blob: a17f190be58e8d1a08636022148c5ef6b70b3138 [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>
David Matlack7889ddd2014-09-19 16:03:25 -070055#include <asm/ioctl.h>
Avi Kivitye4956062007-06-28 14:15:57 -040056#include <asm/uaccess.h>
Izik Eidus3e021bf2007-11-19 11:16:57 +020057#include <asm/pgtable.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080058
Laurent Vivier5f94c172008-05-30 16:05:54 +020059#include "coalesced_mmio.h"
Gleb Natapovaf585b92010-10-14 11:22:46 +020060#include "async_pf.h"
Laurent Vivier5f94c172008-05-30 16:05:54 +020061
Marcelo Tosatti229456f2009-06-17 09:22:14 -030062#define CREATE_TRACE_POINTS
63#include <trace/events/kvm.h>
64
Avi Kivity6aa8b732006-12-10 02:21:36 -080065MODULE_AUTHOR("Qumranet");
66MODULE_LICENSE("GPL");
67
Marcelo Tosattifa40a822009-06-04 15:08:24 -030068/*
69 * Ordering of locks:
70 *
Sheng Yangfae3a352009-12-15 10:28:07 +080071 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
Marcelo Tosattifa40a822009-06-04 15:08:24 -030072 */
73
Jan Kiszkae935b832011-02-08 12:55:33 +010074DEFINE_RAW_SPINLOCK(kvm_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{
Joerg Roedelfc5659c2009-02-18 14:08:58 +0100106 if (pfn_valid(pfn)) {
Andrea Arcangeli22e5c472011-01-13 15:47:20 -0800107 int reserved;
Andrea Arcangeli936a5fe2011-01-13 15:46:48 -0800108 struct page *tail = pfn_to_page(pfn);
David Rientjesdef52ac2014-03-03 15:38:18 -0800109 struct page *head = compound_head(tail);
Andrea Arcangeli22e5c472011-01-13 15:47:20 -0800110 reserved = PageReserved(head);
Andrea Arcangeli936a5fe2011-01-13 15:46:48 -0800111 if (head != tail) {
Andrea Arcangeli936a5fe2011-01-13 15:46:48 -0800112 /*
Andrea Arcangeli22e5c472011-01-13 15:47:20 -0800113 * "head" is not a dangling pointer
David Rientjesdef52ac2014-03-03 15:38:18 -0800114 * (compound_head takes care of that)
Andrea Arcangeli22e5c472011-01-13 15:47:20 -0800115 * but the hugepage may have been splitted
116 * from under us (and we may not hold a
117 * reference count on the head page so it can
118 * be reused before we run PageReferenced), so
119 * we've to check PageTail before returning
120 * what we just read.
Andrea Arcangeli936a5fe2011-01-13 15:46:48 -0800121 */
Andrea Arcangeli22e5c472011-01-13 15:47:20 -0800122 smp_rmb();
123 if (PageTail(tail))
124 return reserved;
Andrea Arcangeli936a5fe2011-01-13 15:46:48 -0800125 }
126 return PageReserved(tail);
Joerg Roedelfc5659c2009-02-18 14:08:58 +0100127 }
Ben-Ami Yassourcbff90a2008-07-28 19:26:24 +0300128
129 return true;
130}
131
Avi Kivity6aa8b732006-12-10 02:21:36 -0800132/*
133 * Switches to specified vcpu, until a matching vcpu_put()
134 */
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +0300135int vcpu_load(struct kvm_vcpu *vcpu)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800136{
Avi Kivity15ad7142007-07-11 18:17:21 +0300137 int cpu;
138
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +0300139 if (mutex_lock_killable(&vcpu->mutex))
140 return -EINTR;
Rik van Riel34bb10b2011-02-01 09:52:41 -0500141 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
142 /* The thread running this VCPU changed. */
143 struct pid *oldpid = vcpu->pid;
144 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
145 rcu_assign_pointer(vcpu->pid, newpid);
146 synchronize_rcu();
147 put_pid(oldpid);
148 }
Avi Kivity15ad7142007-07-11 18:17:21 +0300149 cpu = get_cpu();
150 preempt_notifier_register(&vcpu->preempt_notifier);
Carsten Otte313a3dc2007-10-11 19:16:52 +0200151 kvm_arch_vcpu_load(vcpu, cpu);
Avi Kivity15ad7142007-07-11 18:17:21 +0300152 put_cpu();
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +0300153 return 0;
Avi Kivitybccf2152007-02-21 18:04:26 +0200154}
155
Carsten Otte313a3dc2007-10-11 19:16:52 +0200156void vcpu_put(struct kvm_vcpu *vcpu)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800157{
Avi Kivity15ad7142007-07-11 18:17:21 +0300158 preempt_disable();
Carsten Otte313a3dc2007-10-11 19:16:52 +0200159 kvm_arch_vcpu_put(vcpu);
Avi Kivity15ad7142007-07-11 18:17:21 +0300160 preempt_notifier_unregister(&vcpu->preempt_notifier);
161 preempt_enable();
Avi Kivity6aa8b732006-12-10 02:21:36 -0800162 mutex_unlock(&vcpu->mutex);
163}
164
Avi Kivityd9e368d2007-06-07 19:18:30 +0300165static void ack_flush(void *_completed)
166{
Avi Kivityd9e368d2007-06-07 19:18:30 +0300167}
168
Rusty Russell49846892008-12-08 20:26:24 +1030169static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
Avi Kivityd9e368d2007-06-07 19:18:30 +0300170{
Avi Kivity597a5f52008-07-20 14:24:22 +0300171 int i, cpu, me;
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030172 cpumask_var_t cpus;
173 bool called = true;
Avi Kivityd9e368d2007-06-07 19:18:30 +0300174 struct kvm_vcpu *vcpu;
Avi Kivityd9e368d2007-06-07 19:18:30 +0300175
Li Zefan79f55992009-06-15 14:58:26 +0800176 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030177
Xiao Guangrong3cba4132011-01-12 15:41:22 +0800178 me = get_cpu();
Gleb Natapov988a2ca2009-06-09 15:56:29 +0300179 kvm_for_each_vcpu(i, vcpu, kvm) {
Xiao Guangrong3cba4132011-01-12 15:41:22 +0800180 kvm_make_request(req, vcpu);
Avi Kivityd9e368d2007-06-07 19:18:30 +0300181 cpu = vcpu->cpu;
Xiao Guangrong6b7e2d02011-01-12 15:40:31 +0800182
183 /* Set ->requests bit before we read ->mode */
184 smp_mb();
185
186 if (cpus != NULL && cpu != -1 && cpu != me &&
187 kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030188 cpumask_set_cpu(cpu, cpus);
Avi Kivityd9e368d2007-06-07 19:18:30 +0300189 }
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030190 if (unlikely(cpus == NULL))
191 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
192 else if (!cpumask_empty(cpus))
193 smp_call_function_many(cpus, ack_flush, NULL, 1);
194 else
195 called = false;
Xiao Guangrong3cba4132011-01-12 15:41:22 +0800196 put_cpu();
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030197 free_cpumask_var(cpus);
Rusty Russell49846892008-12-08 20:26:24 +1030198 return called;
199}
200
201void kvm_flush_remote_tlbs(struct kvm *kvm)
202{
Alex Shibec87d62012-03-04 13:13:38 +0800203 long dirty_count = kvm->tlbs_dirty;
Xiao Guangronga4ee1ca2010-11-23 11:13:00 +0800204
205 smp_mb();
Rusty Russell49846892008-12-08 20:26:24 +1030206 if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
207 ++kvm->stat.remote_tlb_flush;
Xiao Guangronga4ee1ca2010-11-23 11:13:00 +0800208 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
Avi Kivityd9e368d2007-06-07 19:18:30 +0300209}
210
Marcelo Tosatti2e53d632008-02-20 14:47:24 -0500211void kvm_reload_remote_mmus(struct kvm *kvm)
212{
Rusty Russell49846892008-12-08 20:26:24 +1030213 make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
Marcelo Tosatti2e53d632008-02-20 14:47:24 -0500214}
215
Marcelo Tosattid8281992012-11-27 23:29:01 -0200216void kvm_make_mclock_inprogress_request(struct kvm *kvm)
217{
218 make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
219}
220
Yang Zhang3d81bc72013-04-11 19:25:13 +0800221void kvm_make_scan_ioapic_request(struct kvm *kvm)
Yang Zhangc7c9c562013-01-25 10:18:51 +0800222{
Yang Zhang3d81bc72013-04-11 19:25:13 +0800223 make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
Yang Zhangc7c9c562013-01-25 10:18:51 +0800224}
225
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000226int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
227{
228 struct page *page;
229 int r;
230
231 mutex_init(&vcpu->mutex);
232 vcpu->cpu = -1;
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000233 vcpu->kvm = kvm;
234 vcpu->vcpu_id = id;
Rik van Riel34bb10b2011-02-01 09:52:41 -0500235 vcpu->pid = NULL;
Eddie Dongb6958ce2007-07-18 12:15:21 +0300236 init_waitqueue_head(&vcpu->wq);
Gleb Natapovaf585b92010-10-14 11:22:46 +0200237 kvm_async_pf_vcpu_init(vcpu);
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000238
239 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
240 if (!page) {
241 r = -ENOMEM;
242 goto fail;
243 }
244 vcpu->run = page_address(page);
245
Raghavendra K T4c088492012-07-18 19:07:46 +0530246 kvm_vcpu_set_in_spin_loop(vcpu, false);
247 kvm_vcpu_set_dy_eligible(vcpu, false);
Raghavendra K T3a08a8f2013-03-04 23:32:07 +0530248 vcpu->preempted = false;
Raghavendra K T4c088492012-07-18 19:07:46 +0530249
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +0800250 r = kvm_arch_vcpu_init(vcpu);
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000251 if (r < 0)
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +0800252 goto fail_free_run;
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000253 return 0;
254
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000255fail_free_run:
256 free_page((unsigned long)vcpu->run);
257fail:
Rusty Russell76fafa52007-10-08 10:50:48 +1000258 return r;
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000259}
260EXPORT_SYMBOL_GPL(kvm_vcpu_init);
261
262void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
263{
Rik van Riel34bb10b2011-02-01 09:52:41 -0500264 put_pid(vcpu->pid);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +0800265 kvm_arch_vcpu_uninit(vcpu);
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000266 free_page((unsigned long)vcpu->run);
267}
268EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
269
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200270#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
271static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
272{
273 return container_of(mn, struct kvm, mmu_notifier);
274}
275
276static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
277 struct mm_struct *mm,
278 unsigned long address)
279{
280 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200281 int need_tlb_flush, idx;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200282
283 /*
284 * When ->invalidate_page runs, the linux pte has been zapped
285 * already but the page is still allocated until
286 * ->invalidate_page returns. So if we increase the sequence
287 * here the kvm page fault will notice if the spte can't be
288 * established because the page is going to be freed. If
289 * instead the kvm page fault establishes the spte before
290 * ->invalidate_page runs, kvm_unmap_hva will release it
291 * before returning.
292 *
293 * The sequence increase only need to be seen at spin_unlock
294 * time, and not at spin_lock time.
295 *
296 * Increasing the sequence after the spin_unlock would be
297 * unsafe because the kvm page fault could then establish the
298 * pte after kvm_unmap_hva returned, without noticing the page
299 * is going to be freed.
300 */
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200301 idx = srcu_read_lock(&kvm->srcu);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200302 spin_lock(&kvm->mmu_lock);
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900303
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200304 kvm->mmu_notifier_seq++;
Xiao Guangronga4ee1ca2010-11-23 11:13:00 +0800305 need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200306 /* we've to flush the tlb before the pages can be freed */
307 if (need_tlb_flush)
308 kvm_flush_remote_tlbs(kvm);
309
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900310 spin_unlock(&kvm->mmu_lock);
311 srcu_read_unlock(&kvm->srcu, idx);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200312}
313
Izik Eidus3da0dd42009-09-23 21:47:18 +0300314static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
315 struct mm_struct *mm,
316 unsigned long address,
317 pte_t pte)
318{
319 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200320 int idx;
Izik Eidus3da0dd42009-09-23 21:47:18 +0300321
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200322 idx = srcu_read_lock(&kvm->srcu);
Izik Eidus3da0dd42009-09-23 21:47:18 +0300323 spin_lock(&kvm->mmu_lock);
324 kvm->mmu_notifier_seq++;
325 kvm_set_spte_hva(kvm, address, pte);
326 spin_unlock(&kvm->mmu_lock);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200327 srcu_read_unlock(&kvm->srcu, idx);
Izik Eidus3da0dd42009-09-23 21:47:18 +0300328}
329
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200330static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
331 struct mm_struct *mm,
332 unsigned long start,
333 unsigned long end)
334{
335 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200336 int need_tlb_flush = 0, idx;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200337
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200338 idx = srcu_read_lock(&kvm->srcu);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200339 spin_lock(&kvm->mmu_lock);
340 /*
341 * The count increase must become visible at unlock time as no
342 * spte can be established without taking the mmu_lock and
343 * count is also read inside the mmu_lock critical section.
344 */
345 kvm->mmu_notifier_count++;
Takuya Yoshikawab3ae2092012-07-02 17:56:33 +0900346 need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
Xiao Guangronga4ee1ca2010-11-23 11:13:00 +0800347 need_tlb_flush |= kvm->tlbs_dirty;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200348 /* we've to flush the tlb before the pages can be freed */
349 if (need_tlb_flush)
350 kvm_flush_remote_tlbs(kvm);
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900351
352 spin_unlock(&kvm->mmu_lock);
353 srcu_read_unlock(&kvm->srcu, idx);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200354}
355
356static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
357 struct mm_struct *mm,
358 unsigned long start,
359 unsigned long end)
360{
361 struct kvm *kvm = mmu_notifier_to_kvm(mn);
362
363 spin_lock(&kvm->mmu_lock);
364 /*
365 * This sequence increase will notify the kvm page fault that
366 * the page that is going to be mapped in the spte could have
367 * been freed.
368 */
369 kvm->mmu_notifier_seq++;
Paul Mackerrasa355aa52011-12-12 12:37:21 +0000370 smp_wmb();
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200371 /*
372 * The above sequence increase must be visible before the
Paul Mackerrasa355aa52011-12-12 12:37:21 +0000373 * below count decrease, which is ensured by the smp_wmb above
374 * in conjunction with the smp_rmb in mmu_notifier_retry().
