blob: 662f34c3287e59e8f0707d80b0c88396d402c8c0 [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
Paolo Bonzini2f303b72013-09-25 13:53:07 +020073DEFINE_SPINLOCK(kvm_lock);
Paolo Bonzini4a937f92013-09-10 12:58:35 +020074static DEFINE_RAW_SPINLOCK(kvm_count_lock);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +080075LIST_HEAD(vm_list);
Avi Kivity133de902007-02-12 00:54:44 -080076
Rusty Russell7f59f492008-12-07 21:25:45 +103077static cpumask_var_t cpus_hardware_enabled;
Alexander Graf10474ae2009-09-15 11:37:46 +020078static int kvm_usage_count = 0;
79static atomic_t hardware_enable_failed;
Avi Kivity1b6c0162007-05-24 13:03:52 +030080
Rusty Russellc16f8622007-07-30 21:12:19 +100081struct kmem_cache *kvm_vcpu_cache;
82EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
Avi Kivity1165f5f2007-04-19 17:27:43 +030083
Avi Kivity15ad7142007-07-11 18:17:21 +030084static __read_mostly struct preempt_ops kvm_preempt_ops;
85
Hollis Blanchard76f7c872008-04-15 16:05:42 -050086struct dentry *kvm_debugfs_dir;
Avi Kivity6aa8b732006-12-10 02:21:36 -080087
Avi Kivitybccf2152007-02-21 18:04:26 +020088static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
89 unsigned long arg);
Alexander Graf1dda6062011-06-08 02:45:37 +020090#ifdef CONFIG_COMPAT
91static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
92 unsigned long arg);
93#endif
Alexander Graf10474ae2009-09-15 11:37:46 +020094static int hardware_enable_all(void);
95static void hardware_disable_all(void);
Avi Kivitybccf2152007-02-21 18:04:26 +020096
Marcelo Tosattie93f8a02009-12-23 14:35:24 -020097static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
98
Avi Kivityb7c41452010-12-02 17:52:50 +020099bool kvm_rebooting;
100EXPORT_SYMBOL_GPL(kvm_rebooting);
Avi Kivity4ecac3f2008-05-13 13:23:38 +0300101
Marcelo Tosatti54dee992009-06-11 12:07:44 -0300102static bool largepages_enabled = true;
103
Xiao Guangronga2766322012-07-26 11:58:59 +0800104bool kvm_is_mmio_pfn(pfn_t pfn)
Ben-Ami Yassourcbff90a2008-07-28 19:26:24 +0300105{
Andrea Arcangeli11feeb42013-07-25 03:04:38 +0200106 if (pfn_valid(pfn))
107 return PageReserved(pfn_to_page(pfn));
Ben-Ami Yassourcbff90a2008-07-28 19:26:24 +0300108
109 return true;
110}
111
Avi Kivity6aa8b732006-12-10 02:21:36 -0800112/*
113 * Switches to specified vcpu, until a matching vcpu_put()
114 */
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +0300115int vcpu_load(struct kvm_vcpu *vcpu)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800116{
Avi Kivity15ad7142007-07-11 18:17:21 +0300117 int cpu;
118
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +0300119 if (mutex_lock_killable(&vcpu->mutex))
120 return -EINTR;
Rik van Riel34bb10b2011-02-01 09:52:41 -0500121 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
122 /* The thread running this VCPU changed. */
123 struct pid *oldpid = vcpu->pid;
124 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
125 rcu_assign_pointer(vcpu->pid, newpid);
126 synchronize_rcu();
127 put_pid(oldpid);
128 }
Avi Kivity15ad7142007-07-11 18:17:21 +0300129 cpu = get_cpu();
130 preempt_notifier_register(&vcpu->preempt_notifier);
Carsten Otte313a3dc2007-10-11 19:16:52 +0200131 kvm_arch_vcpu_load(vcpu, cpu);
Avi Kivity15ad7142007-07-11 18:17:21 +0300132 put_cpu();
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +0300133 return 0;
Avi Kivitybccf2152007-02-21 18:04:26 +0200134}
135
Carsten Otte313a3dc2007-10-11 19:16:52 +0200136void vcpu_put(struct kvm_vcpu *vcpu)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800137{
Avi Kivity15ad7142007-07-11 18:17:21 +0300138 preempt_disable();
Carsten Otte313a3dc2007-10-11 19:16:52 +0200139 kvm_arch_vcpu_put(vcpu);
Avi Kivity15ad7142007-07-11 18:17:21 +0300140 preempt_notifier_unregister(&vcpu->preempt_notifier);
141 preempt_enable();
Avi Kivity6aa8b732006-12-10 02:21:36 -0800142 mutex_unlock(&vcpu->mutex);
143}
144
Avi Kivityd9e368d2007-06-07 19:18:30 +0300145static void ack_flush(void *_completed)
146{
Avi Kivityd9e368d2007-06-07 19:18:30 +0300147}
148
Rusty Russell49846892008-12-08 20:26:24 +1030149static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
Avi Kivityd9e368d2007-06-07 19:18:30 +0300150{
Avi Kivity597a5f52008-07-20 14:24:22 +0300151 int i, cpu, me;
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030152 cpumask_var_t cpus;
153 bool called = true;
Avi Kivityd9e368d2007-06-07 19:18:30 +0300154 struct kvm_vcpu *vcpu;
Avi Kivityd9e368d2007-06-07 19:18:30 +0300155
Li Zefan79f55992009-06-15 14:58:26 +0800156 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030157
Xiao Guangrong3cba4132011-01-12 15:41:22 +0800158 me = get_cpu();
Gleb Natapov988a2ca2009-06-09 15:56:29 +0300159 kvm_for_each_vcpu(i, vcpu, kvm) {
Xiao Guangrong3cba4132011-01-12 15:41:22 +0800160 kvm_make_request(req, vcpu);
Avi Kivityd9e368d2007-06-07 19:18:30 +0300161 cpu = vcpu->cpu;
Xiao Guangrong6b7e2d02011-01-12 15:40:31 +0800162
163 /* Set ->requests bit before we read ->mode */
164 smp_mb();
165
166 if (cpus != NULL && cpu != -1 && cpu != me &&
167 kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030168 cpumask_set_cpu(cpu, cpus);
Avi Kivityd9e368d2007-06-07 19:18:30 +0300169 }
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030170 if (unlikely(cpus == NULL))
171 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
172 else if (!cpumask_empty(cpus))
173 smp_call_function_many(cpus, ack_flush, NULL, 1);
174 else
175 called = false;
Xiao Guangrong3cba4132011-01-12 15:41:22 +0800176 put_cpu();
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030177 free_cpumask_var(cpus);
Rusty Russell49846892008-12-08 20:26:24 +1030178 return called;
179}
180
181void kvm_flush_remote_tlbs(struct kvm *kvm)
182{
Alex Shibec87d62012-03-04 13:13:38 +0800183 long dirty_count = kvm->tlbs_dirty;
Xiao Guangronga4ee1ca2010-11-23 11:13:00 +0800184
185 smp_mb();
Rusty Russell49846892008-12-08 20:26:24 +1030186 if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
187 ++kvm->stat.remote_tlb_flush;
Xiao Guangronga4ee1ca2010-11-23 11:13:00 +0800188 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
Avi Kivityd9e368d2007-06-07 19:18:30 +0300189}
Aneesh Kumar K.V2ba9f0d2013-10-07 22:17:59 +0530190EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
Avi Kivityd9e368d2007-06-07 19:18:30 +0300191
Marcelo Tosatti2e53d632008-02-20 14:47:24 -0500192void kvm_reload_remote_mmus(struct kvm *kvm)
193{
Rusty Russell49846892008-12-08 20:26:24 +1030194 make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
Marcelo Tosatti2e53d632008-02-20 14:47:24 -0500195}
196
Marcelo Tosattid8281992012-11-27 23:29:01 -0200197void kvm_make_mclock_inprogress_request(struct kvm *kvm)
198{
199 make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
200}
201
Yang Zhang3d81bc72013-04-11 19:25:13 +0800202void kvm_make_scan_ioapic_request(struct kvm *kvm)
Yang Zhangc7c9c562013-01-25 10:18:51 +0800203{
Yang Zhang3d81bc72013-04-11 19:25:13 +0800204 make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
Yang Zhangc7c9c562013-01-25 10:18:51 +0800205}
206
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000207int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
208{
209 struct page *page;
210 int r;
211
212 mutex_init(&vcpu->mutex);
213 vcpu->cpu = -1;
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000214 vcpu->kvm = kvm;
215 vcpu->vcpu_id = id;
Rik van Riel34bb10b2011-02-01 09:52:41 -0500216 vcpu->pid = NULL;
Eddie Dongb6958ce2007-07-18 12:15:21 +0300217 init_waitqueue_head(&vcpu->wq);
Gleb Natapovaf585b92010-10-14 11:22:46 +0200218 kvm_async_pf_vcpu_init(vcpu);
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000219
220 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
221 if (!page) {
222 r = -ENOMEM;
223 goto fail;
224 }
225 vcpu->run = page_address(page);
226
Raghavendra K T4c088492012-07-18 19:07:46 +0530227 kvm_vcpu_set_in_spin_loop(vcpu, false);
228 kvm_vcpu_set_dy_eligible(vcpu, false);
Raghavendra K T3a08a8f2013-03-04 23:32:07 +0530229 vcpu->preempted = false;
Raghavendra K T4c088492012-07-18 19:07:46 +0530230
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +0800231 r = kvm_arch_vcpu_init(vcpu);
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000232 if (r < 0)
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +0800233 goto fail_free_run;
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000234 return 0;
235
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000236fail_free_run:
237 free_page((unsigned long)vcpu->run);
238fail:
Rusty Russell76fafa52007-10-08 10:50:48 +1000239 return r;
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000240}
241EXPORT_SYMBOL_GPL(kvm_vcpu_init);
242
243void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
244{
Rik van Riel34bb10b2011-02-01 09:52:41 -0500245 put_pid(vcpu->pid);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +0800246 kvm_arch_vcpu_uninit(vcpu);
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000247 free_page((unsigned long)vcpu->run);
248}
249EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
250
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200251#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
252static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
253{
254 return container_of(mn, struct kvm, mmu_notifier);
255}
256
257static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
258 struct mm_struct *mm,
259 unsigned long address)
260{
261 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200262 int need_tlb_flush, idx;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200263
264 /*
265 * When ->invalidate_page runs, the linux pte has been zapped
266 * already but the page is still allocated until
267 * ->invalidate_page returns. So if we increase the sequence
268 * here the kvm page fault will notice if the spte can't be
269 * established because the page is going to be freed. If
270 * instead the kvm page fault establishes the spte before
271 * ->invalidate_page runs, kvm_unmap_hva will release it
272 * before returning.
273 *
274 * The sequence increase only need to be seen at spin_unlock
275 * time, and not at spin_lock time.
276 *
277 * Increasing the sequence after the spin_unlock would be
278 * unsafe because the kvm page fault could then establish the
279 * pte after kvm_unmap_hva returned, without noticing the page
280 * is going to be freed.
281 */
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200282 idx = srcu_read_lock(&kvm->srcu);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200283 spin_lock(&kvm->mmu_lock);
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900284
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200285 kvm->mmu_notifier_seq++;
Xiao Guangronga4ee1ca2010-11-23 11:13:00 +0800286 need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200287 /* we've to flush the tlb before the pages can be freed */
288 if (need_tlb_flush)
289 kvm_flush_remote_tlbs(kvm);
290
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900291 spin_unlock(&kvm->mmu_lock);
292 srcu_read_unlock(&kvm->srcu, idx);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200293}
294
Izik Eidus3da0dd42009-09-23 21:47:18 +0300295static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
296 struct mm_struct *mm,
297 unsigned long address,
298 pte_t pte)
299{
300 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200301 int idx;
Izik Eidus3da0dd42009-09-23 21:47:18 +0300302
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200303 idx = srcu_read_lock(&kvm->srcu);
Izik Eidus3da0dd42009-09-23 21:47:18 +0300304 spin_lock(&kvm->mmu_lock);
305 kvm->mmu_notifier_seq++;
306 kvm_set_spte_hva(kvm, address, pte);
307 spin_unlock(&kvm->mmu_lock);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200308 srcu_read_unlock(&kvm->srcu, idx);
Izik Eidus3da0dd42009-09-23 21:47:18 +0300309}
310
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200311static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
312 struct mm_struct *mm,
313 unsigned long start,
314 unsigned long end)
315{
316 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200317 int need_tlb_flush = 0, idx;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200318
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200319 idx = srcu_read_lock(&kvm->srcu);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200320 spin_lock(&kvm->mmu_lock);
321 /*
322 * The count increase must become visible at unlock time as no
323 * spte can be established without taking the mmu_lock and
324 * count is also read inside the mmu_lock critical section.
325 */
326 kvm->mmu_notifier_count++;
Takuya Yoshikawab3ae2092012-07-02 17:56:33 +0900327 need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
Xiao Guangronga4ee1ca2010-11-23 11:13:00 +0800328 need_tlb_flush |= kvm->tlbs_dirty;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200329 /* we've to flush the tlb before the pages can be freed */
330 if (need_tlb_flush)
331 kvm_flush_remote_tlbs(kvm);
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900332
333 spin_unlock(&kvm->mmu_lock);
334 srcu_read_unlock(&kvm->srcu, idx);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200335}
336
337static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
338 struct mm_struct *mm,
339 unsigned long start,
340 unsigned long end)
341{
342 struct kvm *kvm = mmu_notifier_to_kvm(mn);
343
344 spin_lock(&kvm->mmu_lock);
345 /*
346 * This sequence increase will notify the kvm page fault that
347 * the page that is going to be mapped in the spte could have
348 * been freed.
349 */
350 kvm->mmu_notifier_seq++;
Paul Mackerrasa355aa52011-12-12 12:37:21 +0000351 smp_wmb();
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200352 /*
353 * The above sequence increase must be visible before the
Paul Mackerrasa355aa52011-12-12 12:37:21 +0000354 * below count decrease, which is ensured by the smp_wmb above
355 * in conjunction with the smp_rmb in mmu_notifier_retry().
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200356 */
357 kvm->mmu_notifier_count--;
358 spin_unlock(&kvm->mmu_lock);
359
360 BUG_ON(kvm->mmu_notifier_count < 0);
361}
362
363static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
364 struct mm_struct *mm,
365 unsigned long address)
366{
367 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200368 int young, idx;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200369
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200370 idx = srcu_read_lock(&kvm->srcu);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200371 spin_lock(&kvm->mmu_lock);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200372
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900373 young = kvm_age_hva(kvm, address);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200374 if (young)
375 kvm_flush_remote_tlbs(kvm);
376
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900377 spin_unlock(&kvm->mmu_lock);
378 srcu_read_unlock(&kvm->srcu, idx);
379
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200380 return young;
381}
382
Andrea Arcangeli8ee53822011-01-13 15:47:10 -0800383static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
384 struct mm_struct *mm,
385 unsigned long address)
386{
387 struct kvm *kvm = mmu_notifier_to_kvm(mn);
388 int young, idx;
389
390 idx = srcu_read_lock(&kvm->srcu);
391 spin_lock(&kvm->mmu_lock);
392 young = kvm_test_age_hva(kvm, address);
393 spin_unlock(&kvm->mmu_lock);
394 srcu_read_unlock(&kvm->srcu, idx);
395
396 return young;
397}
398
Marcelo Tosatti85db06e2008-12-10 21:23:26 +0100399static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
400 struct mm_struct *mm)
401{
402 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Lai Jiangshaneda2bed2010-04-20 14:29:29 +0800403 int idx;
404
405 idx = srcu_read_lock(&kvm->srcu);
Marcelo Tosatti2df72e92012-08-24 15:54:57 -0300406 kvm_arch_flush_shadow_all(kvm);
Lai Jiangshaneda2bed2010-04-20 14:29:29 +0800407 srcu_read_unlock(&kvm->srcu, idx);
Marcelo Tosatti85db06e2008-12-10 21:23:26 +0100408}
409
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200410static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
411 .invalidate_page = kvm_mmu_notifier_invalidate_page,
412 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
413 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
414 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
Andrea Arcangeli8ee53822011-01-13 15:47:10 -0800415 .test_young = kvm_mmu_notifier_test_young,
Izik Eidus3da0dd42009-09-23 21:47:18 +0300416 .change_pte = kvm_mmu_notifier_change_pte,
Marcelo Tosatti85db06e2008-12-10 21:23:26 +0100417 .release = kvm_mmu_notifier_release,
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200418};
Avi Kivity4c07b0a2009-12-20 14:54:04 +0200419
420static int kvm_init_mmu_notifier(struct kvm *kvm)
421{
422 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
423 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
424}
425
426#else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
427
428static int kvm_init_mmu_notifier(struct kvm *kvm)
429{
430 return 0;
431}
432
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200433#endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
434
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800435static void kvm_init_memslots_id(struct kvm *kvm)
436{
437 int i;
438 struct kvm_memslots *slots = kvm->memslots;
439
440 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
Xiao Guangrongf85e2cb2011-11-24 17:41:54 +0800441 slots->id_to_index[i] = slots->memslots[i].id = i;
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800442}
443
Carsten Ottee08b9632012-01-04 10:25:20 +0100444static struct kvm *kvm_create_vm(unsigned long type)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800445{
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100446 int r, i;
447 struct kvm *kvm = kvm_arch_alloc_vm();
Avi Kivity6aa8b732006-12-10 02:21:36 -0800448
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100449 if (!kvm)
450 return ERR_PTR(-ENOMEM);
451
Carsten Ottee08b9632012-01-04 10:25:20 +0100452 r = kvm_arch_init_vm(kvm, type);
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100453 if (r)
454 goto out_err_nodisable;
Alexander Graf10474ae2009-09-15 11:37:46 +0200455
456 r = hardware_enable_all();
457 if (r)
458 goto out_err_nodisable;
459
Avi Kivity75858a82009-01-04 17:10:50 +0200460#ifdef CONFIG_HAVE_KVM_IRQCHIP
461 INIT_HLIST_HEAD(&kvm->mask_notifier_list);
Gleb Natapov136bdfe2009-08-24 11:54:23 +0300462 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
Avi Kivity75858a82009-01-04 17:10:50 +0200463#endif
Avi Kivity6aa8b732006-12-10 02:21:36 -0800464
Alex Williamson1e702d92012-12-10 10:33:32 -0700465 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
466
Marcelo Tosatti46a26bf2009-12-23 14:35:16 -0200467 r = -ENOMEM;
468 kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
469 if (!kvm->memslots)
Jan Kiszka57e7fbe2010-11-09 12:42:12 +0100470 goto out_err_nosrcu;
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800471 kvm_init_memslots_id(kvm);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200472 if (init_srcu_struct(&kvm->srcu))
Jan Kiszka57e7fbe2010-11-09 12:42:12 +0100473 goto out_err_nosrcu;
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200474 for (i = 0; i < KVM_NR_BUSES; i++) {
475 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
476 GFP_KERNEL);
Jan Kiszka57e7fbe2010-11-09 12:42:12 +0100477 if (!kvm->buses[i])
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200478 goto out_err;
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200479 }
Mike Waychison74b5c5b2011-06-03 13:04:53 -0700480
OGAWA Hirofumi85722cd2011-05-11 09:28:28 +0900481 spin_lock_init(&kvm->mmu_lock);
Avi Kivity6d4e4c42007-11-21 16:41:05 +0200482 kvm->mm = current->mm;
483 atomic_inc(&kvm->mm->mm_count);
Gregory Haskinsd34e6b12009-07-07 17:08:49 -0400484 kvm_eventfd_init(kvm);
Shaohua Li11ec2802007-07-23 14:51:37 +0800485 mutex_init(&kvm->lock);
Marcelo Tosatti60eead72009-06-04 15:08:23 -0300486 mutex_init(&kvm->irq_lock);
Marcelo Tosatti79fac952009-12-23 14:35:26 -0200487 mutex_init(&kvm->slots_lock);
Izik Eidusd39f13b2008-03-30 16:01:25 +0300488 atomic_set(&kvm->users_count, 1);
Scott Wood07f0a7b2013-04-25 14:11:23 +0000489 INIT_LIST_HEAD(&kvm->devices);
Mike Waychison74b5c5b2011-06-03 13:04:53 -0700490
491 r = kvm_init_mmu_notifier(kvm);
492 if (r)
493 goto out_err;
494
Paolo Bonzini2f303b72013-09-25 13:53:07 +0200495 spin_lock(&kvm_lock);
Rusty Russell5e58cfe2007-07-23 17:08:21 +1000496 list_add(&kvm->vm_list, &vm_list);
Paolo Bonzini2f303b72013-09-25 13:53:07 +0200497 spin_unlock(&kvm_lock);
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100498
Avi Kivityf17abe92007-02-21 19:28:04 +0200499 return kvm;
Alexander Graf10474ae2009-09-15 11:37:46 +0200500
501out_err:
Jan Kiszka57e7fbe2010-11-09 12:42:12 +0100502 cleanup_srcu_struct(&kvm->srcu);
503out_err_nosrcu:
Alexander Graf10474ae2009-09-15 11:37:46 +0200504 hardware_disable_all();
505out_err_nodisable:
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200506 for (i = 0; i < KVM_NR_BUSES; i++)
507 kfree(kvm->buses[i]);
Marcelo Tosatti46a26bf2009-12-23 14:35:16 -0200508 kfree(kvm->memslots);
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100509 kvm_arch_free_vm(kvm);
Alexander Graf10474ae2009-09-15 11:37:46 +0200510 return ERR_PTR(r);
Avi Kivityf17abe92007-02-21 19:28:04 +0200511}
512
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900513/*
514 * Avoid using vmalloc for a small buffer.
