;
2073 r = 0;
2074 break;
2075 }
Jan Kiszkad0bfb942008-12-15 13:52:10 +01002076 case KVM_SET_GUEST_DEBUG: {
2077 struct kvm_guest_debug dbg;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002078
2079 r = -EFAULT;
Al Viro2f3669872007-02-09 16:38:35 +00002080 if (copy_from_user(&dbg, argp, sizeof dbg))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002081 goto out;
Jan Kiszkad0bfb942008-12-15 13:52:10 +01002082 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002083 break;
2084 }
Avi Kivity1961d272007-03-05 19:46:05 +02002085 case KVM_SET_SIGNAL_MASK: {
2086 struct kvm_signal_mask __user *sigmask_arg = argp;
2087 struct kvm_signal_mask kvm_sigmask;
2088 sigset_t sigset, *p;
2089
2090 p = NULL;
2091 if (argp) {
2092 r = -EFAULT;
2093 if (copy_from_user(&kvm_sigmask, argp,
2094 sizeof kvm_sigmask))
2095 goto out;
2096 r = -EINVAL;
2097 if (kvm_sigmask.len != sizeof sigset)
2098 goto out;
2099 r = -EFAULT;
2100 if (copy_from_user(&sigset, sigmask_arg->sigset,
2101 sizeof sigset))
2102 goto out;
2103 p = &sigset;
2104 }
Andi Kleen376d41f2010-06-10 13:10:47 +02002105 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
Avi Kivity1961d272007-03-05 19:46:05 +02002106 break;
2107 }
Avi Kivityb8836732007-04-01 16:34:31 +03002108 case KVM_GET_FPU: {
Dave Hansenfa3795a2008-08-11 10:01:46 -07002109 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2110 r = -ENOMEM;
2111 if (!fpu)
2112 goto out;
2113 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
Avi Kivityb8836732007-04-01 16:34:31 +03002114 if (r)
2115 goto out;
2116 r = -EFAULT;
Dave Hansenfa3795a2008-08-11 10:01:46 -07002117 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
Avi Kivityb8836732007-04-01 16:34:31 +03002118 goto out;
2119 r = 0;
2120 break;
2121 }
2122 case KVM_SET_FPU: {
Sasha Levinff5c2c02011-12-04 19:36:29 +02002123 fpu = memdup_user(argp, sizeof(*fpu));
2124 if (IS_ERR(fpu)) {
2125 r = PTR_ERR(fpu);
Guo Chao18595412012-11-02 18:33:21 +08002126 fpu = NULL;
Avi Kivityb8836732007-04-01 16:34:31 +03002127 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002128 }
Dave Hansenfa3795a2008-08-11 10:01:46 -07002129 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
Avi Kivityb8836732007-04-01 16:34:31 +03002130 break;
2131 }
Avi Kivitybccf2152007-02-21 18:04:26 +02002132 default:
Carsten Otte313a3dc2007-10-11 19:16:52 +02002133 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
Avi Kivitybccf2152007-02-21 18:04:26 +02002134 }
2135out:
Avi Kivity2122ff52010-05-13 11:25:04 +03002136 vcpu_put(vcpu);
Dave Hansenfa3795a2008-08-11 10:01:46 -07002137 kfree(fpu);
2138 kfree(kvm_sregs);
Avi Kivitybccf2152007-02-21 18:04:26 +02002139 return r;
2140}
2141
Alexander Graf1dda6062011-06-08 02:45:37 +02002142#ifdef CONFIG_COMPAT
2143static long kvm_vcpu_compat_ioctl(struct file *filp,
2144 unsigned int ioctl, unsigned long arg)
2145{
2146 struct kvm_vcpu *vcpu = filp->private_data;
2147 void __user *argp = compat_ptr(arg);
2148 int r;
2149
2150 if (vcpu->kvm->mm != current->mm)
2151 return -EIO;
2152
2153 switch (ioctl) {
2154 case KVM_SET_SIGNAL_MASK: {
2155 struct kvm_signal_mask __user *sigmask_arg = argp;
2156 struct kvm_signal_mask kvm_sigmask;
2157 compat_sigset_t csigset;
2158 sigset_t sigset;
2159
2160 if (argp) {
2161 r = -EFAULT;
2162 if (copy_from_user(&kvm_sigmask, argp,
2163 sizeof kvm_sigmask))
2164 goto out;
2165 r = -EINVAL;
2166 if (kvm_sigmask.len != sizeof csigset)
2167 goto out;
2168 r = -EFAULT;
2169 if (copy_from_user(&csigset, sigmask_arg->sigset,
2170 sizeof csigset))
2171 goto out;
Alan Cox760a9a32012-08-22 14:34:11 +01002172 sigset_from_compat(&sigset, &csigset);
2173 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2174 } else
2175 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
Alexander Graf1dda6062011-06-08 02:45:37 +02002176 break;
2177 }
2178 default:
2179 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2180 }
2181
2182out:
2183 return r;
2184}
2185#endif
2186
Scott Wood852b6d52013-04-12 14:08:42 +00002187static int kvm_device_ioctl_attr(struct kvm_device *dev,
2188 int (*accessor)(struct kvm_device *dev,
2189 struct kvm_device_attr *attr),
2190 unsigned long arg)
2191{
2192 struct kvm_device_attr attr;
2193
2194 if (!accessor)
2195 return -EPERM;
2196
2197 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2198 return -EFAULT;
2199
2200 return accessor(dev, &attr);
2201}
2202
2203static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2204 unsigned long arg)
2205{
2206 struct kvm_device *dev = filp->private_data;
2207
2208 switch (ioctl) {
2209 case KVM_SET_DEVICE_ATTR:
2210 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2211 case KVM_GET_DEVICE_ATTR:
2212 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2213 case KVM_HAS_DEVICE_ATTR:
2214 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2215 default:
2216 if (dev->ops->ioctl)
2217 return dev->ops->ioctl(dev, ioctl, arg);
2218
2219 return -ENOTTY;
2220 }
2221}
2222
Scott Wood852b6d52013-04-12 14:08:42 +00002223static int kvm_device_release(struct inode *inode, struct file *filp)
2224{
2225 struct kvm_device *dev = filp->private_data;
2226 struct kvm *kvm = dev->kvm;
2227
Scott Wood852b6d52013-04-12 14:08:42 +00002228 kvm_put_kvm(kvm);
2229 return 0;
2230}
2231
2232static const struct file_operations kvm_device_fops = {
2233 .unlocked_ioctl = kvm_device_ioctl,
Scott Wooddb6ae612013-04-30 20:00:45 -05002234#ifdef CONFIG_COMPAT
2235 .compat_ioctl = kvm_device_ioctl,
2236#endif
Scott Wood852b6d52013-04-12 14:08:42 +00002237 .release = kvm_device_release,
2238};
2239
2240struct kvm_device *kvm_device_from_filp(struct file *filp)
2241{
2242 if (filp->f_op != &kvm_device_fops)
2243 return NULL;
2244
2245 return filp->private_data;
2246}
2247
2248static int kvm_ioctl_create_device(struct kvm *kvm,
2249 struct kvm_create_device *cd)
2250{
2251 struct kvm_device_ops *ops = NULL;
2252 struct kvm_device *dev;
2253 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2254 int ret;
2255
2256 switch (cd->type) {
Scott Wood5df554ad2013-04-12 14:08:46 +00002257#ifdef CONFIG_KVM_MPIC
2258 case KVM_DEV_TYPE_FSL_MPIC_20:
2259 case KVM_DEV_TYPE_FSL_MPIC_42:
2260 ops = &kvm_mpic_ops;
2261 break;
2262#endif
Paul Mackerras5975a2e2013-04-27 00:28:37 +00002263#ifdef CONFIG_KVM_XICS