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200375 */
376 kvm->mmu_notifier_count--;
377 spin_unlock(&kvm->mmu_lock);
378
379 BUG_ON(kvm->mmu_notifier_count < 0);
380}
381
382static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
383 struct mm_struct *mm,
384 unsigned long address)
385{
386 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200387 int young, idx;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200388
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200389 idx = srcu_read_lock(&kvm->srcu);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200390 spin_lock(&kvm->mmu_lock);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200391
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900392 young = kvm_age_hva(kvm, address);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200393 if (young)
394 kvm_flush_remote_tlbs(kvm);
395
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900396 spin_unlock(&kvm->mmu_lock);
397 srcu_read_unlock(&kvm->srcu, idx);
398
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200399 return young;
400}
401
Andrea Arcangeli8ee53822011-01-13 15:47:10 -0800402static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
403 struct mm_struct *mm,
404 unsigned long address)
405{
406 struct kvm *kvm = mmu_notifier_to_kvm(mn);
407 int young, idx;
408
409 idx = srcu_read_lock(&kvm->srcu);
410 spin_lock(&kvm->mmu_lock);
411 young = kvm_test_age_hva(kvm, address);
412 spin_unlock(&kvm->mmu_lock);
413 srcu_read_unlock(&kvm->srcu, idx);
414
415 return young;
416}
417
Marcelo Tosatti85db06e2008-12-10 21:23:26 +0100418static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
419 struct mm_struct *mm)
420{
421 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Lai Jiangshaneda2bed2010-04-20 14:29:29 +0800422 int idx;
423
424 idx = srcu_read_lock(&kvm->srcu);
Marcelo Tosatti2df72e92012-08-24 15:54:57 -0300425 kvm_arch_flush_shadow_all(kvm);
Lai Jiangshaneda2bed2010-04-20 14:29:29 +0800426 srcu_read_unlock(&kvm->srcu, idx);
Marcelo Tosatti85db06e2008-12-10 21:23:26 +0100427}
428
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200429static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
430 .invalidate_page = kvm_mmu_notifier_invalidate_page,
431 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
432 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
433 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
Andrea Arcangeli8ee53822011-01-13 15:47:10 -0800434 .test_young = kvm_mmu_notifier_test_young,
Izik Eidus3da0dd42009-09-23 21:47:18 +0300435 .change_pte = kvm_mmu_notifier_change_pte,
Marcelo Tosatti85db06e2008-12-10 21:23:26 +0100436 .release = kvm_mmu_notifier_release,
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200437};
Avi Kivity4c07b0a2009-12-20 14:54:04 +0200438
439static int kvm_init_mmu_notifier(struct kvm *kvm)
440{
441 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
442 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
443}
444
445#else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
446
447static int kvm_init_mmu_notifier(struct kvm *kvm)
448{
449 return 0;
450}
451
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200452#endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
453
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800454static void kvm_init_memslots_id(struct kvm *kvm)
455{
456 int i;
457 struct kvm_memslots *slots = kvm->memslots;
458
459 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
Xiao Guangrongf85e2cb2011-11-24 17:41:54 +0800460 slots->id_to_index[i] = slots->memslots[i].id = i;
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800461}
462
Carsten Ottee08b9632012-01-04 10:25:20 +0100463static struct kvm *kvm_create_vm(unsigned long type)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800464{
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100465 int r, i;
466 struct kvm *kvm = kvm_arch_alloc_vm();
Avi Kivity6aa8b732006-12-10 02:21:36 -0800467
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100468 if (!kvm)
469 return ERR_PTR(-ENOMEM);
470
Carsten Ottee08b9632012-01-04 10:25:20 +0100471 r = kvm_arch_init_vm(kvm, type);
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100472 if (r)
473 goto out_err_nodisable;
Alexander Graf10474ae2009-09-15 11:37:46 +0200474
475 r = hardware_enable_all();
476 if (r)
477 goto out_err_nodisable;
478
Avi Kivity75858a82009-01-04 17:10:50 +0200479#ifdef CONFIG_HAVE_KVM_IRQCHIP
480 INIT_HLIST_HEAD(&kvm->mask_notifier_list);
Gleb Natapov136bdfe2009-08-24 11:54:23 +0300481 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
Avi Kivity75858a82009-01-04 17:10:50 +0200482#endif
Avi Kivity6aa8b732006-12-10 02:21:36 -0800483
Alex Williamson1e702d92012-12-10 10:33:32 -0700484 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
485
Marcelo Tosatti46a26bf2009-12-23 14:35:16 -0200486 r = -ENOMEM;
487 kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
488 if (!kvm->memslots)
Jan Kiszka57e7fbe2010-11-09 12:42:12 +0100489 goto out_err_nosrcu;
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800490 kvm_init_memslots_id(kvm);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200491 if (init_srcu_struct(&kvm->srcu))
Jan Kiszka57e7fbe2010-11-09 12:42:12 +0100492 goto out_err_nosrcu;
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200493 for (i = 0; i < KVM_NR_BUSES; i++) {
494 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
495 GFP_KERNEL);
Jan Kiszka57e7fbe2010-11-09 12:42:12 +0100496 if (!kvm->buses[i])
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200497 goto out_err;
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200498 }
Mike Waychison74b5c5b2011-06-03 13:04:53 -0700499
OGAWA Hirofumi85722cd2011-05-11 09:28:28 +0900500 spin_lock_init(&kvm->mmu_lock);
Avi Kivity6d4e4c42007-11-21 16:41:05 +0200501 kvm->mm = current->mm;
502 atomic_inc(&kvm->mm->mm_count);
Gregory Haskinsd34e6b12009-07-07 17:08:49 -0400503 kvm_eventfd_init(kvm);
Shaohua Li11ec2802007-07-23 14:51:37 +0800504 mutex_init(&kvm->lock);
Marcelo Tosatti60eead72009-06-04 15:08:23 -0300505 mutex_init(&kvm->irq_lock);
Marcelo Tosatti79fac952009-12-23 14:35:26 -0200506 mutex_init(&kvm->slots_lock);
Izik Eidusd39f13b2008-03-30 16:01:25 +0300507 atomic_set(&kvm->users_count, 1);
Scott Wood07f0a7b2013-04-25 14:11:23 +0000508 INIT_LIST_HEAD(&kvm->devices);
Mike Waychison74b5c5b2011-06-03 13:04:53 -0700509
510 r = kvm_init_mmu_notifier(kvm);
511 if (r)
512 goto out_err;
513
Jan Kiszkae935b832011-02-08 12:55:33 +0100514 raw_spin_lock(&kvm_lock);
Rusty Russell5e58cfe2007-07-23 17:08:21 +1000515 list_add(&kvm->vm_list, &vm_list);
Jan Kiszkae935b832011-02-08 12:55:33 +0100516 raw_spin_unlock(&kvm_lock);
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100517
Avi Kivityf17abe92007-02-21 19:28:04 +0200518 return kvm;
Alexander Graf10474ae2009-09-15 11:37:46 +0200519
520out_err:
Jan Kiszka57e7fbe2010-11-09 12:42:12 +0100521 cleanup_srcu_struct(&kvm->srcu);
522out_err_nosrcu:
Alexander Graf10474ae2009-09-15 11:37:46 +0200523 hardware_disable_all();
524out_err_nodisable:
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200525 for (i = 0; i < KVM_NR_BUSES; i++)
526 kfree(kvm->buses[i]);
Marcelo Tosatti46a26bf2009-12-23 14:35:16 -0200527 kfree(kvm->memslots);
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100528 kvm_arch_free_vm(kvm);
Alexander Graf10474ae2009-09-15 11:37:46 +0200529 return ERR_PTR(r);
Avi Kivityf17abe92007-02-21 19:28:04 +0200530}
531
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900532/*
533 * Avoid using vmalloc for a small buffer.
534 * Should not be used when the size is statically known.
535 */
Takuya Yoshikawac1a7b322012-05-20 13:15:07 +0900536void *kvm_kvzalloc(unsigned long size)
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900537{
538 if (size > PAGE_SIZE)
539 return vzalloc(size);
540 else
541 return kzalloc(size, GFP_KERNEL);
542}
543
Takuya Yoshikawac1a7b322012-05-20 13:15:07 +0900544void kvm_kvfree(const void *addr)
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900545{
546 if (is_vmalloc_addr(addr))
547 vfree(addr);
548 else
549 kfree(addr);
550}
551
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900552static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
553{
554 if (!memslot->dirty_bitmap)
555 return;
556
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900557 kvm_kvfree(memslot->dirty_bitmap);
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900558 memslot->dirty_bitmap = NULL;
559}
560
Avi Kivity6aa8b732006-12-10 02:21:36 -0800561/*
562 * Free any memory in @free but not in @dont.
563 */
564static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
565 struct kvm_memory_slot *dont)
566{
Avi Kivity6aa8b732006-12-10 02:21:36 -0800567 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900568 kvm_destroy_dirty_bitmap(free);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800569
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +0900570 kvm_arch_free_memslot(free, dont);
Marcelo Tosatti05da4552008-02-23 11:44:30 -0300571
Avi Kivity6aa8b732006-12-10 02:21:36 -0800572 free->npages = 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800573}
574
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +0800575void kvm_free_physmem(struct kvm *kvm)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800576{
Marcelo Tosatti46a26bf2009-12-23 14:35:16 -0200577 struct kvm_memslots *slots = kvm->memslots;
Xiao Guangrongbe6ba0f2011-11-24 17:39:18 +0800578 struct kvm_memory_slot *memslot;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800579
Xiao Guangrongbe6ba0f2011-11-24 17:39:18 +0800580 kvm_for_each_memslot(memslot, slots)
581 kvm_free_physmem_slot(memslot, NULL);
Marcelo Tosatti46a26bf2009-12-23 14:35:16 -0200582
583 kfree(kvm->memslots);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800584}
585
Scott Wood07f0a7b2013-04-25 14:11:23 +0000586static void kvm_destroy_devices(struct kvm *kvm)
587{
588 struct list_head *node, *tmp;
589
590 list_for_each_safe(node, tmp, &kvm->devices) {
591 struct kvm_device *dev =
592 list_entry(node, struct kvm_device, vm_node);
593
594 list_del(node);
595 dev->ops->destroy(dev);
596 }
597}
598
Avi Kivityf17abe92007-02-21 19:28:04 +0200599static void kvm_destroy_vm(struct kvm *kvm)
600{
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200601 int i;
Avi Kivity6d4e4c42007-11-21 16:41:05 +0200602 struct mm_struct *mm = kvm->mm;
603
Sheng Yangad8ba2c2009-01-06 10:03:02 +0800604 kvm_arch_sync_events(kvm);
Jan Kiszkae935b832011-02-08 12:55:33 +0100605 raw_spin_lock(&kvm_lock);
Avi Kivity133de902007-02-12 00:54:44 -0800606 list_del(&kvm->vm_list);
Jan Kiszkae935b832011-02-08 12:55:33 +0100607 raw_spin_unlock(&kvm_lock);
Avi Kivity399ec802008-11-19 13:58:46 +0200608 kvm_free_irq_routing(kvm);
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200609 for (i = 0; i < KVM_NR_BUSES; i++)
610 kvm_io_bus_destroy(kvm->buses[i]);
Avi Kivity980da6c2009-12-20 15:13:43 +0200611 kvm_coalesced_mmio_free(kvm);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200612#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
613 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
Gleb Natapovf00be0c2009-03-19 12:20:36 +0200614#else
Marcelo Tosatti2df72e92012-08-24 15:54:57 -0300615 kvm_arch_flush_shadow_all(kvm);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200616#endif
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +0800617 kvm_arch_destroy_vm(kvm);
Scott Wood07f0a7b2013-04-25 14:11:23 +0000618 kvm_destroy_devices(kvm);
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100619 kvm_free_physmem(kvm);
620 cleanup_srcu_struct(&kvm->srcu);
621 kvm_arch_free_vm(kvm);
Alexander Graf10474ae2009-09-15 11:37:46 +0200622 hardware_disable_all();
Avi Kivity6d4e4c42007-11-21 16:41:05 +0200623 mmdrop(mm);
Avi Kivityf17abe92007-02-21 19:28:04 +0200624}
625
Izik Eidusd39f13b2008-03-30 16:01:25 +0300626void kvm_get_kvm(struct kvm *kvm)
627{
628 atomic_inc(&kvm->users_count);
629}
630EXPORT_SYMBOL_GPL(kvm_get_kvm);
631
632void kvm_put_kvm(struct kvm *kvm)
633{
634 if (atomic_dec_and_test(&kvm->users_count))
635 kvm_destroy_vm(kvm);
636}
637EXPORT_SYMBOL_GPL(kvm_put_kvm);
638
639
Avi Kivityf17abe92007-02-21 19:28:04 +0200640static int kvm_vm_release(struct inode *inode, struct file *filp)
641{
642 struct kvm *kvm = filp->private_data;
643
Gregory Haskins721eecbf2009-05-20 10:30:49 -0400644 kvm_irqfd_release(kvm);
645
Izik Eidusd39f13b2008-03-30 16:01:25 +0300646 kvm_put_kvm(kvm);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800647 return 0;
648}
649
Takuya Yoshikawa515a0122010-10-27 18:23:54 +0900650/*
651 * Allocation size is twice as large as the actual dirty bitmap size.
Takuya Yoshikawa93474b22012-03-01 19:34:45 +0900652 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
Takuya Yoshikawa515a0122010-10-27 18:23:54 +0900653 */
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900654static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
655{
Takuya Yoshikawa189a2f72012-02-08 13:01:09 +0900656#ifndef CONFIG_S390
Takuya Yoshikawa515a0122010-10-27 18:23:54 +0900657 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900658
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900659 memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900660 if (!memslot->dirty_bitmap)
661 return -ENOMEM;
662
Takuya Yoshikawa189a2f72012-02-08 13:01:09 +0900663#endif /* !CONFIG_S390 */
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900664 return 0;
665}
666
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800667static int cmp_memslot(const void *slot1, const void *slot2)
668{
669 struct kvm_memory_slot *s1, *s2;
670
671 s1 = (struct kvm_memory_slot *)slot1;
672 s2 = (struct kvm_memory_slot *)slot2;
673
674 if (s1->npages < s2->npages)
675 return 1;
676 if (s1->npages > s2->npages)
677 return -1;
678
679 return 0;
680}
681
682/*
683 * Sort the memslots base on its size, so the larger slots
684 * will get better fit.
685 */
686static void sort_memslots(struct kvm_memslots *slots)
687{
Xiao Guangrongf85e2cb2011-11-24 17:41:54 +0800688 int i;
689
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800690 sort(slots->memslots, KVM_MEM_SLOTS_NUM,
691 sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
Xiao Guangrongf85e2cb2011-11-24 17:41:54 +0800692
693 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
694 slots->id_to_index[slots->memslots[i].id] = i;
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800695}
696
Alex Williamson116c14c2012-12-21 08:20:16 -0700697void update_memslots(struct kvm_memslots *slots, struct kvm_memory_slot *new,
698 u64 last_generation)
Xiao Guangrongbe593d62011-11-24 17:38:24 +0800699{
700 if (new) {
701 int id = new->id;
Xiao Guangrong28a37542011-11-24 19:04:35 +0800702 struct kvm_memory_slot *old = id_to_memslot(slots, id);
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800703 unsigned long npages = old->npages;
Xiao Guangrongbe593d62011-11-24 17:38:24 +0800704
Xiao Guangrong28a37542011-11-24 19:04:35 +0800705 *old = *new;
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800706 if (new->npages != npages)
707 sort_memslots(slots);
Xiao Guangrongbe593d62011-11-24 17:38:24 +0800708 }
709
Alex Williamson116c14c2012-12-21 08:20:16 -0700710 slots->generation = last_generation + 1;
Xiao Guangrongbe593d62011-11-24 17:38:24 +0800711}
712
Xiao Guangronga50d64d2012-08-21 10:58:13 +0800713static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
714{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +0800715 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
716
717#ifdef KVM_CAP_READONLY_MEM
718 valid_flags |= KVM_MEM_READONLY;
719#endif
720
721 if (mem->flags & ~valid_flags)
Xiao Guangronga50d64d2012-08-21 10:58:13 +0800722 return -EINVAL;
723
724 return 0;
725}
726
Gleb Natapov7ec4fb42012-12-24 17:49:30 +0200727static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
728 struct kvm_memslots *slots, struct kvm_memory_slot *new)
729{
730 struct kvm_memslots *old_memslots = kvm->memslots;
731
732 update_memslots(slots, new, kvm->memslots->generation);
733 rcu_assign_pointer(kvm->memslots, slots);
734 synchronize_srcu_expedited(&kvm->srcu);
735 return old_memslots;
736}
737
Avi Kivity6aa8b732006-12-10 02:21:36 -0800738/*
Avi Kivity6aa8b732006-12-10 02:21:36 -0800739 * Allocate some memory and give it an address in the guest physical address
740 * space.
741 *
742 * Discontiguous memory is allowed, mostly for framebuffers.
Sheng Yangf78e0e22007-10-29 09:40:42 +0800743 *
Marcelo Tosatti10589a42007-12-20 19:18:22 -0500744 * Must be called holding mmap_sem for write.
Avi Kivity6aa8b732006-12-10 02:21:36 -0800745 */
Sheng Yangf78e0e22007-10-29 09:40:42 +0800746int __kvm_set_memory_region(struct kvm *kvm,
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +0900747 struct kvm_userspace_memory_region *mem)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800748{
Avi Kivity8234b222010-12-27 12:08:45 +0200749 int r;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800750 gfn_t base_gfn;
Heiko Carstens28bcb112009-09-03 17:35:35 +0200751 unsigned long npages;
Takuya Yoshikawaa843fac2013-01-11 18:27:43 +0900752 struct kvm_memory_slot *slot;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800753 struct kvm_memory_slot old, new;
Alex Williamsonb7f69c52012-12-10 10:33:03 -0700754 struct kvm_memslots *slots = NULL, *old_memslots;
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900755 enum kvm_mr_change change;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800756
Xiao Guangronga50d64d2012-08-21 10:58:13 +0800757 r = check_memory_region_flags(mem);
758 if (r)
759 goto out;
760
Avi Kivity6aa8b732006-12-10 02:21:36 -0800761 r = -EINVAL;
762 /* General sanity checks */
763 if (mem->memory_size & (PAGE_SIZE - 1))
764 goto out;
765 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
766 goto out;
Takuya Yoshikawafa3d3152011-05-07 16:35:38 +0900767 /* We can read the guest memory with __xxx_user() later on. */
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +0900768 if ((mem->slot < KVM_USER_MEM_SLOTS) &&
Takuya Yoshikawafa3d3152011-05-07 16:35:38 +0900769 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
Heiko Carstens9e3bb6b2011-05-24 07:51:27 +0200770 !access_ok(VERIFY_WRITE,
771 (void __user *)(unsigned long)mem->userspace_addr,
772 mem->memory_size)))
Hollis Blanchard78749802008-11-07 13:32:12 -0600773 goto out;
Xiao Guangrong93a5cef2011-11-24 17:37:48 +0800774 if (mem->slot >= KVM_MEM_SLOTS_NUM)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800775 goto out;
776 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
777 goto out;
778
Takuya Yoshikawaa843fac2013-01-11 18:27:43 +0900779 slot = id_to_memslot(kvm->memslots, mem->slot);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800780 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
781 npages = mem->memory_size >> PAGE_SHIFT;
782
Takuya Yoshikawa660c22c2010-04-13 22:47:24 +0900783 r = -EINVAL;
784 if (npages > KVM_MEM_MAX_NR_PAGES)
785 goto out;
786
Avi Kivity6aa8b732006-12-10 02:21:36 -0800787 if (!npages)
788 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
789
Takuya Yoshikawaa843fac2013-01-11 18:27:43 +0900790 new = old = *slot;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800791
Avi Kivitye36d96f2010-06-21 10:56:36 +0300792 new.id = mem->slot;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800793 new.base_gfn = base_gfn;
794 new.npages = npages;
795 new.flags = mem->flags;
796
Avi Kivity6aa8b732006-12-10 02:21:36 -0800797 r = -EINVAL;
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900798 if (npages) {
799 if (!old.npages)
800 change = KVM_MR_CREATE;
801 else { /* Modify an existing slot. */
802 if ((mem->userspace_addr != old.userspace_addr) ||
Takuya Yoshikawa75d61fb2013-01-30 19:40:41 +0900803 (npages != old.npages) ||
804 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900805 goto out;
806
807 if (base_gfn != old.base_gfn)
808 change = KVM_MR_MOVE;
809 else if (new.flags != old.flags)
810 change = KVM_MR_FLAGS_ONLY;
811 else { /* Nothing to change. */
812 r = 0;
813 goto out;
814 }
815 }
816 } else if (old.npages) {
817 change = KVM_MR_DELETE;
818 } else /* Modify a non-existent slot: disallowed. */
Takuya Yoshikawa0ea75e12013-01-11 18:26:10 +0900819 goto out;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800820
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900821 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
Takuya Yoshikawa0a706be2013-01-11 18:26:55 +0900822 /* Check for overlaps */
823 r = -EEXIST;
824 kvm_for_each_memslot(slot, kvm->memslots) {
Takuya Yoshikawaa843fac2013-01-11 18:27:43 +0900825 if ((slot->id >= KVM_USER_MEM_SLOTS) ||
826 (slot->id == mem->slot))
Takuya Yoshikawa0a706be2013-01-11 18:26:55 +0900827 continue;
828 if (!((base_gfn + npages <= slot->base_gfn) ||
829 (base_gfn >= slot->base_gfn + slot->npages)))
830 goto out;
831 }
Avi Kivity6aa8b732006-12-10 02:21:36 -0800832 }
Avi Kivity6aa8b732006-12-10 02:21:36 -0800833
Avi Kivity6aa8b732006-12-10 02:21:36 -0800834 /* Free page dirty bitmap if unneeded */
835 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
Al Viro8b6d44c2007-02-09 16:38:40 +0000836 new.dirty_bitmap = NULL;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800837
838 r = -ENOMEM;
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900839 if (change == KVM_MR_CREATE) {
Takuya Yoshikawa189a2f72012-02-08 13:01:09 +0900840 new.userspace_addr = mem->userspace_addr;
Takuya Yoshikawad89cc612012-08-01 18:03:28 +0900841
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +0900842 if (kvm_arch_create_memslot(&new, npages))
843 goto out_free;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800844 }
Joerg Roedelec04b262009-06-19 15:16:23 +0200845
Avi Kivity6aa8b732006-12-10 02:21:36 -0800846 /* Allocate page dirty bitmap if needed */
847 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900848 if (kvm_create_dirty_bitmap(&new) < 0)
Sheng Yangf78e0e22007-10-29 09:40:42 +0800849 goto out_free;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800850 }
851
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900852 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200853 r = -ENOMEM;
Thomas Meyer6da64fd2011-11-08 20:32:19 +0100854 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
855 GFP_KERNEL);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200856 if (!slots)
857 goto out_free;
Xiao Guangrong28a37542011-11-24 19:04:35 +0800858 slot = id_to_memslot(slots, mem->slot);
859 slot->flags |= KVM_MEMSLOT_INVALID;
860
Gleb Natapov7ec4fb42012-12-24 17:49:30 +0200861 old_memslots = install_new_memslots(kvm, slots, NULL);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200862
Alex Williamsone40f1932012-12-10 10:32:57 -0700863 /* slot was deleted or moved, clear iommu mapping */
864 kvm_iommu_unmap_pages(kvm, &old);
Marcelo Tosatti12d6e752012-08-24 15:54:58 -0300865 /* From this point no new shadow pages pointing to a deleted,
866 * or moved, memslot will be created.