515 * Should not be used when the size is statically known.
516 */
Takuya Yoshikawac1a7b322012-05-20 13:15:07 +0900517void *kvm_kvzalloc(unsigned long size)
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900518{
519 if (size > PAGE_SIZE)
520 return vzalloc(size);
521 else
522 return kzalloc(size, GFP_KERNEL);
523}
524
Takuya Yoshikawac1a7b322012-05-20 13:15:07 +0900525void kvm_kvfree(const void *addr)
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900526{
527 if (is_vmalloc_addr(addr))
528 vfree(addr);
529 else
530 kfree(addr);
531}
532
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900533static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
534{
535 if (!memslot->dirty_bitmap)
536 return;
537
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900538 kvm_kvfree(memslot->dirty_bitmap);
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900539 memslot->dirty_bitmap = NULL;
540}
541
Avi Kivity6aa8b732006-12-10 02:21:36 -0800542/*
543 * Free any memory in @free but not in @dont.
544 */
Aneesh Kumar K.V55870272013-10-07 22:18:00 +0530545static void kvm_free_physmem_slot(struct kvm *kvm, struct kvm_memory_slot *free,
Avi Kivity6aa8b732006-12-10 02:21:36 -0800546 struct kvm_memory_slot *dont)
547{
Avi Kivity6aa8b732006-12-10 02:21:36 -0800548 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900549 kvm_destroy_dirty_bitmap(free);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800550
Aneesh Kumar K.V55870272013-10-07 22:18:00 +0530551 kvm_arch_free_memslot(kvm, free, dont);
Marcelo Tosatti05da4552008-02-23 11:44:30 -0300552
Avi Kivity6aa8b732006-12-10 02:21:36 -0800553 free->npages = 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800554}
555
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +0800556void kvm_free_physmem(struct kvm *kvm)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800557{
Marcelo Tosatti46a26bf2009-12-23 14:35:16 -0200558 struct kvm_memslots *slots = kvm->memslots;
Xiao Guangrongbe6ba0f2011-11-24 17:39:18 +0800559 struct kvm_memory_slot *memslot;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800560
Xiao Guangrongbe6ba0f2011-11-24 17:39:18 +0800561 kvm_for_each_memslot(memslot, slots)
Aneesh Kumar K.V55870272013-10-07 22:18:00 +0530562 kvm_free_physmem_slot(kvm, memslot, NULL);
Marcelo Tosatti46a26bf2009-12-23 14:35:16 -0200563
564 kfree(kvm->memslots);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800565}
566
Scott Wood07f0a7b2013-04-25 14:11:23 +0000567static void kvm_destroy_devices(struct kvm *kvm)
568{
569 struct list_head *node, *tmp;
570
571 list_for_each_safe(node, tmp, &kvm->devices) {
572 struct kvm_device *dev =
573 list_entry(node, struct kvm_device, vm_node);
574
575 list_del(node);
576 dev->ops->destroy(dev);
577 }
578}
579
Avi Kivityf17abe92007-02-21 19:28:04 +0200580static void kvm_destroy_vm(struct kvm *kvm)
581{
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200582 int i;
Avi Kivity6d4e4c42007-11-21 16:41:05 +0200583 struct mm_struct *mm = kvm->mm;
584
Sheng Yangad8ba2c2009-01-06 10:03:02 +0800585 kvm_arch_sync_events(kvm);
Paolo Bonzini2f303b72013-09-25 13:53:07 +0200586 spin_lock(&kvm_lock);
Avi Kivity133de902007-02-12 00:54:44 -0800587 list_del(&kvm->vm_list);
Paolo Bonzini2f303b72013-09-25 13:53:07 +0200588 spin_unlock(&kvm_lock);
Avi Kivity399ec802008-11-19 13:58:46 +0200589 kvm_free_irq_routing(kvm);
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200590 for (i = 0; i < KVM_NR_BUSES; i++)
591 kvm_io_bus_destroy(kvm->buses[i]);
Avi Kivity980da6c2009-12-20 15:13:43 +0200592 kvm_coalesced_mmio_free(kvm);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200593#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
594 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
Gleb Natapovf00be0c2009-03-19 12:20:36 +0200595#else
Marcelo Tosatti2df72e92012-08-24 15:54:57 -0300596 kvm_arch_flush_shadow_all(kvm);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200597#endif
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +0800598 kvm_arch_destroy_vm(kvm);
Scott Wood07f0a7b2013-04-25 14:11:23 +0000599 kvm_destroy_devices(kvm);
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100600 kvm_free_physmem(kvm);
601 cleanup_srcu_struct(&kvm->srcu);
602 kvm_arch_free_vm(kvm);
Alexander Graf10474ae2009-09-15 11:37:46 +0200603 hardware_disable_all();
Avi Kivity6d4e4c42007-11-21 16:41:05 +0200604 mmdrop(mm);
Avi Kivityf17abe92007-02-21 19:28:04 +0200605}
606
Izik Eidusd39f13b2008-03-30 16:01:25 +0300607void kvm_get_kvm(struct kvm *kvm)
608{
609 atomic_inc(&kvm->users_count);
610}
611EXPORT_SYMBOL_GPL(kvm_get_kvm);
612
613void kvm_put_kvm(struct kvm *kvm)
614{
615 if (atomic_dec_and_test(&kvm->users_count))
616 kvm_destroy_vm(kvm);
617}
618EXPORT_SYMBOL_GPL(kvm_put_kvm);
619
620
Avi Kivityf17abe92007-02-21 19:28:04 +0200621static int kvm_vm_release(struct inode *inode, struct file *filp)
622{
623 struct kvm *kvm = filp->private_data;
624
Gregory Haskins721eecb2009-05-20 10:30:49 -0400625 kvm_irqfd_release(kvm);
626
Izik Eidusd39f13b2008-03-30 16:01:25 +0300627 kvm_put_kvm(kvm);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800628 return 0;
629}
630
Takuya Yoshikawa515a0122010-10-27 18:23:54 +0900631/*
632 * Allocation size is twice as large as the actual dirty bitmap size.
Takuya Yoshikawa93474b22012-03-01 19:34:45 +0900633 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
Takuya Yoshikawa515a0122010-10-27 18:23:54 +0900634 */
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900635static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
636{
Takuya Yoshikawa189a2f72012-02-08 13:01:09 +0900637#ifndef CONFIG_S390
Takuya Yoshikawa515a0122010-10-27 18:23:54 +0900638 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900639
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900640 memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900641 if (!memslot->dirty_bitmap)
642 return -ENOMEM;
643
Takuya Yoshikawa189a2f72012-02-08 13:01:09 +0900644#endif /* !CONFIG_S390 */
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900645 return 0;
646}
647
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800648static int cmp_memslot(const void *slot1, const void *slot2)
649{
650 struct kvm_memory_slot *s1, *s2;
651
652 s1 = (struct kvm_memory_slot *)slot1;
653 s2 = (struct kvm_memory_slot *)slot2;
654
655 if (s1->npages < s2->npages)
656 return 1;
657 if (s1->npages > s2->npages)
658 return -1;
659
660 return 0;
661}
662
663/*
664 * Sort the memslots base on its size, so the larger slots
665 * will get better fit.
666 */
667static void sort_memslots(struct kvm_memslots *slots)
668{
Xiao Guangrongf85e2cb2011-11-24 17:41:54 +0800669 int i;
670
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800671 sort(slots->memslots, KVM_MEM_SLOTS_NUM,
672 sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
Xiao Guangrongf85e2cb2011-11-24 17:41:54 +0800673
674 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
675 slots->id_to_index[slots->memslots[i].id] = i;
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800676}
677
Alex Williamson116c14c2012-12-21 08:20:16 -0700678void update_memslots(struct kvm_memslots *slots, struct kvm_memory_slot *new,
679 u64 last_generation)
Xiao Guangrongbe593d62011-11-24 17:38:24 +0800680{
681 if (new) {
682 int id = new->id;
Xiao Guangrong28a37542011-11-24 19:04:35 +0800683 struct kvm_memory_slot *old = id_to_memslot(slots, id);
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800684 unsigned long npages = old->npages;
Xiao Guangrongbe593d62011-11-24 17:38:24 +0800685
Xiao Guangrong28a37542011-11-24 19:04:35 +0800686 *old = *new;
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800687 if (new->npages != npages)
688 sort_memslots(slots);
Xiao Guangrongbe593d62011-11-24 17:38:24 +0800689 }
690
Alex Williamson116c14c2012-12-21 08:20:16 -0700691 slots->generation = last_generation + 1;
Xiao Guangrongbe593d62011-11-24 17:38:24 +0800692}
693
Xiao Guangronga50d64d2012-08-21 10:58:13 +0800694static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
695{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +0800696 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
697
698#ifdef KVM_CAP_READONLY_MEM
699 valid_flags |= KVM_MEM_READONLY;
700#endif
701
702 if (mem->flags & ~valid_flags)
Xiao Guangronga50d64d2012-08-21 10:58:13 +0800703 return -EINVAL;
704
705 return 0;
706}
707
Gleb Natapov7ec4fb42012-12-24 17:49:30 +0200708static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
709 struct kvm_memslots *slots, struct kvm_memory_slot *new)
710{
711 struct kvm_memslots *old_memslots = kvm->memslots;
712
713 update_memslots(slots, new, kvm->memslots->generation);
714 rcu_assign_pointer(kvm->memslots, slots);
715 synchronize_srcu_expedited(&kvm->srcu);
Takuya Yoshikawae59dbe02013-07-04 13:40:29 +0900716
717 kvm_arch_memslots_updated(kvm);
718
719 return old_memslots;
Gleb Natapov7ec4fb42012-12-24 17:49:30 +0200720}
721
Avi Kivity6aa8b732006-12-10 02:21:36 -0800722/*
Avi Kivity6aa8b732006-12-10 02:21:36 -0800723 * Allocate some memory and give it an address in the guest physical address
724 * space.
725 *
726 * Discontiguous memory is allowed, mostly for framebuffers.
Sheng Yangf78e0e22007-10-29 09:40:42 +0800727 *
Marcelo Tosatti10589a42007-12-20 19:18:22 -0500728 * Must be called holding mmap_sem for write.
Avi Kivity6aa8b732006-12-10 02:21:36 -0800729 */
Sheng Yangf78e0e22007-10-29 09:40:42 +0800730int __kvm_set_memory_region(struct kvm *kvm,
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +0900731 struct kvm_userspace_memory_region *mem)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800732{
Avi Kivity8234b222010-12-27 12:08:45 +0200733 int r;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800734 gfn_t base_gfn;
Heiko Carstens28bcb112009-09-03 17:35:35 +0200735 unsigned long npages;
Takuya Yoshikawaa843fac2013-01-11 18:27:43 +0900736 struct kvm_memory_slot *slot;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800737 struct kvm_memory_slot old, new;
Alex Williamsonb7f69c52012-12-10 10:33:03 -0700738 struct kvm_memslots *slots = NULL, *old_memslots;
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900739 enum kvm_mr_change change;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800740
Xiao Guangronga50d64d2012-08-21 10:58:13 +0800741 r = check_memory_region_flags(mem);
742 if (r)
743 goto out;
744
Avi Kivity6aa8b732006-12-10 02:21:36 -0800745 r = -EINVAL;
746 /* General sanity checks */
747 if (mem->memory_size & (PAGE_SIZE - 1))
748 goto out;
749 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
750 goto out;
Takuya Yoshikawafa3d3152011-05-07 16:35:38 +0900751 /* We can read the guest memory with __xxx_user() later on. */
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +0900752 if ((mem->slot < KVM_USER_MEM_SLOTS) &&
Takuya Yoshikawafa3d3152011-05-07 16:35:38 +0900753 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
Heiko Carstens9e3bb6b2011-05-24 07:51:27 +0200754 !access_ok(VERIFY_WRITE,
755 (void __user *)(unsigned long)mem->userspace_addr,
756 mem->memory_size)))
Hollis Blanchard78749802008-11-07 13:32:12 -0600757 goto out;
Xiao Guangrong93a5cef2011-11-24 17:37:48 +0800758 if (mem->slot >= KVM_MEM_SLOTS_NUM)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800759 goto out;
760 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
761 goto out;
762
Takuya Yoshikawaa843fac2013-01-11 18:27:43 +0900763 slot = id_to_memslot(kvm->memslots, mem->slot);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800764 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
765 npages = mem->memory_size >> PAGE_SHIFT;
766
Takuya Yoshikawa660c22c2010-04-13 22:47:24 +0900767 r = -EINVAL;
768 if (npages > KVM_MEM_MAX_NR_PAGES)
769 goto out;
770
Avi Kivity6aa8b732006-12-10 02:21:36 -0800771 if (!npages)
772 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
773
Takuya Yoshikawaa843fac2013-01-11 18:27:43 +0900774 new = old = *slot;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800775
Avi Kivitye36d96f2010-06-21 10:56:36 +0300776 new.id = mem->slot;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800777 new.base_gfn = base_gfn;
778 new.npages = npages;
779 new.flags = mem->flags;
780
Avi Kivity6aa8b732006-12-10 02:21:36 -0800781 r = -EINVAL;
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900782 if (npages) {
783 if (!old.npages)
784 change = KVM_MR_CREATE;
785 else { /* Modify an existing slot. */
786 if ((mem->userspace_addr != old.userspace_addr) ||
Takuya Yoshikawa75d61fb2013-01-30 19:40:41 +0900787 (npages != old.npages) ||
788 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900789 goto out;
790
791 if (base_gfn != old.base_gfn)
792 change = KVM_MR_MOVE;
793 else if (new.flags != old.flags)
794 change = KVM_MR_FLAGS_ONLY;
795 else { /* Nothing to change. */
796 r = 0;
797 goto out;
798 }
799 }
800 } else if (old.npages) {
801 change = KVM_MR_DELETE;
802 } else /* Modify a non-existent slot: disallowed. */
Takuya Yoshikawa0ea75e12013-01-11 18:26:10 +0900803 goto out;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800804
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900805 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
Takuya Yoshikawa0a706be2013-01-11 18:26:55 +0900806 /* Check for overlaps */
807 r = -EEXIST;
808 kvm_for_each_memslot(slot, kvm->memslots) {
Takuya Yoshikawaa843fac2013-01-11 18:27:43 +0900809 if ((slot->id >= KVM_USER_MEM_SLOTS) ||
810 (slot->id == mem->slot))
Takuya Yoshikawa0a706be2013-01-11 18:26:55 +0900811 continue;
812 if (!((base_gfn + npages <= slot->base_gfn) ||
813 (base_gfn >= slot->base_gfn + slot->npages)))
814 goto out;
815 }
Avi Kivity6aa8b732006-12-10 02:21:36 -0800816 }
Avi Kivity6aa8b732006-12-10 02:21:36 -0800817
Avi Kivity6aa8b732006-12-10 02:21:36 -0800818 /* Free page dirty bitmap if unneeded */
819 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
Al Viro8b6d44c2007-02-09 16:38:40 +0000820 new.dirty_bitmap = NULL;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800821
822 r = -ENOMEM;
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900823 if (change == KVM_MR_CREATE) {
Takuya Yoshikawa189a2f72012-02-08 13:01:09 +0900824 new.userspace_addr = mem->userspace_addr;
Takuya Yoshikawad89cc612012-08-01 18:03:28 +0900825
Aneesh Kumar K.V55870272013-10-07 22:18:00 +0530826 if (kvm_arch_create_memslot(kvm, &new, npages))
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +0900827 goto out_free;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800828 }
Joerg Roedelec04b262009-06-19 15:16:23 +0200829
Avi Kivity6aa8b732006-12-10 02:21:36 -0800830 /* Allocate page dirty bitmap if needed */
831 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900832 if (kvm_create_dirty_bitmap(&new) < 0)
Sheng Yangf78e0e22007-10-29 09:40:42 +0800833 goto out_free;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800834 }
835
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900836 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200837 r = -ENOMEM;
Thomas Meyer6da64fd2011-11-08 20:32:19 +0100838 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
839 GFP_KERNEL);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200840 if (!slots)
841 goto out_free;
Xiao Guangrong28a37542011-11-24 19:04:35 +0800842 slot = id_to_memslot(slots, mem->slot);
843 slot->flags |= KVM_MEMSLOT_INVALID;
844
Gleb Natapov7ec4fb42012-12-24 17:49:30 +0200845 old_memslots = install_new_memslots(kvm, slots, NULL);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200846
Alex Williamsone40f1932012-12-10 10:32:57 -0700847 /* slot was deleted or moved, clear iommu mapping */
848 kvm_iommu_unmap_pages(kvm, &old);
Marcelo Tosatti12d6e752012-08-24 15:54:58 -0300849 /* From this point no new shadow pages pointing to a deleted,
850 * or moved, memslot will be created.