2264 case KVM_DEV_TYPE_XICS:
2265 ops = &kvm_xics_ops;
2266 break;
2267#endif
Scott Wood852b6d52013-04-12 14:08:42 +00002268 default:
2269 return -ENODEV;
2270 }
2271
2272 if (test)
2273 return 0;
2274
2275 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2276 if (!dev)
2277 return -ENOMEM;
2278
2279 dev->ops = ops;
2280 dev->kvm = kvm;
Scott Wood852b6d52013-04-12 14:08:42 +00002281
2282 ret = ops->create(dev, cd->type);
2283 if (ret < 0) {
2284 kfree(dev);
2285 return ret;
2286 }
2287
Yann Droneaud24009b02013-08-24 22:14:07 +02002288 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
Scott Wood852b6d52013-04-12 14:08:42 +00002289 if (ret < 0) {
2290 ops->destroy(dev);
2291 return ret;
2292 }
2293
Scott Wood07f0a7b2013-04-25 14:11:23 +00002294 list_add(&dev->vm_node, &kvm->devices);
Scott Wood852b6d52013-04-12 14:08:42 +00002295 kvm_get_kvm(kvm);
2296 cd->fd = ret;
2297 return 0;
2298}
2299
Avi Kivitybccf2152007-02-21 18:04:26 +02002300static long kvm_vm_ioctl(struct file *filp,
2301 unsigned int ioctl, unsigned long arg)
2302{
2303 struct kvm *kvm = filp->private_data;
2304 void __user *argp = (void __user *)arg;
Carsten Otte1fe779f2007-10-29 16:08:35 +01002305 int r;
Avi Kivitybccf2152007-02-21 18:04:26 +02002306
Avi Kivity6d4e4c42007-11-21 16:41:05 +02002307 if (kvm->mm != current->mm)
2308 return -EIO;
Avi Kivitybccf2152007-02-21 18:04:26 +02002309 switch (ioctl) {
2310 case KVM_CREATE_VCPU:
2311 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
Avi Kivitybccf2152007-02-21 18:04:26 +02002312 break;
Izik Eidus6fc138d2007-10-09 19:20:39 +02002313 case KVM_SET_USER_MEMORY_REGION: {
2314 struct kvm_userspace_memory_region kvm_userspace_mem;
2315
2316 r = -EFAULT;
2317 if (copy_from_user(&kvm_userspace_mem, argp,
2318 sizeof kvm_userspace_mem))
2319 goto out;
2320
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +09002321 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002322 break;
2323 }
2324 case KVM_GET_DIRTY_LOG: {
2325 struct kvm_dirty_log log;
2326
2327 r = -EFAULT;
Al Viro2f3669872007-02-09 16:38:35 +00002328 if (copy_from_user(&log, argp, sizeof log))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002329 goto out;
Avi Kivity2c6f5df2007-02-20 18:27:58 +02002330 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002331 break;
2332 }
Laurent Vivier5f94c172008-05-30 16:05:54 +02002333#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2334 case KVM_REGISTER_COALESCED_MMIO: {
2335 struct kvm_coalesced_mmio_zone zone;
2336 r = -EFAULT;
2337 if (copy_from_user(&zone, argp, sizeof zone))
2338 goto out;
Laurent Vivier5f94c172008-05-30 16:05:54 +02002339 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
Laurent Vivier5f94c172008-05-30 16:05:54 +02002340 break;
2341 }
2342 case KVM_UNREGISTER_COALESCED_MMIO: {
2343 struct kvm_coalesced_mmio_zone zone;
2344 r = -EFAULT;
2345 if (copy_from_user(&zone, argp, sizeof zone))
2346 goto out;
Laurent Vivier5f94c172008-05-30 16:05:54 +02002347 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
Laurent Vivier5f94c172008-05-30 16:05:54 +02002348 break;
2349 }
2350#endif
Gregory Haskins721eecbf2009-05-20 10:30:49 -04002351 case KVM_IRQFD: {
2352 struct kvm_irqfd data;
2353
2354 r = -EFAULT;
2355 if (copy_from_user(&data, argp, sizeof data))
2356 goto out;
Alex Williamsond4db2932012-06-29 09:56:08 -06002357 r = kvm_irqfd(kvm, &data);
Gregory Haskins721eecbf2009-05-20 10:30:49 -04002358 break;
2359 }
Gregory Haskinsd34e6b12009-07-07 17:08:49 -04002360 case KVM_IOEVENTFD: {
2361 struct kvm_ioeventfd data;
2362
2363 r = -EFAULT;
2364 if (copy_from_user(&data, argp, sizeof data))
2365 goto out;
2366 r = kvm_ioeventfd(kvm, &data);
2367 break;
2368 }
Gleb Natapov73880c82009-06-09 15:56:28 +03002369#ifdef CONFIG_KVM_APIC_ARCHITECTURE
2370 case KVM_SET_BOOT_CPU_ID:
2371 r = 0;
Marcelo Tosatti894a9c52009-06-23 15:05:14 -03002372 mutex_lock(&kvm->lock);
Gleb Natapov73880c82009-06-09 15:56:28 +03002373 if (atomic_read(&kvm->online_vcpus) != 0)
2374 r = -EBUSY;
2375 else
2376 kvm->bsp_vcpu_id = arg;
Marcelo Tosatti894a9c52009-06-23 15:05:14 -03002377 mutex_unlock(&kvm->lock);
Gleb Natapov73880c82009-06-09 15:56:28 +03002378 break;
2379#endif
Jan Kiszka07975ad2012-03-29 21:14:12 +02002380#ifdef CONFIG_HAVE_KVM_MSI
2381 case KVM_SIGNAL_MSI: {
2382 struct kvm_msi msi;
2383
2384 r = -EFAULT;
2385 if (copy_from_user(&msi, argp, sizeof msi))
2386 goto out;
2387 r = kvm_send_userspace_msi(kvm, &msi);
2388 break;
2389 }
2390#endif
Christoffer Dall23d43cf2012-07-24 08:51:20 -04002391#ifdef __KVM_HAVE_IRQ_LINE
2392 case KVM_IRQ_LINE_STATUS:
2393 case KVM_IRQ_LINE: {
2394 struct kvm_irq_level irq_event;
2395
2396 r = -EFAULT;
2397 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2398 goto out;
2399
Yang Zhangaa2fbe62013-04-11 19:21:40 +08002400 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2401 ioctl == KVM_IRQ_LINE_STATUS);
Christoffer Dall23d43cf2012-07-24 08:51:20 -04002402 if (r)
2403 goto out;
2404
2405 r = -EFAULT;
2406 if (ioctl == KVM_IRQ_LINE_STATUS) {
2407 if (copy_to_user(argp, &irq_event, sizeof irq_event))
2408 goto out;
2409 }
2410
2411 r = 0;
2412 break;
2413 }
2414#endif
Alexander Grafaa8d5942013-04-15 21:12:53 +02002415#ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2416 case KVM_SET_GSI_ROUTING: {
2417 struct kvm_irq_routing routing;
2418 struct kvm_irq_routing __user *urouting;
2419 struct kvm_irq_routing_entry *entries;
2420
2421 r = -EFAULT;
2422 if (copy_from_user(&routing, argp, sizeof(routing)))
2423 goto out;
2424 r = -EINVAL;
2425 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2426 goto out;
2427 if (routing.flags)
2428 goto out;
2429 r = -ENOMEM;
2430 entries = vmalloc(routing.nr * sizeof(*entries));
2431 if (!entries)
2432 goto out;
2433 r = -EFAULT;
2434 urouting = argp;
2435 if (copy_from_user(entries, urouting->entries,
2436 routing.nr * sizeof(*entries)))
2437 goto out_free_irq_routing;
2438 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2439 routing.flags);
2440 out_free_irq_routing:
2441 vfree(entries);
2442 break;
2443 }
2444#endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
Scott Wood852b6d52013-04-12 14:08:42 +00002445 case KVM_CREATE_DEVICE: {
2446 struct kvm_create_device cd;
2447
2448 r = -EFAULT;
2449 if (copy_from_user(&cd, argp, sizeof(cd)))