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200867 *
868 * validation of sp->gfn happens in:
869 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
870 * - kvm_is_visible_gfn (mmu_check_roots)
871 */
Marcelo Tosatti2df72e92012-08-24 15:54:57 -0300872 kvm_arch_flush_shadow_memslot(kvm, slot);
Alex Williamsonb7f69c52012-12-10 10:33:03 -0700873 slots = old_memslots;
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200874 }
Marcelo Tosatti34d4cb82008-07-10 20:49:31 -0300875
Takuya Yoshikawa7b6195a2013-02-27 19:44:34 +0900876 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
Marcelo Tosattif7784b82009-12-23 14:35:18 -0200877 if (r)
Alex Williamsonb7f69c52012-12-10 10:33:03 -0700878 goto out_slots;
Marcelo Tosattif7784b82009-12-23 14:35:18 -0200879
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200880 r = -ENOMEM;
Alex Williamsonb7f69c52012-12-10 10:33:03 -0700881 /*
882 * We can re-use the old_memslots from above, the only difference
883 * from the currently installed memslots is the invalid flag. This
884 * will get overwritten by update_memslots anyway.
885 */
886 if (!slots) {
887 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
888 GFP_KERNEL);
889 if (!slots)
890 goto out_free;
891 }
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200892
Alex Williamson261874b2013-01-24 15:04:03 -0700893 /*
894 * IOMMU mapping: New slots need to be mapped. Old slots need to be
Takuya Yoshikawa75d61fb2013-01-30 19:40:41 +0900895 * un-mapped and re-mapped if their base changes. Since base change
896 * unmapping is handled above with slot deletion, mapping alone is
897 * needed here. Anything else the iommu might care about for existing
898 * slots (size changes, userspace addr changes and read-only flag
899 * changes) is disallowed above, so any other attribute changes getting
900 * here can be skipped.
Alex Williamson261874b2013-01-24 15:04:03 -0700901 */
Takuya Yoshikawa75d61fb2013-01-30 19:40:41 +0900902 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
903 r = kvm_iommu_map_pages(kvm, &new);
904 if (r)
905 goto out_slots;
Alex Williamsone40f1932012-12-10 10:32:57 -0700906 }
907
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200908 /* actual memory is freed via old in kvm_free_physmem_slot below */
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900909 if (change == KVM_MR_DELETE) {
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200910 new.dirty_bitmap = NULL;
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +0900911 memset(&new.arch, 0, sizeof(new.arch));
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200912 }
913
Gleb Natapov7ec4fb42012-12-24 17:49:30 +0200914 old_memslots = install_new_memslots(kvm, slots, &new);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800915
Takuya Yoshikawa84826442013-02-27 19:45:25 +0900916 kvm_arch_commit_memory_region(kvm, mem, &old, change);
Zhang Xiantao3ad82a72007-11-20 13:11:38 +0800917
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200918 kvm_free_physmem_slot(&old, &new);
919 kfree(old_memslots);
920
Avi Kivity6aa8b732006-12-10 02:21:36 -0800921 return 0;
922
Alex Williamsone40f1932012-12-10 10:32:57 -0700923out_slots:
924 kfree(slots);
Sheng Yangf78e0e22007-10-29 09:40:42 +0800925out_free:
Avi Kivity6aa8b732006-12-10 02:21:36 -0800926 kvm_free_physmem_slot(&new, &old);
927out:
928 return r;
Izik Eidus210c7c42007-10-24 23:52:57 +0200929}
Sheng Yangf78e0e22007-10-29 09:40:42 +0800930EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
931
932int kvm_set_memory_region(struct kvm *kvm,
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +0900933 struct kvm_userspace_memory_region *mem)
Sheng Yangf78e0e22007-10-29 09:40:42 +0800934{
935 int r;
936
Marcelo Tosatti79fac952009-12-23 14:35:26 -0200937 mutex_lock(&kvm->slots_lock);
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +0900938 r = __kvm_set_memory_region(kvm, mem);
Marcelo Tosatti79fac952009-12-23 14:35:26 -0200939 mutex_unlock(&kvm->slots_lock);
Sheng Yangf78e0e22007-10-29 09:40:42 +0800940 return r;
941}
Izik Eidus210c7c42007-10-24 23:52:57 +0200942EXPORT_SYMBOL_GPL(kvm_set_memory_region);
943
Carsten Otte1fe779f2007-10-29 16:08:35 +0100944int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +0900945 struct kvm_userspace_memory_region *mem)
Izik Eidus210c7c42007-10-24 23:52:57 +0200946{
Alex Williamsonbbacc0c2012-12-10 10:33:09 -0700947 if (mem->slot >= KVM_USER_MEM_SLOTS)
Izik Eiduse0d62c72007-10-24 23:57:46 +0200948 return -EINVAL;
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +0900949 return kvm_set_memory_region(kvm, mem);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800950}
951
Zhang Xiantao5bb064d2007-11-18 20:29:43 +0800952int kvm_get_dirty_log(struct kvm *kvm,
953 struct kvm_dirty_log *log, int *is_dirty)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800954{
955 struct kvm_memory_slot *memslot;
956 int r, i;
Takuya Yoshikawa87bf6e72010-04-12 19:35:35 +0900957 unsigned long n;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800958 unsigned long any = 0;
959
Avi Kivity6aa8b732006-12-10 02:21:36 -0800960 r = -EINVAL;
Alex Williamsonbbacc0c2012-12-10 10:33:09 -0700961 if (log->slot >= KVM_USER_MEM_SLOTS)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800962 goto out;
963
Xiao Guangrong28a37542011-11-24 19:04:35 +0800964 memslot = id_to_memslot(kvm->memslots, log->slot);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800965 r = -ENOENT;
966 if (!memslot->dirty_bitmap)
967 goto out;
968
Takuya Yoshikawa87bf6e72010-04-12 19:35:35 +0900969 n = kvm_dirty_bitmap_bytes(memslot);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800970
Uri Lublincd1a4a92007-02-22 16:43:09 +0200971 for (i = 0; !any && i < n/sizeof(long); ++i)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800972 any = memslot->dirty_bitmap[i];
973
974 r = -EFAULT;
975 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
976 goto out;
977
Zhang Xiantao5bb064d2007-11-18 20:29:43 +0800978 if (any)
979 *is_dirty = 1;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800980
981 r = 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800982out:
Avi Kivity6aa8b732006-12-10 02:21:36 -0800983 return r;
984}
985
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +0900986bool kvm_largepages_enabled(void)
987{
988 return largepages_enabled;
989}
990
Marcelo Tosatti54dee992009-06-11 12:07:44 -0300991void kvm_disable_largepages(void)
992{
993 largepages_enabled = false;
994}
995EXPORT_SYMBOL_GPL(kvm_disable_largepages);
996
Gleb Natapov49c77542010-10-18 15:22:23 +0200997struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
998{
999 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1000}
Avi Kivitya1f4d3952010-06-21 11:44:20 +03001001EXPORT_SYMBOL_GPL(gfn_to_memslot);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001002
Izik Eiduse0d62c72007-10-24 23:57:46 +02001003int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1004{
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +08001005 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
Izik Eiduse0d62c72007-10-24 23:57:46 +02001006
Alex Williamsonbbacc0c2012-12-10 10:33:09 -07001007 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +08001008 memslot->flags & KVM_MEMSLOT_INVALID)
1009 return 0;
Izik Eiduse0d62c72007-10-24 23:57:46 +02001010
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +08001011 return 1;
Izik Eiduse0d62c72007-10-24 23:57:46 +02001012}
1013EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1014
Joerg Roedel8f0b1ab2010-01-28 12:37:56 +01001015unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1016{
1017 struct vm_area_struct *vma;
1018 unsigned long addr, size;
1019
1020 size = PAGE_SIZE;
1021
1022 addr = gfn_to_hva(kvm, gfn);
1023 if (kvm_is_error_hva(addr))
1024 return PAGE_SIZE;
1025
1026 down_read(&current->mm->mmap_sem);
1027 vma = find_vma(current->mm, addr);
1028 if (!vma)
1029 goto out;
1030
1031 size = vma_kernel_pagesize(vma);
1032
1033out:
1034 up_read(&current->mm->mmap_sem);
1035
1036 return size;
1037}
1038
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001039static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1040{
1041 return slot->flags & KVM_MEM_READONLY;
1042}
1043
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001044static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1045 gfn_t *nr_pages, bool write)
Izik Eidus539cb662007-11-11 22:05:04 +02001046{
Marcelo Tosattibc6678a2009-12-23 14:35:21 -02001047 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
Xiao Guangrongca3a4902012-08-21 11:01:50 +08001048 return KVM_HVA_ERR_BAD;
Xiao Guangrong48987782010-08-22 19:11:43 +08001049
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001050 if (memslot_is_readonly(slot) && write)
1051 return KVM_HVA_ERR_RO_BAD;
Xiao Guangrong48987782010-08-22 19:11:43 +08001052
1053 if (nr_pages)
1054 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1055
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001056 return __gfn_to_hva_memslot(slot, gfn);
Izik Eidus539cb662007-11-11 22:05:04 +02001057}
Xiao Guangrong48987782010-08-22 19:11:43 +08001058
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001059static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1060 gfn_t *nr_pages)
1061{
1062 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1063}
1064
1065unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1066 gfn_t gfn)
1067{
1068 return gfn_to_hva_many(slot, gfn, NULL);
1069}
1070EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1071
Xiao Guangrong48987782010-08-22 19:11:43 +08001072unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1073{
Gleb Natapov49c77542010-10-18 15:22:23 +02001074 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
Xiao Guangrong48987782010-08-22 19:11:43 +08001075}
Sheng Yang0d150292008-04-25 21:44:50 +08001076EXPORT_SYMBOL_GPL(gfn_to_hva);
Izik Eidus539cb662007-11-11 22:05:04 +02001077
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001078/*
1079 * The hva returned by this function is only allowed to be read.
1080 * It should pair with kvm_read_hva() or kvm_read_hva_atomic().
1081 */
1082static unsigned long gfn_to_hva_read(struct kvm *kvm, gfn_t gfn)
Gleb Natapov80300892010-10-19 18:13:41 +02001083{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001084 return __gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL, false);
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001085}
1086
1087static int kvm_read_hva(void *data, void __user *hva, int len)
1088{
1089 return __copy_from_user(data, hva, len);
1090}
1091
1092static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1093{
1094 return __copy_from_user_inatomic(data, hva, len);
Gleb Natapov80300892010-10-19 18:13:41 +02001095}
1096
Geoff Levand39369f72013-04-05 19:20:30 +00001097static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
Gleb Natapov0857b9e2011-02-01 13:21:47 +02001098 unsigned long start, int write, struct page **page)
1099{
1100 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1101
1102 if (write)
1103 flags |= FOLL_WRITE;
1104
1105 return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1106}
1107
Huang Yingfafc3db2011-01-30 11:15:49 +08001108static inline int check_user_page_hwpoison(unsigned long addr)
1109{
1110 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1111
1112 rc = __get_user_pages(current, current->mm, addr, 1,
1113 flags, NULL, NULL, NULL);
1114 return rc == -EHWPOISON;
1115}
1116
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001117/*
1118 * The atomic path to get the writable pfn which will be stored in @pfn,
1119 * true indicates success, otherwise false is returned.
1120 */
1121static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1122 bool write_fault, bool *writable, pfn_t *pfn)
1123{
1124 struct page *page[1];
1125 int npages;
1126
1127 if (!(async || atomic))
1128 return false;
1129
Xiao Guangrong12ce13f2012-08-21 11:00:49 +08001130 /*
1131 * Fast pin a writable pfn only if it is a write fault request
1132 * or the caller allows to map a writable pfn for a read fault
1133 * request.
1134 */
1135 if (!(write_fault || writable))
1136 return false;
1137
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001138 npages = __get_user_pages_fast(addr, 1, 1, page);
1139 if (npages == 1) {
1140 *pfn = page_to_pfn(page[0]);
1141
1142 if (writable)
1143 *writable = true;
1144 return true;
1145 }
1146
1147 return false;
1148}
1149
1150/*
1151 * The slow path to get the pfn of the specified host virtual address,
1152 * 1 indicates success, -errno is returned if error is detected.
1153 */
1154static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1155 bool *writable, pfn_t *pfn)
Avi Kivity954bbbc2007-03-30 14:02:32 +03001156{
Anthony Liguori8d4e1282007-10-18 09:59:34 -05001157 struct page *page[1];
Gleb Natapovaf585b92010-10-14 11:22:46 +02001158 int npages = 0;
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001159
1160 might_sleep();
1161
1162 if (writable)
1163 *writable = write_fault;
1164
1165 if (async) {
1166 down_read(&current->mm->mmap_sem);
1167 npages = get_user_page_nowait(current, current->mm,
1168 addr, write_fault, page);
1169 up_read(&current->mm->mmap_sem);
1170 } else
1171 npages = get_user_pages_fast(addr, 1, write_fault,
1172 page);
1173 if (npages != 1)
1174 return npages;
1175
1176 /* map read fault as writable if possible */
Xiao Guangrong12ce13f2012-08-21 11:00:49 +08001177 if (unlikely(!write_fault) && writable) {
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001178 struct page *wpage[1];
1179
1180 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1181 if (npages == 1) {
1182 *writable = true;
1183 put_page(page[0]);
1184 page[0] = wpage[0];
1185 }
1186
1187 npages = 1;
1188 }
1189 *pfn = page_to_pfn(page[0]);
1190 return npages;
1191}
1192
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001193static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1194{
1195 if (unlikely(!(vma->vm_flags & VM_READ)))
1196 return false;
1197
1198 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1199 return false;
1200
1201 return true;
1202}
1203
Xiao Guangrong12ce13f2012-08-21 11:00:49 +08001204/*
1205 * Pin guest page in memory and return its pfn.
1206 * @addr: host virtual address which maps memory to the guest
1207 * @atomic: whether this function can sleep
1208 * @async: whether this function need to wait IO complete if the
1209 * host page is not in the memory
1210 * @write_fault: whether we should get a writable host page
1211 * @writable: whether it allows to map a writable host page for !@write_fault
1212 *
1213 * The function will map a writable host page for these two cases:
1214 * 1): @write_fault = true
1215 * 2): @write_fault = false && @writable, @writable will tell the caller
1216 * whether the mapping is writable.