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200851 *
852 * validation of sp->gfn happens in:
853 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
854 * - kvm_is_visible_gfn (mmu_check_roots)
855 */
Marcelo Tosatti2df72e92012-08-24 15:54:57 -0300856 kvm_arch_flush_shadow_memslot(kvm, slot);
Alex Williamsonb7f69c52012-12-10 10:33:03 -0700857 slots = old_memslots;
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200858 }
Marcelo Tosatti34d4cb82008-07-10 20:49:31 -0300859
Takuya Yoshikawa7b6195a2013-02-27 19:44:34 +0900860 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
Marcelo Tosattif7784b82009-12-23 14:35:18 -0200861 if (r)
Alex Williamsonb7f69c52012-12-10 10:33:03 -0700862 goto out_slots;
Marcelo Tosattif7784b82009-12-23 14:35:18 -0200863
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200864 r = -ENOMEM;
Alex Williamsonb7f69c52012-12-10 10:33:03 -0700865 /*
866 * We can re-use the old_memslots from above, the only difference
867 * from the currently installed memslots is the invalid flag. This
868 * will get overwritten by update_memslots anyway.
869 */
870 if (!slots) {
871 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
872 GFP_KERNEL);
873 if (!slots)
874 goto out_free;
875 }
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200876
877 /* actual memory is freed via old in kvm_free_physmem_slot below */
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900878 if (change == KVM_MR_DELETE) {
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200879 new.dirty_bitmap = NULL;
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +0900880 memset(&new.arch, 0, sizeof(new.arch));
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200881 }
882
Gleb Natapov7ec4fb42012-12-24 17:49:30 +0200883 old_memslots = install_new_memslots(kvm, slots, &new);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800884
Takuya Yoshikawa84826442013-02-27 19:45:25 +0900885 kvm_arch_commit_memory_region(kvm, mem, &old, change);
Zhang Xiantao3ad82a72007-11-20 13:11:38 +0800886
Aneesh Kumar K.V55870272013-10-07 22:18:00 +0530887 kvm_free_physmem_slot(kvm, &old, &new);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200888 kfree(old_memslots);
889
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200890 /*
891 * IOMMU mapping: New slots need to be mapped. Old slots need to be
Avi Kivity6aa8b732006-12-10 02:21:36 -0800892 * un-mapped and re-mapped if their base changes. Since base change
893 * unmapping is handled above with slot deletion, mapping alone is
Sheng Yangf78e0e22007-10-29 09:40:42 +0800894 * needed here. Anything else the iommu might care about for existing
Avi Kivity6aa8b732006-12-10 02:21:36 -0800895 * slots (size changes, userspace addr changes and read-only flag
896 * changes) is disallowed above, so any other attribute changes getting
897 * here can be skipped.
Izik Eidus210c7c42007-10-24 23:52:57 +0200898 */
899 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
Sheng Yangf78e0e22007-10-29 09:40:42 +0800900 r = kvm_iommu_map_pages(kvm, &new);
Yang Zhange0230e12013-10-24 09:56:39 +0800901 return r;
Sheng Yangf78e0e22007-10-29 09:40:42 +0800902 }
903
Avi Kivity6aa8b732006-12-10 02:21:36 -0800904 return 0;
905
906out_slots:
907 kfree(slots);
908out_free:
Aneesh Kumar K.V55870272013-10-07 22:18:00 +0530909 kvm_free_physmem_slot(kvm, &new, &old);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800910out:
911 return r;
Izik Eidus210c7c42007-10-24 23:52:57 +0200912}
913EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
Sheng Yangf78e0e22007-10-29 09:40:42 +0800914
915int kvm_set_memory_region(struct kvm *kvm,
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +0900916 struct kvm_userspace_memory_region *mem)
Sheng Yangf78e0e22007-10-29 09:40:42 +0800917{
918 int r;
919
Marcelo Tosatti79fac952009-12-23 14:35:26 -0200920 mutex_lock(&kvm->slots_lock);
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +0900921 r = __kvm_set_memory_region(kvm, mem);
Marcelo Tosatti79fac952009-12-23 14:35:26 -0200922 mutex_unlock(&kvm->slots_lock);
Sheng Yangf78e0e22007-10-29 09:40:42 +0800923 return r;
924}
Izik Eidus210c7c42007-10-24 23:52:57 +0200925EXPORT_SYMBOL_GPL(kvm_set_memory_region);
926
Carsten Otte1fe779f2007-10-29 16:08:35 +0100927int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +0900928 struct kvm_userspace_memory_region *mem)
Izik Eidus210c7c42007-10-24 23:52:57 +0200929{
Alex Williamsonbbacc0c2012-12-10 10:33:09 -0700930 if (mem->slot >= KVM_USER_MEM_SLOTS)
Izik Eiduse0d62c72007-10-24 23:57:46 +0200931 return -EINVAL;
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +0900932 return kvm_set_memory_region(kvm, mem);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800933}
934
Zhang Xiantao5bb064d2007-11-18 20:29:43 +0800935int kvm_get_dirty_log(struct kvm *kvm,
936 struct kvm_dirty_log *log, int *is_dirty)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800937{
938 struct kvm_memory_slot *memslot;
939 int r, i;
Takuya Yoshikawa87bf6e72010-04-12 19:35:35 +0900940 unsigned long n;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800941 unsigned long any = 0;
942
Avi Kivity6aa8b732006-12-10 02:21:36 -0800943 r = -EINVAL;
Alex Williamsonbbacc0c2012-12-10 10:33:09 -0700944 if (log->slot >= KVM_USER_MEM_SLOTS)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800945 goto out;
946
Xiao Guangrong28a37542011-11-24 19:04:35 +0800947 memslot = id_to_memslot(kvm->memslots, log->slot);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800948 r = -ENOENT;
949 if (!memslot->dirty_bitmap)
950 goto out;
951
Takuya Yoshikawa87bf6e72010-04-12 19:35:35 +0900952 n = kvm_dirty_bitmap_bytes(memslot);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800953
Uri Lublincd1a4a92007-02-22 16:43:09 +0200954 for (i = 0; !any && i < n/sizeof(long); ++i)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800955 any = memslot->dirty_bitmap[i];
956
957 r = -EFAULT;
958 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
959 goto out;
960
Zhang Xiantao5bb064d2007-11-18 20:29:43 +0800961 if (any)
962 *is_dirty = 1;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800963
964 r = 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800965out:
Avi Kivity6aa8b732006-12-10 02:21:36 -0800966 return r;
967}
Aneesh Kumar K.V2ba9f0d2013-10-07 22:17:59 +0530968EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800969
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +0900970bool kvm_largepages_enabled(void)
971{
972 return largepages_enabled;
973}
974
Marcelo Tosatti54dee992009-06-11 12:07:44 -0300975void kvm_disable_largepages(void)
976{
977 largepages_enabled = false;
978}
979EXPORT_SYMBOL_GPL(kvm_disable_largepages);
980
Gleb Natapov49c77542010-10-18 15:22:23 +0200981struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
982{
983 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
984}
Avi Kivitya1f4d3952010-06-21 11:44:20 +0300985EXPORT_SYMBOL_GPL(gfn_to_memslot);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800986
Izik Eiduse0d62c72007-10-24 23:57:46 +0200987int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
988{
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800989 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
Izik Eiduse0d62c72007-10-24 23:57:46 +0200990
Alex Williamsonbbacc0c2012-12-10 10:33:09 -0700991 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800992 memslot->flags & KVM_MEMSLOT_INVALID)
993 return 0;
Izik Eiduse0d62c72007-10-24 23:57:46 +0200994
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800995 return 1;
Izik Eiduse0d62c72007-10-24 23:57:46 +0200996}
997EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
998
Joerg Roedel8f0b1ab2010-01-28 12:37:56 +0100999unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1000{
1001 struct vm_area_struct *vma;
1002 unsigned long addr, size;
1003
1004 size = PAGE_SIZE;
1005
1006 addr = gfn_to_hva(kvm, gfn);
1007 if (kvm_is_error_hva(addr))
1008 return PAGE_SIZE;
1009
1010 down_read(&current->mm->mmap_sem);
1011 vma = find_vma(current->mm, addr);
1012 if (!vma)
1013 goto out;
1014
1015 size = vma_kernel_pagesize(vma);
1016
1017out:
1018 up_read(&current->mm->mmap_sem);
1019
1020 return size;
1021}
1022
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001023static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1024{
1025 return slot->flags & KVM_MEM_READONLY;
1026}
1027
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001028static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1029 gfn_t *nr_pages, bool write)
Izik Eidus539cb662007-11-11 22:05:04 +02001030{
Marcelo Tosattibc6678a2009-12-23 14:35:21 -02001031 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
Xiao Guangrongca3a4902012-08-21 11:01:50 +08001032 return KVM_HVA_ERR_BAD;
Xiao Guangrong48987782010-08-22 19:11:43 +08001033
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001034 if (memslot_is_readonly(slot) && write)
1035 return KVM_HVA_ERR_RO_BAD;
Xiao Guangrong48987782010-08-22 19:11:43 +08001036
1037 if (nr_pages)
1038 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1039
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001040 return __gfn_to_hva_memslot(slot, gfn);
Izik Eidus539cb662007-11-11 22:05:04 +02001041}
Xiao Guangrong48987782010-08-22 19:11:43 +08001042
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001043static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1044 gfn_t *nr_pages)
1045{
1046 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1047}
1048
1049unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1050 gfn_t gfn)
1051{
1052 return gfn_to_hva_many(slot, gfn, NULL);
1053}
1054EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1055
Xiao Guangrong48987782010-08-22 19:11:43 +08001056unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1057{
Gleb Natapov49c77542010-10-18 15:22:23 +02001058 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
Xiao Guangrong48987782010-08-22 19:11:43 +08001059}
Sheng Yang0d150292008-04-25 21:44:50 +08001060EXPORT_SYMBOL_GPL(gfn_to_hva);
Izik Eidus539cb662007-11-11 22:05:04 +02001061
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001062/*
Paolo Bonziniba6a3542013-09-09 13:52:33 +02001063 * If writable is set to false, the hva returned by this function is only
1064 * allowed to be read.
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001065 */
Paolo Bonziniba6a3542013-09-09 13:52:33 +02001066unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
Gleb Natapov80300892010-10-19 18:13:41 +02001067{
Paolo Bonziniba6a3542013-09-09 13:52:33 +02001068 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
Gleb Natapova2ac07f2013-10-01 19:58:36 +03001069 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1070
1071 if (!kvm_is_error_hva(hva) && writable)
Paolo Bonziniba6a3542013-09-09 13:52:33 +02001072 *writable = !memslot_is_readonly(slot);
1073
Gleb Natapova2ac07f2013-10-01 19:58:36 +03001074 return hva;
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001075}
1076
1077static int kvm_read_hva(void *data, void __user *hva, int len)
1078{
1079 return __copy_from_user(data, hva, len);
1080}
1081
1082static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1083{
1084 return __copy_from_user_inatomic(data, hva, len);
Gleb Natapov80300892010-10-19 18:13:41 +02001085}
1086
Geoff Levand39369f72013-04-05 19:20:30 +00001087static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
Gleb Natapov0857b9e2011-02-01 13:21:47 +02001088 unsigned long start, int write, struct page **page)
1089{
1090 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1091
1092 if (write)
1093 flags |= FOLL_WRITE;
1094
1095 return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1096}
1097
Huang Yingfafc3db2011-01-30 11:15:49 +08001098static inline int check_user_page_hwpoison(unsigned long addr)
1099{
1100 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1101
1102 rc = __get_user_pages(current, current->mm, addr, 1,
1103 flags, NULL, NULL, NULL);
1104 return rc == -EHWPOISON;
1105}
1106
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001107/*
1108 * The atomic path to get the writable pfn which will be stored in @pfn,
1109 * true indicates success, otherwise false is returned.
1110 */
1111static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1112 bool write_fault, bool *writable, pfn_t *pfn)
1113{
1114 struct page *page[1];
1115 int npages;
1116
1117 if (!(async || atomic))
1118 return false;
1119
Xiao Guangrong12ce13f2012-08-21 11:00:49 +08001120 /*
1121 * Fast pin a writable pfn only if it is a write fault request
1122 * or the caller allows to map a writable pfn for a read fault
1123 * request.
1124 */
1125 if (!(write_fault || writable))
1126 return false;
1127
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001128 npages = __get_user_pages_fast(addr, 1, 1, page);
1129 if (npages == 1) {
1130 *pfn = page_to_pfn(page[0]);
1131
1132 if (writable)
1133 *writable = true;
1134 return true;
1135 }
1136
1137 return false;
1138}
1139
1140/*
1141 * The slow path to get the pfn of the specified host virtual address,
1142 * 1 indicates success, -errno is returned if error is detected.
1143 */
1144static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1145 bool *writable, pfn_t *pfn)
Avi Kivity954bbbc2007-03-30 14:02:32 +03001146{
Anthony Liguori8d4e1282007-10-18 09:59:34 -05001147 struct page *page[1];
Gleb Natapovaf585b92010-10-14 11:22:46 +02001148 int npages = 0;
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001149
1150 might_sleep();
1151
1152 if (writable)
1153 *writable = write_fault;
1154
1155 if (async) {
1156 down_read(&current->mm->mmap_sem);
1157 npages = get_user_page_nowait(current, current->mm,
1158 addr, write_fault, page);
1159 up_read(&current->mm->mmap_sem);
1160 } else
1161 npages = get_user_pages_fast(addr, 1, write_fault,
1162 page);
1163 if (npages != 1)
1164 return npages;
1165
1166 /* map read fault as writable if possible */
Xiao Guangrong12ce13f2012-08-21 11:00:49 +08001167 if (unlikely(!write_fault) && writable) {
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001168 struct page *wpage[1];
1169
1170 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1171 if (npages == 1) {
1172 *writable = true;
1173 put_page(page[0]);
1174 page[0] = wpage[0];
1175 }
1176
1177 npages = 1;
1178 }
1179 *pfn = page_to_pfn(page[0]);
1180 return npages;
1181}
1182
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001183static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1184{
1185 if (unlikely(!(vma->vm_flags & VM_READ)))
1186 return false;
1187
1188 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1189 return false;
1190
1191 return true;
1192}
1193
Xiao Guangrong12ce13f2012-08-21 11:00:49 +08001194/*
1195 * Pin guest page in memory and return its pfn.
1196 * @addr: host virtual address which maps memory to the guest
1197 * @atomic: whether this function can sleep
1198 * @async: whether this function need to wait IO complete if the
1199 * host page is not in the memory
1200 * @write_fault: whether we should get a writable host page
1201 * @writable: whether it allows to map a writable host page for !@write_fault
1202 *
1203 * The function will map a writable host page for these two cases:
1204 * 1): @write_fault = true
1205 * 2): @write_fault = false && @writable, @writable will tell the caller
1206 * whether the mapping is writable.