2450 goto out;
2451
2452 r = kvm_ioctl_create_device(kvm, &cd);
2453 if (r)
2454 goto out;
2455
2456 r = -EFAULT;
2457 if (copy_to_user(argp, &cd, sizeof(cd)))
2458 goto out;
2459
2460 r = 0;
2461 break;
2462 }
Avi Kivityf17abe92007-02-21 19:28:04 +02002463 default:
Carsten Otte1fe779f2007-10-29 16:08:35 +01002464 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
Avi Kivitybfd99ff2009-08-26 14:57:50 +03002465 if (r == -ENOTTY)
2466 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
Avi Kivityf17abe92007-02-21 19:28:04 +02002467 }
2468out:
2469 return r;
2470}
2471
Arnd Bergmann6ff58942009-10-22 14:19:27 +02002472#ifdef CONFIG_COMPAT
2473struct compat_kvm_dirty_log {
2474 __u32 slot;
2475 __u32 padding1;
2476 union {
2477 compat_uptr_t dirty_bitmap; /* one bit per page */
2478 __u64 padding2;
2479 };
2480};
2481
2482static long kvm_vm_compat_ioctl(struct file *filp,
2483 unsigned int ioctl, unsigned long arg)
2484{
2485 struct kvm *kvm = filp->private_data;
2486 int r;
2487
2488 if (kvm->mm != current->mm)
2489 return -EIO;
2490 switch (ioctl) {
2491 case KVM_GET_DIRTY_LOG: {
2492 struct compat_kvm_dirty_log compat_log;
2493 struct kvm_dirty_log log;
2494
2495 r = -EFAULT;
2496 if (copy_from_user(&compat_log, (void __user *)arg,
2497 sizeof(compat_log)))
2498 goto out;
2499 log.slot = compat_log.slot;
2500 log.padding1 = compat_log.padding1;
2501 log.padding2 = compat_log.padding2;
2502 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2503
2504 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
Arnd Bergmann6ff58942009-10-22 14:19:27 +02002505 break;
2506 }
2507 default:
2508 r = kvm_vm_ioctl(filp, ioctl, arg);
2509 }
2510
2511out:
2512 return r;
2513}
2514#endif
2515
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11002516static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Avi Kivityf17abe92007-02-21 19:28:04 +02002517{
Marcelo Tosatti777b3f42008-09-16 20:54:46 -03002518 struct page *page[1];
2519 unsigned long addr;
2520 int npages;
2521 gfn_t gfn = vmf->pgoff;
Avi Kivityf17abe92007-02-21 19:28:04 +02002522 struct kvm *kvm = vma->vm_file->private_data;
Avi Kivityf17abe92007-02-21 19:28:04 +02002523
Marcelo Tosatti777b3f42008-09-16 20:54:46 -03002524 addr = gfn_to_hva(kvm, gfn);
2525 if (kvm_is_error_hva(addr))
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11002526 return VM_FAULT_SIGBUS;
Marcelo Tosatti777b3f42008-09-16 20:54:46 -03002527
2528 npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2529 NULL);
2530 if (unlikely(npages != 1))
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11002531 return VM_FAULT_SIGBUS;
Marcelo Tosatti777b3f42008-09-16 20:54:46 -03002532
2533 vmf->page = page[0];
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11002534 return 0;
Avi Kivityf17abe92007-02-21 19:28:04 +02002535}
2536
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04002537static const struct vm_operations_struct kvm_vm_vm_ops = {
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11002538 .fault = kvm_vm_fault,
Avi Kivityf17abe92007-02-21 19:28:04 +02002539};
2540
2541static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2542{
2543 vma->vm_ops = &kvm_vm_vm_ops;
2544 return 0;
2545}
2546
Christian Borntraeger3d3aab12008-12-02 11:17:32 +01002547static struct file_operations kvm_vm_fops = {
Avi Kivityf17abe92007-02-21 19:28:04 +02002548 .release = kvm_vm_release,
2549 .unlocked_ioctl = kvm_vm_ioctl,
Arnd Bergmann6ff58942009-10-22 14:19:27 +02002550#ifdef CONFIG_COMPAT
2551 .compat_ioctl = kvm_vm_compat_ioctl,
2552#endif
Avi Kivityf17abe92007-02-21 19:28:04 +02002553 .mmap = kvm_vm_mmap,
Arnd Bergmann6038f372010-08-15 18:52:59 +02002554 .llseek = noop_llseek,
Avi Kivityf17abe92007-02-21 19:28:04 +02002555};
2556
Carsten Ottee08b9632012-01-04 10:25:20 +01002557static int kvm_dev_ioctl_create_vm(unsigned long type)
Avi Kivityf17abe92007-02-21 19:28:04 +02002558{
Heiko Carstensaac87632010-10-27 17:22:10 +02002559 int r;
Avi Kivityf17abe92007-02-21 19:28:04 +02002560 struct kvm *kvm;
2561
Carsten Ottee08b9632012-01-04 10:25:20 +01002562 kvm = kvm_create_vm(type);
Avi Kivityd6d28162007-06-28 08:38:16 -04002563 if (IS_ERR(kvm))
2564 return PTR_ERR(kvm);
Takuya Yoshikawa6ce5a092010-03-15 22:13:30 +09002565#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2566 r = kvm_coalesced_mmio_init(kvm);
2567 if (r < 0) {
2568 kvm_put_kvm(kvm);
2569 return r;
2570 }
2571#endif
Yann Droneaud24009b02013-08-24 22:14:07 +02002572 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
Heiko Carstensaac87632010-10-27 17:22:10 +02002573 if (r < 0)
Al Viro66c0b392008-04-19 20:33:56 +01002574 kvm_put_kvm(kvm);
Avi Kivityf17abe92007-02-21 19:28:04 +02002575
Heiko Carstensaac87632010-10-27 17:22:10 +02002576 return r;
Avi Kivityf17abe92007-02-21 19:28:04 +02002577}
2578
Avi Kivity1a811b62008-12-08 18:25:27 +02002579static long kvm_dev_ioctl_check_extension_generic(long arg)
2580{
2581 switch (arg) {
Avi Kivityca9edae2008-12-08 18:29:29 +02002582 case KVM_CAP_USER_MEMORY:
Avi Kivity1a811b62008-12-08 18:25:27 +02002583 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
Jan Kiszka4cd481f2009-04-13 11:59:32 +02002584 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
Gleb Natapov73880c82009-06-09 15:56:28 +03002585#ifdef CONFIG_KVM_APIC_ARCHITECTURE
2586 case KVM_CAP_SET_BOOT_CPU_ID:
2587#endif
Avi Kivitya9c73992009-11-04 11:54:59 +02002588 case KVM_CAP_INTERNAL_ERROR_DATA:
Jan Kiszka07975ad2012-03-29 21:14:12 +02002589#ifdef CONFIG_HAVE_KVM_MSI
2590 case KVM_CAP_SIGNAL_MSI:
2591#endif
Alexander Graf7df35f52013-04-16 12:12:49 +02002592#ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2593 case KVM_CAP_IRQFD_RESAMPLE:
2594#endif
Avi Kivity1a811b62008-12-08 18:25:27 +02002595 return 1;
Alexander Grafa725d562013-04-17 13:29:30 +02002596#ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
Avi Kivity399ec802008-11-19 13:58:46 +02002597 case KVM_CAP_IRQ_ROUTING:
Sheng Yang36463142009-03-16 16:33:43 +08002598 return KVM_MAX_IRQ_ROUTES;
Avi Kivity399ec802008-11-19 13:58:46 +02002599#endif
Avi Kivity1a811b62008-12-08 18:25:27 +02002600 default:
2601 break;
2602 }
2603 return kvm_dev_ioctl_check_extension(arg);
2604}
2605
Avi Kivityf17abe92007-02-21 19:28:04 +02002606static long kvm_dev_ioctl(struct file *filp,