1217 */
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001218static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1219 bool write_fault, bool *writable)
1220{
1221 struct vm_area_struct *vma;
1222 pfn_t pfn = 0;
1223 int npages;
Avi Kivity954bbbc2007-03-30 14:02:32 +03001224
Gleb Natapovaf585b92010-10-14 11:22:46 +02001225 /* we can do it either atomically or asynchronously, not both */
1226 BUG_ON(atomic && async);
1227
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001228 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1229 return pfn;
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001230
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001231 if (atomic)
1232 return KVM_PFN_ERR_FAULT;
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001233
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001234 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1235 if (npages == 1)
1236 return pfn;
Gleb Natapovaf585b92010-10-14 11:22:46 +02001237
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001238 down_read(&current->mm->mmap_sem);
1239 if (npages == -EHWPOISON ||
1240 (!async && check_user_page_hwpoison(addr))) {
1241 pfn = KVM_PFN_ERR_HWPOISON;
1242 goto exit;
Xiao Guangrong887c08a2010-08-22 19:10:28 +08001243 }
Izik Eidus539cb662007-11-11 22:05:04 +02001244
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001245 vma = find_vma_intersection(current->mm, addr, addr + 1);
Anthony Liguori8d4e1282007-10-18 09:59:34 -05001246
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001247 if (vma == NULL)
1248 pfn = KVM_PFN_ERR_FAULT;
1249 else if ((vma->vm_flags & VM_PFNMAP)) {
1250 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1251 vma->vm_pgoff;
1252 BUG_ON(!kvm_is_mmio_pfn(pfn));
1253 } else {
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001254 if (async && vma_is_valid(vma, write_fault))
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001255 *async = true;
1256 pfn = KVM_PFN_ERR_FAULT;
1257 }
1258exit:
1259 up_read(&current->mm->mmap_sem);
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001260 return pfn;
Anthony Liguori35149e22008-04-02 14:46:56 -05001261}
1262
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001263static pfn_t
1264__gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1265 bool *async, bool write_fault, bool *writable)
Xiao Guangrong887c08a2010-08-22 19:10:28 +08001266{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001267 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1268
1269 if (addr == KVM_HVA_ERR_RO_BAD)
1270 return KVM_PFN_ERR_RO_FAULT;
1271
1272 if (kvm_is_error_hva(addr))
Xiao Guangrong81c52c52012-10-16 20:10:59 +08001273 return KVM_PFN_NOSLOT;
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001274
1275 /* Do not map writable pfn in the readonly memslot. */
1276 if (writable && memslot_is_readonly(slot)) {
1277 *writable = false;
1278 writable = NULL;
1279 }
1280
1281 return hva_to_pfn(addr, atomic, async, write_fault,
1282 writable);
Xiao Guangrong887c08a2010-08-22 19:10:28 +08001283}
Xiao Guangrong887c08a2010-08-22 19:10:28 +08001284
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001285static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1286 bool write_fault, bool *writable)
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001287{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001288 struct kvm_memory_slot *slot;
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001289
Gleb Natapovaf585b92010-10-14 11:22:46 +02001290 if (async)
1291 *async = false;
1292
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001293 slot = gfn_to_memslot(kvm, gfn);
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001294
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001295 return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1296 writable);
Xiao Guangrong365fb3f2010-08-28 19:24:13 +08001297}
1298
1299pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1300{
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001301 return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
Xiao Guangrong365fb3f2010-08-28 19:24:13 +08001302}
1303EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1304
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001305pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1306 bool write_fault, bool *writable)
Gleb Natapovaf585b92010-10-14 11:22:46 +02001307{
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001308 return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
Gleb Natapovaf585b92010-10-14 11:22:46 +02001309}
1310EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1311
Xiao Guangrong365fb3f2010-08-28 19:24:13 +08001312pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1313{
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001314 return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001315}
Anthony Liguori35149e22008-04-02 14:46:56 -05001316EXPORT_SYMBOL_GPL(gfn_to_pfn);
1317
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001318pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1319 bool *writable)
1320{
1321 return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1322}
1323EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1324
Xiao Guangrongd5661042012-07-17 21:56:16 +08001325pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001326{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001327 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001328}
1329
Xiao Guangrong037d92d2012-08-21 10:59:12 +08001330pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1331{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001332 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
Xiao Guangrong037d92d2012-08-21 10:59:12 +08001333}
1334EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1335
Xiao Guangrong48987782010-08-22 19:11:43 +08001336int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1337 int nr_pages)
1338{
1339 unsigned long addr;
1340 gfn_t entry;
1341
Gleb Natapov49c77542010-10-18 15:22:23 +02001342 addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
Xiao Guangrong48987782010-08-22 19:11:43 +08001343 if (kvm_is_error_hva(addr))
1344 return -1;
1345
1346 if (entry < nr_pages)
1347 return 0;
1348
1349 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1350}
1351EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1352
Xiao Guangronga2766322012-07-26 11:58:59 +08001353static struct page *kvm_pfn_to_page(pfn_t pfn)
1354{
Xiao Guangrong81c52c52012-10-16 20:10:59 +08001355 if (is_error_noslot_pfn(pfn))
Xiao Guangrong6cede2e2012-08-03 15:41:22 +08001356 return KVM_ERR_PTR_BAD_PAGE;
Xiao Guangronga2766322012-07-26 11:58:59 +08001357
Xiao Guangrongcb9aaa32012-08-03 15:42:10 +08001358 if (kvm_is_mmio_pfn(pfn)) {
1359 WARN_ON(1);
1360 return KVM_ERR_PTR_BAD_PAGE;
1361 }
1362
Xiao Guangronga2766322012-07-26 11:58:59 +08001363 return pfn_to_page(pfn);
1364}
1365
Anthony Liguori35149e22008-04-02 14:46:56 -05001366struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1367{
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001368 pfn_t pfn;
1369
1370 pfn = gfn_to_pfn(kvm, gfn);
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001371
Xiao Guangronga2766322012-07-26 11:58:59 +08001372 return kvm_pfn_to_page(pfn);
Avi Kivity954bbbc2007-03-30 14:02:32 +03001373}
Anthony Liguoriaab61cc2007-10-29 15:15:20 -05001374
Avi Kivity954bbbc2007-03-30 14:02:32 +03001375EXPORT_SYMBOL_GPL(gfn_to_page);
1376
Izik Eidusb4231d62007-11-20 11:49:33 +02001377void kvm_release_page_clean(struct page *page)
1378{
Xiao Guangrong32cad842012-08-03 15:42:52 +08001379 WARN_ON(is_error_page(page));
1380
Anthony Liguori35149e22008-04-02 14:46:56 -05001381 kvm_release_pfn_clean(page_to_pfn(page));
Izik Eidusb4231d62007-11-20 11:49:33 +02001382}
1383EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1384
Anthony Liguori35149e22008-04-02 14:46:56 -05001385void kvm_release_pfn_clean(pfn_t pfn)
1386{
Xiao Guangrong81c52c52012-10-16 20:10:59 +08001387 if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001388 put_page(pfn_to_page(pfn));
Anthony Liguori35149e22008-04-02 14:46:56 -05001389}
1390EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1391
Izik Eidusb4231d62007-11-20 11:49:33 +02001392void kvm_release_page_dirty(struct page *page)
Izik Eidus8a7ae052007-10-18 11:09:33 +02001393{
Xiao Guangronga2766322012-07-26 11:58:59 +08001394 WARN_ON(is_error_page(page));
1395
Anthony Liguori35149e22008-04-02 14:46:56 -05001396 kvm_release_pfn_dirty(page_to_pfn(page));
Izik Eidus8a7ae052007-10-18 11:09:33 +02001397}
Izik Eidusb4231d62007-11-20 11:49:33 +02001398EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
Izik Eidus8a7ae052007-10-18 11:09:33 +02001399
Anthony Liguori35149e22008-04-02 14:46:56 -05001400void kvm_release_pfn_dirty(pfn_t pfn)
1401{
1402 kvm_set_pfn_dirty(pfn);
1403 kvm_release_pfn_clean(pfn);
1404}
1405EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1406
1407void kvm_set_page_dirty(struct page *page)
1408{
1409 kvm_set_pfn_dirty(page_to_pfn(page));
1410}
1411EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1412
1413void kvm_set_pfn_dirty(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 struct page *page = pfn_to_page(pfn);
1417 if (!PageReserved(page))
1418 SetPageDirty(page);
1419 }
Anthony Liguori35149e22008-04-02 14:46:56 -05001420}
1421EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1422
1423void kvm_set_pfn_accessed(pfn_t pfn)
1424{
Xiantao Zhangc77fb9d2008-09-27 10:55:40 +08001425 if (!kvm_is_mmio_pfn(pfn))
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001426 mark_page_accessed(pfn_to_page(pfn));
Anthony Liguori35149e22008-04-02 14:46:56 -05001427}
1428EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1429
1430void kvm_get_pfn(pfn_t pfn)
1431{
Xiantao Zhangc77fb9d2008-09-27 10:55:40 +08001432 if (!kvm_is_mmio_pfn(pfn))
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001433 get_page(pfn_to_page(pfn));
Anthony Liguori35149e22008-04-02 14:46:56 -05001434}
1435EXPORT_SYMBOL_GPL(kvm_get_pfn);
1436
Izik Eidus195aefd2007-10-01 22:14:18 +02001437static int next_segment(unsigned long len, int offset)
1438{
1439 if (len > PAGE_SIZE - offset)
1440 return PAGE_SIZE - offset;
1441 else
1442 return len;
1443}
1444
1445int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1446 int len)
1447{
Izik Eiduse0506bc2007-11-11 22:10:22 +02001448 int r;
1449 unsigned long addr;
Izik Eidus195aefd2007-10-01 22:14:18 +02001450
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001451 addr = gfn_to_hva_read(kvm, gfn);
Izik Eiduse0506bc2007-11-11 22:10:22 +02001452 if (kvm_is_error_hva(addr))
Izik Eidus195aefd2007-10-01 22:14:18 +02001453 return -EFAULT;
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001454 r = kvm_read_hva(data, (void __user *)addr + offset, len);
Izik Eiduse0506bc2007-11-11 22:10:22 +02001455 if (r)
1456 return -EFAULT;
Izik Eidus195aefd2007-10-01 22:14:18 +02001457 return 0;
1458}
1459EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1460
1461int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1462{
1463 gfn_t gfn = gpa >> PAGE_SHIFT;
1464 int seg;
1465 int offset = offset_in_page(gpa);
1466 int ret;
1467
1468 while ((seg = next_segment(len, offset)) != 0) {
1469 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1470 if (ret < 0)
1471 return ret;
1472 offset = 0;
1473 len -= seg;
1474 data += seg;
1475 ++gfn;
1476 }
1477 return 0;
1478}
1479EXPORT_SYMBOL_GPL(kvm_read_guest);
1480
Marcelo Tosatti7ec54582007-12-20 19:18:23 -05001481int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1482 unsigned long len)
1483{
1484 int r;
1485 unsigned long addr;
1486 gfn_t gfn = gpa >> PAGE_SHIFT;
1487 int offset = offset_in_page(gpa);
1488
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001489 addr = gfn_to_hva_read(kvm, gfn);
Marcelo Tosatti7ec54582007-12-20 19:18:23 -05001490 if (kvm_is_error_hva(addr))
1491 return -EFAULT;
Andrea Arcangeli0aac03f2008-01-30 19:57:35 +01001492 pagefault_disable();
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001493 r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
Andrea Arcangeli0aac03f2008-01-30 19:57:35 +01001494 pagefault_enable();
Marcelo Tosatti7ec54582007-12-20 19:18:23 -05001495 if (r)
1496 return -EFAULT;
1497 return 0;
1498}
1499EXPORT_SYMBOL(kvm_read_guest_atomic);
1500
Izik Eidus195aefd2007-10-01 22:14:18 +02001501int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1502 int offset, int len)
1503{
Izik Eiduse0506bc2007-11-11 22:10:22 +02001504 int r;
1505 unsigned long addr;
Izik Eidus195aefd2007-10-01 22:14:18 +02001506
Izik Eiduse0506bc2007-11-11 22:10:22 +02001507 addr = gfn_to_hva(kvm, gfn);
1508 if (kvm_is_error_hva(addr))
Izik Eidus195aefd2007-10-01 22:14:18 +02001509 return -EFAULT;
Xiao Guangrong8b0cedf2011-05-15 23:22:04 +08001510 r = __copy_to_user((void __user *)addr + offset, data, len);
Izik Eiduse0506bc2007-11-11 22:10:22 +02001511 if (r)
1512 return -EFAULT;
Izik Eidus195aefd2007-10-01 22:14:18 +02001513 mark_page_dirty(kvm, gfn);
1514 return 0;
1515}
1516EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1517
1518int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1519 unsigned long len)
1520{
1521 gfn_t gfn = gpa >> PAGE_SHIFT;
1522 int seg;
1523 int offset = offset_in_page(gpa);
1524 int ret;
1525
1526 while ((seg = next_segment(len, offset)) != 0) {
1527 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1528 if (ret < 0)
1529 return ret;
1530 offset = 0;
1531 len -= seg;
1532 data += seg;
1533 ++gfn;
1534 }
1535 return 0;
1536}
1537
Gleb Natapov49c77542010-10-18 15:22:23 +02001538int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
Andrew Honig8f964522013-03-29 09:35:21 -07001539 gpa_t gpa, unsigned long len)
Gleb Natapov49c77542010-10-18 15:22:23 +02001540{
1541 struct kvm_memslots *slots = kvm_memslots(kvm);
1542 int offset = offset_in_page(gpa);
Andrew Honig8f964522013-03-29 09:35:21 -07001543 gfn_t start_gfn = gpa >> PAGE_SHIFT;
1544 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1545 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1546 gfn_t nr_pages_avail;
Gleb Natapov49c77542010-10-18 15:22:23 +02001547
1548 ghc->gpa = gpa;
1549 ghc->generation = slots->generation;
Andrew Honig8f964522013-03-29 09:35:21 -07001550 ghc->len = len;
1551 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1552 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, &nr_pages_avail);
1553 if (!kvm_is_error_hva(ghc->hva) && nr_pages_avail >= nr_pages_needed) {
Gleb Natapov49c77542010-10-18 15:22:23 +02001554 ghc->hva += offset;
Andrew Honig8f964522013-03-29 09:35:21 -07001555 } else {
1556 /*
1557 * If the requested region crosses two memslots, we still
1558 * verify that the entire region is valid here.
1559 */
1560 while (start_gfn <= end_gfn) {
1561 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1562 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1563 &nr_pages_avail);
1564 if (kvm_is_error_hva(ghc->hva))
1565 return -EFAULT;
1566 start_gfn += nr_pages_avail;
1567 }
1568 /* Use the slow path for cross page reads and writes. */
1569 ghc->memslot = NULL;
1570 }
Gleb Natapov49c77542010-10-18 15:22:23 +02001571 return 0;
1572}
1573EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1574
1575int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1576 void *data, unsigned long len)
1577{
1578 struct kvm_memslots *slots = kvm_memslots(kvm);
1579 int r;
1580
Andrew Honig8f964522013-03-29 09:35:21 -07001581 BUG_ON(len > ghc->len);
1582
Gleb Natapov49c77542010-10-18 15:22:23 +02001583 if (slots->generation != ghc->generation)
Andrew Honig8f964522013-03-29 09:35:21 -07001584 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1585
1586 if (unlikely(!ghc->memslot))
1587 return kvm_write_guest(kvm, ghc->gpa, data, len);
Gleb Natapov49c77542010-10-18 15:22:23 +02001588
1589 if (kvm_is_error_hva(ghc->hva))
1590 return -EFAULT;
1591
Xiao Guangrong8b0cedf2011-05-15 23:22:04 +08001592 r = __copy_to_user((void __user *)ghc->hva, data, len);
Gleb Natapov49c77542010-10-18 15:22:23 +02001593 if (r)
1594 return -EFAULT;
1595 mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1596
1597 return 0;
1598}
1599EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1600
Gleb Natapove03b6442011-07-11 15:28:11 -04001601int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1602 void *data, unsigned long len)
1603{
1604 struct kvm_memslots *slots = kvm_memslots(kvm);
1605 int r;
1606
Andrew Honig8f964522013-03-29 09:35:21 -07001607 BUG_ON(len > ghc->len);
1608
Gleb Natapove03b6442011-07-11 15:28:11 -04001609 if (slots->generation != ghc->generation)
Andrew Honig8f964522013-03-29 09:35:21 -07001610 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1611
1612 if (unlikely(!ghc->memslot))
1613 return kvm_read_guest(kvm, ghc->gpa, data, len);
Gleb Natapove03b6442011-07-11 15:28:11 -04001614
1615 if (kvm_is_error_hva(ghc->hva))
1616 return -EFAULT;
1617
1618 r = __copy_from_user(data, (void __user *)ghc->hva, len);
1619 if (r)
1620 return -EFAULT;
1621
1622 return 0;
1623}
1624EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1625
Izik Eidus195aefd2007-10-01 22:14:18 +02001626int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1627{
Heiko Carstens3bcc8a82010-10-27 17:21:21 +02001628 return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1629 offset, len);
Izik Eidus195aefd2007-10-01 22:14:18 +02001630}
1631EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1632
1633int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1634{
1635 gfn_t gfn = gpa >> PAGE_SHIFT;
1636 int seg;
1637 int offset = offset_in_page(gpa);
1638 int ret;
1639
1640 while ((seg = next_segment(len, offset)) != 0) {
1641 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1642 if (ret < 0)
1643 return ret;
1644 offset = 0;
1645 len -= seg;
1646 ++gfn;
1647 }
1648 return 0;
1649}
1650EXPORT_SYMBOL_GPL(kvm_clear_guest);
1651
Gleb Natapov49c77542010-10-18 15:22:23 +02001652void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1653 gfn_t gfn)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001654{
Rusty Russell7e9d6192007-07-31 20:41:14 +10001655 if (memslot && memslot->dirty_bitmap) {
1656 unsigned long rel_gfn = gfn - memslot->base_gfn;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001657
Takuya Yoshikawab74ca3b2012-10-04 17:13:12 -07001658 set_bit_le(rel_gfn, memslot->dirty_bitmap);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001659 }
1660}
1661
Gleb Natapov49c77542010-10-18 15:22:23 +02001662void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1663{
1664 struct kvm_memory_slot *memslot;
1665
1666 memslot = gfn_to_memslot(kvm, gfn);
1667 mark_page_dirty_in_slot(kvm, memslot, gfn);
1668}
1669
Eddie Dongb6958ce2007-07-18 12:15:21 +03001670/*
1671 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1672 */
Hollis Blanchard8776e512007-10-31 17:24:24 -05001673void kvm_vcpu_block(struct kvm_vcpu *vcpu)
Eddie Dongb6958ce2007-07-18 12:15:21 +03001674{
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001675 DEFINE_WAIT(wait);
Eddie Dongb6958ce2007-07-18 12:15:21 +03001676
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001677 for (;;) {
1678 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
Eddie Dongb6958ce2007-07-18 12:15:21 +03001679
Gleb Natapova1b37102009-07-09 15:33:52 +03001680 if (kvm_arch_vcpu_runnable(vcpu)) {
Avi Kivitya8eeb042010-05-10 12:34:53 +03001681 kvm_make_request(KVM_REQ_UNHALT, vcpu);
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001682 break;
Marcelo Tosattid7690172008-09-08 15:23:48 -03001683 }
Gleb Natapov09cec752009-03-23 15:11:44 +02001684 if (kvm_cpu_has_pending_timer(vcpu))
1685 break;
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001686 if (signal_pending(current))
1687 break;
1688
Eddie Dongb6958ce2007-07-18 12:15:21 +03001689 schedule();
Eddie Dongb6958ce2007-07-18 12:15:21 +03001690 }
1691
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001692 finish_wait(&vcpu->wq, &wait);
Eddie Dongb6958ce2007-07-18 12:15:21 +03001693}
1694
Marcelo Tosatti8c847802012-03-14 17:58:48 -03001695#ifndef CONFIG_S390
Christoffer Dallb6d33832012-03-08 16:44:24 -05001696/*
1697 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1698 */
1699void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1700{
1701 int me;
1702 int cpu = vcpu->cpu;
1703 wait_queue_head_t *wqp;
1704
1705 wqp = kvm_arch_vcpu_wq(vcpu);
1706 if (waitqueue_active(wqp)) {
1707 wake_up_interruptible(wqp);
1708 ++vcpu->stat.halt_wakeup;
1709 }
1710
1711 me = get_cpu();
1712 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1713 if (kvm_arch_vcpu_should_kick(vcpu))
1714 smp_send_reschedule(cpu);
1715 put_cpu();
1716}
Yang Zhanga20ed542013-04-11 19:25:15 +08001717EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
Marcelo Tosatti8c847802012-03-14 17:58:48 -03001718#endif /* !CONFIG_S390 */
Christoffer Dallb6d33832012-03-08 16:44:24 -05001719
Avi Kivity6aa8b732006-12-10 02:21:36 -08001720void kvm_resched(struct kvm_vcpu *vcpu)
1721{
Yaozu Dong3fca0362007-04-25 16:49:19 +03001722 if (!need_resched())
1723 return;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001724 cond_resched();
Avi Kivity6aa8b732006-12-10 02:21:36 -08001725}
1726EXPORT_SYMBOL_GPL(kvm_resched);
1727
Konstantin Weitz41628d32012-04-25 15:30:38 +02001728bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1729{
1730 struct pid *pid;
1731 struct task_struct *task = NULL;
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301732 bool ret = false;
Konstantin Weitz41628d32012-04-25 15:30:38 +02001733
1734 rcu_read_lock();
1735 pid = rcu_dereference(target->pid);
1736 if (pid)
1737 task = get_pid_task(target->pid, PIDTYPE_PID);
1738 rcu_read_unlock();
1739 if (!task)
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301740 return ret;
Konstantin Weitz41628d32012-04-25 15:30:38 +02001741 if (task->flags & PF_VCPU) {
1742 put_task_struct(task);
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301743 return ret;
Konstantin Weitz41628d32012-04-25 15:30:38 +02001744 }
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301745 ret = yield_to(task, 1);
Konstantin Weitz41628d32012-04-25 15:30:38 +02001746 put_task_struct(task);
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301747
1748 return ret;
Konstantin Weitz41628d32012-04-25 15:30:38 +02001749}
1750EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1751
Raghavendra K T06e48c52012-07-19 15:17:52 +05301752#ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1753/*
1754 * Helper that checks whether a VCPU is eligible for directed yield.