1207 */
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001208static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1209 bool write_fault, bool *writable)
1210{
1211 struct vm_area_struct *vma;
1212 pfn_t pfn = 0;
1213 int npages;
Avi Kivity954bbbc2007-03-30 14:02:32 +03001214
Gleb Natapovaf585b92010-10-14 11:22:46 +02001215 /* we can do it either atomically or asynchronously, not both */
1216 BUG_ON(atomic && async);
1217
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001218 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1219 return pfn;
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001220
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001221 if (atomic)
1222 return KVM_PFN_ERR_FAULT;
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001223
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001224 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1225 if (npages == 1)
1226 return pfn;
Gleb Natapovaf585b92010-10-14 11:22:46 +02001227
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001228 down_read(&current->mm->mmap_sem);
1229 if (npages == -EHWPOISON ||
1230 (!async && check_user_page_hwpoison(addr))) {
1231 pfn = KVM_PFN_ERR_HWPOISON;
1232 goto exit;
Xiao Guangrong887c08a2010-08-22 19:10:28 +08001233 }
Izik Eidus539cb662007-11-11 22:05:04 +02001234
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001235 vma = find_vma_intersection(current->mm, addr, addr + 1);
Anthony Liguori8d4e1282007-10-18 09:59:34 -05001236
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001237 if (vma == NULL)
1238 pfn = KVM_PFN_ERR_FAULT;
1239 else if ((vma->vm_flags & VM_PFNMAP)) {
1240 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1241 vma->vm_pgoff;
1242 BUG_ON(!kvm_is_mmio_pfn(pfn));
1243 } else {
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001244 if (async && vma_is_valid(vma, write_fault))
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001245 *async = true;
1246 pfn = KVM_PFN_ERR_FAULT;
1247 }
1248exit:
1249 up_read(&current->mm->mmap_sem);
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001250 return pfn;
Anthony Liguori35149e22008-04-02 14:46:56 -05001251}
1252
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001253static pfn_t
1254__gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1255 bool *async, bool write_fault, bool *writable)
Xiao Guangrong887c08a2010-08-22 19:10:28 +08001256{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001257 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1258
1259 if (addr == KVM_HVA_ERR_RO_BAD)
1260 return KVM_PFN_ERR_RO_FAULT;
1261
1262 if (kvm_is_error_hva(addr))
Xiao Guangrong81c52c52012-10-16 20:10:59 +08001263 return KVM_PFN_NOSLOT;
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001264
1265 /* Do not map writable pfn in the readonly memslot. */
1266 if (writable && memslot_is_readonly(slot)) {
1267 *writable = false;
1268 writable = NULL;
1269 }
1270
1271 return hva_to_pfn(addr, atomic, async, write_fault,
1272 writable);
Xiao Guangrong887c08a2010-08-22 19:10:28 +08001273}
Xiao Guangrong887c08a2010-08-22 19:10:28 +08001274
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001275static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1276 bool write_fault, bool *writable)
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001277{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001278 struct kvm_memory_slot *slot;
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001279
Gleb Natapovaf585b92010-10-14 11:22:46 +02001280 if (async)
1281 *async = false;
1282
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001283 slot = gfn_to_memslot(kvm, gfn);
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001284
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001285 return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1286 writable);
Xiao Guangrong365fb3f2010-08-28 19:24:13 +08001287}
1288
1289pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1290{
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001291 return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
Xiao Guangrong365fb3f2010-08-28 19:24:13 +08001292}
1293EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1294
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001295pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1296 bool write_fault, bool *writable)
Gleb Natapovaf585b92010-10-14 11:22:46 +02001297{
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001298 return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
Gleb Natapovaf585b92010-10-14 11:22:46 +02001299}
1300EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1301
Xiao Guangrong365fb3f2010-08-28 19:24:13 +08001302pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1303{
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001304 return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001305}
Anthony Liguori35149e22008-04-02 14:46:56 -05001306EXPORT_SYMBOL_GPL(gfn_to_pfn);
1307
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001308pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1309 bool *writable)
1310{
1311 return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1312}
1313EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1314
Xiao Guangrongd5661042012-07-17 21:56:16 +08001315pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001316{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001317 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001318}
1319
Xiao Guangrong037d92d2012-08-21 10:59:12 +08001320pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1321{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001322 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
Xiao Guangrong037d92d2012-08-21 10:59:12 +08001323}
1324EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1325
Xiao Guangrong48987782010-08-22 19:11:43 +08001326int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1327 int nr_pages)
1328{
1329 unsigned long addr;
1330 gfn_t entry;
1331
Gleb Natapov49c77542010-10-18 15:22:23 +02001332 addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
Xiao Guangrong48987782010-08-22 19:11:43 +08001333 if (kvm_is_error_hva(addr))
1334 return -1;
1335
1336 if (entry < nr_pages)
1337 return 0;
1338
1339 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1340}
1341EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1342
Xiao Guangronga2766322012-07-26 11:58:59 +08001343static struct page *kvm_pfn_to_page(pfn_t pfn)
1344{
Xiao Guangrong81c52c52012-10-16 20:10:59 +08001345 if (is_error_noslot_pfn(pfn))
Xiao Guangrong6cede2e2012-08-03 15:41:22 +08001346 return KVM_ERR_PTR_BAD_PAGE;
Xiao Guangronga2766322012-07-26 11:58:59 +08001347
Xiao Guangrongcb9aaa32012-08-03 15:42:10 +08001348 if (kvm_is_mmio_pfn(pfn)) {
1349 WARN_ON(1);
1350 return KVM_ERR_PTR_BAD_PAGE;
1351 }
1352
Xiao Guangronga2766322012-07-26 11:58:59 +08001353 return pfn_to_page(pfn);
1354}
1355
Anthony Liguori35149e22008-04-02 14:46:56 -05001356struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1357{
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001358 pfn_t pfn;
1359
1360 pfn = gfn_to_pfn(kvm, gfn);
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001361
Xiao Guangronga2766322012-07-26 11:58:59 +08001362 return kvm_pfn_to_page(pfn);
Avi Kivity954bbbc2007-03-30 14:02:32 +03001363}
Anthony Liguoriaab61cc2007-10-29 15:15:20 -05001364
Avi Kivity954bbbc2007-03-30 14:02:32 +03001365EXPORT_SYMBOL_GPL(gfn_to_page);
1366
Izik Eidusb4231d62007-11-20 11:49:33 +02001367void kvm_release_page_clean(struct page *page)
1368{
Xiao Guangrong32cad842012-08-03 15:42:52 +08001369 WARN_ON(is_error_page(page));
1370
Anthony Liguori35149e22008-04-02 14:46:56 -05001371 kvm_release_pfn_clean(page_to_pfn(page));
Izik Eidusb4231d62007-11-20 11:49:33 +02001372}
1373EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1374
Anthony Liguori35149e22008-04-02 14:46:56 -05001375void kvm_release_pfn_clean(pfn_t pfn)
1376{
Xiao Guangrong81c52c52012-10-16 20:10:59 +08001377 if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001378 put_page(pfn_to_page(pfn));
Anthony Liguori35149e22008-04-02 14:46:56 -05001379}
1380EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1381
Izik Eidusb4231d62007-11-20 11:49:33 +02001382void kvm_release_page_dirty(struct page *page)
Izik Eidus8a7ae052007-10-18 11:09:33 +02001383{
Xiao Guangronga2766322012-07-26 11:58:59 +08001384 WARN_ON(is_error_page(page));
1385
Anthony Liguori35149e22008-04-02 14:46:56 -05001386 kvm_release_pfn_dirty(page_to_pfn(page));
Izik Eidus8a7ae052007-10-18 11:09:33 +02001387}
Izik Eidusb4231d62007-11-20 11:49:33 +02001388EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
Izik Eidus8a7ae052007-10-18 11:09:33 +02001389
Anthony Liguori35149e22008-04-02 14:46:56 -05001390void kvm_release_pfn_dirty(pfn_t pfn)
1391{
1392 kvm_set_pfn_dirty(pfn);
1393 kvm_release_pfn_clean(pfn);
1394}
1395EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1396
1397void kvm_set_page_dirty(struct page *page)
1398{
1399 kvm_set_pfn_dirty(page_to_pfn(page));
1400}
1401EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1402
1403void kvm_set_pfn_dirty(pfn_t pfn)
1404{
Xiantao Zhangc77fb9d2008-09-27 10:55:40 +08001405 if (!kvm_is_mmio_pfn(pfn)) {
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001406 struct page *page = pfn_to_page(pfn);
1407 if (!PageReserved(page))
1408 SetPageDirty(page);
1409 }
Anthony Liguori35149e22008-04-02 14:46:56 -05001410}
1411EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1412
1413void kvm_set_pfn_accessed(pfn_t pfn)
1414{
Xiantao Zhangc77fb9d2008-09-27 10:55:40 +08001415 if (!kvm_is_mmio_pfn(pfn))
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001416 mark_page_accessed(pfn_to_page(pfn));
Anthony Liguori35149e22008-04-02 14:46:56 -05001417}
1418EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1419
1420void kvm_get_pfn(pfn_t pfn)
1421{
Xiantao Zhangc77fb9d2008-09-27 10:55:40 +08001422 if (!kvm_is_mmio_pfn(pfn))
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001423 get_page(pfn_to_page(pfn));
Anthony Liguori35149e22008-04-02 14:46:56 -05001424}
1425EXPORT_SYMBOL_GPL(kvm_get_pfn);
1426
Izik Eidus195aefd2007-10-01 22:14:18 +02001427static int next_segment(unsigned long len, int offset)
1428{
1429 if (len > PAGE_SIZE - offset)
1430 return PAGE_SIZE - offset;
1431 else
1432 return len;
1433}
1434
1435int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1436 int len)
1437{
Izik Eiduse0506bc2007-11-11 22:10:22 +02001438 int r;
1439 unsigned long addr;
Izik Eidus195aefd2007-10-01 22:14:18 +02001440
Paolo Bonziniba6a3542013-09-09 13:52:33 +02001441 addr = gfn_to_hva_prot(kvm, gfn, NULL);
Izik Eiduse0506bc2007-11-11 22:10:22 +02001442 if (kvm_is_error_hva(addr))
Izik Eidus195aefd2007-10-01 22:14:18 +02001443 return -EFAULT;
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001444 r = kvm_read_hva(data, (void __user *)addr + offset, len);
Izik Eiduse0506bc2007-11-11 22:10:22 +02001445 if (r)
1446 return -EFAULT;
Izik Eidus195aefd2007-10-01 22:14:18 +02001447 return 0;
1448}
1449EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1450
1451int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1452{
1453 gfn_t gfn = gpa >> PAGE_SHIFT;
1454 int seg;
1455 int offset = offset_in_page(gpa);
1456 int ret;
1457
1458 while ((seg = next_segment(len, offset)) != 0) {
1459 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1460 if (ret < 0)
1461 return ret;
1462 offset = 0;
1463 len -= seg;
1464 data += seg;
1465 ++gfn;
1466 }
1467 return 0;
1468}
1469EXPORT_SYMBOL_GPL(kvm_read_guest);
1470
Marcelo Tosatti7ec54582007-12-20 19:18:23 -05001471int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1472 unsigned long len)
1473{
1474 int r;
1475 unsigned long addr;
1476 gfn_t gfn = gpa >> PAGE_SHIFT;
1477 int offset = offset_in_page(gpa);
1478
Paolo Bonziniba6a3542013-09-09 13:52:33 +02001479 addr = gfn_to_hva_prot(kvm, gfn, NULL);
Marcelo Tosatti7ec54582007-12-20 19:18:23 -05001480 if (kvm_is_error_hva(addr))
1481 return -EFAULT;
Andrea Arcangeli0aac03f2008-01-30 19:57:35 +01001482 pagefault_disable();
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001483 r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
Andrea Arcangeli0aac03f2008-01-30 19:57:35 +01001484 pagefault_enable();
Marcelo Tosatti7ec54582007-12-20 19:18:23 -05001485 if (r)
1486 return -EFAULT;
1487 return 0;
1488}
1489EXPORT_SYMBOL(kvm_read_guest_atomic);
1490
Izik Eidus195aefd2007-10-01 22:14:18 +02001491int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1492 int offset, int len)
1493{
Izik Eiduse0506bc2007-11-11 22:10:22 +02001494 int r;
1495 unsigned long addr;
Izik Eidus195aefd2007-10-01 22:14:18 +02001496
Izik Eiduse0506bc2007-11-11 22:10:22 +02001497 addr = gfn_to_hva(kvm, gfn);
1498 if (kvm_is_error_hva(addr))
Izik Eidus195aefd2007-10-01 22:14:18 +02001499 return -EFAULT;
Xiao Guangrong8b0cedf2011-05-15 23:22:04 +08001500 r = __copy_to_user((void __user *)addr + offset, data, len);
Izik Eiduse0506bc2007-11-11 22:10:22 +02001501 if (r)
1502 return -EFAULT;
Izik Eidus195aefd2007-10-01 22:14:18 +02001503 mark_page_dirty(kvm, gfn);
1504 return 0;
1505}
1506EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1507
1508int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1509 unsigned long len)
1510{
1511 gfn_t gfn = gpa >> PAGE_SHIFT;
1512 int seg;
1513 int offset = offset_in_page(gpa);
1514 int ret;
1515
1516 while ((seg = next_segment(len, offset)) != 0) {
1517 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1518 if (ret < 0)
1519 return ret;
1520 offset = 0;
1521 len -= seg;
1522 data += seg;
1523 ++gfn;
1524 }
1525 return 0;
1526}
1527
Gleb Natapov49c77542010-10-18 15:22:23 +02001528int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
Andrew Honig8f964522013-03-29 09:35:21 -07001529 gpa_t gpa, unsigned long len)
Gleb Natapov49c77542010-10-18 15:22:23 +02001530{
1531 struct kvm_memslots *slots = kvm_memslots(kvm);
1532 int offset = offset_in_page(gpa);
Andrew Honig8f964522013-03-29 09:35:21 -07001533 gfn_t start_gfn = gpa >> PAGE_SHIFT;
1534 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1535 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1536 gfn_t nr_pages_avail;
Gleb Natapov49c77542010-10-18 15:22:23 +02001537
1538 ghc->gpa = gpa;
1539 ghc->generation = slots->generation;
Andrew Honig8f964522013-03-29 09:35:21 -07001540 ghc->len = len;
1541 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1542 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, &nr_pages_avail);
1543 if (!kvm_is_error_hva(ghc->hva) && nr_pages_avail >= nr_pages_needed) {
Gleb Natapov49c77542010-10-18 15:22:23 +02001544 ghc->hva += offset;
Andrew Honig8f964522013-03-29 09:35:21 -07001545 } else {
1546 /*
1547 * If the requested region crosses two memslots, we still
1548 * verify that the entire region is valid here.
1549 */
1550 while (start_gfn <= end_gfn) {
1551 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1552 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1553 &nr_pages_avail);
1554 if (kvm_is_error_hva(ghc->hva))
1555 return -EFAULT;
1556 start_gfn += nr_pages_avail;
1557 }
1558 /* Use the slow path for cross page reads and writes. */
1559 ghc->memslot = NULL;
1560 }
Gleb Natapov49c77542010-10-18 15:22:23 +02001561 return 0;
1562}
1563EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1564
1565int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1566 void *data, unsigned long len)
1567{
1568 struct kvm_memslots *slots = kvm_memslots(kvm);
1569 int r;
1570
Andrew Honig8f964522013-03-29 09:35:21 -07001571 BUG_ON(len > ghc->len);
1572
Gleb Natapov49c77542010-10-18 15:22:23 +02001573 if (slots->generation != ghc->generation)
Andrew Honig8f964522013-03-29 09:35:21 -07001574 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1575
1576 if (unlikely(!ghc->memslot))
1577 return kvm_write_guest(kvm, ghc->gpa, data, len);
Gleb Natapov49c77542010-10-18 15:22:23 +02001578
1579 if (kvm_is_error_hva(ghc->hva))
1580 return -EFAULT;
1581
Xiao Guangrong8b0cedf2011-05-15 23:22:04 +08001582 r = __copy_to_user((void __user *)ghc->hva, data, len);
Gleb Natapov49c77542010-10-18 15:22:23 +02001583 if (r)
1584 return -EFAULT;
1585 mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1586
1587 return 0;
1588}
1589EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1590
Gleb Natapove03b6442011-07-11 15:28:11 -04001591int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1592 void *data, unsigned long len)
1593{
1594 struct kvm_memslots *slots = kvm_memslots(kvm);
1595 int r;
1596
Andrew Honig8f964522013-03-29 09:35:21 -07001597 BUG_ON(len > ghc->len);
1598
Gleb Natapove03b6442011-07-11 15:28:11 -04001599 if (slots->generation != ghc->generation)
Andrew Honig8f964522013-03-29 09:35:21 -07001600 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1601
1602 if (unlikely(!ghc->memslot))
1603 return kvm_read_guest(kvm, ghc->gpa, data, len);
Gleb Natapove03b6442011-07-11 15:28:11 -04001604
1605 if (kvm_is_error_hva(ghc->hva))
1606 return -EFAULT;
1607
1608 r = __copy_from_user(data, (void __user *)ghc->hva, len);
1609 if (r)
1610 return -EFAULT;
1611
1612 return 0;
1613}
1614EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1615
Izik Eidus195aefd2007-10-01 22:14:18 +02001616int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1617{
Heiko Carstens3bcc8a82010-10-27 17:21:21 +02001618 return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1619 offset, len);
Izik Eidus195aefd2007-10-01 22:14:18 +02001620}
1621EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1622
1623int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1624{
1625 gfn_t gfn = gpa >> PAGE_SHIFT;
1626 int seg;
1627 int offset = offset_in_page(gpa);
1628 int ret;
1629
1630 while ((seg = next_segment(len, offset)) != 0) {
1631 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1632 if (ret < 0)
1633 return ret;
1634 offset = 0;
1635 len -= seg;
1636 ++gfn;
1637 }
1638 return 0;
1639}
1640EXPORT_SYMBOL_GPL(kvm_clear_guest);
1641
Gleb Natapov49c77542010-10-18 15:22:23 +02001642void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1643 gfn_t gfn)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001644{
Rusty Russell7e9d6192007-07-31 20:41:14 +10001645 if (memslot && memslot->dirty_bitmap) {
1646 unsigned long rel_gfn = gfn - memslot->base_gfn;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001647
Takuya Yoshikawab74ca3b2012-10-04 17:13:12 -07001648 set_bit_le(rel_gfn, memslot->dirty_bitmap);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001649 }
1650}
1651
Gleb Natapov49c77542010-10-18 15:22:23 +02001652void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1653{
1654 struct kvm_memory_slot *memslot;
1655
1656 memslot = gfn_to_memslot(kvm, gfn);
1657 mark_page_dirty_in_slot(kvm, memslot, gfn);
1658}
Aneesh Kumar K.V2ba9f0d2013-10-07 22:17:59 +05301659EXPORT_SYMBOL_GPL(mark_page_dirty);
Gleb Natapov49c77542010-10-18 15:22:23 +02001660
Eddie Dongb6958ce2007-07-18 12:15:21 +03001661/*
1662 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1663 */
Hollis Blanchard8776e512007-10-31 17:24:24 -05001664void kvm_vcpu_block(struct kvm_vcpu *vcpu)
Eddie Dongb6958ce2007-07-18 12:15:21 +03001665{
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001666 DEFINE_WAIT(wait);
Eddie Dongb6958ce2007-07-18 12:15:21 +03001667
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001668 for (;;) {
1669 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
Eddie Dongb6958ce2007-07-18 12:15:21 +03001670
Gleb Natapova1b37102009-07-09 15:33:52 +03001671 if (kvm_arch_vcpu_runnable(vcpu)) {
Avi Kivitya8eeb042010-05-10 12:34:53 +03001672 kvm_make_request(KVM_REQ_UNHALT, vcpu);
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001673 break;
Marcelo Tosattid7690172008-09-08 15:23:48 -03001674 }
Gleb Natapov09cec752009-03-23 15:11:44 +02001675 if (kvm_cpu_has_pending_timer(vcpu))
1676 break;
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001677 if (signal_pending(current))
1678 break;
1679
Eddie Dongb6958ce2007-07-18 12:15:21 +03001680 schedule();
Eddie Dongb6958ce2007-07-18 12:15:21 +03001681 }
1682
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001683 finish_wait(&vcpu->wq, &wait);
Eddie Dongb6958ce2007-07-18 12:15:21 +03001684}
Aneesh Kumar K.V2ba9f0d2013-10-07 22:17:59 +05301685EXPORT_SYMBOL_GPL(kvm_vcpu_block);
Eddie Dongb6958ce2007-07-18 12:15:21 +03001686
Marcelo Tosatti8c847802012-03-14 17:58:48 -03001687#ifndef CONFIG_S390
Christoffer Dallb6d33832012-03-08 16:44:24 -05001688/*
1689 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1690 */
1691void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1692{
1693 int me;
1694 int cpu = vcpu->cpu;
1695 wait_queue_head_t *wqp;
1696
1697 wqp = kvm_arch_vcpu_wq(vcpu);
1698 if (waitqueue_active(wqp)) {
1699 wake_up_interruptible(wqp);
1700 ++vcpu->stat.halt_wakeup;
1701 }
1702
1703 me = get_cpu();
1704 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1705 if (kvm_arch_vcpu_should_kick(vcpu))
1706 smp_send_reschedule(cpu);
1707 put_cpu();
1708}
Yang Zhanga20ed542013-04-11 19:25:15 +08001709EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
Marcelo Tosatti8c847802012-03-14 17:58:48 -03001710#endif /* !CONFIG_S390 */
Christoffer Dallb6d33832012-03-08 16:44:24 -05001711
Avi Kivity6aa8b732006-12-10 02:21:36 -08001712void kvm_resched(struct kvm_vcpu *vcpu)
1713{
Yaozu Dong3fca0362007-04-25 16:49:19 +03001714 if (!need_resched())
1715 return;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001716 cond_resched();
Avi Kivity6aa8b732006-12-10 02:21:36 -08001717}
1718EXPORT_SYMBOL_GPL(kvm_resched);
1719
Konstantin Weitz41628d32012-04-25 15:30:38 +02001720bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1721{
1722 struct pid *pid;
1723 struct task_struct *task = NULL;
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301724 bool ret = false;
Konstantin Weitz41628d32012-04-25 15:30:38 +02001725
1726 rcu_read_lock();
1727 pid = rcu_dereference(target->pid);
1728 if (pid)
1729 task = get_pid_task(target->pid, PIDTYPE_PID);
1730 rcu_read_unlock();
1731 if (!task)
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301732 return ret;
Konstantin Weitz41628d32012-04-25 15:30:38 +02001733 if (task->flags & PF_VCPU) {
1734 put_task_struct(task);
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301735 return ret;
Konstantin Weitz41628d32012-04-25 15:30:38 +02001736 }
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301737 ret = yield_to(task, 1);
Konstantin Weitz41628d32012-04-25 15:30:38 +02001738 put_task_struct(task);
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301739
1740 return ret;
Konstantin Weitz41628d32012-04-25 15:30:38 +02001741}
1742EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1743
Raghavendra K T06e48c52012-07-19 15:17:52 +05301744#ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1745/*
1746 * Helper that checks whether a VCPU is eligible for directed yield.