2607 unsigned int ioctl, unsigned long arg)
2608{
Avi Kivity07c45a32007-03-07 13:05:38 +02002609 long r = -EINVAL;
Avi Kivityf17abe92007-02-21 19:28:04 +02002610
2611 switch (ioctl) {
2612 case KVM_GET_API_VERSION:
Avi Kivityf0fe5102007-03-07 13:11:17 +02002613 r = -EINVAL;
2614 if (arg)
2615 goto out;
Avi Kivityf17abe92007-02-21 19:28:04 +02002616 r = KVM_API_VERSION;
2617 break;
2618 case KVM_CREATE_VM:
Carsten Ottee08b9632012-01-04 10:25:20 +01002619 r = kvm_dev_ioctl_create_vm(arg);
Avi Kivityf17abe92007-02-21 19:28:04 +02002620 break;
Zhang Xiantao018d00d2007-11-15 23:07:47 +08002621 case KVM_CHECK_EXTENSION:
Avi Kivity1a811b62008-12-08 18:25:27 +02002622 r = kvm_dev_ioctl_check_extension_generic(arg);
Avi Kivity5d308f42007-03-01 17:56:20 +02002623 break;
Avi Kivity07c45a32007-03-07 13:05:38 +02002624 case KVM_GET_VCPU_MMAP_SIZE:
2625 r = -EINVAL;
2626 if (arg)
2627 goto out;
Avi Kivityadb1ff42008-01-24 15:13:08 +02002628 r = PAGE_SIZE; /* struct kvm_run */
2629#ifdef CONFIG_X86
2630 r += PAGE_SIZE; /* pio data page */
2631#endif
Laurent Vivier5f94c172008-05-30 16:05:54 +02002632#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2633 r += PAGE_SIZE; /* coalesced mmio ring page */
2634#endif
Avi Kivity07c45a32007-03-07 13:05:38 +02002635 break;
Feng(Eric) Liud4c9ff22008-04-10 08:47:53 -04002636 case KVM_TRACE_ENABLE:
2637 case KVM_TRACE_PAUSE:
2638 case KVM_TRACE_DISABLE:
Marcelo Tosatti2023a292009-06-18 11:47:28 -03002639 r = -EOPNOTSUPP;
Feng(Eric) Liud4c9ff22008-04-10 08:47:53 -04002640 break;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002641 default:
Carsten Otte043405e2007-10-10 17:16:19 +02002642 return kvm_arch_dev_ioctl(filp, ioctl, arg);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002643 }
2644out:
2645 return r;
2646}
2647
Avi Kivity6aa8b732006-12-10 02:21:36 -08002648static struct file_operations kvm_chardev_ops = {
Avi Kivity6aa8b732006-12-10 02:21:36 -08002649 .unlocked_ioctl = kvm_dev_ioctl,
2650 .compat_ioctl = kvm_dev_ioctl,
Arnd Bergmann6038f372010-08-15 18:52:59 +02002651 .llseek = noop_llseek,
Avi Kivity6aa8b732006-12-10 02:21:36 -08002652};
2653
2654static struct miscdevice kvm_dev = {
Avi Kivitybbe44322007-03-04 13:27:36 +02002655 KVM_MINOR,
Avi Kivity6aa8b732006-12-10 02:21:36 -08002656 "kvm",
2657 &kvm_chardev_ops,
2658};
2659
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002660static void hardware_enable_nolock(void *junk)
Avi Kivity1b6c0162007-05-24 13:03:52 +03002661{
2662 int cpu = raw_smp_processor_id();
Alexander Graf10474ae2009-09-15 11:37:46 +02002663 int r;
Avi Kivity1b6c0162007-05-24 13:03:52 +03002664
Rusty Russell7f59f492008-12-07 21:25:45 +10302665 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
Avi Kivity1b6c0162007-05-24 13:03:52 +03002666 return;
Alexander Graf10474ae2009-09-15 11:37:46 +02002667
Rusty Russell7f59f492008-12-07 21:25:45 +10302668 cpumask_set_cpu(cpu, cpus_hardware_enabled);
Alexander Graf10474ae2009-09-15 11:37:46 +02002669
2670 r = kvm_arch_hardware_enable(NULL);
2671
2672 if (r) {
2673 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2674 atomic_inc(&hardware_enable_failed);
2675 printk(KERN_INFO "kvm: enabling virtualization on "
2676 "CPU%d failed\n", cpu);
2677 }
Avi Kivity1b6c0162007-05-24 13:03:52 +03002678}
2679
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002680static void hardware_enable(void *junk)
2681{
Jan Kiszkae935b832011-02-08 12:55:33 +01002682 raw_spin_lock(&kvm_lock);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002683 hardware_enable_nolock(junk);
Jan Kiszkae935b832011-02-08 12:55:33 +01002684 raw_spin_unlock(&kvm_lock);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002685}
2686
2687static void hardware_disable_nolock(void *junk)
Avi Kivity1b6c0162007-05-24 13:03:52 +03002688{
2689 int cpu = raw_smp_processor_id();
2690
Rusty Russell7f59f492008-12-07 21:25:45 +10302691 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
Avi Kivity1b6c0162007-05-24 13:03:52 +03002692 return;
Rusty Russell7f59f492008-12-07 21:25:45 +10302693 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08002694 kvm_arch_hardware_disable(NULL);
Avi Kivity1b6c0162007-05-24 13:03:52 +03002695}
2696
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002697static void hardware_disable(void *junk)
2698{
Jan Kiszkae935b832011-02-08 12:55:33 +01002699 raw_spin_lock(&kvm_lock);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002700 hardware_disable_nolock(junk);
Jan Kiszkae935b832011-02-08 12:55:33 +01002701 raw_spin_unlock(&kvm_lock);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002702}
2703
Alexander Graf10474ae2009-09-15 11:37:46 +02002704static void hardware_disable_all_nolock(void)
2705{
2706 BUG_ON(!kvm_usage_count);
2707
2708 kvm_usage_count--;
2709 if (!kvm_usage_count)
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002710 on_each_cpu(hardware_disable_nolock, NULL, 1);
Alexander Graf10474ae2009-09-15 11:37:46 +02002711}
2712
2713static void hardware_disable_all(void)
2714{
Jan Kiszkae935b832011-02-08 12:55:33 +01002715 raw_spin_lock(&kvm_lock);
Alexander Graf10474ae2009-09-15 11:37:46 +02002716 hardware_disable_all_nolock();
Jan Kiszkae935b832011-02-08 12:55:33 +01002717 raw_spin_unlock(&kvm_lock);
Alexander Graf10474ae2009-09-15 11:37:46 +02002718}
2719
2720static int hardware_enable_all(void)
2721{
2722 int r = 0;
2723
Jan Kiszkae935b832011-02-08 12:55:33 +01002724 raw_spin_lock(&kvm_lock);
Alexander Graf10474ae2009-09-15 11:37:46 +02002725
2726 kvm_usage_count++;
2727 if (kvm_usage_count == 1) {
2728 atomic_set(&hardware_enable_failed, 0);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002729 on_each_cpu(hardware_enable_nolock, NULL, 1);
Alexander Graf10474ae2009-09-15 11:37:46 +02002730
2731 if (atomic_read(&hardware_enable_failed)) {
2732 hardware_disable_all_nolock();
2733 r = -EBUSY;
2734 }
2735 }
2736
Jan Kiszkae935b832011-02-08 12:55:33 +01002737 raw_spin_unlock(&kvm_lock);
Alexander Graf10474ae2009-09-15 11:37:46 +02002738
2739 return r;
2740}
2741
Avi Kivity774c47f2007-02-12 00:54:47 -08002742static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2743 void *v)
2744{
2745 int cpu = (long)v;
2746
Alexander Graf10474ae2009-09-15 11:37:46 +02002747 if (!kvm_usage_count)
2748 return NOTIFY_OK;
2749