1755 * Most eligible candidate to yield is decided by following heuristics:
1756 *
1757 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1758 * (preempted lock holder), indicated by @in_spin_loop.
1759 * Set at the beiginning and cleared at the end of interception/PLE handler.
1760 *
1761 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1762 * chance last time (mostly it has become eligible now since we have probably
1763 * yielded to lockholder in last iteration. This is done by toggling
1764 * @dy_eligible each time a VCPU checked for eligibility.)
1765 *
1766 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1767 * to preempted lock-holder could result in wrong VCPU selection and CPU
1768 * burning. Giving priority for a potential lock-holder increases lock
1769 * progress.
1770 *
1771 * Since algorithm is based on heuristics, accessing another VCPU data without
1772 * locking does not harm. It may result in trying to yield to same VCPU, fail
1773 * and continue with next VCPU and so on.
1774 */
1775bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1776{
1777 bool eligible;
1778
1779 eligible = !vcpu->spin_loop.in_spin_loop ||
1780 (vcpu->spin_loop.in_spin_loop &&
1781 vcpu->spin_loop.dy_eligible);
1782
1783 if (vcpu->spin_loop.in_spin_loop)
1784 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1785
1786 return eligible;
1787}
1788#endif
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301789
Rik van Riel217ece62011-02-01 09:53:28 -05001790void kvm_vcpu_on_spin(struct kvm_vcpu *me)
Zhai, Edwind255f4f2009-10-09 18:03:20 +08001791{
Rik van Riel217ece62011-02-01 09:53:28 -05001792 struct kvm *kvm = me->kvm;
1793 struct kvm_vcpu *vcpu;
1794 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1795 int yielded = 0;
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301796 int try = 3;
Rik van Riel217ece62011-02-01 09:53:28 -05001797 int pass;
1798 int i;
Zhai, Edwind255f4f2009-10-09 18:03:20 +08001799
Raghavendra K T4c088492012-07-18 19:07:46 +05301800 kvm_vcpu_set_in_spin_loop(me, true);
Rik van Riel217ece62011-02-01 09:53:28 -05001801 /*
1802 * We boost the priority of a VCPU that is runnable but not
1803 * currently running, because it got preempted by something
1804 * else and called schedule in __vcpu_run. Hopefully that
1805 * VCPU is holding the lock that we need and will release it.
1806 * We approximate round-robin by starting at the last boosted VCPU.
1807 */
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301808 for (pass = 0; pass < 2 && !yielded && try; pass++) {
Rik van Riel217ece62011-02-01 09:53:28 -05001809 kvm_for_each_vcpu(i, vcpu, kvm) {
Rik van Riel5cfc2aa2012-06-19 16:51:04 -04001810 if (!pass && i <= last_boosted_vcpu) {
Rik van Riel217ece62011-02-01 09:53:28 -05001811 i = last_boosted_vcpu;
1812 continue;
1813 } else if (pass && i > last_boosted_vcpu)
1814 break;
Raghavendra K T7bc7ae22013-03-04 23:32:27 +05301815 if (!ACCESS_ONCE(vcpu->preempted))
1816 continue;
Rik van Riel217ece62011-02-01 09:53:28 -05001817 if (vcpu == me)
1818 continue;
1819 if (waitqueue_active(&vcpu->wq))
1820 continue;
Raghavendra K T06e48c52012-07-19 15:17:52 +05301821 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1822 continue;
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301823
1824 yielded = kvm_vcpu_yield_to(vcpu);
1825 if (yielded > 0) {
Rik van Riel217ece62011-02-01 09:53:28 -05001826 kvm->last_boosted_vcpu = i;
Rik van Riel217ece62011-02-01 09:53:28 -05001827 break;
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301828 } else if (yielded < 0) {
1829 try--;
1830 if (!try)
1831 break;
Rik van Riel217ece62011-02-01 09:53:28 -05001832 }
Rik van Riel217ece62011-02-01 09:53:28 -05001833 }
1834 }
Raghavendra K T4c088492012-07-18 19:07:46 +05301835 kvm_vcpu_set_in_spin_loop(me, false);
Raghavendra K T06e48c52012-07-19 15:17:52 +05301836
1837 /* Ensure vcpu is not eligible during next spinloop */
1838 kvm_vcpu_set_dy_eligible(me, false);
Zhai, Edwind255f4f2009-10-09 18:03:20 +08001839}
1840EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1841
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001842static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001843{
1844 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001845 struct page *page;
1846
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001847 if (vmf->pgoff == 0)
Avi Kivity039576c2007-03-20 12:46:50 +02001848 page = virt_to_page(vcpu->run);
Avi Kivity09566762008-01-23 18:14:23 +02001849#ifdef CONFIG_X86
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001850 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
Zhang Xiantaoad312c72007-12-13 23:50:52 +08001851 page = virt_to_page(vcpu->arch.pio_data);
Avi Kivity09566762008-01-23 18:14:23 +02001852#endif
Laurent Vivier5f94c172008-05-30 16:05:54 +02001853#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1854 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1855 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1856#endif
Avi Kivity039576c2007-03-20 12:46:50 +02001857 else
Carsten Otte5b1c1492012-01-04 10:25:23 +01001858 return kvm_arch_vcpu_fault(vcpu, vmf);
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001859 get_page(page);
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001860 vmf->page = page;
1861 return 0;
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001862}
1863
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04001864static const struct vm_operations_struct kvm_vcpu_vm_ops = {
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001865 .fault = kvm_vcpu_fault,
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001866};
1867
1868static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1869{
1870 vma->vm_ops = &kvm_vcpu_vm_ops;
1871 return 0;
1872}
1873
Avi Kivitybccf2152007-02-21 18:04:26 +02001874static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1875{
1876 struct kvm_vcpu *vcpu = filp->private_data;
1877
Al Viro66c0b392008-04-19 20:33:56 +01001878 kvm_put_kvm(vcpu->kvm);
Avi Kivitybccf2152007-02-21 18:04:26 +02001879 return 0;
1880}
1881
Christian Borntraeger3d3aab12008-12-02 11:17:32 +01001882static struct file_operations kvm_vcpu_fops = {
Avi Kivitybccf2152007-02-21 18:04:26 +02001883 .release = kvm_vcpu_release,
1884 .unlocked_ioctl = kvm_vcpu_ioctl,
Alexander Graf1dda6062011-06-08 02:45:37 +02001885#ifdef CONFIG_COMPAT
1886 .compat_ioctl = kvm_vcpu_compat_ioctl,
1887#endif
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001888 .mmap = kvm_vcpu_mmap,
Arnd Bergmann6038f372010-08-15 18:52:59 +02001889 .llseek = noop_llseek,
Avi Kivitybccf2152007-02-21 18:04:26 +02001890};
1891
1892/*
1893 * Allocates an inode for the vcpu.
1894 */
1895static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1896{
Roland Dreier628ff7c2009-12-18 09:41:24 -08001897 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
Avi Kivitybccf2152007-02-21 18:04:26 +02001898}
1899
Avi Kivityc5ea7662007-02-20 18:41:05 +02001900/*
1901 * Creates some virtual cpus. Good luck creating more than one.
1902 */
Gleb Natapov73880c82009-06-09 15:56:28 +03001903static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
Avi Kivityc5ea7662007-02-20 18:41:05 +02001904{
1905 int r;
Gleb Natapov988a2ca2009-06-09 15:56:29 +03001906 struct kvm_vcpu *vcpu, *v;
Avi Kivityc5ea7662007-02-20 18:41:05 +02001907
Andy Honigf4d83822013-11-18 16:09:22 -08001908 if (id >= KVM_MAX_VCPUS)
1909 return -EINVAL;
1910
Gleb Natapov73880c82009-06-09 15:56:28 +03001911 vcpu = kvm_arch_vcpu_create(kvm, id);
Rusty Russellfb3f0f52007-07-27 17:16:56 +10001912 if (IS_ERR(vcpu))
1913 return PTR_ERR(vcpu);
Avi Kivityc5ea7662007-02-20 18:41:05 +02001914
Avi Kivity15ad7142007-07-11 18:17:21 +03001915 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1916
Avi Kivity26e52152007-11-20 15:30:24 +02001917 r = kvm_arch_vcpu_setup(vcpu);
1918 if (r)
Jan Kiszkad7805922011-05-23 10:33:05 +02001919 goto vcpu_destroy;
Avi Kivity26e52152007-11-20 15:30:24 +02001920
Shaohua Li11ec2802007-07-23 14:51:37 +08001921 mutex_lock(&kvm->lock);
Avi Kivity3e515702012-03-05 14:23:29 +02001922 if (!kvm_vcpu_compatible(vcpu)) {
1923 r = -EINVAL;
1924 goto unlock_vcpu_destroy;
1925 }
Gleb Natapov73880c82009-06-09 15:56:28 +03001926 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1927 r = -EINVAL;
Jan Kiszkad7805922011-05-23 10:33:05 +02001928 goto unlock_vcpu_destroy;
Rusty Russellfb3f0f52007-07-27 17:16:56 +10001929 }
Gleb Natapov73880c82009-06-09 15:56:28 +03001930
Gleb Natapov988a2ca2009-06-09 15:56:29 +03001931 kvm_for_each_vcpu(r, v, kvm)
1932 if (v->vcpu_id == id) {
Gleb Natapov73880c82009-06-09 15:56:28 +03001933 r = -EEXIST;
Jan Kiszkad7805922011-05-23 10:33:05 +02001934 goto unlock_vcpu_destroy;
Gleb Natapov73880c82009-06-09 15:56:28 +03001935 }
1936
1937 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
Rusty Russellfb3f0f52007-07-27 17:16:56 +10001938
1939 /* Now it's all set up, let userspace reach it */
Al Viro66c0b392008-04-19 20:33:56 +01001940 kvm_get_kvm(kvm);
Avi Kivitybccf2152007-02-21 18:04:26 +02001941 r = create_vcpu_fd(vcpu);
Gleb Natapov73880c82009-06-09 15:56:28 +03001942 if (r < 0) {
1943 kvm_put_kvm(kvm);
Jan Kiszkad7805922011-05-23 10:33:05 +02001944 goto unlock_vcpu_destroy;
Gleb Natapov73880c82009-06-09 15:56:28 +03001945 }
1946
1947 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1948 smp_wmb();
1949 atomic_inc(&kvm->online_vcpus);
1950
Gleb Natapov73880c82009-06-09 15:56:28 +03001951 mutex_unlock(&kvm->lock);
Marcelo Tosatti42897d82012-11-27 23:29:02 -02001952 kvm_arch_vcpu_postcreate(vcpu);
Avi Kivitybccf2152007-02-21 18:04:26 +02001953 return r;
Avi Kivityc5ea7662007-02-20 18:41:05 +02001954
Jan Kiszkad7805922011-05-23 10:33:05 +02001955unlock_vcpu_destroy:
Glauber Costa7d8fece2008-09-17 23:16:59 -03001956 mutex_unlock(&kvm->lock);
Jan Kiszkad7805922011-05-23 10:33:05 +02001957vcpu_destroy:
Hollis Blanchardd40ccc62007-11-19 14:04:43 -06001958 kvm_arch_vcpu_destroy(vcpu);
Avi Kivityc5ea7662007-02-20 18:41:05 +02001959 return r;
1960}
1961
Avi Kivity1961d272007-03-05 19:46:05 +02001962static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1963{
1964 if (sigset) {
1965 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1966 vcpu->sigset_active = 1;
1967 vcpu->sigset = *sigset;
1968 } else
1969 vcpu->sigset_active = 0;
1970 return 0;
1971}
1972
Avi Kivitybccf2152007-02-21 18:04:26 +02001973static long kvm_vcpu_ioctl(struct file *filp,
1974 unsigned int ioctl, unsigned long arg)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001975{
Avi Kivitybccf2152007-02-21 18:04:26 +02001976 struct kvm_vcpu *vcpu = filp->private_data;
Al Viro2f3669872007-02-09 16:38:35 +00001977 void __user *argp = (void __user *)arg;
Carsten Otte313a3dc2007-10-11 19:16:52 +02001978 int r;
Dave Hansenfa3795a2008-08-11 10:01:46 -07001979 struct kvm_fpu *fpu = NULL;
1980 struct kvm_sregs *kvm_sregs = NULL;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001981
Avi Kivity6d4e4c42007-11-21 16:41:05 +02001982 if (vcpu->kvm->mm != current->mm)
1983 return -EIO;
Avi Kivity2122ff52010-05-13 11:25:04 +03001984
David Matlack7889ddd2014-09-19 16:03:25 -07001985 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
1986 return -EINVAL;
1987
Sanjay Lal2f4d9b52012-11-21 18:34:15 -08001988#if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
Avi Kivity2122ff52010-05-13 11:25:04 +03001989 /*
1990 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1991 * so vcpu_load() would break it.