1747 * Most eligible candidate to yield is decided by following heuristics:
1748 *
1749 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1750 * (preempted lock holder), indicated by @in_spin_loop.
1751 * Set at the beiginning and cleared at the end of interception/PLE handler.
1752 *
1753 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1754 * chance last time (mostly it has become eligible now since we have probably
1755 * yielded to lockholder in last iteration. This is done by toggling
1756 * @dy_eligible each time a VCPU checked for eligibility.)
1757 *
1758 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1759 * to preempted lock-holder could result in wrong VCPU selection and CPU
1760 * burning. Giving priority for a potential lock-holder increases lock
1761 * progress.
1762 *
1763 * Since algorithm is based on heuristics, accessing another VCPU data without
1764 * locking does not harm. It may result in trying to yield to same VCPU, fail
1765 * and continue with next VCPU and so on.
1766 */
1767bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1768{
1769 bool eligible;
1770
1771 eligible = !vcpu->spin_loop.in_spin_loop ||
1772 (vcpu->spin_loop.in_spin_loop &&
1773 vcpu->spin_loop.dy_eligible);
1774
1775 if (vcpu->spin_loop.in_spin_loop)
1776 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1777
1778 return eligible;
1779}
1780#endif
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301781
Rik van Riel217ece62011-02-01 09:53:28 -05001782void kvm_vcpu_on_spin(struct kvm_vcpu *me)
Zhai, Edwind255f4f2009-10-09 18:03:20 +08001783{
Rik van Riel217ece62011-02-01 09:53:28 -05001784 struct kvm *kvm = me->kvm;
1785 struct kvm_vcpu *vcpu;
1786 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1787 int yielded = 0;
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301788 int try = 3;
Rik van Riel217ece62011-02-01 09:53:28 -05001789 int pass;
1790 int i;
Zhai, Edwind255f4f2009-10-09 18:03:20 +08001791
Raghavendra K T4c088492012-07-18 19:07:46 +05301792 kvm_vcpu_set_in_spin_loop(me, true);
Rik van Riel217ece62011-02-01 09:53:28 -05001793 /*
1794 * We boost the priority of a VCPU that is runnable but not
1795 * currently running, because it got preempted by something
1796 * else and called schedule in __vcpu_run. Hopefully that
1797 * VCPU is holding the lock that we need and will release it.
1798 * We approximate round-robin by starting at the last boosted VCPU.
1799 */
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301800 for (pass = 0; pass < 2 && !yielded && try; pass++) {
Rik van Riel217ece62011-02-01 09:53:28 -05001801 kvm_for_each_vcpu(i, vcpu, kvm) {
Rik van Riel5cfc2aa2012-06-19 16:51:04 -04001802 if (!pass && i <= last_boosted_vcpu) {
Rik van Riel217ece62011-02-01 09:53:28 -05001803 i = last_boosted_vcpu;
1804 continue;
1805 } else if (pass && i > last_boosted_vcpu)
1806 break;
Raghavendra K T7bc7ae22013-03-04 23:32:27 +05301807 if (!ACCESS_ONCE(vcpu->preempted))
1808 continue;
Rik van Riel217ece62011-02-01 09:53:28 -05001809 if (vcpu == me)
1810 continue;
1811 if (waitqueue_active(&vcpu->wq))
1812 continue;
Raghavendra K T06e48c52012-07-19 15:17:52 +05301813 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1814 continue;
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301815
1816 yielded = kvm_vcpu_yield_to(vcpu);
1817 if (yielded > 0) {
Rik van Riel217ece62011-02-01 09:53:28 -05001818 kvm->last_boosted_vcpu = i;
Rik van Riel217ece62011-02-01 09:53:28 -05001819 break;
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301820 } else if (yielded < 0) {
1821 try--;
1822 if (!try)
1823 break;
Rik van Riel217ece62011-02-01 09:53:28 -05001824 }
Rik van Riel217ece62011-02-01 09:53:28 -05001825 }
1826 }
Raghavendra K T4c088492012-07-18 19:07:46 +05301827 kvm_vcpu_set_in_spin_loop(me, false);
Raghavendra K T06e48c52012-07-19 15:17:52 +05301828
1829 /* Ensure vcpu is not eligible during next spinloop */
1830 kvm_vcpu_set_dy_eligible(me, false);
Zhai, Edwind255f4f2009-10-09 18:03:20 +08001831}
1832EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1833
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001834static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001835{
1836 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001837 struct page *page;
1838
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001839 if (vmf->pgoff == 0)
Avi Kivity039576c2007-03-20 12:46:50 +02001840 page = virt_to_page(vcpu->run);
Avi Kivity09566762008-01-23 18:14:23 +02001841#ifdef CONFIG_X86
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001842 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
Zhang Xiantaoad312c72007-12-13 23:50:52 +08001843 page = virt_to_page(vcpu->arch.pio_data);
Avi Kivity09566762008-01-23 18:14:23 +02001844#endif
Laurent Vivier5f94c172008-05-30 16:05:54 +02001845#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1846 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1847 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1848#endif
Avi Kivity039576c2007-03-20 12:46:50 +02001849 else
Carsten Otte5b1c1492012-01-04 10:25:23 +01001850 return kvm_arch_vcpu_fault(vcpu, vmf);
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001851 get_page(page);
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001852 vmf->page = page;
1853 return 0;
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001854}
1855
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04001856static const struct vm_operations_struct kvm_vcpu_vm_ops = {
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001857 .fault = kvm_vcpu_fault,
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001858};
1859
1860static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1861{
1862 vma->vm_ops = &kvm_vcpu_vm_ops;
1863 return 0;
1864}
1865
Avi Kivitybccf2152007-02-21 18:04:26 +02001866static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1867{
1868 struct kvm_vcpu *vcpu = filp->private_data;
1869
Al Viro66c0b392008-04-19 20:33:56 +01001870 kvm_put_kvm(vcpu->kvm);
Avi Kivitybccf2152007-02-21 18:04:26 +02001871 return 0;
1872}
1873
Christian Borntraeger3d3aab12008-12-02 11:17:32 +01001874static struct file_operations kvm_vcpu_fops = {
Avi Kivitybccf2152007-02-21 18:04:26 +02001875 .release = kvm_vcpu_release,
1876 .unlocked_ioctl = kvm_vcpu_ioctl,
Alexander Graf1dda6062011-06-08 02:45:37 +02001877#ifdef CONFIG_COMPAT
1878 .compat_ioctl = kvm_vcpu_compat_ioctl,
1879#endif
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001880 .mmap = kvm_vcpu_mmap,
Arnd Bergmann6038f372010-08-15 18:52:59 +02001881 .llseek = noop_llseek,
Avi Kivitybccf2152007-02-21 18:04:26 +02001882};
1883
1884/*
1885 * Allocates an inode for the vcpu.
1886 */
1887static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1888{
Yann Droneaud24009b02013-08-24 22:14:07 +02001889 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
Avi Kivitybccf2152007-02-21 18:04:26 +02001890}
1891
Avi Kivityc5ea7662007-02-20 18:41:05 +02001892/*
1893 * Creates some virtual cpus. Good luck creating more than one.
1894 */
Gleb Natapov73880c82009-06-09 15:56:28 +03001895static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
Avi Kivityc5ea7662007-02-20 18:41:05 +02001896{
1897 int r;
Gleb Natapov988a2ca2009-06-09 15:56:29 +03001898 struct kvm_vcpu *vcpu, *v;
Avi Kivityc5ea7662007-02-20 18:41:05 +02001899
Gleb Natapov73880c82009-06-09 15:56:28 +03001900 vcpu = kvm_arch_vcpu_create(kvm, id);
Rusty Russellfb3f0f52007-07-27 17:16:56 +10001901 if (IS_ERR(vcpu))
1902 return PTR_ERR(vcpu);
Avi Kivityc5ea7662007-02-20 18:41:05 +02001903
Avi Kivity15ad7142007-07-11 18:17:21 +03001904 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1905
Avi Kivity26e52152007-11-20 15:30:24 +02001906 r = kvm_arch_vcpu_setup(vcpu);
1907 if (r)
Jan Kiszkad7805922011-05-23 10:33:05 +02001908 goto vcpu_destroy;
Avi Kivity26e52152007-11-20 15:30:24 +02001909
Shaohua Li11ec2802007-07-23 14:51:37 +08001910 mutex_lock(&kvm->lock);
Avi Kivity3e515702012-03-05 14:23:29 +02001911 if (!kvm_vcpu_compatible(vcpu)) {
1912 r = -EINVAL;
1913 goto unlock_vcpu_destroy;
1914 }
Gleb Natapov73880c82009-06-09 15:56:28 +03001915 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1916 r = -EINVAL;
Jan Kiszkad7805922011-05-23 10:33:05 +02001917 goto unlock_vcpu_destroy;
Rusty Russellfb3f0f52007-07-27 17:16:56 +10001918 }
Gleb Natapov73880c82009-06-09 15:56:28 +03001919
Gleb Natapov988a2ca2009-06-09 15:56:29 +03001920 kvm_for_each_vcpu(r, v, kvm)
1921 if (v->vcpu_id == id) {
Gleb Natapov73880c82009-06-09 15:56:28 +03001922 r = -EEXIST;
Jan Kiszkad7805922011-05-23 10:33:05 +02001923 goto unlock_vcpu_destroy;
Gleb Natapov73880c82009-06-09 15:56:28 +03001924 }
1925
1926 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
Rusty Russellfb3f0f52007-07-27 17:16:56 +10001927
1928 /* Now it's all set up, let userspace reach it */
Al Viro66c0b392008-04-19 20:33:56 +01001929 kvm_get_kvm(kvm);
Avi Kivitybccf2152007-02-21 18:04:26 +02001930 r = create_vcpu_fd(vcpu);
Gleb Natapov73880c82009-06-09 15:56:28 +03001931 if (r < 0) {
1932 kvm_put_kvm(kvm);
Jan Kiszkad7805922011-05-23 10:33:05 +02001933 goto unlock_vcpu_destroy;
Gleb Natapov73880c82009-06-09 15:56:28 +03001934 }
1935
1936 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1937 smp_wmb();
1938 atomic_inc(&kvm->online_vcpus);
1939
Gleb Natapov73880c82009-06-09 15:56:28 +03001940 mutex_unlock(&kvm->lock);
Marcelo Tosatti42897d82012-11-27 23:29:02 -02001941 kvm_arch_vcpu_postcreate(vcpu);
Avi Kivitybccf2152007-02-21 18:04:26 +02001942 return r;
Avi Kivityc5ea7662007-02-20 18:41:05 +02001943
Jan Kiszkad7805922011-05-23 10:33:05 +02001944unlock_vcpu_destroy:
Glauber Costa7d8fece2008-09-17 23:16:59 -03001945 mutex_unlock(&kvm->lock);
Jan Kiszkad7805922011-05-23 10:33:05 +02001946vcpu_destroy:
Hollis Blanchardd40ccc62007-11-19 14:04:43 -06001947 kvm_arch_vcpu_destroy(vcpu);
Avi Kivityc5ea7662007-02-20 18:41:05 +02001948 return r;
1949}
1950
Avi Kivity1961d272007-03-05 19:46:05 +02001951static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1952{
1953 if (sigset) {
1954 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1955 vcpu->sigset_active = 1;
1956 vcpu->sigset = *sigset;
1957 } else
1958 vcpu->sigset_active = 0;
1959 return 0;
1960}
1961
Avi Kivitybccf2152007-02-21 18:04:26 +02001962static long kvm_vcpu_ioctl(struct file *filp,
1963 unsigned int ioctl, unsigned long arg)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001964{
Avi Kivitybccf2152007-02-21 18:04:26 +02001965 struct kvm_vcpu *vcpu = filp->private_data;
Al Viro2f3669872007-02-09 16:38:35 +00001966 void __user *argp = (void __user *)arg;
Carsten Otte313a3dc2007-10-11 19:16:52 +02001967 int r;
Dave Hansenfa3795a2008-08-11 10:01:46 -07001968 struct kvm_fpu *fpu = NULL;
1969 struct kvm_sregs *kvm_sregs = NULL;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001970
Avi Kivity6d4e4c42007-11-21 16:41:05 +02001971 if (vcpu->kvm->mm != current->mm)
1972 return -EIO;
Avi Kivity2122ff52010-05-13 11:25:04 +03001973
Sanjay Lal2f4d9b52012-11-21 18:34:15 -08001974#if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
Avi Kivity2122ff52010-05-13 11:25:04 +03001975 /*
1976 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1977 * so vcpu_load() would break it.