Avi Kivity1a6f4d72007-11-11 18:37:32 +02002750 val &= ~CPU_TASKS_FROZEN;
Avi Kivity774c47f2007-02-12 00:54:47 -08002751 switch (val) {
Avi Kivitycec9ad22007-05-24 13:11:41 +03002752 case CPU_DYING:
Avi Kivity6ec8a852007-08-19 15:57:26 +03002753 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2754 cpu);
2755 hardware_disable(NULL);
2756 break;
Zachary Amsdenda908f22010-08-19 22:07:27 -10002757 case CPU_STARTING:
Jeremy Katz43934a32007-02-19 14:37:46 +02002758 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2759 cpu);
Zachary Amsdenda908f22010-08-19 22:07:27 -10002760 hardware_enable(NULL);
Avi Kivity774c47f2007-02-12 00:54:47 -08002761 break;
2762 }
2763 return NOTIFY_OK;
2764}
2765
Rusty Russell9a2b85c2007-07-17 23:17:55 +10002766static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
Mike Dayd77c26f2007-10-08 09:02:08 -04002767 void *v)
Rusty Russell9a2b85c2007-07-17 23:17:55 +10002768{
Sheng Yang8e1c1812009-04-29 11:09:04 +08002769 /*
2770 * Some (well, at least mine) BIOSes hang on reboot if
2771 * in vmx root mode.
2772 *
2773 * And Intel TXT required VMX off for all cpu when system shutdown.
2774 */
2775 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2776 kvm_rebooting = true;
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002777 on_each_cpu(hardware_disable_nolock, NULL, 1);
Rusty Russell9a2b85c2007-07-17 23:17:55 +10002778 return NOTIFY_OK;
2779}
2780
2781static struct notifier_block kvm_reboot_notifier = {
2782 .notifier_call = kvm_reboot,
2783 .priority = 0,
2784};
2785
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002786static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002787{
2788 int i;
2789
2790 for (i = 0; i < bus->dev_count; i++) {
Sasha Levin743eeb02011-07-27 16:00:48 +03002791 struct kvm_io_device *pos = bus->range[i].dev;
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002792
2793 kvm_iodevice_destructor(pos);
2794 }
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002795 kfree(bus);
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002796}
2797
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02002798static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
2799 const struct kvm_io_range *r2)
Sasha Levin743eeb02011-07-27 16:00:48 +03002800{
Sasha Levin743eeb02011-07-27 16:00:48 +03002801 if (r1->addr < r2->addr)
2802 return -1;
2803 if (r1->addr + r1->len > r2->addr + r2->len)
2804 return 1;
2805 return 0;
2806}
2807
Paolo Bonzinia343c9b2013-07-16 13:03:29 +02002808static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2809{
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02002810 return kvm_io_bus_cmp(p1, p2);
Paolo Bonzinia343c9b2013-07-16 13:03:29 +02002811}
2812
Geoff Levand39369f72013-04-05 19:20:30 +00002813static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
Sasha Levin743eeb02011-07-27 16:00:48 +03002814 gpa_t addr, int len)
2815{
Sasha Levin743eeb02011-07-27 16:00:48 +03002816 bus->range[bus->dev_count++] = (struct kvm_io_range) {
2817 .addr = addr,
2818 .len = len,
2819 .dev = dev,
2820 };
2821
2822 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2823 kvm_io_bus_sort_cmp, NULL);
2824
2825 return 0;
2826}
2827
Geoff Levand39369f72013-04-05 19:20:30 +00002828static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
Sasha Levin743eeb02011-07-27 16:00:48 +03002829 gpa_t addr, int len)
2830{
2831 struct kvm_io_range *range, key;
2832 int off;
2833
2834 key = (struct kvm_io_range) {
2835 .addr = addr,
2836 .len = len,
2837 };
2838
2839 range = bsearch(&key, bus->range, bus->dev_count,
2840 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2841 if (range == NULL)
2842 return -ENOENT;
2843
2844 off = range - bus->range;
2845
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02002846 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
Sasha Levin743eeb02011-07-27 16:00:48 +03002847 off--;
2848
2849 return off;
2850}
2851
Cornelia Huck126a5af2013-07-03 16:30:53 +02002852static int __kvm_io_bus_write(struct kvm_io_bus *bus,
2853 struct kvm_io_range *range, const void *val)
2854{
2855 int idx;
2856
2857 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2858 if (idx < 0)
2859 return -EOPNOTSUPP;
2860
2861 while (idx < bus->dev_count &&
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02002862 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
Cornelia Huck126a5af2013-07-03 16:30:53 +02002863 if (!kvm_iodevice_write(bus->range[idx].dev, range->addr,
2864 range->len, val))
2865 return idx;
2866 idx++;
2867 }
2868
2869 return -EOPNOTSUPP;
2870}
2871
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002872/* kvm_io_bus_write - called under kvm->slots_lock */
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002873int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002874 int len, const void *val)
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002875{
Cornelia Huck126a5af2013-07-03 16:30:53 +02002876 struct kvm_io_bus *bus;
2877 struct kvm_io_range range;
2878 int r;
2879
2880 range = (struct kvm_io_range) {
2881 .addr = addr,
2882 .len = len,
2883 };
2884
2885 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2886 r = __kvm_io_bus_write(bus, &range, val);
2887 return r < 0 ? r : 0;
2888}
2889
2890/* kvm_io_bus_write_cookie - called under kvm->slots_lock */
2891int kvm_io_bus_write_cookie(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2892 int len, const void *val, long cookie)
2893{
Lai Jiangshan90d83dc2010-04-19 17:41:23 +08002894 struct kvm_io_bus *bus;
Sasha Levin743eeb02011-07-27 16:00:48 +03002895 struct kvm_io_range range;
2896
2897 range = (struct kvm_io_range) {
2898 .addr = addr,
2899 .len = len,
2900 };
Lai Jiangshan90d83dc2010-04-19 17:41:23 +08002901
2902 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
Cornelia Huck126a5af2013-07-03 16:30:53 +02002903
2904 /* First try the device referenced by cookie. */
2905 if ((cookie >= 0) && (cookie < bus->dev_count) &&
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02002906 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
Cornelia Huck126a5af2013-07-03 16:30:53 +02002907 if (!kvm_iodevice_write(bus->range[cookie].dev, addr, len,
2908 val))
2909 return cookie;
2910
2911 /*
2912 * cookie contained garbage; fall back to search and return the
2913 * correct cookie value.