1992 */
1993 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1994 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1995#endif
1996
1997
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +03001998 r = vcpu_load(vcpu);
1999 if (r)
2000 return r;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002001 switch (ioctl) {
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02002002 case KVM_RUN:
Avi Kivityf0fe5102007-03-07 13:11:17 +02002003 r = -EINVAL;
2004 if (arg)
2005 goto out;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05002006 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
Gleb Natapov64be5002010-10-24 16:49:08 +02002007 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002008 break;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002009 case KVM_GET_REGS: {
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002010 struct kvm_regs *kvm_regs;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002011
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002012 r = -ENOMEM;
2013 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2014 if (!kvm_regs)
2015 goto out;
2016 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002017 if (r)
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002018 goto out_free1;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002019 r = -EFAULT;
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002020 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2021 goto out_free1;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002022 r = 0;
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002023out_free1:
2024 kfree(kvm_regs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002025 break;
2026 }
2027 case KVM_SET_REGS: {
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002028 struct kvm_regs *kvm_regs;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002029
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002030 r = -ENOMEM;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002031 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2032 if (IS_ERR(kvm_regs)) {
2033 r = PTR_ERR(kvm_regs);
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002034 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002035 }
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002036 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002037 kfree(kvm_regs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002038 break;
2039 }
2040 case KVM_GET_SREGS: {
Dave Hansenfa3795a2008-08-11 10:01:46 -07002041 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2042 r = -ENOMEM;
2043 if (!kvm_sregs)
2044 goto out;
2045 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002046 if (r)
2047 goto out;
2048 r = -EFAULT;
Dave Hansenfa3795a2008-08-11 10:01:46 -07002049 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002050 goto out;
2051 r = 0;
2052 break;
2053 }
2054 case KVM_SET_SREGS: {
Sasha Levinff5c2c02011-12-04 19:36:29 +02002055 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2056 if (IS_ERR(kvm_sregs)) {
2057 r = PTR_ERR(kvm_sregs);
Guo Chao18595412012-11-02 18:33:21 +08002058 kvm_sregs = NULL;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002059 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002060 }
Dave Hansenfa3795a2008-08-11 10:01:46 -07002061 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002062 break;
2063 }
Marcelo Tosatti62d9f0d2008-04-11 13:24:45 -03002064 case KVM_GET_MP_STATE: {
2065 struct kvm_mp_state mp_state;
2066
2067 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2068 if (r)
2069 goto out;
2070 r = -EFAULT;
2071 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2072 goto out;
2073 r = 0;
2074 break;
2075 }
2076 case KVM_SET_MP_STATE: {
2077 struct kvm_mp_state mp_state;
2078
2079 r = -EFAULT;
2080 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2081 goto out;
2082 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
Marcelo Tosatti62d9f0d2008-04-11 13:24:45 -03002083 break;
2084 }
Avi Kivity6aa8b732006-12-10 02:21:36 -08002085 case KVM_TRANSLATE: {
2086 struct kvm_translation tr;
2087
2088 r = -EFAULT;
Al Viro2f3669872007-02-09 16:38:35 +00002089 if (copy_from_user(&tr, argp, sizeof tr))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002090 goto out;
Zhang Xiantao8b006792007-11-16 13:05:55 +08002091 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002092 if (r)
2093 goto out;
2094 r = -EFAULT;
Al Viro2f3669872007-02-09 16:38:35 +00002095 if (copy_to_user(argp, &tr, sizeof tr))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002096 goto out;
2097 r = 0;
2098 break;
2099 }
Jan Kiszkad0bfb942008-12-15 13:52:10 +01002100 case KVM_SET_GUEST_DEBUG: {
2101 struct kvm_guest_debug dbg;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002102
2103 r = -EFAULT;
Al Viro2f3669872007-02-09 16:38:35 +00002104 if (copy_from_user(&dbg, argp, sizeof dbg))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002105 goto out;
Jan Kiszkad0bfb942008-12-15 13:52:10 +01002106 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002107 break;
2108 }
Avi Kivity1961d272007-03-05 19:46:05 +02002109 case KVM_SET_SIGNAL_MASK: {
2110 struct kvm_signal_mask __user *sigmask_arg = argp;
2111 struct kvm_signal_mask kvm_sigmask;
2112 sigset_t sigset, *p;
2113
2114 p = NULL;
2115 if (argp) {
2116 r = -EFAULT;
2117 if (copy_from_user(&kvm_sigmask, argp,
2118 sizeof kvm_sigmask))
2119 goto out;
2120 r = -EINVAL;
2121 if (kvm_sigmask.len != sizeof sigset)
2122 goto out;
2123 r = -EFAULT;
2124 if (copy_from_user(&sigset, sigmask_arg->sigset,
2125 sizeof sigset))
2126 goto out;
2127 p = &sigset;
2128 }
Andi Kleen376d41f2010-06-10 13:10:47 +02002129 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
Avi Kivity1961d272007-03-05 19:46:05 +02002130 break;
2131 }
Avi Kivityb8836732007-04-01 16:34:31 +03002132 case KVM_GET_FPU: {
Dave Hansenfa3795a2008-08-11 10:01:46 -07002133 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2134 r = -ENOMEM;
2135 if (!fpu)
2136 goto out;
2137 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
Avi Kivityb8836732007-04-01 16:34:31 +03002138 if (r)
2139 goto out;
2140 r = -EFAULT;
Dave Hansenfa3795a2008-08-11 10:01:46 -07002141 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
Avi Kivityb8836732007-04-01 16:34:31 +03002142 goto out;
2143 r = 0;
2144 break;
2145 }
2146 case KVM_SET_FPU: {
Sasha Levinff5c2c02011-12-04 19:36:29 +02002147 fpu = memdup_user(argp, sizeof(*fpu));
2148 if (IS_ERR(fpu)) {
2149 r = PTR_ERR(fpu);
Guo Chao18595412012-11-02 18:33:21 +08002150 fpu = NULL;
Avi Kivityb8836732007-04-01 16:34:31 +03002151 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002152 }
Dave Hansenfa3795a2008-08-11 10:01:46 -07002153 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
Avi Kivityb8836732007-04-01 16:34:31 +03002154 break;
2155 }
Avi Kivitybccf2152007-02-21 18:04:26 +02002156 default:
Carsten Otte313a3dc2007-10-11 19:16:52 +02002157 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
Avi Kivitybccf2152007-02-21 18:04:26 +02002158 }
2159out:
Avi Kivity2122ff52010-05-13 11:25:04 +03002160 vcpu_put(vcpu);
Dave Hansenfa3795a2008-08-11 10:01:46 -07002161 kfree(fpu);
2162 kfree(kvm_sregs);
Avi Kivitybccf2152007-02-21 18:04:26 +02002163 return r;
2164}
2165
Alexander Graf1dda6062011-06-08 02:45:37 +02002166#ifdef CONFIG_COMPAT
2167static long kvm_vcpu_compat_ioctl(struct file *filp,
2168 unsigned int ioctl, unsigned long arg)
2169{
2170 struct kvm_vcpu *vcpu = filp->private_data;
2171 void __user *argp = compat_ptr(arg);
2172 int r;
2173
2174 if (vcpu->kvm->mm != current->mm)
2175 return -EIO;
2176
2177 switch (ioctl) {
2178 case KVM_SET_SIGNAL_MASK: {
2179 struct kvm_signal_mask __user *sigmask_arg = argp;
2180 struct kvm_signal_mask kvm_sigmask;
2181 compat_sigset_t csigset;
2182 sigset_t sigset;
2183
2184 if (argp) {
2185 r = -EFAULT;
2186 if (copy_from_user(&kvm_sigmask, argp,
2187 sizeof kvm_sigmask))
2188 goto out;
2189 r = -EINVAL;
2190 if (kvm_sigmask.len != sizeof csigset)
2191 goto out;
2192 r = -EFAULT;
2193 if (copy_from_user(&csigset, sigmask_arg->sigset,
2194 sizeof csigset))
2195 goto out;
Alan Cox760a9a32012-08-22 14:34:11 +01002196 sigset_from_compat(&sigset, &csigset);
2197 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2198 } else
2199 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
Alexander Graf1dda6062011-06-08 02:45:37 +02002200 break;
2201 }
2202 default:
2203 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2204 }
2205
2206out:
2207 return r;
2208}
2209#endif
2210
Scott Wood852b6d52013-04-12 14:08:42 +00002211static int kvm_device_ioctl_attr(struct kvm_device *dev,
2212 int (*accessor)(struct kvm_device *dev,
2213 struct kvm_device_attr *attr),
2214 unsigned long arg)
2215{
2216 struct kvm_device_attr attr;
2217
2218 if (!accessor)
2219 return -EPERM;
2220
2221 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2222 return -EFAULT;
2223
2224 return accessor(dev, &attr);
2225}
2226
2227static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2228 unsigned long arg)
2229{
2230 struct kvm_device *dev = filp->private_data;
2231
2232 switch (ioctl) {
2233 case KVM_SET_DEVICE_ATTR:
2234 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2235 case KVM_GET_DEVICE_ATTR:
2236 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2237 case KVM_HAS_DEVICE_ATTR:
2238 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2239 default:
2240 if (dev->ops->ioctl)
2241 return dev->ops->ioctl(dev, ioctl, arg);
2242
2243 return -ENOTTY;
2244 }
2245}
2246
Scott Wood852b6d52013-04-12 14:08:42 +00002247static int kvm_device_release(struct inode *inode, struct file *filp)
2248{
2249 struct kvm_device *dev = filp->private_data;
2250 struct kvm *kvm = dev->kvm;
2251
Scott Wood852b6d52013-04-12 14:08:42 +00002252 kvm_put_kvm(kvm);
2253 return 0;
2254}
2255
2256static const struct file_operations kvm_device_fops = {
2257 .unlocked_ioctl = kvm_device_ioctl,
Scott Wooddb6ae612013-04-30 20:00:45 -05002258#ifdef CONFIG_COMPAT
2259 .compat_ioctl = kvm_device_ioctl,
2260#endif
Scott Wood852b6d52013-04-12 14:08:42 +00002261 .release = kvm_device_release,
2262};
2263
2264struct kvm_device *kvm_device_from_filp(struct file *filp)
2265{
2266 if (filp->f_op != &kvm_device_fops)
2267 return NULL;
2268
2269 return filp->private_data;
2270}
2271
2272static int kvm_ioctl_create_device(struct kvm *kvm,
2273 struct kvm_create_device *cd)
2274{
2275 struct kvm_device_ops *ops = NULL;
2276 struct kvm_device *dev;
2277 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2278 int ret;
2279
2280 switch (cd->type) {
Scott Wood5df554ad2013-04-12 14:08:46 +00002281#ifdef CONFIG_KVM_MPIC
2282 case KVM_DEV_TYPE_FSL_MPIC_20:
2283 case KVM_DEV_TYPE_FSL_MPIC_42:
2284 ops = &kvm_mpic_ops;
2285 break;
2286#endif
Paul Mackerras5975a2e2013-04-27 00:28:37 +00002287#ifdef CONFIG_KVM_XICS
2288 case KVM_DEV_TYPE_XICS:
2289 ops = &kvm_xics_ops;
2290 break;
2291#endif
Scott Wood852b6d52013-04-12 14:08:42 +00002292 default:
2293 return -ENODEV;
2294 }
2295
2296 if (test)
2297 return 0;
2298
2299 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2300 if (!dev)
2301 return -ENOMEM;
2302
2303 dev->ops = ops;
2304 dev->kvm = kvm;
Scott Wood852b6d52013-04-12 14:08:42 +00002305
2306 ret = ops->create(dev, cd->type);
2307 if (ret < 0) {
2308 kfree(dev);
2309 return ret;
2310 }
2311
2312 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR);
2313 if (ret < 0) {
2314 ops->destroy(dev);
2315 return ret;
2316 }
2317
Scott Wood07f0a7b2013-04-25 14:11:23 +00002318 list_add(&dev->vm_node, &kvm->devices);
Scott Wood852b6d52013-04-12 14:08:42 +00002319 kvm_get_kvm(kvm);
2320 cd->fd = ret;
2321 return 0;
2322}
2323
Avi Kivitybccf2152007-02-21 18:04:26 +02002324static long kvm_vm_ioctl(struct file *filp,
2325 unsigned int ioctl, unsigned long arg)
2326{
2327 struct kvm *kvm = filp->private_data;
2328 void __user *argp = (void __user *)arg;
Carsten Otte1fe779f2007-10-29 16:08:35 +01002329 int r;
Avi Kivitybccf2152007-02-21 18:04:26 +02002330
Avi Kivity6d4e4c42007-11-21 16:41:05 +02002331 if (kvm->mm != current->mm)
2332 return -EIO;
Avi Kivitybccf2152007-02-21 18:04:26 +02002333 switch (ioctl) {
2334 case KVM_CREATE_VCPU:
2335 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
Avi Kivitybccf2152007-02-21 18:04:26 +02002336 break;
Izik Eidus6fc138d2007-10-09 19:20:39 +02002337 case KVM_SET_USER_MEMORY_REGION: {
2338 struct kvm_userspace_memory_region kvm_userspace_mem;
2339
2340 r = -EFAULT;
2341 if (copy_from_user(&kvm_userspace_mem, argp,
2342 sizeof kvm_userspace_mem))
2343 goto out;
2344
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +09002345 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002346 break;
2347 }
2348 case KVM_GET_DIRTY_LOG: {
2349 struct kvm_dirty_log log;
2350
2351 r = -EFAULT;
Al Viro2f3669872007-02-09 16:38:35 +00002352 if (copy_from_user(&log, argp, sizeof log))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002353 goto out;
Avi Kivity2c6f5df2007-02-20 18:27:58 +02002354 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002355 break;
2356 }
Laurent Vivier5f94c172008-05-30 16:05:54 +02002357#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2358 case KVM_REGISTER_COALESCED_MMIO: {
2359 struct kvm_coalesced_mmio_zone zone;
2360 r = -EFAULT;
2361 if (copy_from_user(&zone, argp, sizeof zone))
2362 goto out;
Laurent Vivier5f94c172008-05-30 16:05:54 +02002363 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
Laurent Vivier5f94c172008-05-30 16:05:54 +02002364 break;
2365 }
2366 case KVM_UNREGISTER_COALESCED_MMIO: {
2367 struct kvm_coalesced_mmio_zone zone;
2368 r = -EFAULT;
2369 if (copy_from_user(&zone, argp, sizeof zone))
2370 goto out;
Laurent Vivier5f94c172008-05-30 16:05:54 +02002371 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
Laurent Vivier5f94c172008-05-30 16:05:54 +02002372 break;
2373 }
2374#endif
Gregory Haskins721eecbf2009-05-20 10:30:49 -04002375 case KVM_IRQFD: {
2376 struct kvm_irqfd data;
2377
2378 r = -EFAULT;
2379 if (copy_from_user(&data, argp, sizeof data))
2380 goto out;
Alex Williamsond4db2932012-06-29 09:56:08 -06002381 r = kvm_irqfd(kvm, &data);
Gregory Haskins721eecbf2009-05-20 10:30:49 -04002382 break;
2383 }
Gregory Haskinsd34e6b12009-07-07 17:08:49 -04002384 case KVM_IOEVENTFD: {
2385 struct kvm_ioeventfd data;
2386
2387 r = -EFAULT;
2388 if (copy_from_user(&data, argp, sizeof data))
2389 goto out;
2390 r = kvm_ioeventfd(kvm, &data);
2391 break;
2392 }
Gleb Natapov73880c82009-06-09 15:56:28 +03002393#ifdef CONFIG_KVM_APIC_ARCHITECTURE
2394 case KVM_SET_BOOT_CPU_ID:
2395 r = 0;
Marcelo Tosatti894a9c52009-06-23 15:05:14 -03002396 mutex_lock(&kvm->lock);
Gleb Natapov73880c82009-06-09 15:56:28 +03002397 if (atomic_read(&kvm->online_vcpus) != 0)
2398 r = -EBUSY;
2399 else
2400 kvm->bsp_vcpu_id = arg;
Marcelo Tosatti894a9c52009-06-23 15:05:14 -03002401 mutex_unlock(&kvm->lock);
Gleb Natapov73880c82009-06-09 15:56:28 +03002402 break;
2403#endif
Jan Kiszka07975ad2012-03-29 21:14:12 +02002404#ifdef CONFIG_HAVE_KVM_MSI
2405 case KVM_SIGNAL_MSI: {
2406 struct kvm_msi msi;
2407
2408 r = -EFAULT;
2409 if (copy_from_user(&msi, argp, sizeof msi))
2410 goto out;
2411 r = kvm_send_userspace_msi(kvm, &msi);
2412 break;
2413 }
2414#endif
Christoffer Dall23d43cf2012-07-24 08:51:20 -04002415#ifdef __KVM_HAVE_IRQ_LINE
2416 case KVM_IRQ_LINE_STATUS:
2417 case KVM_IRQ_LINE: {
2418 struct kvm_irq_level irq_event;
2419
2420 r = -EFAULT;
2421 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2422 goto out;
2423
Yang Zhangaa2fbe62013-04-11 19:21:40 +08002424 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2425 ioctl == KVM_IRQ_LINE_STATUS);
Christoffer Dall23d43cf2012-07-24 08:51:20 -04002426 if (r)
2427 goto out;
2428
2429 r = -EFAULT;
2430 if (ioctl == KVM_IRQ_LINE_STATUS) {
2431 if (copy_to_user(argp, &irq_event, sizeof irq_event))
2432 goto out;
2433 }
2434
2435 r = 0;
2436 break;
2437 }
2438#endif
Alexander Grafaa8d5942013-04-15 21:12:53 +02002439#ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2440 case KVM_SET_GSI_ROUTING: {
2441 struct kvm_irq_routing routing;
2442 struct kvm_irq_routing __user *urouting;
2443 struct kvm_irq_routing_entry *entries;
2444
2445 r = -EFAULT;
2446 if (copy_from_user(&routing, argp, sizeof(routing)))
2447 goto out;
2448 r = -EINVAL;
2449 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2450 goto out;
2451 if (routing.flags)
2452 goto out;
2453 r = -ENOMEM;
2454 entries = vmalloc(routing.nr * sizeof(*entries));
2455 if (!entries)
2456 goto out;
2457 r = -EFAULT;
2458 urouting = argp;
2459 if (copy_from_user(entries, urouting->entries,
2460 routing.nr * sizeof(*entries)))
2461 goto out_free_irq_routing;
2462 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2463 routing.flags);
2464 out_free_irq_routing:
2465 vfree(entries);
2466 break;
2467 }
2468#endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
Scott Wood852b6d52013-04-12 14:08:42 +00002469 case KVM_CREATE_DEVICE: {
2470 struct kvm_create_device cd;
2471
2472 r = -EFAULT;
2473 if (copy_from_user(&cd, argp, sizeof(cd)))
2474 goto out;
2475
2476 r = kvm_ioctl_create_device(kvm, &cd);
2477 if (r)
2478 goto out;
2479
2480 r = -EFAULT;
2481 if (copy_to_user(argp, &cd, sizeof(cd)))
2482 goto out;
2483
2484 r = 0;
2485 break;
2486 }
Avi Kivityf17abe92007-02-21 19:28:04 +02002487 default:
Carsten Otte1fe779f2007-10-29 16:08:35 +01002488 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
Avi Kivitybfd99ff2009-08-26 14:57:50 +03002489 if (r == -ENOTTY)
2490 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
Avi Kivityf17abe92007-02-21 19:28:04 +02002491 }
2492out:
2493 return r;
2494}
2495
Arnd Bergmann6ff58942009-10-22 14:19:27 +02002496#ifdef CONFIG_COMPAT
2497struct compat_kvm_dirty_log {
2498 __u32 slot;
2499 __u32 padding1;
2500 union {
2501 compat_uptr_t dirty_bitmap; /* one bit per page */
2502 __u64 padding2;
2503 };
2504};
2505
2506static long kvm_vm_compat_ioctl(struct file *filp,
2507 unsigned int ioctl, unsigned long arg)
2508{
2509 struct kvm *kvm = filp->private_data;
2510 int r;
2511
2512 if (kvm->mm != current->mm)
2513 return -EIO;
2514 switch (ioctl) {
2515 case KVM_GET_DIRTY_LOG: {
2516 struct compat_kvm_dirty_log compat_log;
2517 struct kvm_dirty_log log;
2518
2519 r = -EFAULT;
2520 if (copy_from_user(&compat_log, (void __user *)arg,
2521 sizeof(compat_log)))
2522 goto out;
2523 log.slot = compat_log.slot;
2524 log.padding1 = compat_log.padding1;
2525 log.padding2 = compat_log.padding2;
2526 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2527
2528 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
Arnd Bergmann6ff58942009-10-22 14:19:27 +02002529 break;
2530 }
2531 default:
2532 r = kvm_vm_ioctl(filp, ioctl, arg);
2533 }
2534
2535out:
2536 return r;
2537}
2538#endif
2539
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11002540static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Avi Kivityf17abe92007-02-21 19:28:04 +02002541{
Marcelo Tosatti777b3f42008-09-16 20:54:46 -03002542 struct page *page[1];
2543 unsigned long addr;
2544 int npages;
2545 gfn_t gfn = vmf->pgoff;
Avi Kivityf17abe92007-02-21 19:28:04 +02002546 struct kvm *kvm = vma->vm_file->private_data;
Avi Kivityf17abe92007-02-21 19:28:04 +02002547
Marcelo Tosatti777b3f42008-09-16 20:54:46 -03002548 addr = gfn_to_hva(kvm, gfn);
2549 if (kvm_is_error_hva(addr))
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11002550 return VM_FAULT_SIGBUS;
Marcelo Tosatti777b3f42008-09-16 20:54:46 -03002551
2552 npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2553 NULL);
2554 if (unlikely(npages != 1))
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11002555 return VM_FAULT_SIGBUS;
Marcelo Tosatti777b3f42008-09-16 20:54:46 -03002556
2557 vmf->page = page[0];
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11002558 return 0;
Avi Kivityf17abe92007-02-21 19:28:04 +02002559}
2560
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04002561static const struct vm_operations_struct kvm_vm_vm_ops = {
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11002562 .fault = kvm_vm_fault,
Avi Kivityf17abe92007-02-21 19:28:04 +02002563};
2564
2565static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2566{
2567 vma->vm_ops = &kvm_vm_vm_ops;
2568 return 0;
2569}
2570
Christian Borntraeger3d3aab12008-12-02 11:17:32 +01002571static struct file_operations kvm_vm_fops = {
Avi Kivityf17abe92007-02-21 19:28:04 +02002572 .release = kvm_vm_release,
2573 .unlocked_ioctl = kvm_vm_ioctl,
Arnd Bergmann6ff58942009-10-22 14:19:27 +02002574#ifdef CONFIG_COMPAT
2575 .compat_ioctl = kvm_vm_compat_ioctl,
2576#endif
Avi Kivityf17abe92007-02-21 19:28:04 +02002577 .mmap = kvm_vm_mmap,
Arnd Bergmann6038f372010-08-15 18:52:59 +02002578 .llseek = noop_llseek,
Avi Kivityf17abe92007-02-21 19:28:04 +02002579};
2580
Carsten Ottee08b9632012-01-04 10:25:20 +01002581static int kvm_dev_ioctl_create_vm(unsigned long type)
Avi Kivityf17abe92007-02-21 19:28:04 +02002582{
Heiko Carstensaac87632010-10-27 17:22:10 +02002583 int r;
Avi Kivityf17abe92007-02-21 19:28:04 +02002584 struct kvm *kvm;
2585
Carsten Ottee08b9632012-01-04 10:25:20 +01002586 kvm = kvm_create_vm(type);
Avi Kivityd6d28162007-06-28 08:38:16 -04002587 if (IS_ERR(kvm))
2588 return PTR_ERR(kvm);
Takuya Yoshikawa6ce5a092010-03-15 22:13:30 +09002589#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2590 r = kvm_coalesced_mmio_init(kvm);
2591 if (r < 0) {
2592 kvm_put_kvm(kvm);
2593 return r;
2594 }
2595#endif
Heiko Carstensaac87632010-10-27 17:22:10 +02002596 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2597 if (r < 0)
Al Viro66c0b392008-04-19 20:33:56 +01002598 kvm_put_kvm(kvm);
Avi Kivityf17abe92007-02-21 19:28:04 +02002599
Heiko Carstensaac87632010-10-27 17:22:10 +02002600 return r;
Avi Kivityf17abe92007-02-21 19:28:04 +02002601}
2602
Avi Kivity1a811b62008-12-08 18:25:27 +02002603static long kvm_dev_ioctl_check_extension_generic(long arg)
2604{
2605 switch (arg) {
Avi Kivityca9edae2008-12-08 18:29:29 +02002606 case KVM_CAP_USER_MEMORY:
Avi Kivity1a811b62008-12-08 18:25:27 +02002607 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
Jan Kiszka4cd481f2009-04-13 11:59:32 +02002608 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
Gleb Natapov73880c82009-06-09 15:56:28 +03002609#ifdef CONFIG_KVM_APIC_ARCHITECTURE
2610 case KVM_CAP_SET_BOOT_CPU_ID:
2611#endif
Avi Kivitya9c73992009-11-04 11:54:59 +02002612 case KVM_CAP_INTERNAL_ERROR_DATA:
Jan Kiszka07975ad2012-03-29 21:14:12 +02002613#ifdef CONFIG_HAVE_KVM_MSI
2614 case KVM_CAP_SIGNAL_MSI:
2615#endif
Alexander Graf7df35f52013-04-16 12:12:49 +02002616#ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2617 case KVM_CAP_IRQFD_RESAMPLE:
2618#endif
Avi Kivity1a811b62008-12-08 18:25:27 +02002619 return 1;
Alexander Grafa725d562013-04-17 13:29:30 +02002620#ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
Avi Kivity399ec802008-11-19 13:58:46 +02002621 case KVM_CAP_IRQ_ROUTING:
Sheng Yang36463142009-03-16 16:33:43 +08002622 return KVM_MAX_IRQ_ROUTES;
Avi Kivity399ec802008-11-19 13:58:46 +02002623#endif
Avi Kivity1a811b62008-12-08 18:25:27 +02002624 default:
2625 break;
2626 }
2627 return kvm_dev_ioctl_check_extension(arg);
2628}
2629
Avi Kivityf17abe92007-02-21 19:28:04 +02002630static long kvm_dev_ioctl(struct file *filp,
2631 unsigned int ioctl, unsigned long arg)
2632{
Avi Kivity07c45a32007-03-07 13:05:38 +02002633 long r = -EINVAL;
Avi Kivityf17abe92007-02-21 19:28:04 +02002634
2635 switch (ioctl) {
2636 case KVM_GET_API_VERSION:
Avi Kivityf0fe5102007-03-07 13:11:17 +02002637 r = -EINVAL;
2638 if (arg)
2639 goto out;
Avi Kivityf17abe92007-02-21 19:28:04 +02002640 r = KVM_API_VERSION;
2641 break;
2642 case KVM_CREATE_VM:
Carsten Ottee08b9632012-01-04 10:25:20 +01002643 r = kvm_dev_ioctl_create_vm(arg);
Avi Kivityf17abe92007-02-21 19:28:04 +02002644 break;
Zhang Xiantao018d00d2007-11-15 23:07:47 +08002645 case KVM_CHECK_EXTENSION:
Avi Kivity1a811b62008-12-08 18:25:27 +02002646 r = kvm_dev_ioctl_check_extension_generic(arg);
Avi Kivity5d308f42007-03-01 17:56:20 +02002647 break;
Avi Kivity07c45a32007-03-07 13:05:38 +02002648 case KVM_GET_VCPU_MMAP_SIZE:
2649 r = -EINVAL;
2650 if (arg)
2651 goto out;
Avi Kivityadb1ff42008-01-24 15:13:08 +02002652 r = PAGE_SIZE; /* struct kvm_run */
2653#ifdef CONFIG_X86
2654 r += PAGE_SIZE; /* pio data page */
2655#endif
Laurent Vivier5f94c172008-05-30 16:05:54 +02002656#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2657 r += PAGE_SIZE; /* coalesced mmio ring page */
2658#endif
Avi Kivity07c45a32007-03-07 13:05:38 +02002659 break;
Feng(Eric) Liud4c9ff22008-04-10 08:47:53 -04002660 case KVM_TRACE_ENABLE:
2661 case KVM_TRACE_PAUSE:
2662 case KVM_TRACE_DISABLE:
Marcelo Tosatti2023a292009-06-18 11:47:28 -03002663 r = -EOPNOTSUPP;
Feng(Eric) Liud4c9ff22008-04-10 08:47:53 -04002664 break;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002665 default:
Carsten Otte043405e2007-10-10 17:16:19 +02002666 return kvm_arch_dev_ioctl(filp, ioctl, arg);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002667 }
2668out:
2669 return r;
2670}
2671
Avi Kivity6aa8b732006-12-10 02:21:36 -08002672static struct file_operations kvm_chardev_ops = {
Avi Kivity6aa8b732006-12-10 02:21:36 -08002673 .unlocked_ioctl = kvm_dev_ioctl,
2674 .compat_ioctl = kvm_dev_ioctl,
Arnd Bergmann6038f372010-08-15 18:52:59 +02002675 .llseek = noop_llseek,
Avi Kivity6aa8b732006-12-10 02:21:36 -08002676};
2677
2678static struct miscdevice kvm_dev = {
Avi Kivitybbe44322007-03-04 13:27:36 +02002679 KVM_MINOR,
Avi Kivity6aa8b732006-12-10 02:21:36 -08002680 "kvm",
2681 &kvm_chardev_ops,
2682};
2683
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002684static void hardware_enable_nolock(void *junk)
Avi Kivity1b6c0162007-05-24 13:03:52 +03002685{
2686 int cpu = raw_smp_processor_id();
Alexander Graf10474ae2009-09-15 11:37:46 +02002687 int r;
Avi Kivity1b6c0162007-05-24 13:03:52 +03002688
Rusty Russell7f59f492008-12-07 21:25:45 +10302689 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
Avi Kivity1b6c0162007-05-24 13:03:52 +03002690 return;
Alexander Graf10474ae2009-09-15 11:37:46 +02002691
Rusty Russell7f59f492008-12-07 21:25:45 +10302692 cpumask_set_cpu(cpu, cpus_hardware_enabled);
Alexander Graf10474ae2009-09-15 11:37:46 +02002693
2694 r = kvm_arch_hardware_enable(NULL);
2695
2696 if (r) {
2697 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2698 atomic_inc(&hardware_enable_failed);
2699 printk(KERN_INFO "kvm: enabling virtualization on "
2700 "CPU%d failed\n", cpu);
2701 }
Avi Kivity1b6c0162007-05-24 13:03:52 +03002702}
2703
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002704static void hardware_enable(void *junk)
2705{
Jan Kiszkae935b832011-02-08 12:55:33 +01002706 raw_spin_lock(&kvm_lock);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002707 hardware_enable_nolock(junk);
Jan Kiszkae935b832011-02-08 12:55:33 +01002708 raw_spin_unlock(&kvm_lock);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002709}
2710
2711static void hardware_disable_nolock(void *junk)
Avi Kivity1b6c0162007-05-24 13:03:52 +03002712{
2713 int cpu = raw_smp_processor_id();
2714
Rusty Russell7f59f492008-12-07 21:25:45 +10302715 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
Avi Kivity1b6c0162007-05-24 13:03:52 +03002716 return;
Rusty Russell7f59f492008-12-07 21:25:45 +10302717 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08002718 kvm_arch_hardware_disable(NULL);
Avi Kivity1b6c0162007-05-24 13:03:52 +03002719}
2720
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002721static void hardware_disable(void *junk)
2722{
Jan Kiszkae935b832011-02-08 12:55:33 +01002723 raw_spin_lock(&kvm_lock);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002724 hardware_disable_nolock(junk);
Jan Kiszkae935b832011-02-08 12:55:33 +01002725 raw_spin_unlock(&kvm_lock);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002726}
2727
Alexander Graf10474ae2009-09-15 11:37:46 +02002728static void hardware_disable_all_nolock(void)
2729{
2730 BUG_ON(!kvm_usage_count);
2731
2732 kvm_usage_count--;
2733 if (!kvm_usage_count)
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002734 on_each_cpu(hardware_disable_nolock, NULL, 1);
Alexander Graf10474ae2009-09-15 11:37:46 +02002735}
2736
2737static void hardware_disable_all(void)
2738{
Jan Kiszkae935b832011-02-08 12:55:33 +01002739 raw_spin_lock(&kvm_lock);
Alexander Graf10474ae2009-09-15 11:37:46 +02002740 hardware_disable_all_nolock();
Jan Kiszkae935b832011-02-08 12:55:33 +01002741 raw_spin_unlock(&kvm_lock);
Alexander Graf10474ae2009-09-15 11:37:46 +02002742}
2743
2744static int hardware_enable_all(void)
2745{
2746 int r = 0;
2747
Jan Kiszkae935b832011-02-08 12:55:33 +01002748 raw_spin_lock(&kvm_lock);
Alexander Graf10474ae2009-09-15 11:37:46 +02002749
2750 kvm_usage_count++;
2751 if (kvm_usage_count == 1) {
2752 atomic_set(&hardware_enable_failed, 0);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002753 on_each_cpu(hardware_enable_nolock, NULL, 1);
Alexander Graf10474ae2009-09-15 11:37:46 +02002754
2755 if (atomic_read(&hardware_enable_failed)) {
2756 hardware_disable_all_nolock();
2757 r = -EBUSY;
2758 }
2759 }
2760
Jan Kiszkae935b832011-02-08 12:55:33 +01002761 raw_spin_unlock(&kvm_lock);
Alexander Graf10474ae2009-09-15 11:37:46 +02002762
2763 return r;
2764}
2765
Avi Kivity774c47f2007-02-12 00:54:47 -08002766static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2767 void *v)
2768{
2769 int cpu = (long)v;
2770
Alexander Graf10474ae2009-09-15 11:37:46 +02002771 if (!kvm_usage_count)
2772 return NOTIFY_OK;
2773
Avi Kivity1a6f4d72007-11-11 18:37:32 +02002774 val &= ~CPU_TASKS_FROZEN;
Avi Kivity774c47f2007-02-12 00:54:47 -08002775 switch (val) {
Avi Kivitycec9ad22007-05-24 13:11:41 +03002776 case CPU_DYING:
Avi Kivity6ec8a8562007-08-19 15:57:26 +03002777 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2778 cpu);
2779 hardware_disable(NULL);
2780 break;
Zachary Amsdenda908f22010-08-19 22:07:27 -10002781 case CPU_STARTING:
Jeremy Katz43934a32007-02-19 14:37:46 +02002782 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2783 cpu);
Zachary Amsdenda908f22010-08-19 22:07:27 -10002784 hardware_enable(NULL);
Avi Kivity774c47f2007-02-12 00:54:47 -08002785 break;
2786 }
2787 return NOTIFY_OK;
2788}
2789
Rusty Russell9a2b85c2007-07-17 23:17:55 +10002790static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
Mike Dayd77c26f2007-10-08 09:02:08 -04002791 void *v)
Rusty Russell9a2b85c2007-07-17 23:17:55 +10002792{
Sheng Yang8e1c1812009-04-29 11:09:04 +08002793 /*
2794 * Some (well, at least mine) BIOSes hang on reboot if
2795 * in vmx root mode.
2796 *
2797 * And Intel TXT required VMX off for all cpu when system shutdown.