1978 */
1979 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1980 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1981#endif
1982
1983
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +03001984 r = vcpu_load(vcpu);
1985 if (r)
1986 return r;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001987 switch (ioctl) {
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001988 case KVM_RUN:
Avi Kivityf0fe5102007-03-07 13:11:17 +02001989 r = -EINVAL;
1990 if (arg)
1991 goto out;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05001992 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
Gleb Natapov64be5002010-10-24 16:49:08 +02001993 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001994 break;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001995 case KVM_GET_REGS: {
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08001996 struct kvm_regs *kvm_regs;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001997
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08001998 r = -ENOMEM;
1999 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2000 if (!kvm_regs)
2001 goto out;
2002 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002003 if (r)
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002004 goto out_free1;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002005 r = -EFAULT;
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002006 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2007 goto out_free1;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002008 r = 0;
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002009out_free1:
2010 kfree(kvm_regs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002011 break;
2012 }
2013 case KVM_SET_REGS: {
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002014 struct kvm_regs *kvm_regs;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002015
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002016 r = -ENOMEM;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002017 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2018 if (IS_ERR(kvm_regs)) {
2019 r = PTR_ERR(kvm_regs);
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002020 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002021 }
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002022 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002023 kfree(kvm_regs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002024 break;
2025 }
2026 case KVM_GET_SREGS: {
Dave Hansenfa3795a2008-08-11 10:01:46 -07002027 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2028 r = -ENOMEM;
2029 if (!kvm_sregs)
2030 goto out;
2031 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002032 if (r)
2033 goto out;
2034 r = -EFAULT;
Dave Hansenfa3795a2008-08-11 10:01:46 -07002035 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002036 goto out;
2037 r = 0;
2038 break;
2039 }
2040 case KVM_SET_SREGS: {
Sasha Levinff5c2c02011-12-04 19:36:29 +02002041 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2042 if (IS_ERR(kvm_sregs)) {
2043 r = PTR_ERR(kvm_sregs);
Guo Chao18595412012-11-02 18:33:21 +08002044 kvm_sregs = NULL;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002045 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002046 }
Dave Hansenfa3795a2008-08-11 10:01:46 -07002047 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002048 break;
2049 }
Marcelo Tosatti62d9f0d2008-04-11 13:24:45 -03002050 case KVM_GET_MP_STATE: {
2051 struct kvm_mp_state mp_state;
2052
2053 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2054 if (r)
2055 goto out;
2056 r = -EFAULT;
2057 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2058 goto out;
2059 r = 0;
2060 break;
2061 }
2062 case KVM_SET_MP_STATE: {
2063 struct kvm_mp_state mp_state;
2064
2065 r = -EFAULT;
2066 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2067 goto out;
2068 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
Marcelo Tosatti62d9f0d2008-04-11 13:24:45 -03002069 break;
2070 }
Avi Kivity6aa8b732006-12-10 02:21:36 -08002071 case KVM_TRANSLATE: {
2072 struct kvm_translation tr;
2073
2074 r = -EFAULT;
Al Viro2f3669872007-02-09 16:38:35 +00002075 if (copy_from_user(&tr, argp, sizeof tr))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002076 goto out;
Zhang Xiantao8b006792007-11-16 13:05:55 +08002077 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002078 if (r)
2079 goto out;
2080 r = -EFAULT;
Al Viro2f3669872007-02-09 16:38:35 +00002081 if (copy_to_user(argp, &tr, sizeof tr))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002082 goto out;
2083 r = 0;
2084 break;
2085 }
Jan Kiszkad0bfb942008-12-15 13:52:10 +01002086 case KVM_SET_GUEST_DEBUG: {
2087 struct kvm_guest_debug dbg;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002088
2089 r = -EFAULT;
Al Viro2f3669872007-02-09 16:38:35 +00002090 if (copy_from_user(&dbg, argp, sizeof dbg))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002091 goto out;
Jan Kiszkad0bfb942008-12-15 13:52:10 +01002092 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002093 break;
2094 }
Avi Kivity1961d272007-03-05 19:46:05 +02002095 case KVM_SET_SIGNAL_MASK: {
2096 struct kvm_signal_mask __user *sigmask_arg = argp;
2097 struct kvm_signal_mask kvm_sigmask;
2098 sigset_t sigset, *p;
2099
2100 p = NULL;
2101 if (argp) {
2102 r = -EFAULT;
2103 if (copy_from_user(&kvm_sigmask, argp,
2104 sizeof kvm_sigmask))
2105 goto out;
2106 r = -EINVAL;
2107 if (kvm_sigmask.len != sizeof sigset)
2108 goto out;
2109 r = -EFAULT;
2110 if (copy_from_user(&sigset, sigmask_arg->sigset,
2111 sizeof sigset))
2112 goto out;
2113 p = &sigset;
2114 }
Andi Kleen376d41f2010-06-10 13:10:47 +02002115 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
Avi Kivity1961d272007-03-05 19:46:05 +02002116 break;
2117 }
Avi Kivityb8836732007-04-01 16:34:31 +03002118 case KVM_GET_FPU: {
Dave Hansenfa3795a2008-08-11 10:01:46 -07002119 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2120 r = -ENOMEM;
2121 if (!fpu)
2122 goto out;
2123 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
Avi Kivityb8836732007-04-01 16:34:31 +03002124 if (r)
2125 goto out;
2126 r = -EFAULT;
Dave Hansenfa3795a2008-08-11 10:01:46 -07002127 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
Avi Kivityb8836732007-04-01 16:34:31 +03002128 goto out;
2129 r = 0;
2130 break;
2131 }
2132 case KVM_SET_FPU: {
Sasha Levinff5c2c02011-12-04 19:36:29 +02002133 fpu = memdup_user(argp, sizeof(*fpu));
2134 if (IS_ERR(fpu)) {
2135 r = PTR_ERR(fpu);
Guo Chao18595412012-11-02 18:33:21 +08002136 fpu = NULL;
Avi Kivityb8836732007-04-01 16:34:31 +03002137 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002138 }
Dave Hansenfa3795a2008-08-11 10:01:46 -07002139 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
Avi Kivityb8836732007-04-01 16:34:31 +03002140 break;
2141 }
Avi Kivitybccf2152007-02-21 18:04:26 +02002142 default:
Carsten Otte313a3dc2007-10-11 19:16:52 +02002143 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
Avi Kivitybccf2152007-02-21 18:04:26 +02002144 }
2145out:
Avi Kivity2122ff52010-05-13 11:25:04 +03002146 vcpu_put(vcpu);
Dave Hansenfa3795a2008-08-11 10:01:46 -07002147 kfree(fpu);
2148 kfree(kvm_sregs);
Avi Kivitybccf2152007-02-21 18:04:26 +02002149 return r;
2150}
2151
Alexander Graf1dda6062011-06-08 02:45:37 +02002152#ifdef CONFIG_COMPAT
2153static long kvm_vcpu_compat_ioctl(struct file *filp,
2154 unsigned int ioctl, unsigned long arg)
2155{
2156 struct kvm_vcpu *vcpu = filp->private_data;
2157 void __user *argp = compat_ptr(arg);
2158 int r;
2159
2160 if (vcpu->kvm->mm != current->mm)
2161 return -EIO;
2162
2163 switch (ioctl) {
2164 case KVM_SET_SIGNAL_MASK: {
2165 struct kvm_signal_mask __user *sigmask_arg = argp;
2166 struct kvm_signal_mask kvm_sigmask;
2167 compat_sigset_t csigset;
2168 sigset_t sigset;
2169
2170 if (argp) {
2171 r = -EFAULT;
2172 if (copy_from_user(&kvm_sigmask, argp,
2173 sizeof kvm_sigmask))
2174 goto out;
2175 r = -EINVAL;
2176 if (kvm_sigmask.len != sizeof csigset)
2177 goto out;
2178 r = -EFAULT;
2179 if (copy_from_user(&csigset, sigmask_arg->sigset,
2180 sizeof csigset))
2181 goto out;
Alan Cox760a9a32012-08-22 14:34:11 +01002182 sigset_from_compat(&sigset, &csigset);
2183 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2184 } else
2185 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
Alexander Graf1dda6062011-06-08 02:45:37 +02002186 break;
2187 }
2188 default:
2189 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2190 }
2191
2192out:
2193 return r;
2194}
2195#endif
2196
Scott Wood852b6d52013-04-12 14:08:42 +00002197static int kvm_device_ioctl_attr(struct kvm_device *dev,
2198 int (*accessor)(struct kvm_device *dev,
2199 struct kvm_device_attr *attr),
2200 unsigned long arg)
2201{
2202 struct kvm_device_attr attr;
2203
2204 if (!accessor)
2205 return -EPERM;
2206
2207 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2208 return -EFAULT;
2209
2210 return accessor(dev, &attr);
2211}
2212
2213static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2214 unsigned long arg)
2215{
2216 struct kvm_device *dev = filp->private_data;
2217
2218 switch (ioctl) {
2219 case KVM_SET_DEVICE_ATTR:
2220 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2221 case KVM_GET_DEVICE_ATTR:
2222 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2223 case KVM_HAS_DEVICE_ATTR:
2224 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2225 default:
2226 if (dev->ops->ioctl)
2227 return dev->ops->ioctl(dev, ioctl, arg);
2228
2229 return -ENOTTY;
2230 }
2231}
2232
Scott Wood852b6d52013-04-12 14:08:42 +00002233static int kvm_device_release(struct inode *inode, struct file *filp)
2234{
2235 struct kvm_device *dev = filp->private_data;
2236 struct kvm *kvm = dev->kvm;
2237
Scott Wood852b6d52013-04-12 14:08:42 +00002238 kvm_put_kvm(kvm);
2239 return 0;
2240}
2241
2242static const struct file_operations kvm_device_fops = {
2243 .unlocked_ioctl = kvm_device_ioctl,
Scott Wooddb6ae612013-04-30 20:00:45 -05002244#ifdef CONFIG_COMPAT
2245 .compat_ioctl = kvm_device_ioctl,
2246#endif
Scott Wood852b6d52013-04-12 14:08:42 +00002247 .release = kvm_device_release,
2248};
2249
2250struct kvm_device *kvm_device_from_filp(struct file *filp)
2251{
2252 if (filp->f_op != &kvm_device_fops)
2253 return NULL;
2254
2255 return filp->private_data;
2256}
2257
2258static int kvm_ioctl_create_device(struct kvm *kvm,
2259 struct kvm_create_device *cd)
2260{
2261 struct kvm_device_ops *ops = NULL;
2262 struct kvm_device *dev;
2263 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2264 int ret;
2265
2266 switch (cd->type) {
Scott Wood5df554a2013-04-12 14:08:46 +00002267#ifdef CONFIG_KVM_MPIC
2268 case KVM_DEV_TYPE_FSL_MPIC_20:
2269 case KVM_DEV_TYPE_FSL_MPIC_42:
2270 ops = &kvm_mpic_ops;
2271 break;
2272#endif
Paul Mackerras5975a2e2013-04-27 00:28:37 +00002273#ifdef CONFIG_KVM_XICS
2274 case KVM_DEV_TYPE_XICS:
2275 ops = &kvm_xics_ops;
2276 break;
2277#endif
Alex Williamsonec535002013-10-30 11:02:17 -06002278#ifdef CONFIG_KVM_VFIO
2279 case KVM_DEV_TYPE_VFIO:
2280 ops = &kvm_vfio_ops;
2281 break;
2282#endif
Scott Wood852b6d52013-04-12 14:08:42 +00002283 default:
2284 return -ENODEV;
2285 }
2286
2287 if (test)
2288 return 0;
2289
2290 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2291 if (!dev)
2292 return -ENOMEM;
2293
2294 dev->ops = ops;
2295 dev->kvm = kvm;
Scott Wood852b6d52013-04-12 14:08:42 +00002296
2297 ret = ops->create(dev, cd->type);
2298 if (ret < 0) {
2299 kfree(dev);
2300 return ret;
2301 }
2302
Yann Droneaud24009b02013-08-24 22:14:07 +02002303 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
Scott Wood852b6d52013-04-12 14:08:42 +00002304 if (ret < 0) {
2305 ops->destroy(dev);
2306 return ret;
2307 }
2308
Scott Wood07f0a7b2013-04-25 14:11:23 +00002309 list_add(&dev->vm_node, &kvm->devices);
Scott Wood852b6d52013-04-12 14:08:42 +00002310 kvm_get_kvm(kvm);
2311 cd->fd = ret;
2312 return 0;
2313}
2314
Avi Kivitybccf2152007-02-21 18:04:26 +02002315static long kvm_vm_ioctl(struct file *filp,
2316 unsigned int ioctl, unsigned long arg)
2317{
2318 struct kvm *kvm = filp->private_data;
2319 void __user *argp = (void __user *)arg;
Carsten Otte1fe779f2007-10-29 16:08:35 +01002320 int r;
Avi Kivitybccf2152007-02-21 18:04:26 +02002321
Avi Kivity6d4e4c42007-11-21 16:41:05 +02002322 if (kvm->mm != current->mm)
2323 return -EIO;
Avi Kivitybccf2152007-02-21 18:04:26 +02002324 switch (ioctl) {
2325 case KVM_CREATE_VCPU:
2326 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
Avi Kivitybccf2152007-02-21 18:04:26 +02002327 break;
Izik Eidus6fc138d2007-10-09 19:20:39 +02002328 case KVM_SET_USER_MEMORY_REGION: {
2329 struct kvm_userspace_memory_region kvm_userspace_mem;
2330
2331 r = -EFAULT;
2332 if (copy_from_user(&kvm_userspace_mem, argp,
2333 sizeof kvm_userspace_mem))
2334 goto out;
2335
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +09002336 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002337 break;
2338 }
2339 case KVM_GET_DIRTY_LOG: {
2340 struct kvm_dirty_log log;
2341
2342 r = -EFAULT;
Al Viro2f3669872007-02-09 16:38:35 +00002343 if (copy_from_user(&log, argp, sizeof log))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002344 goto out;
Avi Kivity2c6f5df2007-02-20 18:27:58 +02002345 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002346 break;
2347 }
Laurent Vivier5f94c172008-05-30 16:05:54 +02002348#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2349 case KVM_REGISTER_COALESCED_MMIO: {
2350 struct kvm_coalesced_mmio_zone zone;
2351 r = -EFAULT;
2352 if (copy_from_user(&zone, argp, sizeof zone))
2353 goto out;
Laurent Vivier5f94c172008-05-30 16:05:54 +02002354 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
Laurent Vivier5f94c172008-05-30 16:05:54 +02002355 break;
2356 }
2357 case KVM_UNREGISTER_COALESCED_MMIO: {
2358 struct kvm_coalesced_mmio_zone zone;
2359 r = -EFAULT;
2360 if (copy_from_user(&zone, argp, sizeof zone))
2361 goto out;
Laurent Vivier5f94c172008-05-30 16:05:54 +02002362 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
Laurent Vivier5f94c172008-05-30 16:05:54 +02002363 break;
2364 }
2365#endif
Gregory Haskins721eecb2009-05-20 10:30:49 -04002366 case KVM_IRQFD: {
2367 struct kvm_irqfd data;
2368
2369 r = -EFAULT;
2370 if (copy_from_user(&data, argp, sizeof data))
2371 goto out;
Alex Williamsond4db2932012-06-29 09:56:08 -06002372 r = kvm_irqfd(kvm, &data);
Gregory Haskins721eecb2009-05-20 10:30:49 -04002373 break;
2374 }
Gregory Haskinsd34e6b12009-07-07 17:08:49 -04002375 case KVM_IOEVENTFD: {
2376 struct kvm_ioeventfd data;
2377
2378 r = -EFAULT;
2379 if (copy_from_user(&data, argp, sizeof data))
2380 goto out;
2381 r = kvm_ioeventfd(kvm, &data);
2382 break;
2383 }
Gleb Natapov73880c82009-06-09 15:56:28 +03002384#ifdef CONFIG_KVM_APIC_ARCHITECTURE
2385 case KVM_SET_BOOT_CPU_ID:
2386 r = 0;
Marcelo Tosatti894a9c52009-06-23 15:05:14 -03002387 mutex_lock(&kvm->lock);
Gleb Natapov73880c82009-06-09 15:56:28 +03002388 if (atomic_read(&kvm->online_vcpus) != 0)
2389 r = -EBUSY;
2390 else
2391 kvm->bsp_vcpu_id = arg;
Marcelo Tosatti894a9c52009-06-23 15:05:14 -03002392 mutex_unlock(&kvm->lock);
Gleb Natapov73880c82009-06-09 15:56:28 +03002393 break;
2394#endif
Jan Kiszka07975ad2012-03-29 21:14:12 +02002395#ifdef CONFIG_HAVE_KVM_MSI
2396 case KVM_SIGNAL_MSI: {
2397 struct kvm_msi msi;
2398
2399 r = -EFAULT;
2400 if (copy_from_user(&msi, argp, sizeof msi))
2401 goto out;
2402 r = kvm_send_userspace_msi(kvm, &msi);
2403 break;
2404 }
2405#endif
Christoffer Dall23d43cf2012-07-24 08:51:20 -04002406#ifdef __KVM_HAVE_IRQ_LINE
2407 case KVM_IRQ_LINE_STATUS:
2408 case KVM_IRQ_LINE: {
2409 struct kvm_irq_level irq_event;
2410
2411 r = -EFAULT;
2412 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2413 goto out;
2414
Yang Zhangaa2fbe62013-04-11 19:21:40 +08002415 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2416 ioctl == KVM_IRQ_LINE_STATUS);
Christoffer Dall23d43cf2012-07-24 08:51:20 -04002417 if (r)
2418 goto out;
2419
2420 r = -EFAULT;
2421 if (ioctl == KVM_IRQ_LINE_STATUS) {
2422 if (copy_to_user(argp, &irq_event, sizeof irq_event))
2423 goto out;
2424 }
2425
2426 r = 0;
2427 break;
2428 }
2429#endif
Alexander Grafaa8d5942013-04-15 21:12:53 +02002430#ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2431 case KVM_SET_GSI_ROUTING: {
2432 struct kvm_irq_routing routing;
2433 struct kvm_irq_routing __user *urouting;
2434 struct kvm_irq_routing_entry *entries;
2435
2436 r = -EFAULT;
2437 if (copy_from_user(&routing, argp, sizeof(routing)))
2438 goto out;
2439 r = -EINVAL;
2440 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2441 goto out;
2442 if (routing.flags)
2443 goto out;
2444 r = -ENOMEM;
2445 entries = vmalloc(routing.nr * sizeof(*entries));
2446 if (!entries)
2447 goto out;
2448 r = -EFAULT;
2449 urouting = argp;
2450 if (copy_from_user(entries, urouting->entries,
2451 routing.nr * sizeof(*entries)))
2452 goto out_free_irq_routing;
2453 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2454 routing.flags);
2455 out_free_irq_routing:
2456 vfree(entries);
2457 break;
2458 }
2459#endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
Scott Wood852b6d52013-04-12 14:08:42 +00002460 case KVM_CREATE_DEVICE: {
2461 struct kvm_create_device cd;
2462
2463 r = -EFAULT;
2464 if (copy_from_user(&cd, argp, sizeof(cd)))
2465 goto out;
2466
2467 r = kvm_ioctl_create_device(kvm, &cd);
2468 if (r)