2914 */
2915 return __kvm_io_bus_write(bus, &range, val);
2916}
2917
2918static int __kvm_io_bus_read(struct kvm_io_bus *bus, struct kvm_io_range *range,
2919 void *val)
2920{
2921 int idx;
2922
2923 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
Sasha Levin743eeb02011-07-27 16:00:48 +03002924 if (idx < 0)
2925 return -EOPNOTSUPP;
2926
2927 while (idx < bus->dev_count &&
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02002928 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
Cornelia Huck126a5af2013-07-03 16:30:53 +02002929 if (!kvm_iodevice_read(bus->range[idx].dev, range->addr,
2930 range->len, val))
2931 return idx;
Sasha Levin743eeb02011-07-27 16:00:48 +03002932 idx++;
2933 }
2934
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002935 return -EOPNOTSUPP;
2936}
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002937
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002938/* kvm_io_bus_read - called under kvm->slots_lock */
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002939int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2940 int len, void *val)
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002941{
Cornelia Huck126a5af2013-07-03 16:30:53 +02002942 struct kvm_io_bus *bus;
2943 struct kvm_io_range range;
2944 int r;
2945
2946 range = (struct kvm_io_range) {
2947 .addr = addr,
2948 .len = len,
2949 };
2950
2951 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2952 r = __kvm_io_bus_read(bus, &range, val);
2953 return r < 0 ? r : 0;
2954}
2955
2956/* kvm_io_bus_read_cookie - called under kvm->slots_lock */
2957int kvm_io_bus_read_cookie(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2958 int len, void *val, long cookie)
2959{
Lai Jiangshan90d83dc2010-04-19 17:41:23 +08002960 struct kvm_io_bus *bus;
Sasha Levin743eeb02011-07-27 16:00:48 +03002961 struct kvm_io_range range;
2962
2963 range = (struct kvm_io_range) {
2964 .addr = addr,
2965 .len = len,
2966 };
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002967
Lai Jiangshan90d83dc2010-04-19 17:41:23 +08002968 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
Sasha Levin743eeb02011-07-27 16:00:48 +03002969
Cornelia Huck126a5af2013-07-03 16:30:53 +02002970 /* First try the device referenced by cookie. */
2971 if ((cookie >= 0) && (cookie < bus->dev_count) &&
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02002972 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
Cornelia Huck126a5af2013-07-03 16:30:53 +02002973 if (!kvm_iodevice_read(bus->range[cookie].dev, addr, len,
2974 val))
2975 return cookie;
Sasha Levin743eeb02011-07-27 16:00:48 +03002976
Cornelia Huck126a5af2013-07-03 16:30:53 +02002977 /*
2978 * cookie contained garbage; fall back to search and return the
2979 * correct cookie value.
2980 */
2981 return __kvm_io_bus_read(bus, &range, val);
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002982}
2983
Marcelo Tosatti79fac952009-12-23 14:35:26 -02002984/* Caller must hold slots_lock. */
Sasha Levin743eeb02011-07-27 16:00:48 +03002985int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2986 int len, struct kvm_io_device *dev)
Michael S. Tsirkin6c474692009-06-29 22:24:26 +03002987{
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002988 struct kvm_io_bus *new_bus, *bus;
Gregory Haskins090b7af2009-07-07 17:08:44 -04002989
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002990 bus = kvm->buses[bus_idx];
Amos Kong6ea34c92013-05-25 06:44:15 +08002991 /* exclude ioeventfd which is limited by maximum fd */
2992 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
Gregory Haskins090b7af2009-07-07 17:08:44 -04002993 return -ENOSPC;
2994
Amos Konga13007162012-03-09 12:17:32 +08002995 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2996 sizeof(struct kvm_io_range)), GFP_KERNEL);
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002997 if (!new_bus)
2998 return -ENOMEM;
Amos Konga13007162012-03-09 12:17:32 +08002999 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
3000 sizeof(struct kvm_io_range)));
Sasha Levin743eeb02011-07-27 16:00:48 +03003001 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003002 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3003 synchronize_srcu_expedited(&kvm->srcu);
3004 kfree(bus);
Gregory Haskins090b7af2009-07-07 17:08:44 -04003005
3006 return 0;
3007}
3008
Marcelo Tosatti79fac952009-12-23 14:35:26 -02003009/* Caller must hold slots_lock. */
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003010int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3011 struct kvm_io_device *dev)
Gregory Haskins090b7af2009-07-07 17:08:44 -04003012{
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003013 int i, r;
3014 struct kvm_io_bus *new_bus, *bus;
Michael S. Tsirkin6c474692009-06-29 22:24:26 +03003015
Sasha Levincdfca7b2011-12-04 19:36:28 +02003016 bus = kvm->buses[bus_idx];
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003017 r = -ENOENT;
Amos Konga13007162012-03-09 12:17:32 +08003018 for (i = 0; i < bus->dev_count; i++)
3019 if (bus->range[i].dev == dev) {
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003020 r = 0;
Gregory Haskins090b7af2009-07-07 17:08:44 -04003021 break;
3022 }
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003023
Amos Konga13007162012-03-09 12:17:32 +08003024 if (r)
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003025 return r;
Amos Konga13007162012-03-09 12:17:32 +08003026
3027 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3028 sizeof(struct kvm_io_range)), GFP_KERNEL);
3029 if (!new_bus)
3030 return -ENOMEM;
3031
3032 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3033 new_bus->dev_count--;
3034 memcpy(new_bus->range + i, bus->range + i + 1,
3035 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003036
3037 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3038 synchronize_srcu_expedited(&kvm->srcu);
3039 kfree(bus);
3040 return r;
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04003041}
3042
Avi Kivity774c47f2007-02-12 00:54:47 -08003043static struct notifier_block kvm_cpu_notifier = {
3044 .notifier_call = kvm_cpu_hotplug,
Avi Kivity774c47f2007-02-12 00:54:47 -08003045};
3046
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003047static int vm_stat_get(void *_offset, u64 *val)