2798 */
2799 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2800 kvm_rebooting = true;
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002801 on_each_cpu(hardware_disable_nolock, NULL, 1);
Rusty Russell9a2b85c2007-07-17 23:17:55 +10002802 return NOTIFY_OK;
2803}
2804
2805static struct notifier_block kvm_reboot_notifier = {
2806 .notifier_call = kvm_reboot,
2807 .priority = 0,
2808};
2809
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002810static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002811{
2812 int i;
2813
2814 for (i = 0; i < bus->dev_count; i++) {
Sasha Levin743eeb02011-07-27 16:00:48 +03002815 struct kvm_io_device *pos = bus->range[i].dev;
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002816
2817 kvm_iodevice_destructor(pos);
2818 }
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002819 kfree(bus);
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002820}
2821
Geoff Levand39369f72013-04-05 19:20:30 +00002822static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
Sasha Levin743eeb02011-07-27 16:00:48 +03002823{
2824 const struct kvm_io_range *r1 = p1;
2825 const struct kvm_io_range *r2 = p2;
2826
2827 if (r1->addr < r2->addr)
2828 return -1;
2829 if (r1->addr + r1->len > r2->addr + r2->len)
2830 return 1;
2831 return 0;
2832}
2833
Geoff Levand39369f72013-04-05 19:20:30 +00002834static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
Sasha Levin743eeb02011-07-27 16:00:48 +03002835 gpa_t addr, int len)
2836{
Sasha Levin743eeb02011-07-27 16:00:48 +03002837 bus->range[bus->dev_count++] = (struct kvm_io_range) {
2838 .addr = addr,
2839 .len = len,
2840 .dev = dev,
2841 };
2842
2843 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2844 kvm_io_bus_sort_cmp, NULL);
2845
2846 return 0;
2847}
2848
Geoff Levand39369f72013-04-05 19:20:30 +00002849static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
Sasha Levin743eeb02011-07-27 16:00:48 +03002850 gpa_t addr, int len)
2851{
2852 struct kvm_io_range *range, key;
2853 int off;
2854
2855 key = (struct kvm_io_range) {
2856 .addr = addr,
2857 .len = len,
2858 };
2859
2860 range = bsearch(&key, bus->range, bus->dev_count,
2861 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2862 if (range == NULL)
2863 return -ENOENT;
2864
2865 off = range - bus->range;
2866
2867 while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2868 off--;
2869
2870 return off;
2871}
2872
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002873/* kvm_io_bus_write - called under kvm->slots_lock */
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002874int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002875 int len, const void *val)
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002876{
Sasha Levin743eeb02011-07-27 16:00:48 +03002877 int idx;
Lai Jiangshan90d83dc2010-04-19 17:41:23 +08002878 struct kvm_io_bus *bus;
Sasha Levin743eeb02011-07-27 16:00:48 +03002879 struct kvm_io_range range;
2880
2881 range = (struct kvm_io_range) {
2882 .addr = addr,
2883 .len = len,
2884 };
Lai Jiangshan90d83dc2010-04-19 17:41:23 +08002885
2886 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
Sasha Levin743eeb02011-07-27 16:00:48 +03002887 idx = kvm_io_bus_get_first_dev(bus, addr, len);
2888 if (idx < 0)
2889 return -EOPNOTSUPP;
2890
2891 while (idx < bus->dev_count &&
2892 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2893 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002894 return 0;
Sasha Levin743eeb02011-07-27 16:00:48 +03002895 idx++;
2896 }
2897
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002898 return -EOPNOTSUPP;
2899}
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002900
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002901/* kvm_io_bus_read - called under kvm->slots_lock */
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002902int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2903 int len, void *val)
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002904{
Sasha Levin743eeb02011-07-27 16:00:48 +03002905 int idx;
Lai Jiangshan90d83dc2010-04-19 17:41:23 +08002906 struct kvm_io_bus *bus;
Sasha Levin743eeb02011-07-27 16:00:48 +03002907 struct kvm_io_range range;
2908
2909 range = (struct kvm_io_range) {
2910 .addr = addr,
2911 .len = len,
2912 };
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002913
Lai Jiangshan90d83dc2010-04-19 17:41:23 +08002914 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
Sasha Levin743eeb02011-07-27 16:00:48 +03002915 idx = kvm_io_bus_get_first_dev(bus, addr, len);
2916 if (idx < 0)
2917 return -EOPNOTSUPP;
2918
2919 while (idx < bus->dev_count &&
2920 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2921 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002922 return 0;
Sasha Levin743eeb02011-07-27 16:00:48 +03002923 idx++;
2924 }
2925
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002926 return -EOPNOTSUPP;
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002927}
2928
Marcelo Tosatti79fac952009-12-23 14:35:26 -02002929/* Caller must hold slots_lock. */
Sasha Levin743eeb02011-07-27 16:00:48 +03002930int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2931 int len, struct kvm_io_device *dev)
Michael S. Tsirkin6c474692009-06-29 22:24:26 +03002932{
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002933 struct kvm_io_bus *new_bus, *bus;
Gregory Haskins090b7af2009-07-07 17:08:44 -04002934
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002935 bus = kvm->buses[bus_idx];
Amos Konga13007162012-03-09 12:17:32 +08002936 if (bus->dev_count > NR_IOBUS_DEVS - 1)
Gregory Haskins090b7af2009-07-07 17:08:44 -04002937 return -ENOSPC;
2938
Amos Konga13007162012-03-09 12:17:32 +08002939 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2940 sizeof(struct kvm_io_range)), GFP_KERNEL);
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002941 if (!new_bus)
2942 return -ENOMEM;
Amos Konga13007162012-03-09 12:17:32 +08002943 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2944 sizeof(struct kvm_io_range)));
Sasha Levin743eeb02011-07-27 16:00:48 +03002945 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002946 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2947 synchronize_srcu_expedited(&kvm->srcu);
2948 kfree(bus);
Gregory Haskins090b7af2009-07-07 17:08:44 -04002949
2950 return 0;
2951}
2952
Marcelo Tosatti79fac952009-12-23 14:35:26 -02002953/* Caller must hold slots_lock. */
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002954int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2955 struct kvm_io_device *dev)
Gregory Haskins090b7af2009-07-07 17:08:44 -04002956{
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002957 int i, r;
2958 struct kvm_io_bus *new_bus, *bus;
Michael S. Tsirkin6c474692009-06-29 22:24:26 +03002959
Sasha Levincdfca7b2011-12-04 19:36:28 +02002960 bus = kvm->buses[bus_idx];
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002961 r = -ENOENT;
Amos Konga13007162012-03-09 12:17:32 +08002962 for (i = 0; i < bus->dev_count; i++)
2963 if (bus->range[i].dev == dev) {
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002964 r = 0;
Gregory Haskins090b7af2009-07-07 17:08:44 -04002965 break;
2966 }
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002967
Amos Konga13007162012-03-09 12:17:32 +08002968 if (r)
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002969 return r;
Amos Konga13007162012-03-09 12:17:32 +08002970
2971 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
2972 sizeof(struct kvm_io_range)), GFP_KERNEL);
2973 if (!new_bus)
2974 return -ENOMEM;
2975
2976 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
2977 new_bus->dev_count--;
2978 memcpy(new_bus->range + i, bus->range + i + 1,
2979 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002980
2981 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2982 synchronize_srcu_expedited(&kvm->srcu);
2983 kfree(bus);
2984 return r;
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002985}
2986
Avi Kivity774c47f2007-02-12 00:54:47 -08002987static struct notifier_block kvm_cpu_notifier = {
2988 .notifier_call = kvm_cpu_hotplug,
Avi Kivity774c47f2007-02-12 00:54:47 -08002989};
2990
Christoph Hellwig8b88b092008-02-08 04:20:26 -08002991static int vm_stat_get(void *_offset, u64 *val)
Avi Kivityba1389b2007-11-18 16:24:12 +02002992{
2993 unsigned offset = (long)_offset;
Avi Kivityba1389b2007-11-18 16:24:12 +02002994 struct kvm *kvm;
2995
Christoph Hellwig8b88b092008-02-08 04:20:26 -08002996 *val = 0;
Jan Kiszkae935b832011-02-08 12:55:33 +01002997 raw_spin_lock(&kvm_lock);
Avi Kivityba1389b2007-11-18 16:24:12 +02002998 list_for_each_entry(kvm, &vm_list, vm_list)
Christoph Hellwig8b88b092008-02-08 04:20:26 -08002999 *val += *(u32 *)((void *)kvm + offset);
Jan Kiszkae935b832011-02-08 12:55:33 +01003000 raw_spin_unlock(&kvm_lock);
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003001 return 0;
Avi Kivityba1389b2007-11-18 16:24:12 +02003002}
3003
3004DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3005
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003006static int vcpu_stat_get(void *_offset, u64 *val)
Avi Kivity1165f5f2007-04-19 17:27:43 +03003007{
3008 unsigned offset = (long)_offset;
Avi Kivity1165f5f2007-04-19 17:27:43 +03003009 struct kvm *kvm;
3010 struct kvm_vcpu *vcpu;
3011 int i;
3012
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003013 *val = 0;
Jan Kiszkae935b832011-02-08 12:55:33 +01003014 raw_spin_lock(&kvm_lock);
Avi Kivity1165f5f2007-04-19 17:27:43 +03003015 list_for_each_entry(kvm, &vm_list, vm_list)
Gleb Natapov988a2ca2009-06-09 15:56:29 +03003016 kvm_for_each_vcpu(i, vcpu, kvm)
3017 *val += *(u32 *)((void *)vcpu + offset);
3018
Jan Kiszkae935b832011-02-08 12:55:33 +01003019 raw_spin_unlock(&kvm_lock);
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003020 return 0;
Avi Kivity1165f5f2007-04-19 17:27:43 +03003021}
3022
Avi Kivityba1389b2007-11-18 16:24:12 +02003023DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3024
Alexey Dobriyan828c0952009-10-01 15:43:56 -07003025static const struct file_operations *stat_fops[] = {
Avi Kivityba1389b2007-11-18 16:24:12 +02003026 [KVM_STAT_VCPU] = &vcpu_stat_fops,
3027 [KVM_STAT_VM] = &vm_stat_fops,
3028};
Avi Kivity1165f5f2007-04-19 17:27:43 +03003029
Hamo4f69b682011-12-15 14:23:16 +08003030static int kvm_init_debug(void)
Avi Kivity6aa8b732006-12-10 02:21:36 -08003031{
Hamo4f69b682011-12-15 14:23:16 +08003032 int r = -EFAULT;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003033 struct kvm_stats_debugfs_item *p;
3034
Hollis Blanchard76f7c872008-04-15 16:05:42 -05003035 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
Hamo4f69b682011-12-15 14:23:16 +08003036 if (kvm_debugfs_dir == NULL)
3037 goto out;
3038
3039 for (p = debugfs_entries; p->name; ++p) {
Hollis Blanchard76f7c872008-04-15 16:05:42 -05003040 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
Avi Kivity1165f5f2007-04-19 17:27:43 +03003041 (void *)(long)p->offset,
Avi Kivityba1389b2007-11-18 16:24:12 +02003042 stat_fops[p->kind]);
Hamo4f69b682011-12-15 14:23:16 +08003043 if (p->dentry == NULL)
3044 goto out_dir;
3045 }
3046
3047 return 0;
3048
3049out_dir:
3050 debugfs_remove_recursive(kvm_debugfs_dir);
3051out:
3052 return r;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003053}
3054
3055static void kvm_exit_debug(void)
3056{
3057 struct kvm_stats_debugfs_item *p;
3058
3059 for (p = debugfs_entries; p->name; ++p)
3060 debugfs_remove(p->dentry);
Hollis Blanchard76f7c872008-04-15 16:05:42 -05003061 debugfs_remove(kvm_debugfs_dir);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003062}
3063
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003064static int kvm_suspend(void)
Avi Kivity59ae6c62007-02-12 00:54:48 -08003065{
Alexander Graf10474ae2009-09-15 11:37:46 +02003066 if (kvm_usage_count)
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09003067 hardware_disable_nolock(NULL);
Avi Kivity59ae6c62007-02-12 00:54:48 -08003068 return 0;
3069}
3070
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003071static void kvm_resume(void)
Avi Kivity59ae6c62007-02-12 00:54:48 -08003072{
Zachary Amsdenca84d1a2010-08-19 22:07:28 -10003073 if (kvm_usage_count) {
Jan Kiszkae935b832011-02-08 12:55:33 +01003074 WARN_ON(raw_spin_is_locked(&kvm_lock));
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09003075 hardware_enable_nolock(NULL);
Zachary Amsdenca84d1a2010-08-19 22:07:28 -10003076 }
Avi Kivity59ae6c62007-02-12 00:54:48 -08003077}
3078
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003079static struct syscore_ops kvm_syscore_ops = {
Avi Kivity59ae6c62007-02-12 00:54:48 -08003080 .suspend = kvm_suspend,
3081 .resume = kvm_resume,
3082};
3083
Avi Kivity15ad7142007-07-11 18:17:21 +03003084static inline
3085struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3086{
3087 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3088}
3089
3090static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3091{
3092 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
Raghavendra K T3a08a8f2013-03-04 23:32:07 +05303093 if (vcpu->preempted)
3094 vcpu->preempted = false;
Avi Kivity15ad7142007-07-11 18:17:21 +03003095
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003096 kvm_arch_vcpu_load(vcpu, cpu);
Avi Kivity15ad7142007-07-11 18:17:21 +03003097}
3098
3099static void kvm_sched_out(struct preempt_notifier *pn,
3100 struct task_struct *next)
3101{
3102 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3103
Raghavendra K T3a08a8f2013-03-04 23:32:07 +05303104 if (current->state == TASK_RUNNING)
3105 vcpu->preempted = true;
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003106 kvm_arch_vcpu_put(vcpu);
Avi Kivity15ad7142007-07-11 18:17:21 +03003107}
3108
Avi Kivity0ee75be2010-04-28 15:39:01 +03003109int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
Rusty Russellc16f8622007-07-30 21:12:19 +10003110 struct module *module)
Avi Kivity6aa8b732006-12-10 02:21:36 -08003111{
3112 int r;
Yang, Sheng002c7f72007-07-31 14:23:01 +03003113 int cpu;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003114
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08003115 r = kvm_arch_init(opaque);
3116 if (r)
Zhang Xiantaod23087842007-11-29 15:35:39 +08003117 goto out_fail;
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08003118
Asias He7dac16c2013-05-08 10:57:29 +08003119 /*
3120 * kvm_arch_init makes sure there's at most one caller
3121 * for architectures that support multiple implementations,
3122 * like intel and amd on x86.
3123 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3124 * conflicts in case kvm is already setup for another implementation.
3125 */
3126 r = kvm_irqfd_init();
3127 if (r)
3128 goto out_irqfd;
3129
Avi Kivity8437a612009-06-06 14:52:35 -07003130 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
Rusty Russell7f59f492008-12-07 21:25:45 +10303131 r = -ENOMEM;
3132 goto out_free_0;
3133 }
3134
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003135 r = kvm_arch_hardware_setup();
Avi Kivity6aa8b732006-12-10 02:21:36 -08003136 if (r < 0)
Rusty Russell7f59f492008-12-07 21:25:45 +10303137 goto out_free_0a;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003138
Yang, Sheng002c7f72007-07-31 14:23:01 +03003139 for_each_online_cpu(cpu) {
3140 smp_call_function_single(cpu,
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003141 kvm_arch_check_processor_compat,
Jens Axboe8691e5a2008-06-06 11:18:06 +02003142 &r, 1);
Yang, Sheng002c7f72007-07-31 14:23:01 +03003143 if (r < 0)
Zhang Xiantaod23087842007-11-29 15:35:39 +08003144 goto out_free_1;
Yang, Sheng002c7f72007-07-31 14:23:01 +03003145 }
3146
Avi Kivity774c47f2007-02-12 00:54:47 -08003147 r = register_cpu_notifier(&kvm_cpu_notifier);
3148 if (r)
Zhang Xiantaod23087842007-11-29 15:35:39 +08003149 goto out_free_2;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003150 register_reboot_notifier(&kvm_reboot_notifier);
3151
Rusty Russellc16f8622007-07-30 21:12:19 +10003152 /* A kmem cache lets us meet the alignment requirements of fx_save. */
Avi Kivity0ee75be2010-04-28 15:39:01 +03003153 if (!vcpu_align)
3154 vcpu_align = __alignof__(struct kvm_vcpu);
3155 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
Joe Perches56919c52007-11-12 20:06:51 -08003156 0, NULL);
Rusty Russellc16f8622007-07-30 21:12:19 +10003157 if (!kvm_vcpu_cache) {
3158 r = -ENOMEM;
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003159 goto out_free_3;
Rusty Russellc16f8622007-07-30 21:12:19 +10003160 }
3161
Gleb Natapovaf585b92010-10-14 11:22:46 +02003162 r = kvm_async_pf_init();
3163 if (r)
3164 goto out_free;
3165
Avi Kivity6aa8b732006-12-10 02:21:36 -08003166 kvm_chardev_ops.owner = module;
Christian Borntraeger3d3aab12008-12-02 11:17:32 +01003167 kvm_vm_fops.owner = module;
3168 kvm_vcpu_fops.owner = module;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003169
3170 r = misc_register(&kvm_dev);
3171 if (r) {
Mike Dayd77c26f2007-10-08 09:02:08 -04003172 printk(KERN_ERR "kvm: misc device register failed\n");
Gleb Natapovaf585b92010-10-14 11:22:46 +02003173 goto out_unreg;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003174 }
3175
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003176 register_syscore_ops(&kvm_syscore_ops);
3177
Avi Kivity15ad7142007-07-11 18:17:21 +03003178 kvm_preempt_ops.sched_in = kvm_sched_in;
3179 kvm_preempt_ops.sched_out = kvm_sched_out;
3180
Hamo4f69b682011-12-15 14:23:16 +08003181 r = kvm_init_debug();
3182 if (r) {
3183 printk(KERN_ERR "kvm: create debugfs files failed\n");
3184 goto out_undebugfs;
3185 }
Darrick J. Wong0ea4ed82009-10-14 16:21:00 -07003186
Avi Kivityc7addb92007-09-16 18:58:32 +02003187 return 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003188
Hamo4f69b682011-12-15 14:23:16 +08003189out_undebugfs:
3190 unregister_syscore_ops(&kvm_syscore_ops);
Gleb Natapovaf585b92010-10-14 11:22:46 +02003191out_unreg:
3192 kvm_async_pf_deinit();
Avi Kivity6aa8b732006-12-10 02:21:36 -08003193out_free:
Rusty Russellc16f8622007-07-30 21:12:19 +10003194 kmem_cache_destroy(kvm_vcpu_cache);
Zhang Xiantaod23087842007-11-29 15:35:39 +08003195out_free_3:
Avi Kivity6aa8b732006-12-10 02:21:36 -08003196 unregister_reboot_notifier(&kvm_reboot_notifier);
Avi Kivity774c47f2007-02-12 00:54:47 -08003197 unregister_cpu_notifier(&kvm_cpu_notifier);
Zhang Xiantaod23087842007-11-29 15:35:39 +08003198out_free_2:
Zhang Xiantaod23087842007-11-29 15:35:39 +08003199out_free_1:
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003200 kvm_arch_hardware_unsetup();
Rusty Russell7f59f492008-12-07 21:25:45 +10303201out_free_0a:
3202 free_cpumask_var(cpus_hardware_enabled);
Zhang Xiantaod23087842007-11-29 15:35:39 +08003203out_free_0:
Cornelia Hucka0f155e2013-02-28 12:33:18 +01003204 kvm_irqfd_exit();
3205out_irqfd:
Asias He7dac16c2013-05-08 10:57:29 +08003206 kvm_arch_exit();
3207out_fail:
Avi Kivity6aa8b732006-12-10 02:21:36 -08003208 return r;
3209}
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08003210EXPORT_SYMBOL_GPL(kvm_init);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003211
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08003212void kvm_exit(void)
Avi Kivity6aa8b732006-12-10 02:21:36 -08003213{
Darrick J. Wong0ea4ed82009-10-14 16:21:00 -07003214 kvm_exit_debug();
Avi Kivity6aa8b732006-12-10 02:21:36 -08003215 misc_deregister(&kvm_dev);
Rusty Russellc16f8622007-07-30 21:12:19 +10003216 kmem_cache_destroy(kvm_vcpu_cache);
Gleb Natapovaf585b92010-10-14 11:22:46 +02003217 kvm_async_pf_deinit();
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003218 unregister_syscore_ops(&kvm_syscore_ops);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003219 unregister_reboot_notifier(&kvm_reboot_notifier);
Avi Kivity59ae6c62007-02-12 00:54:48 -08003220 unregister_cpu_notifier(&kvm_cpu_notifier);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09003221 on_each_cpu(hardware_disable_nolock, NULL, 1);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003222 kvm_arch_hardware_unsetup();
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08003223 kvm_arch_exit();
Cornelia Hucka0f155e2013-02-28 12:33:18 +01003224 kvm_irqfd_exit();
Rusty Russell7f59f492008-12-07 21:25:45 +10303225 free_cpumask_var(cpus_hardware_enabled);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003226}
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08003227EXPORT_SYMBOL_GPL(kvm_exit);