2469 goto out;
2470
2471 r = -EFAULT;
2472 if (copy_to_user(argp, &cd, sizeof(cd)))
2473 goto out;
2474
2475 r = 0;
2476 break;
2477 }
Avi Kivityf17abe92007-02-21 19:28:04 +02002478 default:
Carsten Otte1fe779f2007-10-29 16:08:35 +01002479 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
Avi Kivitybfd99ff2009-08-26 14:57:50 +03002480 if (r == -ENOTTY)
2481 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
Avi Kivityf17abe92007-02-21 19:28:04 +02002482 }
2483out:
2484 return r;
2485}
2486
Arnd Bergmann6ff58942009-10-22 14:19:27 +02002487#ifdef CONFIG_COMPAT
2488struct compat_kvm_dirty_log {
2489 __u32 slot;
2490 __u32 padding1;
2491 union {
2492 compat_uptr_t dirty_bitmap; /* one bit per page */
2493 __u64 padding2;
2494 };
2495};
2496
2497static long kvm_vm_compat_ioctl(struct file *filp,
2498 unsigned int ioctl, unsigned long arg)
2499{
2500 struct kvm *kvm = filp->private_data;
2501 int r;
2502
2503 if (kvm->mm != current->mm)
2504 return -EIO;
2505 switch (ioctl) {
2506 case KVM_GET_DIRTY_LOG: {
2507 struct compat_kvm_dirty_log compat_log;
2508 struct kvm_dirty_log log;
2509
2510 r = -EFAULT;
2511 if (copy_from_user(&compat_log, (void __user *)arg,
2512 sizeof(compat_log)))
2513 goto out;
2514 log.slot = compat_log.slot;
2515 log.padding1 = compat_log.padding1;
2516 log.padding2 = compat_log.padding2;
2517 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2518
2519 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
Arnd Bergmann6ff58942009-10-22 14:19:27 +02002520 break;
2521 }
2522 default:
2523 r = kvm_vm_ioctl(filp, ioctl, arg);
2524 }
2525
2526out:
2527 return r;
2528}
2529#endif
2530
Christian Borntraeger3d3aab12008-12-02 11:17:32 +01002531static struct file_operations kvm_vm_fops = {
Avi Kivityf17abe92007-02-21 19:28:04 +02002532 .release = kvm_vm_release,
2533 .unlocked_ioctl = kvm_vm_ioctl,
Arnd Bergmann6ff58942009-10-22 14:19:27 +02002534#ifdef CONFIG_COMPAT
2535 .compat_ioctl = kvm_vm_compat_ioctl,
2536#endif
Arnd Bergmann6038f372010-08-15 18:52:59 +02002537 .llseek = noop_llseek,
Avi Kivityf17abe92007-02-21 19:28:04 +02002538};
2539
Carsten Ottee08b9632012-01-04 10:25:20 +01002540static int kvm_dev_ioctl_create_vm(unsigned long type)
Avi Kivityf17abe92007-02-21 19:28:04 +02002541{
Heiko Carstensaac87632010-10-27 17:22:10 +02002542 int r;
Avi Kivityf17abe92007-02-21 19:28:04 +02002543 struct kvm *kvm;
2544
Carsten Ottee08b9632012-01-04 10:25:20 +01002545 kvm = kvm_create_vm(type);
Avi Kivityd6d28162007-06-28 08:38:16 -04002546 if (IS_ERR(kvm))
2547 return PTR_ERR(kvm);
Takuya Yoshikawa6ce5a092010-03-15 22:13:30 +09002548#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2549 r = kvm_coalesced_mmio_init(kvm);
2550 if (r < 0) {
2551 kvm_put_kvm(kvm);
2552 return r;
2553 }
2554#endif
Yann Droneaud24009b02013-08-24 22:14:07 +02002555 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
Heiko Carstensaac87632010-10-27 17:22:10 +02002556 if (r < 0)
Al Viro66c0b392008-04-19 20:33:56 +01002557 kvm_put_kvm(kvm);
Avi Kivityf17abe92007-02-21 19:28:04 +02002558
Heiko Carstensaac87632010-10-27 17:22:10 +02002559 return r;
Avi Kivityf17abe92007-02-21 19:28:04 +02002560}
2561
Avi Kivity1a811b62008-12-08 18:25:27 +02002562static long kvm_dev_ioctl_check_extension_generic(long arg)
2563{
2564 switch (arg) {
Avi Kivityca9edae2008-12-08 18:29:29 +02002565 case KVM_CAP_USER_MEMORY:
Avi Kivity1a811b62008-12-08 18:25:27 +02002566 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
Jan Kiszka4cd481f2009-04-13 11:59:32 +02002567 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
Gleb Natapov73880c82009-06-09 15:56:28 +03002568#ifdef CONFIG_KVM_APIC_ARCHITECTURE
2569 case KVM_CAP_SET_BOOT_CPU_ID:
2570#endif
Avi Kivitya9c73992009-11-04 11:54:59 +02002571 case KVM_CAP_INTERNAL_ERROR_DATA:
Jan Kiszka07975ad2012-03-29 21:14:12 +02002572#ifdef CONFIG_HAVE_KVM_MSI
2573 case KVM_CAP_SIGNAL_MSI:
2574#endif
Alexander Graf7df35f52013-04-16 12:12:49 +02002575#ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2576 case KVM_CAP_IRQFD_RESAMPLE:
2577#endif
Avi Kivity1a811b62008-12-08 18:25:27 +02002578 return 1;
Alexander Grafa725d562013-04-17 13:29:30 +02002579#ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
Avi Kivity399ec802008-11-19 13:58:46 +02002580 case KVM_CAP_IRQ_ROUTING:
Sheng Yang36463142009-03-16 16:33:43 +08002581 return KVM_MAX_IRQ_ROUTES;
Avi Kivity399ec802008-11-19 13:58:46 +02002582#endif
Avi Kivity1a811b62008-12-08 18:25:27 +02002583 default:
2584 break;
2585 }
2586 return kvm_dev_ioctl_check_extension(arg);
2587}
2588
Avi Kivityf17abe92007-02-21 19:28:04 +02002589static long kvm_dev_ioctl(struct file *filp,
2590 unsigned int ioctl, unsigned long arg)
2591{
Avi Kivity07c45a32007-03-07 13:05:38 +02002592 long r = -EINVAL;
Avi Kivityf17abe92007-02-21 19:28:04 +02002593
2594 switch (ioctl) {
2595 case KVM_GET_API_VERSION:
Avi Kivityf0fe5102007-03-07 13:11:17 +02002596 r = -EINVAL;
2597 if (arg)
2598 goto out;
Avi Kivityf17abe92007-02-21 19:28:04 +02002599 r = KVM_API_VERSION;
2600 break;
2601 case KVM_CREATE_VM:
Carsten Ottee08b9632012-01-04 10:25:20 +01002602 r = kvm_dev_ioctl_create_vm(arg);
Avi Kivityf17abe92007-02-21 19:28:04 +02002603 break;
Zhang Xiantao018d00d2007-11-15 23:07:47 +08002604 case KVM_CHECK_EXTENSION:
Avi Kivity1a811b62008-12-08 18:25:27 +02002605 r = kvm_dev_ioctl_check_extension_generic(arg);
Avi Kivity5d308f42007-03-01 17:56:20 +02002606 break;
Avi Kivity07c45a32007-03-07 13:05:38 +02002607 case KVM_GET_VCPU_MMAP_SIZE:
2608 r = -EINVAL;
2609 if (arg)
2610 goto out;
Avi Kivityadb1ff42008-01-24 15:13:08 +02002611 r = PAGE_SIZE; /* struct kvm_run */
2612#ifdef CONFIG_X86
2613 r += PAGE_SIZE; /* pio data page */
2614#endif
Laurent Vivier5f94c172008-05-30 16:05:54 +02002615#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2616 r += PAGE_SIZE; /* coalesced mmio ring page */
2617#endif
Avi Kivity07c45a32007-03-07 13:05:38 +02002618 break;
Feng(Eric) Liud4c9ff22008-04-10 08:47:53 -04002619 case KVM_TRACE_ENABLE:
2620 case KVM_TRACE_PAUSE:
2621 case KVM_TRACE_DISABLE:
Marcelo Tosatti2023a292009-06-18 11:47:28 -03002622 r = -EOPNOTSUPP;
Feng(Eric) Liud4c9ff22008-04-10 08:47:53 -04002623 break;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002624 default:
Carsten Otte043405e2007-10-10 17:16:19 +02002625 return kvm_arch_dev_ioctl(filp, ioctl, arg);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002626 }
2627out:
2628 return r;
2629}
2630
Avi Kivity6aa8b732006-12-10 02:21:36 -08002631static struct file_operations kvm_chardev_ops = {
Avi Kivity6aa8b732006-12-10 02:21:36 -08002632 .unlocked_ioctl = kvm_dev_ioctl,
2633 .compat_ioctl = kvm_dev_ioctl,
Arnd Bergmann6038f372010-08-15 18:52:59 +02002634 .llseek = noop_llseek,
Avi Kivity6aa8b732006-12-10 02:21:36 -08002635};
2636
2637static struct miscdevice kvm_dev = {
Avi Kivitybbe44322007-03-04 13:27:36 +02002638 KVM_MINOR,
Avi Kivity6aa8b732006-12-10 02:21:36 -08002639 "kvm",
2640 &kvm_chardev_ops,
2641};
2642
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002643static void hardware_enable_nolock(void *junk)
Avi Kivity1b6c0162007-05-24 13:03:52 +03002644{
2645 int cpu = raw_smp_processor_id();
Alexander Graf10474ae2009-09-15 11:37:46 +02002646 int r;
Avi Kivity1b6c0162007-05-24 13:03:52 +03002647
Rusty Russell7f59f492008-12-07 21:25:45 +10302648 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
Avi Kivity1b6c0162007-05-24 13:03:52 +03002649 return;
Alexander Graf10474ae2009-09-15 11:37:46 +02002650
Rusty Russell7f59f492008-12-07 21:25:45 +10302651 cpumask_set_cpu(cpu, cpus_hardware_enabled);
Alexander Graf10474ae2009-09-15 11:37:46 +02002652
2653 r = kvm_arch_hardware_enable(NULL);
2654
2655 if (r) {
2656 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2657 atomic_inc(&hardware_enable_failed);
2658 printk(KERN_INFO "kvm: enabling virtualization on "
2659 "CPU%d failed\n", cpu);
2660 }
Avi Kivity1b6c0162007-05-24 13:03:52 +03002661}
2662
Paolo Bonzini4fa92fb2013-09-10 12:57:17 +02002663static void hardware_enable(void)
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002664{
Paolo Bonzini4a937f92013-09-10 12:58:35 +02002665 raw_spin_lock(&kvm_count_lock);
Paolo Bonzini4fa92fb2013-09-10 12:57:17 +02002666 if (kvm_usage_count)
2667 hardware_enable_nolock(NULL);
Paolo Bonzini4a937f92013-09-10 12:58:35 +02002668 raw_spin_unlock(&kvm_count_lock);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002669}
2670
2671static void hardware_disable_nolock(void *junk)
Avi Kivity1b6c0162007-05-24 13:03:52 +03002672{
2673 int cpu = raw_smp_processor_id();
2674
Rusty Russell7f59f492008-12-07 21:25:45 +10302675 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
Avi Kivity1b6c0162007-05-24 13:03:52 +03002676 return;
Rusty Russell7f59f492008-12-07 21:25:45 +10302677 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08002678 kvm_arch_hardware_disable(NULL);
Avi Kivity1b6c0162007-05-24 13:03:52 +03002679}
2680
Paolo Bonzini4fa92fb2013-09-10 12:57:17 +02002681static void hardware_disable(void)
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002682{
Paolo Bonzini4a937f92013-09-10 12:58:35 +02002683 raw_spin_lock(&kvm_count_lock);
Paolo Bonzini4fa92fb2013-09-10 12:57:17 +02002684 if (kvm_usage_count)
2685 hardware_disable_nolock(NULL);
Paolo Bonzini4a937f92013-09-10 12:58:35 +02002686 raw_spin_unlock(&kvm_count_lock);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002687}
2688
Alexander Graf10474ae2009-09-15 11:37:46 +02002689static void hardware_disable_all_nolock(void)
2690{
2691 BUG_ON(!kvm_usage_count);
2692
2693 kvm_usage_count--;
2694 if (!kvm_usage_count)
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002695 on_each_cpu(hardware_disable_nolock, NULL, 1);
Alexander Graf10474ae2009-09-15 11:37:46 +02002696}
2697
2698static void hardware_disable_all(void)
2699{
Paolo Bonzini4a937f92013-09-10 12:58:35 +02002700 raw_spin_lock(&kvm_count_lock);
Alexander Graf10474ae2009-09-15 11:37:46 +02002701 hardware_disable_all_nolock();
Paolo Bonzini4a937f92013-09-10 12:58:35 +02002702 raw_spin_unlock(&kvm_count_lock);
Alexander Graf10474ae2009-09-15 11:37:46 +02002703}
2704
2705static int hardware_enable_all(void)
2706{
2707 int r = 0;
2708
Paolo Bonzini4a937f92013-09-10 12:58:35 +02002709 raw_spin_lock(&kvm_count_lock);
Alexander Graf10474ae2009-09-15 11:37:46 +02002710
2711 kvm_usage_count++;
2712 if (kvm_usage_count == 1) {
2713 atomic_set(&hardware_enable_failed, 0);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002714 on_each_cpu(hardware_enable_nolock, NULL, 1);
Alexander Graf10474ae2009-09-15 11:37:46 +02002715
2716 if (atomic_read(&hardware_enable_failed)) {
2717 hardware_disable_all_nolock();
2718 r = -EBUSY;
2719 }
2720 }
2721
Paolo Bonzini4a937f92013-09-10 12:58:35 +02002722 raw_spin_unlock(&kvm_count_lock);
Alexander Graf10474ae2009-09-15 11:37:46 +02002723
2724 return r;
2725}
2726
Avi Kivity774c47f2007-02-12 00:54:47 -08002727static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2728 void *v)
2729{
2730 int cpu = (long)v;
2731
Avi Kivity1a6f4d72007-11-11 18:37:32 +02002732 val &= ~CPU_TASKS_FROZEN;
Avi Kivity774c47f2007-02-12 00:54:47 -08002733 switch (val) {
Avi Kivitycec9ad22007-05-24 13:11:41 +03002734 case CPU_DYING:
Avi Kivity6ec8a8562007-08-19 15:57:26 +03002735 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2736 cpu);
Paolo Bonzini4fa92fb2013-09-10 12:57:17 +02002737 hardware_disable();
Avi Kivity6ec8a8562007-08-19 15:57:26 +03002738 break;
Zachary Amsdenda908f22010-08-19 22:07:27 -10002739 case CPU_STARTING:
Jeremy Katz43934a32007-02-19 14:37:46 +02002740 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2741 cpu);
Paolo Bonzini4fa92fb2013-09-10 12:57:17 +02002742 hardware_enable();
Avi Kivity774c47f2007-02-12 00:54:47 -08002743 break;
2744 }
2745 return NOTIFY_OK;
2746}
2747
Rusty Russell9a2b85c2007-07-17 23:17:55 +10002748static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
Mike Dayd77c26f2007-10-08 09:02:08 -04002749 void *v)
Rusty Russell9a2b85c2007-07-17 23:17:55 +10002750{
Sheng Yang8e1c1812009-04-29 11:09:04 +08002751 /*
2752 * Some (well, at least mine) BIOSes hang on reboot if
2753 * in vmx root mode.
2754 *
2755 * And Intel TXT required VMX off for all cpu when system shutdown.
2756 */
2757 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2758 kvm_rebooting = true;
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002759 on_each_cpu(hardware_disable_nolock, NULL, 1);
Rusty Russell9a2b85c2007-07-17 23:17:55 +10002760 return NOTIFY_OK;
2761}
2762
2763static struct notifier_block kvm_reboot_notifier = {
2764 .notifier_call = kvm_reboot,
2765 .priority = 0,
2766};
2767
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002768static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002769{
2770 int i;
2771
2772 for (i = 0; i < bus->dev_count; i++) {
Sasha Levin743eeb02011-07-27 16:00:48 +03002773 struct kvm_io_device *pos = bus->range[i].dev;
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002774
2775 kvm_iodevice_destructor(pos);
2776 }
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002777 kfree(bus);
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002778}
2779
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02002780static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
2781 const struct kvm_io_range *r2)
Sasha Levin743eeb02011-07-27 16:00:48 +03002782{
Sasha Levin743eeb02011-07-27 16:00:48 +03002783 if (r1->addr < r2->addr)
2784 return -1;
2785 if (r1->addr + r1->len > r2->addr + r2->len)
2786 return 1;
2787 return 0;
2788}
2789
Paolo Bonzinia343c9b2013-07-16 13:03:29 +02002790static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2791{
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02002792 return kvm_io_bus_cmp(p1, p2);
Paolo Bonzinia343c9b2013-07-16 13:03:29 +02002793}
2794
Geoff Levand39369f72013-04-05 19:20:30 +00002795static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
Sasha Levin743eeb02011-07-27 16:00:48 +03002796 gpa_t addr, int len)
2797{
Sasha Levin743eeb02011-07-27 16:00:48 +03002798 bus->range[bus->dev_count++] = (struct kvm_io_range) {
2799 .addr = addr,
2800 .len = len,
2801 .dev = dev,
2802 };
2803
2804 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2805 kvm_io_bus_sort_cmp, NULL);
2806
2807 return 0;
2808}
2809
Geoff Levand39369f72013-04-05 19:20:30 +00002810static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
Sasha Levin743eeb02011-07-27 16:00:48 +03002811 gpa_t addr, int len)
2812{
2813 struct kvm_io_range *range, key;
2814 int off;
2815
2816 key = (struct kvm_io_range) {
2817 .addr = addr,
2818 .len = len,
2819 };
2820
2821 range = bsearch(&key, bus->range, bus->dev_count,
2822 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2823 if (range == NULL)
2824 return -ENOENT;
2825
2826 off = range - bus->range;
2827
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02002828 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
Sasha Levin743eeb02011-07-27 16:00:48 +03002829 off--;
2830
2831 return off;
2832}
2833
Cornelia Huck126a5af2013-07-03 16:30:53 +02002834static int __kvm_io_bus_write(struct kvm_io_bus *bus,
2835 struct kvm_io_range *range, const void *val)
2836{
2837 int idx;
2838
2839 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2840 if (idx < 0)
2841 return -EOPNOTSUPP;
2842
2843 while (idx < bus->dev_count &&
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02002844 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
Cornelia Huck126a5af2013-07-03 16:30:53 +02002845 if (!kvm_iodevice_write(bus->range[idx].dev, range->addr,
2846 range->len, val))
2847 return idx;
2848 idx++;
2849 }
2850
2851 return -EOPNOTSUPP;
2852}
2853
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002854/* kvm_io_bus_write - called under kvm->slots_lock */
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002855int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002856 int len, const void *val)
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002857{
Cornelia Huck126a5af2013-07-03 16:30:53 +02002858 struct kvm_io_bus *bus;
2859 struct kvm_io_range range;
2860 int r;
2861
2862 range = (struct kvm_io_range) {
2863 .addr = addr,
2864 .len = len,
2865 };
2866
2867 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2868 r = __kvm_io_bus_write(bus, &range, val);
2869 return r < 0 ? r : 0;
2870}
2871
2872/* kvm_io_bus_write_cookie - called under kvm->slots_lock */
2873int kvm_io_bus_write_cookie(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2874 int len, const void *val, long cookie)
2875{
Lai Jiangshan90d83dc2010-04-19 17:41:23 +08002876 struct kvm_io_bus *bus;
Sasha Levin743eeb02011-07-27 16:00:48 +03002877 struct kvm_io_range range;
2878
2879 range = (struct kvm_io_range) {
2880 .addr = addr,
2881 .len = len,
2882 };
Lai Jiangshan90d83dc2010-04-19 17:41:23 +08002883
2884 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
Cornelia Huck126a5af2013-07-03 16:30:53 +02002885
2886 /* First try the device referenced by cookie. */
2887 if ((cookie >= 0) && (cookie < bus->dev_count) &&
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02002888 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
Cornelia Huck126a5af2013-07-03 16:30:53 +02002889 if (!kvm_iodevice_write(bus->range[cookie].dev, addr, len,
2890 val))
2891 return cookie;
2892
2893 /*
2894 * cookie contained garbage; fall back to search and return the
2895 * correct cookie value.