Avi Kivityba1389b2007-11-18 16:24:12 +02003048{
3049 unsigned offset = (long)_offset;
Avi Kivityba1389b2007-11-18 16:24:12 +02003050 struct kvm *kvm;
3051
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003052 *val = 0;
Jan Kiszkae935b832011-02-08 12:55:33 +01003053 raw_spin_lock(&kvm_lock);
Avi Kivityba1389b2007-11-18 16:24:12 +02003054 list_for_each_entry(kvm, &vm_list, vm_list)
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003055 *val += *(u32 *)((void *)kvm + offset);
Jan Kiszkae935b832011-02-08 12:55:33 +01003056 raw_spin_unlock(&kvm_lock);
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003057 return 0;
Avi Kivityba1389b2007-11-18 16:24:12 +02003058}
3059
3060DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3061
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003062static int vcpu_stat_get(void *_offset, u64 *val)
Avi Kivity1165f5f2007-04-19 17:27:43 +03003063{
3064 unsigned offset = (long)_offset;
Avi Kivity1165f5f2007-04-19 17:27:43 +03003065 struct kvm *kvm;
3066 struct kvm_vcpu *vcpu;
3067 int i;
3068
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003069 *val = 0;
Jan Kiszkae935b832011-02-08 12:55:33 +01003070 raw_spin_lock(&kvm_lock);
Avi Kivity1165f5f2007-04-19 17:27:43 +03003071 list_for_each_entry(kvm, &vm_list, vm_list)
Gleb Natapov988a2ca2009-06-09 15:56:29 +03003072 kvm_for_each_vcpu(i, vcpu, kvm)
3073 *val += *(u32 *)((void *)vcpu + offset);
3074
Jan Kiszkae935b832011-02-08 12:55:33 +01003075 raw_spin_unlock(&kvm_lock);
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003076 return 0;
Avi Kivity1165f5f2007-04-19 17:27:43 +03003077}
3078
Avi Kivityba1389b2007-11-18 16:24:12 +02003079DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3080
Alexey Dobriyan828c0952009-10-01 15:43:56 -07003081static const struct file_operations *stat_fops[] = {
Avi Kivityba1389b2007-11-18 16:24:12 +02003082 [KVM_STAT_VCPU] = &vcpu_stat_fops,
3083 [KVM_STAT_VM] = &vm_stat_fops,
3084};
Avi Kivity1165f5f2007-04-19 17:27:43 +03003085
Hamo4f69b682011-12-15 14:23:16 +08003086static int kvm_init_debug(void)
Avi Kivity6aa8b732006-12-10 02:21:36 -08003087{
Hamo4f69b682011-12-15 14:23:16 +08003088 int r = -EFAULT;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003089 struct kvm_stats_debugfs_item *p;
3090
Hollis Blanchard76f7c872008-04-15 16:05:42 -05003091 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
Hamo4f69b682011-12-15 14:23:16 +08003092 if (kvm_debugfs_dir == NULL)
3093 goto out;
3094
3095 for (p = debugfs_entries; p->name; ++p) {
Hollis Blanchard76f7c872008-04-15 16:05:42 -05003096 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
Avi Kivity1165f5f2007-04-19 17:27:43 +03003097 (void *)(long)p->offset,
Avi Kivityba1389b2007-11-18 16:24:12 +02003098 stat_fops[p->kind]);
Hamo4f69b682011-12-15 14:23:16 +08003099 if (p->dentry == NULL)
3100 goto out_dir;
3101 }
3102
3103 return 0;
3104
3105out_dir:
3106 debugfs_remove_recursive(kvm_debugfs_dir);
3107out:
3108 return r;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003109}
3110
3111static void kvm_exit_debug(void)
3112{
3113 struct kvm_stats_debugfs_item *p;
3114
3115 for (p = debugfs_entries; p->name; ++p)
3116 debugfs_remove(p->dentry);
Hollis Blanchard76f7c872008-04-15 16:05:42 -05003117 debugfs_remove(kvm_debugfs_dir);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003118}
3119
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003120static int kvm_suspend(void)
Avi Kivity59ae6c62007-02-12 00:54:48 -08003121{
Alexander Graf10474ae2009-09-15 11:37:46 +02003122 if (kvm_usage_count)
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09003123 hardware_disable_nolock(NULL);
Avi Kivity59ae6c62007-02-12 00:54:48 -08003124 return 0;
3125}
3126
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003127static void kvm_resume(void)
Avi Kivity59ae6c62007-02-12 00:54:48 -08003128{
Zachary Amsdenca84d1a2010-08-19 22:07:28 -10003129 if (kvm_usage_count) {
Jan Kiszkae935b832011-02-08 12:55:33 +01003130 WARN_ON(raw_spin_is_locked(&kvm_lock));
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09003131 hardware_enable_nolock(NULL);
Zachary Amsdenca84d1a2010-08-19 22:07:28 -10003132 }
Avi Kivity59ae6c62007-02-12 00:54:48 -08003133}
3134
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003135static struct syscore_ops kvm_syscore_ops = {
Avi Kivity59ae6c62007-02-12 00:54:48 -08003136 .suspend = kvm_suspend,
3137 .resume = kvm_resume,
3138};
3139
Avi Kivity15ad7142007-07-11 18:17:21 +03003140static inline
3141struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3142{
3143 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3144}
3145
3146static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3147{
3148 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
Raghavendra K T3a08a8f2013-03-04 23:32:07 +05303149 if (vcpu->preempted)
3150 vcpu->preempted = false;
Avi Kivity15ad7142007-07-11 18:17:21 +03003151
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003152 kvm_arch_vcpu_load(vcpu, cpu);
Avi Kivity15ad7142007-07-11 18:17:21 +03003153}
3154
3155static void kvm_sched_out(struct preempt_notifier *pn,
3156 struct task_struct *next)
3157{
3158 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3159
Raghavendra K T3a08a8f2013-03-04 23:32:07 +05303160 if (current->state == TASK_RUNNING)
3161 vcpu->preempted = true;
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003162 kvm_arch_vcpu_put(vcpu);
Avi Kivity15ad7142007-07-11 18:17:21 +03003163}
3164
Avi Kivity0ee75be2010-04-28 15:39:01 +03003165int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
Rusty Russellc16f8622007-07-30 21:12:19 +10003166 struct module *module)
Avi Kivity6aa8b732006-12-10 02:21:36 -08003167{
3168 int r;
Yang, Sheng002c7f72007-07-31 14:23:01 +03003169 int cpu;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003170
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08003171 r = kvm_arch_init(opaque);
3172 if (r)
Zhang Xiantaod23087842007-11-29 15:35:39 +08003173 goto out_fail;
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08003174
Asias He7dac16c2013-05-08 10:57:29 +08003175 /*
3176 * kvm_arch_init makes sure there's at most one caller
3177 * for architectures that support multiple implementations,
3178 * like intel and amd on x86.
3179 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3180 * conflicts in case kvm is already setup for another implementation.
3181 */
3182 r = kvm_irqfd_init();
3183 if (r)
3184 goto out_irqfd;
3185
Avi Kivity8437a612009-06-06 14:52:35 -07003186 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
Rusty Russell7f59f492008-12-07 21:25:45 +10303187 r = -ENOMEM;
3188 goto out_free_0;
3189 }
3190
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003191 r = kvm_arch_hardware_setup();
Avi Kivity6aa8b732006-12-10 02:21:36 -08003192 if (r < 0)
Rusty Russell7f59f492008-12-07 21:25:45 +10303193 goto out_free_0a;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003194
Yang, Sheng002c7f72007-07-31 14:23:01 +03003195 for_each_online_cpu(cpu) {
3196 smp_call_function_single(cpu,
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003197 kvm_arch_check_processor_compat,
Jens Axboe8691e5a2008-06-06 11:18:06 +02003198 &r, 1);
Yang, Sheng002c7f72007-07-31 14:23:01 +03003199 if (r < 0)
Zhang Xiantaod23087842007-11-29 15:35:39 +08003200 goto out_free_1;
Yang, Sheng002c7f72007-07-31 14:23:01 +03003201 }
3202
Avi Kivity774c47f2007-02-12 00:54:47 -08003203 r = register_cpu_notifier(&kvm_cpu_notifier);
3204 if (r)
Zhang Xiantaod23087842007-11-29 15:35:39 +08003205 goto out_free_2;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003206 register_reboot_notifier(&kvm_reboot_notifier);
3207
Rusty Russellc16f8622007-07-30 21:12:19 +10003208 /* A kmem cache lets us meet the alignment requirements of fx_save. */
Avi Kivity0ee75be2010-04-28 15:39:01 +03003209 if (!vcpu_align)
3210 vcpu_align = __alignof__(struct kvm_vcpu);
3211 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
Joe Perches56919c52007-11-12 20:06:51 -08003212 0, NULL);
Rusty Russellc16f8622007-07-30 21:12:19 +10003213 if (!kvm_vcpu_cache) {
3214 r = -ENOMEM;
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003215 goto out_free_3;
Rusty Russellc16f8622007-07-30 21:12:19 +10003216 }
3217
Gleb Natapovaf585b92010-10-14 11:22:46 +02003218 r = kvm_async_pf_init();
3219 if (r)
3220 goto out_free;
3221
Avi Kivity6aa8b732006-12-10 02:21:36 -08003222 kvm_chardev_ops.owner = module;
Christian Borntraeger3d3aab12008-12-02 11:17:32 +01003223 kvm_vm_fops.owner = module;
3224 kvm_vcpu_fops.owner = module;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003225
3226 r = misc_register(&kvm_dev);
3227 if (r) {
Mike Dayd77c26f2007-10-08 09:02:08 -04003228 printk(KERN_ERR "kvm: misc device register failed\n");
Gleb Natapovaf585b92010-10-14 11:22:46 +02003229 goto out_unreg;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003230 }
3231
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003232 register_syscore_ops(&kvm_syscore_ops);
3233
Avi Kivity15ad7142007-07-11 18:17:21 +03003234 kvm_preempt_ops.sched_in = kvm_sched_in;
3235 kvm_preempt_ops.sched_out = kvm_sched_out;
3236
Hamo4f69b682011-12-15 14:23:16 +08003237 r = kvm_init_debug();
3238 if (r) {
3239 printk(KERN_ERR "kvm: create debugfs files failed\n");
3240 goto out_undebugfs;
3241 }
Darrick J. Wong0ea4ed82009-10-14 16:21:00 -07003242
Avi Kivityc7addb92007-09-16 18:58:32 +02003243 return 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003244
Hamo4f69b682011-12-15 14:23:16 +08003245out_undebugfs:
3246 unregister_syscore_ops(&kvm_syscore_ops);
Wei Yongjunafc2f792013-05-05 20:03:40 +08003247 misc_deregister(&kvm_dev);
Gleb Natapovaf585b92010-10-14 11:22:46 +02003248out_unreg:
3249 kvm_async_pf_deinit();
Avi Kivity6aa8b732006-12-10 02:21:36 -08003250out_free:
Rusty Russellc16f8622007-07-30 21:12:19 +10003251 kmem_cache_destroy(kvm_vcpu_cache);
Zhang Xiantaod23087842007-11-29 15:35:39 +08003252out_free_3:
Avi Kivity6aa8b732006-12-10 02:21:36 -08003253 unregister_reboot_notifier(&kvm_reboot_notifier);
Avi Kivity774c47f2007-02-12 00:54:47 -08003254 unregister_cpu_notifier(&kvm_cpu_notifier);
Zhang Xiantaod23087842007-11-29 15:35:39 +08003255out_free_2:
Zhang Xiantaod23087842007-11-29 15:35:39 +08003256out_free_1:
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003257 kvm_arch_hardware_unsetup();
Rusty Russell7f59f492008-12-07 21:25:45 +10303258out_free_0a:
3259 free_cpumask_var(cpus_hardware_enabled);
Zhang Xiantaod23087842007-11-29 15:35:39 +08003260out_free_0:
Cornelia Hucka0f155e2013-02-28 12:33:18 +01003261 kvm_irqfd_exit();
3262out_irqfd:
Asias He7dac16c2013-05-08 10:57:29 +08003263 kvm_arch_exit();
3264out_fail:
Avi Kivity6aa8b732006-12-10 02:21:36 -08003265 return r;
3266}
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08003267EXPORT_SYMBOL_GPL(kvm_init);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003268
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08003269void kvm_exit(void)
Avi Kivity6aa8b732006-12-10 02:21:36 -08003270{
Darrick J. Wong0ea4ed82009-10-14 16:21:00 -07003271 kvm_exit_debug();
Avi Kivity6aa8b732006-12-10 02:21:36 -08003272 misc_deregister(&kvm_dev);
Rusty Russellc16f8622007-07-30 21:12:19 +10003273 kmem_cache_destroy(kvm_vcpu_cache);
Gleb Natapovaf585b92010-10-14 11:22:46 +02003274 kvm_async_pf_deinit();
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003275 unregister_syscore_ops(&kvm_syscore_ops);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003276 unregister_reboot_notifier(&kvm_reboot_notifier);
Avi Kivity59ae6c62007-02-12 00:54:48 -08003277 unregister_cpu_notifier(&kvm_cpu_notifier);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09003278 on_each_cpu(hardware_disable_nolock, NULL, 1);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003279 kvm_arch_hardware_unsetup();
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08003280 kvm_arch_exit();
Cornelia Hucka0f155e2013-02-28 12:33:18 +01003281 kvm_irqfd_exit();
Rusty Russell7f59f492008-12-07 21:25:45 +10303282 free_cpumask_var(cpus_hardware_enabled);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003283}
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08003284EXPORT_SYMBOL_GPL(kvm_exit);