2896 */
2897 return __kvm_io_bus_write(bus, &range, val);
2898}
2899
2900static int __kvm_io_bus_read(struct kvm_io_bus *bus, struct kvm_io_range *range,
2901 void *val)
2902{
2903 int idx;
2904
2905 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
Sasha Levin743eeb02011-07-27 16:00:48 +03002906 if (idx < 0)
2907 return -EOPNOTSUPP;
2908
2909 while (idx < bus->dev_count &&
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02002910 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
Cornelia Huck126a5af2013-07-03 16:30:53 +02002911 if (!kvm_iodevice_read(bus->range[idx].dev, range->addr,
2912 range->len, val))
2913 return idx;
Sasha Levin743eeb02011-07-27 16:00:48 +03002914 idx++;
2915 }
2916
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002917 return -EOPNOTSUPP;
2918}
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002919
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002920/* kvm_io_bus_read - called under kvm->slots_lock */
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002921int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2922 int len, void *val)
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002923{
Cornelia Huck126a5af2013-07-03 16:30:53 +02002924 struct kvm_io_bus *bus;
2925 struct kvm_io_range range;
2926 int r;
2927
2928 range = (struct kvm_io_range) {
2929 .addr = addr,
2930 .len = len,
2931 };
2932
2933 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2934 r = __kvm_io_bus_read(bus, &range, val);
2935 return r < 0 ? r : 0;
2936}
2937
2938/* kvm_io_bus_read_cookie - called under kvm->slots_lock */
2939int kvm_io_bus_read_cookie(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2940 int len, void *val, long cookie)
2941{
Lai Jiangshan90d83dc2010-04-19 17:41:23 +08002942 struct kvm_io_bus *bus;
Sasha Levin743eeb02011-07-27 16:00:48 +03002943 struct kvm_io_range range;
2944
2945 range = (struct kvm_io_range) {
2946 .addr = addr,
2947 .len = len,
2948 };
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002949
Lai Jiangshan90d83dc2010-04-19 17:41:23 +08002950 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
Sasha Levin743eeb02011-07-27 16:00:48 +03002951
Cornelia Huck126a5af2013-07-03 16:30:53 +02002952 /* First try the device referenced by cookie. */
2953 if ((cookie >= 0) && (cookie < bus->dev_count) &&
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02002954 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
Cornelia Huck126a5af2013-07-03 16:30:53 +02002955 if (!kvm_iodevice_read(bus->range[cookie].dev, addr, len,
2956 val))
2957 return cookie;
Sasha Levin743eeb02011-07-27 16:00:48 +03002958
Cornelia Huck126a5af2013-07-03 16:30:53 +02002959 /*
2960 * cookie contained garbage; fall back to search and return the
2961 * correct cookie value.
2962 */
2963 return __kvm_io_bus_read(bus, &range, val);
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002964}
2965
Marcelo Tosatti79fac952009-12-23 14:35:26 -02002966/* Caller must hold slots_lock. */
Sasha Levin743eeb02011-07-27 16:00:48 +03002967int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2968 int len, struct kvm_io_device *dev)
Michael S. Tsirkin6c474692009-06-29 22:24:26 +03002969{
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002970 struct kvm_io_bus *new_bus, *bus;
Gregory Haskins090b7af2009-07-07 17:08:44 -04002971
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002972 bus = kvm->buses[bus_idx];
Amos Kong6ea34c92013-05-25 06:44:15 +08002973 /* exclude ioeventfd which is limited by maximum fd */
2974 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
Gregory Haskins090b7af2009-07-07 17:08:44 -04002975 return -ENOSPC;
2976
Amos Konga13007162012-03-09 12:17:32 +08002977 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2978 sizeof(struct kvm_io_range)), GFP_KERNEL);
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002979 if (!new_bus)
2980 return -ENOMEM;
Amos Konga13007162012-03-09 12:17:32 +08002981 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2982 sizeof(struct kvm_io_range)));
Sasha Levin743eeb02011-07-27 16:00:48 +03002983 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002984 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2985 synchronize_srcu_expedited(&kvm->srcu);
2986 kfree(bus);
Gregory Haskins090b7af2009-07-07 17:08:44 -04002987
2988 return 0;
2989}
2990
Marcelo Tosatti79fac952009-12-23 14:35:26 -02002991/* Caller must hold slots_lock. */
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002992int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2993 struct kvm_io_device *dev)
Gregory Haskins090b7af2009-07-07 17:08:44 -04002994{
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002995 int i, r;
2996 struct kvm_io_bus *new_bus, *bus;
Michael S. Tsirkin6c474692009-06-29 22:24:26 +03002997
Sasha Levincdfca7b2011-12-04 19:36:28 +02002998 bus = kvm->buses[bus_idx];
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002999 r = -ENOENT;
Amos Konga13007162012-03-09 12:17:32 +08003000 for (i = 0; i < bus->dev_count; i++)
3001 if (bus->range[i].dev == dev) {
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003002 r = 0;
Gregory Haskins090b7af2009-07-07 17:08:44 -04003003 break;
3004 }
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003005
Amos Konga13007162012-03-09 12:17:32 +08003006 if (r)
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003007 return r;
Amos Konga13007162012-03-09 12:17:32 +08003008
3009 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3010 sizeof(struct kvm_io_range)), GFP_KERNEL);
3011 if (!new_bus)
3012 return -ENOMEM;
3013
3014 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3015 new_bus->dev_count--;
3016 memcpy(new_bus->range + i, bus->range + i + 1,
3017 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003018
3019 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3020 synchronize_srcu_expedited(&kvm->srcu);
3021 kfree(bus);
3022 return r;
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04003023}
3024
Avi Kivity774c47f2007-02-12 00:54:47 -08003025static struct notifier_block kvm_cpu_notifier = {
3026 .notifier_call = kvm_cpu_hotplug,
Avi Kivity774c47f2007-02-12 00:54:47 -08003027};
3028
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003029static int vm_stat_get(void *_offset, u64 *val)
Avi Kivityba1389b2007-11-18 16:24:12 +02003030{
3031 unsigned offset = (long)_offset;
Avi Kivityba1389b2007-11-18 16:24:12 +02003032 struct kvm *kvm;
3033
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003034 *val = 0;
Paolo Bonzini2f303b72013-09-25 13:53:07 +02003035 spin_lock(&kvm_lock);
Avi Kivityba1389b2007-11-18 16:24:12 +02003036 list_for_each_entry(kvm, &vm_list, vm_list)
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003037 *val += *(u32 *)((void *)kvm + offset);
Paolo Bonzini2f303b72013-09-25 13:53:07 +02003038 spin_unlock(&kvm_lock);
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003039 return 0;
Avi Kivityba1389b2007-11-18 16:24:12 +02003040}
3041
3042DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3043
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003044static int vcpu_stat_get(void *_offset, u64 *val)
Avi Kivity1165f5f2007-04-19 17:27:43 +03003045{
3046 unsigned offset = (long)_offset;
Avi Kivity1165f5f2007-04-19 17:27:43 +03003047 struct kvm *kvm;
3048 struct kvm_vcpu *vcpu;
3049 int i;
3050
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003051 *val = 0;
Paolo Bonzini2f303b72013-09-25 13:53:07 +02003052 spin_lock(&kvm_lock);
Avi Kivity1165f5f2007-04-19 17:27:43 +03003053 list_for_each_entry(kvm, &vm_list, vm_list)
Gleb Natapov988a2ca2009-06-09 15:56:29 +03003054 kvm_for_each_vcpu(i, vcpu, kvm)
3055 *val += *(u32 *)((void *)vcpu + offset);
3056
Paolo Bonzini2f303b72013-09-25 13:53:07 +02003057 spin_unlock(&kvm_lock);
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003058 return 0;
Avi Kivity1165f5f2007-04-19 17:27:43 +03003059}
3060
Avi Kivityba1389b2007-11-18 16:24:12 +02003061DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3062
Alexey Dobriyan828c0952009-10-01 15:43:56 -07003063static const struct file_operations *stat_fops[] = {
Avi Kivityba1389b2007-11-18 16:24:12 +02003064 [KVM_STAT_VCPU] = &vcpu_stat_fops,
3065 [KVM_STAT_VM] = &vm_stat_fops,
3066};
Avi Kivity1165f5f2007-04-19 17:27:43 +03003067
Hamo4f69b682011-12-15 14:23:16 +08003068static int kvm_init_debug(void)
Avi Kivity6aa8b732006-12-10 02:21:36 -08003069{
Paolo Bonzini0c8eb042013-10-30 12:12:13 +01003070 int r = -EEXIST;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003071 struct kvm_stats_debugfs_item *p;
3072
Hollis Blanchard76f7c872008-04-15 16:05:42 -05003073 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
Hamo4f69b682011-12-15 14:23:16 +08003074 if (kvm_debugfs_dir == NULL)
3075 goto out;
3076
3077 for (p = debugfs_entries; p->name; ++p) {
Hollis Blanchard76f7c872008-04-15 16:05:42 -05003078 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
Avi Kivity1165f5f2007-04-19 17:27:43 +03003079 (void *)(long)p->offset,
Avi Kivityba1389b2007-11-18 16:24:12 +02003080 stat_fops[p->kind]);
Hamo4f69b682011-12-15 14:23:16 +08003081 if (p->dentry == NULL)
3082 goto out_dir;
3083 }
3084
3085 return 0;
3086
3087out_dir:
3088 debugfs_remove_recursive(kvm_debugfs_dir);
3089out:
3090 return r;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003091}
3092
3093static void kvm_exit_debug(void)
3094{
3095 struct kvm_stats_debugfs_item *p;
3096
3097 for (p = debugfs_entries; p->name; ++p)
3098 debugfs_remove(p->dentry);
Hollis Blanchard76f7c872008-04-15 16:05:42 -05003099 debugfs_remove(kvm_debugfs_dir);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003100}
3101
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003102static int kvm_suspend(void)
Avi Kivity59ae6c62007-02-12 00:54:48 -08003103{
Alexander Graf10474ae2009-09-15 11:37:46 +02003104 if (kvm_usage_count)
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09003105 hardware_disable_nolock(NULL);
Avi Kivity59ae6c62007-02-12 00:54:48 -08003106 return 0;
3107}
3108
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003109static void kvm_resume(void)
Avi Kivity59ae6c62007-02-12 00:54:48 -08003110{
Zachary Amsdenca84d1a2010-08-19 22:07:28 -10003111 if (kvm_usage_count) {
Paolo Bonzini4a937f92013-09-10 12:58:35 +02003112 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09003113 hardware_enable_nolock(NULL);
Zachary Amsdenca84d1a2010-08-19 22:07:28 -10003114 }
Avi Kivity59ae6c62007-02-12 00:54:48 -08003115}
3116
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003117static struct syscore_ops kvm_syscore_ops = {
Avi Kivity59ae6c62007-02-12 00:54:48 -08003118 .suspend = kvm_suspend,
3119 .resume = kvm_resume,
3120};
3121
Avi Kivity15ad7142007-07-11 18:17:21 +03003122static inline
3123struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3124{
3125 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3126}
3127
3128static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3129{
3130 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
Raghavendra K T3a08a8f2013-03-04 23:32:07 +05303131 if (vcpu->preempted)
3132 vcpu->preempted = false;
Avi Kivity15ad7142007-07-11 18:17:21 +03003133
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003134 kvm_arch_vcpu_load(vcpu, cpu);
Avi Kivity15ad7142007-07-11 18:17:21 +03003135}
3136
3137static void kvm_sched_out(struct preempt_notifier *pn,
3138 struct task_struct *next)
3139{
3140 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3141
Raghavendra K T3a08a8f2013-03-04 23:32:07 +05303142 if (current->state == TASK_RUNNING)
3143 vcpu->preempted = true;
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003144 kvm_arch_vcpu_put(vcpu);
Avi Kivity15ad7142007-07-11 18:17:21 +03003145}
3146
Avi Kivity0ee75be2010-04-28 15:39:01 +03003147int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
Rusty Russellc16f8622007-07-30 21:12:19 +10003148 struct module *module)
Avi Kivity6aa8b732006-12-10 02:21:36 -08003149{
3150 int r;
Yang, Sheng002c7f72007-07-31 14:23:01 +03003151 int cpu;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003152
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08003153 r = kvm_arch_init(opaque);
3154 if (r)
Zhang Xiantaod23087842007-11-29 15:35:39 +08003155 goto out_fail;
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08003156
Asias He7dac16c2013-05-08 10:57:29 +08003157 /*
3158 * kvm_arch_init makes sure there's at most one caller
3159 * for architectures that support multiple implementations,
3160 * like intel and amd on x86.
3161 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3162 * conflicts in case kvm is already setup for another implementation.
3163 */
3164 r = kvm_irqfd_init();
3165 if (r)
3166 goto out_irqfd;
3167
Avi Kivity8437a612009-06-06 14:52:35 -07003168 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
Rusty Russell7f59f492008-12-07 21:25:45 +10303169 r = -ENOMEM;
3170 goto out_free_0;
3171 }
3172
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003173 r = kvm_arch_hardware_setup();
Avi Kivity6aa8b732006-12-10 02:21:36 -08003174 if (r < 0)
Rusty Russell7f59f492008-12-07 21:25:45 +10303175 goto out_free_0a;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003176
Yang, Sheng002c7f72007-07-31 14:23:01 +03003177 for_each_online_cpu(cpu) {
3178 smp_call_function_single(cpu,
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003179 kvm_arch_check_processor_compat,
Jens Axboe8691e5a2008-06-06 11:18:06 +02003180 &r, 1);
Yang, Sheng002c7f72007-07-31 14:23:01 +03003181 if (r < 0)
Zhang Xiantaod23087842007-11-29 15:35:39 +08003182 goto out_free_1;
Yang, Sheng002c7f72007-07-31 14:23:01 +03003183 }
3184
Avi Kivity774c47f2007-02-12 00:54:47 -08003185 r = register_cpu_notifier(&kvm_cpu_notifier);
3186 if (r)
Zhang Xiantaod23087842007-11-29 15:35:39 +08003187 goto out_free_2;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003188 register_reboot_notifier(&kvm_reboot_notifier);
3189
Rusty Russellc16f8622007-07-30 21:12:19 +10003190 /* A kmem cache lets us meet the alignment requirements of fx_save. */
Avi Kivity0ee75be2010-04-28 15:39:01 +03003191 if (!vcpu_align)
3192 vcpu_align = __alignof__(struct kvm_vcpu);
3193 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
Joe Perches56919c52007-11-12 20:06:51 -08003194 0, NULL);
Rusty Russellc16f8622007-07-30 21:12:19 +10003195 if (!kvm_vcpu_cache) {
3196 r = -ENOMEM;
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003197 goto out_free_3;
Rusty Russellc16f8622007-07-30 21:12:19 +10003198 }
3199
Gleb Natapovaf585b92010-10-14 11:22:46 +02003200 r = kvm_async_pf_init();
3201 if (r)
3202 goto out_free;
3203
Avi Kivity6aa8b732006-12-10 02:21:36 -08003204 kvm_chardev_ops.owner = module;
Christian Borntraeger3d3aab12008-12-02 11:17:32 +01003205 kvm_vm_fops.owner = module;
3206 kvm_vcpu_fops.owner = module;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003207
3208 r = misc_register(&kvm_dev);
3209 if (r) {
Mike Dayd77c26f2007-10-08 09:02:08 -04003210 printk(KERN_ERR "kvm: misc device register failed\n");
Gleb Natapovaf585b92010-10-14 11:22:46 +02003211 goto out_unreg;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003212 }
3213
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003214 register_syscore_ops(&kvm_syscore_ops);
3215
Avi Kivity15ad7142007-07-11 18:17:21 +03003216 kvm_preempt_ops.sched_in = kvm_sched_in;
3217 kvm_preempt_ops.sched_out = kvm_sched_out;
3218
Hamo4f69b682011-12-15 14:23:16 +08003219 r = kvm_init_debug();
3220 if (r) {
3221 printk(KERN_ERR "kvm: create debugfs files failed\n");
3222 goto out_undebugfs;
3223 }
Darrick J. Wong0ea4ed82009-10-14 16:21:00 -07003224
Avi Kivityc7addb92007-09-16 18:58:32 +02003225 return 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003226
Hamo4f69b682011-12-15 14:23:16 +08003227out_undebugfs:
3228 unregister_syscore_ops(&kvm_syscore_ops);
Wei Yongjunafc2f792013-05-05 20:03:40 +08003229 misc_deregister(&kvm_dev);
Gleb Natapovaf585b92010-10-14 11:22:46 +02003230out_unreg:
3231 kvm_async_pf_deinit();
Avi Kivity6aa8b732006-12-10 02:21:36 -08003232out_free:
Rusty Russellc16f8622007-07-30 21:12:19 +10003233 kmem_cache_destroy(kvm_vcpu_cache);
Zhang Xiantaod23087842007-11-29 15:35:39 +08003234out_free_3:
Avi Kivity6aa8b732006-12-10 02:21:36 -08003235 unregister_reboot_notifier(&kvm_reboot_notifier);
Avi Kivity774c47f2007-02-12 00:54:47 -08003236 unregister_cpu_notifier(&kvm_cpu_notifier);
Zhang Xiantaod23087842007-11-29 15:35:39 +08003237out_free_2:
Zhang Xiantaod23087842007-11-29 15:35:39 +08003238out_free_1:
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003239 kvm_arch_hardware_unsetup();
Rusty Russell7f59f492008-12-07 21:25:45 +10303240out_free_0a:
3241 free_cpumask_var(cpus_hardware_enabled);
Zhang Xiantaod23087842007-11-29 15:35:39 +08003242out_free_0:
Cornelia Hucka0f155e2013-02-28 12:33:18 +01003243 kvm_irqfd_exit();
3244out_irqfd:
Asias He7dac16c2013-05-08 10:57:29 +08003245 kvm_arch_exit();
3246out_fail:
Avi Kivity6aa8b732006-12-10 02:21:36 -08003247 return r;
3248}
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08003249EXPORT_SYMBOL_GPL(kvm_init);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003250
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08003251void kvm_exit(void)
Avi Kivity6aa8b732006-12-10 02:21:36 -08003252{
Darrick J. Wong0ea4ed82009-10-14 16:21:00 -07003253 kvm_exit_debug();
Avi Kivity6aa8b732006-12-10 02:21:36 -08003254 misc_deregister(&kvm_dev);
Rusty Russellc16f8622007-07-30 21:12:19 +10003255 kmem_cache_destroy(kvm_vcpu_cache);
Gleb Natapovaf585b92010-10-14 11:22:46 +02003256 kvm_async_pf_deinit();
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003257 unregister_syscore_ops(&kvm_syscore_ops);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003258 unregister_reboot_notifier(&kvm_reboot_notifier);
Avi Kivity59ae6c62007-02-12 00:54:48 -08003259 unregister_cpu_notifier(&kvm_cpu_notifier);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09003260 on_each_cpu(hardware_disable_nolock, NULL, 1);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003261 kvm_arch_hardware_unsetup();
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08003262 kvm_arch_exit();
Cornelia Hucka0f155e2013-02-28 12:33:18 +01003263 kvm_irqfd_exit();
Rusty Russell7f59f492008-12-07 21:25:45 +10303264 free_cpumask_var(cpus_hardware_enabled);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003265}
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08003266EXPORT_SYMBOL_GPL(kvm_exit);